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
return self.get_sha1s([version_id])[0]
591
def get_sha1s(self, version_ids):
592
"""See VersionedFile.get_sha1()."""
593
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
594
record_map = self._get_record_map(version_ids)
595
# record entry 2 is the 'digest'.
596
return [record_map[v][2] for v in version_ids]
600
"""See VersionedFile.get_suffixes()."""
601
return [DATA_SUFFIX, INDEX_SUFFIX]
603
def has_ghost(self, version_id):
604
"""True if there is a ghost reference in the file to version_id."""
605
version_id = osutils.safe_revision_id(version_id)
607
if self.has_version(version_id):
609
# optimisable if needed by memoising the _ghosts set.
610
items = self._index.get_graph()
611
for node, parents in items:
612
for parent in parents:
613
if parent not in self._index._cache:
614
if parent == version_id:
619
"""See VersionedFile.versions."""
620
return self._index.get_versions()
622
def has_version(self, version_id):
623
"""See VersionedFile.has_version."""
624
version_id = osutils.safe_revision_id(version_id)
625
return self._index.has_version(version_id)
627
__contains__ = has_version
629
def _merge_annotations(self, content, parents, parent_texts={},
630
delta=None, annotated=None):
631
"""Merge annotations for content. This is done by comparing
632
the annotations based on changed to the text.
636
for parent_id in parents:
637
merge_content = self._get_content(parent_id, parent_texts)
638
seq = patiencediff.PatienceSequenceMatcher(
639
None, merge_content.text(), content.text())
640
if delta_seq is None:
641
# setup a delta seq to reuse.
643
for i, j, n in seq.get_matching_blocks():
646
# this appears to copy (origin, text) pairs across to the new
647
# content for any line that matches the last-checked parent.
648
# FIXME: save the sequence control data for delta compression
649
# against the most relevant parent rather than rediffing.
650
content._lines[j:j+n] = merge_content._lines[i:i+n]
653
reference_content = self._get_content(parents[0], parent_texts)
654
new_texts = content.text()
655
old_texts = reference_content.text()
656
delta_seq = patiencediff.PatienceSequenceMatcher(
657
None, old_texts, new_texts)
658
return self._make_line_delta(delta_seq, content)
660
def _make_line_delta(self, delta_seq, new_content):
661
"""Generate a line delta from delta_seq and new_content."""
663
for op in delta_seq.get_opcodes():
666
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
669
def _get_components_positions(self, version_ids):
670
"""Produce a map of position data for the components of versions.
672
This data is intended to be used for retrieving the knit records.
674
A dict of version_id to (method, data_pos, data_size, next) is
676
method is the way referenced data should be applied.
677
data_pos is the position of the data in the knit.
678
data_size is the size of the data in the knit.
679
next is the build-parent of the version, or None for fulltexts.
682
for version_id in version_ids:
685
while cursor is not None and cursor not in component_data:
686
method = self._index.get_method(cursor)
687
if method == 'fulltext':
690
next = self.get_parents(cursor)[0]
691
data_pos, data_size = self._index.get_position(cursor)
692
component_data[cursor] = (method, data_pos, data_size, next)
694
return component_data
696
def _get_content(self, version_id, parent_texts={}):
697
"""Returns a content object that makes up the specified
699
if not self.has_version(version_id):
700
raise RevisionNotPresent(version_id, self.filename)
702
cached_version = parent_texts.get(version_id, None)
703
if cached_version is not None:
704
return cached_version
706
text_map, contents_map = self._get_content_maps([version_id])
707
return contents_map[version_id]
709
def _check_versions_present(self, version_ids):
710
"""Check that all specified versions are present."""
711
self._index.check_versions_present(version_ids)
713
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
714
"""See VersionedFile.add_lines_with_ghosts()."""
715
self._check_add(version_id, lines)
716
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
718
def _add_lines(self, version_id, parents, lines, parent_texts):
719
"""See VersionedFile.add_lines."""
720
self._check_add(version_id, lines)
721
self._check_versions_present(parents)
722
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
724
def _check_add(self, version_id, lines):
725
"""check that version_id and lines are safe to add."""
726
assert self.writable, "knit is not opened for write"
727
### FIXME escape. RBC 20060228
728
if contains_whitespace(version_id):
729
raise InvalidRevisionId(version_id, self.filename)
730
self.check_not_reserved_id(version_id)
731
if self.has_version(version_id):
732
raise RevisionAlreadyPresent(version_id, self.filename)
733
self._check_lines_not_unicode(lines)
734
self._check_lines_are_lines(lines)
736
def _add(self, version_id, lines, parents, delta, parent_texts):
737
"""Add a set of lines on top of version specified by parents.
739
If delta is true, compress the text as a line-delta against
742
Any versions not present will be converted into ghosts.
