1
# Copyright (C) 2005, 2006, 2007 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""Knit versionedfile implementation.
19
A knit is a versioned file implementation that supports efficient append only
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
end-marker; simply "end VERSION"
28
delta can be line or full contents.a
29
... the 8's there are the index number of the annotation.
30
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
8 e.set('executable', 'yes')
36
8 if elt.get('executable') == 'yes':
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
43
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
44
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
45
09:33 < lifeless> right
46
09:33 < jrydberg> lifeless: the position and size is the range in the data file
49
so the index sequence is the dictionary compressed sequence number used
50
in the deltas to provide line annotation
55
# 10:16 < lifeless> make partial index writes safe
56
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
57
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
59
# move sha1 out of the content so that join is faster at verifying parents
60
# record content length ?
64
from cStringIO import StringIO
66
from itertools import izip, chain
73
from bzrlib.lazy_import import lazy_import
74
lazy_import(globals(), """
89
from bzrlib.errors import (
97
RevisionAlreadyPresent,
99
from bzrlib.tuned_gzip import GzipFile
100
from bzrlib.trace import mutter
101
from bzrlib.osutils import (
106
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
107
from bzrlib.tsort import topo_sort
110
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
113
# TODO: Split out code specific to this format into an associated object.
115
# TODO: Can we put in some kind of value to check that the index and data
116
# files belong together?
118
# TODO: accommodate binaries, perhaps by storing a byte count
120
# TODO: function to check whole file
122
# TODO: atomically append data, then measure backwards from the cursor
123
# position after writing to work out where it was located. we may need to
124
# bypass python file buffering.
126
DATA_SUFFIX = '.knit'
127
INDEX_SUFFIX = '.kndx'
130
class KnitContent(object):
131
"""Content of a knit version to which deltas can be applied."""
133
def __init__(self, lines):
136
def annotate_iter(self):
137
"""Yield tuples of (origin, text) for each content line."""
138
return iter(self._lines)
141
"""Return a list of (origin, text) tuples."""
142
return list(self.annotate_iter())
144
def line_delta_iter(self, new_lines):
145
"""Generate line-based delta from this content to new_lines."""
146
new_texts = new_lines.text()
147
old_texts = self.text()
148
s = KnitSequenceMatcher(None, old_texts, new_texts)
149
for tag, i1, i2, j1, j2 in s.get_opcodes():
152
# ofrom, oto, length, data
153
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
155
def line_delta(self, new_lines):
156
return list(self.line_delta_iter(new_lines))
159
return [text for origin, text in self._lines]
162
return KnitContent(self._lines[:])
165
def get_line_delta_blocks(knit_delta, source, target):
166
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
167
target_len = len(target)
170
for s_begin, s_end, t_len, new_text in knit_delta:
171
true_n = s_begin - s_pos
174
# knit deltas do not provide reliable info about whether the
175
# last line of a file matches, due to eol handling.
176
if source[s_pos + n -1] != target[t_pos + n -1]:
179
yield s_pos, t_pos, n
180
t_pos += t_len + true_n
182
n = target_len - t_pos
184
if source[s_pos + n -1] != target[t_pos + n -1]:
187
yield s_pos, t_pos, n
188
yield s_pos + (target_len - t_pos), target_len, 0
191
class _KnitFactory(object):
192
"""Base factory for creating content objects."""
194
def make(self, lines, version_id):
195
num_lines = len(lines)
196
return KnitContent(zip([version_id] * num_lines, lines))
199
class KnitAnnotateFactory(_KnitFactory):
200
"""Factory for creating annotated Content objects."""
204
def parse_fulltext(self, content, version_id):
205
"""Convert fulltext to internal representation
207
fulltext content is of the format
208
revid(utf8) plaintext\n
209
internal representation is of the format:
212
# TODO: jam 20070209 The tests expect this to be returned as tuples,
213
# but the code itself doesn't really depend on that.
214
# Figure out a way to not require the overhead of turning the
215
# list back into tuples.
216
lines = [tuple(line.split(' ', 1)) for line in content]
217
return KnitContent(lines)
219
def parse_line_delta_iter(self, lines):
220
return iter(self.parse_line_delta(lines))
222
def parse_line_delta(self, lines, version_id):
223
"""Convert a line based delta into internal representation.
225
line delta is in the form of:
226
intstart intend intcount
228
revid(utf8) newline\n
229
internal representation is
230
(start, end, count, [1..count tuples (revid, newline)])
237
def cache_and_return(line):
238
origin, text = line.split(' ', 1)
239
return cache.setdefault(origin, origin), text
241
# walk through the lines parsing.
243
start, end, count = [int(n) for n in header.split(',')]
244
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
245
result.append((start, end, count, contents))
248
def get_fulltext_content(self, lines):
249
"""Extract just the content lines from a fulltext."""
250
return (line.split(' ', 1)[1] for line in lines)
252
def get_linedelta_content(self, lines):
253
"""Extract just the content from a line delta.
255
This doesn't return all of the extra information stored in a delta.
256
Only the actual content lines.
261
header = header.split(',')
262
count = int(header[2])
263
for i in xrange(count):
264
origin, text = next().split(' ', 1)
267
def lower_fulltext(self, content):
268
"""convert a fulltext content record into a serializable form.
270
see parse_fulltext which this inverts.
272
# TODO: jam 20070209 We only do the caching thing to make sure that
273
# the origin is a valid utf-8 line, eventually we could remove it
274
return ['%s %s' % (o, t) for o, t in content._lines]
276
def lower_line_delta(self, delta):
277
"""convert a delta into a serializable form.
279
See parse_line_delta which this inverts.
281
# TODO: jam 20070209 We only do the caching thing to make sure that
282
# the origin is a valid utf-8 line, eventually we could remove it
284
for start, end, c, lines in delta:
285
out.append('%d,%d,%d\n' % (start, end, c))
286
out.extend(origin + ' ' + text
287
for origin, text in lines)
291
class KnitPlainFactory(_KnitFactory):
292
"""Factory for creating plain Content objects."""
296
def parse_fulltext(self, content, version_id):
297
"""This parses an unannotated fulltext.
299
Note that this is not a noop - the internal representation
300
has (versionid, line) - its just a constant versionid.
302
return self.make(content, version_id)
304
def parse_line_delta_iter(self, lines, version_id):
306
num_lines = len(lines)
307
while cur < num_lines:
310
start, end, c = [int(n) for n in header.split(',')]
311
yield start, end, c, zip([version_id] * c, lines[cur:cur+c])
314
def parse_line_delta(self, lines, version_id):
315
return list(self.parse_line_delta_iter(lines, version_id))
317
def get_fulltext_content(self, lines):
318
"""Extract just the content lines from a fulltext."""
321
def get_linedelta_content(self, lines):
322
"""Extract just the content from a line delta.
324
This doesn't return all of the extra information stored in a delta.
325
Only the actual content lines.
330
header = header.split(',')
331
count = int(header[2])
332
for i in xrange(count):
335
def lower_fulltext(self, content):
336
return content.text()
338
def lower_line_delta(self, delta):
340
for start, end, c, lines in delta:
341
out.append('%d,%d,%d\n' % (start, end, c))
342
out.extend([text for origin, text in lines])
346
def make_empty_knit(transport, relpath):
347
"""Construct a empty knit at the specified location."""
348
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
351
class KnitVersionedFile(VersionedFile):
352
"""Weave-like structure with faster random access.
354
A knit stores a number of texts and a summary of the relationships
355
between them. Texts are identified by a string version-id. Texts
356
are normally stored and retrieved as a series of lines, but can
357
also be passed as single strings.
359
Lines are stored with the trailing newline (if any) included, to
360
avoid special cases for files with no final newline. Lines are
361
composed of 8-bit characters, not unicode. The combination of
362
these approaches should mean any 'binary' file can be safely
363
stored and retrieved.
366
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
367
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
368
create=False, create_parent_dir=False, delay_create=False,
369
dir_mode=None, index=None, access_method=None):
370
"""Construct a knit at location specified by relpath.
372
:param create: If not True, only open an existing knit.
373
:param create_parent_dir: If True, create the parent directory if
374
creating the file fails. (This is used for stores with
375
hash-prefixes that may not exist yet)
376
:param delay_create: The calling code is aware that the knit won't
377
actually be created until the first data is stored.
378
:param index: An index to use for the knit.
