1
# Copyright (C) 2005, 2006 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""Knit versionedfile implementation.
19
A knit is a versioned file implementation that supports efficient append only
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
end-marker; simply "end VERSION"
28
delta can be line or full contents.a
29
... the 8's there are the index number of the annotation.
30
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
8 e.set('executable', 'yes')
36
8 if elt.get('executable') == 'yes':
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
43
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
44
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
45
09:33 < lifeless> right
46
09:33 < jrydberg> lifeless: the position and size is the range in the data file
49
so the index sequence is the dictionary compressed sequence number used
50
in the deltas to provide line annotation
55
# 10:16 < lifeless> make partial index writes safe
56
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
57
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
59
# move sha1 out of the content so that join is faster at verifying parents
60
# record content length ?
64
from cStringIO import StringIO
66
from itertools import izip, chain
81
from bzrlib.errors import (
89
RevisionAlreadyPresent,
91
from bzrlib.tuned_gzip import GzipFile
92
from bzrlib.trace import mutter
93
from bzrlib.osutils import (
98
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
99
from bzrlib.tsort import topo_sort
102
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
105
# TODO: Split out code specific to this format into an associated object.
107
# TODO: Can we put in some kind of value to check that the index and data
108
# files belong together?
110
# TODO: accommodate binaries, perhaps by storing a byte count
112
# TODO: function to check whole file
114
# TODO: atomically append data, then measure backwards from the cursor
115
# position after writing to work out where it was located. we may need to
116
# bypass python file buffering.
118
DATA_SUFFIX = '.knit'
119
INDEX_SUFFIX = '.kndx'
122
class KnitContent(object):
123
"""Content of a knit version to which deltas can be applied."""
125
def __init__(self, lines):
128
def annotate_iter(self):
129
"""Yield tuples of (origin, text) for each content line."""
130
return iter(self._lines)
133
"""Return a list of (origin, text) tuples."""
134
return list(self.annotate_iter())
136
def line_delta_iter(self, new_lines):
137
"""Generate line-based delta from this content to new_lines."""
138
new_texts = new_lines.text()
139
old_texts = self.text()
140
s = KnitSequenceMatcher(None, old_texts, new_texts)
141
for tag, i1, i2, j1, j2 in s.get_opcodes():
144
# ofrom, oto, length, data
145
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
147
def line_delta(self, new_lines):
148
return list(self.line_delta_iter(new_lines))
151
return [text for origin, text in self._lines]
154
return KnitContent(self._lines[:])
157
def get_line_delta_blocks(knit_delta, source, target):
158
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
159
target_len = len(target)
162
for s_begin, s_end, t_len, new_text in knit_delta:
163
true_n = s_begin - s_pos
166
# knit deltas do not provide reliable info about whether the
167
# last line of a file matches, due to eol handling.
168
if source[s_pos + n -1] != target[t_pos + n -1]:
171
yield s_pos, t_pos, n
172
t_pos += t_len + true_n
174
n = target_len - t_pos
176
if source[s_pos + n -1] != target[t_pos + n -1]:
179
yield s_pos, t_pos, n
180
yield s_pos + (target_len - t_pos), target_len, 0
183
class _KnitFactory(object):
184
"""Base factory for creating content objects."""
186
def make(self, lines, version_id):
187
num_lines = len(lines)
188
return KnitContent(zip([version_id] * num_lines, lines))
191
class KnitAnnotateFactory(_KnitFactory):
192
"""Factory for creating annotated Content objects."""
196
def parse_fulltext(self, content, version_id):
197
"""Convert fulltext to internal representation
199
fulltext content is of the format
200
revid(utf8) plaintext\n
201
internal representation is of the format:
204
# TODO: jam 20070209 The tests expect this to be returned as tuples,
205
# but the code itself doesn't really depend on that.
206
# Figure out a way to not require the overhead of turning the
207
# list back into tuples.
208
lines = [tuple(line.split(' ', 1)) for line in content]
209
return KnitContent(lines)
211
def parse_line_delta_iter(self, lines):
212
return iter(self.parse_line_delta(lines))
214
def parse_line_delta(self, lines, version_id):
215
"""Convert a line based delta into internal representation.
217
line delta is in the form of:
218
intstart intend intcount
220
revid(utf8) newline\n
221
internal representation is
222
(start, end, count, [1..count tuples (revid, newline)])
229
def cache_and_return(line):
230
origin, text = line.split(' ', 1)
231
return cache.setdefault(origin, origin), text
233
# walk through the lines parsing.
235
start, end, count = [int(n) for n in header.split(',')]
236
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
237
result.append((start, end, count, contents))
240
def get_fulltext_content(self, lines):
241
"""Extract just the content lines from a fulltext."""
242
return (line.split(' ', 1)[1] for line in lines)
244
def get_linedelta_content(self, lines):
245
"""Extract just the content from a line delta.
247
This doesn't return all of the extra information stored in a delta.
248
Only the actual content lines.
253
header = header.split(',')
254
count = int(header[2])
255
for i in xrange(count):
256
origin, text = next().split(' ', 1)
259
def lower_fulltext(self, content):
260
"""convert a fulltext content record into a serializable form.
262
see parse_fulltext which this inverts.
264
# TODO: jam 20070209 We only do the caching thing to make sure that
265
# the origin is a valid utf-8 line, eventually we could remove it
266
return ['%s %s' % (o, t) for o, t in content._lines]
268
def lower_line_delta(self, delta):
269
"""convert a delta into a serializable form.
271
See parse_line_delta which this inverts.
273
# TODO: jam 20070209 We only do the caching thing to make sure that
274
# the origin is a valid utf-8 line, eventually we could remove it
276
for start, end, c, lines in delta:
277
out.append('%d,%d,%d\n' % (start, end, c))
278
out.extend(origin + ' ' + text
279
for origin, text in lines)
283
class KnitPlainFactory(_KnitFactory):
284
"""Factory for creating plain Content objects."""
288
def parse_fulltext(self, content, version_id):
289
"""This parses an unannotated fulltext.
291
Note that this is not a noop - the internal representation
292
has (versionid, line) - its just a constant versionid.
294
return self.make(content, version_id)
296
def parse_line_delta_iter(self, lines, version_id):
298
num_lines = len(lines)
299
while cur < num_lines:
302
start, end, c = [int(n) for n in header.split(',')]
303
yield start, end, c, zip([version_id] * c, lines[cur:cur+c])
306
def parse_line_delta(self, lines, version_id):
307
return list(self.parse_line_delta_iter(lines, version_id))
309
def get_fulltext_content(self, lines):
310
"""Extract just the content lines from a fulltext."""
313
def get_linedelta_content(self, lines):
314
"""Extract just the content from a line delta.
316
This doesn't return all of the extra information stored in a delta.
317
Only the actual content lines.
