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(), """
93
from bzrlib.errors import (
99
KnitDataStreamIncompatible,
102
RevisionAlreadyPresent,
104
from bzrlib.tuned_gzip import GzipFile
105
from bzrlib.osutils import (
110
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
111
from bzrlib.tsort import topo_sort
114
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
117
# TODO: Split out code specific to this format into an associated object.
119
# TODO: Can we put in some kind of value to check that the index and data
120
# files belong together?
122
# TODO: accommodate binaries, perhaps by storing a byte count
124
# TODO: function to check whole file
126
# TODO: atomically append data, then measure backwards from the cursor
127
# position after writing to work out where it was located. we may need to
128
# bypass python file buffering.
130
DATA_SUFFIX = '.knit'
131
INDEX_SUFFIX = '.kndx'
134
class KnitContent(object):
135
"""Content of a knit version to which deltas can be applied."""
137
def __init__(self, lines):
140
def annotate_iter(self):
141
"""Yield tuples of (origin, text) for each content line."""
142
return iter(self._lines)
145
"""Return a list of (origin, text) tuples."""
146
return list(self.annotate_iter())
148
def line_delta_iter(self, new_lines):
149
"""Generate line-based delta from this content to new_lines."""
150
new_texts = new_lines.text()
151
old_texts = self.text()
152
s = KnitSequenceMatcher(None, old_texts, new_texts)
153
for tag, i1, i2, j1, j2 in s.get_opcodes():
156
# ofrom, oto, length, data
157
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
159
def line_delta(self, new_lines):
160
return list(self.line_delta_iter(new_lines))
163
return [text for origin, text in self._lines]
166
return KnitContent(self._lines[:])
169
def get_line_delta_blocks(knit_delta, source, target):
170
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
171
target_len = len(target)
174
for s_begin, s_end, t_len, new_text in knit_delta:
175
true_n = s_begin - s_pos
178
# knit deltas do not provide reliable info about whether the
179
# last line of a file matches, due to eol handling.
180
if source[s_pos + n -1] != target[t_pos + n -1]:
183
yield s_pos, t_pos, n
184
t_pos += t_len + true_n
186
n = target_len - t_pos
188
if source[s_pos + n -1] != target[t_pos + n -1]:
191
yield s_pos, t_pos, n
192
yield s_pos + (target_len - t_pos), target_len, 0
195
class _KnitFactory(object):
196
"""Base factory for creating content objects."""
198
def make(self, lines, version_id):
199
num_lines = len(lines)
200
return KnitContent(zip([version_id] * num_lines, lines))
203
class KnitAnnotateFactory(_KnitFactory):
204
"""Factory for creating annotated Content objects."""
208
def parse_fulltext(self, content, version_id):
209
"""Convert fulltext to internal representation
211
fulltext content is of the format
212
revid(utf8) plaintext\n
213
internal representation is of the format:
216
# TODO: jam 20070209 The tests expect this to be returned as tuples,
217
# but the code itself doesn't really depend on that.
218
# Figure out a way to not require the overhead of turning the
219
# list back into tuples.
220
lines = [tuple(line.split(' ', 1)) for line in content]
221
return KnitContent(lines)
223
def parse_line_delta_iter(self, lines):
224
return iter(self.parse_line_delta(lines))
226
def parse_line_delta(self, lines, version_id):
227
"""Convert a line based delta into internal representation.
229
line delta is in the form of:
230
intstart intend intcount
232
revid(utf8) newline\n
233
internal representation is
234
(start, end, count, [1..count tuples (revid, newline)])
241
def cache_and_return(line):
242
origin, text = line.split(' ', 1)
243
return cache.setdefault(origin, origin), text
245
# walk through the lines parsing.
247
start, end, count = [int(n) for n in header.split(',')]
248
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
249
result.append((start, end, count, contents))
252
def get_fulltext_content(self, lines):
253
"""Extract just the content lines from a fulltext."""
254
return (line.split(' ', 1)[1] for line in lines)
256
def get_linedelta_content(self, lines):
257
"""Extract just the content from a line delta.
259
This doesn't return all of the extra information stored in a delta.
260
Only the actual content lines.
265
header = header.split(',')
266
count = int(header[2])
267
for i in xrange(count):
268
origin, text = next().split(' ', 1)
271
def lower_fulltext(self, content):
272
"""convert a fulltext content record into a serializable form.
274
see parse_fulltext which this inverts.
276
# TODO: jam 20070209 We only do the caching thing to make sure that
277
# the origin is a valid utf-8 line, eventually we could remove it
278
return ['%s %s' % (o, t) for o, t in content._lines]
280
def lower_line_delta(self, delta):
281
"""convert a delta into a serializable form.
283
See parse_line_delta which this inverts.
285
# TODO: jam 20070209 We only do the caching thing to make sure that
286
# the origin is a valid utf-8 line, eventually we could remove it
288
for start, end, c, lines in delta:
289
out.append('%d,%d,%d\n' % (start, end, c))
290
out.extend(origin + ' ' + text
291
for origin, text in lines)
294
def annotate_iter(self, knit, version_id):
295
content = knit._get_content(version_id)
296
return content.annotate_iter()
299
class KnitPlainFactory(_KnitFactory):
300
"""Factory for creating plain Content objects."""
304
def parse_fulltext(self, content, version_id):
305
"""This parses an unannotated fulltext.
307
Note that this is not a noop - the internal representation
308
has (versionid, line) - its just a constant versionid.
310
return self.make(content, version_id)
312
def parse_line_delta_iter(self, lines, version_id):
314
num_lines = len(lines)
315
while cur < num_lines:
318
start, end, c = [int(n) for n in header.split(',')]
319
yield start, end, c, zip([version_id] * c, lines[cur:cur+c])
322
def parse_line_delta(self, lines, version_id):
323
return list(self.parse_line_delta_iter(lines, version_id))
325
def get_fulltext_content(self, lines):
326
"""Extract just the content lines from a fulltext."""
329
def get_linedelta_content(self, lines):
330
"""Extract just the content from a line delta.
332
This doesn't return all of the extra information stored in a delta.
333
Only the actual content lines.
338
header = header.split(',')
339
count = int(header[2])
340
for i in xrange(count):
343
def lower_fulltext(self, content):
344
return content.text()
346
def lower_line_delta(self, delta):
348
for start, end, c, lines in delta:
349
out.append('%d,%d,%d\n' % (start, end, c))
350
out.extend([text for origin, text in lines])
353
def annotate_iter(self, knit, version_id):
354
return annotate_knit(knit, version_id)
357
def make_empty_knit(transport, relpath):
358
"""Construct a empty knit at the specified location."""
359
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
362
class KnitVersionedFile(VersionedFile):
363
"""Weave-like structure with faster random access.
365
A knit stores a number of texts and a summary of the relationships
366
between them. Texts are identified by a string version-id. Texts
367
are normally stored and retrieved as a series of lines, but can
368
also be passed as single strings.
370
Lines are stored with the trailing newline (if any) included, to
371
avoid special cases for files with no final newline. Lines are
372
composed of 8-bit characters, not unicode. The combination of
373
these approaches should mean any 'binary' file can be safely
374
stored and retrieved.
377
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
378
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
379
create=False, create_parent_dir=False, delay_create=False,
380
dir_mode=None, index=None, access_method=None):
381
"""Construct a knit at location specified by relpath.
383
:param create: If not True, only open an existing knit.
384
:param create_parent_dir: If True, create the parent directory if
385
creating the file fails. (This is used for stores with
386
hash-prefixes that may not exist yet)
387
:param delay_create: The calling code is aware that the knit won't
388
actually be created until the first data is stored.
389
:param index: An index to use for the knit.
391
if deprecated_passed(basis_knit):
392
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
393
" deprecated as of bzr 0.9.",
394
DeprecationWarning, stacklevel=2)
395
if access_mode is None:
397
super(KnitVersionedFile, self).__init__(access_mode)
398
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
399
self.transport = transport
400
self.filename = relpath
401
self.factory = factory or KnitAnnotateFactory()
402
self.writable = (access_mode == 'w')
405
self._max_delta_chain = 200
408
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
409
access_mode, create=create, file_mode=file_mode,
410
create_parent_dir=create_parent_dir, delay_create=delay_create,
414
if access_method is None:
415
_access = _KnitAccess(transport, relpath + DATA_SUFFIX, file_mode, dir_mode,
416
((create and not len(self)) and delay_create), create_parent_dir)
418
_access = access_method
419
if create and not len(self) and not delay_create:
421
self._data = _KnitData(_access)
424
return '%s(%s)' % (self.__class__.__name__,
425
self.transport.abspath(self.filename))
427
def _check_should_delta(self, first_parents):
428
"""Iterate back through the parent listing, looking for a fulltext.
430
This is used when we want to decide whether to add a delta or a new
431
fulltext. It searches for _max_delta_chain parents. When it finds a
432
fulltext parent, it sees if the total size of the deltas leading up to
433
it is large enough to indicate that we want a new full text anyway.
