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
80
from bzrlib.errors import (
88
RevisionAlreadyPresent,
90
from bzrlib.tuned_gzip import GzipFile
91
from bzrlib.trace import mutter
92
from bzrlib.osutils import (
97
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
98
from bzrlib.tsort import topo_sort
101
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
104
# TODO: Split out code specific to this format into an associated object.
106
# TODO: Can we put in some kind of value to check that the index and data
107
# files belong together?
109
# TODO: accommodate binaries, perhaps by storing a byte count
111
# TODO: function to check whole file
113
# TODO: atomically append data, then measure backwards from the cursor
114
# position after writing to work out where it was located. we may need to
115
# bypass python file buffering.
117
DATA_SUFFIX = '.knit'
118
INDEX_SUFFIX = '.kndx'
121
class KnitContent(object):
122
"""Content of a knit version to which deltas can be applied."""
124
def __init__(self, lines):
127
def annotate_iter(self):
128
"""Yield tuples of (origin, text) for each content line."""
129
return iter(self._lines)
132
"""Return a list of (origin, text) tuples."""
133
return list(self.annotate_iter())
135
def line_delta_iter(self, new_lines):
136
"""Generate line-based delta from this content to new_lines."""
137
new_texts = new_lines.text()
138
old_texts = self.text()
139
s = KnitSequenceMatcher(None, old_texts, new_texts)
140
for tag, i1, i2, j1, j2 in s.get_opcodes():
143
# ofrom, oto, length, data
144
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
146
def line_delta(self, new_lines):
147
return list(self.line_delta_iter(new_lines))
150
return [text for origin, text in self._lines]
153
return KnitContent(self._lines[:])
156
class _KnitFactory(object):
157
"""Base factory for creating content objects."""
159
def make(self, lines, version):
160
num_lines = len(lines)
161
return KnitContent(zip([version] * num_lines, lines))
164
class KnitAnnotateFactory(_KnitFactory):
165
"""Factory for creating annotated Content objects."""
169
def parse_fulltext(self, content, version):
170
"""Convert fulltext to internal representation
172
fulltext content is of the format
173
revid(utf8) plaintext\n
174
internal representation is of the format:
177
decode_utf8 = cache_utf8.decode
180
origin, text = line.split(' ', 1)
181
lines.append((decode_utf8(origin), text))
182
return KnitContent(lines)
184
def parse_line_delta_iter(self, lines):
185
return iter(self.parse_line_delta(lines))
187
def parse_line_delta(self, lines, version):
188
"""Convert a line based delta into internal representation.
190
line delta is in the form of:
191
intstart intend intcount
193
revid(utf8) newline\n
194
internal representation is
195
(start, end, count, [1..count tuples (revid, newline)])
197
decode_utf8 = cache_utf8.decode
201
# walk through the lines parsing.
203
start, end, count = [int(n) for n in header.split(',')]
207
origin, text = next().split(' ', 1)
209
contents.append((decode_utf8(origin), text))
210
result.append((start, end, count, contents))
213
def get_fulltext_content(self, lines):
214
"""Extract just the content lines from a fulltext."""
215
return (line.split(' ', 1)[1] for line in lines)
217
def get_linedelta_content(self, lines):
218
"""Extract just the content from a line delta.
220
This doesn't return all of the extra information stored in a delta.
221
Only the actual content lines.
226
header = header.split(',')
227
count = int(header[2])
228
for i in xrange(count):
229
origin, text = next().split(' ', 1)
232
def lower_fulltext(self, content):
233
"""convert a fulltext content record into a serializable form.
235
see parse_fulltext which this inverts.
237
encode_utf8 = cache_utf8.encode
238
return ['%s %s' % (encode_utf8(o), t) for o, t in content._lines]
240
def lower_line_delta(self, delta):
241
"""convert a delta into a serializable form.
243
See parse_line_delta which this inverts.
245
encode_utf8 = cache_utf8.encode
247
for start, end, c, lines in delta:
248
out.append('%d,%d,%d\n' % (start, end, c))
249
out.extend(encode_utf8(origin) + ' ' + text
250
for origin, text in lines)
254
class KnitPlainFactory(_KnitFactory):
255
"""Factory for creating plain Content objects."""
259
def parse_fulltext(self, content, version):
260
"""This parses an unannotated fulltext.
262
Note that this is not a noop - the internal representation
263
has (versionid, line) - its just a constant versionid.
265
return self.make(content, version)
267
def parse_line_delta_iter(self, lines, version):
269
num_lines = len(lines)
270
while cur < num_lines:
273
start, end, c = [int(n) for n in header.split(',')]
274
yield start, end, c, zip([version] * c, lines[cur:cur+c])
277
def parse_line_delta(self, lines, version):
278
return list(self.parse_line_delta_iter(lines, version))
280
def get_fulltext_content(self, lines):
281
"""Extract just the content lines from a fulltext."""
284
def get_linedelta_content(self, lines):
285
"""Extract just the content from a line delta.
287
This doesn't return all of the extra information stored in a delta.
288
Only the actual content lines.
293
header = header.split(',')
294
count = int(header[2])
295
for i in xrange(count):
298
def lower_fulltext(self, content):
299
return content.text()
301
def lower_line_delta(self, delta):
303
for start, end, c, lines in delta:
304
out.append('%d,%d,%d\n' % (start, end, c))
305
out.extend([text for origin, text in lines])
309
def make_empty_knit(transport, relpath):
310
"""Construct a empty knit at the specified location."""
311
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
315
class KnitVersionedFile(VersionedFile):
316
"""Weave-like structure with faster random access.
318
A knit stores a number of texts and a summary of the relationships
319
between them. Texts are identified by a string version-id. Texts
320
are normally stored and retrieved as a series of lines, but can
321
also be passed as single strings.
323
Lines are stored with the trailing newline (if any) included, to
324
avoid special cases for files with no final newline. Lines are
325
composed of 8-bit characters, not unicode. The combination of
326
these approaches should mean any 'binary' file can be safely
327
stored and retrieved.
330
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
331
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
332
create=False, create_parent_dir=False, delay_create=False,
334
"""Construct a knit at location specified by relpath.
336
:param create: If not True, only open an existing knit.
337
:param create_parent_dir: If True, create the parent directory if
338
creating the file fails. (This is used for stores with
339
hash-prefixes that may not exist yet)
340
:param delay_create: The calling code is aware that the knit won't
341
actually be created until the first data is stored.
