1
# Copyright (C) 2005, 2006, 2007 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
73
from bzrlib.lazy_import import lazy_import
74
lazy_import(globals(), """
91
from bzrlib.errors import (
97
KnitDataStreamIncompatible,
100
RevisionAlreadyPresent,
102
from bzrlib.tuned_gzip import GzipFile
103
from bzrlib.osutils import (
108
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
109
from bzrlib.tsort import topo_sort
111
from bzrlib.util import bencode
113
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
116
# TODO: Split out code specific to this format into an associated object.
118
# TODO: Can we put in some kind of value to check that the index and data
119
# files belong together?
121
# TODO: accommodate binaries, perhaps by storing a byte count
123
# TODO: function to check whole file
125
# TODO: atomically append data, then measure backwards from the cursor
126
# position after writing to work out where it was located. we may need to
127
# bypass python file buffering.
129
DATA_SUFFIX = '.knit'
130
INDEX_SUFFIX = '.kndx'
133
class KnitContent(object):
134
"""Content of a knit version to which deltas can be applied."""
136
def __init__(self, lines):
139
def annotate_iter(self):
140
"""Yield tuples of (origin, text) for each content line."""
141
return iter(self._lines)
144
"""Return a list of (origin, text) tuples."""
145
return list(self.annotate_iter())
147
def line_delta_iter(self, new_lines):
148
"""Generate line-based delta from this content to new_lines."""
149
new_texts = new_lines.text()
150
old_texts = self.text()
151
s = KnitSequenceMatcher(None, old_texts, new_texts)
152
for tag, i1, i2, j1, j2 in s.get_opcodes():
155
# ofrom, oto, length, data
156
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
158
def line_delta(self, new_lines):
159
return list(self.line_delta_iter(new_lines))
162
return [text for origin, text in self._lines]
165
return KnitContent(self._lines[:])
168
def get_line_delta_blocks(knit_delta, source, target):
169
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
170
target_len = len(target)
173
for s_begin, s_end, t_len, new_text in knit_delta:
174
true_n = s_begin - s_pos
177
# knit deltas do not provide reliable info about whether the
178
# last line of a file matches, due to eol handling.
179
if source[s_pos + n -1] != target[t_pos + n -1]:
182
yield s_pos, t_pos, n
183
t_pos += t_len + true_n
185
n = target_len - t_pos
187
if source[s_pos + n -1] != target[t_pos + n -1]:
190
yield s_pos, t_pos, n
191
yield s_pos + (target_len - t_pos), target_len, 0
194
class _KnitFactory(object):
195
"""Base factory for creating content objects."""
197
def make(self, lines, version_id):
198
num_lines = len(lines)
199
return KnitContent(zip([version_id] * num_lines, lines))
202
class KnitAnnotateFactory(_KnitFactory):
203
"""Factory for creating annotated Content objects."""
207
def parse_fulltext(self, content, version_id):
208
"""Convert fulltext to internal representation
210
fulltext content is of the format
211
revid(utf8) plaintext\n
212
internal representation is of the format:
215
# TODO: jam 20070209 The tests expect this to be returned as tuples,
216
# but the code itself doesn't really depend on that.
217
# Figure out a way to not require the overhead of turning the
218
# list back into tuples.
219
lines = [tuple(line.split(' ', 1)) for line in content]
220
return KnitContent(lines)
222
def parse_line_delta_iter(self, lines):
223
return iter(self.parse_line_delta(lines))
225
def parse_line_delta(self, lines, version_id):
226
"""Convert a line based delta into internal representation.
228
line delta is in the form of:
229
intstart intend intcount
231
revid(utf8) newline\n
232
internal representation is
233
(start, end, count, [1..count tuples (revid, newline)])
240
def cache_and_return(line):
241
origin, text = line.split(' ', 1)
242
return cache.setdefault(origin, origin), text
244
# walk through the lines parsing.
246
start, end, count = [int(n) for n in header.split(',')]
247
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
248
result.append((start, end, count, contents))
251
def get_fulltext_content(self, lines):
252
"""Extract just the content lines from a fulltext."""
253
return (line.split(' ', 1)[1] for line in lines)
255
def get_linedelta_content(self, lines):
256
"""Extract just the content from a line delta.
258
This doesn't return all of the extra information stored in a delta.
259
Only the actual content lines.
264
header = header.split(',')
265
count = int(header[2])
266
for i in xrange(count):
267
origin, text = next().split(' ', 1)
270
def lower_fulltext(self, content):
271
"""convert a fulltext content record into a serializable form.
273
see parse_fulltext which this inverts.
275
# TODO: jam 20070209 We only do the caching thing to make sure that
276
# the origin is a valid utf-8 line, eventually we could remove it
277
return ['%s %s' % (o, t) for o, t in content._lines]
279
def lower_line_delta(self, delta):
280
"""convert a delta into a serializable form.
282
See parse_line_delta which this inverts.
284
# TODO: jam 20070209 We only do the caching thing to make sure that
285
# the origin is a valid utf-8 line, eventually we could remove it
287
for start, end, c, lines in delta:
288
out.append('%d,%d,%d\n' % (start, end, c))
289
out.extend(origin + ' ' + text
290
for origin, text in lines)
294
class KnitPlainFactory(_KnitFactory):
295
"""Factory for creating plain Content objects."""
299
def parse_fulltext(self, content, version_id):
300
"""This parses an unannotated fulltext.
302
Note that this is not a noop - the internal representation
303
has (versionid, line) - its just a constant versionid.
305
return self.make(content, version_id)
307
def parse_line_delta_iter(self, lines, version_id):
309
num_lines = len(lines)
310
while cur < num_lines:
313
start, end, c = [int(n) for n in header.split(',')]
314
yield start, end, c, zip([version_id] * c, lines[cur:cur+c])
317
def parse_line_delta(self, lines, version_id):
318
return list(self.parse_line_delta_iter(lines, version_id))
320
def get_fulltext_content(self, lines):
321
"""Extract just the content lines from a fulltext."""
324
def get_linedelta_content(self, lines):
325
"""Extract just the content from a line delta.
327
This doesn't return all of the extra information stored in a delta.
328
Only the actual content lines.
333
header = header.split(',')
334
count = int(header[2])
335
for i in xrange(count):
338
def lower_fulltext(self, content):
339
return content.text()
341
def lower_line_delta(self, delta):
343
for start, end, c, lines in delta:
344
out.append('%d,%d,%d\n' % (start, end, c))
345
out.extend([text for origin, text in lines])
349
def make_empty_knit(transport, relpath):
350
"""Construct a empty knit at the specified location."""
351
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
354
class KnitVersionedFile(VersionedFile):
355
"""Weave-like structure with faster random access.
357
A knit stores a number of texts and a summary of the relationships
358
between them. Texts are identified by a string version-id. Texts
359
are normally stored and retrieved as a series of lines, but can
360
also be passed as single strings.
362
Lines are stored with the trailing newline (if any) included, to
363
avoid special cases for files with no final newline. Lines are
364
composed of 8-bit characters, not unicode. The combination of
365
these approaches should mean any 'binary' file can be safely
366
stored and retrieved.
369
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
370
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
371
create=False, create_parent_dir=False, delay_create=False,
372
dir_mode=None, index=None, access_method=None):
373
"""Construct a knit at location specified by relpath.
375
:param create: If not True, only open an existing knit.
376
:param create_parent_dir: If True, create the parent directory if
377
creating the file fails. (This is used for stores with
378
hash-prefixes that may not exist yet)
379
:param delay_create: The calling code is aware that the knit won't
380
actually be created until the first data is stored.
381
:param index: An index to use for the knit.
383
if deprecated_passed(basis_knit):
384
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
385
" deprecated as of bzr 0.9.",
386
DeprecationWarning, stacklevel=2)
387
if access_mode is None:
389
super(KnitVersionedFile, self).__init__(access_mode)
390
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
391
self.transport = transport
392
self.filename = relpath
393
self.factory = factory or KnitAnnotateFactory()
394
self.writable = (access_mode == 'w')
397
self._max_delta_chain = 200
400
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
401
access_mode, create=create, file_mode=file_mode,
402
create_parent_dir=create_parent_dir, delay_create=delay_create,
406
if access_method is None:
407
_access = _KnitAccess(transport, relpath + DATA_SUFFIX, file_mode, dir_mode,
408
((create and not len(self)) and delay_create), create_parent_dir)
410
_access = access_method
411
if create and not len(self) and not delay_create:
413
self._data = _KnitData(_access)
416
return '%s(%s)' % (self.__class__.__name__,
417
self.transport.abspath(self.filename))
419
def _check_should_delta(self, first_parents):
420
"""Iterate back through the parent listing, looking for a fulltext.
422
This is used when we want to decide whether to add a delta or a new
423
fulltext. It searches for _max_delta_chain parents. When it finds a
424
fulltext parent, it sees if the total size of the deltas leading up to
425
it is large enough to indicate that we want a new full text anyway.
