« back to all changes in this revision
Viewing changes to bzrlib/knit.py
- browse files at revision 2245.4.7
- compare with another revision
- download diff
- download tarball
- view history from revision 2245.4.7
- Committer: Alexander Belchenko
- Date: 2007-01-30 23:05:35 UTC
- mto: This revision was merged to the branch mainline in revision 2259.
- Revision ID: bialix@ukr.net-20070130230535-kx1rd478rtigyc3v
standalone installer: win98 support
- files added:
- NEWS.developers
- bzrlib/bundle/old
- files removed:
- bzrlib/_dirstate_helpers_c.h
- bzrlib/help_topics
- bzrlib/help_topics/en
- bzrlib/smart
- bzrlib/tests/commands
- bzrlib/tests/inventory_implementations
- bzrlib/tests/per_lock
- bzrlib/util/tests
- doc/developers
- doc/en
- doc/en/admin-guide
- doc/en/developer-guide
- doc/en/mini-tutorial
- doc/en/quick-reference
- doc/en/release-notes
- doc/en/tutorials
- doc/en/user-guide
- doc/en/user-guide/images
- doc/en/user-reference
- man1
- files renamed:
- doc/developers/HACKING.txt => HACKING
- bzrlib/deprecated_graph.py => bzrlib/graph.py
- bzrlib/help_topics/__init__.py => bzrlib/help_topics.py
- bzrlib/_patiencediff_py.py => bzrlib/patiencediff.py
- bzrlib/tests/http_utils.py => bzrlib/tests/HTTPTestUtil.py
- bzrlib/tests/http_server.py => bzrlib/tests/HttpServer.py
- bzrlib/tests/blackbox/test_send.py => bzrlib/tests/blackbox/test_bundle.py
- bzrlib/tests/test_deprecated_graph.py => bzrlib/tests/test_graph.py
- bzrlib/transport/remote.py => bzrlib/transport/smart.py
- doc/en/tutorials/centralized_workflow.txt => doc/centralized_workflow.txt
- bzrlib/help_topics/en/configuration.txt => doc/configuration.txt
- doc/en/user-guide/http_smart_server.txt => doc/http_smart_server.txt
- doc/en/user-guide/index.txt => doc/index.txt
- doc/en/user-guide/plugins.txt => doc/plugins.txt
- doc/en/user-guide/server.txt => doc/server.txt
- doc/en/user-guide/setting_up_email.txt => doc/setting_up_email.txt
- doc/en/user-guide/specifying_revisions.txt => doc/specifying_revisions.txt
- doc/en/tutorials/tutorial.txt => doc/tutorial.txt
- doc/en/user-guide/using_aliases.txt => doc/using_aliases.txt
- doc/en/user-guide/version_info.txt => doc/version_info.txt
- files modified:
- .bzrignore
- setup.py *
added
removed
bzrlib/knit.py
1
# Copyright (C) 2005, 2006, 2007 Canonical Ltd
1
# Copyright (C) 2005, 2006 Canonical Ltd
2
2
#
3
3
# This program is free software; you can redistribute it and/or modify
4
4
# it under the terms of the GNU General Public License as published by
62
62
63
63
from copy import copy
64
64
from cStringIO import StringIO
65
import difflib
65
66
from itertools import izip, chain
66
67
import operator
67
68
import os
68
69
import sys
69
70
import warnings
70
from zlib import Z_DEFAULT_COMPRESSION
71
71
72
72
import bzrlib
73
from bzrlib.lazy_import import lazy_import
74
lazy_import(globals(), """
75
from bzrlib import (
76
annotate,
77
graph as _mod_graph,
78
lru_cache,
79
pack,
80
trace,
81
)
82
""")
83
73
from bzrlib import (
84
74
cache_utf8,
85
debug,
86
diff,
87
75
errors,
88
osutils,
89
76
patiencediff,
90
77
progress,
91
merge,
92
78
ui,
93
79
)
94
80
from bzrlib.errors import (
101
87
RevisionNotPresent,
102
88
RevisionAlreadyPresent,
103
89
)
104
from bzrlib.tuned_gzip import GzipFile, bytes_to_gzip
90
from bzrlib.tuned_gzip import GzipFile
91
from bzrlib.trace import mutter
105
92
from bzrlib.osutils import (
106
93
contains_whitespace,
107
94
contains_linebreaks,
108
sha_string,
109
95
sha_strings,
110
96
)
111
97
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
135
121
class KnitContent(object):
136
122
"""Content of a knit version to which deltas can be applied."""
137
123
138
def __init__(self):
139
self._should_strip_eol = False
124
def __init__(self, lines):
125
self._lines = lines
126
127
def annotate_iter(self):
128
"""Yield tuples of (origin, text) for each content line."""
129
return iter(self._lines)
140
130
141
131
def annotate(self):
142
132
"""Return a list of (origin, text) tuples."""
143
133
return list(self.annotate_iter())
144
134
145
def apply_delta(self, delta, new_version_id):
146
"""Apply delta to this object to become new_version_id."""
147
raise NotImplementedError(self.apply_delta)
148
149
def cleanup_eol(self, copy_on_mutate=True):
150
if self._should_strip_eol:
151
if copy_on_mutate:
152
self._lines = self._lines[:]
153
self.strip_last_line_newline()
154
155
135
def line_delta_iter(self, new_lines):
156
136
"""Generate line-based delta from this content to new_lines."""
157
137
new_texts = new_lines.text()
158
138
old_texts = self.text()
159
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
139
s = KnitSequenceMatcher(None, old_texts, new_texts)
160
140
for tag, i1, i2, j1, j2 in s.get_opcodes():
161
141
if tag == 'equal':
162
142
continue
166
146
def line_delta(self, new_lines):
167
147
return list(self.line_delta_iter(new_lines))
168
148
169
@staticmethod
170
def get_line_delta_blocks(knit_delta, source, target):
171
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
172
target_len = len(target)
173
s_pos = 0
174
t_pos = 0
175
for s_begin, s_end, t_len, new_text in knit_delta:
176
true_n = s_begin - s_pos
177
n = true_n
178
if n > 0:
179
# knit deltas do not provide reliable info about whether the
180
# last line of a file matches, due to eol handling.
181
if source[s_pos + n -1] != target[t_pos + n -1]:
182
n-=1
183
if n > 0:
184
yield s_pos, t_pos, n
185
t_pos += t_len + true_n
186
s_pos = s_end
187
n = target_len - t_pos
188
if n > 0:
189
if source[s_pos + n -1] != target[t_pos + n -1]:
190
n-=1
191
if n > 0:
192
yield s_pos, t_pos, n
193
yield s_pos + (target_len - t_pos), target_len, 0
194
195
196
class AnnotatedKnitContent(KnitContent):
197
"""Annotated content."""
198
199
def __init__(self, lines):
200
KnitContent.__init__(self)
201
self._lines = lines
202
203
def annotate_iter(self):
204
"""Yield tuples of (origin, text) for each content line."""
205
return iter(self._lines)
206
207
def apply_delta(self, delta, new_version_id):
208
"""Apply delta to this object to become new_version_id."""
209
offset = 0
210
lines = self._lines
211
for start, end, count, delta_lines in delta:
212
lines[offset+start:offset+end] = delta_lines
213
offset = offset + (start - end) + count
214
215
def strip_last_line_newline(self):
216
line = self._lines[-1][1].rstrip('\n')
217
self._lines[-1] = (self._lines[-1][0], line)
218
self._should_strip_eol = False
219
220
def text(self):
221
try:
222
lines = [text for origin, text in self._lines]
223
except ValueError, e:
224
# most commonly (only?) caused by the internal form of the knit
225
# missing annotation information because of a bug - see thread
226
# around 20071015
227
raise KnitCorrupt(self,
228
"line in annotated knit missing annotation information: %s"
229
% (e,))
230
231
if self._should_strip_eol:
232
anno, line = lines[-1]
233
lines[-1] = (anno, line.rstrip('\n'))
234
return lines
235
236
def copy(self):
237
return AnnotatedKnitContent(self._lines[:])
238
239
240
class PlainKnitContent(KnitContent):
241
"""Unannotated content.
242
243
When annotate[_iter] is called on this content, the same version is reported
244
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
245
objects.
246
"""
247
248
def __init__(self, lines, version_id):
249
KnitContent.__init__(self)
250
self._lines = lines
251
self._version_id = version_id
252
253
def annotate_iter(self):
254
"""Yield tuples of (origin, text) for each content line."""
255
for line in self._lines:
256
yield self._version_id, line
257
258
def apply_delta(self, delta, new_version_id):
259
"""Apply delta to this object to become new_version_id."""
260
offset = 0
261
lines = self._lines
262
for start, end, count, delta_lines in delta:
263
lines[offset+start:offset+end] = delta_lines
264
offset = offset + (start - end) + count
265
self._version_id = new_version_id
266
267
def copy(self):
268
return PlainKnitContent(self._lines[:], self._version_id)
269
270
def strip_last_line_newline(self):
271
self._lines[-1] = self._lines[-1].rstrip('\n')
272
self._should_strip_eol = False
273
274
def text(self):
275
lines = self._lines
276
if self._should_strip_eol:
277
lines = lines[:]
278
lines[-1] = lines[-1].rstrip('\n')
279
return lines
149
def text(self):
150
return [text for origin, text in self._lines]
151
152
def copy(self):
153
return KnitContent(self._lines[:])
280
154
281
155
282
156
class _KnitFactory(object):
283
"""Base class for common Factory functions."""
284
285
def parse_record(self, version_id, record, record_details,
286
base_content, copy_base_content=True):
287
"""Parse a record into a full content object.
288
289
:param version_id: The official version id for this content
290
:param record: The data returned by read_records_iter()
291
:param record_details: Details about the record returned by
292
get_build_details
293
:param base_content: If get_build_details returns a compression_parent,
294
you must return a base_content here, else use None
295
:param copy_base_content: When building from the base_content, decide
296
you can either copy it and return a new object, or modify it in
297
place.
298
:return: (content, delta) A Content object and possibly a line-delta,
299
delta may be None
300
"""
301
method, noeol = record_details
302
if method == 'line-delta':
303
assert base_content is not None
304
if copy_base_content:
305
content = base_content.copy()
306
else:
307
content = base_content
308
delta = self.parse_line_delta(record, version_id)
309
content.apply_delta(delta, version_id)
310
else:
311
content = self.parse_fulltext(record, version_id)
312
delta = None
313
content._should_strip_eol = noeol
314
return (content, delta)
157
"""Base factory for creating content objects."""
158
159
def make(self, lines, version):
160
num_lines = len(lines)
161
return KnitContent(zip([version] * num_lines, lines))
315
162
316
163
317
164
class KnitAnnotateFactory(_KnitFactory):
319
166
320
167
annotated = True
321
168
322
def make(self, lines, version_id):
323
num_lines = len(lines)
324
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
325
326
def parse_fulltext(self, content, version_id):
169
def parse_fulltext(self, content, version):
327
170
"""Convert fulltext to internal representation
328
171
329
172
fulltext content is of the format
331
174
internal representation is of the format:
332
175
(revid, plaintext)
333
176
"""
334
# TODO: jam 20070209 The tests expect this to be returned as tuples,
335
# but the code itself doesn't really depend on that.
336
# Figure out a way to not require the overhead of turning the
337
# list back into tuples.
338
lines = [tuple(line.split(' ', 1)) for line in content]
339
return AnnotatedKnitContent(lines)
177
decode_utf8 = cache_utf8.decode
178
lines = []
179
for line in content:
180
origin, text = line.split(' ', 1)
181
lines.append((decode_utf8(origin), text))
182
return KnitContent(lines)
340
183
341
184
def parse_line_delta_iter(self, lines):
342
185
return iter(self.parse_line_delta(lines))
343
186
344
def parse_line_delta(self, lines, version_id, plain=False):
187
def parse_line_delta(self, lines, version):
345
188
"""Convert a line based delta into internal representation.
346
189
347
190
line delta is in the form of:
350
193
revid(utf8) newline\n
351
194
internal representation is
352
195
(start, end, count, [1..count tuples (revid, newline)])
353
354
:param plain: If True, the lines are returned as a plain
355
list without annotations, not as a list of (origin, content) tuples, i.e.
356
(start, end, count, [1..count newline])
357
196
"""
197
decode_utf8 = cache_utf8.decode
358
198
result = []
359
199
lines = iter(lines)
360
200
next = lines.next
361
362
cache = {}
363
def cache_and_return(line):
364
origin, text = line.split(' ', 1)
365
return cache.setdefault(origin, origin), text
366
367
201
# walk through the lines parsing.
368
# Note that the plain test is explicitly pulled out of the
369
# loop to minimise any performance impact
370
if plain:
371
for header in lines:
372
start, end, count = [int(n) for n in header.split(',')]
373
contents = [next().split(' ', 1)[1] for i in xrange(count)]
374
result.append((start, end, count, contents))
375
else:
376
for header in lines:
377
start, end, count = [int(n) for n in header.split(',')]
378
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
379
result.append((start, end, count, contents))
202
for header in lines:
203
start, end, count = [int(n) for n in header.split(',')]
204
contents = []
205
remaining = count
206
while remaining:
207
origin, text = next().split(' ', 1)
208
remaining -= 1
209
contents.append((decode_utf8(origin), text))
210
result.append((start, end, count, contents))
380
211
return result
381
212
382
213
def get_fulltext_content(self, lines):
403
234
404
235
see parse_fulltext which this inverts.
405
236
"""
406
# TODO: jam 20070209 We only do the caching thing to make sure that
407
# the origin is a valid utf-8 line, eventually we could remove it
408
return ['%s %s' % (o, t) for o, t in content._lines]
237
encode_utf8 = cache_utf8.encode
238
return ['%s %s' % (encode_utf8(o), t) for o, t in content._lines]
409
239
410
240
def lower_line_delta(self, delta):
411
241
"""convert a delta into a serializable form.
412
242
413
243
See parse_line_delta which this inverts.
414
244
"""
415
# TODO: jam 20070209 We only do the caching thing to make sure that
416
# the origin is a valid utf-8 line, eventually we could remove it
245
encode_utf8 = cache_utf8.encode
417
246
out = []
418
247
for start, end, c, lines in delta:
419
248
out.append('%d,%d,%d\n' % (start, end, c))
420
out.extend(origin + ' ' + text
249
out.extend(encode_utf8(origin) + ' ' + text
421
250
for origin, text in lines)
422
251
return out
423
252
424
def annotate_iter(self, knit, version_id):
425
content = knit._get_content(version_id)
426
return content.annotate_iter()
427
428
253
429
254
class KnitPlainFactory(_KnitFactory):
430
255
"""Factory for creating plain Content objects."""
431
256
432
257
annotated = False
433
258
434
def make(self, lines, version_id):
435
return PlainKnitContent(lines, version_id)
436
437
def parse_fulltext(self, content, version_id):
259
def parse_fulltext(self, content, version):
438
260
"""This parses an unannotated fulltext.
