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
81
from bzrlib.errors import (
89
RevisionAlreadyPresent,
91
from bzrlib.tuned_gzip import GzipFile
92
from bzrlib.trace import mutter
93
from bzrlib.osutils import (
98
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
99
from bzrlib.tsort import topo_sort
102
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
105
# TODO: Split out code specific to this format into an associated object.
107
# TODO: Can we put in some kind of value to check that the index and data
108
# files belong together?
110
# TODO: accommodate binaries, perhaps by storing a byte count
112
# TODO: function to check whole file
114
# TODO: atomically append data, then measure backwards from the cursor
115
# position after writing to work out where it was located. we may need to
116
# bypass python file buffering.
118
DATA_SUFFIX = '.knit'
119
INDEX_SUFFIX = '.kndx'
122
class KnitContent(object):
123
"""Content of a knit version to which deltas can be applied."""
125
def __init__(self, lines):
128
def annotate_iter(self):
129
"""Yield tuples of (origin, text) for each content line."""
130
return iter(self._lines)
133
"""Return a list of (origin, text) tuples."""
134
return list(self.annotate_iter())
136
def line_delta_iter(self, new_lines):
137
"""Generate line-based delta from this content to new_lines."""
138
new_texts = new_lines.text()
139
old_texts = self.text()
140
s = KnitSequenceMatcher(None, old_texts, new_texts)
141
for tag, i1, i2, j1, j2 in s.get_opcodes():
144
# ofrom, oto, length, data
145
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
147
def line_delta(self, new_lines):
148
return list(self.line_delta_iter(new_lines))
151
return [text for origin, text in self._lines]
154
return KnitContent(self._lines[:])
157
class _KnitFactory(object):
158
"""Base factory for creating content objects."""
160
def make(self, lines, version_id):
161
num_lines = len(lines)
162
return KnitContent(zip([version_id] * num_lines, lines))
165
class KnitAnnotateFactory(_KnitFactory):
166
"""Factory for creating annotated Content objects."""
170
def parse_fulltext(self, content, version_id):
171
"""Convert fulltext to internal representation
173
fulltext content is of the format
174
revid(utf8) plaintext\n
175
internal representation is of the format:
178
# TODO: jam 20070209 The tests expect this to be returned as tuples,
179
# but the code itself doesn't really depend on that.
180
# Figure out a way to not require the overhead of turning the
181
# list back into tuples.
182
lines = [tuple(line.split(' ', 1)) for line in content]
183
return KnitContent(lines)
185
def parse_line_delta_iter(self, lines):
186
return iter(self.parse_line_delta(lines))
188
def parse_line_delta(self, lines, version_id):
189
"""Convert a line based delta into internal representation.
191
line delta is in the form of:
192
intstart intend intcount
194
revid(utf8) newline\n
195
internal representation is
196
(start, end, count, [1..count tuples (revid, newline)])
203
def cache_and_return(line):
204
origin, text = line.split(' ', 1)
205
return cache.setdefault(origin, origin), text
207
# walk through the lines parsing.
209
start, end, count = [int(n) for n in header.split(',')]
210
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
211
result.append((start, end, count, contents))
214
def get_fulltext_content(self, lines):
215
"""Extract just the content lines from a fulltext."""
216
return (line.split(' ', 1)[1] for line in lines)
218
def get_linedelta_content(self, lines):
219
"""Extract just the content from a line delta.
221
This doesn't return all of the extra information stored in a delta.
222
Only the actual content lines.
227
header = header.split(',')
228
count = int(header[2])
229
for i in xrange(count):
230
origin, text = next().split(' ', 1)
233
def lower_fulltext(self, content):
234
"""convert a fulltext content record into a serializable form.
236
see parse_fulltext which this inverts.
238
# TODO: jam 20070209 We only do the caching thing to make sure that
239
# the origin is a valid utf-8 line, eventually we could remove it
240
return ['%s %s' % (o, t) for o, t in content._lines]
242
def lower_line_delta(self, delta):
243
"""convert a delta into a serializable form.
245
See parse_line_delta which this inverts.
247
# TODO: jam 20070209 We only do the caching thing to make sure that
248
# the origin is a valid utf-8 line, eventually we could remove it
250
for start, end, c, lines in delta:
251
out.append('%d,%d,%d\n' % (start, end, c))
252
out.extend(origin + ' ' + text
253
for origin, text in lines)
257
class KnitPlainFactory(_KnitFactory):
258
"""Factory for creating plain Content objects."""
262
def parse_fulltext(self, content, version_id):
263
"""This parses an unannotated fulltext.
265
Note that this is not a noop - the internal representation
266
has (versionid, line) - its just a constant versionid.
268
return self.make(content, version_id)
270
def parse_line_delta_iter(self, lines, version_id):
272
num_lines = len(lines)
273
while cur < num_lines:
276
start, end, c = [int(n) for n in header.split(',')]
277
yield start, end, c, zip([version_id] * c, lines[cur:cur+c])
280
def parse_line_delta(self, lines, version_id):
281
return list(self.parse_line_delta_iter(lines, version_id))
283
def get_fulltext_content(self, lines):
284
"""Extract just the content lines from a fulltext."""
287
def get_linedelta_content(self, lines):
288
"""Extract just the content from a line delta.
290
This doesn't return all of the extra information stored in a delta.
291
Only the actual content lines.
296
header = header.split(',')
297
count = int(header[2])
298
for i in xrange(count):
301
def lower_fulltext(self, content):
302
return content.text()
304
def lower_line_delta(self, delta):
306
for start, end, c, lines in delta:
307
out.append('%d,%d,%d\n' % (start, end, c))
308
out.extend([text for origin, text in lines])
312
def make_empty_knit(transport, relpath):
313
"""Construct a empty knit at the specified location."""
314
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
318
class KnitVersionedFile(VersionedFile):
319
"""Weave-like structure with faster random access.
321
A knit stores a number of texts and a summary of the relationships
322
between them. Texts are identified by a string version-id. Texts
323
are normally stored and retrieved as a series of lines, but can
324
also be passed as single strings.
326
Lines are stored with the trailing newline (if any) included, to
327
avoid special cases for files with no final newline. Lines are
328
composed of 8-bit characters, not unicode. The combination of
329
these approaches should mean any 'binary' file can be safely
330
stored and retrieved.
333
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
334
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
335
create=False, create_parent_dir=False, delay_create=False,
336
dir_mode=None, index=None):
337
"""Construct a knit at location specified by relpath.
339
:param create: If not True, only open an existing knit.
340
:param create_parent_dir: If True, create the parent directory if
341
creating the file fails. (This is used for stores with
342
hash-prefixes that may not exist yet)
343
:param delay_create: The calling code is aware that the knit won't
344
actually be created until the first data is stored.
345
:param index: An index to use for the knit.
347
if deprecated_passed(basis_knit):
348
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
349
" deprecated as of bzr 0.9.",
350
DeprecationWarning, stacklevel=2)
351
if access_mode is None:
353
super(KnitVersionedFile, self).__init__(access_mode)
354
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
355
self.transport = transport
356
self.filename = relpath
357
self.factory = factory or KnitAnnotateFactory()
358
self.writable = (access_mode == 'w')
361
self._max_delta_chain = 200
364
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
365
access_mode, create=create, file_mode=file_mode,
366
create_parent_dir=create_parent_dir, delay_create=delay_create,
370
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
371
access_mode, create=create and not len(self), file_mode=file_mode,
372
create_parent_dir=create_parent_dir, delay_create=delay_create,
376
return '%s(%s)' % (self.__class__.__name__,
377
self.transport.abspath(self.filename))
379
def _check_should_delta(self, first_parents):
380
"""Iterate back through the parent listing, looking for a fulltext.
