1
# Copyright (C) 2005, 2006 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
81
from bzrlib.errors import (
89
RevisionAlreadyPresent,
91
from bzrlib.tuned_gzip import GzipFile
92
from bzrlib.trace import mutter
93
from bzrlib.osutils import (
98
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
99
from bzrlib.tsort import topo_sort
102
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
105
# TODO: Split out code specific to this format into an associated object.
107
# TODO: Can we put in some kind of value to check that the index and data
108
# files belong together?
110
# TODO: accommodate binaries, perhaps by storing a byte count
112
# TODO: function to check whole file
114
# TODO: atomically append data, then measure backwards from the cursor
115
# position after writing to work out where it was located. we may need to
116
# bypass python file buffering.
118
DATA_SUFFIX = '.knit'
119
INDEX_SUFFIX = '.kndx'
122
class KnitContent(object):
123
"""Content of a knit version to which deltas can be applied."""
125
def __init__(self, lines):
128
def annotate_iter(self):
129
"""Yield tuples of (origin, text) for each content line."""
130
return iter(self._lines)
133
"""Return a list of (origin, text) tuples."""
134
return list(self.annotate_iter())
136
def line_delta_iter(self, new_lines):
137
"""Generate line-based delta from this content to new_lines."""
138
new_texts = new_lines.text()
139
old_texts = self.text()
140
s = KnitSequenceMatcher(None, old_texts, new_texts)
141
for tag, i1, i2, j1, j2 in s.get_opcodes():
144
# ofrom, oto, length, data
145
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
147
def line_delta(self, new_lines):
148
return list(self.line_delta_iter(new_lines))
151
return [text for origin, text in self._lines]
154
return KnitContent(self._lines[:])
157
class _KnitFactory(object):
158
"""Base factory for creating content objects."""
160
def make(self, lines, version_id):
161
num_lines = len(lines)
162
return KnitContent(zip([version_id] * num_lines, lines))
165
class KnitAnnotateFactory(_KnitFactory):
166
"""Factory for creating annotated Content objects."""
170
def parse_fulltext(self, content, version_id):
171
"""Convert fulltext to internal representation
173
fulltext content is of the format
174
revid(utf8) plaintext\n
175
internal representation is of the format:
178
# TODO: jam 20070209 The tests expect this to be returned as tuples,
179
# but the code itself doesn't really depend on that.
180
# Figure out a way to not require the overhead of turning the
181
# list back into tuples.
182
lines = [tuple(line.split(' ', 1)) for line in content]
183
return KnitContent(lines)
185
def parse_line_delta_iter(self, lines):
186
return iter(self.parse_line_delta(lines))
188
def parse_line_delta(self, lines, version_id):
189
"""Convert a line based delta into internal representation.
191
line delta is in the form of:
192
intstart intend intcount
194
revid(utf8) newline\n
195
internal representation is
196
(start, end, count, [1..count tuples (revid, newline)])
203
def cache_and_return(line):
204
origin, text = line.split(' ', 1)
205
return cache.setdefault(origin, origin), text
207
# walk through the lines parsing.
209
start, end, count = [int(n) for n in header.split(',')]
210
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
211
result.append((start, end, count, contents))
214
def get_fulltext_content(self, lines):
215
"""Extract just the content lines from a fulltext."""
216
return (line.split(' ', 1)[1] for line in lines)
218
def get_linedelta_content(self, lines):
219
"""Extract just the content from a line delta.
221
This doesn't return all of the extra information stored in a delta.
222
Only the actual content lines.
227
header = header.split(',')
228
count = int(header[2])
229
for i in xrange(count):
230
origin, text = next().split(' ', 1)
233
def lower_fulltext(self, content):
234
"""convert a fulltext content record into a serializable form.
236
see parse_fulltext which this inverts.
238
# TODO: jam 20070209 We only do the caching thing to make sure that
239
# the origin is a valid utf-8 line, eventually we could remove it
240
return ['%s %s' % (o, t) for o, t in content._lines]
242
def lower_line_delta(self, delta):
243
"""convert a delta into a serializable form.
245
See parse_line_delta which this inverts.
247
# TODO: jam 20070209 We only do the caching thing to make sure that
248
# the origin is a valid utf-8 line, eventually we could remove it
250
for start, end, c, lines in delta:
251
out.append('%d,%d,%d\n' % (start, end, c))
252
out.extend(origin + ' ' + text
253
for origin, text in lines)
257
class KnitPlainFactory(_KnitFactory):
258
"""Factory for creating plain Content objects."""
262
def parse_fulltext(self, content, version_id):
263
"""This parses an unannotated fulltext.
265
Note that this is not a noop - the internal representation
266
has (versionid, line) - its just a constant versionid.
268
return self.make(content, version_id)
270
def parse_line_delta_iter(self, lines, version_id):
272
num_lines = len(lines)
273
while cur < num_lines:
276
start, end, c = [int(n) for n in header.split(',')]
277
yield start, end, c, zip([version_id] * c, lines[cur:cur+c])
280
def parse_line_delta(self, lines, version_id):
281
return list(self.parse_line_delta_iter(lines, version_id))
283
def get_fulltext_content(self, lines):
284
"""Extract just the content lines from a fulltext."""
287
def get_linedelta_content(self, lines):
288
"""Extract just the content from a line delta.
290
This doesn't return all of the extra information stored in a delta.
291
Only the actual content lines.
296
header = header.split(',')
297
count = int(header[2])
298
for i in xrange(count):
301
def lower_fulltext(self, content):
302
return content.text()
304
def lower_line_delta(self, delta):
306
for start, end, c, lines in delta:
307
out.append('%d,%d,%d\n' % (start, end, c))
308
out.extend([text for origin, text in lines])
312
def make_empty_knit(transport, relpath):
313
"""Construct a empty knit at the specified location."""
314
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
318
class KnitVersionedFile(VersionedFile):
319
"""Weave-like structure with faster random access.
321
A knit stores a number of texts and a summary of the relationships
322
between them. Texts are identified by a string version-id. Texts
323
are normally stored and retrieved as a series of lines, but can
324
also be passed as single strings.
326
Lines are stored with the trailing newline (if any) included, to
327
avoid special cases for files with no final newline. Lines are
328
composed of 8-bit characters, not unicode. The combination of
329
these approaches should mean any 'binary' file can be safely
330
stored and retrieved.
333
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
334
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
335
create=False, create_parent_dir=False, delay_create=False,
337
"""Construct a knit at location specified by relpath.
339
:param create: If not True, only open an existing knit.
