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
for result_item in self.parse_line_delta[lines]:
181
def parse_line_delta(self, lines, version):
182
"""Convert a line based delta into internal representation.
184
line delta is in the form of:
185
intstart intend intcount
187
revid(utf8) newline\n
188
internal representation is
189
(start, end, count, [1..count tuples (revid, newline)])
191
decode_utf8 = cache_utf8.decode
195
# walk through the lines parsing.
197
start, end, count = [int(n) for n in header.split(',')]
201
origin, text = next().split(' ', 1)
203
contents.append((decode_utf8(origin), text))
204
result.append((start, end, count, contents))
207
def lower_fulltext(self, content):
208
"""convert a fulltext content record into a serializable form.
210
see parse_fulltext which this inverts.
212
encode_utf8 = cache_utf8.encode
213
return ['%s %s' % (encode_utf8(o), t) for o, t in content._lines]
215
def lower_line_delta(self, delta):
216
"""convert a delta into a serializable form.
218
See parse_line_delta which this inverts.
220
encode_utf8 = cache_utf8.encode
222
for start, end, c, lines in delta:
223
out.append('%d,%d,%d\n' % (start, end, c))
224
out.extend(encode_utf8(origin) + ' ' + text
225
for origin, text in lines)
229
class KnitPlainFactory(_KnitFactory):
230
"""Factory for creating plain Content objects."""
234
def parse_fulltext(self, content, version):
235
"""This parses an unannotated fulltext.
237
Note that this is not a noop - the internal representation
238
has (versionid, line) - its just a constant versionid.
240
return self.make(content, version)
242
def parse_line_delta_iter(self, lines, version):
244
header = lines.pop(0)
245
start, end, c = [int(n) for n in header.split(',')]
246
yield start, end, c, zip([version] * c, lines[:c])
249
def parse_line_delta(self, lines, version):
250
return list(self.parse_line_delta_iter(lines, version))
252
def lower_fulltext(self, content):
253
return content.text()
255
def lower_line_delta(self, delta):
257
for start, end, c, lines in delta:
258
out.append('%d,%d,%d\n' % (start, end, c))
259
out.extend([text for origin, text in lines])
263
def make_empty_knit(transport, relpath):
264
"""Construct a empty knit at the specified location."""
265
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
269
class KnitVersionedFile(VersionedFile):
270
"""Weave-like structure with faster random access.
272
A knit stores a number of texts and a summary of the relationships
273
between them. Texts are identified by a string version-id. Texts
274
are normally stored and retrieved as a series of lines, but can
275
also be passed as single strings.
277
Lines are stored with the trailing newline (if any) included, to
278
avoid special cases for files with no final newline. Lines are
279
composed of 8-bit characters, not unicode. The combination of
280
these approaches should mean any 'binary' file can be safely
281
stored and retrieved.
284
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
285
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
286
create=False, create_parent_dir=False, delay_create=False,
288
"""Construct a knit at location specified by relpath.
290
:param create: If not True, only open an existing knit.
291
:param create_parent_dir: If True, create the parent directory if
292
creating the file fails. (This is used for stores with
293
hash-prefixes that may not exist yet)
294
:param delay_create: The calling code is aware that the knit won't
295
actually be created until the first data is stored.
297
if deprecated_passed(basis_knit):
298
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
299
" deprecated as of bzr 0.9.",
300
DeprecationWarning, stacklevel=2)
301
if access_mode is None:
303
super(KnitVersionedFile, self).__init__(access_mode)
304
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
305
self.transport = transport
306
self.filename = relpath
307
self.factory = factory or KnitAnnotateFactory()
308
self.writable = (access_mode == 'w')
311
self._max_delta_chain = 200
313
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
314
access_mode, create=create, file_mode=file_mode,
315
create_parent_dir=create_parent_dir, delay_create=delay_create,
317
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
318
access_mode, create=create and not len(self), file_mode=file_mode,
319
create_parent_dir=create_parent_dir, delay_create=delay_create,
323
return '%s(%s)' % (self.__class__.__name__,
324
self.transport.abspath(self.filename))
326
def _check_should_delta(self, first_parents):
327
"""Iterate back through the parent listing, looking for a fulltext.
329
This is used when we want to decide whether to add a delta or a new
330
fulltext. It searches for _max_delta_chain parents. When it finds a
331
fulltext parent, it sees if the total size of the deltas leading up to
332
it is large enough to indicate that we want a new full text anyway.
334
Return True if we should create a new delta, False if we should use a
339
delta_parents = first_parents
340
for count in xrange(self._max_delta_chain):
341
parent = delta_parents[0]
342
method = self._index.get_method(parent)
343
pos, size = self._index.get_position(parent)
344
if method == 'fulltext':
348
delta_parents = self._index.get_parents(parent)
350
# We couldn't find a fulltext, so we must create a new one
353
return fulltext_size > delta_size
355
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
356
"""See VersionedFile._add_delta()."""
357
self._check_add(version_id, []) # should we check the lines ?
358
self._check_versions_present(parents)
362
for parent in parents:
363
if not self.has_version(parent):
364
ghosts.append(parent)
366
present_parents.append(parent)
368
if delta_parent is None:
369
# reconstitute as full text.
370
assert len(delta) == 1 or len(delta) == 0
372
assert delta[0][0] == 0
373
assert delta[0][1] == 0, delta[0][1]
374
return super(KnitVersionedFile, self)._add_delta(version_id,
385
options.append('no-eol')
387
if delta_parent is not None:
388
# determine the current delta chain length.
389
# To speed the extract of texts the delta chain is limited
390
# to a fixed number of deltas. This should minimize both
391
# I/O and the time spend applying deltas.
392
# The window was changed to a maximum of 200 deltas, but also added
393
# was a check that the total compressed size of the deltas is
394
# smaller than the compressed size of the fulltext.
395
if not self._check_should_delta([delta_parent]):
396
# We don't want a delta here, just do a normal insertion.
397
return super(KnitVersionedFile, self)._add_delta(version_id,
404
options.append('line-delta')
405
store_lines = self.factory.lower_line_delta(delta)
407
where, size = self._data.add_record(version_id, digest, store_lines)
408
self._index.add_version(version_id, options, where, size, parents)
410
def _add_raw_records(self, records, data):
411
"""Add all the records 'records' with data pre-joined in 'data'.
