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.trace import mutter
110
from bzrlib.tsort import topo_sort
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 _extract_blocks(self, version_id, source, target):
627
if self._index.get_method(version_id) != 'line-delta':
629
parent, sha1, noeol, delta = self.get_delta(version_id)
630
return KnitContent.get_line_delta_blocks(delta, source, target)
632
def get_delta(self, version_id):
633
"""Get a delta for constructing version from some other version."""
634
version_id = osutils.safe_revision_id(version_id)
635
self.check_not_reserved_id(version_id)
636
if not self.has_version(version_id):
637
raise RevisionNotPresent(version_id, self.filename)
639
parents = self.get_parents(version_id)
644
index_memo = self._index.get_position(version_id)
645
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
646
noeol = 'no-eol' in self._index.get_options(version_id)
647
if 'fulltext' == self._index.get_method(version_id):
648
new_content = self.factory.parse_fulltext(data, version_id)
649
if parent is not None:
650
reference_content = self._get_content(parent)
651
old_texts = reference_content.text()
654
new_texts = new_content.text()
655
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
656
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
658
delta = self.factory.parse_line_delta(data, version_id)
659
return parent, sha1, noeol, delta
661
def get_format_signature(self):
662
"""See VersionedFile.get_format_signature()."""
663
if self.factory.annotated:
664
annotated_part = "annotated"
666
annotated_part = "plain"
667
return "knit-%s" % (annotated_part,)
669
def get_graph_with_ghosts(self):
670
"""See VersionedFile.get_graph_with_ghosts()."""
671
graph_items = self._index.get_graph()
672
return dict(graph_items)
674
def get_sha1(self, version_id):
675
return self.get_sha1s([version_id])[0]
677
def get_sha1s(self, version_ids):
678
"""See VersionedFile.get_sha1()."""
679
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
680
record_map = self._get_record_map(version_ids)
681
# record entry 2 is the 'digest'.
682
return [record_map[v][2] for v in version_ids]
686
"""See VersionedFile.get_suffixes()."""
687
return [DATA_SUFFIX, INDEX_SUFFIX]
689
def has_ghost(self, version_id):
690
"""True if there is a ghost reference in the file to version_id."""
691
version_id = osutils.safe_revision_id(version_id)
693
if self.has_version(version_id):
695
# optimisable if needed by memoising the _ghosts set.
696
items = self._index.get_graph()
697
for node, parents in items:
698
for parent in parents:
699
if parent not in self._index._cache:
700
if parent == version_id:
704
def insert_data_stream(self, (format, data_list, reader_callable)):
705
"""Insert knit records from a data stream into this knit.
707
If a version in the stream is already present in this knit, it will not
708
be inserted a second time. It will be checked for consistency with the
709
stored version however, and may cause a KnitCorrupt error to be raised
710
if the data in the stream disagrees with the already stored data.
712
:seealso: get_data_stream
714
if format != self.get_format_signature():
715
mutter('incompatible format signature inserting to %r', self)
716
raise KnitDataStreamIncompatible(
717
format, self.get_format_signature())
719
for version_id, options, length, parents in data_list:
720
if self.has_version(version_id):
721
# First check: the list of parents.
722
my_parents = self.get_parents_with_ghosts(version_id)
723
if my_parents != parents:
724
# XXX: KnitCorrupt is not quite the right exception here.
727
'parents list %r from data stream does not match '
728
'already recorded parents %r for %s'
729
% (parents, my_parents, version_id))
731
# Also check the SHA-1 of the fulltext this content will
733
raw_data = reader_callable(length)
734
my_fulltext_sha1 = self.get_sha1(version_id)
735
df, rec = self._data._parse_record_header(version_id, raw_data)
736
stream_fulltext_sha1 = rec[3]
737
if my_fulltext_sha1 != stream_fulltext_sha1:
738
# Actually, we don't know if it's this knit that's corrupt,
739
# or the data stream we're trying to insert.
741
self.filename, 'sha-1 does not match %s' % version_id)
743
self._add_raw_records(
744
[(version_id, options, parents, length)],
745
reader_callable(length))
748
"""See VersionedFile.versions."""
749
if 'evil' in debug.debug_flags:
750
trace.mutter_callsite(2, "versions scales with size of history")
751
return self._index.get_versions()
753
def has_version(self, version_id):
754
"""See VersionedFile.has_version."""
755
if 'evil' in debug.debug_flags:
756
trace.mutter_callsite(2, "has_version is a LBYL scenario")
757
version_id = osutils.safe_revision_id(version_id)
758
return self._index.has_version(version_id)
760
__contains__ = has_version
762
def _merge_annotations(self, content, parents, parent_texts={},
763
delta=None, annotated=None,
764
left_matching_blocks=None):
765
"""Merge annotations for content. This is done by comparing
766
the annotations based on changed to the text.
768
if left_matching_blocks is not None:
769
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
773
for parent_id in parents:
774
merge_content = self._get_content(parent_id, parent_texts)
775
if (parent_id == parents[0] and delta_seq is not None):
778
seq = patiencediff.PatienceSequenceMatcher(
779
None, merge_content.text(), content.text())
780
for i, j, n in seq.get_matching_blocks():
783
# this appears to copy (origin, text) pairs across to the
784
# new content for any line that matches the last-checked
786
content._lines[j:j+n] = merge_content._lines[i:i+n]
788
if delta_seq is None:
789
reference_content = self._get_content(parents[0], parent_texts)
790
new_texts = content.text()
791
old_texts = reference_content.text()
792
delta_seq = patiencediff.PatienceSequenceMatcher(
793
None, old_texts, new_texts)
794
return self._make_line_delta(delta_seq, content)
796
def _make_line_delta(self, delta_seq, new_content):
797
"""Generate a line delta from delta_seq and new_content."""
799
for op in delta_seq.get_opcodes():
802
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
805
def _get_components_positions(self, version_ids):
806
"""Produce a map of position data for the components of versions.
808
This data is intended to be used for retrieving the knit records.
810
A dict of version_id to (method, data_pos, data_size, next) is
812
method is the way referenced data should be applied.
813
data_pos is the position of the data in the knit.
814
data_size is the size of the data in the knit.
815
next is the build-parent of the version, or None for fulltexts.
818
for version_id in version_ids:
821
while cursor is not None and cursor not in component_data:
822
method = self._index.get_method(cursor)
823
if method == 'fulltext':
826
next = self.get_parents(cursor)[0]
827
index_memo = self._index.get_position(cursor)
828
component_data[cursor] = (method, index_memo, next)
830
return component_data
832
def _get_content(self, version_id, parent_texts={}):
833
"""Returns a content object that makes up the specified
835
if not self.has_version(version_id):
836
raise RevisionNotPresent(version_id, self.filename)
838
cached_version = parent_texts.get(version_id, None)
839
if cached_version is not None:
840
return cached_version
842
text_map, contents_map = self._get_content_maps([version_id])
843
return contents_map[version_id]
845
def _check_versions_present(self, version_ids):
846
"""Check that all specified versions are present."""
847
self._index.check_versions_present(version_ids)
849
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
850
"""See VersionedFile.add_lines_with_ghosts()."""