340
:param create_parent_dir: If True, create the parent directory if
341
creating the file fails. (This is used for stores with
342
hash-prefixes that may not exist yet)
343
:param delay_create: The calling code is aware that the knit won't
344
actually be created until the first data is stored.
346
if deprecated_passed(basis_knit):
347
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
348
" deprecated as of bzr 0.9.",
349
DeprecationWarning, stacklevel=2)
350
if access_mode is None:
352
super(KnitVersionedFile, self).__init__(access_mode)
353
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
354
self.transport = transport
355
self.filename = relpath
356
self.factory = factory or KnitAnnotateFactory()
357
self.writable = (access_mode == 'w')
360
self._max_delta_chain = 200
362
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
363
access_mode, create=create, file_mode=file_mode,
364
create_parent_dir=create_parent_dir, delay_create=delay_create,
366
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
367
access_mode, create=create and not len(self), file_mode=file_mode,
368
create_parent_dir=create_parent_dir, delay_create=delay_create,
372
return '%s(%s)' % (self.__class__.__name__,
373
self.transport.abspath(self.filename))
375
def _check_should_delta(self, first_parents):
376
"""Iterate back through the parent listing, looking for a fulltext.
378
This is used when we want to decide whether to add a delta or a new
379
fulltext. It searches for _max_delta_chain parents. When it finds a
380
fulltext parent, it sees if the total size of the deltas leading up to
381
it is large enough to indicate that we want a new full text anyway.
383
Return True if we should create a new delta, False if we should use a
388
delta_parents = first_parents
389
for count in xrange(self._max_delta_chain):
390
parent = delta_parents[0]
391
method = self._index.get_method(parent)
392
pos, size = self._index.get_position(parent)
393
if method == 'fulltext':
397
delta_parents = self._index.get_parents(parent)
399
# We couldn't find a fulltext, so we must create a new one
402
return fulltext_size > delta_size
404
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
405
"""See VersionedFile._add_delta()."""
406
self._check_add(version_id, []) # should we check the lines ?
407
self._check_versions_present(parents)
411
for parent in parents:
412
if not self.has_version(parent):
413
ghosts.append(parent)
415
present_parents.append(parent)
417
if delta_parent is None:
418
# reconstitute as full text.
419
assert len(delta) == 1 or len(delta) == 0
421
assert delta[0][0] == 0
422
assert delta[0][1] == 0, delta[0][1]
423
return super(KnitVersionedFile, self)._add_delta(version_id,
434
options.append('no-eol')
436
if delta_parent is not None:
437
# determine the current delta chain length.
438
# To speed the extract of texts the delta chain is limited
439
# to a fixed number of deltas. This should minimize both
440
# I/O and the time spend applying deltas.
441
# The window was changed to a maximum of 200 deltas, but also added
442
# was a check that the total compressed size of the deltas is
443
# smaller than the compressed size of the fulltext.
444
if not self._check_should_delta([delta_parent]):
445
# We don't want a delta here, just do a normal insertion.
446
return super(KnitVersionedFile, self)._add_delta(version_id,
453
options.append('line-delta')
454
store_lines = self.factory.lower_line_delta(delta)
456
where, size = self._data.add_record(version_id, digest, store_lines)
457
self._index.add_version(version_id, options, where, size, parents)
459
def _add_raw_records(self, records, data):
460
"""Add all the records 'records' with data pre-joined in 'data'.
462
:param records: A list of tuples(version_id, options, parents, size).
463
:param data: The data for the records. When it is written, the records
464
are adjusted to have pos pointing into data by the sum of
465
the preceding records sizes.
468
pos = self._data.add_raw_record(data)
471
for (version_id, options, parents, size) in records:
472
index_entries.append((version_id, options, pos+offset,
474
if self._data._do_cache:
475
self._data._cache[version_id] = data[offset:offset+size]
477
self._index.add_versions(index_entries)
479
def enable_cache(self):
480
"""Start caching data for this knit"""
481
self._data.enable_cache()
483
def clear_cache(self):
484
"""Clear the data cache only."""
485
self._data.clear_cache()
487
def copy_to(self, name, transport):
488
"""See VersionedFile.copy_to()."""
489
# copy the current index to a temp index to avoid racing with local
491
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
492
self.transport.get(self._index._filename))
494
f = self._data._open_file()
496
transport.put_file(name + DATA_SUFFIX, f)
499
# move the copied index into place
500
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
502
def create_empty(self, name, transport, mode=None):
503
return KnitVersionedFile(name, transport, factory=self.factory,
504
delta=self.delta, create=True)
506
def _fix_parents(self, version_id, new_parents):
507
"""Fix the parents list for version.
509
This is done by appending a new version to the index
510
with identical data except for the parents list.
511
the parents list must be a superset of the current
514
current_values = self._index._cache[version_id]
515
assert set(current_values[4]).difference(set(new_parents)) == set()
516
self._index.add_version(version_id,
522
def get_delta(self, version_id):
523
"""Get a delta for constructing version from some other version."""
524
version_id = osutils.safe_revision_id(version_id)
525
self.check_not_reserved_id(version_id)
526
if not self.has_version(version_id):
527
raise RevisionNotPresent(version_id, self.filename)
529
parents = self.get_parents(version_id)
534
data_pos, data_size = self._index.get_position(version_id)
535
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
536
noeol = 'no-eol' in self._index.get_options(version_id)
537
if 'fulltext' == self._index.get_method(version_id):
538
new_content = self.factory.parse_fulltext(data, version_id)
539
if parent is not None:
540
reference_content = self._get_content(parent)
541
old_texts = reference_content.text()
544
new_texts = new_content.text()
545
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
546
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
548
delta = self.factory.parse_line_delta(data, version_id)
549
return parent, sha1, noeol, delta
551
def get_graph_with_ghosts(self):
552
"""See VersionedFile.get_graph_with_ghosts()."""
553
graph_items = self._index.get_graph()
554
return dict(graph_items)
556
def get_sha1(self, version_id):
557
"""See VersionedFile.get_sha1()."""
558
version_id = osutils.safe_revision_id(version_id)
559
record_map = self._get_record_map([version_id])
560
method, content, digest, next = record_map[version_id]
565
"""See VersionedFile.get_suffixes()."""
566
return [DATA_SUFFIX, INDEX_SUFFIX]
568
def has_ghost(self, version_id):
569
"""True if there is a ghost reference in the file to version_id."""
570
version_id = osutils.safe_revision_id(version_id)
572
if self.has_version(version_id):
574
# optimisable if needed by memoising the _ghosts set.
575
items = self._index.get_graph()
576
for node, parents in items:
577
for parent in parents:
578
if parent not in self._index._cache:
579
if parent == version_id:
584
"""See VersionedFile.versions."""
