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
71
from zlib import Z_DEFAULT_COMPRESSION
74
from bzrlib.lazy_import import lazy_import
75
lazy_import(globals(), """
92
from bzrlib.errors import (
98
KnitDataStreamIncompatible,
101
RevisionAlreadyPresent,
103
from bzrlib.tuned_gzip import GzipFile
104
from bzrlib.osutils import (
109
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
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 extracted.
573
Unlike some other knit methods, this is not used to generate a
574
transitive closure, rather it is used precisely as given.
576
:returns: format_signature, list of (version, options, length, parents),
579
required_versions = set([osutils.safe_revision_id(v) for v in
581
# we don't care about inclusions, the caller cares.
582
# but we need to setup a list of records to visit.
583
for version_id in required_versions:
584
if not self.has_version(version_id):
585
raise RevisionNotPresent(version_id, self.filename)
586
# Pick the desired versions out of the index in oldest-to-newest order
588
for version_id in self.versions():
589
if version_id in required_versions:
590
version_list.append(version_id)
592
# create the list of version information for the result
593
copy_queue_records = []
595
result_version_list = []
596
for version_id in version_list:
597
options = self._index.get_options(version_id)
598
parents = self._index.get_parents_with_ghosts(version_id)
599
index_memo = self._index.get_position(version_id)
600
copy_queue_records.append((version_id, index_memo))
601
none, data_pos, data_size = index_memo
602
copy_set.add(version_id)
603
# version, options, length, parents
604
result_version_list.append((version_id, options, data_size,
607
# Read the compressed record data.
609
# From here down to the return should really be logic in the returned
610
# callable -- in a class that adapts read_records_iter_raw to read
613
for (version_id, raw_data), \
614
(version_id2, options, _, parents) in \
615
izip(self._data.read_records_iter_raw(copy_queue_records),
616
result_version_list):
617
assert version_id == version_id2, 'logic error, inconsistent results'
618
raw_datum.append(raw_data)
619
pseudo_file = StringIO(''.join(raw_datum))
622
return pseudo_file.read()
624
return pseudo_file.read(length)
625
return (self.get_format_signature(), result_version_list, read)
627
def _extract_blocks(self, version_id, source, target):
628
if self._index.get_method(version_id) != 'line-delta':
630
parent, sha1, noeol, delta = self.get_delta(version_id)
631
return KnitContent.get_line_delta_blocks(delta, source, target)
633
def get_delta(self, version_id):
634
"""Get a delta for constructing version from some other version."""
635
version_id = osutils.safe_revision_id(version_id)
636
self.check_not_reserved_id(version_id)
637
if not self.has_version(version_id):
638
raise RevisionNotPresent(version_id, self.filename)
640
parents = self.get_parents(version_id)
645
index_memo = self._index.get_position(version_id)
646
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
647
noeol = 'no-eol' in self._index.get_options(version_id)
648
if 'fulltext' == self._index.get_method(version_id):
649
new_content = self.factory.parse_fulltext(data, version_id)
650
if parent is not None:
651
reference_content = self._get_content(parent)
652
old_texts = reference_content.text()
655
new_texts = new_content.text()
656
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
657
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
659
delta = self.factory.parse_line_delta(data, version_id)
660
return parent, sha1, noeol, delta
662
def get_format_signature(self):
663
"""See VersionedFile.get_format_signature()."""
664
if self.factory.annotated:
665
annotated_part = "annotated"
667
annotated_part = "plain"
668
return "knit-%s" % (annotated_part,)
670
def get_graph_with_ghosts(self):
671
"""See VersionedFile.get_graph_with_ghosts()."""
672
graph_items = self._index.get_graph()
673
return dict(graph_items)
675
def get_sha1(self, version_id):
676
return self.get_sha1s([version_id])[0]
678
def get_sha1s(self, version_ids):
679
"""See VersionedFile.get_sha1()."""
680
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
681
record_map = self._get_record_map(version_ids)
682
# record entry 2 is the 'digest'.
683
return [record_map[v][2] for v in version_ids]
687
"""See VersionedFile.get_suffixes()."""
688
return [DATA_SUFFIX, INDEX_SUFFIX]
690
def has_ghost(self, version_id):
691
"""True if there is a ghost reference in the file to version_id."""
692
version_id = osutils.safe_revision_id(version_id)
694
if self.has_version(version_id):
696
# optimisable if needed by memoising the _ghosts set.
697
items = self._index.get_graph()
698
for node, parents in items:
699
for parent in parents:
700
if parent not in self._index._cache:
701
if parent == version_id:
705
def insert_data_stream(self, (format, data_list, reader_callable)):
706
"""Insert knit records from a data stream into this knit.
708
If a version in the stream is already present in this knit, it will not
709
be inserted a second time. It will be checked for consistency with the
710
stored version however, and may cause a KnitCorrupt error to be raised
711
if the data in the stream disagrees with the already stored data.
713
:seealso: get_data_stream
715
if format != self.get_format_signature():
716
trace.mutter('incompatible format signature inserting to %r', self)
717
raise KnitDataStreamIncompatible(
718
format, self.get_format_signature())
720
for version_id, options, length, parents in data_list:
721
if self.has_version(version_id):
722
# First check: the list of parents.
723
my_parents = self.get_parents_with_ghosts(version_id)
724
if my_parents != parents:
725
# XXX: KnitCorrupt is not quite the right exception here.
728
'parents list %r from data stream does not match '
729
'already recorded parents %r for %s'
730
% (parents, my_parents, version_id))
732
# Also check the SHA-1 of the fulltext this content will
734
raw_data = reader_callable(length)
735
my_fulltext_sha1 = self.get_sha1(version_id)
736
df, rec = self._data._parse_record_header(version_id, raw_data)
737
stream_fulltext_sha1 = rec[3]
738
if my_fulltext_sha1 != stream_fulltext_sha1:
739
# Actually, we don't know if it's this knit that's corrupt,
740
# or the data stream we're trying to insert.
742
self.filename, 'sha-1 does not match %s' % version_id)
744
self._add_raw_records(
745
[(version_id, options, parents, length)],
746
reader_callable(length))
749
"""See VersionedFile.versions."""
750
if 'evil' in debug.debug_flags:
751
trace.mutter_callsite(2, "versions scales with size of history")
752
return self._index.get_versions()
754
def has_version(self, version_id):
755
"""See VersionedFile.has_version."""
756
if 'evil' in debug.debug_flags:
757
trace.mutter_callsite(2, "has_version is a LBYL scenario")
758
version_id = osutils.safe_revision_id(version_id)
759
return self._index.has_version(version_id)
761
__contains__ = has_version
763
def _merge_annotations(self, content, parents, parent_texts={},
764
delta=None, annotated=None,
765
left_matching_blocks=None):
766
"""Merge annotations for content. This is done by comparing
767
the annotations based on changed to the text.
769
if left_matching_blocks is not None:
770
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
774
for parent_id in parents:
775
merge_content = self._get_content(parent_id, parent_texts)
776
if (parent_id == parents[0] and delta_seq is not None):
779
seq = patiencediff.PatienceSequenceMatcher(
780
None, merge_content.text(), content.text())
781
for i, j, n in seq.get_matching_blocks():
784
# this appears to copy (origin, text) pairs across to the
785
# new content for any line that matches the last-checked
787
content._lines[j:j+n] = merge_content._lines[i:i+n]
789
if delta_seq is None:
790
reference_content = self._get_content(parents[0], parent_texts)
791
new_texts = content.text()
792
old_texts = reference_content.text()
793
delta_seq = patiencediff.PatienceSequenceMatcher(
794
None, old_texts, new_texts)
795
return self._make_line_delta(delta_seq, content)
797
def _make_line_delta(self, delta_seq, new_content):
798
"""Generate a line delta from delta_seq and new_content."""