343
if deprecated_passed(basis_knit):
344
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
345
" deprecated as of bzr 0.9.",
346
DeprecationWarning, stacklevel=2)
347
if access_mode is None:
349
super(KnitVersionedFile, self).__init__(access_mode)
350
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
351
self.transport = transport
352
self.filename = relpath
353
self.factory = factory or KnitAnnotateFactory()
354
self.writable = (access_mode == 'w')
357
self._max_delta_chain = 200
359
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
360
access_mode, create=create, file_mode=file_mode,
361
create_parent_dir=create_parent_dir, delay_create=delay_create,
363
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
364
access_mode, create=create and not len(self), file_mode=file_mode,
365
create_parent_dir=create_parent_dir, delay_create=delay_create,
369
return '%s(%s)' % (self.__class__.__name__,
370
self.transport.abspath(self.filename))
372
def _check_should_delta(self, first_parents):
373
"""Iterate back through the parent listing, looking for a fulltext.
375
This is used when we want to decide whether to add a delta or a new
376
fulltext. It searches for _max_delta_chain parents. When it finds a
377
fulltext parent, it sees if the total size of the deltas leading up to
378
it is large enough to indicate that we want a new full text anyway.
380
Return True if we should create a new delta, False if we should use a
385
delta_parents = first_parents
386
for count in xrange(self._max_delta_chain):
387
parent = delta_parents[0]
388
method = self._index.get_method(parent)
389
pos, size = self._index.get_position(parent)
390
if method == 'fulltext':
394
delta_parents = self._index.get_parents(parent)
396
# We couldn't find a fulltext, so we must create a new one
399
return fulltext_size > delta_size
401
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
402
"""See VersionedFile._add_delta()."""
403
self._check_add(version_id, []) # should we check the lines ?
404
self._check_versions_present(parents)
408
for parent in parents:
409
if not self.has_version(parent):
410
ghosts.append(parent)
412
present_parents.append(parent)
414
if delta_parent is None:
415
# reconstitute as full text.
416
assert len(delta) == 1 or len(delta) == 0
418
assert delta[0][0] == 0
419
assert delta[0][1] == 0, delta[0][1]
420
return super(KnitVersionedFile, self)._add_delta(version_id,
431
options.append('no-eol')
433
if delta_parent is not None:
434
# determine the current delta chain length.
435
# To speed the extract of texts the delta chain is limited
436
# to a fixed number of deltas. This should minimize both
437
# I/O and the time spend applying deltas.
438
# The window was changed to a maximum of 200 deltas, but also added
439
# was a check that the total compressed size of the deltas is
440
# smaller than the compressed size of the fulltext.
441
if not self._check_should_delta([delta_parent]):
442
# We don't want a delta here, just do a normal insertion.
443
return super(KnitVersionedFile, self)._add_delta(version_id,
450
options.append('line-delta')
451
store_lines = self.factory.lower_line_delta(delta)
453
where, size = self._data.add_record(version_id, digest, store_lines)
454
self._index.add_version(version_id, options, where, size, parents)
456
def _add_raw_records(self, records, data):
457
"""Add all the records 'records' with data pre-joined in 'data'.
459
:param records: A list of tuples(version_id, options, parents, size).
460
:param data: The data for the records. When it is written, the records
461
are adjusted to have pos pointing into data by the sum of
462
the preceding records sizes.
465
pos = self._data.add_raw_record(data)
468
for (version_id, options, parents, size) in records:
469
index_entries.append((version_id, options, pos+offset,
471
if self._data._do_cache:
472
self._data._cache[version_id] = data[offset:offset+size]
474
self._index.add_versions(index_entries)
476
def enable_cache(self):
477
"""Start caching data for this knit"""
478
self._data.enable_cache()
480
def clear_cache(self):
481
"""Clear the data cache only."""
482
self._data.clear_cache()
484
def copy_to(self, name, transport):
485
"""See VersionedFile.copy_to()."""
486
# copy the current index to a temp index to avoid racing with local
488
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
489
self.transport.get(self._index._filename))
491
f = self._data._open_file()
493
transport.put_file(name + DATA_SUFFIX, f)
496
# move the copied index into place
497
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
499
def create_empty(self, name, transport, mode=None):
500
return KnitVersionedFile(name, transport, factory=self.factory,
501
delta=self.delta, create=True)
503
def _fix_parents(self, version, new_parents):
504
"""Fix the parents list for version.
506
This is done by appending a new version to the index
507
with identical data except for the parents list.
508
the parents list must be a superset of the current
511
current_values = self._index._cache[version]
512
assert set(current_values[4]).difference(set(new_parents)) == set()
513
self._index.add_version(version,
519
def get_delta(self, version_id):
520
"""Get a delta for constructing version from some other version."""
521
self.check_not_reserved_id(version_id)
522
if not self.has_version(version_id):
523
raise RevisionNotPresent(version_id, self.filename)
525
parents = self.get_parents(version_id)
530
data_pos, data_size = self._index.get_position(version_id)
531
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
532
version_idx = self._index.lookup(version_id)
533
noeol = 'no-eol' in self._index.get_options(version_id)
534
if 'fulltext' == self._index.get_method(version_id):
535
new_content = self.factory.parse_fulltext(data, version_idx)
536
if parent is not None:
537
reference_content = self._get_content(parent)
538
old_texts = reference_content.text()
541
new_texts = new_content.text()
542
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
543
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
545
delta = self.factory.parse_line_delta(data, version_idx)
546
return parent, sha1, noeol, delta
548
def get_graph_with_ghosts(self):
549
"""See VersionedFile.get_graph_with_ghosts()."""
550
graph_items = self._index.get_graph()
551
return dict(graph_items)
553
def get_sha1(self, version_id):
554
"""See VersionedFile.get_sha1()."""
555
record_map = self._get_record_map([version_id])
556
method, content, digest, next = record_map[version_id]
561
"""See VersionedFile.get_suffixes()."""
562
return [DATA_SUFFIX, INDEX_SUFFIX]
564
def has_ghost(self, version_id):
565
"""True if there is a ghost reference in the file to version_id."""
567
if self.has_version(version_id):
569
# optimisable if needed by memoising the _ghosts set.
570
items = self._index.get_graph()
571
for node, parents in items:
572
for parent in parents:
573
if parent not in self._index._cache:
574
if parent == version_id:
579
"""See VersionedFile.versions."""
580
return self._index.get_versions()
582
def has_version(self, version_id):
583
"""See VersionedFile.has_version."""