322
header = header.split(',')
323
count = int(header[2])
324
for i in xrange(count):
327
def lower_fulltext(self, content):
328
return content.text()
330
def lower_line_delta(self, delta):
332
for start, end, c, lines in delta:
333
out.append('%d,%d,%d\n' % (start, end, c))
334
out.extend([text for origin, text in lines])
338
def make_empty_knit(transport, relpath):
339
"""Construct a empty knit at the specified location."""
340
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
344
class KnitVersionedFile(VersionedFile):
345
"""Weave-like structure with faster random access.
347
A knit stores a number of texts and a summary of the relationships
348
between them. Texts are identified by a string version-id. Texts
349
are normally stored and retrieved as a series of lines, but can
350
also be passed as single strings.
352
Lines are stored with the trailing newline (if any) included, to
353
avoid special cases for files with no final newline. Lines are
354
composed of 8-bit characters, not unicode. The combination of
355
these approaches should mean any 'binary' file can be safely
356
stored and retrieved.
359
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
360
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
361
create=False, create_parent_dir=False, delay_create=False,
363
"""Construct a knit at location specified by relpath.
365
:param create: If not True, only open an existing knit.
366
:param create_parent_dir: If True, create the parent directory if
367
creating the file fails. (This is used for stores with
368
hash-prefixes that may not exist yet)
369
:param delay_create: The calling code is aware that the knit won't
370
actually be created until the first data is stored.
372
if deprecated_passed(basis_knit):
373
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
374
" deprecated as of bzr 0.9.",
375
DeprecationWarning, stacklevel=2)
376
if access_mode is None:
378
super(KnitVersionedFile, self).__init__(access_mode)
379
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
380
self.transport = transport
381
self.filename = relpath
382
self.factory = factory or KnitAnnotateFactory()
383
self.writable = (access_mode == 'w')
386
self._max_delta_chain = 200
388
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
389
access_mode, create=create, file_mode=file_mode,
390
create_parent_dir=create_parent_dir, delay_create=delay_create,
392
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
393
access_mode, create=create and not len(self), file_mode=file_mode,
394
create_parent_dir=create_parent_dir, delay_create=delay_create,
398
return '%s(%s)' % (self.__class__.__name__,
399
self.transport.abspath(self.filename))
401
def _check_should_delta(self, first_parents):
402
"""Iterate back through the parent listing, looking for a fulltext.
404
This is used when we want to decide whether to add a delta or a new
405
fulltext. It searches for _max_delta_chain parents. When it finds a
406
fulltext parent, it sees if the total size of the deltas leading up to
407
it is large enough to indicate that we want a new full text anyway.
409
Return True if we should create a new delta, False if we should use a
414
delta_parents = first_parents
415
for count in xrange(self._max_delta_chain):
416
parent = delta_parents[0]
417
method = self._index.get_method(parent)
418
pos, size = self._index.get_position(parent)
419
if method == 'fulltext':
423
delta_parents = self._index.get_parents(parent)
425
# We couldn't find a fulltext, so we must create a new one
428
return fulltext_size > delta_size
430
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
431
"""See VersionedFile._add_delta()."""
432
self._check_add(version_id, []) # should we check the lines ?
433
self._check_versions_present(parents)
437
for parent in parents:
438
if not self.has_version(parent):
439
ghosts.append(parent)
441
present_parents.append(parent)
443
if delta_parent is None:
444
# reconstitute as full text.
445
assert len(delta) == 1 or len(delta) == 0
447
assert delta[0][0] == 0
448
assert delta[0][1] == 0, delta[0][1]
449
return super(KnitVersionedFile, self)._add_delta(version_id,
460
options.append('no-eol')
462
if delta_parent is not None:
463
# determine the current delta chain length.
464
# To speed the extract of texts the delta chain is limited
465
# to a fixed number of deltas. This should minimize both
466
# I/O and the time spend applying deltas.
467
# The window was changed to a maximum of 200 deltas, but also added
468
# was a check that the total compressed size of the deltas is
469
# smaller than the compressed size of the fulltext.
470
if not self._check_should_delta([delta_parent]):
471
# We don't want a delta here, just do a normal insertion.
472
return super(KnitVersionedFile, self)._add_delta(version_id,
479
options.append('line-delta')
480
store_lines = self.factory.lower_line_delta(delta)
482
where, size = self._data.add_record(version_id, digest, store_lines)
483
self._index.add_version(version_id, options, where, size, parents)
485
def _add_raw_records(self, records, data):
486
"""Add all the records 'records' with data pre-joined in 'data'.
488
:param records: A list of tuples(version_id, options, parents, size).
489
:param data: The data for the records. When it is written, the records
490
are adjusted to have pos pointing into data by the sum of
491
the preceding records sizes.
494
pos = self._data.add_raw_record(data)
497
for (version_id, options, parents, size) in records:
498
index_entries.append((version_id, options, pos+offset,
500
if self._data._do_cache:
501
self._data._cache[version_id] = data[offset:offset+size]
503
self._index.add_versions(index_entries)
505
def enable_cache(self):
506
"""Start caching data for this knit"""
507
self._data.enable_cache()
509
def clear_cache(self):
510
"""Clear the data cache only."""
511
self._data.clear_cache()
513
def copy_to(self, name, transport):
514
"""See VersionedFile.copy_to()."""
515
# copy the current index to a temp index to avoid racing with local
517
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
518
self.transport.get(self._index._filename))
520
f = self._data._open_file()
522
transport.put_file(name + DATA_SUFFIX, f)
525
# move the copied index into place
526
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
528
def create_empty(self, name, transport, mode=None):
529
return KnitVersionedFile(name, transport, factory=self.factory,
530
delta=self.delta, create=True)
532
def _fix_parents(self, version_id, new_parents):
533
"""Fix the parents list for version.
535
This is done by appending a new version to the index
536
with identical data except for the parents list.
537
the parents list must be a superset of the current
540
current_values = self._index._cache[version_id]
541
assert set(current_values[4]).difference(set(new_parents)) == set()
542
self._index.add_version(version_id,
548
def _extract_blocks(self, version_id, source, target):
549
if self._index.get_method(version_id) != 'line-delta':
551
parent, sha1, noeol, delta = self.get_delta(version_id)
552
return KnitContent.get_line_delta_blocks(delta, source, target)
554
def get_delta(self, version_id):
555
"""Get a delta for constructing version from some other version."""
556
version_id = osutils.safe_revision_id(version_id)
557
self.check_not_reserved_id(version_id)
558
if not self.has_version(version_id):
559
raise RevisionNotPresent(version_id, self.filename)
561
parents = self.get_parents(version_id)
566
data_pos, data_size = self._index.get_position(version_id)
567
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
568
noeol = 'no-eol' in self._index.get_options(version_id)
569
if 'fulltext' == self._index.get_method(version_id):
570
new_content = self.factory.parse_fulltext(data, version_id)
571
if parent is not None:
572
reference_content = self._get_content(parent)
573
old_texts = reference_content.text()
576
new_texts = new_content.text()
577
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
578
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
580
delta = self.factory.parse_line_delta(data, version_id)
581
return parent, sha1, noeol, delta
583
def get_graph_with_ghosts(self):
584
"""See VersionedFile.get_graph_with_ghosts()."""
