1
# Copyright (C) 2005, 2006 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""Knit versionedfile implementation.
19
A knit is a versioned file implementation that supports efficient append only
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
end-marker; simply "end VERSION"
28
delta can be line or full contents.a
29
... the 8's there are the index number of the annotation.
30
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
8 e.set('executable', 'yes')
36
8 if elt.get('executable') == 'yes':
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
43
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
44
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
45
09:33 < lifeless> right
46
09:33 < jrydberg> lifeless: the position and size is the range in the data file
49
so the index sequence is the dictionary compressed sequence number used
50
in the deltas to provide line annotation
55
# 10:16 < lifeless> make partial index writes safe
56
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
57
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
59
# move sha1 out of the content so that join is faster at verifying parents
60
# record content length ?
64
from cStringIO import StringIO
66
from itertools import izip, chain
80
from bzrlib.errors import (
88
RevisionAlreadyPresent,
90
from bzrlib.tuned_gzip import GzipFile
91
from bzrlib.trace import mutter
92
from bzrlib.osutils import (
97
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
98
from bzrlib.tsort import topo_sort
101
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
104
# TODO: Split out code specific to this format into an associated object.
106
# TODO: Can we put in some kind of value to check that the index and data
107
# files belong together?
109
# TODO: accommodate binaries, perhaps by storing a byte count
111
# TODO: function to check whole file
113
# TODO: atomically append data, then measure backwards from the cursor
114
# position after writing to work out where it was located. we may need to
115
# bypass python file buffering.
117
DATA_SUFFIX = '.knit'
118
INDEX_SUFFIX = '.kndx'
121
class KnitContent(object):
122
"""Content of a knit version to which deltas can be applied."""
124
def __init__(self, lines):
127
def annotate_iter(self):
128
"""Yield tuples of (origin, text) for each content line."""
129
return iter(self._lines)
132
"""Return a list of (origin, text) tuples."""
133
return list(self.annotate_iter())
135
def line_delta_iter(self, new_lines):
136
"""Generate line-based delta from this content to new_lines."""
137
new_texts = new_lines.text()
138
old_texts = self.text()
139
s = KnitSequenceMatcher(None, old_texts, new_texts)
140
for tag, i1, i2, j1, j2 in s.get_opcodes():
143
# ofrom, oto, length, data
144
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
146
def line_delta(self, new_lines):
147
return list(self.line_delta_iter(new_lines))
150
return [text for origin, text in self._lines]
153
return KnitContent(self._lines[:])
156
class _KnitFactory(object):
157
"""Base factory for creating content objects."""
159
def make(self, lines, version):
160
num_lines = len(lines)
161
return KnitContent(zip([version] * num_lines, lines))
164
class KnitAnnotateFactory(_KnitFactory):
165
"""Factory for creating annotated Content objects."""
169
def parse_fulltext(self, content, version):
170
"""Convert fulltext to internal representation
172
fulltext content is of the format
173
revid(utf8) plaintext\n
174
internal representation is of the format:
177
lines = [line.split(' ', 1) for line in content]
178
return KnitContent(lines)
180
def parse_line_delta_iter(self, lines):
181
return iter(self.parse_line_delta(lines))
183
def parse_line_delta(self, lines, version):
184
"""Convert a line based delta into internal representation.
186
line delta is in the form of:
187
intstart intend intcount
189
revid(utf8) newline\n
190
internal representation is
191
(start, end, count, [1..count tuples (revid, newline)])
197
# walk through the lines parsing.
199
start, end, count = [int(n) for n in header.split(',')]
200
contents = [next().split(' ', 1) for i in xrange(count)]
201
result.append((start, end, count, contents))
204
def get_fulltext_content(self, lines):
205
"""Extract just the content lines from a fulltext."""
206
return (line.split(' ', 1)[1] for line in lines)
208
def get_linedelta_content(self, lines):
209
"""Extract just the content from a line delta.
211
This doesn't return all of the extra information stored in a delta.
212
Only the actual content lines.
217
header = header.split(',')
218
count = int(header[2])
219
for i in xrange(count):
220
origin, text = next().split(' ', 1)
223
def lower_fulltext(self, content):
224
"""convert a fulltext content record into a serializable form.
226
see parse_fulltext which this inverts.
228
encode_utf8 = cache_utf8.encode
229
return ['%s %s' % (encode_utf8(o), t) for o, t in content._lines]
231
def lower_line_delta(self, delta):
232
"""convert a delta into a serializable form.
234
See parse_line_delta which this inverts.
236
encode_utf8 = cache_utf8.encode
238
for start, end, c, lines in delta:
239
out.append('%d,%d,%d\n' % (start, end, c))
240
out.extend(encode_utf8(origin) + ' ' + text
241
for origin, text in lines)
245
class KnitPlainFactory(_KnitFactory):
246
"""Factory for creating plain Content objects."""
250
def parse_fulltext(self, content, version):
251
"""This parses an unannotated fulltext.
253
Note that this is not a noop - the internal representation
254
has (versionid, line) - its just a constant versionid.
256
return self.make(content, version)
258
def parse_line_delta_iter(self, lines, version):
260
num_lines = len(lines)
261
while cur < num_lines:
264
start, end, c = [int(n) for n in header.split(',')]
265
yield start, end, c, zip([version] * c, lines[cur:cur+c])
268
def parse_line_delta(self, lines, version):
269
return list(self.parse_line_delta_iter(lines, version))
271
def get_fulltext_content(self, lines):
272
"""Extract just the content lines from a fulltext."""
275
def get_linedelta_content(self, lines):
276
"""Extract just the content from a line delta.
278
This doesn't return all of the extra information stored in a delta.
279
Only the actual content lines.
284
header = header.split(',')
285
count = int(header[2])
286
for i in xrange(count):
289
def lower_fulltext(self, content):
290
return content.text()
292
def lower_line_delta(self, delta):
294
for start, end, c, lines in delta:
295
out.append('%d,%d,%d\n' % (start, end, c))
296
out.extend([text for origin, text in lines])
300
def make_empty_knit(transport, relpath):
301
"""Construct a empty knit at the specified location."""
302
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
306
class KnitVersionedFile(VersionedFile):
307
"""Weave-like structure with faster random access.
309
A knit stores a number of texts and a summary of the relationships
310
between them. Texts are identified by a string version-id. Texts
311
are normally stored and retrieved as a series of lines, but can
312
also be passed as single strings.
314
Lines are stored with the trailing newline (if any) included, to
315
avoid special cases for files with no final newline. Lines are
316
composed of 8-bit characters, not unicode. The combination of
317
these approaches should mean any 'binary' file can be safely
318
stored and retrieved.