382
This is used when we want to decide whether to add a delta or a new
383
fulltext. It searches for _max_delta_chain parents. When it finds a
384
fulltext parent, it sees if the total size of the deltas leading up to
385
it is large enough to indicate that we want a new full text anyway.
387
Return True if we should create a new delta, False if we should use a
392
delta_parents = first_parents
393
for count in xrange(self._max_delta_chain):
394
parent = delta_parents[0]
395
method = self._index.get_method(parent)
396
pos, size = self._index.get_position(parent)
397
if method == 'fulltext':
401
delta_parents = self._index.get_parents(parent)
403
# We couldn't find a fulltext, so we must create a new one
406
return fulltext_size > delta_size
408
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
409
"""See VersionedFile._add_delta()."""
410
self._check_add(version_id, []) # should we check the lines ?
411
self._check_versions_present(parents)
415
for parent in parents:
416
if not self.has_version(parent):
417
ghosts.append(parent)
419
present_parents.append(parent)
421
if delta_parent is None:
422
# reconstitute as full text.
423
assert len(delta) == 1 or len(delta) == 0
425
assert delta[0][0] == 0
426
assert delta[0][1] == 0, delta[0][1]
427
return super(KnitVersionedFile, self)._add_delta(version_id,
438
options.append('no-eol')
440
if delta_parent is not None:
441
# determine the current delta chain length.
442
# To speed the extract of texts the delta chain is limited
443
# to a fixed number of deltas. This should minimize both
444
# I/O and the time spend applying deltas.
445
# The window was changed to a maximum of 200 deltas, but also added
446
# was a check that the total compressed size of the deltas is
447
# smaller than the compressed size of the fulltext.
448
if not self._check_should_delta([delta_parent]):
449
# We don't want a delta here, just do a normal insertion.
450
return super(KnitVersionedFile, self)._add_delta(version_id,
457
options.append('line-delta')
458
store_lines = self.factory.lower_line_delta(delta)
460
where, size = self._data.add_record(version_id, digest, store_lines)
461
self._index.add_version(version_id, options, where, size, parents)
463
def _add_raw_records(self, records, data):
464
"""Add all the records 'records' with data pre-joined in 'data'.
466
:param records: A list of tuples(version_id, options, parents, size).
467
:param data: The data for the records. When it is written, the records
468
are adjusted to have pos pointing into data by the sum of
469
the preceding records sizes.
472
pos = self._data.add_raw_record(data)
475
for (version_id, options, parents, size) in records:
476
index_entries.append((version_id, options, pos+offset,
478
if self._data._do_cache:
479
self._data._cache[version_id] = data[offset:offset+size]
481
self._index.add_versions(index_entries)
483
def enable_cache(self):
484
"""Start caching data for this knit"""
485
self._data.enable_cache()
487
def clear_cache(self):
488
"""Clear the data cache only."""
489
self._data.clear_cache()
491
def copy_to(self, name, transport):
492
"""See VersionedFile.copy_to()."""
493
# copy the current index to a temp index to avoid racing with local
495
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
496
self.transport.get(self._index._filename))
498
f = self._data._open_file()
500
transport.put_file(name + DATA_SUFFIX, f)
503
# move the copied index into place
504
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
506
def create_empty(self, name, transport, mode=None):
507
return KnitVersionedFile(name, transport, factory=self.factory,
508
delta=self.delta, create=True)
510
def _fix_parents(self, version_id, new_parents):
511
"""Fix the parents list for version.
513
This is done by appending a new version to the index
514
with identical data except for the parents list.
515
the parents list must be a superset of the current
518
current_values = self._index._cache[version_id]
519
assert set(current_values[4]).difference(set(new_parents)) == set()
520
self._index.add_version(version_id,
526
def get_delta(self, version_id):
527
"""Get a delta for constructing version from some other version."""
528
version_id = osutils.safe_revision_id(version_id)
529
self.check_not_reserved_id(version_id)
530
if not self.has_version(version_id):
531
raise RevisionNotPresent(version_id, self.filename)
533
parents = self.get_parents(version_id)
538
data_pos, data_size = self._index.get_position(version_id)
539
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
540
noeol = 'no-eol' in self._index.get_options(version_id)
541
if 'fulltext' == self._index.get_method(version_id):
542
new_content = self.factory.parse_fulltext(data, version_id)
543
if parent is not None:
544
reference_content = self._get_content(parent)
545
old_texts = reference_content.text()
548
new_texts = new_content.text()
549
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
550
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
552
delta = self.factory.parse_line_delta(data, version_id)
553
return parent, sha1, noeol, delta
555
def get_graph_with_ghosts(self):
556
"""See VersionedFile.get_graph_with_ghosts()."""
557
graph_items = self._index.get_graph()
558
return dict(graph_items)
560
def get_sha1(self, version_id):
561
"""See VersionedFile.get_sha1()."""
562
version_id = osutils.safe_revision_id(version_id)
563
record_map = self._get_record_map([version_id])
564
method, content, digest, next = record_map[version_id]
569
"""See VersionedFile.get_suffixes()."""
570
return [DATA_SUFFIX, INDEX_SUFFIX]
572
def has_ghost(self, version_id):
573
"""True if there is a ghost reference in the file to version_id."""
574
version_id = osutils.safe_revision_id(version_id)
576
if self.has_version(version_id):
578
# optimisable if needed by memoising the _ghosts set.
579
items = self._index.get_graph()
580
for node, parents in items:
581
for parent in parents:
582
if parent not in self._index._cache:
583
if parent == version_id:
588
"""See VersionedFile.versions."""
589
return self._index.get_versions()
591
def has_version(self, version_id):
592
"""See VersionedFile.has_version."""
593
version_id = osutils.safe_revision_id(version_id)
594
return self._index.has_version(version_id)
596
__contains__ = has_version
598
def _merge_annotations(self, content, parents, parent_texts={},
599
delta=None, annotated=None):
600
"""Merge annotations for content. This is done by comparing
601
the annotations based on changed to the text.
605
for parent_id in parents:
606
merge_content = self._get_content(parent_id, parent_texts)
607
seq = patiencediff.PatienceSequenceMatcher(
608
None, merge_content.text(), content.text())
609
if delta_seq is None:
610
# setup a delta seq to reuse.
612
for i, j, n in seq.get_matching_blocks():
615
# this appears to copy (origin, text) pairs across to the new
616
# content for any line that matches the last-checked parent.
617
# FIXME: save the sequence control data for delta compression
618
# against the most relevant parent rather than rediffing.
619
content._lines[j:j+n] = merge_content._lines[i:i+n]
622
reference_content = self._get_content(parents[0], parent_texts)
623
new_texts = content.text()
624
old_texts = reference_content.text()
625
delta_seq = patiencediff.PatienceSequenceMatcher(
626
None, old_texts, new_texts)
627
return self._make_line_delta(delta_seq, content)
629
def _make_line_delta(self, delta_seq, new_content):
630
"""Generate a line delta from delta_seq and new_content."""
632
for op in delta_seq.get_opcodes():
635
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
638
def _get_components_positions(self, version_ids):
639
"""Produce a map of position data for the components of versions.
641
This data is intended to be used for retrieving the knit records.
643
A dict of version_id to (method, data_pos, data_size, next) is
645
method is the way referenced data should be applied.
646
data_pos is the position of the data in the knit.
647
data_size is the size of the data in the knit.
648
next is the build-parent of the version, or None for fulltexts.