585
graph_items = self._index.get_graph()
586
return dict(graph_items)
588
def get_sha1(self, version_id):
589
"""See VersionedFile.get_sha1()."""
590
version_id = osutils.safe_revision_id(version_id)
591
record_map = self._get_record_map([version_id])
592
method, content, digest, next = record_map[version_id]
597
"""See VersionedFile.get_suffixes()."""
598
return [DATA_SUFFIX, INDEX_SUFFIX]
600
def has_ghost(self, version_id):
601
"""True if there is a ghost reference in the file to version_id."""
602
version_id = osutils.safe_revision_id(version_id)
604
if self.has_version(version_id):
606
# optimisable if needed by memoising the _ghosts set.
607
items = self._index.get_graph()
608
for node, parents in items:
609
for parent in parents:
610
if parent not in self._index._cache:
611
if parent == version_id:
616
"""See VersionedFile.versions."""
617
return self._index.get_versions()
619
def has_version(self, version_id):
620
"""See VersionedFile.has_version."""
621
version_id = osutils.safe_revision_id(version_id)
622
return self._index.has_version(version_id)
624
__contains__ = has_version
626
def _merge_annotations(self, content, parents, parent_texts={},
627
delta=None, annotated=None):
628
"""Merge annotations for content. This is done by comparing
629
the annotations based on changed to the text.
633
for parent_id in parents:
634
merge_content = self._get_content(parent_id, parent_texts)
635
seq = patiencediff.PatienceSequenceMatcher(
636
None, merge_content.text(), content.text())
637
if delta_seq is None:
638
# setup a delta seq to reuse.
640
for i, j, n in seq.get_matching_blocks():
643
# this appears to copy (origin, text) pairs across to the new
644
# content for any line that matches the last-checked parent.
645
# FIXME: save the sequence control data for delta compression
646
# against the most relevant parent rather than rediffing.
647
content._lines[j:j+n] = merge_content._lines[i:i+n]
650
reference_content = self._get_content(parents[0], parent_texts)
651
new_texts = content.text()
652
old_texts = reference_content.text()
653
delta_seq = patiencediff.PatienceSequenceMatcher(
654
None, old_texts, new_texts)
655
return self._make_line_delta(delta_seq, content)
657
def _make_line_delta(self, delta_seq, new_content):
658
"""Generate a line delta from delta_seq and new_content."""
660
for op in delta_seq.get_opcodes():
663
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
666
def _get_components_positions(self, version_ids):
667
"""Produce a map of position data for the components of versions.
669
This data is intended to be used for retrieving the knit records.
671
A dict of version_id to (method, data_pos, data_size, next) is
673
method is the way referenced data should be applied.
674
data_pos is the position of the data in the knit.
675
data_size is the size of the data in the knit.
676
next is the build-parent of the version, or None for fulltexts.
679
for version_id in version_ids:
682
while cursor is not None and cursor not in component_data:
683
method = self._index.get_method(cursor)
684
if method == 'fulltext':
687
next = self.get_parents(cursor)[0]
688
data_pos, data_size = self._index.get_position(cursor)
689
component_data[cursor] = (method, data_pos, data_size, next)
691
return component_data
693
def _get_content(self, version_id, parent_texts={}):
694
"""Returns a content object that makes up the specified
696
if not self.has_version(version_id):
697
raise RevisionNotPresent(version_id, self.filename)
699
cached_version = parent_texts.get(version_id, None)
700
if cached_version is not None:
701
return cached_version
703
text_map, contents_map = self._get_content_maps([version_id])
704
return contents_map[version_id]
706
def _check_versions_present(self, version_ids):
707
"""Check that all specified versions are present."""
708
self._index.check_versions_present(version_ids)
710
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
711
"""See VersionedFile.add_lines_with_ghosts()."""
712
self._check_add(version_id, lines)
713
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
715
def _add_lines(self, version_id, parents, lines, parent_texts):
716
"""See VersionedFile.add_lines."""
717
self._check_add(version_id, lines)
718
self._check_versions_present(parents)
719
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
721
def _check_add(self, version_id, lines):
722
"""check that version_id and lines are safe to add."""
723
assert self.writable, "knit is not opened for write"
724
### FIXME escape. RBC 20060228
725
if contains_whitespace(version_id):
726
raise InvalidRevisionId(version_id, self.filename)
727
self.check_not_reserved_id(version_id)
728
if self.has_version(version_id):
729
raise RevisionAlreadyPresent(version_id, self.filename)
730
self._check_lines_not_unicode(lines)
731
self._check_lines_are_lines(lines)
733
def _add(self, version_id, lines, parents, delta, parent_texts):
734
"""Add a set of lines on top of version specified by parents.
736
If delta is true, compress the text as a line-delta against
739
Any versions not present will be converted into ghosts.
741
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
742
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
743
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
744
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
745
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
746
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
747
# +1383 0 8.0370 8.0370 +<len>
748
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
749
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
750
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
751
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
755
if parent_texts is None:
757
for parent in parents:
758
if not self.has_version(parent):
759
ghosts.append(parent)
761
present_parents.append(parent)
763
if delta and not len(present_parents):
766
digest = sha_strings(lines)
769
if lines[-1][-1] != '\n':
770
options.append('no-eol')
771
lines[-1] = lines[-1] + '\n'
773
if len(present_parents) and delta:
774
# To speed the extract of texts the delta chain is limited
775
# to a fixed number of deltas. This should minimize both
776
# I/O and the time spend applying deltas.
777
delta = self._check_should_delta(present_parents)
779
assert isinstance(version_id, str)
780
lines = self.factory.make(lines, version_id)
781
if delta or (self.factory.annotated and len(present_parents) > 0):
782
# Merge annotations from parent texts if so is needed.
783
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
784
delta, self.factory.annotated)
787
options.append('line-delta')
788
store_lines = self.factory.lower_line_delta(delta_hunks)
790
options.append('fulltext')
791
store_lines = self.factory.lower_fulltext(lines)
793
where, size = self._data.add_record(version_id, digest, store_lines)
794
self._index.add_version(version_id, options, where, size, parents)
797
def check(self, progress_bar=None):
798
"""See VersionedFile.check()."""
800
def _clone_text(self, new_version_id, old_version_id, parents):
801
"""See VersionedFile.clone_text()."""
802
# FIXME RBC 20060228 make fast by only inserting an index with null
804
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
806
def get_lines(self, version_id):
807
"""See VersionedFile.get_lines()."""