321
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
322
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
323
create=False, create_parent_dir=False, delay_create=False,
325
"""Construct a knit at location specified by relpath.
327
:param create: If not True, only open an existing knit.
328
:param create_parent_dir: If True, create the parent directory if
329
creating the file fails. (This is used for stores with
330
hash-prefixes that may not exist yet)
331
:param delay_create: The calling code is aware that the knit won't
332
actually be created until the first data is stored.
334
if deprecated_passed(basis_knit):
335
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
336
" deprecated as of bzr 0.9.",
337
DeprecationWarning, stacklevel=2)
338
if access_mode is None:
340
super(KnitVersionedFile, self).__init__(access_mode)
341
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
342
self.transport = transport
343
self.filename = relpath
344
self.factory = factory or KnitAnnotateFactory()
345
self.writable = (access_mode == 'w')
348
self._max_delta_chain = 200
350
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
351
access_mode, create=create, file_mode=file_mode,
352
create_parent_dir=create_parent_dir, delay_create=delay_create,
354
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
355
access_mode, create=create and not len(self), file_mode=file_mode,
356
create_parent_dir=create_parent_dir, delay_create=delay_create,
360
return '%s(%s)' % (self.__class__.__name__,
361
self.transport.abspath(self.filename))
363
def _check_should_delta(self, first_parents):
364
"""Iterate back through the parent listing, looking for a fulltext.
366
This is used when we want to decide whether to add a delta or a new
367
fulltext. It searches for _max_delta_chain parents. When it finds a
368
fulltext parent, it sees if the total size of the deltas leading up to
369
it is large enough to indicate that we want a new full text anyway.
371
Return True if we should create a new delta, False if we should use a
376
delta_parents = first_parents
377
for count in xrange(self._max_delta_chain):
378
parent = delta_parents[0]
379
method = self._index.get_method(parent)
380
pos, size = self._index.get_position(parent)
381
if method == 'fulltext':
385
delta_parents = self._index.get_parents(parent)
387
# We couldn't find a fulltext, so we must create a new one
390
return fulltext_size > delta_size
392
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
393
"""See VersionedFile._add_delta()."""
394
self._check_add(version_id, []) # should we check the lines ?
395
self._check_versions_present(parents)
399
for parent in parents:
400
if not self.has_version(parent):
401
ghosts.append(parent)
403
present_parents.append(parent)
405
if delta_parent is None:
406
# reconstitute as full text.
407
assert len(delta) == 1 or len(delta) == 0
409
assert delta[0][0] == 0
410
assert delta[0][1] == 0, delta[0][1]
411
return super(KnitVersionedFile, self)._add_delta(version_id,
422
options.append('no-eol')
424
if delta_parent is not None:
425
# determine the current delta chain length.
426
# To speed the extract of texts the delta chain is limited
427
# to a fixed number of deltas. This should minimize both
428
# I/O and the time spend applying deltas.
429
# The window was changed to a maximum of 200 deltas, but also added
430
# was a check that the total compressed size of the deltas is
431
# smaller than the compressed size of the fulltext.
432
if not self._check_should_delta([delta_parent]):
433
# We don't want a delta here, just do a normal insertion.
434
return super(KnitVersionedFile, self)._add_delta(version_id,
441
options.append('line-delta')
442
store_lines = self.factory.lower_line_delta(delta)
444
where, size = self._data.add_record(version_id, digest, store_lines)
445
self._index.add_version(version_id, options, where, size, parents)
447
def _add_raw_records(self, records, data):
448
"""Add all the records 'records' with data pre-joined in 'data'.
450
:param records: A list of tuples(version_id, options, parents, size).
451
:param data: The data for the records. When it is written, the records
452
are adjusted to have pos pointing into data by the sum of
453
the preceding records sizes.
456
pos = self._data.add_raw_record(data)
459
for (version_id, options, parents, size) in records:
460
index_entries.append((version_id, options, pos+offset,
462
if self._data._do_cache:
463
self._data._cache[version_id] = data[offset:offset+size]
465
self._index.add_versions(index_entries)
467
def enable_cache(self):
468
"""Start caching data for this knit"""
469
self._data.enable_cache()
471
def clear_cache(self):
472
"""Clear the data cache only."""
473
self._data.clear_cache()
475
def copy_to(self, name, transport):
476
"""See VersionedFile.copy_to()."""
477
# copy the current index to a temp index to avoid racing with local
479
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
480
self.transport.get(self._index._filename))
482
f = self._data._open_file()
484
transport.put_file(name + DATA_SUFFIX, f)
487
# move the copied index into place
488
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
490
def create_empty(self, name, transport, mode=None):
491
return KnitVersionedFile(name, transport, factory=self.factory,
492
delta=self.delta, create=True)
494
def _fix_parents(self, version, new_parents):
495
"""Fix the parents list for version.
497
This is done by appending a new version to the index
498
with identical data except for the parents list.
499
the parents list must be a superset of the current
502
current_values = self._index._cache[version]
503
assert set(current_values[4]).difference(set(new_parents)) == set()
504
self._index.add_version(version,
510
def get_delta(self, version_id):
511
"""Get a delta for constructing version from some other version."""
512
self.check_not_reserved_id(version_id)
513
if not self.has_version(version_id):
514
raise RevisionNotPresent(version_id, self.filename)
516
parents = self.get_parents(version_id)
521
data_pos, data_size = self._index.get_position(version_id)
522
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
523
version_idx = self._index.lookup(version_id)
524
noeol = 'no-eol' in self._index.get_options(version_id)
525
if 'fulltext' == self._index.get_method(version_id):
526
new_content = self.factory.parse_fulltext(data, version_idx)
527
if parent is not None:
528
reference_content = self._get_content(parent)
529
old_texts = reference_content.text()
532
new_texts = new_content.text()
533
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
534
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
536
delta = self.factory.parse_line_delta(data, version_idx)
537
return parent, sha1, noeol, delta
539
def get_graph_with_ghosts(self):
540
"""See VersionedFile.get_graph_with_ghosts()."""
541
graph_items = self._index.get_graph()
542
return dict(graph_items)
544
def get_sha1(self, version_id):
545
"""See VersionedFile.get_sha1()."""
546
record_map = self._get_record_map([version_id])
547
method, content, digest, next = record_map[version_id]
552
"""See VersionedFile.get_suffixes()."""
553
return [DATA_SUFFIX, INDEX_SUFFIX]
555
def has_ghost(self, version_id):
556
"""True if there is a ghost reference in the file to version_id."""
