1
# Copyright (C) 2005, 2006, 2007 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""Knit versionedfile implementation.
19
A knit is a versioned file implementation that supports efficient append only
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
end-marker; simply "end VERSION"
28
delta can be line or full contents.a
29
... the 8's there are the index number of the annotation.
30
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
8 e.set('executable', 'yes')
36
8 if elt.get('executable') == 'yes':
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
43
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
44
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
45
09:33 < lifeless> right
46
09:33 < jrydberg> lifeless: the position and size is the range in the data file
49
so the index sequence is the dictionary compressed sequence number used
50
in the deltas to provide line annotation
55
# 10:16 < lifeless> make partial index writes safe
56
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
57
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
59
# move sha1 out of the content so that join is faster at verifying parents
60
# record content length ?
64
from cStringIO import StringIO
66
from itertools import izip, chain
82
from bzrlib.errors import (
90
RevisionAlreadyPresent,
92
from bzrlib.tuned_gzip import GzipFile
93
from bzrlib.trace import mutter
94
from bzrlib.osutils import (
99
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
100
from bzrlib.tsort import topo_sort
103
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
106
# TODO: Split out code specific to this format into an associated object.
108
# TODO: Can we put in some kind of value to check that the index and data
109
# files belong together?
111
# TODO: accommodate binaries, perhaps by storing a byte count
113
# TODO: function to check whole file
115
# TODO: atomically append data, then measure backwards from the cursor
116
# position after writing to work out where it was located. we may need to
117
# bypass python file buffering.
119
DATA_SUFFIX = '.knit'
120
INDEX_SUFFIX = '.kndx'
123
class KnitContent(object):
124
"""Content of a knit version to which deltas can be applied."""
126
def __init__(self, lines):
129
def annotate_iter(self):
130
"""Yield tuples of (origin, text) for each content line."""
131
return iter(self._lines)
134
"""Return a list of (origin, text) tuples."""
135
return list(self.annotate_iter())
137
def line_delta_iter(self, new_lines):
138
"""Generate line-based delta from this content to new_lines."""
139
new_texts = new_lines.text()
140
old_texts = self.text()
141
s = KnitSequenceMatcher(None, old_texts, new_texts)
142
for tag, i1, i2, j1, j2 in s.get_opcodes():
145
# ofrom, oto, length, data
146
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
148
def line_delta(self, new_lines):
149
return list(self.line_delta_iter(new_lines))
152
return [text for origin, text in self._lines]
155
return KnitContent(self._lines[:])
158
def get_line_delta_blocks(knit_delta, source, target):
159
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
160
target_len = len(target)
163
for s_begin, s_end, t_len, new_text in knit_delta:
164
true_n = s_begin - s_pos
167
# knit deltas do not provide reliable info about whether the
168
# last line of a file matches, due to eol handling.
169
if source[s_pos + n -1] != target[t_pos + n -1]:
172
yield s_pos, t_pos, n
173
t_pos += t_len + true_n
175
n = target_len - t_pos
177
if source[s_pos + n -1] != target[t_pos + n -1]:
180
yield s_pos, t_pos, n
181
yield s_pos + (target_len - t_pos), target_len, 0
184
class _KnitFactory(object):
185
"""Base factory for creating content objects."""
187
def make(self, lines, version_id):
188
num_lines = len(lines)
189
return KnitContent(zip([version_id] * num_lines, lines))
192
class KnitAnnotateFactory(_KnitFactory):
193
"""Factory for creating annotated Content objects."""
197
def parse_fulltext(self, content, version_id):
198
"""Convert fulltext to internal representation
200
fulltext content is of the format
201
revid(utf8) plaintext\n
202
internal representation is of the format:
205
# TODO: jam 20070209 The tests expect this to be returned as tuples,
206
# but the code itself doesn't really depend on that.
207
# Figure out a way to not require the overhead of turning the
208
# list back into tuples.
209
lines = [tuple(line.split(' ', 1)) for line in content]
210
return KnitContent(lines)
212
def parse_line_delta_iter(self, lines):
213
return iter(self.parse_line_delta(lines))
215
def parse_line_delta(self, lines, version_id):
216
"""Convert a line based delta into internal representation.
218
line delta is in the form of:
219
intstart intend intcount
221
revid(utf8) newline\n
222
internal representation is
223
(start, end, count, [1..count tuples (revid, newline)])
230
def cache_and_return(line):
231
origin, text = line.split(' ', 1)
232
return cache.setdefault(origin, origin), text
234
# walk through the lines parsing.
236
start, end, count = [int(n) for n in header.split(',')]
237
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
238
result.append((start, end, count, contents))
241
def get_fulltext_content(self, lines):
242
"""Extract just the content lines from a fulltext."""
243
return (line.split(' ', 1)[1] for line in lines)
245
def get_linedelta_content(self, lines):
246
"""Extract just the content from a line delta.
248
This doesn't return all of the extra information stored in a delta.
249
Only the actual content lines.
254
header = header.split(',')
255
count = int(header[2])
256
for i in xrange(count):
257
origin, text = next().split(' ', 1)
260
def lower_fulltext(self, content):
261
"""convert a fulltext content record into a serializable form.
263
see parse_fulltext which this inverts.
265
# TODO: jam 20070209 We only do the caching thing to make sure that
266
# the origin is a valid utf-8 line, eventually we could remove it
267
return ['%s %s' % (o, t) for o, t in content._lines]
269
def lower_line_delta(self, delta):
270
"""convert a delta into a serializable form.
272
See parse_line_delta which this inverts.
274
# TODO: jam 20070209 We only do the caching thing to make sure that
275
# the origin is a valid utf-8 line, eventually we could remove it
277
for start, end, c, lines in delta:
278
out.append('%d,%d,%d\n' % (start, end, c))
279
out.extend(origin + ' ' + text
280
for origin, text in lines)
284
class KnitPlainFactory(_KnitFactory):
285
"""Factory for creating plain Content objects."""
289
def parse_fulltext(self, content, version_id):
290
"""This parses an unannotated fulltext.
292
Note that this is not a noop - the internal representation
293
has (versionid, line) - its just a constant versionid.
295
return self.make(content, version_id)
297
def parse_line_delta_iter(self, lines, version_id):
299
num_lines = len(lines)
300
while cur < num_lines:
303
start, end, c = [int(n) for n in header.split(',')]
304
yield start, end, c, zip([version_id] * c, lines[cur:cur+c])
307
def parse_line_delta(self, lines, version_id):
308
return list(self.parse_line_delta_iter(lines, version_id))
310
def get_fulltext_content(self, lines):
311
"""Extract just the content lines from a fulltext."""