439
261
440
262
Note that this is not a noop - the internal representation
441
263
has (versionid, line) - its just a constant versionid.
442
264
"""
443
return self.make(content, version_id)
265
return self.make(content, version)
444
266
445
def parse_line_delta_iter(self, lines, version_id):
267
def parse_line_delta_iter(self, lines, version):
446
268
cur = 0
447
269
num_lines = len(lines)
448
270
while cur < num_lines:
449
271
header = lines[cur]
450
272
cur += 1
451
273
start, end, c = [int(n) for n in header.split(',')]
452
yield start, end, c, lines[cur:cur+c]
274
yield start, end, c, zip([version] * c, lines[cur:cur+c])
453
275
cur += c
454
276
455
def parse_line_delta(self, lines, version_id):
456
return list(self.parse_line_delta_iter(lines, version_id))
277
def parse_line_delta(self, lines, version):
278
return list(self.parse_line_delta_iter(lines, version))
457
279
458
280
def get_fulltext_content(self, lines):
459
281
"""Extract just the content lines from a fulltext."""
480
302
out = []
481
303
for start, end, c, lines in delta:
482
304
out.append('%d,%d,%d\n' % (start, end, c))
483
out.extend(lines)
305
out.extend([text for origin, text in lines])
484
306
return out
485
307
486
def annotate_iter(self, knit, version_id):
487
annotator = _KnitAnnotator(knit)
488
return iter(annotator.annotate(version_id))
489
490
308
491
309
def make_empty_knit(transport, relpath):
492
310
"""Construct a empty knit at the specified location."""
493
311
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
312
k._data._open_file()
494
313
495
314
496
315
class KnitVersionedFile(VersionedFile):
509
328
"""
510
329
511
330
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
512
factory=None, delta=True, create=False, create_parent_dir=False,
513
delay_create=False, dir_mode=None, index=None, access_method=None):
331
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
332
create=False, create_parent_dir=False, delay_create=False,
333
dir_mode=None):
514
334
"""Construct a knit at location specified by relpath.
515
335
516
336
:param create: If not True, only open an existing knit.
519
339
hash-prefixes that may not exist yet)
520
340
:param delay_create: The calling code is aware that the knit won't
521
341
actually be created until the first data is stored.
522
:param index: An index to use for the knit.
523
342
"""
343
if deprecated_passed(basis_knit):
344
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
345
" deprecated as of bzr 0.9.",
346
DeprecationWarning, stacklevel=2)
524
347
if access_mode is None:
525
348
access_mode = 'w'
526
349
super(KnitVersionedFile, self).__init__(access_mode)
533
356
534
357
self._max_delta_chain = 200
535
358
536
if index is None:
537
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
538
access_mode, create=create, file_mode=file_mode,
539
create_parent_dir=create_parent_dir, delay_create=delay_create,
540
dir_mode=dir_mode)
541
else:
542
self._index = index
543
if access_method is None:
544
_access = _KnitAccess(transport, relpath + DATA_SUFFIX, file_mode, dir_mode,
545
((create and not len(self)) and delay_create), create_parent_dir)
546
else:
547
_access = access_method
548
if create and not len(self) and not delay_create:
549
_access.create()
550
self._data = _KnitData(_access)
359
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
360
access_mode, create=create, file_mode=file_mode,
361
create_parent_dir=create_parent_dir, delay_create=delay_create,
362
dir_mode=dir_mode)
363
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
364
access_mode, create=create and not len(self), file_mode=file_mode,
365
create_parent_dir=create_parent_dir, delay_create=delay_create,
366
dir_mode=dir_mode)
551
367
552
368
def __repr__(self):
553
return '%s(%s)' % (self.__class__.__name__,
369
return '%s(%s)' % (self.__class__.__name__,
554
370
self.transport.abspath(self.filename))
555
371
556
372
def _check_should_delta(self, first_parents):
570
386
for count in xrange(self._max_delta_chain):
571
387
parent = delta_parents[0]
572
388
method = self._index.get_method(parent)
573
index, pos, size = self._index.get_position(parent)
389
pos, size = self._index.get_position(parent)
574
390
if method == 'fulltext':
575
391
fulltext_size = size
576
392
break
582
398
583
399
return fulltext_size > delta_size
584
400
401
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
402
"""See VersionedFile._add_delta()."""
403
self._check_add(version_id, []) # should we check the lines ?
404
self._check_versions_present(parents)
405
present_parents = []
406
ghosts = []
407
parent_texts = {}
408
for parent in parents:
409
if not self.has_version(parent):
410
ghosts.append(parent)
411
else:
412
present_parents.append(parent)
413
414
if delta_parent is None:
415
# reconstitute as full text.
416
assert len(delta) == 1 or len(delta) == 0
417
if len(delta):
418
assert delta[0][0] == 0
419
assert delta[0][1] == 0, delta[0][1]
420
return super(KnitVersionedFile, self)._add_delta(version_id,
421
parents,
422
delta_parent,
423
sha1,
424
noeol,
425
delta)
426
427
digest = sha1
428
429
options = []
430
if noeol:
431
options.append('no-eol')
432
433
if delta_parent is not None:
434
# determine the current delta chain length.
435
# To speed the extract of texts the delta chain is limited
436
# to a fixed number of deltas. This should minimize both
437
# I/O and the time spend applying deltas.
438
# The window was changed to a maximum of 200 deltas, but also added
439
# was a check that the total compressed size of the deltas is
440
# smaller than the compressed size of the fulltext.
441
if not self._check_should_delta([delta_parent]):
442
# We don't want a delta here, just do a normal insertion.
443
return super(KnitVersionedFile, self)._add_delta(version_id,
444
parents,
445
delta_parent,
446
sha1,
447
noeol,
448
delta)
449
450
options.append('line-delta')
451
store_lines = self.factory.lower_line_delta(delta)
452
453
where, size = self._data.add_record(version_id, digest, store_lines)
454
self._index.add_version(version_id, options, where, size, parents)
455
585
456
def _add_raw_records(self, records, data):
586
457
"""Add all the records 'records' with data pre-joined in 'data'.
587
458
591
462
the preceding records sizes.
592
463
"""
593
464
# write all the data
594
raw_record_sizes = [record[3] for record in records]
595
positions = self._data.add_raw_records(raw_record_sizes, data)
465
pos = self._data.add_raw_record(data)
596
466
offset = 0
597
467
index_entries = []
598
for (version_id, options, parents, size), access_memo in zip(
599
records, positions):
600
index_entries.append((version_id, options, access_memo, parents))
468
for (version_id, options, parents, size) in records:
469
index_entries.append((version_id, options, pos+offset,
470
size, parents))
601
471
if self._data._do_cache:
602
472
self._data._cache[version_id] = data[offset:offset+size]
603
473
offset += size
630
500
return KnitVersionedFile(name, transport, factory=self.factory,
631
501
delta=self.delta, create=True)
632
502
633
def get_data_stream(self, required_versions):
634
"""Get a data stream for the specified versions.
635
636
Versions may be returned in any order, not necessarily the order
637
specified. They are returned in a partial order by compression
638
parent, so that the deltas can be applied as the data stream is
639
inserted; however note that compression parents will not be sent
640
unless they were specifically requested, as the client may already
641
have them.
642
643
:param required_versions: The exact set of versions to be extracted.
644
Unlike some other knit methods, this is not used to generate a
645
transitive closure, rather it is used precisely as given.
503
def _fix_parents(self, version, new_parents):
504
"""Fix the parents list for version.
646
505
647
:returns: format_signature, list of (version, options, length, parents),
648
reader_callable.
506
This is done by appending a new version to the index
507
with identical data except for the parents list.
508
the parents list must be a superset of the current
509
list.
649
510
"""
650
required_version_set = frozenset(required_versions)
651
version_index = {}
652
# list of revisions that can just be sent without waiting for their
653
# compression parent
654
ready_to_send = []
655
# map from revision to the children based on it
656
deferred = {}
657
# first, read all relevant index data, enough to sort into the right
658
# order to return
659
for version_id in required_versions:
660
options = self._index.get_options(version_id)
661
parents = self._index.get_parents_with_ghosts(version_id)
662
index_memo = self._index.get_position(version_id)
663
version_index[version_id] = (index_memo, options, parents)
664
if ('line-delta' in options
665
and parents[0] in required_version_set):
666
# must wait until the parent has been sent
667
deferred.setdefault(parents[0], []). \
668
append(version_id)
669
else:
670
# either a fulltext, or a delta whose parent the client did
671
# not ask for and presumably already has
672
ready_to_send.append(version_id)
673
# build a list of results to return, plus instructions for data to
674
# read from the file
675
copy_queue_records = []
676
temp_version_list = []
677
while ready_to_send:
678
# XXX: pushing and popping lists may be a bit inefficient
679
version_id = ready_to_send.pop(0)
680
(index_memo, options, parents) = version_index[version_id]
681
copy_queue_records.append((version_id, index_memo))
682
none, data_pos, data_size = index_memo
683
temp_version_list.append((version_id, options, data_size,
684
parents))
685
if version_id in deferred:
686
# now we can send all the children of this revision - we could
687
# put them in anywhere, but we hope that sending them soon
688
# after the fulltext will give good locality in the receiver
689
ready_to_send[:0] = deferred.pop(version_id)
690
assert len(deferred) == 0, \
691
"Still have compressed child versions waiting to be sent"
692
# XXX: The stream format is such that we cannot stream it - we have to
693
# know the length of all the data a-priori.
694
raw_datum = []
695
result_version_list = []
696
for (version_id, raw_data), \
697
(version_id2, options, _, parents) in \
698
izip(self._data.read_records_iter_raw(copy_queue_records),
699
temp_version_list):
700
assert version_id == version_id2, \
701
'logic error, inconsistent results'
702
raw_datum.append(raw_data)
703
result_version_list.append(
704
(version_id, options, len(raw_data), parents))
705
# provide a callback to get data incrementally.
706
pseudo_file = StringIO(''.join(raw_datum))
707
def read(length):
708
if length is None:
709
return pseudo_file.read()
710
else:
711
return pseudo_file.read(length)
712
return (self.get_format_signature(), result_version_list, read)
713
714
def _extract_blocks(self, version_id, source, target):
715
if self._index.get_method(version_id) != 'line-delta':
716
return None
717
parent, sha1, noeol, delta = self.get_delta(version_id)
718
return KnitContent.get_line_delta_blocks(delta, source, target)
511
current_values = self._index._cache[version]
512
assert set(current_values[4]).difference(set(new_parents)) == set()
513
self._index.add_version(version,
514
current_values[1],
515
current_values[2],
516
current_values[3],
517
new_parents)
719
518
720
519
def get_delta(self, version_id):
721
520
"""Get a delta for constructing version from some other version."""
722
521
self.check_not_reserved_id(version_id)
522
if not self.has_version(version_id):
523
raise RevisionNotPresent(version_id, self.filename)
524
723
525
parents = self.get_parents(version_id)
724
526
if len(parents):
725
527
parent = parents[0]
726
528
else:
727
529
parent = None
728
index_memo = self._index.get_position(version_id)
729
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
530
data_pos, data_size = self._index.get_position(version_id)
531
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
532
version_idx = self._index.lookup(version_id)
730
533
noeol = 'no-eol' in self._index.get_options(version_id)
731
534
if 'fulltext' == self._index.get_method(version_id):
732
new_content = self.factory.parse_fulltext(data, version_id)
535
new_content = self.factory.parse_fulltext(data, version_idx)
733
536
if parent is not None:
734
537
reference_content = self._get_content(parent)
735
538
old_texts = reference_content.text()
736
539
else:
737
540
old_texts = []
738
541
new_texts = new_content.text()
739
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
740
new_texts)
542
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
741
543
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
742
544
else:
743
delta = self.factory.parse_line_delta(data, version_id)
545
delta = self.factory.parse_line_delta(data, version_idx)
744
546
return parent, sha1, noeol, delta
745
746
def get_format_signature(self):
747
"""See VersionedFile.get_format_signature()."""
748
if self.factory.annotated:
749
annotated_part = "annotated"
750
else:
751
annotated_part = "plain"
752
return "knit-%s" % (annotated_part,)
753
547
754
548
def get_graph_with_ghosts(self):
755
549
"""See VersionedFile.get_graph_with_ghosts()."""
757
551
return dict(graph_items)
758
552
759
553
def get_sha1(self, version_id):
760
return self.get_sha1s([version_id])[0]
761
762
def get_sha1s(self, version_ids):
763
554
"""See VersionedFile.get_sha1()."""
764
record_map = self._get_record_map(version_ids)
765
# record entry 2 is the 'digest'.
766
return [record_map[v][2] for v in version_ids]
555
record_map = self._get_record_map([version_id])
556
method, content, digest, next = record_map[version_id]
557
return digest
767
558
768
559
@staticmethod
769
560
def get_suffixes():
784
575
return True
785
576
return False
786
577
787
def insert_data_stream(self, (format, data_list, reader_callable)):
788
"""Insert knit records from a data stream into this knit.
789
790
If a version in the stream is already present in this knit, it will not
791
be inserted a second time. It will be checked for consistency with the
792
stored version however, and may cause a KnitCorrupt error to be raised
793
if the data in the stream disagrees with the already stored data.
794
795
:seealso: get_data_stream
796
"""
797
if format != self.get_format_signature():
798
if 'knit' in debug.debug_flags:
799
trace.mutter(
800
'incompatible format signature inserting to %r', self)
801
source = self._knit_from_datastream(
802
(format, data_list, reader_callable))
803
self.join(source)
804
return
805
806
for version_id, options, length, parents in data_list:
807
if self.has_version(version_id):
808
# First check: the list of parents.
809
my_parents = self.get_parents_with_ghosts(version_id)
810
if tuple(my_parents) != tuple(parents):
811
# XXX: KnitCorrupt is not quite the right exception here.
812
raise KnitCorrupt(
813
self.filename,
814
'parents list %r from data stream does not match '
815
'already recorded parents %r for %s'
816
% (parents, my_parents, version_id))
817
818
# Also check the SHA-1 of the fulltext this content will
819
# produce.
820
raw_data = reader_callable(length)
821
my_fulltext_sha1 = self.get_sha1(version_id)
822
df, rec = self._data._parse_record_header(version_id, raw_data)
823
stream_fulltext_sha1 = rec[3]
824
if my_fulltext_sha1 != stream_fulltext_sha1:
825
# Actually, we don't know if it's this knit that's corrupt,
826
# or the data stream we're trying to insert.
827
raise KnitCorrupt(
828
self.filename, 'sha-1 does not match %s' % version_id)
829
else:
830
if 'line-delta' in options:
831
# Make sure that this knit record is actually useful: a
832
# line-delta is no use unless we have its parent.
833
# Fetching from a broken repository with this problem
834
# shouldn't break the target repository.
835
#
836
# See https://bugs.launchpad.net/bzr/+bug/164443
837
if not self._index.has_version(parents[0]):
838
raise KnitCorrupt(
839
self.filename,
840
'line-delta from stream '
841
'for version %s '
842
'references '
843
'missing parent %s\n'
844
'Try running "bzr check" '
845
'on the source repository, and "bzr reconcile" '
846
'if necessary.' %
847
(version_id, parents[0]))
848
self._add_raw_records(
849
[(version_id, options, parents, length)],
850
reader_callable(length))
851
852
def _knit_from_datastream(self, (format, data_list, reader_callable)):
853
"""Create a knit object from a data stream.