585
return self._index.get_versions()
587
def has_version(self, version_id):
588
"""See VersionedFile.has_version."""
589
version_id = osutils.safe_revision_id(version_id)
590
return self._index.has_version(version_id)
592
__contains__ = has_version
594
def _merge_annotations(self, content, parents, parent_texts={},
595
delta=None, annotated=None):
596
"""Merge annotations for content. This is done by comparing
597
the annotations based on changed to the text.
601
for parent_id in parents:
602
merge_content = self._get_content(parent_id, parent_texts)
603
seq = patiencediff.PatienceSequenceMatcher(
604
None, merge_content.text(), content.text())
605
if delta_seq is None:
606
# setup a delta seq to reuse.
608
for i, j, n in seq.get_matching_blocks():
611
# this appears to copy (origin, text) pairs across to the new
612
# content for any line that matches the last-checked parent.
613
# FIXME: save the sequence control data for delta compression
614
# against the most relevant parent rather than rediffing.
615
content._lines[j:j+n] = merge_content._lines[i:i+n]
618
reference_content = self._get_content(parents[0], parent_texts)
619
new_texts = content.text()
620
old_texts = reference_content.text()
621
delta_seq = patiencediff.PatienceSequenceMatcher(
622
None, old_texts, new_texts)
623
return self._make_line_delta(delta_seq, content)
625
def _make_line_delta(self, delta_seq, new_content):
626
"""Generate a line delta from delta_seq and new_content."""
628
for op in delta_seq.get_opcodes():
631
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
634
def _get_components_positions(self, version_ids):
635
"""Produce a map of position data for the components of versions.
637
This data is intended to be used for retrieving the knit records.
639
A dict of version_id to (method, data_pos, data_size, next) is
641
method is the way referenced data should be applied.
642
data_pos is the position of the data in the knit.
643
data_size is the size of the data in the knit.
644
next is the build-parent of the version, or None for fulltexts.
647
for version_id in version_ids:
650
while cursor is not None and cursor not in component_data:
651
method = self._index.get_method(cursor)
652
if method == 'fulltext':
655
next = self.get_parents(cursor)[0]
656
data_pos, data_size = self._index.get_position(cursor)
657
component_data[cursor] = (method, data_pos, data_size, next)
659
return component_data
661
def _get_content(self, version_id, parent_texts={}):
662
"""Returns a content object that makes up the specified
664
if not self.has_version(version_id):
665
raise RevisionNotPresent(version_id, self.filename)
667
cached_version = parent_texts.get(version_id, None)
668
if cached_version is not None:
669
return cached_version
671
text_map, contents_map = self._get_content_maps([version_id])
672
return contents_map[version_id]
674
def _check_versions_present(self, version_ids):
675
"""Check that all specified versions are present."""
676
self._index.check_versions_present(version_ids)
678
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
679
"""See VersionedFile.add_lines_with_ghosts()."""
680
self._check_add(version_id, lines)
681
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
683
def _add_lines(self, version_id, parents, lines, parent_texts):
684
"""See VersionedFile.add_lines."""
685
self._check_add(version_id, lines)
686
self._check_versions_present(parents)
687
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
689
def _check_add(self, version_id, lines):
690
"""check that version_id and lines are safe to add."""
691
assert self.writable, "knit is not opened for write"
692
### FIXME escape. RBC 20060228
693
if contains_whitespace(version_id):
694
raise InvalidRevisionId(version_id, self.filename)
695
self.check_not_reserved_id(version_id)
696
if self.has_version(version_id):
697
raise RevisionAlreadyPresent(version_id, self.filename)
698
self._check_lines_not_unicode(lines)
699
self._check_lines_are_lines(lines)
701
def _add(self, version_id, lines, parents, delta, parent_texts):
702
"""Add a set of lines on top of version specified by parents.
704
If delta is true, compress the text as a line-delta against
707
Any versions not present will be converted into ghosts.
709
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
710
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
711
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
712
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
713
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
714
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
715
# +1383 0 8.0370 8.0370 +<len>
716
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
717
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
718
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
719
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
723
if parent_texts is None:
725
for parent in parents:
726
if not self.has_version(parent):
727
ghosts.append(parent)
729
present_parents.append(parent)
731
if delta and not len(present_parents):
734
digest = sha_strings(lines)
737
if lines[-1][-1] != '\n':
738
options.append('no-eol')
739
lines[-1] = lines[-1] + '\n'
741
if len(present_parents) and delta:
742
# To speed the extract of texts the delta chain is limited
743
# to a fixed number of deltas. This should minimize both
744
# I/O and the time spend applying deltas.
745
delta = self._check_should_delta(present_parents)
747
assert isinstance(version_id, str)
748
lines = self.factory.make(lines, version_id)
749
if delta or (self.factory.annotated and len(present_parents) > 0):
750
# Merge annotations from parent texts if so is needed.
751
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
752
delta, self.factory.annotated)
755
options.append('line-delta')
756
store_lines = self.factory.lower_line_delta(delta_hunks)
758
options.append('fulltext')
759
store_lines = self.factory.lower_fulltext(lines)
761
where, size = self._data.add_record(version_id, digest, store_lines)
762
self._index.add_version(version_id, options, where, size, parents)
765
def check(self, progress_bar=None):
766
"""See VersionedFile.check()."""
768
def _clone_text(self, new_version_id, old_version_id, parents):
769
"""See VersionedFile.clone_text()."""
770
# FIXME RBC 20060228 make fast by only inserting an index with null
772
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
774
def get_lines(self, version_id):
775
"""See VersionedFile.get_lines()."""
776
return self.get_line_list([version_id])[0]
778
def _get_record_map(self, version_ids):
779
"""Produce a dictionary of knit records.
781
The keys are version_ids, the values are tuples of (method, content,
783
method is the way the content should be applied.
784
content is a KnitContent object.
785
digest is the SHA1 digest of this version id after all steps are done
786
next is the build-parent of the version, i.e. the leftmost ancestor.
787
If the method is fulltext, next will be None.
789
position_map = self._get_components_positions(version_ids)
790
# c = component_id, m = method, p = position, s = size, n = next
791
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
793
for component_id, content, digest in \
794
self._data.read_records_iter(records):
795
method, position, size, next = position_map[component_id]
796
record_map[component_id] = method, content, digest, next
800
def get_text(self, version_id):
801
"""See VersionedFile.get_text"""
802
return self.get_texts([version_id])[0]
804
def get_texts(self, version_ids):
805
return [''.join(l) for l in self.get_line_list(version_ids)]
807
def get_line_list(self, version_ids):
808
"""Return the texts of listed versions as a list of strings."""
