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
82
from bzrlib.errors import (
90
RevisionAlreadyPresent,
92
from bzrlib.tuned_gzip import GzipFile
93
from bzrlib.trace import mutter
94
from bzrlib.osutils import (
99
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
100
from bzrlib.tsort import topo_sort
103
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
106
# TODO: Split out code specific to this format into an associated object.
108
# TODO: Can we put in some kind of value to check that the index and data
109
# files belong together?
111
# TODO: accommodate binaries, perhaps by storing a byte count
113
# TODO: function to check whole file
115
# TODO: atomically append data, then measure backwards from the cursor
116
# position after writing to work out where it was located. we may need to
117
# bypass python file buffering.
119
DATA_SUFFIX = '.knit'
120
INDEX_SUFFIX = '.kndx'
123
class KnitContent(object):
124
"""Content of a knit version to which deltas can be applied."""
126
def __init__(self, lines):
129
def annotate_iter(self):
130
"""Yield tuples of (origin, text) for each content line."""
131
return iter(self._lines)
134
"""Return a list of (origin, text) tuples."""
135
return list(self.annotate_iter())
137
def line_delta_iter(self, new_lines):
138
"""Generate line-based delta from this content to new_lines."""
139
new_texts = new_lines.text()
140
old_texts = self.text()
141
s = KnitSequenceMatcher(None, old_texts, new_texts)
142
for tag, i1, i2, j1, j2 in s.get_opcodes():
145
# ofrom, oto, length, data
146
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
148
def line_delta(self, new_lines):
149
return list(self.line_delta_iter(new_lines))
152
return [text for origin, text in self._lines]
155
return KnitContent(self._lines[:])
158
def get_line_delta_blocks(knit_delta, source, target):
159
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
160
target_len = len(target)
163
for s_begin, s_end, t_len, new_text in knit_delta:
164
true_n = s_begin - s_pos
167
# knit deltas do not provide reliable info about whether the
168
# last line of a file matches, due to eol handling.
169
if source[s_pos + n -1] != target[t_pos + n -1]:
172
yield s_pos, t_pos, n
173
t_pos += t_len + true_n
175
n = target_len - t_pos
177
if source[s_pos + n -1] != target[t_pos + n -1]:
180
yield s_pos, t_pos, n
181
yield s_pos + (target_len - t_pos), target_len, 0
184
class _KnitFactory(object):
185
"""Base factory for creating content objects."""
187
def make(self, lines, version_id):
188
num_lines = len(lines)
189
return KnitContent(zip([version_id] * num_lines, lines))
192
class KnitAnnotateFactory(_KnitFactory):
193
"""Factory for creating annotated Content objects."""
197
def parse_fulltext(self, content, version_id):
198
"""Convert fulltext to internal representation
200
fulltext content is of the format
201
revid(utf8) plaintext\n
202
internal representation is of the format:
205
# TODO: jam 20070209 The tests expect this to be returned as tuples,
206
# but the code itself doesn't really depend on that.
207
# Figure out a way to not require the overhead of turning the
208
# list back into tuples.
209
lines = [tuple(line.split(' ', 1)) for line in content]
210
return KnitContent(lines)
212
def parse_line_delta_iter(self, lines):
213
return iter(self.parse_line_delta(lines))
215
def parse_line_delta(self, lines, version_id):
216
"""Convert a line based delta into internal representation.
218
line delta is in the form of:
219
intstart intend intcount
221
revid(utf8) newline\n
222
internal representation is
223
(start, end, count, [1..count tuples (revid, newline)])
230
def cache_and_return(line):
231
origin, text = line.split(' ', 1)
232
return cache.setdefault(origin, origin), text
234
# walk through the lines parsing.
236
start, end, count = [int(n) for n in header.split(',')]
237
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
238
result.append((start, end, count, contents))
241
def get_fulltext_content(self, lines):
242
"""Extract just the content lines from a fulltext."""
243
return (line.split(' ', 1)[1] for line in lines)
245
def get_linedelta_content(self, lines):
246
"""Extract just the content from a line delta.
248
This doesn't return all of the extra information stored in a delta.
249
Only the actual content lines.
254
header = header.split(',')
255
count = int(header[2])
256
for i in xrange(count):
257
origin, text = next().split(' ', 1)
260
def lower_fulltext(self, content):
261
"""convert a fulltext content record into a serializable form.
263
see parse_fulltext which this inverts.
265
# TODO: jam 20070209 We only do the caching thing to make sure that
266
# the origin is a valid utf-8 line, eventually we could remove it
267
return ['%s %s' % (o, t) for o, t in content._lines]
269
def lower_line_delta(self, delta):
270
"""convert a delta into a serializable form.
272
See parse_line_delta which this inverts.
274
# TODO: jam 20070209 We only do the caching thing to make sure that
275
# the origin is a valid utf-8 line, eventually we could remove it
277
for start, end, c, lines in delta:
278
out.append('%d,%d,%d\n' % (start, end, c))
279
out.extend(origin + ' ' + text
280
for origin, text in lines)
284
class KnitPlainFactory(_KnitFactory):
285
"""Factory for creating plain Content objects."""
289
def parse_fulltext(self, content, version_id):
290
"""This parses an unannotated fulltext.
292
Note that this is not a noop - the internal representation
293
has (versionid, line) - its just a constant versionid.
295
return self.make(content, version_id)
297
def parse_line_delta_iter(self, lines, version_id):
299
num_lines = len(lines)
300
while cur < num_lines:
303
start, end, c = [int(n) for n in header.split(',')]
304
yield start, end, c, zip([version_id] * c, lines[cur:cur+c])
307
def parse_line_delta(self, lines, version_id):
308
return list(self.parse_line_delta_iter(lines, version_id))
310
def get_fulltext_content(self, lines):
311
"""Extract just the content lines from a fulltext."""
314
def get_linedelta_content(self, lines):
315
"""Extract just the content from a line delta.
317
This doesn't return all of the extra information stored in a delta.
318
Only the actual content lines.
323
header = header.split(',')
324
count = int(header[2])
325
for i in xrange(count):
328
def lower_fulltext(self, content):
329
return content.text()
331
def lower_line_delta(self, delta):
333
for start, end, c, lines in delta:
334
out.append('%d,%d,%d\n' % (start, end, c))
335
out.extend([text for origin, text in lines])
339
def make_empty_knit(transport, relpath):
340
"""Construct a empty knit at the specified location."""
341
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
345
class KnitVersionedFile(VersionedFile):
346
"""Weave-like structure with faster random access.
348
A knit stores a number of texts and a summary of the relationships
349
between them. Texts are identified by a string version-id. Texts
350
are normally stored and retrieved as a series of lines, but can
351
also be passed as single strings.
353
Lines are stored with the trailing newline (if any) included, to
354
avoid special cases for files with no final newline. Lines are
355
composed of 8-bit characters, not unicode. The combination of
356
these approaches should mean any 'binary' file can be safely
357
stored and retrieved.
360
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
361
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
362
create=False, create_parent_dir=False, delay_create=False,
363
dir_mode=None, index=None):
364
"""Construct a knit at location specified by relpath.
366
:param create: If not True, only open an existing knit.
367
:param create_parent_dir: If True, create the parent directory if
368
creating the file fails. (This is used for stores with
369
hash-prefixes that may not exist yet)
370
:param delay_create: The calling code is aware that the knit won't
371
actually be created until the first data is stored.
372
:param index: An index to use for the knit.