744
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
745
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
746
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
747
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
748
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
749
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
750
# +1383 0 8.0370 8.0370 +<len>
751
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
752
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
753
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
754
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
758
if parent_texts is None:
760
for parent in parents:
761
if not self.has_version(parent):
762
ghosts.append(parent)
764
present_parents.append(parent)
766
if delta and not len(present_parents):
769
digest = sha_strings(lines)
772
if lines[-1][-1] != '\n':
773
options.append('no-eol')
774
lines[-1] = lines[-1] + '\n'
776
if len(present_parents) and delta:
777
# To speed the extract of texts the delta chain is limited
778
# to a fixed number of deltas. This should minimize both
779
# I/O and the time spend applying deltas.
780
delta = self._check_should_delta(present_parents)
782
assert isinstance(version_id, str)
783
lines = self.factory.make(lines, version_id)
784
if delta or (self.factory.annotated and len(present_parents) > 0):
785
# Merge annotations from parent texts if so is needed.
786
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
787
delta, self.factory.annotated)
790
options.append('line-delta')
791
store_lines = self.factory.lower_line_delta(delta_hunks)
793
options.append('fulltext')
794
store_lines = self.factory.lower_fulltext(lines)
796
where, size = self._data.add_record(version_id, digest, store_lines)
797
self._index.add_version(version_id, options, where, size, parents)
800
def check(self, progress_bar=None):
801
"""See VersionedFile.check()."""
803
def _clone_text(self, new_version_id, old_version_id, parents):
804
"""See VersionedFile.clone_text()."""
805
# FIXME RBC 20060228 make fast by only inserting an index with null
807
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
809
def get_lines(self, version_id):
810
"""See VersionedFile.get_lines()."""
811
return self.get_line_list([version_id])[0]
813
def _get_record_map(self, version_ids):
814
"""Produce a dictionary of knit records.
816
The keys are version_ids, the values are tuples of (method, content,
818
method is the way the content should be applied.
819
content is a KnitContent object.
820
digest is the SHA1 digest of this version id after all steps are done
821
next is the build-parent of the version, i.e. the leftmost ancestor.
822
If the method is fulltext, next will be None.
824
position_map = self._get_components_positions(version_ids)
825
# c = component_id, m = method, p = position, s = size, n = next
826
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
828
for component_id, content, digest in \
829
self._data.read_records_iter(records):
830
method, position, size, next = position_map[component_id]
831
record_map[component_id] = method, content, digest, next
835
def get_text(self, version_id):
836
"""See VersionedFile.get_text"""
837
return self.get_texts([version_id])[0]
839
def get_texts(self, version_ids):
840
return [''.join(l) for l in self.get_line_list(version_ids)]
842
def get_line_list(self, version_ids):
843
"""Return the texts of listed versions as a list of strings."""
844
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
845
for version_id in version_ids:
846
self.check_not_reserved_id(version_id)
847
text_map, content_map = self._get_content_maps(version_ids)
848
return [text_map[v] for v in version_ids]
850
_get_lf_split_line_list = get_line_list
852
def _get_content_maps(self, version_ids):
853
"""Produce maps of text and KnitContents
855
:return: (text_map, content_map) where text_map contains the texts for
856
the requested versions and content_map contains the KnitContents.
857
Both dicts take version_ids as their keys.
859
for version_id in version_ids:
860
if not self.has_version(version_id):
861
raise RevisionNotPresent(version_id, self.filename)
862
record_map = self._get_record_map(version_ids)
867
for version_id in version_ids:
870
while cursor is not None:
871
method, data, digest, next = record_map[cursor]
872
components.append((cursor, method, data, digest))
873
if cursor in content_map:
878
for component_id, method, data, digest in reversed(components):
879
if component_id in content_map:
880
content = content_map[component_id]
882
if method == 'fulltext':
883
assert content is None
884
content = self.factory.parse_fulltext(data, version_id)
885
elif method == 'line-delta':
886
delta = self.factory.parse_line_delta(data, version_id)
887
content = content.copy()
888
content._lines = self._apply_delta(content._lines,
890
content_map[component_id] = content
892
if 'no-eol' in self._index.get_options(version_id):
893
content = content.copy()
894
line = content._lines[-1][1].rstrip('\n')
895
content._lines[-1] = (content._lines[-1][0], line)
896
final_content[version_id] = content
898
# digest here is the digest from the last applied component.
899
text = content.text()
900
if sha_strings(text) != digest:
901
raise KnitCorrupt(self.filename,
902
'sha-1 does not match %s' % version_id)
904
text_map[version_id] = text
905
return text_map, final_content
907
def iter_lines_added_or_present_in_versions(self, version_ids=None,
909
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
910
if version_ids is None:
911
version_ids = self.versions()
913
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
915
pb = progress.DummyProgress()
916
# we don't care about inclusions, the caller cares.
917
# but we need to setup a list of records to visit.
918
# we need version_id, position, length
919
version_id_records = []
920
requested_versions = set(version_ids)
921
# filter for available versions
922
for version_id in requested_versions:
923
if not self.has_version(version_id):
924
raise RevisionNotPresent(version_id, self.filename)
925
# get a in-component-order queue:
926
for version_id in self.versions():
927
if version_id in requested_versions:
928
data_pos, length = self._index.get_position(version_id)
929
version_id_records.append((version_id, data_pos, length))
931
total = len(version_id_records)
932
for version_idx, (version_id, data, sha_value) in \
933
enumerate(self._data.read_records_iter(version_id_records)):
934
pb.update('Walking content.', version_idx, total)
935
method = self._index.get_method(version_id)
937
assert method in ('fulltext', 'line-delta')
938
if method == 'fulltext':
939
line_iterator = self.factory.get_fulltext_content(data)
941
line_iterator = self.factory.get_linedelta_content(data)
942
for line in line_iterator:
945
pb.update('Walking content.', total, total)
947
def num_versions(self):
948
"""See VersionedFile.num_versions()."""