380
if deprecated_passed(basis_knit):
381
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
382
" deprecated as of bzr 0.9.",
383
DeprecationWarning, stacklevel=2)
384
if access_mode is None:
386
super(KnitVersionedFile, self).__init__(access_mode)
387
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
388
self.transport = transport
389
self.filename = relpath
390
self.factory = factory or KnitAnnotateFactory()
391
self.writable = (access_mode == 'w')
394
self._max_delta_chain = 200
397
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
398
access_mode, create=create, file_mode=file_mode,
399
create_parent_dir=create_parent_dir, delay_create=delay_create,
403
if access_method is None:
404
_access = _KnitAccess(transport, relpath + DATA_SUFFIX, file_mode, dir_mode,
405
((create and not len(self)) and delay_create), create_parent_dir)
407
_access = access_method
408
if create and not len(self) and not delay_create:
410
self._data = _KnitData(_access)
413
return '%s(%s)' % (self.__class__.__name__,
414
self.transport.abspath(self.filename))
416
def _check_should_delta(self, first_parents):
417
"""Iterate back through the parent listing, looking for a fulltext.
419
This is used when we want to decide whether to add a delta or a new
420
fulltext. It searches for _max_delta_chain parents. When it finds a
421
fulltext parent, it sees if the total size of the deltas leading up to
422
it is large enough to indicate that we want a new full text anyway.
424
Return True if we should create a new delta, False if we should use a
429
delta_parents = first_parents
430
for count in xrange(self._max_delta_chain):
431
parent = delta_parents[0]
432
method = self._index.get_method(parent)
433
index, pos, size = self._index.get_position(parent)
434
if method == 'fulltext':
438
delta_parents = self._index.get_parents(parent)
440
# We couldn't find a fulltext, so we must create a new one
443
return fulltext_size > delta_size
445
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
446
"""See VersionedFile._add_delta()."""
447
self._check_add(version_id, []) # should we check the lines ?
448
self._check_versions_present(parents)
452
for parent in parents:
453
if not self.has_version(parent):
454
ghosts.append(parent)
456
present_parents.append(parent)
458
if delta_parent is None:
459
# reconstitute as full text.
460
assert len(delta) == 1 or len(delta) == 0
462
assert delta[0][0] == 0
463
assert delta[0][1] == 0, delta[0][1]
464
return super(KnitVersionedFile, self)._add_delta(version_id,
475
options.append('no-eol')
477
if delta_parent is not None:
478
# determine the current delta chain length.
479
# To speed the extract of texts the delta chain is limited
480
# to a fixed number of deltas. This should minimize both
481
# I/O and the time spend applying deltas.
482
# The window was changed to a maximum of 200 deltas, but also added
483
# was a check that the total compressed size of the deltas is
484
# smaller than the compressed size of the fulltext.
485
if not self._check_should_delta([delta_parent]):
486
# We don't want a delta here, just do a normal insertion.
487
return super(KnitVersionedFile, self)._add_delta(version_id,
494
options.append('line-delta')
495
store_lines = self.factory.lower_line_delta(delta)
497
access_memo = self._data.add_record(version_id, digest, store_lines)
498
self._index.add_version(version_id, options, access_memo, parents)
500
def _add_raw_records(self, records, data):
501
"""Add all the records 'records' with data pre-joined in 'data'.
503
:param records: A list of tuples(version_id, options, parents, size).
504
:param data: The data for the records. When it is written, the records
505
are adjusted to have pos pointing into data by the sum of
506
the preceding records sizes.
509
raw_record_sizes = [record[3] for record in records]
510
positions = self._data.add_raw_records(raw_record_sizes, data)
513
for (version_id, options, parents, size), access_memo in zip(
515
index_entries.append((version_id, options, access_memo, parents))
516
if self._data._do_cache:
517
self._data._cache[version_id] = data[offset:offset+size]
519
self._index.add_versions(index_entries)
521
def enable_cache(self):
522
"""Start caching data for this knit"""
523
self._data.enable_cache()
525
def clear_cache(self):
526
"""Clear the data cache only."""
527
self._data.clear_cache()
529
def copy_to(self, name, transport):
530
"""See VersionedFile.copy_to()."""
531
# copy the current index to a temp index to avoid racing with local
533
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
534
self.transport.get(self._index._filename))
536
f = self._data._open_file()
538
transport.put_file(name + DATA_SUFFIX, f)
541
# move the copied index into place
542
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
544
def create_empty(self, name, transport, mode=None):
545
return KnitVersionedFile(name, transport, factory=self.factory,
546
delta=self.delta, create=True)
548
def _fix_parents(self, version_id, new_parents):
549
"""Fix the parents list for version.
551
This is done by appending a new version to the index
552
with identical data except for the parents list.
553
the parents list must be a superset of the current
556
current_values = self._index._cache[version_id]
557
assert set(current_values[4]).difference(set(new_parents)) == set()
558
self._index.add_version(version_id,
560
(None, current_values[2], current_values[3]),
563
def _extract_blocks(self, version_id, source, target):
564
if self._index.get_method(version_id) != 'line-delta':
566
parent, sha1, noeol, delta = self.get_delta(version_id)
567
return KnitContent.get_line_delta_blocks(delta, source, target)
569
def get_delta(self, version_id):
570
"""Get a delta for constructing version from some other version."""
571
version_id = osutils.safe_revision_id(version_id)
572
self.check_not_reserved_id(version_id)
573
if not self.has_version(version_id):
574
raise RevisionNotPresent(version_id, self.filename)
576
parents = self.get_parents(version_id)
581
index_memo = self._index.get_position(version_id)
582
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
583
noeol = 'no-eol' in self._index.get_options(version_id)
584
if 'fulltext' == self._index.get_method(version_id):
585
new_content = self.factory.parse_fulltext(data, version_id)
586
if parent is not None:
587
reference_content = self._get_content(parent)
588
old_texts = reference_content.text()
591
new_texts = new_content.text()
592
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
593
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
595
delta = self.factory.parse_line_delta(data, version_id)
596
return parent, sha1, noeol, delta
598
def get_graph_with_ghosts(self):
599
"""See VersionedFile.get_graph_with_ghosts()."""
600
graph_items = self._index.get_graph()
601
return dict(graph_items)
603
def get_sha1(self, version_id):
604
return self.get_sha1s([version_id])[0]
606
def get_sha1s(self, version_ids):
607
"""See VersionedFile.get_sha1()."""
608
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
609
record_map = self._get_record_map(version_ids)
610
# record entry 2 is the 'digest'.
611
return [record_map[v][2] for v in version_ids]
615
"""See VersionedFile.get_suffixes()."""
616
return [DATA_SUFFIX, INDEX_SUFFIX]
618
def has_ghost(self, version_id):
619
"""True if there is a ghost reference in the file to version_id."""
620
version_id = osutils.safe_revision_id(version_id)
622
if self.has_version(version_id):
624
# optimisable if needed by memoising the _ghosts set.
625
items = self._index.get_graph()
626
for node, parents in items:
627
for parent in parents:
628
if parent not in self._index._cache:
629
if parent == version_id:
634
"""See VersionedFile.versions."""
635
return self._index.get_versions()
637
def has_version(self, version_id):
638
"""See VersionedFile.has_version."""
639
version_id = osutils.safe_revision_id(version_id)
640
return self._index.has_version(version_id)
642
__contains__ = has_version
644
def _merge_annotations(self, content, parents, parent_texts={},
645
delta=None, annotated=None,
646
left_matching_blocks=None):
647
"""Merge annotations for content. This is done by comparing
648
the annotations based on changed to the text.
650
if left_matching_blocks is not None:
651
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
655
for parent_id in parents:
656
merge_content = self._get_content(parent_id, parent_texts)
657
if (parent_id == parents[0] and delta_seq is not None):
660
seq = patiencediff.PatienceSequenceMatcher(
661
None, merge_content.text(), content.text())
662
for i, j, n in seq.get_matching_blocks():
665
# this appears to copy (origin, text) pairs across to the
666
# new content for any line that matches the last-checked
668
content._lines[j:j+n] = merge_content._lines[i:i+n]
670
if delta_seq is None:
671
reference_content = self._get_content(parents[0], parent_texts)
672
new_texts = content.text()
673
old_texts = reference_content.text()
674
delta_seq = patiencediff.PatienceSequenceMatcher(
675
None, old_texts, new_texts)
676
return self._make_line_delta(delta_seq, content)
678
def _make_line_delta(self, delta_seq, new_content):
679
"""Generate a line delta from delta_seq and new_content."""
