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
if not self.has_version(version_id):
522
raise RevisionNotPresent(version_id, self.filename)
524
parents = self.get_parents(version_id)
529
data_pos, data_size = self._index.get_position(version_id)
530
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
531
version_idx = self._index.lookup(version_id)
532
noeol = 'no-eol' in self._index.get_options(version_id)
533
if 'fulltext' == self._index.get_method(version_id):
534
new_content = self.factory.parse_fulltext(data, version_idx)
535
if parent is not None:
536
reference_content = self._get_content(parent)
537
old_texts = reference_content.text()
540
new_texts = new_content.text()
541
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
542
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
544
delta = self.factory.parse_line_delta(data, version_idx)
545
return parent, sha1, noeol, delta
547
def get_graph_with_ghosts(self):
548
"""See VersionedFile.get_graph_with_ghosts()."""
549
graph_items = self._index.get_graph()
550
return dict(graph_items)
552
def get_sha1(self, version_id):
553
"""See VersionedFile.get_sha1()."""
554
record_map = self._get_record_map([version_id])
555
method, content, digest, next = record_map[version_id]
560
"""See VersionedFile.get_suffixes()."""
561
return [DATA_SUFFIX, INDEX_SUFFIX]
563
def has_ghost(self, version_id):
564
"""True if there is a ghost reference in the file to version_id."""
566
if self.has_version(version_id):
568
# optimisable if needed by memoising the _ghosts set.
569
items = self._index.get_graph()
570
for node, parents in items:
571
for parent in parents:
572
if parent not in self._index._cache:
573
if parent == version_id:
578
"""See VersionedFile.versions."""
579
return self._index.get_versions()
581
def has_version(self, version_id):
582
"""See VersionedFile.has_version."""
583
return self._index.has_version(version_id)
585
__contains__ = has_version
587
def _merge_annotations(self, content, parents, parent_texts={},
588
delta=None, annotated=None):
589
"""Merge annotations for content. This is done by comparing
590
the annotations based on changed to the text.
594
for parent_id in parents:
595
merge_content = self._get_content(parent_id, parent_texts)
596
seq = patiencediff.PatienceSequenceMatcher(
597
None, merge_content.text(), content.text())
598
if delta_seq is None:
599
# setup a delta seq to reuse.
601
for i, j, n in seq.get_matching_blocks():
604
# this appears to copy (origin, text) pairs across to the new
605
# content for any line that matches the last-checked parent.
606
# FIXME: save the sequence control data for delta compression
607
# against the most relevant parent rather than rediffing.
608
content._lines[j:j+n] = merge_content._lines[i:i+n]
611
reference_content = self._get_content(parents[0], parent_texts)
612
new_texts = content.text()
613
old_texts = reference_content.text()
614
delta_seq = patiencediff.PatienceSequenceMatcher(
615
None, old_texts, new_texts)
616
return self._make_line_delta(delta_seq, content)
618
def _make_line_delta(self, delta_seq, new_content):
619
"""Generate a line delta from delta_seq and new_content."""
621
for op in delta_seq.get_opcodes():
624
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
627
def _get_components_positions(self, version_ids):
628
"""Produce a map of position data for the components of versions.
630
This data is intended to be used for retrieving the knit records.
632
A dict of version_id to (method, data_pos, data_size, next) is
634
method is the way referenced data should be applied.
635
data_pos is the position of the data in the knit.
636
data_size is the size of the data in the knit.
637
next is the build-parent of the version, or None for fulltexts.
640
for version_id in version_ids:
643
while cursor is not None and cursor not in component_data:
644
method = self._index.get_method(cursor)
645
if method == 'fulltext':
648
next = self.get_parents(cursor)[0]
649
data_pos, data_size = self._index.get_position(cursor)
650
component_data[cursor] = (method, data_pos, data_size, next)
652
return component_data
654
def _get_content(self, version_id, parent_texts={}):
655
"""Returns a content object that makes up the specified
657
if not self.has_version(version_id):
658
raise RevisionNotPresent(version_id, self.filename)
660
cached_version = parent_texts.get(version_id, None)
661
if cached_version is not None:
662
return cached_version
664
text_map, contents_map = self._get_content_maps([version_id])
665
return contents_map[version_id]
667
def _check_versions_present(self, version_ids):
668
"""Check that all specified versions are present."""
669
self._index.check_versions_present(version_ids)
671
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
672
"""See VersionedFile.add_lines_with_ghosts()."""
673
self._check_add(version_id, lines)
674
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
676
def _add_lines(self, version_id, parents, lines, parent_texts):
677
"""See VersionedFile.add_lines."""
678
self._check_add(version_id, lines)
679
self._check_versions_present(parents)
680
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
682
def _check_add(self, version_id, lines):
683
"""check that version_id and lines are safe to add."""
684
assert self.writable, "knit is not opened for write"
685
### FIXME escape. RBC 20060228
686
if contains_whitespace(version_id):
687
raise InvalidRevisionId(version_id, self.filename)
688
if self.has_version(version_id):
689
raise RevisionAlreadyPresent(version_id, self.filename)
690
self._check_lines_not_unicode(lines)
691
self._check_lines_are_lines(lines)
693
def _add(self, version_id, lines, parents, delta, parent_texts):
694
"""Add a set of lines on top of version specified by parents.
