1
# Copyright (C) 2005, 2006 Canonical Ltd
2
# Written by Martin Pool.
3
# Modified by Johan Rydberg <jrydberg@gnu.org>
4
# Modified by Robert Collins <robert.collins@canonical.com>
5
# Modified by Aaron Bentley <aaron.bentley@utoronto.ca>
7
# This program is free software; you can redistribute it and/or modify
8
# it under the terms of the GNU General Public License as published by
9
# the Free Software Foundation; either version 2 of the License, or
10
# (at your option) any later version.
12
# This program is distributed in the hope that it will be useful,
13
# but WITHOUT ANY WARRANTY; without even the implied warranty of
14
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15
# GNU General Public License for more details.
17
# You should have received a copy of the GNU General Public License
18
# along with this program; if not, write to the Free Software
19
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21
"""Knit versionedfile implementation.
23
A knit is a versioned file implementation that supports efficient append only
27
lifeless: the data file is made up of "delta records". each delta record has a delta header
28
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
29
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
30
end-marker; simply "end VERSION"
32
delta can be line or full contents.a
33
... the 8's there are the index number of the annotation.
34
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
38
8 e.set('executable', 'yes')
40
8 if elt.get('executable') == 'yes':
41
8 ie.executable = True
42
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
46
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
47
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
48
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
49
09:33 < lifeless> right
50
09:33 < jrydberg> lifeless: the position and size is the range in the data file
53
so the index sequence is the dictionary compressed sequence number used
54
in the deltas to provide line annotation
59
# 10:16 < lifeless> make partial index writes safe
60
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
61
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
63
# move sha1 out of the content so that join is faster at verifying parents
64
# record content length ?
68
from cStringIO import StringIO
70
from itertools import izip, chain
83
from bzrlib.errors import FileExists, NoSuchFile, KnitError, \
84
InvalidRevisionId, KnitCorrupt, KnitHeaderError, \
85
RevisionNotPresent, RevisionAlreadyPresent
86
from bzrlib.tuned_gzip import GzipFile
87
from bzrlib.trace import mutter
88
from bzrlib.osutils import contains_whitespace, contains_linebreaks, \
90
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
91
from bzrlib.tsort import topo_sort
94
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
97
# TODO: Split out code specific to this format into an associated object.
99
# TODO: Can we put in some kind of value to check that the index and data
100
# files belong together?
102
# TODO: accommodate binaries, perhaps by storing a byte count
104
# TODO: function to check whole file
106
# TODO: atomically append data, then measure backwards from the cursor
107
# position after writing to work out where it was located. we may need to
108
# bypass python file buffering.
110
DATA_SUFFIX = '.knit'
111
INDEX_SUFFIX = '.kndx'
114
class KnitContent(object):
115
"""Content of a knit version to which deltas can be applied."""
117
def __init__(self, lines):
120
def annotate_iter(self):
121
"""Yield tuples of (origin, text) for each content line."""
122
return iter(self._lines)
125
"""Return a list of (origin, text) tuples."""
126
return list(self.annotate_iter())
128
def line_delta_iter(self, new_lines):
129
"""Generate line-based delta from this content to new_lines."""
130
new_texts = new_lines.text()
131
old_texts = self.text()
132
s = KnitSequenceMatcher(None, old_texts, new_texts)
133
for tag, i1, i2, j1, j2 in s.get_opcodes():
136
# ofrom, oto, length, data
137
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
139
def line_delta(self, new_lines):
140
return list(self.line_delta_iter(new_lines))
143
return [text for origin, text in self._lines]
146
return KnitContent(self._lines[:])
149
class _KnitFactory(object):
150
"""Base factory for creating content objects."""
152
def make(self, lines, version):
153
num_lines = len(lines)
154
return KnitContent(zip([version] * num_lines, lines))
157
class KnitAnnotateFactory(_KnitFactory):
158
"""Factory for creating annotated Content objects."""
162
def parse_fulltext(self, content, version):
163
"""Convert fulltext to internal representation
165
fulltext content is of the format
166
revid(utf8) plaintext\n
167
internal representation is of the format:
170
decode_utf8 = cache_utf8.decode
173
origin, text = line.split(' ', 1)
174
lines.append((decode_utf8(origin), text))
175
return KnitContent(lines)
177
def parse_line_delta_iter(self, lines):
178
return iter(self.parse_line_delta(lines))
180
def parse_line_delta(self, lines, version):
181
"""Convert a line based delta into internal representation.
183
line delta is in the form of:
184
intstart intend intcount
186
revid(utf8) newline\n
187
internal representation is
188
(start, end, count, [1..count tuples (revid, newline)])
190
decode_utf8 = cache_utf8.decode
194
# walk through the lines parsing.
196
start, end, count = [int(n) for n in header.split(',')]
200
origin, text = next().split(' ', 1)
202
contents.append((decode_utf8(origin), text))
203
result.append((start, end, count, contents))
206
def get_fulltext_content(self, lines):
207
"""Extract just the content lines from a fulltext."""
208
return (line.split(' ', 1)[1] for line in lines)
210
def get_linedelta_content(self, lines):
211
"""Extract just the content from a line delta.
213
This doesn't return all of the extra information stored in a delta.
214
Only the actual content lines.
219
header = header.split(',')
220
count = int(header[2])
221
for i in xrange(count):
222
origin, text = next().split(' ', 1)
225
def lower_fulltext(self, content):
226
"""convert a fulltext content record into a serializable form.
228
see parse_fulltext which this inverts.
230
encode_utf8 = cache_utf8.encode
231
return ['%s %s' % (encode_utf8(o), t) for o, t in content._lines]
233
def lower_line_delta(self, delta):
234
"""convert a delta into a serializable form.
236
See parse_line_delta which this inverts.
238
encode_utf8 = cache_utf8.encode
240
for start, end, c, lines in delta:
241
out.append('%d,%d,%d\n' % (start, end, c))
242
out.extend(encode_utf8(origin) + ' ' + text
243
for origin, text in lines)
247
class KnitPlainFactory(_KnitFactory):
248
"""Factory for creating plain Content objects."""
252
def parse_fulltext(self, content, version):
253
"""This parses an unannotated fulltext.
255
Note that this is not a noop - the internal representation
256
has (versionid, line) - its just a constant versionid.
258
return self.make(content, version)
260
def parse_line_delta_iter(self, lines, version):
262
num_lines = len(lines)
263
while cur < num_lines:
266
start, end, c = [int(n) for n in header.split(',')]
267
yield start, end, c, zip([version] * c, lines[cur:cur+c])
270
def parse_line_delta(self, lines, version):
271
return list(self.parse_line_delta_iter(lines, version))
273
def get_fulltext_content(self, lines):
274
"""Extract just the content lines from a fulltext."""
277
def get_linedelta_content(self, lines):
278
"""Extract just the content from a line delta.