427
Return True if we should create a new delta, False if we should use a
432
delta_parents = first_parents
433
for count in xrange(self._max_delta_chain):
434
parent = delta_parents[0]
435
method = self._index.get_method(parent)
436
index, pos, size = self._index.get_position(parent)
437
if method == 'fulltext':
441
delta_parents = self._index.get_parents(parent)
443
# We couldn't find a fulltext, so we must create a new one
446
return fulltext_size > delta_size
448
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
449
"""See VersionedFile._add_delta()."""
450
self._check_add(version_id, []) # should we check the lines ?
451
self._check_versions_present(parents)
455
for parent in parents:
456
if not self.has_version(parent):
457
ghosts.append(parent)
459
present_parents.append(parent)
461
if delta_parent is None:
462
# reconstitute as full text.
463
assert len(delta) == 1 or len(delta) == 0
465
assert delta[0][0] == 0
466
assert delta[0][1] == 0, delta[0][1]
467
return super(KnitVersionedFile, self)._add_delta(version_id,
478
options.append('no-eol')
480
if delta_parent is not None:
481
# determine the current delta chain length.
482
# To speed the extract of texts the delta chain is limited
483
# to a fixed number of deltas. This should minimize both
484
# I/O and the time spend applying deltas.
485
# The window was changed to a maximum of 200 deltas, but also added
486
# was a check that the total compressed size of the deltas is
487
# smaller than the compressed size of the fulltext.
488
if not self._check_should_delta([delta_parent]):
489
# We don't want a delta here, just do a normal insertion.
490
return super(KnitVersionedFile, self)._add_delta(version_id,
497
options.append('line-delta')
498
store_lines = self.factory.lower_line_delta(delta)
500
access_memo = self._data.add_record(version_id, digest, store_lines)
501
self._index.add_version(version_id, options, access_memo, parents)
503
def _add_raw_records(self, records, data):
504
"""Add all the records 'records' with data pre-joined in 'data'.
506
:param records: A list of tuples(version_id, options, parents, size).
507
:param data: The data for the records. When it is written, the records
508
are adjusted to have pos pointing into data by the sum of
509
the preceding records sizes.
512
raw_record_sizes = [record[3] for record in records]
513
positions = self._data.add_raw_records(raw_record_sizes, data)
516
for (version_id, options, parents, size), access_memo in zip(
518
index_entries.append((version_id, options, access_memo, parents))
519
if self._data._do_cache:
520
self._data._cache[version_id] = data[offset:offset+size]
522
self._index.add_versions(index_entries)
524
def enable_cache(self):
525
"""Start caching data for this knit"""
526
self._data.enable_cache()
528
def clear_cache(self):
529
"""Clear the data cache only."""
530
self._data.clear_cache()
532
def copy_to(self, name, transport):
533
"""See VersionedFile.copy_to()."""
534
# copy the current index to a temp index to avoid racing with local
536
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
537
self.transport.get(self._index._filename))
539
f = self._data._open_file()
541
transport.put_file(name + DATA_SUFFIX, f)
544
# move the copied index into place
545
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
547
def create_empty(self, name, transport, mode=None):
548
return KnitVersionedFile(name, transport, factory=self.factory,
549
delta=self.delta, create=True)
551
def _fix_parents(self, version_id, new_parents):
552
"""Fix the parents list for version.
554
This is done by appending a new version to the index
555
with identical data except for the parents list.
556
the parents list must be a superset of the current
559
current_values = self._index._cache[version_id]
560
assert set(current_values[4]).difference(set(new_parents)) == set()
561
self._index.add_version(version_id,
563
(None, current_values[2], current_values[3]),
566
def get_data_stream(self, required_versions):
567
"""Get a data stream for the specified versions.
569
Versions may be returned in any order, not necessarily the order
572
:param required_versions: the exact set of versions to be returned, i.e.
573
not a transitive closure.
575
:returns: format_signature, list of (version, options, length, parents),
578
required_versions = set([osutils.safe_revision_id(v) for v in
580
# we don't care about inclusions, the caller cares.
581
# but we need to setup a list of records to visit.
582
for version_id in required_versions:
583
if not self.has_version(version_id):
584
raise RevisionNotPresent(version_id, self.filename)
585
# Pick the desired versions out of the index in oldest-to-newest order
587
for version_id in self.versions():
588
if version_id in required_versions:
589
version_list.append(version_id)
591
# create the list of version information for the result
592
copy_queue_records = []
594
result_version_list = []
595
for version_id in version_list:
596
options = self._index.get_options(version_id)
597
parents = self._index.get_parents_with_ghosts(version_id)
598
index_memo = self._index.get_position(version_id)
599
copy_queue_records.append((version_id, index_memo))
600
none, data_pos, data_size = index_memo
601
copy_set.add(version_id)
602
# version, options, length, parents
603
result_version_list.append((version_id, options, data_size,
606
# Read the compressed record data.
608
# From here down to the return should really be logic in the returned
609
# callable -- in a class that adapts read_records_iter_raw to read
612
for (version_id, raw_data), \
613
(version_id2, options, _, parents) in \
614
izip(self._data.read_records_iter_raw(copy_queue_records),
615
result_version_list):
616
assert version_id == version_id2, 'logic error, inconsistent results'
617
raw_datum.append(raw_data)
618
pseudo_file = StringIO(''.join(raw_datum))
621
return pseudo_file.read()
623
return pseudo_file.read(length)
624
return (self.get_format_signature(), result_version_list, read)
626
def get_stream_as_bytes(self, required_versions):
627
"""Generate a serialised data stream.
629
The format is a bencoding of a list. The first element of the list is a
630
string of the format signature, then each subsequent element is a list
631
corresponding to a record. Those lists contain:
638
:returns: a bencoded list.
640
knit_stream = self.get_data_stream(required_versions)
641
format_signature, data_list, callable = knit_stream
643
data.append(format_signature)
644
for version, options, length, parents in data_list:
645
data.append([version, options, parents, callable(length)])
646
return bencode.bencode(data)
648
def _extract_blocks(self, version_id, source, target):
649
if self._index.get_method(version_id) != 'line-delta':
651
parent, sha1, noeol, delta = self.get_delta(version_id)
652
return KnitContent.get_line_delta_blocks(delta, source, target)
654
def get_delta(self, version_id):
655
"""Get a delta for constructing version from some other version."""
656
version_id = osutils.safe_revision_id(version_id)
657
self.check_not_reserved_id(version_id)
658
if not self.has_version(version_id):
659
raise RevisionNotPresent(version_id, self.filename)
661
parents = self.get_parents(version_id)
666
index_memo = self._index.get_position(version_id)
667
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
668
noeol = 'no-eol' in self._index.get_options(version_id)
669
if 'fulltext' == self._index.get_method(version_id):
670
new_content = self.factory.parse_fulltext(data, version_id)
671
if parent is not None:
672
reference_content = self._get_content(parent)
673
old_texts = reference_content.text()
676
new_texts = new_content.text()
677
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
678
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
680
delta = self.factory.parse_line_delta(data, version_id)
681
return parent, sha1, noeol, delta
683
def get_format_signature(self):
684
"""See VersionedFile.get_format_signature()."""
685
if self.factory.annotated:
686
annotated_part = "annotated"
688
annotated_part = "plain"
689
return "knit-%s" % (annotated_part,)
691
def get_graph_with_ghosts(self):
692
"""See VersionedFile.get_graph_with_ghosts()."""
693
graph_items = self._index.get_graph()
694
return dict(graph_items)
696
def get_sha1(self, version_id):
697
return self.get_sha1s([version_id])[0]
699
def get_sha1s(self, version_ids):
700
"""See VersionedFile.get_sha1()."""
701
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
702
record_map = self._get_record_map(version_ids)
703
# record entry 2 is the 'digest'.
704
return [record_map[v][2] for v in version_ids]
708
"""See VersionedFile.get_suffixes()."""
709
return [DATA_SUFFIX, INDEX_SUFFIX]
711
def has_ghost(self, version_id):
712
"""True if there is a ghost reference in the file to version_id."""
713
version_id = osutils.safe_revision_id(version_id)
715
if self.has_version(version_id):
717
# optimisable if needed by memoising the _ghosts set.
718
items = self._index.get_graph()
719
for node, parents in items:
720
for parent in parents:
721
if parent not in self._index._cache:
722
if parent == version_id:
726
def insert_data_stream(self, (format, data_list, reader_callable)):
727
"""Insert knit records from a data stream into this knit.
729
If a version in the stream is already present in this knit, it will not
730
be inserted a second time. It will be checked for consistency with the
731
stored version however, and may cause a KnitCorrupt error to be raised
732
if the data in the stream disagrees with the already stored data.