696
If delta is true, compress the text as a line-delta against
699
Any versions not present will be converted into ghosts.
701
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
702
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
703
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
704
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
705
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
706
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
707
# +1383 0 8.0370 8.0370 +<len>
708
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
709
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
710
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
711
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
715
if parent_texts is None:
717
for parent in parents:
718
if not self.has_version(parent):
719
ghosts.append(parent)
721
present_parents.append(parent)
723
if delta and not len(present_parents):
726
digest = sha_strings(lines)
729
if lines[-1][-1] != '\n':
730
options.append('no-eol')
731
lines[-1] = lines[-1] + '\n'
733
if len(present_parents) and delta:
734
# To speed the extract of texts the delta chain is limited
735
# to a fixed number of deltas. This should minimize both
736
# I/O and the time spend applying deltas.
737
delta = self._check_should_delta(present_parents)
739
lines = self.factory.make(lines, version_id)
740
if delta or (self.factory.annotated and len(present_parents) > 0):
741
# Merge annotations from parent texts if so is needed.
742
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
743
delta, self.factory.annotated)
746
options.append('line-delta')
747
store_lines = self.factory.lower_line_delta(delta_hunks)
749
options.append('fulltext')
750
store_lines = self.factory.lower_fulltext(lines)
752
where, size = self._data.add_record(version_id, digest, store_lines)
753
self._index.add_version(version_id, options, where, size, parents)
756
def check(self, progress_bar=None):
757
"""See VersionedFile.check()."""
759
def _clone_text(self, new_version_id, old_version_id, parents):
760
"""See VersionedFile.clone_text()."""
761
# FIXME RBC 20060228 make fast by only inserting an index with null
763
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
765
def get_lines(self, version_id):
766
"""See VersionedFile.get_lines()."""
767
return self.get_line_list([version_id])[0]
769
def _get_record_map(self, version_ids):
770
"""Produce a dictionary of knit records.
772
The keys are version_ids, the values are tuples of (method, content,
774
method is the way the content should be applied.
775
content is a KnitContent object.
776
digest is the SHA1 digest of this version id after all steps are done
777
next is the build-parent of the version, i.e. the leftmost ancestor.
778
If the method is fulltext, next will be None.
780
position_map = self._get_components_positions(version_ids)
781
# c = component_id, m = method, p = position, s = size, n = next
782
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
784
for component_id, content, digest in \
785
self._data.read_records_iter(records):
786
method, position, size, next = position_map[component_id]
787
record_map[component_id] = method, content, digest, next
791
def get_text(self, version_id):
792
"""See VersionedFile.get_text"""
793
return self.get_texts([version_id])[0]
795
def get_texts(self, version_ids):
796
return [''.join(l) for l in self.get_line_list(version_ids)]
798
def get_line_list(self, version_ids):
799
"""Return the texts of listed versions as a list of strings."""
800
text_map, content_map = self._get_content_maps(version_ids)
801
return [text_map[v] for v in version_ids]
803
def _get_content_maps(self, version_ids):
804
"""Produce maps of text and KnitContents
806
:return: (text_map, content_map) where text_map contains the texts for
807
the requested versions and content_map contains the KnitContents.
808
Both dicts take version_ids as their keys.
810
for version_id in version_ids:
811
if not self.has_version(version_id):
812
raise RevisionNotPresent(version_id, self.filename)
813
record_map = self._get_record_map(version_ids)
818
for version_id in version_ids:
821
while cursor is not None:
822
method, data, digest, next = record_map[cursor]
823
components.append((cursor, method, data, digest))
824
if cursor in content_map:
829
for component_id, method, data, digest in reversed(components):
830
if component_id in content_map:
831
content = content_map[component_id]
833
version_idx = self._index.lookup(component_id)
834
if method == 'fulltext':
835
assert content is None
836
content = self.factory.parse_fulltext(data, version_idx)
837
elif method == 'line-delta':
838
delta = self.factory.parse_line_delta(data, version_idx)
839
content = content.copy()
840
content._lines = self._apply_delta(content._lines,
842
content_map[component_id] = content
844
if 'no-eol' in self._index.get_options(version_id):
845
content = content.copy()
846
line = content._lines[-1][1].rstrip('\n')
847
content._lines[-1] = (content._lines[-1][0], line)
848
final_content[version_id] = content
850
# digest here is the digest from the last applied component.
851
text = content.text()
852
if sha_strings(text) != digest:
853
raise KnitCorrupt(self.filename,
854
'sha-1 does not match %s' % version_id)
856
text_map[version_id] = text
857
return text_map, final_content
859
def iter_lines_added_or_present_in_versions(self, version_ids=None,
861
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
862
if version_ids is None:
863
version_ids = self.versions()
865
pb = progress.DummyProgress()