800
for op in delta_seq.get_opcodes():
803
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
806
def _get_components_positions(self, version_ids):
807
"""Produce a map of position data for the components of versions.
809
This data is intended to be used for retrieving the knit records.
811
A dict of version_id to (method, data_pos, data_size, next) is
813
method is the way referenced data should be applied.
814
data_pos is the position of the data in the knit.
815
data_size is the size of the data in the knit.
816
next is the build-parent of the version, or None for fulltexts.
819
for version_id in version_ids:
822
while cursor is not None and cursor not in component_data:
823
method = self._index.get_method(cursor)
824
if method == 'fulltext':
827
next = self.get_parents(cursor)[0]
828
index_memo = self._index.get_position(cursor)
829
component_data[cursor] = (method, index_memo, next)
831
return component_data
833
def _get_content(self, version_id, parent_texts={}):
834
"""Returns a content object that makes up the specified
836
if not self.has_version(version_id):
837
raise RevisionNotPresent(version_id, self.filename)
839
cached_version = parent_texts.get(version_id, None)
840
if cached_version is not None:
841
return cached_version
843
text_map, contents_map = self._get_content_maps([version_id])
844
return contents_map[version_id]
846
def _check_versions_present(self, version_ids):
847
"""Check that all specified versions are present."""
848
self._index.check_versions_present(version_ids)
850
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
851
"""See VersionedFile.add_lines_with_ghosts()."""
852
self._check_add(version_id, lines)
853
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
855
def _add_lines(self, version_id, parents, lines, parent_texts,
856
left_matching_blocks=None):
857
"""See VersionedFile.add_lines."""
858
self._check_add(version_id, lines)
859
self._check_versions_present(parents)
860
return self._add(version_id, lines[:], parents, self.delta,
861
parent_texts, left_matching_blocks)
863
def _check_add(self, version_id, lines):
864
"""check that version_id and lines are safe to add."""
865
assert self.writable, "knit is not opened for write"
866
### FIXME escape. RBC 20060228
867
if contains_whitespace(version_id):
868
raise InvalidRevisionId(version_id, self.filename)
869
self.check_not_reserved_id(version_id)
870
if self.has_version(version_id):
871
raise RevisionAlreadyPresent(version_id, self.filename)
872
self._check_lines_not_unicode(lines)
873
self._check_lines_are_lines(lines)
875
def _add(self, version_id, lines, parents, delta, parent_texts,
876
left_matching_blocks=None):
877
"""Add a set of lines on top of version specified by parents.
879
If delta is true, compress the text as a line-delta against
882
Any versions not present will be converted into ghosts.
884
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
885
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
886
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
887
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
888
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
889
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
890
# +1383 0 8.0370 8.0370 +<len>
891
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
892
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
893
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
894
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
898
if parent_texts is None:
900
for parent in parents:
901
if not self.has_version(parent):
902
ghosts.append(parent)
904
present_parents.append(parent)
906
if delta and not len(present_parents):
909
digest = sha_strings(lines)
912
if lines[-1][-1] != '\n':
913
options.append('no-eol')
914
lines[-1] = lines[-1] + '\n'
916
if len(present_parents) and delta:
917
# To speed the extract of texts the delta chain is limited
918
# to a fixed number of deltas. This should minimize both
919
# I/O and the time spend applying deltas.
920
delta = self._check_should_delta(present_parents)
922
assert isinstance(version_id, str)
923
lines = self.factory.make(lines, version_id)
924
if delta or (self.factory.annotated and len(present_parents) > 0):
925
# Merge annotations from parent texts if so is needed.
926
delta_hunks = self._merge_annotations(lines, present_parents,
927
parent_texts, delta, self.factory.annotated,
928
left_matching_blocks)
931
options.append('line-delta')
932
store_lines = self.factory.lower_line_delta(delta_hunks)
934
options.append('fulltext')
935
store_lines = self.factory.lower_fulltext(lines)
937
access_memo = self._data.add_record(version_id, digest, store_lines)
938
self._index.add_version(version_id, options, access_memo, parents)
941
def check(self, progress_bar=None):
942
"""See VersionedFile.check()."""
944
def _clone_text(self, new_version_id, old_version_id, parents):
945
"""See VersionedFile.clone_text()."""
946
# FIXME RBC 20060228 make fast by only inserting an index with null
948
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
950
def get_lines(self, version_id):
951
"""See VersionedFile.get_lines()."""
952
return self.get_line_list([version_id])[0]
954
def _get_record_map(self, version_ids):
955
"""Produce a dictionary of knit records.
957
The keys are version_ids, the values are tuples of (method, content,
959
method is the way the content should be applied.
960
content is a KnitContent object.
961
digest is the SHA1 digest of this version id after all steps are done
962
next is the build-parent of the version, i.e. the leftmost ancestor.
963
If the method is fulltext, next will be None.
965
position_map = self._get_components_positions(version_ids)
966
# c = component_id, m = method, i_m = index_memo, n = next
967
records = [(c, i_m) for c, (m, i_m, n) in position_map.iteritems()]
969
for component_id, content, digest in \
970
self._data.read_records_iter(records):
971
method, index_memo, next = position_map[component_id]
972
record_map[component_id] = method, content, digest, next
976
def get_text(self, version_id):
977
"""See VersionedFile.get_text"""
978
return self.get_texts([version_id])[0]
980
def get_texts(self, version_ids):
981
return [''.join(l) for l in self.get_line_list(version_ids)]
983
def get_line_list(self, version_ids):
984
"""Return the texts of listed versions as a list of strings."""
985
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
986
for version_id in version_ids:
987
self.check_not_reserved_id(version_id)
988
text_map, content_map = self._get_content_maps(version_ids)
989
return [text_map[v] for v in version_ids]
991
_get_lf_split_line_list = get_line_list
993
def _get_content_maps(self, version_ids):
994
"""Produce maps of text and KnitContents
996
:return: (text_map, content_map) where text_map contains the texts for
997
the requested versions and content_map contains the KnitContents.
998
Both dicts take version_ids as their keys.
