1
# Copyright (C) 2005, 2006 by Canonical Ltd
2
# Written by Martin Pool.
3
# Modified by Johan Rydberg <jrydberg@gnu.org>
4
# Modified by Robert Collins <robert.collins@canonical.com>
5
# Modified by Aaron Bentley <aaron.bentley@utoronto.ca>
7
# This program is free software; you can redistribute it and/or modify
8
# it under the terms of the GNU General Public License as published by
9
# the Free Software Foundation; either version 2 of the License, or
10
# (at your option) any later version.
12
# This program is distributed in the hope that it will be useful,
13
# but WITHOUT ANY WARRANTY; without even the implied warranty of
14
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15
# GNU General Public License for more details.
17
# You should have received a copy of the GNU General Public License
18
# along with this program; if not, write to the Free Software
19
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21
"""Knit versionedfile implementation.
23
A knit is a versioned file implementation that supports efficient append only
27
lifeless: the data file is made up of "delta records". each delta record has a delta header
28
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
29
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
30
end-marker; simply "end VERSION"
32
delta can be line or full contents.a
33
... the 8's there are the index number of the annotation.
34
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
38
8 e.set('executable', 'yes')
40
8 if elt.get('executable') == 'yes':
41
8 ie.executable = True
42
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
46
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
47
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
48
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
49
09:33 < lifeless> right
50
09:33 < jrydberg> lifeless: the position and size is the range in the data file
53
so the index sequence is the dictionary compressed sequence number used
54
in the deltas to provide line annotation
59
# 10:16 < lifeless> make partial index writes safe
60
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
61
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
63
# move sha1 out of the content so that join is faster at verifying parents
64
# record content length ?
68
from cStringIO import StringIO
70
from itertools import izip, chain
81
from bzrlib.errors import FileExists, NoSuchFile, KnitError, \
82
InvalidRevisionId, KnitCorrupt, KnitHeaderError, \
83
RevisionNotPresent, RevisionAlreadyPresent
84
from bzrlib.tuned_gzip import GzipFile
85
from bzrlib.trace import mutter
86
from bzrlib.osutils import contains_whitespace, contains_linebreaks, \
88
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
89
from bzrlib.tsort import topo_sort
91
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
94
# TODO: Split out code specific to this format into an associated object.
96
# TODO: Can we put in some kind of value to check that the index and data
97
# files belong together?
99
# TODO: accommodate binaries, perhaps by storing a byte count
101
# TODO: function to check whole file
103
# TODO: atomically append data, then measure backwards from the cursor
104
# position after writing to work out where it was located. we may need to
105
# bypass python file buffering.
107
DATA_SUFFIX = '.knit'
108
INDEX_SUFFIX = '.kndx'
111
class KnitContent(object):
112
"""Content of a knit version to which deltas can be applied."""
114
def __init__(self, lines):
117
def annotate_iter(self):
118
"""Yield tuples of (origin, text) for each content line."""
119
for origin, text in self._lines:
123
"""Return a list of (origin, text) tuples."""
124
return list(self.annotate_iter())
126
def line_delta_iter(self, new_lines):
127
"""Generate line-based delta from this content to new_lines."""
128
new_texts = [text for origin, text in new_lines._lines]
129
old_texts = [text for origin, text in self._lines]
130
s = KnitSequenceMatcher(None, old_texts, new_texts)
131
for op in s.get_opcodes():
134
# ofrom oto length data
135
yield (op[1], op[2], op[4]-op[3], new_lines._lines[op[3]:op[4]])
137
def line_delta(self, new_lines):
138
return list(self.line_delta_iter(new_lines))
141
return [text for origin, text in self._lines]
144
return KnitContent(self._lines[:])
147
class _KnitFactory(object):
148
"""Base factory for creating content objects."""
150
def make(self, lines, version):
151
num_lines = len(lines)
152
return KnitContent(zip([version] * num_lines, lines))
155
class KnitAnnotateFactory(_KnitFactory):
156
"""Factory for creating annotated Content objects."""
160
def parse_fulltext(self, content, version):
161
"""Convert fulltext to internal representation
163
fulltext content is of the format
164
revid(utf8) plaintext\n
165
internal representation is of the format:
168
decode_utf8 = cache_utf8.decode
171
origin, text = line.split(' ', 1)
172
lines.append((decode_utf8(origin), text))
173
return KnitContent(lines)
175
def parse_line_delta_iter(self, lines):
176
for result_item in self.parse_line_delta[lines]:
179
def parse_line_delta(self, lines, version):
180
"""Convert a line based delta into internal representation.
182
line delta is in the form of:
183
intstart intend intcount
185
revid(utf8) newline\n
186
internal representation is
187
(start, end, count, [1..count tuples (revid, newline)])
189
decode_utf8 = cache_utf8.decode
193
# walk through the lines parsing.
195
start, end, count = [int(n) for n in header.split(',')]
199
origin, text = next().split(' ', 1)
201
contents.append((decode_utf8(origin), text))
202
result.append((start, end, count, contents))
205
def lower_fulltext(self, content):
206
"""convert a fulltext content record into a serializable form.
208
see parse_fulltext which this inverts.
210
encode_utf8 = cache_utf8.encode
211
return ['%s %s' % (encode_utf8(o), t) for o, t in content._lines]
213
def lower_line_delta(self, delta):
214
"""convert a delta into a serializable form.
216
See parse_line_delta which this inverts.
218
encode_utf8 = cache_utf8.encode
220
for start, end, c, lines in delta:
221
out.append('%d,%d,%d\n' % (start, end, c))
222
out.extend(encode_utf8(origin) + ' ' + text
223
for origin, text in lines)
227
class KnitPlainFactory(_KnitFactory):
228
"""Factory for creating plain Content objects."""
232
def parse_fulltext(self, content, version):
233
"""This parses an unannotated fulltext.
235
Note that this is not a noop - the internal representation
236
has (versionid, line) - its just a constant versionid.
238
return self.make(content, version)
240
def parse_line_delta_iter(self, lines, version):
242
header = lines.pop(0)
243
start, end, c = [int(n) for n in header.split(',')]
244
yield start, end, c, zip([version] * c, lines[:c])
247
def parse_line_delta(self, lines, version):
248
return list(self.parse_line_delta_iter(lines, version))
250
def lower_fulltext(self, content):
251
return content.text()
253
def lower_line_delta(self, delta):
255
for start, end, c, lines in delta:
256
out.append('%d,%d,%d\n' % (start, end, c))
257
out.extend([text for origin, text in lines])
261
def make_empty_knit(transport, relpath):
262
"""Construct a empty knit at the specified location."""
263
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
267
class KnitVersionedFile(VersionedFile):
268
"""Weave-like structure with faster random access.