949
return self._index.num_versions()
951
__len__ = num_versions
953
def annotate_iter(self, version_id):
954
"""See VersionedFile.annotate_iter."""
955
version_id = osutils.safe_revision_id(version_id)
956
content = self._get_content(version_id)
957
for origin, text in content.annotate_iter():
960
def get_parents(self, version_id):
961
"""See VersionedFile.get_parents."""
964
# 52554 calls in 1264 872 internal down from 3674
965
version_id = osutils.safe_revision_id(version_id)
967
return self._index.get_parents(version_id)
969
raise RevisionNotPresent(version_id, self.filename)
971
def get_parents_with_ghosts(self, version_id):
972
"""See VersionedFile.get_parents."""
973
version_id = osutils.safe_revision_id(version_id)
975
return self._index.get_parents_with_ghosts(version_id)
977
raise RevisionNotPresent(version_id, self.filename)
979
def get_ancestry(self, versions, topo_sorted=True):
980
"""See VersionedFile.get_ancestry."""
981
if isinstance(versions, basestring):
982
versions = [versions]
985
versions = [osutils.safe_revision_id(v) for v in versions]
986
return self._index.get_ancestry(versions, topo_sorted)
988
def get_ancestry_with_ghosts(self, versions):
989
"""See VersionedFile.get_ancestry_with_ghosts."""
990
if isinstance(versions, basestring):
991
versions = [versions]
994
versions = [osutils.safe_revision_id(v) for v in versions]
995
return self._index.get_ancestry_with_ghosts(versions)
997
#@deprecated_method(zero_eight)
998
def walk(self, version_ids):
999
"""See VersionedFile.walk."""
1000
# We take the short path here, and extract all relevant texts
1001
# and put them in a weave and let that do all the work. Far
1002
# from optimal, but is much simpler.
1003
# FIXME RB 20060228 this really is inefficient!
1004
from bzrlib.weave import Weave
1006
w = Weave(self.filename)
1007
ancestry = set(self.get_ancestry(version_ids, topo_sorted=False))
1008
sorted_graph = topo_sort(self._index.get_graph())
1009
version_list = [vid for vid in sorted_graph if vid in ancestry]
1011
for version_id in version_list:
1012
lines = self.get_lines(version_id)
1013
w.add_lines(version_id, self.get_parents(version_id), lines)
1015
for lineno, insert_id, dset, line in w.walk(version_ids):
1016
yield lineno, insert_id, dset, line
1018
def plan_merge(self, ver_a, ver_b):
1019
"""See VersionedFile.plan_merge."""
1020
ver_a = osutils.safe_revision_id(ver_a)
1021
ver_b = osutils.safe_revision_id(ver_b)
1022
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1023
def status_a(revision, text):
1024
if revision in ancestors_b:
1025
return 'killed-b', text
1027
return 'new-a', text
1029
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1030
def status_b(revision, text):
1031
if revision in ancestors_a:
1032
return 'killed-a', text
1034
return 'new-b', text
1036
annotated_a = self.annotate(ver_a)
1037
annotated_b = self.annotate(ver_b)
1038
plain_a = [t for (a, t) in annotated_a]
1039
plain_b = [t for (a, t) in annotated_b]
1040
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
1043
for ai, bi, l in blocks:
1044
# process all mismatched sections
1045
# (last mismatched section is handled because blocks always
1046
# includes a 0-length last block)
1047
for revision, text in annotated_a[a_cur:ai]:
1048
yield status_a(revision, text)
1049
for revision, text in annotated_b[b_cur:bi]:
1050
yield status_b(revision, text)
1052
# and now the matched section
1055
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
1056
assert text_a == text_b
1057
yield "unchanged", text_a
1060
class _KnitComponentFile(object):
1061
"""One of the files used to implement a knit database"""
1063
def __init__(self, transport, filename, mode, file_mode=None,
1064
create_parent_dir=False, dir_mode=None):
1065
self._transport = transport
1066
self._filename = filename
1068
self._file_mode = file_mode
1069
self._dir_mode = dir_mode
1070
self._create_parent_dir = create_parent_dir
1071
self._need_to_create = False
1073
def _full_path(self):
1074
"""Return the full path to this file."""
1075
return self._transport.base + self._filename
1077
def check_header(self, fp):
1078
line = fp.readline()
1080
# An empty file can actually be treated as though the file doesn't
1082
raise errors.NoSuchFile(self._full_path())
1083
if line != self.HEADER:
1084
raise KnitHeaderError(badline=line,
1085
filename=self._transport.abspath(self._filename))
1088
"""Commit is a nop."""
1091
return '%s(%s)' % (self.__class__.__name__, self._filename)
1094
class _KnitIndex(_KnitComponentFile):
1095
"""Manages knit index file.