1000
for version_id in version_ids:
1001
if not self.has_version(version_id):
1002
raise RevisionNotPresent(version_id, self.filename)
1003
record_map = self._get_record_map(version_ids)
1008
for version_id in version_ids:
1011
while cursor is not None:
1012
method, data, digest, next = record_map[cursor]
1013
components.append((cursor, method, data, digest))
1014
if cursor in content_map:
1019
for component_id, method, data, digest in reversed(components):
1020
if component_id in content_map:
1021
content = content_map[component_id]
1023
if method == 'fulltext':
1024
assert content is None
1025
content = self.factory.parse_fulltext(data, version_id)
1026
elif method == 'line-delta':
1027
delta = self.factory.parse_line_delta(data, version_id)
1028
content = content.copy()
1029
content._lines = self._apply_delta(content._lines,
1031
content_map[component_id] = content
1033
if 'no-eol' in self._index.get_options(version_id):
1034
content = content.copy()
1035
line = content._lines[-1][1].rstrip('\n')
1036
content._lines[-1] = (content._lines[-1][0], line)
1037
final_content[version_id] = content
1039
# digest here is the digest from the last applied component.
1040
text = content.text()
1041
if sha_strings(text) != digest:
1042
raise KnitCorrupt(self.filename,
1043
'sha-1 does not match %s' % version_id)
1045
text_map[version_id] = text
1046
return text_map, final_content
1048
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1050
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1051
if version_ids is None:
1052
version_ids = self.versions()
1054
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
1056
pb = progress.DummyProgress()
1057
# we don't care about inclusions, the caller cares.
1058
# but we need to setup a list of records to visit.
1059
# we need version_id, position, length
1060
version_id_records = []
1061
requested_versions = set(version_ids)
1062
# filter for available versions
1063
for version_id in requested_versions:
1064
if not self.has_version(version_id):
1065
raise RevisionNotPresent(version_id, self.filename)
1066
# get a in-component-order queue:
1067
for version_id in self.versions():
1068
if version_id in requested_versions:
1069
index_memo = self._index.get_position(version_id)
1070
version_id_records.append((version_id, index_memo))
1072
total = len(version_id_records)
1073
for version_idx, (version_id, data, sha_value) in \
1074
enumerate(self._data.read_records_iter(version_id_records)):
1075
pb.update('Walking content.', version_idx, total)
1076
method = self._index.get_method(version_id)
1078
assert method in ('fulltext', 'line-delta')
1079
if method == 'fulltext':
1080
line_iterator = self.factory.get_fulltext_content(data)
1082
line_iterator = self.factory.get_linedelta_content(data)
1083
for line in line_iterator:
1086
pb.update('Walking content.', total, total)
1088
def iter_parents(self, version_ids):
1089
"""Iterate through the parents for many version ids.
1091
:param version_ids: An iterable yielding version_ids.
1092
:return: An iterator that yields (version_id, parents). Requested
1093
version_ids not present in the versioned file are simply skipped.
1094
The order is undefined, allowing for different optimisations in
1095
the underlying implementation.
1097
version_ids = [osutils.safe_revision_id(version_id) for
1098
version_id in version_ids]
1099
return self._index.iter_parents(version_ids)
1101
def num_versions(self):
1102
"""See VersionedFile.num_versions()."""
1103
return self._index.num_versions()
1105
__len__ = num_versions
1107
def annotate_iter(self, version_id):
1108
"""See VersionedFile.annotate_iter."""
1109
version_id = osutils.safe_revision_id(version_id)
1110
content = self._get_content(version_id)
1111
for origin, text in content.annotate_iter():
1114
def get_parents(self, version_id):
1115
"""See VersionedFile.get_parents."""
1118
# 52554 calls in 1264 872 internal down from 3674
1119
version_id = osutils.safe_revision_id(version_id)
1121
return self._index.get_parents(version_id)
1123
raise RevisionNotPresent(version_id, self.filename)
1125
def get_parents_with_ghosts(self, version_id):
1126
"""See VersionedFile.get_parents."""
1127
version_id = osutils.safe_revision_id(version_id)
1129
return self._index.get_parents_with_ghosts(version_id)
1131
raise RevisionNotPresent(version_id, self.filename)
1133
def get_ancestry(self, versions, topo_sorted=True):
1134
"""See VersionedFile.get_ancestry."""
1135
if isinstance(versions, basestring):
1136
versions = [versions]
1139
versions = [osutils.safe_revision_id(v) for v in versions]
1140
return self._index.get_ancestry(versions, topo_sorted)
1142
def get_ancestry_with_ghosts(self, versions):
1143
"""See VersionedFile.get_ancestry_with_ghosts."""
1144
if isinstance(versions, basestring):
1145
versions = [versions]
1148
versions = [osutils.safe_revision_id(v) for v in versions]
1149
return self._index.get_ancestry_with_ghosts(versions)
1151
def plan_merge(self, ver_a, ver_b):
1152
"""See VersionedFile.plan_merge."""
1153
ver_a = osutils.safe_revision_id(ver_a)
1154
ver_b = osutils.safe_revision_id(ver_b)
1155
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1157
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1158
annotated_a = self.annotate(ver_a)
1159
annotated_b = self.annotate(ver_b)
1160
return merge._plan_annotate_merge(annotated_a, annotated_b,
1161
ancestors_a, ancestors_b)
1164
class _KnitComponentFile(object):
1165
"""One of the files used to implement a knit database"""
1167
def __init__(self, transport, filename, mode, file_mode=None,
1168
create_parent_dir=False, dir_mode=None):
1169
self._transport = transport
1170
self._filename = filename
1172
self._file_mode = file_mode
1173
self._dir_mode = dir_mode
1174
self._create_parent_dir = create_parent_dir
1175
self._need_to_create = False
1177
def _full_path(self):
1178
"""Return the full path to this file."""
1179
return self._transport.base + self._filename
1181
def check_header(self, fp):
1182
line = fp.readline()
1184
# An empty file can actually be treated as though the file doesn't
1186
raise errors.NoSuchFile(self._full_path())
1187
if line != self.HEADER:
1188
raise KnitHeaderError(badline=line,
1189
filename=self._transport.abspath(self._filename))
1192
return '%s(%s)' % (self.__class__.__name__, self._filename)
1195
class _KnitIndex(_KnitComponentFile):
1196
"""Manages knit index file.
1198
The index is already kept in memory and read on startup, to enable
1199
fast lookups of revision information. The cursor of the index
1200
file is always pointing to the end, making it easy to append
1203
_cache is a cache for fast mapping from version id to a Index
1206
_history is a cache for fast mapping from indexes to version ids.
1208
The index data format is dictionary compressed when it comes to
1209
parent references; a index entry may only have parents that with a
1210
lover index number. As a result, the index is topological sorted.
1212
Duplicate entries may be written to the index for a single version id
1213
if this is done then the latter one completely replaces the former:
1214
this allows updates to correct version and parent information.
1215
Note that the two entries may share the delta, and that successive
1216
annotations and references MUST point to the first entry.
1218
The index file on disc contains a header, followed by one line per knit
1219
record. The same revision can be present in an index file more than once.
1220
The first occurrence gets assigned a sequence number starting from 0.
1222
The format of a single line is
1223
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1224
REVISION_ID is a utf8-encoded revision id
1225
FLAGS is a comma separated list of flags about the record. Values include
1226
no-eol, line-delta, fulltext.
1227
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1228
that the the compressed data starts at.
1229
LENGTH is the ascii representation of the length of the data file.
1230
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1232
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1233
revision id already in the knit that is a parent of REVISION_ID.
1234
The ' :' marker is the end of record marker.
1237
when a write is interrupted to the index file, it will result in a line
1238
that does not end in ' :'. If the ' :' is not present at the end of a line,
1239
or at the end of the file, then the record that is missing it will be
1240
ignored by the parser.
1242
When writing new records to the index file, the data is preceded by '\n'
1243
to ensure that records always start on new lines even if the last write was
1244
interrupted. As a result its normal for the last line in the index to be
1245
missing a trailing newline. One can be added with no harmful effects.
1248
HEADER = "# bzr knit index 8\n"
1250
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1251
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1253
def _cache_version(self, version_id, options, pos, size, parents):
1254
"""Cache a version record in the history array and index cache.