374
if deprecated_passed(basis_knit):
375
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
376
" deprecated as of bzr 0.9.",
377
DeprecationWarning, stacklevel=2)
378
if access_mode is None:
380
super(KnitVersionedFile, self).__init__(access_mode)
381
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
382
self.transport = transport
383
self.filename = relpath
384
self.factory = factory or KnitAnnotateFactory()
385
self.writable = (access_mode == 'w')
388
self._max_delta_chain = 200
391
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
392
access_mode, create=create, file_mode=file_mode,
393
create_parent_dir=create_parent_dir, delay_create=delay_create,
397
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
398
access_mode, create=create and not len(self), file_mode=file_mode,
399
create_parent_dir=create_parent_dir, delay_create=delay_create,
403
return '%s(%s)' % (self.__class__.__name__,
404
self.transport.abspath(self.filename))
406
def _check_should_delta(self, first_parents):
407
"""Iterate back through the parent listing, looking for a fulltext.
409
This is used when we want to decide whether to add a delta or a new
410
fulltext. It searches for _max_delta_chain parents. When it finds a
411
fulltext parent, it sees if the total size of the deltas leading up to
412
it is large enough to indicate that we want a new full text anyway.
414
Return True if we should create a new delta, False if we should use a
419
delta_parents = first_parents
420
for count in xrange(self._max_delta_chain):
421
parent = delta_parents[0]
422
method = self._index.get_method(parent)
423
pos, size = self._index.get_position(parent)
424
if method == 'fulltext':
428
delta_parents = self._index.get_parents(parent)
430
# We couldn't find a fulltext, so we must create a new one
433
return fulltext_size > delta_size
435
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
436
"""See VersionedFile._add_delta()."""
437
self._check_add(version_id, []) # should we check the lines ?
438
self._check_versions_present(parents)
442
for parent in parents:
443
if not self.has_version(parent):
444
ghosts.append(parent)
446
present_parents.append(parent)
448
if delta_parent is None:
449
# reconstitute as full text.
450
assert len(delta) == 1 or len(delta) == 0
452
assert delta[0][0] == 0
453
assert delta[0][1] == 0, delta[0][1]
454
return super(KnitVersionedFile, self)._add_delta(version_id,
465
options.append('no-eol')
467
if delta_parent is not None:
468
# determine the current delta chain length.
469
# To speed the extract of texts the delta chain is limited
470
# to a fixed number of deltas. This should minimize both
471
# I/O and the time spend applying deltas.
472
# The window was changed to a maximum of 200 deltas, but also added
473
# was a check that the total compressed size of the deltas is
474
# smaller than the compressed size of the fulltext.
475
if not self._check_should_delta([delta_parent]):
476
# We don't want a delta here, just do a normal insertion.
477
return super(KnitVersionedFile, self)._add_delta(version_id,
484
options.append('line-delta')
485
store_lines = self.factory.lower_line_delta(delta)
487
where, size = self._data.add_record(version_id, digest, store_lines)
488
self._index.add_version(version_id, options, where, size, parents)
490
def _add_raw_records(self, records, data):
491
"""Add all the records 'records' with data pre-joined in 'data'.
493
:param records: A list of tuples(version_id, options, parents, size).
494
:param data: The data for the records. When it is written, the records
495
are adjusted to have pos pointing into data by the sum of
496
the preceding records sizes.
499
pos = self._data.add_raw_record(data)
502
for (version_id, options, parents, size) in records:
503
index_entries.append((version_id, options, pos+offset,
505
if self._data._do_cache:
506
self._data._cache[version_id] = data[offset:offset+size]
508
self._index.add_versions(index_entries)
510
def enable_cache(self):
511
"""Start caching data for this knit"""
512
self._data.enable_cache()
514
def clear_cache(self):
515
"""Clear the data cache only."""
516
self._data.clear_cache()
518
def copy_to(self, name, transport):
519
"""See VersionedFile.copy_to()."""
520
# copy the current index to a temp index to avoid racing with local
522
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
523
self.transport.get(self._index._filename))
525
f = self._data._open_file()
527
transport.put_file(name + DATA_SUFFIX, f)
530
# move the copied index into place
531
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
533
def create_empty(self, name, transport, mode=None):
534
return KnitVersionedFile(name, transport, factory=self.factory,
535
delta=self.delta, create=True)
537
def _fix_parents(self, version_id, new_parents):
538
"""Fix the parents list for version.
540
This is done by appending a new version to the index
541
with identical data except for the parents list.
542
the parents list must be a superset of the current
545
current_values = self._index._cache[version_id]
546
assert set(current_values[4]).difference(set(new_parents)) == set()
547
self._index.add_version(version_id,
553
def _extract_blocks(self, version_id, source, target):
554
if self._index.get_method(version_id) != 'line-delta':
556
parent, sha1, noeol, delta = self.get_delta(version_id)
557
return KnitContent.get_line_delta_blocks(delta, source, target)
559
def get_delta(self, version_id):
560
"""Get a delta for constructing version from some other version."""
561
version_id = osutils.safe_revision_id(version_id)
562
self.check_not_reserved_id(version_id)
563
if not self.has_version(version_id):
564
raise RevisionNotPresent(version_id, self.filename)
566
parents = self.get_parents(version_id)
571
data_pos, data_size = self._index.get_position(version_id)
572
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
573
noeol = 'no-eol' in self._index.get_options(version_id)
574
if 'fulltext' == self._index.get_method(version_id):
575
new_content = self.factory.parse_fulltext(data, version_id)
576
if parent is not None:
577
reference_content = self._get_content(parent)
578
old_texts = reference_content.text()
581
new_texts = new_content.text()
582
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
583
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
585
delta = self.factory.parse_line_delta(data, version_id)
586
return parent, sha1, noeol, delta
588
def get_graph_with_ghosts(self):
589
"""See VersionedFile.get_graph_with_ghosts()."""
590
graph_items = self._index.get_graph()
591
return dict(graph_items)
593
def get_sha1(self, version_id):
594
return self.get_sha1s([version_id])[0]
596
def get_sha1s(self, version_ids):
597
"""See VersionedFile.get_sha1()."""
598
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
599
record_map = self._get_record_map(version_ids)
600
# record entry 2 is the 'digest'.
601
return [record_map[v][2] for v in version_ids]
605
"""See VersionedFile.get_suffixes()."""
606
return [DATA_SUFFIX, INDEX_SUFFIX]
608
def has_ghost(self, version_id):
609
"""True if there is a ghost reference in the file to version_id."""
610
version_id = osutils.safe_revision_id(version_id)
612
if self.has_version(version_id):
614
# optimisable if needed by memoising the _ghosts set.
615
items = self._index.get_graph()
616
for node, parents in items:
617
for parent in parents:
618
if parent not in self._index._cache:
619
if parent == version_id:
624
"""See VersionedFile.versions."""
625
return self._index.get_versions()
627
def has_version(self, version_id):
628
"""See VersionedFile.has_version."""
629
version_id = osutils.safe_revision_id(version_id)
630
return self._index.has_version(version_id)
632
__contains__ = has_version
634
def _merge_annotations(self, content, parents, parent_texts={},
635
delta=None, annotated=None):
636
"""Merge annotations for content. This is done by comparing
637
the annotations based on changed to the text.