413
:param records: A list of tuples(version_id, options, parents, size).
414
:param data: The data for the records. When it is written, the records
415
are adjusted to have pos pointing into data by the sum of
416
the preceding records sizes.
419
pos = self._data.add_raw_record(data)
422
for (version_id, options, parents, size) in records:
423
index_entries.append((version_id, options, pos+offset,
425
if self._data._do_cache:
426
self._data._cache[version_id] = data[offset:offset+size]
428
self._index.add_versions(index_entries)
430
def enable_cache(self):
431
"""Start caching data for this knit"""
432
self._data.enable_cache()
434
def clear_cache(self):
435
"""Clear the data cache only."""
436
self._data.clear_cache()
438
def copy_to(self, name, transport):
439
"""See VersionedFile.copy_to()."""
440
# copy the current index to a temp index to avoid racing with local
442
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
443
self.transport.get(self._index._filename))
445
f = self._data._open_file()
447
transport.put_file(name + DATA_SUFFIX, f)
450
# move the copied index into place
451
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
453
def create_empty(self, name, transport, mode=None):
454
return KnitVersionedFile(name, transport, factory=self.factory,
455
delta=self.delta, create=True)
457
def _fix_parents(self, version, new_parents):
458
"""Fix the parents list for version.
460
This is done by appending a new version to the index
461
with identical data except for the parents list.
462
the parents list must be a superset of the current
465
current_values = self._index._cache[version]
466
assert set(current_values[4]).difference(set(new_parents)) == set()
467
self._index.add_version(version,
473
def get_delta(self, version_id):
474
"""Get a delta for constructing version from some other version."""
475
if not self.has_version(version_id):
476
raise RevisionNotPresent(version_id, self.filename)
478
parents = self.get_parents(version_id)
483
data_pos, data_size = self._index.get_position(version_id)
484
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
485
version_idx = self._index.lookup(version_id)
486
noeol = 'no-eol' in self._index.get_options(version_id)
487
if 'fulltext' == self._index.get_method(version_id):
488
new_content = self.factory.parse_fulltext(data, version_idx)
489
if parent is not None:
490
reference_content = self._get_content(parent)
491
old_texts = reference_content.text()
494
new_texts = new_content.text()
495
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
496
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
498
delta = self.factory.parse_line_delta(data, version_idx)
499
return parent, sha1, noeol, delta
501
def get_graph_with_ghosts(self):
502
"""See VersionedFile.get_graph_with_ghosts()."""
503
graph_items = self._index.get_graph()
504
return dict(graph_items)
506
def get_sha1(self, version_id):
507
"""See VersionedFile.get_sha1()."""
508
record_map = self._get_record_map([version_id])
509
method, content, digest, next = record_map[version_id]
514
"""See VersionedFile.get_suffixes()."""
515
return [DATA_SUFFIX, INDEX_SUFFIX]
517
def has_ghost(self, version_id):
518
"""True if there is a ghost reference in the file to version_id."""
520
if self.has_version(version_id):
522
# optimisable if needed by memoising the _ghosts set.
523
items = self._index.get_graph()
524
for node, parents in items:
525
for parent in parents:
526
if parent not in self._index._cache:
527
if parent == version_id:
532
"""See VersionedFile.versions."""
533
return self._index.get_versions()
535
def has_version(self, version_id):
536
"""See VersionedFile.has_version."""
537
return self._index.has_version(version_id)
539
__contains__ = has_version
541
def _merge_annotations(self, content, parents, parent_texts={},
542
delta=None, annotated=None):
543
"""Merge annotations for content. This is done by comparing
544
the annotations based on changed to the text.
548
for parent_id in parents:
549
merge_content = self._get_content(parent_id, parent_texts)
550
seq = patiencediff.PatienceSequenceMatcher(
551
None, merge_content.text(), content.text())
552
if delta_seq is None:
553
# setup a delta seq to reuse.
555
for i, j, n in seq.get_matching_blocks():
558
# this appears to copy (origin, text) pairs across to the new
559
# content for any line that matches the last-checked parent.
560
# FIXME: save the sequence control data for delta compression
561
# against the most relevant parent rather than rediffing.
562
content._lines[j:j+n] = merge_content._lines[i:i+n]
565
reference_content = self._get_content(parents[0], parent_texts)
566
new_texts = content.text()
567
old_texts = reference_content.text()
568
delta_seq = patiencediff.PatienceSequenceMatcher(
569
None, old_texts, new_texts)
570
return self._make_line_delta(delta_seq, content)
572
def _make_line_delta(self, delta_seq, new_content):
573
"""Generate a line delta from delta_seq and new_content."""
575
for op in delta_seq.get_opcodes():
578
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
581
def _get_components_positions(self, version_ids):
582
"""Produce a map of position data for the components of versions.
584
This data is intended to be used for retrieving the knit records.
586
A dict of version_id to (method, data_pos, data_size, next) is
588
method is the way referenced data should be applied.
589
data_pos is the position of the data in the knit.
590
data_size is the size of the data in the knit.
591
next is the build-parent of the version, or None for fulltexts.
594
for version_id in version_ids:
597
while cursor is not None and cursor not in component_data:
598
method = self._index.get_method(cursor)
599
if method == 'fulltext':
602
next = self.get_parents(cursor)[0]
603
data_pos, data_size = self._index.get_position(cursor)
604
component_data[cursor] = (method, data_pos, data_size, next)
606
return component_data
608
def _get_content(self, version_id, parent_texts={}):
609
"""Returns a content object that makes up the specified
611
if not self.has_version(version_id):
612
raise RevisionNotPresent(version_id, self.filename)
614
cached_version = parent_texts.get(version_id, None)
615
if cached_version is not None:
616
return cached_version
618
text_map, contents_map = self._get_content_maps([version_id])
619
return contents_map[version_id]
621
def _check_versions_present(self, version_ids):
622
"""Check that all specified versions are present."""
