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,
286
"""Construct a knit at location specified by relpath.
288
:param create: If not True, only open an existing knit.
289
:param create_parent_dir: If True, create the parent directory if
290
creating the file fails. (This is used for stores with
291
hash-prefixes that may not exist yet)
292
:param delay_create: The calling code is aware that the knit won't
293
actually be created until the first data is stored.
295
if deprecated_passed(basis_knit):
296
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
297
" deprecated as of bzr 0.9.",
298
DeprecationWarning, stacklevel=2)
299
if access_mode is None:
301
super(KnitVersionedFile, self).__init__(access_mode)
302
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
303
self.transport = transport
304
self.filename = relpath
305
self.factory = factory or KnitAnnotateFactory()
306
self.writable = (access_mode == 'w')
309
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
310
access_mode, create=create, file_mode=file_mode,
311
create_parent_dir=create_parent_dir, delay_create=delay_create,
313
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
314
access_mode, create=create and not len(self), file_mode=file_mode,
315
create_parent_dir=create_parent_dir, delay_create=delay_create,
319
return '%s(%s)' % (self.__class__.__name__,
320
self.transport.abspath(self.filename))
322
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
323
"""See VersionedFile._add_delta()."""
324
self._check_add(version_id, []) # should we check the lines ?
325
self._check_versions_present(parents)
329
for parent in parents:
330
if not self.has_version(parent):
331
ghosts.append(parent)
333
present_parents.append(parent)
335
if delta_parent is None:
336
# reconstitute as full text.
337
assert len(delta) == 1 or len(delta) == 0
339
assert delta[0][0] == 0
340
assert delta[0][1] == 0, delta[0][1]
341
return super(KnitVersionedFile, self)._add_delta(version_id,
352
options.append('no-eol')
354
if delta_parent is not None:
355
# determine the current delta chain length.
356
# To speed the extract of texts the delta chain is limited
357
# to a fixed number of deltas. This should minimize both
358
# I/O and the time spend applying deltas.
360
delta_parents = [delta_parent]
362
parent = delta_parents[0]
363
method = self._index.get_method(parent)
364
if method == 'fulltext':
366
delta_parents = self._index.get_parents(parent)
368
if method == 'line-delta':
369
# did not find a fulltext in the delta limit.
370
# just do a normal insertion.
371
return super(KnitVersionedFile, self)._add_delta(version_id,
378
options.append('line-delta')
379
store_lines = self.factory.lower_line_delta(delta)
381
where, size = self._data.add_record(version_id, digest, store_lines)
382
self._index.add_version(version_id, options, where, size, parents)
384
def _add_raw_records(self, records, data):
385
"""Add all the records 'records' with data pre-joined in 'data'.
387
:param records: A list of tuples(version_id, options, parents, size).
388
:param data: The data for the records. When it is written, the records
389
are adjusted to have pos pointing into data by the sum of
390
the preceding records sizes.
393
pos = self._data.add_raw_record(data)
396
for (version_id, options, parents, size) in records:
397
index_entries.append((version_id, options, pos+offset,
399
if self._data._do_cache:
400
self._data._cache[version_id] = data[offset:offset+size]
402
self._index.add_versions(index_entries)
404
def enable_cache(self):
405
"""Start caching data for this knit"""
406
self._data.enable_cache()
408
def clear_cache(self):
409
"""Clear the data cache only."""
410
self._data.clear_cache()
412
def copy_to(self, name, transport):
413
"""See VersionedFile.copy_to()."""
414
# copy the current index to a temp index to avoid racing with local
416
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
417
self.transport.get(self._index._filename))
419
f = self._data._open_file()
421
transport.put_file(name + DATA_SUFFIX, f)
424
# move the copied index into place
425
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
427
def create_empty(self, name, transport, mode=None):
428
return KnitVersionedFile(name, transport, factory=self.factory,
429
delta=self.delta, create=True)
431
def _fix_parents(self, version, new_parents):
432
"""Fix the parents list for version.
434
This is done by appending a new version to the index
435
with identical data except for the parents list.
436
the parents list must be a superset of the current
439
current_values = self._index._cache[version]
440
assert set(current_values[4]).difference(set(new_parents)) == set()
441
self._index.add_version(version,
447
def get_delta(self, version_id):
448
"""Get a delta for constructing version from some other version."""
449
if not self.has_version(version_id):
450
raise RevisionNotPresent(version_id, self.filename)
452
parents = self.get_parents(version_id)
457
data_pos, data_size = self._index.get_position(version_id)
458
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
459
version_idx = self._index.lookup(version_id)
460
noeol = 'no-eol' in self._index.get_options(version_id)
461
if 'fulltext' == self._index.get_method(version_id):
462
new_content = self.factory.parse_fulltext(data, version_idx)
463
if parent is not None:
464
reference_content = self._get_content(parent)
465
old_texts = reference_content.text()
468
new_texts = new_content.text()
469
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
470
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
472
delta = self.factory.parse_line_delta(data, version_idx)
473
return parent, sha1, noeol, delta
475
def get_graph_with_ghosts(self):
476
"""See VersionedFile.get_graph_with_ghosts()."""
477
graph_items = self._index.get_graph()
478
return dict(graph_items)
480
def get_sha1(self, version_id):
481
"""See VersionedFile.get_sha1()."""
482
record_map = self._get_record_map([version_id])
483
method, content, digest, next = record_map[version_id]
488
"""See VersionedFile.get_suffixes()."""
489
return [DATA_SUFFIX, INDEX_SUFFIX]
491
def has_ghost(self, version_id):
492
"""True if there is a ghost reference in the file to version_id."""