270
A knit stores a number of texts and a summary of the relationships
271
between them. Texts are identified by a string version-id. Texts
272
are normally stored and retrieved as a series of lines, but can
273
also be passed as single strings.
275
Lines are stored with the trailing newline (if any) included, to
276
avoid special cases for files with no final newline. Lines are
277
composed of 8-bit characters, not unicode. The combination of
278
these approaches should mean any 'binary' file can be safely
279
stored and retrieved.
282
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
283
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
284
create=False, create_parent_dir=False, delay_create=False,
286
"""Construct a knit at location specified by relpath.
288
:param create: If not True, only open an existing knit.
289
:param create_parent_dir: If True, create the parent directory if
290
creating the file fails. (This is used for stores with
291
hash-prefixes that may not exist yet)
292
:param delay_create: The calling code is aware that the knit won't
293
actually be created until the first data is stored.
295
if deprecated_passed(basis_knit):
296
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
297
" deprecated as of bzr 0.9.",
298
DeprecationWarning, stacklevel=2)
299
if access_mode is None:
301
super(KnitVersionedFile, self).__init__(access_mode)
302
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
303
self.transport = transport
304
self.filename = relpath
305
self.factory = factory or KnitAnnotateFactory()
306
self.writable = (access_mode == 'w')
309
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
310
access_mode, create=create, file_mode=file_mode,
311
create_parent_dir=create_parent_dir, delay_create=delay_create,
313
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
314
access_mode, create=create and not len(self), file_mode=file_mode,
315
create_parent_dir=create_parent_dir, delay_create=delay_create,
319
return '%s(%s)' % (self.__class__.__name__,
320
self.transport.abspath(self.filename))
322
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
323
"""See VersionedFile._add_delta()."""
324
self._check_add(version_id, []) # should we check the lines ?
325
self._check_versions_present(parents)
329
for parent in parents:
330
if not self.has_version(parent):
331
ghosts.append(parent)
333
present_parents.append(parent)
335
if delta_parent is None:
336
# reconstitute as full text.
337
assert len(delta) == 1 or len(delta) == 0
339
assert delta[0][0] == 0
340
assert delta[0][1] == 0, delta[0][1]
341
return super(KnitVersionedFile, self)._add_delta(version_id,
352
options.append('no-eol')
354
if delta_parent is not None:
355
# determine the current delta chain length.
356
# To speed the extract of texts the delta chain is limited
357
# to a fixed number of deltas. This should minimize both
358
# I/O and the time spend applying deltas.
360
delta_parents = [delta_parent]
362
parent = delta_parents[0]
363
method = self._index.get_method(parent)
364
if method == 'fulltext':
366
delta_parents = self._index.get_parents(parent)
368
if method == 'line-delta':
369
# did not find a fulltext in the delta limit.
370
# just do a normal insertion.
371
return super(KnitVersionedFile, self)._add_delta(version_id,
378
options.append('line-delta')
379
store_lines = self.factory.lower_line_delta(delta)
381
where, size = self._data.add_record(version_id, digest, store_lines)
382
self._index.add_version(version_id, options, where, size, parents)
384
def _add_raw_records(self, records, data):
385
"""Add all the records 'records' with data pre-joined in 'data'.
387
:param records: A list of tuples(version_id, options, parents, size).
388
:param data: The data for the records. When it is written, the records
389
are adjusted to have pos pointing into data by the sum of
390
the preceding records sizes.
393
pos = self._data.add_raw_record(data)
396
for (version_id, options, parents, size) in records:
397
index_entries.append((version_id, options, pos+offset,
399
if self._data._do_cache:
400
self._data._cache[version_id] = data[offset:offset+size]
402
self._index.add_versions(index_entries)
404
def enable_cache(self):
405
"""Start caching data for this knit"""
406
self._data.enable_cache()
408
def clear_cache(self):
409
"""Clear the data cache only."""
410
self._data.clear_cache()
412
def copy_to(self, name, transport):
413
"""See VersionedFile.copy_to()."""
414
# copy the current index to a temp index to avoid racing with local
416
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
417
self.transport.get(self._index._filename))
419
f = self._data._open_file()
421
transport.put_file(name + DATA_SUFFIX, f)
424
# move the copied index into place
425
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
427
def create_empty(self, name, transport, mode=None):
428
return KnitVersionedFile(name, transport, factory=self.factory,
429
delta=self.delta, create=True)
431
def _fix_parents(self, version, new_parents):
432
"""Fix the parents list for version.
434
This is done by appending a new version to the index
435
with identical data except for the parents list.
436
the parents list must be a superset of the current
439
current_values = self._index._cache[version]
440
assert set(current_values[4]).difference(set(new_parents)) == set()
441
self._index.add_version(version,
447
def get_delta(self, version_id):
448
"""Get a delta for constructing version from some other version."""
449
if not self.has_version(version_id):
450
raise RevisionNotPresent(version_id, self.filename)
452
parents = self.get_parents(version_id)
457
data_pos, data_size = self._index.get_position(version_id)
458
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
459
version_idx = self._index.lookup(version_id)
460
noeol = 'no-eol' in self._index.get_options(version_id)
461
if 'fulltext' == self._index.get_method(version_id):
462
new_content = self.factory.parse_fulltext(data, version_idx)
463
if parent is not None:
464
reference_content = self._get_content(parent)
465
old_texts = reference_content.text()
468
new_texts = new_content.text()
469
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
470
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
472
delta = self.factory.parse_line_delta(data, version_idx)
473
return parent, sha1, noeol, delta
475
def get_graph_with_ghosts(self):
476
"""See VersionedFile.get_graph_with_ghosts()."""
477
graph_items = self._index.get_graph()
478
return dict(graph_items)
480
def get_sha1s(self, version_ids):
481
"""Return a map of sha1 digests for the specified versions"""
482
record_map = self._get_record_map(version_ids)
483
return dict((v, record_map[v][2]) for v in version_ids)
485
def get_sha1(self, version_id):
486
"""See VersionedFile.get_sha1()."""
487
return self.get_sha1s([version_id])[version_id]
491
"""See VersionedFile.get_suffixes()."""
492
return [DATA_SUFFIX, INDEX_SUFFIX]
494
def has_ghost(self, version_id):
495
"""True if there is a ghost reference in the file to version_id."""
497
if self.has_version(version_id):
499
# optimisable if needed by memoising the _ghosts set.
500
items = self._index.get_graph()
501
for node, parents in items:
502
for parent in parents:
503
if parent not in self._index._cache:
504
if parent == version_id:
509
"""See VersionedFile.versions."""
510
return self._index.get_versions()
512
def has_version(self, version_id):
513
"""See VersionedFile.has_version."""