808
return self.get_line_list([version_id])[0]
810
def _get_record_map(self, version_ids):
811
"""Produce a dictionary of knit records.
813
The keys are version_ids, the values are tuples of (method, content,
815
method is the way the content should be applied.
816
content is a KnitContent object.
817
digest is the SHA1 digest of this version id after all steps are done
818
next is the build-parent of the version, i.e. the leftmost ancestor.
819
If the method is fulltext, next will be None.
821
position_map = self._get_components_positions(version_ids)
822
# c = component_id, m = method, p = position, s = size, n = next
823
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
825
for component_id, content, digest in \
826
self._data.read_records_iter(records):
827
method, position, size, next = position_map[component_id]
828
record_map[component_id] = method, content, digest, next
832
def get_text(self, version_id):
833
"""See VersionedFile.get_text"""
834
return self.get_texts([version_id])[0]
836
def get_texts(self, version_ids):
837
return [''.join(l) for l in self.get_line_list(version_ids)]
839
def get_line_list(self, version_ids):
840
"""Return the texts of listed versions as a list of strings."""
841
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
842
for version_id in version_ids:
843
self.check_not_reserved_id(version_id)
844
text_map, content_map = self._get_content_maps(version_ids)
845
return [text_map[v] for v in version_ids]
847
_get_line_list = get_line_list
849
def _get_content_maps(self, version_ids):
850
"""Produce maps of text and KnitContents
852
:return: (text_map, content_map) where text_map contains the texts for
853
the requested versions and content_map contains the KnitContents.
854
Both dicts take version_ids as their keys.
856
for version_id in version_ids:
857
if not self.has_version(version_id):
858
raise RevisionNotPresent(version_id, self.filename)
859
record_map = self._get_record_map(version_ids)
864
for version_id in version_ids:
867
while cursor is not None:
868
method, data, digest, next = record_map[cursor]
869
components.append((cursor, method, data, digest))
870
if cursor in content_map:
875
for component_id, method, data, digest in reversed(components):
876
if component_id in content_map:
877
content = content_map[component_id]
879
if method == 'fulltext':
880
assert content is None
881
content = self.factory.parse_fulltext(data, version_id)
882
elif method == 'line-delta':
883
delta = self.factory.parse_line_delta(data, version_id)
884
content = content.copy()
885
content._lines = self._apply_delta(content._lines,
887
content_map[component_id] = content
889
if 'no-eol' in self._index.get_options(version_id):
890
content = content.copy()
891
line = content._lines[-1][1].rstrip('\n')
892
content._lines[-1] = (content._lines[-1][0], line)
893
final_content[version_id] = content
895
# digest here is the digest from the last applied component.
896
text = content.text()
897
if sha_strings(text) != digest:
898
raise KnitCorrupt(self.filename,
899
'sha-1 does not match %s' % version_id)
901
text_map[version_id] = text
902
return text_map, final_content
904
def iter_lines_added_or_present_in_versions(self, version_ids=None,
906
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
907
if version_ids is None:
908
version_ids = self.versions()
910
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
912
pb = progress.DummyProgress()
913
# we don't care about inclusions, the caller cares.
914
# but we need to setup a list of records to visit.
915
# we need version_id, position, length
916
version_id_records = []
917
requested_versions = set(version_ids)
918
# filter for available versions
919
for version_id in requested_versions:
920
if not self.has_version(version_id):
921
raise RevisionNotPresent(version_id, self.filename)
922
# get a in-component-order queue:
923
for version_id in self.versions():
924
if version_id in requested_versions:
925
data_pos, length = self._index.get_position(version_id)
926
version_id_records.append((version_id, data_pos, length))
928
total = len(version_id_records)
929
for version_idx, (version_id, data, sha_value) in \
930
enumerate(self._data.read_records_iter(version_id_records)):
931
pb.update('Walking content.', version_idx, total)
932
method = self._index.get_method(version_id)
934
assert method in ('fulltext', 'line-delta')
935
if method == 'fulltext':
936
line_iterator = self.factory.get_fulltext_content(data)
938
line_iterator = self.factory.get_linedelta_content(data)
939
for line in line_iterator:
942
pb.update('Walking content.', total, total)
944
def num_versions(self):
945
"""See VersionedFile.num_versions()."""
946
return self._index.num_versions()
948
__len__ = num_versions
950
def annotate_iter(self, version_id):
951
"""See VersionedFile.annotate_iter."""
952
version_id = osutils.safe_revision_id(version_id)
953
content = self._get_content(version_id)
954
for origin, text in content.annotate_iter():
957
def get_parents(self, version_id):
958
"""See VersionedFile.get_parents."""
961
# 52554 calls in 1264 872 internal down from 3674
962
version_id = osutils.safe_revision_id(version_id)
964
return self._index.get_parents(version_id)
966
raise RevisionNotPresent(version_id, self.filename)
968
def get_parents_with_ghosts(self, version_id):
969
"""See VersionedFile.get_parents."""
970
version_id = osutils.safe_revision_id(version_id)
972
return self._index.get_parents_with_ghosts(version_id)
974
raise RevisionNotPresent(version_id, self.filename)
976
def get_ancestry(self, versions, topo_sorted=True):
977
"""See VersionedFile.get_ancestry."""
978
if isinstance(versions, basestring):
979
versions = [versions]
982
versions = [osutils.safe_revision_id(v) for v in versions]
983
return self._index.get_ancestry(versions, topo_sorted)
985
def get_ancestry_with_ghosts(self, versions):
986
"""See VersionedFile.get_ancestry_with_ghosts."""
987
if isinstance(versions, basestring):
988
versions = [versions]
991
versions = [osutils.safe_revision_id(v) for v in versions]
992
return self._index.get_ancestry_with_ghosts(versions)
994
#@deprecated_method(zero_eight)
995
def walk(self, version_ids):
996
"""See VersionedFile.walk."""
997
# We take the short path here, and extract all relevant texts
998
# and put them in a weave and let that do all the work. Far
999
# from optimal, but is much simpler.
1000
# FIXME RB 20060228 this really is inefficient!
1001
from bzrlib.weave import Weave
1003
w = Weave(self.filename)
1004
ancestry = set(self.get_ancestry(version_ids, topo_sorted=False))
1005
sorted_graph = topo_sort(self._index.get_graph())
1006
version_list = [vid for vid in sorted_graph if vid in ancestry]
1008
for version_id in version_list:
1009
lines = self.get_lines(version_id)
1010
w.add_lines(version_id, self.get_parents(version_id), lines)
1012
for lineno, insert_id, dset, line in w.walk(version_ids):
1013
yield lineno, insert_id, dset, line
1015
def plan_merge(self, ver_a, ver_b):
1016
"""See VersionedFile.plan_merge."""