584
return self._index.has_version(version_id)
586
__contains__ = has_version
588
def _merge_annotations(self, content, parents, parent_texts={},
589
delta=None, annotated=None):
590
"""Merge annotations for content. This is done by comparing
591
the annotations based on changed to the text.
595
for parent_id in parents:
596
merge_content = self._get_content(parent_id, parent_texts)
597
seq = patiencediff.PatienceSequenceMatcher(
598
None, merge_content.text(), content.text())
599
if delta_seq is None:
600
# setup a delta seq to reuse.
602
for i, j, n in seq.get_matching_blocks():
605
# this appears to copy (origin, text) pairs across to the new
606
# content for any line that matches the last-checked parent.
607
# FIXME: save the sequence control data for delta compression
608
# against the most relevant parent rather than rediffing.
609
content._lines[j:j+n] = merge_content._lines[i:i+n]
612
reference_content = self._get_content(parents[0], parent_texts)
613
new_texts = content.text()
614
old_texts = reference_content.text()
615
delta_seq = patiencediff.PatienceSequenceMatcher(
616
None, old_texts, new_texts)
617
return self._make_line_delta(delta_seq, content)
619
def _make_line_delta(self, delta_seq, new_content):
620
"""Generate a line delta from delta_seq and new_content."""
622
for op in delta_seq.get_opcodes():
625
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
628
def _get_components_positions(self, version_ids):
629
"""Produce a map of position data for the components of versions.
631
This data is intended to be used for retrieving the knit records.
633
A dict of version_id to (method, data_pos, data_size, next) is
635
method is the way referenced data should be applied.
636
data_pos is the position of the data in the knit.
637
data_size is the size of the data in the knit.
638
next is the build-parent of the version, or None for fulltexts.
641
for version_id in version_ids:
644
while cursor is not None and cursor not in component_data:
645
method = self._index.get_method(cursor)
646
if method == 'fulltext':
649
next = self.get_parents(cursor)[0]
650
data_pos, data_size = self._index.get_position(cursor)
651
component_data[cursor] = (method, data_pos, data_size, next)
653
return component_data
655
def _get_content(self, version_id, parent_texts={}):
656
"""Returns a content object that makes up the specified
658
if not self.has_version(version_id):
659
raise RevisionNotPresent(version_id, self.filename)
661
cached_version = parent_texts.get(version_id, None)
662
if cached_version is not None:
663
return cached_version
665
text_map, contents_map = self._get_content_maps([version_id])
666
return contents_map[version_id]
668
def _check_versions_present(self, version_ids):
669
"""Check that all specified versions are present."""
670
self._index.check_versions_present(version_ids)
672
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
673
"""See VersionedFile.add_lines_with_ghosts()."""
674
self._check_add(version_id, lines)
675
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
677
def _add_lines(self, version_id, parents, lines, parent_texts):
678
"""See VersionedFile.add_lines."""
679
self._check_add(version_id, lines)
680
self._check_versions_present(parents)
681
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
683
def _check_add(self, version_id, lines):
684
"""check that version_id and lines are safe to add."""
685
assert self.writable, "knit is not opened for write"
686
### FIXME escape. RBC 20060228
687
if contains_whitespace(version_id):
688
raise InvalidRevisionId(version_id, self.filename)
689
self.check_not_reserved_id(version_id)
690
if self.has_version(version_id):
691
raise RevisionAlreadyPresent(version_id, self.filename)
692
self._check_lines_not_unicode(lines)
693
self._check_lines_are_lines(lines)
695
def _add(self, version_id, lines, parents, delta, parent_texts):
696
"""Add a set of lines on top of version specified by parents.
698
If delta is true, compress the text as a line-delta against
701
Any versions not present will be converted into ghosts.
703
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
704
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
705
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
706
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
707
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
708
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
709
# +1383 0 8.0370 8.0370 +<len>
710
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
711
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
712
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
713
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
717
if parent_texts is None:
719
for parent in parents:
720
if not self.has_version(parent):
721
ghosts.append(parent)
723
present_parents.append(parent)
725
if delta and not len(present_parents):
728
digest = sha_strings(lines)
731
if lines[-1][-1] != '\n':
732
options.append('no-eol')
733
lines[-1] = lines[-1] + '\n'
735
if len(present_parents) and delta:
736
# To speed the extract of texts the delta chain is limited
737
# to a fixed number of deltas. This should minimize both
738
# I/O and the time spend applying deltas.
739
delta = self._check_should_delta(present_parents)
741
lines = self.factory.make(lines, version_id)
742
if delta or (self.factory.annotated and len(present_parents) > 0):
743
# Merge annotations from parent texts if so is needed.
744
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
745
delta, self.factory.annotated)
748
options.append('line-delta')
749
store_lines = self.factory.lower_line_delta(delta_hunks)
751
options.append('fulltext')
752
store_lines = self.factory.lower_fulltext(lines)
754
where, size = self._data.add_record(version_id, digest, store_lines)
755
self._index.add_version(version_id, options, where, size, parents)
758
def check(self, progress_bar=None):
759
"""See VersionedFile.check()."""
761
def _clone_text(self, new_version_id, old_version_id, parents):
762
"""See VersionedFile.clone_text()."""
763
# FIXME RBC 20060228 make fast by only inserting an index with null
765
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
767
def get_lines(self, version_id):
768
"""See VersionedFile.get_lines()."""
769
return self.get_line_list([version_id])[0]
771
def _get_record_map(self, version_ids):
772
"""Produce a dictionary of knit records.
774
The keys are version_ids, the values are tuples of (method, content,
776
method is the way the content should be applied.
777
content is a KnitContent object.
778
digest is the SHA1 digest of this version id after all steps are done
779
next is the build-parent of the version, i.e. the leftmost ancestor.
780
If the method is fulltext, next will be None.
782
position_map = self._get_components_positions(version_ids)
783
# c = component_id, m = method, p = position, s = size, n = next
784
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
786
for component_id, content, digest in \
787
self._data.read_records_iter(records):
788
method, position, size, next = position_map[component_id]
789
record_map[component_id] = method, content, digest, next
793
def get_text(self, version_id):
794
"""See VersionedFile.get_text"""
795
return self.get_texts([version_id])[0]
797
def get_texts(self, version_ids):
798
return [''.join(l) for l in self.get_line_list(version_ids)]
800
def get_line_list(self, version_ids):
801
"""Return the texts of listed versions as a list of strings."""