641
for parent_id in parents:
642
merge_content = self._get_content(parent_id, parent_texts)
643
seq = patiencediff.PatienceSequenceMatcher(
644
None, merge_content.text(), content.text())
645
if delta_seq is None:
646
# setup a delta seq to reuse.
648
for i, j, n in seq.get_matching_blocks():
651
# this appears to copy (origin, text) pairs across to the new
652
# content for any line that matches the last-checked parent.
653
# FIXME: save the sequence control data for delta compression
654
# against the most relevant parent rather than rediffing.
655
content._lines[j:j+n] = merge_content._lines[i:i+n]
658
reference_content = self._get_content(parents[0], parent_texts)
659
new_texts = content.text()
660
old_texts = reference_content.text()
661
delta_seq = patiencediff.PatienceSequenceMatcher(
662
None, old_texts, new_texts)
663
return self._make_line_delta(delta_seq, content)
665
def _make_line_delta(self, delta_seq, new_content):
666
"""Generate a line delta from delta_seq and new_content."""
668
for op in delta_seq.get_opcodes():
671
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
674
def _get_components_positions(self, version_ids):
675
"""Produce a map of position data for the components of versions.
677
This data is intended to be used for retrieving the knit records.
679
A dict of version_id to (method, data_pos, data_size, next) is
681
method is the way referenced data should be applied.
682
data_pos is the position of the data in the knit.
683
data_size is the size of the data in the knit.
684
next is the build-parent of the version, or None for fulltexts.
687
for version_id in version_ids:
690
while cursor is not None and cursor not in component_data:
691
method = self._index.get_method(cursor)
692
if method == 'fulltext':
695
next = self.get_parents(cursor)[0]
696
data_pos, data_size = self._index.get_position(cursor)
697
component_data[cursor] = (method, data_pos, data_size, next)
699
return component_data
701
def _get_content(self, version_id, parent_texts={}):
702
"""Returns a content object that makes up the specified
704
if not self.has_version(version_id):
705
raise RevisionNotPresent(version_id, self.filename)
707
cached_version = parent_texts.get(version_id, None)
708
if cached_version is not None:
709
return cached_version
711
text_map, contents_map = self._get_content_maps([version_id])
712
return contents_map[version_id]
714
def _check_versions_present(self, version_ids):
715
"""Check that all specified versions are present."""
716
self._index.check_versions_present(version_ids)
718
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
719
"""See VersionedFile.add_lines_with_ghosts()."""
720
self._check_add(version_id, lines)
721
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
723
def _add_lines(self, version_id, parents, lines, parent_texts):
724
"""See VersionedFile.add_lines."""
725
self._check_add(version_id, lines)
726
self._check_versions_present(parents)
727
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
729
def _check_add(self, version_id, lines):
730
"""check that version_id and lines are safe to add."""
731
assert self.writable, "knit is not opened for write"
732
### FIXME escape. RBC 20060228
733
if contains_whitespace(version_id):
734
raise InvalidRevisionId(version_id, self.filename)
735
self.check_not_reserved_id(version_id)
736
if self.has_version(version_id):
737
raise RevisionAlreadyPresent(version_id, self.filename)
738
self._check_lines_not_unicode(lines)
739
self._check_lines_are_lines(lines)
741
def _add(self, version_id, lines, parents, delta, parent_texts):
742
"""Add a set of lines on top of version specified by parents.
744
If delta is true, compress the text as a line-delta against
747
Any versions not present will be converted into ghosts.
749
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
750
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
751
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
752
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
753
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
754
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
755
# +1383 0 8.0370 8.0370 +<len>
756
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
757
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
758
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
759
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
763
if parent_texts is None:
765
for parent in parents:
766
if not self.has_version(parent):
767
ghosts.append(parent)
769
present_parents.append(parent)
771
if delta and not len(present_parents):
774
digest = sha_strings(lines)
777
if lines[-1][-1] != '\n':
778
options.append('no-eol')
779
lines[-1] = lines[-1] + '\n'
781
if len(present_parents) and delta:
782
# To speed the extract of texts the delta chain is limited
783
# to a fixed number of deltas. This should minimize both
784
# I/O and the time spend applying deltas.
785
delta = self._check_should_delta(present_parents)
787
assert isinstance(version_id, str)
788
lines = self.factory.make(lines, version_id)
789
if delta or (self.factory.annotated and len(present_parents) > 0):
790
# Merge annotations from parent texts if so is needed.
791
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
792
delta, self.factory.annotated)
795
options.append('line-delta')
796
store_lines = self.factory.lower_line_delta(delta_hunks)
798
options.append('fulltext')
799
store_lines = self.factory.lower_fulltext(lines)
801
where, size = self._data.add_record(version_id, digest, store_lines)
802
self._index.add_version(version_id, options, where, size, parents)
805
def check(self, progress_bar=None):
806
"""See VersionedFile.check()."""
808
def _clone_text(self, new_version_id, old_version_id, parents):
809
"""See VersionedFile.clone_text()."""
810
# FIXME RBC 20060228 make fast by only inserting an index with null
812
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
814
def get_lines(self, version_id):
815
"""See VersionedFile.get_lines()."""
816
return self.get_line_list([version_id])[0]
818
def _get_record_map(self, version_ids):
819
"""Produce a dictionary of knit records.
821
The keys are version_ids, the values are tuples of (method, content,
823
method is the way the content should be applied.
824
content is a KnitContent object.
825
digest is the SHA1 digest of this version id after all steps are done
826
next is the build-parent of the version, i.e. the leftmost ancestor.
827
If the method is fulltext, next will be None.
829
position_map = self._get_components_positions(version_ids)
830
# c = component_id, m = method, p = position, s = size, n = next
831
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
833
for component_id, content, digest in \
834
self._data.read_records_iter(records):
835
method, position, size, next = position_map[component_id]
836
record_map[component_id] = method, content, digest, next
840
def get_text(self, version_id):
841
"""See VersionedFile.get_text"""
842
return self.get_texts([version_id])[0]
844
def get_texts(self, version_ids):
845
return [''.join(l) for l in self.get_line_list(version_ids)]
847
def get_line_list(self, version_ids):
848
"""Return the texts of listed versions as a list of strings."""
849
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
850
for version_id in version_ids:
851
self.check_not_reserved_id(version_id)
852
text_map, content_map = self._get_content_maps(version_ids)
853
return [text_map[v] for v in version_ids]
855
_get_lf_split_line_list = get_line_list
857
def _get_content_maps(self, version_ids):
858
"""Produce maps of text and KnitContents
860
:return: (text_map, content_map) where text_map contains the texts for
861
the requested versions and content_map contains the KnitContents.
862
Both dicts take version_ids as their keys.
864
for version_id in version_ids:
865
if not self.has_version(version_id):
866
raise RevisionNotPresent(version_id, self.filename)
867
record_map = self._get_record_map(version_ids)
872
for version_id in version_ids:
875
while cursor is not None:
876
method, data, digest, next = record_map[cursor]
877
components.append((cursor, method, data, digest))
878
if cursor in content_map:
883
for component_id, method, data, digest in reversed(components):
884
if component_id in content_map:
885
content = content_map[component_id]
887
if method == 'fulltext':
888
assert content is None
889
content = self.factory.parse_fulltext(data, version_id)
890
elif method == 'line-delta':
891
delta = self.factory.parse_line_delta(data, version_id)
892
content = content.copy()
893
content._lines = self._apply_delta(content._lines,
895
content_map[component_id] = content
897
if 'no-eol' in self._index.get_options(version_id):
898
content = content.copy()
899
line = content._lines[-1][1].rstrip('\n')
900
content._lines[-1] = (content._lines[-1][0], line)