1017
ver_a = osutils.safe_revision_id(ver_a)
1018
ver_b = osutils.safe_revision_id(ver_b)
1019
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1020
def status_a(revision, text):
1021
if revision in ancestors_b:
1022
return 'killed-b', text
1024
return 'new-a', text
1026
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1027
def status_b(revision, text):
1028
if revision in ancestors_a:
1029
return 'killed-a', text
1031
return 'new-b', text
1033
annotated_a = self.annotate(ver_a)
1034
annotated_b = self.annotate(ver_b)
1035
plain_a = [t for (a, t) in annotated_a]
1036
plain_b = [t for (a, t) in annotated_b]
1037
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
1040
for ai, bi, l in blocks:
1041
# process all mismatched sections
1042
# (last mismatched section is handled because blocks always
1043
# includes a 0-length last block)
1044
for revision, text in annotated_a[a_cur:ai]:
1045
yield status_a(revision, text)
1046
for revision, text in annotated_b[b_cur:bi]:
1047
yield status_b(revision, text)
1049
# and now the matched section
1052
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
1053
assert text_a == text_b
1054
yield "unchanged", text_a
1057
class _KnitComponentFile(object):
1058
"""One of the files used to implement a knit database"""
1060
def __init__(self, transport, filename, mode, file_mode=None,
1061
create_parent_dir=False, dir_mode=None):
1062
self._transport = transport
1063
self._filename = filename
1065
self._file_mode = file_mode
1066
self._dir_mode = dir_mode
1067
self._create_parent_dir = create_parent_dir
1068
self._need_to_create = False
1070
def _full_path(self):
1071
"""Return the full path to this file."""
1072
return self._transport.base + self._filename
1074
def check_header(self, fp):
1075
line = fp.readline()
1077
# An empty file can actually be treated as though the file doesn't
1079
raise errors.NoSuchFile(self._full_path())
1080
if line != self.HEADER:
1081
raise KnitHeaderError(badline=line,
1082
filename=self._transport.abspath(self._filename))
1085
"""Commit is a nop."""
1088
return '%s(%s)' % (self.__class__.__name__, self._filename)
1091
class _KnitIndex(_KnitComponentFile):
1092
"""Manages knit index file.
1094
The index is already kept in memory and read on startup, to enable
1095
fast lookups of revision information. The cursor of the index
1096
file is always pointing to the end, making it easy to append
1099
_cache is a cache for fast mapping from version id to a Index
1102
_history is a cache for fast mapping from indexes to version ids.
1104
The index data format is dictionary compressed when it comes to
1105
parent references; a index entry may only have parents that with a
1106
lover index number. As a result, the index is topological sorted.
1108
Duplicate entries may be written to the index for a single version id
1109
if this is done then the latter one completely replaces the former:
1110
this allows updates to correct version and parent information.
1111
Note that the two entries may share the delta, and that successive
1112
annotations and references MUST point to the first entry.
1114
The index file on disc contains a header, followed by one line per knit
1115
record. The same revision can be present in an index file more than once.
1116
The first occurrence gets assigned a sequence number starting from 0.
1118
The format of a single line is
1119
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1120
REVISION_ID is a utf8-encoded revision id
1121
FLAGS is a comma separated list of flags about the record. Values include
1122
no-eol, line-delta, fulltext.
1123
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1124
that the the compressed data starts at.
1125
LENGTH is the ascii representation of the length of the data file.
1126
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1128
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1129
revision id already in the knit that is a parent of REVISION_ID.
1130
The ' :' marker is the end of record marker.
1133
when a write is interrupted to the index file, it will result in a line
1134
that does not end in ' :'. If the ' :' is not present at the end of a line,
1135
or at the end of the file, then the record that is missing it will be
1136
ignored by the parser.
1138
When writing new records to the index file, the data is preceded by '\n'
1139
to ensure that records always start on new lines even if the last write was
1140
interrupted. As a result its normal for the last line in the index to be
1141
missing a trailing newline. One can be added with no harmful effects.
1144
HEADER = "# bzr knit index 8\n"
1146
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1147
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1149
def _cache_version(self, version_id, options, pos, size, parents):
1150
"""Cache a version record in the history array and index cache.
1152
This is inlined into _load_data for performance. KEEP IN SYNC.
1153
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1156
# only want the _history index to reference the 1st index entry
1158
if version_id not in self._cache:
1159
index = len(self._history)
1160
self._history.append(version_id)
1162
index = self._cache[version_id][5]
1163
self._cache[version_id] = (version_id,
1170
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1171
create_parent_dir=False, delay_create=False, dir_mode=None):
1172
_KnitComponentFile.__init__(self, transport, filename, mode,
1173
file_mode=file_mode,
1174
create_parent_dir=create_parent_dir,
1177
# position in _history is the 'official' index for a revision
1178
# but the values may have come from a newer entry.
1179
# so - wc -l of a knit index is != the number of unique names
1183
fp = self._transport.get(self._filename)
1185
# _load_data may raise NoSuchFile if the target knit is
1191
if mode != 'w' or not create:
1194
self._need_to_create = True
1196
self._transport.put_bytes_non_atomic(
1197
self._filename, self.HEADER, mode=self._file_mode)
1199
def _load_data(self, fp):
1201
history = self._history
1203
self.check_header(fp)
1204
# readlines reads the whole file at once:
1205
# bad for transports like http, good for local disk
1206
# we save 60 ms doing this one change (
1207
# from calling readline each time to calling
1209
# probably what we want for nice behaviour on
1210
# http is a incremental readlines that yields, or
1211
# a check for local vs non local indexes,
1212
history_top = len(history) - 1
1213
for line in fp.readlines():
1215
if len(rec) < 5 or rec[-1] != ':':
1217
# FIXME: in the future we should determine if its a
1218
# short write - and ignore it
1219
# or a different failure, and raise. RBC 20060407
1223
for value in rec[4:-1]:
1225
# uncompressed reference
1226
parent_id = value[1:]
1228
parent_id = history[int(value)]
1229
parents.append(parent_id)
1231
version_id, options, pos, size = rec[:4]
1232
version_id = version_id
1234
# See self._cache_version
1235
# only want the _history index to reference the 1st
1236
# index entry for version_id
1237
if version_id not in cache:
1240
history.append(version_id)
1242
index = cache[version_id][5]
1243
cache[version_id] = (version_id,
1249
# end self._cache_version
1251
def get_graph(self):
1252
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1254
def get_ancestry(self, versions, topo_sorted=True):
1255
"""See VersionedFile.get_ancestry."""
1256
# get a graph of all the mentioned versions:
1258
pending = set(versions)
1261
version = pending.pop()
1264
parents = [p for p in cache[version][4] if p in cache]
1266
raise RevisionNotPresent(version, self._filename)
1267
# if not completed and not a ghost
1268
pending.update([p for p in parents if p not in graph])
1269
graph[version] = parents
1272
return topo_sort(graph.items())
1274
def get_ancestry_with_ghosts(self, versions):
1275
"""See VersionedFile.get_ancestry_with_ghosts."""