1097
The index is already kept in memory and read on startup, to enable
1098
fast lookups of revision information. The cursor of the index
1099
file is always pointing to the end, making it easy to append
1102
_cache is a cache for fast mapping from version id to a Index
1105
_history is a cache for fast mapping from indexes to version ids.
1107
The index data format is dictionary compressed when it comes to
1108
parent references; a index entry may only have parents that with a
1109
lover index number. As a result, the index is topological sorted.
1111
Duplicate entries may be written to the index for a single version id
1112
if this is done then the latter one completely replaces the former:
1113
this allows updates to correct version and parent information.
1114
Note that the two entries may share the delta, and that successive
1115
annotations and references MUST point to the first entry.
1117
The index file on disc contains a header, followed by one line per knit
1118
record. The same revision can be present in an index file more than once.
1119
The first occurrence gets assigned a sequence number starting from 0.
1121
The format of a single line is
1122
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1123
REVISION_ID is a utf8-encoded revision id
1124
FLAGS is a comma separated list of flags about the record. Values include
1125
no-eol, line-delta, fulltext.
1126
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1127
that the the compressed data starts at.
1128
LENGTH is the ascii representation of the length of the data file.
1129
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1131
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1132
revision id already in the knit that is a parent of REVISION_ID.
1133
The ' :' marker is the end of record marker.
1136
when a write is interrupted to the index file, it will result in a line
1137
that does not end in ' :'. If the ' :' is not present at the end of a line,
1138
or at the end of the file, then the record that is missing it will be
1139
ignored by the parser.
1141
When writing new records to the index file, the data is preceded by '\n'
1142
to ensure that records always start on new lines even if the last write was
1143
interrupted. As a result its normal for the last line in the index to be
1144
missing a trailing newline. One can be added with no harmful effects.
1147
HEADER = "# bzr knit index 8\n"
1149
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1150
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1152
def _cache_version(self, version_id, options, pos, size, parents):
1153
"""Cache a version record in the history array and index cache.
1155
This is inlined into _load_data for performance. KEEP IN SYNC.
1156
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1159
# only want the _history index to reference the 1st index entry
1161
if version_id not in self._cache:
1162
index = len(self._history)
1163
self._history.append(version_id)
1165
index = self._cache[version_id][5]
1166
self._cache[version_id] = (version_id,
1173
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1174
create_parent_dir=False, delay_create=False, dir_mode=None):
1175
_KnitComponentFile.__init__(self, transport, filename, mode,
1176
file_mode=file_mode,
1177
create_parent_dir=create_parent_dir,
1180
# position in _history is the 'official' index for a revision
1181
# but the values may have come from a newer entry.
1182
# so - wc -l of a knit index is != the number of unique names
1186
fp = self._transport.get(self._filename)
1188
# _load_data may raise NoSuchFile if the target knit is
1194
if mode != 'w' or not create:
1197
self._need_to_create = True
1199
self._transport.put_bytes_non_atomic(
1200
self._filename, self.HEADER, mode=self._file_mode)
1202
def _load_data(self, fp):
1204
history = self._history
1206
self.check_header(fp)
1207
# readlines reads the whole file at once:
1208
# bad for transports like http, good for local disk
1209
# we save 60 ms doing this one change (
1210
# from calling readline each time to calling
1212
# probably what we want for nice behaviour on
1213
# http is a incremental readlines that yields, or
1214
# a check for local vs non local indexes,
1215
history_top = len(history) - 1
1216
for line in fp.readlines():
1218
if len(rec) < 5 or rec[-1] != ':':
1220
# FIXME: in the future we should determine if its a
1221
# short write - and ignore it
1222
# or a different failure, and raise. RBC 20060407
1227
for value in rec[4:-1]:
1229
# uncompressed reference
1230
parent_id = value[1:]
1232
parent_id = history[int(value)]
1233
parents.append(parent_id)
1234
except (IndexError, ValueError), e:
1235
# The parent could not be decoded to get its parent row. This
1236
# at a minimum will cause this row to have wrong parents, or
1237
# even to apply a delta to the wrong base and decode
1238
# incorrectly. its therefore not usable, and because we have
1239
# encountered a situation where a new knit index had this
1240
# corrupt we can't asssume that no other rows referring to the
1241
# index of this record actually mean the subsequent uncorrupt
1243
raise errors.KnitCorrupt(self._filename,
1244
"line %r: %s" % (rec, e))
1246
version_id, options, pos, size = rec[:4]
1247
version_id = version_id
1249
# See self._cache_version
1250
# only want the _history index to reference the 1st
1251
# index entry for version_id
1252
if version_id not in cache:
1255
history.append(version_id)
1257
index = cache[version_id][5]
1258
cache[version_id] = (version_id,
1264
# end self._cache_version
1266
def get_graph(self):
1267
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1269
def get_ancestry(self, versions, topo_sorted=True):
1270
"""See VersionedFile.get_ancestry."""
1271
# get a graph of all the mentioned versions:
1273
pending = set(versions)
1276
version = pending.pop()
1279
parents = [p for p in cache[version][4] if p in cache]
1281
raise RevisionNotPresent(version, self._filename)
1282
# if not completed and not a ghost
1283
pending.update([p for p in parents if p not in graph])
1284
graph[version] = parents
1287
return topo_sort(graph.items())
1289
def get_ancestry_with_ghosts(self, versions):
1290
"""See VersionedFile.get_ancestry_with_ghosts."""
