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):
859
"""See VersionedFile.add_lines_with_ghosts()."""
860
self._check_add(version_id, lines)
861
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
863
def _add_lines(self, version_id, parents, lines, parent_texts,
864
left_matching_blocks=None):
865
"""See VersionedFile.add_lines."""
866
self._check_add(version_id, lines)
867
self._check_versions_present(parents)
868
return self._add(version_id, lines[:], parents, self.delta,
869
parent_texts, left_matching_blocks)
871
def _check_add(self, version_id, lines):
872
"""check that version_id and lines are safe to add."""
873
assert self.writable, "knit is not opened for write"
874
### FIXME escape. RBC 20060228
875
if contains_whitespace(version_id):
876
raise InvalidRevisionId(version_id, self.filename)
877
self.check_not_reserved_id(version_id)
878
if self.has_version(version_id):
879
raise RevisionAlreadyPresent(version_id, self.filename)
880
self._check_lines_not_unicode(lines)
881
self._check_lines_are_lines(lines)
883
def _add(self, version_id, lines, parents, delta, parent_texts,
884
left_matching_blocks=None):
885
"""Add a set of lines on top of version specified by parents.
887
If delta is true, compress the text as a line-delta against
890
Any versions not present will be converted into ghosts.
892
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
893
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
894
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
895
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
896
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
897
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
898
# +1383 0 8.0370 8.0370 +<len>
899
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
900
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
901
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
902
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
906
if parent_texts is None:
908
for parent in parents:
909
if not self.has_version(parent):
910
ghosts.append(parent)
912
present_parents.append(parent)
914
if delta and not len(present_parents):
917
digest = sha_strings(lines)
918
text_length = sum(map(len, lines))
921
if lines[-1][-1] != '\n':
922
options.append('no-eol')
923
lines[-1] = lines[-1] + '\n'
925
if len(present_parents) and delta:
926
# To speed the extract of texts the delta chain is limited
927
# to a fixed number of deltas. This should minimize both
928
# I/O and the time spend applying deltas.
929
delta = self._check_should_delta(present_parents)
931
assert isinstance(version_id, str)
932
lines = self.factory.make(lines, version_id)
933
if delta or (self.factory.annotated and len(present_parents) > 0):
934
# Merge annotations from parent texts if so is needed.
935
delta_hunks = self._merge_annotations(lines, present_parents,
936
parent_texts, delta, self.factory.annotated,
937
left_matching_blocks)
940
options.append('line-delta')
941
store_lines = self.factory.lower_line_delta(delta_hunks)
943
options.append('fulltext')
944
store_lines = self.factory.lower_fulltext(lines)
946
access_memo = self._data.add_record(version_id, digest, store_lines)
947
self._index.add_version(version_id, options, access_memo, parents)
948
return digest, text_length, lines
950
def check(self, progress_bar=None):
951
"""See VersionedFile.check()."""
953
def _clone_text(self, new_version_id, old_version_id, parents):
954
"""See VersionedFile.clone_text()."""
955
# FIXME RBC 20060228 make fast by only inserting an index with null
957
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
959
def get_lines(self, version_id):
960
"""See VersionedFile.get_lines()."""
961
return self.get_line_list([version_id])[0]
963
def _get_record_map(self, version_ids):
964
"""Produce a dictionary of knit records.
966
The keys are version_ids, the values are tuples of (method, content,
968
method is the way the content should be applied.
969
content is a KnitContent object.
970
digest is the SHA1 digest of this version id after all steps are done
971
next is the build-parent of the version, i.e. the leftmost ancestor.
972
If the method is fulltext, next will be None.
974
position_map = self._get_components_positions(version_ids)
975
# c = component_id, m = method, i_m = index_memo, n = next
976
records = [(c, i_m) for c, (m, i_m, n) in position_map.iteritems()]
978
for component_id, content, digest in \
979
self._data.read_records_iter(records):
980
method, index_memo, next = position_map[component_id]
981
record_map[component_id] = method, content, digest, next
985
def get_text(self, version_id):
986
"""See VersionedFile.get_text"""
987
return self.get_texts([version_id])[0]
989
def get_texts(self, version_ids):
990
return [''.join(l) for l in self.get_line_list(version_ids)]
992
def get_line_list(self, version_ids):
993
"""Return the texts of listed versions as a list of strings."""
994
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
995
for version_id in version_ids:
996
self.check_not_reserved_id(version_id)
997
text_map, content_map = self._get_content_maps(version_ids)
998
return [text_map[v] for v in version_ids]
1000
_get_lf_split_line_list = get_line_list
1002
def _get_content_maps(self, version_ids):
1003
"""Produce maps of text and KnitContents
1005
:return: (text_map, content_map) where text_map contains the texts for
1006
the requested versions and content_map contains the KnitContents.
1007
Both dicts take version_ids as their keys.
1009
for version_id in version_ids:
1010
if not self.has_version(version_id):
1011
raise RevisionNotPresent(version_id, self.filename)
1012
record_map = self._get_record_map(version_ids)
1017
for version_id in version_ids:
1020
while cursor is not None:
1021
method, data, digest, next = record_map[cursor]
1022
components.append((cursor, method, data, digest))
1023
if cursor in content_map:
1028
for component_id, method, data, digest in reversed(components):
1029
if component_id in content_map:
1030
content = content_map[component_id]
1032
if method == 'fulltext':
1033
assert content is None
1034
content = self.factory.parse_fulltext(data, version_id)
1035
elif method == 'line-delta':
1036
delta = self.factory.parse_line_delta(data, version_id)
1037
content = content.copy()
1038
content._lines = self._apply_delta(content._lines,
1040
content_map[component_id] = content
1042
if 'no-eol' in self._index.get_options(version_id):
1043
content = content.copy()
1044
line = content._lines[-1][1].rstrip('\n')
1045
content._lines[-1] = (content._lines[-1][0], line)
1046
final_content[version_id] = content
1048
# digest here is the digest from the last applied component.
1049
text = content.text()
1050
if sha_strings(text) != digest:
1051
raise KnitCorrupt(self.filename,
1052
'sha-1 does not match %s' % version_id)
1054
text_map[version_id] = text
1055
return text_map, final_content
1057
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1059
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1060
if version_ids is None:
1061
version_ids = self.versions()
1063
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
1065
pb = progress.DummyProgress()