681
for op in delta_seq.get_opcodes():
684
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
687
def _get_components_positions(self, version_ids):
688
"""Produce a map of position data for the components of versions.
690
This data is intended to be used for retrieving the knit records.
692
A dict of version_id to (method, data_pos, data_size, next) is
694
method is the way referenced data should be applied.
695
data_pos is the position of the data in the knit.
696
data_size is the size of the data in the knit.
697
next is the build-parent of the version, or None for fulltexts.
700
for version_id in version_ids:
703
while cursor is not None and cursor not in component_data:
704
method = self._index.get_method(cursor)
705
if method == 'fulltext':
708
next = self.get_parents(cursor)[0]
709
index_memo = self._index.get_position(cursor)
710
component_data[cursor] = (method, index_memo, next)
712
return component_data
714
def _get_content(self, version_id, parent_texts={}):
715
"""Returns a content object that makes up the specified
717
if not self.has_version(version_id):
718
raise RevisionNotPresent(version_id, self.filename)
720
cached_version = parent_texts.get(version_id, None)
721
if cached_version is not None:
722
return cached_version
724
text_map, contents_map = self._get_content_maps([version_id])
725
return contents_map[version_id]
727
def _check_versions_present(self, version_ids):
728
"""Check that all specified versions are present."""
729
self._index.check_versions_present(version_ids)
731
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
732
"""See VersionedFile.add_lines_with_ghosts()."""
733
self._check_add(version_id, lines)
734
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
736
def _add_lines(self, version_id, parents, lines, parent_texts,
737
left_matching_blocks=None):
738
"""See VersionedFile.add_lines."""
739
self._check_add(version_id, lines)
740
self._check_versions_present(parents)
741
return self._add(version_id, lines[:], parents, self.delta,
742
parent_texts, left_matching_blocks)
744
def _check_add(self, version_id, lines):
745
"""check that version_id and lines are safe to add."""
746
assert self.writable, "knit is not opened for write"
747
### FIXME escape. RBC 20060228
748
if contains_whitespace(version_id):
749
raise InvalidRevisionId(version_id, self.filename)
750
self.check_not_reserved_id(version_id)
751
if self.has_version(version_id):
752
raise RevisionAlreadyPresent(version_id, self.filename)
753
self._check_lines_not_unicode(lines)
754
self._check_lines_are_lines(lines)
756
def _add(self, version_id, lines, parents, delta, parent_texts,
757
left_matching_blocks=None):
758
"""Add a set of lines on top of version specified by parents.
760
If delta is true, compress the text as a line-delta against
763
Any versions not present will be converted into ghosts.
765
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
766
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
767
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
768
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
769
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
770
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
771
# +1383 0 8.0370 8.0370 +<len>
772
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
773
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
774
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
775
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
779
if parent_texts is None:
781
for parent in parents:
782
if not self.has_version(parent):
783
ghosts.append(parent)
785
present_parents.append(parent)
787
if delta and not len(present_parents):
790
digest = sha_strings(lines)
793
if lines[-1][-1] != '\n':
794
options.append('no-eol')
795
lines[-1] = lines[-1] + '\n'
797
if len(present_parents) and delta:
798
# To speed the extract of texts the delta chain is limited
799
# to a fixed number of deltas. This should minimize both
800
# I/O and the time spend applying deltas.
801
delta = self._check_should_delta(present_parents)
803
assert isinstance(version_id, str)
804
lines = self.factory.make(lines, version_id)
805
if delta or (self.factory.annotated and len(present_parents) > 0):
806
# Merge annotations from parent texts if so is needed.
807
delta_hunks = self._merge_annotations(lines, present_parents,
808
parent_texts, delta, self.factory.annotated,
809
left_matching_blocks)
812
options.append('line-delta')
813
store_lines = self.factory.lower_line_delta(delta_hunks)
815
options.append('fulltext')
816
store_lines = self.factory.lower_fulltext(lines)
818
access_memo = self._data.add_record(version_id, digest, store_lines)
819
self._index.add_version(version_id, options, access_memo, parents)
822
def check(self, progress_bar=None):
823
"""See VersionedFile.check()."""
825
def _clone_text(self, new_version_id, old_version_id, parents):
826
"""See VersionedFile.clone_text()."""
827
# FIXME RBC 20060228 make fast by only inserting an index with null
829
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
831
def get_lines(self, version_id):
832
"""See VersionedFile.get_lines()."""
833
return self.get_line_list([version_id])[0]
835
def _get_record_map(self, version_ids):
836
"""Produce a dictionary of knit records.
838
The keys are version_ids, the values are tuples of (method, content,
840
method is the way the content should be applied.
841
content is a KnitContent object.
842
digest is the SHA1 digest of this version id after all steps are done
843
next is the build-parent of the version, i.e. the leftmost ancestor.
844
If the method is fulltext, next will be None.
846
position_map = self._get_components_positions(version_ids)
847
# c = component_id, m = method, i_m = index_memo, n = next
848
records = [(c, i_m) for c, (m, i_m, n) in position_map.iteritems()]
850
for component_id, content, digest in \
851
self._data.read_records_iter(records):
852
method, index_memo, next = position_map[component_id]
853
record_map[component_id] = method, content, digest, next
857
def get_text(self, version_id):
858
"""See VersionedFile.get_text"""
859
return self.get_texts([version_id])[0]
861
def get_texts(self, version_ids):
862
return [''.join(l) for l in self.get_line_list(version_ids)]
864
def get_line_list(self, version_ids):
865
"""Return the texts of listed versions as a list of strings."""
866
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
867
for version_id in version_ids:
868
self.check_not_reserved_id(version_id)
869
text_map, content_map = self._get_content_maps(version_ids)
870
return [text_map[v] for v in version_ids]
872
_get_lf_split_line_list = get_line_list
874
def _get_content_maps(self, version_ids):
875
"""Produce maps of text and KnitContents
877
:return: (text_map, content_map) where text_map contains the texts for
878
the requested versions and content_map contains the KnitContents.
879
Both dicts take version_ids as their keys.
881
for version_id in version_ids:
882
if not self.has_version(version_id):
883
raise RevisionNotPresent(version_id, self.filename)
884
record_map = self._get_record_map(version_ids)
889
for version_id in version_ids:
892
while cursor is not None:
893
method, data, digest, next = record_map[cursor]
894
components.append((cursor, method, data, digest))
895
if cursor in content_map:
900
for component_id, method, data, digest in reversed(components):
901
if component_id in content_map:
902
content = content_map[component_id]
904
if method == 'fulltext':
905
assert content is None
906
content = self.factory.parse_fulltext(data, version_id)
907
elif method == 'line-delta':
908
delta = self.factory.parse_line_delta(data, version_id)
909
content = content.copy()
910
content._lines = self._apply_delta(content._lines,
912
content_map[component_id] = content
914
if 'no-eol' in self._index.get_options(version_id):
915
content = content.copy()
916
line = content._lines[-1][1].rstrip('\n')
917
content._lines[-1] = (content._lines[-1][0], line)
918
final_content[version_id] = content
920
# digest here is the digest from the last applied component.
921
text = content.text()
922
if sha_strings(text) != digest:
923
raise KnitCorrupt(self.filename,
924
'sha-1 does not match %s' % version_id)
926
text_map[version_id] = text
927
return text_map, final_content
929
def iter_lines_added_or_present_in_versions(self, version_ids=None,
931
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
932
if version_ids is None:
933
version_ids = self.versions()
935
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
937
pb = progress.DummyProgress()
938
# we don't care about inclusions, the caller cares.
939
# but we need to setup a list of records to visit.
940
# we need version_id, position, length
941
version_id_records = []
942
requested_versions = set(version_ids)
943
# filter for available versions
944
for version_id in requested_versions:
945
if not self.has_version(version_id):
946
raise RevisionNotPresent(version_id, self.filename)
947
# get a in-component-order queue:
948
for version_id in self.versions():
949
if version_id in requested_versions:
950
index_memo = self._index.get_position(version_id)
951
version_id_records.append((version_id, index_memo))
953
total = len(version_id_records)
954
for version_idx, (version_id, data, sha_value) in \
955
enumerate(self._data.read_records_iter(version_id_records)):
956
pb.update('Walking content.', version_idx, total)
957
method = self._index.get_method(version_id)
959
assert method in ('fulltext', 'line-delta')
960
if method == 'fulltext':
961
line_iterator = self.factory.get_fulltext_content(data)
963
line_iterator = self.factory.get_linedelta_content(data)
964
for line in line_iterator:
967
pb.update('Walking content.', total, total)
969
def iter_parents(self, version_ids):
970
"""Iterate through the parents for many version ids.