802
for version_id in version_ids:
803
self.check_not_reserved_id(version_id)
804
text_map, content_map = self._get_content_maps(version_ids)
805
return [text_map[v] for v in version_ids]
807
def _get_content_maps(self, version_ids):
808
"""Produce maps of text and KnitContents
810
:return: (text_map, content_map) where text_map contains the texts for
811
the requested versions and content_map contains the KnitContents.
812
Both dicts take version_ids as their keys.
814
for version_id in version_ids:
815
if not self.has_version(version_id):
816
raise RevisionNotPresent(version_id, self.filename)
817
record_map = self._get_record_map(version_ids)
822
for version_id in version_ids:
825
while cursor is not None:
826
method, data, digest, next = record_map[cursor]
827
components.append((cursor, method, data, digest))
828
if cursor in content_map:
833
for component_id, method, data, digest in reversed(components):
834
if component_id in content_map:
835
content = content_map[component_id]
837
version_idx = self._index.lookup(component_id)
838
if method == 'fulltext':
839
assert content is None
840
content = self.factory.parse_fulltext(data, version_idx)
841
elif method == 'line-delta':
842
delta = self.factory.parse_line_delta(data, version_idx)
843
content = content.copy()
844
content._lines = self._apply_delta(content._lines,
846
content_map[component_id] = content
848
if 'no-eol' in self._index.get_options(version_id):
849
content = content.copy()
850
line = content._lines[-1][1].rstrip('\n')
851
content._lines[-1] = (content._lines[-1][0], line)
852
final_content[version_id] = content
854
# digest here is the digest from the last applied component.
855
text = content.text()
856
if sha_strings(text) != digest:
857
raise KnitCorrupt(self.filename,
858
'sha-1 does not match %s' % version_id)
860
text_map[version_id] = text
861
return text_map, final_content
863
def iter_lines_added_or_present_in_versions(self, version_ids=None,
865
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
866
if version_ids is None:
867
version_ids = self.versions()
869
pb = progress.DummyProgress()
870
# we don't care about inclusions, the caller cares.
871
# but we need to setup a list of records to visit.
872
# we need version_id, position, length
873
version_id_records = []
874
requested_versions = set(version_ids)
875
# filter for available versions
876
for version_id in requested_versions:
877
if not self.has_version(version_id):
878
raise RevisionNotPresent(version_id, self.filename)
879
# get a in-component-order queue:
880
for version_id in self.versions():
881
if version_id in requested_versions:
882
data_pos, length = self._index.get_position(version_id)
883
version_id_records.append((version_id, data_pos, length))
885
total = len(version_id_records)
886
for version_idx, (version_id, data, sha_value) in \
887
enumerate(self._data.read_records_iter(version_id_records)):
888
pb.update('Walking content.', version_idx, total)
889
method = self._index.get_method(version_id)
890
version_idx = self._index.lookup(version_id)
892
assert method in ('fulltext', 'line-delta')
893
if method == 'fulltext':
894
line_iterator = self.factory.get_fulltext_content(data)
896
line_iterator = self.factory.get_linedelta_content(data)
897
for line in line_iterator:
900
pb.update('Walking content.', total, total)
902
def num_versions(self):
903
"""See VersionedFile.num_versions()."""
904
return self._index.num_versions()
906
__len__ = num_versions
908
def annotate_iter(self, version_id):
909
"""See VersionedFile.annotate_iter."""
910
content = self._get_content(version_id)
911
for origin, text in content.annotate_iter():
914
def get_parents(self, version_id):
915
"""See VersionedFile.get_parents."""
918
# 52554 calls in 1264 872 internal down from 3674
920
return self._index.get_parents(version_id)
922
raise RevisionNotPresent(version_id, self.filename)
924
def get_parents_with_ghosts(self, version_id):
925
"""See VersionedFile.get_parents."""
927
return self._index.get_parents_with_ghosts(version_id)
929
raise RevisionNotPresent(version_id, self.filename)
931
def get_ancestry(self, versions):
932
"""See VersionedFile.get_ancestry."""
933
if isinstance(versions, basestring):
934
versions = [versions]
937
return self._index.get_ancestry(versions)
939
def get_ancestry_with_ghosts(self, versions):
940
"""See VersionedFile.get_ancestry_with_ghosts."""
941
if isinstance(versions, basestring):
942
versions = [versions]
945
return self._index.get_ancestry_with_ghosts(versions)
947
#@deprecated_method(zero_eight)
948
def walk(self, version_ids):
949
"""See VersionedFile.walk."""
950
# We take the short path here, and extract all relevant texts
951
# and put them in a weave and let that do all the work. Far
952
# from optimal, but is much simpler.
953
# FIXME RB 20060228 this really is inefficient!
954
from bzrlib.weave import Weave
956
w = Weave(self.filename)
957
ancestry = self.get_ancestry(version_ids)
958
sorted_graph = topo_sort(self._index.get_graph())
959
version_list = [vid for vid in sorted_graph if vid in ancestry]
961
for version_id in version_list:
962
lines = self.get_lines(version_id)
963
w.add_lines(version_id, self.get_parents(version_id), lines)
965
for lineno, insert_id, dset, line in w.walk(version_ids):
966
yield lineno, insert_id, dset, line
968
def plan_merge(self, ver_a, ver_b):
969
"""See VersionedFile.plan_merge."""
970
ancestors_b = set(self.get_ancestry(ver_b))
971
def status_a(revision, text):
972
if revision in ancestors_b:
973
return 'killed-b', text
977
ancestors_a = set(self.get_ancestry(ver_a))
978
def status_b(revision, text):
979
if revision in ancestors_a:
980
return 'killed-a', text
984
annotated_a = self.annotate(ver_a)
985
annotated_b = self.annotate(ver_b)
986
plain_a = [t for (a, t) in annotated_a]
987
plain_b = [t for (a, t) in annotated_b]
988
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
991
for ai, bi, l in blocks:
992
# process all mismatched sections
993
# (last mismatched section is handled because blocks always
994
# includes a 0-length last block)
995
for revision, text in annotated_a[a_cur:ai]:
996
yield status_a(revision, text)
997
for revision, text in annotated_b[b_cur:bi]:
998
yield status_b(revision, text)
1000
# and now the matched section
1003
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
1004
assert text_a == text_b
1005
yield "unchanged", text_a
1008
class _KnitComponentFile(object):
1009
"""One of the files used to implement a knit database"""
1011
def __init__(self, transport, filename, mode, file_mode=None,
1012
create_parent_dir=False, dir_mode=None):
1013
self._transport = transport
1014
self._filename = filename
1016
self._file_mode = file_mode
1017
self._dir_mode = dir_mode
1018
self._create_parent_dir = create_parent_dir
1019
self._need_to_create = False
1021
def _full_path(self):
1022
"""Return the full path to this file."""