435
Return True if we should create a new delta, False if we should use a
440
delta_parents = first_parents
441
for count in xrange(self._max_delta_chain):
442
parent = delta_parents[0]
443
method = self._index.get_method(parent)
444
index, pos, size = self._index.get_position(parent)
445
if method == 'fulltext':
449
delta_parents = self._index.get_parents(parent)
451
# We couldn't find a fulltext, so we must create a new one
454
return fulltext_size > delta_size
456
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
457
"""See VersionedFile._add_delta()."""
458
self._check_add(version_id, []) # should we check the lines ?
459
self._check_versions_present(parents)
463
for parent in parents:
464
if not self.has_version(parent):
465
ghosts.append(parent)
467
present_parents.append(parent)
469
if delta_parent is None:
470
# reconstitute as full text.
471
assert len(delta) == 1 or len(delta) == 0
473
assert delta[0][0] == 0
474
assert delta[0][1] == 0, delta[0][1]
475
return super(KnitVersionedFile, self)._add_delta(version_id,
486
options.append('no-eol')
488
if delta_parent is not None:
489
# determine the current delta chain length.
490
# To speed the extract of texts the delta chain is limited
491
# to a fixed number of deltas. This should minimize both
492
# I/O and the time spend applying deltas.
493
# The window was changed to a maximum of 200 deltas, but also added
494
# was a check that the total compressed size of the deltas is
495
# smaller than the compressed size of the fulltext.
496
if not self._check_should_delta([delta_parent]):
497
# We don't want a delta here, just do a normal insertion.
498
return super(KnitVersionedFile, self)._add_delta(version_id,
505
options.append('line-delta')
506
store_lines = self.factory.lower_line_delta(delta)
508
access_memo = self._data.add_record(version_id, digest, store_lines)
509
self._index.add_version(version_id, options, access_memo, parents)
511
def _add_raw_records(self, records, data):
512
"""Add all the records 'records' with data pre-joined in 'data'.
514
:param records: A list of tuples(version_id, options, parents, size).
515
:param data: The data for the records. When it is written, the records
516
are adjusted to have pos pointing into data by the sum of
517
the preceding records sizes.
520
raw_record_sizes = [record[3] for record in records]
521
positions = self._data.add_raw_records(raw_record_sizes, data)
524
for (version_id, options, parents, size), access_memo in zip(
526
index_entries.append((version_id, options, access_memo, parents))
527
if self._data._do_cache:
528
self._data._cache[version_id] = data[offset:offset+size]
530
self._index.add_versions(index_entries)
532
def enable_cache(self):
533
"""Start caching data for this knit"""
534
self._data.enable_cache()
536
def clear_cache(self):
537
"""Clear the data cache only."""
538
self._data.clear_cache()
540
def copy_to(self, name, transport):
541
"""See VersionedFile.copy_to()."""
542
# copy the current index to a temp index to avoid racing with local
544
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
545
self.transport.get(self._index._filename))
547
f = self._data._open_file()
549
transport.put_file(name + DATA_SUFFIX, f)
552
# move the copied index into place
553
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
555
def create_empty(self, name, transport, mode=None):
556
return KnitVersionedFile(name, transport, factory=self.factory,
557
delta=self.delta, create=True)
559
def _fix_parents(self, version_id, new_parents):
560
"""Fix the parents list for version.
562
This is done by appending a new version to the index
563
with identical data except for the parents list.
564
the parents list must be a superset of the current
567
current_values = self._index._cache[version_id]
568
assert set(current_values[4]).difference(set(new_parents)) == set()
569
self._index.add_version(version_id,
571
(None, current_values[2], current_values[3]),
574
def get_data_stream(self, required_versions):
575
"""Get a data stream for the specified versions.
577
Versions may be returned in any order, not necessarily the order
580
:param required_versions: The exact set of versions to be extracted.
581
Unlike some other knit methods, this is not used to generate a
582
transitive closure, rather it is used precisely as given.
584
:returns: format_signature, list of (version, options, length, parents),
587
required_versions = set([osutils.safe_revision_id(v) for v in
589
# we don't care about inclusions, the caller cares.
590
# but we need to setup a list of records to visit.
591
for version_id in required_versions:
592
if not self.has_version(version_id):
593
raise RevisionNotPresent(version_id, self.filename)
594
# Pick the desired versions out of the index in oldest-to-newest order
596
for version_id in self.versions():
597
if version_id in required_versions:
598
version_list.append(version_id)
600
# create the list of version information for the result
601
copy_queue_records = []
603
result_version_list = []
604
for version_id in version_list:
605
options = self._index.get_options(version_id)
606
parents = self._index.get_parents_with_ghosts(version_id)
607
index_memo = self._index.get_position(version_id)
608
copy_queue_records.append((version_id, index_memo))
609
none, data_pos, data_size = index_memo
610
copy_set.add(version_id)
611
# version, options, length, parents
612
result_version_list.append((version_id, options, data_size,
615
# Read the compressed record data.
617
# From here down to the return should really be logic in the returned
618
# callable -- in a class that adapts read_records_iter_raw to read
621
for (version_id, raw_data), \
622
(version_id2, options, _, parents) in \
623
izip(self._data.read_records_iter_raw(copy_queue_records),
624
result_version_list):
625
assert version_id == version_id2, 'logic error, inconsistent results'
626
raw_datum.append(raw_data)
627
pseudo_file = StringIO(''.join(raw_datum))
630
return pseudo_file.read()
632
return pseudo_file.read(length)
633
return (self.get_format_signature(), result_version_list, read)
635
def _extract_blocks(self, version_id, source, target):
636
if self._index.get_method(version_id) != 'line-delta':
638
parent, sha1, noeol, delta = self.get_delta(version_id)
639
return KnitContent.get_line_delta_blocks(delta, source, target)
641
def get_delta(self, version_id):
642
"""Get a delta for constructing version from some other version."""
643
version_id = osutils.safe_revision_id(version_id)
644
self.check_not_reserved_id(version_id)
645
if not self.has_version(version_id):
646
raise RevisionNotPresent(version_id, self.filename)
648
parents = self.get_parents(version_id)
653
index_memo = self._index.get_position(version_id)
654
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
655
noeol = 'no-eol' in self._index.get_options(version_id)
656
if 'fulltext' == self._index.get_method(version_id):
657
new_content = self.factory.parse_fulltext(data, version_id)
658
if parent is not None:
659
reference_content = self._get_content(parent)
660
old_texts = reference_content.text()
663
new_texts = new_content.text()
664
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
665
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
667
delta = self.factory.parse_line_delta(data, version_id)
668
return parent, sha1, noeol, delta
670
def get_format_signature(self):
671
"""See VersionedFile.get_format_signature()."""
672
if self.factory.annotated:
673
annotated_part = "annotated"
675
annotated_part = "plain"
676
return "knit-%s" % (annotated_part,)
678
def get_graph_with_ghosts(self):
679
"""See VersionedFile.get_graph_with_ghosts()."""
680
graph_items = self._index.get_graph()
681
return dict(graph_items)
683
def get_sha1(self, version_id):
684
return self.get_sha1s([version_id])[0]
686
def get_sha1s(self, version_ids):
687
"""See VersionedFile.get_sha1()."""
688
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
689
record_map = self._get_record_map(version_ids)
690
# record entry 2 is the 'digest'.
691
return [record_map[v][2] for v in version_ids]
695
"""See VersionedFile.get_suffixes()."""
696
return [DATA_SUFFIX, INDEX_SUFFIX]
698
def has_ghost(self, version_id):
699
"""True if there is a ghost reference in the file to version_id."""
700
version_id = osutils.safe_revision_id(version_id)
702
if self.has_version(version_id):
704
# optimisable if needed by memoising the _ghosts set.
705
items = self._index.get_graph()
706
for node, parents in items:
707
for parent in parents:
708
if parent not in self._index._cache:
709
if parent == version_id:
713
def insert_data_stream(self, (format, data_list, reader_callable)):
714
"""Insert knit records from a data stream into this knit.
716
If a version in the stream is already present in this knit, it will not
717
be inserted a second time. It will be checked for consistency with the
718
stored version however, and may cause a KnitCorrupt error to be raised
719
if the data in the stream disagrees with the already stored data.
721
:seealso: get_data_stream
723
if format != self.get_format_signature():
724
trace.mutter('incompatible format signature inserting to %r', self)
725
raise KnitDataStreamIncompatible(
726
format, self.get_format_signature())
728
for version_id, options, length, parents in data_list:
729
if self.has_version(version_id):
730
# First check: the list of parents.
731
my_parents = self.get_parents_with_ghosts(version_id)
732
if my_parents != parents:
733
# XXX: KnitCorrupt is not quite the right exception here.
736
'parents list %r from data stream does not match '
737
'already recorded parents %r for %s'
738
% (parents, my_parents, version_id))
740
# Also check the SHA-1 of the fulltext this content will
742
raw_data = reader_callable(length)
743
my_fulltext_sha1 = self.get_sha1(version_id)
744
df, rec = self._data._parse_record_header(version_id, raw_data)
745
stream_fulltext_sha1 = rec[3]
746
if my_fulltext_sha1 != stream_fulltext_sha1:
747
# Actually, we don't know if it's this knit that's corrupt,
748
# or the data stream we're trying to insert.
750
self.filename, 'sha-1 does not match %s' % version_id)
752
self._add_raw_records(
753
[(version_id, options, parents, length)],
754
reader_callable(length))
757
"""See VersionedFile.versions."""
