« back to all changes in this revision
Viewing changes to bzrlib/knit.py
- browse files at revision 3441.5.3
- compare with another revision
- download diff
- download tarball
- view history from revision 3441.5.3
- Committer: Andrew Bennetts
- Date: 2008-05-21 11:58:09 UTC
- mto: (3452.2.9 inter-remote-pack)
- mto: This revision was merged to the branch mainline in revision 3511.
- Revision ID: andrew.bennetts@canonical.com-20080521115809-6cw3t8gn4qm0bpg9
Remove a bit more debugging cruft.
- files added:
- COPYING.txt
- bzrlib/benchmarks/tree_creator
- bzrlib/help_topics
- bzrlib/help_topics/en
- bzrlib/repofmt
- bzrlib/smart
- bzrlib/tests/commands
- bzrlib/tests/intertree_implementations
- bzrlib/tests/inventory_implementations
- bzrlib/tests/per_lock
- bzrlib/tests/tree_implementations
- bzrlib/util/tests
- bzrlib/version_info_formats
- 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
- doc/es
- doc/es/guia-desarrollador
- doc/es/guia-usuario
- doc/es/mini-tutorial
- doc/es/notas-version
- doc/es/referencia
- doc/es/referencia-rapida
- man1
- tools/win32
- files removed:
- NEWS.developers
- bzrlib/bundle/old
- bzrlib/util/urlgrabber
- files renamed:
- bzrlib/patiencediff.py => bzrlib/_patiencediff_py.py
- bzrlib/bundle/read_bundle.py => bzrlib/bundle/bundle_data.py
- bzrlib/bundle/serializer/v07.py => bzrlib/bundle/serializer/v08.py
- bzrlib/graph.py => bzrlib/deprecated_graph.py
- doc/configuration.txt => bzrlib/help_topics/en/configuration.txt
- bzrlib/tests/HTTPTestUtil.py => bzrlib/tests/http_utils.py
- bzrlib/tests/test_revprops.py => bzrlib/tests/repository_implementations/test_revision.py
- bzrlib/tests/test_command.py => bzrlib/tests/test_commands.py
- bzrlib/tests/test_graph.py => bzrlib/tests/test_deprecated_graph.py
- bzrlib/tests/test_revisionnamespaces.py => bzrlib/tests/test_revisionspec.py
- bzrlib/win32console.py => bzrlib/win32utils.py
- bzrlib/xml5.py => bzrlib/xml8.py
- HACKING => doc/developers/HACKING.txt
- doc/tutorial.txt => doc/en/tutorials/tutorial.txt
- doc/plugins.txt => doc/en/user-guide/plugins.txt
- doc/setting_up_email.txt => doc/en/user-guide/setting_up_email.txt
- doc/specifying_revisions.txt => doc/en/user-guide/specifying_revisions.txt
- doc/using_aliases.txt => doc/en/user-guide/using_aliases.txt
- files modified:
- .bzrignore
added
removed
bzrlib/knit.py
1
# Copyright (C) 2005, 2006 by Canonical Ltd
2
# Written by Martin Pool.
3
# Modified by Johan Rydberg <jrydberg@gnu.org>
4
# Modified by Robert Collins <robert.collins@canonical.com>
1
# Copyright (C) 2005, 2006, 2007 Canonical Ltd
5
2
#
6
3
# This program is free software; you can redistribute it and/or modify
7
4
# it under the terms of the GNU General Public License as published by
65
62
66
63
from copy import copy
67
64
from cStringIO import StringIO
68
import difflib
69
65
from itertools import izip, chain
66
import operator
70
67
import os
71
68
import sys
69
import warnings
70
from zlib import Z_DEFAULT_COMPRESSION
72
71
73
72
import bzrlib
74
import bzrlib.errors as errors
75
from bzrlib.errors import FileExists, NoSuchFile, KnitError, \
76
InvalidRevisionId, KnitCorrupt, KnitHeaderError, \
77
RevisionNotPresent, RevisionAlreadyPresent
78
from bzrlib.tuned_gzip import *
79
from bzrlib.trace import mutter
80
from bzrlib.osutils import contains_whitespace, contains_linebreaks, \
81
sha_strings
82
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
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
from bzrlib import (
84
cache_utf8,
85
debug,
86
diff,
87
errors,
88
osutils,
89
patiencediff,
90
progress,
91
merge,
92
ui,
93
)
94
from bzrlib.errors import (
95
FileExists,
96
NoSuchFile,
97
KnitError,
98
InvalidRevisionId,
99
KnitCorrupt,
100
KnitHeaderError,
101
RevisionNotPresent,
102
RevisionAlreadyPresent,
103
)
104
from bzrlib.graph import Graph
105
from bzrlib.osutils import (
106
contains_whitespace,
107
contains_linebreaks,
108
sha_string,
109
sha_strings,
110
split_lines,
111
)
83
112
from bzrlib.tsort import topo_sort
113
from bzrlib.tuned_gzip import GzipFile, bytes_to_gzip
114
import bzrlib.ui
115
from bzrlib.versionedfile import (
116
AbsentContentFactory,
117
adapter_registry,
118
ContentFactory,
119
InterVersionedFile,
120
VersionedFile,
121
)
84
122
import bzrlib.weave
85
123
86
124
101
139
INDEX_SUFFIX = '.kndx'
102
140
103
141
142
class KnitAdapter(object):
143
"""Base class for knit record adaption."""
144
145
def __init__(self, basis_vf):
146
"""Create an adapter which accesses full texts from basis_vf.
147
148
:param basis_vf: A versioned file to access basis texts of deltas from.
149
May be None for adapters that do not need to access basis texts.
150
"""
151
self._data = _KnitData(None)
152
self._annotate_factory = KnitAnnotateFactory()
153
self._plain_factory = KnitPlainFactory()
154
self._basis_vf = basis_vf
155
156
157
class FTAnnotatedToUnannotated(KnitAdapter):
158
"""An adapter from FT annotated knits to unannotated ones."""
159
160
def get_bytes(self, factory, annotated_compressed_bytes):
161
rec, contents = \
162
self._data._parse_record_unchecked(annotated_compressed_bytes)
163
content = self._annotate_factory.parse_fulltext(contents, rec[1])
164
size, bytes = self._data._record_to_data(rec[1], rec[3], content.text())
165
return bytes
166
167
168
class DeltaAnnotatedToUnannotated(KnitAdapter):
169
"""An adapter for deltas from annotated to unannotated."""
170
171
def get_bytes(self, factory, annotated_compressed_bytes):
172
rec, contents = \
173
self._data._parse_record_unchecked(annotated_compressed_bytes)
174
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
175
plain=True)
176
contents = self._plain_factory.lower_line_delta(delta)
177
size, bytes = self._data._record_to_data(rec[1], rec[3], contents)
178
return bytes
179
180
181
class FTAnnotatedToFullText(KnitAdapter):
182
"""An adapter from FT annotated knits to unannotated ones."""
183
184
def get_bytes(self, factory, annotated_compressed_bytes):
185
rec, contents = \
186
self._data._parse_record_unchecked(annotated_compressed_bytes)
187
content, delta = self._annotate_factory.parse_record(factory.key[0],
188
contents, factory._build_details, None)
189
return ''.join(content.text())
190
191
192
class DeltaAnnotatedToFullText(KnitAdapter):
193
"""An adapter for deltas from annotated to unannotated."""
194
195
def get_bytes(self, factory, annotated_compressed_bytes):
196
rec, contents = \
197
self._data._parse_record_unchecked(annotated_compressed_bytes)
198
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
199
plain=True)
200
compression_parent = factory.parents[0][0]
201
basis_lines = self._basis_vf.get_lines(compression_parent)
202
# Manually apply the delta because we have one annotated content and
203
# one plain.
204
basis_content = PlainKnitContent(basis_lines, compression_parent)
205
basis_content.apply_delta(delta, rec[1])
206
basis_content._should_strip_eol = factory._build_details[1]
207
return ''.join(basis_content.text())
208
209
210
class FTPlainToFullText(KnitAdapter):
211
"""An adapter from FT plain knits to unannotated ones."""
212
213
def get_bytes(self, factory, compressed_bytes):
214
rec, contents = \
215
self._data._parse_record_unchecked(compressed_bytes)
216
content, delta = self._plain_factory.parse_record(factory.key[0],
217
contents, factory._build_details, None)
218
return ''.join(content.text())
219
220
221
class DeltaPlainToFullText(KnitAdapter):
222
"""An adapter for deltas from annotated to unannotated."""