1276
# get a graph of all the mentioned versions:
1277
self.check_versions_present(versions)
1280
pending = set(versions)
1282
version = pending.pop()
1284
parents = cache[version][4]
1290
pending.update([p for p in parents if p not in graph])
1291
graph[version] = parents
1292
return topo_sort(graph.items())
1294
def num_versions(self):
1295
return len(self._history)
1297
__len__ = num_versions
1299
def get_versions(self):
1300
return self._history
1302
def idx_to_name(self, idx):
1303
return self._history[idx]
1305
def lookup(self, version_id):
1306
assert version_id in self._cache
1307
return self._cache[version_id][5]
1309
def _version_list_to_index(self, versions):
1312
for version in versions:
1313
if version in cache:
1314
# -- inlined lookup() --
1315
result_list.append(str(cache[version][5]))
1316
# -- end lookup () --
1318
result_list.append('.' + version)
1319
return ' '.join(result_list)
1321
def add_version(self, version_id, options, pos, size, parents):
1322
"""Add a version record to the index."""
1323
self.add_versions(((version_id, options, pos, size, parents),))
1325
def add_versions(self, versions):
1326
"""Add multiple versions to the index.
1328
:param versions: a list of tuples:
1329
(version_id, options, pos, size, parents).
1332
orig_history = self._history[:]
1333
orig_cache = self._cache.copy()
1336
for version_id, options, pos, size, parents in versions:
1337
line = "\n%s %s %s %s %s :" % (version_id,
1341
self._version_list_to_index(parents))
1342
assert isinstance(line, str), \
1343
'content must be utf-8 encoded: %r' % (line,)
1345
self._cache_version(version_id, options, pos, size, parents)
1346
if not self._need_to_create:
1347
self._transport.append_bytes(self._filename, ''.join(lines))
1350
sio.write(self.HEADER)
1351
sio.writelines(lines)
1353
self._transport.put_file_non_atomic(self._filename, sio,
1354
create_parent_dir=self._create_parent_dir,
1355
mode=self._file_mode,
1356
dir_mode=self._dir_mode)
1357
self._need_to_create = False
1359
# If any problems happen, restore the original values and re-raise
1360
self._history = orig_history
1361
self._cache = orig_cache
1364
def has_version(self, version_id):
1365
"""True if the version is in the index."""
1366
return version_id in self._cache
1368
def get_position(self, version_id):
1369
"""Return data position and size of specified version."""
1370
entry = self._cache[version_id]
1371
return entry[2], entry[3]
1373
def get_method(self, version_id):
1374
"""Return compression method of specified version."""
1375
options = self._cache[version_id][1]
1376
if 'fulltext' in options:
1379
if 'line-delta' not in options:
1380
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1383
def get_options(self, version_id):
1384
return self._cache[version_id][1]
1386
def get_parents(self, version_id):
1387
"""Return parents of specified version ignoring ghosts."""
1388
return [parent for parent in self._cache[version_id][4]
1389
if parent in self._cache]
1391
def get_parents_with_ghosts(self, version_id):
1392
"""Return parents of specified version with ghosts."""
1393
return self._cache[version_id][4]
1395
def check_versions_present(self, version_ids):
1396
"""Check that all specified versions are present."""
1398
for version_id in version_ids:
1399
if version_id not in cache:
1400
raise RevisionNotPresent(version_id, self._filename)
1403
class _KnitData(_KnitComponentFile):
1404
"""Contents of the knit data file"""
1406
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1407
create_parent_dir=False, delay_create=False,
1409
_KnitComponentFile.__init__(self, transport, filename, mode,
1410
file_mode=file_mode,
1411
create_parent_dir=create_parent_dir,
1413
self._checked = False
1414
# TODO: jam 20060713 conceptually, this could spill to disk
1415
# if the cached size gets larger than a certain amount
1416
# but it complicates the model a bit, so for now just use
1417
# a simple dictionary
1419
self._do_cache = False
1422
self._need_to_create = create
1424
self._transport.put_bytes_non_atomic(self._filename, '',
1425
mode=self._file_mode)
1427
def enable_cache(self):
1428
"""Enable caching of reads."""
1429
self._do_cache = True
1431
def clear_cache(self):
1432
"""Clear the record cache."""
1433
self._do_cache = False
1436
def _open_file(self):
1438
return self._transport.get(self._filename)
1443
def _record_to_data(self, version_id, digest, lines):
1444
"""Convert version_id, digest, lines into a raw data block.
1446
:return: (len, a StringIO instance with the raw data ready to read.)
1449
data_file = GzipFile(None, mode='wb', fileobj=sio)
1451
assert isinstance(version_id, str)
1452
data_file.writelines(chain(
1453
["version %s %d %s\n" % (version_id,
1457
["end %s\n" % version_id]))
1464
def add_raw_record(self, raw_data):
1465
"""Append a prepared record to the data file.
1467
:return: the offset in the data file raw_data was written.
1469
assert isinstance(raw_data, str), 'data must be plain bytes'
1470
if not self._need_to_create:
1471
return self._transport.append_bytes(self._filename, raw_data)
1473
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1474
create_parent_dir=self._create_parent_dir,
1475
mode=self._file_mode,
1476
dir_mode=self._dir_mode)
1477
self._need_to_create = False
1480
def add_record(self, version_id, digest, lines):
1481
"""Write new text record to disk. Returns the position in the
1482
file where it was written."""
1483
size, sio = self._record_to_data(version_id, digest, lines)
1485
if not self._need_to_create:
1486
start_pos = self._transport.append_file(self._filename, sio)
1488
self._transport.put_file_non_atomic(self._filename, sio,
1489
create_parent_dir=self._create_parent_dir,
1490
mode=self._file_mode,
1491
dir_mode=self._dir_mode)
1492
self._need_to_create = False
1495
self._cache[version_id] = sio.getvalue()
1496
return start_pos, size
1498
def _parse_record_header(self, version_id, raw_data):
1499
"""Parse a record header for consistency.
1501
:return: the header and the decompressor stream.