1256
This is inlined into _load_data for performance. KEEP IN SYNC.
1257
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1260
# only want the _history index to reference the 1st index entry
1262
if version_id not in self._cache:
1263
index = len(self._history)
1264
self._history.append(version_id)
1266
index = self._cache[version_id][5]
1267
self._cache[version_id] = (version_id,
1274
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1275
create_parent_dir=False, delay_create=False, dir_mode=None):
1276
_KnitComponentFile.__init__(self, transport, filename, mode,
1277
file_mode=file_mode,
1278
create_parent_dir=create_parent_dir,
1281
# position in _history is the 'official' index for a revision
1282
# but the values may have come from a newer entry.
1283
# so - wc -l of a knit index is != the number of unique names
1287
fp = self._transport.get(self._filename)
1289
# _load_data may raise NoSuchFile if the target knit is
1291
_load_data(self, fp)
1295
if mode != 'w' or not create:
1298
self._need_to_create = True
1300
self._transport.put_bytes_non_atomic(
1301
self._filename, self.HEADER, mode=self._file_mode)
1303
def get_graph(self):
1304
"""Return a list of the node:parents lists from this knit index."""
1305
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1307
def get_ancestry(self, versions, topo_sorted=True):
1308
"""See VersionedFile.get_ancestry."""
1309
# get a graph of all the mentioned versions:
1311
pending = set(versions)
1314
version = pending.pop()
1317
parents = [p for p in cache[version][4] if p in cache]
1319
raise RevisionNotPresent(version, self._filename)
1320
# if not completed and not a ghost
1321
pending.update([p for p in parents if p not in graph])
1322
graph[version] = parents
1325
return topo_sort(graph.items())
1327
def get_ancestry_with_ghosts(self, versions):
1328
"""See VersionedFile.get_ancestry_with_ghosts."""
1329
# get a graph of all the mentioned versions:
1330
self.check_versions_present(versions)
1333
pending = set(versions)
1335
version = pending.pop()
1337
parents = cache[version][4]
1343
pending.update([p for p in parents if p not in graph])
1344
graph[version] = parents
1345
return topo_sort(graph.items())
1347
def iter_parents(self, version_ids):
1348
"""Iterate through the parents for many version ids.
1350
:param version_ids: An iterable yielding version_ids.
1351
:return: An iterator that yields (version_id, parents). Requested
1352
version_ids not present in the versioned file are simply skipped.
1353
The order is undefined, allowing for different optimisations in
1354
the underlying implementation.
1356
for version_id in version_ids:
1358
yield version_id, tuple(self.get_parents(version_id))
1362
def num_versions(self):
1363
return len(self._history)
1365
__len__ = num_versions
1367
def get_versions(self):
1368
"""Get all the versions in the file. not topologically sorted."""
1369
return self._history
1371
def _version_list_to_index(self, versions):
1374
for version in versions:
1375
if version in cache:
1376
# -- inlined lookup() --
1377
result_list.append(str(cache[version][5]))
1378
# -- end lookup () --
1380
result_list.append('.' + version)
1381
return ' '.join(result_list)
1383
def add_version(self, version_id, options, index_memo, parents):
1384
"""Add a version record to the index."""
1385
self.add_versions(((version_id, options, index_memo, parents),))
1387
def add_versions(self, versions):
1388
"""Add multiple versions to the index.
1390
:param versions: a list of tuples:
1391
(version_id, options, pos, size, parents).
1394
orig_history = self._history[:]
1395
orig_cache = self._cache.copy()
1398
for version_id, options, (index, pos, size), parents in versions:
1399
line = "\n%s %s %s %s %s :" % (version_id,
1403
self._version_list_to_index(parents))
1404
assert isinstance(line, str), \
1405
'content must be utf-8 encoded: %r' % (line,)
1407
self._cache_version(version_id, options, pos, size, parents)
1408
if not self._need_to_create:
1409
self._transport.append_bytes(self._filename, ''.join(lines))
1412
sio.write(self.HEADER)
1413
sio.writelines(lines)
1415
self._transport.put_file_non_atomic(self._filename, sio,
1416
create_parent_dir=self._create_parent_dir,
1417
mode=self._file_mode,
1418
dir_mode=self._dir_mode)
1419
self._need_to_create = False
1421
# If any problems happen, restore the original values and re-raise
1422
self._history = orig_history
1423
self._cache = orig_cache
1426
def has_version(self, version_id):
1427
"""True if the version is in the index."""
1428
return version_id in self._cache
1430
def get_position(self, version_id):
1431
"""Return details needed to access the version.
1433
.kndx indices do not support split-out data, so return None for the
1436
:return: a tuple (None, data position, size) to hand to the access
1437
logic to get the record.
1439
entry = self._cache[version_id]
1440
return None, entry[2], entry[3]
1442
def get_method(self, version_id):
1443
"""Return compression method of specified version."""
1444
options = self._cache[version_id][1]
1445
if 'fulltext' in options:
1448
if 'line-delta' not in options:
1449
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1452
def get_options(self, version_id):
1453
"""Return a string represention options.
1457
return self._cache[version_id][1]
1459
def get_parents(self, version_id):
1460
"""Return parents of specified version ignoring ghosts."""
1461
return [parent for parent in self._cache[version_id][4]
1462
if parent in self._cache]
1464
def get_parents_with_ghosts(self, version_id):
1465
"""Return parents of specified version with ghosts."""
1466
return self._cache[version_id][4]
1468
def check_versions_present(self, version_ids):
1469
"""Check that all specified versions are present."""
1471
for version_id in version_ids:
1472
if version_id not in cache:
1473
raise RevisionNotPresent(version_id, self._filename)
1476
class KnitGraphIndex(object):
1477
"""A knit index that builds on GraphIndex."""
1479
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1480
"""Construct a KnitGraphIndex on a graph_index.
1482
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1483
:param deltas: Allow delta-compressed records.
1484
:param add_callback: If not None, allow additions to the index and call
1485
this callback with a list of added GraphIndex nodes:
1486
[(node, value, node_refs), ...]
1487
:param parents: If True, record knits parents, if not do not record
1490
self._graph_index = graph_index
1491
self._deltas = deltas
1492
self._add_callback = add_callback
1493
self._parents = parents
1494
if deltas and not parents:
1495
raise KnitCorrupt(self, "Cannot do delta compression without "
1498
def _get_entries(self, keys, check_present=False):
1499
"""Get the entries for keys.
1501
:param keys: An iterable of index keys, - 1-tuples.
1506
for node in self._graph_index.iter_entries(keys):
1508
found_keys.add(node[1])
1510
# adapt parentless index to the rest of the code.
1511
for node in self._graph_index.iter_entries(keys):
1512
yield node[0], node[1], node[2], ()
1513
found_keys.add(node[1])
1515
missing_keys = keys.difference(found_keys)
1517
raise RevisionNotPresent(missing_keys.pop(), self)
1519
def _present_keys(self, version_ids):
1521
node[1] for node in self._get_entries(version_ids)])
1523
def _parentless_ancestry(self, versions):
1524
"""Honour the get_ancestry API for parentless knit indices."""