223
224
def get_bytes(self, factory, compressed_bytes):
225
rec, contents = \
226
self._data._parse_record_unchecked(compressed_bytes)
227
delta = self._plain_factory.parse_line_delta(contents, rec[1])
228
compression_parent = factory.parents[0][0]
229
basis_lines = self._basis_vf.get_lines(compression_parent)
230
basis_content = PlainKnitContent(basis_lines, compression_parent)
231
# Manually apply the delta because we have one annotated content and
232
# one plain.
233
content, _ = self._plain_factory.parse_record(rec[1], contents,
234
factory._build_details, basis_content)
235
return ''.join(content.text())
236
237
238
class KnitContentFactory(ContentFactory):
239
"""Content factory for streaming from knits.
240
241
:seealso ContentFactory:
242
"""
243
244
def __init__(self, version, parents, build_details, sha1, raw_record,
245
annotated, knit=None):
246
"""Create a KnitContentFactory for version.
247
248
:param version: The version.
249
:param parents: The parents.
250
:param build_details: The build details as returned from
251
get_build_details.
252
:param sha1: The sha1 expected from the full text of this object.
253
:param raw_record: The bytes of the knit data from disk.
254
:param annotated: True if the raw data is annotated.
255
"""
256
ContentFactory.__init__(self)
257
self.sha1 = sha1
258
self.key = (version,)
259
self.parents = tuple((parent,) for parent in parents)
260
if build_details[0] == 'line-delta':
261
kind = 'delta'
262
else:
263
kind = 'ft'
264
if annotated:
265
annotated_kind = 'annotated-'
266
else:
267
annotated_kind = ''
268
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
269
self._raw_record = raw_record
270
self._build_details = build_details
271
self._knit = knit
272
273
def get_bytes_as(self, storage_kind):
274
if storage_kind == self.storage_kind:
275
return self._raw_record
276
if storage_kind == 'fulltext' and self._knit is not None:
277
return self._knit.get_text(self.key[0])
278
else:
279
raise errors.UnavailableRepresentation(self.key, storage_kind,
280
self.storage_kind)
281
282
104
283
class KnitContent(object):
105
284
"""Content of a knit version to which deltas can be applied."""
106
285
107
def __init__(self, lines):
108
self._lines = lines
109
110
def annotate_iter(self):
111
"""Yield tuples of (origin, text) for each content line."""
112
for origin, text in self._lines:
113
yield origin, text
114
115
def annotate(self):
116
"""Return a list of (origin, text) tuples."""
117
return list(self.annotate_iter())
286
def __init__(self):
287
self._should_strip_eol = False
288
289
def apply_delta(self, delta, new_version_id):
290
"""Apply delta to this object to become new_version_id."""
291
raise NotImplementedError(self.apply_delta)
292
293
def cleanup_eol(self, copy_on_mutate=True):
294
if self._should_strip_eol:
295
if copy_on_mutate:
296
self._lines = self._lines[:]
297
self.strip_last_line_newline()
118
298
119
299
def line_delta_iter(self, new_lines):
120
300
"""Generate line-based delta from this content to new_lines."""
121
new_texts = [text for origin, text in new_lines._lines]
122
old_texts = [text for origin, text in self._lines]
123
s = KnitSequenceMatcher(None, old_texts, new_texts)
124
for op in s.get_opcodes():
125
if op[0] == 'equal':
301
new_texts = new_lines.text()
302
old_texts = self.text()
303
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
304
for tag, i1, i2, j1, j2 in s.get_opcodes():
305
if tag == 'equal':
126
306
continue
127
# ofrom oto length data
128
yield (op[1], op[2], op[4]-op[3], new_lines._lines[op[3]:op[4]])
307
# ofrom, oto, length, data
308
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
129
309
130
310
def line_delta(self, new_lines):
131
311
return list(self.line_delta_iter(new_lines))
132
312
133
def text(self):
134
return [text for origin, text in self._lines]
313
@staticmethod
314
def get_line_delta_blocks(knit_delta, source, target):
315
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
316
target_len = len(target)
317
s_pos = 0
318
t_pos = 0
319
for s_begin, s_end, t_len, new_text in knit_delta:
320
true_n = s_begin - s_pos
321
n = true_n
322
if n > 0:
323
# knit deltas do not provide reliable info about whether the
324
# last line of a file matches, due to eol handling.
325
if source[s_pos + n -1] != target[t_pos + n -1]:
326
n-=1
327
if n > 0:
328
yield s_pos, t_pos, n
329
t_pos += t_len + true_n
330
s_pos = s_end
331
n = target_len - t_pos
332
if n > 0:
333
if source[s_pos + n -1] != target[t_pos + n -1]:
334
n-=1
335
if n > 0:
336
yield s_pos, t_pos, n
337
yield s_pos + (target_len - t_pos), target_len, 0
338
339
340
class AnnotatedKnitContent(KnitContent):
341
"""Annotated content."""
342
343
def __init__(self, lines):
344
KnitContent.__init__(self)
345
self._lines = lines
346
347
def annotate(self):
348
"""Return a list of (origin, text) for each content line."""
349
return list(self._lines)
350
351
def apply_delta(self, delta, new_version_id):
352
"""Apply delta to this object to become new_version_id."""
353
offset = 0
354
lines = self._lines
355
for start, end, count, delta_lines in delta:
356
lines[offset+start:offset+end] = delta_lines
357
offset = offset + (start - end) + count
358
359
def strip_last_line_newline(self):
360
line = self._lines[-1][1].rstrip('\n')
361
self._lines[-1] = (self._lines[-1][0], line)
362
self._should_strip_eol = False
363
364
def text(self):
365
try:
366
lines = [text for origin, text in self._lines]
367
except ValueError, e:
368
# most commonly (only?) caused by the internal form of the knit
369
# missing annotation information because of a bug - see thread
370
# around 20071015
371
raise KnitCorrupt(self,
372
"line in annotated knit missing annotation information: %s"
373
% (e,))
374
375
if self._should_strip_eol:
376
lines[-1] = lines[-1].rstrip('\n')
377
return lines
378
379
def copy(self):
380
return AnnotatedKnitContent(self._lines[:])
381
382
383
class PlainKnitContent(KnitContent):
384
"""Unannotated content.
385
386
When annotate[_iter] is called on this content, the same version is reported
387
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
388
objects.
389
"""
390
391
def __init__(self, lines, version_id):
392
KnitContent.__init__(self)
393
self._lines = lines
394
self._version_id = version_id
395
396
def annotate(self):
397
"""Return a list of (origin, text) for each content line."""
398
return [(self._version_id, line) for line in self._lines]
399
400
def apply_delta(self, delta, new_version_id):
401
"""Apply delta to this object to become new_version_id."""
402
offset = 0
403
lines = self._lines
404
for start, end, count, delta_lines in delta:
405
lines[offset+start:offset+end] = delta_lines
406
offset = offset + (start - end) + count
407
self._version_id = new_version_id
408
409
def copy(self):
410
return PlainKnitContent(self._lines[:], self._version_id)
411
412
def strip_last_line_newline(self):
413
self._lines[-1] = self._lines[-1].rstrip('\n')
414
self._should_strip_eol = False
415
416
def text(self):
417
lines = self._lines
418
if self._should_strip_eol:
419
lines = lines[:]
420
lines[-1] = lines[-1].rstrip('\n')
421
return lines
135
422
136
423
137
424
class _KnitFactory(object):
138
"""Base factory for creating content objects."""
139
140
def make(self, lines, version):
141
num_lines = len(lines)
142
return KnitContent(zip([version] * num_lines, lines))
425
"""Base class for common Factory functions."""
426
427
def parse_record(self, version_id, record, record_details,
428
base_content, copy_base_content=True):
429
"""Parse a record into a full content object.
430
431
:param version_id: The official version id for this content
432
:param record: The data returned by read_records_iter()
433
:param record_details: Details about the record returned by
434
get_build_details
435
:param base_content: If get_build_details returns a compression_parent,
436
you must return a base_content here, else use None
437
:param copy_base_content: When building from the base_content, decide
438
you can either copy it and return a new object, or modify it in
439
place.