972
:param version_ids: An iterable yielding version_ids.
973
:return: An iterator that yields (version_id, parents). Requested
974
version_ids not present in the versioned file are simply skipped.
975
The order is undefined, allowing for different optimisations in
976
the underlying implementation.
978
version_ids = [osutils.safe_revision_id(version_id) for
979
version_id in version_ids]
980
return self._index.iter_parents(version_ids)
982
def num_versions(self):
983
"""See VersionedFile.num_versions()."""
984
return self._index.num_versions()
986
__len__ = num_versions
988
def annotate_iter(self, version_id):
989
"""See VersionedFile.annotate_iter."""
990
version_id = osutils.safe_revision_id(version_id)
991
content = self._get_content(version_id)
992
for origin, text in content.annotate_iter():
995
def get_parents(self, version_id):
996
"""See VersionedFile.get_parents."""
999
# 52554 calls in 1264 872 internal down from 3674
1000
version_id = osutils.safe_revision_id(version_id)
1002
return self._index.get_parents(version_id)
1004
raise RevisionNotPresent(version_id, self.filename)
1006
def get_parents_with_ghosts(self, version_id):
1007
"""See VersionedFile.get_parents."""
1008
version_id = osutils.safe_revision_id(version_id)
1010
return self._index.get_parents_with_ghosts(version_id)
1012
raise RevisionNotPresent(version_id, self.filename)
1014
def get_ancestry(self, versions, topo_sorted=True):
1015
"""See VersionedFile.get_ancestry."""
1016
if isinstance(versions, basestring):
1017
versions = [versions]
1020
versions = [osutils.safe_revision_id(v) for v in versions]
1021
return self._index.get_ancestry(versions, topo_sorted)
1023
def get_ancestry_with_ghosts(self, versions):
1024
"""See VersionedFile.get_ancestry_with_ghosts."""
1025
if isinstance(versions, basestring):
1026
versions = [versions]
1029
versions = [osutils.safe_revision_id(v) for v in versions]
1030
return self._index.get_ancestry_with_ghosts(versions)
1032
def plan_merge(self, ver_a, ver_b):
1033
"""See VersionedFile.plan_merge."""
1034
ver_a = osutils.safe_revision_id(ver_a)
1035
ver_b = osutils.safe_revision_id(ver_b)
1036
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1038
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1039
annotated_a = self.annotate(ver_a)
1040
annotated_b = self.annotate(ver_b)
1041
return merge._plan_annotate_merge(annotated_a, annotated_b,
1042
ancestors_a, ancestors_b)
1045
class _KnitComponentFile(object):
1046
"""One of the files used to implement a knit database"""
1048
def __init__(self, transport, filename, mode, file_mode=None,
1049
create_parent_dir=False, dir_mode=None):
1050
self._transport = transport
1051
self._filename = filename
1053
self._file_mode = file_mode
1054
self._dir_mode = dir_mode
1055
self._create_parent_dir = create_parent_dir
1056
self._need_to_create = False
1058
def _full_path(self):
1059
"""Return the full path to this file."""
1060
return self._transport.base + self._filename
1062
def check_header(self, fp):
1063
line = fp.readline()
1065
# An empty file can actually be treated as though the file doesn't
1067
raise errors.NoSuchFile(self._full_path())
1068
if line != self.HEADER:
1069
raise KnitHeaderError(badline=line,
1070
filename=self._transport.abspath(self._filename))
1073
return '%s(%s)' % (self.__class__.__name__, self._filename)
1076
class _KnitIndex(_KnitComponentFile):
1077
"""Manages knit index file.
1079
The index is already kept in memory and read on startup, to enable
1080
fast lookups of revision information. The cursor of the index
1081
file is always pointing to the end, making it easy to append
1084
_cache is a cache for fast mapping from version id to a Index
1087
_history is a cache for fast mapping from indexes to version ids.
1089
The index data format is dictionary compressed when it comes to
1090
parent references; a index entry may only have parents that with a
1091
lover index number. As a result, the index is topological sorted.
1093
Duplicate entries may be written to the index for a single version id
1094
if this is done then the latter one completely replaces the former:
1095
this allows updates to correct version and parent information.
1096
Note that the two entries may share the delta, and that successive
1097
annotations and references MUST point to the first entry.
1099
The index file on disc contains a header, followed by one line per knit
1100
record. The same revision can be present in an index file more than once.
1101
The first occurrence gets assigned a sequence number starting from 0.
1103
The format of a single line is
1104
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1105
REVISION_ID is a utf8-encoded revision id
1106
FLAGS is a comma separated list of flags about the record. Values include
1107
no-eol, line-delta, fulltext.
1108
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1109
that the the compressed data starts at.
1110
LENGTH is the ascii representation of the length of the data file.
1111
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1113
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1114
revision id already in the knit that is a parent of REVISION_ID.
1115
The ' :' marker is the end of record marker.
1118
when a write is interrupted to the index file, it will result in a line
1119
that does not end in ' :'. If the ' :' is not present at the end of a line,
1120
or at the end of the file, then the record that is missing it will be
1121
ignored by the parser.
1123
When writing new records to the index file, the data is preceded by '\n'
1124
to ensure that records always start on new lines even if the last write was
1125
interrupted. As a result its normal for the last line in the index to be
1126
missing a trailing newline. One can be added with no harmful effects.
1129
HEADER = "# bzr knit index 8\n"
1131
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1132
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1134
def _cache_version(self, version_id, options, pos, size, parents):
1135
"""Cache a version record in the history array and index cache.
1137
This is inlined into _load_data for performance. KEEP IN SYNC.
1138
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1141
# only want the _history index to reference the 1st index entry
1143
if version_id not in self._cache:
1144
index = len(self._history)
1145
self._history.append(version_id)
1147
index = self._cache[version_id][5]
1148
self._cache[version_id] = (version_id,
1155
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1156
create_parent_dir=False, delay_create=False, dir_mode=None):
1157
_KnitComponentFile.__init__(self, transport, filename, mode,
1158
file_mode=file_mode,
1159
create_parent_dir=create_parent_dir,
1162
# position in _history is the 'official' index for a revision
1163
# but the values may have come from a newer entry.
1164
# so - wc -l of a knit index is != the number of unique names
1168
fp = self._transport.get(self._filename)
1170
# _load_data may raise NoSuchFile if the target knit is
1172
_load_data(self, fp)
1176
if mode != 'w' or not create:
1179
self._need_to_create = True
1181
self._transport.put_bytes_non_atomic(
1182
self._filename, self.HEADER, mode=self._file_mode)
1184
def get_graph(self):
1185
"""Return a list of the node:parents lists from this knit index."""
1186
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1188
def get_ancestry(self, versions, topo_sorted=True):
1189
"""See VersionedFile.get_ancestry."""
1190
# get a graph of all the mentioned versions:
1192
pending = set(versions)
1195
version = pending.pop()
1198
parents = [p for p in cache[version][4] if p in cache]
1200
raise RevisionNotPresent(version, self._filename)
1201
# if not completed and not a ghost
1202
pending.update([p for p in parents if p not in graph])
1203
graph[version] = parents
1206
return topo_sort(graph.items())
1208
def get_ancestry_with_ghosts(self, versions):
1209
"""See VersionedFile.get_ancestry_with_ghosts."""
1210
# get a graph of all the mentioned versions:
1211
self.check_versions_present(versions)
1214
pending = set(versions)
1216
version = pending.pop()
1218
parents = cache[version][4]
1224
pending.update([p for p in parents if p not in graph])
1225
graph[version] = parents
1226
return topo_sort(graph.items())
1228
def iter_parents(self, version_ids):
1229
"""Iterate through the parents for many version ids.
1231
:param version_ids: An iterable yielding version_ids.
1232
:return: An iterator that yields (version_id, parents). Requested
1233
version_ids not present in the versioned file are simply skipped.
1234
The order is undefined, allowing for different optimisations in
1235
the underlying implementation.
1237
for version_id in version_ids:
1239
yield version_id, tuple(self.get_parents(version_id))
1243
def num_versions(self):
1244
return len(self._history)
1246
__len__ = num_versions
1248
def get_versions(self):
1249
"""Get all the versions in the file. not topologically sorted."""
