« back to all changes in this revision
Viewing changes to bzrlib/knit.py
- browse files at revision 1
- compare with another revision
- download diff
- download tarball
- view history from revision 1
- Committer: mbp at sourcefrog
- Date: 2005-03-09 04:08:15 UTC
- Revision ID: mbp@sourcefrog.net-20050309040815-13242001617e4a06
import from baz patch-364
- files added:
- bzrlib/tests.py
- files removed:
- .bzrignore
- bzrlib/doc
- bzrlib/doc/api
- bzrlib/export
- bzrlib/plugins
- bzrlib/store
- bzrlib/store/revision
- bzrlib/store/versioned
- bzrlib/tests
- bzrlib/tests/blackbox
- bzrlib/tests/branch_implementations
- bzrlib/tests/bzrdir_implementations
- bzrlib/tests/interrepository_implementations
- bzrlib/tests/interversionedfile_implementations
- bzrlib/tests/repository_implementations
- bzrlib/tests/revisionstore_implementations
- bzrlib/tests/workingtree_implementations
- bzrlib/transport
- bzrlib/transport/http
- bzrlib/ui
- bzrlib/util
- bzrlib/util/configobj
- bzrlib/util/configobj/docs
- bzrlib/util/effbot
- bzrlib/util/effbot/org
- bzrlib/util/elementtree
- bzrlib/util/urlgrabber
- contrib
- contrib/bash
- contrib/emacs
- contrib/zsh
- doc
- notes
- tools
- tools/doc_generate
- files renamed:
- bzrlib/commands.py => bzr.py
- bzrlib/store/__init__.py => bzrlib/store.py
- bzrlib/xml_serializer.py => bzrlib/xml.py
- files modified:
- README
added
removed
bzrlib/knit.py
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
#
6
# This program is free software; you can redistribute it and/or modify
7
# it under the terms of the GNU General Public License as published by
8
# the Free Software Foundation; either version 2 of the License, or
9
# (at your option) any later version.
10
#
11
# This program is distributed in the hope that it will be useful,
12
# but WITHOUT ANY WARRANTY; without even the implied warranty of
13
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14
# GNU General Public License for more details.
15
#
16
# You should have received a copy of the GNU General Public License
17
# along with this program; if not, write to the Free Software
18
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19
20
"""Knit versionedfile implementation.
21
22
A knit is a versioned file implementation that supports efficient append only
23
updates.
24
25
Knit file layout:
26
lifeless: the data file is made up of "delta records". each delta record has a delta header
27
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
28
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
29
end-marker; simply "end VERSION"
30
31
delta can be line or full contents.a
32
... the 8's there are the index number of the annotation.
33
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
59,59,3
35
8
36
8 if ie.executable:
37
8 e.set('executable', 'yes')
38
130,130,2
39
8 if elt.get('executable') == 'yes':
40
8 ie.executable = True
41
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
43
44
whats in an index:
45
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
46
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
47
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
48
09:33 < lifeless> right
49
09:33 < jrydberg> lifeless: the position and size is the range in the data file
50
51
52
so the index sequence is the dictionary compressed sequence number used
53
in the deltas to provide line annotation
54
55
"""
56
57
# TODOS:
58
# 10:16 < lifeless> make partial index writes safe
59
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
60
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
61
# always' approach.
62
# move sha1 out of the content so that join is faster at verifying parents
63
# record content length ?
64
65
66
from copy import copy
67
from cStringIO import StringIO
68
import difflib
69
from itertools import izip, chain
70
import os
71
import sys
72
73
import bzrlib
74
import bzrlib.errors as errors
75
from bzrlib.errors import FileExists, NoSuchFile, KnitError, \
76
InvalidRevisionId, KnitCorrupt, KnitHeaderError, \
77
RevisionNotPresent, RevisionAlreadyPresent
78
from bzrlib.tuned_gzip import *
79
from bzrlib.trace import mutter
80
from bzrlib.osutils import contains_whitespace, contains_linebreaks, \
81
sha_strings
82
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
83
from bzrlib.tsort import topo_sort
84
85
86
# TODO: Split out code specific to this format into an associated object.
87
88
# TODO: Can we put in some kind of value to check that the index and data
89
# files belong together?
90
91
# TODO: accomodate binaries, perhaps by storing a byte count
92
93
# TODO: function to check whole file
94
95
# TODO: atomically append data, then measure backwards from the cursor
96
# position after writing to work out where it was located. we may need to
97
# bypass python file buffering.
98
99
DATA_SUFFIX = '.knit'
100
INDEX_SUFFIX = '.kndx'
101
102
103
class KnitContent(object):
104
"""Content of a knit version to which deltas can be applied."""
105
106
def __init__(self, lines):
107
self._lines = lines
108
109
def annotate_iter(self):
110
"""Yield tuples of (origin, text) for each content line."""
111
for origin, text in self._lines:
112
yield origin, text
113
114
def annotate(self):
115
"""Return a list of (origin, text) tuples."""
116
return list(self.annotate_iter())
117
118
def line_delta_iter(self, new_lines):
119
"""Generate line-based delta from this content to new_lines."""
120
new_texts = [text for origin, text in new_lines._lines]
121
old_texts = [text for origin, text in self._lines]
122
s = SequenceMatcher(None, old_texts, new_texts)
123
for op in s.get_opcodes():
124
if op[0] == 'equal':
125
continue
126
# ofrom oto length data
127
yield (op[1], op[2], op[4]-op[3], new_lines._lines[op[3]:op[4]])
128
129
def line_delta(self, new_lines):
130
return list(self.line_delta_iter(new_lines))
131
132
def text(self):
133
return [text for origin, text in self._lines]
134
135
136
class _KnitFactory(object):
137
"""Base factory for creating content objects."""
138
139
def make(self, lines, version):
140
num_lines = len(lines)
141
return KnitContent(zip([version] * num_lines, lines))
142
143
144
class KnitAnnotateFactory(_KnitFactory):
145
"""Factory for creating annotated Content objects."""
146
147
annotated = True
148
149
def parse_fulltext(self, content, version):
150
"""Convert fulltext to internal representation
151
152
fulltext content is of the format
153
revid(utf8) plaintext\n
154
internal representation is of the format:
155
(revid, plaintext)
156
"""
157
lines = []
158
for line in content:
159
origin, text = line.split(' ', 1)
160
lines.append((origin.decode('utf-8'), text))
161
return KnitContent(lines)
162
163
def parse_line_delta_iter(self, lines):
164
for result_item in self.parse_line_delta[lines]:
165
yield result_item
166
167
def parse_line_delta(self, lines, version):
168
"""Convert a line based delta into internal representation.