1291
# get a graph of all the mentioned versions:
1292
self.check_versions_present(versions)
1295
pending = set(versions)
1297
version = pending.pop()
1299
parents = cache[version][4]
1305
pending.update([p for p in parents if p not in graph])
1306
graph[version] = parents
1307
return topo_sort(graph.items())
1309
def num_versions(self):
1310
return len(self._history)
1312
__len__ = num_versions
1314
def get_versions(self):
1315
return self._history
1317
def idx_to_name(self, idx):
1318
return self._history[idx]
1320
def lookup(self, version_id):
1321
assert version_id in self._cache
1322
return self._cache[version_id][5]
1324
def _version_list_to_index(self, versions):
1327
for version in versions:
1328
if version in cache:
1329
# -- inlined lookup() --
1330
result_list.append(str(cache[version][5]))
1331
# -- end lookup () --
1333
result_list.append('.' + version)
1334
return ' '.join(result_list)
1336
def add_version(self, version_id, options, pos, size, parents):
1337
"""Add a version record to the index."""
1338
self.add_versions(((version_id, options, pos, size, parents),))
1340
def add_versions(self, versions):
1341
"""Add multiple versions to the index.
1343
:param versions: a list of tuples:
1344
(version_id, options, pos, size, parents).
1347
orig_history = self._history[:]
1348
orig_cache = self._cache.copy()
1351
for version_id, options, pos, size, parents in versions:
1352
line = "\n%s %s %s %s %s :" % (version_id,
1356
self._version_list_to_index(parents))
1357
assert isinstance(line, str), \
1358
'content must be utf-8 encoded: %r' % (line,)
1360
self._cache_version(version_id, options, pos, size, parents)
1361
if not self._need_to_create:
1362
self._transport.append_bytes(self._filename, ''.join(lines))
1365
sio.write(self.HEADER)
1366
sio.writelines(lines)
1368
self._transport.put_file_non_atomic(self._filename, sio,
1369
create_parent_dir=self._create_parent_dir,
1370
mode=self._file_mode,
1371
dir_mode=self._dir_mode)
1372
self._need_to_create = False
1374
# If any problems happen, restore the original values and re-raise
1375
self._history = orig_history
1376
self._cache = orig_cache
1379
def has_version(self, version_id):
1380
"""True if the version is in the index."""
1381
return version_id in self._cache
1383
def get_position(self, version_id):
1384
"""Return data position and size of specified version."""
1385
entry = self._cache[version_id]
1386
return entry[2], entry[3]
1388
def get_method(self, version_id):
1389
"""Return compression method of specified version."""
1390
options = self._cache[version_id][1]
1391
if 'fulltext' in options:
1394
if 'line-delta' not in options:
1395
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1398
def get_options(self, version_id):
1399
return self._cache[version_id][1]
1401
def get_parents(self, version_id):
1402
"""Return parents of specified version ignoring ghosts."""
1403
return [parent for parent in self._cache[version_id][4]
1404
if parent in self._cache]
1406
def get_parents_with_ghosts(self, version_id):
1407
"""Return parents of specified version with ghosts."""
1408
return self._cache[version_id][4]
1410
def check_versions_present(self, version_ids):
1411
"""Check that all specified versions are present."""
1413
for version_id in version_ids:
1414
if version_id not in cache:
1415
raise RevisionNotPresent(version_id, self._filename)
1418
class _KnitData(_KnitComponentFile):
1419
"""Contents of the knit data file"""
1421
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1422
create_parent_dir=False, delay_create=False,
1424
_KnitComponentFile.__init__(self, transport, filename, mode,
1425
file_mode=file_mode,
1426
create_parent_dir=create_parent_dir,
1428
self._checked = False
1429
# TODO: jam 20060713 conceptually, this could spill to disk
1430
# if the cached size gets larger than a certain amount
1431
# but it complicates the model a bit, so for now just use
1432
# a simple dictionary
1434
self._do_cache = False
1437
self._need_to_create = create
1439
self._transport.put_bytes_non_atomic(self._filename, '',
1440
mode=self._file_mode)
1442
def enable_cache(self):
1443
"""Enable caching of reads."""
1444
self._do_cache = True
1446
def clear_cache(self):
1447
"""Clear the record cache."""
1448
self._do_cache = False
1451
def _open_file(self):
1453
return self._transport.get(self._filename)
1458
def _record_to_data(self, version_id, digest, lines):
1459
"""Convert version_id, digest, lines into a raw data block.
1461
:return: (len, a StringIO instance with the raw data ready to read.)
1464
data_file = GzipFile(None, mode='wb', fileobj=sio)
1466
assert isinstance(version_id, str)
1467
data_file.writelines(chain(
1468
["version %s %d %s\n" % (version_id,
1472
["end %s\n" % version_id]))
1479
def add_raw_record(self, raw_data):
1480
"""Append a prepared record to the data file.
1482
:return: the offset in the data file raw_data was written.