1250
return self._history
1252
def _version_list_to_index(self, versions):
1255
for version in versions:
1256
if version in cache:
1257
# -- inlined lookup() --
1258
result_list.append(str(cache[version][5]))
1259
# -- end lookup () --
1261
result_list.append('.' + version)
1262
return ' '.join(result_list)
1264
def add_version(self, version_id, options, index_memo, parents):
1265
"""Add a version record to the index."""
1266
self.add_versions(((version_id, options, index_memo, parents),))
1268
def add_versions(self, versions):
1269
"""Add multiple versions to the index.
1271
:param versions: a list of tuples:
1272
(version_id, options, pos, size, parents).
1275
orig_history = self._history[:]
1276
orig_cache = self._cache.copy()
1279
for version_id, options, (index, pos, size), parents in versions:
1280
line = "\n%s %s %s %s %s :" % (version_id,
1284
self._version_list_to_index(parents))
1285
assert isinstance(line, str), \
1286
'content must be utf-8 encoded: %r' % (line,)
1288
self._cache_version(version_id, options, pos, size, parents)
1289
if not self._need_to_create:
1290
self._transport.append_bytes(self._filename, ''.join(lines))
1293
sio.write(self.HEADER)
1294
sio.writelines(lines)
1296
self._transport.put_file_non_atomic(self._filename, sio,
1297
create_parent_dir=self._create_parent_dir,
1298
mode=self._file_mode,
1299
dir_mode=self._dir_mode)
1300
self._need_to_create = False
1302
# If any problems happen, restore the original values and re-raise
1303
self._history = orig_history
1304
self._cache = orig_cache
1307
def has_version(self, version_id):
1308
"""True if the version is in the index."""
1309
return version_id in self._cache
1311
def get_position(self, version_id):
1312
"""Return details needed to access the version.
1314
.kndx indices do not support split-out data, so return None for the
1317
:return: a tuple (None, data position, size) to hand to the access
1318
logic to get the record.
1320
entry = self._cache[version_id]
1321
return None, entry[2], entry[3]
1323
def get_method(self, version_id):
1324
"""Return compression method of specified version."""
1325
options = self._cache[version_id][1]
1326
if 'fulltext' in options:
1329
if 'line-delta' not in options:
1330
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1333
def get_options(self, version_id):
1334
"""Return a string represention options.
1338
return self._cache[version_id][1]
1340
def get_parents(self, version_id):
1341
"""Return parents of specified version ignoring ghosts."""
1342
return [parent for parent in self._cache[version_id][4]
1343
if parent in self._cache]
1345
def get_parents_with_ghosts(self, version_id):
1346
"""Return parents of specified version with ghosts."""
1347
return self._cache[version_id][4]
1349
def check_versions_present(self, version_ids):
1350
"""Check that all specified versions are present."""
1352
for version_id in version_ids:
1353
if version_id not in cache:
1354
raise RevisionNotPresent(version_id, self._filename)
1357
class KnitGraphIndex(object):
1358
"""A knit index that builds on GraphIndex."""
1360
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1361
"""Construct a KnitGraphIndex on a graph_index.
1363
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1364
:param deltas: Allow delta-compressed records.
1365
:param add_callback: If not None, allow additions to the index and call
1366
this callback with a list of added GraphIndex nodes:
1367
[(node, value, node_refs), ...]
1368
:param parents: If True, record knits parents, if not do not record
1371
self._graph_index = graph_index
1372
self._deltas = deltas
1373
self._add_callback = add_callback
1374
self._parents = parents
1375
if deltas and not parents:
1376
raise KnitCorrupt(self, "Cannot do delta compression without "
1379
def _get_entries(self, keys, check_present=False):
1380
"""Get the entries for keys.
1382
:param keys: An iterable of index keys, - 1-tuples.
1387
for node in self._graph_index.iter_entries(keys):
1389
found_keys.add(node[1])
1391
# adapt parentless index to the rest of the code.
1392
for node in self._graph_index.iter_entries(keys):
1393
yield node[0], node[1], node[2], ()
1394
found_keys.add(node[1])
1396
missing_keys = keys.difference(found_keys)
1398
raise RevisionNotPresent(missing_keys.pop(), self)
1400
def _present_keys(self, version_ids):
1402
node[1] for node in self._get_entries(version_ids)])
1404
def _parentless_ancestry(self, versions):
1405
"""Honour the get_ancestry API for parentless knit indices."""
1406
wanted_keys = self._version_ids_to_keys(versions)
1407
present_keys = self._present_keys(wanted_keys)
1408
missing = set(wanted_keys).difference(present_keys)
1410
raise RevisionNotPresent(missing.pop(), self)
1411
return list(self._keys_to_version_ids(present_keys))
1413
def get_ancestry(self, versions, topo_sorted=True):
1414
"""See VersionedFile.get_ancestry."""
1415
if not self._parents:
1416
return self._parentless_ancestry(versions)
1417
# XXX: This will do len(history) index calls - perhaps
1418
# it should be altered to be a index core feature?
1419
# get a graph of all the mentioned versions:
1422
versions = self._version_ids_to_keys(versions)
1423
pending = set(versions)
1425
# get all pending nodes
1426
this_iteration = pending
1427
new_nodes = self._get_entries(this_iteration)
1430
for (index, key, value, node_refs) in new_nodes:
1431
# dont ask for ghosties - otherwise
1432
# we we can end up looping with pending
1433
# being entirely ghosted.
1434
graph[key] = [parent for parent in node_refs[0]
1435
if parent not in ghosts]
1437
for parent in graph[key]:
1438
# dont examine known nodes again
1443
ghosts.update(this_iteration.difference(found))
1444
if versions.difference(graph):
1445
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1447
result_keys = topo_sort(graph.items())
1449
result_keys = graph.iterkeys()
1450
return [key[0] for key in result_keys]
1452
def get_ancestry_with_ghosts(self, versions):
1453
"""See VersionedFile.get_ancestry."""
1454
if not self._parents:
1455
return self._parentless_ancestry(versions)
1456
# XXX: This will do len(history) index calls - perhaps
1457
# it should be altered to be a index core feature?
1458
# get a graph of all the mentioned versions:
1460
versions = self._version_ids_to_keys(versions)
1461
pending = set(versions)
1463
# get all pending nodes
1464
this_iteration = pending
1465
new_nodes = self._get_entries(this_iteration)
1467
for (index, key, value, node_refs) in new_nodes:
1468
graph[key] = node_refs[0]
1470
for parent in graph[key]:
1471
# dont examine known nodes again
1475
missing_versions = this_iteration.difference(graph)
1476
missing_needed = versions.intersection(missing_versions)
1478
raise RevisionNotPresent(missing_needed.pop(), self)
1479
for missing_version in missing_versions:
1480
# add a key, no parents
1481
graph[missing_version] = []
1482
pending.discard(missing_version) # don't look for it
1483
result_keys = topo_sort(graph.items())
1484
return [key[0] for key in result_keys]
1486
def get_graph(self):
1487
"""Return a list of the node:parents lists from this knit index."""
1488
if not self._parents:
1489
return [(key, ()) for key in self.get_versions()]
1491
for index, key, value, refs in self._graph_index.iter_all_entries():
1492
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1495
def iter_parents(self, version_ids):
1496
"""Iterate through the parents for many version ids.
1498
:param version_ids: An iterable yielding version_ids.
1499
:return: An iterator that yields (version_id, parents). Requested
1500
version_ids not present in the versioned file are simply skipped.
1501
The order is undefined, allowing for different optimisations in
1502
the underlying implementation.
1505
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1507
present_parents = set()
1508
for node in all_nodes:
1509
all_parents.update(node[3][0])
1510
# any node we are querying must be present
1511
present_parents.add(node[1])
1512
unknown_parents = all_parents.difference(present_parents)
1513
present_parents.update(self._present_keys(unknown_parents))
1514
for node in all_nodes:
1516
for parent in node[3][0]:
1517
if parent in present_parents:
1518
parents.append(parent[0])
1519
yield node[1][0], tuple(parents)
1521
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1522
yield node[1][0], ()
1524
def num_versions(self):
1525
return len(list(self._graph_index.iter_all_entries()))
1527
__len__ = num_versions
1529
def get_versions(self):
1530
"""Get all the versions in the file. not topologically sorted."""
1531
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1533
def has_version(self, version_id):
1534
"""True if the version is in the index."""