314
def get_linedelta_content(self, lines):
315
"""Extract just the content from a line delta.
317
This doesn't return all of the extra information stored in a delta.
318
Only the actual content lines.
323
header = header.split(',')
324
count = int(header[2])
325
for i in xrange(count):
328
def lower_fulltext(self, content):
329
return content.text()
331
def lower_line_delta(self, delta):
333
for start, end, c, lines in delta:
334
out.append('%d,%d,%d\n' % (start, end, c))
335
out.extend([text for origin, text in lines])
339
def make_empty_knit(transport, relpath):
340
"""Construct a empty knit at the specified location."""
341
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
345
class KnitVersionedFile(VersionedFile):
346
"""Weave-like structure with faster random access.
348
A knit stores a number of texts and a summary of the relationships
349
between them. Texts are identified by a string version-id. Texts
350
are normally stored and retrieved as a series of lines, but can
351
also be passed as single strings.
353
Lines are stored with the trailing newline (if any) included, to
354
avoid special cases for files with no final newline. Lines are
355
composed of 8-bit characters, not unicode. The combination of
356
these approaches should mean any 'binary' file can be safely
357
stored and retrieved.
360
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
361
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
362
create=False, create_parent_dir=False, delay_create=False,
364
"""Construct a knit at location specified by relpath.
366
:param create: If not True, only open an existing knit.
367
:param create_parent_dir: If True, create the parent directory if
368
creating the file fails. (This is used for stores with
369
hash-prefixes that may not exist yet)
370
:param delay_create: The calling code is aware that the knit won't
371
actually be created until the first data is stored.
373
if deprecated_passed(basis_knit):
374
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
375
" deprecated as of bzr 0.9.",
376
DeprecationWarning, stacklevel=2)
377
if access_mode is None:
379
super(KnitVersionedFile, self).__init__(access_mode)
380
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
381
self.transport = transport
382
self.filename = relpath
383
self.factory = factory or KnitAnnotateFactory()
384
self.writable = (access_mode == 'w')
387
self._max_delta_chain = 200
389
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
390
access_mode, create=create, file_mode=file_mode,
391
create_parent_dir=create_parent_dir, delay_create=delay_create,
393
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
394
access_mode, create=create and not len(self), file_mode=file_mode,
395
create_parent_dir=create_parent_dir, delay_create=delay_create,
399
return '%s(%s)' % (self.__class__.__name__,
400
self.transport.abspath(self.filename))
402
def _check_should_delta(self, first_parents):
403
"""Iterate back through the parent listing, looking for a fulltext.
405
This is used when we want to decide whether to add a delta or a new
406
fulltext. It searches for _max_delta_chain parents. When it finds a
407
fulltext parent, it sees if the total size of the deltas leading up to
408
it is large enough to indicate that we want a new full text anyway.
410
Return True if we should create a new delta, False if we should use a
415
delta_parents = first_parents
416
for count in xrange(self._max_delta_chain):
417
parent = delta_parents[0]
418
method = self._index.get_method(parent)
419
pos, size = self._index.get_position(parent)
420
if method == 'fulltext':
424
delta_parents = self._index.get_parents(parent)
426
# We couldn't find a fulltext, so we must create a new one
429
return fulltext_size > delta_size
431
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
432
"""See VersionedFile._add_delta()."""
433
self._check_add(version_id, []) # should we check the lines ?
434
self._check_versions_present(parents)
438
for parent in parents:
439
if not self.has_version(parent):
440
ghosts.append(parent)
442
present_parents.append(parent)
444
if delta_parent is None:
445
# reconstitute as full text.
446
assert len(delta) == 1 or len(delta) == 0
448
assert delta[0][0] == 0
449
assert delta[0][1] == 0, delta[0][1]
450
return super(KnitVersionedFile, self)._add_delta(version_id,
461
options.append('no-eol')
463
if delta_parent is not None:
464
# determine the current delta chain length.
465
# To speed the extract of texts the delta chain is limited
466
# to a fixed number of deltas. This should minimize both
467
# I/O and the time spend applying deltas.
468
# The window was changed to a maximum of 200 deltas, but also added
469
# was a check that the total compressed size of the deltas is
470
# smaller than the compressed size of the fulltext.
471
if not self._check_should_delta([delta_parent]):
472
# We don't want a delta here, just do a normal insertion.
473
return super(KnitVersionedFile, self)._add_delta(version_id,
480
options.append('line-delta')
481
store_lines = self.factory.lower_line_delta(delta)
483
where, size = self._data.add_record(version_id, digest, store_lines)
484
self._index.add_version(version_id, options, where, size, parents)
486
def _add_raw_records(self, records, data):
487
"""Add all the records 'records' with data pre-joined in 'data'.
489
:param records: A list of tuples(version_id, options, parents, size).
490
:param data: The data for the records. When it is written, the records
491
are adjusted to have pos pointing into data by the sum of
492
the preceding records sizes.
495
pos = self._data.add_raw_record(data)
498
for (version_id, options, parents, size) in records:
499
index_entries.append((version_id, options, pos+offset,
501
if self._data._do_cache:
502
self._data._cache[version_id] = data[offset:offset+size]
504
self._index.add_versions(index_entries)
506
def enable_cache(self):
507
"""Start caching data for this knit"""
508
self._data.enable_cache()
510
def clear_cache(self):
511
"""Clear the data cache only."""
512
self._data.clear_cache()
514
def copy_to(self, name, transport):
515
"""See VersionedFile.copy_to()."""
516
# copy the current index to a temp index to avoid racing with local
518
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
519
self.transport.get(self._index._filename))
521
f = self._data._open_file()
523
transport.put_file(name + DATA_SUFFIX, f)
526
# move the copied index into place
527
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
529
def create_empty(self, name, transport, mode=None):
530
return KnitVersionedFile(name, transport, factory=self.factory,
531
delta=self.delta, create=True)
533
def _fix_parents(self, version_id, new_parents):
534
"""Fix the parents list for version.
536
This is done by appending a new version to the index
537
with identical data except for the parents list.
538
the parents list must be a superset of the current
541
current_values = self._index._cache[version_id]
542
assert set(current_values[4]).difference(set(new_parents)) == set()
543
self._index.add_version(version_id,
549
def _extract_blocks(self, version_id, source, target):
550
if self._index.get_method(version_id) != 'line-delta':
552
parent, sha1, noeol, delta = self.get_delta(version_id)
553
return KnitContent.get_line_delta_blocks(delta, source, target)
555
def get_delta(self, version_id):
556
"""Get a delta for constructing version from some other version."""