734
:seealso: get_data_stream
736
if format != self.get_format_signature():
737
mutter('incompatible format signature inserting to %r', self)
738
raise KnitDataStreamIncompatible(
739
format, self.get_format_signature())
741
for version_id, options, length, parents in data_list:
742
if self.has_version(version_id):
743
# First check: the list of parents.
744
my_parents = self.get_parents_with_ghosts(version_id)
745
if my_parents != parents:
746
# XXX: KnitCorrupt is not quite the right exception here.
749
'parents list %r from data stream does not match '
750
'already recorded parents %r for %s'
751
% (parents, my_parents, version_id))
753
# Also check the SHA-1 of the fulltext this content will
755
raw_data = reader_callable(length)
756
my_fulltext_sha1 = self.get_sha1(version_id)
757
df, rec = self._data._parse_record_header(version_id, raw_data)
758
stream_fulltext_sha1 = rec[3]
759
if my_fulltext_sha1 != stream_fulltext_sha1:
760
# Actually, we don't know if it's this knit that's corrupt,
761
# or the data stream we're trying to insert.
763
self.filename, 'sha-1 does not match %s' % version_id)
765
self._add_raw_records(
766
[(version_id, options, parents, length)],
767
reader_callable(length))
770
"""See VersionedFile.versions."""
771
if 'evil' in debug.debug_flags:
772
trace.mutter_callsite(2, "versions scales with size of history")
773
return self._index.get_versions()
775
def has_version(self, version_id):
776
"""See VersionedFile.has_version."""
777
if 'evil' in debug.debug_flags:
778
trace.mutter_callsite(2, "has_version is a LBYL scenario")
779
version_id = osutils.safe_revision_id(version_id)
780
return self._index.has_version(version_id)
782
__contains__ = has_version
784
def _merge_annotations(self, content, parents, parent_texts={},
785
delta=None, annotated=None,
786
left_matching_blocks=None):
787
"""Merge annotations for content. This is done by comparing
788
the annotations based on changed to the text.
790
if left_matching_blocks is not None:
791
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
795
for parent_id in parents:
796
merge_content = self._get_content(parent_id, parent_texts)
797
if (parent_id == parents[0] and delta_seq is not None):
800
seq = patiencediff.PatienceSequenceMatcher(
801
None, merge_content.text(), content.text())
802
for i, j, n in seq.get_matching_blocks():
805
# this appears to copy (origin, text) pairs across to the
806
# new content for any line that matches the last-checked
808
content._lines[j:j+n] = merge_content._lines[i:i+n]
810
if delta_seq is None:
811
reference_content = self._get_content(parents[0], parent_texts)
812
new_texts = content.text()
813
old_texts = reference_content.text()
814
delta_seq = patiencediff.PatienceSequenceMatcher(
815
None, old_texts, new_texts)
816
return self._make_line_delta(delta_seq, content)
818
def _make_line_delta(self, delta_seq, new_content):
819
"""Generate a line delta from delta_seq and new_content."""
821
for op in delta_seq.get_opcodes():
824
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
827
def _get_components_positions(self, version_ids):
828
"""Produce a map of position data for the components of versions.
830
This data is intended to be used for retrieving the knit records.
832
A dict of version_id to (method, data_pos, data_size, next) is
834
method is the way referenced data should be applied.
835
data_pos is the position of the data in the knit.
836
data_size is the size of the data in the knit.
837
next is the build-parent of the version, or None for fulltexts.
840
for version_id in version_ids:
843
while cursor is not None and cursor not in component_data:
844
method = self._index.get_method(cursor)
845
if method == 'fulltext':
848
next = self.get_parents(cursor)[0]
849
index_memo = self._index.get_position(cursor)
850
component_data[cursor] = (method, index_memo, next)
852
return component_data
854
def _get_content(self, version_id, parent_texts={}):
855
"""Returns a content object that makes up the specified
857
if not self.has_version(version_id):
858
raise RevisionNotPresent(version_id, self.filename)
860
cached_version = parent_texts.get(version_id, None)
861
if cached_version is not None:
862
return cached_version
864
text_map, contents_map = self._get_content_maps([version_id])
865
return contents_map[version_id]
867
def _check_versions_present(self, version_ids):
868
"""Check that all specified versions are present."""
869
self._index.check_versions_present(version_ids)
871
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
872
"""See VersionedFile.add_lines_with_ghosts()."""
873
self._check_add(version_id, lines)
874
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
876
def _add_lines(self, version_id, parents, lines, parent_texts,
877
left_matching_blocks=None):
878
"""See VersionedFile.add_lines."""
879
self._check_add(version_id, lines)
880
self._check_versions_present(parents)
881
return self._add(version_id, lines[:], parents, self.delta,
882
parent_texts, left_matching_blocks)
884
def _check_add(self, version_id, lines):
885
"""check that version_id and lines are safe to add."""
886
assert self.writable, "knit is not opened for write"
887
### FIXME escape. RBC 20060228
888
if contains_whitespace(version_id):
889
raise InvalidRevisionId(version_id, self.filename)
890
self.check_not_reserved_id(version_id)
891
if self.has_version(version_id):
892
raise RevisionAlreadyPresent(version_id, self.filename)
893
self._check_lines_not_unicode(lines)
894
self._check_lines_are_lines(lines)
896
def _add(self, version_id, lines, parents, delta, parent_texts,
897
left_matching_blocks=None):
898
"""Add a set of lines on top of version specified by parents.
900
If delta is true, compress the text as a line-delta against
903
Any versions not present will be converted into ghosts.
905
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
906
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
907
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
908
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
909
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
910
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
911
# +1383 0 8.0370 8.0370 +<len>
912
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
913
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
914
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
915
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
919
if parent_texts is None:
921
for parent in parents:
922
if not self.has_version(parent):
923
ghosts.append(parent)
925
present_parents.append(parent)
927
if delta and not len(present_parents):
930
digest = sha_strings(lines)
933
if lines[-1][-1] != '\n':
934
options.append('no-eol')
935
lines[-1] = lines[-1] + '\n'
937
if len(present_parents) and delta:
938
# To speed the extract of texts the delta chain is limited
939
# to a fixed number of deltas. This should minimize both
940
# I/O and the time spend applying deltas.
941
delta = self._check_should_delta(present_parents)
943
assert isinstance(version_id, str)
944
lines = self.factory.make(lines, version_id)
945
if delta or (self.factory.annotated and len(present_parents) > 0):
946
# Merge annotations from parent texts if so is needed.
947
delta_hunks = self._merge_annotations(lines, present_parents,
948
parent_texts, delta, self.factory.annotated,
949
left_matching_blocks)
952
options.append('line-delta')
953
store_lines = self.factory.lower_line_delta(delta_hunks)
955
options.append('fulltext')
956
store_lines = self.factory.lower_fulltext(lines)
958
access_memo = self._data.add_record(version_id, digest, store_lines)
959
self._index.add_version(version_id, options, access_memo, parents)
962
def check(self, progress_bar=None):
963
"""See VersionedFile.check()."""
965
def _clone_text(self, new_version_id, old_version_id, parents):
966
"""See VersionedFile.clone_text()."""
967
# FIXME RBC 20060228 make fast by only inserting an index with null
969
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
971
def get_lines(self, version_id):
972
"""See VersionedFile.get_lines()."""
973
return self.get_line_list([version_id])[0]
975
def _get_record_map(self, version_ids):
976
"""Produce a dictionary of knit records.
978
The keys are version_ids, the values are tuples of (method, content,
980
method is the way the content should be applied.
981
content is a KnitContent object.
982
digest is the SHA1 digest of this version id after all steps are done
983
next is the build-parent of the version, i.e. the leftmost ancestor.
984
If the method is fulltext, next will be None.
986
position_map = self._get_components_positions(version_ids)
987
# c = component_id, m = method, i_m = index_memo, n = next
988
records = [(c, i_m) for c, (m, i_m, n) in position_map.iteritems()]
990
for component_id, content, digest in \
991
self._data.read_records_iter(records):
992
method, index_memo, next = position_map[component_id]
993
record_map[component_id] = method, content, digest, next
997
def get_text(self, version_id):
998
"""See VersionedFile.get_text"""
999
return self.get_texts([version_id])[0]
1001
def get_texts(self, version_ids):
1002
return [''.join(l) for l in self.get_line_list(version_ids)]
1004
def get_line_list(self, version_ids):
1005
"""Return the texts of listed versions as a list of strings."""
1006
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
1007
for version_id in version_ids:
1008
self.check_not_reserved_id(version_id)
1009
text_map, content_map = self._get_content_maps(version_ids)
1010
return [text_map[v] for v in version_ids]
1012
_get_lf_split_line_list = get_line_list
1014
def _get_content_maps(self, version_ids):
1015
"""Produce maps of text and KnitContents
1017
:return: (text_map, content_map) where text_map contains the texts for
1018
the requested versions and content_map contains the KnitContents.
1019
Both dicts take version_ids as their keys.
