1
# Copyright (C) 2005, 2006 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
83
from bzrlib.errors import FileExists, NoSuchFile, KnitError, \
84
InvalidRevisionId, KnitCorrupt, KnitHeaderError, \
85
RevisionNotPresent, RevisionAlreadyPresent
86
from bzrlib.tuned_gzip import GzipFile
87
from bzrlib.trace import mutter
88
from bzrlib.osutils import contains_whitespace, contains_linebreaks, \
90
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
91
from bzrlib.tsort import topo_sort
94
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
97
# TODO: Split out code specific to this format into an associated object.
99
# TODO: Can we put in some kind of value to check that the index and data
100
# files belong together?
102
# TODO: accommodate binaries, perhaps by storing a byte count
104
# TODO: function to check whole file
106
# TODO: atomically append data, then measure backwards from the cursor
107
# position after writing to work out where it was located. we may need to
108
# bypass python file buffering.
110
DATA_SUFFIX = '.knit'
111
INDEX_SUFFIX = '.kndx'
114
class KnitContent(object):
115
"""Content of a knit version to which deltas can be applied."""
117
def __init__(self, lines):
120
def annotate_iter(self):
121
"""Yield tuples of (origin, text) for each content line."""
122
return iter(self._lines)
125
"""Return a list of (origin, text) tuples."""
126
return list(self.annotate_iter())
128
def line_delta_iter(self, new_lines):
129
"""Generate line-based delta from this content to new_lines."""
130
new_texts = new_lines.text()
131
old_texts = self.text()
132
s = KnitSequenceMatcher(None, old_texts, new_texts)
133
for tag, i1, i2, j1, j2 in s.get_opcodes():
136
# ofrom, oto, length, data
137
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
139
def line_delta(self, new_lines):
140
return list(self.line_delta_iter(new_lines))
143
return [text for origin, text in self._lines]
146
return KnitContent(self._lines[:])
149
class _KnitFactory(object):
150
"""Base factory for creating content objects."""
152
def make(self, lines, version):
153
num_lines = len(lines)
154
return KnitContent(zip([version] * num_lines, lines))
157
class KnitAnnotateFactory(_KnitFactory):
158
"""Factory for creating annotated Content objects."""
162
def parse_fulltext(self, content, version):
163
"""Convert fulltext to internal representation
165
fulltext content is of the format
166
revid(utf8) plaintext\n
167
internal representation is of the format:
170
decode_utf8 = cache_utf8.decode
173
origin, text = line.split(' ', 1)
174
lines.append((decode_utf8(origin), text))
175
return KnitContent(lines)
177
def parse_line_delta_iter(self, lines):
178
return iter(self.parse_line_delta(lines))
180
def parse_line_delta(self, lines, version):
181
"""Convert a line based delta into internal representation.
183
line delta is in the form of:
184
intstart intend intcount
186
revid(utf8) newline\n
187
internal representation is
188
(start, end, count, [1..count tuples (revid, newline)])
190
decode_utf8 = cache_utf8.decode
194
# walk through the lines parsing.
196
start, end, count = [int(n) for n in header.split(',')]
200
origin, text = next().split(' ', 1)
202
contents.append((decode_utf8(origin), text))
203
result.append((start, end, count, contents))
206
def lower_fulltext(self, content):
207
"""convert a fulltext content record into a serializable form.
209
see parse_fulltext which this inverts.
211
encode_utf8 = cache_utf8.encode
212
return ['%s %s' % (encode_utf8(o), t) for o, t in content._lines]
214
def lower_line_delta(self, delta):
215
"""convert a delta into a serializable form.
217
See parse_line_delta which this inverts.
219
encode_utf8 = cache_utf8.encode
221
for start, end, c, lines in delta:
222
out.append('%d,%d,%d\n' % (start, end, c))
223
out.extend(encode_utf8(origin) + ' ' + text
224
for origin, text in lines)
228
class KnitPlainFactory(_KnitFactory):
229
"""Factory for creating plain Content objects."""
233
def parse_fulltext(self, content, version):
234
"""This parses an unannotated fulltext.
236
Note that this is not a noop - the internal representation
237
has (versionid, line) - its just a constant versionid.
239
return self.make(content, version)
241
def parse_line_delta_iter(self, lines, version):
243
num_lines = len(lines)
244
while cur < num_lines:
247
start, end, c = [int(n) for n in header.split(',')]
248
yield start, end, c, zip([version] * c, lines[cur:cur+c])
251
def parse_line_delta(self, lines, version):
252
return list(self.parse_line_delta_iter(lines, version))
254
def lower_fulltext(self, content):
255
return content.text()
257
def lower_line_delta(self, delta):
259
for start, end, c, lines in delta:
260
out.append('%d,%d,%d\n' % (start, end, c))
261
out.extend([text for origin, text in lines])
265
def make_empty_knit(transport, relpath):
266
"""Construct a empty knit at the specified location."""
267
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
271
class KnitVersionedFile(VersionedFile):
272
"""Weave-like structure with faster random access.
274
A knit stores a number of texts and a summary of the relationships
275
between them. Texts are identified by a string version-id. Texts
276
are normally stored and retrieved as a series of lines, but can
277
also be passed as single strings.
279
Lines are stored with the trailing newline (if any) included, to
280
avoid special cases for files with no final newline. Lines are
281
composed of 8-bit characters, not unicode. The combination of
282
these approaches should mean any 'binary' file can be safely
283
stored and retrieved.
286
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
287
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
288
create=False, create_parent_dir=False, delay_create=False,
290
"""Construct a knit at location specified by relpath.
292
:param create: If not True, only open an existing knit.
293
:param create_parent_dir: If True, create the parent directory if
294
creating the file fails. (This is used for stores with
295
hash-prefixes that may not exist yet)
296
:param delay_create: The calling code is aware that the knit won't
297
actually be created until the first data is stored.
299
if deprecated_passed(basis_knit):
300
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
301
" deprecated as of bzr 0.9.",
302
DeprecationWarning, stacklevel=2)
303
if access_mode is None:
305
super(KnitVersionedFile, self).__init__(access_mode)
306
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
307
self.transport = transport
308
self.filename = relpath
309
self.factory = factory or KnitAnnotateFactory()
310
self.writable = (access_mode == 'w')
313
self._max_delta_chain = 200
315
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
316
access_mode, create=create, file_mode=file_mode,
317
create_parent_dir=create_parent_dir, delay_create=delay_create,
319
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
320
access_mode, create=create and not len(self), file_mode=file_mode,
321
create_parent_dir=create_parent_dir, delay_create=delay_create,
325
return '%s(%s)' % (self.__class__.__name__,
326
self.transport.abspath(self.filename))
328
def _check_should_delta(self, first_parents):
329
"""Iterate back through the parent listing, looking for a fulltext.