1066
# we don't care about inclusions, the caller cares.
1067
# but we need to setup a list of records to visit.
1068
# we need version_id, position, length
1069
version_id_records = []
1070
requested_versions = set(version_ids)
1071
# filter for available versions
1072
for version_id in requested_versions:
1073
if not self.has_version(version_id):
1074
raise RevisionNotPresent(version_id, self.filename)
1075
# get a in-component-order queue:
1076
for version_id in self.versions():
1077
if version_id in requested_versions:
1078
index_memo = self._index.get_position(version_id)
1079
version_id_records.append((version_id, index_memo))
1081
total = len(version_id_records)
1082
for version_idx, (version_id, data, sha_value) in \
1083
enumerate(self._data.read_records_iter(version_id_records)):
1084
pb.update('Walking content.', version_idx, total)
1085
method = self._index.get_method(version_id)
1087
assert method in ('fulltext', 'line-delta')
1088
if method == 'fulltext':
1089
line_iterator = self.factory.get_fulltext_content(data)
1091
line_iterator = self.factory.get_linedelta_content(data)
1092
for line in line_iterator:
1095
pb.update('Walking content.', total, total)
1097
def iter_parents(self, version_ids):
1098
"""Iterate through the parents for many version ids.
1100
:param version_ids: An iterable yielding version_ids.
1101
:return: An iterator that yields (version_id, parents). Requested
1102
version_ids not present in the versioned file are simply skipped.
1103
The order is undefined, allowing for different optimisations in
1104
the underlying implementation.
1106
version_ids = [osutils.safe_revision_id(version_id) for
1107
version_id in version_ids]
1108
return self._index.iter_parents(version_ids)
1110
def num_versions(self):
1111
"""See VersionedFile.num_versions()."""
1112
return self._index.num_versions()
1114
__len__ = num_versions
1116
def annotate_iter(self, version_id):
1117
"""See VersionedFile.annotate_iter."""
1118
version_id = osutils.safe_revision_id(version_id)
1119
return self.factory.annotate_iter(self, version_id)
1121
def get_parents(self, version_id):
1122
"""See VersionedFile.get_parents."""
1125
# 52554 calls in 1264 872 internal down from 3674
1126
version_id = osutils.safe_revision_id(version_id)
1128
return self._index.get_parents(version_id)
1130
raise RevisionNotPresent(version_id, self.filename)
1132
def get_parents_with_ghosts(self, version_id):
1133
"""See VersionedFile.get_parents."""
1134
version_id = osutils.safe_revision_id(version_id)
1136
return self._index.get_parents_with_ghosts(version_id)
1138
raise RevisionNotPresent(version_id, self.filename)
1140
def get_ancestry(self, versions, topo_sorted=True):
1141
"""See VersionedFile.get_ancestry."""
1142
if isinstance(versions, basestring):
1143
versions = [versions]
1146
versions = [osutils.safe_revision_id(v) for v in versions]
1147
return self._index.get_ancestry(versions, topo_sorted)
1149
def get_ancestry_with_ghosts(self, versions):
1150
"""See VersionedFile.get_ancestry_with_ghosts."""
1151
if isinstance(versions, basestring):
1152
versions = [versions]
1155
versions = [osutils.safe_revision_id(v) for v in versions]
1156
return self._index.get_ancestry_with_ghosts(versions)
1158
def plan_merge(self, ver_a, ver_b):
1159
"""See VersionedFile.plan_merge."""
1160
ver_a = osutils.safe_revision_id(ver_a)
1161
ver_b = osutils.safe_revision_id(ver_b)
1162
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1164
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1165
annotated_a = self.annotate(ver_a)
1166
annotated_b = self.annotate(ver_b)
1167
return merge._plan_annotate_merge(annotated_a, annotated_b,
1168
ancestors_a, ancestors_b)
1171
class _KnitComponentFile(object):
1172
"""One of the files used to implement a knit database"""
1174
def __init__(self, transport, filename, mode, file_mode=None,
1175
create_parent_dir=False, dir_mode=None):
1176
self._transport = transport
1177
self._filename = filename
1179
self._file_mode = file_mode
1180
self._dir_mode = dir_mode
1181
self._create_parent_dir = create_parent_dir
1182
self._need_to_create = False
1184
def _full_path(self):
1185
"""Return the full path to this file."""
1186
return self._transport.base + self._filename
1188
def check_header(self, fp):
1189
line = fp.readline()
1191
# An empty file can actually be treated as though the file doesn't
1193
raise errors.NoSuchFile(self._full_path())
1194
if line != self.HEADER:
1195
raise KnitHeaderError(badline=line,
1196
filename=self._transport.abspath(self._filename))
1199
return '%s(%s)' % (self.__class__.__name__, self._filename)
1202
class _KnitIndex(_KnitComponentFile):
1203
"""Manages knit index file.
1205
The index is already kept in memory and read on startup, to enable
1206
fast lookups of revision information. The cursor of the index
1207
file is always pointing to the end, making it easy to append
1210
_cache is a cache for fast mapping from version id to a Index
1213
_history is a cache for fast mapping from indexes to version ids.
1215
The index data format is dictionary compressed when it comes to
1216
parent references; a index entry may only have parents that with a
1217
lover index number. As a result, the index is topological sorted.
1219
Duplicate entries may be written to the index for a single version id
1220
if this is done then the latter one completely replaces the former:
1221
this allows updates to correct version and parent information.
1222
Note that the two entries may share the delta, and that successive
1223
annotations and references MUST point to the first entry.
1225
The index file on disc contains a header, followed by one line per knit
1226
record. The same revision can be present in an index file more than once.
1227
The first occurrence gets assigned a sequence number starting from 0.
1229
The format of a single line is
1230
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1231
REVISION_ID is a utf8-encoded revision id
1232
FLAGS is a comma separated list of flags about the record. Values include
1233
no-eol, line-delta, fulltext.
1234
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1235
that the the compressed data starts at.
1236
LENGTH is the ascii representation of the length of the data file.
1237
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1239
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1240
revision id already in the knit that is a parent of REVISION_ID.
1241
The ' :' marker is the end of record marker.
1244
when a write is interrupted to the index file, it will result in a line
1245
that does not end in ' :'. If the ' :' is not present at the end of a line,
1246
or at the end of the file, then the record that is missing it will be
1247
ignored by the parser.