866
# we don't care about inclusions, the caller cares.
867
# but we need to setup a list of records to visit.
868
# we need version_id, position, length
869
version_id_records = []
870
requested_versions = set(version_ids)
871
# filter for available versions
872
for version_id in requested_versions:
873
if not self.has_version(version_id):
874
raise RevisionNotPresent(version_id, self.filename)
875
# get a in-component-order queue:
876
for version_id in self.versions():
877
if version_id in requested_versions:
878
data_pos, length = self._index.get_position(version_id)
879
version_id_records.append((version_id, data_pos, length))
881
total = len(version_id_records)
882
for version_idx, (version_id, data, sha_value) in \
883
enumerate(self._data.read_records_iter(version_id_records)):
884
pb.update('Walking content.', version_idx, total)
885
method = self._index.get_method(version_id)
886
version_idx = self._index.lookup(version_id)
888
assert method in ('fulltext', 'line-delta')
889
if method == 'fulltext':
890
line_iterator = self.factory.get_fulltext_content(data)
892
line_iterator = self.factory.get_linedelta_content(data)
893
for line in line_iterator:
896
pb.update('Walking content.', total, total)
898
def num_versions(self):
899
"""See VersionedFile.num_versions()."""
900
return self._index.num_versions()
902
__len__ = num_versions
904
def annotate_iter(self, version_id):
905
"""See VersionedFile.annotate_iter."""
906
content = self._get_content(version_id)
907
for origin, text in content.annotate_iter():
910
def get_parents(self, version_id):
911
"""See VersionedFile.get_parents."""
914
# 52554 calls in 1264 872 internal down from 3674
916
return self._index.get_parents(version_id)
918
raise RevisionNotPresent(version_id, self.filename)
920
def get_parents_with_ghosts(self, version_id):
921
"""See VersionedFile.get_parents."""
923
return self._index.get_parents_with_ghosts(version_id)
925
raise RevisionNotPresent(version_id, self.filename)
927
def get_ancestry(self, versions):
928
"""See VersionedFile.get_ancestry."""
929
if isinstance(versions, basestring):
930
versions = [versions]
933
return self._index.get_ancestry(versions)
935
def get_ancestry_with_ghosts(self, versions):
936
"""See VersionedFile.get_ancestry_with_ghosts."""
937
if isinstance(versions, basestring):
938
versions = [versions]
941
return self._index.get_ancestry_with_ghosts(versions)
943
#@deprecated_method(zero_eight)
944
def walk(self, version_ids):
945
"""See VersionedFile.walk."""
946
# We take the short path here, and extract all relevant texts
947
# and put them in a weave and let that do all the work. Far
948
# from optimal, but is much simpler.
949
# FIXME RB 20060228 this really is inefficient!
950
from bzrlib.weave import Weave
952
w = Weave(self.filename)
953
ancestry = self.get_ancestry(version_ids)
954
sorted_graph = topo_sort(self._index.get_graph())
955
version_list = [vid for vid in sorted_graph if vid in ancestry]
957
for version_id in version_list:
958
lines = self.get_lines(version_id)
959
w.add_lines(version_id, self.get_parents(version_id), lines)
961
for lineno, insert_id, dset, line in w.walk(version_ids):
962
yield lineno, insert_id, dset, line
964
def plan_merge(self, ver_a, ver_b):
965
"""See VersionedFile.plan_merge."""
966
ancestors_b = set(self.get_ancestry(ver_b))
967
def status_a(revision, text):
968
if revision in ancestors_b:
969
return 'killed-b', text
973
ancestors_a = set(self.get_ancestry(ver_a))
974
def status_b(revision, text):
975
if revision in ancestors_a:
976
return 'killed-a', text
980
annotated_a = self.annotate(ver_a)
981
annotated_b = self.annotate(ver_b)
982
plain_a = [t for (a, t) in annotated_a]
983
plain_b = [t for (a, t) in annotated_b]
984
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
987
for ai, bi, l in blocks:
988
# process all mismatched sections
989
# (last mismatched section is handled because blocks always
990
# includes a 0-length last block)
991
for revision, text in annotated_a[a_cur:ai]:
992
yield status_a(revision, text)
993
for revision, text in annotated_b[b_cur:bi]:
994
yield status_b(revision, text)
996
# and now the matched section
999
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
1000
assert text_a == text_b
1001
yield "unchanged", text_a
1004
class _KnitComponentFile(object):
1005
"""One of the files used to implement a knit database"""
1007
def __init__(self, transport, filename, mode, file_mode=None,
1008
create_parent_dir=False, dir_mode=None):
1009
self._transport = transport
1010
self._filename = filename
1012
self._file_mode = file_mode
1013
self._dir_mode = dir_mode
1014
self._create_parent_dir = create_parent_dir
1015
self._need_to_create = False
1017
def _full_path(self):
1018
"""Return the full path to this file."""
1019
return self._transport.base + self._filename
1021
def check_header(self, fp):
1022
line = fp.readline()
1024
# An empty file can actually be treated as though the file doesn't
1026
raise errors.NoSuchFile(self._full_path())
1027
if line != self.HEADER:
1028
raise KnitHeaderError(badline=line,
1029
filename=self._transport.abspath(self._filename))
1032
"""Commit is a nop."""