558
if self.has_version(version_id):
560
# optimisable if needed by memoising the _ghosts set.
561
items = self._index.get_graph()
562
for node, parents in items:
563
for parent in parents:
564
if parent not in self._index._cache:
565
if parent == version_id:
570
"""See VersionedFile.versions."""
571
return self._index.get_versions()
573
def has_version(self, version_id):
574
"""See VersionedFile.has_version."""
575
return self._index.has_version(version_id)
577
__contains__ = has_version
579
def _merge_annotations(self, content, parents, parent_texts={},
580
delta=None, annotated=None):
581
"""Merge annotations for content. This is done by comparing
582
the annotations based on changed to the text.
586
for parent_id in parents:
587
merge_content = self._get_content(parent_id, parent_texts)
588
seq = patiencediff.PatienceSequenceMatcher(
589
None, merge_content.text(), content.text())
590
if delta_seq is None:
591
# setup a delta seq to reuse.
593
for i, j, n in seq.get_matching_blocks():
596
# this appears to copy (origin, text) pairs across to the new
597
# content for any line that matches the last-checked parent.
598
# FIXME: save the sequence control data for delta compression
599
# against the most relevant parent rather than rediffing.
600
content._lines[j:j+n] = merge_content._lines[i:i+n]
603
reference_content = self._get_content(parents[0], parent_texts)
604
new_texts = content.text()
605
old_texts = reference_content.text()
606
delta_seq = patiencediff.PatienceSequenceMatcher(
607
None, old_texts, new_texts)
608
return self._make_line_delta(delta_seq, content)
610
def _make_line_delta(self, delta_seq, new_content):
611
"""Generate a line delta from delta_seq and new_content."""
613
for op in delta_seq.get_opcodes():
616
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
619
def _get_components_positions(self, version_ids):
620
"""Produce a map of position data for the components of versions.
622
This data is intended to be used for retrieving the knit records.
624
A dict of version_id to (method, data_pos, data_size, next) is
626
method is the way referenced data should be applied.
627
data_pos is the position of the data in the knit.
628
data_size is the size of the data in the knit.
629
next is the build-parent of the version, or None for fulltexts.
632
for version_id in version_ids:
635
while cursor is not None and cursor not in component_data:
636
method = self._index.get_method(cursor)
637
if method == 'fulltext':
640
next = self.get_parents(cursor)[0]
641
data_pos, data_size = self._index.get_position(cursor)
642
component_data[cursor] = (method, data_pos, data_size, next)
644
return component_data
646
def _get_content(self, version_id, parent_texts={}):
647
"""Returns a content object that makes up the specified
649
if not self.has_version(version_id):
650
raise RevisionNotPresent(version_id, self.filename)
652
cached_version = parent_texts.get(version_id, None)
653
if cached_version is not None:
654
return cached_version
656
text_map, contents_map = self._get_content_maps([version_id])
657
return contents_map[version_id]
659
def _check_versions_present(self, version_ids):
660
"""Check that all specified versions are present."""
661
self._index.check_versions_present(version_ids)
663
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
664
"""See VersionedFile.add_lines_with_ghosts()."""
665
self._check_add(version_id, lines)
666
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
668
def _add_lines(self, version_id, parents, lines, parent_texts):
669
"""See VersionedFile.add_lines."""
670
self._check_add(version_id, lines)
671
self._check_versions_present(parents)
672
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
674
def _check_add(self, version_id, lines):
675
"""check that version_id and lines are safe to add."""
676
assert self.writable, "knit is not opened for write"
677
### FIXME escape. RBC 20060228
678
if contains_whitespace(version_id):
679
raise InvalidRevisionId(version_id, self.filename)
680
self.check_not_reserved_id(version_id)
681
if self.has_version(version_id):
682
raise RevisionAlreadyPresent(version_id, self.filename)
683
self._check_lines_not_unicode(lines)
684
self._check_lines_are_lines(lines)
686
def _add(self, version_id, lines, parents, delta, parent_texts):
687
"""Add a set of lines on top of version specified by parents.
689
If delta is true, compress the text as a line-delta against
692
Any versions not present will be converted into ghosts.
694
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
695
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
696
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
697
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
698
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
699
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
700
# +1383 0 8.0370 8.0370 +<len>
701
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
702
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
703
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
704
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
708
if parent_texts is None:
710
for parent in parents:
711
if not self.has_version(parent):
712
ghosts.append(parent)
714
present_parents.append(parent)
716
if delta and not len(present_parents):
719
digest = sha_strings(lines)
722
if lines[-1][-1] != '\n':
723
options.append('no-eol')
724
lines[-1] = lines[-1] + '\n'
726
if len(present_parents) and delta:
727
# To speed the extract of texts the delta chain is limited
728
# to a fixed number of deltas. This should minimize both
729
# I/O and the time spend applying deltas.
730
delta = self._check_should_delta(present_parents)
732
lines = self.factory.make(lines, version_id)
733
if delta or (self.factory.annotated and len(present_parents) > 0):
734
# Merge annotations from parent texts if so is needed.
735
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
736
delta, self.factory.annotated)
739
options.append('line-delta')
740
store_lines = self.factory.lower_line_delta(delta_hunks)
742
options.append('fulltext')
743
store_lines = self.factory.lower_fulltext(lines)
745
where, size = self._data.add_record(version_id, digest, store_lines)
746
self._index.add_version(version_id, options, where, size, parents)
749
def check(self, progress_bar=None):
750
"""See VersionedFile.check()."""
752
def _clone_text(self, new_version_id, old_version_id, parents):
753
"""See VersionedFile.clone_text()."""
754
# FIXME RBC 20060228 make fast by only inserting an index with null
756
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
758
def get_lines(self, version_id):
759
"""See VersionedFile.get_lines()."""
760
return self.get_line_list([version_id])[0]
762
def _get_record_map(self, version_ids):
763
"""Produce a dictionary of knit records.
765
The keys are version_ids, the values are tuples of (method, content,
767
method is the way the content should be applied.
768
content is a KnitContent object.
769
digest is the SHA1 digest of this version id after all steps are done
770
next is the build-parent of the version, i.e. the leftmost ancestor.
771
If the method is fulltext, next will be None.
773
position_map = self._get_components_positions(version_ids)
774
# c = component_id, m = method, p = position, s = size, n = next
775
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
777
for component_id, content, digest in \
778
self._data.read_records_iter(records):
779
method, position, size, next = position_map[component_id]
780
record_map[component_id] = method, content, digest, next
784
def get_text(self, version_id):
785
"""See VersionedFile.get_text"""
786
return self.get_texts([version_id])[0]
788
def get_texts(self, version_ids):
789
return [''.join(l) for l in self.get_line_list(version_ids)]
791
def get_line_list(self, version_ids):
792
"""Return the texts of listed versions as a list of strings."""