851
self._check_add(version_id, lines)
852
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
854
def _add_lines(self, version_id, parents, lines, parent_texts,
855
left_matching_blocks=None):
856
"""See VersionedFile.add_lines."""
857
self._check_add(version_id, lines)
858
self._check_versions_present(parents)
859
return self._add(version_id, lines[:], parents, self.delta,
860
parent_texts, left_matching_blocks)
862
def _check_add(self, version_id, lines):
863
"""check that version_id and lines are safe to add."""
864
assert self.writable, "knit is not opened for write"
865
### FIXME escape. RBC 20060228
866
if contains_whitespace(version_id):
867
raise InvalidRevisionId(version_id, self.filename)
868
self.check_not_reserved_id(version_id)
869
if self.has_version(version_id):
870
raise RevisionAlreadyPresent(version_id, self.filename)
871
self._check_lines_not_unicode(lines)
872
self._check_lines_are_lines(lines)
874
def _add(self, version_id, lines, parents, delta, parent_texts,
875
left_matching_blocks=None):
876
"""Add a set of lines on top of version specified by parents.
878
If delta is true, compress the text as a line-delta against
881
Any versions not present will be converted into ghosts.
883
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
884
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
885
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
886
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
887
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
888
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
889
# +1383 0 8.0370 8.0370 +<len>
890
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
891
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
892
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
893
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
897
if parent_texts is None:
899
for parent in parents:
900
if not self.has_version(parent):
901
ghosts.append(parent)
903
present_parents.append(parent)
905
if delta and not len(present_parents):
908
digest = sha_strings(lines)
911
if lines[-1][-1] != '\n':
912
options.append('no-eol')
913
lines[-1] = lines[-1] + '\n'
915
if len(present_parents) and delta:
916
# To speed the extract of texts the delta chain is limited
917
# to a fixed number of deltas. This should minimize both
918
# I/O and the time spend applying deltas.
919
delta = self._check_should_delta(present_parents)
921
assert isinstance(version_id, str)
922
lines = self.factory.make(lines, version_id)
923
if delta or (self.factory.annotated and len(present_parents) > 0):
924
# Merge annotations from parent texts if so is needed.
925
delta_hunks = self._merge_annotations(lines, present_parents,
926
parent_texts, delta, self.factory.annotated,
927
left_matching_blocks)
930
options.append('line-delta')
931
store_lines = self.factory.lower_line_delta(delta_hunks)
933
options.append('fulltext')
934
store_lines = self.factory.lower_fulltext(lines)
936
access_memo = self._data.add_record(version_id, digest, store_lines)
937
self._index.add_version(version_id, options, access_memo, parents)
940
def check(self, progress_bar=None):
941
"""See VersionedFile.check()."""
943
def _clone_text(self, new_version_id, old_version_id, parents):
944
"""See VersionedFile.clone_text()."""
945
# FIXME RBC 20060228 make fast by only inserting an index with null
947
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
949
def get_lines(self, version_id):
950
"""See VersionedFile.get_lines()."""
951
return self.get_line_list([version_id])[0]
953
def _get_record_map(self, version_ids):
954
"""Produce a dictionary of knit records.
956
The keys are version_ids, the values are tuples of (method, content,
958
method is the way the content should be applied.
959
content is a KnitContent object.
960
digest is the SHA1 digest of this version id after all steps are done
961
next is the build-parent of the version, i.e. the leftmost ancestor.
962
If the method is fulltext, next will be None.
964
position_map = self._get_components_positions(version_ids)
965
# c = component_id, m = method, i_m = index_memo, n = next
966
records = [(c, i_m) for c, (m, i_m, n) in position_map.iteritems()]
968
for component_id, content, digest in \
969
self._data.read_records_iter(records):
970
method, index_memo, next = position_map[component_id]
971
record_map[component_id] = method, content, digest, next
975
def get_text(self, version_id):
976
"""See VersionedFile.get_text"""
977
return self.get_texts([version_id])[0]
979
def get_texts(self, version_ids):
980
return [''.join(l) for l in self.get_line_list(version_ids)]
982
def get_line_list(self, version_ids):
983
"""Return the texts of listed versions as a list of strings."""
984
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
985
for version_id in version_ids:
986
self.check_not_reserved_id(version_id)
987
text_map, content_map = self._get_content_maps(version_ids)
988
return [text_map[v] for v in version_ids]
990
_get_lf_split_line_list = get_line_list
992
def _get_content_maps(self, version_ids):
993
"""Produce maps of text and KnitContents
995
:return: (text_map, content_map) where text_map contains the texts for
996
the requested versions and content_map contains the KnitContents.
997
Both dicts take version_ids as their keys.
999
for version_id in version_ids:
1000
if not self.has_version(version_id):
1001
raise RevisionNotPresent(version_id, self.filename)
1002
record_map = self._get_record_map(version_ids)
1007
for version_id in version_ids:
1010
while cursor is not None:
1011
method, data, digest, next = record_map[cursor]
1012
components.append((cursor, method, data, digest))
1013
if cursor in content_map:
1018
for component_id, method, data, digest in reversed(components):
1019
if component_id in content_map:
1020
content = content_map[component_id]
1022
if method == 'fulltext':
1023
assert content is None
1024
content = self.factory.parse_fulltext(data, version_id)
1025
elif method == 'line-delta':
1026
delta = self.factory.parse_line_delta(data, version_id)
1027
content = content.copy()
1028
content._lines = self._apply_delta(content._lines,
1030
content_map[component_id] = content
1032
if 'no-eol' in self._index.get_options(version_id):
1033
content = content.copy()
1034
line = content._lines[-1][1].rstrip('\n')
1035
content._lines[-1] = (content._lines[-1][0], line)
1036
final_content[version_id] = content
1038
# digest here is the digest from the last applied component.
1039
text = content.text()
1040
if sha_strings(text) != digest:
1041
raise KnitCorrupt(self.filename,
1042
'sha-1 does not match %s' % version_id)
1044
text_map[version_id] = text
1045
return text_map, final_content
1047
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1049
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1050
if version_ids is None:
1051
version_ids = self.versions()
1053
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
1055
pb = progress.DummyProgress()
1056
# we don't care about inclusions, the caller cares.
1057
# but we need to setup a list of records to visit.
1058
# we need version_id, position, length
1059
version_id_records = []
1060
requested_versions = set(version_ids)
1061
# filter for available versions
1062
for version_id in requested_versions:
1063
if not self.has_version(version_id):
1064
raise RevisionNotPresent(version_id, self.filename)
1065
# get a in-component-order queue:
1066
for version_id in self.versions():
1067
if version_id in requested_versions:
1068
index_memo = self._index.get_position(version_id)
1069
version_id_records.append((version_id, index_memo))
1071
total = len(version_id_records)
1072
for version_idx, (version_id, data, sha_value) in \
1073
enumerate(self._data.read_records_iter(version_id_records)):
1074
pb.update('Walking content.', version_idx, total)
1075
method = self._index.get_method(version_id)
1077
assert method in ('fulltext', 'line-delta')
1078
if method == 'fulltext':
1079
line_iterator = self.factory.get_fulltext_content(data)
1081
line_iterator = self.factory.get_linedelta_content(data)
1082
for line in line_iterator:
1085
pb.update('Walking content.', total, total)
1087
def iter_parents(self, version_ids):
1088
"""Iterate through the parents for many version ids.