901
final_content[version_id] = content
903
# digest here is the digest from the last applied component.
904
text = content.text()
905
if sha_strings(text) != digest:
906
raise KnitCorrupt(self.filename,
907
'sha-1 does not match %s' % version_id)
909
text_map[version_id] = text
910
return text_map, final_content
912
def iter_lines_added_or_present_in_versions(self, version_ids=None,
914
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
915
if version_ids is None:
916
version_ids = self.versions()
918
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
920
pb = progress.DummyProgress()
921
# we don't care about inclusions, the caller cares.
922
# but we need to setup a list of records to visit.
923
# we need version_id, position, length
924
version_id_records = []
925
requested_versions = set(version_ids)
926
# filter for available versions
927
for version_id in requested_versions:
928
if not self.has_version(version_id):
929
raise RevisionNotPresent(version_id, self.filename)
930
# get a in-component-order queue:
931
for version_id in self.versions():
932
if version_id in requested_versions:
933
data_pos, length = self._index.get_position(version_id)
934
version_id_records.append((version_id, data_pos, length))
936
total = len(version_id_records)
937
for version_idx, (version_id, data, sha_value) in \
938
enumerate(self._data.read_records_iter(version_id_records)):
939
pb.update('Walking content.', version_idx, total)
940
method = self._index.get_method(version_id)
942
assert method in ('fulltext', 'line-delta')
943
if method == 'fulltext':
944
line_iterator = self.factory.get_fulltext_content(data)
946
line_iterator = self.factory.get_linedelta_content(data)
947
for line in line_iterator:
950
pb.update('Walking content.', total, total)
952
def iter_parents(self, version_ids):
953
"""Iterate through the parents for many version ids.
955
:param version_ids: An iterable yielding version_ids.
956
:return: An iterator that yields (version_id, parents). Requested
957
version_ids not present in the versioned file are simply skipped.
958
The order is undefined, allowing for different optimisations in
959
the underlying implementation.
961
version_ids = [osutils.safe_revision_id(version_id) for
962
version_id in version_ids]
963
return self._index.iter_parents(version_ids)
965
def num_versions(self):
966
"""See VersionedFile.num_versions()."""
967
return self._index.num_versions()
969
__len__ = num_versions
971
def annotate_iter(self, version_id):
972
"""See VersionedFile.annotate_iter."""
973
version_id = osutils.safe_revision_id(version_id)
974
content = self._get_content(version_id)
975
for origin, text in content.annotate_iter():
978
def get_parents(self, version_id):
979
"""See VersionedFile.get_parents."""
982
# 52554 calls in 1264 872 internal down from 3674
983
version_id = osutils.safe_revision_id(version_id)
985
return self._index.get_parents(version_id)
987
raise RevisionNotPresent(version_id, self.filename)
989
def get_parents_with_ghosts(self, version_id):
990
"""See VersionedFile.get_parents."""
991
version_id = osutils.safe_revision_id(version_id)
993
return self._index.get_parents_with_ghosts(version_id)
995
raise RevisionNotPresent(version_id, self.filename)
997
def get_ancestry(self, versions, topo_sorted=True):
998
"""See VersionedFile.get_ancestry."""
999
if isinstance(versions, basestring):
1000
versions = [versions]
1003
versions = [osutils.safe_revision_id(v) for v in versions]
1004
return self._index.get_ancestry(versions, topo_sorted)
1006
def get_ancestry_with_ghosts(self, versions):
1007
"""See VersionedFile.get_ancestry_with_ghosts."""
1008
if isinstance(versions, basestring):
1009
versions = [versions]
1012
versions = [osutils.safe_revision_id(v) for v in versions]
1013
return self._index.get_ancestry_with_ghosts(versions)
1015
#@deprecated_method(zero_eight)
1016
def walk(self, version_ids):
1017
"""See VersionedFile.walk."""
1018
# We take the short path here, and extract all relevant texts
1019
# and put them in a weave and let that do all the work. Far
1020
# from optimal, but is much simpler.
1021
# FIXME RB 20060228 this really is inefficient!
1022
from bzrlib.weave import Weave
1024
w = Weave(self.filename)
1025
ancestry = set(self.get_ancestry(version_ids, topo_sorted=False))
1026
sorted_graph = topo_sort(self._index.get_graph())
1027
version_list = [vid for vid in sorted_graph if vid in ancestry]
1029
for version_id in version_list:
1030
lines = self.get_lines(version_id)
1031
w.add_lines(version_id, self.get_parents(version_id), lines)
1033
for lineno, insert_id, dset, line in w.walk(version_ids):
1034
yield lineno, insert_id, dset, line
1036
def plan_merge(self, ver_a, ver_b):
1037
"""See VersionedFile.plan_merge."""
1038
ver_a = osutils.safe_revision_id(ver_a)
1039
ver_b = osutils.safe_revision_id(ver_b)
1040
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1042
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1043
annotated_a = self.annotate(ver_a)
1044
annotated_b = self.annotate(ver_b)
1045
return merge._plan_annotate_merge(annotated_a, annotated_b,
1046
ancestors_a, ancestors_b)
1049
class _KnitComponentFile(object):
1050
"""One of the files used to implement a knit database"""
1052
def __init__(self, transport, filename, mode, file_mode=None,
1053
create_parent_dir=False, dir_mode=None):
1054
self._transport = transport
1055
self._filename = filename
1057
self._file_mode = file_mode
1058
self._dir_mode = dir_mode
1059
self._create_parent_dir = create_parent_dir
1060
self._need_to_create = False
1062
def _full_path(self):
1063
"""Return the full path to this file."""
1064
return self._transport.base + self._filename
1066
def check_header(self, fp):
1067
line = fp.readline()
1069
# An empty file can actually be treated as though the file doesn't
1071
raise errors.NoSuchFile(self._full_path())
1072
if line != self.HEADER:
1073
raise KnitHeaderError(badline=line,
1074
filename=self._transport.abspath(self._filename))
1077
"""Commit is a nop."""
1080
return '%s(%s)' % (self.__class__.__name__, self._filename)
1083
class _KnitIndex(_KnitComponentFile):
1084
"""Manages knit index file.
1086
The index is already kept in memory and read on startup, to enable
1087
fast lookups of revision information. The cursor of the index
1088
file is always pointing to the end, making it easy to append
1091
_cache is a cache for fast mapping from version id to a Index
1094
_history is a cache for fast mapping from indexes to version ids.
1096
The index data format is dictionary compressed when it comes to
1097
parent references; a index entry may only have parents that with a
1098
lover index number. As a result, the index is topological sorted.
1100
Duplicate entries may be written to the index for a single version id
1101
if this is done then the latter one completely replaces the former:
1102
this allows updates to correct version and parent information.
1103
Note that the two entries may share the delta, and that successive
1104
annotations and references MUST point to the first entry.
1106
The index file on disc contains a header, followed by one line per knit
1107
record. The same revision can be present in an index file more than once.
1108
The first occurrence gets assigned a sequence number starting from 0.
1110
The format of a single line is
1111
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1112
REVISION_ID is a utf8-encoded revision id
1113
FLAGS is a comma separated list of flags about the record. Values include
1114
no-eol, line-delta, fulltext.
1115
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1116
that the the compressed data starts at.
1117
LENGTH is the ascii representation of the length of the data file.
1118
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1120
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1121
revision id already in the knit that is a parent of REVISION_ID.
1122
The ' :' marker is the end of record marker.
1125
when a write is interrupted to the index file, it will result in a line
1126
that does not end in ' :'. If the ' :' is not present at the end of a line,
1127
or at the end of the file, then the record that is missing it will be
1128
ignored by the parser.
1130
When writing new records to the index file, the data is preceded by '\n'
1131
to ensure that records always start on new lines even if the last write was
1132
interrupted. As a result its normal for the last line in the index to be
1133
missing a trailing newline. One can be added with no harmful effects.