1502
as (stream, header_record)
1504
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1506
rec = self._check_header(version_id, df.readline())
1507
except Exception, e:
1508
raise KnitCorrupt(self._filename,
1509
"While reading {%s} got %s(%s)"
1510
% (version_id, e.__class__.__name__, str(e)))
1513
def _check_header(self, version_id, line):
1516
raise KnitCorrupt(self._filename,
1517
'unexpected number of elements in record header')
1518
if rec[1] != version_id:
1519
raise KnitCorrupt(self._filename,
1520
'unexpected version, wanted %r, got %r'
1521
% (version_id, rec[1]))
1524
def _parse_record(self, version_id, data):
1526
# 4168 calls in 2880 217 internal
1527
# 4168 calls to _parse_record_header in 2121
1528
# 4168 calls to readlines in 330
1529
df = GzipFile(mode='rb', fileobj=StringIO(data))
1532
record_contents = df.readlines()
1533
except Exception, e:
1534
raise KnitCorrupt(self._filename,
1535
"While reading {%s} got %s(%s)"
1536
% (version_id, e.__class__.__name__, str(e)))
1537
header = record_contents.pop(0)
1538
rec = self._check_header(version_id, header)
1540
last_line = record_contents.pop()
1541
if len(record_contents) != int(rec[2]):
1542
raise KnitCorrupt(self._filename,
1543
'incorrect number of lines %s != %s'
1545
% (len(record_contents), int(rec[2]),
1547
if last_line != 'end %s\n' % rec[1]:
1548
raise KnitCorrupt(self._filename,
1549
'unexpected version end line %r, wanted %r'
1550
% (last_line, version_id))
1552
return record_contents, rec[3]
1554
def read_records_iter_raw(self, records):
1555
"""Read text records from data file and yield raw data.
1557
This unpacks enough of the text record to validate the id is
1558
as expected but thats all.
1560
# setup an iterator of the external records:
1561
# uses readv so nice and fast we hope.
1563
# grab the disk data needed.
1565
# Don't check _cache if it is empty
1566
needed_offsets = [(pos, size) for version_id, pos, size
1568
if version_id not in self._cache]
1570
needed_offsets = [(pos, size) for version_id, pos, size
1573
raw_records = self._transport.readv(self._filename, needed_offsets)
1575
for version_id, pos, size in records:
1576
if version_id in self._cache:
1577
# This data has already been validated
1578
data = self._cache[version_id]
1580
pos, data = raw_records.next()
1582
self._cache[version_id] = data
1584
# validate the header
1585
df, rec = self._parse_record_header(version_id, data)
1587
yield version_id, data
1589
def read_records_iter(self, records):
1590
"""Read text records from data file and yield result.
1592
The result will be returned in whatever is the fastest to read.
1593
Not by the order requested. Also, multiple requests for the same
1594
record will only yield 1 response.
1595
:param records: A list of (version_id, pos, len) entries
1596
:return: Yields (version_id, contents, digest) in the order
1597
read, not the order requested
1603
# Skip records we have alread seen
1604
yielded_records = set()
1605
needed_records = set()
1606
for record in records:
1607
if record[0] in self._cache:
1608
if record[0] in yielded_records:
1610
yielded_records.add(record[0])
1611
data = self._cache[record[0]]
1612
content, digest = self._parse_record(record[0], data)
1613
yield (record[0], content, digest)
1615
needed_records.add(record)
1616
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1618
needed_records = sorted(set(records), key=operator.itemgetter(1))
1620
if not needed_records:
1623
# The transport optimizes the fetching as well
1624
# (ie, reads continuous ranges.)
1625
readv_response = self._transport.readv(self._filename,
1626
[(pos, size) for version_id, pos, size in needed_records])
1628
for (version_id, pos, size), (pos, data) in \
1629
izip(iter(needed_records), readv_response):
1630
content, digest = self._parse_record(version_id, data)
1632
self._cache[version_id] = data
1633
yield version_id, content, digest
1635
def read_records(self, records):
1636
"""Read records into a dictionary."""
1638
for record_id, content, digest in \
1639
self.read_records_iter(records):
1640
components[record_id] = (content, digest)
1644
class InterKnit(InterVersionedFile):
1645
"""Optimised code paths for knit to knit operations."""
1647
_matching_file_from_factory = KnitVersionedFile
1648
_matching_file_to_factory = KnitVersionedFile
1651
def is_compatible(source, target):
1652
"""Be compatible with knits. """
1654
return (isinstance(source, KnitVersionedFile) and
1655
isinstance(target, KnitVersionedFile))
1656
except AttributeError:
1659
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1660
"""See InterVersionedFile.join."""
1661
assert isinstance(self.source, KnitVersionedFile)
1662
assert isinstance(self.target, KnitVersionedFile)
1664
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1669
pb = ui.ui_factory.nested_progress_bar()
1671
version_ids = list(version_ids)
1672
if None in version_ids:
1673
version_ids.remove(None)
1675
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1676
this_versions = set(self.target._index.get_versions())
1677
needed_versions = self.source_ancestry - this_versions
1678
cross_check_versions = self.source_ancestry.intersection(this_versions)
1679
mismatched_versions = set()
1680
for version in cross_check_versions:
1681
# scan to include needed parents.
1682
n1 = set(self.target.get_parents_with_ghosts(version))
1683
n2 = set(self.source.get_parents_with_ghosts(version))
1685
# FIXME TEST this check for cycles being introduced works
1686
# the logic is we have a cycle if in our graph we are an
1687
# ancestor of any of the n2 revisions.
1693
parent_ancestors = self.source.get_ancestry(parent)
1694
if version in parent_ancestors:
1695
raise errors.GraphCycleError([parent, version])
1696
# ensure this parent will be available later.
1697
new_parents = n2.difference(n1)
1698
needed_versions.update(new_parents.difference(this_versions))
1699
mismatched_versions.add(version)
1701
if not needed_versions and not mismatched_versions:
1703
full_list = topo_sort(self.source.get_graph())
1705
version_list = [i for i in full_list if (not self.target.has_version(i)
1706
and i in needed_versions)]
1710
copy_queue_records = []
1712
for version_id in version_list:
1713
options = self.source._index.get_options(version_id)
1714
parents = self.source._index.get_parents_with_ghosts(version_id)
1715
# check that its will be a consistent copy:
1716
for parent in parents:
1717
# if source has the parent, we must :
1718
# * already have it or
1719
# * have it scheduled already
1720
# otherwise we don't care
1721
assert (self.target.has_version(parent) or
1722
parent in copy_set or
1723
not self.source.has_version(parent))
1724
data_pos, data_size = self.source._index.get_position(version_id)
1725
copy_queue_records.append((version_id, data_pos, data_size))
1726
copy_queue.append((version_id, options, parents))
1727
copy_set.add(version_id)
1729
# data suck the join:
1731
total = len(version_list)
1734
for (version_id, raw_data), \
1735
(version_id2, options, parents) in \
1736
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1738
assert version_id == version_id2, 'logic error, inconsistent results'
1740
pb.update("Joining knit", count, total)
1741
raw_records.append((version_id, options, parents, len(raw_data)))
1742
raw_datum.append(raw_data)
1743
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1745
for version in mismatched_versions:
1746
# FIXME RBC 20060309 is this needed?
1747
n1 = set(self.target.get_parents_with_ghosts(version))
1748
n2 = set(self.source.get_parents_with_ghosts(version))
1749
# write a combined record to our history preserving the current
1750
# parents as first in the list
1751
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1752
self.target.fix_parents(version, new_parents)
1758
InterVersionedFile.register_optimiser(InterKnit)
1761
class WeaveToKnit(InterVersionedFile):
1762
"""Optimised code paths for weave to knit operations."""