1090
:param version_ids: An iterable yielding version_ids.
1091
:return: An iterator that yields (version_id, parents). Requested
1092
version_ids not present in the versioned file are simply skipped.
1093
The order is undefined, allowing for different optimisations in
1094
the underlying implementation.
1096
version_ids = [osutils.safe_revision_id(version_id) for
1097
version_id in version_ids]
1098
return self._index.iter_parents(version_ids)
1100
def num_versions(self):
1101
"""See VersionedFile.num_versions()."""
1102
return self._index.num_versions()
1104
__len__ = num_versions
1106
def annotate_iter(self, version_id):
1107
"""See VersionedFile.annotate_iter."""
1108
version_id = osutils.safe_revision_id(version_id)
1109
content = self._get_content(version_id)
1110
for origin, text in content.annotate_iter():
1113
def get_parents(self, version_id):
1114
"""See VersionedFile.get_parents."""
1117
# 52554 calls in 1264 872 internal down from 3674
1118
version_id = osutils.safe_revision_id(version_id)
1120
return self._index.get_parents(version_id)
1122
raise RevisionNotPresent(version_id, self.filename)
1124
def get_parents_with_ghosts(self, version_id):
1125
"""See VersionedFile.get_parents."""
1126
version_id = osutils.safe_revision_id(version_id)
1128
return self._index.get_parents_with_ghosts(version_id)
1130
raise RevisionNotPresent(version_id, self.filename)
1132
def get_ancestry(self, versions, topo_sorted=True):
1133
"""See VersionedFile.get_ancestry."""
1134
if isinstance(versions, basestring):
1135
versions = [versions]
1138
versions = [osutils.safe_revision_id(v) for v in versions]
1139
return self._index.get_ancestry(versions, topo_sorted)
1141
def get_ancestry_with_ghosts(self, versions):
1142
"""See VersionedFile.get_ancestry_with_ghosts."""
1143
if isinstance(versions, basestring):
1144
versions = [versions]
1147
versions = [osutils.safe_revision_id(v) for v in versions]
1148
return self._index.get_ancestry_with_ghosts(versions)
1150
def plan_merge(self, ver_a, ver_b):
1151
"""See VersionedFile.plan_merge."""
1152
ver_a = osutils.safe_revision_id(ver_a)
1153
ver_b = osutils.safe_revision_id(ver_b)
1154
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1156
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1157
annotated_a = self.annotate(ver_a)
1158
annotated_b = self.annotate(ver_b)
1159
return merge._plan_annotate_merge(annotated_a, annotated_b,
1160
ancestors_a, ancestors_b)
1163
class _KnitComponentFile(object):
1164
"""One of the files used to implement a knit database"""
1166
def __init__(self, transport, filename, mode, file_mode=None,
1167
create_parent_dir=False, dir_mode=None):
1168
self._transport = transport
1169
self._filename = filename
1171
self._file_mode = file_mode
1172
self._dir_mode = dir_mode
1173
self._create_parent_dir = create_parent_dir
1174
self._need_to_create = False
1176
def _full_path(self):
1177
"""Return the full path to this file."""
1178
return self._transport.base + self._filename
1180
def check_header(self, fp):
1181
line = fp.readline()
1183
# An empty file can actually be treated as though the file doesn't
1185
raise errors.NoSuchFile(self._full_path())
1186
if line != self.HEADER:
1187
raise KnitHeaderError(badline=line,
1188
filename=self._transport.abspath(self._filename))
1191
return '%s(%s)' % (self.__class__.__name__, self._filename)
1194
class _KnitIndex(_KnitComponentFile):
1195
"""Manages knit index file.
1197
The index is already kept in memory and read on startup, to enable
1198
fast lookups of revision information. The cursor of the index
1199
file is always pointing to the end, making it easy to append
1202
_cache is a cache for fast mapping from version id to a Index
1205
_history is a cache for fast mapping from indexes to version ids.
1207
The index data format is dictionary compressed when it comes to
1208
parent references; a index entry may only have parents that with a
1209
lover index number. As a result, the index is topological sorted.
1211
Duplicate entries may be written to the index for a single version id
1212
if this is done then the latter one completely replaces the former:
1213
this allows updates to correct version and parent information.
1214
Note that the two entries may share the delta, and that successive
1215
annotations and references MUST point to the first entry.
1217
The index file on disc contains a header, followed by one line per knit
1218
record. The same revision can be present in an index file more than once.
1219
The first occurrence gets assigned a sequence number starting from 0.
1221
The format of a single line is
1222
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1223
REVISION_ID is a utf8-encoded revision id
1224
FLAGS is a comma separated list of flags about the record. Values include
1225
no-eol, line-delta, fulltext.
1226
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1227
that the the compressed data starts at.
1228
LENGTH is the ascii representation of the length of the data file.
1229
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1231
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1232
revision id already in the knit that is a parent of REVISION_ID.
1233
The ' :' marker is the end of record marker.
1236
when a write is interrupted to the index file, it will result in a line
1237
that does not end in ' :'. If the ' :' is not present at the end of a line,
1238
or at the end of the file, then the record that is missing it will be
1239
ignored by the parser.
1241
When writing new records to the index file, the data is preceded by '\n'
1242
to ensure that records always start on new lines even if the last write was
1243
interrupted. As a result its normal for the last line in the index to be
1244
missing a trailing newline. One can be added with no harmful effects.
1247
HEADER = "# bzr knit index 8\n"
1249
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1250
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1252
def _cache_version(self, version_id, options, pos, size, parents):
1253
"""Cache a version record in the history array and index cache.
1255
This is inlined into _load_data for performance. KEEP IN SYNC.
1256
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1259
# only want the _history index to reference the 1st index entry
1261
if version_id not in self._cache:
1262
index = len(self._history)
1263
self._history.append(version_id)
1265
index = self._cache[version_id][5]
1266
self._cache[version_id] = (version_id,
1273
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1274
create_parent_dir=False, delay_create=False, dir_mode=None):
1275
_KnitComponentFile.__init__(self, transport, filename, mode,
1276
file_mode=file_mode,
1277
create_parent_dir=create_parent_dir,
1280
# position in _history is the 'official' index for a revision
1281
# but the values may have come from a newer entry.
1282
# so - wc -l of a knit index is != the number of unique names
1286
fp = self._transport.get(self._filename)
1288
# _load_data may raise NoSuchFile if the target knit is
1290
_load_data(self, fp)
1294
if mode != 'w' or not create:
1297
self._need_to_create = True
1299
self._transport.put_bytes_non_atomic(
1300
self._filename, self.HEADER, mode=self._file_mode)
1302
def get_graph(self):
1303
"""Return a list of the node:parents lists from this knit index."""