331
This is used when we want to decide whether to add a delta or a new
332
fulltext. It searches for _max_delta_chain parents. When it finds a
333
fulltext parent, it sees if the total size of the deltas leading up to
334
it is large enough to indicate that we want a new full text anyway.
336
Return True if we should create a new delta, False if we should use a
341
delta_parents = first_parents
342
for count in xrange(self._max_delta_chain):
343
parent = delta_parents[0]
344
method = self._index.get_method(parent)
345
pos, size = self._index.get_position(parent)
346
if method == 'fulltext':
350
delta_parents = self._index.get_parents(parent)
352
# We couldn't find a fulltext, so we must create a new one
355
return fulltext_size > delta_size
357
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
358
"""See VersionedFile._add_delta()."""
359
self._check_add(version_id, []) # should we check the lines ?
360
self._check_versions_present(parents)
364
for parent in parents:
365
if not self.has_version(parent):
366
ghosts.append(parent)
368
present_parents.append(parent)
370
if delta_parent is None:
371
# reconstitute as full text.
372
assert len(delta) == 1 or len(delta) == 0
374
assert delta[0][0] == 0
375
assert delta[0][1] == 0, delta[0][1]
376
return super(KnitVersionedFile, self)._add_delta(version_id,
387
options.append('no-eol')
389
if delta_parent is not None:
390
# determine the current delta chain length.
391
# To speed the extract of texts the delta chain is limited
392
# to a fixed number of deltas. This should minimize both
393
# I/O and the time spend applying deltas.
394
# The window was changed to a maximum of 200 deltas, but also added
395
# was a check that the total compressed size of the deltas is
396
# smaller than the compressed size of the fulltext.
397
if not self._check_should_delta([delta_parent]):
398
# We don't want a delta here, just do a normal insertion.
399
return super(KnitVersionedFile, self)._add_delta(version_id,
406
options.append('line-delta')
407
store_lines = self.factory.lower_line_delta(delta)
409
where, size = self._data.add_record(version_id, digest, store_lines)
410
self._index.add_version(version_id, options, where, size, parents)
412
def _add_raw_records(self, records, data):
413
"""Add all the records 'records' with data pre-joined in 'data'.
415
:param records: A list of tuples(version_id, options, parents, size).
416
:param data: The data for the records. When it is written, the records
417
are adjusted to have pos pointing into data by the sum of
418
the preceding records sizes.
421
pos = self._data.add_raw_record(data)
424
for (version_id, options, parents, size) in records:
425
index_entries.append((version_id, options, pos+offset,
427
if self._data._do_cache:
428
self._data._cache[version_id] = data[offset:offset+size]
430
self._index.add_versions(index_entries)
432
def enable_cache(self):
433
"""Start caching data for this knit"""
434
self._data.enable_cache()
436
def clear_cache(self):
437
"""Clear the data cache only."""
438
self._data.clear_cache()
440
def copy_to(self, name, transport):
441
"""See VersionedFile.copy_to()."""
442
# copy the current index to a temp index to avoid racing with local
444
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
445
self.transport.get(self._index._filename))
447
f = self._data._open_file()
449
transport.put_file(name + DATA_SUFFIX, f)
452
# move the copied index into place
453
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
455
def create_empty(self, name, transport, mode=None):
456
return KnitVersionedFile(name, transport, factory=self.factory,
457
delta=self.delta, create=True)
459
def _fix_parents(self, version, new_parents):
460
"""Fix the parents list for version.
462
This is done by appending a new version to the index
463
with identical data except for the parents list.
464
the parents list must be a superset of the current
467
current_values = self._index._cache[version]
468
assert set(current_values[4]).difference(set(new_parents)) == set()
469
self._index.add_version(version,
475
def get_delta(self, version_id):
476
"""Get a delta for constructing version from some other version."""
477
if not self.has_version(version_id):
478
raise RevisionNotPresent(version_id, self.filename)
480
parents = self.get_parents(version_id)
485
data_pos, data_size = self._index.get_position(version_id)
486
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
487
version_idx = self._index.lookup(version_id)
488
noeol = 'no-eol' in self._index.get_options(version_id)
489
if 'fulltext' == self._index.get_method(version_id):
490
new_content = self.factory.parse_fulltext(data, version_idx)
491
if parent is not None:
492
reference_content = self._get_content(parent)
493
old_texts = reference_content.text()
496
new_texts = new_content.text()
497
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
498
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
500
delta = self.factory.parse_line_delta(data, version_idx)
501
return parent, sha1, noeol, delta
503
def get_graph_with_ghosts(self):
504
"""See VersionedFile.get_graph_with_ghosts()."""
505
graph_items = self._index.get_graph()
506
return dict(graph_items)
508
def get_sha1(self, version_id):
509
"""See VersionedFile.get_sha1()."""
510
record_map = self._get_record_map([version_id])
511
method, content, digest, next = record_map[version_id]
516
"""See VersionedFile.get_suffixes()."""
517
return [DATA_SUFFIX, INDEX_SUFFIX]
519
def has_ghost(self, version_id):
520
"""True if there is a ghost reference in the file to version_id."""
522
if self.has_version(version_id):
524
# optimisable if needed by memoising the _ghosts set.
525
items = self._index.get_graph()
526
for node, parents in items:
527
for parent in parents:
528
if parent not in self._index._cache:
529
if parent == version_id:
534
"""See VersionedFile.versions."""
535
return self._index.get_versions()
537
def has_version(self, version_id):
538
"""See VersionedFile.has_version."""
539
return self._index.has_version(version_id)
541
__contains__ = has_version
543
def _merge_annotations(self, content, parents, parent_texts={},
544
delta=None, annotated=None):
545
"""Merge annotations for content. This is done by comparing
546
the annotations based on changed to the text.
