1
# Copyright (C) 2005, 2006 by Canonical Ltd
2
# Written by Martin Pool.
3
# Modified by Johan Rydberg <jrydberg@gnu.org>
4
# Modified by Robert Collins <robert.collins@canonical.com>
5
# Modified by Aaron Bentley <aaron.bentley@utoronto.ca>
7
# This program is free software; you can redistribute it and/or modify
8
# it under the terms of the GNU General Public License as published by
9
# the Free Software Foundation; either version 2 of the License, or
10
# (at your option) any later version.
12
# This program is distributed in the hope that it will be useful,
13
# but WITHOUT ANY WARRANTY; without even the implied warranty of
14
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15
# GNU General Public License for more details.
17
# You should have received a copy of the GNU General Public License
18
# along with this program; if not, write to the Free Software
19
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21
"""Knit versionedfile implementation.
23
A knit is a versioned file implementation that supports efficient append only
27
lifeless: the data file is made up of "delta records". each delta record has a delta header
28
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
29
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
30
end-marker; simply "end VERSION"
32
delta can be line or full contents.a
33
... the 8's there are the index number of the annotation.
34
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
38
8 e.set('executable', 'yes')
40
8 if elt.get('executable') == 'yes':
41
8 ie.executable = True
42
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
46
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
47
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
48
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
49
09:33 < lifeless> right
50
09:33 < jrydberg> lifeless: the position and size is the range in the data file
53
so the index sequence is the dictionary compressed sequence number used
54
in the deltas to provide line annotation
59
# 10:16 < lifeless> make partial index writes safe
60
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
61
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
63
# move sha1 out of the content so that join is faster at verifying parents
64
# record content length ?
68
from cStringIO import StringIO
70
from itertools import izip, chain
81
from bzrlib.errors import FileExists, NoSuchFile, KnitError, \
82
InvalidRevisionId, KnitCorrupt, KnitHeaderError, \
83
RevisionNotPresent, RevisionAlreadyPresent
84
from bzrlib.tuned_gzip import GzipFile
85
from bzrlib.trace import mutter
86
from bzrlib.osutils import contains_whitespace, contains_linebreaks, \
88
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
89
from bzrlib.tsort import topo_sort
91
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
94
# TODO: Split out code specific to this format into an associated object.
96
# TODO: Can we put in some kind of value to check that the index and data
97
# files belong together?
99
# TODO: accommodate binaries, perhaps by storing a byte count
101
# TODO: function to check whole file
103
# TODO: atomically append data, then measure backwards from the cursor
104
# position after writing to work out where it was located. we may need to
105
# bypass python file buffering.
107
DATA_SUFFIX = '.knit'
108
INDEX_SUFFIX = '.kndx'
111
class KnitContent(object):
112
"""Content of a knit version to which deltas can be applied."""
114
def __init__(self, lines):
117
def annotate_iter(self):
118
"""Yield tuples of (origin, text) for each content line."""
119
for origin, text in self._lines:
123
"""Return a list of (origin, text) tuples."""
124
return list(self.annotate_iter())
126
def line_delta_iter(self, new_lines):
127
"""Generate line-based delta from this content to new_lines."""
128
new_texts = [text for origin, text in new_lines._lines]
129
old_texts = [text for origin, text in self._lines]
130
s = KnitSequenceMatcher(None, old_texts, new_texts)
131
for op in s.get_opcodes():
134
# ofrom oto length data
135
yield (op[1], op[2], op[4]-op[3], new_lines._lines[op[3]:op[4]])
137
def line_delta(self, new_lines):
138
return list(self.line_delta_iter(new_lines))
141
return [text for origin, text in self._lines]
144
return KnitContent(self._lines[:])
147
class _KnitFactory(object):
148
"""Base factory for creating content objects."""
150
def make(self, lines, version):
151
num_lines = len(lines)
152
return KnitContent(zip([version] * num_lines, lines))
155
class KnitAnnotateFactory(_KnitFactory):
156
"""Factory for creating annotated Content objects."""
160
def parse_fulltext(self, content, version):
161
"""Convert fulltext to internal representation
163
fulltext content is of the format
164
revid(utf8) plaintext\n
165
internal representation is of the format:
168
decode_utf8 = cache_utf8.decode
171
origin, text = line.split(' ', 1)
172
lines.append((decode_utf8(origin), text))
173
return KnitContent(lines)
175
def parse_line_delta_iter(self, lines):
176
for result_item in self.parse_line_delta[lines]:
179
def parse_line_delta(self, lines, version):
180
"""Convert a line based delta into internal representation.
182
line delta is in the form of:
183
intstart intend intcount
185
revid(utf8) newline\n
186
internal representation is
187
(start, end, count, [1..count tuples (revid, newline)])
189
decode_utf8 = cache_utf8.decode
193
# walk through the lines parsing.
195
start, end, count = [int(n) for n in header.split(',')]
199
origin, text = next().split(' ', 1)
201
contents.append((decode_utf8(origin), text))
202
result.append((start, end, count, contents))
205
def lower_fulltext(self, content):
206
"""convert a fulltext content record into a serializable form.
208
see parse_fulltext which this inverts.
210
encode_utf8 = cache_utf8.encode
211
return ['%s %s' % (encode_utf8(o), t) for o, t in content._lines]
213
def lower_line_delta(self, delta):
214
"""convert a delta into a serializable form.
216
See parse_line_delta which this inverts.
218
encode_utf8 = cache_utf8.encode
220
for start, end, c, lines in delta:
221
out.append('%d,%d,%d\n' % (start, end, c))
222
out.extend(encode_utf8(origin) + ' ' + text
223
for origin, text in lines)
227
class KnitPlainFactory(_KnitFactory):
228
"""Factory for creating plain Content objects."""
232
def parse_fulltext(self, content, version):
233
"""This parses an unannotated fulltext.
235
Note that this is not a noop - the internal representation
236
has (versionid, line) - its just a constant versionid.
238
return self.make(content, version)
240
def parse_line_delta_iter(self, lines, version):
242
header = lines.pop(0)
243
start, end, c = [int(n) for n in header.split(',')]
244
yield start, end, c, zip([version] * c, lines[:c])
247
def parse_line_delta(self, lines, version):
248
return list(self.parse_line_delta_iter(lines, version))
250
def lower_fulltext(self, content):
251
return content.text()
253
def lower_line_delta(self, delta):
255
for start, end, c, lines in delta:
256
out.append('%d,%d,%d\n' % (start, end, c))
257
out.extend([text for origin, text in lines])
261
def make_empty_knit(transport, relpath):
262
"""Construct a empty knit at the specified location."""