1304
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1306
def get_ancestry(self, versions, topo_sorted=True):
1307
"""See VersionedFile.get_ancestry."""
1308
# get a graph of all the mentioned versions:
1310
pending = set(versions)
1313
version = pending.pop()
1316
parents = [p for p in cache[version][4] if p in cache]
1318
raise RevisionNotPresent(version, self._filename)
1319
# if not completed and not a ghost
1320
pending.update([p for p in parents if p not in graph])
1321
graph[version] = parents
1324
return topo_sort(graph.items())
1326
def get_ancestry_with_ghosts(self, versions):
1327
"""See VersionedFile.get_ancestry_with_ghosts."""
1328
# get a graph of all the mentioned versions:
1329
self.check_versions_present(versions)
1332
pending = set(versions)
1334
version = pending.pop()
1336
parents = cache[version][4]
1342
pending.update([p for p in parents if p not in graph])
1343
graph[version] = parents
1344
return topo_sort(graph.items())
1346
def iter_parents(self, version_ids):
1347
"""Iterate through the parents for many version ids.
1349
:param version_ids: An iterable yielding version_ids.
1350
:return: An iterator that yields (version_id, parents). Requested
1351
version_ids not present in the versioned file are simply skipped.
1352
The order is undefined, allowing for different optimisations in
1353
the underlying implementation.
1355
for version_id in version_ids:
1357
yield version_id, tuple(self.get_parents(version_id))
1361
def num_versions(self):
1362
return len(self._history)
1364
__len__ = num_versions
1366
def get_versions(self):
1367
"""Get all the versions in the file. not topologically sorted."""
1368
return self._history
1370
def _version_list_to_index(self, versions):
1373
for version in versions:
1374
if version in cache:
1375
# -- inlined lookup() --
1376
result_list.append(str(cache[version][5]))
1377
# -- end lookup () --
1379
result_list.append('.' + version)
1380
return ' '.join(result_list)
1382
def add_version(self, version_id, options, index_memo, parents):
1383
"""Add a version record to the index."""
1384
self.add_versions(((version_id, options, index_memo, parents),))
1386
def add_versions(self, versions):
1387
"""Add multiple versions to the index.
1389
:param versions: a list of tuples:
1390
(version_id, options, pos, size, parents).
1393
orig_history = self._history[:]
1394
orig_cache = self._cache.copy()
1397
for version_id, options, (index, pos, size), parents in versions:
1398
line = "\n%s %s %s %s %s :" % (version_id,
1402
self._version_list_to_index(parents))
1403
assert isinstance(line, str), \
1404
'content must be utf-8 encoded: %r' % (line,)
1406
self._cache_version(version_id, options, pos, size, parents)
1407
if not self._need_to_create:
1408
self._transport.append_bytes(self._filename, ''.join(lines))
1411
sio.write(self.HEADER)
1412
sio.writelines(lines)
1414
self._transport.put_file_non_atomic(self._filename, sio,
1415
create_parent_dir=self._create_parent_dir,
1416
mode=self._file_mode,
1417
dir_mode=self._dir_mode)
1418
self._need_to_create = False
1420
# If any problems happen, restore the original values and re-raise
1421
self._history = orig_history
1422
self._cache = orig_cache
1425
def has_version(self, version_id):
1426
"""True if the version is in the index."""
1427
return version_id in self._cache
1429
def get_position(self, version_id):
1430
"""Return details needed to access the version.
1432
.kndx indices do not support split-out data, so return None for the
1435
:return: a tuple (None, data position, size) to hand to the access
1436
logic to get the record.
1438
entry = self._cache[version_id]
1439
return None, entry[2], entry[3]
1441
def get_method(self, version_id):
1442
"""Return compression method of specified version."""
1443
options = self._cache[version_id][1]
1444
if 'fulltext' in options:
1447
if 'line-delta' not in options:
1448
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1451
def get_options(self, version_id):
1452
"""Return a string represention options.
1456
return self._cache[version_id][1]
1458
def get_parents(self, version_id):
1459
"""Return parents of specified version ignoring ghosts."""
1460
return [parent for parent in self._cache[version_id][4]
1461
if parent in self._cache]
1463
def get_parents_with_ghosts(self, version_id):
1464
"""Return parents of specified version with ghosts."""
1465
return self._cache[version_id][4]
1467
def check_versions_present(self, version_ids):
1468
"""Check that all specified versions are present."""
1470
for version_id in version_ids:
1471
if version_id not in cache:
1472
raise RevisionNotPresent(version_id, self._filename)
1475
class KnitGraphIndex(object):
1476
"""A knit index that builds on GraphIndex."""
1478
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1479
"""Construct a KnitGraphIndex on a graph_index.
1481
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1482
:param deltas: Allow delta-compressed records.
1483
:param add_callback: If not None, allow additions to the index and call
1484
this callback with a list of added GraphIndex nodes:
1485
[(node, value, node_refs), ...]
1486
:param parents: If True, record knits parents, if not do not record
1489
self._graph_index = graph_index
1490
self._deltas = deltas
1491
self._add_callback = add_callback
1492
self._parents = parents
1493
if deltas and not parents:
1494
raise KnitCorrupt(self, "Cannot do delta compression without "
1497
def _get_entries(self, keys, check_present=False):
1498
"""Get the entries for keys.
1500
:param keys: An iterable of index keys, - 1-tuples.
1505
for node in self._graph_index.iter_entries(keys):
1507
found_keys.add(node[1])
1509
# adapt parentless index to the rest of the code.
1510
for node in self._graph_index.iter_entries(keys):
1511
yield node[0], node[1], node[2], ()
1512
found_keys.add(node[1])
1514
missing_keys = keys.difference(found_keys)
1516
raise RevisionNotPresent(missing_keys.pop(), self)
1518
def _present_keys(self, version_ids):
1520
node[1] for node in self._get_entries(version_ids)])
1522
def _parentless_ancestry(self, versions):
1523
"""Honour the get_ancestry API for parentless knit indices."""
1524
wanted_keys = self._version_ids_to_keys(versions)
1525
present_keys = self._present_keys(wanted_keys)
1526
missing = set(wanted_keys).difference(present_keys)
1528
raise RevisionNotPresent(missing.pop(), self)
1529
return list(self._keys_to_version_ids(present_keys))
1531
def get_ancestry(self, versions, topo_sorted=True):
1532
"""See VersionedFile.get_ancestry."""
1533
if not self._parents:
1534
return self._parentless_ancestry(versions)
1535
# XXX: This will do len(history) index calls - perhaps
1536
# it should be altered to be a index core feature?
1537
# get a graph of all the mentioned versions:
1540
versions = self._version_ids_to_keys(versions)
1541
pending = set(versions)
1543
# get all pending nodes
1544
this_iteration = pending
1545
new_nodes = self._get_entries(this_iteration)
1548
for (index, key, value, node_refs) in new_nodes:
1549
# dont ask for ghosties - otherwise
1550
# we we can end up looping with pending
1551
# being entirely ghosted.
1552
graph[key] = [parent for parent in node_refs[0]
1553
if parent not in ghosts]
1555
for parent in graph[key]:
1556
# dont examine known nodes again
1561
ghosts.update(this_iteration.difference(found))
1562
if versions.difference(graph):
1563
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1565
result_keys = topo_sort(graph.items())
1567
result_keys = graph.iterkeys()
1568
return [key[0] for key in result_keys]
1570
def get_ancestry_with_ghosts(self, versions):
1571
"""See VersionedFile.get_ancestry."""