1136
HEADER = "# bzr knit index 8\n"
1138
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1139
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1141
def _cache_version(self, version_id, options, pos, size, parents):
1142
"""Cache a version record in the history array and index cache.
1144
This is inlined into _load_data for performance. KEEP IN SYNC.
1145
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1148
# only want the _history index to reference the 1st index entry
1150
if version_id not in self._cache:
1151
index = len(self._history)
1152
self._history.append(version_id)
1154
index = self._cache[version_id][5]
1155
self._cache[version_id] = (version_id,
1162
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1163
create_parent_dir=False, delay_create=False, dir_mode=None):
1164
_KnitComponentFile.__init__(self, transport, filename, mode,
1165
file_mode=file_mode,
1166
create_parent_dir=create_parent_dir,
1169
# position in _history is the 'official' index for a revision
1170
# but the values may have come from a newer entry.
1171
# so - wc -l of a knit index is != the number of unique names
1175
fp = self._transport.get(self._filename)
1177
# _load_data may raise NoSuchFile if the target knit is
1179
_load_data(self, fp)
1183
if mode != 'w' or not create:
1186
self._need_to_create = True
1188
self._transport.put_bytes_non_atomic(
1189
self._filename, self.HEADER, mode=self._file_mode)
1191
def get_graph(self):
1192
"""Return a list of the node:parents lists from this knit index."""
1193
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1195
def get_ancestry(self, versions, topo_sorted=True):
1196
"""See VersionedFile.get_ancestry."""
1197
# get a graph of all the mentioned versions:
1199
pending = set(versions)
1202
version = pending.pop()
1205
parents = [p for p in cache[version][4] if p in cache]
1207
raise RevisionNotPresent(version, self._filename)
1208
# if not completed and not a ghost
1209
pending.update([p for p in parents if p not in graph])
1210
graph[version] = parents
1213
return topo_sort(graph.items())
1215
def get_ancestry_with_ghosts(self, versions):
1216
"""See VersionedFile.get_ancestry_with_ghosts."""
1217
# get a graph of all the mentioned versions:
1218
self.check_versions_present(versions)
1221
pending = set(versions)
1223
version = pending.pop()
1225
parents = cache[version][4]
1231
pending.update([p for p in parents if p not in graph])
1232
graph[version] = parents
1233
return topo_sort(graph.items())
1235
def iter_parents(self, version_ids):
1236
"""Iterate through the parents for many version ids.
1238
:param version_ids: An iterable yielding version_ids.
1239
:return: An iterator that yields (version_id, parents). Requested
1240
version_ids not present in the versioned file are simply skipped.
1241
The order is undefined, allowing for different optimisations in
1242
the underlying implementation.
1244
for version_id in version_ids:
1246
yield version_id, tuple(self.get_parents(version_id))
1250
def num_versions(self):
1251
return len(self._history)
1253
__len__ = num_versions
1255
def get_versions(self):
1256
"""Get all the versions in the file. not topologically sorted."""
1257
return self._history
1259
def _version_list_to_index(self, versions):
1262
for version in versions:
1263
if version in cache:
1264
# -- inlined lookup() --
1265
result_list.append(str(cache[version][5]))
1266
# -- end lookup () --
1268
result_list.append('.' + version)
1269
return ' '.join(result_list)
1271
def add_version(self, version_id, options, pos, size, parents):
1272
"""Add a version record to the index."""
1273
self.add_versions(((version_id, options, pos, size, parents),))
1275
def add_versions(self, versions):
1276
"""Add multiple versions to the index.
1278
:param versions: a list of tuples:
1279
(version_id, options, pos, size, parents).
1282
orig_history = self._history[:]
1283
orig_cache = self._cache.copy()
1286
for version_id, options, pos, size, parents in versions:
1287
line = "\n%s %s %s %s %s :" % (version_id,
1291
self._version_list_to_index(parents))
1292
assert isinstance(line, str), \
1293
'content must be utf-8 encoded: %r' % (line,)
1295
self._cache_version(version_id, options, pos, size, parents)
1296
if not self._need_to_create:
1297
self._transport.append_bytes(self._filename, ''.join(lines))
1300
sio.write(self.HEADER)
1301
sio.writelines(lines)
1303
self._transport.put_file_non_atomic(self._filename, sio,
1304
create_parent_dir=self._create_parent_dir,
1305
mode=self._file_mode,
1306
dir_mode=self._dir_mode)
1307
self._need_to_create = False
1309
# If any problems happen, restore the original values and re-raise
1310
self._history = orig_history
1311
self._cache = orig_cache
1314
def has_version(self, version_id):
1315
"""True if the version is in the index."""
1316
return version_id in self._cache
1318
def get_position(self, version_id):
1319
"""Return data position and size of specified version."""
1320
entry = self._cache[version_id]
1321
return 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, version_ids, check_present=False):
1380
"""Get the entries for version_ids."""
1381
version_ids = set(version_ids)
1384
for node in self._graph_index.iter_entries(version_ids):
1386
found_keys.add(node[0])
1388
# adapt parentless index to the rest of the code.
1389
for node in self._graph_index.iter_entries(version_ids):
1390
yield node[0], node[1], ()
1391
found_keys.add(node[0])
1393
missing_keys = version_ids.difference(found_keys)
1395
raise RevisionNotPresent(missing_keys.pop(), self)
1397
def _present_keys(self, version_ids):
1399
node[0] for node in self._get_entries(version_ids)])
1401
def _parentless_ancestry(self, versions):
1402
"""Honour the get_ancestry API for parentless knit indices."""
1403
present_keys = self._present_keys(versions)
1404
missing = set(versions).difference(present_keys)
1406
raise RevisionNotPresent(missing.pop(), self)
1407
return list(present_keys)
1409
def get_ancestry(self, versions, topo_sorted=True):
1410
"""See VersionedFile.get_ancestry."""
1411
if not self._parents:
1412
return self._parentless_ancestry(versions)
1413
# XXX: This will do len(history) index calls - perhaps
1414
# it should be altered to be a index core feature?
1415
# get a graph of all the mentioned versions:
1418
versions = set(versions)
1419
pending = set(versions)
1421
# get all pending nodes
1422
this_iteration = pending
1423
new_nodes = self._get_entries(this_iteration)
1425
for (key, value, node_refs) in new_nodes:
1426
# dont ask for ghosties - otherwise
1427
# we we can end up looping with pending
1428
# being entirely ghosted.
1429
graph[key] = [parent for parent in node_refs[0]
1430
if parent not in ghosts]
1432
pending.update(graph[key])
1433
ghosts.difference_update(graph)
1434
# dont examine known nodes
1435
pending.difference_update(graph)
1436
if versions.difference(graph):
1437
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1440
return topo_sort(graph.items())
1442
def get_ancestry_with_ghosts(self, versions):
1443
"""See VersionedFile.get_ancestry."""
1444
if not self._parents:
1445
return self._parentless_ancestry(versions)
1446
# XXX: This will do len(history) index calls - perhaps
1447
# it should be altered to be a index core feature?
1448
# get a graph of all the mentioned versions:
1450
versions = set(versions)
1451
pending = set(versions)
1453
# get all pending nodes
1454
this_iteration = pending
1455
new_nodes = self._get_entries(this_iteration)
1457
for (key, value, node_refs) in new_nodes:
1458
graph[key] = node_refs[0]
1460
pending.update(graph[key])
1461
missing_versions = this_iteration.difference(graph)
1462
missing_needed = versions.intersection(missing_versions)
1464
raise RevisionNotPresent(missing_needed.pop(), self)
1465
for missing_version in missing_versions:
1466
# add a key, no parents
1467
graph[missing_version] = []
1468
# dont examine known nodes
1469
pending.difference_update(graph)
1470
return topo_sort(graph.items())
1472
def get_graph(self):
1473
"""Return a list of the node:parents lists from this knit index."""