1021
for version_id in version_ids:
1022
if not self.has_version(version_id):
1023
raise RevisionNotPresent(version_id, self.filename)
1024
record_map = self._get_record_map(version_ids)
1029
for version_id in version_ids:
1032
while cursor is not None:
1033
method, data, digest, next = record_map[cursor]
1034
components.append((cursor, method, data, digest))
1035
if cursor in content_map:
1040
for component_id, method, data, digest in reversed(components):
1041
if component_id in content_map:
1042
content = content_map[component_id]
1044
if method == 'fulltext':
1045
assert content is None
1046
content = self.factory.parse_fulltext(data, version_id)
1047
elif method == 'line-delta':
1048
delta = self.factory.parse_line_delta(data, version_id)
1049
content = content.copy()
1050
content._lines = self._apply_delta(content._lines,
1052
content_map[component_id] = content
1054
if 'no-eol' in self._index.get_options(version_id):
1055
content = content.copy()
1056
line = content._lines[-1][1].rstrip('\n')
1057
content._lines[-1] = (content._lines[-1][0], line)
1058
final_content[version_id] = content
1060
# digest here is the digest from the last applied component.
1061
text = content.text()
1062
if sha_strings(text) != digest:
1063
raise KnitCorrupt(self.filename,
1064
'sha-1 does not match %s' % version_id)
1066
text_map[version_id] = text
1067
return text_map, final_content
1069
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1071
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1072
if version_ids is None:
1073
version_ids = self.versions()
1075
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
1077
pb = progress.DummyProgress()
1078
# we don't care about inclusions, the caller cares.
1079
# but we need to setup a list of records to visit.
1080
# we need version_id, position, length
1081
version_id_records = []
1082
requested_versions = set(version_ids)
1083
# filter for available versions
1084
for version_id in requested_versions:
1085
if not self.has_version(version_id):
1086
raise RevisionNotPresent(version_id, self.filename)
1087
# get a in-component-order queue:
1088
for version_id in self.versions():
1089
if version_id in requested_versions:
1090
index_memo = self._index.get_position(version_id)
1091
version_id_records.append((version_id, index_memo))
1093
total = len(version_id_records)
1094
for version_idx, (version_id, data, sha_value) in \
1095
enumerate(self._data.read_records_iter(version_id_records)):
1096
pb.update('Walking content.', version_idx, total)
1097
method = self._index.get_method(version_id)
1099
assert method in ('fulltext', 'line-delta')
1100
if method == 'fulltext':
1101
line_iterator = self.factory.get_fulltext_content(data)
1103
line_iterator = self.factory.get_linedelta_content(data)
1104
for line in line_iterator:
1107
pb.update('Walking content.', total, total)
1109
def iter_parents(self, version_ids):
1110
"""Iterate through the parents for many version ids.
1112
:param version_ids: An iterable yielding version_ids.
1113
:return: An iterator that yields (version_id, parents). Requested
1114
version_ids not present in the versioned file are simply skipped.
1115
The order is undefined, allowing for different optimisations in
1116
the underlying implementation.
1118
version_ids = [osutils.safe_revision_id(version_id) for
1119
version_id in version_ids]
1120
return self._index.iter_parents(version_ids)
1122
def num_versions(self):
1123
"""See VersionedFile.num_versions()."""
1124
return self._index.num_versions()
1126
__len__ = num_versions
1128
def annotate_iter(self, version_id):
1129
"""See VersionedFile.annotate_iter."""
1130
version_id = osutils.safe_revision_id(version_id)
1131
content = self._get_content(version_id)
1132
for origin, text in content.annotate_iter():
1135
def get_parents(self, version_id):
1136
"""See VersionedFile.get_parents."""
1139
# 52554 calls in 1264 872 internal down from 3674
1140
version_id = osutils.safe_revision_id(version_id)
1142
return self._index.get_parents(version_id)
1144
raise RevisionNotPresent(version_id, self.filename)
1146
def get_parents_with_ghosts(self, version_id):
1147
"""See VersionedFile.get_parents."""
1148
version_id = osutils.safe_revision_id(version_id)
1150
return self._index.get_parents_with_ghosts(version_id)
1152
raise RevisionNotPresent(version_id, self.filename)
1154
def get_ancestry(self, versions, topo_sorted=True):
1155
"""See VersionedFile.get_ancestry."""
1156
if isinstance(versions, basestring):
1157
versions = [versions]
1160
versions = [osutils.safe_revision_id(v) for v in versions]
1161
return self._index.get_ancestry(versions, topo_sorted)
1163
def get_ancestry_with_ghosts(self, versions):
1164
"""See VersionedFile.get_ancestry_with_ghosts."""
1165
if isinstance(versions, basestring):
1166
versions = [versions]
1169
versions = [osutils.safe_revision_id(v) for v in versions]
1170
return self._index.get_ancestry_with_ghosts(versions)
1172
def plan_merge(self, ver_a, ver_b):
1173
"""See VersionedFile.plan_merge."""
1174
ver_a = osutils.safe_revision_id(ver_a)
1175
ver_b = osutils.safe_revision_id(ver_b)
1176
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1178
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1179
annotated_a = self.annotate(ver_a)
1180
annotated_b = self.annotate(ver_b)
1181
return merge._plan_annotate_merge(annotated_a, annotated_b,
1182
ancestors_a, ancestors_b)
1185
class _KnitComponentFile(object):
1186
"""One of the files used to implement a knit database"""
1188
def __init__(self, transport, filename, mode, file_mode=None,
1189
create_parent_dir=False, dir_mode=None):
1190
self._transport = transport
1191
self._filename = filename
1193
self._file_mode = file_mode
1194
self._dir_mode = dir_mode
1195
self._create_parent_dir = create_parent_dir
1196
self._need_to_create = False
1198
def _full_path(self):
1199
"""Return the full path to this file."""
1200
return self._transport.base + self._filename
1202
def check_header(self, fp):
1203
line = fp.readline()
1205
# An empty file can actually be treated as though the file doesn't
1207
raise errors.NoSuchFile(self._full_path())
1208
if line != self.HEADER:
1209
raise KnitHeaderError(badline=line,
1210
filename=self._transport.abspath(self._filename))
1213
return '%s(%s)' % (self.__class__.__name__, self._filename)
1216
class _KnitIndex(_KnitComponentFile):
1217
"""Manages knit index file.
1219
The index is already kept in memory and read on startup, to enable
1220
fast lookups of revision information. The cursor of the index
1221
file is always pointing to the end, making it easy to append
1224
_cache is a cache for fast mapping from version id to a Index
1227
_history is a cache for fast mapping from indexes to version ids.
1229
The index data format is dictionary compressed when it comes to
1230
parent references; a index entry may only have parents that with a
1231
lover index number. As a result, the index is topological sorted.
1233
Duplicate entries may be written to the index for a single version id
1234
if this is done then the latter one completely replaces the former:
1235
this allows updates to correct version and parent information.
1236
Note that the two entries may share the delta, and that successive
1237
annotations and references MUST point to the first entry.
1239
The index file on disc contains a header, followed by one line per knit
1240
record. The same revision can be present in an index file more than once.
1241
The first occurrence gets assigned a sequence number starting from 0.
1243
The format of a single line is
1244
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1245
REVISION_ID is a utf8-encoded revision id
1246
FLAGS is a comma separated list of flags about the record. Values include
1247
no-eol, line-delta, fulltext.
1248
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1249
that the the compressed data starts at.
1250
LENGTH is the ascii representation of the length of the data file.
1251
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1253
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1254
revision id already in the knit that is a parent of REVISION_ID.
1255
The ' :' marker is the end of record marker.
1258
when a write is interrupted to the index file, it will result in a line
1259
that does not end in ' :'. If the ' :' is not present at the end of a line,
1260
or at the end of the file, then the record that is missing it will be
1261
ignored by the parser.
1263
When writing new records to the index file, the data is preceded by '\n'
1264
to ensure that records always start on new lines even if the last write was
1265
interrupted. As a result its normal for the last line in the index to be
1266
missing a trailing newline. One can be added with no harmful effects.
1269
HEADER = "# bzr knit index 8\n"
1271
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1272
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1274
def _cache_version(self, version_id, options, pos, size, parents):
1275
"""Cache a version record in the history array and index cache.
1277
This is inlined into _load_data for performance. KEEP IN SYNC.
1278
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1281
# only want the _history index to reference the 1st index entry
1283
if version_id not in self._cache:
1284
index = len(self._history)
1285
self._history.append(version_id)
1287
index = self._cache[version_id][5]
1288
self._cache[version_id] = (version_id,
1295
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1296
create_parent_dir=False, delay_create=False, dir_mode=None):
1297
_KnitComponentFile.__init__(self, transport, filename, mode,
1298
file_mode=file_mode,
1299
create_parent_dir=create_parent_dir,
1302
# position in _history is the 'official' index for a revision
1303
# but the values may have come from a newer entry.
1304
# so - wc -l of a knit index is != the number of unique names
1308
fp = self._transport.get(self._filename)
1310
# _load_data may raise NoSuchFile if the target knit is
1312
_load_data(self, fp)
1316
if mode != 'w' or not create:
1319
self._need_to_create = True
1321
self._transport.put_bytes_non_atomic(
1322
self._filename, self.HEADER, mode=self._file_mode)
1324
def get_graph(self):
1325
"""Return a list of the node:parents lists from this knit index."""