1249
When writing new records to the index file, the data is preceded by '\n'
1250
to ensure that records always start on new lines even if the last write was
1251
interrupted. As a result its normal for the last line in the index to be
1252
missing a trailing newline. One can be added with no harmful effects.
1255
HEADER = "# bzr knit index 8\n"
1257
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1258
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1260
def _cache_version(self, version_id, options, pos, size, parents):
1261
"""Cache a version record in the history array and index cache.
1263
This is inlined into _load_data for performance. KEEP IN SYNC.
1264
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1267
# only want the _history index to reference the 1st index entry
1269
if version_id not in self._cache:
1270
index = len(self._history)
1271
self._history.append(version_id)
1273
index = self._cache[version_id][5]
1274
self._cache[version_id] = (version_id,
1281
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1282
create_parent_dir=False, delay_create=False, dir_mode=None):
1283
_KnitComponentFile.__init__(self, transport, filename, mode,
1284
file_mode=file_mode,
1285
create_parent_dir=create_parent_dir,
1288
# position in _history is the 'official' index for a revision
1289
# but the values may have come from a newer entry.
1290
# so - wc -l of a knit index is != the number of unique names
1294
fp = self._transport.get(self._filename)
1296
# _load_data may raise NoSuchFile if the target knit is
1298
_load_data(self, fp)
1302
if mode != 'w' or not create:
1305
self._need_to_create = True
1307
self._transport.put_bytes_non_atomic(
1308
self._filename, self.HEADER, mode=self._file_mode)
1310
def get_graph(self):
1311
"""Return a list of the node:parents lists from this knit index."""
1312
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1314
def get_ancestry(self, versions, topo_sorted=True):
1315
"""See VersionedFile.get_ancestry."""
1316
# get a graph of all the mentioned versions:
1318
pending = set(versions)
1321
version = pending.pop()
1324
parents = [p for p in cache[version][4] if p in cache]
1326
raise RevisionNotPresent(version, self._filename)
1327
# if not completed and not a ghost
1328
pending.update([p for p in parents if p not in graph])
1329
graph[version] = parents
1332
return topo_sort(graph.items())
1334
def get_ancestry_with_ghosts(self, versions):
1335
"""See VersionedFile.get_ancestry_with_ghosts."""
1336
# get a graph of all the mentioned versions:
1337
self.check_versions_present(versions)
1340
pending = set(versions)
1342
version = pending.pop()
1344
parents = cache[version][4]
1350
pending.update([p for p in parents if p not in graph])
1351
graph[version] = parents
1352
return topo_sort(graph.items())
1354
def iter_parents(self, version_ids):
1355
"""Iterate through the parents for many version ids.
1357
:param version_ids: An iterable yielding version_ids.
1358
:return: An iterator that yields (version_id, parents). Requested
1359
version_ids not present in the versioned file are simply skipped.
1360
The order is undefined, allowing for different optimisations in
1361
the underlying implementation.
1363
for version_id in version_ids:
1365
yield version_id, tuple(self.get_parents(version_id))
1369
def num_versions(self):
1370
return len(self._history)
1372
__len__ = num_versions
1374
def get_versions(self):
1375
"""Get all the versions in the file. not topologically sorted."""
1376
return self._history
1378
def _version_list_to_index(self, versions):
1381
for version in versions:
1382
if version in cache:
1383
# -- inlined lookup() --
1384
result_list.append(str(cache[version][5]))
1385
# -- end lookup () --
1387
result_list.append('.' + version)
1388
return ' '.join(result_list)
1390
def add_version(self, version_id, options, index_memo, parents):
1391
"""Add a version record to the index."""
1392
self.add_versions(((version_id, options, index_memo, parents),))
1394
def add_versions(self, versions):
1395
"""Add multiple versions to the index.
1397
:param versions: a list of tuples:
1398
(version_id, options, pos, size, parents).
1401
orig_history = self._history[:]
1402
orig_cache = self._cache.copy()
1405
for version_id, options, (index, pos, size), parents in versions:
1406
line = "\n%s %s %s %s %s :" % (version_id,
1410
self._version_list_to_index(parents))
1411
assert isinstance(line, str), \
1412
'content must be utf-8 encoded: %r' % (line,)
1414
self._cache_version(version_id, options, pos, size, parents)
1415
if not self._need_to_create:
1416
self._transport.append_bytes(self._filename, ''.join(lines))
1419
sio.write(self.HEADER)
1420
sio.writelines(lines)
1422
self._transport.put_file_non_atomic(self._filename, sio,
1423
create_parent_dir=self._create_parent_dir,
1424
mode=self._file_mode,
1425
dir_mode=self._dir_mode)
1426
self._need_to_create = False
1428
# If any problems happen, restore the original values and re-raise
1429
self._history = orig_history
1430
self._cache = orig_cache
1433
def has_version(self, version_id):
1434
"""True if the version is in the index."""
1435
return version_id in self._cache
1437
def get_position(self, version_id):
1438
"""Return details needed to access the version.
1440
.kndx indices do not support split-out data, so return None for the
1443
:return: a tuple (None, data position, size) to hand to the access
1444
logic to get the record.
1446
entry = self._cache[version_id]
1447
return None, entry[2], entry[3]
1449
def get_method(self, version_id):
1450
"""Return compression method of specified version."""
1451
options = self._cache[version_id][1]
1452
if 'fulltext' in options:
1455
if 'line-delta' not in options:
1456
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1459
def get_options(self, version_id):
1460
"""Return a string represention options.
1464
return self._cache[version_id][1]
1466
def get_parents(self, version_id):
1467
"""Return parents of specified version ignoring ghosts."""
1468
return [parent for parent in self._cache[version_id][4]
1469
if parent in self._cache]
1471
def get_parents_with_ghosts(self, version_id):
1472
"""Return parents of specified version with ghosts."""
1473
return self._cache[version_id][4]
1475
def check_versions_present(self, version_ids):
1476
"""Check that all specified versions are present."""
1478
for version_id in version_ids:
1479
if version_id not in cache:
1480
raise RevisionNotPresent(version_id, self._filename)
1483
class KnitGraphIndex(object):
1484
"""A knit index that builds on GraphIndex."""
1486
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1487
"""Construct a KnitGraphIndex on a graph_index.
1489
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1490
:param deltas: Allow delta-compressed records.
1491
:param add_callback: If not None, allow additions to the index and call
1492
this callback with a list of added GraphIndex nodes:
1493
[(node, value, node_refs), ...]