758
if 'evil' in debug.debug_flags:
759
trace.mutter_callsite(2, "versions scales with size of history")
760
return self._index.get_versions()
762
def has_version(self, version_id):
763
"""See VersionedFile.has_version."""
764
if 'evil' in debug.debug_flags:
765
trace.mutter_callsite(2, "has_version is a LBYL scenario")
766
version_id = osutils.safe_revision_id(version_id)
767
return self._index.has_version(version_id)
769
__contains__ = has_version
771
def _merge_annotations(self, content, parents, parent_texts={},
772
delta=None, annotated=None,
773
left_matching_blocks=None):
774
"""Merge annotations for content. This is done by comparing
775
the annotations based on changed to the text.
777
if left_matching_blocks is not None:
778
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
782
for parent_id in parents:
783
merge_content = self._get_content(parent_id, parent_texts)
784
if (parent_id == parents[0] and delta_seq is not None):
787
seq = patiencediff.PatienceSequenceMatcher(
788
None, merge_content.text(), content.text())
789
for i, j, n in seq.get_matching_blocks():
792
# this appears to copy (origin, text) pairs across to the
793
# new content for any line that matches the last-checked
795
content._lines[j:j+n] = merge_content._lines[i:i+n]
797
if delta_seq is None:
798
reference_content = self._get_content(parents[0], parent_texts)
799
new_texts = content.text()
800
old_texts = reference_content.text()
801
delta_seq = patiencediff.PatienceSequenceMatcher(
802
None, old_texts, new_texts)
803
return self._make_line_delta(delta_seq, content)
805
def _make_line_delta(self, delta_seq, new_content):
806
"""Generate a line delta from delta_seq and new_content."""
808
for op in delta_seq.get_opcodes():
811
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
814
def _get_components_positions(self, version_ids):
815
"""Produce a map of position data for the components of versions.
817
This data is intended to be used for retrieving the knit records.
819
A dict of version_id to (method, data_pos, data_size, next) is
821
method is the way referenced data should be applied.
822
data_pos is the position of the data in the knit.
823
data_size is the size of the data in the knit.
824
next is the build-parent of the version, or None for fulltexts.
827
for version_id in version_ids:
830
while cursor is not None and cursor not in component_data:
831
method = self._index.get_method(cursor)
832
if method == 'fulltext':
835
next = self.get_parents(cursor)[0]
836
index_memo = self._index.get_position(cursor)
837
component_data[cursor] = (method, index_memo, next)
839
return component_data
841
def _get_content(self, version_id, parent_texts={}):
842
"""Returns a content object that makes up the specified
844
if not self.has_version(version_id):
845
raise RevisionNotPresent(version_id, self.filename)
847
cached_version = parent_texts.get(version_id, None)
848
if cached_version is not None:
849
return cached_version
851
text_map, contents_map = self._get_content_maps([version_id])
852
return contents_map[version_id]
854
def _check_versions_present(self, version_ids):
855
"""Check that all specified versions are present."""
856
self._index.check_versions_present(version_ids)
858
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
860
"""See VersionedFile.add_lines_with_ghosts()."""
861
self._check_add(version_id, lines)
862
return self._add(version_id, lines[:], parents, self.delta,
863
parent_texts, None, nostore_sha)
865
def _add_lines(self, version_id, parents, lines, parent_texts,
866
left_matching_blocks, nostore_sha):
867
"""See VersionedFile.add_lines."""
868
self._check_add(version_id, lines)
869
self._check_versions_present(parents)
870
return self._add(version_id, lines[:], parents, self.delta,
871
parent_texts, left_matching_blocks, nostore_sha)
873
def _check_add(self, version_id, lines):
874
"""check that version_id and lines are safe to add."""
875
assert self.writable, "knit is not opened for write"
876
### FIXME escape. RBC 20060228
877
if contains_whitespace(version_id):
878
raise InvalidRevisionId(version_id, self.filename)
879
self.check_not_reserved_id(version_id)
880
if self.has_version(version_id):
881
raise RevisionAlreadyPresent(version_id, self.filename)
882
self._check_lines_not_unicode(lines)
883
self._check_lines_are_lines(lines)
885
def _add(self, version_id, lines, parents, delta, parent_texts,
886
left_matching_blocks, nostore_sha):
887
"""Add a set of lines on top of version specified by parents.
889
If delta is true, compress the text as a line-delta against
892
Any versions not present will be converted into ghosts.
894
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
895
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
896
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
897
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
898
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
899
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
900
# +1383 0 8.0370 8.0370 +<len>
901
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
902
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
903
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
904
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
908
if parent_texts is None:
910
for parent in parents:
911
if not self.has_version(parent):
912
ghosts.append(parent)
914
present_parents.append(parent)
916
if delta and not len(present_parents):
919
digest = sha_strings(lines)
920
if nostore_sha == digest:
921
raise errors.ExistingContent
922
text_length = sum(map(len, lines))
925
if lines[-1][-1] != '\n':
926
options.append('no-eol')
927
lines[-1] = lines[-1] + '\n'
929
if len(present_parents) and delta:
930
# To speed the extract of texts the delta chain is limited
931
# to a fixed number of deltas. This should minimize both
932
# I/O and the time spend applying deltas.
933
delta = self._check_should_delta(present_parents)
935
assert isinstance(version_id, str)
936
lines = self.factory.make(lines, version_id)
937
if delta or (self.factory.annotated and len(present_parents) > 0):
938
# Merge annotations from parent texts if so is needed.
939
delta_hunks = self._merge_annotations(lines, present_parents,
940
parent_texts, delta, self.factory.annotated,
941
left_matching_blocks)
944
options.append('line-delta')
945
store_lines = self.factory.lower_line_delta(delta_hunks)
947
options.append('fulltext')
948
store_lines = self.factory.lower_fulltext(lines)
950
access_memo = self._data.add_record(version_id, digest, store_lines)
951
self._index.add_version(version_id, options, access_memo, parents)
952
return digest, text_length, lines
954
def check(self, progress_bar=None):
955
"""See VersionedFile.check()."""
957
def _clone_text(self, new_version_id, old_version_id, parents):
958
"""See VersionedFile.clone_text()."""
959
# FIXME RBC 20060228 make fast by only inserting an index with null
961
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
963
def get_lines(self, version_id):
964
"""See VersionedFile.get_lines()."""
965
return self.get_line_list([version_id])[0]
967
def _get_record_map(self, version_ids):
968
"""Produce a dictionary of knit records.
970
The keys are version_ids, the values are tuples of (method, content,
972
method is the way the content should be applied.
973
content is a KnitContent object.
974
digest is the SHA1 digest of this version id after all steps are done
975
next is the build-parent of the version, i.e. the leftmost ancestor.
976
If the method is fulltext, next will be None.
978
position_map = self._get_components_positions(version_ids)
979
# c = component_id, m = method, i_m = index_memo, n = next
980
records = [(c, i_m) for c, (m, i_m, n) in position_map.iteritems()]
982
for component_id, content, digest in \
983
self._data.read_records_iter(records):
984
method, index_memo, next = position_map[component_id]
985
record_map[component_id] = method, content, digest, next
989
def get_text(self, version_id):
990
"""See VersionedFile.get_text"""
991
return self.get_texts([version_id])[0]
993
def get_texts(self, version_ids):
994
return [''.join(l) for l in self.get_line_list(version_ids)]
996
def get_line_list(self, version_ids):
997
"""Return the texts of listed versions as a list of strings."""
998
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
999
for version_id in version_ids:
1000
self.check_not_reserved_id(version_id)
1001
text_map, content_map = self._get_content_maps(version_ids)
1002
return [text_map[v] for v in version_ids]
1004
_get_lf_split_line_list = get_line_list
1006
def _get_content_maps(self, version_ids):
1007
"""Produce maps of text and KnitContents
1009
:return: (text_map, content_map) where text_map contains the texts for
1010
the requested versions and content_map contains the KnitContents.
1011
Both dicts take version_ids as their keys.