623
version_ids = set(version_ids)
624
for r in list(version_ids):
625
if self._index.has_version(r):
626
version_ids.remove(r)
628
raise RevisionNotPresent(list(version_ids)[0], self.filename)
630
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
631
"""See VersionedFile.add_lines_with_ghosts()."""
632
self._check_add(version_id, lines)
633
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
635
def _add_lines(self, version_id, parents, lines, parent_texts):
636
"""See VersionedFile.add_lines."""
637
self._check_add(version_id, lines)
638
self._check_versions_present(parents)
639
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
641
def _check_add(self, version_id, lines):
642
"""check that version_id and lines are safe to add."""
643
assert self.writable, "knit is not opened for write"
644
### FIXME escape. RBC 20060228
645
if contains_whitespace(version_id):
646
raise InvalidRevisionId(version_id, self.filename)
647
if self.has_version(version_id):
648
raise RevisionAlreadyPresent(version_id, self.filename)
649
self._check_lines_not_unicode(lines)
650
self._check_lines_are_lines(lines)
652
def _add(self, version_id, lines, parents, delta, parent_texts):
653
"""Add a set of lines on top of version specified by parents.
655
If delta is true, compress the text as a line-delta against
658
Any versions not present will be converted into ghosts.
660
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
661
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
662
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
663
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
664
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
665
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
666
# +1383 0 8.0370 8.0370 +<len>
667
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
668
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
669
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
670
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
674
if parent_texts is None:
676
for parent in parents:
677
if not self.has_version(parent):
678
ghosts.append(parent)
680
present_parents.append(parent)
682
if delta and not len(present_parents):
685
digest = sha_strings(lines)
688
if lines[-1][-1] != '\n':
689
options.append('no-eol')
690
lines[-1] = lines[-1] + '\n'
692
if len(present_parents) and delta:
693
# To speed the extract of texts the delta chain is limited
694
# to a fixed number of deltas. This should minimize both
695
# I/O and the time spend applying deltas.
696
delta = self._check_should_delta(present_parents)
698
lines = self.factory.make(lines, version_id)
699
if delta or (self.factory.annotated and len(present_parents) > 0):
700
# Merge annotations from parent texts if so is needed.
701
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
702
delta, self.factory.annotated)
705
options.append('line-delta')
706
store_lines = self.factory.lower_line_delta(delta_hunks)
708
options.append('fulltext')
709
store_lines = self.factory.lower_fulltext(lines)
711
where, size = self._data.add_record(version_id, digest, store_lines)
712
self._index.add_version(version_id, options, where, size, parents)
715
def check(self, progress_bar=None):
716
"""See VersionedFile.check()."""
718
def _clone_text(self, new_version_id, old_version_id, parents):
719
"""See VersionedFile.clone_text()."""
720
# FIXME RBC 20060228 make fast by only inserting an index with null
722
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
724
def get_lines(self, version_id):
725
"""See VersionedFile.get_lines()."""
726
return self.get_line_list([version_id])[0]
728
def _get_record_map(self, version_ids):
729
"""Produce a dictionary of knit records.
731
The keys are version_ids, the values are tuples of (method, content,
733
method is the way the content should be applied.
734
content is a KnitContent object.
735
digest is the SHA1 digest of this version id after all steps are done
736
next is the build-parent of the version, i.e. the leftmost ancestor.
737
If the method is fulltext, next will be None.
739
position_map = self._get_components_positions(version_ids)
740
# c = component_id, m = method, p = position, s = size, n = next
741
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
743
for component_id, content, digest in \
744
self._data.read_records_iter(records):
745
method, position, size, next = position_map[component_id]
746
record_map[component_id] = method, content, digest, next
750
def get_text(self, version_id):
751
"""See VersionedFile.get_text"""
752
return self.get_texts([version_id])[0]
754
def get_texts(self, version_ids):
755
return [''.join(l) for l in self.get_line_list(version_ids)]
757
def get_line_list(self, version_ids):
758
"""Return the texts of listed versions as a list of strings."""
759
text_map, content_map = self._get_content_maps(version_ids)
760
return [text_map[v] for v in version_ids]
762
def _get_content_maps(self, version_ids):
763
"""Produce maps of text and KnitContents
765
:return: (text_map, content_map) where text_map contains the texts for
766
the requested versions and content_map contains the KnitContents.
767
Both dicts take version_ids as their keys.
769
for version_id in version_ids:
770
if not self.has_version(version_id):
771
raise RevisionNotPresent(version_id, self.filename)
772
record_map = self._get_record_map(version_ids)
777
for version_id in version_ids:
780
while cursor is not None:
781
method, data, digest, next = record_map[cursor]
782
components.append((cursor, method, data, digest))
783
if cursor in content_map:
788
for component_id, method, data, digest in reversed(components):
789
if component_id in content_map:
790
content = content_map[component_id]
792
version_idx = self._index.lookup(component_id)
793
if method == 'fulltext':
794
assert content is None
795
content = self.factory.parse_fulltext(data, version_idx)
796
elif method == 'line-delta':
797
delta = self.factory.parse_line_delta(data[:],
799
content = content.copy()
800
content._lines = self._apply_delta(content._lines,
802
content_map[component_id] = content
804
if 'no-eol' in self._index.get_options(version_id):
805
content = content.copy()
806
line = content._lines[-1][1].rstrip('\n')
807
content._lines[-1] = (content._lines[-1][0], line)
808
final_content[version_id] = content
810
# digest here is the digest from the last applied component.
811
text = content.text()
812
if sha_strings(text) != digest:
813
raise KnitCorrupt(self.filename,
814
'sha-1 does not match %s' % version_id)
816
text_map[version_id] = text
817
return text_map, final_content
819
def iter_lines_added_or_present_in_versions(self, version_ids=None,
821
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
822
if version_ids is None:
823
version_ids = self.versions()
825
pb = progress.DummyProgress()