1525
wanted_keys = self._version_ids_to_keys(versions)
1526
present_keys = self._present_keys(wanted_keys)
1527
missing = set(wanted_keys).difference(present_keys)
1529
raise RevisionNotPresent(missing.pop(), self)
1530
return list(self._keys_to_version_ids(present_keys))
1532
def get_ancestry(self, versions, topo_sorted=True):
1533
"""See VersionedFile.get_ancestry."""
1534
if not self._parents:
1535
return self._parentless_ancestry(versions)
1536
# XXX: This will do len(history) index calls - perhaps
1537
# it should be altered to be a index core feature?
1538
# get a graph of all the mentioned versions:
1541
versions = self._version_ids_to_keys(versions)
1542
pending = set(versions)
1544
# get all pending nodes
1545
this_iteration = pending
1546
new_nodes = self._get_entries(this_iteration)
1549
for (index, key, value, node_refs) in new_nodes:
1550
# dont ask for ghosties - otherwise
1551
# we we can end up looping with pending
1552
# being entirely ghosted.
1553
graph[key] = [parent for parent in node_refs[0]
1554
if parent not in ghosts]
1556
for parent in graph[key]:
1557
# dont examine known nodes again
1562
ghosts.update(this_iteration.difference(found))
1563
if versions.difference(graph):
1564
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1566
result_keys = topo_sort(graph.items())
1568
result_keys = graph.iterkeys()
1569
return [key[0] for key in result_keys]
1571
def get_ancestry_with_ghosts(self, versions):
1572
"""See VersionedFile.get_ancestry."""
1573
if not self._parents:
1574
return self._parentless_ancestry(versions)
1575
# XXX: This will do len(history) index calls - perhaps
1576
# it should be altered to be a index core feature?
1577
# get a graph of all the mentioned versions:
1579
versions = self._version_ids_to_keys(versions)
1580
pending = set(versions)
1582
# get all pending nodes
1583
this_iteration = pending
1584
new_nodes = self._get_entries(this_iteration)
1586
for (index, key, value, node_refs) in new_nodes:
1587
graph[key] = node_refs[0]
1589
for parent in graph[key]:
1590
# dont examine known nodes again
1594
missing_versions = this_iteration.difference(graph)
1595
missing_needed = versions.intersection(missing_versions)
1597
raise RevisionNotPresent(missing_needed.pop(), self)
1598
for missing_version in missing_versions:
1599
# add a key, no parents
1600
graph[missing_version] = []
1601
pending.discard(missing_version) # don't look for it
1602
result_keys = topo_sort(graph.items())
1603
return [key[0] for key in result_keys]
1605
def get_graph(self):
1606
"""Return a list of the node:parents lists from this knit index."""
1607
if not self._parents:
1608
return [(key, ()) for key in self.get_versions()]
1610
for index, key, value, refs in self._graph_index.iter_all_entries():
1611
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1614
def iter_parents(self, version_ids):
1615
"""Iterate through the parents for many version ids.
1617
:param version_ids: An iterable yielding version_ids.
1618
:return: An iterator that yields (version_id, parents). Requested
1619
version_ids not present in the versioned file are simply skipped.
1620
The order is undefined, allowing for different optimisations in
1621
the underlying implementation.
1624
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1626
present_parents = set()
1627
for node in all_nodes:
1628
all_parents.update(node[3][0])
1629
# any node we are querying must be present
1630
present_parents.add(node[1])
1631
unknown_parents = all_parents.difference(present_parents)
1632
present_parents.update(self._present_keys(unknown_parents))
1633
for node in all_nodes:
1635
for parent in node[3][0]:
1636
if parent in present_parents:
1637
parents.append(parent[0])
1638
yield node[1][0], tuple(parents)
1640
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1641
yield node[1][0], ()
1643
def num_versions(self):
1644
return len(list(self._graph_index.iter_all_entries()))
1646
__len__ = num_versions
1648
def get_versions(self):
1649
"""Get all the versions in the file. not topologically sorted."""
1650
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1652
def has_version(self, version_id):
1653
"""True if the version is in the index."""
1654
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1656
def _keys_to_version_ids(self, keys):
1657
return tuple(key[0] for key in keys)
1659
def get_position(self, version_id):
1660
"""Return details needed to access the version.
1662
:return: a tuple (index, data position, size) to hand to the access
1663
logic to get the record.
1665
node = self._get_node(version_id)
1666
bits = node[2][1:].split(' ')
1667
return node[0], int(bits[0]), int(bits[1])
1669
def get_method(self, version_id):
1670
"""Return compression method of specified version."""
1671
if not self._deltas:
1673
return self._parent_compression(self._get_node(version_id)[3][1])
1675
def _parent_compression(self, reference_list):
1676
# use the second reference list to decide if this is delta'd or not.
1677
if len(reference_list):
1682
def _get_node(self, version_id):
1683
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1685
def get_options(self, version_id):
1686
"""Return a string represention options.
1690
node = self._get_node(version_id)
1691
if not self._deltas:
1692
options = ['fulltext']
1694
options = [self._parent_compression(node[3][1])]
1695
if node[2][0] == 'N':
1696
options.append('no-eol')
1699
def get_parents(self, version_id):
1700
"""Return parents of specified version ignoring ghosts."""
1701
parents = list(self.iter_parents([version_id]))
1704
raise errors.RevisionNotPresent(version_id, self)
1705
return parents[0][1]
1707
def get_parents_with_ghosts(self, version_id):
1708
"""Return parents of specified version with ghosts."""
1709
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1710
check_present=True))
1711
if not self._parents:
1713
return self._keys_to_version_ids(nodes[0][3][0])
1715
def check_versions_present(self, version_ids):
1716
"""Check that all specified versions are present."""
1717
keys = self._version_ids_to_keys(version_ids)
1718
present = self._present_keys(keys)
1719
missing = keys.difference(present)
1721
raise RevisionNotPresent(missing.pop(), self)
1723
def add_version(self, version_id, options, access_memo, parents):
1724
"""Add a version record to the index."""
1725
return self.add_versions(((version_id, options, access_memo, parents),))
1727
def add_versions(self, versions):
1728
"""Add multiple versions to the index.
1730
This function does not insert data into the Immutable GraphIndex
1731
backing the KnitGraphIndex, instead it prepares data for insertion by
1732
the caller and checks that it is safe to insert then calls
1733
self._add_callback with the prepared GraphIndex nodes.
1735
:param versions: a list of tuples:
1736
(version_id, options, pos, size, parents).
1738
if not self._add_callback:
1739
raise errors.ReadOnlyError(self)
1740
# we hope there are no repositories with inconsistent parentage
1745
for (version_id, options, access_memo, parents) in versions:
1746
index, pos, size = access_memo
1747
key = (version_id, )
1748
parents = tuple((parent, ) for parent in parents)
1749
if 'no-eol' in options:
1753
value += "%d %d" % (pos, size)
1754
if not self._deltas:
1755
if 'line-delta' in options:
1756
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1759
if 'line-delta' in options:
1760
node_refs = (parents, (parents[0],))
1762
node_refs = (parents, ())
1764
node_refs = (parents, )
1767
raise KnitCorrupt(self, "attempt to add node with parents "
1768
"in parentless index.")