854
855
This method exists to allow conversion of data streams that do not
856
match the signature of this knit. Generally it will be slower and use
857
more memory to use this method to insert data, but it will work.
858
859
:seealso: get_data_stream for details on datastreams.
860
:return: A knit versioned file which can be used to join the datastream
861
into self.
862
"""
863
if format == "knit-plain":
864
factory = KnitPlainFactory()
865
elif format == "knit-annotated":
866
factory = KnitAnnotateFactory()
867
else:
868
raise errors.KnitDataStreamUnknown(format)
869
index = _StreamIndex(data_list, self._index)
870
access = _StreamAccess(reader_callable, index, self, factory)
871
return KnitVersionedFile(self.filename, self.transport,
872
factory=factory, index=index, access_method=access)
873
874
578
def versions(self):
875
579
"""See VersionedFile.versions."""
876
if 'evil' in debug.debug_flags:
877
trace.mutter_callsite(2, "versions scales with size of history")
878
580
return self._index.get_versions()
879
581
880
582
def has_version(self, version_id):
881
583
"""See VersionedFile.has_version."""
882
if 'evil' in debug.debug_flags:
883
trace.mutter_callsite(2, "has_version is a LBYL scenario")
884
584
return self._index.has_version(version_id)
885
585
886
586
__contains__ = has_version
887
587
888
588
def _merge_annotations(self, content, parents, parent_texts={},
889
delta=None, annotated=None,
890
left_matching_blocks=None):
589
delta=None, annotated=None):
891
590
"""Merge annotations for content. This is done by comparing
892
591
the annotations based on changed to the text.
893
592
"""
894
if left_matching_blocks is not None:
895
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
896
else:
593
if annotated:
897
594
delta_seq = None
898
if annotated:
899
595
for parent_id in parents:
900
596
merge_content = self._get_content(parent_id, parent_texts)
901
if (parent_id == parents[0] and delta_seq is not None):
902
seq = delta_seq
903
else:
904
seq = patiencediff.PatienceSequenceMatcher(
905
None, merge_content.text(), content.text())
597
seq = patiencediff.PatienceSequenceMatcher(
598
None, merge_content.text(), content.text())
599
if delta_seq is None:
600
# setup a delta seq to reuse.
601
delta_seq = seq
906
602
for i, j, n in seq.get_matching_blocks():
907
603
if n == 0:
908
604
continue
909
# this appears to copy (origin, text) pairs across to the
910
# new content for any line that matches the last-checked
911
# parent.
605
# this appears to copy (origin, text) pairs across to the new
606
# content for any line that matches the last-checked parent.
607
# FIXME: save the sequence control data for delta compression
608
# against the most relevant parent rather than rediffing.
912
609
content._lines[j:j+n] = merge_content._lines[i:i+n]
913
610
if delta:
914
if delta_seq is None:
611
if not annotated:
915
612
reference_content = self._get_content(parents[0], parent_texts)
916
613
new_texts = content.text()
917
614
old_texts = reference_content.text()
933
630
934
631
This data is intended to be used for retrieving the knit records.
935
632
936
A dict of version_id to (record_details, index_memo, next, parents) is
633
A dict of version_id to (method, data_pos, data_size, next) is
937
634
returned.
938
635
method is the way referenced data should be applied.
939
index_memo is the handle to pass to the data access to actually get the
940
data
636
data_pos is the position of the data in the knit.
637
data_size is the size of the data in the knit.
941
638
next is the build-parent of the version, or None for fulltexts.
942
parents is the version_ids of the parents of this version
943
639
"""
944
640
component_data = {}
945
pending_components = version_ids
946
while pending_components:
947
build_details = self._index.get_build_details(pending_components)
948
current_components = set(pending_components)
949
pending_components = set()
950
for version_id, details in build_details.iteritems():
951
(index_memo, compression_parent, parents,
952
record_details) = details
953
method = record_details[0]
954
if compression_parent is not None:
955
pending_components.add(compression_parent)
956
component_data[version_id] = (record_details, index_memo,
957
compression_parent)
958
missing = current_components.difference(build_details)
959
if missing:
960
raise errors.RevisionNotPresent(missing.pop(), self.filename)
641
for version_id in version_ids:
642
cursor = version_id
643
644
while cursor is not None and cursor not in component_data:
645
method = self._index.get_method(cursor)
646
if method == 'fulltext':
647
next = None
648
else:
649
next = self.get_parents(cursor)[0]
650
data_pos, data_size = self._index.get_position(cursor)
651
component_data[cursor] = (method, data_pos, data_size, next)
652
cursor = next
961
653
return component_data
962
654
963
655
def _get_content(self, version_id, parent_texts={}):
964
656
"""Returns a content object that makes up the specified
965
657
version."""
658
if not self.has_version(version_id):
659
raise RevisionNotPresent(version_id, self.filename)
660
966
661
cached_version = parent_texts.get(version_id, None)
967
662
if cached_version is not None:
968
if not self.has_version(version_id):
969
raise RevisionNotPresent(version_id, self.filename)
970
663
return cached_version
971
664
972
665
text_map, contents_map = self._get_content_maps([version_id])
976
669
"""Check that all specified versions are present."""
977
670
self._index.check_versions_present(version_ids)
978
671
979
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
980
nostore_sha, random_id, check_content):
672
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
981
673
"""See VersionedFile.add_lines_with_ghosts()."""
982
self._check_add(version_id, lines, random_id, check_content)
983
return self._add(version_id, lines, parents, self.delta,
984
parent_texts, None, nostore_sha, random_id)
674
self._check_add(version_id, lines)
675
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
985
676
986
def _add_lines(self, version_id, parents, lines, parent_texts,
987
left_matching_blocks, nostore_sha, random_id, check_content):
677
def _add_lines(self, version_id, parents, lines, parent_texts):
988
678
"""See VersionedFile.add_lines."""
989
self._check_add(version_id, lines, random_id, check_content)
679
self._check_add(version_id, lines)
990
680
self._check_versions_present(parents)
991
return self._add(version_id, lines[:], parents, self.delta,
992
parent_texts, left_matching_blocks, nostore_sha, random_id)
681
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
993
682
994
def _check_add(self, version_id, lines, random_id, check_content):
683
def _check_add(self, version_id, lines):
995
684
"""check that version_id and lines are safe to add."""
685
assert self.writable, "knit is not opened for write"
686
### FIXME escape. RBC 20060228
996
687
if contains_whitespace(version_id):
997
688
raise InvalidRevisionId(version_id, self.filename)
998
689
self.check_not_reserved_id(version_id)
999
# Technically this could be avoided if we are happy to allow duplicate
1000
# id insertion when other things than bzr core insert texts, but it
1001
# seems useful for folk using the knit api directly to have some safety
1002
# blanket that we can disable.
1003
if not random_id and self.has_version(version_id):
690
if self.has_version(version_id):
1004
691
raise RevisionAlreadyPresent(version_id, self.filename)
1005
if check_content:
1006
self._check_lines_not_unicode(lines)
1007
self._check_lines_are_lines(lines)
692
self._check_lines_not_unicode(lines)
693
self._check_lines_are_lines(lines)
1008
694
1009
def _add(self, version_id, lines, parents, delta, parent_texts,
1010
left_matching_blocks, nostore_sha, random_id):
695
def _add(self, version_id, lines, parents, delta, parent_texts):
1011
696
"""Add a set of lines on top of version specified by parents.
1012
697
1013
698
If delta is true, compress the text as a line-delta against
1015
700
1016
701
Any versions not present will be converted into ghosts.
1017
702
"""
1018
# first thing, if the content is something we don't need to store, find
1019
# that out.
1020
line_bytes = ''.join(lines)
1021
digest = sha_string(line_bytes)
1022
if nostore_sha == digest:
1023
raise errors.ExistingContent
703
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
704
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
705
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
706
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
707
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
708
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
709
# +1383 0 8.0370 8.0370 +<len>
710
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
711
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
712
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
713
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
1024
714
1025
715
present_parents = []
716
ghosts = []
1026
717
if parent_texts is None:
1027
718
parent_texts = {}
1028
719
for parent in parents:
1029
if self.has_version(parent):
720
if not self.has_version(parent):
721
ghosts.append(parent)
722
else:
1030
723
present_parents.append(parent)
1031
724
1032
# can only compress against the left most present parent.
1033
if (delta and
1034
(len(present_parents) == 0 or
1035
present_parents[0] != parents[0])):
725
if delta and not len(present_parents):
1036
726
delta = False
1037
727
1038
text_length = len(line_bytes)
728
digest = sha_strings(lines)
1039
729
options = []
1040
730
if lines:
1041
731
if lines[-1][-1] != '\n':
1042
# copy the contents of lines.
1043
lines = lines[:]
1044
732
options.append('no-eol')
1045
733
lines[-1] = lines[-1] + '\n'
1046
line_bytes += '\n'
1047
734
1048
if delta:
735
if len(present_parents) and delta:
1049
736
# To speed the extract of texts the delta chain is limited
1050
737
# to a fixed number of deltas. This should minimize both
1051
738
# I/O and the time spend applying deltas.
1052
739
delta = self._check_should_delta(present_parents)
1053
740
1054
assert isinstance(version_id, str)
1055
content = self.factory.make(lines, version_id)
741
lines = self.factory.make(lines, version_id)
1056
742
if delta or (self.factory.annotated and len(present_parents) > 0):
1057
# Merge annotations from parent texts if needed.
1058
delta_hunks = self._merge_annotations(content, present_parents,
1059
parent_texts, delta, self.factory.annotated,
1060
left_matching_blocks)
743
# Merge annotations from parent texts if so is needed.
744
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
745
delta, self.factory.annotated)
1061
746
1062
747
if delta:
1063
748
options.append('line-delta')
1064
749
store_lines = self.factory.lower_line_delta(delta_hunks)
1065
size, bytes = self._data._record_to_data(version_id, digest,
1066
store_lines)
1067
750
else:
1068
751
options.append('fulltext')
1069
# isinstance is slower and we have no hierarchy.
1070
if self.factory.__class__ == KnitPlainFactory:
1071
# Use the already joined bytes saving iteration time in
1072
# _record_to_data.
1073
size, bytes = self._data._record_to_data(version_id, digest,
1074
lines, [line_bytes])
1075
else:
1076
# get mixed annotation + content and feed it into the
1077
# serialiser.
1078
store_lines = self.factory.lower_fulltext(content)
1079
size, bytes = self._data._record_to_data(version_id, digest,
1080
store_lines)
752
store_lines = self.factory.lower_fulltext(lines)
1081
753
1082
access_memo = self._data.add_raw_records([size], bytes)[0]
1083
self._index.add_versions(
1084
((version_id, options, access_memo, parents),),
1085
random_id=random_id)
1086
return digest, text_length, content
754
where, size = self._data.add_record(version_id, digest, store_lines)
755
self._index.add_version(version_id, options, where, size, parents)
756
return lines
1087
757
1088
758
def check(self, progress_bar=None):
1089
759
"""See VersionedFile.check()."""
1101
771
def _get_record_map(self, version_ids):
1102
772
"""Produce a dictionary of knit records.
1103
773
1104
:return: {version_id:(record, record_details, digest, next)}
1105
record
1106
data returned from read_records
1107
record_details
1108
opaque information to pass to parse_record
1109
digest
1110
SHA1 digest of the full text after all steps are done
1111
next
1112
build-parent of the version, i.e. the leftmost ancestor.
1113
Will be None if the record is not a delta.
774
The keys are version_ids, the values are tuples of (method, content,
775
digest, next).
776
method is the way the content should be applied.
777
content is a KnitContent object.
778
digest is the SHA1 digest of this version id after all steps are done
779
next is the build-parent of the version, i.e. the leftmost ancestor.
780
If the method is fulltext, next will be None.
1114
781
"""
1115
782
position_map = self._get_components_positions(version_ids)
1116
# c = component_id, r = record_details, i_m = index_memo, n = next
1117
records = [(c, i_m) for c, (r, i_m, n)
1118
in position_map.iteritems()]
783
# c = component_id, m = method, p = position, s = size, n = next
784
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
1119
785
record_map = {}
1120
for component_id, record, digest in \
786
for component_id, content, digest in \
1121
787
self._data.read_records_iter(records):
1122
(record_details, index_memo, next) = position_map[component_id]
1123
record_map[component_id] = record, record_details, digest, next
1124
788
method, position, size, next = position_map[component_id]
789
record_map[component_id] = method, content, digest, next
790
1125
791
return record_map
1126
792
1127
793
def get_text(self, version_id):
1138
804
text_map, content_map = self._get_content_maps(version_ids)
1139
805
return [text_map[v] for v in version_ids]
1140
806
1141
_get_lf_split_line_list = get_line_list
1142
1143
807
def _get_content_maps(self, version_ids):
1144
808
"""Produce maps of text and KnitContents
1145
809
1147
811
the requested versions and content_map contains the KnitContents.
1148
812
Both dicts take version_ids as their keys.