280
This doesn't return all of the extra information stored in a delta.
281
Only the actual content lines.
286
header = header.split(',')
287
count = int(header[2])
288
for i in xrange(count):
291
def lower_fulltext(self, content):
292
return content.text()
294
def lower_line_delta(self, delta):
296
for start, end, c, lines in delta:
297
out.append('%d,%d,%d\n' % (start, end, c))
298
out.extend([text for origin, text in lines])
302
def make_empty_knit(transport, relpath):
303
"""Construct a empty knit at the specified location."""
304
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
308
class KnitVersionedFile(VersionedFile):
309
"""Weave-like structure with faster random access.
311
A knit stores a number of texts and a summary of the relationships
312
between them. Texts are identified by a string version-id. Texts
313
are normally stored and retrieved as a series of lines, but can
314
also be passed as single strings.
316
Lines are stored with the trailing newline (if any) included, to
317
avoid special cases for files with no final newline. Lines are
318
composed of 8-bit characters, not unicode. The combination of
319
these approaches should mean any 'binary' file can be safely
320
stored and retrieved.
323
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
324
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
325
create=False, create_parent_dir=False, delay_create=False,
327
"""Construct a knit at location specified by relpath.
329
:param create: If not True, only open an existing knit.
330
:param create_parent_dir: If True, create the parent directory if
331
creating the file fails. (This is used for stores with
332
hash-prefixes that may not exist yet)
333
:param delay_create: The calling code is aware that the knit won't
334
actually be created until the first data is stored.
336
if deprecated_passed(basis_knit):
337
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
338
" deprecated as of bzr 0.9.",
339
DeprecationWarning, stacklevel=2)
340
if access_mode is None:
342
super(KnitVersionedFile, self).__init__(access_mode)
343
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
344
self.transport = transport
345
self.filename = relpath
346
self.factory = factory or KnitAnnotateFactory()
347
self.writable = (access_mode == 'w')
350
self._max_delta_chain = 200
352
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
353
access_mode, create=create, file_mode=file_mode,
354
create_parent_dir=create_parent_dir, delay_create=delay_create,
356
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
357
access_mode, create=create and not len(self), file_mode=file_mode,
358
create_parent_dir=create_parent_dir, delay_create=delay_create,
362
return '%s(%s)' % (self.__class__.__name__,
363
self.transport.abspath(self.filename))
365
def _check_should_delta(self, first_parents):
366
"""Iterate back through the parent listing, looking for a fulltext.
368
This is used when we want to decide whether to add a delta or a new
369
fulltext. It searches for _max_delta_chain parents. When it finds a
370
fulltext parent, it sees if the total size of the deltas leading up to
371
it is large enough to indicate that we want a new full text anyway.
373
Return True if we should create a new delta, False if we should use a
378
delta_parents = first_parents
379
for count in xrange(self._max_delta_chain):
380
parent = delta_parents[0]
381
method = self._index.get_method(parent)
382
pos, size = self._index.get_position(parent)
383
if method == 'fulltext':
387
delta_parents = self._index.get_parents(parent)
389
# We couldn't find a fulltext, so we must create a new one
392
return fulltext_size > delta_size
394
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
395
"""See VersionedFile._add_delta()."""
396
self._check_add(version_id, []) # should we check the lines ?
397
self._check_versions_present(parents)
401
for parent in parents:
402
if not self.has_version(parent):
403
ghosts.append(parent)
405
present_parents.append(parent)
407
if delta_parent is None:
408
# reconstitute as full text.
409
assert len(delta) == 1 or len(delta) == 0
411
assert delta[0][0] == 0
412
assert delta[0][1] == 0, delta[0][1]
413
return super(KnitVersionedFile, self)._add_delta(version_id,
424
options.append('no-eol')
426
if delta_parent is not None:
427
# determine the current delta chain length.
428
# To speed the extract of texts the delta chain is limited
429
# to a fixed number of deltas. This should minimize both
430
# I/O and the time spend applying deltas.
431
# The window was changed to a maximum of 200 deltas, but also added
432
# was a check that the total compressed size of the deltas is
433
# smaller than the compressed size of the fulltext.
434
if not self._check_should_delta([delta_parent]):
435
# We don't want a delta here, just do a normal insertion.
436
return super(KnitVersionedFile, self)._add_delta(version_id,
443
options.append('line-delta')
444
store_lines = self.factory.lower_line_delta(delta)
446
where, size = self._data.add_record(version_id, digest, store_lines)
447
self._index.add_version(version_id, options, where, size, parents)
449
def _add_raw_records(self, records, data):
450
"""Add all the records 'records' with data pre-joined in 'data'.
452
:param records: A list of tuples(version_id, options, parents, size).
453
:param data: The data for the records. When it is written, the records
454
are adjusted to have pos pointing into data by the sum of
455
the preceding records sizes.
458
pos = self._data.add_raw_record(data)
461
for (version_id, options, parents, size) in records:
462
index_entries.append((version_id, options, pos+offset,
464
if self._data._do_cache:
465
self._data._cache[version_id] = data[offset:offset+size]
467
self._index.add_versions(index_entries)
469
def enable_cache(self):
470
"""Start caching data for this knit"""
471
self._data.enable_cache()
473
def clear_cache(self):
474
"""Clear the data cache only."""
475
self._data.clear_cache()
477
def copy_to(self, name, transport):
478
"""See VersionedFile.copy_to()."""
479
# copy the current index to a temp index to avoid racing with local
481
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
482
self.transport.get(self._index._filename))
484
f = self._data._open_file()
486
transport.put_file(name + DATA_SUFFIX, f)
489
# move the copied index into place
490
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
492
def create_empty(self, name, transport, mode=None):
493
return KnitVersionedFile(name, transport, factory=self.factory,
494
delta=self.delta, create=True)
496
def _fix_parents(self, version, new_parents):
497
"""Fix the parents list for version.
499
This is done by appending a new version to the index
500
with identical data except for the parents list.
501
the parents list must be a superset of the current
504
current_values = self._index._cache[version]
505
assert set(current_values[4]).difference(set(new_parents)) == set()
506
self._index.add_version(version,
512
def get_delta(self, version_id):
513
"""Get a delta for constructing version from some other version."""
514
if not self.has_version(version_id):
515
raise RevisionNotPresent(version_id, self.filename)
517
parents = self.get_parents(version_id)
522
data_pos, data_size = self._index.get_position(version_id)
523
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
524
version_idx = self._index.lookup(version_id)
525
noeol = 'no-eol' in self._index.get_options(version_id)
526
if 'fulltext' == self._index.get_method(version_id):
527
new_content = self.factory.parse_fulltext(data, version_idx)
528
if parent is not None:
529
reference_content = self._get_content(parent)
530
old_texts = reference_content.text()
533
new_texts = new_content.text()
534
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
535
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
537
delta = self.factory.parse_line_delta(data, version_idx)
538
return parent, sha1, noeol, delta
540
def get_graph_with_ghosts(self):
541
"""See VersionedFile.get_graph_with_ghosts()."""