440
:return: (content, delta) A Content object and possibly a line-delta,
441
delta may be None
442
"""
443
method, noeol = record_details
444
if method == 'line-delta':
445
if copy_base_content:
446
content = base_content.copy()
447
else:
448
content = base_content
449
delta = self.parse_line_delta(record, version_id)
450
content.apply_delta(delta, version_id)
451
else:
452
content = self.parse_fulltext(record, version_id)
453
delta = None
454
content._should_strip_eol = noeol
455
return (content, delta)
143
456
144
457
145
458
class KnitAnnotateFactory(_KnitFactory):
147
460
148
461
annotated = True
149
462
150
def parse_fulltext(self, content, version):
463
def make(self, lines, version_id):
464
num_lines = len(lines)
465
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
466
467
def parse_fulltext(self, content, version_id):
151
468
"""Convert fulltext to internal representation
152
469
153
470
fulltext content is of the format
155
472
internal representation is of the format:
156
473
(revid, plaintext)
157
474
"""
158
lines = []
159
for line in content:
160
origin, text = line.split(' ', 1)
161
lines.append((origin.decode('utf-8'), text))
162
return KnitContent(lines)
475
# TODO: jam 20070209 The tests expect this to be returned as tuples,
476
# but the code itself doesn't really depend on that.
477
# Figure out a way to not require the overhead of turning the
478
# list back into tuples.
479
lines = [tuple(line.split(' ', 1)) for line in content]
480
return AnnotatedKnitContent(lines)
163
481
164
482
def parse_line_delta_iter(self, lines):
165
for result_item in self.parse_line_delta[lines]:
166
yield result_item
483
return iter(self.parse_line_delta(lines))
167
484
168
def parse_line_delta(self, lines, version):
485
def parse_line_delta(self, lines, version_id, plain=False):
169
486
"""Convert a line based delta into internal representation.
170
487
171
488
line delta is in the form of:
174
491
revid(utf8) newline\n
175
492
internal representation is
176
493
(start, end, count, [1..count tuples (revid, newline)])
494
495
:param plain: If True, the lines are returned as a plain
496
list without annotations, not as a list of (origin, content) tuples, i.e.
497
(start, end, count, [1..count newline])
177
498
"""
178
499
result = []
179
500
lines = iter(lines)
180
501
next = lines.next
502
503
cache = {}
504
def cache_and_return(line):
505
origin, text = line.split(' ', 1)
506
return cache.setdefault(origin, origin), text
507
181
508
# walk through the lines parsing.
509
# Note that the plain test is explicitly pulled out of the
510
# loop to minimise any performance impact
511
if plain:
512
for header in lines:
513
start, end, count = [int(n) for n in header.split(',')]
514
contents = [next().split(' ', 1)[1] for i in xrange(count)]
515
result.append((start, end, count, contents))
516
else:
517
for header in lines:
518
start, end, count = [int(n) for n in header.split(',')]
519
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
520
result.append((start, end, count, contents))
521
return result
522
523
def get_fulltext_content(self, lines):
524
"""Extract just the content lines from a fulltext."""
525
return (line.split(' ', 1)[1] for line in lines)
526
527
def get_linedelta_content(self, lines):
528
"""Extract just the content from a line delta.
529
530
This doesn't return all of the extra information stored in a delta.
531
Only the actual content lines.
532
"""
533
lines = iter(lines)
534
next = lines.next
182
535
for header in lines:
183
start, end, count = [int(n) for n in header.split(',')]
184
contents = []
185
remaining = count
186
while remaining:
536
header = header.split(',')
537
count = int(header[2])
538
for i in xrange(count):
187
539
origin, text = next().split(' ', 1)
188
remaining -= 1
189
contents.append((origin.decode('utf-8'), text))
190
result.append((start, end, count, contents))
191
return result
540
yield text
192
541
193
542
def lower_fulltext(self, content):
194
543
"""convert a fulltext content record into a serializable form.
195
544
196
545
see parse_fulltext which this inverts.
197
546
"""
198
return ['%s %s' % (o.encode('utf-8'), t) for o, t in content._lines]
547
# TODO: jam 20070209 We only do the caching thing to make sure that
548
# the origin is a valid utf-8 line, eventually we could remove it
549
return ['%s %s' % (o, t) for o, t in content._lines]
199
550
200
551
def lower_line_delta(self, delta):
201
552
"""convert a delta into a serializable form.
202
553
203
554
See parse_line_delta which this inverts.
204
555
"""
556
# TODO: jam 20070209 We only do the caching thing to make sure that
557
# the origin is a valid utf-8 line, eventually we could remove it
205
558
out = []
206
559
for start, end, c, lines in delta:
207
560
out.append('%d,%d,%d\n' % (start, end, c))
208
for origin, text in lines:
209
out.append('%s %s' % (origin.encode('utf-8'), text))
561
out.extend(origin + ' ' + text
562
for origin, text in lines)
210
563
return out
211
564
565
def annotate(self, knit, version_id):
566
content = knit._get_content(version_id)
567
return content.annotate()
568
212
569
213
570
class KnitPlainFactory(_KnitFactory):
214
571
"""Factory for creating plain Content objects."""
215
572
216
573
annotated = False
217
574
218
def parse_fulltext(self, content, version):
575
def make(self, lines, version_id):
576
return PlainKnitContent(lines, version_id)
577
578
def parse_fulltext(self, content, version_id):
219
579
"""This parses an unannotated fulltext.
220
580
221
581
Note that this is not a noop - the internal representation
222
582
has (versionid, line) - its just a constant versionid.
223
583
"""
224
return self.make(content, version)
584
return self.make(content, version_id)
225
585
226
def parse_line_delta_iter(self, lines, version):
227
while lines:
228
header = lines.pop(0)
586
def parse_line_delta_iter(self, lines, version_id):
587
cur = 0
588
num_lines = len(lines)
589
while cur < num_lines:
590
header = lines[cur]
591
cur += 1
229
592
start, end, c = [int(n) for n in header.split(',')]
230
yield start, end, c, zip([version] * c, lines[:c])
231
del lines[:c]
232
233
def parse_line_delta(self, lines, version):
234
return list(self.parse_line_delta_iter(lines, version))
235
593
yield start, end, c, lines[cur:cur+c]
594
cur += c
595
596
def parse_line_delta(self, lines, version_id):
597
return list(self.parse_line_delta_iter(lines, version_id))
598
599
def get_fulltext_content(self, lines):
600
"""Extract just the content lines from a fulltext."""
601
return iter(lines)
602
603
def get_linedelta_content(self, lines):
604
"""Extract just the content from a line delta.
605
606
This doesn't return all of the extra information stored in a delta.
607
Only the actual content lines.
608
"""
609
lines = iter(lines)
610
next = lines.next
611
for header in lines:
612
header = header.split(',')
613
count = int(header[2])
614
for i in xrange(count):
615
yield next()
616
236
617
def lower_fulltext(self, content):
237
618
return content.text()
238
619
240
621
out = []
241
622
for start, end, c, lines in delta:
242
623
out.append('%d,%d,%d\n' % (start, end, c))
243
out.extend([text for origin, text in lines])
624
out.extend(lines)
244
625
return out
245
626
627
def annotate(self, knit, version_id):
628
annotator = _KnitAnnotator(knit)
629
return annotator.annotate(version_id)
630
246
631
247
632
def make_empty_knit(transport, relpath):
248
633
"""Construct a empty knit at the specified location."""
249
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
250
k._data._open_file()
634
k = make_file_knit(transport, relpath, 'w', KnitPlainFactory)
635
636
637
def make_file_knit(name, transport, file_mode=None, access_mode='w',
638
factory=None, delta=True, create=False, create_parent_dir=False,
639
delay_create=False, dir_mode=None, get_scope=None):
640
"""Factory to create a KnitVersionedFile for a .knit/.kndx file pair."""
641
if factory is None:
642
factory = KnitAnnotateFactory()
643
if get_scope is None:
644
get_scope = lambda:None
645
index = _KnitIndex(transport, name + INDEX_SUFFIX,
646
access_mode, create=create, file_mode=file_mode,
647
create_parent_dir=create_parent_dir, delay_create=delay_create,
648
dir_mode=dir_mode, get_scope=get_scope)
649
access = _KnitAccess(transport, name + DATA_SUFFIX, file_mode,
650
dir_mode, ((create and not len(index)) and delay_create),
651
create_parent_dir)
652
return KnitVersionedFile(name, transport, factory=factory,
653
create=create, delay_create=delay_create, index=index,
654
access_method=access)
655
656
657
def get_suffixes():
658
"""Return the suffixes used by file based knits."""