494
if self.has_version(version_id):
496
# optimisable if needed by memoising the _ghosts set.
497
items = self._index.get_graph()
498
for node, parents in items:
499
for parent in parents:
500
if parent not in self._index._cache:
501
if parent == version_id:
506
"""See VersionedFile.versions."""
507
return self._index.get_versions()
509
def has_version(self, version_id):
510
"""See VersionedFile.has_version."""
511
return self._index.has_version(version_id)
513
__contains__ = has_version
515
def _merge_annotations(self, content, parents, parent_texts={},
516
delta=None, annotated=None):
517
"""Merge annotations for content. This is done by comparing
518
the annotations based on changed to the text.
522
for parent_id in parents:
523
merge_content = self._get_content(parent_id, parent_texts)
524
seq = KnitSequenceMatcher(None, merge_content.text(), content.text())
525
if delta_seq is None:
526
# setup a delta seq to reuse.
528
for i, j, n in seq.get_matching_blocks():
531
# this appears to copy (origin, text) pairs across to the new
532
# content for any line that matches the last-checked parent.
533
# FIXME: save the sequence control data for delta compression
534
# against the most relevant parent rather than rediffing.
535
content._lines[j:j+n] = merge_content._lines[i:i+n]
538
reference_content = self._get_content(parents[0], parent_texts)
539
new_texts = content.text()
540
old_texts = reference_content.text()
541
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
542
return self._make_line_delta(delta_seq, content)
544
def _make_line_delta(self, delta_seq, new_content):
545
"""Generate a line delta from delta_seq and new_content."""
547
for op in delta_seq.get_opcodes():
550
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
553
def _get_components_positions(self, version_ids):
554
"""Produce a map of position data for the components of versions.
556
This data is intended to be used for retrieving the knit records.
558
A dict of version_id to (method, data_pos, data_size, next) is
560
method is the way referenced data should be applied.
561
data_pos is the position of the data in the knit.
562
data_size is the size of the data in the knit.
563
next is the build-parent of the version, or None for fulltexts.
566
for version_id in version_ids:
569
while cursor is not None and cursor not in component_data:
570
method = self._index.get_method(cursor)
571
if method == 'fulltext':
574
next = self.get_parents(cursor)[0]
575
data_pos, data_size = self._index.get_position(cursor)
576
component_data[cursor] = (method, data_pos, data_size, next)
578
return component_data
580
def _get_content(self, version_id, parent_texts={}):
581
"""Returns a content object that makes up the specified
583
if not self.has_version(version_id):
584
raise RevisionNotPresent(version_id, self.filename)
586
cached_version = parent_texts.get(version_id, None)
587
if cached_version is not None:
588
return cached_version
590
text_map, contents_map = self._get_content_maps([version_id])
591
return contents_map[version_id]
593
def _check_versions_present(self, version_ids):
594
"""Check that all specified versions are present."""
595
version_ids = set(version_ids)
596
for r in list(version_ids):
597
if self._index.has_version(r):
598
version_ids.remove(r)
600
raise RevisionNotPresent(list(version_ids)[0], self.filename)
602
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
603
"""See VersionedFile.add_lines_with_ghosts()."""
604
self._check_add(version_id, lines)
605
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
607
def _add_lines(self, version_id, parents, lines, parent_texts):
608
"""See VersionedFile.add_lines."""
609
self._check_add(version_id, lines)
610
self._check_versions_present(parents)
611
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
613
def _check_add(self, version_id, lines):
614
"""check that version_id and lines are safe to add."""
615
assert self.writable, "knit is not opened for write"
616
### FIXME escape. RBC 20060228
617
if contains_whitespace(version_id):
618
raise InvalidRevisionId(version_id, self.filename)
619
if self.has_version(version_id):
620
raise RevisionAlreadyPresent(version_id, self.filename)
621
self._check_lines_not_unicode(lines)
622
self._check_lines_are_lines(lines)
624
def _add(self, version_id, lines, parents, delta, parent_texts):
625
"""Add a set of lines on top of version specified by parents.
627
If delta is true, compress the text as a line-delta against
630
Any versions not present will be converted into ghosts.
632
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
633
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
634
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
635
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
636
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
637
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
638
# +1383 0 8.0370 8.0370 +<len>
639
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
640
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
641
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
642
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
646
if parent_texts is None:
648
for parent in parents:
649
if not self.has_version(parent):
650
ghosts.append(parent)
652
present_parents.append(parent)
654
if delta and not len(present_parents):
657
digest = sha_strings(lines)
660
if lines[-1][-1] != '\n':
661
options.append('no-eol')
662
lines[-1] = lines[-1] + '\n'
664
if len(present_parents) and delta:
665
# To speed the extract of texts the delta chain is limited
666
# to a fixed number of deltas. This should minimize both
667
# I/O and the time spend applying deltas.
669
delta_parents = present_parents
671
parent = delta_parents[0]
672
method = self._index.get_method(parent)
673
if method == 'fulltext':
675
delta_parents = self._index.get_parents(parent)
677
if method == 'line-delta':
680
lines = self.factory.make(lines, version_id)
681
if delta or (self.factory.annotated and len(present_parents) > 0):
682
# Merge annotations from parent texts if so is needed.
683
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
684
delta, self.factory.annotated)
687
options.append('line-delta')
688
store_lines = self.factory.lower_line_delta(delta_hunks)
690
options.append('fulltext')
691
store_lines = self.factory.lower_fulltext(lines)
693
where, size = self._data.add_record(version_id, digest, store_lines)
694
self._index.add_version(version_id, options, where, size, parents)
697
def check(self, progress_bar=None):
698
"""See VersionedFile.check()."""
700
def _clone_text(self, new_version_id, old_version_id, parents):
701
"""See VersionedFile.clone_text()."""