542
graph_items = self._index.get_graph()
543
return dict(graph_items)
545
def get_sha1(self, version_id):
546
"""See VersionedFile.get_sha1()."""
547
record_map = self._get_record_map([version_id])
548
method, content, digest, next = record_map[version_id]
553
"""See VersionedFile.get_suffixes()."""
554
return [DATA_SUFFIX, INDEX_SUFFIX]
556
def has_ghost(self, version_id):
557
"""True if there is a ghost reference in the file to version_id."""
559
if self.has_version(version_id):
561
# optimisable if needed by memoising the _ghosts set.
562
items = self._index.get_graph()
563
for node, parents in items:
564
for parent in parents:
565
if parent not in self._index._cache:
566
if parent == version_id:
571
"""See VersionedFile.versions."""
572
return self._index.get_versions()
574
def has_version(self, version_id):
575
"""See VersionedFile.has_version."""
576
return self._index.has_version(version_id)
578
__contains__ = has_version
580
def _merge_annotations(self, content, parents, parent_texts={},
581
delta=None, annotated=None):
582
"""Merge annotations for content. This is done by comparing
583
the annotations based on changed to the text.
587
for parent_id in parents:
588
merge_content = self._get_content(parent_id, parent_texts)
589
seq = patiencediff.PatienceSequenceMatcher(
590
None, merge_content.text(), content.text())
591
if delta_seq is None:
592
# setup a delta seq to reuse.
594
for i, j, n in seq.get_matching_blocks():
597
# this appears to copy (origin, text) pairs across to the new
598
# content for any line that matches the last-checked parent.
599
# FIXME: save the sequence control data for delta compression
600
# against the most relevant parent rather than rediffing.
601
content._lines[j:j+n] = merge_content._lines[i:i+n]
604
reference_content = self._get_content(parents[0], parent_texts)
605
new_texts = content.text()
606
old_texts = reference_content.text()
607
delta_seq = patiencediff.PatienceSequenceMatcher(
608
None, old_texts, new_texts)
609
return self._make_line_delta(delta_seq, content)
611
def _make_line_delta(self, delta_seq, new_content):
612
"""Generate a line delta from delta_seq and new_content."""
614
for op in delta_seq.get_opcodes():
617
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
620
def _get_components_positions(self, version_ids):
621
"""Produce a map of position data for the components of versions.
623
This data is intended to be used for retrieving the knit records.
625
A dict of version_id to (method, data_pos, data_size, next) is
627
method is the way referenced data should be applied.
628
data_pos is the position of the data in the knit.
629
data_size is the size of the data in the knit.
630
next is the build-parent of the version, or None for fulltexts.
633
for version_id in version_ids:
636
while cursor is not None and cursor not in component_data:
637
method = self._index.get_method(cursor)
638
if method == 'fulltext':
641
next = self.get_parents(cursor)[0]
642
data_pos, data_size = self._index.get_position(cursor)
643
component_data[cursor] = (method, data_pos, data_size, next)
645
return component_data
647
def _get_content(self, version_id, parent_texts={}):
648
"""Returns a content object that makes up the specified
650
if not self.has_version(version_id):
651
raise RevisionNotPresent(version_id, self.filename)
653
cached_version = parent_texts.get(version_id, None)
654
if cached_version is not None:
655
return cached_version
657
text_map, contents_map = self._get_content_maps([version_id])
658
return contents_map[version_id]
660
def _check_versions_present(self, version_ids):
661
"""Check that all specified versions are present."""
662
version_ids = set(version_ids)
663
for r in list(version_ids):
664
if self._index.has_version(r):
665
version_ids.remove(r)
667
raise RevisionNotPresent(list(version_ids)[0], self.filename)
669
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
670
"""See VersionedFile.add_lines_with_ghosts()."""
671
self._check_add(version_id, lines)
672
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
674
def _add_lines(self, version_id, parents, lines, parent_texts):
675
"""See VersionedFile.add_lines."""
676
self._check_add(version_id, lines)
677
self._check_versions_present(parents)
678
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
680
def _check_add(self, version_id, lines):
681
"""check that version_id and lines are safe to add."""
682
assert self.writable, "knit is not opened for write"
683
### FIXME escape. RBC 20060228
684
if contains_whitespace(version_id):
685
raise InvalidRevisionId(version_id, self.filename)
686
if self.has_version(version_id):
687
raise RevisionAlreadyPresent(version_id, self.filename)
688
self._check_lines_not_unicode(lines)
689
self._check_lines_are_lines(lines)
691
def _add(self, version_id, lines, parents, delta, parent_texts):
692
"""Add a set of lines on top of version specified by parents.
694
If delta is true, compress the text as a line-delta against
697
Any versions not present will be converted into ghosts.
699
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
700
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
701
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
702
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
703
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
704
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
705
# +1383 0 8.0370 8.0370 +<len>
706
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
707
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
708
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
709
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
713
if parent_texts is None:
715
for parent in parents:
716
if not self.has_version(parent):
717
ghosts.append(parent)
719
present_parents.append(parent)
721
if delta and not len(present_parents):
724
digest = sha_strings(lines)
727
if lines[-1][-1] != '\n':
728
options.append('no-eol')
729
lines[-1] = lines[-1] + '\n'
731
if len(present_parents) and delta:
732
# To speed the extract of texts the delta chain is limited
733
# to a fixed number of deltas. This should minimize both
734
# I/O and the time spend applying deltas.
735
delta = self._check_should_delta(present_parents)
737
lines = self.factory.make(lines, version_id)
738
if delta or (self.factory.annotated and len(present_parents) > 0):
739
# Merge annotations from parent texts if so is needed.
740
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
741
delta, self.factory.annotated)
744
options.append('line-delta')
745
store_lines = self.factory.lower_line_delta(delta_hunks)
747
options.append('fulltext')
748
store_lines = self.factory.lower_fulltext(lines)
750
where, size = self._data.add_record(version_id, digest, store_lines)
751
self._index.add_version(version_id, options, where, size, parents)
754
def check(self, progress_bar=None):
755
"""See VersionedFile.check()."""
757
def _clone_text(self, new_version_id, old_version_id, parents):
758
"""See VersionedFile.clone_text()."""
759
# FIXME RBC 20060228 make fast by only inserting an index with null
761
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
763
def get_lines(self, version_id):
764
"""See VersionedFile.get_lines()."""
765
return self.get_line_list([version_id])[0]
767
def _get_record_map(self, version_ids):
768
"""Produce a dictionary of knit records.