1149
813
"""
1150
# FUTURE: This function could be improved for the 'extract many' case
1151
# by tracking each component and only doing the copy when the number of
1152
# children than need to apply delta's to it is > 1 or it is part of the
1153
# final output.
1154
version_ids = list(version_ids)
1155
multiple_versions = len(version_ids) != 1
814
for version_id in version_ids:
815
if not self.has_version(version_id):
816
raise RevisionNotPresent(version_id, self.filename)
1156
817
record_map = self._get_record_map(version_ids)
1157
818
1158
819
text_map = {}
1162
823
components = []
1163
824
cursor = version_id
1164
825
while cursor is not None:
1165
record, record_details, digest, next = record_map[cursor]
1166
components.append((cursor, record, record_details, digest))
826
method, data, digest, next = record_map[cursor]
827
components.append((cursor, method, data, digest))
1167
828
if cursor in content_map:
1168
829
break
1169
830
cursor = next
1170
831
1171
832
content = None
1172
for (component_id, record, record_details,
1173
digest) in reversed(components):
833
for component_id, method, data, digest in reversed(components):
1174
834
if component_id in content_map:
1175
835
content = content_map[component_id]
1176
836
else:
1177
content, delta = self.factory.parse_record(version_id,
1178
record, record_details, content,
1179
copy_base_content=multiple_versions)
1180
if multiple_versions:
1181
content_map[component_id] = content
837
version_idx = self._index.lookup(component_id)
838
if method == 'fulltext':
839
assert content is None
840
content = self.factory.parse_fulltext(data, version_idx)
841
elif method == 'line-delta':
842
delta = self.factory.parse_line_delta(data, version_idx)
843
content = content.copy()
844
content._lines = self._apply_delta(content._lines,
845
delta)
846
content_map[component_id] = content
1182
847
1183
content.cleanup_eol(copy_on_mutate=multiple_versions)
848
if 'no-eol' in self._index.get_options(version_id):
849
content = content.copy()
850
line = content._lines[-1][1].rstrip('\n')
851
content._lines[-1] = (content._lines[-1][0], line)
1184
852
final_content[version_id] = content
1185
853
1186
854
# digest here is the digest from the last applied component.
1187
855
text = content.text()
1188
actual_sha = sha_strings(text)
1189
if actual_sha != digest:
1190
raise KnitCorrupt(self.filename,
1191
'\n sha-1 %s'
1192
'\n of reconstructed text does not match'
1193
'\n expected %s'
1194
'\n for version %s' %
1195
(actual_sha, digest, version_id))
1196
text_map[version_id] = text
1197
return text_map, final_content
856
if sha_strings(text) != digest:
857
raise KnitCorrupt(self.filename,
858
'sha-1 does not match %s' % version_id)
859
860
text_map[version_id] = text
861
return text_map, final_content
1198
862
1199
863
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1200
864
pb=None):
1215
879
# get a in-component-order queue:
1216
880
for version_id in self.versions():
1217
881
if version_id in requested_versions:
1218
index_memo = self._index.get_position(version_id)
1219
version_id_records.append((version_id, index_memo))
882
data_pos, length = self._index.get_position(version_id)
883
version_id_records.append((version_id, data_pos, length))
1220
884
1221
885
total = len(version_id_records)
1222
886
for version_idx, (version_id, data, sha_value) in \
1223
887
enumerate(self._data.read_records_iter(version_id_records)):
1224
888
pb.update('Walking content.', version_idx, total)
1225
889
method = self._index.get_method(version_id)
890
version_idx = self._index.lookup(version_id)
1226
891
1227
892
assert method in ('fulltext', 'line-delta')
1228
893
if method == 'fulltext':
1229
894
line_iterator = self.factory.get_fulltext_content(data)
1230
895
else:
1231
896
line_iterator = self.factory.get_linedelta_content(data)
1232
# XXX: It might be more efficient to yield (version_id,
1233
# line_iterator) in the future. However for now, this is a simpler
1234
# change to integrate into the rest of the codebase. RBC 20071110
1235
897
for line in line_iterator:
1236
yield line, version_id
898
yield line
1237
899
1238
900
pb.update('Walking content.', total, total)
1239
901
1240
def iter_parents(self, version_ids):
1241
"""Iterate through the parents for many version ids.
1242
1243
:param version_ids: An iterable yielding version_ids.
1244
:return: An iterator that yields (version_id, parents). Requested
1245
version_ids not present in the versioned file are simply skipped.
1246
The order is undefined, allowing for different optimisations in
1247
the underlying implementation.
1248
"""
1249
return self._index.iter_parents(version_ids)
1250
1251
902
def num_versions(self):
1252
903
"""See VersionedFile.num_versions()."""
1253
904
return self._index.num_versions()
1256
907
1257
908
def annotate_iter(self, version_id):
1258
909
"""See VersionedFile.annotate_iter."""
1259
return self.factory.annotate_iter(self, version_id)
910
content = self._get_content(version_id)
911
for origin, text in content.annotate_iter():
912
yield origin, text
1260
913
1261
914
def get_parents(self, version_id):
1262
915
"""See VersionedFile.get_parents."""
1275
928
except KeyError:
1276
929
raise RevisionNotPresent(version_id, self.filename)
1277
930
1278
def get_ancestry(self, versions, topo_sorted=True):
931
def get_ancestry(self, versions):
1279
932
"""See VersionedFile.get_ancestry."""
1280
933
if isinstance(versions, basestring):
1281
934
versions = [versions]
1282
935
if not versions:
1283
936
return []
1284
return self._index.get_ancestry(versions, topo_sorted)
937
return self._index.get_ancestry(versions)
1285
938
1286
939
def get_ancestry_with_ghosts(self, versions):
1287
940
"""See VersionedFile.get_ancestry_with_ghosts."""
1291
944
return []
1292
945
return self._index.get_ancestry_with_ghosts(versions)
1293
946
947
#@deprecated_method(zero_eight)
948
def walk(self, version_ids):
949
"""See VersionedFile.walk."""
950
# We take the short path here, and extract all relevant texts
951
# and put them in a weave and let that do all the work. Far
952
# from optimal, but is much simpler.
953
# FIXME RB 20060228 this really is inefficient!
954
from bzrlib.weave import Weave
955
956
w = Weave(self.filename)
957
ancestry = self.get_ancestry(version_ids)
958
sorted_graph = topo_sort(self._index.get_graph())
959
version_list = [vid for vid in sorted_graph if vid in ancestry]
960
961
for version_id in version_list:
962
lines = self.get_lines(version_id)
963
w.add_lines(version_id, self.get_parents(version_id), lines)
964
965
for lineno, insert_id, dset, line in w.walk(version_ids):
966
yield lineno, insert_id, dset, line
967
1294
968
def plan_merge(self, ver_a, ver_b):
1295
969
"""See VersionedFile.plan_merge."""
1296
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1297
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
970
ancestors_b = set(self.get_ancestry(ver_b))
971
def status_a(revision, text):
972
if revision in ancestors_b:
973
return 'killed-b', text
974
else:
975
return 'new-a', text
976
977
ancestors_a = set(self.get_ancestry(ver_a))
978
def status_b(revision, text):
979
if revision in ancestors_a:
980
return 'killed-a', text
981
else:
982
return 'new-b', text
983
1298
984
annotated_a = self.annotate(ver_a)
1299
985
annotated_b = self.annotate(ver_b)
1300
return merge._plan_annotate_merge(annotated_a, annotated_b,
1301
ancestors_a, ancestors_b)
986
plain_a = [t for (a, t) in annotated_a]
987
plain_b = [t for (a, t) in annotated_b]
988
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
989
a_cur = 0
990
b_cur = 0
991
for ai, bi, l in blocks:
992
# process all mismatched sections
993
# (last mismatched section is handled because blocks always
994
# includes a 0-length last block)
995
for revision, text in annotated_a[a_cur:ai]:
996
yield status_a(revision, text)
997
for revision, text in annotated_b[b_cur:bi]:
998
yield status_b(revision, text)
999
1000
# and now the matched section
1001
a_cur = ai + l
1002
b_cur = bi + l
1003
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
1004
assert text_a == text_b
1005
yield "unchanged", text_a
1302
1006
1303
1007
1304
1008
class _KnitComponentFile(object):
1328
1032
raise KnitHeaderError(badline=line,
1329
1033
filename=self._transport.abspath(self._filename))
1330
1034
1035
def commit(self):
1036
"""Commit is a nop."""
1037
1331
1038
def __repr__(self):
1332
1039
return '%s(%s)' % (self.__class__.__name__, self._filename)
1333
1040
1423
1130
# so - wc -l of a knit index is != the number of unique names
1424
1131
# in the knit.
1425
1132
self._history = []
1133
decode_utf8 = cache_utf8.decode
1134
pb = ui.ui_factory.nested_progress_bar()
1426
1135
try:
1427
fp = self._transport.get(self._filename)
1136
pb.update('read knit index', 0, 1)
1428
1137
try:
1429
# _load_data may raise NoSuchFile if the target knit is
1430
# completely empty.
1431
_load_data(self, fp)
1432
finally:
1433
fp.close()
1434
except NoSuchFile:
1435
if mode != 'w' or not create:
1436
raise
1437
elif delay_create:
1438
self._need_to_create = True
1138
fp = self._transport.get(self._filename)
1139
try:
1140
# _load_data may raise NoSuchFile if the target knit is
1141
# completely empty.
1142
self._load_data(fp)
1143
finally:
1144
fp.close()
1145
except NoSuchFile:
1146
if mode != 'w' or not create:
1147
raise
1148
elif delay_create:
1149
self._need_to_create = True
1150
else:
1151
self._transport.put_bytes_non_atomic(
1152
self._filename, self.HEADER, mode=self._file_mode)
1153
finally:
1154
pb.update('read knit index', 1, 1)
1155
pb.finished()
1156
1157
def _load_data(self, fp):
1158
cache = self._cache
1159
history = self._history
1160
decode_utf8 = cache_utf8.decode
1161
1162
self.check_header(fp)
1163
# readlines reads the whole file at once:
1164
# bad for transports like http, good for local disk
1165
# we save 60 ms doing this one change (
1166
# from calling readline each time to calling
1167
# readlines once.
1168
# probably what we want for nice behaviour on
1169
# http is a incremental readlines that yields, or
1170
# a check for local vs non local indexes,
1171
history_top = len(history) - 1
1172
for line in fp.readlines():
1173
rec = line.split()
1174
if len(rec) < 5 or rec[-1] != ':':
1175
# corrupt line.
1176
# FIXME: in the future we should determine if its a
1177
# short write - and ignore it
1178
# or a different failure, and raise. RBC 20060407
1179
continue
1180
1181
parents = []
1182
for value in rec[4:-1]:
1183
if value[0] == '.':
1184
# uncompressed reference
1185
parents.append(decode_utf8(value[1:]))
1186
else:
1187
parents.append(history[int(value)])
1188
1189
version_id, options, pos, size = rec[:4]
1190
version_id = decode_utf8(version_id)
1191
1192
# See self._cache_version
1193
# only want the _history index to reference the 1st
1194
# index entry for version_id
1195
if version_id not in cache:
1196
history_top += 1
1197
index = history_top
1198
history.append(version_id)
1439
1199
else:
1440
self._transport.put_bytes_non_atomic(
1441
self._filename, self.HEADER, mode=self._file_mode)
1200
index = cache[version_id][5]
1201
cache[version_id] = (version_id,
1202
options.split(','),
1203
int(pos),
1204
int(size),
1205
parents,
1206
index)
1207
# end self._cache_version
1442
1208
1443
1209
def get_graph(self):
1444
"""Return a list of the node:parents lists from this knit index."""
1445
1210
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1446
1211
1447
def get_ancestry(self, versions, topo_sorted=True):
1212
def get_ancestry(self, versions):
1448
1213
"""See VersionedFile.get_ancestry."""
1449
1214
# get a graph of all the mentioned versions:
1450
1215
graph = {}
1460
1225
# if not completed and not a ghost
1461
1226
pending.update([p for p in parents if p not in graph])
1462
1227
graph[version] = parents
1463
if not topo_sorted:
1464
return graph.keys()
1465
1228
return topo_sort(graph.items())
1466
1229
1467
1230
def get_ancestry_with_ghosts(self, versions):
1484
1247
graph[version] = parents
1485
1248
return topo_sort(graph.items())
1486
1249
1487
def get_build_details(self, version_ids):
1488
"""Get the method, index_memo and compression parent for version_ids.
1489
1490
Ghosts are omitted from the result.
1491
1492
:param version_ids: An iterable of version_ids.
1493
:return: A dict of version_id:(index_memo, compression_parent,
1494
parents, record_details).
1495
index_memo
1496
opaque structure to pass to read_records to extract the raw
1497
data
1498
compression_parent
1499
Content that this record is built upon, may be None
1500
parents
1501
Logical parents of this node
1502
record_details
1503
extra information about the content which needs to be passed to
1504
Factory.parse_record
1505
"""
1506
result = {}
1507
for version_id in version_ids:
1508
if version_id not in self._cache:
1509
# ghosts are omitted
1510
continue
1511
method = self.get_method(version_id)
1512
parents = self.get_parents_with_ghosts(version_id)
1513
if method == 'fulltext':
1514
compression_parent = None
1515
else:
1516
compression_parent = parents[0]
1517
noeol = 'no-eol' in self.get_options(version_id)
1518
index_memo = self.get_position(version_id)
1519
result[version_id] = (index_memo, compression_parent,
1520
parents, (method, noeol))
1521
return result
1522
1523
def iter_parents(self, version_ids):
1524
"""Iterate through the parents for many version ids.
1525
1526
:param version_ids: An iterable yielding version_ids.
1527
:return: An iterator that yields (version_id, parents). Requested
1528
version_ids not present in the versioned file are simply skipped.
1529
The order is undefined, allowing for different optimisations in
1530
the underlying implementation.
1531
"""
1532
for version_id in version_ids:
1533
try:
1534
yield version_id, tuple(self.get_parents(version_id))
1535
except KeyError:
1536
pass
1537
1538
1250
def num_versions(self):
1539
1251
return len(self._history)
1540
1252
1541
1253
__len__ = num_versions
1542
1254
1543
1255
def get_versions(self):
1544
"""Get all the versions in the file. not topologically sorted."""
1545
1256
return self._history
1546
1257
1258
def idx_to_name(self, idx):
1259
return self._history[idx]
1260
1261
def lookup(self, version_id):
1262
assert version_id in self._cache
1263
return self._cache[version_id][5]
1264
1547
1265
def _version_list_to_index(self, versions):
1266
encode_utf8 = cache_utf8.encode
1548
1267
result_list = []
1549
1268
cache = self._cache
1550
1269
for version in versions:
1553
1272
result_list.append(str(cache[version][5]))
1554
1273
# -- end lookup () --
1555
1274
else:
1556
result_list.append('.' + version)
1275
result_list.append('.' + encode_utf8(version))
1557
1276
return ' '.join(result_list)
1558
1277
1559
def add_version(self, version_id, options, index_memo, parents):
1278
def add_version(self, version_id, options, pos, size, parents):
1560
1279
"""Add a version record to the index."""
1561
self.add_versions(((version_id, options, index_memo, parents),))
1280
self.add_versions(((version_id, options, pos, size, parents),))
1562
1281
1563
def add_versions(self, versions, random_id=False):
1282
def add_versions(self, versions):
1564
1283
"""Add multiple versions to the index.
1565
1284
1566
1285
:param versions: a list of tuples:
1567
1286
(version_id, options, pos, size, parents).
1568
:param random_id: If True the ids being added were randomly generated
1569
and no check for existence will be performed.
1570
1287
"""
1571
1288
lines = []
1289
encode_utf8 = cache_utf8.encode
1572
1290
orig_history = self._history[:]
1573
1291
orig_cache = self._cache.copy()
1574
1292
1575
1293
try:
1576
for version_id, options, (index, pos, size), parents in versions:
1577
line = "\n%s %s %s %s %s :" % (version_id,
1294
for version_id, options, pos, size, parents in versions:
1295
line = "\n%s %s %s %s %s :" % (encode_utf8(version_id),
1578
1296
','.join(options),
1579
1297
pos,
1580
1298
size,
1606
1324
return version_id in self._cache
1607
1325
1608
1326
def get_position(self, version_id):
1609
"""Return details needed to access the version.
1610
1611
.kndx indices do not support split-out data, so return None for the
1612
index field.
1613
1614
:return: a tuple (None, data position, size) to hand to the access
1615
logic to get the record.
1616
"""
1327
"""Return data position and size of specified version."""