826
# we don't care about inclusions, the caller cares.
827
# but we need to setup a list of records to visit.
828
# we need version_id, position, length
829
version_id_records = []
830
requested_versions = list(version_ids)
831
# filter for available versions
832
for version_id in requested_versions:
833
if not self.has_version(version_id):
834
raise RevisionNotPresent(version_id, self.filename)
835
# get a in-component-order queue:
837
for version_id in self.versions():
838
if version_id in requested_versions:
839
version_ids.append(version_id)
840
data_pos, length = self._index.get_position(version_id)
841
version_id_records.append((version_id, data_pos, length))
843
total = len(version_id_records)
844
for version_idx, (version_id, data, sha_value) in \
845
enumerate(self._data.read_records_iter(version_id_records)):
846
pb.update('Walking content.', version_idx, total)
847
method = self._index.get_method(version_id)
848
version_idx = self._index.lookup(version_id)
849
assert method in ('fulltext', 'line-delta')
850
if method == 'fulltext':
851
content = self.factory.parse_fulltext(data, version_idx)
852
for line in content.text():
855
delta = self.factory.parse_line_delta(data, version_idx)
856
for start, end, count, lines in delta:
857
for origin, line in lines:
859
pb.update('Walking content.', total, total)
861
def num_versions(self):
862
"""See VersionedFile.num_versions()."""
863
return self._index.num_versions()
865
__len__ = num_versions
867
def annotate_iter(self, version_id):
868
"""See VersionedFile.annotate_iter."""
869
content = self._get_content(version_id)
870
for origin, text in content.annotate_iter():
873
def get_parents(self, version_id):
874
"""See VersionedFile.get_parents."""
877
# 52554 calls in 1264 872 internal down from 3674
879
return self._index.get_parents(version_id)
881
raise RevisionNotPresent(version_id, self.filename)
883
def get_parents_with_ghosts(self, version_id):
884
"""See VersionedFile.get_parents."""
886
return self._index.get_parents_with_ghosts(version_id)
888
raise RevisionNotPresent(version_id, self.filename)
890
def get_ancestry(self, versions):
891
"""See VersionedFile.get_ancestry."""
892
if isinstance(versions, basestring):
893
versions = [versions]
896
self._check_versions_present(versions)
897
return self._index.get_ancestry(versions)
899
def get_ancestry_with_ghosts(self, versions):
900
"""See VersionedFile.get_ancestry_with_ghosts."""
901
if isinstance(versions, basestring):
902
versions = [versions]
905
self._check_versions_present(versions)
906
return self._index.get_ancestry_with_ghosts(versions)
908
#@deprecated_method(zero_eight)
909
def walk(self, version_ids):
910
"""See VersionedFile.walk."""
911
# We take the short path here, and extract all relevant texts
912
# and put them in a weave and let that do all the work. Far
913
# from optimal, but is much simpler.
914
# FIXME RB 20060228 this really is inefficient!
915
from bzrlib.weave import Weave
917
w = Weave(self.filename)
918
ancestry = self.get_ancestry(version_ids)
919
sorted_graph = topo_sort(self._index.get_graph())
920
version_list = [vid for vid in sorted_graph if vid in ancestry]
922
for version_id in version_list:
923
lines = self.get_lines(version_id)
924
w.add_lines(version_id, self.get_parents(version_id), lines)
926
for lineno, insert_id, dset, line in w.walk(version_ids):
927
yield lineno, insert_id, dset, line
929
def plan_merge(self, ver_a, ver_b):
930
"""See VersionedFile.plan_merge."""
931
ancestors_b = set(self.get_ancestry(ver_b))
932
def status_a(revision, text):
933
if revision in ancestors_b:
934
return 'killed-b', text
938
ancestors_a = set(self.get_ancestry(ver_a))
939
def status_b(revision, text):
940
if revision in ancestors_a:
941
return 'killed-a', text
945
annotated_a = self.annotate(ver_a)
946
annotated_b = self.annotate(ver_b)
947
plain_a = [t for (a, t) in annotated_a]
948
plain_b = [t for (a, t) in annotated_b]
949
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
952
for ai, bi, l in blocks:
953
# process all mismatched sections
954
# (last mismatched section is handled because blocks always
955
# includes a 0-length last block)
956
for revision, text in annotated_a[a_cur:ai]:
957
yield status_a(revision, text)
958
for revision, text in annotated_b[b_cur:bi]:
959
yield status_b(revision, text)
961
# and now the matched section
964
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
965
assert text_a == text_b
966
yield "unchanged", text_a
969
class _KnitComponentFile(object):
970
"""One of the files used to implement a knit database"""
972
def __init__(self, transport, filename, mode, file_mode=None,
973
create_parent_dir=False, dir_mode=None):
974
self._transport = transport
975
self._filename = filename
977
self._file_mode = file_mode
978
self._dir_mode = dir_mode
979
self._create_parent_dir = create_parent_dir
980
self._need_to_create = False
982
def check_header(self, fp):
984
if line != self.HEADER:
985
raise KnitHeaderError(badline=line)
988
"""Commit is a nop."""
991
return '%s(%s)' % (self.__class__.__name__, self._filename)
994
class _KnitIndex(_KnitComponentFile):
995
"""Manages knit index file.
997
The index is already kept in memory and read on startup, to enable
998
fast lookups of revision information. The cursor of the index
999
file is always pointing to the end, making it easy to append
1002
_cache is a cache for fast mapping from version id to a Index
1005
_history is a cache for fast mapping from indexes to version ids.
1007
The index data format is dictionary compressed when it comes to
1008
parent references; a index entry may only have parents that with a
1009
lover index number. As a result, the index is topological sorted.
1011
Duplicate entries may be written to the index for a single version id
1012
if this is done then the latter one completely replaces the former:
1013
this allows updates to correct version and parent information.
1014
Note that the two entries may share the delta, and that successive
1015
annotations and references MUST point to the first entry.
1017
The index file on disc contains a header, followed by one line per knit
1018
record. The same revision can be present in an index file more than once.
1019
The first occurrence gets assigned a sequence number starting from 0.
1021
The format of a single line is
1022
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1023
REVISION_ID is a utf8-encoded revision id
1024
FLAGS is a comma separated list of flags about the record. Values include
1025
no-eol, line-delta, fulltext.
1026
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1027
that the the compressed data starts at.
1028
LENGTH is the ascii representation of the length of the data file.