1013
for version_id in version_ids:
1014
if not self.has_version(version_id):
1015
raise RevisionNotPresent(version_id, self.filename)
1016
record_map = self._get_record_map(version_ids)
1021
for version_id in version_ids:
1024
while cursor is not None:
1025
method, data, digest, next = record_map[cursor]
1026
components.append((cursor, method, data, digest))
1027
if cursor in content_map:
1032
for component_id, method, data, digest in reversed(components):
1033
if component_id in content_map:
1034
content = content_map[component_id]
1036
if method == 'fulltext':
1037
assert content is None
1038
content = self.factory.parse_fulltext(data, version_id)
1039
elif method == 'line-delta':
1040
delta = self.factory.parse_line_delta(data, version_id)
1041
content = content.copy()
1042
content._lines = self._apply_delta(content._lines,
1044
content_map[component_id] = content
1046
if 'no-eol' in self._index.get_options(version_id):
1047
content = content.copy()
1048
line = content._lines[-1][1].rstrip('\n')
1049
content._lines[-1] = (content._lines[-1][0], line)
1050
final_content[version_id] = content
1052
# digest here is the digest from the last applied component.
1053
text = content.text()
1054
if sha_strings(text) != digest:
1055
raise KnitCorrupt(self.filename,
1056
'sha-1 does not match %s' % version_id)
1058
text_map[version_id] = text
1059
return text_map, final_content
1061
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1063
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1064
if version_ids is None:
1065
version_ids = self.versions()
1067
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
1069
pb = progress.DummyProgress()
1070
# we don't care about inclusions, the caller cares.
1071
# but we need to setup a list of records to visit.
1072
# we need version_id, position, length
1073
version_id_records = []
1074
requested_versions = set(version_ids)
1075
# filter for available versions
1076
for version_id in requested_versions:
1077
if not self.has_version(version_id):
1078
raise RevisionNotPresent(version_id, self.filename)
1079
# get a in-component-order queue:
1080
for version_id in self.versions():
1081
if version_id in requested_versions:
1082
index_memo = self._index.get_position(version_id)
1083
version_id_records.append((version_id, index_memo))
1085
total = len(version_id_records)
1086
for version_idx, (version_id, data, sha_value) in \
1087
enumerate(self._data.read_records_iter(version_id_records)):
1088
pb.update('Walking content.', version_idx, total)
1089
method = self._index.get_method(version_id)
1091
assert method in ('fulltext', 'line-delta')
1092
if method == 'fulltext':
1093
line_iterator = self.factory.get_fulltext_content(data)
1095
line_iterator = self.factory.get_linedelta_content(data)
1096
for line in line_iterator:
1099
pb.update('Walking content.', total, total)
1101
def iter_parents(self, version_ids):
1102
"""Iterate through the parents for many version ids.
1104
:param version_ids: An iterable yielding version_ids.
1105
:return: An iterator that yields (version_id, parents). Requested
1106
version_ids not present in the versioned file are simply skipped.
1107
The order is undefined, allowing for different optimisations in
1108
the underlying implementation.
1110
version_ids = [osutils.safe_revision_id(version_id) for
1111
version_id in version_ids]
1112
return self._index.iter_parents(version_ids)
1114
def num_versions(self):
1115
"""See VersionedFile.num_versions()."""
1116
return self._index.num_versions()
1118
__len__ = num_versions
1120
def annotate_iter(self, version_id):
1121
"""See VersionedFile.annotate_iter."""
1122
version_id = osutils.safe_revision_id(version_id)
1123
return self.factory.annotate_iter(self, version_id)
1125
def get_parents(self, version_id):
1126
"""See VersionedFile.get_parents."""
1129
# 52554 calls in 1264 872 internal down from 3674
1130
version_id = osutils.safe_revision_id(version_id)
1132
return self._index.get_parents(version_id)
1134
raise RevisionNotPresent(version_id, self.filename)
1136
def get_parents_with_ghosts(self, version_id):
1137
"""See VersionedFile.get_parents."""
1138
version_id = osutils.safe_revision_id(version_id)
1140
return self._index.get_parents_with_ghosts(version_id)
1142
raise RevisionNotPresent(version_id, self.filename)
1144
def get_ancestry(self, versions, topo_sorted=True):
1145
"""See VersionedFile.get_ancestry."""
1146
if isinstance(versions, basestring):
1147
versions = [versions]
1150
versions = [osutils.safe_revision_id(v) for v in versions]
1151
return self._index.get_ancestry(versions, topo_sorted)
1153
def get_ancestry_with_ghosts(self, versions):
1154
"""See VersionedFile.get_ancestry_with_ghosts."""
1155
if isinstance(versions, basestring):
1156
versions = [versions]
1159
versions = [osutils.safe_revision_id(v) for v in versions]
1160
return self._index.get_ancestry_with_ghosts(versions)
1162
def plan_merge(self, ver_a, ver_b):
1163
"""See VersionedFile.plan_merge."""
1164
ver_a = osutils.safe_revision_id(ver_a)
1165
ver_b = osutils.safe_revision_id(ver_b)
1166
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1168
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1169
annotated_a = self.annotate(ver_a)
1170
annotated_b = self.annotate(ver_b)
1171
return merge._plan_annotate_merge(annotated_a, annotated_b,
1172
ancestors_a, ancestors_b)
1175
class _KnitComponentFile(object):
1176
"""One of the files used to implement a knit database"""
1178
def __init__(self, transport, filename, mode, file_mode=None,
1179
create_parent_dir=False, dir_mode=None):
1180
self._transport = transport
1181
self._filename = filename
1183
self._file_mode = file_mode
1184
self._dir_mode = dir_mode
1185
self._create_parent_dir = create_parent_dir
1186
self._need_to_create = False
1188
def _full_path(self):
1189
"""Return the full path to this file."""
1190
return self._transport.base + self._filename
1192
def check_header(self, fp):
1193
line = fp.readline()
1195
# An empty file can actually be treated as though the file doesn't
1197
raise errors.NoSuchFile(self._full_path())
1198
if line != self.HEADER:
1199
raise KnitHeaderError(badline=line,
1200
filename=self._transport.abspath(self._filename))
1203
return '%s(%s)' % (self.__class__.__name__, self._filename)
1206
class _KnitIndex(_KnitComponentFile):
1207
"""Manages knit index file.
1209
The index is already kept in memory and read on startup, to enable
1210
fast lookups of revision information. The cursor of the index
1211
file is always pointing to the end, making it easy to append
1214
_cache is a cache for fast mapping from version id to a Index
1217
_history is a cache for fast mapping from indexes to version ids.
1219
The index data format is dictionary compressed when it comes to
1220
parent references; a index entry may only have parents that with a
1221
lover index number. As a result, the index is topological sorted.
1223
Duplicate entries may be written to the index for a single version id
1224
if this is done then the latter one completely replaces the former:
1225
this allows updates to correct version and parent information.
1226
Note that the two entries may share the delta, and that successive
1227
annotations and references MUST point to the first entry.
1229
The index file on disc contains a header, followed by one line per knit
1230
record. The same revision can be present in an index file more than once.
1231
The first occurrence gets assigned a sequence number starting from 0.
1233
The format of a single line is
1234
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1235
REVISION_ID is a utf8-encoded revision id
1236
FLAGS is a comma separated list of flags about the record. Values include
1237
no-eol, line-delta, fulltext.
1238
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1239
that the the compressed data starts at.
1240
LENGTH is the ascii representation of the length of the data file.
1241
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1243
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1244
revision id already in the knit that is a parent of REVISION_ID.
1245
The ' :' marker is the end of record marker.
1248
when a write is interrupted to the index file, it will result in a line
1249
that does not end in ' :'. If the ' :' is not present at the end of a line,
1250
or at the end of the file, then the record that is missing it will be
1251
ignored by the parser.
1253
When writing new records to the index file, the data is preceded by '\n'
1254
to ensure that records always start on new lines even if the last write was
1255
interrupted. As a result its normal for the last line in the index to be
1256
missing a trailing newline. One can be added with no harmful effects.
1259
HEADER = "# bzr knit index 8\n"
1261
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1262
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1264
def _cache_version(self, version_id, options, pos, size, parents):
1265
"""Cache a version record in the history array and index cache.
1267
This is inlined into _load_data for performance. KEEP IN SYNC.
1268
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1271
# only want the _history index to reference the 1st index entry
1273
if version_id not in self._cache:
1274
index = len(self._history)
1275
self._history.append(version_id)
1277
index = self._cache[version_id][5]
1278
self._cache[version_id] = (version_id,
1285
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1286
create_parent_dir=False, delay_create=False, dir_mode=None):
1287
_KnitComponentFile.__init__(self, transport, filename, mode,
1288
file_mode=file_mode,
1289
create_parent_dir=create_parent_dir,
1292
# position in _history is the 'official' index for a revision
1293
# but the values may have come from a newer entry.
1294
# so - wc -l of a knit index is != the number of unique names
1298
fp = self._transport.get(self._filename)
1300
# _load_data may raise NoSuchFile if the target knit is
1302
_load_data(self, fp)
1306
if mode != 'w' or not create:
1309
self._need_to_create = True
1311
self._transport.put_bytes_non_atomic(
1312
self._filename, self.HEADER, mode=self._file_mode)
1314
def get_graph(self):
1315
"""Return a list of the node:parents lists from this knit index."""
1316
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1318
def get_ancestry(self, versions, topo_sorted=True):
1319
"""See VersionedFile.get_ancestry."""