651
for version_id in version_ids:
654
while cursor is not None and cursor not in component_data:
655
method = self._index.get_method(cursor)
656
if method == 'fulltext':
659
next = self.get_parents(cursor)[0]
660
data_pos, data_size = self._index.get_position(cursor)
661
component_data[cursor] = (method, data_pos, data_size, next)
663
return component_data
665
def _get_content(self, version_id, parent_texts={}):
666
"""Returns a content object that makes up the specified
668
if not self.has_version(version_id):
669
raise RevisionNotPresent(version_id, self.filename)
671
cached_version = parent_texts.get(version_id, None)
672
if cached_version is not None:
673
return cached_version
675
text_map, contents_map = self._get_content_maps([version_id])
676
return contents_map[version_id]
678
def _check_versions_present(self, version_ids):
679
"""Check that all specified versions are present."""
680
self._index.check_versions_present(version_ids)
682
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
683
"""See VersionedFile.add_lines_with_ghosts()."""
684
self._check_add(version_id, lines)
685
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
687
def _add_lines(self, version_id, parents, lines, parent_texts):
688
"""See VersionedFile.add_lines."""
689
self._check_add(version_id, lines)
690
self._check_versions_present(parents)
691
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
693
def _check_add(self, version_id, lines):
694
"""check that version_id and lines are safe to add."""
695
assert self.writable, "knit is not opened for write"
696
### FIXME escape. RBC 20060228
697
if contains_whitespace(version_id):
698
raise InvalidRevisionId(version_id, self.filename)
699
self.check_not_reserved_id(version_id)
700
if self.has_version(version_id):
701
raise RevisionAlreadyPresent(version_id, self.filename)
702
self._check_lines_not_unicode(lines)
703
self._check_lines_are_lines(lines)
705
def _add(self, version_id, lines, parents, delta, parent_texts):
706
"""Add a set of lines on top of version specified by parents.
708
If delta is true, compress the text as a line-delta against
711
Any versions not present will be converted into ghosts.
713
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
714
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
715
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
716
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
717
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
718
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
719
# +1383 0 8.0370 8.0370 +<len>
720
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
721
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
722
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
723
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
727
if parent_texts is None:
729
for parent in parents:
730
if not self.has_version(parent):
731
ghosts.append(parent)
733
present_parents.append(parent)
735
if delta and not len(present_parents):
738
digest = sha_strings(lines)
741
if lines[-1][-1] != '\n':
742
options.append('no-eol')
743
lines[-1] = lines[-1] + '\n'
745
if len(present_parents) and delta:
746
# To speed the extract of texts the delta chain is limited
747
# to a fixed number of deltas. This should minimize both
748
# I/O and the time spend applying deltas.
749
delta = self._check_should_delta(present_parents)
751
assert isinstance(version_id, str)
752
lines = self.factory.make(lines, version_id)
753
if delta or (self.factory.annotated and len(present_parents) > 0):
754
# Merge annotations from parent texts if so is needed.
755
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
756
delta, self.factory.annotated)
759
options.append('line-delta')
760
store_lines = self.factory.lower_line_delta(delta_hunks)
762
options.append('fulltext')
763
store_lines = self.factory.lower_fulltext(lines)
765
where, size = self._data.add_record(version_id, digest, store_lines)
766
self._index.add_version(version_id, options, where, size, parents)
769
def check(self, progress_bar=None):
770
"""See VersionedFile.check()."""
772
def _clone_text(self, new_version_id, old_version_id, parents):
773
"""See VersionedFile.clone_text()."""
774
# FIXME RBC 20060228 make fast by only inserting an index with null
776
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
778
def get_lines(self, version_id):
779
"""See VersionedFile.get_lines()."""
780
return self.get_line_list([version_id])[0]
782
def _get_record_map(self, version_ids):
783
"""Produce a dictionary of knit records.
785
The keys are version_ids, the values are tuples of (method, content,
787
method is the way the content should be applied.
788
content is a KnitContent object.
789
digest is the SHA1 digest of this version id after all steps are done
790
next is the build-parent of the version, i.e. the leftmost ancestor.
791
If the method is fulltext, next will be None.
793
position_map = self._get_components_positions(version_ids)
794
# c = component_id, m = method, p = position, s = size, n = next
795
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
797
for component_id, content, digest in \
798
self._data.read_records_iter(records):
799
method, position, size, next = position_map[component_id]
800
record_map[component_id] = method, content, digest, next
804
def get_text(self, version_id):
805
"""See VersionedFile.get_text"""
806
return self.get_texts([version_id])[0]
808
def get_texts(self, version_ids):
809
return [''.join(l) for l in self.get_line_list(version_ids)]
811
def get_line_list(self, version_ids):
812
"""Return the texts of listed versions as a list of strings."""
813
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
814
for version_id in version_ids:
815
self.check_not_reserved_id(version_id)
816
text_map, content_map = self._get_content_maps(version_ids)
817
return [text_map[v] for v in version_ids]
819
def _get_content_maps(self, version_ids):
820
"""Produce maps of text and KnitContents
822
:return: (text_map, content_map) where text_map contains the texts for
823
the requested versions and content_map contains the KnitContents.
824
Both dicts take version_ids as their keys.
826
for version_id in version_ids:
827
if not self.has_version(version_id):
828
raise RevisionNotPresent(version_id, self.filename)
829
record_map = self._get_record_map(version_ids)
834
for version_id in version_ids:
837
while cursor is not None:
838
method, data, digest, next = record_map[cursor]
839
components.append((cursor, method, data, digest))
840
if cursor in content_map:
845
for component_id, method, data, digest in reversed(components):
846
if component_id in content_map:
847
content = content_map[component_id]
849
if method == 'fulltext':
850
assert content is None
851
content = self.factory.parse_fulltext(data, version_id)
852
elif method == 'line-delta':
853
delta = self.factory.parse_line_delta(data, version_id)
854
content = content.copy()
855
content._lines = self._apply_delta(content._lines,
857
content_map[component_id] = content
859
if 'no-eol' in self._index.get_options(version_id):
860
content = content.copy()
861
line = content._lines[-1][1].rstrip('\n')
862
content._lines[-1] = (content._lines[-1][0], line)
863
final_content[version_id] = content
865
# digest here is the digest from the last applied component.
866
text = content.text()
867
if sha_strings(text) != digest:
868
raise KnitCorrupt(self.filename,
869
'sha-1 does not match %s' % version_id)
871
text_map[version_id] = text
872
return text_map, final_content
874
def iter_lines_added_or_present_in_versions(self, version_ids=None,
876
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
877
if version_ids is None:
878
version_ids = self.versions()
880
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
882
pb = progress.DummyProgress()
883
# we don't care about inclusions, the caller cares.
884
# but we need to setup a list of records to visit.
885
# we need version_id, position, length
886
version_id_records = []
887
requested_versions = set(version_ids)
888
# filter for available versions
889
for version_id in requested_versions:
890
if not self.has_version(version_id):
891
raise RevisionNotPresent(version_id, self.filename)
892
# get a in-component-order queue:
893
for version_id in self.versions():
894
if version_id in requested_versions:
895
data_pos, length = self._index.get_position(version_id)
896
version_id_records.append((version_id, data_pos, length))
898
total = len(version_id_records)
899
for version_idx, (version_id, data, sha_value) in \
900
enumerate(self._data.read_records_iter(version_id_records)):
901
pb.update('Walking content.', version_idx, total)
902
method = self._index.get_method(version_id)
904
assert method in ('fulltext', 'line-delta')
905
if method == 'fulltext':
906
line_iterator = self.factory.get_fulltext_content(data)
908
line_iterator = self.factory.get_linedelta_content(data)
909
for line in line_iterator:
912
pb.update('Walking content.', total, total)
914
def iter_parents(self, version_ids):
915
"""Iterate through the parents for many version ids.
917
:param version_ids: An iterable yielding version_ids.