1029
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1031
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1032
revision id already in the knit that is a parent of REVISION_ID.
1033
The ' :' marker is the end of record marker.
1036
when a write is interrupted to the index file, it will result in a line that
1037
does not end in ' :'. If the ' :' is not present at the end of a line, or at
1038
the end of the file, then the record that is missing it will be ignored by
1041
When writing new records to the index file, the data is preceded by '\n'
1042
to ensure that records always start on new lines even if the last write was
1043
interrupted. As a result its normal for the last line in the index to be
1044
missing a trailing newline. One can be added with no harmful effects.
1047
HEADER = "# bzr knit index 8\n"
1049
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1050
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1052
def _cache_version(self, version_id, options, pos, size, parents):
1053
"""Cache a version record in the history array and index cache.
1055
This is inlined into __init__ for performance. KEEP IN SYNC.
1056
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1059
# only want the _history index to reference the 1st index entry
1061
if version_id not in self._cache:
1062
index = len(self._history)
1063
self._history.append(version_id)
1065
index = self._cache[version_id][5]
1066
self._cache[version_id] = (version_id,
1073
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1074
create_parent_dir=False, delay_create=False, dir_mode=None):
1075
_KnitComponentFile.__init__(self, transport, filename, mode,
1076
file_mode=file_mode,
1077
create_parent_dir=create_parent_dir,
1080
# position in _history is the 'official' index for a revision
1081
# but the values may have come from a newer entry.
1082
# so - wc -l of a knit index is != the number of unique names
1085
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1090
pb.update('read knit index', count, total)
1091
fp = self._transport.get(self._filename)
1093
self.check_header(fp)
1094
# readlines reads the whole file at once:
1095
# bad for transports like http, good for local disk
1096
# we save 60 ms doing this one change (
1097
# from calling readline each time to calling
1099
# probably what we want for nice behaviour on
1100
# http is a incremental readlines that yields, or
1101
# a check for local vs non local indexes,
1102
for l in fp.readlines():
1104
if len(rec) < 5 or rec[-1] != ':':
1106
# FIXME: in the future we should determine if its a
1107
# short write - and ignore it
1108
# or a different failure, and raise. RBC 20060407
1112
#pb.update('read knit index', count, total)
1113
# See self._parse_parents
1115
for value in rec[4:-1]:
1117
# uncompressed reference
1118
parents.append(value[1:])
1120
# this is 15/4000ms faster than isinstance,
1122
# this function is called thousands of times a
1123
# second so small variations add up.
1124
assert value.__class__ is str
1125
parents.append(self._history[int(value)])
1126
# end self._parse_parents
1127
# self._cache_version(rec[0],
1128
# rec[1].split(','),
1132
# --- self._cache_version
1133
# only want the _history index to reference the 1st
1134
# index entry for version_id
1136
if version_id not in self._cache:
1137
index = len(self._history)
1138
self._history.append(version_id)
1140
index = self._cache[version_id][5]
1141
self._cache[version_id] = (version_id,
1147
# --- self._cache_version
1150
except NoSuchFile, e:
1151
if mode != 'w' or not create:
1154
self._need_to_create = True
1156
self._transport.put_bytes_non_atomic(self._filename,
1157
self.HEADER, mode=self._file_mode)
1160
pb.update('read knit index', total, total)
1163
def _parse_parents(self, compressed_parents):
1164
"""convert a list of string parent values into version ids.
1166
ints are looked up in the index.
1167
.FOO values are ghosts and converted in to FOO.
1169
NOTE: the function is retained here for clarity, and for possible
1170
use in partial index reads. However bulk processing now has
1171
it inlined in __init__ for inner-loop optimisation.
1174
for value in compressed_parents:
1175
if value[-1] == '.':
1176
# uncompressed reference
1177
result.append(value[1:])
1179
# this is 15/4000ms faster than isinstance,
1180
# this function is called thousands of times a
1181
# second so small variations add up.
1182
assert value.__class__ is str
1183
result.append(self._history[int(value)])
1186
def get_graph(self):
1188
for version_id, index in self._cache.iteritems():
1189
graph.append((version_id, index[4]))
1192
def get_ancestry(self, versions):
1193
"""See VersionedFile.get_ancestry."""
1194
# get a graph of all the mentioned versions:
1196
pending = set(versions)
1198
version = pending.pop()
1199
parents = self._cache[version][4]
1200
# got the parents ok
1202
parents = [parent for parent in parents if parent in self._cache]
1203
for parent in parents:
1204
# if not completed and not a ghost
1205
if parent not in graph:
1207
graph[version] = parents
1208
return topo_sort(graph.items())
1210
def get_ancestry_with_ghosts(self, versions):
1211
"""See VersionedFile.get_ancestry_with_ghosts."""
1212
# get a graph of all the mentioned versions:
1214
pending = set(versions)
1216
version = pending.pop()
1218
parents = self._cache[version][4]
1224
# got the parents ok
1225
for parent in parents:
1226
if parent not in graph:
1228
graph[version] = parents
1229
return topo_sort(graph.items())
1231
def num_versions(self):
1232
return len(self._history)
1234
__len__ = num_versions
1236
def get_versions(self):
1237
return self._history
1239
def idx_to_name(self, idx):
1240
return self._history[idx]
1242
def lookup(self, version_id):
1243
assert version_id in self._cache
1244
return self._cache[version_id][5]
1246
def _version_list_to_index(self, versions):
1247
encode_utf8 = cache_utf8.encode
1249
for version in versions:
1250
if version in self._cache:
1251
# -- inlined lookup() --
1252
result_list.append(str(self._cache[version][5]))
1253
# -- end lookup () --
1255
result_list.append('.' + encode_utf8(version))
1256
return ' '.join(result_list)
1258
def add_version(self, version_id, options, pos, size, parents):
1259
"""Add a version record to the index."""
1260
self.add_versions(((version_id, options, pos, size, parents),))
1262
def add_versions(self, versions):
1263
"""Add multiple versions to the index.
1265
:param versions: a list of tuples:
1266
(version_id, options, pos, size, parents).