514
return self._index.has_version(version_id)
516
__contains__ = has_version
518
def _merge_annotations(self, content, parents, parent_texts={},
519
delta=None, annotated=None):
520
"""Merge annotations for content. This is done by comparing
521
the annotations based on changed to the text.
525
for parent_id in parents:
526
merge_content = self._get_content(parent_id, parent_texts)
527
seq = KnitSequenceMatcher(None, merge_content.text(), content.text())
528
if delta_seq is None:
529
# setup a delta seq to reuse.
531
for i, j, n in seq.get_matching_blocks():
534
# this appears to copy (origin, text) pairs across to the new
535
# content for any line that matches the last-checked parent.
536
# FIXME: save the sequence control data for delta compression
537
# against the most relevant parent rather than rediffing.
538
content._lines[j:j+n] = merge_content._lines[i:i+n]
541
reference_content = self._get_content(parents[0], parent_texts)
542
new_texts = content.text()
543
old_texts = reference_content.text()
544
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
545
return self._make_line_delta(delta_seq, content)
547
def _make_line_delta(self, delta_seq, new_content):
548
"""Generate a line delta from delta_seq and new_content."""
550
for op in delta_seq.get_opcodes():
553
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
556
def _get_components_positions(self, version_ids):
557
"""Produce a map of position data for the components of versions.
559
This data is intended to be used for retrieving the knit records.
561
A dict of version_id to (method, data_pos, data_size, next) is
563
method is the way referenced data should be applied.
564
data_pos is the position of the data in the knit.
565
data_size is the size of the data in the knit.
566
next is the build-parent of the version, or None for fulltexts.
569
for version_id in version_ids:
572
while cursor is not None and cursor not in component_data:
573
method = self._index.get_method(cursor)
574
if method == 'fulltext':
577
next = self.get_parents(cursor)[0]
578
data_pos, data_size = self._index.get_position(cursor)
579
component_data[cursor] = (method, data_pos, data_size, next)
581
return component_data
583
def _get_content(self, version_id, parent_texts={}):
584
"""Returns a content object that makes up the specified
586
if not self.has_version(version_id):
587
raise RevisionNotPresent(version_id, self.filename)
589
cached_version = parent_texts.get(version_id, None)
590
if cached_version is not None:
591
return cached_version
593
text_map, contents_map = self._get_content_maps([version_id])
594
return contents_map[version_id]
596
def _check_versions_present(self, version_ids):
597
"""Check that all specified versions are present."""
598
version_ids = set(version_ids)
599
for r in list(version_ids):
600
if self._index.has_version(r):
601
version_ids.remove(r)
603
raise RevisionNotPresent(list(version_ids)[0], self.filename)
605
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
606
"""See VersionedFile.add_lines_with_ghosts()."""
607
self._check_add(version_id, lines)
608
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
610
def _add_lines(self, version_id, parents, lines, parent_texts):
611
"""See VersionedFile.add_lines."""
612
self._check_add(version_id, lines)
613
self._check_versions_present(parents)
614
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
616
def _check_add(self, version_id, lines):
617
"""check that version_id and lines are safe to add."""
618
assert self.writable, "knit is not opened for write"
619
### FIXME escape. RBC 20060228
620
if contains_whitespace(version_id):
621
raise InvalidRevisionId(version_id, self.filename)
622
if self.has_version(version_id):
623
raise RevisionAlreadyPresent(version_id, self.filename)
624
self._check_lines_not_unicode(lines)
625
self._check_lines_are_lines(lines)
627
def _add(self, version_id, lines, parents, delta, parent_texts):
628
"""Add a set of lines on top of version specified by parents.
630
If delta is true, compress the text as a line-delta against
633
Any versions not present will be converted into ghosts.
635
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
636
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
637
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
638
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
639
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
640
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
641
# +1383 0 8.0370 8.0370 +<len>
642
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
643
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
644
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
645
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
649
if parent_texts is None:
651
for parent in parents:
652
if not self.has_version(parent):
653
ghosts.append(parent)
655
present_parents.append(parent)
657
if delta and not len(present_parents):
660
digest = sha_strings(lines)
663
if lines[-1][-1] != '\n':
664
options.append('no-eol')
665
lines[-1] = lines[-1] + '\n'
667
if len(present_parents) and delta:
668
# To speed the extract of texts the delta chain is limited
669
# to a fixed number of deltas. This should minimize both
670
# I/O and the time spend applying deltas.
672
delta_parents = present_parents
674
parent = delta_parents[0]
675
method = self._index.get_method(parent)
676
if method == 'fulltext':
678
delta_parents = self._index.get_parents(parent)
680
if method == 'line-delta':
683
lines = self.factory.make(lines, version_id)
684
if delta or (self.factory.annotated and len(present_parents) > 0):
685
# Merge annotations from parent texts if so is needed.
686
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
687
delta, self.factory.annotated)
690
options.append('line-delta')
691
store_lines = self.factory.lower_line_delta(delta_hunks)
693
options.append('fulltext')
694
store_lines = self.factory.lower_fulltext(lines)
696
where, size = self._data.add_record(version_id, digest, store_lines)
697
self._index.add_version(version_id, options, where, size, parents)
700
def check(self, progress_bar=None):
701
"""See VersionedFile.check()."""
703
def _clone_text(self, new_version_id, old_version_id, parents):
704
"""See VersionedFile.clone_text()."""
705
# FIXME RBC 20060228 make fast by only inserting an index with null
707
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
709
def get_lines(self, version_id):
710
"""See VersionedFile.get_lines()."""
711
return self.get_line_list([version_id])[0]
713
def _get_record_map(self, version_ids):
714
"""Produce a dictionary of knit records.
716
The keys are version_ids, the values are tuples of (method, content,
718
method is the way the content should be applied.
719
content is a KnitContent object.
720
digest is the SHA1 digest of this version id after all steps are done
721
next is the build-parent of the version, i.e. the leftmost ancestor.
722
If the method is fulltext, next will be None.
724
position_map = self._get_components_positions(version_ids)
725
# c = component_id, m = method, p = position, s = size, n = next
726
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
728
for component_id, content, digest in \
729
self._data.read_records_iter(records):
730
method, position, size, next = position_map[component_id]
731
record_map[component_id] = method, content, digest, next
735
def get_text(self, version_id):
736
"""See VersionedFile.get_text"""
737
return self.get_texts([version_id])[0]
739
def get_texts(self, version_ids):
740
return [''.join(l) for l in self.get_line_list(version_ids)]
742
def get_line_list(self, version_ids):
743
"""Return the texts of listed versions as a list of strings."""