918
:return: An iterator that yields (version_id, parents). Requested
919
version_ids not present in the versioned file are simply skipped.
920
The order is undefined, allowing for different optimisations in
921
the underlying implementation.
923
version_ids = [osutils.safe_revision_id(version_id) for
924
version_id in version_ids]
925
return self._index.iter_parents(version_ids)
927
def num_versions(self):
928
"""See VersionedFile.num_versions()."""
929
return self._index.num_versions()
931
__len__ = num_versions
933
def annotate_iter(self, version_id):
934
"""See VersionedFile.annotate_iter."""
935
version_id = osutils.safe_revision_id(version_id)
936
content = self._get_content(version_id)
937
for origin, text in content.annotate_iter():
940
def get_parents(self, version_id):
941
"""See VersionedFile.get_parents."""
944
# 52554 calls in 1264 872 internal down from 3674
945
version_id = osutils.safe_revision_id(version_id)
947
return self._index.get_parents(version_id)
949
raise RevisionNotPresent(version_id, self.filename)
951
def get_parents_with_ghosts(self, version_id):
952
"""See VersionedFile.get_parents."""
953
version_id = osutils.safe_revision_id(version_id)
955
return self._index.get_parents_with_ghosts(version_id)
957
raise RevisionNotPresent(version_id, self.filename)
959
def get_ancestry(self, versions, topo_sorted=True):
960
"""See VersionedFile.get_ancestry."""
961
if isinstance(versions, basestring):
962
versions = [versions]
965
versions = [osutils.safe_revision_id(v) for v in versions]
966
return self._index.get_ancestry(versions, topo_sorted)
968
def get_ancestry_with_ghosts(self, versions):
969
"""See VersionedFile.get_ancestry_with_ghosts."""
970
if isinstance(versions, basestring):
971
versions = [versions]
974
versions = [osutils.safe_revision_id(v) for v in versions]
975
return self._index.get_ancestry_with_ghosts(versions)
977
#@deprecated_method(zero_eight)
978
def walk(self, version_ids):
979
"""See VersionedFile.walk."""
980
# We take the short path here, and extract all relevant texts
981
# and put them in a weave and let that do all the work. Far
982
# from optimal, but is much simpler.
983
# FIXME RB 20060228 this really is inefficient!
984
from bzrlib.weave import Weave
986
w = Weave(self.filename)
987
ancestry = set(self.get_ancestry(version_ids, topo_sorted=False))
988
sorted_graph = topo_sort(self._index.get_graph())
989
version_list = [vid for vid in sorted_graph if vid in ancestry]
991
for version_id in version_list:
992
lines = self.get_lines(version_id)
993
w.add_lines(version_id, self.get_parents(version_id), lines)
995
for lineno, insert_id, dset, line in w.walk(version_ids):
996
yield lineno, insert_id, dset, line
998
def plan_merge(self, ver_a, ver_b):
999
"""See VersionedFile.plan_merge."""
1000
ver_a = osutils.safe_revision_id(ver_a)
1001
ver_b = osutils.safe_revision_id(ver_b)
1002
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1003
def status_a(revision, text):
1004
if revision in ancestors_b:
1005
return 'killed-b', text
1007
return 'new-a', text
1009
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1010
def status_b(revision, text):
1011
if revision in ancestors_a:
1012
return 'killed-a', text
1014
return 'new-b', text
1016
annotated_a = self.annotate(ver_a)
1017
annotated_b = self.annotate(ver_b)
1018
plain_a = [t for (a, t) in annotated_a]
1019
plain_b = [t for (a, t) in annotated_b]
1020
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
1023
for ai, bi, l in blocks:
1024
# process all mismatched sections
1025
# (last mismatched section is handled because blocks always
1026
# includes a 0-length last block)
1027
for revision, text in annotated_a[a_cur:ai]:
1028
yield status_a(revision, text)
1029
for revision, text in annotated_b[b_cur:bi]:
1030
yield status_b(revision, text)
1032
# and now the matched section
1035
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
1036
assert text_a == text_b
1037
yield "unchanged", text_a
1040
class _KnitComponentFile(object):
1041
"""One of the files used to implement a knit database"""
1043
def __init__(self, transport, filename, mode, file_mode=None,
1044
create_parent_dir=False, dir_mode=None):
1045
self._transport = transport
1046
self._filename = filename
1048
self._file_mode = file_mode
1049
self._dir_mode = dir_mode
1050
self._create_parent_dir = create_parent_dir
1051
self._need_to_create = False
1053
def _full_path(self):
1054
"""Return the full path to this file."""
1055
return self._transport.base + self._filename
1057
def check_header(self, fp):
1058
line = fp.readline()
1060
# An empty file can actually be treated as though the file doesn't
1062
raise errors.NoSuchFile(self._full_path())
1063
if line != self.HEADER:
1064
raise KnitHeaderError(badline=line,
1065
filename=self._transport.abspath(self._filename))
1068
"""Commit is a nop."""
1071
return '%s(%s)' % (self.__class__.__name__, self._filename)
1074
class _KnitIndex(_KnitComponentFile):
1075
"""Manages knit index file.
1077
The index is already kept in memory and read on startup, to enable
1078
fast lookups of revision information. The cursor of the index
1079
file is always pointing to the end, making it easy to append
1082
_cache is a cache for fast mapping from version id to a Index
1085
_history is a cache for fast mapping from indexes to version ids.
1087
The index data format is dictionary compressed when it comes to
1088
parent references; a index entry may only have parents that with a
1089
lover index number. As a result, the index is topological sorted.
1091
Duplicate entries may be written to the index for a single version id
1092
if this is done then the latter one completely replaces the former:
1093
this allows updates to correct version and parent information.
1094
Note that the two entries may share the delta, and that successive
1095
annotations and references MUST point to the first entry.
1097
The index file on disc contains a header, followed by one line per knit
1098
record. The same revision can be present in an index file more than once.
1099
The first occurrence gets assigned a sequence number starting from 0.
1101
The format of a single line is
1102
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1103
REVISION_ID is a utf8-encoded revision id
1104
FLAGS is a comma separated list of flags about the record. Values include
1105
no-eol, line-delta, fulltext.
1106
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1107
that the the compressed data starts at.
1108
LENGTH is the ascii representation of the length of the data file.
1109
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1111
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1112
revision id already in the knit that is a parent of REVISION_ID.
1113
The ' :' marker is the end of record marker.
1116
when a write is interrupted to the index file, it will result in a line
1117
that does not end in ' :'. If the ' :' is not present at the end of a line,
1118
or at the end of the file, then the record that is missing it will be
1119
ignored by the parser.
1121
When writing new records to the index file, the data is preceded by '\n'
1122
to ensure that records always start on new lines even if the last write was
1123
interrupted. As a result its normal for the last line in the index to be
1124
missing a trailing newline. One can be added with no harmful effects.
1127
HEADER = "# bzr knit index 8\n"
1129
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1130
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1132
def _cache_version(self, version_id, options, pos, size, parents):
1133
"""Cache a version record in the history array and index cache.
1135
This is inlined into _load_data for performance. KEEP IN SYNC.
1136
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1139
# only want the _history index to reference the 1st index entry
1141
if version_id not in self._cache:
1142
index = len(self._history)
1143
self._history.append(version_id)
1145
index = self._cache[version_id][5]
1146
self._cache[version_id] = (version_id,
1153
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1154
create_parent_dir=False, delay_create=False, dir_mode=None):
1155
_KnitComponentFile.__init__(self, transport, filename, mode,
1156
file_mode=file_mode,
1157
create_parent_dir=create_parent_dir,
1160
# position in _history is the 'official' index for a revision
1161
# but the values may have come from a newer entry.
1162
# so - wc -l of a knit index is != the number of unique names
1166
fp = self._transport.get(self._filename)
1168
# _load_data may raise NoSuchFile if the target knit is
1170
_load_data(self, fp)
1174
if mode != 'w' or not create:
1177
self._need_to_create = True
1179
self._transport.put_bytes_non_atomic(
1180
self._filename, self.HEADER, mode=self._file_mode)
1182
def get_graph(self):
1183
"""Return a list of the node:parents lists from this knit index."""