1770
keys[key] = (value, node_refs)
1771
present_nodes = self._get_entries(keys)
1772
for (index, key, value, node_refs) in present_nodes:
1773
if (value, node_refs) != keys[key]:
1774
raise KnitCorrupt(self, "inconsistent details in add_versions"
1775
": %s %s" % ((value, node_refs), keys[key]))
1779
for key, (value, node_refs) in keys.iteritems():
1780
result.append((key, value, node_refs))
1782
for key, (value, node_refs) in keys.iteritems():
1783
result.append((key, value))
1784
self._add_callback(result)
1786
def _version_ids_to_keys(self, version_ids):
1787
return set((version_id, ) for version_id in version_ids)
1790
class _KnitAccess(object):
1791
"""Access to knit records in a .knit file."""
1793
def __init__(self, transport, filename, _file_mode, _dir_mode,
1794
_need_to_create, _create_parent_dir):
1795
"""Create a _KnitAccess for accessing and inserting data.
1797
:param transport: The transport the .knit is located on.
1798
:param filename: The filename of the .knit.
1800
self._transport = transport
1801
self._filename = filename
1802
self._file_mode = _file_mode
1803
self._dir_mode = _dir_mode
1804
self._need_to_create = _need_to_create
1805
self._create_parent_dir = _create_parent_dir
1807
def add_raw_records(self, sizes, raw_data):
1808
"""Add raw knit bytes to a storage area.
1810
The data is spooled to whereever the access method is storing data.
1812
:param sizes: An iterable containing the size of each raw data segment.
1813
:param raw_data: A bytestring containing the data.
1814
:return: A list of memos to retrieve the record later. Each memo is a
1815
tuple - (index, pos, length), where the index field is always None
1816
for the .knit access method.
1818
assert type(raw_data) == str, \
1819
'data must be plain bytes was %s' % type(raw_data)
1820
if not self._need_to_create:
1821
base = self._transport.append_bytes(self._filename, raw_data)
1823
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1824
create_parent_dir=self._create_parent_dir,
1825
mode=self._file_mode,
1826
dir_mode=self._dir_mode)
1827
self._need_to_create = False
1831
result.append((None, base, size))
1836
"""IFF this data access has its own storage area, initialise it.
1840
self._transport.put_bytes_non_atomic(self._filename, '',
1841
mode=self._file_mode)
1843
def open_file(self):
1844
"""IFF this data access can be represented as a single file, open it.
1846
For knits that are not mapped to a single file on disk this will
1849
:return: None or a file handle.
1852
return self._transport.get(self._filename)
1857
def get_raw_records(self, memos_for_retrieval):
1858
"""Get the raw bytes for a records.
1860
:param memos_for_retrieval: An iterable containing the (index, pos,
1861
length) memo for retrieving the bytes. The .knit method ignores
1862
the index as there is always only a single file.
1863
:return: An iterator over the bytes of the records.
1865
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
1866
for pos, data in self._transport.readv(self._filename, read_vector):
1870
class _PackAccess(object):
1871
"""Access to knit records via a collection of packs."""
1873
def __init__(self, index_to_packs, writer=None):
1874
"""Create a _PackAccess object.
1876
:param index_to_packs: A dict mapping index objects to the transport
1877
and file names for obtaining data.
1878
:param writer: A tuple (pack.ContainerWriter, write_index) which
1879
contains the pack to write, and the index that reads from it will
1883
self.container_writer = writer[0]
1884
self.write_index = writer[1]
1886
self.container_writer = None
1887
self.write_index = None
1888
self.indices = index_to_packs
1890
def add_raw_records(self, sizes, raw_data):
1891
"""Add raw knit bytes to a storage area.
1893
The data is spooled to the container writer in one bytes-record per
1896
:param sizes: An iterable containing the size of each raw data segment.
1897
:param raw_data: A bytestring containing the data.
1898
:return: A list of memos to retrieve the record later. Each memo is a
1899
tuple - (index, pos, length), where the index field is the
1900
write_index object supplied to the PackAccess object.
1902
assert type(raw_data) == str, \
1903
'data must be plain bytes was %s' % type(raw_data)
1907
p_offset, p_length = self.container_writer.add_bytes_record(
1908
raw_data[offset:offset+size], [])
1910
result.append((self.write_index, p_offset, p_length))
1914
"""Pack based knits do not get individually created."""
1916
def get_raw_records(self, memos_for_retrieval):
1917
"""Get the raw bytes for a records.
1919
:param memos_for_retrieval: An iterable containing the (index, pos,
1920
length) memo for retrieving the bytes. The Pack access method
1921
looks up the pack to use for a given record in its index_to_pack
1923
:return: An iterator over the bytes of the records.
1925
# first pass, group into same-index requests
1927
current_index = None
1928
for (index, offset, length) in memos_for_retrieval:
1929
if current_index == index:
1930
current_list.append((offset, length))
1932
if current_index is not None:
1933
request_lists.append((current_index, current_list))
1934
current_index = index
1935
current_list = [(offset, length)]
1936
# handle the last entry
1937
if current_index is not None:
1938
request_lists.append((current_index, current_list))
1939
for index, offsets in request_lists:
1940
transport, path = self.indices[index]
1941
reader = pack.make_readv_reader(transport, path, offsets)
1942
for names, read_func in reader.iter_records():
1943
yield read_func(None)
1945
def open_file(self):
1946
"""Pack based knits have no single file."""
1949
def set_writer(self, writer, index, (transport, packname)):
1950
"""Set a writer to use for adding data."""
1951
self.indices[index] = (transport, packname)
1952
self.container_writer = writer
1953
self.write_index = index
1956
class _KnitData(object):
1957
"""Manage extraction of data from a KnitAccess, caching and decompressing.
1959
The KnitData class provides the logic for parsing and using knit records,
1960
making use of an access method for the low level read and write operations.
1963
def __init__(self, access):
1964
"""Create a KnitData object.
1966
:param access: The access method to use. Access methods such as
1967
_KnitAccess manage the insertion of raw records and the subsequent
1968
retrieval of the same.
1970
self._access = access
1971
self._checked = False
1972
# TODO: jam 20060713 conceptually, this could spill to disk
1973
# if the cached size gets larger than a certain amount
1974
# but it complicates the model a bit, so for now just use
1975
# a simple dictionary
1977
self._do_cache = False
1979
def enable_cache(self):
1980
"""Enable caching of reads."""
1981
self._do_cache = True
1983
def clear_cache(self):
1984
"""Clear the record cache."""
1985
self._do_cache = False
1988
def _open_file(self):
1989
return self._access.open_file()
1991
def _record_to_data(self, version_id, digest, lines):
1992
"""Convert version_id, digest, lines into a raw data block.
1994
:return: (len, a StringIO instance with the raw data ready to read.)
1997
data_file = GzipFile(None, mode='wb', fileobj=sio,
1998
compresslevel=Z_DEFAULT_COMPRESSION)
2000
assert isinstance(version_id, str)
2001
data_file.writelines(chain(
2002
["version %s %d %s\n" % (version_id,
2006
["end %s\n" % version_id]))
2013
def add_raw_records(self, sizes, raw_data):
2014
"""Append a prepared record to the data file.
2016
:param sizes: An iterable containing the size of each raw data segment.
2017
:param raw_data: A bytestring containing the data.
2018
:return: a list of index data for the way the data was stored.
2019
See the access method add_raw_records documentation for more
2022
return self._access.add_raw_records(sizes, raw_data)
2024
def add_record(self, version_id, digest, lines):
2025
"""Write new text record to disk.
2027
Returns index data for retrieving it later, as per add_raw_records.