263
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
267
class KnitVersionedFile(VersionedFile):
268
"""Weave-like structure with faster random access.
270
A knit stores a number of texts and a summary of the relationships
271
between them. Texts are identified by a string version-id. Texts
272
are normally stored and retrieved as a series of lines, but can
273
also be passed as single strings.
275
Lines are stored with the trailing newline (if any) included, to
276
avoid special cases for files with no final newline. Lines are
277
composed of 8-bit characters, not unicode. The combination of
278
these approaches should mean any 'binary' file can be safely
279
stored and retrieved.
282
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
283
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
284
create=False, create_parent_dir=False, delay_create=False):
285
"""Construct a knit at location specified by relpath.
287
:param create: If not True, only open an existing knit.
288
:param create_parent_dir: If True, create the parent directory if
289
creating the file fails. (This is used for stores with
290
hash-prefixes that may not exist yet)
291
:param delay_create: The calling code is aware that the knit won't
292
actually be created until the first data is stored.
294
if deprecated_passed(basis_knit):
295
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
296
" deprecated as of bzr 0.9.",
297
DeprecationWarning, stacklevel=2)
298
if access_mode is None:
300
super(KnitVersionedFile, self).__init__(access_mode)
301
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
302
self.transport = transport
303
self.filename = relpath
304
self.factory = factory or KnitAnnotateFactory()
305
self.writable = (access_mode == 'w')
308
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
309
access_mode, create=create, file_mode=file_mode,
310
create_parent_dir=create_parent_dir, delay_create=delay_create)
311
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
312
access_mode, create=create and not len(self), file_mode=file_mode,
313
create_parent_dir=create_parent_dir, delay_create=delay_create)
316
return '%s(%s)' % (self.__class__.__name__,
317
self.transport.abspath(self.filename))
319
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
320
"""See VersionedFile._add_delta()."""
321
self._check_add(version_id, []) # should we check the lines ?
322
self._check_versions_present(parents)
326
for parent in parents:
327
if not self.has_version(parent):
328
ghosts.append(parent)
330
present_parents.append(parent)
332
if delta_parent is None:
333
# reconstitute as full text.
334
assert len(delta) == 1 or len(delta) == 0
336
assert delta[0][0] == 0
337
assert delta[0][1] == 0, delta[0][1]
338
return super(KnitVersionedFile, self)._add_delta(version_id,
349
options.append('no-eol')
351
if delta_parent is not None:
352
# determine the current delta chain length.
353
# To speed the extract of texts the delta chain is limited
354
# to a fixed number of deltas. This should minimize both
355
# I/O and the time spend applying deltas.
357
delta_parents = [delta_parent]
359
parent = delta_parents[0]
360
method = self._index.get_method(parent)
361
if method == 'fulltext':
363
delta_parents = self._index.get_parents(parent)
365
if method == 'line-delta':
366
# did not find a fulltext in the delta limit.
367
# just do a normal insertion.
368
return super(KnitVersionedFile, self)._add_delta(version_id,
375
options.append('line-delta')
376
store_lines = self.factory.lower_line_delta(delta)
378
where, size = self._data.add_record(version_id, digest, store_lines)
379
self._index.add_version(version_id, options, where, size, parents)
381
def _add_raw_records(self, records, data):
382
"""Add all the records 'records' with data pre-joined in 'data'.
384
:param records: A list of tuples(version_id, options, parents, size).
385
:param data: The data for the records. When it is written, the records
386
are adjusted to have pos pointing into data by the sum of
387
the preceding records sizes.
390
pos = self._data.add_raw_record(data)
393
for (version_id, options, parents, size) in records:
394
index_entries.append((version_id, options, pos+offset,
396
if self._data._do_cache:
397
self._data._cache[version_id] = data[offset:offset+size]
399
self._index.add_versions(index_entries)
401
def enable_cache(self):
402
"""Start caching data for this knit"""
403
self._data.enable_cache()
405
def clear_cache(self):
406
"""Clear the data cache only."""
407
self._data.clear_cache()
409
def copy_to(self, name, transport):
410
"""See VersionedFile.copy_to()."""
411
# copy the current index to a temp index to avoid racing with local
413
transport.put(name + INDEX_SUFFIX + '.tmp', self.transport.get(self._index._filename),)
415
f = self._data._open_file()
417
transport.put(name + DATA_SUFFIX, f)
420
# move the copied index into place
421
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
423
def create_empty(self, name, transport, mode=None):
424
return KnitVersionedFile(name, transport, factory=self.factory, delta=self.delta, create=True)
426
def _fix_parents(self, version, new_parents):
427
"""Fix the parents list for version.
429
This is done by appending a new version to the index
430
with identical data except for the parents list.
431
the parents list must be a superset of the current
434
current_values = self._index._cache[version]
435
assert set(current_values[4]).difference(set(new_parents)) == set()
436
self._index.add_version(version,
442
def get_delta(self, version_id):
443
"""Get a delta for constructing version from some other version."""
444
if not self.has_version(version_id):
445
raise RevisionNotPresent(version_id, self.filename)
447
parents = self.get_parents(version_id)
452
data_pos, data_size = self._index.get_position(version_id)
453
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
454
version_idx = self._index.lookup(version_id)
455
noeol = 'no-eol' in self._index.get_options(version_id)
456
if 'fulltext' == self._index.get_method(version_id):
457
new_content = self.factory.parse_fulltext(data, version_idx)
458
if parent is not None:
459
reference_content = self._get_content(parent)
460
old_texts = reference_content.text()
463
new_texts = new_content.text()
464
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
465
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
467
delta = self.factory.parse_line_delta(data, version_idx)
468
return parent, sha1, noeol, delta
470
def get_graph_with_ghosts(self):
471
"""See VersionedFile.get_graph_with_ghosts()."""
472
graph_items = self._index.get_graph()
473
return dict(graph_items)
475
def get_sha1(self, version_id):
476
"""See VersionedFile.get_sha1()."""
477
record_map = self._get_record_map([version_id])
478
method, content, digest, next = record_map[version_id]
483
"""See VersionedFile.get_suffixes()."""
484
return [DATA_SUFFIX, INDEX_SUFFIX]
486
def has_ghost(self, version_id):
487
"""True if there is a ghost reference in the file to version_id."""
489
if self.has_version(version_id):
491
# optimisable if needed by memoising the _ghosts set.
492
items = self._index.get_graph()
493
for node, parents in items:
494
for parent in parents:
495
if parent not in self._index._cache:
496
if parent == version_id:
501
"""See VersionedFile.versions."""