1023
return self._transport.base + self._filename
1025
def check_header(self, fp):
1026
line = fp.readline()
1028
# An empty file can actually be treated as though the file doesn't
1030
raise errors.NoSuchFile(self._full_path())
1031
if line != self.HEADER:
1032
raise KnitHeaderError(badline=line,
1033
filename=self._transport.abspath(self._filename))
1036
"""Commit is a nop."""
1039
return '%s(%s)' % (self.__class__.__name__, self._filename)
1042
class _KnitIndex(_KnitComponentFile):
1043
"""Manages knit index file.
1045
The index is already kept in memory and read on startup, to enable
1046
fast lookups of revision information. The cursor of the index
1047
file is always pointing to the end, making it easy to append
1050
_cache is a cache for fast mapping from version id to a Index
1053
_history is a cache for fast mapping from indexes to version ids.
1055
The index data format is dictionary compressed when it comes to
1056
parent references; a index entry may only have parents that with a
1057
lover index number. As a result, the index is topological sorted.
1059
Duplicate entries may be written to the index for a single version id
1060
if this is done then the latter one completely replaces the former:
1061
this allows updates to correct version and parent information.
1062
Note that the two entries may share the delta, and that successive
1063
annotations and references MUST point to the first entry.
1065
The index file on disc contains a header, followed by one line per knit
1066
record. The same revision can be present in an index file more than once.
1067
The first occurrence gets assigned a sequence number starting from 0.
1069
The format of a single line is
1070
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1071
REVISION_ID is a utf8-encoded revision id
1072
FLAGS is a comma separated list of flags about the record. Values include
1073
no-eol, line-delta, fulltext.
1074
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1075
that the the compressed data starts at.
1076
LENGTH is the ascii representation of the length of the data file.
1077
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1079
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1080
revision id already in the knit that is a parent of REVISION_ID.
1081
The ' :' marker is the end of record marker.
1084
when a write is interrupted to the index file, it will result in a line
1085
that does not end in ' :'. If the ' :' is not present at the end of a line,
1086
or at the end of the file, then the record that is missing it will be
1087
ignored by the parser.
1089
When writing new records to the index file, the data is preceded by '\n'
1090
to ensure that records always start on new lines even if the last write was
1091
interrupted. As a result its normal for the last line in the index to be
1092
missing a trailing newline. One can be added with no harmful effects.
1095
HEADER = "# bzr knit index 8\n"
1097
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1098
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1100
def _cache_version(self, version_id, options, pos, size, parents):
1101
"""Cache a version record in the history array and index cache.
1103
This is inlined into _load_data for performance. KEEP IN SYNC.
1104
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1107
# only want the _history index to reference the 1st index entry
1109
if version_id not in self._cache:
1110
index = len(self._history)
1111
self._history.append(version_id)
1113
index = self._cache[version_id][5]
1114
self._cache[version_id] = (version_id,
1121
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1122
create_parent_dir=False, delay_create=False, dir_mode=None):
1123
_KnitComponentFile.__init__(self, transport, filename, mode,
1124
file_mode=file_mode,
1125
create_parent_dir=create_parent_dir,
1128
# position in _history is the 'official' index for a revision
1129
# but the values may have come from a newer entry.
1130
# so - wc -l of a knit index is != the number of unique names
1134
fp = self._transport.get(self._filename)
1136
# _load_data may raise NoSuchFile if the target knit is
1142
if mode != 'w' or not create:
1145
self._need_to_create = True
1147
self._transport.put_bytes_non_atomic(
1148
self._filename, self.HEADER, mode=self._file_mode)
1150
def _load_data(self, fp):
1152
history = self._history
1153
decode_utf8 = cache_utf8.decode
1155
self.check_header(fp)
1156
# readlines reads the whole file at once:
1157
# bad for transports like http, good for local disk
1158
# we save 60 ms doing this one change (
1159
# from calling readline each time to calling
1161
# probably what we want for nice behaviour on
1162
# http is a incremental readlines that yields, or
1163
# a check for local vs non local indexes,
1164
history_top = len(history) - 1
1165
for line in fp.readlines():
1167
if len(rec) < 5 or rec[-1] != ':':
1169
# FIXME: in the future we should determine if its a
1170
# short write - and ignore it
1171
# or a different failure, and raise. RBC 20060407
1175
for value in rec[4:-1]:
1177
# uncompressed reference
1178
parents.append(decode_utf8(value[1:]))
1180
parents.append(history[int(value)])
1182
version_id, options, pos, size = rec[:4]
1183
version_id = decode_utf8(version_id)
1185
# See self._cache_version
1186
# only want the _history index to reference the 1st
1187
# index entry for version_id
1188
if version_id not in cache:
1191
history.append(version_id)
1193
index = cache[version_id][5]
1194
cache[version_id] = (version_id,
1200
# end self._cache_version
1202
def get_graph(self):
1203
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1205
def get_ancestry(self, versions):
1206
"""See VersionedFile.get_ancestry."""
1207
# get a graph of all the mentioned versions:
1209
pending = set(versions)
1212
version = pending.pop()
1215
parents = [p for p in cache[version][4] if p in cache]
1217
raise RevisionNotPresent(version, self._filename)
1218
# if not completed and not a ghost
1219
pending.update([p for p in parents if p not in graph])
1220
graph[version] = parents
1221
return topo_sort(graph.items())
1223
def get_ancestry_with_ghosts(self, versions):
1224
"""See VersionedFile.get_ancestry_with_ghosts."""
1225
# get a graph of all the mentioned versions:
1226
self.check_versions_present(versions)
1229
pending = set(versions)
1231
version = pending.pop()
1233
parents = cache[version][4]
1239
pending.update([p for p in parents if p not in graph])
1240
graph[version] = parents
1241
return topo_sort(graph.items())
1243
def num_versions(self):
1244
return len(self._history)
1246
__len__ = num_versions
1248
def get_versions(self):
1249
return self._history
1251
def idx_to_name(self, idx):
1252
return self._history[idx]
1254
def lookup(self, version_id):
1255
assert version_id in self._cache
1256
return self._cache[version_id][5]
1258
def _version_list_to_index(self, versions):
1259
encode_utf8 = cache_utf8.encode
1262
for version in versions:
1263
if version in cache:
1264
# -- inlined lookup() --
1265
result_list.append(str(cache[version][5]))
1266
# -- end lookup () --
1268
result_list.append('.' + encode_utf8(version))
1269
return ' '.join(result_list)
1271
def add_version(self, version_id, options, pos, size, parents):
1272
"""Add a version record to the index."""