1617
1328
entry = self._cache[version_id]
1618
return None, entry[2], entry[3]
1329
return entry[2], entry[3]
1619
1330
1620
1331
def get_method(self, version_id):
1621
1332
"""Return compression method of specified version."""
1622
try:
1623
options = self._cache[version_id][1]
1624
except KeyError:
1625
raise RevisionNotPresent(version_id, self._filename)
1333
options = self._cache[version_id][1]
1626
1334
if 'fulltext' in options:
1627
1335
return 'fulltext'
1628
1336
else:
1631
1339
return 'line-delta'
1632
1340
1633
1341
def get_options(self, version_id):
1634
"""Return a list representing options.
1635
1636
e.g. ['foo', 'bar']
1637
"""
1638
1342
return self._cache[version_id][1]
1639
1343
1640
1344
def get_parents(self, version_id):
1654
1358
raise RevisionNotPresent(version_id, self._filename)
1655
1359
1656
1360
1657
class KnitGraphIndex(object):
1658
"""A knit index that builds on GraphIndex."""
1659
1660
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1661
"""Construct a KnitGraphIndex on a graph_index.
1662
1663
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1664
:param deltas: Allow delta-compressed records.
1665
:param add_callback: If not None, allow additions to the index and call
1666
this callback with a list of added GraphIndex nodes:
1667
[(node, value, node_refs), ...]
1668
:param parents: If True, record knits parents, if not do not record
1669
parents.
1670
"""
1671
self._graph_index = graph_index
1672
self._deltas = deltas
1673
self._add_callback = add_callback
1674
self._parents = parents
1675
if deltas and not parents:
1676
raise KnitCorrupt(self, "Cannot do delta compression without "
1677
"parent tracking.")
1678
1679
def _get_entries(self, keys, check_present=False):
1680
"""Get the entries for keys.
1681
1682
:param keys: An iterable of index keys, - 1-tuples.
1683
"""
1684
keys = set(keys)
1685
found_keys = set()
1686
if self._parents:
1687
for node in self._graph_index.iter_entries(keys):
1688
yield node
1689
found_keys.add(node[1])
1690
else:
1691
# adapt parentless index to the rest of the code.
1692
for node in self._graph_index.iter_entries(keys):
1693
yield node[0], node[1], node[2], ()
1694
found_keys.add(node[1])
1695
if check_present:
1696
missing_keys = keys.difference(found_keys)
1697
if missing_keys:
1698
raise RevisionNotPresent(missing_keys.pop(), self)
1699
1700
def _present_keys(self, version_ids):
1701
return set([
1702
node[1] for node in self._get_entries(version_ids)])
1703
1704
def _parentless_ancestry(self, versions):
1705
"""Honour the get_ancestry API for parentless knit indices."""
1706
wanted_keys = self._version_ids_to_keys(versions)
1707
present_keys = self._present_keys(wanted_keys)
1708
missing = set(wanted_keys).difference(present_keys)
1709
if missing:
1710
raise RevisionNotPresent(missing.pop(), self)
1711
return list(self._keys_to_version_ids(present_keys))
1712
1713
def get_ancestry(self, versions, topo_sorted=True):
1714
"""See VersionedFile.get_ancestry."""
1715
if not self._parents:
1716
return self._parentless_ancestry(versions)
1717
# XXX: This will do len(history) index calls - perhaps
1718
# it should be altered to be a index core feature?
1719
# get a graph of all the mentioned versions:
1720
graph = {}
1721
ghosts = set()
1722
versions = self._version_ids_to_keys(versions)
1723
pending = set(versions)
1724
while pending:
1725
# get all pending nodes
1726
this_iteration = pending
1727
new_nodes = self._get_entries(this_iteration)
1728
found = set()
1729
pending = set()
1730
for (index, key, value, node_refs) in new_nodes:
1731
# dont ask for ghosties - otherwise
1732
# we we can end up looping with pending
1733
# being entirely ghosted.
1734
graph[key] = [parent for parent in node_refs[0]
1735
if parent not in ghosts]
1736
# queue parents
1737
for parent in graph[key]:
1738
# dont examine known nodes again
1739
if parent in graph:
1740
continue
1741
pending.add(parent)
1742
found.add(key)
1743
ghosts.update(this_iteration.difference(found))
1744
if versions.difference(graph):
1745
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1746
if topo_sorted:
1747
result_keys = topo_sort(graph.items())
1748
else:
1749
result_keys = graph.iterkeys()
1750
return [key[0] for key in result_keys]
1751
1752
def get_ancestry_with_ghosts(self, versions):
1753
"""See VersionedFile.get_ancestry."""
1754
if not self._parents:
1755
return self._parentless_ancestry(versions)
1756
# XXX: This will do len(history) index calls - perhaps
1757
# it should be altered to be a index core feature?
1758
# get a graph of all the mentioned versions:
1759
graph = {}
1760
versions = self._version_ids_to_keys(versions)
1761
pending = set(versions)
1762
while pending:
1763
# get all pending nodes
1764
this_iteration = pending
1765
new_nodes = self._get_entries(this_iteration)
1766
pending = set()
1767
for (index, key, value, node_refs) in new_nodes:
1768
graph[key] = node_refs[0]
1769
# queue parents
1770
for parent in graph[key]:
1771
# dont examine known nodes again
1772
if parent in graph:
1773
continue
1774
pending.add(parent)
1775
missing_versions = this_iteration.difference(graph)
1776
missing_needed = versions.intersection(missing_versions)
1777
if missing_needed:
1778
raise RevisionNotPresent(missing_needed.pop(), self)
1779
for missing_version in missing_versions:
1780
# add a key, no parents
1781
graph[missing_version] = []
1782
pending.discard(missing_version) # don't look for it
1783
result_keys = topo_sort(graph.items())
1784
return [key[0] for key in result_keys]
1785
1786
def get_build_details(self, version_ids):
1787
"""Get the method, index_memo and compression parent for version_ids.
1788
1789
Ghosts are omitted from the result.
1790
1791
:param version_ids: An iterable of version_ids.
1792
:return: A dict of version_id:(index_memo, compression_parent,
1793
parents, record_details).
1794
index_memo
1795
opaque structure to pass to read_records to extract the raw
1796
data
1797
compression_parent
1798
Content that this record is built upon, may be None
1799
parents
1800
Logical parents of this node
1801
record_details
1802
extra information about the content which needs to be passed to
1803
Factory.parse_record
1804
"""
1805
result = {}
1806
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
1807
for entry in entries:
1808
version_id = self._keys_to_version_ids((entry[1],))[0]
1809
if not self._parents:
1810
parents = ()
1811
else:
1812
parents = self._keys_to_version_ids(entry[3][0])
1813
if not self._deltas:
1814
compression_parent = None
1815
else:
1816
compression_parent_key = self._compression_parent(entry)
1817
if compression_parent_key:
1818
compression_parent = self._keys_to_version_ids(
1819
(compression_parent_key,))[0]
1820
else:
1821
compression_parent = None
1822
noeol = (entry[2][0] == 'N')
1823
if compression_parent:
1824
method = 'line-delta'
1825
else:
1826
method = 'fulltext'
1827
result[version_id] = (self._node_to_position(entry),
1828
compression_parent, parents,
1829
(method, noeol))
1830
return result
1831
1832
def _compression_parent(self, an_entry):
1833
# return the key that an_entry is compressed against, or None
1834
# Grab the second parent list (as deltas implies parents currently)
1835
compression_parents = an_entry[3][1]
1836
if not compression_parents:
1837
return None
1838
assert len(compression_parents) == 1
1839
return compression_parents[0]
1840
1841
def _get_method(self, node):
1842
if not self._deltas:
1843
return 'fulltext'
1844
if self._compression_parent(node):
1845
return 'line-delta'
1846
else:
1847
return 'fulltext'
1848
1849
def get_graph(self):
1850
"""Return a list of the node:parents lists from this knit index."""
1851
if not self._parents:
1852
return [(key, ()) for key in self.get_versions()]
1853
result = []
1854
for index, key, value, refs in self._graph_index.iter_all_entries():
1855
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1856
return result
1857
1858
def iter_parents(self, version_ids):
1859
"""Iterate through the parents for many version ids.
1860
1861
:param version_ids: An iterable yielding version_ids.
1862
:return: An iterator that yields (version_id, parents). Requested
1863
version_ids not present in the versioned file are simply skipped.
1864
The order is undefined, allowing for different optimisations in
1865
the underlying implementation.
1866
"""
1867
if self._parents:
1868
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1869
all_parents = set()
1870
present_parents = set()
1871
for node in all_nodes:
1872
all_parents.update(node[3][0])
1873
# any node we are querying must be present
1874
present_parents.add(node[1])
1875
unknown_parents = all_parents.difference(present_parents)
1876
present_parents.update(self._present_keys(unknown_parents))
1877
for node in all_nodes:
1878
parents = []
1879
for parent in node[3][0]:
1880
if parent in present_parents:
1881
parents.append(parent[0])
1882
yield node[1][0], tuple(parents)
1883
else:
1884
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1885
yield node[1][0], ()
1886
1887
def num_versions(self):
1888
return len(list(self._graph_index.iter_all_entries()))
1889
1890
__len__ = num_versions
1891
1892
def get_versions(self):
1893
"""Get all the versions in the file. not topologically sorted."""
1894
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1895
1896
def has_version(self, version_id):
1897
"""True if the version is in the index."""
1898
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1899
1900
def _keys_to_version_ids(self, keys):
1901
return tuple(key[0] for key in keys)
1902
1903
def get_position(self, version_id):
1904
"""Return details needed to access the version.
1905
1906
:return: a tuple (index, data position, size) to hand to the access
1907
logic to get the record.
1908
"""
1909
node = self._get_node(version_id)
1910
return self._node_to_position(node)
1911
1912
def _node_to_position(self, node):
1913
"""Convert an index value to position details."""
1914
bits = node[2][1:].split(' ')
1915
return node[0], int(bits[0]), int(bits[1])
1916
1917
def get_method(self, version_id):
1918
"""Return compression method of specified version."""
1919
return self._get_method(self._get_node(version_id))
1920
1921
def _get_node(self, version_id):
1922
try:
1923
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1924
except IndexError:
1925
raise RevisionNotPresent(version_id, self)
1926
1927
def get_options(self, version_id):
1928
"""Return a list representing options.
1929
1930
e.g. ['foo', 'bar']
1931
"""
1932
node = self._get_node(version_id)
1933
options = [self._get_method(node)]
1934
if node[2][0] == 'N':
1935
options.append('no-eol')
1936
return options
1937
1938
def get_parents(self, version_id):
1939
"""Return parents of specified version ignoring ghosts."""
1940
parents = list(self.iter_parents([version_id]))
1941
if not parents:
1942
# missing key
1943
raise errors.RevisionNotPresent(version_id, self)
1944
return parents[0][1]
1945
1946
def get_parents_with_ghosts(self, version_id):
1947
"""Return parents of specified version with ghosts."""
1948
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1949
check_present=True))
1950
if not self._parents:
1951
return ()
1952
return self._keys_to_version_ids(nodes[0][3][0])
1953
1954
def check_versions_present(self, version_ids):
1955
"""Check that all specified versions are present."""
1956
keys = self._version_ids_to_keys(version_ids)
1957
present = self._present_keys(keys)
1958
missing = keys.difference(present)
1959
if missing:
1960
raise RevisionNotPresent(missing.pop(), self)
1961
1962
def add_version(self, version_id, options, access_memo, parents):
1963
"""Add a version record to the index."""
1964
return self.add_versions(((version_id, options, access_memo, parents),))
1965
1966
def add_versions(self, versions, random_id=False):
1967
"""Add multiple versions to the index.
1968
1969
This function does not insert data into the Immutable GraphIndex
1970
backing the KnitGraphIndex, instead it prepares data for insertion by
1971
the caller and checks that it is safe to insert then calls
1972
self._add_callback with the prepared GraphIndex nodes.
1973
1974
:param versions: a list of tuples:
1975
(version_id, options, pos, size, parents).
1976
:param random_id: If True the ids being added were randomly generated
1977
and no check for existence will be performed.
1978
"""
1979
if not self._add_callback:
1980
raise errors.ReadOnlyError(self)
1981
# we hope there are no repositories with inconsistent parentage
1982
# anymore.
1983
# check for dups
1984
1985
keys = {}
1986
for (version_id, options, access_memo, parents) in versions:
1987
index, pos, size = access_memo
1988
key = (version_id, )
1989
parents = tuple((parent, ) for parent in parents)
1990
if 'no-eol' in options:
1991
value = 'N'
1992
else:
1993
value = ' '
1994
value += "%d %d" % (pos, size)
1995
if not self._deltas:
1996
if 'line-delta' in options:
1997
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1998
if self._parents:
1999
if self._deltas:
2000
if 'line-delta' in options:
2001
node_refs = (parents, (parents[0],))
2002
else:
2003
node_refs = (parents, ())
2004
else:
2005
node_refs = (parents, )
2006
else:
2007
if parents:
2008
raise KnitCorrupt(self, "attempt to add node with parents "
2009
"in parentless index.")
2010
node_refs = ()
2011
keys[key] = (value, node_refs)
2012
if not random_id:
2013
present_nodes = self._get_entries(keys)
2014
for (index, key, value, node_refs) in present_nodes:
2015
if (value, node_refs) != keys[key]:
2016
raise KnitCorrupt(self, "inconsistent details in add_versions"
2017
": %s %s" % ((value, node_refs), keys[key]))
2018
del keys[key]
2019
result = []
2020
if self._parents:
2021
for key, (value, node_refs) in keys.iteritems():
2022
result.append((key, value, node_refs))
2023
else:
2024
for key, (value, node_refs) in keys.iteritems():
2025
result.append((key, value))
2026
self._add_callback(result)
2027
2028
def _version_ids_to_keys(self, version_ids):
2029
return set((version_id, ) for version_id in version_ids)
2030
2031
2032
class _KnitAccess(object):
2033
"""Access to knit records in a .knit file."""
2034
2035
def __init__(self, transport, filename, _file_mode, _dir_mode,
2036
_need_to_create, _create_parent_dir):
2037
"""Create a _KnitAccess for accessing and inserting data.
2038
2039
:param transport: The transport the .knit is located on.
2040
:param filename: The filename of the .knit.
2041
"""
2042
self._transport = transport
2043
self._filename = filename
2044
self._file_mode = _file_mode
2045
self._dir_mode = _dir_mode
2046
self._need_to_create = _need_to_create
2047
self._create_parent_dir = _create_parent_dir
2048
2049
def add_raw_records(self, sizes, raw_data):
2050
"""Add raw knit bytes to a storage area.
2051
2052
The data is spooled to whereever the access method is storing data.
2053
2054
:param sizes: An iterable containing the size of each raw data segment.
2055
:param raw_data: A bytestring containing the data.
2056
:return: A list of memos to retrieve the record later. Each memo is a
2057
tuple - (index, pos, length), where the index field is always None
2058
for the .knit access method.