702
# FIXME RBC 20060228 make fast by only inserting an index with null
704
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
706
def get_lines(self, version_id):
707
"""See VersionedFile.get_lines()."""
708
return self.get_line_list([version_id])[0]
710
def _get_record_map(self, version_ids):
711
"""Produce a dictionary of knit records.
713
The keys are version_ids, the values are tuples of (method, content,
715
method is the way the content should be applied.
716
content is a KnitContent object.
717
digest is the SHA1 digest of this version id after all steps are done
718
next is the build-parent of the version, i.e. the leftmost ancestor.
719
If the method is fulltext, next will be None.
721
position_map = self._get_components_positions(version_ids)
722
# c = component_id, m = method, p = position, s = size, n = next
723
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
725
for component_id, content, digest in \
726
self._data.read_records_iter(records):
727
method, position, size, next = position_map[component_id]
728
record_map[component_id] = method, content, digest, next
732
def get_text(self, version_id):
733
"""See VersionedFile.get_text"""
734
return self.get_texts([version_id])[0]
736
def get_texts(self, version_ids):
737
return [''.join(l) for l in self.get_line_list(version_ids)]
739
def get_line_list(self, version_ids):
740
"""Return the texts of listed versions as a list of strings."""
741
text_map, content_map = self._get_content_maps(version_ids)
742
return [text_map[v] for v in version_ids]
744
def _get_content_maps(self, version_ids):
745
"""Produce maps of text and KnitContents
747
:return: (text_map, content_map) where text_map contains the texts for
748
the requested versions and content_map contains the KnitContents.
749
Both dicts take version_ids as their keys.
751
for version_id in version_ids:
752
if not self.has_version(version_id):
753
raise RevisionNotPresent(version_id, self.filename)
754
record_map = self._get_record_map(version_ids)
759
for version_id in version_ids:
762
while cursor is not None:
763
method, data, digest, next = record_map[cursor]
764
components.append((cursor, method, data, digest))
765
if cursor in content_map:
770
for component_id, method, data, digest in reversed(components):
771
if component_id in content_map:
772
content = content_map[component_id]
774
version_idx = self._index.lookup(component_id)
775
if method == 'fulltext':
776
assert content is None
777
content = self.factory.parse_fulltext(data, version_idx)
778
elif method == 'line-delta':
779
delta = self.factory.parse_line_delta(data[:],
781
content = content.copy()
782
content._lines = self._apply_delta(content._lines,
784
content_map[component_id] = content
786
if 'no-eol' in self._index.get_options(version_id):
787
content = content.copy()
788
line = content._lines[-1][1].rstrip('\n')
789
content._lines[-1] = (content._lines[-1][0], line)
790
final_content[version_id] = content
792
# digest here is the digest from the last applied component.
793
text = content.text()
794
if sha_strings(text) != digest:
795
raise KnitCorrupt(self.filename,
796
'sha-1 does not match %s' % version_id)
798
text_map[version_id] = text
799
return text_map, final_content
801
def iter_lines_added_or_present_in_versions(self, version_ids=None):
802
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
803
if version_ids is None:
804
version_ids = self.versions()
805
# we don't care about inclusions, the caller cares.
806
# but we need to setup a list of records to visit.
807
# we need version_id, position, length
808
version_id_records = []
809
requested_versions = list(version_ids)
810
# filter for available versions
811
for version_id in requested_versions:
812
if not self.has_version(version_id):
813
raise RevisionNotPresent(version_id, self.filename)
814
# get a in-component-order queue:
816
for version_id in self.versions():
817
if version_id in requested_versions:
818
version_ids.append(version_id)
819
data_pos, length = self._index.get_position(version_id)
820
version_id_records.append((version_id, data_pos, length))
822
pb = bzrlib.ui.ui_factory.nested_progress_bar()
824
total = len(version_id_records)
826
pb.update('Walking content.', count, total)
827
for version_id, data, sha_value in \
828
self._data.read_records_iter(version_id_records):
829
pb.update('Walking content.', count, total)
830
method = self._index.get_method(version_id)
831
version_idx = self._index.lookup(version_id)
832
assert method in ('fulltext', 'line-delta')
833
if method == 'fulltext':
834
content = self.factory.parse_fulltext(data, version_idx)
835
for line in content.text():
838
delta = self.factory.parse_line_delta(data, version_idx)
839
for start, end, count, lines in delta:
840
for origin, line in lines:
843
pb.update('Walking content.', total, total)
846
pb.update('Walking content.', total, total)
850
def num_versions(self):
851
"""See VersionedFile.num_versions()."""
852
return self._index.num_versions()
854
__len__ = num_versions
856
def annotate_iter(self, version_id):
857
"""See VersionedFile.annotate_iter."""
858
content = self._get_content(version_id)
859
for origin, text in content.annotate_iter():
862
def get_parents(self, version_id):
863
"""See VersionedFile.get_parents."""
866
# 52554 calls in 1264 872 internal down from 3674
868
return self._index.get_parents(version_id)
870
raise RevisionNotPresent(version_id, self.filename)
872
def get_parents_with_ghosts(self, version_id):
873
"""See VersionedFile.get_parents."""
875
return self._index.get_parents_with_ghosts(version_id)
877
raise RevisionNotPresent(version_id, self.filename)
879
def get_ancestry(self, versions):
880
"""See VersionedFile.get_ancestry."""
881
if isinstance(versions, basestring):
882
versions = [versions]
885
self._check_versions_present(versions)
886
return self._index.get_ancestry(versions)
888
def get_ancestry_with_ghosts(self, versions):
889
"""See VersionedFile.get_ancestry_with_ghosts."""