1273
self.add_versions(((version_id, options, pos, size, parents),))
1275
def add_versions(self, versions):
1276
"""Add multiple versions to the index.
1278
:param versions: a list of tuples:
1279
(version_id, options, pos, size, parents).
1282
encode_utf8 = cache_utf8.encode
1283
orig_history = self._history[:]
1284
orig_cache = self._cache.copy()
1287
for version_id, options, pos, size, parents in versions:
1288
line = "\n%s %s %s %s %s :" % (encode_utf8(version_id),
1292
self._version_list_to_index(parents))
1293
assert isinstance(line, str), \
1294
'content must be utf-8 encoded: %r' % (line,)
1296
self._cache_version(version_id, options, pos, size, parents)
1297
if not self._need_to_create:
1298
self._transport.append_bytes(self._filename, ''.join(lines))
1301
sio.write(self.HEADER)
1302
sio.writelines(lines)
1304
self._transport.put_file_non_atomic(self._filename, sio,
1305
create_parent_dir=self._create_parent_dir,
1306
mode=self._file_mode,
1307
dir_mode=self._dir_mode)
1308
self._need_to_create = False
1310
# If any problems happen, restore the original values and re-raise
1311
self._history = orig_history
1312
self._cache = orig_cache
1315
def has_version(self, version_id):
1316
"""True if the version is in the index."""
1317
return version_id in self._cache
1319
def get_position(self, version_id):
1320
"""Return data position and size of specified version."""
1321
entry = self._cache[version_id]
1322
return entry[2], entry[3]
1324
def get_method(self, version_id):
1325
"""Return compression method of specified version."""
1326
options = self._cache[version_id][1]
1327
if 'fulltext' in options:
1330
if 'line-delta' not in options:
1331
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1334
def get_options(self, version_id):
1335
return self._cache[version_id][1]
1337
def get_parents(self, version_id):
1338
"""Return parents of specified version ignoring ghosts."""
1339
return [parent for parent in self._cache[version_id][4]
1340
if parent in self._cache]
1342
def get_parents_with_ghosts(self, version_id):
1343
"""Return parents of specified version with ghosts."""
1344
return self._cache[version_id][4]
1346
def check_versions_present(self, version_ids):
1347
"""Check that all specified versions are present."""
1349
for version_id in version_ids:
1350
if version_id not in cache:
1351
raise RevisionNotPresent(version_id, self._filename)
1354
class _KnitData(_KnitComponentFile):
1355
"""Contents of the knit data file"""
1357
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1358
create_parent_dir=False, delay_create=False,
1360
_KnitComponentFile.__init__(self, transport, filename, mode,
1361
file_mode=file_mode,
1362
create_parent_dir=create_parent_dir,
1364
self._checked = False
1365
# TODO: jam 20060713 conceptually, this could spill to disk
1366
# if the cached size gets larger than a certain amount
1367
# but it complicates the model a bit, so for now just use
1368
# a simple dictionary
1370
self._do_cache = False
1373
self._need_to_create = create
1375
self._transport.put_bytes_non_atomic(self._filename, '',
1376
mode=self._file_mode)
1378
def enable_cache(self):
1379
"""Enable caching of reads."""
1380
self._do_cache = True
1382
def clear_cache(self):
1383
"""Clear the record cache."""
1384
self._do_cache = False
1387
def _open_file(self):
1389
return self._transport.get(self._filename)
1394
def _record_to_data(self, version_id, digest, lines):
1395
"""Convert version_id, digest, lines into a raw data block.
1397
:return: (len, a StringIO instance with the raw data ready to read.)
1400
data_file = GzipFile(None, mode='wb', fileobj=sio)
1402
version_id_utf8 = cache_utf8.encode(version_id)
1403
data_file.writelines(chain(
1404
["version %s %d %s\n" % (version_id_utf8,
1408
["end %s\n" % version_id_utf8]))
1415
def add_raw_record(self, raw_data):
1416
"""Append a prepared record to the data file.
1418
:return: the offset in the data file raw_data was written.
1420
assert isinstance(raw_data, str), 'data must be plain bytes'
1421
if not self._need_to_create:
1422
return self._transport.append_bytes(self._filename, raw_data)
1424
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1425
create_parent_dir=self._create_parent_dir,
1426
mode=self._file_mode,
1427
dir_mode=self._dir_mode)
1428
self._need_to_create = False
1431
def add_record(self, version_id, digest, lines):
1432
"""Write new text record to disk. Returns the position in the
1433
file where it was written."""
1434
size, sio = self._record_to_data(version_id, digest, lines)
1436
if not self._need_to_create:
1437
start_pos = self._transport.append_file(self._filename, sio)
1439
self._transport.put_file_non_atomic(self._filename, sio,
1440
create_parent_dir=self._create_parent_dir,
1441
mode=self._file_mode,
1442
dir_mode=self._dir_mode)
1443
self._need_to_create = False
1446
self._cache[version_id] = sio.getvalue()
1447
return start_pos, size
1449
def _parse_record_header(self, version_id, raw_data):
1450
"""Parse a record header for consistency.
1452
:return: the header and the decompressor stream.
1453
as (stream, header_record)
1455
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1456
rec = self._check_header(version_id, df.readline())
1459
def _check_header(self, version_id, line):
1462
raise KnitCorrupt(self._filename,
1463
'unexpected number of elements in record header')
1464
if cache_utf8.decode(rec[1]) != version_id:
1465
raise KnitCorrupt(self._filename,
1466
'unexpected version, wanted %r, got %r'
1467
% (version_id, rec[1]))
1470
def _parse_record(self, version_id, data):
1472
# 4168 calls in 2880 217 internal
1473
# 4168 calls to _parse_record_header in 2121
1474
# 4168 calls to readlines in 330
1475
df = GzipFile(mode='rb', fileobj=StringIO(data))
1477
record_contents = df.readlines()
1478
header = record_contents.pop(0)
1479
rec = self._check_header(version_id, header)
1481
last_line = record_contents.pop()
1482
assert len(record_contents) == int(rec[2])
1483
if last_line != 'end %s\n' % rec[1]:
1484
raise KnitCorrupt(self._filename,
1485
'unexpected version end line %r, wanted %r'
1486
% (last_line, version_id))
1488
return record_contents, rec[3]
1490
def read_records_iter_raw(self, records):
1491
"""Read text records from data file and yield raw data.