1535
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1537
def _keys_to_version_ids(self, keys):
1538
return tuple(key[0] for key in keys)
1540
def get_position(self, version_id):
1541
"""Return details needed to access the version.
1543
:return: a tuple (index, data position, size) to hand to the access
1544
logic to get the record.
1546
node = self._get_node(version_id)
1547
bits = node[2][1:].split(' ')
1548
return node[0], int(bits[0]), int(bits[1])
1550
def get_method(self, version_id):
1551
"""Return compression method of specified version."""
1552
if not self._deltas:
1554
return self._parent_compression(self._get_node(version_id)[3][1])
1556
def _parent_compression(self, reference_list):
1557
# use the second reference list to decide if this is delta'd or not.
1558
if len(reference_list):
1563
def _get_node(self, version_id):
1564
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1566
def get_options(self, version_id):
1567
"""Return a string represention options.
1571
node = self._get_node(version_id)
1572
if not self._deltas:
1573
options = ['fulltext']
1575
options = [self._parent_compression(node[3][1])]
1576
if node[2][0] == 'N':
1577
options.append('no-eol')
1580
def get_parents(self, version_id):
1581
"""Return parents of specified version ignoring ghosts."""
1582
parents = list(self.iter_parents([version_id]))
1585
raise errors.RevisionNotPresent(version_id, self)
1586
return parents[0][1]
1588
def get_parents_with_ghosts(self, version_id):
1589
"""Return parents of specified version with ghosts."""
1590
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1591
check_present=True))
1592
if not self._parents:
1594
return self._keys_to_version_ids(nodes[0][3][0])
1596
def check_versions_present(self, version_ids):
1597
"""Check that all specified versions are present."""
1598
keys = self._version_ids_to_keys(version_ids)
1599
present = self._present_keys(keys)
1600
missing = keys.difference(present)
1602
raise RevisionNotPresent(missing.pop(), self)
1604
def add_version(self, version_id, options, access_memo, parents):
1605
"""Add a version record to the index."""
1606
return self.add_versions(((version_id, options, access_memo, parents),))
1608
def add_versions(self, versions):
1609
"""Add multiple versions to the index.
1611
This function does not insert data into the Immutable GraphIndex
1612
backing the KnitGraphIndex, instead it prepares data for insertion by
1613
the caller and checks that it is safe to insert then calls
1614
self._add_callback with the prepared GraphIndex nodes.
1616
:param versions: a list of tuples:
1617
(version_id, options, pos, size, parents).
1619
if not self._add_callback:
1620
raise errors.ReadOnlyError(self)
1621
# we hope there are no repositories with inconsistent parentage
1626
for (version_id, options, access_memo, parents) in versions:
1627
index, pos, size = access_memo
1628
key = (version_id, )
1629
parents = tuple((parent, ) for parent in parents)
1630
if 'no-eol' in options:
1634
value += "%d %d" % (pos, size)
1635
if not self._deltas:
1636
if 'line-delta' in options:
1637
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1640
if 'line-delta' in options:
1641
node_refs = (parents, (parents[0],))
1643
node_refs = (parents, ())
1645
node_refs = (parents, )
1648
raise KnitCorrupt(self, "attempt to add node with parents "
1649
"in parentless index.")
1651
keys[key] = (value, node_refs)
1652
present_nodes = self._get_entries(keys)
1653
for (index, key, value, node_refs) in present_nodes:
1654
if (value, node_refs) != keys[key]:
1655
raise KnitCorrupt(self, "inconsistent details in add_versions"
1656
": %s %s" % ((value, node_refs), keys[key]))
1660
for key, (value, node_refs) in keys.iteritems():
1661
result.append((key, value, node_refs))
1663
for key, (value, node_refs) in keys.iteritems():
1664
result.append((key, value))
1665
self._add_callback(result)
1667
def _version_ids_to_keys(self, version_ids):
1668
return set((version_id, ) for version_id in version_ids)
1671
class _KnitAccess(object):
1672
"""Access to knit records in a .knit file."""
1674
def __init__(self, transport, filename, _file_mode, _dir_mode,
1675
_need_to_create, _create_parent_dir):
1676
"""Create a _KnitAccess for accessing and inserting data.
1678
:param transport: The transport the .knit is located on.
1679
:param filename: The filename of the .knit.
1681
self._transport = transport
1682
self._filename = filename
1683
self._file_mode = _file_mode
1684
self._dir_mode = _dir_mode
1685
self._need_to_create = _need_to_create
1686
self._create_parent_dir = _create_parent_dir
1688
def add_raw_records(self, sizes, raw_data):
1689
"""Add raw knit bytes to a storage area.
1691
The data is spooled to whereever the access method is storing data.
1693
:param sizes: An iterable containing the size of each raw data segment.
1694
:param raw_data: A bytestring containing the data.
1695
:return: A list of memos to retrieve the record later. Each memo is a
1696
tuple - (index, pos, length), where the index field is always None
1697
for the .knit access method.
1699
assert type(raw_data) == str, \
1700
'data must be plain bytes was %s' % type(raw_data)
1701
if not self._need_to_create:
1702
base = self._transport.append_bytes(self._filename, raw_data)
1704
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1705
create_parent_dir=self._create_parent_dir,
1706
mode=self._file_mode,
1707
dir_mode=self._dir_mode)
1708
self._need_to_create = False
1712
result.append((None, base, size))
1717
"""IFF this data access has its own storage area, initialise it.
1721
self._transport.put_bytes_non_atomic(self._filename, '',
1722
mode=self._file_mode)
1724
def open_file(self):
1725
"""IFF this data access can be represented as a single file, open it.
1727
For knits that are not mapped to a single file on disk this will
1730
:return: None or a file handle.
1733
return self._transport.get(self._filename)
1738
def get_raw_records(self, memos_for_retrieval):
1739
"""Get the raw bytes for a records.
1741
:param memos_for_retrieval: An iterable containing the (index, pos,
1742
length) memo for retrieving the bytes. The .knit method ignores
1743
the index as there is always only a single file.
1744
:return: An iterator over the bytes of the records.
1746
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
1747
for pos, data in self._transport.readv(self._filename, read_vector):
1751
class _PackAccess(object):
1752
"""Access to knit records via a collection of packs."""
1754
def __init__(self, index_to_packs, writer=None):
1755
"""Create a _PackAccess object.
1757
:param index_to_packs: A dict mapping index objects to the transport
1758
and file names for obtaining data.
1759
:param writer: A tuple (pack.ContainerWriter, write_index) which
1760
contains the pack to write, and the index that reads from it will
1764
self.container_writer = writer[0]
1765
self.write_index = writer[1]
1767
self.container_writer = None
1768
self.write_index = None
1769
self.indices = index_to_packs
1771
def add_raw_records(self, sizes, raw_data):
1772
"""Add raw knit bytes to a storage area.
1774
The data is spooled to the container writer in one bytes-record per
1777
:param sizes: An iterable containing the size of each raw data segment.
1778
:param raw_data: A bytestring containing the data.
1779
:return: A list of memos to retrieve the record later. Each memo is a
1780
tuple - (index, pos, length), where the index field is the
1781
write_index object supplied to the PackAccess object.
1783
assert type(raw_data) == str, \
1784
'data must be plain bytes was %s' % type(raw_data)
1788
p_offset, p_length = self.container_writer.add_bytes_record(
1789
raw_data[offset:offset+size], [])
1791
result.append((self.write_index, p_offset, p_length))
1795
"""Pack based knits do not get individually created."""
1797
def get_raw_records(self, memos_for_retrieval):
1798
"""Get the raw bytes for a records.
1800
:param memos_for_retrieval: An iterable containing the (index, pos,
1801
length) memo for retrieving the bytes. The Pack access method
1802
looks up the pack to use for a given record in its index_to_pack
1804
:return: An iterator over the bytes of the records.
1806
# first pass, group into same-index requests
1808
current_index = None
1809
for (index, offset, length) in memos_for_retrieval:
1810
if current_index == index:
1811
current_list.append((offset, length))
1813
if current_index is not None:
1814
request_lists.append((current_index, current_list))
1815
current_index = index
1816
current_list = [(offset, length)]
1817
# handle the last entry
1818
if current_index is not None:
1819
request_lists.append((current_index, current_list))
1820
for index, offsets in request_lists:
1821
transport, path = self.indices[index]
1822
reader = pack.make_readv_reader(transport, path, offsets)
1823
for names, read_func in reader.iter_records():
1824
yield read_func(None)
1826
def open_file(self):
1827
"""Pack based knits have no single file."""