809
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
810
for version_id in version_ids:
811
self.check_not_reserved_id(version_id)
812
text_map, content_map = self._get_content_maps(version_ids)
813
return [text_map[v] for v in version_ids]
815
def _get_content_maps(self, version_ids):
816
"""Produce maps of text and KnitContents
818
:return: (text_map, content_map) where text_map contains the texts for
819
the requested versions and content_map contains the KnitContents.
820
Both dicts take version_ids as their keys.
822
for version_id in version_ids:
823
if not self.has_version(version_id):
824
raise RevisionNotPresent(version_id, self.filename)
825
record_map = self._get_record_map(version_ids)
830
for version_id in version_ids:
833
while cursor is not None:
834
method, data, digest, next = record_map[cursor]
835
components.append((cursor, method, data, digest))
836
if cursor in content_map:
841
for component_id, method, data, digest in reversed(components):
842
if component_id in content_map:
843
content = content_map[component_id]
845
if method == 'fulltext':
846
assert content is None
847
content = self.factory.parse_fulltext(data, version_id)
848
elif method == 'line-delta':
849
delta = self.factory.parse_line_delta(data, version_id)
850
content = content.copy()
851
content._lines = self._apply_delta(content._lines,
853
content_map[component_id] = content
855
if 'no-eol' in self._index.get_options(version_id):
856
content = content.copy()
857
line = content._lines[-1][1].rstrip('\n')
858
content._lines[-1] = (content._lines[-1][0], line)
859
final_content[version_id] = content
861
# digest here is the digest from the last applied component.
862
text = content.text()
863
if sha_strings(text) != digest:
864
raise KnitCorrupt(self.filename,
865
'sha-1 does not match %s' % version_id)
867
text_map[version_id] = text
868
return text_map, final_content
870
def iter_lines_added_or_present_in_versions(self, version_ids=None,
872
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
873
if version_ids is None:
874
version_ids = self.versions()
876
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
878
pb = progress.DummyProgress()
879
# we don't care about inclusions, the caller cares.
880
# but we need to setup a list of records to visit.
881
# we need version_id, position, length
882
version_id_records = []
883
requested_versions = set(version_ids)
884
# filter for available versions
885
for version_id in requested_versions:
886
if not self.has_version(version_id):
887
raise RevisionNotPresent(version_id, self.filename)
888
# get a in-component-order queue:
889
for version_id in self.versions():
890
if version_id in requested_versions:
891
data_pos, length = self._index.get_position(version_id)
892
version_id_records.append((version_id, data_pos, length))
894
total = len(version_id_records)
895
for version_idx, (version_id, data, sha_value) in \
896
enumerate(self._data.read_records_iter(version_id_records)):
897
pb.update('Walking content.', version_idx, total)
898
method = self._index.get_method(version_id)
900
assert method in ('fulltext', 'line-delta')
901
if method == 'fulltext':
902
line_iterator = self.factory.get_fulltext_content(data)
904
line_iterator = self.factory.get_linedelta_content(data)
905
for line in line_iterator:
908
pb.update('Walking content.', total, total)
910
def num_versions(self):
911
"""See VersionedFile.num_versions()."""
912
return self._index.num_versions()
914
__len__ = num_versions
916
def annotate_iter(self, version_id):
917
"""See VersionedFile.annotate_iter."""
918
version_id = osutils.safe_revision_id(version_id)
919
content = self._get_content(version_id)
920
for origin, text in content.annotate_iter():
923
def get_parents(self, version_id):
924
"""See VersionedFile.get_parents."""
927
# 52554 calls in 1264 872 internal down from 3674
928
version_id = osutils.safe_revision_id(version_id)
930
return self._index.get_parents(version_id)
932
raise RevisionNotPresent(version_id, self.filename)
934
def get_parents_with_ghosts(self, version_id):
935
"""See VersionedFile.get_parents."""
936
version_id = osutils.safe_revision_id(version_id)
938
return self._index.get_parents_with_ghosts(version_id)
940
raise RevisionNotPresent(version_id, self.filename)
942
def get_ancestry(self, versions):
943
"""See VersionedFile.get_ancestry."""
944
if isinstance(versions, basestring):
945
versions = [versions]
948
versions = [osutils.safe_revision_id(v) for v in versions]
949
return self._index.get_ancestry(versions)
951
def get_ancestry_with_ghosts(self, versions):
952
"""See VersionedFile.get_ancestry_with_ghosts."""
953
if isinstance(versions, basestring):
954
versions = [versions]
957
versions = [osutils.safe_revision_id(v) for v in versions]
958
return self._index.get_ancestry_with_ghosts(versions)
960
#@deprecated_method(zero_eight)
961
def walk(self, version_ids):
962
"""See VersionedFile.walk."""
963
# We take the short path here, and extract all relevant texts
964
# and put them in a weave and let that do all the work. Far
965
# from optimal, but is much simpler.
966
# FIXME RB 20060228 this really is inefficient!
967
from bzrlib.weave import Weave
969
w = Weave(self.filename)
970
ancestry = self.get_ancestry(version_ids)
971
sorted_graph = topo_sort(self._index.get_graph())
972
version_list = [vid for vid in sorted_graph if vid in ancestry]
974
for version_id in version_list:
975
lines = self.get_lines(version_id)
976
w.add_lines(version_id, self.get_parents(version_id), lines)
978
for lineno, insert_id, dset, line in w.walk(version_ids):
979
yield lineno, insert_id, dset, line
981
def plan_merge(self, ver_a, ver_b):
982
"""See VersionedFile.plan_merge."""