659
return [DATA_SUFFIX, INDEX_SUFFIX]
660
make_file_knit.get_suffixes = get_suffixes
251
661
252
662
253
663
class KnitVersionedFile(VersionedFile):
265
675
stored and retrieved.
266
676
"""
267
677
268
def __init__(self, relpath, transport, file_mode=None, access_mode=None, factory=None,
269
basis_knit=None, delta=True, create=False):
678
def __init__(self, relpath, transport, file_mode=None,
679
factory=None, delta=True, create=False, create_parent_dir=False,
680
delay_create=False, dir_mode=None, index=None, access_method=None):
270
681
"""Construct a knit at location specified by relpath.
271
682
272
683
:param create: If not True, only open an existing knit.
684
:param create_parent_dir: If True, create the parent directory if
685
creating the file fails. (This is used for stores with
686
hash-prefixes that may not exist yet)
687
:param delay_create: The calling code is aware that the knit won't
688
actually be created until the first data is stored.
689
:param index: An index to use for the knit.
273
690
"""
274
if access_mode is None:
275
access_mode = 'w'
276
super(KnitVersionedFile, self).__init__(access_mode)
277
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
278
assert not basis_knit or isinstance(basis_knit, KnitVersionedFile), \
279
type(basis_knit)
280
691
super(KnitVersionedFile, self).__init__()
281
692
self.transport = transport
282
693
self.filename = relpath
283
self.basis_knit = basis_knit
284
694
self.factory = factory or KnitAnnotateFactory()
285
self.writable = (access_mode == 'w')
286
695
self.delta = delta
287
696
288
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
289
access_mode, create=create, file_mode=file_mode)
290
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
291
access_mode, create=create and not len(self), file_mode=file_mode)
697
self._max_delta_chain = 200
698
699
if None in (access_method, index):
700
raise ValueError("No default access_method or index any more")
701
self._index = index
702
_access = access_method
703
if create and not len(self) and not delay_create:
704
_access.create()
705
self._data = _KnitData(_access)
292
706
293
707
def __repr__(self):
294
return '%s(%s)' % (self.__class__.__name__,
708
return '%s(%s)' % (self.__class__.__name__,
295
709
self.transport.abspath(self.filename))
296
710
297
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
298
"""See VersionedFile._add_delta()."""
299
self._check_add(version_id, []) # should we check the lines ?
300
self._check_versions_present(parents)
301
present_parents = []
302
ghosts = []
303
parent_texts = {}
304
for parent in parents:
305
if not self.has_version(parent):
306
ghosts.append(parent)
307
else:
308
present_parents.append(parent)
309
310
if delta_parent is None:
311
# reconstitute as full text.
312
assert len(delta) == 1 or len(delta) == 0
313
if len(delta):
314
assert delta[0][0] == 0
315
assert delta[0][1] == 0, delta[0][1]
316
return super(KnitVersionedFile, self)._add_delta(version_id,
317
parents,
318
delta_parent,
319
sha1,
320
noeol,
321
delta)
322
323
digest = sha1
324
325
options = []
326
if noeol:
327
options.append('no-eol')
328
329
if delta_parent is not None:
330
# determine the current delta chain length.
331
# To speed the extract of texts the delta chain is limited
332
# to a fixed number of deltas. This should minimize both
333
# I/O and the time spend applying deltas.
334
count = 0
335
delta_parents = [delta_parent]
336
while count < 25:
337
parent = delta_parents[0]
338
method = self._index.get_method(parent)
339
if method == 'fulltext':
340
break
341
delta_parents = self._index.get_parents(parent)
342
count = count + 1
343
if method == 'line-delta':
344
# did not find a fulltext in the delta limit.
345
# just do a normal insertion.
346
return super(KnitVersionedFile, self)._add_delta(version_id,
347
parents,
348
delta_parent,
349
sha1,
350
noeol,
351
delta)
352
353
options.append('line-delta')
354
store_lines = self.factory.lower_line_delta(delta)
355
356
where, size = self._data.add_record(version_id, digest, store_lines)
357
self._index.add_version(version_id, options, where, size, parents)
711
def _check_should_delta(self, first_parents):
712
"""Iterate back through the parent listing, looking for a fulltext.
713
714
This is used when we want to decide whether to add a delta or a new
715
fulltext. It searches for _max_delta_chain parents. When it finds a
716
fulltext parent, it sees if the total size of the deltas leading up to
717
it is large enough to indicate that we want a new full text anyway.
718
719
Return True if we should create a new delta, False if we should use a
720
full text.
721
"""
722
delta_size = 0
723
fulltext_size = None
724
delta_parents = first_parents
725
for count in xrange(self._max_delta_chain):
726
parent = delta_parents[0]
727
method = self._index.get_method(parent)
728
index, pos, size = self._index.get_position(parent)
729
if method == 'fulltext':
730
fulltext_size = size
731
break
732
delta_size += size
733
delta_parents = self._index.get_parent_map([parent])[parent]
734
else:
735
# We couldn't find a fulltext, so we must create a new one
736
return False
737
738
return fulltext_size > delta_size
739
740
def _check_write_ok(self):
741
return self._index._check_write_ok()
358
742
359
743
def _add_raw_records(self, records, data):
360
744
"""Add all the records 'records' with data pre-joined in 'data'.
365
749
the preceding records sizes.
366
750
"""
367
751
# write all the data
368
pos = self._data.add_raw_record(data)
752
raw_record_sizes = [record[3] for record in records]
753
positions = self._data.add_raw_records(raw_record_sizes, data)
369
754
index_entries = []
370
for (version_id, options, parents, size) in records:
371
index_entries.append((version_id, options, pos, size, parents))
372
pos += size
755
for (version_id, options, parents, _), access_memo in zip(
756
records, positions):
757
index_entries.append((version_id, options, access_memo, parents))
373
758
self._index.add_versions(index_entries)
374
759
375
def clear_cache(self):
376
"""Clear the data cache only."""
377
self._data.clear_cache()
378
379
760
def copy_to(self, name, transport):
380
761
"""See VersionedFile.copy_to()."""
381
762
# copy the current index to a temp index to avoid racing with local
382
763
# writes
383
transport.put(name + INDEX_SUFFIX + '.tmp', self.transport.get(self._index._filename),)
764
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
765
self.transport.get(self._index._filename))
384
766
# copy the data file
385
transport.put(name + DATA_SUFFIX, self._data._open_file())
386
# rename the copied index into place
387
transport.rename(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
388
389
def create_empty(self, name, transport, mode=None):
390
return KnitVersionedFile(name, transport, factory=self.factory, delta=self.delta, create=True)
391
392
def _fix_parents(self, version, new_parents):
393
"""Fix the parents list for version.
767
f = self._data._open_file()
768
try:
769
transport.put_file(name + DATA_SUFFIX, f)
770
finally:
771
f.close()
772
# move the copied index into place
773
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
774
775
def get_data_stream(self, required_versions):
776
"""Get a data stream for the specified versions.
777
778
Versions may be returned in any order, not necessarily the order
779
specified. They are returned in a partial order by compression
780
parent, so that the deltas can be applied as the data stream is
781
inserted; however note that compression parents will not be sent
782
unless they were specifically requested, as the client may already
783
have them.
784
785
:param required_versions: The exact set of versions to be extracted.
786
Unlike some other knit methods, this is not used to generate a
787
transitive closure, rather it is used precisely as given.
394
788
395
This is done by appending a new version to the index
396
with identical data except for the parents list.
397
the parents list must be a superset of the current
398
list.
399
"""
400
current_values = self._index._cache[version]
401
assert set(current_values[4]).difference(set(new_parents)) == set()
402
self._index.add_version(version,
403
current_values[1],
404
current_values[2],
405
current_values[3],
406
new_parents)
789
:returns: format_signature, list of (version, options, length, parents),
790
reader_callable.