1484
assert isinstance(raw_data, str), 'data must be plain bytes'
1485
if not self._need_to_create:
1486
return self._transport.append_bytes(self._filename, raw_data)
1488
self._transport.put_bytes_non_atomic(self._filename, raw_data,
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
def add_record(self, version_id, digest, lines):
1496
"""Write new text record to disk. Returns the position in the
1497
file where it was written."""
1498
size, sio = self._record_to_data(version_id, digest, lines)
1500
if not self._need_to_create:
1501
start_pos = self._transport.append_file(self._filename, sio)
1503
self._transport.put_file_non_atomic(self._filename, sio,
1504
create_parent_dir=self._create_parent_dir,
1505
mode=self._file_mode,
1506
dir_mode=self._dir_mode)
1507
self._need_to_create = False
1510
self._cache[version_id] = sio.getvalue()
1511
return start_pos, size
1513
def _parse_record_header(self, version_id, raw_data):
1514
"""Parse a record header for consistency.
1516
:return: the header and the decompressor stream.
1517
as (stream, header_record)
1519
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1521
rec = self._check_header(version_id, df.readline())
1522
except Exception, e:
1523
raise KnitCorrupt(self._filename,
1524
"While reading {%s} got %s(%s)"
1525
% (version_id, e.__class__.__name__, str(e)))
1528
def _check_header(self, version_id, line):
1531
raise KnitCorrupt(self._filename,
1532
'unexpected number of elements in record header')
1533
if rec[1] != version_id:
1534
raise KnitCorrupt(self._filename,
1535
'unexpected version, wanted %r, got %r'
1536
% (version_id, rec[1]))
1539
def _parse_record(self, version_id, data):
1541
# 4168 calls in 2880 217 internal
1542
# 4168 calls to _parse_record_header in 2121
1543
# 4168 calls to readlines in 330
1544
df = GzipFile(mode='rb', fileobj=StringIO(data))
1547
record_contents = df.readlines()
1548
except Exception, e:
1549
raise KnitCorrupt(self._filename,
1550
"While reading {%s} got %s(%s)"
1551
% (version_id, e.__class__.__name__, str(e)))
1552
header = record_contents.pop(0)
1553
rec = self._check_header(version_id, header)
1555
last_line = record_contents.pop()
1556
if len(record_contents) != int(rec[2]):
1557
raise KnitCorrupt(self._filename,
1558
'incorrect number of lines %s != %s'
1560
% (len(record_contents), int(rec[2]),
1562
if last_line != 'end %s\n' % rec[1]:
1563
raise KnitCorrupt(self._filename,
1564
'unexpected version end line %r, wanted %r'
1565
% (last_line, version_id))
1567
return record_contents, rec[3]
1569
def read_records_iter_raw(self, records):
1570
"""Read text records from data file and yield raw data.
1572
This unpacks enough of the text record to validate the id is
1573
as expected but thats all.
1575
# setup an iterator of the external records:
1576
# uses readv so nice and fast we hope.
1578
# grab the disk data needed.
1580
# Don't check _cache if it is empty
1581
needed_offsets = [(pos, size) for version_id, pos, size
1583
if version_id not in self._cache]
1585
needed_offsets = [(pos, size) for version_id, pos, size
1588
raw_records = self._transport.readv(self._filename, needed_offsets)
1590
for version_id, pos, size in records:
1591
if version_id in self._cache:
1592
# This data has already been validated
1593
data = self._cache[version_id]
1595
pos, data = raw_records.next()
1597
self._cache[version_id] = data
1599
# validate the header
1600
df, rec = self._parse_record_header(version_id, data)
1602
yield version_id, data
1604
def read_records_iter(self, records):
1605
"""Read text records from data file and yield result.
1607
The result will be returned in whatever is the fastest to read.
1608
Not by the order requested. Also, multiple requests for the same
1609
record will only yield 1 response.
1610
:param records: A list of (version_id, pos, len) entries
1611
:return: Yields (version_id, contents, digest) in the order
1612
read, not the order requested
1618
# Skip records we have alread seen
1619
yielded_records = set()
1620
needed_records = set()
1621
for record in records:
1622
if record[0] in self._cache:
1623
if record[0] in yielded_records:
1625
yielded_records.add(record[0])
1626
data = self._cache[record[0]]
1627
content, digest = self._parse_record(record[0], data)
1628
yield (record[0], content, digest)
1630
needed_records.add(record)
1631
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1633
needed_records = sorted(set(records), key=operator.itemgetter(1))
1635
if not needed_records:
1638
# The transport optimizes the fetching as well
1639
# (ie, reads continuous ranges.)
1640
readv_response = self._transport.readv(self._filename,
1641
[(pos, size) for version_id, pos, size in needed_records])
1643
for (version_id, pos, size), (pos, data) in \
1644
izip(iter(needed_records), readv_response):
1645
content, digest = self._parse_record(version_id, data)
1647
self._cache[version_id] = data
1648
yield version_id, content, digest
1650
def read_records(self, records):
1651
"""Read records into a dictionary."""
1653
for record_id, content, digest in \
1654
self.read_records_iter(records):
1655
components[record_id] = (content, digest)
1659
class InterKnit(InterVersionedFile):
1660
"""Optimised code paths for knit to knit operations."""