1474
if not self._parents:
1475
return [(key, ()) for key in self.get_versions()]
1476
return [(key, refs[0]) for (key, value, refs) in
1477
self._graph_index.iter_all_entries()]
1479
def iter_parents(self, version_ids):
1480
"""Iterate through the parents for many version ids.
1482
:param version_ids: An iterable yielding version_ids.
1483
:return: An iterator that yields (version_id, parents). Requested
1484
version_ids not present in the versioned file are simply skipped.
1485
The order is undefined, allowing for different optimisations in
1486
the underlying implementation.
1489
all_nodes = set(self._get_entries(version_ids))
1491
present_parents = set()
1492
for node in all_nodes:
1493
all_parents.update(node[2][0])
1494
# any node we are querying must be present
1495
present_parents.add(node[0])
1496
unknown_parents = all_parents.difference(present_parents)
1497
present_parents.update(self._present_keys(unknown_parents))
1498
for node in all_nodes:
1500
for parent in node[2][0]:
1501
if parent in present_parents:
1502
parents.append(parent)
1503
yield node[0], tuple(parents)
1505
for node in self._get_entries(version_ids):
1508
def num_versions(self):
1509
return len(list(self._graph_index.iter_all_entries()))
1511
__len__ = num_versions
1513
def get_versions(self):
1514
"""Get all the versions in the file. not topologically sorted."""
1515
return [node[0] for node in self._graph_index.iter_all_entries()]
1517
def has_version(self, version_id):
1518
"""True if the version is in the index."""
1519
return len(self._present_keys([version_id])) == 1
1521
def get_position(self, version_id):
1522
"""Return data position and size of specified version."""
1523
bits = self._get_node(version_id)[1][1:].split(' ')
1524
return int(bits[0]), int(bits[1])
1526
def get_method(self, version_id):
1527
"""Return compression method of specified version."""
1528
if not self._deltas:
1530
return self._parent_compression(self._get_node(version_id)[2][1])
1532
def _parent_compression(self, reference_list):
1533
# use the second reference list to decide if this is delta'd or not.
1534
if len(reference_list):
1539
def _get_node(self, version_id):
1540
return list(self._get_entries([version_id]))[0]
1542
def get_options(self, version_id):
1543
"""Return a string represention options.
1547
node = self._get_node(version_id)
1548
if not self._deltas:
1549
options = ['fulltext']
1551
options = [self._parent_compression(node[2][1])]
1552
if node[1][0] == 'N':
1553
options.append('no-eol')
1554
return ','.join(options)
1556
def get_parents(self, version_id):
1557
"""Return parents of specified version ignoring ghosts."""
1558
parents = list(self.iter_parents([version_id]))
1561
raise errors.RevisionNotPresent(version_id, self)
1562
return parents[0][1]
1564
def get_parents_with_ghosts(self, version_id):
1565
"""Return parents of specified version with ghosts."""
1566
nodes = list(self._get_entries([version_id], check_present=True))
1567
if not self._parents:
1569
return nodes[0][2][0]
1571
def check_versions_present(self, version_ids):
1572
"""Check that all specified versions are present."""
1573
version_ids = set(version_ids)
1574
present = self._present_keys(version_ids)
1575
missing = version_ids.difference(present)
1577
raise RevisionNotPresent(missing.pop(), self)
1579
def add_version(self, version_id, options, pos, size, parents):
1580
"""Add a version record to the index."""
1581
return self.add_versions(((version_id, options, pos, size, parents),))
1583
def add_versions(self, versions):
1584
"""Add multiple versions to the index.
1586
This function does not insert data into the Immutable GraphIndex
1587
backing the KnitGraphIndex, instead it prepares data for insertion by
1588
the caller and checks that it is safe to insert then calls
1589
self._add_callback with the prepared GraphIndex nodes.
1591
:param versions: a list of tuples:
1592
(version_id, options, pos, size, parents).
1594
if not self._add_callback:
1595
raise errors.ReadOnlyError(self)
1596
# we hope there are no repositories with inconsistent parentage
1601
for (version_id, options, pos, size, parents) in versions:
1602
if 'no-eol' in options:
1606
value += "%d %d" % (pos, size)
1607
if not self._deltas:
1608
if 'line-delta' in options:
1609
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1612
if 'line-delta' in options:
1613
node_refs = (tuple(parents), (parents[0],))
1615
node_refs = (tuple(parents), ())
1617
node_refs = (tuple(parents), )
1620
raise KnitCorrupt(self, "attempt to add node with parents "
1621
"in parentless index.")
1623
keys[version_id] = (value, node_refs)
1624
present_nodes = self._get_entries(keys)
1625
for (key, value, node_refs) in present_nodes:
1626
if (value, node_refs) != keys[key]:
1627
raise KnitCorrupt(self, "inconsistent details in add_versions"
1628
": %s %s" % ((value, node_refs), keys[key]))
1632
for key, (value, node_refs) in keys.iteritems():
1633
result.append((key, value, node_refs))
1635
for key, (value, node_refs) in keys.iteritems():
1636
result.append((key, value))
1637
self._add_callback(result)
1640
class _KnitData(_KnitComponentFile):
1641
"""Contents of the knit data file"""
1643
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1644
create_parent_dir=False, delay_create=False,
1646
_KnitComponentFile.__init__(self, transport, filename, mode,
1647
file_mode=file_mode,
1648
create_parent_dir=create_parent_dir,
1650
self._checked = False
1651
# TODO: jam 20060713 conceptually, this could spill to disk
1652
# if the cached size gets larger than a certain amount
1653
# but it complicates the model a bit, so for now just use
1654
# a simple dictionary
1656
self._do_cache = False
1659
self._need_to_create = create
1661
self._transport.put_bytes_non_atomic(self._filename, '',
1662
mode=self._file_mode)
1664
def enable_cache(self):
1665
"""Enable caching of reads."""
1666
self._do_cache = True
1668
def clear_cache(self):
1669
"""Clear the record cache."""
1670
self._do_cache = False
1673
def _open_file(self):
1675
return self._transport.get(self._filename)
1680
def _record_to_data(self, version_id, digest, lines):
1681
"""Convert version_id, digest, lines into a raw data block.
1683
:return: (len, a StringIO instance with the raw data ready to read.)
1686
data_file = GzipFile(None, mode='wb', fileobj=sio)
1688
assert isinstance(version_id, str)
1689
data_file.writelines(chain(
1690
["version %s %d %s\n" % (version_id,
1694
["end %s\n" % version_id]))
1701
def add_raw_record(self, raw_data):
1702
"""Append a prepared record to the data file.
1704
:return: the offset in the data file raw_data was written.
1706
assert isinstance(raw_data, str), 'data must be plain bytes'
1707
if not self._need_to_create:
1708
return self._transport.append_bytes(self._filename, raw_data)
1710
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1711
create_parent_dir=self._create_parent_dir,
1712
mode=self._file_mode,
1713
dir_mode=self._dir_mode)
1714
self._need_to_create = False
1717
def add_record(self, version_id, digest, lines):
1718
"""Write new text record to disk. Returns the position in the
1719
file where it was written."""