1494
:param parents: If True, record knits parents, if not do not record
1497
self._graph_index = graph_index
1498
self._deltas = deltas
1499
self._add_callback = add_callback
1500
self._parents = parents
1501
if deltas and not parents:
1502
raise KnitCorrupt(self, "Cannot do delta compression without "
1505
def _get_entries(self, keys, check_present=False):
1506
"""Get the entries for keys.
1508
:param keys: An iterable of index keys, - 1-tuples.
1513
for node in self._graph_index.iter_entries(keys):
1515
found_keys.add(node[1])
1517
# adapt parentless index to the rest of the code.
1518
for node in self._graph_index.iter_entries(keys):
1519
yield node[0], node[1], node[2], ()
1520
found_keys.add(node[1])
1522
missing_keys = keys.difference(found_keys)
1524
raise RevisionNotPresent(missing_keys.pop(), self)
1526
def _present_keys(self, version_ids):
1528
node[1] for node in self._get_entries(version_ids)])
1530
def _parentless_ancestry(self, versions):
1531
"""Honour the get_ancestry API for parentless knit indices."""
1532
wanted_keys = self._version_ids_to_keys(versions)
1533
present_keys = self._present_keys(wanted_keys)
1534
missing = set(wanted_keys).difference(present_keys)
1536
raise RevisionNotPresent(missing.pop(), self)
1537
return list(self._keys_to_version_ids(present_keys))
1539
def get_ancestry(self, versions, topo_sorted=True):
1540
"""See VersionedFile.get_ancestry."""
1541
if not self._parents:
1542
return self._parentless_ancestry(versions)
1543
# XXX: This will do len(history) index calls - perhaps
1544
# it should be altered to be a index core feature?
1545
# get a graph of all the mentioned versions:
1548
versions = self._version_ids_to_keys(versions)
1549
pending = set(versions)
1551
# get all pending nodes
1552
this_iteration = pending
1553
new_nodes = self._get_entries(this_iteration)
1556
for (index, key, value, node_refs) in new_nodes:
1557
# dont ask for ghosties - otherwise
1558
# we we can end up looping with pending
1559
# being entirely ghosted.
1560
graph[key] = [parent for parent in node_refs[0]
1561
if parent not in ghosts]
1563
for parent in graph[key]:
1564
# dont examine known nodes again
1569
ghosts.update(this_iteration.difference(found))
1570
if versions.difference(graph):
1571
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1573
result_keys = topo_sort(graph.items())
1575
result_keys = graph.iterkeys()
1576
return [key[0] for key in result_keys]
1578
def get_ancestry_with_ghosts(self, versions):
1579
"""See VersionedFile.get_ancestry."""
1580
if not self._parents:
1581
return self._parentless_ancestry(versions)
1582
# XXX: This will do len(history) index calls - perhaps
1583
# it should be altered to be a index core feature?
1584
# get a graph of all the mentioned versions:
1586
versions = self._version_ids_to_keys(versions)
1587
pending = set(versions)
1589
# get all pending nodes
1590
this_iteration = pending
1591
new_nodes = self._get_entries(this_iteration)
1593
for (index, key, value, node_refs) in new_nodes:
1594
graph[key] = node_refs[0]
1596
for parent in graph[key]:
1597
# dont examine known nodes again
1601
missing_versions = this_iteration.difference(graph)
1602
missing_needed = versions.intersection(missing_versions)
1604
raise RevisionNotPresent(missing_needed.pop(), self)
1605
for missing_version in missing_versions:
1606
# add a key, no parents
1607
graph[missing_version] = []
1608
pending.discard(missing_version) # don't look for it
1609
result_keys = topo_sort(graph.items())
1610
return [key[0] for key in result_keys]
1612
def get_graph(self):
1613
"""Return a list of the node:parents lists from this knit index."""
1614
if not self._parents:
1615
return [(key, ()) for key in self.get_versions()]
1617
for index, key, value, refs in self._graph_index.iter_all_entries():
1618
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1621
def iter_parents(self, version_ids):
1622
"""Iterate through the parents for many version ids.
1624
:param version_ids: An iterable yielding version_ids.
1625
:return: An iterator that yields (version_id, parents). Requested
1626
version_ids not present in the versioned file are simply skipped.
1627
The order is undefined, allowing for different optimisations in
1628
the underlying implementation.
1631
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1633
present_parents = set()
1634
for node in all_nodes:
1635
all_parents.update(node[3][0])
1636
# any node we are querying must be present
1637
present_parents.add(node[1])
1638
unknown_parents = all_parents.difference(present_parents)
1639
present_parents.update(self._present_keys(unknown_parents))
1640
for node in all_nodes:
1642
for parent in node[3][0]:
1643
if parent in present_parents:
1644
parents.append(parent[0])
1645
yield node[1][0], tuple(parents)
1647
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1648
yield node[1][0], ()
1650
def num_versions(self):
1651
return len(list(self._graph_index.iter_all_entries()))
1653
__len__ = num_versions
1655
def get_versions(self):
1656
"""Get all the versions in the file. not topologically sorted."""
1657
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1659
def has_version(self, version_id):
1660
"""True if the version is in the index."""
1661
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1663
def _keys_to_version_ids(self, keys):
1664
return tuple(key[0] for key in keys)
1666
def get_position(self, version_id):
1667
"""Return details needed to access the version.
1669
:return: a tuple (index, data position, size) to hand to the access
1670
logic to get the record.
1672
node = self._get_node(version_id)
1673
bits = node[2][1:].split(' ')
1674
return node[0], int(bits[0]), int(bits[1])
1676
def get_method(self, version_id):
1677
"""Return compression method of specified version."""
1678
if not self._deltas:
1680
return self._parent_compression(self._get_node(version_id)[3][1])
1682
def _parent_compression(self, reference_list):
1683
# use the second reference list to decide if this is delta'd or not.
1684
if len(reference_list):
1689
def _get_node(self, version_id):
1690
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1692
def get_options(self, version_id):
1693
"""Return a string represention options.
1697
node = self._get_node(version_id)
1698
if not self._deltas:
1699
options = ['fulltext']
1701
options = [self._parent_compression(node[3][1])]
1702
if node[2][0] == 'N':
1703
options.append('no-eol')
1706
def get_parents(self, version_id):
1707
"""Return parents of specified version ignoring ghosts."""