770
The keys are version_ids, the values are tuples of (method, content,
772
method is the way the content should be applied.
773
content is a KnitContent object.
774
digest is the SHA1 digest of this version id after all steps are done
775
next is the build-parent of the version, i.e. the leftmost ancestor.
776
If the method is fulltext, next will be None.
778
position_map = self._get_components_positions(version_ids)
779
# c = component_id, m = method, p = position, s = size, n = next
780
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
782
for component_id, content, digest in \
783
self._data.read_records_iter(records):
784
method, position, size, next = position_map[component_id]
785
record_map[component_id] = method, content, digest, next
789
def get_text(self, version_id):
790
"""See VersionedFile.get_text"""
791
return self.get_texts([version_id])[0]
793
def get_texts(self, version_ids):
794
return [''.join(l) for l in self.get_line_list(version_ids)]
796
def get_line_list(self, version_ids):
797
"""Return the texts of listed versions as a list of strings."""
798
text_map, content_map = self._get_content_maps(version_ids)
799
return [text_map[v] for v in version_ids]
801
def _get_content_maps(self, version_ids):
802
"""Produce maps of text and KnitContents
804
:return: (text_map, content_map) where text_map contains the texts for
805
the requested versions and content_map contains the KnitContents.
806
Both dicts take version_ids as their keys.
808
for version_id in version_ids:
809
if not self.has_version(version_id):
810
raise RevisionNotPresent(version_id, self.filename)
811
record_map = self._get_record_map(version_ids)
816
for version_id in version_ids:
819
while cursor is not None:
820
method, data, digest, next = record_map[cursor]
821
components.append((cursor, method, data, digest))
822
if cursor in content_map:
827
for component_id, method, data, digest in reversed(components):
828
if component_id in content_map:
829
content = content_map[component_id]
831
version_idx = self._index.lookup(component_id)
832
if method == 'fulltext':
833
assert content is None
834
content = self.factory.parse_fulltext(data, version_idx)
835
elif method == 'line-delta':
836
delta = self.factory.parse_line_delta(data, version_idx)
837
content = content.copy()
838
content._lines = self._apply_delta(content._lines,
840
content_map[component_id] = content
842
if 'no-eol' in self._index.get_options(version_id):
843
content = content.copy()
844
line = content._lines[-1][1].rstrip('\n')
845
content._lines[-1] = (content._lines[-1][0], line)
846
final_content[version_id] = content
848
# digest here is the digest from the last applied component.
849
text = content.text()
850
if sha_strings(text) != digest:
851
raise KnitCorrupt(self.filename,
852
'sha-1 does not match %s' % version_id)
854
text_map[version_id] = text
855
return text_map, final_content
857
def iter_lines_added_or_present_in_versions(self, version_ids=None,
859
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
860
if version_ids is None:
861
version_ids = self.versions()
863
pb = progress.DummyProgress()
864
# we don't care about inclusions, the caller cares.
865
# but we need to setup a list of records to visit.
866
# we need version_id, position, length
867
version_id_records = []
868
requested_versions = set(version_ids)
869
# filter for available versions
870
for version_id in requested_versions:
871
if not self.has_version(version_id):
872
raise RevisionNotPresent(version_id, self.filename)
873
# get a in-component-order queue:
874
for version_id in self.versions():
875
if version_id in requested_versions:
876
data_pos, length = self._index.get_position(version_id)
877
version_id_records.append((version_id, data_pos, length))
879
total = len(version_id_records)
880
for version_idx, (version_id, data, sha_value) in \
881
enumerate(self._data.read_records_iter(version_id_records)):
882
pb.update('Walking content.', version_idx, total)
883
method = self._index.get_method(version_id)
884
version_idx = self._index.lookup(version_id)
886
assert method in ('fulltext', 'line-delta')
887
if method == 'fulltext':
888
line_iterator = self.factory.get_fulltext_content(data)
890
line_iterator = self.factory.get_linedelta_content(data)
891
for line in line_iterator:
894
pb.update('Walking content.', total, total)
896
def num_versions(self):
897
"""See VersionedFile.num_versions()."""
898
return self._index.num_versions()
900
__len__ = num_versions
902
def annotate_iter(self, version_id):
903
"""See VersionedFile.annotate_iter."""
904
content = self._get_content(version_id)
905
for origin, text in content.annotate_iter():
908
def get_parents(self, version_id):
909
"""See VersionedFile.get_parents."""
912
# 52554 calls in 1264 872 internal down from 3674
914
return self._index.get_parents(version_id)
916
raise RevisionNotPresent(version_id, self.filename)
918
def get_parents_with_ghosts(self, version_id):
919
"""See VersionedFile.get_parents."""
921
return self._index.get_parents_with_ghosts(version_id)
923
raise RevisionNotPresent(version_id, self.filename)
925
def get_ancestry(self, versions):
926
"""See VersionedFile.get_ancestry."""
927
if isinstance(versions, basestring):
928
versions = [versions]
931
self._check_versions_present(versions)
932
return self._index.get_ancestry(versions)
934
def get_ancestry_with_ghosts(self, versions):
935
"""See VersionedFile.get_ancestry_with_ghosts."""
936
if isinstance(versions, basestring):
937
versions = [versions]
940
self._check_versions_present(versions)
941
return self._index.get_ancestry_with_ghosts(versions)
943
#@deprecated_method(zero_eight)
944
def walk(self, version_ids):
945
"""See VersionedFile.walk."""
946
# We take the short path here, and extract all relevant texts
947
# and put them in a weave and let that do all the work. Far
948
# from optimal, but is much simpler.
949
# FIXME RB 20060228 this really is inefficient!
950
from bzrlib.weave import Weave
952
w = Weave(self.filename)
953
ancestry = self.get_ancestry(version_ids)
954
sorted_graph = topo_sort(self._index.get_graph())
955
version_list = [vid for vid in sorted_graph if vid in ancestry]
957
for version_id in version_list:
958
lines = self.get_lines(version_id)
959
w.add_lines(version_id, self.get_parents(version_id), lines)
961
for lineno, insert_id, dset, line in w.walk(version_ids):
962
yield lineno, insert_id, dset, line
964
def plan_merge(self, ver_a, ver_b):
965
"""See VersionedFile.plan_merge."""