1269
encode_utf8 = cache_utf8.encode
1270
orig_history = self._history[:]
1271
orig_cache = self._cache.copy()
1274
for version_id, options, pos, size, parents in versions:
1275
line = "\n%s %s %s %s %s :" % (encode_utf8(version_id),
1279
self._version_list_to_index(parents))
1280
assert isinstance(line, str), \
1281
'content must be utf-8 encoded: %r' % (line,)
1283
self._cache_version(version_id, options, pos, size, parents)
1284
if not self._need_to_create:
1285
self._transport.append_bytes(self._filename, ''.join(lines))
1288
sio.write(self.HEADER)
1289
sio.writelines(lines)
1291
self._transport.put_file_non_atomic(self._filename, sio,
1292
create_parent_dir=self._create_parent_dir,
1293
mode=self._file_mode,
1294
dir_mode=self._dir_mode)
1295
self._need_to_create = False
1297
# If any problems happen, restore the original values and re-raise
1298
self._history = orig_history
1299
self._cache = orig_cache
1302
def has_version(self, version_id):
1303
"""True if the version is in the index."""
1304
return (version_id in self._cache)
1306
def get_position(self, version_id):
1307
"""Return data position and size of specified version."""
1308
return (self._cache[version_id][2], \
1309
self._cache[version_id][3])
1311
def get_method(self, version_id):
1312
"""Return compression method of specified version."""
1313
options = self._cache[version_id][1]
1314
if 'fulltext' in options:
1317
assert 'line-delta' in options
1320
def get_options(self, version_id):
1321
return self._cache[version_id][1]
1323
def get_parents(self, version_id):
1324
"""Return parents of specified version ignoring ghosts."""
1325
return [parent for parent in self._cache[version_id][4]
1326
if parent in self._cache]
1328
def get_parents_with_ghosts(self, version_id):
1329
"""Return parents of specified version with ghosts."""
1330
return self._cache[version_id][4]
1332
def check_versions_present(self, version_ids):
1333
"""Check that all specified versions are present."""
1334
version_ids = set(version_ids)
1335
for version_id in list(version_ids):
1336
if version_id in self._cache:
1337
version_ids.remove(version_id)
1339
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1342
class _KnitData(_KnitComponentFile):
1343
"""Contents of the knit data file"""
1345
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1346
create_parent_dir=False, delay_create=False,
1348
_KnitComponentFile.__init__(self, transport, filename, mode,
1349
file_mode=file_mode,
1350
create_parent_dir=create_parent_dir,
1352
self._checked = False
1353
# TODO: jam 20060713 conceptually, this could spill to disk
1354
# if the cached size gets larger than a certain amount
1355
# but it complicates the model a bit, so for now just use
1356
# a simple dictionary
1358
self._do_cache = False
1361
self._need_to_create = create
1363
self._transport.put_bytes_non_atomic(self._filename, '',
1364
mode=self._file_mode)
1366
def enable_cache(self):
1367
"""Enable caching of reads."""
1368
self._do_cache = True
1370
def clear_cache(self):
1371
"""Clear the record cache."""
1372
self._do_cache = False
1375
def _open_file(self):
1377
return self._transport.get(self._filename)
1382
def _record_to_data(self, version_id, digest, lines):
1383
"""Convert version_id, digest, lines into a raw data block.
1385
:return: (len, a StringIO instance with the raw data ready to read.)
1388
data_file = GzipFile(None, mode='wb', fileobj=sio)
1390
version_id_utf8 = cache_utf8.encode(version_id)
1391
data_file.writelines(chain(
1392
["version %s %d %s\n" % (version_id_utf8,
1396
["end %s\n" % version_id_utf8]))
1403
def add_raw_record(self, raw_data):
1404
"""Append a prepared record to the data file.
1406
:return: the offset in the data file raw_data was written.
1408
assert isinstance(raw_data, str), 'data must be plain bytes'
1409
if not self._need_to_create:
1410
return self._transport.append_bytes(self._filename, raw_data)
1412
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1413
create_parent_dir=self._create_parent_dir,
1414
mode=self._file_mode,
1415
dir_mode=self._dir_mode)
1416
self._need_to_create = False
1419
def add_record(self, version_id, digest, lines):
1420
"""Write new text record to disk. Returns the position in the
1421
file where it was written."""
1422
size, sio = self._record_to_data(version_id, digest, lines)
1424
if not self._need_to_create:
1425
start_pos = self._transport.append_file(self._filename, sio)
1427
self._transport.put_file_non_atomic(self._filename, sio,
1428
create_parent_dir=self._create_parent_dir,
1429
mode=self._file_mode,
1430
dir_mode=self._dir_mode)
1431
self._need_to_create = False
1434
self._cache[version_id] = sio.getvalue()
1435
return start_pos, size
1437
def _parse_record_header(self, version_id, raw_data):
1438
"""Parse a record header for consistency.
1440
:return: the header and the decompressor stream.
1441
as (stream, header_record)
1443
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1444
rec = df.readline().split()
1446
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
1447
if cache_utf8.decode(rec[1]) != version_id:
1448
raise KnitCorrupt(self._filename,
1449
'unexpected version, wanted %r, got %r' % (
1450
version_id, rec[1]))
1453
def _parse_record(self, version_id, data):
1455
# 4168 calls in 2880 217 internal
1456
# 4168 calls to _parse_record_header in 2121
1457
# 4168 calls to readlines in 330
1458
df, rec = self._parse_record_header(version_id, data)
1459
record_contents = df.readlines()
1460
l = record_contents.pop()
1461
assert len(record_contents) == int(rec[2])
1462
if l != 'end %s\n' % cache_utf8.encode(version_id):
1463
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
1466
return record_contents, rec[3]
1468
def read_records_iter_raw(self, records):
1469
"""Read text records from data file and yield raw data.
1471
This unpacks enough of the text record to validate the id is
1472
as expected but thats all.