1708
parents = list(self.iter_parents([version_id]))
1711
raise errors.RevisionNotPresent(version_id, self)
1712
return parents[0][1]
1714
def get_parents_with_ghosts(self, version_id):
1715
"""Return parents of specified version with ghosts."""
1716
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1717
check_present=True))
1718
if not self._parents:
1720
return self._keys_to_version_ids(nodes[0][3][0])
1722
def check_versions_present(self, version_ids):
1723
"""Check that all specified versions are present."""
1724
keys = self._version_ids_to_keys(version_ids)
1725
present = self._present_keys(keys)
1726
missing = keys.difference(present)
1728
raise RevisionNotPresent(missing.pop(), self)
1730
def add_version(self, version_id, options, access_memo, parents):
1731
"""Add a version record to the index."""
1732
return self.add_versions(((version_id, options, access_memo, parents),))
1734
def add_versions(self, versions):
1735
"""Add multiple versions to the index.
1737
This function does not insert data into the Immutable GraphIndex
1738
backing the KnitGraphIndex, instead it prepares data for insertion by
1739
the caller and checks that it is safe to insert then calls
1740
self._add_callback with the prepared GraphIndex nodes.
1742
:param versions: a list of tuples:
1743
(version_id, options, pos, size, parents).
1745
if not self._add_callback:
1746
raise errors.ReadOnlyError(self)
1747
# we hope there are no repositories with inconsistent parentage
1752
for (version_id, options, access_memo, parents) in versions:
1753
index, pos, size = access_memo
1754
key = (version_id, )
1755
parents = tuple((parent, ) for parent in parents)
1756
if 'no-eol' in options:
1760
value += "%d %d" % (pos, size)
1761
if not self._deltas:
1762
if 'line-delta' in options:
1763
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1766
if 'line-delta' in options:
1767
node_refs = (parents, (parents[0],))
1769
node_refs = (parents, ())
1771
node_refs = (parents, )
1774
raise KnitCorrupt(self, "attempt to add node with parents "
1775
"in parentless index.")
1777
keys[key] = (value, node_refs)
1778
present_nodes = self._get_entries(keys)
1779
for (index, key, value, node_refs) in present_nodes:
1780
if (value, node_refs) != keys[key]:
1781
raise KnitCorrupt(self, "inconsistent details in add_versions"
1782
": %s %s" % ((value, node_refs), keys[key]))
1786
for key, (value, node_refs) in keys.iteritems():
1787
result.append((key, value, node_refs))
1789
for key, (value, node_refs) in keys.iteritems():
1790
result.append((key, value))
1791
self._add_callback(result)
1793
def _version_ids_to_keys(self, version_ids):
1794
return set((version_id, ) for version_id in version_ids)
1797
class _KnitAccess(object):
1798
"""Access to knit records in a .knit file."""
1800
def __init__(self, transport, filename, _file_mode, _dir_mode,
1801
_need_to_create, _create_parent_dir):
1802
"""Create a _KnitAccess for accessing and inserting data.
1804
:param transport: The transport the .knit is located on.
1805
:param filename: The filename of the .knit.
1807
self._transport = transport
1808
self._filename = filename
1809
self._file_mode = _file_mode
1810
self._dir_mode = _dir_mode
1811
self._need_to_create = _need_to_create
1812
self._create_parent_dir = _create_parent_dir
1814
def add_raw_records(self, sizes, raw_data):
1815
"""Add raw knit bytes to a storage area.
1817
The data is spooled to whereever the access method is storing data.
1819
:param sizes: An iterable containing the size of each raw data segment.
1820
:param raw_data: A bytestring containing the data.
1821
:return: A list of memos to retrieve the record later. Each memo is a
1822
tuple - (index, pos, length), where the index field is always None
1823
for the .knit access method.
1825
assert type(raw_data) == str, \
1826
'data must be plain bytes was %s' % type(raw_data)
1827
if not self._need_to_create:
1828
base = self._transport.append_bytes(self._filename, raw_data)
1830
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1831
create_parent_dir=self._create_parent_dir,
1832
mode=self._file_mode,
1833
dir_mode=self._dir_mode)
1834
self._need_to_create = False
1838
result.append((None, base, size))
1843
"""IFF this data access has its own storage area, initialise it.
1847
self._transport.put_bytes_non_atomic(self._filename, '',
1848
mode=self._file_mode)
1850
def open_file(self):
1851
"""IFF this data access can be represented as a single file, open it.
1853
For knits that are not mapped to a single file on disk this will
1856
:return: None or a file handle.
1859
return self._transport.get(self._filename)
1864
def get_raw_records(self, memos_for_retrieval):
1865
"""Get the raw bytes for a records.
1867
:param memos_for_retrieval: An iterable containing the (index, pos,
1868
length) memo for retrieving the bytes. The .knit method ignores
1869
the index as there is always only a single file.
1870
:return: An iterator over the bytes of the records.
1872
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
1873
for pos, data in self._transport.readv(self._filename, read_vector):
1877
class _PackAccess(object):
1878
"""Access to knit records via a collection of packs."""
1880
def __init__(self, index_to_packs, writer=None):
1881
"""Create a _PackAccess object.
1883
:param index_to_packs: A dict mapping index objects to the transport
1884
and file names for obtaining data.
1885
:param writer: A tuple (pack.ContainerWriter, write_index) which
1886
contains the pack to write, and the index that reads from it will
1890
self.container_writer = writer[0]
1891
self.write_index = writer[1]
1893
self.container_writer = None
1894
self.write_index = None
1895
self.indices = index_to_packs
1897
def add_raw_records(self, sizes, raw_data):
1898
"""Add raw knit bytes to a storage area.
1900
The data is spooled to the container writer in one bytes-record per
1903
:param sizes: An iterable containing the size of each raw data segment.
1904
:param raw_data: A bytestring containing the data.
1905
:return: A list of memos to retrieve the record later. Each memo is a
1906
tuple - (index, pos, length), where the index field is the
1907
write_index object supplied to the PackAccess object.
1909
assert type(raw_data) == str, \
1910
'data must be plain bytes was %s' % type(raw_data)
1914
p_offset, p_length = self.container_writer.add_bytes_record(
1915
raw_data[offset:offset+size], [])
1917
result.append((self.write_index, p_offset, p_length))
1921
"""Pack based knits do not get individually created."""