169
170
line delta is in the form of:
171
intstart intend intcount
172
1..count lines:
173
revid(utf8) newline\n
174
internal represnetation is
175
(start, end, count, [1..count tuples (revid, newline)])
176
"""
177
result = []
178
lines = iter(lines)
179
next = lines.next
180
# walk through the lines parsing.
181
for header in lines:
182
start, end, count = [int(n) for n in header.split(',')]
183
contents = []
184
remaining = count
185
while remaining:
186
origin, text = next().split(' ', 1)
187
remaining -= 1
188
contents.append((origin.decode('utf-8'), text))
189
result.append((start, end, count, contents))
190
return result
191
192
def lower_fulltext(self, content):
193
"""convert a fulltext content record into a serializable form.
194
195
see parse_fulltext which this inverts.
196
"""
197
return ['%s %s' % (o.encode('utf-8'), t) for o, t in content._lines]
198
199
def lower_line_delta(self, delta):
200
"""convert a delta into a serializable form.
201
202
See parse_line_delta which this inverts.
203
"""
204
out = []
205
for start, end, c, lines in delta:
206
out.append('%d,%d,%d\n' % (start, end, c))
207
for origin, text in lines:
208
out.append('%s %s' % (origin.encode('utf-8'), text))
209
return out
210
211
212
class KnitPlainFactory(_KnitFactory):
213
"""Factory for creating plain Content objects."""
214
215
annotated = False
216
217
def parse_fulltext(self, content, version):
218
"""This parses an unannotated fulltext.
219
220
Note that this is not a noop - the internal representation
221
has (versionid, line) - its just a constant versionid.
222
"""
223
return self.make(content, version)
224
225
def parse_line_delta_iter(self, lines, version):
226
while lines:
227
header = lines.pop(0)
228
start, end, c = [int(n) for n in header.split(',')]
229
yield start, end, c, zip([version] * c, lines[:c])
230
del lines[:c]
231
232
def parse_line_delta(self, lines, version):
233
return list(self.parse_line_delta_iter(lines, version))
234
235
def lower_fulltext(self, content):
236
return content.text()
237
238
def lower_line_delta(self, delta):
239
out = []
240
for start, end, c, lines in delta:
241
out.append('%d,%d,%d\n' % (start, end, c))
242
out.extend([text for origin, text in lines])
243
return out
244
245
246
def make_empty_knit(transport, relpath):
247
"""Construct a empty knit at the specified location."""
248
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
249
k._data._open_file()
250
251
252
class KnitVersionedFile(VersionedFile):
253
"""Weave-like structure with faster random access.
254
255
A knit stores a number of texts and a summary of the relationships
256
between them. Texts are identified by a string version-id. Texts
257
are normally stored and retrieved as a series of lines, but can
258
also be passed as single strings.
259
260
Lines are stored with the trailing newline (if any) included, to
261
avoid special cases for files with no final newline. Lines are
262
composed of 8-bit characters, not unicode. The combination of
263
these approaches should mean any 'binary' file can be safely
264
stored and retrieved.
265
"""
266
267
def __init__(self, relpath, transport, file_mode=None, access_mode=None, factory=None,
268
basis_knit=None, delta=True, create=False):
269
"""Construct a knit at location specified by relpath.
270
271
:param create: If not True, only open an existing knit.
272
"""
273
if access_mode is None:
274
access_mode = 'w'
275
super(KnitVersionedFile, self).__init__(access_mode)
276
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
277
assert not basis_knit or isinstance(basis_knit, KnitVersionedFile), \
278
type(basis_knit)
279
280
self.transport = transport
281
self.filename = relpath
282
self.basis_knit = basis_knit
283
self.factory = factory or KnitAnnotateFactory()
284
self.writable = (access_mode == 'w')
285
self.delta = delta
286
287
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
288
access_mode, create=create)
289
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
290
access_mode, create=not len(self.versions()))
291
292
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
293
"""See VersionedFile._add_delta()."""
294
self._check_add(version_id, []) # should we check the lines ?
295
self._check_versions_present(parents)
296
present_parents = []
297
ghosts = []
298
parent_texts = {}
299
for parent in parents:
300
if not self.has_version(parent):
301
ghosts.append(parent)
302
else:
303
present_parents.append(parent)
304
305
if delta_parent is None:
306
# reconstitute as full text.
307
assert len(delta) == 1 or len(delta) == 0
308
if len(delta):
309
assert delta[0][0] == 0
310
assert delta[0][1] == 0, delta[0][1]
311
return super(KnitVersionedFile, self)._add_delta(version_id,
312
parents,
313
delta_parent,
314
sha1,
315
noeol,
316
delta)
317
318
digest = sha1
319
320
options = []
321
if noeol:
322
options.append('no-eol')
323
324
if delta_parent is not None:
325
# determine the current delta chain length.
326
# To speed the extract of texts the delta chain is limited
327
# to a fixed number of deltas. This should minimize both
328
# I/O and the time spend applying deltas.
329
count = 0
330
delta_parents = [delta_parent]
331
while count < 25:
332
parent = delta_parents[0]
333
method = self._index.get_method(parent)
334
if method == 'fulltext':
335
break
336
delta_parents = self._index.get_parents(parent)
337
count = count + 1
338
if method == 'line-delta':
339
# did not find a fulltext in the delta limit.
340
# just do a normal insertion.
341
return super(KnitVersionedFile, self)._add_delta(version_id,
342
parents,
343
delta_parent,
344
sha1,
345
noeol,
346
delta)
347
348
options.append('line-delta')
349
store_lines = self.factory.lower_line_delta(delta)
350
351
where, size = self._data.add_record(version_id, digest, store_lines)
352
self._index.add_version(version_id, options, where, size, parents)
353
354
def clear_cache(self):
355
"""Clear the data cache only."""
356
self._data.clear_cache()
357
358
def copy_to(self, name, transport):
359
"""See VersionedFile.copy_to()."""
360
# copy the current index to a temp index to avoid racing with local
361
# writes
362
transport.put(name + INDEX_SUFFIX + '.tmp', self.transport.get(self._index._filename))
363
# copy the data file
364
transport.put(name + DATA_SUFFIX, self._data._open_file())
365
# rename the copied index into place
366
transport.rename(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
367
368
def create_empty(self, name, transport, mode=None):
369
return KnitVersionedFile(name, transport, factory=self.factory, delta=self.delta, create=True)
370
371
def _fix_parents(self, version, new_parents):
372
"""Fix the parents list for version.
373
374
This is done by appending a new version to the index
375
with identical data except for the parents list.
376
the parents list must be a superset of the current
377
list.
378
"""
379
current_values = self._index._cache[version]
380
assert set(current_values[4]).difference(set(new_parents)) == set()
381
self._index.add_version(version,
382
current_values[1],
383
current_values[2],
384
current_values[3],
385
new_parents)
386
387
def get_delta(self, version_id):
388
"""Get a delta for constructing version from some other version."""