1320
# get a graph of all the mentioned versions:
1322
pending = set(versions)
1325
version = pending.pop()
1328
parents = [p for p in cache[version][4] if p in cache]
1330
raise RevisionNotPresent(version, self._filename)
1331
# if not completed and not a ghost
1332
pending.update([p for p in parents if p not in graph])
1333
graph[version] = parents
1336
return topo_sort(graph.items())
1338
def get_ancestry_with_ghosts(self, versions):
1339
"""See VersionedFile.get_ancestry_with_ghosts."""
1340
# get a graph of all the mentioned versions:
1341
self.check_versions_present(versions)
1344
pending = set(versions)
1346
version = pending.pop()
1348
parents = cache[version][4]
1354
pending.update([p for p in parents if p not in graph])
1355
graph[version] = parents
1356
return topo_sort(graph.items())
1358
def iter_parents(self, version_ids):
1359
"""Iterate through the parents for many version ids.
1361
:param version_ids: An iterable yielding version_ids.
1362
:return: An iterator that yields (version_id, parents). Requested
1363
version_ids not present in the versioned file are simply skipped.
1364
The order is undefined, allowing for different optimisations in
1365
the underlying implementation.
1367
for version_id in version_ids:
1369
yield version_id, tuple(self.get_parents(version_id))
1373
def num_versions(self):
1374
return len(self._history)
1376
__len__ = num_versions
1378
def get_versions(self):
1379
"""Get all the versions in the file. not topologically sorted."""
1380
return self._history
1382
def _version_list_to_index(self, versions):
1385
for version in versions:
1386
if version in cache:
1387
# -- inlined lookup() --
1388
result_list.append(str(cache[version][5]))
1389
# -- end lookup () --
1391
result_list.append('.' + version)
1392
return ' '.join(result_list)
1394
def add_version(self, version_id, options, index_memo, parents):
1395
"""Add a version record to the index."""
1396
self.add_versions(((version_id, options, index_memo, parents),))
1398
def add_versions(self, versions):
1399
"""Add multiple versions to the index.
1401
:param versions: a list of tuples:
1402
(version_id, options, pos, size, parents).
1405
orig_history = self._history[:]
1406
orig_cache = self._cache.copy()
1409
for version_id, options, (index, pos, size), parents in versions:
1410
line = "\n%s %s %s %s %s :" % (version_id,
1414
self._version_list_to_index(parents))
1415
assert isinstance(line, str), \
1416
'content must be utf-8 encoded: %r' % (line,)
1418
self._cache_version(version_id, options, pos, size, parents)
1419
if not self._need_to_create:
1420
self._transport.append_bytes(self._filename, ''.join(lines))
1423
sio.write(self.HEADER)
1424
sio.writelines(lines)
1426
self._transport.put_file_non_atomic(self._filename, sio,
1427
create_parent_dir=self._create_parent_dir,
1428
mode=self._file_mode,
1429
dir_mode=self._dir_mode)
1430
self._need_to_create = False
1432
# If any problems happen, restore the original values and re-raise
1433
self._history = orig_history
1434
self._cache = orig_cache
1437
def has_version(self, version_id):
1438
"""True if the version is in the index."""
1439
return version_id in self._cache
1441
def get_position(self, version_id):
1442
"""Return details needed to access the version.
1444
.kndx indices do not support split-out data, so return None for the
1447
:return: a tuple (None, data position, size) to hand to the access
1448
logic to get the record.
1450
entry = self._cache[version_id]
1451
return None, entry[2], entry[3]
1453
def get_method(self, version_id):
1454
"""Return compression method of specified version."""
1455
options = self._cache[version_id][1]
1456
if 'fulltext' in options:
1459
if 'line-delta' not in options:
1460
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1463
def get_options(self, version_id):
1464
"""Return a string represention options.
1468
return self._cache[version_id][1]
1470
def get_parents(self, version_id):
1471
"""Return parents of specified version ignoring ghosts."""
1472
return [parent for parent in self._cache[version_id][4]
1473
if parent in self._cache]
1475
def get_parents_with_ghosts(self, version_id):
1476
"""Return parents of specified version with ghosts."""
1477
return self._cache[version_id][4]
1479
def check_versions_present(self, version_ids):
1480
"""Check that all specified versions are present."""
1482
for version_id in version_ids:
1483
if version_id not in cache:
1484
raise RevisionNotPresent(version_id, self._filename)
1487
class KnitGraphIndex(object):
1488
"""A knit index that builds on GraphIndex."""
1490
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1491
"""Construct a KnitGraphIndex on a graph_index.
1493
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1494
:param deltas: Allow delta-compressed records.
1495
:param add_callback: If not None, allow additions to the index and call
1496
this callback with a list of added GraphIndex nodes:
1497
[(node, value, node_refs), ...]
1498
:param parents: If True, record knits parents, if not do not record
1501
self._graph_index = graph_index
1502
self._deltas = deltas
1503
self._add_callback = add_callback
1504
self._parents = parents
1505
if deltas and not parents:
1506
raise KnitCorrupt(self, "Cannot do delta compression without "
1509
def _get_entries(self, keys, check_present=False):
1510
"""Get the entries for keys.
1512
:param keys: An iterable of index keys, - 1-tuples.
1517
for node in self._graph_index.iter_entries(keys):
1519
found_keys.add(node[1])
1521
# adapt parentless index to the rest of the code.
1522
for node in self._graph_index.iter_entries(keys):
1523
yield node[0], node[1], node[2], ()
1524
found_keys.add(node[1])
1526
missing_keys = keys.difference(found_keys)
1528
raise RevisionNotPresent(missing_keys.pop(), self)
1530
def _present_keys(self, version_ids):
1532
node[1] for node in self._get_entries(version_ids)])
1534
def _parentless_ancestry(self, versions):
1535
"""Honour the get_ancestry API for parentless knit indices."""
1536
wanted_keys = self._version_ids_to_keys(versions)
1537
present_keys = self._present_keys(wanted_keys)
1538
missing = set(wanted_keys).difference(present_keys)
1540
raise RevisionNotPresent(missing.pop(), self)
1541
return list(self._keys_to_version_ids(present_keys))
1543
def get_ancestry(self, versions, topo_sorted=True):
1544
"""See VersionedFile.get_ancestry."""
1545
if not self._parents:
1546
return self._parentless_ancestry(versions)
1547
# XXX: This will do len(history) index calls - perhaps
1548
# it should be altered to be a index core feature?
1549
# get a graph of all the mentioned versions:
1552
versions = self._version_ids_to_keys(versions)
1553
pending = set(versions)
1555
# get all pending nodes
1556
this_iteration = pending
1557
new_nodes = self._get_entries(this_iteration)
1560
for (index, key, value, node_refs) in new_nodes:
1561
# dont ask for ghosties - otherwise
1562
# we we can end up looping with pending
1563
# being entirely ghosted.
1564
graph[key] = [parent for parent in node_refs[0]
1565
if parent not in ghosts]
1567
for parent in graph[key]:
1568
# dont examine known nodes again
1573
ghosts.update(this_iteration.difference(found))
1574
if versions.difference(graph):
1575
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1577
result_keys = topo_sort(graph.items())
1579
result_keys = graph.iterkeys()
1580
return [key[0] for key in result_keys]
1582
def get_ancestry_with_ghosts(self, versions):
1583
"""See VersionedFile.get_ancestry."""
1584
if not self._parents:
1585
return self._parentless_ancestry(versions)
1586
# XXX: This will do len(history) index calls - perhaps
1587
# it should be altered to be a index core feature?
1588
# get a graph of all the mentioned versions:
1590
versions = self._version_ids_to_keys(versions)
1591
pending = set(versions)
1593
# get all pending nodes
1594
this_iteration = pending
1595
new_nodes = self._get_entries(this_iteration)
1597
for (index, key, value, node_refs) in new_nodes:
1598
graph[key] = node_refs[0]
1600
for parent in graph[key]:
1601
# dont examine known nodes again
1605
missing_versions = this_iteration.difference(graph)
1606
missing_needed = versions.intersection(missing_versions)
1608
raise RevisionNotPresent(missing_needed.pop(), self)
1609
for missing_version in missing_versions:
1610
# add a key, no parents
1611
graph[missing_version] = []
1612
pending.discard(missing_version) # don't look for it
1613
result_keys = topo_sort(graph.items())
1614
return [key[0] for key in result_keys]
1616
def get_graph(self):
1617
"""Return a list of the node:parents lists from this knit index."""
1618
if not self._parents:
1619
return [(key, ()) for key in self.get_versions()]
1621
for index, key, value, refs in self._graph_index.iter_all_entries():
1622
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1625
def iter_parents(self, version_ids):
1626
"""Iterate through the parents for many version ids.
1628
:param version_ids: An iterable yielding version_ids.
1629
:return: An iterator that yields (version_id, parents). Requested
1630
version_ids not present in the versioned file are simply skipped.
1631
The order is undefined, allowing for different optimisations in
1632
the underlying implementation.