1184
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1186
def get_ancestry(self, versions, topo_sorted=True):
1187
"""See VersionedFile.get_ancestry."""
1188
# get a graph of all the mentioned versions:
1190
pending = set(versions)
1193
version = pending.pop()
1196
parents = [p for p in cache[version][4] if p in cache]
1198
raise RevisionNotPresent(version, self._filename)
1199
# if not completed and not a ghost
1200
pending.update([p for p in parents if p not in graph])
1201
graph[version] = parents
1204
return topo_sort(graph.items())
1206
def get_ancestry_with_ghosts(self, versions):
1207
"""See VersionedFile.get_ancestry_with_ghosts."""
1208
# get a graph of all the mentioned versions:
1209
self.check_versions_present(versions)
1212
pending = set(versions)
1214
version = pending.pop()
1216
parents = cache[version][4]
1222
pending.update([p for p in parents if p not in graph])
1223
graph[version] = parents
1224
return topo_sort(graph.items())
1226
def iter_parents(self, version_ids):
1227
"""Iterate through the parents for many version ids.
1229
:param version_ids: An iterable yielding version_ids.
1230
:return: An iterator that yields (version_id, parents). Requested
1231
version_ids not present in the versioned file are simply skipped.
1232
The order is undefined, allowing for different optimisations in
1233
the underlying implementation.
1235
for version_id in version_ids:
1237
yield version_id, tuple(self.get_parents(version_id))
1241
def num_versions(self):
1242
return len(self._history)
1244
__len__ = num_versions
1246
def get_versions(self):
1247
"""Get all the versions in the file. not topologically sorted."""
1248
return self._history
1250
def _version_list_to_index(self, versions):
1253
for version in versions:
1254
if version in cache:
1255
# -- inlined lookup() --
1256
result_list.append(str(cache[version][5]))
1257
# -- end lookup () --
1259
result_list.append('.' + version)
1260
return ' '.join(result_list)
1262
def add_version(self, version_id, options, pos, size, parents):
1263
"""Add a version record to the index."""
1264
self.add_versions(((version_id, options, pos, size, parents),))
1266
def add_versions(self, versions):
1267
"""Add multiple versions to the index.
1269
:param versions: a list of tuples:
1270
(version_id, options, pos, size, parents).
1273
orig_history = self._history[:]
1274
orig_cache = self._cache.copy()
1277
for version_id, options, pos, size, parents in versions:
1278
line = "\n%s %s %s %s %s :" % (version_id,
1282
self._version_list_to_index(parents))
1283
assert isinstance(line, str), \
1284
'content must be utf-8 encoded: %r' % (line,)
1286
self._cache_version(version_id, options, pos, size, parents)
1287
if not self._need_to_create:
1288
self._transport.append_bytes(self._filename, ''.join(lines))
1291
sio.write(self.HEADER)
1292
sio.writelines(lines)
1294
self._transport.put_file_non_atomic(self._filename, sio,
1295
create_parent_dir=self._create_parent_dir,
1296
mode=self._file_mode,
1297
dir_mode=self._dir_mode)
1298
self._need_to_create = False
1300
# If any problems happen, restore the original values and re-raise
1301
self._history = orig_history
1302
self._cache = orig_cache
1305
def has_version(self, version_id):
1306
"""True if the version is in the index."""
1307
return version_id in self._cache
1309
def get_position(self, version_id):
1310
"""Return data position and size of specified version."""
1311
entry = self._cache[version_id]
1312
return entry[2], entry[3]
1314
def get_method(self, version_id):
1315
"""Return compression method of specified version."""
1316
options = self._cache[version_id][1]
1317
if 'fulltext' in options:
1320
if 'line-delta' not in options:
1321
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1324
def get_options(self, version_id):
1325
"""Return a string represention options.
1329
return self._cache[version_id][1]
1331
def get_parents(self, version_id):
1332
"""Return parents of specified version ignoring ghosts."""
1333
return [parent for parent in self._cache[version_id][4]
1334
if parent in self._cache]
1336
def get_parents_with_ghosts(self, version_id):
1337
"""Return parents of specified version with ghosts."""
1338
return self._cache[version_id][4]
1340
def check_versions_present(self, version_ids):
1341
"""Check that all specified versions are present."""
1343
for version_id in version_ids:
1344
if version_id not in cache:
1345
raise RevisionNotPresent(version_id, self._filename)
1348
class KnitGraphIndex(object):
1349
"""A knit index that builds on GraphIndex."""
1351
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1352
"""Construct a KnitGraphIndex on a graph_index.
1354
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1355
:param deltas: Allow delta-compressed records.
1356
:param add_callback: If not None, allow additions to the index and call
1357
this callback with a list of added GraphIndex nodes:
1358
[(node, value, node_refs), ...]
1359
:param parents: If True, record knits parents, if not do not record
1362
self._graph_index = graph_index
1363
self._deltas = deltas
1364
self._add_callback = add_callback
1365
self._parents = parents
1366
if deltas and not parents:
1367
raise KnitCorrupt(self, "Cannot do delta compression without "
1370
def _get_entries(self, version_ids, check_present=False):
1371
"""Get the entries for version_ids."""
1372
version_ids = set(version_ids)
1375
for node in self._graph_index.iter_entries(version_ids):
1377
found_keys.add(node[0])
1379
# adapt parentless index to the rest of the code.
1380
for node in self._graph_index.iter_entries(version_ids):
1381
yield node[0], node[1], ()
1382
found_keys.add(node[0])
1384
missing_keys = version_ids.difference(found_keys)
1386
raise RevisionNotPresent(missing_keys.pop(), self)
1388
def _present_keys(self, version_ids):
1390
node[0] for node in self._get_entries(version_ids)])
1392
def _parentless_ancestry(self, versions):
1393
"""Honour the get_ancestry API for parentless knit indices."""
1394
present_keys = self._present_keys(versions)
1395
missing = set(versions).difference(present_keys)
1397
raise RevisionNotPresent(missing.pop(), self)
1398
return list(present_keys)
1400
def get_ancestry(self, versions, topo_sorted=True):
1401
"""See VersionedFile.get_ancestry."""
1402
if not self._parents:
1403
return self._parentless_ancestry(versions)
1404
# XXX: This will do len(history) index calls - perhaps
1405
# it should be altered to be a index core feature?
1406
# get a graph of all the mentioned versions:
1409
versions = set(versions)
1410
pending = set(versions)
1412
# get all pending nodes
1413
this_iteration = pending
1414
new_nodes = self._get_entries(this_iteration)
1416
for (key, value, node_refs) in new_nodes:
1417
# dont ask for ghosties - otherwise
1418
# we we can end up looping with pending
1419
# being entirely ghosted.
1420
graph[key] = [parent for parent in node_refs[0]
1421
if parent not in ghosts]
1423
pending.update(graph[key])
1424
ghosts.difference_update(graph)
1425
# dont examine known nodes
1426
pending.difference_update(graph)
1427
if versions.difference(graph):
1428
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1431
return topo_sort(graph.items())
1433
def get_ancestry_with_ghosts(self, versions):
1434
"""See VersionedFile.get_ancestry."""