1035
return '%s(%s)' % (self.__class__.__name__, self._filename)
1038
class _KnitIndex(_KnitComponentFile):
1039
"""Manages knit index file.
1041
The index is already kept in memory and read on startup, to enable
1042
fast lookups of revision information. The cursor of the index
1043
file is always pointing to the end, making it easy to append
1046
_cache is a cache for fast mapping from version id to a Index
1049
_history is a cache for fast mapping from indexes to version ids.
1051
The index data format is dictionary compressed when it comes to
1052
parent references; a index entry may only have parents that with a
1053
lover index number. As a result, the index is topological sorted.
1055
Duplicate entries may be written to the index for a single version id
1056
if this is done then the latter one completely replaces the former:
1057
this allows updates to correct version and parent information.
1058
Note that the two entries may share the delta, and that successive
1059
annotations and references MUST point to the first entry.
1061
The index file on disc contains a header, followed by one line per knit
1062
record. The same revision can be present in an index file more than once.
1063
The first occurrence gets assigned a sequence number starting from 0.
1065
The format of a single line is
1066
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1067
REVISION_ID is a utf8-encoded revision id
1068
FLAGS is a comma separated list of flags about the record. Values include
1069
no-eol, line-delta, fulltext.
1070
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1071
that the the compressed data starts at.
1072
LENGTH is the ascii representation of the length of the data file.
1073
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1075
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1076
revision id already in the knit that is a parent of REVISION_ID.
1077
The ' :' marker is the end of record marker.
1080
when a write is interrupted to the index file, it will result in a line
1081
that does not end in ' :'. If the ' :' is not present at the end of a line,
1082
or at the end of the file, then the record that is missing it will be
1083
ignored by the parser.
1085
When writing new records to the index file, the data is preceded by '\n'
1086
to ensure that records always start on new lines even if the last write was
1087
interrupted. As a result its normal for the last line in the index to be
1088
missing a trailing newline. One can be added with no harmful effects.
1091
HEADER = "# bzr knit index 8\n"
1093
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1094
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1096
def _cache_version(self, version_id, options, pos, size, parents):
1097
"""Cache a version record in the history array and index cache.
1099
This is inlined into _load_data for performance. KEEP IN SYNC.
1100
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1103
# only want the _history index to reference the 1st index entry
1105
if version_id not in self._cache:
1106
index = len(self._history)
1107
self._history.append(version_id)
1109
index = self._cache[version_id][5]
1110
self._cache[version_id] = (version_id,
1117
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1118
create_parent_dir=False, delay_create=False, dir_mode=None):
1119
_KnitComponentFile.__init__(self, transport, filename, mode,
1120
file_mode=file_mode,
1121
create_parent_dir=create_parent_dir,
1124
# position in _history is the 'official' index for a revision
1125
# but the values may have come from a newer entry.
1126
# so - wc -l of a knit index is != the number of unique names
1129
decode_utf8 = cache_utf8.decode
1130
pb = ui.ui_factory.nested_progress_bar()
1132
pb.update('read knit index', 0, 1)
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
pb.update('read knit index', 1, 1)
1153
def _load_data(self, fp):
1155
history = self._history
1156
decode_utf8 = cache_utf8.decode
1158
self.check_header(fp)
1159
# readlines reads the whole file at once:
1160
# bad for transports like http, good for local disk
1161
# we save 60 ms doing this one change (
1162
# from calling readline each time to calling
1164
# probably what we want for nice behaviour on
1165
# http is a incremental readlines that yields, or
1166
# a check for local vs non local indexes,
1167
history_top = len(history) - 1
1168
for line in fp.readlines():
1170
if len(rec) < 5 or rec[-1] != ':':
1172
# FIXME: in the future we should determine if its a
1173
# short write - and ignore it
1174
# or a different failure, and raise. RBC 20060407
1178
for value in rec[4:-1]:
1180
# uncompressed reference
1181
parents.append(decode_utf8(value[1:]))
1183
parents.append(history[int(value)])
1185
version_id, options, pos, size = rec[:4]
1186
version_id = decode_utf8(version_id)
1188
# See self._cache_version
1189
# only want the _history index to reference the 1st
1190
# index entry for version_id
1191
if version_id not in cache:
1194
history.append(version_id)
1196
index = cache[version_id][5]
1197
cache[version_id] = (version_id,
1203
# end self._cache_version
1205
def get_graph(self):
1206
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1208
def get_ancestry(self, versions):
1209
"""See VersionedFile.get_ancestry."""
1210
# get a graph of all the mentioned versions:
1212
pending = set(versions)
1215
version = pending.pop()
1218
parents = [p for p in cache[version][4] if p in cache]
1220
raise RevisionNotPresent(version, self._filename)
1221
# if not completed and not a ghost
1222
pending.update([p for p in parents if p not in graph])
1223
graph[version] = parents
1224
return topo_sort(graph.items())
1226
def get_ancestry_with_ghosts(self, versions):
1227
"""See VersionedFile.get_ancestry_with_ghosts."""