389
if not self.has_version(version_id):
390
raise RevisionNotPresent(version_id, self.filename)
391
392
parents = self.get_parents(version_id)
393
if len(parents):
394
parent = parents[0]
395
else:
396
parent = None
397
data_pos, data_size = self._index.get_position(version_id)
398
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
399
version_idx = self._index.lookup(version_id)
400
noeol = 'no-eol' in self._index.get_options(version_id)
401
if 'fulltext' == self._index.get_method(version_id):
402
new_content = self.factory.parse_fulltext(data, version_idx)
403
if parent is not None:
404
reference_content = self._get_content(parent)
405
old_texts = reference_content.text()
406
else:
407
old_texts = []
408
new_texts = new_content.text()
409
delta_seq = SequenceMatcher(None, old_texts, new_texts)
410
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
411
else:
412
delta = self.factory.parse_line_delta(data, version_idx)
413
return parent, sha1, noeol, delta
414
415
def get_graph_with_ghosts(self):
416
"""See VersionedFile.get_graph_with_ghosts()."""
417
graph_items = self._index.get_graph()
418
return dict(graph_items)
419
420
@staticmethod
421
def get_suffixes():
422
"""See VersionedFile.get_suffixes()."""
423
return [DATA_SUFFIX, INDEX_SUFFIX]
424
425
def has_ghost(self, version_id):
426
"""True if there is a ghost reference in the file to version_id."""
427
# maybe we have it
428
if self.has_version(version_id):
429
return False
430
# optimisable if needed by memoising the _ghosts set.
431
items = self._index.get_graph()
432
for node, parents in items:
433
for parent in parents:
434
if parent not in self._index._cache:
435
if parent == version_id:
436
return True
437
return False
438
439
def versions(self):
440
"""See VersionedFile.versions."""
441
return self._index.get_versions()
442
443
def has_version(self, version_id):
444
"""See VersionedFile.has_version."""
445
return self._index.has_version(version_id)
446
447
__contains__ = has_version
448
449
def _merge_annotations(self, content, parents, parent_texts={},
450
delta=None, annotated=None):
451
"""Merge annotations for content. This is done by comparing
452
the annotations based on changed to the text.
453
"""
454
if annotated:
455
delta_seq = None
456
for parent_id in parents:
457
merge_content = self._get_content(parent_id, parent_texts)
458
seq = SequenceMatcher(None, merge_content.text(), content.text())
459
if delta_seq is None:
460
# setup a delta seq to reuse.
461
delta_seq = seq
462
for i, j, n in seq.get_matching_blocks():
463
if n == 0:
464
continue
465
# this appears to copy (origin, text) pairs across to the new
466
# content for any line that matches the last-checked parent.
467
# FIXME: save the sequence control data for delta compression
468
# against the most relevant parent rather than rediffing.
469
content._lines[j:j+n] = merge_content._lines[i:i+n]
470
if delta:
471
if not annotated:
472
reference_content = self._get_content(parents[0], parent_texts)
473
new_texts = content.text()
474
old_texts = reference_content.text()
475
delta_seq = SequenceMatcher(None, old_texts, new_texts)
476
return self._make_line_delta(delta_seq, content)
477
478
def _make_line_delta(self, delta_seq, new_content):
479
"""Generate a line delta from delta_seq and new_content."""
480
diff_hunks = []
481
for op in delta_seq.get_opcodes():
482
if op[0] == 'equal':
483
continue
484
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
485
return diff_hunks
486
487
def _get_components(self, version_id):
488
"""Return a list of (version_id, method, data) tuples that
489
makes up version specified by version_id of the knit.
490
491
The components should be applied in the order of the returned
492
list.
493
494
The basis knit will be used to the largest extent possible
495
since it is assumed that accesses to it is faster.
496
"""
497
#profile notes:
498
# 4168 calls in 14912, 2289 internal
499
# 4168 in 9711 to read_records
500
# 52554 in 1250 to get_parents
501
# 170166 in 865 to list.append
502
503
# needed_revisions holds a list of (method, version_id) of
504
# versions that is needed to be fetched to construct the final
505
# version of the file.
506
#
507
# basis_revisions is a list of versions that needs to be
508
# fetched but exists in the basis knit.
509
510
basis = self.basis_knit
511
needed_versions = []
512
basis_versions = []
513
cursor = version_id
514
515
while 1:
516
picked_knit = self
517
if basis and basis._index.has_version(cursor):
518
picked_knit = basis
519
basis_versions.append(cursor)
520
method = picked_knit._index.get_method(cursor)
521
needed_versions.append((method, cursor))
522
if method == 'fulltext':
523
break
524
cursor = picked_knit.get_parents(cursor)[0]
525
526
components = {}
527
if basis_versions:
528
records = []
529
for comp_id in basis_versions:
530
data_pos, data_size = basis._index.get_data_position(comp_id)
531
records.append((piece_id, data_pos, data_size))
532
components.update(basis._data.read_records(records))
533
534
records = []
535
for comp_id in [vid for method, vid in needed_versions
536
if vid not in basis_versions]:
537
data_pos, data_size = self._index.get_position(comp_id)
538
records.append((comp_id, data_pos, data_size))
539
components.update(self._data.read_records(records))
540
541
# get_data_records returns a mapping with the version id as
542
# index and the value as data. The order the components need
543
# to be applied is held by needed_versions (reversed).
544
out = []
545
for method, comp_id in reversed(needed_versions):
546
out.append((comp_id, method, components[comp_id]))
547
548
return out
549
550
def _get_content(self, version_id, parent_texts={}):
551
"""Returns a content object that makes up the specified
552
version."""
553
if not self.has_version(version_id):
554
raise RevisionNotPresent(version_id, self.filename)
555
556
cached_version = parent_texts.get(version_id, None)
557
if cached_version is not None:
558
return cached_version
559
560
if self.basis_knit and version_id in self.basis_knit:
561
return self.basis_knit._get_content(version_id)
562
563
content = None
564
components = self._get_components(version_id)
565
for component_id, method, (data, digest) in components:
566
version_idx = self._index.lookup(component_id)
567
if method == 'fulltext':
568
assert content is None
569
content = self.factory.parse_fulltext(data, version_idx)
570
elif method == 'line-delta':
571
delta = self.factory.parse_line_delta(data, version_idx)
572
content._lines = self._apply_delta(content._lines, delta)
573
574
if 'no-eol' in self._index.get_options(version_id):
575
line = content._lines[-1][1].rstrip('\n')
576
content._lines[-1] = (content._lines[-1][0], line)
577
578
if sha_strings(content.text()) != digest:
579
import pdb;pdb.set_trace()
580
raise KnitCorrupt(self.filename, 'sha-1 does not match %s' % version_id)
581
582
return content
583
584
def _check_versions_present(self, version_ids):
585
"""Check that all specified versions are present."""
586
version_ids = set(version_ids)
587
for r in list(version_ids):
588
if self._index.has_version(r):
589
version_ids.remove(r)
590
if version_ids:
591
raise RevisionNotPresent(list(version_ids)[0], self.filename)
592
593
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
594
"""See VersionedFile.add_lines_with_ghosts()."""