1635
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1637
present_parents = set()
1638
for node in all_nodes:
1639
all_parents.update(node[3][0])
1640
# any node we are querying must be present
1641
present_parents.add(node[1])
1642
unknown_parents = all_parents.difference(present_parents)
1643
present_parents.update(self._present_keys(unknown_parents))
1644
for node in all_nodes:
1646
for parent in node[3][0]:
1647
if parent in present_parents:
1648
parents.append(parent[0])
1649
yield node[1][0], tuple(parents)
1651
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1652
yield node[1][0], ()
1654
def num_versions(self):
1655
return len(list(self._graph_index.iter_all_entries()))
1657
__len__ = num_versions
1659
def get_versions(self):
1660
"""Get all the versions in the file. not topologically sorted."""
1661
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1663
def has_version(self, version_id):
1664
"""True if the version is in the index."""
1665
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1667
def _keys_to_version_ids(self, keys):
1668
return tuple(key[0] for key in keys)
1670
def get_position(self, version_id):
1671
"""Return details needed to access the version.
1673
:return: a tuple (index, data position, size) to hand to the access
1674
logic to get the record.
1676
node = self._get_node(version_id)
1677
bits = node[2][1:].split(' ')
1678
return node[0], int(bits[0]), int(bits[1])
1680
def get_method(self, version_id):
1681
"""Return compression method of specified version."""
1682
if not self._deltas:
1684
return self._parent_compression(self._get_node(version_id)[3][1])
1686
def _parent_compression(self, reference_list):
1687
# use the second reference list to decide if this is delta'd or not.
1688
if len(reference_list):
1693
def _get_node(self, version_id):
1694
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1696
def get_options(self, version_id):
1697
"""Return a string represention options.
1701
node = self._get_node(version_id)
1702
if not self._deltas:
1703
options = ['fulltext']
1705
options = [self._parent_compression(node[3][1])]
1706
if node[2][0] == 'N':
1707
options.append('no-eol')
1710
def get_parents(self, version_id):
1711
"""Return parents of specified version ignoring ghosts."""
1712
parents = list(self.iter_parents([version_id]))
1715
raise errors.RevisionNotPresent(version_id, self)
1716
return parents[0][1]
1718
def get_parents_with_ghosts(self, version_id):
1719
"""Return parents of specified version with ghosts."""
1720
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1721
check_present=True))
1722
if not self._parents:
1724
return self._keys_to_version_ids(nodes[0][3][0])
1726
def check_versions_present(self, version_ids):
1727
"""Check that all specified versions are present."""
1728
keys = self._version_ids_to_keys(version_ids)
1729
present = self._present_keys(keys)
1730
missing = keys.difference(present)
1732
raise RevisionNotPresent(missing.pop(), self)
1734
def add_version(self, version_id, options, access_memo, parents):
1735
"""Add a version record to the index."""
1736
return self.add_versions(((version_id, options, access_memo, parents),))
1738
def add_versions(self, versions):
1739
"""Add multiple versions to the index.
1741
This function does not insert data into the Immutable GraphIndex
1742
backing the KnitGraphIndex, instead it prepares data for insertion by
1743
the caller and checks that it is safe to insert then calls
1744
self._add_callback with the prepared GraphIndex nodes.
1746
:param versions: a list of tuples:
1747
(version_id, options, pos, size, parents).
1749
if not self._add_callback:
1750
raise errors.ReadOnlyError(self)
1751
# we hope there are no repositories with inconsistent parentage
1756
for (version_id, options, access_memo, parents) in versions:
1757
index, pos, size = access_memo
1758
key = (version_id, )
1759
parents = tuple((parent, ) for parent in parents)
1760
if 'no-eol' in options:
1764
value += "%d %d" % (pos, size)
1765
if not self._deltas:
1766
if 'line-delta' in options:
1767
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1770
if 'line-delta' in options:
1771
node_refs = (parents, (parents[0],))
1773
node_refs = (parents, ())
1775
node_refs = (parents, )
1778
raise KnitCorrupt(self, "attempt to add node with parents "
1779
"in parentless index.")
1781
keys[key] = (value, node_refs)
1782
present_nodes = self._get_entries(keys)
1783
for (index, key, value, node_refs) in present_nodes:
1784
if (value, node_refs) != keys[key]:
1785
raise KnitCorrupt(self, "inconsistent details in add_versions"
1786
": %s %s" % ((value, node_refs), keys[key]))
1790
for key, (value, node_refs) in keys.iteritems():
1791
result.append((key, value, node_refs))
1793
for key, (value, node_refs) in keys.iteritems():
1794
result.append((key, value))
1795
self._add_callback(result)
1797
def _version_ids_to_keys(self, version_ids):
1798
return set((version_id, ) for version_id in version_ids)
1801
class _KnitAccess(object):
1802
"""Access to knit records in a .knit file."""
1804
def __init__(self, transport, filename, _file_mode, _dir_mode,
1805
_need_to_create, _create_parent_dir):
1806
"""Create a _KnitAccess for accessing and inserting data.
1808
:param transport: The transport the .knit is located on.
1809
:param filename: The filename of the .knit.
1811
self._transport = transport
1812
self._filename = filename
1813
self._file_mode = _file_mode
1814
self._dir_mode = _dir_mode
1815
self._need_to_create = _need_to_create
1816
self._create_parent_dir = _create_parent_dir
1818
def add_raw_records(self, sizes, raw_data):
1819
"""Add raw knit bytes to a storage area.
1821
The data is spooled to whereever the access method is storing data.
1823
:param sizes: An iterable containing the size of each raw data segment.
1824
:param raw_data: A bytestring containing the data.
1825
:return: A list of memos to retrieve the record later. Each memo is a
1826
tuple - (index, pos, length), where the index field is always None
1827
for the .knit access method.
1829
assert type(raw_data) == str, \
1830
'data must be plain bytes was %s' % type(raw_data)
1831
if not self._need_to_create:
1832
base = self._transport.append_bytes(self._filename, raw_data)
1834
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1835
create_parent_dir=self._create_parent_dir,
1836
mode=self._file_mode,
1837
dir_mode=self._dir_mode)
1838
self._need_to_create = False
1842
result.append((None, base, size))
1847
"""IFF this data access has its own storage area, initialise it.
1851
self._transport.put_bytes_non_atomic(self._filename, '',
1852
mode=self._file_mode)
1854
def open_file(self):
1855
"""IFF this data access can be represented as a single file, open it.
1857
For knits that are not mapped to a single file on disk this will
1860
:return: None or a file handle.
1863
return self._transport.get(self._filename)
1868
def get_raw_records(self, memos_for_retrieval):
1869
"""Get the raw bytes for a records.
1871
:param memos_for_retrieval: An iterable containing the (index, pos,
1872
length) memo for retrieving the bytes. The .knit method ignores
1873
the index as there is always only a single file.
1874
:return: An iterator over the bytes of the records.
1876
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
1877
for pos, data in self._transport.readv(self._filename, read_vector):
1881
class _PackAccess(object):
1882
"""Access to knit records via a collection of packs."""
1884
def __init__(self, index_to_packs, writer=None):
1885
"""Create a _PackAccess object.
1887
:param index_to_packs: A dict mapping index objects to the transport
1888
and file names for obtaining data.
1889
:param writer: A tuple (pack.ContainerWriter, write_index) which
1890
contains the pack to write, and the index that reads from it will
1894
self.container_writer = writer[0]
1895
self.write_index = writer[1]
1897
self.container_writer = None
1898
self.write_index = None
1899
self.indices = index_to_packs
1901
def add_raw_records(self, sizes, raw_data):
1902
"""Add raw knit bytes to a storage area.
1904
The data is spooled to the container writer in one bytes-record per
1907
:param sizes: An iterable containing the size of each raw data segment.
1908
:param raw_data: A bytestring containing the data.
1909
:return: A list of memos to retrieve the record later. Each memo is a
1910
tuple - (index, pos, length), where the index field is the
1911
write_index object supplied to the PackAccess object.
1913
assert type(raw_data) == str, \
1914
'data must be plain bytes was %s' % type(raw_data)
1918
p_offset, p_length = self.container_writer.add_bytes_record(
1919
raw_data[offset:offset+size], [])
1921
result.append((self.write_index, p_offset, p_length))
1925
"""Pack based knits do not get individually created."""
1927
def get_raw_records(self, memos_for_retrieval):
1928
"""Get the raw bytes for a records.
1930
:param memos_for_retrieval: An iterable containing the (index, pos,
1931
length) memo for retrieving the bytes. The Pack access method
1932
looks up the pack to use for a given record in its index_to_pack
1934
:return: An iterator over the bytes of the records.
1936
# first pass, group into same-index requests
1938
current_index = None
1939
for (index, offset, length) in memos_for_retrieval:
1940
if current_index == index:
1941
current_list.append((offset, length))
1943
if current_index is not None:
1944
request_lists.append((current_index, current_list))
1945
current_index = index
1946
current_list = [(offset, length)]
1947
# handle the last entry
1948
if current_index is not None:
1949
request_lists.append((current_index, current_list))
1950
for index, offsets in request_lists:
1951
transport, path = self.indices[index]
1952
reader = pack.make_readv_reader(transport, path, offsets)
1953
for names, read_func in reader.iter_records():
1954
yield read_func(None)
1956
def open_file(self):
1957
"""Pack based knits have no single file."""