793
for version_id in version_ids:
794
self.check_not_reserved_id(version_id)
795
text_map, content_map = self._get_content_maps(version_ids)
796
return [text_map[v] for v in version_ids]
798
def _get_content_maps(self, version_ids):
799
"""Produce maps of text and KnitContents
801
:return: (text_map, content_map) where text_map contains the texts for
802
the requested versions and content_map contains the KnitContents.
803
Both dicts take version_ids as their keys.
805
for version_id in version_ids:
806
if not self.has_version(version_id):
807
raise RevisionNotPresent(version_id, self.filename)
808
record_map = self._get_record_map(version_ids)
813
for version_id in version_ids:
816
while cursor is not None:
817
method, data, digest, next = record_map[cursor]
818
components.append((cursor, method, data, digest))
819
if cursor in content_map:
824
for component_id, method, data, digest in reversed(components):
825
if component_id in content_map:
826
content = content_map[component_id]
828
version_idx = self._index.lookup(component_id)
829
if method == 'fulltext':
830
assert content is None
831
content = self.factory.parse_fulltext(data, version_idx)
832
elif method == 'line-delta':
833
delta = self.factory.parse_line_delta(data, version_idx)
834
content = content.copy()
835
content._lines = self._apply_delta(content._lines,
837
content_map[component_id] = content
839
if 'no-eol' in self._index.get_options(version_id):
840
content = content.copy()
841
line = content._lines[-1][1].rstrip('\n')
842
content._lines[-1] = (content._lines[-1][0], line)
843
final_content[version_id] = content
845
# digest here is the digest from the last applied component.
846
text = content.text()
847
if sha_strings(text) != digest:
848
raise KnitCorrupt(self.filename,
849
'sha-1 does not match %s' % version_id)
851
text_map[version_id] = text
852
return text_map, final_content
854
def iter_lines_added_or_present_in_versions(self, version_ids=None,
856
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
857
if version_ids is None:
858
version_ids = self.versions()
860
pb = progress.DummyProgress()
861
# we don't care about inclusions, the caller cares.
862
# but we need to setup a list of records to visit.
863
# we need version_id, position, length
864
version_id_records = []
865
requested_versions = set(version_ids)
866
# filter for available versions
867
for version_id in requested_versions:
868
if not self.has_version(version_id):
869
raise RevisionNotPresent(version_id, self.filename)
870
# get a in-component-order queue:
871
for version_id in self.versions():
872
if version_id in requested_versions:
873
data_pos, length = self._index.get_position(version_id)
874
version_id_records.append((version_id, data_pos, length))
876
total = len(version_id_records)
877
for version_idx, (version_id, data, sha_value) in \
878
enumerate(self._data.read_records_iter(version_id_records)):
879
pb.update('Walking content.', version_idx, total)
880
method = self._index.get_method(version_id)
881
version_idx = self._index.lookup(version_id)
883
assert method in ('fulltext', 'line-delta')
884
if method == 'fulltext':
885
line_iterator = self.factory.get_fulltext_content(data)
887
line_iterator = self.factory.get_linedelta_content(data)
888
for line in line_iterator:
891
pb.update('Walking content.', total, total)
893
def num_versions(self):
894
"""See VersionedFile.num_versions()."""
895
return self._index.num_versions()
897
__len__ = num_versions
899
def annotate_iter(self, version_id):
900
"""See VersionedFile.annotate_iter."""
901
content = self._get_content(version_id)
902
for origin, text in content.annotate_iter():
905
def get_parents(self, version_id):
906
"""See VersionedFile.get_parents."""
909
# 52554 calls in 1264 872 internal down from 3674
911
return self._index.get_parents(version_id)
913
raise RevisionNotPresent(version_id, self.filename)
915
def get_parents_with_ghosts(self, version_id):
916
"""See VersionedFile.get_parents."""
918
return self._index.get_parents_with_ghosts(version_id)
920
raise RevisionNotPresent(version_id, self.filename)
922
def get_ancestry(self, versions):
923
"""See VersionedFile.get_ancestry."""
924
if isinstance(versions, basestring):
925
versions = [versions]
928
return self._index.get_ancestry(versions)
930
def get_ancestry_with_ghosts(self, versions):
931
"""See VersionedFile.get_ancestry_with_ghosts."""
932
if isinstance(versions, basestring):
933
versions = [versions]
936
return self._index.get_ancestry_with_ghosts(versions)
938
#@deprecated_method(zero_eight)
939
def walk(self, version_ids):
940
"""See VersionedFile.walk."""
941
# We take the short path here, and extract all relevant texts
942
# and put them in a weave and let that do all the work. Far
943
# from optimal, but is much simpler.
944
# FIXME RB 20060228 this really is inefficient!
945
from bzrlib.weave import Weave
947
w = Weave(self.filename)
948
ancestry = self.get_ancestry(version_ids)
949
sorted_graph = topo_sort(self._index.get_graph())
950
version_list = [vid for vid in sorted_graph if vid in ancestry]
952
for version_id in version_list:
953
lines = self.get_lines(version_id)
954
w.add_lines(version_id, self.get_parents(version_id), lines)
956
for lineno, insert_id, dset, line in w.walk(version_ids):
957
yield lineno, insert_id, dset, line
959
def plan_merge(self, ver_a, ver_b):
960
"""See VersionedFile.plan_merge."""
961
ancestors_b = set(self.get_ancestry(ver_b))
962
def status_a(revision, text):
963
if revision in ancestors_b:
964
return 'killed-b', text
968
ancestors_a = set(self.get_ancestry(ver_a))
969
def status_b(revision, text):
970
if revision in ancestors_a:
971
return 'killed-a', text
975
annotated_a = self.annotate(ver_a)
976
annotated_b = self.annotate(ver_b)
977
plain_a = [t for (a, t) in annotated_a]
978
plain_b = [t for (a, t) in annotated_b]
979
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
982
for ai, bi, l in blocks:
983
# process all mismatched sections
984
# (last mismatched section is handled because blocks always
985
# includes a 0-length last block)
986
for revision, text in annotated_a[a_cur:ai]:
987
yield status_a(revision, text)
988
for revision, text in annotated_b[b_cur:bi]:
989
yield status_b(revision, text)
991
# and now the matched section
994
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
995
assert text_a == text_b
996
yield "unchanged", text_a
999
class _KnitComponentFile(object):
1000
"""One of the files used to implement a knit database"""
1002
def __init__(self, transport, filename, mode, file_mode=None,
1003
create_parent_dir=False, dir_mode=None):
1004
self._transport = transport
1005
self._filename = filename
1007
self._file_mode = file_mode
1008
self._dir_mode = dir_mode
1009
self._create_parent_dir = create_parent_dir
1010
self._need_to_create = False
1012
def _full_path(self):
1013
"""Return the full path to this file."""