1830
def set_writer(self, writer, index, (transport, packname)):
1831
"""Set a writer to use for adding data."""
1832
self.indices[index] = (transport, packname)
1833
self.container_writer = writer
1834
self.write_index = index
1837
class _KnitData(object):
1838
"""Manage extraction of data from a KnitAccess, caching and decompressing.
1840
The KnitData class provides the logic for parsing and using knit records,
1841
making use of an access method for the low level read and write operations.
1844
def __init__(self, access):
1845
"""Create a KnitData object.
1847
:param access: The access method to use. Access methods such as
1848
_KnitAccess manage the insertion of raw records and the subsequent
1849
retrieval of the same.
1851
self._access = access
1852
self._checked = False
1853
# TODO: jam 20060713 conceptually, this could spill to disk
1854
# if the cached size gets larger than a certain amount
1855
# but it complicates the model a bit, so for now just use
1856
# a simple dictionary
1858
self._do_cache = False
1860
def enable_cache(self):
1861
"""Enable caching of reads."""
1862
self._do_cache = True
1864
def clear_cache(self):
1865
"""Clear the record cache."""
1866
self._do_cache = False
1869
def _open_file(self):
1870
return self._access.open_file()
1872
def _record_to_data(self, version_id, digest, lines):
1873
"""Convert version_id, digest, lines into a raw data block.
1875
:return: (len, a StringIO instance with the raw data ready to read.)
1878
data_file = GzipFile(None, mode='wb', fileobj=sio)
1880
assert isinstance(version_id, str)
1881
data_file.writelines(chain(
1882
["version %s %d %s\n" % (version_id,
1886
["end %s\n" % version_id]))
1893
def add_raw_records(self, sizes, raw_data):
1894
"""Append a prepared record to the data file.
1896
:param sizes: An iterable containing the size of each raw data segment.
1897
:param raw_data: A bytestring containing the data.
1898
:return: a list of index data for the way the data was stored.
1899
See the access method add_raw_records documentation for more
1902
return self._access.add_raw_records(sizes, raw_data)
1904
def add_record(self, version_id, digest, lines):
1905
"""Write new text record to disk.
1907
Returns index data for retrieving it later, as per add_raw_records.
1909
size, sio = self._record_to_data(version_id, digest, lines)
1910
result = self.add_raw_records([size], sio.getvalue())
1912
self._cache[version_id] = sio.getvalue()
1915
def _parse_record_header(self, version_id, raw_data):
1916
"""Parse a record header for consistency.
1918
:return: the header and the decompressor stream.
1919
as (stream, header_record)
1921
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1923
rec = self._check_header(version_id, df.readline())
1924
except Exception, e:
1925
raise KnitCorrupt(self._access,
1926
"While reading {%s} got %s(%s)"
1927
% (version_id, e.__class__.__name__, str(e)))
1930
def _check_header(self, version_id, line):
1933
raise KnitCorrupt(self._access,
1934
'unexpected number of elements in record header')
1935
if rec[1] != version_id:
1936
raise KnitCorrupt(self._access,
1937
'unexpected version, wanted %r, got %r'
1938
% (version_id, rec[1]))
1941
def _parse_record(self, version_id, data):
1943
# 4168 calls in 2880 217 internal
1944
# 4168 calls to _parse_record_header in 2121
1945
# 4168 calls to readlines in 330
1946
df = GzipFile(mode='rb', fileobj=StringIO(data))
1949
record_contents = df.readlines()
1950
except Exception, e:
1951
raise KnitCorrupt(self._access,
1952
"While reading {%s} got %s(%s)"
1953
% (version_id, e.__class__.__name__, str(e)))
1954
header = record_contents.pop(0)
1955
rec = self._check_header(version_id, header)
1957
last_line = record_contents.pop()
1958
if len(record_contents) != int(rec[2]):
1959
raise KnitCorrupt(self._access,
1960
'incorrect number of lines %s != %s'
1962
% (len(record_contents), int(rec[2]),
1964
if last_line != 'end %s\n' % rec[1]:
1965
raise KnitCorrupt(self._access,
1966
'unexpected version end line %r, wanted %r'
1967
% (last_line, version_id))
1969
return record_contents, rec[3]
1971
def read_records_iter_raw(self, records):
1972
"""Read text records from data file and yield raw data.
1974
This unpacks enough of the text record to validate the id is
1975
as expected but thats all.
1977
# setup an iterator of the external records:
1978
# uses readv so nice and fast we hope.
1980
# grab the disk data needed.
1982
# Don't check _cache if it is empty
1983
needed_offsets = [index_memo for version_id, index_memo
1985
if version_id not in self._cache]
1987
needed_offsets = [index_memo for version_id, index_memo
1990
raw_records = self._access.get_raw_records(needed_offsets)
1992
for version_id, index_memo in records:
1993
if version_id in self._cache:
1994
# This data has already been validated
1995
data = self._cache[version_id]
1997
data = raw_records.next()
1999
self._cache[version_id] = data
2001
# validate the header
2002
df, rec = self._parse_record_header(version_id, data)
2004
yield version_id, data
2006
def read_records_iter(self, records):
2007
"""Read text records from data file and yield result.
2009
The result will be returned in whatever is the fastest to read.
2010
Not by the order requested. Also, multiple requests for the same
2011
record will only yield 1 response.
2012
:param records: A list of (version_id, pos, len) entries
2013
:return: Yields (version_id, contents, digest) in the order
2014
read, not the order requested
2020
# Skip records we have alread seen
2021
yielded_records = set()
2022
needed_records = set()
2023
for record in records:
2024
if record[0] in self._cache:
2025
if record[0] in yielded_records:
2027
yielded_records.add(record[0])
2028
data = self._cache[record[0]]
2029
content, digest = self._parse_record(record[0], data)
2030
yield (record[0], content, digest)
2032
needed_records.add(record)
2033
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2035
needed_records = sorted(set(records), key=operator.itemgetter(1))
2037
if not needed_records:
2040
# The transport optimizes the fetching as well
2041
# (ie, reads continuous ranges.)
2042
raw_data = self._access.get_raw_records(
2043
[index_memo for version_id, index_memo in needed_records])
2045
for (version_id, index_memo), data in \
2046
izip(iter(needed_records), raw_data):
2047
content, digest = self._parse_record(version_id, data)
2049
self._cache[version_id] = data
2050
yield version_id, content, digest
2052
def read_records(self, records):
2053
"""Read records into a dictionary."""
2055
for record_id, content, digest in \
2056
self.read_records_iter(records):
2057
components[record_id] = (content, digest)
2061
class InterKnit(InterVersionedFile):
2062
"""Optimised code paths for knit to knit operations."""
2064
_matching_file_from_factory = KnitVersionedFile
2065
_matching_file_to_factory = KnitVersionedFile
2068
def is_compatible(source, target):
2069
"""Be compatible with knits. """
2071
return (isinstance(source, KnitVersionedFile) and
2072
isinstance(target, KnitVersionedFile))
2073
except AttributeError:
2076
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2077
"""See InterVersionedFile.join."""
2078
assert isinstance(self.source, KnitVersionedFile)
2079
assert isinstance(self.target, KnitVersionedFile)
2081
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2086
pb = ui.ui_factory.nested_progress_bar()
2088
version_ids = list(version_ids)
2089
if None in version_ids:
2090
version_ids.remove(None)
2092
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2093
this_versions = set(self.target._index.get_versions())
2094
needed_versions = self.source_ancestry - this_versions
2095
cross_check_versions = self.source_ancestry.intersection(this_versions)
2096
mismatched_versions = set()
2097
for version in cross_check_versions:
2098
# scan to include needed parents.
2099
n1 = set(self.target.get_parents_with_ghosts(version))
2100
n2 = set(self.source.get_parents_with_ghosts(version))
2102
# FIXME TEST this check for cycles being introduced works
2103
# the logic is we have a cycle if in our graph we are an
2104
# ancestor of any of the n2 revisions.
2110
parent_ancestors = self.source.get_ancestry(parent)
2111
if version in parent_ancestors:
2112
raise errors.GraphCycleError([parent, version])