550
for parent_id in parents:
551
merge_content = self._get_content(parent_id, parent_texts)
552
seq = patiencediff.PatienceSequenceMatcher(
553
None, merge_content.text(), content.text())
554
if delta_seq is None:
555
# setup a delta seq to reuse.
557
for i, j, n in seq.get_matching_blocks():
560
# this appears to copy (origin, text) pairs across to the new
561
# content for any line that matches the last-checked parent.
562
# FIXME: save the sequence control data for delta compression
563
# against the most relevant parent rather than rediffing.
564
content._lines[j:j+n] = merge_content._lines[i:i+n]
567
reference_content = self._get_content(parents[0], parent_texts)
568
new_texts = content.text()
569
old_texts = reference_content.text()
570
delta_seq = patiencediff.PatienceSequenceMatcher(
571
None, old_texts, new_texts)
572
return self._make_line_delta(delta_seq, content)
574
def _make_line_delta(self, delta_seq, new_content):
575
"""Generate a line delta from delta_seq and new_content."""
577
for op in delta_seq.get_opcodes():
580
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
583
def _get_components_positions(self, version_ids):
584
"""Produce a map of position data for the components of versions.
586
This data is intended to be used for retrieving the knit records.
588
A dict of version_id to (method, data_pos, data_size, next) is
590
method is the way referenced data should be applied.
591
data_pos is the position of the data in the knit.
592
data_size is the size of the data in the knit.
593
next is the build-parent of the version, or None for fulltexts.
596
for version_id in version_ids:
599
while cursor is not None and cursor not in component_data:
600
method = self._index.get_method(cursor)
601
if method == 'fulltext':
604
next = self.get_parents(cursor)[0]
605
data_pos, data_size = self._index.get_position(cursor)
606
component_data[cursor] = (method, data_pos, data_size, next)
608
return component_data
610
def _get_content(self, version_id, parent_texts={}):
611
"""Returns a content object that makes up the specified
613
if not self.has_version(version_id):
614
raise RevisionNotPresent(version_id, self.filename)
616
cached_version = parent_texts.get(version_id, None)
617
if cached_version is not None:
618
return cached_version
620
text_map, contents_map = self._get_content_maps([version_id])
621
return contents_map[version_id]
623
def _check_versions_present(self, version_ids):
624
"""Check that all specified versions are present."""
625
version_ids = set(version_ids)
626
for r in list(version_ids):
627
if self._index.has_version(r):
628
version_ids.remove(r)
630
raise RevisionNotPresent(list(version_ids)[0], self.filename)
632
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
633
"""See VersionedFile.add_lines_with_ghosts()."""
634
self._check_add(version_id, lines)
635
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
637
def _add_lines(self, version_id, parents, lines, parent_texts):
638
"""See VersionedFile.add_lines."""
639
self._check_add(version_id, lines)
640
self._check_versions_present(parents)
641
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
643
def _check_add(self, version_id, lines):
644
"""check that version_id and lines are safe to add."""
645
assert self.writable, "knit is not opened for write"
646
### FIXME escape. RBC 20060228
647
if contains_whitespace(version_id):
648
raise InvalidRevisionId(version_id, self.filename)
649
if self.has_version(version_id):
650
raise RevisionAlreadyPresent(version_id, self.filename)
651
self._check_lines_not_unicode(lines)
652
self._check_lines_are_lines(lines)
654
def _add(self, version_id, lines, parents, delta, parent_texts):
655
"""Add a set of lines on top of version specified by parents.
657
If delta is true, compress the text as a line-delta against
660
Any versions not present will be converted into ghosts.
662
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
663
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
664
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
665
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
666
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
667
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
668
# +1383 0 8.0370 8.0370 +<len>
669
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
670
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
671
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
672
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
676
if parent_texts is None:
678
for parent in parents:
679
if not self.has_version(parent):
680
ghosts.append(parent)
682
present_parents.append(parent)
684
if delta and not len(present_parents):
687
digest = sha_strings(lines)
690
if lines[-1][-1] != '\n':
691
options.append('no-eol')
692
lines[-1] = lines[-1] + '\n'
694
if len(present_parents) and delta:
695
# To speed the extract of texts the delta chain is limited
696
# to a fixed number of deltas. This should minimize both
697
# I/O and the time spend applying deltas.
698
delta = self._check_should_delta(present_parents)
700
lines = self.factory.make(lines, version_id)
701
if delta or (self.factory.annotated and len(present_parents) > 0):
702
# Merge annotations from parent texts if so is needed.
703
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
704
delta, self.factory.annotated)
707
options.append('line-delta')
708
store_lines = self.factory.lower_line_delta(delta_hunks)
710
options.append('fulltext')
711
store_lines = self.factory.lower_fulltext(lines)
713
where, size = self._data.add_record(version_id, digest, store_lines)
714
self._index.add_version(version_id, options, where, size, parents)
717
def check(self, progress_bar=None):
718
"""See VersionedFile.check()."""
720
def _clone_text(self, new_version_id, old_version_id, parents):
721
"""See VersionedFile.clone_text()."""
722
# FIXME RBC 20060228 make fast by only inserting an index with null
724
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
726
def get_lines(self, version_id):
727
"""See VersionedFile.get_lines()."""
728
return self.get_line_list([version_id])[0]
730
def _get_record_map(self, version_ids):
731
"""Produce a dictionary of knit records.
733
The keys are version_ids, the values are tuples of (method, content,
735
method is the way the content should be applied.
736
content is a KnitContent object.
737
digest is the SHA1 digest of this version id after all steps are done
738
next is the build-parent of the version, i.e. the leftmost ancestor.
739
If the method is fulltext, next will be None.
741
position_map = self._get_components_positions(version_ids)
742
# c = component_id, m = method, p = position, s = size, n = next
743
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
745
for component_id, content, digest in \
746
self._data.read_records_iter(records):
747
method, position, size, next = position_map[component_id]
748
record_map[component_id] = method, content, digest, next
752
def get_text(self, version_id):
753
"""See VersionedFile.get_text"""
754
return self.get_texts([version_id])[0]
756
def get_texts(self, version_ids):
757
return [''.join(l) for l in self.get_line_list(version_ids)]
759
def get_line_list(self, version_ids):
760
"""Return the texts of listed versions as a list of strings."""
761
text_map, content_map = self._get_content_maps(version_ids)
762
return [text_map[v] for v in version_ids]
764
def _get_content_maps(self, version_ids):
765
"""Produce maps of text and KnitContents
767
:return: (text_map, content_map) where text_map contains the texts for
768
the requested versions and content_map contains the KnitContents.
769
Both dicts take version_ids as their keys.