1014
return self._transport.base + self._filename
1016
def check_header(self, fp):
1017
line = fp.readline()
1019
# An empty file can actually be treated as though the file doesn't
1021
raise errors.NoSuchFile(self._full_path())
1022
if line != self.HEADER:
1023
raise KnitHeaderError(badline=line,
1024
filename=self._transport.abspath(self._filename))
1027
"""Commit is a nop."""
1030
return '%s(%s)' % (self.__class__.__name__, self._filename)
1033
class _KnitIndex(_KnitComponentFile):
1034
"""Manages knit index file.
1036
The index is already kept in memory and read on startup, to enable
1037
fast lookups of revision information. The cursor of the index
1038
file is always pointing to the end, making it easy to append
1041
_cache is a cache for fast mapping from version id to a Index
1044
_history is a cache for fast mapping from indexes to version ids.
1046
The index data format is dictionary compressed when it comes to
1047
parent references; a index entry may only have parents that with a
1048
lover index number. As a result, the index is topological sorted.
1050
Duplicate entries may be written to the index for a single version id
1051
if this is done then the latter one completely replaces the former:
1052
this allows updates to correct version and parent information.
1053
Note that the two entries may share the delta, and that successive
1054
annotations and references MUST point to the first entry.
1056
The index file on disc contains a header, followed by one line per knit
1057
record. The same revision can be present in an index file more than once.
1058
The first occurrence gets assigned a sequence number starting from 0.
1060
The format of a single line is
1061
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1062
REVISION_ID is a utf8-encoded revision id
1063
FLAGS is a comma separated list of flags about the record. Values include
1064
no-eol, line-delta, fulltext.
1065
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1066
that the the compressed data starts at.
1067
LENGTH is the ascii representation of the length of the data file.
1068
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1070
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1071
revision id already in the knit that is a parent of REVISION_ID.
1072
The ' :' marker is the end of record marker.
1075
when a write is interrupted to the index file, it will result in a line
1076
that does not end in ' :'. If the ' :' is not present at the end of a line,
1077
or at the end of the file, then the record that is missing it will be
1078
ignored by the parser.
1080
When writing new records to the index file, the data is preceded by '\n'
1081
to ensure that records always start on new lines even if the last write was
1082
interrupted. As a result its normal for the last line in the index to be
1083
missing a trailing newline. One can be added with no harmful effects.
1086
HEADER = "# bzr knit index 8\n"
1088
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1089
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1091
def _cache_version(self, version_id, options, pos, size, parents):
1092
"""Cache a version record in the history array and index cache.
1094
This is inlined into _load_data for performance. KEEP IN SYNC.
1095
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1098
# only want the _history index to reference the 1st index entry
1100
if version_id not in self._cache:
1101
index = len(self._history)
1102
self._history.append(version_id)
1104
index = self._cache[version_id][5]
1105
self._cache[version_id] = (version_id,
1112
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1113
create_parent_dir=False, delay_create=False, dir_mode=None):
1114
_KnitComponentFile.__init__(self, transport, filename, mode,
1115
file_mode=file_mode,
1116
create_parent_dir=create_parent_dir,
1119
# position in _history is the 'official' index for a revision
1120
# but the values may have come from a newer entry.
1121
# so - wc -l of a knit index is != the number of unique names
1124
decode_utf8 = cache_utf8.decode
1125
pb = ui.ui_factory.nested_progress_bar()
1127
pb.update('read knit index', 0, 1)
1129
fp = self._transport.get(self._filename)
1131
# _load_data may raise NoSuchFile if the target knit is
1137
if mode != 'w' or not create:
1140
self._need_to_create = True
1142
self._transport.put_bytes_non_atomic(
1143
self._filename, self.HEADER, mode=self._file_mode)
1145
pb.update('read knit index', 1, 1)
1148
def _load_data(self, fp):
1150
history = self._history
1152
self.check_header(fp)
1153
# readlines reads the whole file at once:
1154
# bad for transports like http, good for local disk
1155
# we save 60 ms doing this one change (
1156
# from calling readline each time to calling
1158
# probably what we want for nice behaviour on
1159
# http is a incremental readlines that yields, or
1160
# a check for local vs non local indexes,
1161
history_top = len(history) - 1
1162
for line in fp.readlines():
1164
if len(rec) < 5 or rec[-1] != ':':
1166
# FIXME: in the future we should determine if its a
1167
# short write - and ignore it
1168
# or a different failure, and raise. RBC 20060407
1172
for value in rec[4:-1]:
1174
# uncompressed reference
1175
parents.append(value[1:])
1177
parents.append(history[int(value)])
1179
version_id, options, pos, size = rec[:4]
1181
# See self._cache_version
1182
# only want the _history index to reference the 1st
1183
# index entry for version_id
1184
if version_id not in cache:
1187
history.append(version_id)
1189
index = cache[version_id][5]
1190
cache[version_id] = (version_id,
1196
# end self._cache_version
1198
def get_graph(self):
1199
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1201
def get_ancestry(self, versions):
1202
"""See VersionedFile.get_ancestry."""
1203
# get a graph of all the mentioned versions:
1205
pending = set(versions)
1208
version = pending.pop()
1211
parents = [p for p in cache[version][4] if p in cache]
1213
raise RevisionNotPresent(version, self._filename)
1214
# if not completed and not a ghost
1215
pending.update([p for p in parents if p not in graph])
1216
graph[version] = parents
1217
return topo_sort(graph.items())
1219
def get_ancestry_with_ghosts(self, versions):
1220
"""See VersionedFile.get_ancestry_with_ghosts."""