744
text_map, content_map = self._get_content_maps(version_ids)
745
return [text_map[v] for v in version_ids]
747
def _get_content_maps(self, version_ids):
748
"""Produce maps of text and KnitContents
750
:return: (text_map, content_map) where text_map contains the texts for
751
the requested versions and content_map contains the KnitContents.
752
Both dicts take version_ids as their keys.
754
for version_id in version_ids:
755
if not self.has_version(version_id):
756
raise RevisionNotPresent(version_id, self.filename)
757
record_map = self._get_record_map(version_ids)
762
for version_id in version_ids:
765
while cursor is not None:
766
method, data, digest, next = record_map[cursor]
767
components.append((cursor, method, data, digest))
768
if cursor in content_map:
773
for component_id, method, data, digest in reversed(components):
774
if component_id in content_map:
775
content = content_map[component_id]
777
version_idx = self._index.lookup(component_id)
778
if method == 'fulltext':
779
assert content is None
780
content = self.factory.parse_fulltext(data, version_idx)
781
elif method == 'line-delta':
782
delta = self.factory.parse_line_delta(data[:],
784
content = content.copy()
785
content._lines = self._apply_delta(content._lines,
787
content_map[component_id] = content
789
if 'no-eol' in self._index.get_options(version_id):
790
content = content.copy()
791
line = content._lines[-1][1].rstrip('\n')
792
content._lines[-1] = (content._lines[-1][0], line)
793
final_content[version_id] = content
795
# digest here is the digest from the last applied component.
796
text = content.text()
797
if sha_strings(text) != digest:
798
raise KnitCorrupt(self.filename,
799
'sha-1 does not match %s' % version_id)
801
text_map[version_id] = text
802
return text_map, final_content
804
def iter_lines_added_or_present_in_versions(self, version_ids=None):
805
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
806
if version_ids is None:
807
version_ids = self.versions()
808
# we don't care about inclusions, the caller cares.
809
# but we need to setup a list of records to visit.
810
# we need version_id, position, length
811
version_id_records = []
812
requested_versions = list(version_ids)
813
# filter for available versions
814
for version_id in requested_versions:
815
if not self.has_version(version_id):
816
raise RevisionNotPresent(version_id, self.filename)
817
# get a in-component-order queue:
819
for version_id in self.versions():
820
if version_id in requested_versions:
821
version_ids.append(version_id)
822
data_pos, length = self._index.get_position(version_id)
823
version_id_records.append((version_id, data_pos, length))
825
pb = bzrlib.ui.ui_factory.nested_progress_bar()
827
total = len(version_id_records)
829
pb.update('Walking content.', count, total)
830
for version_id, data, sha_value in \
831
self._data.read_records_iter(version_id_records):
832
pb.update('Walking content.', count, total)
833
method = self._index.get_method(version_id)
834
version_idx = self._index.lookup(version_id)
835
assert method in ('fulltext', 'line-delta')
836
if method == 'fulltext':
837
content = self.factory.parse_fulltext(data, version_idx)
838
for line in content.text():
841
delta = self.factory.parse_line_delta(data, version_idx)
842
for start, end, count, lines in delta:
843
for origin, line in lines:
846
pb.update('Walking content.', total, total)
849
pb.update('Walking content.', total, total)
853
def num_versions(self):
854
"""See VersionedFile.num_versions()."""
855
return self._index.num_versions()
857
__len__ = num_versions
859
def annotate_iter(self, version_id):
860
"""See VersionedFile.annotate_iter."""
861
content = self._get_content(version_id)
862
for origin, text in content.annotate_iter():
865
def get_parents(self, version_id):
866
"""See VersionedFile.get_parents."""
869
# 52554 calls in 1264 872 internal down from 3674
871
return self._index.get_parents(version_id)
873
raise RevisionNotPresent(version_id, self.filename)
875
def get_parents_with_ghosts(self, version_id):
876
"""See VersionedFile.get_parents."""
878
return self._index.get_parents_with_ghosts(version_id)
880
raise RevisionNotPresent(version_id, self.filename)
882
def get_ancestry(self, versions):
883
"""See VersionedFile.get_ancestry."""
884
if isinstance(versions, basestring):
885
versions = [versions]
888
self._check_versions_present(versions)
889
return self._index.get_ancestry(versions)
891
def get_ancestry_with_ghosts(self, versions):
892
"""See VersionedFile.get_ancestry_with_ghosts."""
893
if isinstance(versions, basestring):
894
versions = [versions]
897
self._check_versions_present(versions)
898
return self._index.get_ancestry_with_ghosts(versions)
900
#@deprecated_method(zero_eight)
901
def walk(self, version_ids):
902
"""See VersionedFile.walk."""
903
# We take the short path here, and extract all relevant texts
904
# and put them in a weave and let that do all the work. Far
905
# from optimal, but is much simpler.
906
# FIXME RB 20060228 this really is inefficient!
907
from bzrlib.weave import Weave
909
w = Weave(self.filename)
910
ancestry = self.get_ancestry(version_ids)
911
sorted_graph = topo_sort(self._index.get_graph())
912
version_list = [vid for vid in sorted_graph if vid in ancestry]
914
for version_id in version_list:
915
lines = self.get_lines(version_id)
916
w.add_lines(version_id, self.get_parents(version_id), lines)
918
for lineno, insert_id, dset, line in w.walk(version_ids):
919
yield lineno, insert_id, dset, line
921
def plan_merge(self, ver_a, ver_b):
922
"""See VersionedFile.plan_merge."""
923
ancestors_b = set(self.get_ancestry(ver_b))
924
def status_a(revision, text):
925
if revision in ancestors_b:
926
return 'killed-b', text
930
ancestors_a = set(self.get_ancestry(ver_a))
931
def status_b(revision, text):
932
if revision in ancestors_a:
933
return 'killed-a', text
937
annotated_a = self.annotate(ver_a)
938
annotated_b = self.annotate(ver_b)
939
plain_a = [t for (a, t) in annotated_a]
940
plain_b = [t for (a, t) in annotated_b]
941
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
944
for ai, bi, l in blocks:
945
# process all mismatched sections
946
# (last mismatched section is handled because blocks always
947
# includes a 0-length last block)
948
for revision, text in annotated_a[a_cur:ai]:
949
yield status_a(revision, text)
950
for revision, text in annotated_b[b_cur:bi]:
951
yield status_b(revision, text)
953
# and now the matched section
956
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
957
assert text_a == text_b
958
yield "unchanged", text_a
961
class _KnitComponentFile(object):
962
"""One of the files used to implement a knit database"""
964
def __init__(self, transport, filename, mode, file_mode=None,
965
create_parent_dir=False, dir_mode=None):
966
self._transport = transport
967
self._filename = filename
969
self._file_mode = file_mode
970
self._dir_mode = dir_mode
971
self._create_parent_dir = create_parent_dir
972
self._need_to_create = False
974
def check_header(self, fp):
976
if line != self.HEADER:
977
raise KnitHeaderError(badline=line)
980
"""Commit is a nop."""