771
for version_id in version_ids:
772
if not self.has_version(version_id):
773
raise RevisionNotPresent(version_id, self.filename)
774
record_map = self._get_record_map(version_ids)
779
for version_id in version_ids:
782
while cursor is not None:
783
method, data, digest, next = record_map[cursor]
784
components.append((cursor, method, data, digest))
785
if cursor in content_map:
790
for component_id, method, data, digest in reversed(components):
791
if component_id in content_map:
792
content = content_map[component_id]
794
version_idx = self._index.lookup(component_id)
795
if method == 'fulltext':
796
assert content is None
797
content = self.factory.parse_fulltext(data, version_idx)
798
elif method == 'line-delta':
799
delta = self.factory.parse_line_delta(data, version_idx)
800
content = content.copy()
801
content._lines = self._apply_delta(content._lines,
803
content_map[component_id] = content
805
if 'no-eol' in self._index.get_options(version_id):
806
content = content.copy()
807
line = content._lines[-1][1].rstrip('\n')
808
content._lines[-1] = (content._lines[-1][0], line)
809
final_content[version_id] = content
811
# digest here is the digest from the last applied component.
812
text = content.text()
813
if sha_strings(text) != digest:
814
raise KnitCorrupt(self.filename,
815
'sha-1 does not match %s' % version_id)
817
text_map[version_id] = text
818
return text_map, final_content
820
def iter_lines_added_or_present_in_versions(self, version_ids=None,
822
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
823
if version_ids is None:
824
version_ids = self.versions()
826
pb = progress.DummyProgress()
827
# we don't care about inclusions, the caller cares.
828
# but we need to setup a list of records to visit.
829
# we need version_id, position, length
830
version_id_records = []
831
requested_versions = set(version_ids)
832
# filter for available versions
833
for version_id in requested_versions:
834
if not self.has_version(version_id):
835
raise RevisionNotPresent(version_id, self.filename)
836
# get a in-component-order queue:
837
for version_id in self.versions():
838
if version_id in requested_versions:
839
data_pos, length = self._index.get_position(version_id)
840
version_id_records.append((version_id, data_pos, length))
842
total = len(version_id_records)
843
for version_idx, (version_id, data, sha_value) in \
844
enumerate(self._data.read_records_iter(version_id_records)):
845
pb.update('Walking content.', version_idx, total)
846
method = self._index.get_method(version_id)
847
version_idx = self._index.lookup(version_id)
848
assert method in ('fulltext', 'line-delta')
849
if method == 'fulltext':
850
content = self.factory.parse_fulltext(data, version_idx)
851
for line in content.text():
854
delta = self.factory.parse_line_delta(data, version_idx)
855
for start, end, count, lines in delta:
856
for origin, line in lines:
858
pb.update('Walking content.', total, total)
860
def num_versions(self):
861
"""See VersionedFile.num_versions()."""
862
return self._index.num_versions()
864
__len__ = num_versions
866
def annotate_iter(self, version_id):
867
"""See VersionedFile.annotate_iter."""
868
content = self._get_content(version_id)
869
for origin, text in content.annotate_iter():
872
def get_parents(self, version_id):
873
"""See VersionedFile.get_parents."""
876
# 52554 calls in 1264 872 internal down from 3674
878
return self._index.get_parents(version_id)
880
raise RevisionNotPresent(version_id, self.filename)
882
def get_parents_with_ghosts(self, version_id):
883
"""See VersionedFile.get_parents."""
885
return self._index.get_parents_with_ghosts(version_id)
887
raise RevisionNotPresent(version_id, self.filename)
889
def get_ancestry(self, versions):
890
"""See VersionedFile.get_ancestry."""
891
if isinstance(versions, basestring):
892
versions = [versions]
895
self._check_versions_present(versions)
896
return self._index.get_ancestry(versions)
898
def get_ancestry_with_ghosts(self, versions):
899
"""See VersionedFile.get_ancestry_with_ghosts."""
900
if isinstance(versions, basestring):
901
versions = [versions]
904
self._check_versions_present(versions)
905
return self._index.get_ancestry_with_ghosts(versions)
907
#@deprecated_method(zero_eight)
908
def walk(self, version_ids):
909
"""See VersionedFile.walk."""
910
# We take the short path here, and extract all relevant texts
911
# and put them in a weave and let that do all the work. Far
912
# from optimal, but is much simpler.
913
# FIXME RB 20060228 this really is inefficient!
914
from bzrlib.weave import Weave
916
w = Weave(self.filename)
917
ancestry = self.get_ancestry(version_ids)
918
sorted_graph = topo_sort(self._index.get_graph())
919
version_list = [vid for vid in sorted_graph if vid in ancestry]
921
for version_id in version_list:
922
lines = self.get_lines(version_id)
923
w.add_lines(version_id, self.get_parents(version_id), lines)
925
for lineno, insert_id, dset, line in w.walk(version_ids):
926
yield lineno, insert_id, dset, line
928
def plan_merge(self, ver_a, ver_b):
929
"""See VersionedFile.plan_merge."""
930
ancestors_b = set(self.get_ancestry(ver_b))
931
def status_a(revision, text):
932
if revision in ancestors_b:
933
return 'killed-b', text
937
ancestors_a = set(self.get_ancestry(ver_a))
938
def status_b(revision, text):
939
if revision in ancestors_a:
940
return 'killed-a', text
944
annotated_a = self.annotate(ver_a)
945
annotated_b = self.annotate(ver_b)
946
plain_a = [t for (a, t) in annotated_a]
947
plain_b = [t for (a, t) in annotated_b]
948
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
951
for ai, bi, l in blocks:
952
# process all mismatched sections
953
# (last mismatched section is handled because blocks always
954
# includes a 0-length last block)
955
for revision, text in annotated_a[a_cur:ai]:
956
yield status_a(revision, text)
957
for revision, text in annotated_b[b_cur:bi]:
958
yield status_b(revision, text)
960
# and now the matched section
963
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
964
assert text_a == text_b
965
yield "unchanged", text_a
968
class _KnitComponentFile(object):
969
"""One of the files used to implement a knit database"""
971
def __init__(self, transport, filename, mode, file_mode=None,
972
create_parent_dir=False, dir_mode=None):
973
self._transport = transport
974
self._filename = filename
976
self._file_mode = file_mode
977
self._dir_mode = dir_mode
978
self._create_parent_dir = create_parent_dir
979
self._need_to_create = False
981
def check_header(self, fp):
983
if line != self.HEADER:
984
raise KnitHeaderError(badline=line)
987
"""Commit is a nop."""