1221
# get a graph of all the mentioned versions:
1222
self.check_versions_present(versions)
1225
pending = set(versions)
1227
version = pending.pop()
1229
parents = cache[version][4]
1235
pending.update([p for p in parents if p not in graph])
1236
graph[version] = parents
1237
return topo_sort(graph.items())
1239
def num_versions(self):
1240
return len(self._history)
1242
__len__ = num_versions
1244
def get_versions(self):
1245
return self._history
1247
def idx_to_name(self, idx):
1248
return self._history[idx]
1250
def lookup(self, version_id):
1251
assert version_id in self._cache
1252
return self._cache[version_id][5]
1254
def _version_list_to_index(self, versions):
1255
encode_utf8 = cache_utf8.encode
1258
for version in versions:
1259
if version in cache:
1260
# -- inlined lookup() --
1261
result_list.append(str(cache[version][5]))
1262
# -- end lookup () --
1264
result_list.append('.' + encode_utf8(version))
1265
return ' '.join(result_list)
1267
def add_version(self, version_id, options, pos, size, parents):
1268
"""Add a version record to the index."""
1269
self.add_versions(((version_id, options, pos, size, parents),))
1271
def add_versions(self, versions):
1272
"""Add multiple versions to the index.
1274
:param versions: a list of tuples:
1275
(version_id, options, pos, size, parents).
1278
encode_utf8 = cache_utf8.encode
1279
orig_history = self._history[:]
1280
orig_cache = self._cache.copy()
1283
for version_id, options, pos, size, parents in versions:
1284
line = "\n%s %s %s %s %s :" % (encode_utf8(version_id),
1288
self._version_list_to_index(parents))
1289
assert isinstance(line, str), \
1290
'content must be utf-8 encoded: %r' % (line,)
1292
self._cache_version(version_id, options, pos, size, parents)
1293
if not self._need_to_create:
1294
self._transport.append_bytes(self._filename, ''.join(lines))
1297
sio.write(self.HEADER)
1298
sio.writelines(lines)
1300
self._transport.put_file_non_atomic(self._filename, sio,
1301
create_parent_dir=self._create_parent_dir,
1302
mode=self._file_mode,
1303
dir_mode=self._dir_mode)
1304
self._need_to_create = False
1306
# If any problems happen, restore the original values and re-raise
1307
self._history = orig_history
1308
self._cache = orig_cache
1311
def has_version(self, version_id):
1312
"""True if the version is in the index."""
1313
return version_id in self._cache
1315
def get_position(self, version_id):
1316
"""Return data position and size of specified version."""
1317
entry = self._cache[version_id]
1318
return entry[2], entry[3]
1320
def get_method(self, version_id):
1321
"""Return compression method of specified version."""
1322
options = self._cache[version_id][1]
1323
if 'fulltext' in options:
1326
if 'line-delta' not in options:
1327
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1330
def get_options(self, version_id):
1331
return self._cache[version_id][1]
1333
def get_parents(self, version_id):
1334
"""Return parents of specified version ignoring ghosts."""
1335
return [parent for parent in self._cache[version_id][4]
1336
if parent in self._cache]
1338
def get_parents_with_ghosts(self, version_id):
1339
"""Return parents of specified version with ghosts."""
1340
return self._cache[version_id][4]
1342
def check_versions_present(self, version_ids):
1343
"""Check that all specified versions are present."""
1345
for version_id in version_ids:
1346
if version_id not in cache:
1347
raise RevisionNotPresent(version_id, self._filename)
1350
class _KnitData(_KnitComponentFile):
1351
"""Contents of the knit data file"""
1353
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1354
create_parent_dir=False, delay_create=False,
1356
_KnitComponentFile.__init__(self, transport, filename, mode,
1357
file_mode=file_mode,
1358
create_parent_dir=create_parent_dir,
1360
self._checked = False
1361
# TODO: jam 20060713 conceptually, this could spill to disk
1362
# if the cached size gets larger than a certain amount
1363
# but it complicates the model a bit, so for now just use
1364
# a simple dictionary
1366
self._do_cache = False
1369
self._need_to_create = create
1371
self._transport.put_bytes_non_atomic(self._filename, '',
1372
mode=self._file_mode)
1374
def enable_cache(self):
1375
"""Enable caching of reads."""
1376
self._do_cache = True
1378
def clear_cache(self):
1379
"""Clear the record cache."""
1380
self._do_cache = False
1383
def _open_file(self):
1385
return self._transport.get(self._filename)
1390
def _record_to_data(self, version_id, digest, lines):
1391
"""Convert version_id, digest, lines into a raw data block.
1393
:return: (len, a StringIO instance with the raw data ready to read.)
1396
data_file = GzipFile(None, mode='wb', fileobj=sio)
1398
version_id_utf8 = cache_utf8.encode(version_id)
1399
data_file.writelines(chain(
1400
["version %s %d %s\n" % (version_id_utf8,
1404
["end %s\n" % version_id_utf8]))
1411
def add_raw_record(self, raw_data):
1412
"""Append a prepared record to the data file.
1414
:return: the offset in the data file raw_data was written.
1416
assert isinstance(raw_data, str), 'data must be plain bytes'
1417
if not self._need_to_create:
1418
return self._transport.append_bytes(self._filename, raw_data)
1420
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1421
create_parent_dir=self._create_parent_dir,
1422
mode=self._file_mode,
1423
dir_mode=self._dir_mode)
1424
self._need_to_create = False
1427
def add_record(self, version_id, digest, lines):
1428
"""Write new text record to disk. Returns the position in the
1429
file where it was written."""
1430
size, sio = self._record_to_data(version_id, digest, lines)
1432
if not self._need_to_create:
1433
start_pos = self._transport.append_file(self._filename, sio)
1435
self._transport.put_file_non_atomic(self._filename, sio,
1436
create_parent_dir=self._create_parent_dir,
1437
mode=self._file_mode,
1438
dir_mode=self._dir_mode)
1439
self._need_to_create = False
1442
self._cache[version_id] = sio.getvalue()
1443
return start_pos, size
1445
def _parse_record_header(self, version_id, raw_data):
1446
"""Parse a record header for consistency.
1448
:return: the header and the decompressor stream.
1449
as (stream, header_record)
1451
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1452
rec = self._check_header(version_id, df.readline())
1455
def _check_header(self, version_id, line):
1458
raise KnitCorrupt(self._filename,
1459
'unexpected number of elements in record header')
1460
if cache_utf8.decode(rec[1]) != version_id:
1461
raise KnitCorrupt(self._filename,
1462
'unexpected version, wanted %r, got %r'
1463
% (version_id, rec[1]))
1466
def _parse_record(self, version_id, data):
1468
# 4168 calls in 2880 217 internal
1469
# 4168 calls to _parse_record_header in 2121
1470
# 4168 calls to readlines in 330
1471
df = GzipFile(mode='rb', fileobj=StringIO(data))
1473
record_contents = df.readlines()
1474
header = record_contents.pop(0)
1475
rec = self._check_header(version_id, header)
1477
last_line = record_contents.pop()
1478
assert len(record_contents) == int(rec[2])
1479
if last_line != 'end %s\n' % rec[1]:
1480
raise KnitCorrupt(self._filename,
1481
'unexpected version end line %r, wanted %r'
1482
% (last_line, version_id))
1484
return record_contents, rec[3]
1486
def read_records_iter_raw(self, records):
1487
"""Read text records from data file and yield raw data.