890
if isinstance(versions, basestring):
891
versions = [versions]
894
self._check_versions_present(versions)
895
return self._index.get_ancestry_with_ghosts(versions)
897
#@deprecated_method(zero_eight)
898
def walk(self, version_ids):
899
"""See VersionedFile.walk."""
900
# We take the short path here, and extract all relevant texts
901
# and put them in a weave and let that do all the work. Far
902
# from optimal, but is much simpler.
903
# FIXME RB 20060228 this really is inefficient!
904
from bzrlib.weave import Weave
906
w = Weave(self.filename)
907
ancestry = self.get_ancestry(version_ids)
908
sorted_graph = topo_sort(self._index.get_graph())
909
version_list = [vid for vid in sorted_graph if vid in ancestry]
911
for version_id in version_list:
912
lines = self.get_lines(version_id)
913
w.add_lines(version_id, self.get_parents(version_id), lines)
915
for lineno, insert_id, dset, line in w.walk(version_ids):
916
yield lineno, insert_id, dset, line
918
def plan_merge(self, ver_a, ver_b):
919
"""See VersionedFile.plan_merge."""
920
ancestors_b = set(self.get_ancestry(ver_b))
921
def status_a(revision, text):
922
if revision in ancestors_b:
923
return 'killed-b', text
927
ancestors_a = set(self.get_ancestry(ver_a))
928
def status_b(revision, text):
929
if revision in ancestors_a:
930
return 'killed-a', text
934
annotated_a = self.annotate(ver_a)
935
annotated_b = self.annotate(ver_b)
936
plain_a = [t for (a, t) in annotated_a]
937
plain_b = [t for (a, t) in annotated_b]
938
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
941
for ai, bi, l in blocks:
942
# process all mismatched sections
943
# (last mismatched section is handled because blocks always
944
# includes a 0-length last block)
945
for revision, text in annotated_a[a_cur:ai]:
946
yield status_a(revision, text)
947
for revision, text in annotated_b[b_cur:bi]:
948
yield status_b(revision, text)
950
# and now the matched section
953
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
954
assert text_a == text_b
955
yield "unchanged", text_a
958
class _KnitComponentFile(object):
959
"""One of the files used to implement a knit database"""
961
def __init__(self, transport, filename, mode, file_mode=None,
962
create_parent_dir=False, dir_mode=None):
963
self._transport = transport
964
self._filename = filename
966
self._file_mode = file_mode
967
self._dir_mode = dir_mode
968
self._create_parent_dir = create_parent_dir
969
self._need_to_create = False
971
def check_header(self, fp):
973
if line != self.HEADER:
974
raise KnitHeaderError(badline=line)
977
"""Commit is a nop."""
980
return '%s(%s)' % (self.__class__.__name__, self._filename)
983
class _KnitIndex(_KnitComponentFile):
984
"""Manages knit index file.
986
The index is already kept in memory and read on startup, to enable
987
fast lookups of revision information. The cursor of the index
988
file is always pointing to the end, making it easy to append
991
_cache is a cache for fast mapping from version id to a Index
994
_history is a cache for fast mapping from indexes to version ids.
996
The index data format is dictionary compressed when it comes to
997
parent references; a index entry may only have parents that with a
998
lover index number. As a result, the index is topological sorted.
1000
Duplicate entries may be written to the index for a single version id
1001
if this is done then the latter one completely replaces the former:
1002
this allows updates to correct version and parent information.
1003
Note that the two entries may share the delta, and that successive
1004
annotations and references MUST point to the first entry.
1006
The index file on disc contains a header, followed by one line per knit
1007
record. The same revision can be present in an index file more than once.
1008
The first occurrence gets assigned a sequence number starting from 0.
1010
The format of a single line is
1011
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1012
REVISION_ID is a utf8-encoded revision id
1013
FLAGS is a comma separated list of flags about the record. Values include
1014
no-eol, line-delta, fulltext.
1015
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1016
that the the compressed data starts at.
1017
LENGTH is the ascii representation of the length of the data file.
1018
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1020
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1021
revision id already in the knit that is a parent of REVISION_ID.
1022
The ' :' marker is the end of record marker.
1025
when a write is interrupted to the index file, it will result in a line that
1026
does not end in ' :'. If the ' :' is not present at the end of a line, or at
1027
the end of the file, then the record that is missing it will be ignored by
1030
When writing new records to the index file, the data is preceded by '\n'
1031
to ensure that records always start on new lines even if the last write was
1032
interrupted. As a result its normal for the last line in the index to be
1033
missing a trailing newline. One can be added with no harmful effects.
1036
HEADER = "# bzr knit index 8\n"
1038
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1039
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1041
def _cache_version(self, version_id, options, pos, size, parents):
1042
"""Cache a version record in the history array and index cache.
1044
This is inlined into __init__ for performance. KEEP IN SYNC.
1045
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1048
# only want the _history index to reference the 1st index entry
1050
if version_id not in self._cache:
1051
index = len(self._history)
1052
self._history.append(version_id)
1054
index = self._cache[version_id][5]
1055
self._cache[version_id] = (version_id,
1062
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1063
create_parent_dir=False, delay_create=False, dir_mode=None):