1326
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1328
def get_ancestry(self, versions, topo_sorted=True):
1329
"""See VersionedFile.get_ancestry."""
1330
# get a graph of all the mentioned versions:
1332
pending = set(versions)
1335
version = pending.pop()
1338
parents = [p for p in cache[version][4] if p in cache]
1340
raise RevisionNotPresent(version, self._filename)
1341
# if not completed and not a ghost
1342
pending.update([p for p in parents if p not in graph])
1343
graph[version] = parents
1346
return topo_sort(graph.items())
1348
def get_ancestry_with_ghosts(self, versions):
1349
"""See VersionedFile.get_ancestry_with_ghosts."""
1350
# get a graph of all the mentioned versions:
1351
self.check_versions_present(versions)
1354
pending = set(versions)
1356
version = pending.pop()
1358
parents = cache[version][4]
1364
pending.update([p for p in parents if p not in graph])
1365
graph[version] = parents
1366
return topo_sort(graph.items())
1368
def iter_parents(self, version_ids):
1369
"""Iterate through the parents for many version ids.
1371
:param version_ids: An iterable yielding version_ids.
1372
:return: An iterator that yields (version_id, parents). Requested
1373
version_ids not present in the versioned file are simply skipped.
1374
The order is undefined, allowing for different optimisations in
1375
the underlying implementation.
1377
for version_id in version_ids:
1379
yield version_id, tuple(self.get_parents(version_id))
1383
def num_versions(self):
1384
return len(self._history)
1386
__len__ = num_versions
1388
def get_versions(self):
1389
"""Get all the versions in the file. not topologically sorted."""
1390
return self._history
1392
def _version_list_to_index(self, versions):
1395
for version in versions:
1396
if version in cache:
1397
# -- inlined lookup() --
1398
result_list.append(str(cache[version][5]))
1399
# -- end lookup () --
1401
result_list.append('.' + version)
1402
return ' '.join(result_list)
1404
def add_version(self, version_id, options, index_memo, parents):
1405
"""Add a version record to the index."""
1406
self.add_versions(((version_id, options, index_memo, parents),))
1408
def add_versions(self, versions):
1409
"""Add multiple versions to the index.
1411
:param versions: a list of tuples:
1412
(version_id, options, pos, size, parents).
1415
orig_history = self._history[:]
1416
orig_cache = self._cache.copy()
1419
for version_id, options, (index, pos, size), parents in versions:
1420
line = "\n%s %s %s %s %s :" % (version_id,
1424
self._version_list_to_index(parents))
1425
assert isinstance(line, str), \
1426
'content must be utf-8 encoded: %r' % (line,)
1428
self._cache_version(version_id, options, pos, size, parents)
1429
if not self._need_to_create:
1430
self._transport.append_bytes(self._filename, ''.join(lines))
1433
sio.write(self.HEADER)
1434
sio.writelines(lines)
1436
self._transport.put_file_non_atomic(self._filename, sio,
1437
create_parent_dir=self._create_parent_dir,
1438
mode=self._file_mode,
1439
dir_mode=self._dir_mode)
1440
self._need_to_create = False
1442
# If any problems happen, restore the original values and re-raise
1443
self._history = orig_history
1444
self._cache = orig_cache
1447
def has_version(self, version_id):
1448
"""True if the version is in the index."""
1449
return version_id in self._cache
1451
def get_position(self, version_id):
1452
"""Return details needed to access the version.
1454
.kndx indices do not support split-out data, so return None for the
1457
:return: a tuple (None, data position, size) to hand to the access
1458
logic to get the record.
1460
entry = self._cache[version_id]
1461
return None, entry[2], entry[3]
1463
def get_method(self, version_id):
1464
"""Return compression method of specified version."""
1465
options = self._cache[version_id][1]
1466
if 'fulltext' in options:
1469
if 'line-delta' not in options:
1470
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1473
def get_options(self, version_id):
1474
"""Return a string represention options.
1478
return self._cache[version_id][1]
1480
def get_parents(self, version_id):
1481
"""Return parents of specified version ignoring ghosts."""
1482
return [parent for parent in self._cache[version_id][4]
1483
if parent in self._cache]
1485
def get_parents_with_ghosts(self, version_id):
1486
"""Return parents of specified version with ghosts."""
1487
return self._cache[version_id][4]
1489
def check_versions_present(self, version_ids):
1490
"""Check that all specified versions are present."""
1492
for version_id in version_ids:
1493
if version_id not in cache:
1494
raise RevisionNotPresent(version_id, self._filename)
1497
class KnitGraphIndex(object):
1498
"""A knit index that builds on GraphIndex."""
1500
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1501
"""Construct a KnitGraphIndex on a graph_index.
1503
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1504
:param deltas: Allow delta-compressed records.
1505
:param add_callback: If not None, allow additions to the index and call
1506
this callback with a list of added GraphIndex nodes:
1507
[(node, value, node_refs), ...]
1508
:param parents: If True, record knits parents, if not do not record
1511
self._graph_index = graph_index
1512
self._deltas = deltas
1513
self._add_callback = add_callback
1514
self._parents = parents
1515
if deltas and not parents:
1516
raise KnitCorrupt(self, "Cannot do delta compression without "
1519
def _get_entries(self, keys, check_present=False):
1520
"""Get the entries for keys.
1522
:param keys: An iterable of index keys, - 1-tuples.
1527
for node in self._graph_index.iter_entries(keys):
1529
found_keys.add(node[1])
1531
# adapt parentless index to the rest of the code.
1532
for node in self._graph_index.iter_entries(keys):
1533
yield node[0], node[1], node[2], ()
1534
found_keys.add(node[1])
1536
missing_keys = keys.difference(found_keys)
1538
raise RevisionNotPresent(missing_keys.pop(), self)
1540
def _present_keys(self, version_ids):
1542
node[1] for node in self._get_entries(version_ids)])
1544
def _parentless_ancestry(self, versions):
1545
"""Honour the get_ancestry API for parentless knit indices."""
1546
wanted_keys = self._version_ids_to_keys(versions)
1547
present_keys = self._present_keys(wanted_keys)
1548
missing = set(wanted_keys).difference(present_keys)
1550
raise RevisionNotPresent(missing.pop(), self)
1551
return list(self._keys_to_version_ids(present_keys))
1553
def get_ancestry(self, versions, topo_sorted=True):
1554
"""See VersionedFile.get_ancestry."""
1555
if not self._parents:
1556
return self._parentless_ancestry(versions)
1557
# XXX: This will do len(history) index calls - perhaps
1558
# it should be altered to be a index core feature?
1559
# get a graph of all the mentioned versions:
1562
versions = self._version_ids_to_keys(versions)
1563
pending = set(versions)
1565
# get all pending nodes
1566
this_iteration = pending
1567
new_nodes = self._get_entries(this_iteration)
1570
for (index, key, value, node_refs) in new_nodes:
1571
# dont ask for ghosties - otherwise
1572
# we we can end up looping with pending
1573
# being entirely ghosted.
1574
graph[key] = [parent for parent in node_refs[0]
1575
if parent not in ghosts]
1577
for parent in graph[key]:
1578
# dont examine known nodes again
1583
ghosts.update(this_iteration.difference(found))
1584
if versions.difference(graph):
1585
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1587
result_keys = topo_sort(graph.items())
1589
result_keys = graph.iterkeys()
1590
return [key[0] for key in result_keys]
1592
def get_ancestry_with_ghosts(self, versions):
1593
"""See VersionedFile.get_ancestry."""
1594
if not self._parents:
1595
return self._parentless_ancestry(versions)
1596
# XXX: This will do len(history) index calls - perhaps
1597
# it should be altered to be a index core feature?
1598
# get a graph of all the mentioned versions:
1600
versions = self._version_ids_to_keys(versions)
1601
pending = set(versions)
1603
# get all pending nodes
1604
this_iteration = pending
1605
new_nodes = self._get_entries(this_iteration)
1607
for (index, key, value, node_refs) in new_nodes:
1608
graph[key] = node_refs[0]
1610
for parent in graph[key]:
1611
# dont examine known nodes again
1615
missing_versions = this_iteration.difference(graph)
1616
missing_needed = versions.intersection(missing_versions)
1618
raise RevisionNotPresent(missing_needed.pop(), self)
1619
for missing_version in missing_versions:
1620
# add a key, no parents
1621
graph[missing_version] = []
1622
pending.discard(missing_version) # don't look for it
1623
result_keys = topo_sort(graph.items())
1624
return [key[0] for key in result_keys]
1626
def get_graph(self):
1627
"""Return a list of the node:parents lists from this knit index."""