1572
if not self._parents:
1573
return self._parentless_ancestry(versions)
1574
# XXX: This will do len(history) index calls - perhaps
1575
# it should be altered to be a index core feature?
1576
# get a graph of all the mentioned versions:
1578
versions = self._version_ids_to_keys(versions)
1579
pending = set(versions)
1581
# get all pending nodes
1582
this_iteration = pending
1583
new_nodes = self._get_entries(this_iteration)
1585
for (index, key, value, node_refs) in new_nodes:
1586
graph[key] = node_refs[0]
1588
for parent in graph[key]:
1589
# dont examine known nodes again
1593
missing_versions = this_iteration.difference(graph)
1594
missing_needed = versions.intersection(missing_versions)
1596
raise RevisionNotPresent(missing_needed.pop(), self)
1597
for missing_version in missing_versions:
1598
# add a key, no parents
1599
graph[missing_version] = []
1600
pending.discard(missing_version) # don't look for it
1601
result_keys = topo_sort(graph.items())
1602
return [key[0] for key in result_keys]
1604
def get_graph(self):
1605
"""Return a list of the node:parents lists from this knit index."""
1606
if not self._parents:
1607
return [(key, ()) for key in self.get_versions()]
1609
for index, key, value, refs in self._graph_index.iter_all_entries():
1610
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1613
def iter_parents(self, version_ids):
1614
"""Iterate through the parents for many version ids.
1616
:param version_ids: An iterable yielding version_ids.
1617
:return: An iterator that yields (version_id, parents). Requested
1618
version_ids not present in the versioned file are simply skipped.
1619
The order is undefined, allowing for different optimisations in
1620
the underlying implementation.
1623
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1625
present_parents = set()
1626
for node in all_nodes:
1627
all_parents.update(node[3][0])
1628
# any node we are querying must be present
1629
present_parents.add(node[1])
1630
unknown_parents = all_parents.difference(present_parents)
1631
present_parents.update(self._present_keys(unknown_parents))
1632
for node in all_nodes:
1634
for parent in node[3][0]:
1635
if parent in present_parents:
1636
parents.append(parent[0])
1637
yield node[1][0], tuple(parents)
1639
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1640
yield node[1][0], ()
1642
def num_versions(self):
1643
return len(list(self._graph_index.iter_all_entries()))
1645
__len__ = num_versions
1647
def get_versions(self):
1648
"""Get all the versions in the file. not topologically sorted."""
1649
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1651
def has_version(self, version_id):
1652
"""True if the version is in the index."""
1653
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1655
def _keys_to_version_ids(self, keys):
1656
return tuple(key[0] for key in keys)
1658
def get_position(self, version_id):
1659
"""Return details needed to access the version.
1661
:return: a tuple (index, data position, size) to hand to the access
1662
logic to get the record.
1664
node = self._get_node(version_id)
1665
bits = node[2][1:].split(' ')
1666
return node[0], int(bits[0]), int(bits[1])
1668
def get_method(self, version_id):
1669
"""Return compression method of specified version."""
1670
if not self._deltas:
1672
return self._parent_compression(self._get_node(version_id)[3][1])
1674
def _parent_compression(self, reference_list):
1675
# use the second reference list to decide if this is delta'd or not.
1676
if len(reference_list):
1681
def _get_node(self, version_id):
1682
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1684
def get_options(self, version_id):
1685
"""Return a string represention options.
1689
node = self._get_node(version_id)
1690
if not self._deltas:
1691
options = ['fulltext']
1693
options = [self._parent_compression(node[3][1])]
1694
if node[2][0] == 'N':
1695
options.append('no-eol')
1698
def get_parents(self, version_id):
1699
"""Return parents of specified version ignoring ghosts."""
1700
parents = list(self.iter_parents([version_id]))
1703
raise errors.RevisionNotPresent(version_id, self)
1704
return parents[0][1]
1706
def get_parents_with_ghosts(self, version_id):
1707
"""Return parents of specified version with ghosts."""
1708
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1709
check_present=True))
1710
if not self._parents:
1712
return self._keys_to_version_ids(nodes[0][3][0])
1714
def check_versions_present(self, version_ids):
1715
"""Check that all specified versions are present."""
1716
keys = self._version_ids_to_keys(version_ids)
1717
present = self._present_keys(keys)
1718
missing = keys.difference(present)
1720
raise RevisionNotPresent(missing.pop(), self)
1722
def add_version(self, version_id, options, access_memo, parents):
1723
"""Add a version record to the index."""
1724
return self.add_versions(((version_id, options, access_memo, parents),))
1726
def add_versions(self, versions):
1727
"""Add multiple versions to the index.
1729
This function does not insert data into the Immutable GraphIndex
1730
backing the KnitGraphIndex, instead it prepares data for insertion by
1731
the caller and checks that it is safe to insert then calls
1732
self._add_callback with the prepared GraphIndex nodes.
1734
:param versions: a list of tuples:
1735
(version_id, options, pos, size, parents).
1737
if not self._add_callback:
1738
raise errors.ReadOnlyError(self)
1739
# we hope there are no repositories with inconsistent parentage
1744
for (version_id, options, access_memo, parents) in versions:
1745
index, pos, size = access_memo
1746
key = (version_id, )
1747
parents = tuple((parent, ) for parent in parents)
1748
if 'no-eol' in options:
1752
value += "%d %d" % (pos, size)
1753
if not self._deltas:
1754
if 'line-delta' in options:
1755
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1758
if 'line-delta' in options:
1759
node_refs = (parents, (parents[0],))
1761
node_refs = (parents, ())
1763
node_refs = (parents, )
1766
raise KnitCorrupt(self, "attempt to add node with parents "
1767
"in parentless index.")
1769
keys[key] = (value, node_refs)
1770
present_nodes = self._get_entries(keys)
1771
for (index, key, value, node_refs) in present_nodes:
1772
if (value, node_refs) != keys[key]:
1773
raise KnitCorrupt(self, "inconsistent details in add_versions"
1774
": %s %s" % ((value, node_refs), keys[key]))
1778
for key, (value, node_refs) in keys.iteritems():
1779
result.append((key, value, node_refs))
1781
for key, (value, node_refs) in keys.iteritems():
1782
result.append((key, value))
1783
self._add_callback(result)
1785
def _version_ids_to_keys(self, version_ids):
1786
return set((version_id, ) for version_id in version_ids)
1789
class _KnitAccess(object):
1790
"""Access to knit records in a .knit file."""
1792
def __init__(self, transport, filename, _file_mode, _dir_mode,
1793
_need_to_create, _create_parent_dir):
1794
"""Create a _KnitAccess for accessing and inserting data.
1796
:param transport: The transport the .knit is located on.
1797
:param filename: The filename of the .knit.