2059
"""
2060
assert type(raw_data) == str, \
2061
'data must be plain bytes was %s' % type(raw_data)
2062
if not self._need_to_create:
2063
base = self._transport.append_bytes(self._filename, raw_data)
2064
else:
2065
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2066
create_parent_dir=self._create_parent_dir,
2067
mode=self._file_mode,
2068
dir_mode=self._dir_mode)
2069
self._need_to_create = False
2070
base = 0
2071
result = []
2072
for size in sizes:
2073
result.append((None, base, size))
2074
base += size
2075
return result
2076
2077
def create(self):
2078
"""IFF this data access has its own storage area, initialise it.
2079
2080
:return: None.
2081
"""
2082
self._transport.put_bytes_non_atomic(self._filename, '',
2083
mode=self._file_mode)
2084
2085
def open_file(self):
2086
"""IFF this data access can be represented as a single file, open it.
2087
2088
For knits that are not mapped to a single file on disk this will
2089
always return None.
2090
2091
:return: None or a file handle.
2092
"""
2093
try:
2094
return self._transport.get(self._filename)
2095
except NoSuchFile:
2096
pass
2097
return None
2098
2099
def get_raw_records(self, memos_for_retrieval):
2100
"""Get the raw bytes for a records.
2101
2102
:param memos_for_retrieval: An iterable containing the (index, pos,
2103
length) memo for retrieving the bytes. The .knit method ignores
2104
the index as there is always only a single file.
2105
:return: An iterator over the bytes of the records.
2106
"""
2107
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2108
for pos, data in self._transport.readv(self._filename, read_vector):
2109
yield data
2110
2111
2112
class _PackAccess(object):
2113
"""Access to knit records via a collection of packs."""
2114
2115
def __init__(self, index_to_packs, writer=None):
2116
"""Create a _PackAccess object.
2117
2118
:param index_to_packs: A dict mapping index objects to the transport
2119
and file names for obtaining data.
2120
:param writer: A tuple (pack.ContainerWriter, write_index) which
2121
contains the pack to write, and the index that reads from it will
2122
be associated with.
2123
"""
2124
if writer:
2125
self.container_writer = writer[0]
2126
self.write_index = writer[1]
2127
else:
2128
self.container_writer = None
2129
self.write_index = None
2130
self.indices = index_to_packs
2131
2132
def add_raw_records(self, sizes, raw_data):
2133
"""Add raw knit bytes to a storage area.
2134
2135
The data is spooled to the container writer in one bytes-record per
2136
raw data item.
2137
2138
:param sizes: An iterable containing the size of each raw data segment.
2139
:param raw_data: A bytestring containing the data.
2140
:return: A list of memos to retrieve the record later. Each memo is a
2141
tuple - (index, pos, length), where the index field is the
2142
write_index object supplied to the PackAccess object.
2143
"""
2144
assert type(raw_data) == str, \
2145
'data must be plain bytes was %s' % type(raw_data)
2146
result = []
2147
offset = 0
2148
for size in sizes:
2149
p_offset, p_length = self.container_writer.add_bytes_record(
2150
raw_data[offset:offset+size], [])
2151
offset += size
2152
result.append((self.write_index, p_offset, p_length))
2153
return result
2154
2155
def create(self):
2156
"""Pack based knits do not get individually created."""
2157
2158
def get_raw_records(self, memos_for_retrieval):
2159
"""Get the raw bytes for a records.
2160
2161
:param memos_for_retrieval: An iterable containing the (index, pos,
2162
length) memo for retrieving the bytes. The Pack access method
2163
looks up the pack to use for a given record in its index_to_pack
2164
map.
2165
:return: An iterator over the bytes of the records.
2166
"""
2167
# first pass, group into same-index requests
2168
request_lists = []
2169
current_index = None
2170
for (index, offset, length) in memos_for_retrieval:
2171
if current_index == index:
2172
current_list.append((offset, length))
2173
else:
2174
if current_index is not None:
2175
request_lists.append((current_index, current_list))
2176
current_index = index
2177
current_list = [(offset, length)]
2178
# handle the last entry
2179
if current_index is not None:
2180
request_lists.append((current_index, current_list))
2181
for index, offsets in request_lists:
2182
transport, path = self.indices[index]
2183
reader = pack.make_readv_reader(transport, path, offsets)
2184
for names, read_func in reader.iter_records():
2185
yield read_func(None)
2186
2187
def open_file(self):
2188
"""Pack based knits have no single file."""
2189
return None
2190
2191
def set_writer(self, writer, index, (transport, packname)):
2192
"""Set a writer to use for adding data."""
2193
if index is not None:
2194
self.indices[index] = (transport, packname)
2195
self.container_writer = writer
2196
self.write_index = index
2197
2198
2199
class _StreamAccess(object):
2200
"""A Knit Access object that provides data from a datastream.
2201
2202
It also provides a fallback to present as unannotated data, annotated data
2203
from a *backing* access object.
2204
2205
This is triggered by a index_memo which is pointing to a different index
2206
than this was constructed with, and is used to allow extracting full
2207
unannotated texts for insertion into annotated knits.
2208
"""
2209
2210
def __init__(self, reader_callable, stream_index, backing_knit,
2211
orig_factory):
2212
"""Create a _StreamAccess object.
2213
2214
:param reader_callable: The reader_callable from the datastream.
2215
This is called to buffer all the data immediately, for
2216
random access.
2217
:param stream_index: The index the data stream this provides access to
2218
which will be present in native index_memo's.
2219
:param backing_knit: The knit object that will provide access to
2220
annotated texts which are not available in the stream, so as to
2221
create unannotated texts.
2222
:param orig_factory: The original content factory used to generate the
2223
stream. This is used for checking whether the thunk code for
2224
supporting _copy_texts will generate the correct form of data.
2225
"""
2226
self.data = reader_callable(None)
2227
self.stream_index = stream_index
2228
self.backing_knit = backing_knit
2229
self.orig_factory = orig_factory
2230
2231
def get_raw_records(self, memos_for_retrieval):
2232
"""Get the raw bytes for a records.
2233
2234
:param memos_for_retrieval: An iterable containing the (thunk_flag,
2235
index, start, end) memo for retrieving the bytes.
2236
:return: An iterator over the bytes of the records.
2237
"""
2238
# use a generator for memory friendliness
2239
for thunk_flag, version_id, start, end in memos_for_retrieval:
2240
if version_id is self.stream_index:
2241
yield self.data[start:end]
2242
continue
2243
# we have been asked to thunk. This thunking only occurs when
2244
# we are obtaining plain texts from an annotated backing knit
2245
# so that _copy_texts will work.
2246
# We could improve performance here by scanning for where we need
2247
# to do this and using get_line_list, then interleaving the output
2248
# as desired. However, for now, this is sufficient.
2249
if self.orig_factory.__class__ != KnitPlainFactory:
2250
raise errors.KnitCorrupt(
2251
self, 'Bad thunk request %r' % version_id)
2252
lines = self.backing_knit.get_lines(version_id)
2253
line_bytes = ''.join(lines)
2254
digest = sha_string(line_bytes)
2255
if lines:
2256
if lines[-1][-1] != '\n':
2257
lines[-1] = lines[-1] + '\n'
2258
line_bytes += '\n'
2259
orig_options = list(self.backing_knit._index.get_options(version_id))
2260
if 'fulltext' not in orig_options:
2261
if 'line-delta' not in orig_options:
2262
raise errors.KnitCorrupt(self,
2263
'Unknown compression method %r' % orig_options)
2264
orig_options.remove('line-delta')
2265
orig_options.append('fulltext')
2266
# We want plain data, because we expect to thunk only to allow text
2267
# extraction.
2268
size, bytes = self.backing_knit._data._record_to_data(version_id,
2269
digest, lines, line_bytes)
2270
yield bytes
2271
2272
2273
class _StreamIndex(object):
2274
"""A Knit Index object that uses the data map from a datastream."""
2275
2276
def __init__(self, data_list, backing_index):
2277
"""Create a _StreamIndex object.
2278
2279
:param data_list: The data_list from the datastream.
2280
:param backing_index: The index which will supply values for nodes
2281
referenced outside of this stream.
2282
"""
2283
self.data_list = data_list
2284
self.backing_index = backing_index
2285
self._by_version = {}
2286
pos = 0
2287
for key, options, length, parents in data_list:
2288
self._by_version[key] = options, (pos, pos + length), parents
2289
pos += length
2290
2291
def get_ancestry(self, versions, topo_sorted):
2292
"""Get an ancestry list for versions."""
2293
if topo_sorted:
2294
# Not needed for basic joins
2295
raise NotImplementedError(self.get_ancestry)
2296
# get a graph of all the mentioned versions:
2297
# Little ugly - basically copied from KnitIndex, but don't want to
2298
# accidentally incorporate too much of that index's code.
2299
ancestry = set()
2300
pending = set(versions)
2301
cache = self._by_version
2302
while pending:
2303
version = pending.pop()
2304
# trim ghosts
2305
try:
2306
parents = [p for p in cache[version][2] if p in cache]
2307
except KeyError:
2308
raise RevisionNotPresent(version, self)
2309
# if not completed and not a ghost
2310
pending.update([p for p in parents if p not in ancestry])
2311
ancestry.add(version)
2312
return list(ancestry)
2313
2314
def get_build_details(self, version_ids):
2315
"""Get the method, index_memo and compression parent for version_ids.
2316
2317
Ghosts are omitted from the result.
2318
2319
:param version_ids: An iterable of version_ids.
2320
:return: A dict of version_id:(index_memo, compression_parent,
2321
parents, record_details).
2322
index_memo
2323
opaque structure to pass to read_records to extract the raw
2324
data
2325
compression_parent
2326
Content that this record is built upon, may be None
2327
parents
2328
Logical parents of this node
2329
record_details
2330
extra information about the content which needs to be passed to
2331
Factory.parse_record
2332
"""
2333
result = {}
2334
for version_id in version_ids:
2335
try:
2336
method = self.get_method(version_id)
2337
except errors.RevisionNotPresent:
2338
# ghosts are omitted
2339
continue
2340
parent_ids = self.get_parents_with_ghosts(version_id)
2341
noeol = ('no-eol' in self.get_options(version_id))
2342
if method == 'fulltext':
2343
compression_parent = None
2344
else:
2345
compression_parent = parent_ids[0]
2346
index_memo = self.get_position(version_id)
2347
result[version_id] = (index_memo, compression_parent,
2348
parent_ids, (method, noeol))
2349
return result
2350
2351
def get_method(self, version_id):
2352
"""Return compression method of specified version."""
2353
try:
2354
options = self._by_version[version_id][0]
2355
except KeyError:
2356
# Strictly speaking this should check in the backing knit, but
2357
# until we have a test to discriminate, this will do.
2358
return self.backing_index.get_method(version_id)
2359
if 'fulltext' in options:
2360
return 'fulltext'
2361
elif 'line-delta' in options:
2362
return 'line-delta'
2363
else:
2364
raise errors.KnitIndexUnknownMethod(self, options)
2365
2366
def get_options(self, version_id):
2367
"""Return a list representing options.
2368
2369
e.g. ['foo', 'bar']
2370
"""
2371
try:
2372
return self._by_version[version_id][0]
2373
except KeyError:
2374
return self.backing_index.get_options(version_id)
2375
2376
def get_parents_with_ghosts(self, version_id):
2377
"""Return parents of specified version with ghosts."""
2378
try:
2379
return self._by_version[version_id][2]
2380
except KeyError:
2381
return self.backing_index.get_parents_with_ghosts(version_id)
2382
2383
def get_position(self, version_id):
2384
"""Return details needed to access the version.
2385
2386
_StreamAccess has the data as a big array, so we return slice
2387
coordinates into that (as index_memo's are opaque outside the
2388
index and matching access class).
2389
2390
:return: a tuple (thunk_flag, index, start, end). If thunk_flag is
2391
False, index will be self, otherwise it will be a version id.
2392
"""
2393
try:
2394
start, end = self._by_version[version_id][1]
2395
return False, self, start, end
2396
except KeyError:
2397
# Signal to the access object to handle this from the backing knit.
2398
return (True, version_id, None, None)
2399
2400
def get_versions(self):
2401
"""Get all the versions in the stream."""
2402
return self._by_version.keys()
2403
2404
def iter_parents(self, version_ids):
2405
"""Iterate through the parents for many version ids.
2406
2407
:param version_ids: An iterable yielding version_ids.
2408
:return: An iterator that yields (version_id, parents). Requested
2409
version_ids not present in the versioned file are simply skipped.
2410
The order is undefined, allowing for different optimisations in
2411
the underlying implementation.
2412
"""
2413
result = []
2414
for version in version_ids:
2415
try:
2416
result.append((version, self._by_version[version][2]))
2417
except KeyError:
2418
pass
2419
return result
2420
2421
2422
class _KnitData(object):
2423
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2424
2425
The KnitData class provides the logic for parsing and using knit records,
2426
making use of an access method for the low level read and write operations.
2427
"""
2428
2429
def __init__(self, access):
2430
"""Create a KnitData object.
2431
2432
:param access: The access method to use. Access methods such as
2433
_KnitAccess manage the insertion of raw records and the subsequent
2434
retrieval of the same.
2435
"""
2436
self._access = access
1361
class _KnitData(_KnitComponentFile):
1362
"""Contents of the knit data file"""
1363
1364
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1365
create_parent_dir=False, delay_create=False,
1366
dir_mode=None):
1367
_KnitComponentFile.__init__(self, transport, filename, mode,
1368
file_mode=file_mode,
1369
create_parent_dir=create_parent_dir,
1370
dir_mode=dir_mode)
2437
1371
self._checked = False
2438
1372
# TODO: jam 20060713 conceptually, this could spill to disk
2439
1373
# if the cached size gets larger than a certain amount
2441
1375
# a simple dictionary
2442
1376
self._cache = {}
2443
1377
self._do_cache = False
1378
if create:
1379
if delay_create:
1380
self._need_to_create = create
1381
else:
1382
self._transport.put_bytes_non_atomic(self._filename, '',
1383
mode=self._file_mode)
2444
1384
2445
1385
def enable_cache(self):
2446
1386
"""Enable caching of reads."""
2452
1392
self._cache = {}
2453
1393
2454
1394
def _open_file(self):
2455
return self._access.open_file()
1395
try:
1396
return self._transport.get(self._filename)
1397
except NoSuchFile:
1398
pass
1399
return None
2456
1400
2457
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
1401
def _record_to_data(self, version_id, digest, lines):
2458
1402
"""Convert version_id, digest, lines into a raw data block.
2459
1403
2460
:param dense_lines: The bytes of lines but in a denser form. For
2461
instance, if lines is a list of 1000 bytestrings each ending in \n,
2462
dense_lines may be a list with one line in it, containing all the
2463
1000's lines and their \n's. Using dense_lines if it is already
2464
known is a win because the string join to create bytes in this
2465
function spends less time resizing the final string.