983
ver_a = osutils.safe_revision_id(ver_a)
984
ver_b = osutils.safe_revision_id(ver_b)
985
ancestors_b = set(self.get_ancestry(ver_b))
986
def status_a(revision, text):
987
if revision in ancestors_b:
988
return 'killed-b', text
992
ancestors_a = set(self.get_ancestry(ver_a))
993
def status_b(revision, text):
994
if revision in ancestors_a:
995
return 'killed-a', text
999
annotated_a = self.annotate(ver_a)
1000
annotated_b = self.annotate(ver_b)
1001
plain_a = [t for (a, t) in annotated_a]
1002
plain_b = [t for (a, t) in annotated_b]
1003
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
1006
for ai, bi, l in blocks:
1007
# process all mismatched sections
1008
# (last mismatched section is handled because blocks always
1009
# includes a 0-length last block)
1010
for revision, text in annotated_a[a_cur:ai]:
1011
yield status_a(revision, text)
1012
for revision, text in annotated_b[b_cur:bi]:
1013
yield status_b(revision, text)
1015
# and now the matched section
1018
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
1019
assert text_a == text_b
1020
yield "unchanged", text_a
1023
class _KnitComponentFile(object):
1024
"""One of the files used to implement a knit database"""
1026
def __init__(self, transport, filename, mode, file_mode=None,
1027
create_parent_dir=False, dir_mode=None):
1028
self._transport = transport
1029
self._filename = filename
1031
self._file_mode = file_mode
1032
self._dir_mode = dir_mode
1033
self._create_parent_dir = create_parent_dir
1034
self._need_to_create = False
1036
def _full_path(self):
1037
"""Return the full path to this file."""
1038
return self._transport.base + self._filename
1040
def check_header(self, fp):
1041
line = fp.readline()
1043
# An empty file can actually be treated as though the file doesn't
1045
raise errors.NoSuchFile(self._full_path())
1046
if line != self.HEADER:
1047
raise KnitHeaderError(badline=line,
1048
filename=self._transport.abspath(self._filename))
1051
"""Commit is a nop."""
1054
return '%s(%s)' % (self.__class__.__name__, self._filename)
1057
class _KnitIndex(_KnitComponentFile):
1058
"""Manages knit index file.
1060
The index is already kept in memory and read on startup, to enable
1061
fast lookups of revision information. The cursor of the index
1062
file is always pointing to the end, making it easy to append
1065
_cache is a cache for fast mapping from version id to a Index
1068
_history is a cache for fast mapping from indexes to version ids.
1070
The index data format is dictionary compressed when it comes to
1071
parent references; a index entry may only have parents that with a
1072
lover index number. As a result, the index is topological sorted.
1074
Duplicate entries may be written to the index for a single version id
1075
if this is done then the latter one completely replaces the former:
1076
this allows updates to correct version and parent information.
1077
Note that the two entries may share the delta, and that successive
1078
annotations and references MUST point to the first entry.
1080
The index file on disc contains a header, followed by one line per knit
1081
record. The same revision can be present in an index file more than once.
1082
The first occurrence gets assigned a sequence number starting from 0.
1084
The format of a single line is
1085
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1086
REVISION_ID is a utf8-encoded revision id
1087
FLAGS is a comma separated list of flags about the record. Values include
1088
no-eol, line-delta, fulltext.
1089
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1090
that the the compressed data starts at.
1091
LENGTH is the ascii representation of the length of the data file.
1092
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1094
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1095
revision id already in the knit that is a parent of REVISION_ID.
1096
The ' :' marker is the end of record marker.
1099
when a write is interrupted to the index file, it will result in a line
1100
that does not end in ' :'. If the ' :' is not present at the end of a line,
1101
or at the end of the file, then the record that is missing it will be
1102
ignored by the parser.
1104
When writing new records to the index file, the data is preceded by '\n'
1105
to ensure that records always start on new lines even if the last write was
1106
interrupted. As a result its normal for the last line in the index to be
1107
missing a trailing newline. One can be added with no harmful effects.
1110
HEADER = "# bzr knit index 8\n"
1112
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1113
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1115
def _cache_version(self, version_id, options, pos, size, parents):
1116
"""Cache a version record in the history array and index cache.
1118
This is inlined into _load_data for performance. KEEP IN SYNC.
1119
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1122
# only want the _history index to reference the 1st index entry
1124
if version_id not in self._cache:
1125
index = len(self._history)
1126
self._history.append(version_id)
1128
index = self._cache[version_id][5]
1129
self._cache[version_id] = (version_id,
1136
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1137
create_parent_dir=False, delay_create=False, dir_mode=None):
1138
_KnitComponentFile.__init__(self, transport, filename, mode,
1139
file_mode=file_mode,
1140
create_parent_dir=create_parent_dir,
1143
# position in _history is the 'official' index for a revision
1144
# but the values may have come from a newer entry.
1145
# so - wc -l of a knit index is != the number of unique names
1149
fp = self._transport.get(self._filename)
1151
# _load_data may raise NoSuchFile if the target knit is
1157
if mode != 'w' or not create:
1160
self._need_to_create = True
1162
self._transport.put_bytes_non_atomic(
1163
self._filename, self.HEADER, mode=self._file_mode)
1165
def _load_data(self, fp):
1167
history = self._history
1169
self.check_header(fp)
1170
# readlines reads the whole file at once:
1171
# bad for transports like http, good for local disk
1172
# we save 60 ms doing this one change (
1173
# from calling readline each time to calling
1175
# probably what we want for nice behaviour on
1176
# http is a incremental readlines that yields, or
1177
# a check for local vs non local indexes,
1178
history_top = len(history) - 1
1179
for line in fp.readlines():
1181
if len(rec) < 5 or rec[-1] != ':':
1183
# FIXME: in the future we should determine if its a
1184
# short write - and ignore it
1185
# or a different failure, and raise. RBC 20060407
1189
for value in rec[4:-1]:
1191
# uncompressed reference
1192
parent_id = value[1:]
1194
parent_id = history[int(value)]
1195
parents.append(parent_id)
1197
version_id, options, pos, size = rec[:4]
1198
version_id = version_id
1200
# See self._cache_version
1201
# only want the _history index to reference the 1st
1202
# index entry for version_id
1203
if version_id not in cache:
1206
history.append(version_id)
1208
index = cache[version_id][5]
1209
cache[version_id] = (version_id,
1215
# end self._cache_version
1217
def get_graph(self):
1218
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1220
def get_ancestry(self, versions):
1221
"""See VersionedFile.get_ancestry."""
1222
# get a graph of all the mentioned versions:
1224
pending = set(versions)
1227
version = pending.pop()
1230
parents = [p for p in cache[version][4] if p in cache]
1232
raise RevisionNotPresent(version, self._filename)
1233
# if not completed and not a ghost
1234
pending.update([p for p in parents if p not in graph])
1235
graph[version] = parents
1236
return topo_sort(graph.items())
1238
def get_ancestry_with_ghosts(self, versions):
1239
"""See VersionedFile.get_ancestry_with_ghosts."""