2029
size, sio = self._record_to_data(version_id, digest, lines)
2030
result = self.add_raw_records([size], sio.getvalue())
2032
self._cache[version_id] = sio.getvalue()
2035
def _parse_record_header(self, version_id, raw_data):
2036
"""Parse a record header for consistency.
2038
:return: the header and the decompressor stream.
2039
as (stream, header_record)
2041
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2043
rec = self._check_header(version_id, df.readline())
2044
except Exception, e:
2045
raise KnitCorrupt(self._access,
2046
"While reading {%s} got %s(%s)"
2047
% (version_id, e.__class__.__name__, str(e)))
2050
def _check_header(self, version_id, line):
2053
raise KnitCorrupt(self._access,
2054
'unexpected number of elements in record header')
2055
if rec[1] != version_id:
2056
raise KnitCorrupt(self._access,
2057
'unexpected version, wanted %r, got %r'
2058
% (version_id, rec[1]))
2061
def _parse_record(self, version_id, data):
2063
# 4168 calls in 2880 217 internal
2064
# 4168 calls to _parse_record_header in 2121
2065
# 4168 calls to readlines in 330
2066
df = GzipFile(mode='rb', fileobj=StringIO(data))
2069
record_contents = df.readlines()
2070
except Exception, e:
2071
raise KnitCorrupt(self._access,
2072
"While reading {%s} got %s(%s)"
2073
% (version_id, e.__class__.__name__, str(e)))
2074
header = record_contents.pop(0)
2075
rec = self._check_header(version_id, header)
2077
last_line = record_contents.pop()
2078
if len(record_contents) != int(rec[2]):
2079
raise KnitCorrupt(self._access,
2080
'incorrect number of lines %s != %s'
2082
% (len(record_contents), int(rec[2]),
2084
if last_line != 'end %s\n' % rec[1]:
2085
raise KnitCorrupt(self._access,
2086
'unexpected version end line %r, wanted %r'
2087
% (last_line, version_id))
2089
return record_contents, rec[3]
2091
def read_records_iter_raw(self, records):
2092
"""Read text records from data file and yield raw data.
2094
This unpacks enough of the text record to validate the id is
2095
as expected but thats all.
2097
# setup an iterator of the external records:
2098
# uses readv so nice and fast we hope.
2100
# grab the disk data needed.
2102
# Don't check _cache if it is empty
2103
needed_offsets = [index_memo for version_id, index_memo
2105
if version_id not in self._cache]
2107
needed_offsets = [index_memo for version_id, index_memo
2110
raw_records = self._access.get_raw_records(needed_offsets)
2112
for version_id, index_memo in records:
2113
if version_id in self._cache:
2114
# This data has already been validated
2115
data = self._cache[version_id]
2117
data = raw_records.next()
2119
self._cache[version_id] = data
2121
# validate the header
2122
df, rec = self._parse_record_header(version_id, data)
2124
yield version_id, data
2126
def read_records_iter(self, records):
2127
"""Read text records from data file and yield result.
2129
The result will be returned in whatever is the fastest to read.
2130
Not by the order requested. Also, multiple requests for the same
2131
record will only yield 1 response.
2132
:param records: A list of (version_id, pos, len) entries
2133
:return: Yields (version_id, contents, digest) in the order
2134
read, not the order requested
2140
# Skip records we have alread seen
2141
yielded_records = set()
2142
needed_records = set()
2143
for record in records:
2144
if record[0] in self._cache:
2145
if record[0] in yielded_records:
2147
yielded_records.add(record[0])
2148
data = self._cache[record[0]]
2149
content, digest = self._parse_record(record[0], data)
2150
yield (record[0], content, digest)
2152
needed_records.add(record)
2153
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2155
needed_records = sorted(set(records), key=operator.itemgetter(1))
2157
if not needed_records:
2160
# The transport optimizes the fetching as well
2161
# (ie, reads continuous ranges.)
2162
raw_data = self._access.get_raw_records(
2163
[index_memo for version_id, index_memo in needed_records])
2165
for (version_id, index_memo), data in \
2166
izip(iter(needed_records), raw_data):
2167
content, digest = self._parse_record(version_id, data)
2169
self._cache[version_id] = data
2170
yield version_id, content, digest
2172
def read_records(self, records):
2173
"""Read records into a dictionary."""
2175
for record_id, content, digest in \
2176
self.read_records_iter(records):
2177
components[record_id] = (content, digest)
2181
class InterKnit(InterVersionedFile):
2182
"""Optimised code paths for knit to knit operations."""
2184
_matching_file_from_factory = KnitVersionedFile
2185
_matching_file_to_factory = KnitVersionedFile
2188
def is_compatible(source, target):
2189
"""Be compatible with knits. """
2191
return (isinstance(source, KnitVersionedFile) and
2192
isinstance(target, KnitVersionedFile))
2193
except AttributeError:
2196
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2197
"""See InterVersionedFile.join."""
2198
assert isinstance(self.source, KnitVersionedFile)
2199
assert isinstance(self.target, KnitVersionedFile)
2201
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2206
pb = ui.ui_factory.nested_progress_bar()
2208
version_ids = list(version_ids)
2209
if None in version_ids:
2210
version_ids.remove(None)
2212
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2213
this_versions = set(self.target._index.get_versions())
2214
needed_versions = self.source_ancestry - this_versions
2215
cross_check_versions = self.source_ancestry.intersection(this_versions)
2216
mismatched_versions = set()
2217
for version in cross_check_versions:
2218
# scan to include needed parents.
2219
n1 = set(self.target.get_parents_with_ghosts(version))
2220
n2 = set(self.source.get_parents_with_ghosts(version))
2222
# FIXME TEST this check for cycles being introduced works
2223
# the logic is we have a cycle if in our graph we are an
2224
# ancestor of any of the n2 revisions.
2230
parent_ancestors = self.source.get_ancestry(parent)
2231
if version in parent_ancestors:
2232
raise errors.GraphCycleError([parent, version])
2233
# ensure this parent will be available later.
2234
new_parents = n2.difference(n1)
2235
needed_versions.update(new_parents.difference(this_versions))
2236
mismatched_versions.add(version)
2238
if not needed_versions and not mismatched_versions:
2240
full_list = topo_sort(self.source.get_graph())
2242
version_list = [i for i in full_list if (not self.target.has_version(i)
2243
and i in needed_versions)]
2247
copy_queue_records = []
2249
for version_id in version_list:
2250
options = self.source._index.get_options(version_id)
2251
parents = self.source._index.get_parents_with_ghosts(version_id)
2252
# check that its will be a consistent copy:
2253
for parent in parents:
2254
# if source has the parent, we must :
2255
# * already have it or
2256
# * have it scheduled already
2257
# otherwise we don't care
2258
assert (self.target.has_version(parent) or
2259
parent in copy_set or
2260
not self.source.has_version(parent))
2261
index_memo = self.source._index.get_position(version_id)
2262
copy_queue_records.append((version_id, index_memo))
2263
copy_queue.append((version_id, options, parents))
2264
copy_set.add(version_id)
2266
# data suck the join:
2268
total = len(version_list)
2271
for (version_id, raw_data), \
2272
(version_id2, options, parents) in \
2273
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2275
assert version_id == version_id2, 'logic error, inconsistent results'
2277
pb.update("Joining knit", count, total)
2278
raw_records.append((version_id, options, parents, len(raw_data)))
2279
raw_datum.append(raw_data)
2280
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2282
for version in mismatched_versions:
2283
# FIXME RBC 20060309 is this needed?
2284
n1 = set(self.target.get_parents_with_ghosts(version))
2285
n2 = set(self.source.get_parents_with_ghosts(version))
2286
# write a combined record to our history preserving the current
2287
# parents as first in the list
2288
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2289
self.target.fix_parents(version, new_parents)
2295
InterVersionedFile.register_optimiser(InterKnit)
2298
class WeaveToKnit(InterVersionedFile):
2299
"""Optimised code paths for weave to knit operations."""