2466
1404
:return: (len, a StringIO instance with the raw data ready to read.)
2467
1405
"""
2468
# Note: using a string copy here increases memory pressure with e.g.
2469
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2470
# when doing the initial commit of a mozilla tree. RBC 20070921
2471
bytes = ''.join(chain(
2472
["version %s %d %s\n" % (version_id,
1406
sio = StringIO()
1407
data_file = GzipFile(None, mode='wb', fileobj=sio)
1408
1409
version_id_utf8 = cache_utf8.encode(version_id)
1410
data_file.writelines(chain(
1411
["version %s %d %s\n" % (version_id_utf8,
2473
1412
len(lines),
2474
1413
digest)],
2475
dense_lines or lines,
2476
["end %s\n" % version_id]))
2477
assert bytes.__class__ == str
2478
compressed_bytes = bytes_to_gzip(bytes)
2479
return len(compressed_bytes), compressed_bytes
2480
2481
def add_raw_records(self, sizes, raw_data):
1414
lines,
1415
["end %s\n" % version_id_utf8]))
1416
data_file.close()
1417
length= sio.tell()
1418
1419
sio.seek(0)
1420
return length, sio
1421
1422
def add_raw_record(self, raw_data):
2482
1423
"""Append a prepared record to the data file.
2483
1424
2484
:param sizes: An iterable containing the size of each raw data segment.
2485
:param raw_data: A bytestring containing the data.
2486
:return: a list of index data for the way the data was stored.
2487
See the access method add_raw_records documentation for more
2488
details.
1425
:return: the offset in the data file raw_data was written.
2489
1426
"""
2490
return self._access.add_raw_records(sizes, raw_data)
1427
assert isinstance(raw_data, str), 'data must be plain bytes'
1428
if not self._need_to_create:
1429
return self._transport.append_bytes(self._filename, raw_data)
1430
else:
1431
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1432
create_parent_dir=self._create_parent_dir,
1433
mode=self._file_mode,
1434
dir_mode=self._dir_mode)
1435
self._need_to_create = False
1436
return 0
2491
1437
1438
def add_record(self, version_id, digest, lines):
1439
"""Write new text record to disk. Returns the position in the
1440
file where it was written."""
1441
size, sio = self._record_to_data(version_id, digest, lines)
1442
# write to disk
1443
if not self._need_to_create:
1444
start_pos = self._transport.append_file(self._filename, sio)
1445
else:
1446
self._transport.put_file_non_atomic(self._filename, sio,
1447
create_parent_dir=self._create_parent_dir,
1448
mode=self._file_mode,
1449
dir_mode=self._dir_mode)
1450
self._need_to_create = False
1451
start_pos = 0
1452
if self._do_cache:
1453
self._cache[version_id] = sio.getvalue()
1454
return start_pos, size
1455
2492
1456
def _parse_record_header(self, version_id, raw_data):
2493
1457
"""Parse a record header for consistency.
2494
1458
2496
1460
as (stream, header_record)
2497
1461
"""
2498
1462
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2499
try:
2500
rec = self._check_header(version_id, df.readline())
2501
except Exception, e:
2502
raise KnitCorrupt(self._access,
2503
"While reading {%s} got %s(%s)"
2504
% (version_id, e.__class__.__name__, str(e)))
1463
rec = self._check_header(version_id, df.readline())
2505
1464
return df, rec
2506
1465
2507
1466
def _check_header(self, version_id, line):
2508
1467
rec = line.split()
2509
1468
if len(rec) != 4:
2510
raise KnitCorrupt(self._access,
1469
raise KnitCorrupt(self._filename,
2511
1470
'unexpected number of elements in record header')
2512
if rec[1] != version_id:
2513
raise KnitCorrupt(self._access,
1471
if cache_utf8.decode(rec[1]) != version_id:
1472
raise KnitCorrupt(self._filename,
2514
1473
'unexpected version, wanted %r, got %r'
2515
1474
% (version_id, rec[1]))
2516
1475
return rec
2522
1481
# 4168 calls to readlines in 330
2523
1482
df = GzipFile(mode='rb', fileobj=StringIO(data))
2524
1483
2525
try:
2526
record_contents = df.readlines()
2527
except Exception, e:
2528
raise KnitCorrupt(self._access,
2529
"While reading {%s} got %s(%s)"
2530
% (version_id, e.__class__.__name__, str(e)))
1484
record_contents = df.readlines()
2531
1485
header = record_contents.pop(0)
2532
1486
rec = self._check_header(version_id, header)
2533
1487
2534
1488
last_line = record_contents.pop()
2535
if len(record_contents) != int(rec[2]):
2536
raise KnitCorrupt(self._access,
2537
'incorrect number of lines %s != %s'
2538
' for version {%s}'
2539
% (len(record_contents), int(rec[2]),
2540
version_id))
1489
assert len(record_contents) == int(rec[2])
2541
1490
if last_line != 'end %s\n' % rec[1]:
2542
raise KnitCorrupt(self._access,
1491
raise KnitCorrupt(self._filename,
2543
1492
'unexpected version end line %r, wanted %r'
2544
1493
% (last_line, version_id))
2545
1494
df.close()
2557
1506
# grab the disk data needed.
2558
1507
if self._cache:
2559
1508
# Don't check _cache if it is empty
2560
needed_offsets = [index_memo for version_id, index_memo
1509
needed_offsets = [(pos, size) for version_id, pos, size
2561
1510
in records
2562
1511
if version_id not in self._cache]
2563
1512
else:
2564
needed_offsets = [index_memo for version_id, index_memo
1513
needed_offsets = [(pos, size) for version_id, pos, size
2565
1514
in records]
2566
1515
2567
raw_records = self._access.get_raw_records(needed_offsets)
1516
raw_records = self._transport.readv(self._filename, needed_offsets)
2568
1517
2569
for version_id, index_memo in records:
1518
for version_id, pos, size in records:
2570
1519
if version_id in self._cache:
2571
1520
# This data has already been validated
2572
1521
data = self._cache[version_id]
2573
1522
else:
2574
data = raw_records.next()
1523
pos, data = raw_records.next()
2575
1524
if self._do_cache:
2576
1525
self._cache[version_id] = data
2577
1526
2616
1565
2617
1566
# The transport optimizes the fetching as well
2618
1567
# (ie, reads continuous ranges.)
2619
raw_data = self._access.get_raw_records(
2620
[index_memo for version_id, index_memo in needed_records])
1568
readv_response = self._transport.readv(self._filename,
1569
[(pos, size) for version_id, pos, size in needed_records])
2621
1570
2622
for (version_id, index_memo), data in \
2623
izip(iter(needed_records), raw_data):
1571
for (version_id, pos, size), (pos, data) in \
1572
izip(iter(needed_records), readv_response):
2624
1573
content, digest = self._parse_record(version_id, data)
2625
1574
if self._do_cache:
2626
1575
self._cache[version_id] = data
2650
1599
except AttributeError:
2651
1600
return False
2652
1601
2653
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2654
"""Copy texts to the target by extracting and adding them one by one.
2655
2656
see join() for the parameter definitions.
2657
"""
2658
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2659
graph = self.source.get_graph(version_ids)
2660
order = topo_sort(graph.items())
2661
2662
def size_of_content(content):
2663
return sum(len(line) for line in content.text())
2664
# Cache at most 10MB of parent texts
2665
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2666
compute_size=size_of_content)
2667
# TODO: jam 20071116 It would be nice to have a streaming interface to
2668
# get multiple texts from a source. The source could be smarter
2669
# about how it handled intermediate stages.
2670
# get_line_list() or make_mpdiffs() seem like a possibility, but
2671
# at the moment they extract all full texts into memory, which
2672
# causes us to store more than our 3x fulltext goal.
2673
# Repository.iter_files_bytes() may be another possibility
2674
to_process = [version for version in order
2675
if version not in self.target]
2676
total = len(to_process)
2677
pb = ui.ui_factory.nested_progress_bar()
2678
try:
2679
for index, version in enumerate(to_process):
2680
pb.update('Converting versioned data', index, total)
2681
sha1, num_bytes, parent_text = self.target.add_lines(version,
2682
self.source.get_parents_with_ghosts(version),
2683
self.source.get_lines(version),
2684
parent_texts=parent_cache)
2685
parent_cache[version] = parent_text
2686
finally:
2687
pb.finished()
2688
return total
2689
2690
1602
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2691
1603
"""See InterVersionedFile.join."""
2692
1604
assert isinstance(self.source, KnitVersionedFile)
2693
1605
assert isinstance(self.target, KnitVersionedFile)
2694
1606
2695
# If the source and target are mismatched w.r.t. annotations vs
2696
# plain, the data needs to be converted accordingly
2697
if self.source.factory.annotated == self.target.factory.annotated:
2698
converter = None
2699
elif self.source.factory.annotated:
2700
converter = self._anno_to_plain_converter
2701
else:
2702
# We're converting from a plain to an annotated knit. Copy them
2703
# across by full texts.
2704
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2705
2706
1607
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1608
2707
1609
if not version_ids:
2708
1610
return 0
2709
1611
2713
1615
if None in version_ids:
2714
1616
version_ids.remove(None)
2715
1617
2716
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2717
topo_sorted=False))
1618
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2718
1619
this_versions = set(self.target._index.get_versions())
2719
# XXX: For efficiency we should not look at the whole index,
2720
# we only need to consider the referenced revisions - they
2721
# must all be present, or the method must be full-text.
2722
# TODO, RBC 20070919
2723
1620
needed_versions = self.source_ancestry - this_versions
1621
cross_check_versions = self.source_ancestry.intersection(this_versions)
1622
mismatched_versions = set()
1623
for version in cross_check_versions:
1624
# scan to include needed parents.
1625
n1 = set(self.target.get_parents_with_ghosts(version))
1626
n2 = set(self.source.get_parents_with_ghosts(version))
1627
if n1 != n2:
1628
# FIXME TEST this check for cycles being introduced works
1629
# the logic is we have a cycle if in our graph we are an
1630
# ancestor of any of the n2 revisions.
1631
for parent in n2:
1632
if parent in n1:
1633
# safe
1634
continue
1635
else:
1636
parent_ancestors = self.source.get_ancestry(parent)
1637
if version in parent_ancestors:
1638
raise errors.GraphCycleError([parent, version])
1639
# ensure this parent will be available later.
1640
new_parents = n2.difference(n1)
1641
needed_versions.update(new_parents.difference(this_versions))
1642
mismatched_versions.add(version)
2724
1643
2725
if not needed_versions:
1644
if not needed_versions and not mismatched_versions:
2726
1645
return 0
2727
1646
full_list = topo_sort(self.source.get_graph())
2728
1647
2745
1664
assert (self.target.has_version(parent) or
2746
1665
parent in copy_set or
2747
1666
not self.source.has_version(parent))
2748
index_memo = self.source._index.get_position(version_id)
2749
copy_queue_records.append((version_id, index_memo))
1667
data_pos, data_size = self.source._index.get_position(version_id)
1668
copy_queue_records.append((version_id, data_pos, data_size))
2750
1669
copy_queue.append((version_id, options, parents))
2751
1670
copy_set.add(version_id)
2752
1671
2762
1681
assert version_id == version_id2, 'logic error, inconsistent results'
2763
1682
count = count + 1
2764
1683
pb.update("Joining knit", count, total)
2765
if converter:
2766
size, raw_data = converter(raw_data, version_id, options,
2767
parents)
2768
else:
2769
size = len(raw_data)
2770
raw_records.append((version_id, options, parents, size))
1684
raw_records.append((version_id, options, parents, len(raw_data)))
2771
1685
raw_datum.append(raw_data)
2772
1686
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1687
1688
for version in mismatched_versions:
1689
# FIXME RBC 20060309 is this needed?
1690
n1 = set(self.target.get_parents_with_ghosts(version))
1691
n2 = set(self.source.get_parents_with_ghosts(version))
1692
# write a combined record to our history preserving the current
1693
# parents as first in the list
1694
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1695
self.target.fix_parents(version, new_parents)
2773
1696
return count
2774
1697
finally:
2775
1698
pb.finished()
2776
1699
2777
def _anno_to_plain_converter(self, raw_data, version_id, options,
2778
parents):
2779
"""Convert annotated content to plain content."""
2780
data, digest = self.source._data._parse_record(version_id, raw_data)
2781
if 'fulltext' in options:
2782
content = self.source.factory.parse_fulltext(data, version_id)
2783
lines = self.target.factory.lower_fulltext(content)
2784
else:
2785
delta = self.source.factory.parse_line_delta(data, version_id,
2786
plain=True)
2787
lines = self.target.factory.lower_line_delta(delta)
2788
return self.target._data._record_to_data(version_id, digest, lines)
2789
2790
1700
2791
1701
InterVersionedFile.register_optimiser(InterKnit)
2792
1702
2823
1733
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2824
1734
this_versions = set(self.target._index.get_versions())
2825
1735
needed_versions = self.source_ancestry - this_versions
1736
cross_check_versions = self.source_ancestry.intersection(this_versions)
1737
mismatched_versions = set()
1738
for version in cross_check_versions:
1739
# scan to include needed parents.
1740
n1 = set(self.target.get_parents_with_ghosts(version))
1741
n2 = set(self.source.get_parents(version))
1742
# if all of n2's parents are in n1, then its fine.
1743
if n2.difference(n1):
1744
# FIXME TEST this check for cycles being introduced works
1745
# the logic is we have a cycle if in our graph we are an
1746
# ancestor of any of the n2 revisions.
1747
for parent in n2:
1748
if parent in n1:
1749
# safe
1750
continue
1751
else:
1752
parent_ancestors = self.source.get_ancestry(parent)
1753
if version in parent_ancestors:
1754
raise errors.GraphCycleError([parent, version])
1755
# ensure this parent will be available later.
1756
new_parents = n2.difference(n1)
1757
needed_versions.update(new_parents.difference(this_versions))
1758
mismatched_versions.add(version)
2826
1759
2827
if not needed_versions:
1760
if not needed_versions and not mismatched_versions:
2828
1761
return 0
2829
1762
full_list = topo_sort(self.source.get_graph())
2830
1763
2844
1777
self.target.add_lines(
2845
1778
version_id, parents, self.source.get_lines(version_id))
2846
1779
count = count + 1
1780
1781
for version in mismatched_versions:
1782
# FIXME RBC 20060309 is this needed?
1783
n1 = set(self.target.get_parents_with_ghosts(version))
1784
n2 = set(self.source.get_parents(version))
1785
# write a combined record to our history preserving the current
1786
# parents as first in the list
1787
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1788
self.target.fix_parents(version, new_parents)
2847
1789
return count
2848
1790
finally:
2849
1791
pb.finished()
2852
1794
InterVersionedFile.register_optimiser(WeaveToKnit)
2853
1795
2854
1796
2855
# Deprecated, use PatienceSequenceMatcher instead
2856
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2857
2858
2859
def annotate_knit(knit, revision_id):
2860
"""Annotate a knit with no cached annotations.
2861
2862
This implementation is for knits with no cached annotations.