1489
This unpacks enough of the text record to validate the id is
1490
as expected but thats all.
1492
# setup an iterator of the external records:
1493
# uses readv so nice and fast we hope.
1495
# grab the disk data needed.
1497
# Don't check _cache if it is empty
1498
needed_offsets = [(pos, size) for version_id, pos, size
1500
if version_id not in self._cache]
1502
needed_offsets = [(pos, size) for version_id, pos, size
1505
raw_records = self._transport.readv(self._filename, needed_offsets)
1507
for version_id, pos, size in records:
1508
if version_id in self._cache:
1509
# This data has already been validated
1510
data = self._cache[version_id]
1512
pos, data = raw_records.next()
1514
self._cache[version_id] = data
1516
# validate the header
1517
df, rec = self._parse_record_header(version_id, data)
1519
yield version_id, data
1521
def read_records_iter(self, records):
1522
"""Read text records from data file and yield result.
1524
The result will be returned in whatever is the fastest to read.
1525
Not by the order requested. Also, multiple requests for the same
1526
record will only yield 1 response.
1527
:param records: A list of (version_id, pos, len) entries
1528
:return: Yields (version_id, contents, digest) in the order
1529
read, not the order requested
1535
# Skip records we have alread seen
1536
yielded_records = set()
1537
needed_records = set()
1538
for record in records:
1539
if record[0] in self._cache:
1540
if record[0] in yielded_records:
1542
yielded_records.add(record[0])
1543
data = self._cache[record[0]]
1544
content, digest = self._parse_record(record[0], data)
1545
yield (record[0], content, digest)
1547
needed_records.add(record)
1548
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1550
needed_records = sorted(set(records), key=operator.itemgetter(1))
1552
if not needed_records:
1555
# The transport optimizes the fetching as well
1556
# (ie, reads continuous ranges.)
1557
readv_response = self._transport.readv(self._filename,
1558
[(pos, size) for version_id, pos, size in needed_records])
1560
for (version_id, pos, size), (pos, data) in \
1561
izip(iter(needed_records), readv_response):
1562
content, digest = self._parse_record(version_id, data)
1564
self._cache[version_id] = data
1565
yield version_id, content, digest
1567
def read_records(self, records):
1568
"""Read records into a dictionary."""
1570
for record_id, content, digest in \
1571
self.read_records_iter(records):
1572
components[record_id] = (content, digest)
1576
class InterKnit(InterVersionedFile):
1577
"""Optimised code paths for knit to knit operations."""
1579
_matching_file_from_factory = KnitVersionedFile
1580
_matching_file_to_factory = KnitVersionedFile
1583
def is_compatible(source, target):
1584
"""Be compatible with knits. """
1586
return (isinstance(source, KnitVersionedFile) and
1587
isinstance(target, KnitVersionedFile))
1588
except AttributeError:
1591
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1592
"""See InterVersionedFile.join."""
1593
assert isinstance(self.source, KnitVersionedFile)
1594
assert isinstance(self.target, KnitVersionedFile)
1596
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1601
pb = ui.ui_factory.nested_progress_bar()
1603
version_ids = list(version_ids)
1604
if None in version_ids:
1605
version_ids.remove(None)
1607
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1608
this_versions = set(self.target._index.get_versions())
1609
needed_versions = self.source_ancestry - this_versions
1610
cross_check_versions = self.source_ancestry.intersection(this_versions)
1611
mismatched_versions = set()
1612
for version in cross_check_versions:
1613
# scan to include needed parents.
1614
n1 = set(self.target.get_parents_with_ghosts(version))
1615
n2 = set(self.source.get_parents_with_ghosts(version))
1617
# FIXME TEST this check for cycles being introduced works
1618
# the logic is we have a cycle if in our graph we are an
1619
# ancestor of any of the n2 revisions.
1625
parent_ancestors = self.source.get_ancestry(parent)
1626
if version in parent_ancestors:
1627
raise errors.GraphCycleError([parent, version])
1628
# ensure this parent will be available later.
1629
new_parents = n2.difference(n1)
1630
needed_versions.update(new_parents.difference(this_versions))
1631
mismatched_versions.add(version)
1633
if not needed_versions and not mismatched_versions:
1635
full_list = topo_sort(self.source.get_graph())
1637
version_list = [i for i in full_list if (not self.target.has_version(i)
1638
and i in needed_versions)]
1642
copy_queue_records = []
1644
for version_id in version_list:
1645
options = self.source._index.get_options(version_id)
1646
parents = self.source._index.get_parents_with_ghosts(version_id)
1647
# check that its will be a consistent copy:
1648
for parent in parents:
1649
# if source has the parent, we must :
1650
# * already have it or
1651
# * have it scheduled already
1652
# otherwise we don't care
1653
assert (self.target.has_version(parent) or
1654
parent in copy_set or
1655
not self.source.has_version(parent))
1656
data_pos, data_size = self.source._index.get_position(version_id)
1657
copy_queue_records.append((version_id, data_pos, data_size))
1658
copy_queue.append((version_id, options, parents))
1659
copy_set.add(version_id)
1661
# data suck the join:
1663
total = len(version_list)
1666
for (version_id, raw_data), \
1667
(version_id2, options, parents) in \
1668
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1670
assert version_id == version_id2, 'logic error, inconsistent results'
1672
pb.update("Joining knit", count, total)
1673
raw_records.append((version_id, options, parents, len(raw_data)))
1674
raw_datum.append(raw_data)
1675
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1677
for version in mismatched_versions:
1678
# FIXME RBC 20060309 is this needed?
1679
n1 = set(self.target.get_parents_with_ghosts(version))
1680
n2 = set(self.source.get_parents_with_ghosts(version))
1681
# write a combined record to our history preserving the current
1682
# parents as first in the list
1683
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1684
self.target.fix_parents(version, new_parents)
1690
InterVersionedFile.register_optimiser(InterKnit)
1693
class WeaveToKnit(InterVersionedFile):
1694
"""Optimised code paths for weave to knit operations."""
1696
_matching_file_from_factory = bzrlib.weave.WeaveFile
1697
_matching_file_to_factory = KnitVersionedFile
1700
def is_compatible(source, target):
1701
"""Be compatible with weaves to knits."""