1474
# setup an iterator of the external records:
1475
# uses readv so nice and fast we hope.
1477
# grab the disk data needed.
1479
# Don't check _cache if it is empty
1480
needed_offsets = [(pos, size) for version_id, pos, size
1482
if version_id not in self._cache]
1484
needed_offsets = [(pos, size) for version_id, pos, size
1487
raw_records = self._transport.readv(self._filename, needed_offsets)
1490
for version_id, pos, size in records:
1491
if version_id in self._cache:
1492
# This data has already been validated
1493
data = self._cache[version_id]
1495
pos, data = raw_records.next()
1497
self._cache[version_id] = data
1499
# validate the header
1500
df, rec = self._parse_record_header(version_id, data)
1502
yield version_id, data
1504
def read_records_iter(self, records):
1505
"""Read text records from data file and yield result.
1507
The result will be returned in whatever is the fastest to read.
1508
Not by the order requested. Also, multiple requests for the same
1509
record will only yield 1 response.
1510
:param records: A list of (version_id, pos, len) entries
1511
:return: Yields (version_id, contents, digest) in the order
1512
read, not the order requested
1518
# Skip records we have alread seen
1519
yielded_records = set()
1520
needed_records = set()
1521
for record in records:
1522
if record[0] in self._cache:
1523
if record[0] in yielded_records:
1525
yielded_records.add(record[0])
1526
data = self._cache[record[0]]
1527
content, digest = self._parse_record(record[0], data)
1528
yield (record[0], content, digest)
1530
needed_records.add(record)
1531
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1533
needed_records = sorted(set(records), key=operator.itemgetter(1))
1535
if not needed_records:
1538
# The transport optimizes the fetching as well
1539
# (ie, reads continuous ranges.)
1540
readv_response = self._transport.readv(self._filename,
1541
[(pos, size) for version_id, pos, size in needed_records])
1543
for (version_id, pos, size), (pos, data) in \
1544
izip(iter(needed_records), readv_response):
1545
content, digest = self._parse_record(version_id, data)
1547
self._cache[version_id] = data
1548
yield version_id, content, digest
1550
def read_records(self, records):
1551
"""Read records into a dictionary."""
1553
for record_id, content, digest in \
1554
self.read_records_iter(records):
1555
components[record_id] = (content, digest)
1559
class InterKnit(InterVersionedFile):
1560
"""Optimised code paths for knit to knit operations."""
1562
_matching_file_from_factory = KnitVersionedFile
1563
_matching_file_to_factory = KnitVersionedFile
1566
def is_compatible(source, target):
1567
"""Be compatible with knits. """
1569
return (isinstance(source, KnitVersionedFile) and
1570
isinstance(target, KnitVersionedFile))
1571
except AttributeError:
1574
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1575
"""See InterVersionedFile.join."""
1576
assert isinstance(self.source, KnitVersionedFile)
1577
assert isinstance(self.target, KnitVersionedFile)
1579
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1584
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1586
version_ids = list(version_ids)
1587
if None in version_ids:
1588
version_ids.remove(None)
1590
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1591
this_versions = set(self.target._index.get_versions())
1592
needed_versions = self.source_ancestry - this_versions
1593
cross_check_versions = self.source_ancestry.intersection(this_versions)
1594
mismatched_versions = set()
1595
for version in cross_check_versions:
1596
# scan to include needed parents.
1597
n1 = set(self.target.get_parents_with_ghosts(version))
1598
n2 = set(self.source.get_parents_with_ghosts(version))
1600
# FIXME TEST this check for cycles being introduced works
1601
# the logic is we have a cycle if in our graph we are an
1602
# ancestor of any of the n2 revisions.
1608
parent_ancestors = self.source.get_ancestry(parent)
1609
if version in parent_ancestors:
1610
raise errors.GraphCycleError([parent, version])
1611
# ensure this parent will be available later.
1612
new_parents = n2.difference(n1)
1613
needed_versions.update(new_parents.difference(this_versions))
1614
mismatched_versions.add(version)
1616
if not needed_versions and not mismatched_versions:
1618
full_list = topo_sort(self.source.get_graph())
1620
version_list = [i for i in full_list if (not self.target.has_version(i)
1621
and i in needed_versions)]
1625
copy_queue_records = []
1627
for version_id in version_list:
1628
options = self.source._index.get_options(version_id)
1629
parents = self.source._index.get_parents_with_ghosts(version_id)
1630
# check that its will be a consistent copy:
1631
for parent in parents:
1632
# if source has the parent, we must :
1633
# * already have it or
1634
# * have it scheduled already
1635
# otherwise we don't care
1636
assert (self.target.has_version(parent) or
1637
parent in copy_set or
1638
not self.source.has_version(parent))
1639
data_pos, data_size = self.source._index.get_position(version_id)
1640
copy_queue_records.append((version_id, data_pos, data_size))
1641
copy_queue.append((version_id, options, parents))
1642
copy_set.add(version_id)
1644
# data suck the join:
1646
total = len(version_list)
1649
for (version_id, raw_data), \
1650
(version_id2, options, parents) in \
1651
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1653
assert version_id == version_id2, 'logic error, inconsistent results'
1655
pb.update("Joining knit", count, total)
1656
raw_records.append((version_id, options, parents, len(raw_data)))
1657
raw_datum.append(raw_data)
1658
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1660
for version in mismatched_versions:
1661
# FIXME RBC 20060309 is this needed?
1662
n1 = set(self.target.get_parents_with_ghosts(version))
1663
n2 = set(self.source.get_parents_with_ghosts(version))
1664
# write a combined record to our history preserving the current
1665
# parents as first in the list
1666
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1667
self.target.fix_parents(version, new_parents)
1673
InterVersionedFile.register_optimiser(InterKnit)
1676
class WeaveToKnit(InterVersionedFile):
1677
"""Optimised code paths for weave to knit operations."""
1679
_matching_file_from_factory = bzrlib.weave.WeaveFile
1680
_matching_file_to_factory = KnitVersionedFile
1683
def is_compatible(source, target):
1684
"""Be compatible with weaves to knits."""
1686
return (isinstance(source, bzrlib.weave.Weave) and
1687
isinstance(target, KnitVersionedFile))
1688
except AttributeError:
1691
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1692
"""See InterVersionedFile.join."""