502
return self._index.get_versions()
504
def has_version(self, version_id):
505
"""See VersionedFile.has_version."""
506
return self._index.has_version(version_id)
508
__contains__ = has_version
510
def _merge_annotations(self, content, parents, parent_texts={},
511
delta=None, annotated=None):
512
"""Merge annotations for content. This is done by comparing
513
the annotations based on changed to the text.
517
for parent_id in parents:
518
merge_content = self._get_content(parent_id, parent_texts)
519
seq = KnitSequenceMatcher(None, merge_content.text(), content.text())
520
if delta_seq is None:
521
# setup a delta seq to reuse.
523
for i, j, n in seq.get_matching_blocks():
526
# this appears to copy (origin, text) pairs across to the new
527
# content for any line that matches the last-checked parent.
528
# FIXME: save the sequence control data for delta compression
529
# against the most relevant parent rather than rediffing.
530
content._lines[j:j+n] = merge_content._lines[i:i+n]
533
reference_content = self._get_content(parents[0], parent_texts)
534
new_texts = content.text()
535
old_texts = reference_content.text()
536
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
537
return self._make_line_delta(delta_seq, content)
539
def _make_line_delta(self, delta_seq, new_content):
540
"""Generate a line delta from delta_seq and new_content."""
542
for op in delta_seq.get_opcodes():
545
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
548
def _get_components_positions(self, version_ids):
549
"""Produce a map of position data for the components of versions.
551
This data is intended to be used for retrieving the knit records.
553
A dict of version_id to (method, data_pos, data_size, next) is
555
method is the way referenced data should be applied.
556
data_pos is the position of the data in the knit.
557
data_size is the size of the data in the knit.
558
next is the build-parent of the version, or None for fulltexts.
561
for version_id in version_ids:
564
while cursor is not None and cursor not in component_data:
565
method = self._index.get_method(cursor)
566
if method == 'fulltext':
569
next = self.get_parents(cursor)[0]
570
data_pos, data_size = self._index.get_position(cursor)
571
component_data[cursor] = (method, data_pos, data_size, next)
573
return component_data
575
def _get_content(self, version_id, parent_texts={}):
576
"""Returns a content object that makes up the specified
578
if not self.has_version(version_id):
579
raise RevisionNotPresent(version_id, self.filename)
581
cached_version = parent_texts.get(version_id, None)
582
if cached_version is not None:
583
return cached_version
585
text_map, contents_map = self._get_content_maps([version_id])
586
return contents_map[version_id]
588
def _check_versions_present(self, version_ids):
589
"""Check that all specified versions are present."""
590
version_ids = set(version_ids)
591
for r in list(version_ids):
592
if self._index.has_version(r):
593
version_ids.remove(r)
595
raise RevisionNotPresent(list(version_ids)[0], self.filename)
597
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
598
"""See VersionedFile.add_lines_with_ghosts()."""
599
self._check_add(version_id, lines)
600
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
602
def _add_lines(self, version_id, parents, lines, parent_texts):
603
"""See VersionedFile.add_lines."""
604
self._check_add(version_id, lines)
605
self._check_versions_present(parents)
606
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
608
def _check_add(self, version_id, lines):
609
"""check that version_id and lines are safe to add."""
610
assert self.writable, "knit is not opened for write"
611
### FIXME escape. RBC 20060228
612
if contains_whitespace(version_id):
613
raise InvalidRevisionId(version_id, self.filename)
614
if self.has_version(version_id):
615
raise RevisionAlreadyPresent(version_id, self.filename)
616
self._check_lines_not_unicode(lines)
617
self._check_lines_are_lines(lines)
619
def _add(self, version_id, lines, parents, delta, parent_texts):
620
"""Add a set of lines on top of version specified by parents.
622
If delta is true, compress the text as a line-delta against
625
Any versions not present will be converted into ghosts.
627
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
628
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
629
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
630
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
631
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
632
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
633
# +1383 0 8.0370 8.0370 +<len>
634
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
635
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
636
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
637
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
641
if parent_texts is None:
643
for parent in parents:
644
if not self.has_version(parent):
645
ghosts.append(parent)
647
present_parents.append(parent)
649
if delta and not len(present_parents):
652
digest = sha_strings(lines)
655
if lines[-1][-1] != '\n':
656
options.append('no-eol')
657
lines[-1] = lines[-1] + '\n'
659
if len(present_parents) and delta:
660
# To speed the extract of texts the delta chain is limited
661
# to a fixed number of deltas. This should minimize both
662
# I/O and the time spend applying deltas.
664
delta_parents = present_parents
666
parent = delta_parents[0]
667
method = self._index.get_method(parent)
668
if method == 'fulltext':
670
delta_parents = self._index.get_parents(parent)
672
if method == 'line-delta':
675
lines = self.factory.make(lines, version_id)
676
if delta or (self.factory.annotated and len(present_parents) > 0):
677
# Merge annotations from parent texts if so is needed.
678
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
679
delta, self.factory.annotated)
682
options.append('line-delta')
683
store_lines = self.factory.lower_line_delta(delta_hunks)
685
options.append('fulltext')
686
store_lines = self.factory.lower_fulltext(lines)
688
where, size = self._data.add_record(version_id, digest, store_lines)
689
self._index.add_version(version_id, options, where, size, parents)
692
def check(self, progress_bar=None):
693
"""See VersionedFile.check()."""
695
def _clone_text(self, new_version_id, old_version_id, parents):
696
"""See VersionedFile.clone_text()."""
697
# FIXME RBC 20060228 make fast by only inserting an index with null
699
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
701
def get_lines(self, version_id):
702
"""See VersionedFile.get_lines()."""
703
return self.get_line_list([version_id])[0]
705
def _get_record_map(self, version_ids):
706
"""Produce a dictionary of knit records.
708
The keys are version_ids, the values are tuples of (method, content,
710
method is the way the content should be applied.
711
content is a KnitContent object.
712
digest is the SHA1 digest of this version id after all steps are done
713
next is the build-parent of the version, i.e. the leftmost ancestor.
714
If the method is fulltext, next will be None.
716
position_map = self._get_components_positions(version_ids)
717
# c = component_id, m = method, p = position, s = size, n = next
718
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
720
for component_id, content, digest in \
721
self._data.read_records_iter(records):
722
method, position, size, next = position_map[component_id]
723
record_map[component_id] = method, content, digest, next
727
def get_text(self, version_id):
728
"""See VersionedFile.get_text"""
729
return self.get_texts([version_id])[0]
731
def get_texts(self, version_ids):
732
return [''.join(l) for l in self.get_line_list(version_ids)]
734
def get_line_list(self, version_ids):
735
"""Return the texts of listed versions as a list of strings."""