990
return '%s(%s)' % (self.__class__.__name__, self._filename)
993
class _KnitIndex(_KnitComponentFile):
994
"""Manages knit index file.
996
The index is already kept in memory and read on startup, to enable
997
fast lookups of revision information. The cursor of the index
998
file is always pointing to the end, making it easy to append
1001
_cache is a cache for fast mapping from version id to a Index
1004
_history is a cache for fast mapping from indexes to version ids.
1006
The index data format is dictionary compressed when it comes to
1007
parent references; a index entry may only have parents that with a
1008
lover index number. As a result, the index is topological sorted.
1010
Duplicate entries may be written to the index for a single version id
1011
if this is done then the latter one completely replaces the former:
1012
this allows updates to correct version and parent information.
1013
Note that the two entries may share the delta, and that successive
1014
annotations and references MUST point to the first entry.
1016
The index file on disc contains a header, followed by one line per knit
1017
record. The same revision can be present in an index file more than once.
1018
The first occurrence gets assigned a sequence number starting from 0.
1020
The format of a single line is
1021
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1022
REVISION_ID is a utf8-encoded revision id
1023
FLAGS is a comma separated list of flags about the record. Values include
1024
no-eol, line-delta, fulltext.
1025
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1026
that the the compressed data starts at.
1027
LENGTH is the ascii representation of the length of the data file.
1028
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1030
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1031
revision id already in the knit that is a parent of REVISION_ID.
1032
The ' :' marker is the end of record marker.
1035
when a write is interrupted to the index file, it will result in a line that
1036
does not end in ' :'. If the ' :' is not present at the end of a line, or at
1037
the end of the file, then the record that is missing it will be ignored by
1040
When writing new records to the index file, the data is preceded by '\n'
1041
to ensure that records always start on new lines even if the last write was
1042
interrupted. As a result its normal for the last line in the index to be
1043
missing a trailing newline. One can be added with no harmful effects.
1046
HEADER = "# bzr knit index 8\n"
1048
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1049
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1051
def _cache_version(self, version_id, options, pos, size, parents):
1052
"""Cache a version record in the history array and index cache.
1054
This is inlined into __init__ for performance. KEEP IN SYNC.
1055
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1058
# only want the _history index to reference the 1st index entry
1060
if version_id not in self._cache:
1061
index = len(self._history)
1062
self._history.append(version_id)
1064
index = self._cache[version_id][5]
1065
self._cache[version_id] = (version_id,
1072
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1073
create_parent_dir=False, delay_create=False, dir_mode=None):
1074
_KnitComponentFile.__init__(self, transport, filename, mode,
1075
file_mode=file_mode,
1076
create_parent_dir=create_parent_dir,
1079
# position in _history is the 'official' index for a revision
1080
# but the values may have come from a newer entry.
1081
# so - wc -l of a knit index is != the number of unique names
1084
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1089
pb.update('read knit index', count, total)
1090
fp = self._transport.get(self._filename)
1092
self.check_header(fp)
1093
# readlines reads the whole file at once:
1094
# bad for transports like http, good for local disk
1095
# we save 60 ms doing this one change (
1096
# from calling readline each time to calling
1098
# probably what we want for nice behaviour on
1099
# http is a incremental readlines that yields, or
1100
# a check for local vs non local indexes,
1101
for l in fp.readlines():
1103
if len(rec) < 5 or rec[-1] != ':':
1105
# FIXME: in the future we should determine if its a
1106
# short write - and ignore it
1107
# or a different failure, and raise. RBC 20060407
1111
#pb.update('read knit index', count, total)
1112
# See self._parse_parents
1114
for value in rec[4:-1]:
1116
# uncompressed reference
1117
parents.append(value[1:])
1119
# this is 15/4000ms faster than isinstance,
1121
# this function is called thousands of times a
1122
# second so small variations add up.
1123
assert value.__class__ is str
1124
parents.append(self._history[int(value)])
1125
# end self._parse_parents
1126
# self._cache_version(rec[0],
1127
# rec[1].split(','),
1131
# --- self._cache_version
1132
# only want the _history index to reference the 1st
1133
# index entry for version_id
1135
if version_id not in self._cache:
1136
index = len(self._history)
1137
self._history.append(version_id)
1139
index = self._cache[version_id][5]
1140
self._cache[version_id] = (version_id,
1146
# --- self._cache_version
1149
except NoSuchFile, e:
1150
if mode != 'w' or not create:
1153
self._need_to_create = True
1155
self._transport.put_bytes_non_atomic(self._filename,
1156
self.HEADER, mode=self._file_mode)
1159
pb.update('read knit index', total, total)
1162
def _parse_parents(self, compressed_parents):
1163
"""convert a list of string parent values into version ids.
1165
ints are looked up in the index.
1166
.FOO values are ghosts and converted in to FOO.
1168
NOTE: the function is retained here for clarity, and for possible
1169
use in partial index reads. However bulk processing now has
1170
it inlined in __init__ for inner-loop optimisation.
1173
for value in compressed_parents:
1174
if value[-1] == '.':
1175
# uncompressed reference
1176
result.append(value[1:])
1178
# this is 15/4000ms faster than isinstance,
1179
# this function is called thousands of times a
1180
# second so small variations add up.
1181
assert value.__class__ is str
1182
result.append(self._history[int(value)])
1185
def get_graph(self):
1187
for version_id, index in self._cache.iteritems():
1188
graph.append((version_id, index[4]))
1191
def get_ancestry(self, versions):
1192
"""See VersionedFile.get_ancestry."""
1193
# get a graph of all the mentioned versions:
1195
pending = set(versions)
1197
version = pending.pop()
1198
parents = self._cache[version][4]
1199
# got the parents ok
1201
parents = [parent for parent in parents if parent in self._cache]
1202
for parent in parents:
1203
# if not completed and not a ghost
1204
if parent not in graph:
1206
graph[version] = parents
1207
return topo_sort(graph.items())
1209
def get_ancestry_with_ghosts(self, versions):
1210
"""See VersionedFile.get_ancestry_with_ghosts."""