1064
_KnitComponentFile.__init__(self, transport, filename, mode,
1065
file_mode=file_mode,
1066
create_parent_dir=create_parent_dir,
1069
# position in _history is the 'official' index for a revision
1070
# but the values may have come from a newer entry.
1071
# so - wc -l of a knit index is != the number of unique names
1074
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1079
pb.update('read knit index', count, total)
1080
fp = self._transport.get(self._filename)
1082
self.check_header(fp)
1083
# readlines reads the whole file at once:
1084
# bad for transports like http, good for local disk
1085
# we save 60 ms doing this one change (
1086
# from calling readline each time to calling
1088
# probably what we want for nice behaviour on
1089
# http is a incremental readlines that yields, or
1090
# a check for local vs non local indexes,
1091
for l in fp.readlines():
1093
if len(rec) < 5 or rec[-1] != ':':
1095
# FIXME: in the future we should determine if its a
1096
# short write - and ignore it
1097
# or a different failure, and raise. RBC 20060407
1101
#pb.update('read knit index', count, total)
1102
# See self._parse_parents
1104
for value in rec[4:-1]:
1106
# uncompressed reference
1107
parents.append(value[1:])
1109
# this is 15/4000ms faster than isinstance,
1111
# this function is called thousands of times a
1112
# second so small variations add up.
1113
assert value.__class__ is str
1114
parents.append(self._history[int(value)])
1115
# end self._parse_parents
1116
# self._cache_version(rec[0],
1117
# rec[1].split(','),
1121
# --- self._cache_version
1122
# only want the _history index to reference the 1st
1123
# index entry for version_id
1125
if version_id not in self._cache:
1126
index = len(self._history)
1127
self._history.append(version_id)
1129
index = self._cache[version_id][5]
1130
self._cache[version_id] = (version_id,
1136
# --- self._cache_version
1139
except NoSuchFile, e:
1140
if mode != 'w' or not create:
1143
self._need_to_create = True
1145
self._transport.put_bytes_non_atomic(self._filename,
1146
self.HEADER, mode=self._file_mode)
1149
pb.update('read knit index', total, total)
1152
def _parse_parents(self, compressed_parents):
1153
"""convert a list of string parent values into version ids.
1155
ints are looked up in the index.
1156
.FOO values are ghosts and converted in to FOO.
1158
NOTE: the function is retained here for clarity, and for possible
1159
use in partial index reads. However bulk processing now has
1160
it inlined in __init__ for inner-loop optimisation.
1163
for value in compressed_parents:
1164
if value[-1] == '.':
1165
# uncompressed reference
1166
result.append(value[1:])
1168
# this is 15/4000ms faster than isinstance,
1169
# this function is called thousands of times a
1170
# second so small variations add up.
1171
assert value.__class__ is str
1172
result.append(self._history[int(value)])
1175
def get_graph(self):
1177
for version_id, index in self._cache.iteritems():
1178
graph.append((version_id, index[4]))
1181
def get_ancestry(self, versions):
1182
"""See VersionedFile.get_ancestry."""
1183
# get a graph of all the mentioned versions:
1185
pending = set(versions)
1187
version = pending.pop()
1188
parents = self._cache[version][4]
1189
# got the parents ok
1191
parents = [parent for parent in parents if parent in self._cache]
1192
for parent in parents:
1193
# if not completed and not a ghost
1194
if parent not in graph:
1196
graph[version] = parents
1197
return topo_sort(graph.items())
1199
def get_ancestry_with_ghosts(self, versions):
1200
"""See VersionedFile.get_ancestry_with_ghosts."""
1201
# get a graph of all the mentioned versions:
1203
pending = set(versions)
1205
version = pending.pop()
1207
parents = self._cache[version][4]
1213
# got the parents ok
1214
for parent in parents:
1215
if parent not in graph:
1217
graph[version] = parents
1218
return topo_sort(graph.items())
1220
def num_versions(self):
1221
return len(self._history)
1223
__len__ = num_versions
1225
def get_versions(self):
1226
return self._history
1228
def idx_to_name(self, idx):
1229
return self._history[idx]
1231
def lookup(self, version_id):
1232
assert version_id in self._cache
1233
return self._cache[version_id][5]
1235
def _version_list_to_index(self, versions):
1236
encode_utf8 = cache_utf8.encode
1238
for version in versions:
1239
if version in self._cache:
1240
# -- inlined lookup() --
1241
result_list.append(str(self._cache[version][5]))
1242
# -- end lookup () --
1244
result_list.append('.' + encode_utf8(version))
1245
return ' '.join(result_list)
1247
def add_version(self, version_id, options, pos, size, parents):
1248
"""Add a version record to the index."""
1249
self.add_versions(((version_id, options, pos, size, parents),))
1251
def add_versions(self, versions):
1252
"""Add multiple versions to the index.
1254
:param versions: a list of tuples:
1255
(version_id, options, pos, size, parents).
1258
encode_utf8 = cache_utf8.encode
1259
for version_id, options, pos, size, parents in versions:
1260
line = "\n%s %s %s %s %s :" % (encode_utf8(version_id),
1264
self._version_list_to_index(parents))
1265
assert isinstance(line, str), \
1266
'content must be utf-8 encoded: %r' % (line,)
1268
if not self._need_to_create:
1269
self._transport.append_bytes(self._filename, ''.join(lines))
1272
sio.write(self.HEADER)
1273
sio.writelines(lines)
1275
self._transport.put_file_non_atomic(self._filename, sio,
1276
create_parent_dir=self._create_parent_dir,
1277
mode=self._file_mode,
1278
dir_mode=self._dir_mode)
1279
self._need_to_create = False
1281
# cache after writing, so that a failed write leads to missing cache
1282
# entries not extra ones. XXX TODO: RBC 20060502 in the event of a
1283
# failure, reload the index or flush it or some such, to prevent
1284
# writing records that did complete twice.
1285
for version_id, options, pos, size, parents in versions:
1286
self._cache_version(version_id, options, pos, size, parents)
1288
def has_version(self, version_id):
1289
"""True if the version is in the index."""
1290
return (version_id in self._cache)
1292
def get_position(self, version_id):
1293
"""Return data position and size of specified version."""
1294
return (self._cache[version_id][2], \
1295
self._cache[version_id][3])
1297
def get_method(self, version_id):
1298
"""Return compression method of specified version."""