1960
def set_writer(self, writer, index, (transport, packname)):
1961
"""Set a writer to use for adding data."""
1962
self.indices[index] = (transport, packname)
1963
self.container_writer = writer
1964
self.write_index = index
1967
class _KnitData(object):
1968
"""Manage extraction of data from a KnitAccess, caching and decompressing.
1970
The KnitData class provides the logic for parsing and using knit records,
1971
making use of an access method for the low level read and write operations.
1974
def __init__(self, access):
1975
"""Create a KnitData object.
1977
:param access: The access method to use. Access methods such as
1978
_KnitAccess manage the insertion of raw records and the subsequent
1979
retrieval of the same.
1981
self._access = access
1982
self._checked = False
1983
# TODO: jam 20060713 conceptually, this could spill to disk
1984
# if the cached size gets larger than a certain amount
1985
# but it complicates the model a bit, so for now just use
1986
# a simple dictionary
1988
self._do_cache = False
1990
def enable_cache(self):
1991
"""Enable caching of reads."""
1992
self._do_cache = True
1994
def clear_cache(self):
1995
"""Clear the record cache."""
1996
self._do_cache = False
1999
def _open_file(self):
2000
return self._access.open_file()
2002
def _record_to_data(self, version_id, digest, lines):
2003
"""Convert version_id, digest, lines into a raw data block.
2005
:return: (len, a StringIO instance with the raw data ready to read.)
2008
data_file = GzipFile(None, mode='wb', fileobj=sio,
2009
compresslevel=Z_DEFAULT_COMPRESSION)
2011
assert isinstance(version_id, str)
2012
data_file.writelines(chain(
2013
["version %s %d %s\n" % (version_id,
2017
["end %s\n" % version_id]))
2024
def add_raw_records(self, sizes, raw_data):
2025
"""Append a prepared record to the data file.
2027
:param sizes: An iterable containing the size of each raw data segment.
2028
:param raw_data: A bytestring containing the data.
2029
:return: a list of index data for the way the data was stored.
2030
See the access method add_raw_records documentation for more
2033
return self._access.add_raw_records(sizes, raw_data)
2035
def add_record(self, version_id, digest, lines):
2036
"""Write new text record to disk.
2038
Returns index data for retrieving it later, as per add_raw_records.
2040
size, sio = self._record_to_data(version_id, digest, lines)
2041
result = self.add_raw_records([size], sio.getvalue())
2043
self._cache[version_id] = sio.getvalue()
2046
def _parse_record_header(self, version_id, raw_data):
2047
"""Parse a record header for consistency.
2049
:return: the header and the decompressor stream.
2050
as (stream, header_record)
2052
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2054
rec = self._check_header(version_id, df.readline())
2055
except Exception, e:
2056
raise KnitCorrupt(self._access,
2057
"While reading {%s} got %s(%s)"
2058
% (version_id, e.__class__.__name__, str(e)))
2061
def _check_header(self, version_id, line):
2064
raise KnitCorrupt(self._access,
2065
'unexpected number of elements in record header')
2066
if rec[1] != version_id:
2067
raise KnitCorrupt(self._access,
2068
'unexpected version, wanted %r, got %r'
2069
% (version_id, rec[1]))
2072
def _parse_record(self, version_id, data):
2074
# 4168 calls in 2880 217 internal
2075
# 4168 calls to _parse_record_header in 2121
2076
# 4168 calls to readlines in 330
2077
df = GzipFile(mode='rb', fileobj=StringIO(data))
2080
record_contents = df.readlines()
2081
except Exception, e:
2082
raise KnitCorrupt(self._access,
2083
"While reading {%s} got %s(%s)"
2084
% (version_id, e.__class__.__name__, str(e)))
2085
header = record_contents.pop(0)
2086
rec = self._check_header(version_id, header)
2088
last_line = record_contents.pop()
2089
if len(record_contents) != int(rec[2]):
2090
raise KnitCorrupt(self._access,
2091
'incorrect number of lines %s != %s'
2093
% (len(record_contents), int(rec[2]),
2095
if last_line != 'end %s\n' % rec[1]:
2096
raise KnitCorrupt(self._access,
2097
'unexpected version end line %r, wanted %r'
2098
% (last_line, version_id))
2100
return record_contents, rec[3]
2102
def read_records_iter_raw(self, records):
2103
"""Read text records from data file and yield raw data.
2105
This unpacks enough of the text record to validate the id is
2106
as expected but thats all.
2108
# setup an iterator of the external records:
2109
# uses readv so nice and fast we hope.
2111
# grab the disk data needed.
2113
# Don't check _cache if it is empty
2114
needed_offsets = [index_memo for version_id, index_memo
2116
if version_id not in self._cache]
2118
needed_offsets = [index_memo for version_id, index_memo
2121
raw_records = self._access.get_raw_records(needed_offsets)
2123
for version_id, index_memo in records:
2124
if version_id in self._cache:
2125
# This data has already been validated
2126
data = self._cache[version_id]
2128
data = raw_records.next()
2130
self._cache[version_id] = data
2132
# validate the header
2133
df, rec = self._parse_record_header(version_id, data)
2135
yield version_id, data
2137
def read_records_iter(self, records):
2138
"""Read text records from data file and yield result.
2140
The result will be returned in whatever is the fastest to read.
2141
Not by the order requested. Also, multiple requests for the same
2142
record will only yield 1 response.
2143
:param records: A list of (version_id, pos, len) entries
2144
:return: Yields (version_id, contents, digest) in the order
2145
read, not the order requested
2151
# Skip records we have alread seen
2152
yielded_records = set()
2153
needed_records = set()
2154
for record in records:
2155
if record[0] in self._cache:
2156
if record[0] in yielded_records:
2158
yielded_records.add(record[0])
2159
data = self._cache[record[0]]
2160
content, digest = self._parse_record(record[0], data)
2161
yield (record[0], content, digest)
2163
needed_records.add(record)
2164
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2166
needed_records = sorted(set(records), key=operator.itemgetter(1))
2168
if not needed_records:
2171
# The transport optimizes the fetching as well
2172
# (ie, reads continuous ranges.)
2173
raw_data = self._access.get_raw_records(
2174
[index_memo for version_id, index_memo in needed_records])
2176
for (version_id, index_memo), data in \
2177
izip(iter(needed_records), raw_data):
2178
content, digest = self._parse_record(version_id, data)
2180
self._cache[version_id] = data
2181
yield version_id, content, digest
2183
def read_records(self, records):
2184
"""Read records into a dictionary."""
2186
for record_id, content, digest in \
2187
self.read_records_iter(records):
2188
components[record_id] = (content, digest)
2192
class InterKnit(InterVersionedFile):
2193
"""Optimised code paths for knit to knit operations."""
2195
_matching_file_from_factory = KnitVersionedFile
2196
_matching_file_to_factory = KnitVersionedFile
2199
def is_compatible(source, target):
2200
"""Be compatible with knits. """
2202
return (isinstance(source, KnitVersionedFile) and
2203
isinstance(target, KnitVersionedFile))
2204
except AttributeError:
2207
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2208
"""See InterVersionedFile.join."""
2209
assert isinstance(self.source, KnitVersionedFile)
2210
assert isinstance(self.target, KnitVersionedFile)
2212
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2217
pb = ui.ui_factory.nested_progress_bar()
2219
version_ids = list(version_ids)
2220
if None in version_ids:
2221
version_ids.remove(None)
2223
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2224
this_versions = set(self.target._index.get_versions())
2225
needed_versions = self.source_ancestry - this_versions
2226
cross_check_versions = self.source_ancestry.intersection(this_versions)
2227
mismatched_versions = set()
2228
for version in cross_check_versions:
2229
# scan to include needed parents.
2230
n1 = set(self.target.get_parents_with_ghosts(version))
2231
n2 = set(self.source.get_parents_with_ghosts(version))
2233
# FIXME TEST this check for cycles being introduced works
2234
# the logic is we have a cycle if in our graph we are an
2235
# ancestor of any of the n2 revisions.
2241
parent_ancestors = self.source.get_ancestry(parent)
2242
if version in parent_ancestors:
2243
raise errors.GraphCycleError([parent, version])
2244
# ensure this parent will be available later.
2245
new_parents = n2.difference(n1)
2246
needed_versions.update(new_parents.difference(this_versions))
2247
mismatched_versions.add(version)
2249
if not needed_versions and not mismatched_versions:
2251
full_list = topo_sort(self.source.get_graph())
2253
version_list = [i for i in full_list if (not self.target.has_version(i)
2254
and i in needed_versions)]
2258
copy_queue_records = []
2260
for version_id in version_list:
2261
options = self.source._index.get_options(version_id)
2262
parents = self.source._index.get_parents_with_ghosts(version_id)
2263
# check that its will be a consistent copy:
2264
for parent in parents:
2265
# if source has the parent, we must :
2266
# * already have it or
2267
# * have it scheduled already
2268
# otherwise we don't care
2269
assert (self.target.has_version(parent) or
2270
parent in copy_set or
2271
not self.source.has_version(parent))
2272
index_memo = self.source._index.get_position(version_id)
2273
copy_queue_records.append((version_id, index_memo))
2274
copy_queue.append((version_id, options, parents))
2275
copy_set.add(version_id)
2277
# data suck the join:
2279
total = len(version_list)
2282
for (version_id, raw_data), \
2283
(version_id2, options, parents) in \
2284
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2286
assert version_id == version_id2, 'logic error, inconsistent results'
2288
pb.update("Joining knit", count, total)
2289
raw_records.append((version_id, options, parents, len(raw_data)))
2290
raw_datum.append(raw_data)
2291
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2293
for version in mismatched_versions:
2294
# FIXME RBC 20060309 is this needed?
2295
n1 = set(self.target.get_parents_with_ghosts(version))
2296
n2 = set(self.source.get_parents_with_ghosts(version))
2297
# write a combined record to our history preserving the current
2298
# parents as first in the list
2299
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2300
self.target.fix_parents(version, new_parents)
2306
InterVersionedFile.register_optimiser(InterKnit)
2309
class WeaveToKnit(InterVersionedFile):
2310
"""Optimised code paths for weave to knit operations."""