966
ancestors_b = set(self.get_ancestry(ver_b))
967
def status_a(revision, text):
968
if revision in ancestors_b:
969
return 'killed-b', text
973
ancestors_a = set(self.get_ancestry(ver_a))
974
def status_b(revision, text):
975
if revision in ancestors_a:
976
return 'killed-a', text
980
annotated_a = self.annotate(ver_a)
981
annotated_b = self.annotate(ver_b)
982
plain_a = [t for (a, t) in annotated_a]
983
plain_b = [t for (a, t) in annotated_b]
984
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
987
for ai, bi, l in blocks:
988
# process all mismatched sections
989
# (last mismatched section is handled because blocks always
990
# includes a 0-length last block)
991
for revision, text in annotated_a[a_cur:ai]:
992
yield status_a(revision, text)
993
for revision, text in annotated_b[b_cur:bi]:
994
yield status_b(revision, text)
996
# and now the matched section
999
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
1000
assert text_a == text_b
1001
yield "unchanged", text_a
1004
class _KnitComponentFile(object):
1005
"""One of the files used to implement a knit database"""
1007
def __init__(self, transport, filename, mode, file_mode=None,
1008
create_parent_dir=False, dir_mode=None):
1009
self._transport = transport
1010
self._filename = filename
1012
self._file_mode = file_mode
1013
self._dir_mode = dir_mode
1014
self._create_parent_dir = create_parent_dir
1015
self._need_to_create = False
1017
def check_header(self, fp):
1018
line = fp.readline()
1020
# An empty file can actually be treated as though the file doesn't
1022
raise errors.NoSuchFile(self._transport.base + self._filename)
1023
if line != self.HEADER:
1024
raise KnitHeaderError(badline=line,
1025
filename=self._transport.abspath(self._filename))
1028
"""Commit is a nop."""
1031
return '%s(%s)' % (self.__class__.__name__, self._filename)
1034
class _KnitIndex(_KnitComponentFile):
1035
"""Manages knit index file.
1037
The index is already kept in memory and read on startup, to enable
1038
fast lookups of revision information. The cursor of the index
1039
file is always pointing to the end, making it easy to append
1042
_cache is a cache for fast mapping from version id to a Index
1045
_history is a cache for fast mapping from indexes to version ids.
1047
The index data format is dictionary compressed when it comes to
1048
parent references; a index entry may only have parents that with a
1049
lover index number. As a result, the index is topological sorted.
1051
Duplicate entries may be written to the index for a single version id
1052
if this is done then the latter one completely replaces the former:
1053
this allows updates to correct version and parent information.
1054
Note that the two entries may share the delta, and that successive
1055
annotations and references MUST point to the first entry.
1057
The index file on disc contains a header, followed by one line per knit
1058
record. The same revision can be present in an index file more than once.
1059
The first occurrence gets assigned a sequence number starting from 0.
1061
The format of a single line is
1062
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1063
REVISION_ID is a utf8-encoded revision id
1064
FLAGS is a comma separated list of flags about the record. Values include
1065
no-eol, line-delta, fulltext.
1066
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1067
that the the compressed data starts at.
1068
LENGTH is the ascii representation of the length of the data file.
1069
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1071
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1072
revision id already in the knit that is a parent of REVISION_ID.
1073
The ' :' marker is the end of record marker.
1076
when a write is interrupted to the index file, it will result in a line that
1077
does not end in ' :'. If the ' :' is not present at the end of a line, or at
1078
the end of the file, then the record that is missing it will be ignored by
1081
When writing new records to the index file, the data is preceded by '\n'
1082
to ensure that records always start on new lines even if the last write was
1083
interrupted. As a result its normal for the last line in the index to be
1084
missing a trailing newline. One can be added with no harmful effects.
1087
HEADER = "# bzr knit index 8\n"
1089
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1090
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1092
def _cache_version(self, version_id, options, pos, size, parents):
1093
"""Cache a version record in the history array and index cache.
1095
This is inlined into __init__ for performance. KEEP IN SYNC.
1096
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1099
# only want the _history index to reference the 1st index entry
1101
if version_id not in self._cache:
1102
index = len(self._history)
1103
self._history.append(version_id)
1105
index = self._cache[version_id][5]
1106
self._cache[version_id] = (version_id,
1113
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1114
create_parent_dir=False, delay_create=False, dir_mode=None):
1115
_KnitComponentFile.__init__(self, transport, filename, mode,
1116
file_mode=file_mode,
1117
create_parent_dir=create_parent_dir,
1120
# position in _history is the 'official' index for a revision
1121
# but the values may have come from a newer entry.
1122
# so - wc -l of a knit index is != the number of unique names
1125
decode_utf8 = cache_utf8.decode
1126
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1131
pb.update('read knit index', count, total)
1132
fp = self._transport.get(self._filename)
1134
self.check_header(fp)
1135
# readlines reads the whole file at once:
1136
# bad for transports like http, good for local disk
1137
# we save 60 ms doing this one change (
1138
# from calling readline each time to calling
1140
# probably what we want for nice behaviour on
1141
# http is a incremental readlines that yields, or
1142
# a check for local vs non local indexes,
1143
for l in fp.readlines():
1145
if len(rec) < 5 or rec[-1] != ':':
1147
# FIXME: in the future we should determine if its a
1148
# short write - and ignore it
1149
# or a different failure, and raise. RBC 20060407
1153
#pb.update('read knit index', count, total)
1154
# See self._parse_parents
1156
for value in rec[4:-1]:
1158
# uncompressed reference
1159
parents.append(decode_utf8(value[1:]))
1161
# this is 15/4000ms faster than isinstance,
1163
# this function is called thousands of times a
1164
# second so small variations add up.
1165
assert value.__class__ is str
1166
parents.append(self._history[int(value)])
1167
# end self._parse_parents
1168
# self._cache_version(decode_utf8(rec[0]),
1169
# rec[1].split(','),
1173
# --- self._cache_version
1174
# only want the _history index to reference the 1st
1175
# index entry for version_id
1176
version_id = decode_utf8(rec[0])
1177
if version_id not in self._cache:
1178
index = len(self._history)
1179
self._history.append(version_id)
1181
index = self._cache[version_id][5]
1182
self._cache[version_id] = (version_id,
1188
# --- self._cache_version
1191
except NoSuchFile, e:
1192
if mode != 'w' or not create:
1195
self._need_to_create = True
1197
self._transport.put_bytes_non_atomic(self._filename,
1198
self.HEADER, mode=self._file_mode)
1201
pb.update('read knit index', total, total)
1204
def _parse_parents(self, compressed_parents):
1205
"""convert a list of string parent values into version ids.
1207
ints are looked up in the index.
1208
.FOO values are ghosts and converted in to FOO.
1210
NOTE: the function is retained here for clarity, and for possible
1211
use in partial index reads. However bulk processing now has
1212
it inlined in __init__ for inner-loop optimisation.
1215
for value in compressed_parents:
1216
if value[-1] == '.':
1217
# uncompressed reference
1218
result.append(cache_utf8.decode_utf8(value[1:]))
1220
# this is 15/4000ms faster than isinstance,
1221
# this function is called thousands of times a
1222
# second so small variations add up.
1223
assert value.__class__ is str
1224
result.append(self._history[int(value)])
1227
def get_graph(self):
1229
for version_id, index in self._cache.iteritems():
1230
graph.append((version_id, index[4]))
1233
def get_ancestry(self, versions):
1234
"""See VersionedFile.get_ancestry."""