1299
options = self._cache[version_id][1]
1300
if 'fulltext' in options:
1303
assert 'line-delta' in options
1306
def get_options(self, version_id):
1307
return self._cache[version_id][1]
1309
def get_parents(self, version_id):
1310
"""Return parents of specified version ignoring ghosts."""
1311
return [parent for parent in self._cache[version_id][4]
1312
if parent in self._cache]
1314
def get_parents_with_ghosts(self, version_id):
1315
"""Return parents of specified version with ghosts."""
1316
return self._cache[version_id][4]
1318
def check_versions_present(self, version_ids):
1319
"""Check that all specified versions are present."""
1320
version_ids = set(version_ids)
1321
for version_id in list(version_ids):
1322
if version_id in self._cache:
1323
version_ids.remove(version_id)
1325
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1328
class _KnitData(_KnitComponentFile):
1329
"""Contents of the knit data file"""
1331
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1332
create_parent_dir=False, delay_create=False,
1334
_KnitComponentFile.__init__(self, transport, filename, mode,
1335
file_mode=file_mode,
1336
create_parent_dir=create_parent_dir,
1338
self._checked = False
1339
# TODO: jam 20060713 conceptually, this could spill to disk
1340
# if the cached size gets larger than a certain amount
1341
# but it complicates the model a bit, so for now just use
1342
# a simple dictionary
1344
self._do_cache = False
1347
self._need_to_create = create
1349
self._transport.put_bytes_non_atomic(self._filename, '',
1350
mode=self._file_mode)
1352
def enable_cache(self):
1353
"""Enable caching of reads."""
1354
self._do_cache = True
1356
def clear_cache(self):
1357
"""Clear the record cache."""
1358
self._do_cache = False
1361
def _open_file(self):
1363
return self._transport.get(self._filename)
1368
def _record_to_data(self, version_id, digest, lines):
1369
"""Convert version_id, digest, lines into a raw data block.
1371
:return: (len, a StringIO instance with the raw data ready to read.)
1374
data_file = GzipFile(None, mode='wb', fileobj=sio)
1376
version_id_utf8 = cache_utf8.encode(version_id)
1377
data_file.writelines(chain(
1378
["version %s %d %s\n" % (version_id_utf8,
1382
["end %s\n" % version_id_utf8]))
1389
def add_raw_record(self, raw_data):
1390
"""Append a prepared record to the data file.
1392
:return: the offset in the data file raw_data was written.
1394
assert isinstance(raw_data, str), 'data must be plain bytes'
1395
if not self._need_to_create:
1396
return self._transport.append_bytes(self._filename, raw_data)
1398
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1399
create_parent_dir=self._create_parent_dir,
1400
mode=self._file_mode,
1401
dir_mode=self._dir_mode)
1402
self._need_to_create = False
1405
def add_record(self, version_id, digest, lines):
1406
"""Write new text record to disk. Returns the position in the
1407
file where it was written."""
1408
size, sio = self._record_to_data(version_id, digest, lines)
1410
if not self._need_to_create:
1411
start_pos = self._transport.append_file(self._filename, sio)
1413
self._transport.put_file_non_atomic(self._filename, sio,
1414
create_parent_dir=self._create_parent_dir,
1415
mode=self._file_mode,
1416
dir_mode=self._dir_mode)
1417
self._need_to_create = False
1420
self._cache[version_id] = sio.getvalue()
1421
return start_pos, size
1423
def _parse_record_header(self, version_id, raw_data):
1424
"""Parse a record header for consistency.
1426
:return: the header and the decompressor stream.
1427
as (stream, header_record)
1429
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1430
rec = df.readline().split()
1432
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
1433
if cache_utf8.decode(rec[1]) != version_id:
1434
raise KnitCorrupt(self._filename,
1435
'unexpected version, wanted %r, got %r' % (
1436
version_id, rec[1]))
1439
def _parse_record(self, version_id, data):
1441
# 4168 calls in 2880 217 internal
1442
# 4168 calls to _parse_record_header in 2121
1443
# 4168 calls to readlines in 330
1444
df, rec = self._parse_record_header(version_id, data)
1445
record_contents = df.readlines()
1446
l = record_contents.pop()
1447
assert len(record_contents) == int(rec[2])
1448
if l != 'end %s\n' % cache_utf8.encode(version_id):
1449
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
1452
return record_contents, rec[3]
1454
def read_records_iter_raw(self, records):
1455
"""Read text records from data file and yield raw data.
1457
This unpacks enough of the text record to validate the id is
1458
as expected but thats all.
1460
# setup an iterator of the external records:
1461
# uses readv so nice and fast we hope.
1463
# grab the disk data needed.
1465
# Don't check _cache if it is empty
1466
needed_offsets = [(pos, size) for version_id, pos, size
1468
if version_id not in self._cache]
1470
needed_offsets = [(pos, size) for version_id, pos, size
1473
raw_records = self._transport.readv(self._filename, needed_offsets)
1476
for version_id, pos, size in records:
1477
if version_id in self._cache:
1478
# This data has already been validated
1479
data = self._cache[version_id]
1481
pos, data = raw_records.next()
1483
self._cache[version_id] = data
1485
# validate the header
1486
df, rec = self._parse_record_header(version_id, data)
1488
yield version_id, data
1490
def read_records_iter(self, records):
1491
"""Read text records from data file and yield result.
1493
The result will be returned in whatever is the fastest to read.
1494
Not by the order requested. Also, multiple requests for the same
1495
record will only yield 1 response.