595
self._check_add(version_id, lines)
596
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
597
598
def _add_lines(self, version_id, parents, lines, parent_texts):
599
"""See VersionedFile.add_lines."""
600
self._check_add(version_id, lines)
601
self._check_versions_present(parents)
602
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
603
604
def _check_add(self, version_id, lines):
605
"""check that version_id and lines are safe to add."""
606
assert self.writable, "knit is not opened for write"
607
### FIXME escape. RBC 20060228
608
if contains_whitespace(version_id):
609
raise InvalidRevisionId(version_id)
610
if self.has_version(version_id):
611
raise RevisionAlreadyPresent(version_id, self.filename)
612
613
if False or __debug__:
614
for l in lines:
615
assert '\n' not in l[:-1]
616
617
def _add(self, version_id, lines, parents, delta, parent_texts):
618
"""Add a set of lines on top of version specified by parents.
619
620
If delta is true, compress the text as a line-delta against
621
the first parent.
622
623
Any versions not present will be converted into ghosts.
624
"""
625
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
626
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
627
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
628
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
629
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
630
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
631
# +1383 0 8.0370 8.0370 +<len>
632
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
633
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
634
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
635
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
636
637
present_parents = []
638
ghosts = []
639
if parent_texts is None:
640
parent_texts = {}
641
for parent in parents:
642
if not self.has_version(parent):
643
ghosts.append(parent)
644
else:
645
present_parents.append(parent)
646
647
if delta and not len(present_parents):
648
delta = False
649
650
digest = sha_strings(lines)
651
options = []
652
if lines:
653
if lines[-1][-1] != '\n':
654
options.append('no-eol')
655
lines[-1] = lines[-1] + '\n'
656
657
if len(present_parents) and delta:
658
# To speed the extract of texts the delta chain is limited
659
# to a fixed number of deltas. This should minimize both
660
# I/O and the time spend applying deltas.
661
count = 0
662
delta_parents = present_parents
663
while count < 25:
664
parent = delta_parents[0]
665
method = self._index.get_method(parent)
666
if method == 'fulltext':
667
break
668
delta_parents = self._index.get_parents(parent)
669
count = count + 1
670
if method == 'line-delta':
671
delta = False
672
673
lines = self.factory.make(lines, version_id)
674
if delta or (self.factory.annotated and len(present_parents) > 0):
675
# Merge annotations from parent texts if so is needed.
676
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
677
delta, self.factory.annotated)
678
679
if delta:
680
options.append('line-delta')
681
store_lines = self.factory.lower_line_delta(delta_hunks)
682
else:
683
options.append('fulltext')
684
store_lines = self.factory.lower_fulltext(lines)
685
686
where, size = self._data.add_record(version_id, digest, store_lines)
687
self._index.add_version(version_id, options, where, size, parents)
688
return lines
689
690
def check(self, progress_bar=None):
691
"""See VersionedFile.check()."""
692
693
def _clone_text(self, new_version_id, old_version_id, parents):
694
"""See VersionedFile.clone_text()."""
695
# FIXME RBC 20060228 make fast by only inserting an index with null delta.
696
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
697
698
def get_lines(self, version_id):
699
"""See VersionedFile.get_lines()."""
700
return self._get_content(version_id).text()
701
702
def iter_lines_added_or_present_in_versions(self, version_ids=None):
703
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
704
if version_ids is None:
705
version_ids = self.versions()
706
# we dont care about inclusions, the caller cares.
707
# but we need to setup a list of records to visit.
708
# we need version_id, position, length
709
version_id_records = []
710
requested_versions = list(version_ids)
711
# filter for available versions
712
for version_id in requested_versions:
713
if not self.has_version(version_id):
714
raise RevisionNotPresent(version_id, self.filename)
715
# get a in-component-order queue:
716
version_ids = []
717
for version_id in self.versions():
718
if version_id in requested_versions:
719
version_ids.append(version_id)
720
data_pos, length = self._index.get_position(version_id)
721
version_id_records.append((version_id, data_pos, length))
722
723
pb = bzrlib.ui.ui_factory.nested_progress_bar()
724
count = 0
725
total = len(version_id_records)
726
try:
727
pb.update('Walking content.', count, total)
728
for version_id, data, sha_value in \
729
self._data.read_records_iter(version_id_records):
730
pb.update('Walking content.', count, total)
731
method = self._index.get_method(version_id)
732
version_idx = self._index.lookup(version_id)
733
assert method in ('fulltext', 'line-delta')
734
if method == 'fulltext':
735
content = self.factory.parse_fulltext(data, version_idx)
736
for line in content.text():
737
yield line
738
else:
739
delta = self.factory.parse_line_delta(data, version_idx)
740
for start, end, count, lines in delta:
741
for origin, line in lines:
742
yield line
743
count +=1
744
pb.update('Walking content.', total, total)
745
pb.finished()
746
except:
747
pb.update('Walking content.', total, total)
748
pb.finished()
749
raise
750
751
def num_versions(self):
752
"""See VersionedFile.num_versions()."""
753
return self._index.num_versions()
754
755
__len__ = num_versions
756
757
def annotate_iter(self, version_id):
758
"""See VersionedFile.annotate_iter."""
759
content = self._get_content(version_id)
760
for origin, text in content.annotate_iter():
761
yield origin, text
762
763
def get_parents(self, version_id):
764
"""See VersionedFile.get_parents."""
765
# perf notes:
766
# optimism counts!
767
# 52554 calls in 1264 872 internal down from 3674
768
try:
769
return self._index.get_parents(version_id)
770
except KeyError:
771
raise RevisionNotPresent(version_id, self.filename)
772
773
def get_parents_with_ghosts(self, version_id):
774
"""See VersionedFile.get_parents."""
775
try:
776
return self._index.get_parents_with_ghosts(version_id)
777
except KeyError:
778
raise RevisionNotPresent(version_id, self.filename)
779
780
def get_ancestry(self, versions):
781
"""See VersionedFile.get_ancestry."""
782
if isinstance(versions, basestring):
783
versions = [versions]
784
if not versions:
785
return []
786
self._check_versions_present(versions)
787
return self._index.get_ancestry(versions)
788
789
def get_ancestry_with_ghosts(self, versions):
790
"""See VersionedFile.get_ancestry_with_ghosts."""
791
if isinstance(versions, basestring):
792
versions = [versions]
793
if not versions:
794
return []
795
self._check_versions_present(versions)
796
return self._index.get_ancestry_with_ghosts(versions)
797
798
#@deprecated_method(zero_eight)
799
def walk(self, version_ids):
800
"""See VersionedFile.walk."""