736
text_map, content_map = self._get_content_maps(version_ids)
737
return [text_map[v] for v in version_ids]
739
def _get_content_maps(self, version_ids):
740
"""Produce maps of text and KnitContents
742
:return: (text_map, content_map) where text_map contains the texts for
743
the requested versions and content_map contains the KnitContents.
744
Both dicts take version_ids as their keys.
746
for version_id in version_ids:
747
if not self.has_version(version_id):
748
raise RevisionNotPresent(version_id, self.filename)
749
record_map = self._get_record_map(version_ids)
754
for version_id in version_ids:
757
while cursor is not None:
758
method, data, digest, next = record_map[cursor]
759
components.append((cursor, method, data, digest))
760
if cursor in content_map:
765
for component_id, method, data, digest in reversed(components):
766
if component_id in content_map:
767
content = content_map[component_id]
769
version_idx = self._index.lookup(component_id)
770
if method == 'fulltext':
771
assert content is None
772
content = self.factory.parse_fulltext(data, version_idx)
773
elif method == 'line-delta':
774
delta = self.factory.parse_line_delta(data[:],
776
content = content.copy()
777
content._lines = self._apply_delta(content._lines,
779
content_map[component_id] = content
781
if 'no-eol' in self._index.get_options(version_id):
782
content = content.copy()
783
line = content._lines[-1][1].rstrip('\n')
784
content._lines[-1] = (content._lines[-1][0], line)
785
final_content[version_id] = content
787
# digest here is the digest from the last applied component.
788
text = content.text()
789
if sha_strings(text) != digest:
790
raise KnitCorrupt(self.filename,
791
'sha-1 does not match %s' % version_id)
793
text_map[version_id] = text
794
return text_map, final_content
796
def iter_lines_added_or_present_in_versions(self, version_ids=None):
797
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
798
if version_ids is None:
799
version_ids = self.versions()
800
# we don't care about inclusions, the caller cares.
801
# but we need to setup a list of records to visit.
802
# we need version_id, position, length
803
version_id_records = []
804
requested_versions = list(version_ids)
805
# filter for available versions
806
for version_id in requested_versions:
807
if not self.has_version(version_id):
808
raise RevisionNotPresent(version_id, self.filename)
809
# get a in-component-order queue:
811
for version_id in self.versions():
812
if version_id in requested_versions:
813
version_ids.append(version_id)
814
data_pos, length = self._index.get_position(version_id)
815
version_id_records.append((version_id, data_pos, length))
817
pb = bzrlib.ui.ui_factory.nested_progress_bar()
819
total = len(version_id_records)
821
pb.update('Walking content.', count, total)
822
for version_id, data, sha_value in \
823
self._data.read_records_iter(version_id_records):
824
pb.update('Walking content.', count, total)
825
method = self._index.get_method(version_id)
826
version_idx = self._index.lookup(version_id)
827
assert method in ('fulltext', 'line-delta')
828
if method == 'fulltext':
829
content = self.factory.parse_fulltext(data, version_idx)
830
for line in content.text():
833
delta = self.factory.parse_line_delta(data, version_idx)
834
for start, end, count, lines in delta:
835
for origin, line in lines:
838
pb.update('Walking content.', total, total)
841
pb.update('Walking content.', total, total)
845
def num_versions(self):
846
"""See VersionedFile.num_versions()."""
847
return self._index.num_versions()
849
__len__ = num_versions
851
def annotate_iter(self, version_id):
852
"""See VersionedFile.annotate_iter."""
853
content = self._get_content(version_id)
854
for origin, text in content.annotate_iter():
857
def get_parents(self, version_id):
858
"""See VersionedFile.get_parents."""
861
# 52554 calls in 1264 872 internal down from 3674
863
return self._index.get_parents(version_id)
865
raise RevisionNotPresent(version_id, self.filename)
867
def get_parents_with_ghosts(self, version_id):
868
"""See VersionedFile.get_parents."""
870
return self._index.get_parents_with_ghosts(version_id)
872
raise RevisionNotPresent(version_id, self.filename)
874
def get_ancestry(self, versions):
875
"""See VersionedFile.get_ancestry."""
876
if isinstance(versions, basestring):
877
versions = [versions]
880
self._check_versions_present(versions)
881
return self._index.get_ancestry(versions)
883
def get_ancestry_with_ghosts(self, versions):
884
"""See VersionedFile.get_ancestry_with_ghosts."""
885
if isinstance(versions, basestring):
886
versions = [versions]
889
self._check_versions_present(versions)
890
return self._index.get_ancestry_with_ghosts(versions)
892
#@deprecated_method(zero_eight)
893
def walk(self, version_ids):
894
"""See VersionedFile.walk."""
895
# We take the short path here, and extract all relevant texts
896
# and put them in a weave and let that do all the work. Far
897
# from optimal, but is much simpler.
898
# FIXME RB 20060228 this really is inefficient!
899
from bzrlib.weave import Weave
901
w = Weave(self.filename)
902
ancestry = self.get_ancestry(version_ids)
903
sorted_graph = topo_sort(self._index.get_graph())
904
version_list = [vid for vid in sorted_graph if vid in ancestry]
906
for version_id in version_list:
907
lines = self.get_lines(version_id)
908
w.add_lines(version_id, self.get_parents(version_id), lines)
910
for lineno, insert_id, dset, line in w.walk(version_ids):
911
yield lineno, insert_id, dset, line
913
def plan_merge(self, ver_a, ver_b):
914
"""See VersionedFile.plan_merge."""
915
ancestors_b = set(self.get_ancestry(ver_b))
916
def status_a(revision, text):
917
if revision in ancestors_b:
918
return 'killed-b', text
922
ancestors_a = set(self.get_ancestry(ver_a))
923
def status_b(revision, text):
924
if revision in ancestors_a:
925
return 'killed-a', text
929
annotated_a = self.annotate(ver_a)
930
annotated_b = self.annotate(ver_b)
931
plain_a = [t for (a, t) in annotated_a]
932
plain_b = [t for (a, t) in annotated_b]
933
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
936
for ai, bi, l in blocks:
937
# process all mismatched sections
938
# (last mismatched section is handled because blocks always
939
# includes a 0-length last block)
940
for revision, text in annotated_a[a_cur:ai]:
941
yield status_a(revision, text)
942
for revision, text in annotated_b[b_cur:bi]:
943
yield status_b(revision, text)
945
# and now the matched section
948
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
949
assert text_a == text_b
950
yield "unchanged", text_a
953
class _KnitComponentFile(object):
954
"""One of the files used to implement a knit database"""
956
def __init__(self, transport, filename, mode, file_mode=None,
957
create_parent_dir=False):
958
self._transport = transport
959
self._filename = filename
961
self._file_mode = file_mode
962
self._create_parent_dir = create_parent_dir
963
self._need_to_create = False
965
def check_header(self, fp):
967
if line != self.HEADER:
968
raise KnitHeaderError(badline=line)
971
"""Commit is a nop."""