1764
_matching_file_from_factory = bzrlib.weave.WeaveFile
1765
_matching_file_to_factory = KnitVersionedFile
1768
def is_compatible(source, target):
1769
"""Be compatible with weaves to knits."""
1771
return (isinstance(source, bzrlib.weave.Weave) and
1772
isinstance(target, KnitVersionedFile))
1773
except AttributeError:
1776
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1777
"""See InterVersionedFile.join."""
1778
assert isinstance(self.source, bzrlib.weave.Weave)
1779
assert isinstance(self.target, KnitVersionedFile)
1781
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1786
pb = ui.ui_factory.nested_progress_bar()
1788
version_ids = list(version_ids)
1790
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1791
this_versions = set(self.target._index.get_versions())
1792
needed_versions = self.source_ancestry - this_versions
1793
cross_check_versions = self.source_ancestry.intersection(this_versions)
1794
mismatched_versions = set()
1795
for version in cross_check_versions:
1796
# scan to include needed parents.
1797
n1 = set(self.target.get_parents_with_ghosts(version))
1798
n2 = set(self.source.get_parents(version))
1799
# if all of n2's parents are in n1, then its fine.
1800
if n2.difference(n1):
1801
# FIXME TEST this check for cycles being introduced works
1802
# the logic is we have a cycle if in our graph we are an
1803
# ancestor of any of the n2 revisions.
1809
parent_ancestors = self.source.get_ancestry(parent)
1810
if version in parent_ancestors:
1811
raise errors.GraphCycleError([parent, version])
1812
# ensure this parent will be available later.
1813
new_parents = n2.difference(n1)
1814
needed_versions.update(new_parents.difference(this_versions))
1815
mismatched_versions.add(version)
1817
if not needed_versions and not mismatched_versions:
1819
full_list = topo_sort(self.source.get_graph())
1821
version_list = [i for i in full_list if (not self.target.has_version(i)
1822
and i in needed_versions)]
1826
total = len(version_list)
1827
for version_id in version_list:
1828
pb.update("Converting to knit", count, total)
1829
parents = self.source.get_parents(version_id)
1830
# check that its will be a consistent copy:
1831
for parent in parents:
1832
# if source has the parent, we must already have it
1833
assert (self.target.has_version(parent))
1834
self.target.add_lines(
1835
version_id, parents, self.source.get_lines(version_id))
1838
for version in mismatched_versions:
1839
# FIXME RBC 20060309 is this needed?
1840
n1 = set(self.target.get_parents_with_ghosts(version))
1841
n2 = set(self.source.get_parents(version))
1842
# write a combined record to our history preserving the current
1843
# parents as first in the list
1844
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1845
self.target.fix_parents(version, new_parents)
1851
InterVersionedFile.register_optimiser(WeaveToKnit)
1854
class KnitSequenceMatcher(difflib.SequenceMatcher):
1855
"""Knit tuned sequence matcher.
1857
This is based on profiling of difflib which indicated some improvements
1858
for our usage pattern.
1861
def find_longest_match(self, alo, ahi, blo, bhi):
1862
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1864
If isjunk is not defined:
1866
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1867
alo <= i <= i+k <= ahi
1868
blo <= j <= j+k <= bhi
1869
and for all (i',j',k') meeting those conditions,
1872
and if i == i', j <= j'
1874
In other words, of all maximal matching blocks, return one that
1875
starts earliest in a, and of all those maximal matching blocks that
1876
start earliest in a, return the one that starts earliest in b.
1878
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1879
>>> s.find_longest_match(0, 5, 0, 9)
1882
If isjunk is defined, first the longest matching block is
1883
determined as above, but with the additional restriction that no
1884
junk element appears in the block. Then that block is extended as
1885
far as possible by matching (only) junk elements on both sides. So
1886
the resulting block never matches on junk except as identical junk
1887
happens to be adjacent to an "interesting" match.
1889
Here's the same example as before, but considering blanks to be
1890
junk. That prevents " abcd" from matching the " abcd" at the tail
1891
end of the second sequence directly. Instead only the "abcd" can
1892
match, and matches the leftmost "abcd" in the second sequence:
1894
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1895
>>> s.find_longest_match(0, 5, 0, 9)
1898
If no blocks match, return (alo, blo, 0).
1900
>>> s = SequenceMatcher(None, "ab", "c")
1901
>>> s.find_longest_match(0, 2, 0, 1)
1905
# CAUTION: stripping common prefix or suffix would be incorrect.
1909
# Longest matching block is "ab", but if common prefix is
1910
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1911
# strip, so ends up claiming that ab is changed to acab by
1912
# inserting "ca" in the middle. That's minimal but unintuitive:
1913
# "it's obvious" that someone inserted "ac" at the front.
1914
# Windiff ends up at the same place as diff, but by pairing up
1915
# the unique 'b's and then matching the first two 'a's.
1917
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1918
besti, bestj, bestsize = alo, blo, 0
1919
# find longest junk-free match
1920
# during an iteration of the loop, j2len[j] = length of longest
1921
# junk-free match ending with a[i-1] and b[j]
1925
for i in xrange(alo, ahi):
1926
# look at all instances of a[i] in b; note that because
1927
# b2j has no junk keys, the loop is skipped if a[i] is junk
1928
j2lenget = j2len.get
1931
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1932
# following improvement
1933
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1934
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1935
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1937
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1938
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1939
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1951
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1953
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1956
# Extend the best by non-junk elements on each end. In particular,
1957
# "popular" non-junk elements aren't in b2j, which greatly speeds
1958
# the inner loop above, but also means "the best" match so far
1959
# doesn't contain any junk *or* popular non-junk elements.
1960
while besti > alo and bestj > blo and \
1961
not isbjunk(b[bestj-1]) and \
1962
a[besti-1] == b[bestj-1]:
1963
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1964
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1965
not isbjunk(b[bestj+bestsize]) and \
1966
a[besti+bestsize] == b[bestj+bestsize]:
1969
# Now that we have a wholly interesting match (albeit possibly
1970
# empty!), we may as well suck up the matching junk on each
1971
# side of it too. Can't think of a good reason not to, and it
1972
# saves post-processing the (possibly considerable) expense of
1973
# figuring out what to do with it. In the case of an empty
1974
# interesting match, this is clearly the right thing to do,
1975
# because no other kind of match is possible in the regions.
1976
while besti > alo and bestj > blo and \
1977
isbjunk(b[bestj-1]) and \
1978
a[besti-1] == b[bestj-1]:
1979
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1980
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1981
isbjunk(b[bestj+bestsize]) and \
1982
a[besti+bestsize] == b[bestj+bestsize]:
1983
bestsize = bestsize + 1
1985
return besti, bestj, bestsize
added
removed
bzrlib/knit.py