1496
:param records: A list of (version_id, pos, len) entries
1497
:return: Yields (version_id, contents, digest) in the order
1498
read, not the order requested
1504
# Skip records we have alread seen
1505
yielded_records = set()
1506
needed_records = set()
1507
for record in records:
1508
if record[0] in self._cache:
1509
if record[0] in yielded_records:
1511
yielded_records.add(record[0])
1512
data = self._cache[record[0]]
1513
content, digest = self._parse_record(record[0], data)
1514
yield (record[0], content, digest)
1516
needed_records.add(record)
1517
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1519
needed_records = sorted(set(records), key=operator.itemgetter(1))
1521
if not needed_records:
1524
# The transport optimizes the fetching as well
1525
# (ie, reads continuous ranges.)
1526
readv_response = self._transport.readv(self._filename,
1527
[(pos, size) for version_id, pos, size in needed_records])
1529
for (version_id, pos, size), (pos, data) in \
1530
izip(iter(needed_records), readv_response):
1531
content, digest = self._parse_record(version_id, data)
1533
self._cache[version_id] = data
1534
yield version_id, content, digest
1536
def read_records(self, records):
1537
"""Read records into a dictionary."""
1539
for record_id, content, digest in \
1540
self.read_records_iter(records):
1541
components[record_id] = (content, digest)
1545
class InterKnit(InterVersionedFile):
1546
"""Optimised code paths for knit to knit operations."""
1548
_matching_file_from_factory = KnitVersionedFile
1549
_matching_file_to_factory = KnitVersionedFile
1552
def is_compatible(source, target):
1553
"""Be compatible with knits. """
1555
return (isinstance(source, KnitVersionedFile) and
1556
isinstance(target, KnitVersionedFile))
1557
except AttributeError:
1560
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1561
"""See InterVersionedFile.join."""
1562
assert isinstance(self.source, KnitVersionedFile)
1563
assert isinstance(self.target, KnitVersionedFile)
1565
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1570
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1572
version_ids = list(version_ids)
1573
if None in version_ids:
1574
version_ids.remove(None)
1576
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1577
this_versions = set(self.target._index.get_versions())
1578
needed_versions = self.source_ancestry - this_versions
1579
cross_check_versions = self.source_ancestry.intersection(this_versions)
1580
mismatched_versions = set()
1581
for version in cross_check_versions:
1582
# scan to include needed parents.
1583
n1 = set(self.target.get_parents_with_ghosts(version))
1584
n2 = set(self.source.get_parents_with_ghosts(version))
1586
# FIXME TEST this check for cycles being introduced works
1587
# the logic is we have a cycle if in our graph we are an
1588
# ancestor of any of the n2 revisions.
1594
parent_ancestors = self.source.get_ancestry(parent)
1595
if version in parent_ancestors:
1596
raise errors.GraphCycleError([parent, version])
1597
# ensure this parent will be available later.
1598
new_parents = n2.difference(n1)
1599
needed_versions.update(new_parents.difference(this_versions))
1600
mismatched_versions.add(version)
1602
if not needed_versions and not mismatched_versions:
1604
full_list = topo_sort(self.source.get_graph())
1606
version_list = [i for i in full_list if (not self.target.has_version(i)
1607
and i in needed_versions)]
1611
copy_queue_records = []
1613
for version_id in version_list:
1614
options = self.source._index.get_options(version_id)
1615
parents = self.source._index.get_parents_with_ghosts(version_id)
1616
# check that its will be a consistent copy:
1617
for parent in parents:
1618
# if source has the parent, we must :
1619
# * already have it or
1620
# * have it scheduled already
1621
# otherwise we don't care
1622
assert (self.target.has_version(parent) or
1623
parent in copy_set or
1624
not self.source.has_version(parent))
1625
data_pos, data_size = self.source._index.get_position(version_id)
1626
copy_queue_records.append((version_id, data_pos, data_size))
1627
copy_queue.append((version_id, options, parents))
1628
copy_set.add(version_id)
1630
# data suck the join:
1632
total = len(version_list)
1635
for (version_id, raw_data), \
1636
(version_id2, options, parents) in \
1637
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1639
assert version_id == version_id2, 'logic error, inconsistent results'
1641
pb.update("Joining knit", count, total)
1642
raw_records.append((version_id, options, parents, len(raw_data)))
1643
raw_datum.append(raw_data)
1644
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1646
for version in mismatched_versions:
1647
# FIXME RBC 20060309 is this needed?
1648
n1 = set(self.target.get_parents_with_ghosts(version))
1649
n2 = set(self.source.get_parents_with_ghosts(version))
1650
# write a combined record to our history preserving the current
1651
# parents as first in the list
1652
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1653
self.target.fix_parents(version, new_parents)
1659
InterVersionedFile.register_optimiser(InterKnit)
1662
class WeaveToKnit(InterVersionedFile):
1663
"""Optimised code paths for weave to knit operations."""
1665
_matching_file_from_factory = bzrlib.weave.WeaveFile
1666
_matching_file_to_factory = KnitVersionedFile
1669
def is_compatible(source, target):
1670
"""Be compatible with weaves to knits."""
1672
return (isinstance(source, bzrlib.weave.Weave) and
1673
isinstance(target, KnitVersionedFile))
1674
except AttributeError:
1677
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1678
"""See InterVersionedFile.join."""