983
return '%s(%s)' % (self.__class__.__name__, self._filename)
986
class _KnitIndex(_KnitComponentFile):
987
"""Manages knit index file.
989
The index is already kept in memory and read on startup, to enable
990
fast lookups of revision information. The cursor of the index
991
file is always pointing to the end, making it easy to append
994
_cache is a cache for fast mapping from version id to a Index
997
_history is a cache for fast mapping from indexes to version ids.
999
The index data format is dictionary compressed when it comes to
1000
parent references; a index entry may only have parents that with a
1001
lover index number. As a result, the index is topological sorted.
1003
Duplicate entries may be written to the index for a single version id
1004
if this is done then the latter one completely replaces the former:
1005
this allows updates to correct version and parent information.
1006
Note that the two entries may share the delta, and that successive
1007
annotations and references MUST point to the first entry.
1009
The index file on disc contains a header, followed by one line per knit
1010
record. The same revision can be present in an index file more than once.
1011
The first occurrence gets assigned a sequence number starting from 0.
1013
The format of a single line is
1014
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1015
REVISION_ID is a utf8-encoded revision id
1016
FLAGS is a comma separated list of flags about the record. Values include
1017
no-eol, line-delta, fulltext.
1018
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1019
that the the compressed data starts at.
1020
LENGTH is the ascii representation of the length of the data file.
1021
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1023
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1024
revision id already in the knit that is a parent of REVISION_ID.
1025
The ' :' marker is the end of record marker.
1028
when a write is interrupted to the index file, it will result in a line that
1029
does not end in ' :'. If the ' :' is not present at the end of a line, or at
1030
the end of the file, then the record that is missing it will be ignored by
1033
When writing new records to the index file, the data is preceded by '\n'
1034
to ensure that records always start on new lines even if the last write was
1035
interrupted. As a result its normal for the last line in the index to be
1036
missing a trailing newline. One can be added with no harmful effects.
1039
HEADER = "# bzr knit index 8\n"
1041
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1042
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1044
def _cache_version(self, version_id, options, pos, size, parents):
1045
"""Cache a version record in the history array and index cache.
1047
This is inlined into __init__ for performance. KEEP IN SYNC.
1048
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1051
# only want the _history index to reference the 1st index entry
1053
if version_id not in self._cache:
1054
index = len(self._history)
1055
self._history.append(version_id)
1057
index = self._cache[version_id][5]
1058
self._cache[version_id] = (version_id,
1065
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1066
create_parent_dir=False, delay_create=False, dir_mode=None):
1067
_KnitComponentFile.__init__(self, transport, filename, mode,
1068
file_mode=file_mode,
1069
create_parent_dir=create_parent_dir,
1072
# position in _history is the 'official' index for a revision
1073
# but the values may have come from a newer entry.
1074
# so - wc -l of a knit index is != the number of unique names
1077
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1082
pb.update('read knit index', count, total)
1083
fp = self._transport.get(self._filename)
1085
self.check_header(fp)
1086
# readlines reads the whole file at once:
1087
# bad for transports like http, good for local disk
1088
# we save 60 ms doing this one change (
1089
# from calling readline each time to calling
1091
# probably what we want for nice behaviour on
1092
# http is a incremental readlines that yields, or
1093
# a check for local vs non local indexes,
1094
for l in fp.readlines():
1096
if len(rec) < 5 or rec[-1] != ':':
1098
# FIXME: in the future we should determine if its a
1099
# short write - and ignore it
1100
# or a different failure, and raise. RBC 20060407
1104
#pb.update('read knit index', count, total)
1105
# See self._parse_parents
1107
for value in rec[4:-1]:
1109
# uncompressed reference
1110
parents.append(value[1:])
1112
# this is 15/4000ms faster than isinstance,
1114
# this function is called thousands of times a
1115
# second so small variations add up.
1116
assert value.__class__ is str
1117
parents.append(self._history[int(value)])
1118
# end self._parse_parents
1119
# self._cache_version(rec[0],
1120
# rec[1].split(','),
1124
# --- self._cache_version
1125
# only want the _history index to reference the 1st
1126
# index entry for version_id
1128
if version_id not in self._cache:
1129
index = len(self._history)
1130
self._history.append(version_id)
1132
index = self._cache[version_id][5]
1133
self._cache[version_id] = (version_id,
1139
# --- self._cache_version
1142
except NoSuchFile, e:
1143
if mode != 'w' or not create:
1146
self._need_to_create = True
1148
self._transport.put_bytes_non_atomic(self._filename,
1149
self.HEADER, mode=self._file_mode)
1152
pb.update('read knit index', total, total)
1155
def _parse_parents(self, compressed_parents):
1156
"""convert a list of string parent values into version ids.
1158
ints are looked up in the index.
1159
.FOO values are ghosts and converted in to FOO.
1161
NOTE: the function is retained here for clarity, and for possible
1162
use in partial index reads. However bulk processing now has
1163
it inlined in __init__ for inner-loop optimisation.
1166
for value in compressed_parents:
1167
if value[-1] == '.':
1168
# uncompressed reference
1169
result.append(value[1:])
1171
# this is 15/4000ms faster than isinstance,
1172
# this function is called thousands of times a
1173
# second so small variations add up.
1174
assert value.__class__ is str
1175
result.append(self._history[int(value)])
1178
def get_graph(self):
1180
for version_id, index in self._cache.iteritems():
1181
graph.append((version_id, index[4]))
1184
def get_ancestry(self, versions):
1185
"""See VersionedFile.get_ancestry."""
1186
# get a graph of all the mentioned versions:
1188
pending = set(versions)
1190
version = pending.pop()
1191
parents = self._cache[version][4]
1192
# got the parents ok
1194
parents = [parent for parent in parents if parent in self._cache]
1195
for parent in parents:
1196
# if not completed and not a ghost
1197
if parent not in graph:
1199
graph[version] = parents
1200
return topo_sort(graph.items())
1202
def get_ancestry_with_ghosts(self, versions):
1203
"""See VersionedFile.get_ancestry_with_ghosts."""