2863
It will work for knits with cached annotations, but this is not
2864
recommended.
1797
class KnitSequenceMatcher(difflib.SequenceMatcher):
1798
"""Knit tuned sequence matcher.
1799
1800
This is based on profiling of difflib which indicated some improvements
1801
for our usage pattern.
2865
1802
"""
2866
annotator = _KnitAnnotator(knit)
2867
return iter(annotator.annotate(revision_id))
2868
2869
2870
class _KnitAnnotator(object):
2871
"""Build up the annotations for a text."""
2872
2873
def __init__(self, knit):
2874
self._knit = knit
2875
2876
# Content objects, differs from fulltexts because of how final newlines
2877
# are treated by knits. the content objects here will always have a
2878
# final newline
2879
self._fulltext_contents = {}
2880
2881
# Annotated lines of specific revisions
2882
self._annotated_lines = {}
2883
2884
# Track the raw data for nodes that we could not process yet.
2885
# This maps the revision_id of the base to a list of children that will
2886
# annotated from it.
2887
self._pending_children = {}
2888
2889
# Nodes which cannot be extracted
2890
self._ghosts = set()
2891
2892
# Track how many children this node has, so we know if we need to keep
2893
# it
2894
self._annotate_children = {}
2895
self._compression_children = {}
2896
2897
self._all_build_details = {}
2898
# The children => parent revision_id graph
2899
self._revision_id_graph = {}
2900
2901
self._heads_provider = None
2902
2903
self._nodes_to_keep_annotations = set()
2904
self._generations_until_keep = 100
2905
2906
def set_generations_until_keep(self, value):
2907
"""Set the number of generations before caching a node.
2908
2909
Setting this to -1 will cache every merge node, setting this higher
2910
will cache fewer nodes.
2911
"""
2912
self._generations_until_keep = value
2913
2914
def _add_fulltext_content(self, revision_id, content_obj):
2915
self._fulltext_contents[revision_id] = content_obj
2916
# TODO: jam 20080305 It might be good to check the sha1digest here
2917
return content_obj.text()
2918
2919
def _check_parents(self, child, nodes_to_annotate):
2920
"""Check if all parents have been processed.
2921
2922
:param child: A tuple of (rev_id, parents, raw_content)
2923
:param nodes_to_annotate: If child is ready, add it to
2924
nodes_to_annotate, otherwise put it back in self._pending_children
2925
"""
2926
for parent_id in child[1]:
2927
if (parent_id not in self._annotated_lines):
2928
# This parent is present, but another parent is missing
2929
self._pending_children.setdefault(parent_id,
2930
[]).append(child)
2931
break
2932
else:
2933
# This one is ready to be processed
2934
nodes_to_annotate.append(child)
2935
2936
def _add_annotation(self, revision_id, fulltext, parent_ids,
2937
left_matching_blocks=None):
2938
"""Add an annotation entry.
2939
2940
All parents should already have been annotated.
2941
:return: A list of children that now have their parents satisfied.
2942
"""
2943
a = self._annotated_lines
2944
annotated_parent_lines = [a[p] for p in parent_ids]
2945
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2946
fulltext, revision_id, left_matching_blocks,
2947
heads_provider=self._get_heads_provider()))
2948
self._annotated_lines[revision_id] = annotated_lines
2949
for p in parent_ids:
2950
ann_children = self._annotate_children[p]
2951
ann_children.remove(revision_id)
2952
if (not ann_children
2953
and p not in self._nodes_to_keep_annotations):
2954
del self._annotated_lines[p]
2955
del self._all_build_details[p]
2956
if p in self._fulltext_contents:
2957
del self._fulltext_contents[p]
2958
# Now that we've added this one, see if there are any pending
2959
# deltas to be done, certainly this parent is finished
2960
nodes_to_annotate = []
2961
for child in self._pending_children.pop(revision_id, []):
2962
self._check_parents(child, nodes_to_annotate)
2963
return nodes_to_annotate
2964
2965
def _get_build_graph(self, revision_id):
2966
"""Get the graphs for building texts and annotations.
2967
2968
The data you need for creating a full text may be different than the
2969
data you need to annotate that text. (At a minimum, you need both
2970
parents to create an annotation, but only need 1 parent to generate the
2971
fulltext.)
2972
2973
:return: A list of (revision_id, index_memo) records, suitable for
2974
passing to read_records_iter to start reading in the raw data fro/
2975
the pack file.
2976
"""
2977
if revision_id in self._annotated_lines:
2978
# Nothing to do
2979
return []
2980
pending = set([revision_id])
2981
records = []
2982
generation = 0
2983
kept_generation = 0
2984
while pending:
2985
# get all pending nodes
2986
generation += 1
2987
this_iteration = pending
2988
build_details = self._knit._index.get_build_details(this_iteration)
2989
self._all_build_details.update(build_details)
2990
# new_nodes = self._knit._index._get_entries(this_iteration)
2991
pending = set()
2992
for rev_id, details in build_details.iteritems():
2993
(index_memo, compression_parent, parents,
2994
record_details) = details
2995
self._revision_id_graph[rev_id] = parents
2996
records.append((rev_id, index_memo))
2997
# Do we actually need to check _annotated_lines?
2998
pending.update(p for p in parents
2999
if p not in self._all_build_details)
3000
if compression_parent:
3001
self._compression_children.setdefault(compression_parent,
3002
[]).append(rev_id)
3003
if parents:
3004
for parent in parents:
3005
self._annotate_children.setdefault(parent,
3006
[]).append(rev_id)
3007
num_gens = generation - kept_generation
3008
if ((num_gens >= self._generations_until_keep)
3009
and len(parents) > 1):
3010
kept_generation = generation
3011
self._nodes_to_keep_annotations.add(rev_id)
3012
3013
missing_versions = this_iteration.difference(build_details.keys())
3014
self._ghosts.update(missing_versions)
3015
for missing_version in missing_versions:
3016
# add a key, no parents
3017
self._revision_id_graph[missing_version] = ()
3018
pending.discard(missing_version) # don't look for it
3019
# XXX: This should probably be a real exception, as it is a data
3020
# inconsistency
3021
assert not self._ghosts.intersection(self._compression_children), \
3022
"We cannot have nodes which have a compression parent of a ghost."
3023
# Cleanout anything that depends on a ghost so that we don't wait for
3024
# the ghost to show up
3025
for node in self._ghosts:
3026
if node in self._annotate_children:
3027
# We won't be building this node
3028
del self._annotate_children[node]
3029
# Generally we will want to read the records in reverse order, because
3030
# we find the parent nodes after the children
3031
records.reverse()
3032
return records
3033
3034
def _annotate_records(self, records):
3035
"""Build the annotations for the listed records."""
3036
# We iterate in the order read, rather than a strict order requested
3037
# However, process what we can, and put off to the side things that
3038
# still need parents, cleaning them up when those parents are
3039
# processed.
3040
for (rev_id, record,
3041
digest) in self._knit._data.read_records_iter(records):
3042
if rev_id in self._annotated_lines:
3043
continue
3044
parent_ids = self._revision_id_graph[rev_id]
3045
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3046
details = self._all_build_details[rev_id]
3047
(index_memo, compression_parent, parents,
3048
record_details) = details
3049
nodes_to_annotate = []
3050
# TODO: Remove the punning between compression parents, and
3051
# parent_ids, we should be able to do this without assuming
3052
# the build order
3053
if len(parent_ids) == 0:
3054
# There are no parents for this node, so just add it
3055
# TODO: This probably needs to be decoupled
3056
assert compression_parent is None
3057
fulltext_content, delta = self._knit.factory.parse_record(
3058
rev_id, record, record_details, None)
3059
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3060
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3061
parent_ids, left_matching_blocks=None))
1803
1804
def find_longest_match(self, alo, ahi, blo, bhi):
1805
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1806
1807
If isjunk is not defined:
1808
1809
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1810
alo <= i <= i+k <= ahi
1811
blo <= j <= j+k <= bhi
1812
and for all (i',j',k') meeting those conditions,
1813
k >= k'
1814
i <= i'
1815
and if i == i', j <= j'
1816
1817
In other words, of all maximal matching blocks, return one that
1818
starts earliest in a, and of all those maximal matching blocks that
1819
start earliest in a, return the one that starts earliest in b.
1820
1821
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1822
>>> s.find_longest_match(0, 5, 0, 9)
1823
(0, 4, 5)
1824
1825
If isjunk is defined, first the longest matching block is
1826
determined as above, but with the additional restriction that no
1827
junk element appears in the block. Then that block is extended as
1828
far as possible by matching (only) junk elements on both sides. So
1829
the resulting block never matches on junk except as identical junk
1830
happens to be adjacent to an "interesting" match.
1831
1832
Here's the same example as before, but considering blanks to be
1833
junk. That prevents " abcd" from matching the " abcd" at the tail
1834
end of the second sequence directly. Instead only the "abcd" can
1835
match, and matches the leftmost "abcd" in the second sequence:
1836
1837
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1838
>>> s.find_longest_match(0, 5, 0, 9)
1839
(1, 0, 4)
1840
1841
If no blocks match, return (alo, blo, 0).
1842
1843
>>> s = SequenceMatcher(None, "ab", "c")
1844
>>> s.find_longest_match(0, 2, 0, 1)
1845
(0, 0, 0)
1846
"""
1847
1848
# CAUTION: stripping common prefix or suffix would be incorrect.
1849
# E.g.,
1850
# ab
1851
# acab
1852
# Longest matching block is "ab", but if common prefix is
1853
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1854
# strip, so ends up claiming that ab is changed to acab by
1855
# inserting "ca" in the middle. That's minimal but unintuitive:
1856
# "it's obvious" that someone inserted "ac" at the front.
1857
# Windiff ends up at the same place as diff, but by pairing up
1858
# the unique 'b's and then matching the first two 'a's.
1859
1860
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1861
besti, bestj, bestsize = alo, blo, 0
1862
# find longest junk-free match
1863
# during an iteration of the loop, j2len[j] = length of longest
1864
# junk-free match ending with a[i-1] and b[j]
1865
j2len = {}
1866
# nothing = []
1867
b2jget = b2j.get
1868
for i in xrange(alo, ahi):
1869
# look at all instances of a[i] in b; note that because
1870
# b2j has no junk keys, the loop is skipped if a[i] is junk
1871
j2lenget = j2len.get
1872
newj2len = {}
1873
1874
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1875
# following improvement
1876
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1877
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1878
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1879
# to
1880
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1881
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1882
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1883
1884
try:
1885
js = b2j[a[i]]
1886
except KeyError:
1887
pass
3062
1888
else:
3063
child = (rev_id, parent_ids, record)
3064
# Check if all the parents are present
3065
self._check_parents(child, nodes_to_annotate)
3066
while nodes_to_annotate:
3067
# Should we use a queue here instead of a stack?
3068
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3069
(index_memo, compression_parent, parents,
3070
record_details) = self._all_build_details[rev_id]
3071
if compression_parent is not None:
3072
comp_children = self._compression_children[compression_parent]
3073
assert rev_id in comp_children
3074
# If there is only 1 child, it is safe to reuse this
3075
# content
3076
reuse_content = (len(comp_children) == 1
3077
and compression_parent not in
3078
self._nodes_to_keep_annotations)
3079
if reuse_content:
3080
# Remove it from the cache since it will be changing
3081
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3082
# Make sure to copy the fulltext since it might be
3083
# modified
3084
parent_fulltext = list(parent_fulltext_content.text())
3085
else:
3086
parent_fulltext_content = self._fulltext_contents[compression_parent]
3087
parent_fulltext = parent_fulltext_content.text()
3088
comp_children.remove(rev_id)
3089
fulltext_content, delta = self._knit.factory.parse_record(
3090
rev_id, record, record_details,
3091
parent_fulltext_content,
3092
copy_base_content=(not reuse_content))
3093
fulltext = self._add_fulltext_content(rev_id,
3094
fulltext_content)
3095
blocks = KnitContent.get_line_delta_blocks(delta,
3096
parent_fulltext, fulltext)
3097
else:
3098
fulltext_content = self._knit.factory.parse_fulltext(
3099
record, rev_id)
3100
fulltext = self._add_fulltext_content(rev_id,
3101
fulltext_content)
3102
blocks = None
3103
nodes_to_annotate.extend(
3104
self._add_annotation(rev_id, fulltext, parent_ids,
3105
left_matching_blocks=blocks))
3106
3107
def _get_heads_provider(self):
3108
"""Create a heads provider for resolving ancestry issues."""
3109
if self._heads_provider is not None:
3110
return self._heads_provider
3111
parent_provider = _mod_graph.DictParentsProvider(
3112
self._revision_id_graph)
3113
graph_obj = _mod_graph.Graph(parent_provider)
3114
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3115
self._heads_provider = head_cache
3116
return head_cache
3117
3118
def annotate(self, revision_id):
3119
"""Return the annotated fulltext at the given revision.
3120
3121
:param revision_id: The revision id for this file
3122
"""
3123
records = self._get_build_graph(revision_id)
3124
if revision_id in self._ghosts:
3125
raise errors.RevisionNotPresent(revision_id, self._knit)
3126
self._annotate_records(records)
3127
return self._annotated_lines[revision_id]
3128
3129
3130
try:
3131
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3132
except ImportError:
3133
from bzrlib._knit_load_data_py import _load_data_py as _load_data
1889
for j in js:
1890
# a[i] matches b[j]
1891
if j >= blo:
1892
if j >= bhi:
1893
break
1894
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1895
if k > bestsize:
1896
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1897
j2len = newj2len
1898
1899
# Extend the best by non-junk elements on each end. In particular,
1900
# "popular" non-junk elements aren't in b2j, which greatly speeds
1901
# the inner loop above, but also means "the best" match so far
1902
# doesn't contain any junk *or* popular non-junk elements.
1903
while besti > alo and bestj > blo and \
1904
not isbjunk(b[bestj-1]) and \
1905
a[besti-1] == b[bestj-1]:
1906
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1907
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1908
not isbjunk(b[bestj+bestsize]) and \
1909
a[besti+bestsize] == b[bestj+bestsize]:
1910
bestsize += 1
1911
1912
# Now that we have a wholly interesting match (albeit possibly
1913
# empty!), we may as well suck up the matching junk on each
1914
# side of it too. Can't think of a good reason not to, and it
1915
# saves post-processing the (possibly considerable) expense of
1916
# figuring out what to do with it. In the case of an empty
1917
# interesting match, this is clearly the right thing to do,
1918
# because no other kind of match is possible in the regions.
1919
while besti > alo and bestj > blo and \
1920
isbjunk(b[bestj-1]) and \
1921
a[besti-1] == b[bestj-1]:
1922
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1923
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1924
isbjunk(b[bestj+bestsize]) and \
1925
a[besti+bestsize] == b[bestj+bestsize]:
1926
bestsize = bestsize + 1
1927
1928
return besti, bestj, bestsize
Loggerhead is a web-based interface for Breezy
Version: 2.0.1