1240
# get a graph of all the mentioned versions:
1241
self.check_versions_present(versions)
1244
pending = set(versions)
1246
version = pending.pop()
1248
parents = cache[version][4]
1254
pending.update([p for p in parents if p not in graph])
1255
graph[version] = parents
1256
return topo_sort(graph.items())
1258
def num_versions(self):
1259
return len(self._history)
1261
__len__ = num_versions
1263
def get_versions(self):
1264
return self._history
1266
def idx_to_name(self, idx):
1267
return self._history[idx]
1269
def lookup(self, version_id):
1270
assert version_id in self._cache
1271
return self._cache[version_id][5]
1273
def _version_list_to_index(self, versions):
1276
for version in versions:
1277
if version in cache:
1278
# -- inlined lookup() --
1279
result_list.append(str(cache[version][5]))
1280
# -- end lookup () --
1282
result_list.append('.' + version)
1283
return ' '.join(result_list)
1285
def add_version(self, version_id, options, pos, size, parents):
1286
"""Add a version record to the index."""
1287
self.add_versions(((version_id, options, pos, size, parents),))
1289
def add_versions(self, versions):
1290
"""Add multiple versions to the index.
1292
:param versions: a list of tuples:
1293
(version_id, options, pos, size, parents).
1296
orig_history = self._history[:]
1297
orig_cache = self._cache.copy()
1300
for version_id, options, pos, size, parents in versions:
1301
line = "\n%s %s %s %s %s :" % (version_id,
1305
self._version_list_to_index(parents))
1306
assert isinstance(line, str), \
1307
'content must be utf-8 encoded: %r' % (line,)
1309
self._cache_version(version_id, options, pos, size, parents)
1310
if not self._need_to_create:
1311
self._transport.append_bytes(self._filename, ''.join(lines))
1314
sio.write(self.HEADER)
1315
sio.writelines(lines)
1317
self._transport.put_file_non_atomic(self._filename, sio,
1318
create_parent_dir=self._create_parent_dir,
1319
mode=self._file_mode,
1320
dir_mode=self._dir_mode)
1321
self._need_to_create = False
1323
# If any problems happen, restore the original values and re-raise
1324
self._history = orig_history
1325
self._cache = orig_cache
1328
def has_version(self, version_id):
1329
"""True if the version is in the index."""
1330
return version_id in self._cache
1332
def get_position(self, version_id):
1333
"""Return data position and size of specified version."""
1334
entry = self._cache[version_id]
1335
return entry[2], entry[3]
1337
def get_method(self, version_id):
1338
"""Return compression method of specified version."""
1339
options = self._cache[version_id][1]
1340
if 'fulltext' in options:
1343
if 'line-delta' not in options:
1344
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1347
def get_options(self, version_id):
1348
return self._cache[version_id][1]
1350
def get_parents(self, version_id):
1351
"""Return parents of specified version ignoring ghosts."""
1352
return [parent for parent in self._cache[version_id][4]
1353
if parent in self._cache]
1355
def get_parents_with_ghosts(self, version_id):
1356
"""Return parents of specified version with ghosts."""
1357
return self._cache[version_id][4]
1359
def check_versions_present(self, version_ids):
1360
"""Check that all specified versions are present."""
1362
for version_id in version_ids:
1363
if version_id not in cache:
1364
raise RevisionNotPresent(version_id, self._filename)
1367
class _KnitData(_KnitComponentFile):
1368
"""Contents of the knit data file"""
1370
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1371
create_parent_dir=False, delay_create=False,
1373
_KnitComponentFile.__init__(self, transport, filename, mode,
1374
file_mode=file_mode,
1375
create_parent_dir=create_parent_dir,
1377
self._checked = False
1378
# TODO: jam 20060713 conceptually, this could spill to disk
1379
# if the cached size gets larger than a certain amount
1380
# but it complicates the model a bit, so for now just use
1381
# a simple dictionary
1383
self._do_cache = False
1386
self._need_to_create = create
1388
self._transport.put_bytes_non_atomic(self._filename, '',
1389
mode=self._file_mode)
1391
def enable_cache(self):
1392
"""Enable caching of reads."""
1393
self._do_cache = True
1395
def clear_cache(self):
1396
"""Clear the record cache."""
1397
self._do_cache = False
1400
def _open_file(self):
1402
return self._transport.get(self._filename)
1407
def _record_to_data(self, version_id, digest, lines):
1408
"""Convert version_id, digest, lines into a raw data block.
1410
:return: (len, a StringIO instance with the raw data ready to read.)
1413
data_file = GzipFile(None, mode='wb', fileobj=sio)
1415
assert isinstance(version_id, str)
1416
data_file.writelines(chain(
1417
["version %s %d %s\n" % (version_id,
1421
["end %s\n" % version_id]))
1428
def add_raw_record(self, raw_data):
1429
"""Append a prepared record to the data file.
1431
:return: the offset in the data file raw_data was written.
1433
assert isinstance(raw_data, str), 'data must be plain bytes'
1434
if not self._need_to_create:
1435
return self._transport.append_bytes(self._filename, raw_data)
1437
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1438
create_parent_dir=self._create_parent_dir,
1439
mode=self._file_mode,
1440
dir_mode=self._dir_mode)
1441
self._need_to_create = False
1444
def add_record(self, version_id, digest, lines):
1445
"""Write new text record to disk. Returns the position in the
1446
file where it was written."""
1447
size, sio = self._record_to_data(version_id, digest, lines)
1449
if not self._need_to_create:
1450
start_pos = self._transport.append_file(self._filename, sio)
1452
self._transport.put_file_non_atomic(self._filename, sio,
1453
create_parent_dir=self._create_parent_dir,
1454
mode=self._file_mode,
1455
dir_mode=self._dir_mode)
1456
self._need_to_create = False
1459
self._cache[version_id] = sio.getvalue()
1460
return start_pos, size
1462
def _parse_record_header(self, version_id, raw_data):
1463
"""Parse a record header for consistency.
1465
:return: the header and the decompressor stream.
1466
as (stream, header_record)
1468
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1469
rec = self._check_header(version_id, df.readline())
1472
def _check_header(self, version_id, line):
1475
raise KnitCorrupt(self._filename,
1476
'unexpected number of elements in record header')
1477
if rec[1] != version_id:
1478
raise KnitCorrupt(self._filename,
1479
'unexpected version, wanted %r, got %r'
1480
% (version_id, rec[1]))
1483
def _parse_record(self, version_id, data):
1485
# 4168 calls in 2880 217 internal
1486
# 4168 calls to _parse_record_header in 2121
1487
# 4168 calls to readlines in 330
1488
df = GzipFile(mode='rb', fileobj=StringIO(data))
1490
record_contents = df.readlines()
1491
header = record_contents.pop(0)
1492
rec = self._check_header(version_id, header)
1494
last_line = record_contents.pop()
1495
assert len(record_contents) == int(rec[2])
1496
if last_line != 'end %s\n' % rec[1]:
1497
raise KnitCorrupt(self._filename,
1498
'unexpected version end line %r, wanted %r'
1499
% (last_line, version_id))
1501
return record_contents, rec[3]
1503
def read_records_iter_raw(self, records):
1504
"""Read text records from data file and yield raw data.