1204
# get a graph of all the mentioned versions:
1206
pending = set(versions)
1208
version = pending.pop()
1210
parents = self._cache[version][4]
1216
# got the parents ok
1217
for parent in parents:
1218
if parent not in graph:
1220
graph[version] = parents
1221
return topo_sort(graph.items())
1223
def num_versions(self):
1224
return len(self._history)
1226
__len__ = num_versions
1228
def get_versions(self):
1229
return self._history
1231
def idx_to_name(self, idx):
1232
return self._history[idx]
1234
def lookup(self, version_id):
1235
assert version_id in self._cache
1236
return self._cache[version_id][5]
1238
def _version_list_to_index(self, versions):
1239
encode_utf8 = cache_utf8.encode
1241
for version in versions:
1242
if version in self._cache:
1243
# -- inlined lookup() --
1244
result_list.append(str(self._cache[version][5]))
1245
# -- end lookup () --
1247
result_list.append('.' + encode_utf8(version))
1248
return ' '.join(result_list)
1250
def add_version(self, version_id, options, pos, size, parents):
1251
"""Add a version record to the index."""
1252
self.add_versions(((version_id, options, pos, size, parents),))
1254
def add_versions(self, versions):
1255
"""Add multiple versions to the index.
1257
:param versions: a list of tuples:
1258
(version_id, options, pos, size, parents).
1261
encode_utf8 = cache_utf8.encode
1262
for version_id, options, pos, size, parents in versions:
1263
line = "\n%s %s %s %s %s :" % (encode_utf8(version_id),
1267
self._version_list_to_index(parents))
1268
assert isinstance(line, str), \
1269
'content must be utf-8 encoded: %r' % (line,)
1271
if not self._need_to_create:
1272
self._transport.append_bytes(self._filename, ''.join(lines))
1275
sio.write(self.HEADER)
1276
sio.writelines(lines)
1278
self._transport.put_file_non_atomic(self._filename, sio,
1279
create_parent_dir=self._create_parent_dir,
1280
mode=self._file_mode,
1281
dir_mode=self._dir_mode)
1282
self._need_to_create = False
1284
# cache after writing, so that a failed write leads to missing cache
1285
# entries not extra ones. XXX TODO: RBC 20060502 in the event of a
1286
# failure, reload the index or flush it or some such, to prevent
1287
# writing records that did complete twice.
1288
for version_id, options, pos, size, parents in versions:
1289
self._cache_version(version_id, options, pos, size, parents)
1291
def has_version(self, version_id):
1292
"""True if the version is in the index."""
1293
return (version_id in self._cache)
1295
def get_position(self, version_id):
1296
"""Return data position and size of specified version."""
1297
return (self._cache[version_id][2], \
1298
self._cache[version_id][3])
1300
def get_method(self, version_id):
1301
"""Return compression method of specified version."""
1302
options = self._cache[version_id][1]
1303
if 'fulltext' in options:
1306
assert 'line-delta' in options
1309
def get_options(self, version_id):
1310
return self._cache[version_id][1]
1312
def get_parents(self, version_id):
1313
"""Return parents of specified version ignoring ghosts."""
1314
return [parent for parent in self._cache[version_id][4]
1315
if parent in self._cache]
1317
def get_parents_with_ghosts(self, version_id):
1318
"""Return parents of specified version with ghosts."""
1319
return self._cache[version_id][4]
1321
def check_versions_present(self, version_ids):
1322
"""Check that all specified versions are present."""
1323
version_ids = set(version_ids)
1324
for version_id in list(version_ids):
1325
if version_id in self._cache:
1326
version_ids.remove(version_id)
1328
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1331
class _KnitData(_KnitComponentFile):
1332
"""Contents of the knit data file"""
1334
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1335
create_parent_dir=False, delay_create=False,
1337
_KnitComponentFile.__init__(self, transport, filename, mode,
1338
file_mode=file_mode,
1339
create_parent_dir=create_parent_dir,
1341
self._checked = False
1342
# TODO: jam 20060713 conceptually, this could spill to disk
1343
# if the cached size gets larger than a certain amount
1344
# but it complicates the model a bit, so for now just use
1345
# a simple dictionary
1347
self._do_cache = False
1350
self._need_to_create = create
1352
self._transport.put_bytes_non_atomic(self._filename, '',
1353
mode=self._file_mode)
1355
def enable_cache(self):
1356
"""Enable caching of reads."""
1357
self._do_cache = True
1359
def clear_cache(self):
1360
"""Clear the record cache."""
1361
self._do_cache = False
1364
def _open_file(self):
1366
return self._transport.get(self._filename)
1371
def _record_to_data(self, version_id, digest, lines):
1372
"""Convert version_id, digest, lines into a raw data block.
1374
:return: (len, a StringIO instance with the raw data ready to read.)
1377
data_file = GzipFile(None, mode='wb', fileobj=sio)
1379
version_id_utf8 = cache_utf8.encode(version_id)
1380
data_file.writelines(chain(
1381
["version %s %d %s\n" % (version_id_utf8,
1385
["end %s\n" % version_id_utf8]))
1392
def add_raw_record(self, raw_data):
1393
"""Append a prepared record to the data file.
1395
:return: the offset in the data file raw_data was written.
1397
assert isinstance(raw_data, str), 'data must be plain bytes'
1398
if not self._need_to_create:
1399
return self._transport.append_bytes(self._filename, raw_data)
1401
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1402
create_parent_dir=self._create_parent_dir,
1403
mode=self._file_mode,
1404
dir_mode=self._dir_mode)
1405
self._need_to_create = False
1408
def add_record(self, version_id, digest, lines):
1409
"""Write new text record to disk. Returns the position in the
1410
file where it was written."""
1411
size, sio = self._record_to_data(version_id, digest, lines)
1413
if not self._need_to_create:
1414
start_pos = self._transport.append_file(self._filename, sio)
1416
self._transport.put_file_non_atomic(self._filename, sio,
1417
create_parent_dir=self._create_parent_dir,
1418
mode=self._file_mode,
1419
dir_mode=self._dir_mode)
1420
self._need_to_create = False
1423
self._cache[version_id] = sio.getvalue()
1424
return start_pos, size
1426
def _parse_record_header(self, version_id, raw_data):
1427
"""Parse a record header for consistency.
1429
:return: the header and the decompressor stream.
1430
as (stream, header_record)
1432
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1433
rec = df.readline().split()
1435
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
1436
if cache_utf8.decode(rec[1]) != version_id:
1437
raise KnitCorrupt(self._filename,
1438
'unexpected version, wanted %r, got %r' % (
1439
version_id, rec[1]))
1442
def _parse_record(self, version_id, data):
1444
# 4168 calls in 2880 217 internal
1445
# 4168 calls to _parse_record_header in 2121
1446
# 4168 calls to readlines in 330
1447
df, rec = self._parse_record_header(version_id, data)
1448
record_contents = df.readlines()
1449
l = record_contents.pop()
1450
assert len(record_contents) == int(rec[2])
1451
if l != 'end %s\n' % cache_utf8.encode(version_id):
1452
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
1455
return record_contents, rec[3]
1457
def read_records_iter_raw(self, records):
1458
"""Read text records from data file and yield raw data.