1662
_matching_file_from_factory = KnitVersionedFile
1663
_matching_file_to_factory = KnitVersionedFile
1666
def is_compatible(source, target):
1667
"""Be compatible with knits. """
1669
return (isinstance(source, KnitVersionedFile) and
1670
isinstance(target, KnitVersionedFile))
1671
except AttributeError:
1674
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1675
"""See InterVersionedFile.join."""
1676
assert isinstance(self.source, KnitVersionedFile)
1677
assert isinstance(self.target, KnitVersionedFile)
1679
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1684
pb = ui.ui_factory.nested_progress_bar()
1686
version_ids = list(version_ids)
1687
if None in version_ids:
1688
version_ids.remove(None)
1690
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1691
this_versions = set(self.target._index.get_versions())
1692
needed_versions = self.source_ancestry - this_versions
1693
cross_check_versions = self.source_ancestry.intersection(this_versions)
1694
mismatched_versions = set()
1695
for version in cross_check_versions:
1696
# scan to include needed parents.
1697
n1 = set(self.target.get_parents_with_ghosts(version))
1698
n2 = set(self.source.get_parents_with_ghosts(version))
1700
# FIXME TEST this check for cycles being introduced works
1701
# the logic is we have a cycle if in our graph we are an
1702
# ancestor of any of the n2 revisions.
1708
parent_ancestors = self.source.get_ancestry(parent)
1709
if version in parent_ancestors:
1710
raise errors.GraphCycleError([parent, version])
1711
# ensure this parent will be available later.
1712
new_parents = n2.difference(n1)
1713
needed_versions.update(new_parents.difference(this_versions))
1714
mismatched_versions.add(version)
1716
if not needed_versions and not mismatched_versions:
1718
full_list = topo_sort(self.source.get_graph())
1720
version_list = [i for i in full_list if (not self.target.has_version(i)
1721
and i in needed_versions)]
1725
copy_queue_records = []
1727
for version_id in version_list:
1728
options = self.source._index.get_options(version_id)
1729
parents = self.source._index.get_parents_with_ghosts(version_id)
1730
# check that its will be a consistent copy:
1731
for parent in parents:
1732
# if source has the parent, we must :
1733
# * already have it or
1734
# * have it scheduled already
1735
# otherwise we don't care
1736
assert (self.target.has_version(parent) or
1737
parent in copy_set or
1738
not self.source.has_version(parent))
1739
data_pos, data_size = self.source._index.get_position(version_id)
1740
copy_queue_records.append((version_id, data_pos, data_size))
1741
copy_queue.append((version_id, options, parents))
1742
copy_set.add(version_id)
1744
# data suck the join:
1746
total = len(version_list)
1749
for (version_id, raw_data), \
1750
(version_id2, options, parents) in \
1751
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1753
assert version_id == version_id2, 'logic error, inconsistent results'
1755
pb.update("Joining knit", count, total)
1756
raw_records.append((version_id, options, parents, len(raw_data)))
1757
raw_datum.append(raw_data)
1758
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1760
for version in mismatched_versions:
1761
# FIXME RBC 20060309 is this needed?
1762
n1 = set(self.target.get_parents_with_ghosts(version))
1763
n2 = set(self.source.get_parents_with_ghosts(version))
1764
# write a combined record to our history preserving the current
1765
# parents as first in the list
1766
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1767
self.target.fix_parents(version, new_parents)
1773
InterVersionedFile.register_optimiser(InterKnit)
1776
class WeaveToKnit(InterVersionedFile):
1777
"""Optimised code paths for weave to knit operations."""
1779
_matching_file_from_factory = bzrlib.weave.WeaveFile
1780
_matching_file_to_factory = KnitVersionedFile
1783
def is_compatible(source, target):
1784
"""Be compatible with weaves to knits."""
1786
return (isinstance(source, bzrlib.weave.Weave) and
1787
isinstance(target, KnitVersionedFile))
1788
except AttributeError:
1791
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1792
"""See InterVersionedFile.join."""
1793
assert isinstance(self.source, bzrlib.weave.Weave)
1794
assert isinstance(self.target, KnitVersionedFile)
1796
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1801
pb = ui.ui_factory.nested_progress_bar()
1803
version_ids = list(version_ids)
1805
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1806
this_versions = set(self.target._index.get_versions())
1807
needed_versions = self.source_ancestry - this_versions
1808
cross_check_versions = self.source_ancestry.intersection(this_versions)
1809
mismatched_versions = set()
1810
for version in cross_check_versions:
1811
# scan to include needed parents.
1812
n1 = set(self.target.get_parents_with_ghosts(version))
1813
n2 = set(self.source.get_parents(version))
1814
# if all of n2's parents are in n1, then its fine.
1815
if n2.difference(n1):
1816
# FIXME TEST this check for cycles being introduced works
1817
# the logic is we have a cycle if in our graph we are an
1818
# ancestor of any of the n2 revisions.
1824
parent_ancestors = self.source.get_ancestry(parent)
1825
if version in parent_ancestors:
1826
raise errors.GraphCycleError([parent, version])