1435
if not self._parents:
1436
return self._parentless_ancestry(versions)
1437
# XXX: This will do len(history) index calls - perhaps
1438
# it should be altered to be a index core feature?
1439
# get a graph of all the mentioned versions:
1441
versions = set(versions)
1442
pending = set(versions)
1444
# get all pending nodes
1445
this_iteration = pending
1446
new_nodes = self._get_entries(this_iteration)
1448
for (key, value, node_refs) in new_nodes:
1449
graph[key] = node_refs[0]
1451
pending.update(graph[key])
1452
missing_versions = this_iteration.difference(graph)
1453
missing_needed = versions.intersection(missing_versions)
1455
raise RevisionNotPresent(missing_needed.pop(), self)
1456
for missing_version in missing_versions:
1457
# add a key, no parents
1458
graph[missing_version] = []
1459
# dont examine known nodes
1460
pending.difference_update(graph)
1461
return topo_sort(graph.items())
1463
def get_graph(self):
1464
"""Return a list of the node:parents lists from this knit index."""
1465
if not self._parents:
1466
return [(key, ()) for key in self.get_versions()]
1467
return [(key, refs[0]) for (key, value, refs) in
1468
self._graph_index.iter_all_entries()]
1470
def iter_parents(self, version_ids):
1471
"""Iterate through the parents for many version ids.
1473
:param version_ids: An iterable yielding version_ids.
1474
:return: An iterator that yields (version_id, parents). Requested
1475
version_ids not present in the versioned file are simply skipped.
1476
The order is undefined, allowing for different optimisations in
1477
the underlying implementation.
1480
all_nodes = set(self._get_entries(version_ids))
1482
present_parents = set()
1483
for node in all_nodes:
1484
all_parents.update(node[2][0])
1485
# any node we are querying must be present
1486
present_parents.add(node[0])
1487
unknown_parents = all_parents.difference(present_parents)
1488
present_parents.update(self._present_keys(unknown_parents))
1489
for node in all_nodes:
1491
for parent in node[2][0]:
1492
if parent in present_parents:
1493
parents.append(parent)
1494
yield node[0], tuple(parents)
1496
for node in self._get_entries(version_ids):
1499
def num_versions(self):
1500
return len(list(self._graph_index.iter_all_entries()))
1502
__len__ = num_versions
1504
def get_versions(self):
1505
"""Get all the versions in the file. not topologically sorted."""
1506
return [node[0] for node in self._graph_index.iter_all_entries()]
1508
def has_version(self, version_id):
1509
"""True if the version is in the index."""
1510
return len(self._present_keys([version_id])) == 1
1512
def get_position(self, version_id):
1513
"""Return data position and size of specified version."""
1514
bits = self._get_node(version_id)[1][1:].split(' ')
1515
return int(bits[0]), int(bits[1])
1517
def get_method(self, version_id):
1518
"""Return compression method of specified version."""
1519
if not self._deltas:
1521
return self._parent_compression(self._get_node(version_id)[2][1])
1523
def _parent_compression(self, reference_list):
1524
# use the second reference list to decide if this is delta'd or not.
1525
if len(reference_list):
1530
def _get_node(self, version_id):
1531
return list(self._get_entries([version_id]))[0]
1533
def get_options(self, version_id):
1534
"""Return a string represention options.
1538
node = self._get_node(version_id)
1539
if not self._deltas:
1540
options = ['fulltext']
1542
options = [self._parent_compression(node[2][1])]
1543
if node[1][0] == 'N':
1544
options.append('no-eol')
1545
return ','.join(options)
1547
def get_parents(self, version_id):
1548
"""Return parents of specified version ignoring ghosts."""
1549
parents = list(self.iter_parents([version_id]))
1552
raise errors.RevisionNotPresent(version_id, self)
1553
return parents[0][1]
1555
def get_parents_with_ghosts(self, version_id):
1556
"""Return parents of specified version with ghosts."""
1557
nodes = list(self._get_entries([version_id], check_present=True))
1558
if not self._parents:
1560
return nodes[0][2][0]
1562
def check_versions_present(self, version_ids):
1563
"""Check that all specified versions are present."""
1564
version_ids = set(version_ids)
1565
present = self._present_keys(version_ids)
1566
missing = version_ids.difference(present)
1568
raise RevisionNotPresent(missing.pop(), self)
1570
def add_version(self, version_id, options, pos, size, parents):
1571
"""Add a version record to the index."""
1572
return self.add_versions(((version_id, options, pos, size, parents),))
1574
def add_versions(self, versions):
1575
"""Add multiple versions to the index.
1577
This function does not insert data into the Immutable GraphIndex
1578
backing the KnitGraphIndex, instead it prepares data for insertion by
1579
the caller and checks that it is safe to insert then calls
1580
self._add_callback with the prepared GraphIndex nodes.
1582
:param versions: a list of tuples:
1583
(version_id, options, pos, size, parents).
1585
if not self._add_callback:
1586
raise errors.ReadOnlyError(self)
1587
# we hope there are no repositories with inconsistent parentage
1592
for (version_id, options, pos, size, parents) in versions:
1593
if 'no-eol' in options:
1597
value += "%d %d" % (pos, size)
1598
if not self._deltas:
1599
if 'line-delta' in options:
1600
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1603
if 'line-delta' in options:
1604
node_refs = (tuple(parents), (parents[0],))
1606
node_refs = (tuple(parents), ())
1608
node_refs = (tuple(parents), )
1611
raise KnitCorrupt(self, "attempt to add node with parents "
1612
"in parentless index.")
1614
keys[version_id] = (value, node_refs)
1615
present_nodes = self._get_entries(keys)
1616
for (key, value, node_refs) in present_nodes:
1617
if (value, node_refs) != keys[key]:
1618
raise KnitCorrupt(self, "inconsistent details in add_versions"
1619
": %s %s" % ((value, node_refs), keys[key]))
1623
for key, (value, node_refs) in keys.iteritems():
1624
result.append((key, value, node_refs))
1626
for key, (value, node_refs) in keys.iteritems():
1627
result.append((key, value))
1628
self._add_callback(result)
1631
class _KnitData(_KnitComponentFile):
1632
"""Contents of the knit data file"""
1634
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1635
create_parent_dir=False, delay_create=False,
1637
_KnitComponentFile.__init__(self, transport, filename, mode,
1638
file_mode=file_mode,
1639
create_parent_dir=create_parent_dir,
1641
self._checked = False
1642
# TODO: jam 20060713 conceptually, this could spill to disk
1643
# if the cached size gets larger than a certain amount
1644
# but it complicates the model a bit, so for now just use
1645
# a simple dictionary
1647
self._do_cache = False
1650
self._need_to_create = create
1652
self._transport.put_bytes_non_atomic(self._filename, '',
1653
mode=self._file_mode)
1655
def enable_cache(self):
1656
"""Enable caching of reads."""
1657
self._do_cache = True
1659
def clear_cache(self):
1660
"""Clear the record cache."""
1661
self._do_cache = False
1664
def _open_file(self):
1666
return self._transport.get(self._filename)
1671
def _record_to_data(self, version_id, digest, lines):
1672
"""Convert version_id, digest, lines into a raw data block.
1674
:return: (len, a StringIO instance with the raw data ready to read.)
1677
data_file = GzipFile(None, mode='wb', fileobj=sio)
1679
assert isinstance(version_id, str)
1680
data_file.writelines(chain(
1681
["version %s %d %s\n" % (version_id,
1685
["end %s\n" % version_id]))
1692
def add_raw_record(self, raw_data):
1693
"""Append a prepared record to the data file.
1695
:return: the offset in the data file raw_data was written.