557
version_id = osutils.safe_revision_id(version_id)
558
self.check_not_reserved_id(version_id)
559
if not self.has_version(version_id):
560
raise RevisionNotPresent(version_id, self.filename)
562
parents = self.get_parents(version_id)
567
data_pos, data_size = self._index.get_position(version_id)
568
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
569
noeol = 'no-eol' in self._index.get_options(version_id)
570
if 'fulltext' == self._index.get_method(version_id):
571
new_content = self.factory.parse_fulltext(data, version_id)
572
if parent is not None:
573
reference_content = self._get_content(parent)
574
old_texts = reference_content.text()
577
new_texts = new_content.text()
578
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
579
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
581
delta = self.factory.parse_line_delta(data, version_id)
582
return parent, sha1, noeol, delta
584
def get_graph_with_ghosts(self):
585
"""See VersionedFile.get_graph_with_ghosts()."""
586
graph_items = self._index.get_graph()
587
return dict(graph_items)
589
def get_sha1(self, version_id):
590
return self.get_sha1s([version_id])[0]
592
def get_sha1s(self, version_ids):
593
"""See VersionedFile.get_sha1()."""
594
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
595
record_map = self._get_record_map(version_ids)
596
# record entry 2 is the 'digest'.
597
return [record_map[v][2] for v in version_ids]
601
"""See VersionedFile.get_suffixes()."""
602
return [DATA_SUFFIX, INDEX_SUFFIX]
604
def has_ghost(self, version_id):
605
"""True if there is a ghost reference in the file to version_id."""
606
version_id = osutils.safe_revision_id(version_id)
608
if self.has_version(version_id):
610
# optimisable if needed by memoising the _ghosts set.
611
items = self._index.get_graph()
612
for node, parents in items:
613
for parent in parents:
614
if parent not in self._index._cache:
615
if parent == version_id:
620
"""See VersionedFile.versions."""
621
return self._index.get_versions()
623
def has_version(self, version_id):
624
"""See VersionedFile.has_version."""
625
version_id = osutils.safe_revision_id(version_id)
626
return self._index.has_version(version_id)
628
__contains__ = has_version
630
def _merge_annotations(self, content, parents, parent_texts={},
631
delta=None, annotated=None):
632
"""Merge annotations for content. This is done by comparing
633
the annotations based on changed to the text.
637
for parent_id in parents:
638
merge_content = self._get_content(parent_id, parent_texts)
639
seq = patiencediff.PatienceSequenceMatcher(
640
None, merge_content.text(), content.text())
641
if delta_seq is None:
642
# setup a delta seq to reuse.
644
for i, j, n in seq.get_matching_blocks():
647
# this appears to copy (origin, text) pairs across to the new
648
# content for any line that matches the last-checked parent.
649
# FIXME: save the sequence control data for delta compression
650
# against the most relevant parent rather than rediffing.
651
content._lines[j:j+n] = merge_content._lines[i:i+n]
654
reference_content = self._get_content(parents[0], parent_texts)
655
new_texts = content.text()
656
old_texts = reference_content.text()
657
delta_seq = patiencediff.PatienceSequenceMatcher(
658
None, old_texts, new_texts)
659
return self._make_line_delta(delta_seq, content)
661
def _make_line_delta(self, delta_seq, new_content):
662
"""Generate a line delta from delta_seq and new_content."""
664
for op in delta_seq.get_opcodes():
667
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
670
def _get_components_positions(self, version_ids):
671
"""Produce a map of position data for the components of versions.
673
This data is intended to be used for retrieving the knit records.
675
A dict of version_id to (method, data_pos, data_size, next) is
677
method is the way referenced data should be applied.
678
data_pos is the position of the data in the knit.
679
data_size is the size of the data in the knit.
680
next is the build-parent of the version, or None for fulltexts.
683
for version_id in version_ids:
686
while cursor is not None and cursor not in component_data:
687
method = self._index.get_method(cursor)
688
if method == 'fulltext':
691
next = self.get_parents(cursor)[0]
692
data_pos, data_size = self._index.get_position(cursor)
693
component_data[cursor] = (method, data_pos, data_size, next)
695
return component_data
697
def _get_content(self, version_id, parent_texts={}):
698
"""Returns a content object that makes up the specified
700
if not self.has_version(version_id):
701
raise RevisionNotPresent(version_id, self.filename)
703
cached_version = parent_texts.get(version_id, None)
704
if cached_version is not None:
705
return cached_version
707
text_map, contents_map = self._get_content_maps([version_id])
708
return contents_map[version_id]
710
def _check_versions_present(self, version_ids):
711
"""Check that all specified versions are present."""
712
self._index.check_versions_present(version_ids)
714
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
715
"""See VersionedFile.add_lines_with_ghosts()."""
716
self._check_add(version_id, lines)
717
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
719
def _add_lines(self, version_id, parents, lines, parent_texts):
720
"""See VersionedFile.add_lines."""
721
self._check_add(version_id, lines)
722
self._check_versions_present(parents)
723
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
725
def _check_add(self, version_id, lines):
726
"""check that version_id and lines are safe to add."""
727
assert self.writable, "knit is not opened for write"
728
### FIXME escape. RBC 20060228
729
if contains_whitespace(version_id):
730
raise InvalidRevisionId(version_id, self.filename)
731
self.check_not_reserved_id(version_id)
732
if self.has_version(version_id):
733
raise RevisionAlreadyPresent(version_id, self.filename)
734
self._check_lines_not_unicode(lines)
735
self._check_lines_are_lines(lines)
737
def _add(self, version_id, lines, parents, delta, parent_texts):
738
"""Add a set of lines on top of version specified by parents.
740
If delta is true, compress the text as a line-delta against
743
Any versions not present will be converted into ghosts.
745
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
746
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
747
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
748
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
749
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
750
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
751
# +1383 0 8.0370 8.0370 +<len>
752
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
753
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
754
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
755
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
759
if parent_texts is None:
761
for parent in parents:
762
if not self.has_version(parent):
763
ghosts.append(parent)
765
present_parents.append(parent)
767
if delta and not len(present_parents):
770
digest = sha_strings(lines)
773
if lines[-1][-1] != '\n':
774
options.append('no-eol')
775
lines[-1] = lines[-1] + '\n'
777
if len(present_parents) and delta:
778
# To speed the extract of texts the delta chain is limited
779
# to a fixed number of deltas. This should minimize both
780
# I/O and the time spend applying deltas.
781
delta = self._check_should_delta(present_parents)
783
assert isinstance(version_id, str)
784
lines = self.factory.make(lines, version_id)
785
if delta or (self.factory.annotated and len(present_parents) > 0):
786
# Merge annotations from parent texts if so is needed.
787
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
788
delta, self.factory.annotated)
791
options.append('line-delta')
792
store_lines = self.factory.lower_line_delta(delta_hunks)
794
options.append('fulltext')
795
store_lines = self.factory.lower_fulltext(lines)
797
where, size = self._data.add_record(version_id, digest, store_lines)
798
self._index.add_version(version_id, options, where, size, parents)
801
def check(self, progress_bar=None):
802
"""See VersionedFile.check()."""