1923
def get_raw_records(self, memos_for_retrieval):
1924
"""Get the raw bytes for a records.
1926
:param memos_for_retrieval: An iterable containing the (index, pos,
1927
length) memo for retrieving the bytes. The Pack access method
1928
looks up the pack to use for a given record in its index_to_pack
1930
:return: An iterator over the bytes of the records.
1932
# first pass, group into same-index requests
1934
current_index = None
1935
for (index, offset, length) in memos_for_retrieval:
1936
if current_index == index:
1937
current_list.append((offset, length))
1939
if current_index is not None:
1940
request_lists.append((current_index, current_list))
1941
current_index = index
1942
current_list = [(offset, length)]
1943
# handle the last entry
1944
if current_index is not None:
1945
request_lists.append((current_index, current_list))
1946
for index, offsets in request_lists:
1947
transport, path = self.indices[index]
1948
reader = pack.make_readv_reader(transport, path, offsets)
1949
for names, read_func in reader.iter_records():
1950
yield read_func(None)
1952
def open_file(self):
1953
"""Pack based knits have no single file."""
1956
def set_writer(self, writer, index, (transport, packname)):
1957
"""Set a writer to use for adding data."""
1958
self.indices[index] = (transport, packname)
1959
self.container_writer = writer
1960
self.write_index = index
1963
class _KnitData(object):
1964
"""Manage extraction of data from a KnitAccess, caching and decompressing.
1966
The KnitData class provides the logic for parsing and using knit records,
1967
making use of an access method for the low level read and write operations.
1970
def __init__(self, access):
1971
"""Create a KnitData object.
1973
:param access: The access method to use. Access methods such as
1974
_KnitAccess manage the insertion of raw records and the subsequent
1975
retrieval of the same.
1977
self._access = access
1978
self._checked = False
1979
# TODO: jam 20060713 conceptually, this could spill to disk
1980
# if the cached size gets larger than a certain amount
1981
# but it complicates the model a bit, so for now just use
1982
# a simple dictionary
1984
self._do_cache = False
1986
def enable_cache(self):
1987
"""Enable caching of reads."""
1988
self._do_cache = True
1990
def clear_cache(self):
1991
"""Clear the record cache."""
1992
self._do_cache = False
1995
def _open_file(self):
1996
return self._access.open_file()
1998
def _record_to_data(self, version_id, digest, lines):
1999
"""Convert version_id, digest, lines into a raw data block.
2001
:return: (len, a StringIO instance with the raw data ready to read.)
2004
data_file = GzipFile(None, mode='wb', fileobj=sio,
2005
compresslevel=Z_DEFAULT_COMPRESSION)
2007
assert isinstance(version_id, str)
2008
data_file.writelines(chain(
2009
["version %s %d %s\n" % (version_id,
2013
["end %s\n" % version_id]))
2020
def add_raw_records(self, sizes, raw_data):
2021
"""Append a prepared record to the data file.
2023
:param sizes: An iterable containing the size of each raw data segment.
2024
:param raw_data: A bytestring containing the data.
2025
:return: a list of index data for the way the data was stored.
2026
See the access method add_raw_records documentation for more
2029
return self._access.add_raw_records(sizes, raw_data)
2031
def add_record(self, version_id, digest, lines):
2032
"""Write new text record to disk.
2034
Returns index data for retrieving it later, as per add_raw_records.
2036
size, sio = self._record_to_data(version_id, digest, lines)
2037
result = self.add_raw_records([size], sio.getvalue())
2039
self._cache[version_id] = sio.getvalue()
2042
def _parse_record_header(self, version_id, raw_data):
2043
"""Parse a record header for consistency.
2045
:return: the header and the decompressor stream.
2046
as (stream, header_record)
2048
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2050
rec = self._check_header(version_id, df.readline())
2051
except Exception, e:
2052
raise KnitCorrupt(self._access,
2053
"While reading {%s} got %s(%s)"
2054
% (version_id, e.__class__.__name__, str(e)))
2057
def _check_header(self, version_id, line):
2060
raise KnitCorrupt(self._access,
2061
'unexpected number of elements in record header')
2062
if rec[1] != version_id:
2063
raise KnitCorrupt(self._access,
2064
'unexpected version, wanted %r, got %r'
2065
% (version_id, rec[1]))
2068
def _parse_record(self, version_id, data):
2070
# 4168 calls in 2880 217 internal
2071
# 4168 calls to _parse_record_header in 2121
2072
# 4168 calls to readlines in 330
2073
df = GzipFile(mode='rb', fileobj=StringIO(data))
2076
record_contents = df.readlines()
2077
except Exception, e:
2078
raise KnitCorrupt(self._access,
2079
"While reading {%s} got %s(%s)"
2080
% (version_id, e.__class__.__name__, str(e)))
2081
header = record_contents.pop(0)
2082
rec = self._check_header(version_id, header)
2084
last_line = record_contents.pop()
2085
if len(record_contents) != int(rec[2]):
2086
raise KnitCorrupt(self._access,
2087
'incorrect number of lines %s != %s'
2089
% (len(record_contents), int(rec[2]),
2091
if last_line != 'end %s\n' % rec[1]:
2092
raise KnitCorrupt(self._access,
2093
'unexpected version end line %r, wanted %r'
2094
% (last_line, version_id))
2096
return record_contents, rec[3]
2098
def read_records_iter_raw(self, records):
2099
"""Read text records from data file and yield raw data.
2101
This unpacks enough of the text record to validate the id is
2102
as expected but thats all.
2104
# setup an iterator of the external records:
2105
# uses readv so nice and fast we hope.
2107
# grab the disk data needed.
2109
# Don't check _cache if it is empty
2110
needed_offsets = [index_memo for version_id, index_memo
2112
if version_id not in self._cache]
2114
needed_offsets = [index_memo for version_id, index_memo
2117
raw_records = self._access.get_raw_records(needed_offsets)
2119
for version_id, index_memo in records:
2120
if version_id in self._cache:
2121
# This data has already been validated
2122
data = self._cache[version_id]
2124
data = raw_records.next()
2126
self._cache[version_id] = data
2128
# validate the header
2129
df, rec = self._parse_record_header(version_id, data)
2131
yield version_id, data
2133
def read_records_iter(self, records):
2134
"""Read text records from data file and yield result.
2136
The result will be returned in whatever is the fastest to read.