1506
This unpacks enough of the text record to validate the id is
1507
as expected but thats all.
1509
# setup an iterator of the external records:
1510
# uses readv so nice and fast we hope.
1512
# grab the disk data needed.
1514
# Don't check _cache if it is empty
1515
needed_offsets = [(pos, size) for version_id, pos, size
1517
if version_id not in self._cache]
1519
needed_offsets = [(pos, size) for version_id, pos, size
1522
raw_records = self._transport.readv(self._filename, needed_offsets)
1524
for version_id, pos, size in records:
1525
if version_id in self._cache:
1526
# This data has already been validated
1527
data = self._cache[version_id]
1529
pos, data = raw_records.next()
1531
self._cache[version_id] = data
1533
# validate the header
1534
df, rec = self._parse_record_header(version_id, data)
1536
yield version_id, data
1538
def read_records_iter(self, records):
1539
"""Read text records from data file and yield result.
1541
The result will be returned in whatever is the fastest to read.
1542
Not by the order requested. Also, multiple requests for the same
1543
record will only yield 1 response.
1544
:param records: A list of (version_id, pos, len) entries
1545
:return: Yields (version_id, contents, digest) in the order
1546
read, not the order requested
1552
# Skip records we have alread seen
1553
yielded_records = set()
1554
needed_records = set()
1555
for record in records:
1556
if record[0] in self._cache:
1557
if record[0] in yielded_records:
1559
yielded_records.add(record[0])
1560
data = self._cache[record[0]]
1561
content, digest = self._parse_record(record[0], data)
1562
yield (record[0], content, digest)
1564
needed_records.add(record)
1565
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1567
needed_records = sorted(set(records), key=operator.itemgetter(1))
1569
if not needed_records:
1572
# The transport optimizes the fetching as well
1573
# (ie, reads continuous ranges.)
1574
readv_response = self._transport.readv(self._filename,
1575
[(pos, size) for version_id, pos, size in needed_records])
1577
for (version_id, pos, size), (pos, data) in \
1578
izip(iter(needed_records), readv_response):
1579
content, digest = self._parse_record(version_id, data)
1581
self._cache[version_id] = data
1582
yield version_id, content, digest
1584
def read_records(self, records):
1585
"""Read records into a dictionary."""
1587
for record_id, content, digest in \
1588
self.read_records_iter(records):
1589
components[record_id] = (content, digest)
1593
class InterKnit(InterVersionedFile):
1594
"""Optimised code paths for knit to knit operations."""
1596
_matching_file_from_factory = KnitVersionedFile
1597
_matching_file_to_factory = KnitVersionedFile
1600
def is_compatible(source, target):
1601
"""Be compatible with knits. """
1603
return (isinstance(source, KnitVersionedFile) and
1604
isinstance(target, KnitVersionedFile))
1605
except AttributeError:
1608
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1609
"""See InterVersionedFile.join."""
1610
assert isinstance(self.source, KnitVersionedFile)
1611
assert isinstance(self.target, KnitVersionedFile)
1613
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1618
pb = ui.ui_factory.nested_progress_bar()
1620
version_ids = list(version_ids)
1621
if None in version_ids:
1622
version_ids.remove(None)
1624
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1625
this_versions = set(self.target._index.get_versions())
1626
needed_versions = self.source_ancestry - this_versions
1627
cross_check_versions = self.source_ancestry.intersection(this_versions)
1628
mismatched_versions = set()
1629
for version in cross_check_versions:
1630
# scan to include needed parents.
1631
n1 = set(self.target.get_parents_with_ghosts(version))
1632
n2 = set(self.source.get_parents_with_ghosts(version))
1634
# FIXME TEST this check for cycles being introduced works
1635
# the logic is we have a cycle if in our graph we are an
1636
# ancestor of any of the n2 revisions.
1642
parent_ancestors = self.source.get_ancestry(parent)
1643
if version in parent_ancestors:
1644
raise errors.GraphCycleError([parent, version])
1645
# ensure this parent will be available later.
1646
new_parents = n2.difference(n1)
1647
needed_versions.update(new_parents.difference(this_versions))
1648
mismatched_versions.add(version)
1650
if not needed_versions and not mismatched_versions:
1652
full_list = topo_sort(self.source.get_graph())
1654
version_list = [i for i in full_list if (not self.target.has_version(i)
1655
and i in needed_versions)]
1659
copy_queue_records = []
1661
for version_id in version_list:
1662
options = self.source._index.get_options(version_id)
1663
parents = self.source._index.get_parents_with_ghosts(version_id)
1664
# check that its will be a consistent copy:
1665
for parent in parents:
1666
# if source has the parent, we must :
1667
# * already have it or
1668
# * have it scheduled already
1669
# otherwise we don't care
1670
assert (self.target.has_version(parent) or
1671
parent in copy_set or
1672
not self.source.has_version(parent))
1673
data_pos, data_size = self.source._index.get_position(version_id)
1674
copy_queue_records.append((version_id, data_pos, data_size))
1675
copy_queue.append((version_id, options, parents))
1676
copy_set.add(version_id)
1678
# data suck the join:
1680
total = len(version_list)
1683
for (version_id, raw_data), \
1684
(version_id2, options, parents) in \
1685
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1687
assert version_id == version_id2, 'logic error, inconsistent results'
1689
pb.update("Joining knit", count, total)
1690
raw_records.append((version_id, options, parents, len(raw_data)))
1691
raw_datum.append(raw_data)
1692
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1694
for version in mismatched_versions:
1695
# FIXME RBC 20060309 is this needed?
1696
n1 = set(self.target.get_parents_with_ghosts(version))
1697
n2 = set(self.source.get_parents_with_ghosts(version))
1698
# write a combined record to our history preserving the current
1699
# parents as first in the list
1700
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1701
self.target.fix_parents(version, new_parents)
1707
InterVersionedFile.register_optimiser(InterKnit)
1710
class WeaveToKnit(InterVersionedFile):
1711
"""Optimised code paths for weave to knit operations."""
1713
_matching_file_from_factory = bzrlib.weave.WeaveFile
1714
_matching_file_to_factory = KnitVersionedFile
1717
def is_compatible(source, target):
1718
"""Be compatible with weaves to knits."""