804
def _clone_text(self, new_version_id, old_version_id, parents):
805
"""See VersionedFile.clone_text()."""
806
# FIXME RBC 20060228 make fast by only inserting an index with null
808
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
810
def get_lines(self, version_id):
811
"""See VersionedFile.get_lines()."""
812
return self.get_line_list([version_id])[0]
814
def _get_record_map(self, version_ids):
815
"""Produce a dictionary of knit records.
817
The keys are version_ids, the values are tuples of (method, content,
819
method is the way the content should be applied.
820
content is a KnitContent object.
821
digest is the SHA1 digest of this version id after all steps are done
822
next is the build-parent of the version, i.e. the leftmost ancestor.
823
If the method is fulltext, next will be None.
825
position_map = self._get_components_positions(version_ids)
826
# c = component_id, m = method, p = position, s = size, n = next
827
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
829
for component_id, content, digest in \
830
self._data.read_records_iter(records):
831
method, position, size, next = position_map[component_id]
832
record_map[component_id] = method, content, digest, next
836
def get_text(self, version_id):
837
"""See VersionedFile.get_text"""
838
return self.get_texts([version_id])[0]
840
def get_texts(self, version_ids):
841
return [''.join(l) for l in self.get_line_list(version_ids)]
843
def get_line_list(self, version_ids):
844
"""Return the texts of listed versions as a list of strings."""
845
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
846
for version_id in version_ids:
847
self.check_not_reserved_id(version_id)
848
text_map, content_map = self._get_content_maps(version_ids)
849
return [text_map[v] for v in version_ids]
851
_get_lf_split_line_list = get_line_list
853
def _get_content_maps(self, version_ids):
854
"""Produce maps of text and KnitContents
856
:return: (text_map, content_map) where text_map contains the texts for
857
the requested versions and content_map contains the KnitContents.
858
Both dicts take version_ids as their keys.
860
for version_id in version_ids:
861
if not self.has_version(version_id):
862
raise RevisionNotPresent(version_id, self.filename)
863
record_map = self._get_record_map(version_ids)
868
for version_id in version_ids:
871
while cursor is not None:
872
method, data, digest, next = record_map[cursor]
873
components.append((cursor, method, data, digest))
874
if cursor in content_map:
879
for component_id, method, data, digest in reversed(components):
880
if component_id in content_map:
881
content = content_map[component_id]
883
if method == 'fulltext':
884
assert content is None
885
content = self.factory.parse_fulltext(data, version_id)
886
elif method == 'line-delta':
887
delta = self.factory.parse_line_delta(data, version_id)
888
content = content.copy()
889
content._lines = self._apply_delta(content._lines,
891
content_map[component_id] = content
893
if 'no-eol' in self._index.get_options(version_id):
894
content = content.copy()
895
line = content._lines[-1][1].rstrip('\n')
896
content._lines[-1] = (content._lines[-1][0], line)
897
final_content[version_id] = content
899
# digest here is the digest from the last applied component.
900
text = content.text()
901
if sha_strings(text) != digest:
902
raise KnitCorrupt(self.filename,
903
'sha-1 does not match %s' % version_id)
905
text_map[version_id] = text
906
return text_map, final_content
908
def iter_lines_added_or_present_in_versions(self, version_ids=None,
910
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
911
if version_ids is None:
912
version_ids = self.versions()
914
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
916
pb = progress.DummyProgress()
917
# we don't care about inclusions, the caller cares.
918
# but we need to setup a list of records to visit.
919
# we need version_id, position, length
920
version_id_records = []
921
requested_versions = set(version_ids)
922
# filter for available versions
923
for version_id in requested_versions:
924
if not self.has_version(version_id):
925
raise RevisionNotPresent(version_id, self.filename)
926
# get a in-component-order queue:
927
for version_id in self.versions():
928
if version_id in requested_versions:
929
data_pos, length = self._index.get_position(version_id)
930
version_id_records.append((version_id, data_pos, length))
932
total = len(version_id_records)
933
for version_idx, (version_id, data, sha_value) in \
934
enumerate(self._data.read_records_iter(version_id_records)):
935
pb.update('Walking content.', version_idx, total)
936
method = self._index.get_method(version_id)
938
assert method in ('fulltext', 'line-delta')
939
if method == 'fulltext':
940
line_iterator = self.factory.get_fulltext_content(data)
942
line_iterator = self.factory.get_linedelta_content(data)
943
for line in line_iterator:
946
pb.update('Walking content.', total, total)
948
def num_versions(self):
949
"""See VersionedFile.num_versions()."""
950
return self._index.num_versions()
952
__len__ = num_versions
954
def annotate_iter(self, version_id):
955
"""See VersionedFile.annotate_iter."""
956
version_id = osutils.safe_revision_id(version_id)
957
content = self._get_content(version_id)
958
for origin, text in content.annotate_iter():
961
def get_parents(self, version_id):
962
"""See VersionedFile.get_parents."""
965
# 52554 calls in 1264 872 internal down from 3674
966
version_id = osutils.safe_revision_id(version_id)
968
return self._index.get_parents(version_id)
970
raise RevisionNotPresent(version_id, self.filename)
972
def get_parents_with_ghosts(self, version_id):
973
"""See VersionedFile.get_parents."""
974
version_id = osutils.safe_revision_id(version_id)
976
return self._index.get_parents_with_ghosts(version_id)
978
raise RevisionNotPresent(version_id, self.filename)
980
def get_ancestry(self, versions, topo_sorted=True):
981
"""See VersionedFile.get_ancestry."""
982
if isinstance(versions, basestring):
983
versions = [versions]
986
versions = [osutils.safe_revision_id(v) for v in versions]
987
return self._index.get_ancestry(versions, topo_sorted)
989
def get_ancestry_with_ghosts(self, versions):
990
"""See VersionedFile.get_ancestry_with_ghosts."""