1228
# get a graph of all the mentioned versions:
1229
self.check_versions_present(versions)
1232
pending = set(versions)
1234
version = pending.pop()
1236
parents = cache[version][4]
1242
pending.update([p for p in parents if p not in graph])
1243
graph[version] = parents
1244
return topo_sort(graph.items())
1246
def num_versions(self):
1247
return len(self._history)
1249
__len__ = num_versions
1251
def get_versions(self):
1252
return self._history
1254
def idx_to_name(self, idx):
1255
return self._history[idx]
1257
def lookup(self, version_id):
1258
assert version_id in self._cache
1259
return self._cache[version_id][5]
1261
def _version_list_to_index(self, versions):
1262
encode_utf8 = cache_utf8.encode
1265
for version in versions:
1266
if version in cache:
1267
# -- inlined lookup() --
1268
result_list.append(str(cache[version][5]))
1269
# -- end lookup () --
1271
result_list.append('.' + encode_utf8(version))
1272
return ' '.join(result_list)
1274
def add_version(self, version_id, options, pos, size, parents):
1275
"""Add a version record to the index."""
1276
self.add_versions(((version_id, options, pos, size, parents),))
1278
def add_versions(self, versions):
1279
"""Add multiple versions to the index.
1281
:param versions: a list of tuples:
1282
(version_id, options, pos, size, parents).
1285
encode_utf8 = cache_utf8.encode
1286
orig_history = self._history[:]
1287
orig_cache = self._cache.copy()
1290
for version_id, options, pos, size, parents in versions:
1291
line = "\n%s %s %s %s %s :" % (encode_utf8(version_id),
1295
self._version_list_to_index(parents))
1296
assert isinstance(line, str), \
1297
'content must be utf-8 encoded: %r' % (line,)
1299
self._cache_version(version_id, options, pos, size, parents)
1300
if not self._need_to_create:
1301
self._transport.append_bytes(self._filename, ''.join(lines))
1304
sio.write(self.HEADER)
1305
sio.writelines(lines)
1307
self._transport.put_file_non_atomic(self._filename, sio,
1308
create_parent_dir=self._create_parent_dir,
1309
mode=self._file_mode,
1310
dir_mode=self._dir_mode)
1311
self._need_to_create = False
1313
# If any problems happen, restore the original values and re-raise
1314
self._history = orig_history
1315
self._cache = orig_cache
1318
def has_version(self, version_id):
1319
"""True if the version is in the index."""
1320
return version_id in self._cache
1322
def get_position(self, version_id):
1323
"""Return data position and size of specified version."""
1324
entry = self._cache[version_id]
1325
return entry[2], entry[3]
1327
def get_method(self, version_id):
1328
"""Return compression method of specified version."""
1329
options = self._cache[version_id][1]
1330
if 'fulltext' in options:
1333
if 'line-delta' not in options:
1334
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1337
def get_options(self, version_id):
1338
return self._cache[version_id][1]
1340
def get_parents(self, version_id):
1341
"""Return parents of specified version ignoring ghosts."""
1342
return [parent for parent in self._cache[version_id][4]
1343
if parent in self._cache]
1345
def get_parents_with_ghosts(self, version_id):
1346
"""Return parents of specified version with ghosts."""
1347
return self._cache[version_id][4]
1349
def check_versions_present(self, version_ids):
1350
"""Check that all specified versions are present."""
1352
for version_id in version_ids:
1353
if version_id not in cache:
1354
raise RevisionNotPresent(version_id, self._filename)
1357
class _KnitData(_KnitComponentFile):
1358
"""Contents of the knit data file"""
1360
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1361
create_parent_dir=False, delay_create=False,
1363
_KnitComponentFile.__init__(self, transport, filename, mode,
1364
file_mode=file_mode,
1365
create_parent_dir=create_parent_dir,
1367
self._checked = False
1368
# TODO: jam 20060713 conceptually, this could spill to disk
1369
# if the cached size gets larger than a certain amount
1370
# but it complicates the model a bit, so for now just use
1371
# a simple dictionary
1373
self._do_cache = False
1376
self._need_to_create = create
1378
self._transport.put_bytes_non_atomic(self._filename, '',
1379
mode=self._file_mode)
1381
def enable_cache(self):
1382
"""Enable caching of reads."""
1383
self._do_cache = True
1385
def clear_cache(self):
1386
"""Clear the record cache."""
1387
self._do_cache = False
1390
def _open_file(self):
1392
return self._transport.get(self._filename)
1397
def _record_to_data(self, version_id, digest, lines):
1398
"""Convert version_id, digest, lines into a raw data block.
1400
:return: (len, a StringIO instance with the raw data ready to read.)
1403
data_file = GzipFile(None, mode='wb', fileobj=sio)
1405
version_id_utf8 = cache_utf8.encode(version_id)
1406
data_file.writelines(chain(
1407
["version %s %d %s\n" % (version_id_utf8,
1411
["end %s\n" % version_id_utf8]))
1418
def add_raw_record(self, raw_data):
1419
"""Append a prepared record to the data file.