1235
# get a graph of all the mentioned versions:
1237
pending = set(versions)
1239
version = pending.pop()
1240
parents = self._cache[version][4]
1241
# got the parents ok
1243
parents = [parent for parent in parents if parent in self._cache]
1244
for parent in parents:
1245
# if not completed and not a ghost
1246
if parent not in graph:
1248
graph[version] = parents
1249
return topo_sort(graph.items())
1251
def get_ancestry_with_ghosts(self, versions):
1252
"""See VersionedFile.get_ancestry_with_ghosts."""
1253
# get a graph of all the mentioned versions:
1255
pending = set(versions)
1257
version = pending.pop()
1259
parents = self._cache[version][4]
1265
# got the parents ok
1266
for parent in parents:
1267
if parent not in graph:
1269
graph[version] = parents
1270
return topo_sort(graph.items())
1272
def num_versions(self):
1273
return len(self._history)
1275
__len__ = num_versions
1277
def get_versions(self):
1278
return self._history
1280
def idx_to_name(self, idx):
1281
return self._history[idx]
1283
def lookup(self, version_id):
1284
assert version_id in self._cache
1285
return self._cache[version_id][5]
1287
def _version_list_to_index(self, versions):
1288
encode_utf8 = cache_utf8.encode
1290
for version in versions:
1291
if version in self._cache:
1292
# -- inlined lookup() --
1293
result_list.append(str(self._cache[version][5]))
1294
# -- end lookup () --
1296
result_list.append('.' + encode_utf8(version))
1297
return ' '.join(result_list)
1299
def add_version(self, version_id, options, pos, size, parents):
1300
"""Add a version record to the index."""
1301
self.add_versions(((version_id, options, pos, size, parents),))
1303
def add_versions(self, versions):
1304
"""Add multiple versions to the index.
1306
:param versions: a list of tuples:
1307
(version_id, options, pos, size, parents).
1310
encode_utf8 = cache_utf8.encode
1311
orig_history = self._history[:]
1312
orig_cache = self._cache.copy()
1315
for version_id, options, pos, size, parents in versions:
1316
line = "\n%s %s %s %s %s :" % (encode_utf8(version_id),
1320
self._version_list_to_index(parents))
1321
assert isinstance(line, str), \
1322
'content must be utf-8 encoded: %r' % (line,)
1324
self._cache_version(version_id, options, pos, size, parents)
1325
if not self._need_to_create:
1326
self._transport.append_bytes(self._filename, ''.join(lines))
1329
sio.write(self.HEADER)
1330
sio.writelines(lines)
1332
self._transport.put_file_non_atomic(self._filename, sio,
1333
create_parent_dir=self._create_parent_dir,
1334
mode=self._file_mode,
1335
dir_mode=self._dir_mode)
1336
self._need_to_create = False
1338
# If any problems happen, restore the original values and re-raise
1339
self._history = orig_history
1340
self._cache = orig_cache
1343
def has_version(self, version_id):
1344
"""True if the version is in the index."""
1345
return (version_id in self._cache)
1347
def get_position(self, version_id):
1348
"""Return data position and size of specified version."""
1349
return (self._cache[version_id][2], \
1350
self._cache[version_id][3])
1352
def get_method(self, version_id):
1353
"""Return compression method of specified version."""
1354
options = self._cache[version_id][1]
1355
if 'fulltext' in options:
1358
assert 'line-delta' in options
1361
def get_options(self, version_id):
1362
return self._cache[version_id][1]
1364
def get_parents(self, version_id):
1365
"""Return parents of specified version ignoring ghosts."""
1366
return [parent for parent in self._cache[version_id][4]
1367
if parent in self._cache]
1369
def get_parents_with_ghosts(self, version_id):
1370
"""Return parents of specified version with ghosts."""
1371
return self._cache[version_id][4]
1373
def check_versions_present(self, version_ids):
1374
"""Check that all specified versions are present."""
1375
version_ids = set(version_ids)
1376
for version_id in list(version_ids):
1377
if version_id in self._cache:
1378
version_ids.remove(version_id)
1380
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1383
class _KnitData(_KnitComponentFile):
1384
"""Contents of the knit data file"""
1386
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1387
create_parent_dir=False, delay_create=False,
1389
_KnitComponentFile.__init__(self, transport, filename, mode,
1390
file_mode=file_mode,
1391
create_parent_dir=create_parent_dir,
1393
self._checked = False
1394
# TODO: jam 20060713 conceptually, this could spill to disk
1395
# if the cached size gets larger than a certain amount
1396
# but it complicates the model a bit, so for now just use
1397
# a simple dictionary
1399
self._do_cache = False
1402
self._need_to_create = create
1404
self._transport.put_bytes_non_atomic(self._filename, '',
1405
mode=self._file_mode)
1407
def enable_cache(self):
1408
"""Enable caching of reads."""
1409
self._do_cache = True
1411
def clear_cache(self):
1412
"""Clear the record cache."""
1413
self._do_cache = False
1416
def _open_file(self):
1418
return self._transport.get(self._filename)
1423
def _record_to_data(self, version_id, digest, lines):
1424
"""Convert version_id, digest, lines into a raw data block.
1426
:return: (len, a StringIO instance with the raw data ready to read.)
1429
data_file = GzipFile(None, mode='wb', fileobj=sio)
1431
version_id_utf8 = cache_utf8.encode(version_id)
1432
data_file.writelines(chain(
1433
["version %s %d %s\n" % (version_id_utf8,
1437
["end %s\n" % version_id_utf8]))
1444
def add_raw_record(self, raw_data):
1445
"""Append a prepared record to the data file.
1447
:return: the offset in the data file raw_data was written.
1449
assert isinstance(raw_data, str), 'data must be plain bytes'
1450
if not self._need_to_create:
1451
return self._transport.append_bytes(self._filename, raw_data)
1453
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1454
create_parent_dir=self._create_parent_dir,
1455
mode=self._file_mode,
1456
dir_mode=self._dir_mode)
1457
self._need_to_create = False
1460
def add_record(self, version_id, digest, lines):
1461
"""Write new text record to disk. Returns the position in the
1462
file where it was written."""
1463
size, sio = self._record_to_data(version_id, digest, lines)
1465
if not self._need_to_create:
1466
start_pos = self._transport.append_file(self._filename, sio)
1468
self._transport.put_file_non_atomic(self._filename, sio,
1469
create_parent_dir=self._create_parent_dir,
1470
mode=self._file_mode,
1471
dir_mode=self._dir_mode)
1472
self._need_to_create = False
1475
self._cache[version_id] = sio.getvalue()
1476
return start_pos, size
1478
def _parse_record_header(self, version_id, raw_data):
1479
"""Parse a record header for consistency.