991
if isinstance(versions, basestring):
992
versions = [versions]
995
versions = [osutils.safe_revision_id(v) for v in versions]
996
return self._index.get_ancestry_with_ghosts(versions)
998
#@deprecated_method(zero_eight)
999
def walk(self, version_ids):
1000
"""See VersionedFile.walk."""
1001
# We take the short path here, and extract all relevant texts
1002
# and put them in a weave and let that do all the work. Far
1003
# from optimal, but is much simpler.
1004
# FIXME RB 20060228 this really is inefficient!
1005
from bzrlib.weave import Weave
1007
w = Weave(self.filename)
1008
ancestry = set(self.get_ancestry(version_ids, topo_sorted=False))
1009
sorted_graph = topo_sort(self._index.get_graph())
1010
version_list = [vid for vid in sorted_graph if vid in ancestry]
1012
for version_id in version_list:
1013
lines = self.get_lines(version_id)
1014
w.add_lines(version_id, self.get_parents(version_id), lines)
1016
for lineno, insert_id, dset, line in w.walk(version_ids):
1017
yield lineno, insert_id, dset, line
1019
def plan_merge(self, ver_a, ver_b):
1020
"""See VersionedFile.plan_merge."""
1021
ver_a = osutils.safe_revision_id(ver_a)
1022
ver_b = osutils.safe_revision_id(ver_b)
1023
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1025
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1026
annotated_a = self.annotate(ver_a)
1027
annotated_b = self.annotate(ver_b)
1028
return merge._plan_annotate_merge(annotated_a, annotated_b,
1029
ancestors_a, ancestors_b)
1032
class _KnitComponentFile(object):
1033
"""One of the files used to implement a knit database"""
1035
def __init__(self, transport, filename, mode, file_mode=None,
1036
create_parent_dir=False, dir_mode=None):
1037
self._transport = transport
1038
self._filename = filename
1040
self._file_mode = file_mode
1041
self._dir_mode = dir_mode
1042
self._create_parent_dir = create_parent_dir
1043
self._need_to_create = False
1045
def _full_path(self):
1046
"""Return the full path to this file."""
1047
return self._transport.base + self._filename
1049
def check_header(self, fp):
1050
line = fp.readline()
1052
# An empty file can actually be treated as though the file doesn't
1054
raise errors.NoSuchFile(self._full_path())
1055
if line != self.HEADER:
1056
raise KnitHeaderError(badline=line,
1057
filename=self._transport.abspath(self._filename))
1060
"""Commit is a nop."""
1063
return '%s(%s)' % (self.__class__.__name__, self._filename)
1066
class _KnitIndex(_KnitComponentFile):
1067
"""Manages knit index file.
1069
The index is already kept in memory and read on startup, to enable
1070
fast lookups of revision information. The cursor of the index
1071
file is always pointing to the end, making it easy to append
1074
_cache is a cache for fast mapping from version id to a Index
1077
_history is a cache for fast mapping from indexes to version ids.
1079
The index data format is dictionary compressed when it comes to
1080
parent references; a index entry may only have parents that with a
1081
lover index number. As a result, the index is topological sorted.
1083
Duplicate entries may be written to the index for a single version id
1084
if this is done then the latter one completely replaces the former:
1085
this allows updates to correct version and parent information.
1086
Note that the two entries may share the delta, and that successive
1087
annotations and references MUST point to the first entry.
1089
The index file on disc contains a header, followed by one line per knit
1090
record. The same revision can be present in an index file more than once.
1091
The first occurrence gets assigned a sequence number starting from 0.
1093
The format of a single line is
1094
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1095
REVISION_ID is a utf8-encoded revision id
1096
FLAGS is a comma separated list of flags about the record. Values include
1097
no-eol, line-delta, fulltext.
1098
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1099
that the the compressed data starts at.
1100
LENGTH is the ascii representation of the length of the data file.
1101
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1103
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1104
revision id already in the knit that is a parent of REVISION_ID.
1105
The ' :' marker is the end of record marker.
1108
when a write is interrupted to the index file, it will result in a line
1109
that does not end in ' :'. If the ' :' is not present at the end of a line,
1110
or at the end of the file, then the record that is missing it will be
1111
ignored by the parser.
1113
When writing new records to the index file, the data is preceded by '\n'
1114
to ensure that records always start on new lines even if the last write was
1115
interrupted. As a result its normal for the last line in the index to be
1116
missing a trailing newline. One can be added with no harmful effects.
1119
HEADER = "# bzr knit index 8\n"
1121
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1122
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1124
def _cache_version(self, version_id, options, pos, size, parents):
1125
"""Cache a version record in the history array and index cache.
1127
This is inlined into _load_data for performance. KEEP IN SYNC.
1128
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1131
# only want the _history index to reference the 1st index entry
1133
if version_id not in self._cache:
1134
index = len(self._history)
1135
self._history.append(version_id)
1137
index = self._cache[version_id][5]
1138
self._cache[version_id] = (version_id,
1145
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1146
create_parent_dir=False, delay_create=False, dir_mode=None):
1147
_KnitComponentFile.__init__(self, transport, filename, mode,
1148
file_mode=file_mode,
1149
create_parent_dir=create_parent_dir,
1152
# position in _history is the 'official' index for a revision
1153
# but the values may have come from a newer entry.
1154
# so - wc -l of a knit index is != the number of unique names
1158
fp = self._transport.get(self._filename)
1160
# _load_data may raise NoSuchFile if the target knit is
1162
_load_data(self, fp)
1166
if mode != 'w' or not create:
1169
self._need_to_create = True
1171
self._transport.put_bytes_non_atomic(
1172
self._filename, self.HEADER, mode=self._file_mode)
1174
def get_graph(self):
1175
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1177
def get_ancestry(self, versions, topo_sorted=True):
1178
"""See VersionedFile.get_ancestry."""
1179
# get a graph of all the mentioned versions:
1181
pending = set(versions)
1184
version = pending.pop()
1187
parents = [p for p in cache[version][4] if p in cache]
1189
raise RevisionNotPresent(version, self._filename)
1190
# if not completed and not a ghost
1191
pending.update([p for p in parents if p not in graph])
1192
graph[version] = parents
1195
return topo_sort(graph.items())
1197
def get_ancestry_with_ghosts(self, versions):
1198
"""See VersionedFile.get_ancestry_with_ghosts."""