1799
self._transport = transport
1800
self._filename = filename
1801
self._file_mode = _file_mode
1802
self._dir_mode = _dir_mode
1803
self._need_to_create = _need_to_create
1804
self._create_parent_dir = _create_parent_dir
1806
def add_raw_records(self, sizes, raw_data):
1807
"""Add raw knit bytes to a storage area.
1809
The data is spooled to whereever the access method is storing data.
1811
:param sizes: An iterable containing the size of each raw data segment.
1812
:param raw_data: A bytestring containing the data.
1813
:return: A list of memos to retrieve the record later. Each memo is a
1814
tuple - (index, pos, length), where the index field is always None
1815
for the .knit access method.
1817
assert type(raw_data) == str, \
1818
'data must be plain bytes was %s' % type(raw_data)
1819
if not self._need_to_create:
1820
base = self._transport.append_bytes(self._filename, raw_data)
1822
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1823
create_parent_dir=self._create_parent_dir,
1824
mode=self._file_mode,
1825
dir_mode=self._dir_mode)
1826
self._need_to_create = False
1830
result.append((None, base, size))
1835
"""IFF this data access has its own storage area, initialise it.
1839
self._transport.put_bytes_non_atomic(self._filename, '',
1840
mode=self._file_mode)
1842
def open_file(self):
1843
"""IFF this data access can be represented as a single file, open it.
1845
For knits that are not mapped to a single file on disk this will
1848
:return: None or a file handle.
1851
return self._transport.get(self._filename)
1856
def get_raw_records(self, memos_for_retrieval):
1857
"""Get the raw bytes for a records.
1859
:param memos_for_retrieval: An iterable containing the (index, pos,
1860
length) memo for retrieving the bytes. The .knit method ignores
1861
the index as there is always only a single file.
1862
:return: An iterator over the bytes of the records.
1864
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
1865
for pos, data in self._transport.readv(self._filename, read_vector):
1869
class _PackAccess(object):
1870
"""Access to knit records via a collection of packs."""
1872
def __init__(self, index_to_packs, writer=None):
1873
"""Create a _PackAccess object.
1875
:param index_to_packs: A dict mapping index objects to the transport
1876
and file names for obtaining data.
1877
:param writer: A tuple (pack.ContainerWriter, write_index) which
1878
contains the pack to write, and the index that reads from it will
1882
self.container_writer = writer[0]
1883
self.write_index = writer[1]
1885
self.container_writer = None
1886
self.write_index = None
1887
self.indices = index_to_packs
1889
def add_raw_records(self, sizes, raw_data):
1890
"""Add raw knit bytes to a storage area.
1892
The data is spooled to the container writer in one bytes-record per
1895
:param sizes: An iterable containing the size of each raw data segment.
1896
:param raw_data: A bytestring containing the data.
1897
:return: A list of memos to retrieve the record later. Each memo is a
1898
tuple - (index, pos, length), where the index field is the
1899
write_index object supplied to the PackAccess object.
1901
assert type(raw_data) == str, \
1902
'data must be plain bytes was %s' % type(raw_data)
1906
p_offset, p_length = self.container_writer.add_bytes_record(
1907
raw_data[offset:offset+size], [])
1909
result.append((self.write_index, p_offset, p_length))
1913
"""Pack based knits do not get individually created."""
1915
def get_raw_records(self, memos_for_retrieval):
1916
"""Get the raw bytes for a records.
1918
:param memos_for_retrieval: An iterable containing the (index, pos,
1919
length) memo for retrieving the bytes. The Pack access method
1920
looks up the pack to use for a given record in its index_to_pack
1922
:return: An iterator over the bytes of the records.
1924
# first pass, group into same-index requests
1926
current_index = None
1927
for (index, offset, length) in memos_for_retrieval:
1928
if current_index == index:
1929
current_list.append((offset, length))
1931
if current_index is not None:
1932
request_lists.append((current_index, current_list))
1933
current_index = index
1934
current_list = [(offset, length)]
1935
# handle the last entry
1936
if current_index is not None:
1937
request_lists.append((current_index, current_list))
1938
for index, offsets in request_lists:
1939
transport, path = self.indices[index]
1940
reader = pack.make_readv_reader(transport, path, offsets)
1941
for names, read_func in reader.iter_records():
1942
yield read_func(None)
1944
def open_file(self):
1945
"""Pack based knits have no single file."""
1948
def set_writer(self, writer, index, (transport, packname)):
1949
"""Set a writer to use for adding data."""
1950
self.indices[index] = (transport, packname)
1951
self.container_writer = writer
1952
self.write_index = index
1955
class _KnitData(object):
1956
"""Manage extraction of data from a KnitAccess, caching and decompressing.
1958
The KnitData class provides the logic for parsing and using knit records,
1959
making use of an access method for the low level read and write operations.
1962
def __init__(self, access):
1963
"""Create a KnitData object.
1965
:param access: The access method to use. Access methods such as
1966
_KnitAccess manage the insertion of raw records and the subsequent
1967
retrieval of the same.
1969
self._access = access
1970
self._checked = False
1971
# TODO: jam 20060713 conceptually, this could spill to disk
1972
# if the cached size gets larger than a certain amount
1973
# but it complicates the model a bit, so for now just use
1974
# a simple dictionary
1976
self._do_cache = False
1978
def enable_cache(self):
1979
"""Enable caching of reads."""
1980
self._do_cache = True
1982
def clear_cache(self):
1983
"""Clear the record cache."""
1984
self._do_cache = False
1987
def _open_file(self):
1988
return self._access.open_file()
1990
def _record_to_data(self, version_id, digest, lines):
1991
"""Convert version_id, digest, lines into a raw data block.
1993
:return: (len, a StringIO instance with the raw data ready to read.)
1996
data_file = GzipFile(None, mode='wb', fileobj=sio)
1998
assert isinstance(version_id, str)
1999
data_file.writelines(chain(
2000
["version %s %d %s\n" % (version_id,
2004
["end %s\n" % version_id]))
2011
def add_raw_records(self, sizes, raw_data):
2012
"""Append a prepared record to the data file.
2014
:param sizes: An iterable containing the size of each raw data segment.
2015
:param raw_data: A bytestring containing the data.
2016
:return: a list of index data for the way the data was stored.
2017
See the access method add_raw_records documentation for more
2020
return self._access.add_raw_records(sizes, raw_data)
2022
def add_record(self, version_id, digest, lines):
2023
"""Write new text record to disk.
2025
Returns index data for retrieving it later, as per add_raw_records.
2027
size, sio = self._record_to_data(version_id, digest, lines)
2028
result = self.add_raw_records([size], sio.getvalue())
2030
self._cache[version_id] = sio.getvalue()
2033
def _parse_record_header(self, version_id, raw_data):
2034
"""Parse a record header for consistency.
2036
:return: the header and the decompressor stream.