2137
Not by the order requested. Also, multiple requests for the same
2138
record will only yield 1 response.
2139
:param records: A list of (version_id, pos, len) entries
2140
:return: Yields (version_id, contents, digest) in the order
2141
read, not the order requested
2147
# Skip records we have alread seen
2148
yielded_records = set()
2149
needed_records = set()
2150
for record in records:
2151
if record[0] in self._cache:
2152
if record[0] in yielded_records:
2154
yielded_records.add(record[0])
2155
data = self._cache[record[0]]
2156
content, digest = self._parse_record(record[0], data)
2157
yield (record[0], content, digest)
2159
needed_records.add(record)
2160
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2162
needed_records = sorted(set(records), key=operator.itemgetter(1))
2164
if not needed_records:
2167
# The transport optimizes the fetching as well
2168
# (ie, reads continuous ranges.)
2169
raw_data = self._access.get_raw_records(
2170
[index_memo for version_id, index_memo in needed_records])
2172
for (version_id, index_memo), data in \
2173
izip(iter(needed_records), raw_data):
2174
content, digest = self._parse_record(version_id, data)
2176
self._cache[version_id] = data
2177
yield version_id, content, digest
2179
def read_records(self, records):
2180
"""Read records into a dictionary."""
2182
for record_id, content, digest in \
2183
self.read_records_iter(records):
2184
components[record_id] = (content, digest)
2188
class InterKnit(InterVersionedFile):
2189
"""Optimised code paths for knit to knit operations."""
2191
_matching_file_from_factory = KnitVersionedFile
2192
_matching_file_to_factory = KnitVersionedFile
2195
def is_compatible(source, target):
2196
"""Be compatible with knits. """
2198
return (isinstance(source, KnitVersionedFile) and
2199
isinstance(target, KnitVersionedFile))
2200
except AttributeError:
2203
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2204
"""See InterVersionedFile.join."""
2205
assert isinstance(self.source, KnitVersionedFile)
2206
assert isinstance(self.target, KnitVersionedFile)
2208
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2213
pb = ui.ui_factory.nested_progress_bar()
2215
version_ids = list(version_ids)
2216
if None in version_ids:
2217
version_ids.remove(None)
2219
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2220
this_versions = set(self.target._index.get_versions())
2221
needed_versions = self.source_ancestry - this_versions
2222
cross_check_versions = self.source_ancestry.intersection(this_versions)
2223
mismatched_versions = set()
2224
for version in cross_check_versions:
2225
# scan to include needed parents.
2226
n1 = set(self.target.get_parents_with_ghosts(version))
2227
n2 = set(self.source.get_parents_with_ghosts(version))
2229
# FIXME TEST this check for cycles being introduced works
2230
# the logic is we have a cycle if in our graph we are an
2231
# ancestor of any of the n2 revisions.
2237
parent_ancestors = self.source.get_ancestry(parent)
2238
if version in parent_ancestors:
2239
raise errors.GraphCycleError([parent, version])
2240
# ensure this parent will be available later.
2241
new_parents = n2.difference(n1)
2242
needed_versions.update(new_parents.difference(this_versions))
2243
mismatched_versions.add(version)
2245
if not needed_versions and not mismatched_versions:
2247
full_list = topo_sort(self.source.get_graph())
2249
version_list = [i for i in full_list if (not self.target.has_version(i)
2250
and i in needed_versions)]
2254
copy_queue_records = []
2256
for version_id in version_list:
2257
options = self.source._index.get_options(version_id)
2258
parents = self.source._index.get_parents_with_ghosts(version_id)
2259
# check that its will be a consistent copy:
2260
for parent in parents:
2261
# if source has the parent, we must :
2262
# * already have it or
2263
# * have it scheduled already
2264
# otherwise we don't care
2265
assert (self.target.has_version(parent) or
2266
parent in copy_set or
2267
not self.source.has_version(parent))
2268
index_memo = self.source._index.get_position(version_id)
2269
copy_queue_records.append((version_id, index_memo))
2270
copy_queue.append((version_id, options, parents))
2271
copy_set.add(version_id)
2273
# data suck the join:
2275
total = len(version_list)
2278
for (version_id, raw_data), \
2279
(version_id2, options, parents) in \
2280
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2282
assert version_id == version_id2, 'logic error, inconsistent results'
2284
pb.update("Joining knit", count, total)
2285
raw_records.append((version_id, options, parents, len(raw_data)))
2286
raw_datum.append(raw_data)
2287
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2289
for version in mismatched_versions:
2290
# FIXME RBC 20060309 is this needed?
2291
n1 = set(self.target.get_parents_with_ghosts(version))
2292
n2 = set(self.source.get_parents_with_ghosts(version))
2293
# write a combined record to our history preserving the current
2294
# parents as first in the list
2295
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2296
self.target.fix_parents(version, new_parents)
2302
InterVersionedFile.register_optimiser(InterKnit)
2305
class WeaveToKnit(InterVersionedFile):
2306
"""Optimised code paths for weave to knit operations."""
2308
_matching_file_from_factory = bzrlib.weave.WeaveFile
2309
_matching_file_to_factory = KnitVersionedFile
2312
def is_compatible(source, target):
2313
"""Be compatible with weaves to knits."""
2315
return (isinstance(source, bzrlib.weave.Weave) and
2316
isinstance(target, KnitVersionedFile))
2317
except AttributeError:
2320
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2321
"""See InterVersionedFile.join."""