791
"""
792
required_version_set = frozenset(required_versions)
793
version_index = {}
794
# list of revisions that can just be sent without waiting for their
795
# compression parent
796
ready_to_send = []
797
# map from revision to the children based on it
798
deferred = {}
799
# first, read all relevant index data, enough to sort into the right
800
# order to return
801
for version_id in required_versions:
802
options = self._index.get_options(version_id)
803
parents = self._index.get_parents_with_ghosts(version_id)
804
index_memo = self._index.get_position(version_id)
805
version_index[version_id] = (index_memo, options, parents)
806
if ('line-delta' in options
807
and parents[0] in required_version_set):
808
# must wait until the parent has been sent
809
deferred.setdefault(parents[0], []). \
810
append(version_id)
811
else:
812
# either a fulltext, or a delta whose parent the client did
813
# not ask for and presumably already has
814
ready_to_send.append(version_id)
815
# build a list of results to return, plus instructions for data to
816
# read from the file
817
copy_queue_records = []
818
temp_version_list = []
819
while ready_to_send:
820
# XXX: pushing and popping lists may be a bit inefficient
821
version_id = ready_to_send.pop(0)
822
(index_memo, options, parents) = version_index[version_id]
823
copy_queue_records.append((version_id, index_memo))
824
none, data_pos, data_size = index_memo
825
temp_version_list.append((version_id, options, data_size,
826
parents))
827
if version_id in deferred:
828
# now we can send all the children of this revision - we could
829
# put them in anywhere, but we hope that sending them soon
830
# after the fulltext will give good locality in the receiver
831
ready_to_send[:0] = deferred.pop(version_id)
832
if not (len(deferred) == 0):
833
raise AssertionError("Still have compressed child versions waiting to be sent")
834
# XXX: The stream format is such that we cannot stream it - we have to
835
# know the length of all the data a-priori.
836
raw_datum = []
837
result_version_list = []
838
for (version_id, raw_data, _), \
839
(version_id2, options, _, parents) in \
840
izip(self._data.read_records_iter_raw(copy_queue_records),
841
temp_version_list):
842
if not (version_id == version_id2):
843
raise AssertionError('logic error, inconsistent results')
844
raw_datum.append(raw_data)
845
result_version_list.append(
846
(version_id, options, len(raw_data), parents))
847
# provide a callback to get data incrementally.
848
pseudo_file = StringIO(''.join(raw_datum))
849
def read(length):
850
if length is None:
851
return pseudo_file.read()
852
else:
853
return pseudo_file.read(length)
854
return (self.get_format_signature(), result_version_list, read)
855
856
def get_record_stream(self, versions, ordering, include_delta_closure):
857
"""Get a stream of records for versions.
858
859
:param versions: The versions to include. Each version is a tuple
860
(version,).
861
:param ordering: Either 'unordered' or 'topological'. A topologically
862
sorted stream has compression parents strictly before their
863
children.
864
:param include_delta_closure: If True then the closure across any
865
compression parents will be included (in the opaque data).
866
:return: An iterator of ContentFactory objects, each of which is only
867
valid until the iterator is advanced.
868
"""
869
if include_delta_closure:
870
# Nb: what we should do is plan the data to stream to allow
871
# reconstruction of all the texts without excessive buffering,
872
# including re-sending common bases as needed. This makes the most
873
# sense when we start serialising these streams though, so for now
874
# we just fallback to individual text construction behind the
875
# abstraction barrier.
876
knit = self
877
else:
878
knit = None
879
# We end up doing multiple index lookups here for parents details and
880
# disk layout details - we need a unified api ?
881
parent_map = self.get_parent_map(versions)
882
absent_versions = set(versions) - set(parent_map)
883
if ordering == 'topological':
884
present_versions = topo_sort(parent_map)
885
else:
886
# List comprehension to keep the requested order (as that seems
887
# marginally useful, at least until we start doing IO optimising
888
# here.
889
present_versions = [version for version in versions if version in
890
parent_map]
891
position_map = self._get_components_positions(present_versions)
892
records = [(version, position_map[version][1]) for version in
893
present_versions]
894
record_map = {}
895
for version in absent_versions:
896
yield AbsentContentFactory((version,))
897
for version, raw_data, sha1 in \
898
self._data.read_records_iter_raw(records):
899
(record_details, index_memo, _) = position_map[version]
900
yield KnitContentFactory(version, parent_map[version],
901
record_details, sha1, raw_data, self.factory.annotated, knit)
902
903
def _extract_blocks(self, version_id, source, target):
904
if self._index.get_method(version_id) != 'line-delta':
905
return None
906
parent, sha1, noeol, delta = self.get_delta(version_id)
907
return KnitContent.get_line_delta_blocks(delta, source, target)
407
908
408
909
def get_delta(self, version_id):
409
910
"""Get a delta for constructing version from some other version."""
410
if not self.has_version(version_id):
411
raise RevisionNotPresent(version_id, self.filename)
412
413
parents = self.get_parents(version_id)
911
self.check_not_reserved_id(version_id)
912
parents = self.get_parent_map([version_id])[version_id]
414
913
if len(parents):
415
914
parent = parents[0]
416
915
else:
417
916
parent = None
418
data_pos, data_size = self._index.get_position(version_id)
419
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
420
version_idx = self._index.lookup(version_id)
917
index_memo = self._index.get_position(version_id)
918
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
421
919
noeol = 'no-eol' in self._index.get_options(version_id)
422
920
if 'fulltext' == self._index.get_method(version_id):
423
new_content = self.factory.parse_fulltext(data, version_idx)
921
new_content = self.factory.parse_fulltext(data, version_id)
424
922
if parent is not None:
425
923
reference_content = self._get_content(parent)
426
924
old_texts = reference_content.text()
427
925
else:
428
926
old_texts = []
429
927
new_texts = new_content.text()
430
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
928
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
929
new_texts)
431
930
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
432
931
else:
433
delta = self.factory.parse_line_delta(data, version_idx)
932
delta = self.factory.parse_line_delta(data, version_id)
434
933
return parent, sha1, noeol, delta
435
436
def get_graph_with_ghosts(self):
437
"""See VersionedFile.get_graph_with_ghosts()."""
438
graph_items = self._index.get_graph()
439
return dict(graph_items)
440
441
def get_sha1(self, version_id):
442
"""See VersionedFile.get_sha1()."""
443
components = self._get_components(version_id)
444
return components[-1][-1][-1]
445
446
@staticmethod
447
def get_suffixes():
448
"""See VersionedFile.get_suffixes()."""
449
return [DATA_SUFFIX, INDEX_SUFFIX]
450
451
def has_ghost(self, version_id):
452
"""True if there is a ghost reference in the file to version_id."""
453
# maybe we have it
454
if self.has_version(version_id):
455
return False
456
# optimisable if needed by memoising the _ghosts set.
457
items = self._index.get_graph()
458
for node, parents in items:
459
for parent in parents:
460
if parent not in self._index._cache:
461
if parent == version_id:
462
return True
463
return False
934
935
def get_format_signature(self):
936
"""See VersionedFile.get_format_signature()."""
937
if self.factory.annotated:
938
annotated_part = "annotated"
939
else:
940
annotated_part = "plain"
941
return "knit-%s" % (annotated_part,)
942
943
def get_sha1s(self, version_ids):
944
"""See VersionedFile.get_sha1s()."""
945
record_map = self._get_record_map(version_ids)
946
# record entry 2 is the 'digest'.
947
return [record_map[v][2] for v in version_ids]
948
949
def insert_data_stream(self, (format, data_list, reader_callable)):
950
"""Insert knit records from a data stream into this knit.
951
952
If a version in the stream is already present in this knit, it will not
953
be inserted a second time. It will be checked for consistency with the
954
stored version however, and may cause a KnitCorrupt error to be raised
955
if the data in the stream disagrees with the already stored data.