1460
This unpacks enough of the text record to validate the id is
1461
as expected but thats all.
1463
# setup an iterator of the external records:
1464
# uses readv so nice and fast we hope.
1466
# grab the disk data needed.
1468
# Don't check _cache if it is empty
1469
needed_offsets = [(pos, size) for version_id, pos, size
1471
if version_id not in self._cache]
1473
needed_offsets = [(pos, size) for version_id, pos, size
1476
raw_records = self._transport.readv(self._filename, needed_offsets)
1479
for version_id, pos, size in records:
1480
if version_id in self._cache:
1481
# This data has already been validated
1482
data = self._cache[version_id]
1484
pos, data = raw_records.next()
1486
self._cache[version_id] = data
1488
# validate the header
1489
df, rec = self._parse_record_header(version_id, data)
1491
yield version_id, data
1493
def read_records_iter(self, records):
1494
"""Read text records from data file and yield result.
1496
The result will be returned in whatever is the fastest to read.
1497
Not by the order requested. Also, multiple requests for the same
1498
record will only yield 1 response.
1499
:param records: A list of (version_id, pos, len) entries
1500
:return: Yields (version_id, contents, digest) in the order
1501
read, not the order requested
1507
# Skip records we have alread seen
1508
yielded_records = set()
1509
needed_records = set()
1510
for record in records:
1511
if record[0] in self._cache:
1512
if record[0] in yielded_records:
1514
yielded_records.add(record[0])
1515
data = self._cache[record[0]]
1516
content, digest = self._parse_record(record[0], data)
1517
yield (record[0], content, digest)
1519
needed_records.add(record)
1520
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1522
needed_records = sorted(set(records), key=operator.itemgetter(1))
1524
if not needed_records:
1527
# The transport optimizes the fetching as well
1528
# (ie, reads continuous ranges.)
1529
readv_response = self._transport.readv(self._filename,
1530
[(pos, size) for version_id, pos, size in needed_records])
1532
for (version_id, pos, size), (pos, data) in \
1533
izip(iter(needed_records), readv_response):
1534
content, digest = self._parse_record(version_id, data)
1536
self._cache[version_id] = data
1537
yield version_id, content, digest
1539
def read_records(self, records):
1540
"""Read records into a dictionary."""
1542
for record_id, content, digest in \
1543
self.read_records_iter(records):
1544
components[record_id] = (content, digest)
1548
class InterKnit(InterVersionedFile):
1549
"""Optimised code paths for knit to knit operations."""
1551
_matching_file_from_factory = KnitVersionedFile
1552
_matching_file_to_factory = KnitVersionedFile
1555
def is_compatible(source, target):
1556
"""Be compatible with knits. """
1558
return (isinstance(source, KnitVersionedFile) and
1559
isinstance(target, KnitVersionedFile))
1560
except AttributeError:
1563
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1564
"""See InterVersionedFile.join."""
1565
assert isinstance(self.source, KnitVersionedFile)
1566
assert isinstance(self.target, KnitVersionedFile)
1568
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1573
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1575
version_ids = list(version_ids)
1576
if None in version_ids:
1577
version_ids.remove(None)
1579
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1580
this_versions = set(self.target._index.get_versions())
1581
needed_versions = self.source_ancestry - this_versions
1582
cross_check_versions = self.source_ancestry.intersection(this_versions)
1583
mismatched_versions = set()
1584
for version in cross_check_versions:
1585
# scan to include needed parents.
1586
n1 = set(self.target.get_parents_with_ghosts(version))
1587
n2 = set(self.source.get_parents_with_ghosts(version))
1589
# FIXME TEST this check for cycles being introduced works
1590
# the logic is we have a cycle if in our graph we are an
1591
# ancestor of any of the n2 revisions.
1597
parent_ancestors = self.source.get_ancestry(parent)
1598
if version in parent_ancestors:
1599
raise errors.GraphCycleError([parent, version])
1600
# ensure this parent will be available later.
1601
new_parents = n2.difference(n1)
1602
needed_versions.update(new_parents.difference(this_versions))
1603
mismatched_versions.add(version)
1605
if not needed_versions and not mismatched_versions:
1608
source_graph = self.source.get_graph()
1609
full_list = topo_sort(source_graph)
1610
version_list = [i for i in full_list if (i not in self.target
1611
and i in needed_versions)]
1612
self.check_matching_parents(version_list, source_graph)
1616
copy_queue_records = []
1618
for version_id in version_list:
1619
options = self.source._index.get_options(version_id)
1620
parents = self.source._index.get_parents_with_ghosts(version_id)
1621
# check that its will be a consistent copy:
1622
for parent in parents:
1623
# if source has the parent, we must :
1624
# * already have it or
1625
# * have it scheduled already
1626
# otherwise we don't care
1627
assert (self.target.has_version(parent) or
1628
parent in copy_set or
1629
not self.source.has_version(parent))
1630
data_pos, data_size = self.source._index.get_position(version_id)
1631
copy_queue_records.append((version_id, data_pos, data_size))
1632
copy_queue.append((version_id, options, parents))
1633
copy_set.add(version_id)
1635
# data suck the join:
1637
total = len(version_list)
1640
for (version_id, raw_data), \
1641
(version_id2, options, parents) in \
1642
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1644
assert version_id == version_id2, 'logic error, inconsistent results'
1646
pb.update("Joining knit", count, total)
1647
raw_records.append((version_id, options, parents, len(raw_data)))
1648
raw_datum.append(raw_data)
1649
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1651
for version in mismatched_versions:
1652
# FIXME RBC 20060309 is this needed?
1653
n1 = set(self.target.get_parents_with_ghosts(version))
1654
n2 = set(self.source.get_parents_with_ghosts(version))
1655
# write a combined record to our history preserving the current
1656
# parents as first in the list
1657
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1658
self.target.fix_parents(version, new_parents)
1663
def check_matching_parents(self, versions, source_graph):
1664
"""Check that parents texts match between knits
1666
This prevents knit corruption when copying raw records.
1668
parents_to_check = set()
1669
for version in versions:
1670
for parent in source_graph[version]:
1671
if parent in self.target and parent in self.source:
1672
parents_to_check.add(parent)
1673
source_sha1s = self.source.get_sha1s(parents_to_check)
1674
target_sha1s = self.target.get_sha1s(parents_to_check)
1675
for parent in parents_to_check:
1676
if source_sha1s[parent] != target_sha1s[parent]:
1677
raise errors.KnitTextsDiffer(parent)
1680
InterVersionedFile.register_optimiser(InterKnit)
1683
class WeaveToKnit(InterVersionedFile):
1684
"""Optimised code paths for weave to knit operations."""