1697
assert isinstance(raw_data, str), 'data must be plain bytes'
1698
if not self._need_to_create:
1699
return self._transport.append_bytes(self._filename, raw_data)
1701
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1702
create_parent_dir=self._create_parent_dir,
1703
mode=self._file_mode,
1704
dir_mode=self._dir_mode)
1705
self._need_to_create = False
1708
def add_record(self, version_id, digest, lines):
1709
"""Write new text record to disk. Returns the position in the
1710
file where it was written."""
1711
size, sio = self._record_to_data(version_id, digest, lines)
1713
if not self._need_to_create:
1714
start_pos = self._transport.append_file(self._filename, sio)
1716
self._transport.put_file_non_atomic(self._filename, sio,
1717
create_parent_dir=self._create_parent_dir,
1718
mode=self._file_mode,
1719
dir_mode=self._dir_mode)
1720
self._need_to_create = False
1723
self._cache[version_id] = sio.getvalue()
1724
return start_pos, size
1726
def _parse_record_header(self, version_id, raw_data):
1727
"""Parse a record header for consistency.
1729
:return: the header and the decompressor stream.
1730
as (stream, header_record)
1732
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1734
rec = self._check_header(version_id, df.readline())
1735
except Exception, e:
1736
raise KnitCorrupt(self._filename,
1737
"While reading {%s} got %s(%s)"
1738
% (version_id, e.__class__.__name__, str(e)))
1741
def _check_header(self, version_id, line):
1744
raise KnitCorrupt(self._filename,
1745
'unexpected number of elements in record header')
1746
if rec[1] != version_id:
1747
raise KnitCorrupt(self._filename,
1748
'unexpected version, wanted %r, got %r'
1749
% (version_id, rec[1]))
1752
def _parse_record(self, version_id, data):
1754
# 4168 calls in 2880 217 internal
1755
# 4168 calls to _parse_record_header in 2121
1756
# 4168 calls to readlines in 330
1757
df = GzipFile(mode='rb', fileobj=StringIO(data))
1760
record_contents = df.readlines()
1761
except Exception, e:
1762
raise KnitCorrupt(self._filename,
1763
"While reading {%s} got %s(%s)"
1764
% (version_id, e.__class__.__name__, str(e)))
1765
header = record_contents.pop(0)
1766
rec = self._check_header(version_id, header)
1768
last_line = record_contents.pop()
1769
if len(record_contents) != int(rec[2]):
1770
raise KnitCorrupt(self._filename,
1771
'incorrect number of lines %s != %s'
1773
% (len(record_contents), int(rec[2]),
1775
if last_line != 'end %s\n' % rec[1]:
1776
raise KnitCorrupt(self._filename,
1777
'unexpected version end line %r, wanted %r'
1778
% (last_line, version_id))
1780
return record_contents, rec[3]
1782
def read_records_iter_raw(self, records):
1783
"""Read text records from data file and yield raw data.
1785
This unpacks enough of the text record to validate the id is
1786
as expected but thats all.
1788
# setup an iterator of the external records:
1789
# uses readv so nice and fast we hope.
1791
# grab the disk data needed.
1793
# Don't check _cache if it is empty
1794
needed_offsets = [(pos, size) for version_id, pos, size
1796
if version_id not in self._cache]
1798
needed_offsets = [(pos, size) for version_id, pos, size
1801
raw_records = self._transport.readv(self._filename, needed_offsets)
1803
for version_id, pos, size in records:
1804
if version_id in self._cache:
1805
# This data has already been validated
1806
data = self._cache[version_id]
1808
pos, data = raw_records.next()
1810
self._cache[version_id] = data
1812
# validate the header
1813
df, rec = self._parse_record_header(version_id, data)
1815
yield version_id, data
1817
def read_records_iter(self, records):
1818
"""Read text records from data file and yield result.
1820
The result will be returned in whatever is the fastest to read.
1821
Not by the order requested. Also, multiple requests for the same
1822
record will only yield 1 response.
1823
:param records: A list of (version_id, pos, len) entries
1824
:return: Yields (version_id, contents, digest) in the order
1825
read, not the order requested
1831
# Skip records we have alread seen
1832
yielded_records = set()
1833
needed_records = set()
1834
for record in records:
1835
if record[0] in self._cache:
1836
if record[0] in yielded_records:
1838
yielded_records.add(record[0])
1839
data = self._cache[record[0]]
1840
content, digest = self._parse_record(record[0], data)
1841
yield (record[0], content, digest)
1843
needed_records.add(record)
1844
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1846
needed_records = sorted(set(records), key=operator.itemgetter(1))
1848
if not needed_records:
1851
# The transport optimizes the fetching as well
1852
# (ie, reads continuous ranges.)
1853
readv_response = self._transport.readv(self._filename,
1854
[(pos, size) for version_id, pos, size in needed_records])
1856
for (version_id, pos, size), (pos, data) in \
1857
izip(iter(needed_records), readv_response):
1858
content, digest = self._parse_record(version_id, data)
1860
self._cache[version_id] = data
1861
yield version_id, content, digest
1863
def read_records(self, records):
1864
"""Read records into a dictionary."""
1866
for record_id, content, digest in \
1867
self.read_records_iter(records):
1868
components[record_id] = (content, digest)
1872
class InterKnit(InterVersionedFile):
1873
"""Optimised code paths for knit to knit operations."""
1875
_matching_file_from_factory = KnitVersionedFile
1876
_matching_file_to_factory = KnitVersionedFile
1879
def is_compatible(source, target):
1880
"""Be compatible with knits. """
1882
return (isinstance(source, KnitVersionedFile) and
1883
isinstance(target, KnitVersionedFile))
1884
except AttributeError:
1887
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1888
"""See InterVersionedFile.join."""
1889
assert isinstance(self.source, KnitVersionedFile)
1890
assert isinstance(self.target, KnitVersionedFile)
1892
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1897
pb = ui.ui_factory.nested_progress_bar()
1899
version_ids = list(version_ids)
1900
if None in version_ids:
1901
version_ids.remove(None)
1903
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1904
this_versions = set(self.target._index.get_versions())
1905
needed_versions = self.source_ancestry - this_versions
1906
cross_check_versions = self.source_ancestry.intersection(this_versions)
1907
mismatched_versions = set()
1908
for version in cross_check_versions:
1909
# scan to include needed parents.
1910
n1 = set(self.target.get_parents_with_ghosts(version))
1911
n2 = set(self.source.get_parents_with_ghosts(version))
1913
# FIXME TEST this check for cycles being introduced works
1914
# the logic is we have a cycle if in our graph we are an
1915
# ancestor of any of the n2 revisions.
1921
parent_ancestors = self.source.get_ancestry(parent)
1922
if version in parent_ancestors:
1923
raise errors.GraphCycleError([parent, version])
1924
# ensure this parent will be available later.
1925
new_parents = n2.difference(n1)
1926
needed_versions.update(new_parents.difference(this_versions))
1927
mismatched_versions.add(version)
1929
if not needed_versions and not mismatched_versions:
1931
full_list = topo_sort(self.source.get_graph())
1933
version_list = [i for i in full_list if (not self.target.has_version(i)
1934
and i in needed_versions)]
1938
copy_queue_records = []
1940
for version_id in version_list:
1941
options = self.source._index.get_options(version_id)
1942
parents = self.source._index.get_parents_with_ghosts(version_id)
1943
# check that its will be a consistent copy:
1944
for parent in parents:
1945
# if source has the parent, we must :
1946
# * already have it or
1947
# * have it scheduled already
1948
# otherwise we don't care
1949
assert (self.target.has_version(parent) or
1950
parent in copy_set or
1951
not self.source.has_version(parent))
1952
data_pos, data_size = self.source._index.get_position(version_id)
1953
copy_queue_records.append((version_id, data_pos, data_size))
1954
copy_queue.append((version_id, options, parents))
1955
copy_set.add(version_id)
1957
# data suck the join:
1959
total = len(version_list)
1962
for (version_id, raw_data), \
1963
(version_id2, options, parents) in \
1964
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1966
assert version_id == version_id2, 'logic error, inconsistent results'
1968
pb.update("Joining knit", count, total)
1969
raw_records.append((version_id, options, parents, len(raw_data)))
1970
raw_datum.append(raw_data)
1971
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1973
for version in mismatched_versions:
1974
# FIXME RBC 20060309 is this needed?
1975
n1 = set(self.target.get_parents_with_ghosts(version))
1976
n2 = set(self.source.get_parents_with_ghosts(version))
1977
# write a combined record to our history preserving the current
1978
# parents as first in the list
1979
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1980
self.target.fix_parents(version, new_parents)
1986
InterVersionedFile.register_optimiser(InterKnit)
1989
class WeaveToKnit(InterVersionedFile):
1990
"""Optimised code paths for weave to knit operations."""