956
957
:seealso: get_data_stream
958
"""
959
if format != self.get_format_signature():
960
if 'knit' in debug.debug_flags:
961
trace.mutter(
962
'incompatible format signature inserting to %r', self)
963
source = self._knit_from_datastream(
964
(format, data_list, reader_callable))
965
stream = source.get_record_stream(source.versions(), 'unordered', False)
966
self.insert_record_stream(stream)
967
return
968
969
for version_id, options, length, parents in data_list:
970
if self.has_version(version_id):
971
# First check: the list of parents.
972
my_parents = self.get_parents_with_ghosts(version_id)
973
if tuple(my_parents) != tuple(parents):
974
# XXX: KnitCorrupt is not quite the right exception here.
975
raise KnitCorrupt(
976
self.filename,
977
'parents list %r from data stream does not match '
978
'already recorded parents %r for %s'
979
% (parents, my_parents, version_id))
980
981
# Also check the SHA-1 of the fulltext this content will
982
# produce.
983
raw_data = reader_callable(length)
984
my_fulltext_sha1 = self.get_sha1s([version_id])[0]
985
df, rec = self._data._parse_record_header(version_id, raw_data)
986
stream_fulltext_sha1 = rec[3]
987
if my_fulltext_sha1 != stream_fulltext_sha1:
988
# Actually, we don't know if it's this knit that's corrupt,
989
# or the data stream we're trying to insert.
990
raise KnitCorrupt(
991
self.filename, 'sha-1 does not match %s' % version_id)
992
else:
993
if 'line-delta' in options:
994
# Make sure that this knit record is actually useful: a
995
# line-delta is no use unless we have its parent.
996
# Fetching from a broken repository with this problem
997
# shouldn't break the target repository.
998
#
999
# See https://bugs.launchpad.net/bzr/+bug/164443
1000
if not self._index.has_version(parents[0]):
1001
raise KnitCorrupt(
1002
self.filename,
1003
'line-delta from stream '
1004
'for version %s '
1005
'references '
1006
'missing parent %s\n'
1007
'Try running "bzr check" '
1008
'on the source repository, and "bzr reconcile" '
1009
'if necessary.' %
1010
(version_id, parents[0]))
1011
if not self.delta:
1012
# We received a line-delta record for a non-delta knit.
1013
# Convert it to a fulltext.
1014
gzip_bytes = reader_callable(length)
1015
self._convert_line_delta_to_fulltext(
1016
gzip_bytes, version_id, parents)
1017
continue
1018
1019
self._add_raw_records(
1020
[(version_id, options, parents, length)],
1021
reader_callable(length))
1022
1023
def _convert_line_delta_to_fulltext(self, gzip_bytes, version_id, parents):
1024
lines, sha1 = self._data._parse_record(version_id, gzip_bytes)
1025
delta = self.factory.parse_line_delta(lines, version_id)
1026
content = self.factory.make(self.get_lines(parents[0]), parents[0])
1027
content.apply_delta(delta, version_id)
1028
digest, len, content = self.add_lines(
1029
version_id, parents, content.text())
1030
if digest != sha1:
1031
raise errors.VersionedFileInvalidChecksum(version_id)
1032
1033
def _knit_from_datastream(self, (format, data_list, reader_callable)):
1034
"""Create a knit object from a data stream.
1035
1036
This method exists to allow conversion of data streams that do not
1037
match the signature of this knit. Generally it will be slower and use
1038
more memory to use this method to insert data, but it will work.
1039
1040
:seealso: get_data_stream for details on datastreams.
1041
:return: A knit versioned file which can be used to join the datastream
1042
into self.
1043
"""
1044
if format == "knit-plain":
1045
factory = KnitPlainFactory()
1046
elif format == "knit-annotated":
1047
factory = KnitAnnotateFactory()
1048
else:
1049
raise errors.KnitDataStreamUnknown(format)
1050
index = _StreamIndex(data_list, self._index)
1051
access = _StreamAccess(reader_callable, index, self, factory)
1052
return KnitVersionedFile(self.filename, self.transport,
1053
factory=factory, index=index, access_method=access)
1054
1055
def insert_record_stream(self, stream):
1056
"""Insert a record stream into this versioned file.
1057
1058
:param stream: A stream of records to insert.
1059
:return: None
1060
:seealso VersionedFile.get_record_stream:
1061
"""
1062
def get_adapter(adapter_key):
1063
try:
1064
return adapters[adapter_key]
1065
except KeyError:
1066
adapter_factory = adapter_registry.get(adapter_key)
1067
adapter = adapter_factory(self)
1068
adapters[adapter_key] = adapter
1069
return adapter
1070
if self.factory.annotated:
1071
# self is annotated, we need annotated knits to use directly.
1072
annotated = "annotated-"
1073
convertibles = []
1074
else:
1075
# self is not annotated, but we can strip annotations cheaply.
1076
annotated = ""
1077
convertibles = set(["knit-annotated-delta-gz",
1078
"knit-annotated-ft-gz"])
1079
# The set of types we can cheaply adapt without needing basis texts.
1080
native_types = set()
1081
native_types.add("knit-%sdelta-gz" % annotated)
1082
native_types.add("knit-%sft-gz" % annotated)
1083
knit_types = native_types.union(convertibles)
1084
adapters = {}
1085
# Buffer all index entries that we can't add immediately because their
1086
# basis parent is missing. We don't buffer all because generating
1087
# annotations may require access to some of the new records. However we
1088
# can't generate annotations from new deltas until their basis parent
1089
# is present anyway, so we get away with not needing an index that
1090
# includes the new keys.
1091
# key = basis_parent, value = index entry to add
1092
buffered_index_entries = {}
1093
for record in stream:
1094
# Raise an error when a record is missing.
1095
if record.storage_kind == 'absent':
1096
raise RevisionNotPresent([record.key[0]], self)
1097
# adapt to non-tuple interface
1098
parents = [parent[0] for parent in record.parents]
1099
if record.storage_kind in knit_types:
1100
if record.storage_kind not in native_types:
1101
try:
1102
adapter_key = (record.storage_kind, "knit-delta-gz")
1103
adapter = get_adapter(adapter_key)
1104
except KeyError:
1105
adapter_key = (record.storage_kind, "knit-ft-gz")
1106
adapter = get_adapter(adapter_key)
1107
bytes = adapter.get_bytes(
1108
record, record.get_bytes_as(record.storage_kind))
1109
else:
1110
bytes = record.get_bytes_as(record.storage_kind)
1111
options = [record._build_details[0]]
1112
if record._build_details[1]:
1113
options.append('no-eol')
1114
# Just blat it across.
1115
# Note: This does end up adding data on duplicate keys. As
1116
# modern repositories use atomic insertions this should not
1117
# lead to excessive growth in the event of interrupted fetches.
1118
# 'knit' repositories may suffer excessive growth, but as a
1119
# deprecated format this is tolerable. It can be fixed if
1120
# needed by in the kndx index support raising on a duplicate
1121
# add with identical parents and options.
1122
access_memo = self._data.add_raw_records([len(bytes)], bytes)[0]
1123
index_entry = (record.key[0], options, access_memo, parents)
1124
buffered = False
1125
if 'fulltext' not in options:
1126
basis_parent = parents[0]
1127
if not self.has_version(basis_parent):
1128
pending = buffered_index_entries.setdefault(
1129
basis_parent, [])
1130
pending.append(index_entry)
1131
buffered = True
1132
if not buffered:
1133
self._index.add_versions([index_entry])
1134
elif record.storage_kind == 'fulltext':
1135
self.add_lines(record.key[0], parents,
1136
split_lines(record.get_bytes_as('fulltext')))
1137
else:
1138
adapter_key = record.storage_kind, 'fulltext'
1139
adapter = get_adapter(adapter_key)
1140
lines = split_lines(adapter.get_bytes(
1141
record, record.get_bytes_as(record.storage_kind)))
1142
try:
1143
self.add_lines(record.key[0], parents, lines)
1144
except errors.RevisionAlreadyPresent:
1145
pass
1146
# Add any records whose basis parent is now available.
1147
added_keys = [record.key[0]]
1148
while added_keys:
1149
key = added_keys.pop(0)
1150
if key in buffered_index_entries:
1151
index_entries = buffered_index_entries[key]
1152
self._index.add_versions(index_entries)
1153
added_keys.extend(
1154
[index_entry[0] for index_entry in index_entries])
1155
del buffered_index_entries[key]
1156
# If there were any deltas which had a missing basis parent, error.
1157
if buffered_index_entries:
1158
raise errors.RevisionNotPresent(buffered_index_entries.keys()[0],
1159
self)
464
1160
465
1161
def versions(self):
466
1162
"""See VersionedFile.versions."""
1163
if 'evil' in debug.debug_flags:
1164
trace.mutter_callsite(2, "versions scales with size of history")
467
1165
return self._index.get_versions()
468
1166
469
1167
def has_version(self, version_id):
470
1168
"""See VersionedFile.has_version."""