1628
if not self._parents:
1629
return [(key, ()) for key in self.get_versions()]
1631
for index, key, value, refs in self._graph_index.iter_all_entries():
1632
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1635
def iter_parents(self, version_ids):
1636
"""Iterate through the parents for many version ids.
1638
:param version_ids: An iterable yielding version_ids.
1639
:return: An iterator that yields (version_id, parents). Requested
1640
version_ids not present in the versioned file are simply skipped.
1641
The order is undefined, allowing for different optimisations in
1642
the underlying implementation.
1645
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1647
present_parents = set()
1648
for node in all_nodes:
1649
all_parents.update(node[3][0])
1650
# any node we are querying must be present
1651
present_parents.add(node[1])
1652
unknown_parents = all_parents.difference(present_parents)
1653
present_parents.update(self._present_keys(unknown_parents))
1654
for node in all_nodes:
1656
for parent in node[3][0]:
1657
if parent in present_parents:
1658
parents.append(parent[0])
1659
yield node[1][0], tuple(parents)
1661
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1662
yield node[1][0], ()
1664
def num_versions(self):
1665
return len(list(self._graph_index.iter_all_entries()))
1667
__len__ = num_versions
1669
def get_versions(self):
1670
"""Get all the versions in the file. not topologically sorted."""
1671
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1673
def has_version(self, version_id):
1674
"""True if the version is in the index."""
1675
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1677
def _keys_to_version_ids(self, keys):
1678
return tuple(key[0] for key in keys)
1680
def get_position(self, version_id):
1681
"""Return details needed to access the version.
1683
:return: a tuple (index, data position, size) to hand to the access
1684
logic to get the record.
1686
node = self._get_node(version_id)
1687
bits = node[2][1:].split(' ')
1688
return node[0], int(bits[0]), int(bits[1])
1690
def get_method(self, version_id):
1691
"""Return compression method of specified version."""
1692
if not self._deltas:
1694
return self._parent_compression(self._get_node(version_id)[3][1])
1696
def _parent_compression(self, reference_list):
1697
# use the second reference list to decide if this is delta'd or not.
1698
if len(reference_list):
1703
def _get_node(self, version_id):
1704
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1706
def get_options(self, version_id):
1707
"""Return a string represention options.
1711
node = self._get_node(version_id)
1712
if not self._deltas:
1713
options = ['fulltext']
1715
options = [self._parent_compression(node[3][1])]
1716
if node[2][0] == 'N':
1717
options.append('no-eol')
1720
def get_parents(self, version_id):
1721
"""Return parents of specified version ignoring ghosts."""
1722
parents = list(self.iter_parents([version_id]))
1725
raise errors.RevisionNotPresent(version_id, self)
1726
return parents[0][1]
1728
def get_parents_with_ghosts(self, version_id):
1729
"""Return parents of specified version with ghosts."""
1730
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1731
check_present=True))
1732
if not self._parents:
1734
return self._keys_to_version_ids(nodes[0][3][0])
1736
def check_versions_present(self, version_ids):
1737
"""Check that all specified versions are present."""
1738
keys = self._version_ids_to_keys(version_ids)
1739
present = self._present_keys(keys)
1740
missing = keys.difference(present)
1742
raise RevisionNotPresent(missing.pop(), self)
1744
def add_version(self, version_id, options, access_memo, parents):
1745
"""Add a version record to the index."""
1746
return self.add_versions(((version_id, options, access_memo, parents),))
1748
def add_versions(self, versions):
1749
"""Add multiple versions to the index.
1751
This function does not insert data into the Immutable GraphIndex
1752
backing the KnitGraphIndex, instead it prepares data for insertion by
1753
the caller and checks that it is safe to insert then calls
1754
self._add_callback with the prepared GraphIndex nodes.
1756
:param versions: a list of tuples:
1757
(version_id, options, pos, size, parents).
1759
if not self._add_callback:
1760
raise errors.ReadOnlyError(self)
1761
# we hope there are no repositories with inconsistent parentage
1766
for (version_id, options, access_memo, parents) in versions:
1767
index, pos, size = access_memo
1768
key = (version_id, )
1769
parents = tuple((parent, ) for parent in parents)
1770
if 'no-eol' in options:
1774
value += "%d %d" % (pos, size)
1775
if not self._deltas:
1776
if 'line-delta' in options:
1777
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1780
if 'line-delta' in options:
1781
node_refs = (parents, (parents[0],))
1783
node_refs = (parents, ())
1785
node_refs = (parents, )
1788
raise KnitCorrupt(self, "attempt to add node with parents "
1789
"in parentless index.")
1791
keys[key] = (value, node_refs)
1792
present_nodes = self._get_entries(keys)
1793
for (index, key, value, node_refs) in present_nodes:
1794
if (value, node_refs) != keys[key]:
1795
raise KnitCorrupt(self, "inconsistent details in add_versions"
1796
": %s %s" % ((value, node_refs), keys[key]))
1800
for key, (value, node_refs) in keys.iteritems():
1801
result.append((key, value, node_refs))
1803
for key, (value, node_refs) in keys.iteritems():
1804
result.append((key, value))
1805
self._add_callback(result)
1807
def _version_ids_to_keys(self, version_ids):
1808
return set((version_id, ) for version_id in version_ids)
1811
class _KnitAccess(object):
1812
"""Access to knit records in a .knit file."""
1814
def __init__(self, transport, filename, _file_mode, _dir_mode,
1815
_need_to_create, _create_parent_dir):
1816
"""Create a _KnitAccess for accessing and inserting data.
1818
:param transport: The transport the .knit is located on.
1819
:param filename: The filename of the .knit.
1821
self._transport = transport
1822
self._filename = filename
1823
self._file_mode = _file_mode
1824
self._dir_mode = _dir_mode
1825
self._need_to_create = _need_to_create
1826
self._create_parent_dir = _create_parent_dir
1828
def add_raw_records(self, sizes, raw_data):
1829
"""Add raw knit bytes to a storage area.
1831
The data is spooled to whereever the access method is storing data.
1833
:param sizes: An iterable containing the size of each raw data segment.
1834
:param raw_data: A bytestring containing the data.
1835
:return: A list of memos to retrieve the record later. Each memo is a
1836
tuple - (index, pos, length), where the index field is always None
1837
for the .knit access method.
1839
assert type(raw_data) == str, \
1840
'data must be plain bytes was %s' % type(raw_data)
1841
if not self._need_to_create:
1842
base = self._transport.append_bytes(self._filename, raw_data)
1844
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1845
create_parent_dir=self._create_parent_dir,
1846
mode=self._file_mode,
1847
dir_mode=self._dir_mode)
1848
self._need_to_create = False
1852
result.append((None, base, size))
1857
"""IFF this data access has its own storage area, initialise it.
1861
self._transport.put_bytes_non_atomic(self._filename, '',
1862
mode=self._file_mode)
1864
def open_file(self):
1865
"""IFF this data access can be represented as a single file, open it.
1867
For knits that are not mapped to a single file on disk this will
1870
:return: None or a file handle.
1873
return self._transport.get(self._filename)
1878
def get_raw_records(self, memos_for_retrieval):
1879
"""Get the raw bytes for a records.
1881
:param memos_for_retrieval: An iterable containing the (index, pos,
1882
length) memo for retrieving the bytes. The .knit method ignores
1883
the index as there is always only a single file.
1884
:return: An iterator over the bytes of the records.
1886
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
1887
for pos, data in self._transport.readv(self._filename, read_vector):
1891
class _PackAccess(object):
1892
"""Access to knit records via a collection of packs."""
1894
def __init__(self, index_to_packs, writer=None):
1895
"""Create a _PackAccess object.
1897
:param index_to_packs: A dict mapping index objects to the transport
1898
and file names for obtaining data.
1899
:param writer: A tuple (pack.ContainerWriter, write_index) which
1900
contains the pack to write, and the index that reads from it will
1904
self.container_writer = writer[0]
1905
self.write_index = writer[1]
1907
self.container_writer = None
1908
self.write_index = None
1909
self.indices = index_to_packs
1911
def add_raw_records(self, sizes, raw_data):
1912
"""Add raw knit bytes to a storage area.
1914
The data is spooled to the container writer in one bytes-record per
1917
:param sizes: An iterable containing the size of each raw data segment.
1918
:param raw_data: A bytestring containing the data.
1919
:return: A list of memos to retrieve the record later. Each memo is a
1920
tuple - (index, pos, length), where the index field is the
1921
write_index object supplied to the PackAccess object.
1923
assert type(raw_data) == str, \
1924
'data must be plain bytes was %s' % type(raw_data)
1928
p_offset, p_length = self.container_writer.add_bytes_record(
1929
raw_data[offset:offset+size], [])
1931
result.append((self.write_index, p_offset, p_length))
1935
"""Pack based knits do not get individually created."""
1937
def get_raw_records(self, memos_for_retrieval):
1938
"""Get the raw bytes for a records.