2312
_matching_file_from_factory = bzrlib.weave.WeaveFile
2313
_matching_file_to_factory = KnitVersionedFile
2316
def is_compatible(source, target):
2317
"""Be compatible with weaves to knits."""
2319
return (isinstance(source, bzrlib.weave.Weave) and
2320
isinstance(target, KnitVersionedFile))
2321
except AttributeError:
2324
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2325
"""See InterVersionedFile.join."""
2326
assert isinstance(self.source, bzrlib.weave.Weave)
2327
assert isinstance(self.target, KnitVersionedFile)
2329
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2334
pb = ui.ui_factory.nested_progress_bar()
2336
version_ids = list(version_ids)
2338
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2339
this_versions = set(self.target._index.get_versions())
2340
needed_versions = self.source_ancestry - this_versions
2341
cross_check_versions = self.source_ancestry.intersection(this_versions)
2342
mismatched_versions = set()
2343
for version in cross_check_versions:
2344
# scan to include needed parents.
2345
n1 = set(self.target.get_parents_with_ghosts(version))
2346
n2 = set(self.source.get_parents(version))
2347
# if all of n2's parents are in n1, then its fine.
2348
if n2.difference(n1):
2349
# FIXME TEST this check for cycles being introduced works
2350
# the logic is we have a cycle if in our graph we are an
2351
# ancestor of any of the n2 revisions.
2357
parent_ancestors = self.source.get_ancestry(parent)
2358
if version in parent_ancestors:
2359
raise errors.GraphCycleError([parent, version])
2360
# ensure this parent will be available later.
2361
new_parents = n2.difference(n1)
2362
needed_versions.update(new_parents.difference(this_versions))
2363
mismatched_versions.add(version)
2365
if not needed_versions and not mismatched_versions:
2367
full_list = topo_sort(self.source.get_graph())
2369
version_list = [i for i in full_list if (not self.target.has_version(i)
2370
and i in needed_versions)]
2374
total = len(version_list)
2375
for version_id in version_list:
2376
pb.update("Converting to knit", count, total)
2377
parents = self.source.get_parents(version_id)
2378
# check that its will be a consistent copy:
2379
for parent in parents:
2380
# if source has the parent, we must already have it
2381
assert (self.target.has_version(parent))
2382
self.target.add_lines(
2383
version_id, parents, self.source.get_lines(version_id))
2386
for version in mismatched_versions:
2387
# FIXME RBC 20060309 is this needed?
2388
n1 = set(self.target.get_parents_with_ghosts(version))
2389
n2 = set(self.source.get_parents(version))
2390
# write a combined record to our history preserving the current
2391
# parents as first in the list
2392
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2393
self.target.fix_parents(version, new_parents)
2399
InterVersionedFile.register_optimiser(WeaveToKnit)
2402
class KnitSequenceMatcher(difflib.SequenceMatcher):
2403
"""Knit tuned sequence matcher.
2405
This is based on profiling of difflib which indicated some improvements
2406
for our usage pattern.
2409
def find_longest_match(self, alo, ahi, blo, bhi):
2410
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
2412
If isjunk is not defined:
2414
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
2415
alo <= i <= i+k <= ahi
2416
blo <= j <= j+k <= bhi
2417
and for all (i',j',k') meeting those conditions,
2420
and if i == i', j <= j'
2422
In other words, of all maximal matching blocks, return one that
2423
starts earliest in a, and of all those maximal matching blocks that
2424
start earliest in a, return the one that starts earliest in b.
2426
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
2427
>>> s.find_longest_match(0, 5, 0, 9)
2430
If isjunk is defined, first the longest matching block is
2431
determined as above, but with the additional restriction that no
2432
junk element appears in the block. Then that block is extended as
2433
far as possible by matching (only) junk elements on both sides. So
2434
the resulting block never matches on junk except as identical junk
2435
happens to be adjacent to an "interesting" match.
2437
Here's the same example as before, but considering blanks to be
2438
junk. That prevents " abcd" from matching the " abcd" at the tail
2439
end of the second sequence directly. Instead only the "abcd" can
2440
match, and matches the leftmost "abcd" in the second sequence:
2442
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
2443
>>> s.find_longest_match(0, 5, 0, 9)
2446
If no blocks match, return (alo, blo, 0).
2448
>>> s = SequenceMatcher(None, "ab", "c")
2449
>>> s.find_longest_match(0, 2, 0, 1)
2453
# CAUTION: stripping common prefix or suffix would be incorrect.
2457
# Longest matching block is "ab", but if common prefix is
2458
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
2459
# strip, so ends up claiming that ab is changed to acab by
2460
# inserting "ca" in the middle. That's minimal but unintuitive:
2461
# "it's obvious" that someone inserted "ac" at the front.
2462
# Windiff ends up at the same place as diff, but by pairing up
2463
# the unique 'b's and then matching the first two 'a's.
2465
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
2466
besti, bestj, bestsize = alo, blo, 0
2467
# find longest junk-free match
2468
# during an iteration of the loop, j2len[j] = length of longest
2469
# junk-free match ending with a[i-1] and b[j]
2473
for i in xrange(alo, ahi):
2474
# look at all instances of a[i] in b; note that because
2475
# b2j has no junk keys, the loop is skipped if a[i] is junk
2476
j2lenget = j2len.get
2479
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
2480
# following improvement
2481
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
2482
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
2483
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
2485
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
2486
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
2487
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
2499
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
2501
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
2504
# Extend the best by non-junk elements on each end. In particular,
2505
# "popular" non-junk elements aren't in b2j, which greatly speeds
2506
# the inner loop above, but also means "the best" match so far
2507
# doesn't contain any junk *or* popular non-junk elements.
2508
while besti > alo and bestj > blo and \
2509
not isbjunk(b[bestj-1]) and \
2510
a[besti-1] == b[bestj-1]:
2511
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2512
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2513
not isbjunk(b[bestj+bestsize]) and \
2514
a[besti+bestsize] == b[bestj+bestsize]:
2517
# Now that we have a wholly interesting match (albeit possibly
2518
# empty!), we may as well suck up the matching junk on each
2519
# side of it too. Can't think of a good reason not to, and it
2520
# saves post-processing the (possibly considerable) expense of
2521
# figuring out what to do with it. In the case of an empty
2522
# interesting match, this is clearly the right thing to do,
2523
# because no other kind of match is possible in the regions.
2524
while besti > alo and bestj > blo and \
2525
isbjunk(b[bestj-1]) and \
2526
a[besti-1] == b[bestj-1]:
2527
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2528
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2529
isbjunk(b[bestj+bestsize]) and \
2530
a[besti+bestsize] == b[bestj+bestsize]:
2531
bestsize = bestsize + 1
2533
return besti, bestj, bestsize
2536
def annotate_knit(knit, revision_id):
2537
"""Annotate a knit with no cached annotations.
2539
This implementation is for knits with no cached annotations.
2540
It will work for knits with cached annotations, but this is not
2543
ancestry = knit.get_ancestry(revision_id)
2544
fulltext = dict(zip(ancestry, knit.get_line_list(ancestry)))
2546
for candidate in ancestry:
2547
if candidate in annotations:
2549
parents = knit.get_parents(candidate)
2550
if len(parents) == 0:
2552
elif knit._index.get_method(candidate) != 'line-delta':
2555
parent, sha1, noeol, delta = knit.get_delta(candidate)
2556
blocks = KnitContent.get_line_delta_blocks(delta,
2557
fulltext[parents[0]], fulltext[candidate])
2558
annotations[candidate] = list(annotate.reannotate([annotations[p]
2559
for p in parents], fulltext[candidate], candidate, blocks))
2560
return iter(annotations[revision_id])
2564
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2566
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