1481
:return: the header and the decompressor stream.
1482
as (stream, header_record)
1484
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1485
rec = df.readline().split()
1487
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
1488
if cache_utf8.decode(rec[1]) != version_id:
1489
raise KnitCorrupt(self._filename,
1490
'unexpected version, wanted %r, got %r' % (
1491
version_id, rec[1]))
1494
def _parse_record(self, version_id, data):
1496
# 4168 calls in 2880 217 internal
1497
# 4168 calls to _parse_record_header in 2121
1498
# 4168 calls to readlines in 330
1499
df, rec = self._parse_record_header(version_id, data)
1500
record_contents = df.readlines()
1501
l = record_contents.pop()
1502
assert len(record_contents) == int(rec[2])
1503
if l != 'end %s\n' % cache_utf8.encode(version_id):
1504
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
1507
return record_contents, rec[3]
1509
def read_records_iter_raw(self, records):
1510
"""Read text records from data file and yield raw data.
1512
This unpacks enough of the text record to validate the id is
1513
as expected but thats all.
1515
# setup an iterator of the external records:
1516
# uses readv so nice and fast we hope.
1518
# grab the disk data needed.
1520
# Don't check _cache if it is empty
1521
needed_offsets = [(pos, size) for version_id, pos, size
1523
if version_id not in self._cache]
1525
needed_offsets = [(pos, size) for version_id, pos, size
1528
raw_records = self._transport.readv(self._filename, needed_offsets)
1531
for version_id, pos, size in records:
1532
if version_id in self._cache:
1533
# This data has already been validated
1534
data = self._cache[version_id]
1536
pos, data = raw_records.next()
1538
self._cache[version_id] = data
1540
# validate the header
1541
df, rec = self._parse_record_header(version_id, data)
1543
yield version_id, data
1545
def read_records_iter(self, records):
1546
"""Read text records from data file and yield result.
1548
The result will be returned in whatever is the fastest to read.
1549
Not by the order requested. Also, multiple requests for the same
1550
record will only yield 1 response.
1551
:param records: A list of (version_id, pos, len) entries
1552
:return: Yields (version_id, contents, digest) in the order
1553
read, not the order requested
1559
# Skip records we have alread seen
1560
yielded_records = set()
1561
needed_records = set()
1562
for record in records:
1563
if record[0] in self._cache:
1564
if record[0] in yielded_records:
1566
yielded_records.add(record[0])
1567
data = self._cache[record[0]]
1568
content, digest = self._parse_record(record[0], data)
1569
yield (record[0], content, digest)
1571
needed_records.add(record)
1572
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1574
needed_records = sorted(set(records), key=operator.itemgetter(1))
1576
if not needed_records:
1579
# The transport optimizes the fetching as well
1580
# (ie, reads continuous ranges.)
1581
readv_response = self._transport.readv(self._filename,
1582
[(pos, size) for version_id, pos, size in needed_records])
1584
for (version_id, pos, size), (pos, data) in \
1585
izip(iter(needed_records), readv_response):
1586
content, digest = self._parse_record(version_id, data)
1588
self._cache[version_id] = data
1589
yield version_id, content, digest
1591
def read_records(self, records):
1592
"""Read records into a dictionary."""
1594
for record_id, content, digest in \
1595
self.read_records_iter(records):
1596
components[record_id] = (content, digest)
1600
class InterKnit(InterVersionedFile):
1601
"""Optimised code paths for knit to knit operations."""
1603
_matching_file_from_factory = KnitVersionedFile
1604
_matching_file_to_factory = KnitVersionedFile
1607
def is_compatible(source, target):
1608
"""Be compatible with knits. """
1610
return (isinstance(source, KnitVersionedFile) and
1611
isinstance(target, KnitVersionedFile))
1612
except AttributeError:
1615
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1616
"""See InterVersionedFile.join."""
1617
assert isinstance(self.source, KnitVersionedFile)
1618
assert isinstance(self.target, KnitVersionedFile)
1620
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1625
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1627
version_ids = list(version_ids)
1628
if None in version_ids:
1629
version_ids.remove(None)
1631
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1632
this_versions = set(self.target._index.get_versions())
1633
needed_versions = self.source_ancestry - this_versions
1634
cross_check_versions = self.source_ancestry.intersection(this_versions)
1635
mismatched_versions = set()
1636
for version in cross_check_versions:
1637
# scan to include needed parents.
1638
n1 = set(self.target.get_parents_with_ghosts(version))
1639
n2 = set(self.source.get_parents_with_ghosts(version))
1641
# FIXME TEST this check for cycles being introduced works
1642
# the logic is we have a cycle if in our graph we are an
1643
# ancestor of any of the n2 revisions.
1649
parent_ancestors = self.source.get_ancestry(parent)
1650
if version in parent_ancestors:
1651
raise errors.GraphCycleError([parent, version])
1652
# ensure this parent will be available later.
1653
new_parents = n2.difference(n1)
1654
needed_versions.update(new_parents.difference(this_versions))
1655
mismatched_versions.add(version)
1657
if not needed_versions and not mismatched_versions:
1659
full_list = topo_sort(self.source.get_graph())
1661
version_list = [i for i in full_list if (not self.target.has_version(i)
1662
and i in needed_versions)]
1666
copy_queue_records = []
1668
for version_id in version_list:
1669
options = self.source._index.get_options(version_id)
1670
parents = self.source._index.get_parents_with_ghosts(version_id)
1671
# check that its will be a consistent copy:
1672
for parent in parents:
1673
# if source has the parent, we must :
1674
# * already have it or
1675
# * have it scheduled already
1676
# otherwise we don't care
1677
assert (self.target.has_version(parent) or
1678
parent in copy_set or
1679
not self.source.has_version(parent))
1680
data_pos, data_size = self.source._index.get_position(version_id)
1681
copy_queue_records.append((version_id, data_pos, data_size))
1682
copy_queue.append((version_id, options, parents))
1683
copy_set.add(version_id)
1685
# data suck the join:
1687
total = len(version_list)
1690
for (version_id, raw_data), \
1691
(version_id2, options, parents) in \
1692
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1694
assert version_id == version_id2, 'logic error, inconsistent results'
1696
pb.update("Joining knit", count, total)
1697
raw_records.append((version_id, options, parents, len(raw_data)))
1698
raw_datum.append(raw_data)
1699
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1701
for version in mismatched_versions:
1702
# FIXME RBC 20060309 is this needed?
1703
n1 = set(self.target.get_parents_with_ghosts(version))
1704
n2 = set(self.source.get_parents_with_ghosts(version))
1705
# write a combined record to our history preserving the current
1706
# parents as first in the list
1707
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1708
self.target.fix_parents(version, new_parents)
1714
InterVersionedFile.register_optimiser(InterKnit)
1717
class WeaveToKnit(InterVersionedFile):
1718
"""Optimised code paths for weave to knit operations."""