2037
as (stream, header_record)
2039
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2041
rec = self._check_header(version_id, df.readline())
2042
except Exception, e:
2043
raise KnitCorrupt(self._access,
2044
"While reading {%s} got %s(%s)"
2045
% (version_id, e.__class__.__name__, str(e)))
2048
def _check_header(self, version_id, line):
2051
raise KnitCorrupt(self._access,
2052
'unexpected number of elements in record header')
2053
if rec[1] != version_id:
2054
raise KnitCorrupt(self._access,
2055
'unexpected version, wanted %r, got %r'
2056
% (version_id, rec[1]))
2059
def _parse_record(self, version_id, data):
2061
# 4168 calls in 2880 217 internal
2062
# 4168 calls to _parse_record_header in 2121
2063
# 4168 calls to readlines in 330
2064
df = GzipFile(mode='rb', fileobj=StringIO(data))
2067
record_contents = df.readlines()
2068
except Exception, e:
2069
raise KnitCorrupt(self._access,
2070
"While reading {%s} got %s(%s)"
2071
% (version_id, e.__class__.__name__, str(e)))
2072
header = record_contents.pop(0)
2073
rec = self._check_header(version_id, header)
2075
last_line = record_contents.pop()
2076
if len(record_contents) != int(rec[2]):
2077
raise KnitCorrupt(self._access,
2078
'incorrect number of lines %s != %s'
2080
% (len(record_contents), int(rec[2]),
2082
if last_line != 'end %s\n' % rec[1]:
2083
raise KnitCorrupt(self._access,
2084
'unexpected version end line %r, wanted %r'
2085
% (last_line, version_id))
2087
return record_contents, rec[3]
2089
def read_records_iter_raw(self, records):
2090
"""Read text records from data file and yield raw data.
2092
This unpacks enough of the text record to validate the id is
2093
as expected but thats all.
2095
# setup an iterator of the external records:
2096
# uses readv so nice and fast we hope.
2098
# grab the disk data needed.
2100
# Don't check _cache if it is empty
2101
needed_offsets = [index_memo for version_id, index_memo
2103
if version_id not in self._cache]
2105
needed_offsets = [index_memo for version_id, index_memo
2108
raw_records = self._access.get_raw_records(needed_offsets)
2110
for version_id, index_memo in records:
2111
if version_id in self._cache:
2112
# This data has already been validated
2113
data = self._cache[version_id]
2115
data = raw_records.next()
2117
self._cache[version_id] = data
2119
# validate the header
2120
df, rec = self._parse_record_header(version_id, data)
2122
yield version_id, data
2124
def read_records_iter(self, records):
2125
"""Read text records from data file and yield result.
2127
The result will be returned in whatever is the fastest to read.
2128
Not by the order requested. Also, multiple requests for the same
2129
record will only yield 1 response.
2130
:param records: A list of (version_id, pos, len) entries
2131
:return: Yields (version_id, contents, digest) in the order
2132
read, not the order requested
2138
# Skip records we have alread seen
2139
yielded_records = set()
2140
needed_records = set()
2141
for record in records:
2142
if record[0] in self._cache:
2143
if record[0] in yielded_records:
2145
yielded_records.add(record[0])
2146
data = self._cache[record[0]]
2147
content, digest = self._parse_record(record[0], data)
2148
yield (record[0], content, digest)
2150
needed_records.add(record)
2151
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2153
needed_records = sorted(set(records), key=operator.itemgetter(1))
2155
if not needed_records:
2158
# The transport optimizes the fetching as well
2159
# (ie, reads continuous ranges.)
2160
raw_data = self._access.get_raw_records(
2161
[index_memo for version_id, index_memo in needed_records])
2163
for (version_id, index_memo), data in \
2164
izip(iter(needed_records), raw_data):
2165
content, digest = self._parse_record(version_id, data)
2167
self._cache[version_id] = data
2168
yield version_id, content, digest
2170
def read_records(self, records):
2171
"""Read records into a dictionary."""
2173
for record_id, content, digest in \
2174
self.read_records_iter(records):
2175
components[record_id] = (content, digest)
2179
class InterKnit(InterVersionedFile):
2180
"""Optimised code paths for knit to knit operations."""
2182
_matching_file_from_factory = KnitVersionedFile
2183
_matching_file_to_factory = KnitVersionedFile
2186
def is_compatible(source, target):
2187
"""Be compatible with knits. """
2189
return (isinstance(source, KnitVersionedFile) and
2190
isinstance(target, KnitVersionedFile))
2191
except AttributeError:
2194
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2195
"""See InterVersionedFile.join."""
2196
assert isinstance(self.source, KnitVersionedFile)
2197
assert isinstance(self.target, KnitVersionedFile)
2199
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2204
pb = ui.ui_factory.nested_progress_bar()
2206
version_ids = list(version_ids)
2207
if None in version_ids:
2208
version_ids.remove(None)
2210
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2211
this_versions = set(self.target._index.get_versions())
2212
needed_versions = self.source_ancestry - this_versions
2213
cross_check_versions = self.source_ancestry.intersection(this_versions)
2214
mismatched_versions = set()
2215
for version in cross_check_versions:
2216
# scan to include needed parents.
2217
n1 = set(self.target.get_parents_with_ghosts(version))
2218
n2 = set(self.source.get_parents_with_ghosts(version))
2220
# FIXME TEST this check for cycles being introduced works
2221
# the logic is we have a cycle if in our graph we are an
2222
# ancestor of any of the n2 revisions.
2228
parent_ancestors = self.source.get_ancestry(parent)
2229
if version in parent_ancestors:
2230
raise errors.GraphCycleError([parent, version])
2231
# ensure this parent will be available later.
2232
new_parents = n2.difference(n1)
2233
needed_versions.update(new_parents.difference(this_versions))
2234
mismatched_versions.add(version)
2236
if not needed_versions and not mismatched_versions:
2238
full_list = topo_sort(self.source.get_graph())
2240
version_list = [i for i in full_list if (not self.target.has_version(i)
2241
and i in needed_versions)]
2245
copy_queue_records = []
2247
for version_id in version_list:
2248
options = self.source._index.get_options(version_id)
2249
parents = self.source._index.get_parents_with_ghosts(version_id)
2250
# check that its will be a consistent copy:
2251
for parent in parents:
2252
# if source has the parent, we must :
2253
# * already have it or
2254
# * have it scheduled already
2255
# otherwise we don't care
2256
assert (self.target.has_version(parent) or
2257
parent in copy_set or
2258
not self.source.has_version(parent))
2259
index_memo = self.source._index.get_position(version_id)
2260
copy_queue_records.append((version_id, index_memo))
2261
copy_queue.append((version_id, options, parents))
2262
copy_set.add(version_id)
2264
# data suck the join:
2266
total = len(version_list)
2269
for (version_id, raw_data), \
2270
(version_id2, options, parents) in \
2271
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2273
assert version_id == version_id2, 'logic error, inconsistent results'
2275
pb.update("Joining knit", count, total)
2276
raw_records.append((version_id, options, parents, len(raw_data)))
2277
raw_datum.append(raw_data)
2278
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2280
for version in mismatched_versions:
2281
# FIXME RBC 20060309 is this needed?
2282
n1 = set(self.target.get_parents_with_ghosts(version))
2283
n2 = set(self.source.get_parents_with_ghosts(version))
2284
# write a combined record to our history preserving the current
2285
# parents as first in the list
2286
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2287
self.target.fix_parents(version, new_parents)
2293
InterVersionedFile.register_optimiser(InterKnit)
2296
class WeaveToKnit(InterVersionedFile):
2297
"""Optimised code paths for weave to knit operations."""