1693
assert isinstance(self.source, bzrlib.weave.Weave)
1694
assert isinstance(self.target, KnitVersionedFile)
1696
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1701
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1703
version_ids = list(version_ids)
1705
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1706
this_versions = set(self.target._index.get_versions())
1707
needed_versions = self.source_ancestry - this_versions
1708
cross_check_versions = self.source_ancestry.intersection(this_versions)
1709
mismatched_versions = set()
1710
for version in cross_check_versions:
1711
# scan to include needed parents.
1712
n1 = set(self.target.get_parents_with_ghosts(version))
1713
n2 = set(self.source.get_parents(version))
1714
# if all of n2's parents are in n1, then its fine.
1715
if n2.difference(n1):
1716
# FIXME TEST this check for cycles being introduced works
1717
# the logic is we have a cycle if in our graph we are an
1718
# ancestor of any of the n2 revisions.
1724
parent_ancestors = self.source.get_ancestry(parent)
1725
if version in parent_ancestors:
1726
raise errors.GraphCycleError([parent, version])
1727
# ensure this parent will be available later.
1728
new_parents = n2.difference(n1)
1729
needed_versions.update(new_parents.difference(this_versions))
1730
mismatched_versions.add(version)
1732
if not needed_versions and not mismatched_versions:
1734
full_list = topo_sort(self.source.get_graph())
1736
version_list = [i for i in full_list if (not self.target.has_version(i)
1737
and i in needed_versions)]
1741
total = len(version_list)
1742
for version_id in version_list:
1743
pb.update("Converting to knit", count, total)
1744
parents = self.source.get_parents(version_id)
1745
# check that its will be a consistent copy:
1746
for parent in parents:
1747
# if source has the parent, we must already have it
1748
assert (self.target.has_version(parent))
1749
self.target.add_lines(
1750
version_id, parents, self.source.get_lines(version_id))
1753
for version in mismatched_versions:
1754
# FIXME RBC 20060309 is this needed?
1755
n1 = set(self.target.get_parents_with_ghosts(version))
1756
n2 = set(self.source.get_parents(version))
1757
# write a combined record to our history preserving the current
1758
# parents as first in the list
1759
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1760
self.target.fix_parents(version, new_parents)
1766
InterVersionedFile.register_optimiser(WeaveToKnit)
1769
class KnitSequenceMatcher(difflib.SequenceMatcher):
1770
"""Knit tuned sequence matcher.
1772
This is based on profiling of difflib which indicated some improvements
1773
for our usage pattern.
1776
def find_longest_match(self, alo, ahi, blo, bhi):
1777
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1779
If isjunk is not defined:
1781
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1782
alo <= i <= i+k <= ahi
1783
blo <= j <= j+k <= bhi
1784
and for all (i',j',k') meeting those conditions,
1787
and if i == i', j <= j'
1789
In other words, of all maximal matching blocks, return one that
1790
starts earliest in a, and of all those maximal matching blocks that
1791
start earliest in a, return the one that starts earliest in b.
1793
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1794
>>> s.find_longest_match(0, 5, 0, 9)
1797
If isjunk is defined, first the longest matching block is
1798
determined as above, but with the additional restriction that no
1799
junk element appears in the block. Then that block is extended as
1800
far as possible by matching (only) junk elements on both sides. So
1801
the resulting block never matches on junk except as identical junk
1802
happens to be adjacent to an "interesting" match.
1804
Here's the same example as before, but considering blanks to be
1805
junk. That prevents " abcd" from matching the " abcd" at the tail
1806
end of the second sequence directly. Instead only the "abcd" can
1807
match, and matches the leftmost "abcd" in the second sequence:
1809
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1810
>>> s.find_longest_match(0, 5, 0, 9)
1813
If no blocks match, return (alo, blo, 0).
1815
>>> s = SequenceMatcher(None, "ab", "c")
1816
>>> s.find_longest_match(0, 2, 0, 1)
1820
# CAUTION: stripping common prefix or suffix would be incorrect.
1824
# Longest matching block is "ab", but if common prefix is
1825
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1826
# strip, so ends up claiming that ab is changed to acab by
1827
# inserting "ca" in the middle. That's minimal but unintuitive:
1828
# "it's obvious" that someone inserted "ac" at the front.
1829
# Windiff ends up at the same place as diff, but by pairing up
1830
# the unique 'b's and then matching the first two 'a's.
1832
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1833
besti, bestj, bestsize = alo, blo, 0
1834
# find longest junk-free match
1835
# during an iteration of the loop, j2len[j] = length of longest
1836
# junk-free match ending with a[i-1] and b[j]
1840
for i in xrange(alo, ahi):
1841
# look at all instances of a[i] in b; note that because
1842
# b2j has no junk keys, the loop is skipped if a[i] is junk
1843
j2lenget = j2len.get
1846
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1847
# following improvement
1848
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1849
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1850
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1852
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1853
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1854
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1866
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1868
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1871
# Extend the best by non-junk elements on each end. In particular,
1872
# "popular" non-junk elements aren't in b2j, which greatly speeds
1873
# the inner loop above, but also means "the best" match so far
1874
# doesn't contain any junk *or* popular non-junk elements.
1875
while besti > alo and bestj > blo and \
1876
not isbjunk(b[bestj-1]) and \
1877
a[besti-1] == b[bestj-1]:
1878
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1879
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1880
not isbjunk(b[bestj+bestsize]) and \
1881
a[besti+bestsize] == b[bestj+bestsize]:
1884
# Now that we have a wholly interesting match (albeit possibly
1885
# empty!), we may as well suck up the matching junk on each
1886
# side of it too. Can't think of a good reason not to, and it
1887
# saves post-processing the (possibly considerable) expense of
1888
# figuring out what to do with it. In the case of an empty
1889
# interesting match, this is clearly the right thing to do,
1890
# because no other kind of match is possible in the regions.
1891
while besti > alo and bestj > blo and \
1892
isbjunk(b[bestj-1]) and \
1893
a[besti-1] == b[bestj-1]:
1894
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1895
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1896
isbjunk(b[bestj+bestsize]) and \
1897
a[besti+bestsize] == b[bestj+bestsize]:
1898
bestsize = bestsize + 1
1900
return besti, bestj, bestsize
added
removed
bzrlib/knit.py