1421
:return: the offset in the data file raw_data was written.
1423
assert isinstance(raw_data, str), 'data must be plain bytes'
1424
if not self._need_to_create:
1425
return self._transport.append_bytes(self._filename, raw_data)
1427
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1428
create_parent_dir=self._create_parent_dir,
1429
mode=self._file_mode,
1430
dir_mode=self._dir_mode)
1431
self._need_to_create = False
1434
def add_record(self, version_id, digest, lines):
1435
"""Write new text record to disk. Returns the position in the
1436
file where it was written."""
1437
size, sio = self._record_to_data(version_id, digest, lines)
1439
if not self._need_to_create:
1440
start_pos = self._transport.append_file(self._filename, sio)
1442
self._transport.put_file_non_atomic(self._filename, sio,
1443
create_parent_dir=self._create_parent_dir,
1444
mode=self._file_mode,
1445
dir_mode=self._dir_mode)
1446
self._need_to_create = False
1449
self._cache[version_id] = sio.getvalue()
1450
return start_pos, size
1452
def _parse_record_header(self, version_id, raw_data):
1453
"""Parse a record header for consistency.
1455
:return: the header and the decompressor stream.
1456
as (stream, header_record)
1458
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1459
rec = self._check_header(version_id, df.readline())
1462
def _check_header(self, version_id, line):
1465
raise KnitCorrupt(self._filename,
1466
'unexpected number of elements in record header')
1467
if cache_utf8.decode(rec[1]) != version_id:
1468
raise KnitCorrupt(self._filename,
1469
'unexpected version, wanted %r, got %r'
1470
% (version_id, rec[1]))
1473
def _parse_record(self, version_id, data):
1475
# 4168 calls in 2880 217 internal
1476
# 4168 calls to _parse_record_header in 2121
1477
# 4168 calls to readlines in 330
1478
df = GzipFile(mode='rb', fileobj=StringIO(data))
1480
record_contents = df.readlines()
1481
header = record_contents.pop(0)
1482
rec = self._check_header(version_id, header)
1484
last_line = record_contents.pop()
1485
assert len(record_contents) == int(rec[2])
1486
if last_line != 'end %s\n' % rec[1]:
1487
raise KnitCorrupt(self._filename,
1488
'unexpected version end line %r, wanted %r'
1489
% (last_line, version_id))
1491
return record_contents, rec[3]
1493
def read_records_iter_raw(self, records):
1494
"""Read text records from data file and yield raw data.
1496
This unpacks enough of the text record to validate the id is
1497
as expected but thats all.
1499
# setup an iterator of the external records:
1500
# uses readv so nice and fast we hope.
1502
# grab the disk data needed.
1504
# Don't check _cache if it is empty
1505
needed_offsets = [(pos, size) for version_id, pos, size
1507
if version_id not in self._cache]
1509
needed_offsets = [(pos, size) for version_id, pos, size
1512
raw_records = self._transport.readv(self._filename, needed_offsets)
1514
for version_id, pos, size in records:
1515
if version_id in self._cache:
1516
# This data has already been validated
1517
data = self._cache[version_id]
1519
pos, data = raw_records.next()
1521
self._cache[version_id] = data
1523
# validate the header
1524
df, rec = self._parse_record_header(version_id, data)
1526
yield version_id, data
1528
def read_records_iter(self, records):
1529
"""Read text records from data file and yield result.
1531
The result will be returned in whatever is the fastest to read.
1532
Not by the order requested. Also, multiple requests for the same
1533
record will only yield 1 response.
1534
:param records: A list of (version_id, pos, len) entries
1535
:return: Yields (version_id, contents, digest) in the order
1536
read, not the order requested
1542
# Skip records we have alread seen
1543
yielded_records = set()
1544
needed_records = set()
1545
for record in records:
1546
if record[0] in self._cache:
1547
if record[0] in yielded_records:
1549
yielded_records.add(record[0])
1550
data = self._cache[record[0]]
1551
content, digest = self._parse_record(record[0], data)
1552
yield (record[0], content, digest)
1554
needed_records.add(record)
1555
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1557
needed_records = sorted(set(records), key=operator.itemgetter(1))
1559
if not needed_records:
1562
# The transport optimizes the fetching as well
1563
# (ie, reads continuous ranges.)
1564
readv_response = self._transport.readv(self._filename,
1565
[(pos, size) for version_id, pos, size in needed_records])
1567
for (version_id, pos, size), (pos, data) in \
1568
izip(iter(needed_records), readv_response):
1569
content, digest = self._parse_record(version_id, data)
1571
self._cache[version_id] = data
1572
yield version_id, content, digest
1574
def read_records(self, records):
1575
"""Read records into a dictionary."""
1577
for record_id, content, digest in \
1578
self.read_records_iter(records):
1579
components[record_id] = (content, digest)
1583
class InterKnit(InterVersionedFile):
1584
"""Optimised code paths for knit to knit operations."""