1169
if 'evil' in debug.debug_flags:
1170
trace.mutter_callsite(2, "has_version is a LBYL scenario")
471
1171
return self._index.has_version(version_id)
472
1172
473
1173
__contains__ = has_version
474
1174
475
1175
def _merge_annotations(self, content, parents, parent_texts={},
476
delta=None, annotated=None):
1176
delta=None, annotated=None,
1177
left_matching_blocks=None):
477
1178
"""Merge annotations for content. This is done by comparing
478
1179
the annotations based on changed to the text.
479
1180
"""
480
if annotated:
1181
if left_matching_blocks is not None:
1182
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1183
else:
481
1184
delta_seq = None
1185
if annotated:
482
1186
for parent_id in parents:
483
1187
merge_content = self._get_content(parent_id, parent_texts)
484
seq = KnitSequenceMatcher(None, merge_content.text(), content.text())
485
if delta_seq is None:
486
# setup a delta seq to reuse.
487
delta_seq = seq
1188
if (parent_id == parents[0] and delta_seq is not None):
1189
seq = delta_seq
1190
else:
1191
seq = patiencediff.PatienceSequenceMatcher(
1192
None, merge_content.text(), content.text())
488
1193
for i, j, n in seq.get_matching_blocks():
489
1194
if n == 0:
490
1195
continue
491
# this appears to copy (origin, text) pairs across to the new
492
# content for any line that matches the last-checked parent.
493
# FIXME: save the sequence control data for delta compression
494
# against the most relevant parent rather than rediffing.
1196
# this appears to copy (origin, text) pairs across to the
1197
# new content for any line that matches the last-checked
1198
# parent.
495
1199
content._lines[j:j+n] = merge_content._lines[i:i+n]
496
1200
if delta:
497
if not annotated:
1201
if delta_seq is None:
498
1202
reference_content = self._get_content(parents[0], parent_texts)
499
1203
new_texts = content.text()
500
1204
old_texts = reference_content.text()
501
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
1205
delta_seq = patiencediff.PatienceSequenceMatcher(
1206
None, old_texts, new_texts)
502
1207
return self._make_line_delta(delta_seq, content)
503
1208
504
1209
def _make_line_delta(self, delta_seq, new_content):
510
1215
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
511
1216
return diff_hunks
512
1217
513
def _get_components(self, version_id):
514
"""Return a list of (version_id, method, data) tuples that
515
makes up version specified by version_id of the knit.
516
517
The components should be applied in the order of the returned
518
list.
519
520
The basis knit will be used to the largest extent possible
521
since it is assumed that accesses to it is faster.
1218
def _get_components_positions(self, version_ids):
1219
"""Produce a map of position data for the components of versions.
1220
1221
This data is intended to be used for retrieving the knit records.
1222
1223
A dict of version_id to (record_details, index_memo, next, parents) is
1224
returned.
1225
method is the way referenced data should be applied.
1226
index_memo is the handle to pass to the data access to actually get the
1227
data
1228
next is the build-parent of the version, or None for fulltexts.
1229
parents is the version_ids of the parents of this version
522
1230
"""
523
#profile notes:
524
# 4168 calls in 14912, 2289 internal
525
# 4168 in 9711 to read_records
526
# 52554 in 1250 to get_parents
527
# 170166 in 865 to list.append
528
529
# needed_revisions holds a list of (method, version_id) of
530
# versions that is needed to be fetched to construct the final
531
# version of the file.
532
#
533
# basis_revisions is a list of versions that needs to be
534
# fetched but exists in the basis knit.
535
536
basis = self.basis_knit
537
needed_versions = []
538
basis_versions = []
539
cursor = version_id
540
541
while 1:
542
picked_knit = self
543
if basis and basis._index.has_version(cursor):
544
picked_knit = basis
545
basis_versions.append(cursor)
546
method = picked_knit._index.get_method(cursor)
547
needed_versions.append((method, cursor))
548
if method == 'fulltext':
549
break
550
cursor = picked_knit.get_parents(cursor)[0]
551
552
components = {}
553
if basis_versions:
554
records = []
555
for comp_id in basis_versions:
556
data_pos, data_size = basis._index.get_data_position(comp_id)
557
records.append((piece_id, data_pos, data_size))
558
components.update(basis._data.read_records(records))
559
560
records = []
561
for comp_id in [vid for method, vid in needed_versions
562
if vid not in basis_versions]:
563
data_pos, data_size = self._index.get_position(comp_id)
564
records.append((comp_id, data_pos, data_size))
565
components.update(self._data.read_records(records))
566
567
# get_data_records returns a mapping with the version id as
568
# index and the value as data. The order the components need
569
# to be applied is held by needed_versions (reversed).
570
out = []
571
for method, comp_id in reversed(needed_versions):
572
out.append((comp_id, method, components[comp_id]))
573
574
return out
575
1231
component_data = {}
1232
pending_components = version_ids
1233
while pending_components:
1234
build_details = self._index.get_build_details(pending_components)
1235
current_components = set(pending_components)
1236
pending_components = set()
1237
for version_id, details in build_details.iteritems():
1238
(index_memo, compression_parent, parents,
1239
record_details) = details
1240
method = record_details[0]
1241
if compression_parent is not None:
1242
pending_components.add(compression_parent)
1243
component_data[version_id] = (record_details, index_memo,
1244
compression_parent)
1245
missing = current_components.difference(build_details)
1246
if missing:
1247
raise errors.RevisionNotPresent(missing.pop(), self.filename)
1248
return component_data
1249
576
1250
def _get_content(self, version_id, parent_texts={}):
577
1251
"""Returns a content object that makes up the specified
578
1252
version."""
579
if not self.has_version(version_id):
580
raise RevisionNotPresent(version_id, self.filename)
581
582
1253
cached_version = parent_texts.get(version_id, None)
583
1254
if cached_version is not None:
1255
if not self.has_version(version_id):
1256
raise RevisionNotPresent(version_id, self.filename)
584
1257
return cached_version
585
1258
586
if self.basis_knit and version_id in self.basis_knit:
587
return self.basis_knit._get_content(version_id)
588
589
content = None
590
components = self._get_components(version_id)
591
for component_id, method, (data, digest) in components:
592
version_idx = self._index.lookup(component_id)
593
if method == 'fulltext':
594
assert content is None
595
content = self.factory.parse_fulltext(data, version_idx)
596
elif method == 'line-delta':
597
delta = self.factory.parse_line_delta(data, version_idx)
598
content._lines = self._apply_delta(content._lines, delta)
599
600
if 'no-eol' in self._index.get_options(version_id):
601
line = content._lines[-1][1].rstrip('\n')
602
content._lines[-1] = (content._lines[-1][0], line)
603
604
# digest here is the digest from the last applied component.
605
if sha_strings(content.text()) != digest:
606
import pdb;pdb.set_trace()
607
raise KnitCorrupt(self.filename, 'sha-1 does not match %s' % version_id)
608
609
return content
1259
text_map, contents_map = self._get_content_maps([version_id])
1260
return contents_map[version_id]
610
1261
611
1262
def _check_versions_present(self, version_ids):
612
1263
"""Check that all specified versions are present."""
613
version_ids = set(version_ids)
614
for r in list(version_ids):
615
if self._index.has_version(r):
616
version_ids.remove(r)
617
if version_ids:
618
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1264
self._index.check_versions_present(version_ids)
619
1265
620
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
1266
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
1267
nostore_sha, random_id, check_content, left_matching_blocks):
621
1268
"""See VersionedFile.add_lines_with_ghosts()."""
622
self._check_add(version_id, lines)
623
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
1269
self._check_add(version_id, lines, random_id, check_content)
1270
return self._add(version_id, lines, parents, self.delta,
1271
parent_texts, left_matching_blocks, nostore_sha, random_id)
624
1272
625
def _add_lines(self, version_id, parents, lines, parent_texts):
1273
def _add_lines(self, version_id, parents, lines, parent_texts,
1274
left_matching_blocks, nostore_sha, random_id, check_content):
626
1275
"""See VersionedFile.add_lines."""
627
self._check_add(version_id, lines)
1276
self._check_add(version_id, lines, random_id, check_content)
628
1277
self._check_versions_present(parents)
629
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
1278
return self._add(version_id, lines[:], parents, self.delta,
1279
parent_texts, left_matching_blocks, nostore_sha, random_id)
630
1280
631
def _check_add(self, version_id, lines):
1281
def _check_add(self, version_id, lines, random_id, check_content):
632
1282
"""check that version_id and lines are safe to add."""