1720
size, sio = self._record_to_data(version_id, digest, lines)
1722
if not self._need_to_create:
1723
start_pos = self._transport.append_file(self._filename, sio)
1725
self._transport.put_file_non_atomic(self._filename, sio,
1726
create_parent_dir=self._create_parent_dir,
1727
mode=self._file_mode,
1728
dir_mode=self._dir_mode)
1729
self._need_to_create = False
1732
self._cache[version_id] = sio.getvalue()
1733
return start_pos, size
1735
def _parse_record_header(self, version_id, raw_data):
1736
"""Parse a record header for consistency.
1738
:return: the header and the decompressor stream.
1739
as (stream, header_record)
1741
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1743
rec = self._check_header(version_id, df.readline())
1744
except Exception, e:
1745
raise KnitCorrupt(self._filename,
1746
"While reading {%s} got %s(%s)"
1747
% (version_id, e.__class__.__name__, str(e)))
1750
def _check_header(self, version_id, line):
1753
raise KnitCorrupt(self._filename,
1754
'unexpected number of elements in record header')
1755
if rec[1] != version_id:
1756
raise KnitCorrupt(self._filename,
1757
'unexpected version, wanted %r, got %r'
1758
% (version_id, rec[1]))
1761
def _parse_record(self, version_id, data):
1763
# 4168 calls in 2880 217 internal
1764
# 4168 calls to _parse_record_header in 2121
1765
# 4168 calls to readlines in 330
1766
df = GzipFile(mode='rb', fileobj=StringIO(data))
1769
record_contents = df.readlines()
1770
except Exception, e:
1771
raise KnitCorrupt(self._filename,
1772
"While reading {%s} got %s(%s)"
1773
% (version_id, e.__class__.__name__, str(e)))
1774
header = record_contents.pop(0)
1775
rec = self._check_header(version_id, header)
1777
last_line = record_contents.pop()
1778
if len(record_contents) != int(rec[2]):
1779
raise KnitCorrupt(self._filename,
1780
'incorrect number of lines %s != %s'
1782
% (len(record_contents), int(rec[2]),
1784
if last_line != 'end %s\n' % rec[1]:
1785
raise KnitCorrupt(self._filename,
1786
'unexpected version end line %r, wanted %r'
1787
% (last_line, version_id))
1789
return record_contents, rec[3]
1791
def read_records_iter_raw(self, records):
1792
"""Read text records from data file and yield raw data.
1794
This unpacks enough of the text record to validate the id is
1795
as expected but thats all.
1797
# setup an iterator of the external records:
1798
# uses readv so nice and fast we hope.
1800
# grab the disk data needed.
1802
# Don't check _cache if it is empty
1803
needed_offsets = [(pos, size) for version_id, pos, size
1805
if version_id not in self._cache]
1807
needed_offsets = [(pos, size) for version_id, pos, size
1810
raw_records = self._transport.readv(self._filename, needed_offsets)
1812
for version_id, pos, size in records:
1813
if version_id in self._cache:
1814
# This data has already been validated
1815
data = self._cache[version_id]
1817
pos, data = raw_records.next()
1819
self._cache[version_id] = data
1821
# validate the header
1822
df, rec = self._parse_record_header(version_id, data)
1824
yield version_id, data
1826
def read_records_iter(self, records):
1827
"""Read text records from data file and yield result.
1829
The result will be returned in whatever is the fastest to read.
1830
Not by the order requested. Also, multiple requests for the same
1831
record will only yield 1 response.
1832
:param records: A list of (version_id, pos, len) entries
1833
:return: Yields (version_id, contents, digest) in the order
1834
read, not the order requested
1840
# Skip records we have alread seen
1841
yielded_records = set()
1842
needed_records = set()
1843
for record in records:
1844
if record[0] in self._cache:
1845
if record[0] in yielded_records:
1847
yielded_records.add(record[0])
1848
data = self._cache[record[0]]
1849
content, digest = self._parse_record(record[0], data)
1850
yield (record[0], content, digest)
1852
needed_records.add(record)
1853
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1855
needed_records = sorted(set(records), key=operator.itemgetter(1))
1857
if not needed_records:
1860
# The transport optimizes the fetching as well
1861
# (ie, reads continuous ranges.)
1862
readv_response = self._transport.readv(self._filename,
1863
[(pos, size) for version_id, pos, size in needed_records])
1865
for (version_id, pos, size), (pos, data) in \
1866
izip(iter(needed_records), readv_response):
1867
content, digest = self._parse_record(version_id, data)
1869
self._cache[version_id] = data
1870
yield version_id, content, digest
1872
def read_records(self, records):
1873
"""Read records into a dictionary."""
1875
for record_id, content, digest in \
1876
self.read_records_iter(records):
1877
components[record_id] = (content, digest)
1881
class InterKnit(InterVersionedFile):
1882
"""Optimised code paths for knit to knit operations."""
1884
_matching_file_from_factory = KnitVersionedFile
1885
_matching_file_to_factory = KnitVersionedFile
1888
def is_compatible(source, target):
1889
"""Be compatible with knits. """
1891
return (isinstance(source, KnitVersionedFile) and
1892
isinstance(target, KnitVersionedFile))
1893
except AttributeError:
1896
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1897
"""See InterVersionedFile.join."""
1898
assert isinstance(self.source, KnitVersionedFile)
1899
assert isinstance(self.target, KnitVersionedFile)
1901
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1906
pb = ui.ui_factory.nested_progress_bar()
1908
version_ids = list(version_ids)
1909
if None in version_ids:
1910
version_ids.remove(None)
1912
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1913
this_versions = set(self.target._index.get_versions())
1914
needed_versions = self.source_ancestry - this_versions
1915
cross_check_versions = self.source_ancestry.intersection(this_versions)
1916
mismatched_versions = set()
1917
for version in cross_check_versions:
1918
# scan to include needed parents.
1919
n1 = set(self.target.get_parents_with_ghosts(version))
1920
n2 = set(self.source.get_parents_with_ghosts(version))
1922
# FIXME TEST this check for cycles being introduced works
1923
# the logic is we have a cycle if in our graph we are an
1924
# ancestor of any of the n2 revisions.
1930
parent_ancestors = self.source.get_ancestry(parent)
1931
if version in parent_ancestors:
1932
raise errors.GraphCycleError([parent, version])
1933
# ensure this parent will be available later.
1934
new_parents = n2.difference(n1)
1935
needed_versions.update(new_parents.difference(this_versions))
1936
mismatched_versions.add(version)
1938
if not needed_versions and not mismatched_versions:
1940
full_list = topo_sort(self.source.get_graph())
1942
version_list = [i for i in full_list if (not self.target.has_version(i)
1943
and i in needed_versions)]
1947
copy_queue_records = []
1949
for version_id in version_list:
1950
options = self.source._index.get_options(version_id)
1951
parents = self.source._index.get_parents_with_ghosts(version_id)
1952
# check that its will be a consistent copy:
1953
for parent in parents:
1954
# if source has the parent, we must :
1955
# * already have it or
1956
# * have it scheduled already
1957
# otherwise we don't care
1958
assert (self.target.has_version(parent) or
1959
parent in copy_set or
1960
not self.source.has_version(parent))
1961
data_pos, data_size = self.source._index.get_position(version_id)
1962
copy_queue_records.append((version_id, data_pos, data_size))
1963
copy_queue.append((version_id, options, parents))
1964
copy_set.add(version_id)
1966
# data suck the join:
1968
total = len(version_list)
1971
for (version_id, raw_data), \
1972
(version_id2, options, parents) in \
1973
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1975
assert version_id == version_id2, 'logic error, inconsistent results'
1977
pb.update("Joining knit", count, total)
1978
raw_records.append((version_id, options, parents, len(raw_data)))
1979
raw_datum.append(raw_data)
1980
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1982
for version in mismatched_versions:
1983
# FIXME RBC 20060309 is this needed?
1984
n1 = set(self.target.get_parents_with_ghosts(version))
1985
n2 = set(self.source.get_parents_with_ghosts(version))
1986
# write a combined record to our history preserving the current
1987
# parents as first in the list
1988
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1989
self.target.fix_parents(version, new_parents)
1995
InterVersionedFile.register_optimiser(InterKnit)
1998
class WeaveToKnit(InterVersionedFile):
1999
"""Optimised code paths for weave to knit operations."""