974
return '%s(%s)' % (self.__class__.__name__, self._filename)
977
class _KnitIndex(_KnitComponentFile):
978
"""Manages knit index file.
980
The index is already kept in memory and read on startup, to enable
981
fast lookups of revision information. The cursor of the index
982
file is always pointing to the end, making it easy to append
985
_cache is a cache for fast mapping from version id to a Index
988
_history is a cache for fast mapping from indexes to version ids.
990
The index data format is dictionary compressed when it comes to
991
parent references; a index entry may only have parents that with a
992
lover index number. As a result, the index is topological sorted.
994
Duplicate entries may be written to the index for a single version id
995
if this is done then the latter one completely replaces the former:
996
this allows updates to correct version and parent information.
997
Note that the two entries may share the delta, and that successive
998
annotations and references MUST point to the first entry.
1000
The index file on disc contains a header, followed by one line per knit
1001
record. The same revision can be present in an index file more than once.
1002
The first occurrence gets assigned a sequence number starting from 0.
1004
The format of a single line is
1005
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1006
REVISION_ID is a utf8-encoded revision id
1007
FLAGS is a comma separated list of flags about the record. Values include
1008
no-eol, line-delta, fulltext.
1009
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1010
that the the compressed data starts at.
1011
LENGTH is the ascii representation of the length of the data file.
1012
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1014
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1015
revision id already in the knit that is a parent of REVISION_ID.
1016
The ' :' marker is the end of record marker.
1019
when a write is interrupted to the index file, it will result in a line that
1020
does not end in ' :'. If the ' :' is not present at the end of a line, or at
1021
the end of the file, then the record that is missing it will be ignored by
1024
When writing new records to the index file, the data is preceded by '\n'
1025
to ensure that records always start on new lines even if the last write was
1026
interrupted. As a result its normal for the last line in the index to be
1027
missing a trailing newline. One can be added with no harmful effects.
1030
HEADER = "# bzr knit index 8\n"
1032
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1033
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1035
def _cache_version(self, version_id, options, pos, size, parents):
1036
"""Cache a version record in the history array and index cache.
1038
This is inlined into __init__ for performance. KEEP IN SYNC.
1039
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1042
# only want the _history index to reference the 1st index entry
1044
if version_id not in self._cache:
1045
index = len(self._history)
1046
self._history.append(version_id)
1048
index = self._cache[version_id][5]
1049
self._cache[version_id] = (version_id,
1056
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1057
create_parent_dir=False, delay_create=False):
1058
_KnitComponentFile.__init__(self, transport, filename, mode, file_mode,
1059
create_parent_dir=create_parent_dir)
1061
# position in _history is the 'official' index for a revision
1062
# but the values may have come from a newer entry.
1063
# so - wc -l of a knit index is != the number of unique names
1066
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1071
pb.update('read knit index', count, total)
1072
fp = self._transport.get(self._filename)
1074
self.check_header(fp)
1075
# readlines reads the whole file at once:
1076
# bad for transports like http, good for local disk
1077
# we save 60 ms doing this one change (
1078
# from calling readline each time to calling
1080
# probably what we want for nice behaviour on
1081
# http is a incremental readlines that yields, or
1082
# a check for local vs non local indexes,
1083
for l in fp.readlines():
1085
if len(rec) < 5 or rec[-1] != ':':
1087
# FIXME: in the future we should determine if its a
1088
# short write - and ignore it
1089
# or a different failure, and raise. RBC 20060407
1093
#pb.update('read knit index', count, total)
1094
# See self._parse_parents
1096
for value in rec[4:-1]:
1098
# uncompressed reference
1099
parents.append(value[1:])
1101
# this is 15/4000ms faster than isinstance,
1103
# this function is called thousands of times a
1104
# second so small variations add up.
1105
assert value.__class__ is str
1106
parents.append(self._history[int(value)])
1107
# end self._parse_parents
1108
# self._cache_version(rec[0],
1109
# rec[1].split(','),
1113
# --- self._cache_version
1114
# only want the _history index to reference the 1st
1115
# index entry for version_id
1117
if version_id not in self._cache:
1118
index = len(self._history)
1119
self._history.append(version_id)
1121
index = self._cache[version_id][5]
1122
self._cache[version_id] = (version_id,
1128
# --- self._cache_version
1131
except NoSuchFile, e:
1132
if mode != 'w' or not create:
1135
self._need_to_create = True
1137
self._transport.non_atomic_put(self._filename,
1138
StringIO(self.HEADER),
1139
mode=self._file_mode)
1142
pb.update('read knit index', total, total)
1145
def _parse_parents(self, compressed_parents):
1146
"""convert a list of string parent values into version ids.
1148
ints are looked up in the index.
1149
.FOO values are ghosts and converted in to FOO.
1151
NOTE: the function is retained here for clarity, and for possible
1152
use in partial index reads. However bulk processing now has
1153
it inlined in __init__ for inner-loop optimisation.
1156
for value in compressed_parents:
1157
if value[-1] == '.':
1158
# uncompressed reference
1159
result.append(value[1:])
1161
# this is 15/4000ms faster than isinstance,
1162
# this function is called thousands of times a
1163
# second so small variations add up.
1164
assert value.__class__ is str
1165
result.append(self._history[int(value)])
1168
def get_graph(self):
1170
for version_id, index in self._cache.iteritems():
1171
graph.append((version_id, index[4]))
1174
def get_ancestry(self, versions):
1175
"""See VersionedFile.get_ancestry."""
1176
# get a graph of all the mentioned versions:
1178
pending = set(versions)
1180
version = pending.pop()
1181
parents = self._cache[version][4]
1182
# got the parents ok
1184
parents = [parent for parent in parents if parent in self._cache]
1185
for parent in parents:
1186
# if not completed and not a ghost
1187
if parent not in graph:
1189
graph[version] = parents
1190
return topo_sort(graph.items())
1192
def get_ancestry_with_ghosts(self, versions):
1193
"""See VersionedFile.get_ancestry_with_ghosts."""