801
# We take the short path here, and extract all relevant texts
802
# and put them in a weave and let that do all the work. Far
803
# from optimal, but is much simpler.
804
# FIXME RB 20060228 this really is inefficient!
805
from bzrlib.weave import Weave
806
807
w = Weave(self.filename)
808
ancestry = self.get_ancestry(version_ids)
809
sorted_graph = topo_sort(self._index.get_graph())
810
version_list = [vid for vid in sorted_graph if vid in ancestry]
811
812
for version_id in version_list:
813
lines = self.get_lines(version_id)
814
w.add_lines(version_id, self.get_parents(version_id), lines)
815
816
for lineno, insert_id, dset, line in w.walk(version_ids):
817
yield lineno, insert_id, dset, line
818
819
def plan_merge(self, ver_a, ver_b):
820
"""See VersionedFile.plan_merge."""
821
ancestors_b = set(self.get_ancestry(ver_b))
822
def status_a(revision, text):
823
if revision in ancestors_b:
824
return 'killed-b', text
825
else:
826
return 'new-a', text
827
828
ancestors_a = set(self.get_ancestry(ver_a))
829
def status_b(revision, text):
830
if revision in ancestors_a:
831
return 'killed-a', text
832
else:
833
return 'new-b', text
834
835
annotated_a = self.annotate(ver_a)
836
annotated_b = self.annotate(ver_b)
837
plain_a = [t for (a, t) in annotated_a]
838
plain_b = [t for (a, t) in annotated_b]
839
blocks = SequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
840
a_cur = 0
841
b_cur = 0
842
for ai, bi, l in blocks:
843
# process all mismatched sections
844
# (last mismatched section is handled because blocks always
845
# includes a 0-length last block)
846
for revision, text in annotated_a[a_cur:ai]:
847
yield status_a(revision, text)
848
for revision, text in annotated_b[b_cur:bi]:
849
yield status_b(revision, text)
850
851
# and now the matched section
852
a_cur = ai + l
853
b_cur = bi + l
854
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
855
assert text_a == text_b
856
yield "unchanged", text_a
857
858
859
class _KnitComponentFile(object):
860
"""One of the files used to implement a knit database"""
861
862
def __init__(self, transport, filename, mode):
863
self._transport = transport
864
self._filename = filename
865
self._mode = mode
866
867
def write_header(self):
868
if self._transport.append(self._filename, StringIO(self.HEADER)):
869
raise KnitCorrupt(self._filename, 'misaligned after writing header')
870
871
def check_header(self, fp):
872
line = fp.readline()
873
if line != self.HEADER:
874
raise KnitHeaderError(badline=line)
875
876
def commit(self):
877
"""Commit is a nop."""
878
879
def __repr__(self):
880
return '%s(%s)' % (self.__class__.__name__, self._filename)
881
882
883
class _KnitIndex(_KnitComponentFile):
884
"""Manages knit index file.
885
886
The index is already kept in memory and read on startup, to enable
887
fast lookups of revision information. The cursor of the index
888
file is always pointing to the end, making it easy to append
889
entries.
890
891
_cache is a cache for fast mapping from version id to a Index
892
object.
893
894
_history is a cache for fast mapping from indexes to version ids.
895
896
The index data format is dictionary compressed when it comes to
897
parent references; a index entry may only have parents that with a
898
lover index number. As a result, the index is topological sorted.
899
900
Duplicate entries may be written to the index for a single version id
901
if this is done then the latter one completely replaces the former:
902
this allows updates to correct version and parent information.
903
Note that the two entries may share the delta, and that successive
904
annotations and references MUST point to the first entry.
905
906
The index file on disc contains a header, followed by one line per knit
907
record. The same revision can be present in an index file more than once.
908
The first occurence gets assigned a sequence number starting from 0.
909
910
The format of a single line is
911
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
912
REVISION_ID is a utf8-encoded revision id
913
FLAGS is a comma separated list of flags about the record. Values include
914
no-eol, line-delta, fulltext.
915
BYTE_OFFSET is the ascii representation of the byte offset in the data file
916
that the the compressed data starts at.
917
LENGTH is the ascii representation of the length of the data file.
918
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
919
REVISION_ID.
920
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
921
revision id already in the knit that is a parent of REVISION_ID.
922
The ' :' marker is the end of record marker.
923
924
partial writes:
925
when a write is interrupted to the index file, it will result in a line that
926
does not end in ' :'. If the ' :' is not present at the end of a line, or at
927
the end of the file, then the record that is missing it will be ignored by
928
the parser.
929
930
When writing new records to the index file, the data is preceeded by '\n'
931
to ensure that records always start on new lines even if the last write was
932
interrupted. As a result its normal for the last line in the index to be
933
missing a trailing newline. One can be added with no harmful effects.
934
"""
935
936
HEADER = "# bzr knit index 7\n"
937
938
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
939
# __slots__ = ['_cache', '_history', '_transport', '_filename']
940
941
def _cache_version(self, version_id, options, pos, size, parents):
942
"""Cache a version record in the history array and index cache.
943
944
This is inlined into __init__ for performance. KEEP IN SYNC.
945
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
946
indexes).
947
"""
948
# only want the _history index to reference the 1st index entry
949
# for version_id
950
if version_id not in self._cache:
951
index = len(self._history)
952
self._history.append(version_id)
953
else:
954
index = self._cache[version_id][5]
955
self._cache[version_id] = (version_id,
956
options,
957
pos,
958
size,
959
parents,
960
index)
961
962
def __init__(self, transport, filename, mode, create=False):
963
_KnitComponentFile.__init__(self, transport, filename, mode)
964
self._cache = {}
965
# position in _history is the 'official' index for a revision
966
# but the values may have come from a newer entry.
967
# so - wc -l of a knit index is != the number of uniqe names
968
# in the weave.
969
self._history = []
970
pb = bzrlib.ui.ui_factory.nested_progress_bar()
971
try:
972
count = 0
973
total = 1
974
try:
975
pb.update('read knit index', count, total)
976
fp = self._transport.get(self._filename)
977
self.check_header(fp)
978
# readlines reads the whole file at once:
979
# bad for transports like http, good for local disk
980
# we save 60 ms doing this one change (
981
# from calling readline each time to calling
982
# readlines once.
983
# probably what we want for nice behaviour on
984
# http is a incremental readlines that yields, or
985
# a check for local vs non local indexes,
986
for l in fp.readlines():
987
rec = l.split()
988
if len(rec) < 5 or rec[-1] != ':':
989
# corrupt line.
990
# FIXME: in the future we should determine if its a
991
# short write - and ignore it
992
# or a different failure, and raise. RBC 20060407
993
continue
994
count += 1
995
total += 1
996
#pb.update('read knit index', count, total)
997
# See self._parse_parents
998
parents = []
999
for value in rec[4:-1]:
1000
if '.' == value[0]:
1001
# uncompressed reference
1002
parents.append(value[1:])
1003
else:
1004
# this is 15/4000ms faster than isinstance,
1005
# (in lsprof)
1006
# this function is called thousands of times a
1007
# second so small variations add up.
1008
assert value.__class__ is str
1009
parents.append(self._history[int(value)])
1010
# end self._parse_parents
1011
# self._cache_version(rec[0],
1012
# rec[1].split(','),
1013
# int(rec[2]),
1014
# int(rec[3]),
1015
# parents)
1016
# --- self._cache_version
1017
# only want the _history index to reference the 1st
1018
# index entry for version_id
1019
version_id = rec[0]
1020
if version_id not in self._cache:
1021
index = len(self._history)
1022
self._history.append(version_id)
1023
else:
1024
index = self._cache[version_id][5]
1025
self._cache[version_id] = (version_id,
1026
rec[1].split(','),
1027
int(rec[2]),
1028
int(rec[3]),
1029
parents,
1030
index)
1031
# --- self._cache_version
1032
except NoSuchFile, e:
1033
if mode != 'w' or not create:
1034
raise
1035
self.write_header()
1036
finally:
1037
pb.update('read knit index', total, total)
1038
pb.finished()
1039
1040
def _parse_parents(self, compressed_parents):
1041
"""convert a list of string parent values into version ids.