1211
# get a graph of all the mentioned versions:
1213
pending = set(versions)
1215
version = pending.pop()
1217
parents = self._cache[version][4]
1223
# got the parents ok
1224
for parent in parents:
1225
if parent not in graph:
1227
graph[version] = parents
1228
return topo_sort(graph.items())
1230
def num_versions(self):
1231
return len(self._history)
1233
__len__ = num_versions
1235
def get_versions(self):
1236
return self._history
1238
def idx_to_name(self, idx):
1239
return self._history[idx]
1241
def lookup(self, version_id):
1242
assert version_id in self._cache
1243
return self._cache[version_id][5]
1245
def _version_list_to_index(self, versions):
1246
encode_utf8 = cache_utf8.encode
1248
for version in versions:
1249
if version in self._cache:
1250
# -- inlined lookup() --
1251
result_list.append(str(self._cache[version][5]))
1252
# -- end lookup () --
1254
result_list.append('.' + encode_utf8(version))
1255
return ' '.join(result_list)
1257
def add_version(self, version_id, options, pos, size, parents):
1258
"""Add a version record to the index."""
1259
self.add_versions(((version_id, options, pos, size, parents),))
1261
def add_versions(self, versions):
1262
"""Add multiple versions to the index.
1264
:param versions: a list of tuples:
1265
(version_id, options, pos, size, parents).
1268
encode_utf8 = cache_utf8.encode
1269
orig_history = self._history[:]
1270
orig_cache = self._cache.copy()
1273
for version_id, options, pos, size, parents in versions:
1274
line = "\n%s %s %s %s %s :" % (encode_utf8(version_id),
1278
self._version_list_to_index(parents))
1279
assert isinstance(line, str), \
1280
'content must be utf-8 encoded: %r' % (line,)
1282
self._cache_version(version_id, options, pos, size, parents)
1283
if not self._need_to_create:
1284
self._transport.append_bytes(self._filename, ''.join(lines))
1287
sio.write(self.HEADER)
1288
sio.writelines(lines)
1290
self._transport.put_file_non_atomic(self._filename, sio,
1291
create_parent_dir=self._create_parent_dir,
1292
mode=self._file_mode,
1293
dir_mode=self._dir_mode)
1294
self._need_to_create = False
1296
# If any problems happen, restore the original values and re-raise
1297
self._history = orig_history
1298
self._cache = orig_cache
1301
def has_version(self, version_id):
1302
"""True if the version is in the index."""
1303
return (version_id in self._cache)
1305
def get_position(self, version_id):
1306
"""Return data position and size of specified version."""
1307
return (self._cache[version_id][2], \
1308
self._cache[version_id][3])
1310
def get_method(self, version_id):
1311
"""Return compression method of specified version."""
1312
options = self._cache[version_id][1]
1313
if 'fulltext' in options:
1316
assert 'line-delta' in options
1319
def get_options(self, version_id):
1320
return self._cache[version_id][1]
1322
def get_parents(self, version_id):
1323
"""Return parents of specified version ignoring ghosts."""
1324
return [parent for parent in self._cache[version_id][4]
1325
if parent in self._cache]
1327
def get_parents_with_ghosts(self, version_id):
1328
"""Return parents of specified version with ghosts."""
1329
return self._cache[version_id][4]
1331
def check_versions_present(self, version_ids):
1332
"""Check that all specified versions are present."""
1333
version_ids = set(version_ids)
1334
for version_id in list(version_ids):
1335
if version_id in self._cache:
1336
version_ids.remove(version_id)
1338
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1341
class _KnitData(_KnitComponentFile):
1342
"""Contents of the knit data file"""
1344
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1345
create_parent_dir=False, delay_create=False,
1347
_KnitComponentFile.__init__(self, transport, filename, mode,
1348
file_mode=file_mode,
1349
create_parent_dir=create_parent_dir,
1351
self._checked = False
1352
# TODO: jam 20060713 conceptually, this could spill to disk
1353
# if the cached size gets larger than a certain amount
1354
# but it complicates the model a bit, so for now just use
1355
# a simple dictionary
1357
self._do_cache = False
1360
self._need_to_create = create
1362
self._transport.put_bytes_non_atomic(self._filename, '',
1363
mode=self._file_mode)
1365
def enable_cache(self):
1366
"""Enable caching of reads."""
1367
self._do_cache = True
1369
def clear_cache(self):
1370
"""Clear the record cache."""
1371
self._do_cache = False
1374
def _open_file(self):
1376
return self._transport.get(self._filename)
1381
def _record_to_data(self, version_id, digest, lines):
1382
"""Convert version_id, digest, lines into a raw data block.
1384
:return: (len, a StringIO instance with the raw data ready to read.)
1387
data_file = GzipFile(None, mode='wb', fileobj=sio)
1389
version_id_utf8 = cache_utf8.encode(version_id)
1390
data_file.writelines(chain(
1391
["version %s %d %s\n" % (version_id_utf8,
1395
["end %s\n" % version_id_utf8]))
1402
def add_raw_record(self, raw_data):
1403
"""Append a prepared record to the data file.
1405
:return: the offset in the data file raw_data was written.
1407
assert isinstance(raw_data, str), 'data must be plain bytes'
1408
if not self._need_to_create:
1409
return self._transport.append_bytes(self._filename, raw_data)
1411
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1412
create_parent_dir=self._create_parent_dir,
1413
mode=self._file_mode,
1414
dir_mode=self._dir_mode)
1415
self._need_to_create = False
1418
def add_record(self, version_id, digest, lines):
1419
"""Write new text record to disk. Returns the position in the
1420
file where it was written."""
1421
size, sio = self._record_to_data(version_id, digest, lines)
1423
if not self._need_to_create:
1424
start_pos = self._transport.append_file(self._filename, sio)
1426
self._transport.put_file_non_atomic(self._filename, sio,
1427
create_parent_dir=self._create_parent_dir,
1428
mode=self._file_mode,
1429
dir_mode=self._dir_mode)
1430
self._need_to_create = False
1433
self._cache[version_id] = sio.getvalue()
1434
return start_pos, size
1436
def _parse_record_header(self, version_id, raw_data):
1437
"""Parse a record header for consistency.
1439
:return: the header and the decompressor stream.
1440
as (stream, header_record)
1442
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1443
rec = df.readline().split()
1445
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
1446
if cache_utf8.decode(rec[1]) != version_id:
1447
raise KnitCorrupt(self._filename,
1448
'unexpected version, wanted %r, got %r' % (
1449
version_id, rec[1]))
1452
def _parse_record(self, version_id, data):
1454
# 4168 calls in 2880 217 internal
1455
# 4168 calls to _parse_record_header in 2121
1456
# 4168 calls to readlines in 330
1457
df, rec = self._parse_record_header(version_id, data)
1458
record_contents = df.readlines()
1459
l = record_contents.pop()
1460
assert len(record_contents) == int(rec[2])
1461
if l != 'end %s\n' % cache_utf8.encode(version_id):
1462
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
1465
return record_contents, rec[3]
1467
def read_records_iter_raw(self, records):
1468
"""Read text records from data file and yield raw data.