2113
# ensure this parent will be available later.
2114
new_parents = n2.difference(n1)
2115
needed_versions.update(new_parents.difference(this_versions))
2116
mismatched_versions.add(version)
2118
if not needed_versions and not mismatched_versions:
2120
full_list = topo_sort(self.source.get_graph())
2122
version_list = [i for i in full_list if (not self.target.has_version(i)
2123
and i in needed_versions)]
2127
copy_queue_records = []
2129
for version_id in version_list:
2130
options = self.source._index.get_options(version_id)
2131
parents = self.source._index.get_parents_with_ghosts(version_id)
2132
# check that its will be a consistent copy:
2133
for parent in parents:
2134
# if source has the parent, we must :
2135
# * already have it or
2136
# * have it scheduled already
2137
# otherwise we don't care
2138
assert (self.target.has_version(parent) or
2139
parent in copy_set or
2140
not self.source.has_version(parent))
2141
index_memo = self.source._index.get_position(version_id)
2142
copy_queue_records.append((version_id, index_memo))
2143
copy_queue.append((version_id, options, parents))
2144
copy_set.add(version_id)
2146
# data suck the join:
2148
total = len(version_list)
2151
for (version_id, raw_data), \
2152
(version_id2, options, parents) in \
2153
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2155
assert version_id == version_id2, 'logic error, inconsistent results'
2157
pb.update("Joining knit", count, total)
2158
raw_records.append((version_id, options, parents, len(raw_data)))
2159
raw_datum.append(raw_data)
2160
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2162
for version in mismatched_versions:
2163
# FIXME RBC 20060309 is this needed?
2164
n1 = set(self.target.get_parents_with_ghosts(version))
2165
n2 = set(self.source.get_parents_with_ghosts(version))
2166
# write a combined record to our history preserving the current
2167
# parents as first in the list
2168
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2169
self.target.fix_parents(version, new_parents)
2175
InterVersionedFile.register_optimiser(InterKnit)
2178
class WeaveToKnit(InterVersionedFile):
2179
"""Optimised code paths for weave to knit operations."""
2181
_matching_file_from_factory = bzrlib.weave.WeaveFile
2182
_matching_file_to_factory = KnitVersionedFile
2185
def is_compatible(source, target):
2186
"""Be compatible with weaves to knits."""
2188
return (isinstance(source, bzrlib.weave.Weave) and
2189
isinstance(target, KnitVersionedFile))
2190
except AttributeError:
2193
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2194
"""See InterVersionedFile.join."""
2195
assert isinstance(self.source, bzrlib.weave.Weave)
2196
assert isinstance(self.target, KnitVersionedFile)
2198
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2203
pb = ui.ui_factory.nested_progress_bar()
2205
version_ids = list(version_ids)
2207
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2208
this_versions = set(self.target._index.get_versions())
2209
needed_versions = self.source_ancestry - this_versions
2210
cross_check_versions = self.source_ancestry.intersection(this_versions)
2211
mismatched_versions = set()
2212
for version in cross_check_versions:
2213
# scan to include needed parents.
2214
n1 = set(self.target.get_parents_with_ghosts(version))
2215
n2 = set(self.source.get_parents(version))
2216
# if all of n2's parents are in n1, then its fine.
2217
if n2.difference(n1):
2218
# FIXME TEST this check for cycles being introduced works
2219
# the logic is we have a cycle if in our graph we are an
2220
# ancestor of any of the n2 revisions.
2226
parent_ancestors = self.source.get_ancestry(parent)
2227
if version in parent_ancestors:
2228
raise errors.GraphCycleError([parent, version])
2229
# ensure this parent will be available later.
2230
new_parents = n2.difference(n1)
2231
needed_versions.update(new_parents.difference(this_versions))
2232
mismatched_versions.add(version)
2234
if not needed_versions and not mismatched_versions:
2236
full_list = topo_sort(self.source.get_graph())
2238
version_list = [i for i in full_list if (not self.target.has_version(i)
2239
and i in needed_versions)]
2243
total = len(version_list)
2244
for version_id in version_list:
2245
pb.update("Converting to knit", count, total)
2246
parents = self.source.get_parents(version_id)
2247
# check that its will be a consistent copy:
2248
for parent in parents:
2249
# if source has the parent, we must already have it
2250
assert (self.target.has_version(parent))
2251
self.target.add_lines(
2252
version_id, parents, self.source.get_lines(version_id))
2255
for version in mismatched_versions:
2256
# FIXME RBC 20060309 is this needed?
2257
n1 = set(self.target.get_parents_with_ghosts(version))
2258
n2 = set(self.source.get_parents(version))
2259
# write a combined record to our history preserving the current
2260
# parents as first in the list
2261
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2262
self.target.fix_parents(version, new_parents)
2268
InterVersionedFile.register_optimiser(WeaveToKnit)
2271
class KnitSequenceMatcher(difflib.SequenceMatcher):
2272
"""Knit tuned sequence matcher.
2274
This is based on profiling of difflib which indicated some improvements
2275
for our usage pattern.
2278
def find_longest_match(self, alo, ahi, blo, bhi):
2279
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
2281
If isjunk is not defined:
2283
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
2284
alo <= i <= i+k <= ahi
2285
blo <= j <= j+k <= bhi
2286
and for all (i',j',k') meeting those conditions,
2289
and if i == i', j <= j'
2291
In other words, of all maximal matching blocks, return one that
2292
starts earliest in a, and of all those maximal matching blocks that
2293
start earliest in a, return the one that starts earliest in b.
2295
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
2296
>>> s.find_longest_match(0, 5, 0, 9)
2299
If isjunk is defined, first the longest matching block is
2300
determined as above, but with the additional restriction that no
2301
junk element appears in the block. Then that block is extended as
2302
far as possible by matching (only) junk elements on both sides. So
2303
the resulting block never matches on junk except as identical junk
2304
happens to be adjacent to an "interesting" match.
2306
Here's the same example as before, but considering blanks to be
2307
junk. That prevents " abcd" from matching the " abcd" at the tail
2308
end of the second sequence directly. Instead only the "abcd" can
2309
match, and matches the leftmost "abcd" in the second sequence:
2311
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
2312
>>> s.find_longest_match(0, 5, 0, 9)
2315
If no blocks match, return (alo, blo, 0).
2317
>>> s = SequenceMatcher(None, "ab", "c")
2318
>>> s.find_longest_match(0, 2, 0, 1)
2322
# CAUTION: stripping common prefix or suffix would be incorrect.
2326
# Longest matching block is "ab", but if common prefix is
2327
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
2328
# strip, so ends up claiming that ab is changed to acab by
2329
# inserting "ca" in the middle. That's minimal but unintuitive:
2330
# "it's obvious" that someone inserted "ac" at the front.
2331
# Windiff ends up at the same place as diff, but by pairing up
2332
# the unique 'b's and then matching the first two 'a's.
2334
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
2335
besti, bestj, bestsize = alo, blo, 0
2336
# find longest junk-free match
2337
# during an iteration of the loop, j2len[j] = length of longest
2338
# junk-free match ending with a[i-1] and b[j]
2342
for i in xrange(alo, ahi):
2343
# look at all instances of a[i] in b; note that because
2344
# b2j has no junk keys, the loop is skipped if a[i] is junk
2345
j2lenget = j2len.get
2348
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
2349
# following improvement
2350
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
2351
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
2352
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
2354
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
2355
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
2356
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
2368
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
2370
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
2373
# Extend the best by non-junk elements on each end. In particular,
2374
# "popular" non-junk elements aren't in b2j, which greatly speeds
2375
# the inner loop above, but also means "the best" match so far
2376
# doesn't contain any junk *or* popular non-junk elements.
2377
while besti > alo and bestj > blo and \
2378
not isbjunk(b[bestj-1]) and \
2379
a[besti-1] == b[bestj-1]:
2380
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2381
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2382
not isbjunk(b[bestj+bestsize]) and \
2383
a[besti+bestsize] == b[bestj+bestsize]:
2386
# Now that we have a wholly interesting match (albeit possibly
2387
# empty!), we may as well suck up the matching junk on each
2388
# side of it too. Can't think of a good reason not to, and it
2389
# saves post-processing the (possibly considerable) expense of
2390
# figuring out what to do with it. In the case of an empty
2391
# interesting match, this is clearly the right thing to do,
2392
# because no other kind of match is possible in the regions.
2393
while besti > alo and bestj > blo and \
2394
isbjunk(b[bestj-1]) and \
2395
a[besti-1] == b[bestj-1]:
2396
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2397
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2398
isbjunk(b[bestj+bestsize]) and \
2399
a[besti+bestsize] == b[bestj+bestsize]:
2400
bestsize = bestsize + 1
2402
return besti, bestj, bestsize
2406
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2408
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