1042
1043
ints are looked up in the index.
1044
.FOO values are ghosts and converted in to FOO.
1045
1046
NOTE: the function is retained here for clarity, and for possible
1047
use in partial index reads. However bulk processing now has
1048
it inlined in __init__ for inner-loop optimisation.
1049
"""
1050
result = []
1051
for value in compressed_parents:
1052
if value[-1] == '.':
1053
# uncompressed reference
1054
result.append(value[1:])
1055
else:
1056
# this is 15/4000ms faster than isinstance,
1057
# this function is called thousands of times a
1058
# second so small variations add up.
1059
assert value.__class__ is str
1060
result.append(self._history[int(value)])
1061
return result
1062
1063
def get_graph(self):
1064
graph = []
1065
for version_id, index in self._cache.iteritems():
1066
graph.append((version_id, index[4]))
1067
return graph
1068
1069
def get_ancestry(self, versions):
1070
"""See VersionedFile.get_ancestry."""
1071
# get a graph of all the mentioned versions:
1072
graph = {}
1073
pending = set(versions)
1074
while len(pending):
1075
version = pending.pop()
1076
parents = self._cache[version][4]
1077
# got the parents ok
1078
# trim ghosts
1079
parents = [parent for parent in parents if parent in self._cache]
1080
for parent in parents:
1081
# if not completed and not a ghost
1082
if parent not in graph:
1083
pending.add(parent)
1084
graph[version] = parents
1085
return topo_sort(graph.items())
1086
1087
def get_ancestry_with_ghosts(self, versions):
1088
"""See VersionedFile.get_ancestry_with_ghosts."""
1089
# get a graph of all the mentioned versions:
1090
graph = {}
1091
pending = set(versions)
1092
while len(pending):
1093
version = pending.pop()
1094
try:
1095
parents = self._cache[version][4]
1096
except KeyError:
1097
# ghost, fake it
1098
graph[version] = []
1099
pass
1100
else:
1101
# got the parents ok
1102
for parent in parents:
1103
if parent not in graph:
1104
pending.add(parent)
1105
graph[version] = parents
1106
return topo_sort(graph.items())
1107
1108
def num_versions(self):
1109
return len(self._history)
1110
1111
__len__ = num_versions
1112
1113
def get_versions(self):
1114
return self._history
1115
1116
def idx_to_name(self, idx):
1117
return self._history[idx]
1118
1119
def lookup(self, version_id):
1120
assert version_id in self._cache
1121
return self._cache[version_id][5]
1122
1123
def _version_list_to_index(self, versions):
1124
result_list = []
1125
for version in versions:
1126
if version in self._cache:
1127
# -- inlined lookup() --
1128
result_list.append(str(self._cache[version][5]))
1129
# -- end lookup () --
1130
else:
1131
result_list.append('.' + version.encode('utf-8'))
1132
return ' '.join(result_list)
1133
1134
def add_version(self, version_id, options, pos, size, parents):
1135
"""Add a version record to the index."""
1136
self._cache_version(version_id, options, pos, size, parents)
1137
1138
content = "\n%s %s %s %s %s :" % (version_id.encode('utf-8'),
1139
','.join(options),
1140
pos,
1141
size,
1142
self._version_list_to_index(parents))
1143
assert isinstance(content, str), 'content must be utf-8 encoded'
1144
self._transport.append(self._filename, StringIO(content))
1145
1146
def has_version(self, version_id):
1147
"""True if the version is in the index."""
1148
return self._cache.has_key(version_id)
1149
1150
def get_position(self, version_id):
1151
"""Return data position and size of specified version."""
1152
return (self._cache[version_id][2], \
1153
self._cache[version_id][3])
1154
1155
def get_method(self, version_id):
1156
"""Return compression method of specified version."""
1157
options = self._cache[version_id][1]
1158
if 'fulltext' in options:
1159
return 'fulltext'
1160
else:
1161
assert 'line-delta' in options
1162
return 'line-delta'
1163
1164
def get_options(self, version_id):
1165
return self._cache[version_id][1]
1166
1167
def get_parents(self, version_id):
1168
"""Return parents of specified version ignoring ghosts."""
1169
return [parent for parent in self._cache[version_id][4]
1170
if parent in self._cache]
1171
1172
def get_parents_with_ghosts(self, version_id):
1173
"""Return parents of specified version wth ghosts."""
1174
return self._cache[version_id][4]
1175
1176
def check_versions_present(self, version_ids):
1177
"""Check that all specified versions are present."""
1178
version_ids = set(version_ids)
1179
for version_id in list(version_ids):
1180
if version_id in self._cache:
1181
version_ids.remove(version_id)
1182
if version_ids:
1183
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1184
1185
1186
class _KnitData(_KnitComponentFile):
1187
"""Contents of the knit data file"""
1188
1189
HEADER = "# bzr knit data 7\n"
1190
1191
def __init__(self, transport, filename, mode, create=False):
1192
_KnitComponentFile.__init__(self, transport, filename, mode)
1193
self._file = None
1194
self._checked = False
1195
if create:
1196
self._transport.put(self._filename, StringIO(''))
1197
self._records = {}
1198
1199
def clear_cache(self):
1200
"""Clear the record cache."""
1201
self._records = {}
1202
1203
def _open_file(self):
1204
if self._file is None:
1205
try:
1206
self._file = self._transport.get(self._filename)
1207
except NoSuchFile:
1208
pass
1209
return self._file
1210
1211
def _record_to_data(self, version_id, digest, lines):
1212
"""Convert version_id, digest, lines into a raw data block.
1213
1214
:return: (len, a StringIO instance with the raw data ready to read.)
1215
"""
1216
sio = StringIO()
1217
data_file = GzipFile(None, mode='wb', fileobj=sio)
1218
data_file.writelines(chain(
1219
["version %s %d %s\n" % (version_id.encode('utf-8'),
1220
len(lines),
1221
digest)],
1222
lines,
1223
["end %s\n" % version_id.encode('utf-8')]))
1224
data_file.close()
1225
length= sio.tell()
1226
1227
sio.seek(0)
1228
return length, sio
1229
1230
def add_raw_record(self, raw_data):
1231
"""Append a prepared record to the data file."""