2299
_matching_file_from_factory = bzrlib.weave.WeaveFile
2300
_matching_file_to_factory = KnitVersionedFile
2303
def is_compatible(source, target):
2304
"""Be compatible with weaves to knits."""
2306
return (isinstance(source, bzrlib.weave.Weave) and
2307
isinstance(target, KnitVersionedFile))
2308
except AttributeError:
2311
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2312
"""See InterVersionedFile.join."""
2313
assert isinstance(self.source, bzrlib.weave.Weave)
2314
assert isinstance(self.target, KnitVersionedFile)
2316
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2321
pb = ui.ui_factory.nested_progress_bar()
2323
version_ids = list(version_ids)
2325
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2326
this_versions = set(self.target._index.get_versions())
2327
needed_versions = self.source_ancestry - this_versions
2328
cross_check_versions = self.source_ancestry.intersection(this_versions)
2329
mismatched_versions = set()
2330
for version in cross_check_versions:
2331
# scan to include needed parents.
2332
n1 = set(self.target.get_parents_with_ghosts(version))
2333
n2 = set(self.source.get_parents(version))
2334
# if all of n2's parents are in n1, then its fine.
2335
if n2.difference(n1):
2336
# FIXME TEST this check for cycles being introduced works
2337
# the logic is we have a cycle if in our graph we are an
2338
# ancestor of any of the n2 revisions.
2344
parent_ancestors = self.source.get_ancestry(parent)
2345
if version in parent_ancestors:
2346
raise errors.GraphCycleError([parent, version])
2347
# ensure this parent will be available later.
2348
new_parents = n2.difference(n1)
2349
needed_versions.update(new_parents.difference(this_versions))
2350
mismatched_versions.add(version)
2352
if not needed_versions and not mismatched_versions:
2354
full_list = topo_sort(self.source.get_graph())
2356
version_list = [i for i in full_list if (not self.target.has_version(i)
2357
and i in needed_versions)]
2361
total = len(version_list)
2362
for version_id in version_list:
2363
pb.update("Converting to knit", count, total)
2364
parents = self.source.get_parents(version_id)
2365
# check that its will be a consistent copy:
2366
for parent in parents:
2367
# if source has the parent, we must already have it
2368
assert (self.target.has_version(parent))
2369
self.target.add_lines(
2370
version_id, parents, self.source.get_lines(version_id))
2373
for version in mismatched_versions:
2374
# FIXME RBC 20060309 is this needed?
2375
n1 = set(self.target.get_parents_with_ghosts(version))
2376
n2 = set(self.source.get_parents(version))
2377
# write a combined record to our history preserving the current
2378
# parents as first in the list
2379
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2380
self.target.fix_parents(version, new_parents)
2386
InterVersionedFile.register_optimiser(WeaveToKnit)
2389
class KnitSequenceMatcher(difflib.SequenceMatcher):
2390
"""Knit tuned sequence matcher.
2392
This is based on profiling of difflib which indicated some improvements
2393
for our usage pattern.
2396
def find_longest_match(self, alo, ahi, blo, bhi):
2397
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
2399
If isjunk is not defined:
2401
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
2402
alo <= i <= i+k <= ahi
2403
blo <= j <= j+k <= bhi
2404
and for all (i',j',k') meeting those conditions,
2407
and if i == i', j <= j'
2409
In other words, of all maximal matching blocks, return one that
2410
starts earliest in a, and of all those maximal matching blocks that
2411
start earliest in a, return the one that starts earliest in b.
2413
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
2414
>>> s.find_longest_match(0, 5, 0, 9)
2417
If isjunk is defined, first the longest matching block is
2418
determined as above, but with the additional restriction that no
2419
junk element appears in the block. Then that block is extended as
2420
far as possible by matching (only) junk elements on both sides. So
2421
the resulting block never matches on junk except as identical junk
2422
happens to be adjacent to an "interesting" match.
2424
Here's the same example as before, but considering blanks to be
2425
junk. That prevents " abcd" from matching the " abcd" at the tail
2426
end of the second sequence directly. Instead only the "abcd" can
2427
match, and matches the leftmost "abcd" in the second sequence:
2429
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
2430
>>> s.find_longest_match(0, 5, 0, 9)
2433
If no blocks match, return (alo, blo, 0).
2435
>>> s = SequenceMatcher(None, "ab", "c")
2436
>>> s.find_longest_match(0, 2, 0, 1)
2440
# CAUTION: stripping common prefix or suffix would be incorrect.
2444
# Longest matching block is "ab", but if common prefix is
2445
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
2446
# strip, so ends up claiming that ab is changed to acab by
2447
# inserting "ca" in the middle. That's minimal but unintuitive:
2448
# "it's obvious" that someone inserted "ac" at the front.
2449
# Windiff ends up at the same place as diff, but by pairing up
2450
# the unique 'b's and then matching the first two 'a's.
2452
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
2453
besti, bestj, bestsize = alo, blo, 0
2454
# find longest junk-free match
2455
# during an iteration of the loop, j2len[j] = length of longest
2456
# junk-free match ending with a[i-1] and b[j]
2460
for i in xrange(alo, ahi):
2461
# look at all instances of a[i] in b; note that because
2462
# b2j has no junk keys, the loop is skipped if a[i] is junk
2463
j2lenget = j2len.get
2466
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
2467
# following improvement
2468
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
2469
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
2470
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
2472
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
2473
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
2474
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
2486
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
2488
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
2491
# Extend the best by non-junk elements on each end. In particular,
2492
# "popular" non-junk elements aren't in b2j, which greatly speeds
2493
# the inner loop above, but also means "the best" match so far
2494
# doesn't contain any junk *or* popular non-junk elements.
2495
while besti > alo and bestj > blo and \
2496
not isbjunk(b[bestj-1]) and \
2497
a[besti-1] == b[bestj-1]:
2498
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2499
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2500
not isbjunk(b[bestj+bestsize]) and \
2501
a[besti+bestsize] == b[bestj+bestsize]:
2504
# Now that we have a wholly interesting match (albeit possibly
2505
# empty!), we may as well suck up the matching junk on each
2506
# side of it too. Can't think of a good reason not to, and it
2507
# saves post-processing the (possibly considerable) expense of
2508
# figuring out what to do with it. In the case of an empty
2509
# interesting match, this is clearly the right thing to do,
2510
# because no other kind of match is possible in the regions.
2511
while besti > alo and bestj > blo and \
2512
isbjunk(b[bestj-1]) and \
2513
a[besti-1] == b[bestj-1]:
2514
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2515
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2516
isbjunk(b[bestj+bestsize]) and \
2517
a[besti+bestsize] == b[bestj+bestsize]:
2518
bestsize = bestsize + 1
2520
return besti, bestj, bestsize
2524
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2526
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