1199
# get a graph of all the mentioned versions:
1200
self.check_versions_present(versions)
1203
pending = set(versions)
1205
version = pending.pop()
1207
parents = cache[version][4]
1213
pending.update([p for p in parents if p not in graph])
1214
graph[version] = parents
1215
return topo_sort(graph.items())
1217
def num_versions(self):
1218
return len(self._history)
1220
__len__ = num_versions
1222
def get_versions(self):
1223
return self._history
1225
def idx_to_name(self, idx):
1226
return self._history[idx]
1228
def lookup(self, version_id):
1229
assert version_id in self._cache
1230
return self._cache[version_id][5]
1232
def _version_list_to_index(self, versions):
1235
for version in versions:
1236
if version in cache:
1237
# -- inlined lookup() --
1238
result_list.append(str(cache[version][5]))
1239
# -- end lookup () --
1241
result_list.append('.' + version)
1242
return ' '.join(result_list)
1244
def add_version(self, version_id, options, pos, size, parents):
1245
"""Add a version record to the index."""
1246
self.add_versions(((version_id, options, pos, size, parents),))
1248
def add_versions(self, versions):
1249
"""Add multiple versions to the index.
1251
:param versions: a list of tuples:
1252
(version_id, options, pos, size, parents).
1255
orig_history = self._history[:]
1256
orig_cache = self._cache.copy()
1259
for version_id, options, pos, size, parents in versions:
1260
line = "\n%s %s %s %s %s :" % (version_id,
1264
self._version_list_to_index(parents))
1265
assert isinstance(line, str), \
1266
'content must be utf-8 encoded: %r' % (line,)
1268
self._cache_version(version_id, options, pos, size, parents)
1269
if not self._need_to_create:
1270
self._transport.append_bytes(self._filename, ''.join(lines))
1273
sio.write(self.HEADER)
1274
sio.writelines(lines)
1276
self._transport.put_file_non_atomic(self._filename, sio,
1277
create_parent_dir=self._create_parent_dir,
1278
mode=self._file_mode,
1279
dir_mode=self._dir_mode)
1280
self._need_to_create = False
1282
# If any problems happen, restore the original values and re-raise
1283
self._history = orig_history
1284
self._cache = orig_cache
1287
def has_version(self, version_id):
1288
"""True if the version is in the index."""
1289
return version_id in self._cache
1291
def get_position(self, version_id):
1292
"""Return data position and size of specified version."""
1293
entry = self._cache[version_id]
1294
return entry[2], entry[3]
1296
def get_method(self, version_id):
1297
"""Return compression method of specified version."""
1298
options = self._cache[version_id][1]
1299
if 'fulltext' in options:
1302
if 'line-delta' not in options:
1303
raise errors.KnitIndexUnknownMethod(self._full_path(), 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."""
1321
for version_id in version_ids:
1322
if version_id not in cache:
1323
raise RevisionNotPresent(version_id, self._filename)
1326
class _KnitData(_KnitComponentFile):
1327
"""Contents of the knit data file"""
1329
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1330
create_parent_dir=False, delay_create=False,
1332
_KnitComponentFile.__init__(self, transport, filename, mode,
1333
file_mode=file_mode,
1334
create_parent_dir=create_parent_dir,
1336
self._checked = False
1337
# TODO: jam 20060713 conceptually, this could spill to disk
1338
# if the cached size gets larger than a certain amount
1339
# but it complicates the model a bit, so for now just use
1340
# a simple dictionary
1342
self._do_cache = False
1345
self._need_to_create = create
1347
self._transport.put_bytes_non_atomic(self._filename, '',
1348
mode=self._file_mode)
1350
def enable_cache(self):
1351
"""Enable caching of reads."""
1352
self._do_cache = True
1354
def clear_cache(self):
1355
"""Clear the record cache."""
1356
self._do_cache = False
1359
def _open_file(self):
1361
return self._transport.get(self._filename)
1366
def _record_to_data(self, version_id, digest, lines):
1367
"""Convert version_id, digest, lines into a raw data block.
1369
:return: (len, a StringIO instance with the raw data ready to read.)
1372
data_file = GzipFile(None, mode='wb', fileobj=sio)
1374
assert isinstance(version_id, str)
1375
data_file.writelines(chain(
1376
["version %s %d %s\n" % (version_id,
1380
["end %s\n" % version_id]))
1387
def add_raw_record(self, raw_data):
1388
"""Append a prepared record to the data file.
1390
:return: the offset in the data file raw_data was written.
1392
assert isinstance(raw_data, str), 'data must be plain bytes'
1393
if not self._need_to_create:
1394
return self._transport.append_bytes(self._filename, raw_data)
1396
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1397
create_parent_dir=self._create_parent_dir,
1398
mode=self._file_mode,
1399
dir_mode=self._dir_mode)
1400
self._need_to_create = False
1403
def add_record(self, version_id, digest, lines):
1404
"""Write new text record to disk. Returns the position in the
1405
file where it was written."""
1406
size, sio = self._record_to_data(version_id, digest, lines)
1408
if not self._need_to_create:
1409
start_pos = self._transport.append_file(self._filename, sio)
1411
self._transport.put_file_non_atomic(self._filename, sio,
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
self._cache[version_id] = sio.getvalue()
1419
return start_pos, size
1421
def _parse_record_header(self, version_id, raw_data):
1422
"""Parse a record header for consistency.
1424
:return: the header and the decompressor stream.
1425
as (stream, header_record)
1427
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1429
rec = self._check_header(version_id, df.readline())
1430
except Exception, e:
1431
raise KnitCorrupt(self._filename,
1432
"While reading {%s} got %s(%s)"
1433
% (version_id, e.__class__.__name__, str(e)))
1436
def _check_header(self, version_id, line):
1439
raise KnitCorrupt(self._filename,
1440
'unexpected number of elements in record header')
1441
if rec[1] != version_id:
1442
raise KnitCorrupt(self._filename,
1443
'unexpected version, wanted %r, got %r'
1444
% (version_id, rec[1]))
1447
def _parse_record(self, version_id, data):
1449
# 4168 calls in 2880 217 internal
1450
# 4168 calls to _parse_record_header in 2121
1451
# 4168 calls to readlines in 330
1452
df = GzipFile(mode='rb', fileobj=StringIO(data))
1455
record_contents = df.readlines()
1456
except Exception, e:
1457
raise KnitCorrupt(self._filename,
1458
"While reading {%s} got %s(%s)"
1459
% (version_id, e.__class__.__name__, str(e)))
1460
header = record_contents.pop(0)
1461
rec = self._check_header(version_id, header)
1463
last_line = record_contents.pop()
1464
if len(record_contents) != int(rec[2]):
1465
raise KnitCorrupt(self._filename,
1466
'incorrect number of lines %s != %s'
1468
% (len(record_contents), int(rec[2]),
1470
if last_line != 'end %s\n' % rec[1]:
1471
raise KnitCorrupt(self._filename,
1472
'unexpected version end line %r, wanted %r'
1473
% (last_line, version_id))
1475
return record_contents, rec[3]
1477
def read_records_iter_raw(self, records):
1478
"""Read text records from data file and yield raw data.