1194
# get a graph of all the mentioned versions:
1196
pending = set(versions)
1198
version = pending.pop()
1200
parents = self._cache[version][4]
1206
# got the parents ok
1207
for parent in parents:
1208
if parent not in graph:
1210
graph[version] = parents
1211
return topo_sort(graph.items())
1213
def num_versions(self):
1214
return len(self._history)
1216
__len__ = num_versions
1218
def get_versions(self):
1219
return self._history
1221
def idx_to_name(self, idx):
1222
return self._history[idx]
1224
def lookup(self, version_id):
1225
assert version_id in self._cache
1226
return self._cache[version_id][5]
1228
def _version_list_to_index(self, versions):
1229
encode_utf8 = cache_utf8.encode
1231
for version in versions:
1232
if version in self._cache:
1233
# -- inlined lookup() --
1234
result_list.append(str(self._cache[version][5]))
1235
# -- end lookup () --
1237
result_list.append('.' + encode_utf8(version))
1238
return ' '.join(result_list)
1240
def add_version(self, version_id, options, pos, size, parents):
1241
"""Add a version record to the index."""
1242
self.add_versions(((version_id, options, pos, size, parents),))
1244
def add_versions(self, versions):
1245
"""Add multiple versions to the index.
1247
:param versions: a list of tuples:
1248
(version_id, options, pos, size, parents).
1251
encode_utf8 = cache_utf8.encode
1252
for version_id, options, pos, size, parents in versions:
1253
line = "\n%s %s %s %s %s :" % (encode_utf8(version_id),
1257
self._version_list_to_index(parents))
1258
assert isinstance(line, str), \
1259
'content must be utf-8 encoded: %r' % (line,)
1261
if not self._need_to_create:
1262
self._transport.append(self._filename, StringIO(''.join(lines)))
1265
sio.write(self.HEADER)
1266
sio.writelines(lines)
1268
self._transport.non_atomic_put(self._filename, sio,
1269
create_parent_dir=self._create_parent_dir,
1270
mode=self._file_mode)
1271
self._need_to_create = False
1273
# cache after writing, so that a failed write leads to missing cache
1274
# entries not extra ones. XXX TODO: RBC 20060502 in the event of a
1275
# failure, reload the index or flush it or some such, to prevent
1276
# writing records that did complete twice.
1277
for version_id, options, pos, size, parents in versions:
1278
self._cache_version(version_id, options, pos, size, parents)
1280
def has_version(self, version_id):
1281
"""True if the version is in the index."""
1282
return self._cache.has_key(version_id)
1284
def get_position(self, version_id):
1285
"""Return data position and size of specified version."""
1286
return (self._cache[version_id][2], \
1287
self._cache[version_id][3])
1289
def get_method(self, version_id):
1290
"""Return compression method of specified version."""
1291
options = self._cache[version_id][1]
1292
if 'fulltext' in options:
1295
assert 'line-delta' in options
1298
def get_options(self, version_id):
1299
return self._cache[version_id][1]
1301
def get_parents(self, version_id):
1302
"""Return parents of specified version ignoring ghosts."""
1303
return [parent for parent in self._cache[version_id][4]
1304
if parent in self._cache]
1306
def get_parents_with_ghosts(self, version_id):
1307
"""Return parents of specified version with ghosts."""
1308
return self._cache[version_id][4]
1310
def check_versions_present(self, version_ids):
1311
"""Check that all specified versions are present."""
1312
version_ids = set(version_ids)
1313
for version_id in list(version_ids):
1314
if version_id in self._cache:
1315
version_ids.remove(version_id)
1317
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1320
class _KnitData(_KnitComponentFile):
1321
"""Contents of the knit data file"""
1323
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1324
create_parent_dir=False, delay_create=False):
1325
_KnitComponentFile.__init__(self, transport, filename, mode,
1326
file_mode=file_mode,
1327
create_parent_dir=create_parent_dir)
1328
self._checked = False
1329
# TODO: jam 20060713 conceptually, this could spill to disk
1330
# if the cached size gets larger than a certain amount
1331
# but it complicates the model a bit, so for now just use
1332
# a simple dictionary
1334
self._do_cache = False
1337
self._need_to_create = create
1339
self._transport.non_atomic_put(self._filename, StringIO(''),
1340
mode=self._file_mode)
1342
def enable_cache(self):
1343
"""Enable caching of reads."""
1344
self._do_cache = True
1346
def clear_cache(self):
1347
"""Clear the record cache."""
1348
self._do_cache = False
1351
def _open_file(self):
1353
return self._transport.get(self._filename)
1358
def _record_to_data(self, version_id, digest, lines):
1359
"""Convert version_id, digest, lines into a raw data block.
1361
:return: (len, a StringIO instance with the raw data ready to read.)
1364
data_file = GzipFile(None, mode='wb', fileobj=sio)
1366
version_id_utf8 = cache_utf8.encode(version_id)
1367
data_file.writelines(chain(
1368
["version %s %d %s\n" % (version_id_utf8,
1372
["end %s\n" % version_id_utf8]))
1379
def add_raw_record(self, raw_data):
1380
"""Append a prepared record to the data file.
1382
:return: the offset in the data file raw_data was written.
1384
assert isinstance(raw_data, str), 'data must be plain bytes'
1385
if not self._need_to_create:
1386
return self._transport.append(self._filename, StringIO(raw_data))
1388
self._transport.non_atomic_put(self._filename, StringIO(raw_data),
1389
create_parent_dir=self._create_parent_dir,
1390
mode=self._file_mode)
1391
self._need_to_create = False
1394
def add_record(self, version_id, digest, lines):
1395
"""Write new text record to disk. Returns the position in the
1396
file where it was written."""
1397
size, sio = self._record_to_data(version_id, digest, lines)
1399
if not self._need_to_create:
1400
start_pos = self._transport.append(self._filename, sio)
1402
self._transport.non_atomic_put(self._filename, sio,
1403
create_parent_dir=self._create_parent_dir,
1404
mode=self._file_mode)
1405
self._need_to_create = False
1408
self._cache[version_id] = sio.getvalue()
1409
return start_pos, size
1411
def _parse_record_header(self, version_id, raw_data):
1412
"""Parse a record header for consistency.
1414
:return: the header and the decompressor stream.