1586
_matching_file_from_factory = KnitVersionedFile
1587
_matching_file_to_factory = KnitVersionedFile
1590
def is_compatible(source, target):
1591
"""Be compatible with knits. """
1593
return (isinstance(source, KnitVersionedFile) and
1594
isinstance(target, KnitVersionedFile))
1595
except AttributeError:
1598
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1599
"""See InterVersionedFile.join."""
1600
assert isinstance(self.source, KnitVersionedFile)
1601
assert isinstance(self.target, KnitVersionedFile)
1603
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1608
pb = ui.ui_factory.nested_progress_bar()
1610
version_ids = list(version_ids)
1611
if None in version_ids:
1612
version_ids.remove(None)
1614
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1615
this_versions = set(self.target._index.get_versions())
1616
needed_versions = self.source_ancestry - this_versions
1617
cross_check_versions = self.source_ancestry.intersection(this_versions)
1618
mismatched_versions = set()
1619
for version in cross_check_versions:
1620
# scan to include needed parents.
1621
n1 = set(self.target.get_parents_with_ghosts(version))
1622
n2 = set(self.source.get_parents_with_ghosts(version))
1624
# FIXME TEST this check for cycles being introduced works
1625
# the logic is we have a cycle if in our graph we are an
1626
# ancestor of any of the n2 revisions.
1632
parent_ancestors = self.source.get_ancestry(parent)
1633
if version in parent_ancestors:
1634
raise errors.GraphCycleError([parent, version])
1635
# ensure this parent will be available later.
1636
new_parents = n2.difference(n1)
1637
needed_versions.update(new_parents.difference(this_versions))
1638
mismatched_versions.add(version)
1640
if not needed_versions and not mismatched_versions:
1642
full_list = topo_sort(self.source.get_graph())
1644
version_list = [i for i in full_list if (not self.target.has_version(i)
1645
and i in needed_versions)]
1649
copy_queue_records = []
1651
for version_id in version_list:
1652
options = self.source._index.get_options(version_id)
1653
parents = self.source._index.get_parents_with_ghosts(version_id)
1654
# check that its will be a consistent copy:
1655
for parent in parents:
1656
# if source has the parent, we must :
1657
# * already have it or
1658
# * have it scheduled already
1659
# otherwise we don't care
1660
assert (self.target.has_version(parent) or
1661
parent in copy_set or
1662
not self.source.has_version(parent))
1663
data_pos, data_size = self.source._index.get_position(version_id)
1664
copy_queue_records.append((version_id, data_pos, data_size))
1665
copy_queue.append((version_id, options, parents))
1666
copy_set.add(version_id)
1668
# data suck the join:
1670
total = len(version_list)
1673
for (version_id, raw_data), \
1674
(version_id2, options, parents) in \
1675
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1677
assert version_id == version_id2, 'logic error, inconsistent results'
1679
pb.update("Joining knit", count, total)
1680
raw_records.append((version_id, options, parents, len(raw_data)))
1681
raw_datum.append(raw_data)
1682
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1684
for version in mismatched_versions:
1685
# FIXME RBC 20060309 is this needed?
1686
n1 = set(self.target.get_parents_with_ghosts(version))
1687
n2 = set(self.source.get_parents_with_ghosts(version))
1688
# write a combined record to our history preserving the current
1689
# parents as first in the list
1690
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1691
self.target.fix_parents(version, new_parents)
1697
InterVersionedFile.register_optimiser(InterKnit)
1700
class WeaveToKnit(InterVersionedFile):
1701
"""Optimised code paths for weave to knit operations."""
1703
_matching_file_from_factory = bzrlib.weave.WeaveFile
1704
_matching_file_to_factory = KnitVersionedFile
1707
def is_compatible(source, target):
1708
"""Be compatible with weaves to knits."""
1710
return (isinstance(source, bzrlib.weave.Weave) and
1711
isinstance(target, KnitVersionedFile))
1712
except AttributeError:
1715
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1716
"""See InterVersionedFile.join."""
1717
assert isinstance(self.source, bzrlib.weave.Weave)
1718
assert isinstance(self.target, KnitVersionedFile)
1720
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1725
pb = ui.ui_factory.nested_progress_bar()
1727
version_ids = list(version_ids)
1729
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1730
this_versions = set(self.target._index.get_versions())
1731
needed_versions = self.source_ancestry - this_versions
1732
cross_check_versions = self.source_ancestry.intersection(this_versions)
1733
mismatched_versions = set()
1734
for version in cross_check_versions:
1735
# scan to include needed parents.
1736
n1 = set(self.target.get_parents_with_ghosts(version))
1737
n2 = set(self.source.get_parents(version))
1738
# if all of n2's parents are in n1, then its fine.
1739
if n2.difference(n1):
1740
# FIXME TEST this check for cycles being introduced works
1741
# the logic is we have a cycle if in our graph we are an
1742
# ancestor of any of the n2 revisions.
1748
parent_ancestors = self.source.get_ancestry(parent)
1749
if version in parent_ancestors:
1750
raise errors.GraphCycleError([parent, version])