1480
This unpacks enough of the text record to validate the id is
1481
as expected but thats all.
1483
# setup an iterator of the external records:
1484
# uses readv so nice and fast we hope.
1486
# grab the disk data needed.
1488
# Don't check _cache if it is empty
1489
needed_offsets = [(pos, size) for version_id, pos, size
1491
if version_id not in self._cache]
1493
needed_offsets = [(pos, size) for version_id, pos, size
1496
raw_records = self._transport.readv(self._filename, needed_offsets)
1498
for version_id, pos, size in records:
1499
if version_id in self._cache:
1500
# This data has already been validated
1501
data = self._cache[version_id]
1503
pos, data = raw_records.next()
1505
self._cache[version_id] = data
1507
# validate the header
1508
df, rec = self._parse_record_header(version_id, data)
1510
yield version_id, data
1512
def read_records_iter(self, records):
1513
"""Read text records from data file and yield result.
1515
The result will be returned in whatever is the fastest to read.
1516
Not by the order requested. Also, multiple requests for the same
1517
record will only yield 1 response.
1518
:param records: A list of (version_id, pos, len) entries
1519
:return: Yields (version_id, contents, digest) in the order
1520
read, not the order requested
1526
# Skip records we have alread seen
1527
yielded_records = set()
1528
needed_records = set()
1529
for record in records:
1530
if record[0] in self._cache:
1531
if record[0] in yielded_records:
1533
yielded_records.add(record[0])
1534
data = self._cache[record[0]]
1535
content, digest = self._parse_record(record[0], data)
1536
yield (record[0], content, digest)
1538
needed_records.add(record)
1539
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1541
needed_records = sorted(set(records), key=operator.itemgetter(1))
1543
if not needed_records:
1546
# The transport optimizes the fetching as well
1547
# (ie, reads continuous ranges.)
1548
readv_response = self._transport.readv(self._filename,
1549
[(pos, size) for version_id, pos, size in needed_records])
1551
for (version_id, pos, size), (pos, data) in \
1552
izip(iter(needed_records), readv_response):
1553
content, digest = self._parse_record(version_id, data)
1555
self._cache[version_id] = data
1556
yield version_id, content, digest
1558
def read_records(self, records):
1559
"""Read records into a dictionary."""
1561
for record_id, content, digest in \
1562
self.read_records_iter(records):
1563
components[record_id] = (content, digest)
1567
class InterKnit(InterVersionedFile):
1568
"""Optimised code paths for knit to knit operations."""
1570
_matching_file_from_factory = KnitVersionedFile
1571
_matching_file_to_factory = KnitVersionedFile
1574
def is_compatible(source, target):
1575
"""Be compatible with knits. """
1577
return (isinstance(source, KnitVersionedFile) and
1578
isinstance(target, KnitVersionedFile))
1579
except AttributeError:
1582
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1583
"""See InterVersionedFile.join."""
1584
assert isinstance(self.source, KnitVersionedFile)
1585
assert isinstance(self.target, KnitVersionedFile)
1587
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1592
pb = ui.ui_factory.nested_progress_bar()
1594
version_ids = list(version_ids)
1595
if None in version_ids:
1596
version_ids.remove(None)
1598
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1599
this_versions = set(self.target._index.get_versions())
1600
needed_versions = self.source_ancestry - this_versions
1601
cross_check_versions = self.source_ancestry.intersection(this_versions)
1602
mismatched_versions = set()
1603
for version in cross_check_versions:
1604
# scan to include needed parents.
1605
n1 = set(self.target.get_parents_with_ghosts(version))
1606
n2 = set(self.source.get_parents_with_ghosts(version))
1608
# FIXME TEST this check for cycles being introduced works
1609
# the logic is we have a cycle if in our graph we are an
1610
# ancestor of any of the n2 revisions.
1616
parent_ancestors = self.source.get_ancestry(parent)
1617
if version in parent_ancestors:
1618
raise errors.GraphCycleError([parent, version])
1619
# ensure this parent will be available later.
1620
new_parents = n2.difference(n1)
1621
needed_versions.update(new_parents.difference(this_versions))
1622
mismatched_versions.add(version)
1624
if not needed_versions and not mismatched_versions:
1626
full_list = topo_sort(self.source.get_graph())
1628
version_list = [i for i in full_list if (not self.target.has_version(i)
1629
and i in needed_versions)]
1633
copy_queue_records = []
1635
for version_id in version_list:
1636
options = self.source._index.get_options(version_id)
1637
parents = self.source._index.get_parents_with_ghosts(version_id)
1638
# check that its will be a consistent copy:
1639
for parent in parents:
1640
# if source has the parent, we must :
1641
# * already have it or
1642
# * have it scheduled already
1643
# otherwise we don't care
1644
assert (self.target.has_version(parent) or
1645
parent in copy_set or
1646
not self.source.has_version(parent))
1647
data_pos, data_size = self.source._index.get_position(version_id)
1648
copy_queue_records.append((version_id, data_pos, data_size))
1649
copy_queue.append((version_id, options, parents))
1650
copy_set.add(version_id)
1652
# data suck the join:
1654
total = len(version_list)
1657
for (version_id, raw_data), \
1658
(version_id2, options, parents) in \
1659
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1661
assert version_id == version_id2, 'logic error, inconsistent results'
1663
pb.update("Joining knit", count, total)
1664
raw_records.append((version_id, options, parents, len(raw_data)))
1665
raw_datum.append(raw_data)
1666
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1668
for version in mismatched_versions:
1669
# FIXME RBC 20060309 is this needed?
1670
n1 = set(self.target.get_parents_with_ghosts(version))
1671
n2 = set(self.source.get_parents_with_ghosts(version))
1672
# write a combined record to our history preserving the current
1673
# parents as first in the list
1674
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1675
self.target.fix_parents(version, new_parents)
1681
InterVersionedFile.register_optimiser(InterKnit)
1684
class WeaveToKnit(InterVersionedFile):
1685
"""Optimised code paths for weave to knit operations."""
1687
_matching_file_from_factory = bzrlib.weave.WeaveFile
1688
_matching_file_to_factory = KnitVersionedFile
1691
def is_compatible(source, target):
1692
"""Be compatible with weaves to knits."""