1703
return (isinstance(source, bzrlib.weave.Weave) and
1704
isinstance(target, KnitVersionedFile))
1705
except AttributeError:
1708
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1709
"""See InterVersionedFile.join."""
1710
assert isinstance(self.source, bzrlib.weave.Weave)
1711
assert isinstance(self.target, KnitVersionedFile)
1713
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1718
pb = ui.ui_factory.nested_progress_bar()
1720
version_ids = list(version_ids)
1722
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1723
this_versions = set(self.target._index.get_versions())
1724
needed_versions = self.source_ancestry - this_versions
1725
cross_check_versions = self.source_ancestry.intersection(this_versions)
1726
mismatched_versions = set()
1727
for version in cross_check_versions:
1728
# scan to include needed parents.
1729
n1 = set(self.target.get_parents_with_ghosts(version))
1730
n2 = set(self.source.get_parents(version))
1731
# if all of n2's parents are in n1, then its fine.
1732
if n2.difference(n1):
1733
# FIXME TEST this check for cycles being introduced works
1734
# the logic is we have a cycle if in our graph we are an
1735
# ancestor of any of the n2 revisions.
1741
parent_ancestors = self.source.get_ancestry(parent)
1742
if version in parent_ancestors:
1743
raise errors.GraphCycleError([parent, version])
1744
# ensure this parent will be available later.
1745
new_parents = n2.difference(n1)
1746
needed_versions.update(new_parents.difference(this_versions))
1747
mismatched_versions.add(version)
1749
if not needed_versions and not mismatched_versions:
1751
full_list = topo_sort(self.source.get_graph())
1753
version_list = [i for i in full_list if (not self.target.has_version(i)
1754
and i in needed_versions)]
1758
total = len(version_list)
1759
for version_id in version_list:
1760
pb.update("Converting to knit", count, total)
1761
parents = self.source.get_parents(version_id)
1762
# check that its will be a consistent copy:
1763
for parent in parents:
1764
# if source has the parent, we must already have it
1765
assert (self.target.has_version(parent))
1766
self.target.add_lines(
1767
version_id, parents, self.source.get_lines(version_id))
1770
for version in mismatched_versions:
1771
# FIXME RBC 20060309 is this needed?
1772
n1 = set(self.target.get_parents_with_ghosts(version))
1773
n2 = set(self.source.get_parents(version))
1774
# write a combined record to our history preserving the current
1775
# parents as first in the list
1776
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1777
self.target.fix_parents(version, new_parents)
1783
InterVersionedFile.register_optimiser(WeaveToKnit)
1786
class KnitSequenceMatcher(difflib.SequenceMatcher):
1787
"""Knit tuned sequence matcher.
1789
This is based on profiling of difflib which indicated some improvements
1790
for our usage pattern.
1793
def find_longest_match(self, alo, ahi, blo, bhi):
1794
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1796
If isjunk is not defined:
1798
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1799
alo <= i <= i+k <= ahi
1800
blo <= j <= j+k <= bhi
1801
and for all (i',j',k') meeting those conditions,
1804
and if i == i', j <= j'
1806
In other words, of all maximal matching blocks, return one that
1807
starts earliest in a, and of all those maximal matching blocks that
1808
start earliest in a, return the one that starts earliest in b.
1810
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1811
>>> s.find_longest_match(0, 5, 0, 9)
1814
If isjunk is defined, first the longest matching block is
1815
determined as above, but with the additional restriction that no
1816
junk element appears in the block. Then that block is extended as
1817
far as possible by matching (only) junk elements on both sides. So
1818
the resulting block never matches on junk except as identical junk
1819
happens to be adjacent to an "interesting" match.
1821
Here's the same example as before, but considering blanks to be
1822
junk. That prevents " abcd" from matching the " abcd" at the tail
1823
end of the second sequence directly. Instead only the "abcd" can
1824
match, and matches the leftmost "abcd" in the second sequence:
1826
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1827
>>> s.find_longest_match(0, 5, 0, 9)
1830
If no blocks match, return (alo, blo, 0).
1832
>>> s = SequenceMatcher(None, "ab", "c")
1833
>>> s.find_longest_match(0, 2, 0, 1)
1837
# CAUTION: stripping common prefix or suffix would be incorrect.
1841
# Longest matching block is "ab", but if common prefix is
1842
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1843
# strip, so ends up claiming that ab is changed to acab by
1844
# inserting "ca" in the middle. That's minimal but unintuitive:
1845
# "it's obvious" that someone inserted "ac" at the front.
1846
# Windiff ends up at the same place as diff, but by pairing up
1847
# the unique 'b's and then matching the first two 'a's.
1849
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1850
besti, bestj, bestsize = alo, blo, 0
1851
# find longest junk-free match
1852
# during an iteration of the loop, j2len[j] = length of longest
1853
# junk-free match ending with a[i-1] and b[j]
1857
for i in xrange(alo, ahi):
1858
# look at all instances of a[i] in b; note that because
1859
# b2j has no junk keys, the loop is skipped if a[i] is junk
1860
j2lenget = j2len.get
1863
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1864
# following improvement
1865
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1866
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1867
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1869
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1870
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1871
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1883
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1885
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1888
# Extend the best by non-junk elements on each end. In particular,
1889
# "popular" non-junk elements aren't in b2j, which greatly speeds
1890
# the inner loop above, but also means "the best" match so far
1891
# doesn't contain any junk *or* popular non-junk elements.
1892
while besti > alo and bestj > blo and \
1893
not isbjunk(b[bestj-1]) and \
1894
a[besti-1] == b[bestj-1]:
1895
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1896
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1897
not isbjunk(b[bestj+bestsize]) and \
1898
a[besti+bestsize] == b[bestj+bestsize]:
1901
# Now that we have a wholly interesting match (albeit possibly
1902
# empty!), we may as well suck up the matching junk on each
1903
# side of it too. Can't think of a good reason not to, and it
1904
# saves post-processing the (possibly considerable) expense of
1905
# figuring out what to do with it. In the case of an empty
1906
# interesting match, this is clearly the right thing to do,
1907
# because no other kind of match is possible in the regions.
1908
while besti > alo and bestj > blo and \
1909
isbjunk(b[bestj-1]) and \
1910
a[besti-1] == b[bestj-1]:
1911
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1912
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1913
isbjunk(b[bestj+bestsize]) and \
1914
a[besti+bestsize] == b[bestj+bestsize]:
1915
bestsize = bestsize + 1
1917
return besti, bestj, bestsize
added
removed
bzrlib/knit.py