1940
:param memos_for_retrieval: An iterable containing the (index, pos,
1941
length) memo for retrieving the bytes. The Pack access method
1942
looks up the pack to use for a given record in its index_to_pack
1944
:return: An iterator over the bytes of the records.
1946
# first pass, group into same-index requests
1948
current_index = None
1949
for (index, offset, length) in memos_for_retrieval:
1950
if current_index == index:
1951
current_list.append((offset, length))
1953
if current_index is not None:
1954
request_lists.append((current_index, current_list))
1955
current_index = index
1956
current_list = [(offset, length)]
1957
# handle the last entry
1958
if current_index is not None:
1959
request_lists.append((current_index, current_list))
1960
for index, offsets in request_lists:
1961
transport, path = self.indices[index]
1962
reader = pack.make_readv_reader(transport, path, offsets)
1963
for names, read_func in reader.iter_records():
1964
yield read_func(None)
1966
def open_file(self):
1967
"""Pack based knits have no single file."""
1970
def set_writer(self, writer, index, (transport, packname)):
1971
"""Set a writer to use for adding data."""
1972
self.indices[index] = (transport, packname)
1973
self.container_writer = writer
1974
self.write_index = index
1977
class _KnitData(object):
1978
"""Manage extraction of data from a KnitAccess, caching and decompressing.
1980
The KnitData class provides the logic for parsing and using knit records,
1981
making use of an access method for the low level read and write operations.
1984
def __init__(self, access):
1985
"""Create a KnitData object.
1987
:param access: The access method to use. Access methods such as
1988
_KnitAccess manage the insertion of raw records and the subsequent
1989
retrieval of the same.
1991
self._access = access
1992
self._checked = False
1993
# TODO: jam 20060713 conceptually, this could spill to disk
1994
# if the cached size gets larger than a certain amount
1995
# but it complicates the model a bit, so for now just use
1996
# a simple dictionary
1998
self._do_cache = False
2000
def enable_cache(self):
2001
"""Enable caching of reads."""
2002
self._do_cache = True
2004
def clear_cache(self):
2005
"""Clear the record cache."""
2006
self._do_cache = False
2009
def _open_file(self):
2010
return self._access.open_file()
2012
def _record_to_data(self, version_id, digest, lines):
2013
"""Convert version_id, digest, lines into a raw data block.
2015
:return: (len, a StringIO instance with the raw data ready to read.)
2018
data_file = GzipFile(None, mode='wb', fileobj=sio)
2020
assert isinstance(version_id, str)
2021
data_file.writelines(chain(
2022
["version %s %d %s\n" % (version_id,
2026
["end %s\n" % version_id]))
2033
def add_raw_records(self, sizes, raw_data):
2034
"""Append a prepared record to the data file.
2036
:param sizes: An iterable containing the size of each raw data segment.
2037
:param raw_data: A bytestring containing the data.
2038
:return: a list of index data for the way the data was stored.
2039
See the access method add_raw_records documentation for more
2042
return self._access.add_raw_records(sizes, raw_data)
2044
def add_record(self, version_id, digest, lines):
2045
"""Write new text record to disk.
2047
Returns index data for retrieving it later, as per add_raw_records.
2049
size, sio = self._record_to_data(version_id, digest, lines)
2050
result = self.add_raw_records([size], sio.getvalue())
2052
self._cache[version_id] = sio.getvalue()
2055
def _parse_record_header(self, version_id, raw_data):
2056
"""Parse a record header for consistency.
2058
:return: the header and the decompressor stream.
2059
as (stream, header_record)
2061
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2063
rec = self._check_header(version_id, df.readline())
2064
except Exception, e:
2065
raise KnitCorrupt(self._access,
2066
"While reading {%s} got %s(%s)"
2067
% (version_id, e.__class__.__name__, str(e)))
2070
def _check_header(self, version_id, line):
2073
raise KnitCorrupt(self._access,
2074
'unexpected number of elements in record header')
2075
if rec[1] != version_id:
2076
raise KnitCorrupt(self._access,
2077
'unexpected version, wanted %r, got %r'
2078
% (version_id, rec[1]))
2081
def _parse_record(self, version_id, data):
2083
# 4168 calls in 2880 217 internal
2084
# 4168 calls to _parse_record_header in 2121
2085
# 4168 calls to readlines in 330
2086
df = GzipFile(mode='rb', fileobj=StringIO(data))
2089
record_contents = df.readlines()
2090
except Exception, e:
2091
raise KnitCorrupt(self._access,
2092
"While reading {%s} got %s(%s)"
2093
% (version_id, e.__class__.__name__, str(e)))
2094
header = record_contents.pop(0)
2095
rec = self._check_header(version_id, header)
2097
last_line = record_contents.pop()
2098
if len(record_contents) != int(rec[2]):
2099
raise KnitCorrupt(self._access,
2100
'incorrect number of lines %s != %s'
2102
% (len(record_contents), int(rec[2]),
2104
if last_line != 'end %s\n' % rec[1]:
2105
raise KnitCorrupt(self._access,
2106
'unexpected version end line %r, wanted %r'
2107
% (last_line, version_id))
2109
return record_contents, rec[3]
2111
def read_records_iter_raw(self, records):
2112
"""Read text records from data file and yield raw data.
2114
This unpacks enough of the text record to validate the id is
2115
as expected but thats all.
2117
# setup an iterator of the external records:
2118
# uses readv so nice and fast we hope.
2120
# grab the disk data needed.
2122
# Don't check _cache if it is empty
2123
needed_offsets = [index_memo for version_id, index_memo
2125
if version_id not in self._cache]
2127
needed_offsets = [index_memo for version_id, index_memo
2130
raw_records = self._access.get_raw_records(needed_offsets)
2132
for version_id, index_memo in records:
2133
if version_id in self._cache:
2134
# This data has already been validated
2135
data = self._cache[version_id]
2137
data = raw_records.next()
2139
self._cache[version_id] = data
2141
# validate the header
2142
df, rec = self._parse_record_header(version_id, data)
2144
yield version_id, data
2146
def read_records_iter(self, records):
2147
"""Read text records from data file and yield result.
2149
The result will be returned in whatever is the fastest to read.
2150
Not by the order requested. Also, multiple requests for the same
2151
record will only yield 1 response.
2152
:param records: A list of (version_id, pos, len) entries
2153
:return: Yields (version_id, contents, digest) in the order
2154
read, not the order requested
2160
# Skip records we have alread seen
2161
yielded_records = set()
2162
needed_records = set()
2163
for record in records:
2164
if record[0] in self._cache:
2165
if record[0] in yielded_records:
2167
yielded_records.add(record[0])
2168
data = self._cache[record[0]]
2169
content, digest = self._parse_record(record[0], data)
2170
yield (record[0], content, digest)
2172
needed_records.add(record)
2173
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2175
needed_records = sorted(set(records), key=operator.itemgetter(1))
2177
if not needed_records:
2180
# The transport optimizes the fetching as well
2181
# (ie, reads continuous ranges.)
2182
raw_data = self._access.get_raw_records(
2183
[index_memo for version_id, index_memo in needed_records])
2185
for (version_id, index_memo), data in \
2186
izip(iter(needed_records), raw_data):
2187
content, digest = self._parse_record(version_id, data)
2189
self._cache[version_id] = data
2190
yield version_id, content, digest
2192
def read_records(self, records):
2193
"""Read records into a dictionary."""
2195
for record_id, content, digest in \
2196
self.read_records_iter(records):
2197
components[record_id] = (content, digest)
2201
class InterKnit(InterVersionedFile):
2202
"""Optimised code paths for knit to knit operations."""
2204
_matching_file_from_factory = KnitVersionedFile
2205
_matching_file_to_factory = KnitVersionedFile
2208
def is_compatible(source, target):
2209
"""Be compatible with knits. """
2211
return (isinstance(source, KnitVersionedFile) and
2212
isinstance(target, KnitVersionedFile))
2213
except AttributeError:
2216
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2217
"""See InterVersionedFile.join."""