1232
assert isinstance(raw_data, str), 'data must be plain bytes'
1233
start_pos = self._transport.append(self._filename, StringIO(raw_data))
1234
return start_pos, len(raw_data)
1235
1236
def add_record(self, version_id, digest, lines):
1237
"""Write new text record to disk. Returns the position in the
1238
file where it was written."""
1239
size, sio = self._record_to_data(version_id, digest, lines)
1240
# cache
1241
self._records[version_id] = (digest, lines)
1242
# write to disk
1243
start_pos = self._transport.append(self._filename, sio)
1244
return start_pos, size
1245
1246
def _parse_record_header(self, version_id, raw_data):
1247
"""Parse a record header for consistency.
1248
1249
:return: the header and the decompressor stream.
1250
as (stream, header_record)
1251
"""
1252
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1253
rec = df.readline().split()
1254
if len(rec) != 4:
1255
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
1256
if rec[1].decode('utf-8')!= version_id:
1257
raise KnitCorrupt(self._filename,
1258
'unexpected version, wanted %r, got %r' % (
1259
version_id, rec[1]))
1260
return df, rec
1261
1262
def _parse_record(self, version_id, data):
1263
# profiling notes:
1264
# 4168 calls in 2880 217 internal
1265
# 4168 calls to _parse_record_header in 2121
1266
# 4168 calls to readlines in 330
1267
df, rec = self._parse_record_header(version_id, data)
1268
record_contents = df.readlines()
1269
l = record_contents.pop()
1270
assert len(record_contents) == int(rec[2])
1271
if l.decode('utf-8') != 'end %s\n' % version_id:
1272
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
1273
% (l, version_id))
1274
df.close()
1275
return record_contents, rec[3]
1276
1277
def read_records_iter_raw(self, records):
1278
"""Read text records from data file and yield raw data.
1279
1280
This unpacks enough of the text record to validate the id is
1281
as expected but thats all.
1282
1283
It will actively recompress currently cached records on the
1284
basis that that is cheaper than I/O activity.
1285
"""
1286
needed_records = []
1287
for version_id, pos, size in records:
1288
if version_id not in self._records:
1289
needed_records.append((version_id, pos, size))
1290
1291
# setup an iterator of the external records:
1292
# uses readv so nice and fast we hope.
1293
if len(needed_records):
1294
# grab the disk data needed.
1295
raw_records = self._transport.readv(self._filename,
1296
[(pos, size) for version_id, pos, size in needed_records])
1297
1298
for version_id, pos, size in records:
1299
if version_id in self._records:
1300
# compress a new version
1301
size, sio = self._record_to_data(version_id,
1302
self._records[version_id][0],
1303
self._records[version_id][1])
1304
yield version_id, sio.getvalue()
1305
else:
1306
pos, data = raw_records.next()
1307
# validate the header
1308
df, rec = self._parse_record_header(version_id, data)
1309
df.close()
1310
yield version_id, data
1311
1312
1313
def read_records_iter(self, records):
1314
"""Read text records from data file and yield result.
1315
1316
Each passed record is a tuple of (version_id, pos, len) and
1317
will be read in the given order. Yields (version_id,
1318
contents, digest).
1319
"""
1320
# profiling notes:
1321
# 60890 calls for 4168 extractions in 5045, 683 internal.
1322
# 4168 calls to readv in 1411
1323
# 4168 calls to parse_record in 2880
1324
1325
needed_records = []
1326
for version_id, pos, size in records:
1327
if version_id not in self._records:
1328
needed_records.append((version_id, pos, size))
1329
1330
if len(needed_records):
1331
# We take it that the transport optimizes the fetching as good
1332
# as possible (ie, reads continous ranges.)
1333
response = self._transport.readv(self._filename,
1334
[(pos, size) for version_id, pos, size in needed_records])
1335
1336
for (record_id, pos, size), (pos, data) in izip(iter(needed_records), response):
1337
content, digest = self._parse_record(record_id, data)
1338
self._records[record_id] = (digest, content)
1339
1340
for version_id, pos, size in records:
1341
yield version_id, list(self._records[version_id][1]), self._records[version_id][0]
1342
1343
def read_records(self, records):
1344
"""Read records into a dictionary."""
1345
components = {}
1346
for record_id, content, digest in self.read_records_iter(records):
1347
components[record_id] = (content, digest)
1348
return components
1349
1350
1351
class InterKnit(InterVersionedFile):
1352
"""Optimised code paths for knit to knit operations."""
1353
1354
_matching_file_factory = KnitVersionedFile
1355
1356
@staticmethod
1357
def is_compatible(source, target):
1358
"""Be compatible with knits. """
1359
try:
1360
return (isinstance(source, KnitVersionedFile) and
1361
isinstance(target, KnitVersionedFile))
1362
except AttributeError:
1363
return False
1364
1365
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1366
"""See InterVersionedFile.join."""
1367
assert isinstance(self.source, KnitVersionedFile)
1368
assert isinstance(self.target, KnitVersionedFile)
1369
1370
if version_ids is None:
1371
version_ids = self.source.versions()
1372
else:
1373
if not ignore_missing:
1374
self.source._check_versions_present(version_ids)
1375
else:
1376
version_ids = set(self.source.versions()).intersection(
1377
set(version_ids))
1378
1379
if not version_ids:
1380
return 0
1381
1382
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1383
try:
1384
version_ids = list(version_ids)
1385
if None in version_ids:
1386
version_ids.remove(None)
1387
1388
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1389
this_versions = set(self.target._index.get_versions())
1390
needed_versions = self.source_ancestry - this_versions
1391
cross_check_versions = self.source_ancestry.intersection(this_versions)
1392
mismatched_versions = set()
1393
for version in cross_check_versions:
1394
# scan to include needed parents.
1395
n1 = set(self.target.get_parents_with_ghosts(version))
1396
n2 = set(self.source.get_parents_with_ghosts(version))
1397
if n1 != n2:
1398
# FIXME TEST this check for cycles being introduced works
1399
# the logic is we have a cycle if in our graph we are an
1400
# ancestor of any of the n2 revisions.
1401
for parent in n2:
1402
if parent in n1:
1403
# safe
1404
continue
1405
else:
1406
parent_ancestors = self.source.get_ancestry(parent)
1407
if version in parent_ancestors:
1408
raise errors.GraphCycleError([parent, version])