1720
return (isinstance(source, bzrlib.weave.Weave) and
1721
isinstance(target, KnitVersionedFile))
1722
except AttributeError:
1725
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1726
"""See InterVersionedFile.join."""
1727
assert isinstance(self.source, bzrlib.weave.Weave)
1728
assert isinstance(self.target, KnitVersionedFile)
1730
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1735
pb = ui.ui_factory.nested_progress_bar()
1737
version_ids = list(version_ids)
1739
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1740
this_versions = set(self.target._index.get_versions())
1741
needed_versions = self.source_ancestry - this_versions
1742
cross_check_versions = self.source_ancestry.intersection(this_versions)
1743
mismatched_versions = set()
1744
for version in cross_check_versions:
1745
# scan to include needed parents.
1746
n1 = set(self.target.get_parents_with_ghosts(version))
1747
n2 = set(self.source.get_parents(version))
1748
# if all of n2's parents are in n1, then its fine.
1749
if n2.difference(n1):
1750
# FIXME TEST this check for cycles being introduced works
1751
# the logic is we have a cycle if in our graph we are an
1752
# ancestor of any of the n2 revisions.
1758
parent_ancestors = self.source.get_ancestry(parent)
1759
if version in parent_ancestors:
1760
raise errors.GraphCycleError([parent, version])
1761
# ensure this parent will be available later.
1762
new_parents = n2.difference(n1)
1763
needed_versions.update(new_parents.difference(this_versions))
1764
mismatched_versions.add(version)
1766
if not needed_versions and not mismatched_versions:
1768
full_list = topo_sort(self.source.get_graph())
1770
version_list = [i for i in full_list if (not self.target.has_version(i)
1771
and i in needed_versions)]
1775
total = len(version_list)
1776
for version_id in version_list:
1777
pb.update("Converting to knit", count, total)
1778
parents = self.source.get_parents(version_id)
1779
# check that its will be a consistent copy:
1780
for parent in parents:
1781
# if source has the parent, we must already have it
1782
assert (self.target.has_version(parent))
1783
self.target.add_lines(
1784
version_id, parents, self.source.get_lines(version_id))
1787
for version in mismatched_versions:
1788
# FIXME RBC 20060309 is this needed?
1789
n1 = set(self.target.get_parents_with_ghosts(version))
1790
n2 = set(self.source.get_parents(version))
1791
# write a combined record to our history preserving the current
1792
# parents as first in the list
1793
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1794
self.target.fix_parents(version, new_parents)
1800
InterVersionedFile.register_optimiser(WeaveToKnit)
1803
class KnitSequenceMatcher(difflib.SequenceMatcher):
1804
"""Knit tuned sequence matcher.
1806
This is based on profiling of difflib which indicated some improvements
1807
for our usage pattern.
1810
def find_longest_match(self, alo, ahi, blo, bhi):
1811
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1813
If isjunk is not defined:
1815
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1816
alo <= i <= i+k <= ahi
1817
blo <= j <= j+k <= bhi
1818
and for all (i',j',k') meeting those conditions,
1821
and if i == i', j <= j'
1823
In other words, of all maximal matching blocks, return one that
1824
starts earliest in a, and of all those maximal matching blocks that
1825
start earliest in a, return the one that starts earliest in b.
1827
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1828
>>> s.find_longest_match(0, 5, 0, 9)
1831
If isjunk is defined, first the longest matching block is
1832
determined as above, but with the additional restriction that no
1833
junk element appears in the block. Then that block is extended as
1834
far as possible by matching (only) junk elements on both sides. So
1835
the resulting block never matches on junk except as identical junk
1836
happens to be adjacent to an "interesting" match.
1838
Here's the same example as before, but considering blanks to be
1839
junk. That prevents " abcd" from matching the " abcd" at the tail
1840
end of the second sequence directly. Instead only the "abcd" can
1841
match, and matches the leftmost "abcd" in the second sequence:
1843
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1844
>>> s.find_longest_match(0, 5, 0, 9)
1847
If no blocks match, return (alo, blo, 0).
1849
>>> s = SequenceMatcher(None, "ab", "c")
1850
>>> s.find_longest_match(0, 2, 0, 1)
1854
# CAUTION: stripping common prefix or suffix would be incorrect.
1858
# Longest matching block is "ab", but if common prefix is
1859
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1860
# strip, so ends up claiming that ab is changed to acab by
1861
# inserting "ca" in the middle. That's minimal but unintuitive:
1862
# "it's obvious" that someone inserted "ac" at the front.
1863
# Windiff ends up at the same place as diff, but by pairing up
1864
# the unique 'b's and then matching the first two 'a's.
1866
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1867
besti, bestj, bestsize = alo, blo, 0
1868
# find longest junk-free match
1869
# during an iteration of the loop, j2len[j] = length of longest
1870
# junk-free match ending with a[i-1] and b[j]
1874
for i in xrange(alo, ahi):
1875
# look at all instances of a[i] in b; note that because
1876
# b2j has no junk keys, the loop is skipped if a[i] is junk
1877
j2lenget = j2len.get
1880
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1881
# following improvement
1882
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1883
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1884
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1886
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1887
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1888
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1900
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1902
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1905
# Extend the best by non-junk elements on each end. In particular,
1906
# "popular" non-junk elements aren't in b2j, which greatly speeds
1907
# the inner loop above, but also means "the best" match so far
1908
# doesn't contain any junk *or* popular non-junk elements.
1909
while besti > alo and bestj > blo and \
1910
not isbjunk(b[bestj-1]) and \
1911
a[besti-1] == b[bestj-1]:
1912
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1913
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1914
not isbjunk(b[bestj+bestsize]) and \
1915
a[besti+bestsize] == b[bestj+bestsize]:
1918
# Now that we have a wholly interesting match (albeit possibly
1919
# empty!), we may as well suck up the matching junk on each
1920
# side of it too. Can't think of a good reason not to, and it
1921
# saves post-processing the (possibly considerable) expense of
1922
# figuring out what to do with it. In the case of an empty
1923
# interesting match, this is clearly the right thing to do,
1924
# because no other kind of match is possible in the regions.
1925
while besti > alo and bestj > blo and \
1926
isbjunk(b[bestj-1]) and \
1927
a[besti-1] == b[bestj-1]:
1928
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1929
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1930
isbjunk(b[bestj+bestsize]) and \
1931
a[besti+bestsize] == b[bestj+bestsize]:
1932
bestsize = bestsize + 1
1934
return besti, bestj, bestsize
added
removed
bzrlib/knit.py