1493
This unpacks enough of the text record to validate the id is
1494
as expected but thats all.
1496
# setup an iterator of the external records:
1497
# uses readv so nice and fast we hope.
1499
# grab the disk data needed.
1501
# Don't check _cache if it is empty
1502
needed_offsets = [(pos, size) for version_id, pos, size
1504
if version_id not in self._cache]
1506
needed_offsets = [(pos, size) for version_id, pos, size
1509
raw_records = self._transport.readv(self._filename, needed_offsets)
1511
for version_id, pos, size in records:
1512
if version_id in self._cache:
1513
# This data has already been validated
1514
data = self._cache[version_id]
1516
pos, data = raw_records.next()
1518
self._cache[version_id] = data
1520
# validate the header
1521
df, rec = self._parse_record_header(version_id, data)
1523
yield version_id, data
1525
def read_records_iter(self, records):
1526
"""Read text records from data file and yield result.
1528
The result will be returned in whatever is the fastest to read.
1529
Not by the order requested. Also, multiple requests for the same
1530
record will only yield 1 response.
1531
:param records: A list of (version_id, pos, len) entries
1532
:return: Yields (version_id, contents, digest) in the order
1533
read, not the order requested
1539
# Skip records we have alread seen
1540
yielded_records = set()
1541
needed_records = set()
1542
for record in records:
1543
if record[0] in self._cache:
1544
if record[0] in yielded_records:
1546
yielded_records.add(record[0])
1547
data = self._cache[record[0]]
1548
content, digest = self._parse_record(record[0], data)
1549
yield (record[0], content, digest)
1551
needed_records.add(record)
1552
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1554
needed_records = sorted(set(records), key=operator.itemgetter(1))
1556
if not needed_records:
1559
# The transport optimizes the fetching as well
1560
# (ie, reads continuous ranges.)
1561
readv_response = self._transport.readv(self._filename,
1562
[(pos, size) for version_id, pos, size in needed_records])
1564
for (version_id, pos, size), (pos, data) in \
1565
izip(iter(needed_records), readv_response):
1566
content, digest = self._parse_record(version_id, data)
1568
self._cache[version_id] = data
1569
yield version_id, content, digest
1571
def read_records(self, records):
1572
"""Read records into a dictionary."""
1574
for record_id, content, digest in \
1575
self.read_records_iter(records):
1576
components[record_id] = (content, digest)
1580
class InterKnit(InterVersionedFile):
1581
"""Optimised code paths for knit to knit operations."""
1583
_matching_file_from_factory = KnitVersionedFile
1584
_matching_file_to_factory = KnitVersionedFile
1587
def is_compatible(source, target):
1588
"""Be compatible with knits. """
1590
return (isinstance(source, KnitVersionedFile) and
1591
isinstance(target, KnitVersionedFile))
1592
except AttributeError:
1595
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1596
"""See InterVersionedFile.join."""
1597
assert isinstance(self.source, KnitVersionedFile)
1598
assert isinstance(self.target, KnitVersionedFile)
1600
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1605
pb = ui.ui_factory.nested_progress_bar()
1607
version_ids = list(version_ids)
1608
if None in version_ids:
1609
version_ids.remove(None)
1611
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1612
this_versions = set(self.target._index.get_versions())
1613
needed_versions = self.source_ancestry - this_versions
1614
cross_check_versions = self.source_ancestry.intersection(this_versions)
1615
mismatched_versions = set()
1616
for version in cross_check_versions:
1617
# scan to include needed parents.
1618
n1 = set(self.target.get_parents_with_ghosts(version))
1619
n2 = set(self.source.get_parents_with_ghosts(version))
1621
# FIXME TEST this check for cycles being introduced works
1622
# the logic is we have a cycle if in our graph we are an
1623
# ancestor of any of the n2 revisions.
1629
parent_ancestors = self.source.get_ancestry(parent)
1630
if version in parent_ancestors:
1631
raise errors.GraphCycleError([parent, version])
1632
# ensure this parent will be available later.
1633
new_parents = n2.difference(n1)
1634
needed_versions.update(new_parents.difference(this_versions))
1635
mismatched_versions.add(version)
1637
if not needed_versions and not mismatched_versions:
1639
full_list = topo_sort(self.source.get_graph())
1641
version_list = [i for i in full_list if (not self.target.has_version(i)
1642
and i in needed_versions)]
1646
copy_queue_records = []
1648
for version_id in version_list:
1649
options = self.source._index.get_options(version_id)
1650
parents = self.source._index.get_parents_with_ghosts(version_id)
1651
# check that its will be a consistent copy:
1652
for parent in parents:
1653
# if source has the parent, we must :
1654
# * already have it or
1655
# * have it scheduled already
1656
# otherwise we don't care
1657
assert (self.target.has_version(parent) or
1658
parent in copy_set or
1659
not self.source.has_version(parent))
1660
data_pos, data_size = self.source._index.get_position(version_id)
1661
copy_queue_records.append((version_id, data_pos, data_size))
1662
copy_queue.append((version_id, options, parents))
1663
copy_set.add(version_id)
1665
# data suck the join:
1667
total = len(version_list)
1670
for (version_id, raw_data), \
1671
(version_id2, options, parents) in \
1672
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1674
assert version_id == version_id2, 'logic error, inconsistent results'
1676
pb.update("Joining knit", count, total)
1677
raw_records.append((version_id, options, parents, len(raw_data)))
1678
raw_datum.append(raw_data)
1679
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1681
for version in mismatched_versions:
1682
# FIXME RBC 20060309 is this needed?
1683
n1 = set(self.target.get_parents_with_ghosts(version))
1684
n2 = set(self.source.get_parents_with_ghosts(version))
1685
# write a combined record to our history preserving the current
1686
# parents as first in the list
1687
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1688
self.target.fix_parents(version, new_parents)
1694
InterVersionedFile.register_optimiser(InterKnit)
1697
class WeaveToKnit(InterVersionedFile):
1698
"""Optimised code paths for weave to knit operations."""
1700
_matching_file_from_factory = bzrlib.weave.WeaveFile
1701
_matching_file_to_factory = KnitVersionedFile
1704
def is_compatible(source, target):
1705
"""Be compatible with weaves to knits."""