1751
# ensure this parent will be available later.
1752
new_parents = n2.difference(n1)
1753
needed_versions.update(new_parents.difference(this_versions))
1754
mismatched_versions.add(version)
1756
if not needed_versions and not mismatched_versions:
1758
full_list = topo_sort(self.source.get_graph())
1760
version_list = [i for i in full_list if (not self.target.has_version(i)
1761
and i in needed_versions)]
1765
total = len(version_list)
1766
for version_id in version_list:
1767
pb.update("Converting to knit", count, total)
1768
parents = self.source.get_parents(version_id)
1769
# check that its will be a consistent copy:
1770
for parent in parents:
1771
# if source has the parent, we must already have it
1772
assert (self.target.has_version(parent))
1773
self.target.add_lines(
1774
version_id, parents, self.source.get_lines(version_id))
1777
for version in mismatched_versions:
1778
# FIXME RBC 20060309 is this needed?
1779
n1 = set(self.target.get_parents_with_ghosts(version))
1780
n2 = set(self.source.get_parents(version))
1781
# write a combined record to our history preserving the current
1782
# parents as first in the list
1783
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1784
self.target.fix_parents(version, new_parents)
1790
InterVersionedFile.register_optimiser(WeaveToKnit)
1793
class KnitSequenceMatcher(difflib.SequenceMatcher):
1794
"""Knit tuned sequence matcher.
1796
This is based on profiling of difflib which indicated some improvements
1797
for our usage pattern.
1800
def find_longest_match(self, alo, ahi, blo, bhi):
1801
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1803
If isjunk is not defined:
1805
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1806
alo <= i <= i+k <= ahi
1807
blo <= j <= j+k <= bhi
1808
and for all (i',j',k') meeting those conditions,
1811
and if i == i', j <= j'
1813
In other words, of all maximal matching blocks, return one that
1814
starts earliest in a, and of all those maximal matching blocks that
1815
start earliest in a, return the one that starts earliest in b.
1817
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1818
>>> s.find_longest_match(0, 5, 0, 9)
1821
If isjunk is defined, first the longest matching block is
1822
determined as above, but with the additional restriction that no
1823
junk element appears in the block. Then that block is extended as
1824
far as possible by matching (only) junk elements on both sides. So
1825
the resulting block never matches on junk except as identical junk
1826
happens to be adjacent to an "interesting" match.
1828
Here's the same example as before, but considering blanks to be
1829
junk. That prevents " abcd" from matching the " abcd" at the tail
1830
end of the second sequence directly. Instead only the "abcd" can
1831
match, and matches the leftmost "abcd" in the second sequence:
1833
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1834
>>> s.find_longest_match(0, 5, 0, 9)
1837
If no blocks match, return (alo, blo, 0).
1839
>>> s = SequenceMatcher(None, "ab", "c")
1840
>>> s.find_longest_match(0, 2, 0, 1)
1844
# CAUTION: stripping common prefix or suffix would be incorrect.
1848
# Longest matching block is "ab", but if common prefix is
1849
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1850
# strip, so ends up claiming that ab is changed to acab by
1851
# inserting "ca" in the middle. That's minimal but unintuitive:
1852
# "it's obvious" that someone inserted "ac" at the front.
1853
# Windiff ends up at the same place as diff, but by pairing up
1854
# the unique 'b's and then matching the first two 'a's.
1856
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1857
besti, bestj, bestsize = alo, blo, 0
1858
# find longest junk-free match
1859
# during an iteration of the loop, j2len[j] = length of longest
1860
# junk-free match ending with a[i-1] and b[j]
1864
for i in xrange(alo, ahi):
1865
# look at all instances of a[i] in b; note that because
1866
# b2j has no junk keys, the loop is skipped if a[i] is junk
1867
j2lenget = j2len.get
1870
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1871
# following improvement
1872
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1873
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1874
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1876
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1877
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1878
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1890
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1892
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1895
# Extend the best by non-junk elements on each end. In particular,
1896
# "popular" non-junk elements aren't in b2j, which greatly speeds
1897
# the inner loop above, but also means "the best" match so far
1898
# doesn't contain any junk *or* popular non-junk elements.
1899
while besti > alo and bestj > blo and \
1900
not isbjunk(b[bestj-1]) and \
1901
a[besti-1] == b[bestj-1]:
1902
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1903
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1904
not isbjunk(b[bestj+bestsize]) and \
1905
a[besti+bestsize] == b[bestj+bestsize]:
1908
# Now that we have a wholly interesting match (albeit possibly
1909
# empty!), we may as well suck up the matching junk on each
1910
# side of it too. Can't think of a good reason not to, and it
1911
# saves post-processing the (possibly considerable) expense of
1912
# figuring out what to do with it. In the case of an empty
1913
# interesting match, this is clearly the right thing to do,
1914
# because no other kind of match is possible in the regions.
1915
while besti > alo and bestj > blo and \
1916
isbjunk(b[bestj-1]) and \
1917
a[besti-1] == b[bestj-1]:
1918
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1919
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1920
isbjunk(b[bestj+bestsize]) and \
1921
a[besti+bestsize] == b[bestj+bestsize]:
1922
bestsize = bestsize + 1
1924
return besti, bestj, bestsize
added
removed
bzrlib/knit.py