2301
_matching_file_from_factory = bzrlib.weave.WeaveFile
2302
_matching_file_to_factory = KnitVersionedFile
2305
def is_compatible(source, target):
2306
"""Be compatible with weaves to knits."""
2308
return (isinstance(source, bzrlib.weave.Weave) and
2309
isinstance(target, KnitVersionedFile))
2310
except AttributeError:
2313
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2314
"""See InterVersionedFile.join."""
2315
assert isinstance(self.source, bzrlib.weave.Weave)
2316
assert isinstance(self.target, KnitVersionedFile)
2318
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2323
pb = ui.ui_factory.nested_progress_bar()
2325
version_ids = list(version_ids)
2327
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2328
this_versions = set(self.target._index.get_versions())
2329
needed_versions = self.source_ancestry - this_versions
2330
cross_check_versions = self.source_ancestry.intersection(this_versions)
2331
mismatched_versions = set()
2332
for version in cross_check_versions:
2333
# scan to include needed parents.
2334
n1 = set(self.target.get_parents_with_ghosts(version))
2335
n2 = set(self.source.get_parents(version))
2336
# if all of n2's parents are in n1, then its fine.
2337
if n2.difference(n1):
2338
# FIXME TEST this check for cycles being introduced works
2339
# the logic is we have a cycle if in our graph we are an
2340
# ancestor of any of the n2 revisions.
2346
parent_ancestors = self.source.get_ancestry(parent)
2347
if version in parent_ancestors:
2348
raise errors.GraphCycleError([parent, version])
2349
# ensure this parent will be available later.
2350
new_parents = n2.difference(n1)
2351
needed_versions.update(new_parents.difference(this_versions))
2352
mismatched_versions.add(version)
2354
if not needed_versions and not mismatched_versions:
2356
full_list = topo_sort(self.source.get_graph())
2358
version_list = [i for i in full_list if (not self.target.has_version(i)
2359
and i in needed_versions)]
2363
total = len(version_list)
2364
for version_id in version_list:
2365
pb.update("Converting to knit", count, total)
2366
parents = self.source.get_parents(version_id)
2367
# check that its will be a consistent copy:
2368
for parent in parents:
2369
# if source has the parent, we must already have it
2370
assert (self.target.has_version(parent))
2371
self.target.add_lines(
2372
version_id, parents, self.source.get_lines(version_id))
2375
for version in mismatched_versions:
2376
# FIXME RBC 20060309 is this needed?
2377
n1 = set(self.target.get_parents_with_ghosts(version))
2378
n2 = set(self.source.get_parents(version))
2379
# write a combined record to our history preserving the current
2380
# parents as first in the list
2381
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2382
self.target.fix_parents(version, new_parents)
2388
InterVersionedFile.register_optimiser(WeaveToKnit)
2391
class KnitSequenceMatcher(difflib.SequenceMatcher):
2392
"""Knit tuned sequence matcher.
2394
This is based on profiling of difflib which indicated some improvements
2395
for our usage pattern.
2398
def find_longest_match(self, alo, ahi, blo, bhi):
2399
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
2401
If isjunk is not defined:
2403
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
2404
alo <= i <= i+k <= ahi
2405
blo <= j <= j+k <= bhi
2406
and for all (i',j',k') meeting those conditions,
2409
and if i == i', j <= j'
2411
In other words, of all maximal matching blocks, return one that
2412
starts earliest in a, and of all those maximal matching blocks that
2413
start earliest in a, return the one that starts earliest in b.
2415
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
2416
>>> s.find_longest_match(0, 5, 0, 9)
2419
If isjunk is defined, first the longest matching block is
2420
determined as above, but with the additional restriction that no
2421
junk element appears in the block. Then that block is extended as
2422
far as possible by matching (only) junk elements on both sides. So
2423
the resulting block never matches on junk except as identical junk
2424
happens to be adjacent to an "interesting" match.
2426
Here's the same example as before, but considering blanks to be
2427
junk. That prevents " abcd" from matching the " abcd" at the tail
2428
end of the second sequence directly. Instead only the "abcd" can
2429
match, and matches the leftmost "abcd" in the second sequence:
2431
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
2432
>>> s.find_longest_match(0, 5, 0, 9)
2435
If no blocks match, return (alo, blo, 0).
2437
>>> s = SequenceMatcher(None, "ab", "c")
2438
>>> s.find_longest_match(0, 2, 0, 1)
2442
# CAUTION: stripping common prefix or suffix would be incorrect.
2446
# Longest matching block is "ab", but if common prefix is
2447
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
2448
# strip, so ends up claiming that ab is changed to acab by
2449
# inserting "ca" in the middle. That's minimal but unintuitive:
2450
# "it's obvious" that someone inserted "ac" at the front.
2451
# Windiff ends up at the same place as diff, but by pairing up
2452
# the unique 'b's and then matching the first two 'a's.
2454
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
2455
besti, bestj, bestsize = alo, blo, 0
2456
# find longest junk-free match
2457
# during an iteration of the loop, j2len[j] = length of longest
2458
# junk-free match ending with a[i-1] and b[j]
2462
for i in xrange(alo, ahi):
2463
# look at all instances of a[i] in b; note that because
2464
# b2j has no junk keys, the loop is skipped if a[i] is junk
2465
j2lenget = j2len.get
2468
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
2469
# following improvement
2470
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
2471
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
2472
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
2474
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
2475
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
2476
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
2488
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
2490
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
2493
# Extend the best by non-junk elements on each end. In particular,
2494
# "popular" non-junk elements aren't in b2j, which greatly speeds
2495
# the inner loop above, but also means "the best" match so far
2496
# doesn't contain any junk *or* popular non-junk elements.
2497
while besti > alo and bestj > blo and \
2498
not isbjunk(b[bestj-1]) and \
2499
a[besti-1] == b[bestj-1]:
2500
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2501
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2502
not isbjunk(b[bestj+bestsize]) and \
2503
a[besti+bestsize] == b[bestj+bestsize]:
2506
# Now that we have a wholly interesting match (albeit possibly
2507
# empty!), we may as well suck up the matching junk on each
2508
# side of it too. Can't think of a good reason not to, and it
2509
# saves post-processing the (possibly considerable) expense of
2510
# figuring out what to do with it. In the case of an empty
2511
# interesting match, this is clearly the right thing to do,
2512
# because no other kind of match is possible in the regions.
2513
while besti > alo and bestj > blo and \
2514
isbjunk(b[bestj-1]) and \
2515
a[besti-1] == b[bestj-1]:
2516
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2517
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2518
isbjunk(b[bestj+bestsize]) and \
2519
a[besti+bestsize] == b[bestj+bestsize]:
2520
bestsize = bestsize + 1
2522
return besti, bestj, bestsize
2526
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2528
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