1415
as (stream, header_record)
1417
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1418
rec = df.readline().split()
1420
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
1421
if cache_utf8.decode(rec[1]) != version_id:
1422
raise KnitCorrupt(self._filename,
1423
'unexpected version, wanted %r, got %r' % (
1424
version_id, rec[1]))
1427
def _parse_record(self, version_id, data):
1429
# 4168 calls in 2880 217 internal
1430
# 4168 calls to _parse_record_header in 2121
1431
# 4168 calls to readlines in 330
1432
df, rec = self._parse_record_header(version_id, data)
1433
record_contents = df.readlines()
1434
l = record_contents.pop()
1435
assert len(record_contents) == int(rec[2])
1436
if l != 'end %s\n' % cache_utf8.encode(version_id):
1437
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
1440
return record_contents, rec[3]
1442
def read_records_iter_raw(self, records):
1443
"""Read text records from data file and yield raw data.
1445
This unpacks enough of the text record to validate the id is
1446
as expected but thats all.
1448
# setup an iterator of the external records:
1449
# uses readv so nice and fast we hope.
1451
# grab the disk data needed.
1453
# Don't check _cache if it is empty
1454
needed_offsets = [(pos, size) for version_id, pos, size
1456
if version_id not in self._cache]
1458
needed_offsets = [(pos, size) for version_id, pos, size
1461
raw_records = self._transport.readv(self._filename, needed_offsets)
1464
for version_id, pos, size in records:
1465
if version_id in self._cache:
1466
# This data has already been validated
1467
data = self._cache[version_id]
1469
pos, data = raw_records.next()
1471
self._cache[version_id] = data
1473
# validate the header
1474
df, rec = self._parse_record_header(version_id, data)
1476
yield version_id, data
1478
def read_records_iter(self, records):
1479
"""Read text records from data file and yield result.
1481
The result will be returned in whatever is the fastest to read.
1482
Not by the order requested. Also, multiple requests for the same
1483
record will only yield 1 response.
1484
:param records: A list of (version_id, pos, len) entries
1485
:return: Yields (version_id, contents, digest) in the order
1486
read, not the order requested
1492
# Skip records we have alread seen
1493
yielded_records = set()
1494
needed_records = set()
1495
for record in records:
1496
if record[0] in self._cache:
1497
if record[0] in yielded_records:
1499
yielded_records.add(record[0])
1500
data = self._cache[record[0]]
1501
content, digest = self._parse_record(record[0], data)
1502
yield (record[0], content, digest)
1504
needed_records.add(record)
1505
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1507
needed_records = sorted(set(records), key=operator.itemgetter(1))
1509
if not needed_records:
1512
# The transport optimizes the fetching as well
1513
# (ie, reads continuous ranges.)
1514
readv_response = self._transport.readv(self._filename,
1515
[(pos, size) for version_id, pos, size in needed_records])
1517
for (version_id, pos, size), (pos, data) in \
1518
izip(iter(needed_records), readv_response):
1519
content, digest = self._parse_record(version_id, data)
1521
self._cache[version_id] = data
1522
yield version_id, content, digest
1524
def read_records(self, records):
1525
"""Read records into a dictionary."""
1527
for record_id, content, digest in \
1528
self.read_records_iter(records):
1529
components[record_id] = (content, digest)
1533
class InterKnit(InterVersionedFile):
1534
"""Optimised code paths for knit to knit operations."""
1536
_matching_file_from_factory = KnitVersionedFile
1537
_matching_file_to_factory = KnitVersionedFile
1540
def is_compatible(source, target):
1541
"""Be compatible with knits. """
1543
return (isinstance(source, KnitVersionedFile) and
1544
isinstance(target, KnitVersionedFile))
1545
except AttributeError:
1548
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1549
"""See InterVersionedFile.join."""
1550
assert isinstance(self.source, KnitVersionedFile)
1551
assert isinstance(self.target, KnitVersionedFile)
1553
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1558
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1560
version_ids = list(version_ids)
1561
if None in version_ids:
1562
version_ids.remove(None)
1564
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1565
this_versions = set(self.target._index.get_versions())
1566
needed_versions = self.source_ancestry - this_versions
1567
cross_check_versions = self.source_ancestry.intersection(this_versions)
1568
mismatched_versions = set()
1569
for version in cross_check_versions:
1570
# scan to include needed parents.
1571
n1 = set(self.target.get_parents_with_ghosts(version))
1572
n2 = set(self.source.get_parents_with_ghosts(version))
1574
# FIXME TEST this check for cycles being introduced works
1575
# the logic is we have a cycle if in our graph we are an
1576
# ancestor of any of the n2 revisions.
1582
parent_ancestors = self.source.get_ancestry(parent)
1583
if version in parent_ancestors:
1584
raise errors.GraphCycleError([parent, version])
1585
# ensure this parent will be available later.
1586
new_parents = n2.difference(n1)
1587
needed_versions.update(new_parents.difference(this_versions))
1588
mismatched_versions.add(version)
1590
if not needed_versions and not mismatched_versions:
1592
full_list = topo_sort(self.source.get_graph())
1594
version_list = [i for i in full_list if (not self.target.has_version(i)
1595
and i in needed_versions)]
1599
copy_queue_records = []
1601
for version_id in version_list:
1602
options = self.source._index.get_options(version_id)
1603
parents = self.source._index.get_parents_with_ghosts(version_id)
1604
# check that its will be a consistent copy:
1605
for parent in parents:
1606
# if source has the parent, we must :
1607
# * already have it or
1608
# * have it scheduled already
1609
# otherwise we don't care
1610
assert (self.target.has_version(parent) or
1611
parent in copy_set or
1612
not self.source.has_version(parent))
1613
data_pos, data_size = self.source._index.get_position(version_id)
1614
copy_queue_records.append((version_id, data_pos, data_size))
1615
copy_queue.append((version_id, options, parents))
1616
copy_set.add(version_id)
1618
# data suck the join:
1620
total = len(version_list)
1623
for (version_id, raw_data), \
1624
(version_id2, options, parents) in \
1625
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1627
assert version_id == version_id2, 'logic error, inconsistent results'
1629
pb.update("Joining knit", count, total)
1630
raw_records.append((version_id, options, parents, len(raw_data)))
1631
raw_datum.append(raw_data)
1632
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1634
for version in mismatched_versions:
1635
# FIXME RBC 20060309 is this needed?
1636
n1 = set(self.target.get_parents_with_ghosts(version))
1637
n2 = set(self.source.get_parents_with_ghosts(version))
1638
# write a combined record to our history preserving the current
1639
# parents as first in the list
1640
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1641
self.target.fix_parents(version, new_parents)
1647
InterVersionedFile.register_optimiser(InterKnit)
1650
class WeaveToKnit(InterVersionedFile):
1651
"""Optimised code paths for weave to knit operations."""
1653
_matching_file_from_factory = bzrlib.weave.WeaveFile
1654
_matching_file_to_factory = KnitVersionedFile
1657
def is_compatible(source, target):
1658
"""Be compatible with weaves to knits."""