1707
return (isinstance(source, bzrlib.weave.Weave) and
1708
isinstance(target, KnitVersionedFile))
1709
except AttributeError:
1712
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1713
"""See InterVersionedFile.join."""
1714
assert isinstance(self.source, bzrlib.weave.Weave)
1715
assert isinstance(self.target, KnitVersionedFile)
1717
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1722
pb = ui.ui_factory.nested_progress_bar()
1724
version_ids = list(version_ids)
1726
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1727
this_versions = set(self.target._index.get_versions())
1728
needed_versions = self.source_ancestry - this_versions
1729
cross_check_versions = self.source_ancestry.intersection(this_versions)
1730
mismatched_versions = set()
1731
for version in cross_check_versions:
1732
# scan to include needed parents.
1733
n1 = set(self.target.get_parents_with_ghosts(version))
1734
n2 = set(self.source.get_parents(version))
1735
# if all of n2's parents are in n1, then its fine.
1736
if n2.difference(n1):
1737
# FIXME TEST this check for cycles being introduced works
1738
# the logic is we have a cycle if in our graph we are an
1739
# ancestor of any of the n2 revisions.
1745
parent_ancestors = self.source.get_ancestry(parent)
1746
if version in parent_ancestors:
1747
raise errors.GraphCycleError([parent, version])
1748
# ensure this parent will be available later.
1749
new_parents = n2.difference(n1)
1750
needed_versions.update(new_parents.difference(this_versions))
1751
mismatched_versions.add(version)
1753
if not needed_versions and not mismatched_versions:
1755
full_list = topo_sort(self.source.get_graph())
1757
version_list = [i for i in full_list if (not self.target.has_version(i)
1758
and i in needed_versions)]
1762
total = len(version_list)
1763
for version_id in version_list:
1764
pb.update("Converting to knit", count, total)
1765
parents = self.source.get_parents(version_id)
1766
# check that its will be a consistent copy:
1767
for parent in parents:
1768
# if source has the parent, we must already have it
1769
assert (self.target.has_version(parent))
1770
self.target.add_lines(
1771
version_id, parents, self.source.get_lines(version_id))
1774
for version in mismatched_versions:
1775
# FIXME RBC 20060309 is this needed?
1776
n1 = set(self.target.get_parents_with_ghosts(version))
1777
n2 = set(self.source.get_parents(version))
1778
# write a combined record to our history preserving the current
1779
# parents as first in the list
1780
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1781
self.target.fix_parents(version, new_parents)
1787
InterVersionedFile.register_optimiser(WeaveToKnit)
1790
class KnitSequenceMatcher(difflib.SequenceMatcher):
1791
"""Knit tuned sequence matcher.
1793
This is based on profiling of difflib which indicated some improvements
1794
for our usage pattern.
1797
def find_longest_match(self, alo, ahi, blo, bhi):
1798
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1800
If isjunk is not defined:
1802
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1803
alo <= i <= i+k <= ahi
1804
blo <= j <= j+k <= bhi
1805
and for all (i',j',k') meeting those conditions,
1808
and if i == i', j <= j'
1810
In other words, of all maximal matching blocks, return one that
1811
starts earliest in a, and of all those maximal matching blocks that
1812
start earliest in a, return the one that starts earliest in b.
1814
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1815
>>> s.find_longest_match(0, 5, 0, 9)
1818
If isjunk is defined, first the longest matching block is
1819
determined as above, but with the additional restriction that no
1820
junk element appears in the block. Then that block is extended as
1821
far as possible by matching (only) junk elements on both sides. So
1822
the resulting block never matches on junk except as identical junk
1823
happens to be adjacent to an "interesting" match.
1825
Here's the same example as before, but considering blanks to be
1826
junk. That prevents " abcd" from matching the " abcd" at the tail
1827
end of the second sequence directly. Instead only the "abcd" can
1828
match, and matches the leftmost "abcd" in the second sequence:
1830
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1831
>>> s.find_longest_match(0, 5, 0, 9)
1834
If no blocks match, return (alo, blo, 0).
1836
>>> s = SequenceMatcher(None, "ab", "c")
1837
>>> s.find_longest_match(0, 2, 0, 1)
1841
# CAUTION: stripping common prefix or suffix would be incorrect.
1845
# Longest matching block is "ab", but if common prefix is
1846
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1847
# strip, so ends up claiming that ab is changed to acab by
1848
# inserting "ca" in the middle. That's minimal but unintuitive:
1849
# "it's obvious" that someone inserted "ac" at the front.
1850
# Windiff ends up at the same place as diff, but by pairing up
1851
# the unique 'b's and then matching the first two 'a's.
1853
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1854
besti, bestj, bestsize = alo, blo, 0
1855
# find longest junk-free match
1856
# during an iteration of the loop, j2len[j] = length of longest
1857
# junk-free match ending with a[i-1] and b[j]
1861
for i in xrange(alo, ahi):
1862
# look at all instances of a[i] in b; note that because
1863
# b2j has no junk keys, the loop is skipped if a[i] is junk
1864
j2lenget = j2len.get
1867
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1868
# following improvement
1869
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1870
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1871
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1873
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1874
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1875
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1887
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1889
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1892
# Extend the best by non-junk elements on each end. In particular,
1893
# "popular" non-junk elements aren't in b2j, which greatly speeds
1894
# the inner loop above, but also means "the best" match so far
1895
# doesn't contain any junk *or* popular non-junk elements.
1896
while besti > alo and bestj > blo and \
1897
not isbjunk(b[bestj-1]) and \
1898
a[besti-1] == b[bestj-1]:
1899
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1900
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1901
not isbjunk(b[bestj+bestsize]) and \
1902
a[besti+bestsize] == b[bestj+bestsize]:
1905
# Now that we have a wholly interesting match (albeit possibly
1906
# empty!), we may as well suck up the matching junk on each
1907
# side of it too. Can't think of a good reason not to, and it
1908
# saves post-processing the (possibly considerable) expense of
1909
# figuring out what to do with it. In the case of an empty
1910
# interesting match, this is clearly the right thing to do,
1911
# because no other kind of match is possible in the regions.
1912
while besti > alo and bestj > blo and \
1913
isbjunk(b[bestj-1]) and \
1914
a[besti-1] == b[bestj-1]:
1915
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1916
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1917
isbjunk(b[bestj+bestsize]) and \
1918
a[besti+bestsize] == b[bestj+bestsize]:
1919
bestsize = bestsize + 1
1921
return besti, bestj, bestsize
added
removed
bzrlib/knit.py