1827
# ensure this parent will be available later.
1828
new_parents = n2.difference(n1)
1829
needed_versions.update(new_parents.difference(this_versions))
1830
mismatched_versions.add(version)
1832
if not needed_versions and not mismatched_versions:
1834
full_list = topo_sort(self.source.get_graph())
1836
version_list = [i for i in full_list if (not self.target.has_version(i)
1837
and i in needed_versions)]
1841
total = len(version_list)
1842
for version_id in version_list:
1843
pb.update("Converting to knit", count, total)
1844
parents = self.source.get_parents(version_id)
1845
# check that its will be a consistent copy:
1846
for parent in parents:
1847
# if source has the parent, we must already have it
1848
assert (self.target.has_version(parent))
1849
self.target.add_lines(
1850
version_id, parents, self.source.get_lines(version_id))
1853
for version in mismatched_versions:
1854
# FIXME RBC 20060309 is this needed?
1855
n1 = set(self.target.get_parents_with_ghosts(version))
1856
n2 = set(self.source.get_parents(version))
1857
# write a combined record to our history preserving the current
1858
# parents as first in the list
1859
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1860
self.target.fix_parents(version, new_parents)
1866
InterVersionedFile.register_optimiser(WeaveToKnit)
1869
class KnitSequenceMatcher(difflib.SequenceMatcher):
1870
"""Knit tuned sequence matcher.
1872
This is based on profiling of difflib which indicated some improvements
1873
for our usage pattern.
1876
def find_longest_match(self, alo, ahi, blo, bhi):
1877
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1879
If isjunk is not defined:
1881
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1882
alo <= i <= i+k <= ahi
1883
blo <= j <= j+k <= bhi
1884
and for all (i',j',k') meeting those conditions,
1887
and if i == i', j <= j'
1889
In other words, of all maximal matching blocks, return one that
1890
starts earliest in a, and of all those maximal matching blocks that
1891
start earliest in a, return the one that starts earliest in b.
1893
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1894
>>> s.find_longest_match(0, 5, 0, 9)
1897
If isjunk is defined, first the longest matching block is
1898
determined as above, but with the additional restriction that no
1899
junk element appears in the block. Then that block is extended as
1900
far as possible by matching (only) junk elements on both sides. So
1901
the resulting block never matches on junk except as identical junk
1902
happens to be adjacent to an "interesting" match.
1904
Here's the same example as before, but considering blanks to be
1905
junk. That prevents " abcd" from matching the " abcd" at the tail
1906
end of the second sequence directly. Instead only the "abcd" can
1907
match, and matches the leftmost "abcd" in the second sequence:
1909
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1910
>>> s.find_longest_match(0, 5, 0, 9)
1913
If no blocks match, return (alo, blo, 0).
1915
>>> s = SequenceMatcher(None, "ab", "c")
1916
>>> s.find_longest_match(0, 2, 0, 1)
1920
# CAUTION: stripping common prefix or suffix would be incorrect.
1924
# Longest matching block is "ab", but if common prefix is
1925
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1926
# strip, so ends up claiming that ab is changed to acab by
1927
# inserting "ca" in the middle. That's minimal but unintuitive:
1928
# "it's obvious" that someone inserted "ac" at the front.
1929
# Windiff ends up at the same place as diff, but by pairing up
1930
# the unique 'b's and then matching the first two 'a's.
1932
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1933
besti, bestj, bestsize = alo, blo, 0
1934
# find longest junk-free match
1935
# during an iteration of the loop, j2len[j] = length of longest
1936
# junk-free match ending with a[i-1] and b[j]
1940
for i in xrange(alo, ahi):
1941
# look at all instances of a[i] in b; note that because
1942
# b2j has no junk keys, the loop is skipped if a[i] is junk
1943
j2lenget = j2len.get
1946
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1947
# following improvement
1948
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1949
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1950
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1952
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1953
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1954
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1966
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1968
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1971
# Extend the best by non-junk elements on each end. In particular,
1972
# "popular" non-junk elements aren't in b2j, which greatly speeds
1973
# the inner loop above, but also means "the best" match so far
1974
# doesn't contain any junk *or* popular non-junk elements.
1975
while besti > alo and bestj > blo and \
1976
not isbjunk(b[bestj-1]) and \
1977
a[besti-1] == b[bestj-1]:
1978
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1979
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1980
not isbjunk(b[bestj+bestsize]) and \
1981
a[besti+bestsize] == b[bestj+bestsize]:
1984
# Now that we have a wholly interesting match (albeit possibly
1985
# empty!), we may as well suck up the matching junk on each
1986
# side of it too. Can't think of a good reason not to, and it
1987
# saves post-processing the (possibly considerable) expense of
1988
# figuring out what to do with it. In the case of an empty
1989
# interesting match, this is clearly the right thing to do,
1990
# because no other kind of match is possible in the regions.
1991
while besti > alo and bestj > blo and \
1992
isbjunk(b[bestj-1]) and \
1993
a[besti-1] == b[bestj-1]:
1994
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1995
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1996
isbjunk(b[bestj+bestsize]) and \
1997
a[besti+bestsize] == b[bestj+bestsize]:
1998
bestsize = bestsize + 1
2000
return besti, bestj, bestsize
added
removed
bzrlib/knit.py