1470
This unpacks enough of the text record to validate the id is
1471
as expected but thats all.
1473
# setup an iterator of the external records:
1474
# uses readv so nice and fast we hope.
1476
# grab the disk data needed.
1478
# Don't check _cache if it is empty
1479
needed_offsets = [(pos, size) for version_id, pos, size
1481
if version_id not in self._cache]
1483
needed_offsets = [(pos, size) for version_id, pos, size
1486
raw_records = self._transport.readv(self._filename, needed_offsets)
1489
for version_id, pos, size in records:
1490
if version_id in self._cache:
1491
# This data has already been validated
1492
data = self._cache[version_id]
1494
pos, data = raw_records.next()
1496
self._cache[version_id] = data
1498
# validate the header
1499
df, rec = self._parse_record_header(version_id, data)
1501
yield version_id, data
1503
def read_records_iter(self, records):
1504
"""Read text records from data file and yield result.
1506
The result will be returned in whatever is the fastest to read.
1507
Not by the order requested. Also, multiple requests for the same
1508
record will only yield 1 response.
1509
:param records: A list of (version_id, pos, len) entries
1510
:return: Yields (version_id, contents, digest) in the order
1511
read, not the order requested
1517
# Skip records we have alread seen
1518
yielded_records = set()
1519
needed_records = set()
1520
for record in records:
1521
if record[0] in self._cache:
1522
if record[0] in yielded_records:
1524
yielded_records.add(record[0])
1525
data = self._cache[record[0]]
1526
content, digest = self._parse_record(record[0], data)
1527
yield (record[0], content, digest)
1529
needed_records.add(record)
1530
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1532
needed_records = sorted(set(records), key=operator.itemgetter(1))
1534
if not needed_records:
1537
# The transport optimizes the fetching as well
1538
# (ie, reads continuous ranges.)
1539
readv_response = self._transport.readv(self._filename,
1540
[(pos, size) for version_id, pos, size in needed_records])
1542
for (version_id, pos, size), (pos, data) in \
1543
izip(iter(needed_records), readv_response):
1544
content, digest = self._parse_record(version_id, data)
1546
self._cache[version_id] = data
1547
yield version_id, content, digest
1549
def read_records(self, records):
1550
"""Read records into a dictionary."""
1552
for record_id, content, digest in \
1553
self.read_records_iter(records):
1554
components[record_id] = (content, digest)
1558
class InterKnit(InterVersionedFile):
1559
"""Optimised code paths for knit to knit operations."""
1561
_matching_file_from_factory = KnitVersionedFile
1562
_matching_file_to_factory = KnitVersionedFile
1565
def is_compatible(source, target):
1566
"""Be compatible with knits. """
1568
return (isinstance(source, KnitVersionedFile) and
1569
isinstance(target, KnitVersionedFile))
1570
except AttributeError:
1573
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1574
"""See InterVersionedFile.join."""
1575
assert isinstance(self.source, KnitVersionedFile)
1576
assert isinstance(self.target, KnitVersionedFile)
1578
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1583
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1585
version_ids = list(version_ids)
1586
if None in version_ids:
1587
version_ids.remove(None)
1589
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1590
this_versions = set(self.target._index.get_versions())
1591
needed_versions = self.source_ancestry - this_versions
1592
cross_check_versions = self.source_ancestry.intersection(this_versions)
1593
mismatched_versions = set()
1594
for version in cross_check_versions:
1595
# scan to include needed parents.
1596
n1 = set(self.target.get_parents_with_ghosts(version))
1597
n2 = set(self.source.get_parents_with_ghosts(version))
1599
# FIXME TEST this check for cycles being introduced works
1600
# the logic is we have a cycle if in our graph we are an
1601
# ancestor of any of the n2 revisions.
1607
parent_ancestors = self.source.get_ancestry(parent)
1608
if version in parent_ancestors:
1609
raise errors.GraphCycleError([parent, version])
1610
# ensure this parent will be available later.
1611
new_parents = n2.difference(n1)
1612
needed_versions.update(new_parents.difference(this_versions))
1613
mismatched_versions.add(version)
1615
if not needed_versions and not mismatched_versions:
1617
full_list = topo_sort(self.source.get_graph())
1619
version_list = [i for i in full_list if (not self.target.has_version(i)
1620
and i in needed_versions)]
1624
copy_queue_records = []
1626
for version_id in version_list:
1627
options = self.source._index.get_options(version_id)
1628
parents = self.source._index.get_parents_with_ghosts(version_id)
1629
# check that its will be a consistent copy:
1630
for parent in parents:
1631
# if source has the parent, we must :
1632
# * already have it or
1633
# * have it scheduled already
1634
# otherwise we don't care
1635
assert (self.target.has_version(parent) or
1636
parent in copy_set or
1637
not self.source.has_version(parent))
1638
data_pos, data_size = self.source._index.get_position(version_id)
1639
copy_queue_records.append((version_id, data_pos, data_size))
1640
copy_queue.append((version_id, options, parents))
1641
copy_set.add(version_id)
1643
# data suck the join:
1645
total = len(version_list)
1648
for (version_id, raw_data), \
1649
(version_id2, options, parents) in \
1650
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1652
assert version_id == version_id2, 'logic error, inconsistent results'
1654
pb.update("Joining knit", count, total)
1655
raw_records.append((version_id, options, parents, len(raw_data)))
1656
raw_datum.append(raw_data)
1657
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1659
for version in mismatched_versions:
1660
# FIXME RBC 20060309 is this needed?
1661
n1 = set(self.target.get_parents_with_ghosts(version))
1662
n2 = set(self.source.get_parents_with_ghosts(version))
1663
# write a combined record to our history preserving the current
1664
# parents as first in the list
1665
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1666
self.target.fix_parents(version, new_parents)
1672
InterVersionedFile.register_optimiser(InterKnit)
1675
class WeaveToKnit(InterVersionedFile):
1676
"""Optimised code paths for weave to knit operations."""
1678
_matching_file_from_factory = bzrlib.weave.WeaveFile
1679
_matching_file_to_factory = KnitVersionedFile
1682
def is_compatible(source, target):
1683
"""Be compatible with weaves to knits."""