1409
# ensure this parent will be available later.
1410
new_parents = n2.difference(n1)
1411
needed_versions.update(new_parents.difference(this_versions))
1412
mismatched_versions.add(version)
1413
1414
if not needed_versions and not cross_check_versions:
1415
return 0
1416
full_list = topo_sort(self.source.get_graph())
1417
1418
version_list = [i for i in full_list if (not self.target.has_version(i)
1419
and i in needed_versions)]
1420
1421
# plan the join:
1422
copy_queue = []
1423
copy_queue_records = []
1424
copy_set = set()
1425
for version_id in version_list:
1426
options = self.source._index.get_options(version_id)
1427
parents = self.source._index.get_parents_with_ghosts(version_id)
1428
# check that its will be a consistent copy:
1429
for parent in parents:
1430
# if source has the parent, we must :
1431
# * already have it or
1432
# * have it scheduled already
1433
# otherwise we dont care
1434
assert (self.target.has_version(parent) or
1435
parent in copy_set or
1436
not self.source.has_version(parent))
1437
data_pos, data_size = self.source._index.get_position(version_id)
1438
copy_queue_records.append((version_id, data_pos, data_size))
1439
copy_queue.append((version_id, options, parents))
1440
copy_set.add(version_id)
1441
1442
# data suck the join:
1443
count = 0
1444
total = len(version_list)
1445
# we want the raw gzip for bulk copying, but the record validated
1446
# just enough to be sure its the right one.
1447
# TODO: consider writev or write combining to reduce
1448
# death of a thousand cuts feeling.
1449
for (version_id, raw_data), \
1450
(version_id2, options, parents) in \
1451
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1452
copy_queue):
1453
assert version_id == version_id2, 'logic error, inconsistent results'
1454
count = count + 1
1455
pb.update("Joining knit", count, total)
1456
pos, size = self.target._data.add_raw_record(raw_data)
1457
self.target._index.add_version(version_id, options, pos, size, parents)
1458
1459
for version in mismatched_versions:
1460
# FIXME RBC 20060309 is this needed?
1461
n1 = set(self.target.get_parents_with_ghosts(version))
1462
n2 = set(self.source.get_parents_with_ghosts(version))
1463
# write a combined record to our history preserving the current
1464
# parents as first in the list
1465
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1466
self.target.fix_parents(version, new_parents)
1467
return count
1468
finally:
1469
pb.finished()
1470
1471
1472
InterVersionedFile.register_optimiser(InterKnit)
1473
1474
1475
class SequenceMatcher(difflib.SequenceMatcher):
1476
"""Knit tuned sequence matcher.
1477
1478
This is based on profiling of difflib which indicated some improvements
1479
for our usage pattern.
1480
"""
1481
1482
def find_longest_match(self, alo, ahi, blo, bhi):
1483
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1484
1485
If isjunk is not defined:
1486
1487
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1488
alo <= i <= i+k <= ahi
1489
blo <= j <= j+k <= bhi
1490
and for all (i',j',k') meeting those conditions,
1491
k >= k'
1492
i <= i'
1493
and if i == i', j <= j'
1494
1495
In other words, of all maximal matching blocks, return one that
1496
starts earliest in a, and of all those maximal matching blocks that
1497
start earliest in a, return the one that starts earliest in b.
1498
1499
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1500
>>> s.find_longest_match(0, 5, 0, 9)
1501
(0, 4, 5)
1502
1503
If isjunk is defined, first the longest matching block is
1504
determined as above, but with the additional restriction that no
1505
junk element appears in the block. Then that block is extended as
1506
far as possible by matching (only) junk elements on both sides. So
1507
the resulting block never matches on junk except as identical junk
1508
happens to be adjacent to an "interesting" match.
1509
1510
Here's the same example as before, but considering blanks to be
1511
junk. That prevents " abcd" from matching the " abcd" at the tail
1512
end of the second sequence directly. Instead only the "abcd" can
1513
match, and matches the leftmost "abcd" in the second sequence:
1514
1515
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1516
>>> s.find_longest_match(0, 5, 0, 9)
1517
(1, 0, 4)
1518
1519
If no blocks match, return (alo, blo, 0).
1520
1521
>>> s = SequenceMatcher(None, "ab", "c")
1522
>>> s.find_longest_match(0, 2, 0, 1)
1523
(0, 0, 0)
1524
"""
1525
1526
# CAUTION: stripping common prefix or suffix would be incorrect.
1527
# E.g.,
1528
# ab
1529
# acab
1530
# Longest matching block is "ab", but if common prefix is
1531
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1532
# strip, so ends up claiming that ab is changed to acab by
1533
# inserting "ca" in the middle. That's minimal but unintuitive:
1534
# "it's obvious" that someone inserted "ac" at the front.
1535
# Windiff ends up at the same place as diff, but by pairing up
1536
# the unique 'b's and then matching the first two 'a's.
1537
1538
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1539
besti, bestj, bestsize = alo, blo, 0
1540
# find longest junk-free match
1541
# during an iteration of the loop, j2len[j] = length of longest
1542
# junk-free match ending with a[i-1] and b[j]
1543
j2len = {}
1544
# nothing = []
1545
b2jget = b2j.get
1546
for i in xrange(alo, ahi):
1547
# look at all instances of a[i] in b; note that because
1548
# b2j has no junk keys, the loop is skipped if a[i] is junk
1549
j2lenget = j2len.get
1550
newj2len = {}
1551
1552
# changing b2j.get(a[i], nothing) to a try:Keyerror pair produced the
1553
# following improvement
1554
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1555
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1556
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1557
# to
1558
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1559
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1560
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1561
1562
try:
1563
js = b2j[a[i]]
1564
except KeyError:
1565
pass
1566
else:
1567
for j in js:
1568
# a[i] matches b[j]
1569
if j >= blo:
1570
if j >= bhi:
1571
break
1572
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1573
if k > bestsize:
1574
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1575
j2len = newj2len
1576
1577
# Extend the best by non-junk elements on each end. In particular,
1578
# "popular" non-junk elements aren't in b2j, which greatly speeds
1579
# the inner loop above, but also means "the best" match so far
1580
# doesn't contain any junk *or* popular non-junk elements.
1581
while besti > alo and bestj > blo and \
1582
not isbjunk(b[bestj-1]) and \
1583
a[besti-1] == b[bestj-1]:
1584
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1585
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1586
not isbjunk(b[bestj+bestsize]) and \
1587
a[besti+bestsize] == b[bestj+bestsize]:
1588
bestsize += 1
1589
1590
# Now that we have a wholly interesting match (albeit possibly
1591
# empty!), we may as well suck up the matching junk on each
1592
# side of it too. Can't think of a good reason not to, and it
1593
# saves post-processing the (possibly considerable) expense of
1594
# figuring out what to do with it. In the case of an empty
1595
# interesting match, this is clearly the right thing to do,
1596
# because no other kind of match is possible in the regions.
1597
while besti > alo and bestj > blo and \
1598
isbjunk(b[bestj-1]) and \
1599
a[besti-1] == b[bestj-1]:
1600
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1601
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1602
isbjunk(b[bestj+bestsize]) and \
1603
a[besti+bestsize] == b[bestj+bestsize]:
1604
bestsize = bestsize + 1
1605
1606
return besti, bestj, bestsize
1607
Loggerhead is a web-based interface for Breezy
Version: 2.0.1