2218
assert isinstance(self.source, KnitVersionedFile)
2219
assert isinstance(self.target, KnitVersionedFile)
2221
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2226
pb = ui.ui_factory.nested_progress_bar()
2228
version_ids = list(version_ids)
2229
if None in version_ids:
2230
version_ids.remove(None)
2232
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2233
this_versions = set(self.target._index.get_versions())
2234
needed_versions = self.source_ancestry - this_versions
2235
cross_check_versions = self.source_ancestry.intersection(this_versions)
2236
mismatched_versions = set()
2237
for version in cross_check_versions:
2238
# scan to include needed parents.
2239
n1 = set(self.target.get_parents_with_ghosts(version))
2240
n2 = set(self.source.get_parents_with_ghosts(version))
2242
# FIXME TEST this check for cycles being introduced works
2243
# the logic is we have a cycle if in our graph we are an
2244
# ancestor of any of the n2 revisions.
2250
parent_ancestors = self.source.get_ancestry(parent)
2251
if version in parent_ancestors:
2252
raise errors.GraphCycleError([parent, version])
2253
# ensure this parent will be available later.
2254
new_parents = n2.difference(n1)
2255
needed_versions.update(new_parents.difference(this_versions))
2256
mismatched_versions.add(version)
2258
if not needed_versions and not mismatched_versions:
2260
full_list = topo_sort(self.source.get_graph())
2262
version_list = [i for i in full_list if (not self.target.has_version(i)
2263
and i in needed_versions)]
2267
copy_queue_records = []
2269
for version_id in version_list:
2270
options = self.source._index.get_options(version_id)
2271
parents = self.source._index.get_parents_with_ghosts(version_id)
2272
# check that its will be a consistent copy:
2273
for parent in parents:
2274
# if source has the parent, we must :
2275
# * already have it or
2276
# * have it scheduled already
2277
# otherwise we don't care
2278
assert (self.target.has_version(parent) or
2279
parent in copy_set or
2280
not self.source.has_version(parent))
2281
index_memo = self.source._index.get_position(version_id)
2282
copy_queue_records.append((version_id, index_memo))
2283
copy_queue.append((version_id, options, parents))
2284
copy_set.add(version_id)
2286
# data suck the join:
2288
total = len(version_list)
2291
for (version_id, raw_data), \
2292
(version_id2, options, parents) in \
2293
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2295
assert version_id == version_id2, 'logic error, inconsistent results'
2297
pb.update("Joining knit", count, total)
2298
raw_records.append((version_id, options, parents, len(raw_data)))
2299
raw_datum.append(raw_data)
2300
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2302
for version in mismatched_versions:
2303
# FIXME RBC 20060309 is this needed?
2304
n1 = set(self.target.get_parents_with_ghosts(version))
2305
n2 = set(self.source.get_parents_with_ghosts(version))
2306
# write a combined record to our history preserving the current
2307
# parents as first in the list
2308
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2309
self.target.fix_parents(version, new_parents)
2315
InterVersionedFile.register_optimiser(InterKnit)
2318
class WeaveToKnit(InterVersionedFile):
2319
"""Optimised code paths for weave to knit operations."""
2321
_matching_file_from_factory = bzrlib.weave.WeaveFile
2322
_matching_file_to_factory = KnitVersionedFile
2325
def is_compatible(source, target):
2326
"""Be compatible with weaves to knits."""
2328
return (isinstance(source, bzrlib.weave.Weave) and
2329
isinstance(target, KnitVersionedFile))
2330
except AttributeError:
2333
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2334
"""See InterVersionedFile.join."""
2335
assert isinstance(self.source, bzrlib.weave.Weave)
2336
assert isinstance(self.target, KnitVersionedFile)
2338
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2343
pb = ui.ui_factory.nested_progress_bar()
2345
version_ids = list(version_ids)
2347
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2348
this_versions = set(self.target._index.get_versions())
2349
needed_versions = self.source_ancestry - this_versions
2350
cross_check_versions = self.source_ancestry.intersection(this_versions)
2351
mismatched_versions = set()
2352
for version in cross_check_versions:
2353
# scan to include needed parents.
2354
n1 = set(self.target.get_parents_with_ghosts(version))
2355
n2 = set(self.source.get_parents(version))
2356
# if all of n2's parents are in n1, then its fine.
2357
if n2.difference(n1):
2358
# FIXME TEST this check for cycles being introduced works
2359
# the logic is we have a cycle if in our graph we are an
2360
# ancestor of any of the n2 revisions.
2366
parent_ancestors = self.source.get_ancestry(parent)
2367
if version in parent_ancestors:
2368
raise errors.GraphCycleError([parent, version])
2369
# ensure this parent will be available later.
2370
new_parents = n2.difference(n1)
2371
needed_versions.update(new_parents.difference(this_versions))
2372
mismatched_versions.add(version)
2374
if not needed_versions and not mismatched_versions:
2376
full_list = topo_sort(self.source.get_graph())
2378
version_list = [i for i in full_list if (not self.target.has_version(i)
2379
and i in needed_versions)]
2383
total = len(version_list)
2384
for version_id in version_list:
2385
pb.update("Converting to knit", count, total)
2386
parents = self.source.get_parents(version_id)
2387
# check that its will be a consistent copy:
2388
for parent in parents:
2389
# if source has the parent, we must already have it
2390
assert (self.target.has_version(parent))
2391
self.target.add_lines(
2392
version_id, parents, self.source.get_lines(version_id))
2395
for version in mismatched_versions:
2396
# FIXME RBC 20060309 is this needed?
2397
n1 = set(self.target.get_parents_with_ghosts(version))
2398
n2 = set(self.source.get_parents(version))
2399
# write a combined record to our history preserving the current
2400
# parents as first in the list
2401
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2402
self.target.fix_parents(version, new_parents)
2408
InterVersionedFile.register_optimiser(WeaveToKnit)
2411
class KnitSequenceMatcher(difflib.SequenceMatcher):
2412
"""Knit tuned sequence matcher.
2414
This is based on profiling of difflib which indicated some improvements
2415
for our usage pattern.
2418
def find_longest_match(self, alo, ahi, blo, bhi):
2419
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
2421
If isjunk is not defined:
2423
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
2424
alo <= i <= i+k <= ahi
2425
blo <= j <= j+k <= bhi
2426
and for all (i',j',k') meeting those conditions,
2429
and if i == i', j <= j'
2431
In other words, of all maximal matching blocks, return one that
2432
starts earliest in a, and of all those maximal matching blocks that
2433
start earliest in a, return the one that starts earliest in b.
2435
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
2436
>>> s.find_longest_match(0, 5, 0, 9)
2439
If isjunk is defined, first the longest matching block is
2440
determined as above, but with the additional restriction that no
2441
junk element appears in the block. Then that block is extended as
2442
far as possible by matching (only) junk elements on both sides. So
2443
the resulting block never matches on junk except as identical junk
2444
happens to be adjacent to an "interesting" match.
2446
Here's the same example as before, but considering blanks to be
2447
junk. That prevents " abcd" from matching the " abcd" at the tail
2448
end of the second sequence directly. Instead only the "abcd" can
2449
match, and matches the leftmost "abcd" in the second sequence:
2451
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
2452
>>> s.find_longest_match(0, 5, 0, 9)
2455
If no blocks match, return (alo, blo, 0).
2457
>>> s = SequenceMatcher(None, "ab", "c")
2458
>>> s.find_longest_match(0, 2, 0, 1)
2462
# CAUTION: stripping common prefix or suffix would be incorrect.
2466
# Longest matching block is "ab", but if common prefix is
2467
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
2468
# strip, so ends up claiming that ab is changed to acab by
2469
# inserting "ca" in the middle. That's minimal but unintuitive:
2470
# "it's obvious" that someone inserted "ac" at the front.
2471
# Windiff ends up at the same place as diff, but by pairing up
2472
# the unique 'b's and then matching the first two 'a's.
2474
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
2475
besti, bestj, bestsize = alo, blo, 0
2476
# find longest junk-free match
2477
# during an iteration of the loop, j2len[j] = length of longest
2478
# junk-free match ending with a[i-1] and b[j]
2482
for i in xrange(alo, ahi):
2483
# look at all instances of a[i] in b; note that because
2484
# b2j has no junk keys, the loop is skipped if a[i] is junk
2485
j2lenget = j2len.get
2488
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
2489
# following improvement
2490
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
2491
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
2492
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
2494
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
2495
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
2496
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
2508
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
2510
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
2513
# Extend the best by non-junk elements on each end. In particular,
2514
# "popular" non-junk elements aren't in b2j, which greatly speeds
2515
# the inner loop above, but also means "the best" match so far
2516
# doesn't contain any junk *or* popular non-junk elements.
2517
while besti > alo and bestj > blo and \
2518
not isbjunk(b[bestj-1]) and \
2519
a[besti-1] == b[bestj-1]:
2520
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2521
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2522
not isbjunk(b[bestj+bestsize]) and \
2523
a[besti+bestsize] == b[bestj+bestsize]:
2526
# Now that we have a wholly interesting match (albeit possibly
2527
# empty!), we may as well suck up the matching junk on each
2528
# side of it too. Can't think of a good reason not to, and it
2529
# saves post-processing the (possibly considerable) expense of
2530
# figuring out what to do with it. In the case of an empty
2531
# interesting match, this is clearly the right thing to do,
2532
# because no other kind of match is possible in the regions.
2533
while besti > alo and bestj > blo and \
2534
isbjunk(b[bestj-1]) and \
2535
a[besti-1] == b[bestj-1]:
2536
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2537
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2538
isbjunk(b[bestj+bestsize]) and \
2539
a[besti+bestsize] == b[bestj+bestsize]:
2540
bestsize = bestsize + 1
2542
return besti, bestj, bestsize
2546
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2548
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