1694
return (isinstance(source, bzrlib.weave.Weave) and
1695
isinstance(target, KnitVersionedFile))
1696
except AttributeError:
1699
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1700
"""See InterVersionedFile.join."""
1701
assert isinstance(self.source, bzrlib.weave.Weave)
1702
assert isinstance(self.target, KnitVersionedFile)
1704
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1709
pb = ui.ui_factory.nested_progress_bar()
1711
version_ids = list(version_ids)
1713
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1714
this_versions = set(self.target._index.get_versions())
1715
needed_versions = self.source_ancestry - this_versions
1716
cross_check_versions = self.source_ancestry.intersection(this_versions)
1717
mismatched_versions = set()
1718
for version in cross_check_versions:
1719
# scan to include needed parents.
1720
n1 = set(self.target.get_parents_with_ghosts(version))
1721
n2 = set(self.source.get_parents(version))
1722
# if all of n2's parents are in n1, then its fine.
1723
if n2.difference(n1):
1724
# FIXME TEST this check for cycles being introduced works
1725
# the logic is we have a cycle if in our graph we are an
1726
# ancestor of any of the n2 revisions.
1732
parent_ancestors = self.source.get_ancestry(parent)
1733
if version in parent_ancestors:
1734
raise errors.GraphCycleError([parent, version])
1735
# ensure this parent will be available later.
1736
new_parents = n2.difference(n1)
1737
needed_versions.update(new_parents.difference(this_versions))
1738
mismatched_versions.add(version)
1740
if not needed_versions and not mismatched_versions:
1742
full_list = topo_sort(self.source.get_graph())
1744
version_list = [i for i in full_list if (not self.target.has_version(i)
1745
and i in needed_versions)]
1749
total = len(version_list)
1750
for version_id in version_list:
1751
pb.update("Converting to knit", count, total)
1752
parents = self.source.get_parents(version_id)
1753
# check that its will be a consistent copy:
1754
for parent in parents:
1755
# if source has the parent, we must already have it
1756
assert (self.target.has_version(parent))
1757
self.target.add_lines(
1758
version_id, parents, self.source.get_lines(version_id))
1761
for version in mismatched_versions:
1762
# FIXME RBC 20060309 is this needed?
1763
n1 = set(self.target.get_parents_with_ghosts(version))
1764
n2 = set(self.source.get_parents(version))
1765
# write a combined record to our history preserving the current
1766
# parents as first in the list
1767
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1768
self.target.fix_parents(version, new_parents)
1774
InterVersionedFile.register_optimiser(WeaveToKnit)
1777
class KnitSequenceMatcher(difflib.SequenceMatcher):
1778
"""Knit tuned sequence matcher.
1780
This is based on profiling of difflib which indicated some improvements
1781
for our usage pattern.
1784
def find_longest_match(self, alo, ahi, blo, bhi):
1785
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1787
If isjunk is not defined:
1789
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1790
alo <= i <= i+k <= ahi
1791
blo <= j <= j+k <= bhi
1792
and for all (i',j',k') meeting those conditions,
1795
and if i == i', j <= j'
1797
In other words, of all maximal matching blocks, return one that
1798
starts earliest in a, and of all those maximal matching blocks that
1799
start earliest in a, return the one that starts earliest in b.
1801
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1802
>>> s.find_longest_match(0, 5, 0, 9)
1805
If isjunk is defined, first the longest matching block is
1806
determined as above, but with the additional restriction that no
1807
junk element appears in the block. Then that block is extended as
1808
far as possible by matching (only) junk elements on both sides. So
1809
the resulting block never matches on junk except as identical junk
1810
happens to be adjacent to an "interesting" match.
1812
Here's the same example as before, but considering blanks to be
1813
junk. That prevents " abcd" from matching the " abcd" at the tail
1814
end of the second sequence directly. Instead only the "abcd" can
1815
match, and matches the leftmost "abcd" in the second sequence:
1817
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1818
>>> s.find_longest_match(0, 5, 0, 9)
1821
If no blocks match, return (alo, blo, 0).
1823
>>> s = SequenceMatcher(None, "ab", "c")
1824
>>> s.find_longest_match(0, 2, 0, 1)
1828
# CAUTION: stripping common prefix or suffix would be incorrect.
1832
# Longest matching block is "ab", but if common prefix is
1833
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1834
# strip, so ends up claiming that ab is changed to acab by
1835
# inserting "ca" in the middle. That's minimal but unintuitive:
1836
# "it's obvious" that someone inserted "ac" at the front.
1837
# Windiff ends up at the same place as diff, but by pairing up
1838
# the unique 'b's and then matching the first two 'a's.
1840
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1841
besti, bestj, bestsize = alo, blo, 0
1842
# find longest junk-free match
1843
# during an iteration of the loop, j2len[j] = length of longest
1844
# junk-free match ending with a[i-1] and b[j]
1848
for i in xrange(alo, ahi):
1849
# look at all instances of a[i] in b; note that because
1850
# b2j has no junk keys, the loop is skipped if a[i] is junk
1851
j2lenget = j2len.get
1854
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1855
# following improvement
1856
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1857
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1858
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1860
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1861
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1862
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1874
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1876
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1879
# Extend the best by non-junk elements on each end. In particular,
1880
# "popular" non-junk elements aren't in b2j, which greatly speeds
1881
# the inner loop above, but also means "the best" match so far
1882
# doesn't contain any junk *or* popular non-junk elements.
1883
while besti > alo and bestj > blo and \
1884
not isbjunk(b[bestj-1]) and \
1885
a[besti-1] == b[bestj-1]:
1886
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1887
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1888
not isbjunk(b[bestj+bestsize]) and \
1889
a[besti+bestsize] == b[bestj+bestsize]:
1892
# Now that we have a wholly interesting match (albeit possibly
1893
# empty!), we may as well suck up the matching junk on each
1894
# side of it too. Can't think of a good reason not to, and it
1895
# saves post-processing the (possibly considerable) expense of
1896
# figuring out what to do with it. In the case of an empty
1897
# interesting match, this is clearly the right thing to do,
1898
# because no other kind of match is possible in the regions.
1899
while besti > alo and bestj > blo and \
1900
isbjunk(b[bestj-1]) and \
1901
a[besti-1] == b[bestj-1]:
1902
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1903
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1904
isbjunk(b[bestj+bestsize]) and \
1905
a[besti+bestsize] == b[bestj+bestsize]:
1906
bestsize = bestsize + 1
1908
return besti, bestj, bestsize
1912
from bzrlib._knit_load_data_c import _load_data_c as _load_data
1914
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