1992
_matching_file_from_factory = bzrlib.weave.WeaveFile
1993
_matching_file_to_factory = KnitVersionedFile
1996
def is_compatible(source, target):
1997
"""Be compatible with weaves to knits."""
1999
return (isinstance(source, bzrlib.weave.Weave) and
2000
isinstance(target, KnitVersionedFile))
2001
except AttributeError:
2004
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2005
"""See InterVersionedFile.join."""
2006
assert isinstance(self.source, bzrlib.weave.Weave)
2007
assert isinstance(self.target, KnitVersionedFile)
2009
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2014
pb = ui.ui_factory.nested_progress_bar()
2016
version_ids = list(version_ids)
2018
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2019
this_versions = set(self.target._index.get_versions())
2020
needed_versions = self.source_ancestry - this_versions
2021
cross_check_versions = self.source_ancestry.intersection(this_versions)
2022
mismatched_versions = set()
2023
for version in cross_check_versions:
2024
# scan to include needed parents.
2025
n1 = set(self.target.get_parents_with_ghosts(version))
2026
n2 = set(self.source.get_parents(version))
2027
# if all of n2's parents are in n1, then its fine.
2028
if n2.difference(n1):
2029
# FIXME TEST this check for cycles being introduced works
2030
# the logic is we have a cycle if in our graph we are an
2031
# ancestor of any of the n2 revisions.
2037
parent_ancestors = self.source.get_ancestry(parent)
2038
if version in parent_ancestors:
2039
raise errors.GraphCycleError([parent, version])
2040
# ensure this parent will be available later.
2041
new_parents = n2.difference(n1)
2042
needed_versions.update(new_parents.difference(this_versions))
2043
mismatched_versions.add(version)
2045
if not needed_versions and not mismatched_versions:
2047
full_list = topo_sort(self.source.get_graph())
2049
version_list = [i for i in full_list if (not self.target.has_version(i)
2050
and i in needed_versions)]
2054
total = len(version_list)
2055
for version_id in version_list:
2056
pb.update("Converting to knit", count, total)
2057
parents = self.source.get_parents(version_id)
2058
# check that its will be a consistent copy:
2059
for parent in parents:
2060
# if source has the parent, we must already have it
2061
assert (self.target.has_version(parent))
2062
self.target.add_lines(
2063
version_id, parents, self.source.get_lines(version_id))
2066
for version in mismatched_versions:
2067
# FIXME RBC 20060309 is this needed?
2068
n1 = set(self.target.get_parents_with_ghosts(version))
2069
n2 = set(self.source.get_parents(version))
2070
# write a combined record to our history preserving the current
2071
# parents as first in the list
2072
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2073
self.target.fix_parents(version, new_parents)
2079
InterVersionedFile.register_optimiser(WeaveToKnit)
2082
class KnitSequenceMatcher(difflib.SequenceMatcher):
2083
"""Knit tuned sequence matcher.
2085
This is based on profiling of difflib which indicated some improvements
2086
for our usage pattern.
2089
def find_longest_match(self, alo, ahi, blo, bhi):
2090
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
2092
If isjunk is not defined:
2094
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
2095
alo <= i <= i+k <= ahi
2096
blo <= j <= j+k <= bhi
2097
and for all (i',j',k') meeting those conditions,
2100
and if i == i', j <= j'
2102
In other words, of all maximal matching blocks, return one that
2103
starts earliest in a, and of all those maximal matching blocks that
2104
start earliest in a, return the one that starts earliest in b.
2106
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
2107
>>> s.find_longest_match(0, 5, 0, 9)
2110
If isjunk is defined, first the longest matching block is
2111
determined as above, but with the additional restriction that no
2112
junk element appears in the block. Then that block is extended as
2113
far as possible by matching (only) junk elements on both sides. So
2114
the resulting block never matches on junk except as identical junk
2115
happens to be adjacent to an "interesting" match.
2117
Here's the same example as before, but considering blanks to be
2118
junk. That prevents " abcd" from matching the " abcd" at the tail
2119
end of the second sequence directly. Instead only the "abcd" can
2120
match, and matches the leftmost "abcd" in the second sequence:
2122
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
2123
>>> s.find_longest_match(0, 5, 0, 9)
2126
If no blocks match, return (alo, blo, 0).
2128
>>> s = SequenceMatcher(None, "ab", "c")
2129
>>> s.find_longest_match(0, 2, 0, 1)
2133
# CAUTION: stripping common prefix or suffix would be incorrect.
2137
# Longest matching block is "ab", but if common prefix is
2138
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
2139
# strip, so ends up claiming that ab is changed to acab by
2140
# inserting "ca" in the middle. That's minimal but unintuitive:
2141
# "it's obvious" that someone inserted "ac" at the front.
2142
# Windiff ends up at the same place as diff, but by pairing up
2143
# the unique 'b's and then matching the first two 'a's.
2145
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
2146
besti, bestj, bestsize = alo, blo, 0
2147
# find longest junk-free match
2148
# during an iteration of the loop, j2len[j] = length of longest
2149
# junk-free match ending with a[i-1] and b[j]
2153
for i in xrange(alo, ahi):
2154
# look at all instances of a[i] in b; note that because
2155
# b2j has no junk keys, the loop is skipped if a[i] is junk
2156
j2lenget = j2len.get
2159
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
2160
# following improvement
2161
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
2162
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
2163
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
2165
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
2166
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
2167
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
2179
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
2181
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
2184
# Extend the best by non-junk elements on each end. In particular,
2185
# "popular" non-junk elements aren't in b2j, which greatly speeds
2186
# the inner loop above, but also means "the best" match so far
2187
# doesn't contain any junk *or* popular non-junk elements.
2188
while besti > alo and bestj > blo and \
2189
not isbjunk(b[bestj-1]) and \
2190
a[besti-1] == b[bestj-1]:
2191
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2192
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2193
not isbjunk(b[bestj+bestsize]) and \
2194
a[besti+bestsize] == b[bestj+bestsize]:
2197
# Now that we have a wholly interesting match (albeit possibly
2198
# empty!), we may as well suck up the matching junk on each
2199
# side of it too. Can't think of a good reason not to, and it
2200
# saves post-processing the (possibly considerable) expense of
2201
# figuring out what to do with it. In the case of an empty
2202
# interesting match, this is clearly the right thing to do,
2203
# because no other kind of match is possible in the regions.
2204
while besti > alo and bestj > blo and \
2205
isbjunk(b[bestj-1]) and \
2206
a[besti-1] == b[bestj-1]:
2207
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2208
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2209
isbjunk(b[bestj+bestsize]) and \
2210
a[besti+bestsize] == b[bestj+bestsize]:
2211
bestsize = bestsize + 1
2213
return besti, bestj, bestsize
2217
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2219
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