1679
assert isinstance(self.source, bzrlib.weave.Weave)
1680
assert isinstance(self.target, KnitVersionedFile)
1682
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1687
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1689
version_ids = list(version_ids)
1691
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1692
this_versions = set(self.target._index.get_versions())
1693
needed_versions = self.source_ancestry - this_versions
1694
cross_check_versions = self.source_ancestry.intersection(this_versions)
1695
mismatched_versions = set()
1696
for version in cross_check_versions:
1697
# scan to include needed parents.
1698
n1 = set(self.target.get_parents_with_ghosts(version))
1699
n2 = set(self.source.get_parents(version))
1700
# if all of n2's parents are in n1, then its fine.
1701
if n2.difference(n1):
1702
# FIXME TEST this check for cycles being introduced works
1703
# the logic is we have a cycle if in our graph we are an
1704
# ancestor of any of the n2 revisions.
1710
parent_ancestors = self.source.get_ancestry(parent)
1711
if version in parent_ancestors:
1712
raise errors.GraphCycleError([parent, version])
1713
# ensure this parent will be available later.
1714
new_parents = n2.difference(n1)
1715
needed_versions.update(new_parents.difference(this_versions))
1716
mismatched_versions.add(version)
1718
if not needed_versions and not mismatched_versions:
1720
full_list = topo_sort(self.source.get_graph())
1722
version_list = [i for i in full_list if (not self.target.has_version(i)
1723
and i in needed_versions)]
1727
total = len(version_list)
1728
for version_id in version_list:
1729
pb.update("Converting to knit", count, total)
1730
parents = self.source.get_parents(version_id)
1731
# check that its will be a consistent copy:
1732
for parent in parents:
1733
# if source has the parent, we must already have it
1734
assert (self.target.has_version(parent))
1735
self.target.add_lines(
1736
version_id, parents, self.source.get_lines(version_id))
1739
for version in mismatched_versions:
1740
# FIXME RBC 20060309 is this needed?
1741
n1 = set(self.target.get_parents_with_ghosts(version))
1742
n2 = set(self.source.get_parents(version))
1743
# write a combined record to our history preserving the current
1744
# parents as first in the list
1745
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1746
self.target.fix_parents(version, new_parents)
1752
InterVersionedFile.register_optimiser(WeaveToKnit)
1755
class KnitSequenceMatcher(difflib.SequenceMatcher):
1756
"""Knit tuned sequence matcher.
1758
This is based on profiling of difflib which indicated some improvements
1759
for our usage pattern.
1762
def find_longest_match(self, alo, ahi, blo, bhi):
1763
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1765
If isjunk is not defined:
1767
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1768
alo <= i <= i+k <= ahi
1769
blo <= j <= j+k <= bhi
1770
and for all (i',j',k') meeting those conditions,
1773
and if i == i', j <= j'
1775
In other words, of all maximal matching blocks, return one that
1776
starts earliest in a, and of all those maximal matching blocks that
1777
start earliest in a, return the one that starts earliest in b.
1779
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1780
>>> s.find_longest_match(0, 5, 0, 9)
1783
If isjunk is defined, first the longest matching block is
1784
determined as above, but with the additional restriction that no
1785
junk element appears in the block. Then that block is extended as
1786
far as possible by matching (only) junk elements on both sides. So
1787
the resulting block never matches on junk except as identical junk
1788
happens to be adjacent to an "interesting" match.
1790
Here's the same example as before, but considering blanks to be
1791
junk. That prevents " abcd" from matching the " abcd" at the tail
1792
end of the second sequence directly. Instead only the "abcd" can
1793
match, and matches the leftmost "abcd" in the second sequence:
1795
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1796
>>> s.find_longest_match(0, 5, 0, 9)
1799
If no blocks match, return (alo, blo, 0).
1801
>>> s = SequenceMatcher(None, "ab", "c")
1802
>>> s.find_longest_match(0, 2, 0, 1)
1806
# CAUTION: stripping common prefix or suffix would be incorrect.
1810
# Longest matching block is "ab", but if common prefix is
1811
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1812
# strip, so ends up claiming that ab is changed to acab by
1813
# inserting "ca" in the middle. That's minimal but unintuitive:
1814
# "it's obvious" that someone inserted "ac" at the front.
1815
# Windiff ends up at the same place as diff, but by pairing up
1816
# the unique 'b's and then matching the first two 'a's.
1818
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1819
besti, bestj, bestsize = alo, blo, 0
1820
# find longest junk-free match
1821
# during an iteration of the loop, j2len[j] = length of longest
1822
# junk-free match ending with a[i-1] and b[j]
1826
for i in xrange(alo, ahi):
1827
# look at all instances of a[i] in b; note that because
1828
# b2j has no junk keys, the loop is skipped if a[i] is junk
1829
j2lenget = j2len.get
1832
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1833
# following improvement
1834
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1835
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1836
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1838
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1839
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1840
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1852
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1854
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1857
# Extend the best by non-junk elements on each end. In particular,
1858
# "popular" non-junk elements aren't in b2j, which greatly speeds
1859
# the inner loop above, but also means "the best" match so far
1860
# doesn't contain any junk *or* popular non-junk elements.
1861
while besti > alo and bestj > blo and \
1862
not isbjunk(b[bestj-1]) and \
1863
a[besti-1] == b[bestj-1]:
1864
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1865
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1866
not isbjunk(b[bestj+bestsize]) and \
1867
a[besti+bestsize] == b[bestj+bestsize]:
1870
# Now that we have a wholly interesting match (albeit possibly
1871
# empty!), we may as well suck up the matching junk on each
1872
# side of it too. Can't think of a good reason not to, and it
1873
# saves post-processing the (possibly considerable) expense of
1874
# figuring out what to do with it. In the case of an empty
1875
# interesting match, this is clearly the right thing to do,
1876
# because no other kind of match is possible in the regions.
1877
while besti > alo and bestj > blo and \
1878
isbjunk(b[bestj-1]) and \
1879
a[besti-1] == b[bestj-1]:
1880
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1881
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1882
isbjunk(b[bestj+bestsize]) and \
1883
a[besti+bestsize] == b[bestj+bestsize]:
1884
bestsize = bestsize + 1
1886
return besti, bestj, bestsize
added
removed
bzrlib/knit.py