2001
_matching_file_from_factory = bzrlib.weave.WeaveFile
2002
_matching_file_to_factory = KnitVersionedFile
2005
def is_compatible(source, target):
2006
"""Be compatible with weaves to knits."""
2008
return (isinstance(source, bzrlib.weave.Weave) and
2009
isinstance(target, KnitVersionedFile))
2010
except AttributeError:
2013
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2014
"""See InterVersionedFile.join."""
2015
assert isinstance(self.source, bzrlib.weave.Weave)
2016
assert isinstance(self.target, KnitVersionedFile)
2018
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2023
pb = ui.ui_factory.nested_progress_bar()
2025
version_ids = list(version_ids)
2027
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2028
this_versions = set(self.target._index.get_versions())
2029
needed_versions = self.source_ancestry - this_versions
2030
cross_check_versions = self.source_ancestry.intersection(this_versions)
2031
mismatched_versions = set()
2032
for version in cross_check_versions:
2033
# scan to include needed parents.
2034
n1 = set(self.target.get_parents_with_ghosts(version))
2035
n2 = set(self.source.get_parents(version))
2036
# if all of n2's parents are in n1, then its fine.
2037
if n2.difference(n1):
2038
# FIXME TEST this check for cycles being introduced works
2039
# the logic is we have a cycle if in our graph we are an
2040
# ancestor of any of the n2 revisions.
2046
parent_ancestors = self.source.get_ancestry(parent)
2047
if version in parent_ancestors:
2048
raise errors.GraphCycleError([parent, version])
2049
# ensure this parent will be available later.
2050
new_parents = n2.difference(n1)
2051
needed_versions.update(new_parents.difference(this_versions))
2052
mismatched_versions.add(version)
2054
if not needed_versions and not mismatched_versions:
2056
full_list = topo_sort(self.source.get_graph())
2058
version_list = [i for i in full_list if (not self.target.has_version(i)
2059
and i in needed_versions)]
2063
total = len(version_list)
2064
for version_id in version_list:
2065
pb.update("Converting to knit", count, total)
2066
parents = self.source.get_parents(version_id)
2067
# check that its will be a consistent copy:
2068
for parent in parents:
2069
# if source has the parent, we must already have it
2070
assert (self.target.has_version(parent))
2071
self.target.add_lines(
2072
version_id, parents, self.source.get_lines(version_id))
2075
for version in mismatched_versions:
2076
# FIXME RBC 20060309 is this needed?
2077
n1 = set(self.target.get_parents_with_ghosts(version))
2078
n2 = set(self.source.get_parents(version))
2079
# write a combined record to our history preserving the current
2080
# parents as first in the list
2081
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2082
self.target.fix_parents(version, new_parents)
2088
InterVersionedFile.register_optimiser(WeaveToKnit)
2091
class KnitSequenceMatcher(difflib.SequenceMatcher):
2092
"""Knit tuned sequence matcher.
2094
This is based on profiling of difflib which indicated some improvements
2095
for our usage pattern.
2098
def find_longest_match(self, alo, ahi, blo, bhi):
2099
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
2101
If isjunk is not defined:
2103
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
2104
alo <= i <= i+k <= ahi
2105
blo <= j <= j+k <= bhi
2106
and for all (i',j',k') meeting those conditions,
2109
and if i == i', j <= j'
2111
In other words, of all maximal matching blocks, return one that
2112
starts earliest in a, and of all those maximal matching blocks that
2113
start earliest in a, return the one that starts earliest in b.
2115
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
2116
>>> s.find_longest_match(0, 5, 0, 9)
2119
If isjunk is defined, first the longest matching block is
2120
determined as above, but with the additional restriction that no
2121
junk element appears in the block. Then that block is extended as
2122
far as possible by matching (only) junk elements on both sides. So
2123
the resulting block never matches on junk except as identical junk
2124
happens to be adjacent to an "interesting" match.
2126
Here's the same example as before, but considering blanks to be
2127
junk. That prevents " abcd" from matching the " abcd" at the tail
2128
end of the second sequence directly. Instead only the "abcd" can
2129
match, and matches the leftmost "abcd" in the second sequence:
2131
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
2132
>>> s.find_longest_match(0, 5, 0, 9)
2135
If no blocks match, return (alo, blo, 0).
2137
>>> s = SequenceMatcher(None, "ab", "c")
2138
>>> s.find_longest_match(0, 2, 0, 1)
2142
# CAUTION: stripping common prefix or suffix would be incorrect.
2146
# Longest matching block is "ab", but if common prefix is
2147
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
2148
# strip, so ends up claiming that ab is changed to acab by
2149
# inserting "ca" in the middle. That's minimal but unintuitive:
2150
# "it's obvious" that someone inserted "ac" at the front.
2151
# Windiff ends up at the same place as diff, but by pairing up
2152
# the unique 'b's and then matching the first two 'a's.
2154
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
2155
besti, bestj, bestsize = alo, blo, 0
2156
# find longest junk-free match
2157
# during an iteration of the loop, j2len[j] = length of longest
2158
# junk-free match ending with a[i-1] and b[j]
2162
for i in xrange(alo, ahi):
2163
# look at all instances of a[i] in b; note that because
2164
# b2j has no junk keys, the loop is skipped if a[i] is junk
2165
j2lenget = j2len.get
2168
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
2169
# following improvement
2170
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
2171
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
2172
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
2174
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
2175
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
2176
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
2188
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
2190
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
2193
# Extend the best by non-junk elements on each end. In particular,
2194
# "popular" non-junk elements aren't in b2j, which greatly speeds
2195
# the inner loop above, but also means "the best" match so far
2196
# doesn't contain any junk *or* popular non-junk elements.
2197
while besti > alo and bestj > blo and \
2198
not isbjunk(b[bestj-1]) and \
2199
a[besti-1] == b[bestj-1]:
2200
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2201
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2202
not isbjunk(b[bestj+bestsize]) and \
2203
a[besti+bestsize] == b[bestj+bestsize]:
2206
# Now that we have a wholly interesting match (albeit possibly
2207
# empty!), we may as well suck up the matching junk on each
2208
# side of it too. Can't think of a good reason not to, and it
2209
# saves post-processing the (possibly considerable) expense of
2210
# figuring out what to do with it. In the case of an empty
2211
# interesting match, this is clearly the right thing to do,
2212
# because no other kind of match is possible in the regions.
2213
while besti > alo and bestj > blo and \
2214
isbjunk(b[bestj-1]) and \
2215
a[besti-1] == b[bestj-1]:
2216
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2217
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2218
isbjunk(b[bestj+bestsize]) and \
2219
a[besti+bestsize] == b[bestj+bestsize]:
2220
bestsize = bestsize + 1
2222
return besti, bestj, bestsize
2226
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2228
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