2322
assert isinstance(self.source, bzrlib.weave.Weave)
2323
assert isinstance(self.target, KnitVersionedFile)
2325
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2330
pb = ui.ui_factory.nested_progress_bar()
2332
version_ids = list(version_ids)
2334
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2335
this_versions = set(self.target._index.get_versions())
2336
needed_versions = self.source_ancestry - this_versions
2337
cross_check_versions = self.source_ancestry.intersection(this_versions)
2338
mismatched_versions = set()
2339
for version in cross_check_versions:
2340
# scan to include needed parents.
2341
n1 = set(self.target.get_parents_with_ghosts(version))
2342
n2 = set(self.source.get_parents(version))
2343
# if all of n2's parents are in n1, then its fine.
2344
if n2.difference(n1):
2345
# FIXME TEST this check for cycles being introduced works
2346
# the logic is we have a cycle if in our graph we are an
2347
# ancestor of any of the n2 revisions.
2353
parent_ancestors = self.source.get_ancestry(parent)
2354
if version in parent_ancestors:
2355
raise errors.GraphCycleError([parent, version])
2356
# ensure this parent will be available later.
2357
new_parents = n2.difference(n1)
2358
needed_versions.update(new_parents.difference(this_versions))
2359
mismatched_versions.add(version)
2361
if not needed_versions and not mismatched_versions:
2363
full_list = topo_sort(self.source.get_graph())
2365
version_list = [i for i in full_list if (not self.target.has_version(i)
2366
and i in needed_versions)]
2370
total = len(version_list)
2371
for version_id in version_list:
2372
pb.update("Converting to knit", count, total)
2373
parents = self.source.get_parents(version_id)
2374
# check that its will be a consistent copy:
2375
for parent in parents:
2376
# if source has the parent, we must already have it
2377
assert (self.target.has_version(parent))
2378
self.target.add_lines(
2379
version_id, parents, self.source.get_lines(version_id))
2382
for version in mismatched_versions:
2383
# FIXME RBC 20060309 is this needed?
2384
n1 = set(self.target.get_parents_with_ghosts(version))
2385
n2 = set(self.source.get_parents(version))
2386
# write a combined record to our history preserving the current
2387
# parents as first in the list
2388
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2389
self.target.fix_parents(version, new_parents)
2395
InterVersionedFile.register_optimiser(WeaveToKnit)
2398
class KnitSequenceMatcher(difflib.SequenceMatcher):
2399
"""Knit tuned sequence matcher.
2401
This is based on profiling of difflib which indicated some improvements
2402
for our usage pattern.
2405
def find_longest_match(self, alo, ahi, blo, bhi):
2406
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
2408
If isjunk is not defined:
2410
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
2411
alo <= i <= i+k <= ahi
2412
blo <= j <= j+k <= bhi
2413
and for all (i',j',k') meeting those conditions,
2416
and if i == i', j <= j'
2418
In other words, of all maximal matching blocks, return one that
2419
starts earliest in a, and of all those maximal matching blocks that
2420
start earliest in a, return the one that starts earliest in b.
2422
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
2423
>>> s.find_longest_match(0, 5, 0, 9)
2426
If isjunk is defined, first the longest matching block is
2427
determined as above, but with the additional restriction that no
2428
junk element appears in the block. Then that block is extended as
2429
far as possible by matching (only) junk elements on both sides. So
2430
the resulting block never matches on junk except as identical junk
2431
happens to be adjacent to an "interesting" match.
2433
Here's the same example as before, but considering blanks to be
2434
junk. That prevents " abcd" from matching the " abcd" at the tail
2435
end of the second sequence directly. Instead only the "abcd" can
2436
match, and matches the leftmost "abcd" in the second sequence:
2438
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
2439
>>> s.find_longest_match(0, 5, 0, 9)
2442
If no blocks match, return (alo, blo, 0).
2444
>>> s = SequenceMatcher(None, "ab", "c")
2445
>>> s.find_longest_match(0, 2, 0, 1)
2449
# CAUTION: stripping common prefix or suffix would be incorrect.
2453
# Longest matching block is "ab", but if common prefix is
2454
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
2455
# strip, so ends up claiming that ab is changed to acab by
2456
# inserting "ca" in the middle. That's minimal but unintuitive:
2457
# "it's obvious" that someone inserted "ac" at the front.
2458
# Windiff ends up at the same place as diff, but by pairing up
2459
# the unique 'b's and then matching the first two 'a's.
2461
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
2462
besti, bestj, bestsize = alo, blo, 0
2463
# find longest junk-free match
2464
# during an iteration of the loop, j2len[j] = length of longest
2465
# junk-free match ending with a[i-1] and b[j]
2469
for i in xrange(alo, ahi):
2470
# look at all instances of a[i] in b; note that because
2471
# b2j has no junk keys, the loop is skipped if a[i] is junk
2472
j2lenget = j2len.get
2475
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
2476
# following improvement
2477
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
2478
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
2479
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
2481
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
2482
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
2483
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
2495
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
2497
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
2500
# Extend the best by non-junk elements on each end. In particular,
2501
# "popular" non-junk elements aren't in b2j, which greatly speeds
2502
# the inner loop above, but also means "the best" match so far
2503
# doesn't contain any junk *or* popular non-junk elements.
2504
while besti > alo and bestj > blo and \
2505
not isbjunk(b[bestj-1]) and \
2506
a[besti-1] == b[bestj-1]:
2507
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2508
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2509
not isbjunk(b[bestj+bestsize]) and \
2510
a[besti+bestsize] == b[bestj+bestsize]:
2513
# Now that we have a wholly interesting match (albeit possibly
2514
# empty!), we may as well suck up the matching junk on each
2515
# side of it too. Can't think of a good reason not to, and it
2516
# saves post-processing the (possibly considerable) expense of
2517
# figuring out what to do with it. In the case of an empty
2518
# interesting match, this is clearly the right thing to do,
2519
# because no other kind of match is possible in the regions.
2520
while besti > alo and bestj > blo and \
2521
isbjunk(b[bestj-1]) and \
2522
a[besti-1] == b[bestj-1]:
2523
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2524
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2525
isbjunk(b[bestj+bestsize]) and \
2526
a[besti+bestsize] == b[bestj+bestsize]:
2527
bestsize = bestsize + 1
2529
return besti, bestj, bestsize
2532
def annotate_knit(knit, revision_id):
2533
"""Annotate a knit with no cached annotations.
2535
This implementation is for knits with no cached annotations.
2536
It will work for knits with cached annotations, but this is not
2539
ancestry = knit.get_ancestry(revision_id)
2540
fulltext = dict(zip(ancestry, knit.get_line_list(ancestry)))
2542
for candidate in ancestry:
2543
if candidate in annotations:
2545
parents = knit.get_parents(candidate)
2546
if len(parents) == 0:
2548
elif knit._index.get_method(candidate) != 'line-delta':
2551
parent, sha1, noeol, delta = knit.get_delta(candidate)
2552
blocks = KnitContent.get_line_delta_blocks(delta,
2553
fulltext[parents[0]], fulltext[candidate])
2554
annotations[candidate] = list(annotate.reannotate([annotations[p]
2555
for p in parents], fulltext[candidate], candidate, blocks))
2556
return iter(annotations[revision_id])
2560
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2562
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