1660
return (isinstance(source, bzrlib.weave.Weave) and
1661
isinstance(target, KnitVersionedFile))
1662
except AttributeError:
1665
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1666
"""See InterVersionedFile.join."""
1667
assert isinstance(self.source, bzrlib.weave.Weave)
1668
assert isinstance(self.target, KnitVersionedFile)
1670
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1675
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1677
version_ids = list(version_ids)
1679
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1680
this_versions = set(self.target._index.get_versions())
1681
needed_versions = self.source_ancestry - this_versions
1682
cross_check_versions = self.source_ancestry.intersection(this_versions)
1683
mismatched_versions = set()
1684
for version in cross_check_versions:
1685
# scan to include needed parents.
1686
n1 = set(self.target.get_parents_with_ghosts(version))
1687
n2 = set(self.source.get_parents(version))
1688
# if all of n2's parents are in n1, then its fine.
1689
if n2.difference(n1):
1690
# FIXME TEST this check for cycles being introduced works
1691
# the logic is we have a cycle if in our graph we are an
1692
# ancestor of any of the n2 revisions.
1698
parent_ancestors = self.source.get_ancestry(parent)
1699
if version in parent_ancestors:
1700
raise errors.GraphCycleError([parent, version])
1701
# ensure this parent will be available later.
1702
new_parents = n2.difference(n1)
1703
needed_versions.update(new_parents.difference(this_versions))
1704
mismatched_versions.add(version)
1706
if not needed_versions and not mismatched_versions:
1708
full_list = topo_sort(self.source.get_graph())
1710
version_list = [i for i in full_list if (not self.target.has_version(i)
1711
and i in needed_versions)]
1715
total = len(version_list)
1716
for version_id in version_list:
1717
pb.update("Converting to knit", count, total)
1718
parents = self.source.get_parents(version_id)
1719
# check that its will be a consistent copy:
1720
for parent in parents:
1721
# if source has the parent, we must already have it
1722
assert (self.target.has_version(parent))
1723
self.target.add_lines(
1724
version_id, parents, self.source.get_lines(version_id))
1727
for version in mismatched_versions:
1728
# FIXME RBC 20060309 is this needed?
1729
n1 = set(self.target.get_parents_with_ghosts(version))
1730
n2 = set(self.source.get_parents(version))
1731
# write a combined record to our history preserving the current
1732
# parents as first in the list
1733
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1734
self.target.fix_parents(version, new_parents)
1740
InterVersionedFile.register_optimiser(WeaveToKnit)
1743
class KnitSequenceMatcher(difflib.SequenceMatcher):
1744
"""Knit tuned sequence matcher.
1746
This is based on profiling of difflib which indicated some improvements
1747
for our usage pattern.
1750
def find_longest_match(self, alo, ahi, blo, bhi):
1751
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1753
If isjunk is not defined:
1755
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1756
alo <= i <= i+k <= ahi
1757
blo <= j <= j+k <= bhi
1758
and for all (i',j',k') meeting those conditions,
1761
and if i == i', j <= j'
1763
In other words, of all maximal matching blocks, return one that
1764
starts earliest in a, and of all those maximal matching blocks that
1765
start earliest in a, return the one that starts earliest in b.
1767
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1768
>>> s.find_longest_match(0, 5, 0, 9)
1771
If isjunk is defined, first the longest matching block is
1772
determined as above, but with the additional restriction that no
1773
junk element appears in the block. Then that block is extended as
1774
far as possible by matching (only) junk elements on both sides. So
1775
the resulting block never matches on junk except as identical junk
1776
happens to be adjacent to an "interesting" match.
1778
Here's the same example as before, but considering blanks to be
1779
junk. That prevents " abcd" from matching the " abcd" at the tail
1780
end of the second sequence directly. Instead only the "abcd" can
1781
match, and matches the leftmost "abcd" in the second sequence:
1783
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1784
>>> s.find_longest_match(0, 5, 0, 9)
1787
If no blocks match, return (alo, blo, 0).
1789
>>> s = SequenceMatcher(None, "ab", "c")
1790
>>> s.find_longest_match(0, 2, 0, 1)
1794
# CAUTION: stripping common prefix or suffix would be incorrect.
1798
# Longest matching block is "ab", but if common prefix is
1799
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1800
# strip, so ends up claiming that ab is changed to acab by
1801
# inserting "ca" in the middle. That's minimal but unintuitive:
1802
# "it's obvious" that someone inserted "ac" at the front.
1803
# Windiff ends up at the same place as diff, but by pairing up
1804
# the unique 'b's and then matching the first two 'a's.
1806
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1807
besti, bestj, bestsize = alo, blo, 0
1808
# find longest junk-free match
1809
# during an iteration of the loop, j2len[j] = length of longest
1810
# junk-free match ending with a[i-1] and b[j]
1814
for i in xrange(alo, ahi):
1815
# look at all instances of a[i] in b; note that because
1816
# b2j has no junk keys, the loop is skipped if a[i] is junk
1817
j2lenget = j2len.get
1820
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1821
# following improvement
1822
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1823
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1824
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1826
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1827
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1828
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1840
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1842
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1845
# Extend the best by non-junk elements on each end. In particular,
1846
# "popular" non-junk elements aren't in b2j, which greatly speeds
1847
# the inner loop above, but also means "the best" match so far
1848
# doesn't contain any junk *or* popular non-junk elements.
1849
while besti > alo and bestj > blo and \
1850
not isbjunk(b[bestj-1]) and \
1851
a[besti-1] == b[bestj-1]:
1852
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1853
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1854
not isbjunk(b[bestj+bestsize]) and \
1855
a[besti+bestsize] == b[bestj+bestsize]:
1858
# Now that we have a wholly interesting match (albeit possibly
1859
# empty!), we may as well suck up the matching junk on each
1860
# side of it too. Can't think of a good reason not to, and it
1861
# saves post-processing the (possibly considerable) expense of
1862
# figuring out what to do with it. In the case of an empty
1863
# interesting match, this is clearly the right thing to do,
1864
# because no other kind of match is possible in the regions.
1865
while besti > alo and bestj > blo and \
1866
isbjunk(b[bestj-1]) and \
1867
a[besti-1] == b[bestj-1]:
1868
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1869
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1870
isbjunk(b[bestj+bestsize]) and \
1871
a[besti+bestsize] == b[bestj+bestsize]:
1872
bestsize = bestsize + 1
1874
return besti, bestj, bestsize
added
removed
bzrlib/knit.py