1720
_matching_file_from_factory = bzrlib.weave.WeaveFile
1721
_matching_file_to_factory = KnitVersionedFile
1724
def is_compatible(source, target):
1725
"""Be compatible with weaves to knits."""
1727
return (isinstance(source, bzrlib.weave.Weave) and
1728
isinstance(target, KnitVersionedFile))
1729
except AttributeError:
1732
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1733
"""See InterVersionedFile.join."""
1734
assert isinstance(self.source, bzrlib.weave.Weave)
1735
assert isinstance(self.target, KnitVersionedFile)
1737
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1742
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1744
version_ids = list(version_ids)
1746
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1747
this_versions = set(self.target._index.get_versions())
1748
needed_versions = self.source_ancestry - this_versions
1749
cross_check_versions = self.source_ancestry.intersection(this_versions)
1750
mismatched_versions = set()
1751
for version in cross_check_versions:
1752
# scan to include needed parents.
1753
n1 = set(self.target.get_parents_with_ghosts(version))
1754
n2 = set(self.source.get_parents(version))
1755
# if all of n2's parents are in n1, then its fine.
1756
if n2.difference(n1):
1757
# FIXME TEST this check for cycles being introduced works
1758
# the logic is we have a cycle if in our graph we are an
1759
# ancestor of any of the n2 revisions.
1765
parent_ancestors = self.source.get_ancestry(parent)
1766
if version in parent_ancestors:
1767
raise errors.GraphCycleError([parent, version])
1768
# ensure this parent will be available later.
1769
new_parents = n2.difference(n1)
1770
needed_versions.update(new_parents.difference(this_versions))
1771
mismatched_versions.add(version)
1773
if not needed_versions and not mismatched_versions:
1775
full_list = topo_sort(self.source.get_graph())
1777
version_list = [i for i in full_list if (not self.target.has_version(i)
1778
and i in needed_versions)]
1782
total = len(version_list)
1783
for version_id in version_list:
1784
pb.update("Converting to knit", count, total)
1785
parents = self.source.get_parents(version_id)
1786
# check that its will be a consistent copy:
1787
for parent in parents:
1788
# if source has the parent, we must already have it
1789
assert (self.target.has_version(parent))
1790
self.target.add_lines(
1791
version_id, parents, self.source.get_lines(version_id))
1794
for version in mismatched_versions:
1795
# FIXME RBC 20060309 is this needed?
1796
n1 = set(self.target.get_parents_with_ghosts(version))
1797
n2 = set(self.source.get_parents(version))
1798
# write a combined record to our history preserving the current
1799
# parents as first in the list
1800
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1801
self.target.fix_parents(version, new_parents)
1807
InterVersionedFile.register_optimiser(WeaveToKnit)
1810
class KnitSequenceMatcher(difflib.SequenceMatcher):
1811
"""Knit tuned sequence matcher.
1813
This is based on profiling of difflib which indicated some improvements
1814
for our usage pattern.
1817
def find_longest_match(self, alo, ahi, blo, bhi):
1818
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1820
If isjunk is not defined:
1822
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1823
alo <= i <= i+k <= ahi
1824
blo <= j <= j+k <= bhi
1825
and for all (i',j',k') meeting those conditions,
1828
and if i == i', j <= j'
1830
In other words, of all maximal matching blocks, return one that
1831
starts earliest in a, and of all those maximal matching blocks that
1832
start earliest in a, return the one that starts earliest in b.
1834
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1835
>>> s.find_longest_match(0, 5, 0, 9)
1838
If isjunk is defined, first the longest matching block is
1839
determined as above, but with the additional restriction that no
1840
junk element appears in the block. Then that block is extended as
1841
far as possible by matching (only) junk elements on both sides. So
1842
the resulting block never matches on junk except as identical junk
1843
happens to be adjacent to an "interesting" match.
1845
Here's the same example as before, but considering blanks to be
1846
junk. That prevents " abcd" from matching the " abcd" at the tail
1847
end of the second sequence directly. Instead only the "abcd" can
1848
match, and matches the leftmost "abcd" in the second sequence:
1850
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1851
>>> s.find_longest_match(0, 5, 0, 9)
1854
If no blocks match, return (alo, blo, 0).
1856
>>> s = SequenceMatcher(None, "ab", "c")
1857
>>> s.find_longest_match(0, 2, 0, 1)
1861
# CAUTION: stripping common prefix or suffix would be incorrect.
1865
# Longest matching block is "ab", but if common prefix is
1866
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1867
# strip, so ends up claiming that ab is changed to acab by
1868
# inserting "ca" in the middle. That's minimal but unintuitive:
1869
# "it's obvious" that someone inserted "ac" at the front.
1870
# Windiff ends up at the same place as diff, but by pairing up
1871
# the unique 'b's and then matching the first two 'a's.
1873
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1874
besti, bestj, bestsize = alo, blo, 0
1875
# find longest junk-free match
1876
# during an iteration of the loop, j2len[j] = length of longest
1877
# junk-free match ending with a[i-1] and b[j]
1881
for i in xrange(alo, ahi):
1882
# look at all instances of a[i] in b; note that because
1883
# b2j has no junk keys, the loop is skipped if a[i] is junk
1884
j2lenget = j2len.get
1887
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1888
# following improvement
1889
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1890
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1891
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1893
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1894
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1895
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1907
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1909
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1912
# Extend the best by non-junk elements on each end. In particular,
1913
# "popular" non-junk elements aren't in b2j, which greatly speeds
1914
# the inner loop above, but also means "the best" match so far
1915
# doesn't contain any junk *or* popular non-junk elements.
1916
while besti > alo and bestj > blo and \
1917
not isbjunk(b[bestj-1]) and \
1918
a[besti-1] == b[bestj-1]:
1919
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1920
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1921
not isbjunk(b[bestj+bestsize]) and \
1922
a[besti+bestsize] == b[bestj+bestsize]:
1925
# Now that we have a wholly interesting match (albeit possibly
1926
# empty!), we may as well suck up the matching junk on each
1927
# side of it too. Can't think of a good reason not to, and it
1928
# saves post-processing the (possibly considerable) expense of
1929
# figuring out what to do with it. In the case of an empty
1930
# interesting match, this is clearly the right thing to do,
1931
# because no other kind of match is possible in the regions.
1932
while besti > alo and bestj > blo and \
1933
isbjunk(b[bestj-1]) and \
1934
a[besti-1] == b[bestj-1]:
1935
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1936
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1937
isbjunk(b[bestj+bestsize]) and \
1938
a[besti+bestsize] == b[bestj+bestsize]:
1939
bestsize = bestsize + 1
1941
return besti, bestj, bestsize
added
removed
bzrlib/knit.py