1686
_matching_file_from_factory = bzrlib.weave.WeaveFile
1687
_matching_file_to_factory = KnitVersionedFile
1690
def is_compatible(source, target):
1691
"""Be compatible with weaves to knits."""
1693
return (isinstance(source, bzrlib.weave.Weave) and
1694
isinstance(target, KnitVersionedFile))
1695
except AttributeError:
1698
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1699
"""See InterVersionedFile.join."""
1700
assert isinstance(self.source, bzrlib.weave.Weave)
1701
assert isinstance(self.target, KnitVersionedFile)
1703
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1708
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1710
version_ids = list(version_ids)
1712
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1713
this_versions = set(self.target._index.get_versions())
1714
needed_versions = self.source_ancestry - this_versions
1715
cross_check_versions = self.source_ancestry.intersection(this_versions)
1716
mismatched_versions = set()
1717
for version in cross_check_versions:
1718
# scan to include needed parents.
1719
n1 = set(self.target.get_parents_with_ghosts(version))
1720
n2 = set(self.source.get_parents(version))
1721
# if all of n2's parents are in n1, then its fine.
1722
if n2.difference(n1):
1723
# FIXME TEST this check for cycles being introduced works
1724
# the logic is we have a cycle if in our graph we are an
1725
# ancestor of any of the n2 revisions.
1731
parent_ancestors = self.source.get_ancestry(parent)
1732
if version in parent_ancestors:
1733
raise errors.GraphCycleError([parent, version])
1734
# ensure this parent will be available later.
1735
new_parents = n2.difference(n1)
1736
needed_versions.update(new_parents.difference(this_versions))
1737
mismatched_versions.add(version)
1739
if not needed_versions and not mismatched_versions:
1741
full_list = topo_sort(self.source.get_graph())
1743
version_list = [i for i in full_list if (not self.target.has_version(i)
1744
and i in needed_versions)]
1748
total = len(version_list)
1749
for version_id in version_list:
1750
pb.update("Converting to knit", count, total)
1751
parents = self.source.get_parents(version_id)
1752
# check that its will be a consistent copy:
1753
for parent in parents:
1754
# if source has the parent, we must already have it
1755
assert (self.target.has_version(parent))
1756
self.target.add_lines(
1757
version_id, parents, self.source.get_lines(version_id))
1760
for version in mismatched_versions:
1761
# FIXME RBC 20060309 is this needed?
1762
n1 = set(self.target.get_parents_with_ghosts(version))
1763
n2 = set(self.source.get_parents(version))
1764
# write a combined record to our history preserving the current
1765
# parents as first in the list
1766
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1767
self.target.fix_parents(version, new_parents)
1773
InterVersionedFile.register_optimiser(WeaveToKnit)
1776
class KnitSequenceMatcher(difflib.SequenceMatcher):
1777
"""Knit tuned sequence matcher.
1779
This is based on profiling of difflib which indicated some improvements
1780
for our usage pattern.
1783
def find_longest_match(self, alo, ahi, blo, bhi):
1784
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1786
If isjunk is not defined:
1788
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1789
alo <= i <= i+k <= ahi
1790
blo <= j <= j+k <= bhi
1791
and for all (i',j',k') meeting those conditions,
1794
and if i == i', j <= j'
1796
In other words, of all maximal matching blocks, return one that
1797
starts earliest in a, and of all those maximal matching blocks that
1798
start earliest in a, return the one that starts earliest in b.
1800
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1801
>>> s.find_longest_match(0, 5, 0, 9)
1804
If isjunk is defined, first the longest matching block is
1805
determined as above, but with the additional restriction that no
1806
junk element appears in the block. Then that block is extended as
1807
far as possible by matching (only) junk elements on both sides. So
1808
the resulting block never matches on junk except as identical junk
1809
happens to be adjacent to an "interesting" match.
1811
Here's the same example as before, but considering blanks to be
1812
junk. That prevents " abcd" from matching the " abcd" at the tail
1813
end of the second sequence directly. Instead only the "abcd" can
1814
match, and matches the leftmost "abcd" in the second sequence:
1816
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1817
>>> s.find_longest_match(0, 5, 0, 9)
1820
If no blocks match, return (alo, blo, 0).
1822
>>> s = SequenceMatcher(None, "ab", "c")
1823
>>> s.find_longest_match(0, 2, 0, 1)
1827
# CAUTION: stripping common prefix or suffix would be incorrect.
1831
# Longest matching block is "ab", but if common prefix is
1832
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1833
# strip, so ends up claiming that ab is changed to acab by
1834
# inserting "ca" in the middle. That's minimal but unintuitive:
1835
# "it's obvious" that someone inserted "ac" at the front.
1836
# Windiff ends up at the same place as diff, but by pairing up
1837
# the unique 'b's and then matching the first two 'a's.
1839
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1840
besti, bestj, bestsize = alo, blo, 0
1841
# find longest junk-free match
1842
# during an iteration of the loop, j2len[j] = length of longest
1843
# junk-free match ending with a[i-1] and b[j]
1847
for i in xrange(alo, ahi):
1848
# look at all instances of a[i] in b; note that because
1849
# b2j has no junk keys, the loop is skipped if a[i] is junk
1850
j2lenget = j2len.get
1853
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1854
# following improvement
1855
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1856
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1857
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1859
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1860
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1861
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1873
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1875
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1878
# Extend the best by non-junk elements on each end. In particular,
1879
# "popular" non-junk elements aren't in b2j, which greatly speeds
1880
# the inner loop above, but also means "the best" match so far
1881
# doesn't contain any junk *or* popular non-junk elements.
1882
while besti > alo and bestj > blo and \
1883
not isbjunk(b[bestj-1]) and \
1884
a[besti-1] == b[bestj-1]:
1885
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1886
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1887
not isbjunk(b[bestj+bestsize]) and \
1888
a[besti+bestsize] == b[bestj+bestsize]:
1891
# Now that we have a wholly interesting match (albeit possibly
1892
# empty!), we may as well suck up the matching junk on each
1893
# side of it too. Can't think of a good reason not to, and it
1894
# saves post-processing the (possibly considerable) expense of
1895
# figuring out what to do with it. In the case of an empty
1896
# interesting match, this is clearly the right thing to do,
1897
# because no other kind of match is possible in the regions.
1898
while besti > alo and bestj > blo and \
1899
isbjunk(b[bestj-1]) and \
1900
a[besti-1] == b[bestj-1]:
1901
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1902
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1903
isbjunk(b[bestj+bestsize]) and \
1904
a[besti+bestsize] == b[bestj+bestsize]:
1905
bestsize = bestsize + 1
1907
return besti, bestj, bestsize
added
removed
bzrlib/knit.py