1685
return (isinstance(source, bzrlib.weave.Weave) and
1686
isinstance(target, KnitVersionedFile))
1687
except AttributeError:
1690
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1691
"""See InterVersionedFile.join."""
1692
assert isinstance(self.source, bzrlib.weave.Weave)
1693
assert isinstance(self.target, KnitVersionedFile)
1695
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1700
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1702
version_ids = list(version_ids)
1704
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1705
this_versions = set(self.target._index.get_versions())
1706
needed_versions = self.source_ancestry - this_versions
1707
cross_check_versions = self.source_ancestry.intersection(this_versions)
1708
mismatched_versions = set()
1709
for version in cross_check_versions:
1710
# scan to include needed parents.
1711
n1 = set(self.target.get_parents_with_ghosts(version))
1712
n2 = set(self.source.get_parents(version))
1713
# if all of n2's parents are in n1, then its fine.
1714
if n2.difference(n1):
1715
# FIXME TEST this check for cycles being introduced works
1716
# the logic is we have a cycle if in our graph we are an
1717
# ancestor of any of the n2 revisions.
1723
parent_ancestors = self.source.get_ancestry(parent)
1724
if version in parent_ancestors:
1725
raise errors.GraphCycleError([parent, version])
1726
# ensure this parent will be available later.
1727
new_parents = n2.difference(n1)
1728
needed_versions.update(new_parents.difference(this_versions))
1729
mismatched_versions.add(version)
1731
if not needed_versions and not mismatched_versions:
1733
full_list = topo_sort(self.source.get_graph())
1735
version_list = [i for i in full_list if (not self.target.has_version(i)
1736
and i in needed_versions)]
1740
total = len(version_list)
1741
for version_id in version_list:
1742
pb.update("Converting to knit", count, total)
1743
parents = self.source.get_parents(version_id)
1744
# check that its will be a consistent copy:
1745
for parent in parents:
1746
# if source has the parent, we must already have it
1747
assert (self.target.has_version(parent))
1748
self.target.add_lines(
1749
version_id, parents, self.source.get_lines(version_id))
1752
for version in mismatched_versions:
1753
# FIXME RBC 20060309 is this needed?
1754
n1 = set(self.target.get_parents_with_ghosts(version))
1755
n2 = set(self.source.get_parents(version))
1756
# write a combined record to our history preserving the current
1757
# parents as first in the list
1758
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1759
self.target.fix_parents(version, new_parents)
1765
InterVersionedFile.register_optimiser(WeaveToKnit)
1768
class KnitSequenceMatcher(difflib.SequenceMatcher):
1769
"""Knit tuned sequence matcher.
1771
This is based on profiling of difflib which indicated some improvements
1772
for our usage pattern.
1775
def find_longest_match(self, alo, ahi, blo, bhi):
1776
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1778
If isjunk is not defined:
1780
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1781
alo <= i <= i+k <= ahi
1782
blo <= j <= j+k <= bhi
1783
and for all (i',j',k') meeting those conditions,
1786
and if i == i', j <= j'
1788
In other words, of all maximal matching blocks, return one that
1789
starts earliest in a, and of all those maximal matching blocks that
1790
start earliest in a, return the one that starts earliest in b.
1792
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1793
>>> s.find_longest_match(0, 5, 0, 9)
1796
If isjunk is defined, first the longest matching block is
1797
determined as above, but with the additional restriction that no
1798
junk element appears in the block. Then that block is extended as
1799
far as possible by matching (only) junk elements on both sides. So
1800
the resulting block never matches on junk except as identical junk
1801
happens to be adjacent to an "interesting" match.
1803
Here's the same example as before, but considering blanks to be
1804
junk. That prevents " abcd" from matching the " abcd" at the tail
1805
end of the second sequence directly. Instead only the "abcd" can
1806
match, and matches the leftmost "abcd" in the second sequence:
1808
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1809
>>> s.find_longest_match(0, 5, 0, 9)
1812
If no blocks match, return (alo, blo, 0).
1814
>>> s = SequenceMatcher(None, "ab", "c")
1815
>>> s.find_longest_match(0, 2, 0, 1)
1819
# CAUTION: stripping common prefix or suffix would be incorrect.
1823
# Longest matching block is "ab", but if common prefix is
1824
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1825
# strip, so ends up claiming that ab is changed to acab by
1826
# inserting "ca" in the middle. That's minimal but unintuitive:
1827
# "it's obvious" that someone inserted "ac" at the front.
1828
# Windiff ends up at the same place as diff, but by pairing up
1829
# the unique 'b's and then matching the first two 'a's.
1831
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1832
besti, bestj, bestsize = alo, blo, 0
1833
# find longest junk-free match
1834
# during an iteration of the loop, j2len[j] = length of longest
1835
# junk-free match ending with a[i-1] and b[j]
1839
for i in xrange(alo, ahi):
1840
# look at all instances of a[i] in b; note that because
1841
# b2j has no junk keys, the loop is skipped if a[i] is junk
1842
j2lenget = j2len.get
1845
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1846
# following improvement
1847
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1848
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1849
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1851
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1852
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1853
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1865
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1867
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1870
# Extend the best by non-junk elements on each end. In particular,
1871
# "popular" non-junk elements aren't in b2j, which greatly speeds
1872
# the inner loop above, but also means "the best" match so far
1873
# doesn't contain any junk *or* popular non-junk elements.
1874
while besti > alo and bestj > blo and \
1875
not isbjunk(b[bestj-1]) and \
1876
a[besti-1] == b[bestj-1]:
1877
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1878
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1879
not isbjunk(b[bestj+bestsize]) and \
1880
a[besti+bestsize] == b[bestj+bestsize]:
1883
# Now that we have a wholly interesting match (albeit possibly
1884
# empty!), we may as well suck up the matching junk on each
1885
# side of it too. Can't think of a good reason not to, and it
1886
# saves post-processing the (possibly considerable) expense of
1887
# figuring out what to do with it. In the case of an empty
1888
# interesting match, this is clearly the right thing to do,
1889
# because no other kind of match is possible in the regions.
1890
while besti > alo and bestj > blo and \
1891
isbjunk(b[bestj-1]) and \
1892
a[besti-1] == b[bestj-1]:
1893
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1894
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1895
isbjunk(b[bestj+bestsize]) and \
1896
a[besti+bestsize] == b[bestj+bestsize]:
1897
bestsize = bestsize + 1
1899
return besti, bestj, bestsize
added
removed
bzrlib/knit.py