633
assert self.writable, "knit is not opened for write"
634
### FIXME escape. RBC 20060228
635
1283
if contains_whitespace(version_id):
636
1284
raise InvalidRevisionId(version_id, self.filename)
637
if self.has_version(version_id):
1285
self.check_not_reserved_id(version_id)
1286
# Technically this could be avoided if we are happy to allow duplicate
1287
# id insertion when other things than bzr core insert texts, but it
1288
# seems useful for folk using the knit api directly to have some safety
1289
# blanket that we can disable.
1290
if not random_id and self.has_version(version_id):
638
1291
raise RevisionAlreadyPresent(version_id, self.filename)
639
self._check_lines_not_unicode(lines)
640
self._check_lines_are_lines(lines)
1292
if check_content:
1293
self._check_lines_not_unicode(lines)
1294
self._check_lines_are_lines(lines)
641
1295
642
def _add(self, version_id, lines, parents, delta, parent_texts):
1296
def _add(self, version_id, lines, parents, delta, parent_texts,
1297
left_matching_blocks, nostore_sha, random_id):
643
1298
"""Add a set of lines on top of version specified by parents.
644
1299
645
1300
If delta is true, compress the text as a line-delta against
647
1302
648
1303
Any versions not present will be converted into ghosts.
649
1304
"""
650
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
651
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
652
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
653
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
654
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
655
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
656
# +1383 0 8.0370 8.0370 +<len>
657
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
658
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
659
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
660
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
1305
# first thing, if the content is something we don't need to store, find
1306
# that out.
1307
line_bytes = ''.join(lines)
1308
digest = sha_string(line_bytes)
1309
if nostore_sha == digest:
1310
raise errors.ExistingContent
661
1311
662
1312
present_parents = []
663
ghosts = []
664
1313
if parent_texts is None:
665
1314
parent_texts = {}
666
1315
for parent in parents:
667
if not self.has_version(parent):
668
ghosts.append(parent)
669
else:
1316
if self.has_version(parent):
670
1317
present_parents.append(parent)
671
1318
672
if delta and not len(present_parents):
1319
# can only compress against the left most present parent.
1320
if (delta and
1321
(len(present_parents) == 0 or
1322
present_parents[0] != parents[0])):
673
1323
delta = False
674
1324
675
digest = sha_strings(lines)
1325
text_length = len(line_bytes)
676
1326
options = []
677
1327
if lines:
678
1328
if lines[-1][-1] != '\n':
1329
# copy the contents of lines.
1330
lines = lines[:]
679
1331
options.append('no-eol')
680
1332
lines[-1] = lines[-1] + '\n'
1333
line_bytes += '\n'
681
1334
682
if len(present_parents) and delta:
1335
if delta:
683
1336
# To speed the extract of texts the delta chain is limited
684
1337
# to a fixed number of deltas. This should minimize both
685
1338
# I/O and the time spend applying deltas.
686
count = 0
687
delta_parents = present_parents
688
while count < 25:
689
parent = delta_parents[0]
690
method = self._index.get_method(parent)
691
if method == 'fulltext':
692
break
693
delta_parents = self._index.get_parents(parent)
694
count = count + 1
695
if method == 'line-delta':
696
delta = False
1339
delta = self._check_should_delta(present_parents)
697
1340
698
lines = self.factory.make(lines, version_id)
1341
content = self.factory.make(lines, version_id)
699
1342
if delta or (self.factory.annotated and len(present_parents) > 0):
700
# Merge annotations from parent texts if so is needed.
701
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
702
delta, self.factory.annotated)
1343
# Merge annotations from parent texts if needed.
1344
delta_hunks = self._merge_annotations(content, present_parents,
1345
parent_texts, delta, self.factory.annotated,
1346
left_matching_blocks)
703
1347
704
1348
if delta:
705
1349
options.append('line-delta')
706
1350
store_lines = self.factory.lower_line_delta(delta_hunks)
1351
size, bytes = self._data._record_to_data(version_id, digest,
1352
store_lines)
707
1353
else:
708
1354
options.append('fulltext')
709
store_lines = self.factory.lower_fulltext(lines)
1355
# isinstance is slower and we have no hierarchy.
1356
if self.factory.__class__ == KnitPlainFactory:
1357
# Use the already joined bytes saving iteration time in
1358
# _record_to_data.
1359
size, bytes = self._data._record_to_data(version_id, digest,
1360
lines, [line_bytes])
1361
else:
1362
# get mixed annotation + content and feed it into the
1363
# serialiser.
1364
store_lines = self.factory.lower_fulltext(content)
1365
size, bytes = self._data._record_to_data(version_id, digest,
1366
store_lines)
710
1367
711
where, size = self._data.add_record(version_id, digest, store_lines)
712
self._index.add_version(version_id, options, where, size, parents)
713
return lines
1368
access_memo = self._data.add_raw_records([size], bytes)[0]
1369
self._index.add_versions(
1370
((version_id, options, access_memo, parents),),
1371
random_id=random_id)
1372
return digest, text_length, content
714
1373
715
1374
def check(self, progress_bar=None):
716
1375
"""See VersionedFile.check()."""
717
718
def _clone_text(self, new_version_id, old_version_id, parents):
719
"""See VersionedFile.clone_text()."""
720
# FIXME RBC 20060228 make fast by only inserting an index with null delta.
721
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
1376
# This doesn't actually test extraction of everything, but that will
1377
# impact 'bzr check' substantially, and needs to be integrated with
1378
# care. However, it does check for the obvious problem of a delta with
1379
# no basis.
1380
versions = self.versions()
1381
parent_map = self.get_parent_map(versions)
1382
for version in versions:
1383
if self._index.get_method(version) != 'fulltext':
1384
compression_parent = parent_map[version][0]
1385
if compression_parent not in parent_map:
1386
raise errors.KnitCorrupt(self,
1387
"Missing basis parent %s for %s" % (
1388
compression_parent, version))
722
1389
723
1390
def get_lines(self, version_id):
724
1391
"""See VersionedFile.get_lines()."""
725
return self._get_content(version_id).text()
726
727
def iter_lines_added_or_present_in_versions(self, version_ids=None):
1392
return self.get_line_list([version_id])[0]
1393
1394
def _get_record_map(self, version_ids):
1395
"""Produce a dictionary of knit records.
1396
1397
:return: {version_id:(record, record_details, digest, next)}
1398
record
1399
data returned from read_records
1400
record_details
1401
opaque information to pass to parse_record
1402
digest
1403
SHA1 digest of the full text after all steps are done
1404
next
1405
build-parent of the version, i.e. the leftmost ancestor.
1406
Will be None if the record is not a delta.
1407
"""
1408
position_map = self._get_components_positions(version_ids)
1409
# c = component_id, r = record_details, i_m = index_memo, n = next
1410
records = [(c, i_m) for c, (r, i_m, n)
1411
in position_map.iteritems()]
1412
record_map = {}
1413
for component_id, record, digest in \
1414
self._data.read_records_iter(records):
1415
(record_details, index_memo, next) = position_map[component_id]
1416
record_map[component_id] = record, record_details, digest, next
1417
1418
return record_map
1419
1420
def get_text(self, version_id):
1421
"""See VersionedFile.get_text"""
1422
return self.get_texts([version_id])[0]
1423
1424
def get_texts(self, version_ids):
1425
return [''.join(l) for l in self.get_line_list(version_ids)]
1426
1427
def get_line_list(self, version_ids):
1428
"""Return the texts of listed versions as a list of strings."""
1429
for version_id in version_ids:
1430
self.check_not_reserved_id(version_id)
1431
text_map, content_map = self._get_content_maps(version_ids)
1432
return [text_map[v] for v in version_ids]
1433
1434
_get_lf_split_line_list = get_line_list
1435
1436
def _get_content_maps(self, version_ids):
1437
"""Produce maps of text and KnitContents
1438
1439
:return: (text_map, content_map) where text_map contains the texts for
1440
the requested versions and content_map contains the KnitContents.
1441
Both dicts take version_ids as their keys.
1442
"""
1443
# FUTURE: This function could be improved for the 'extract many' case
1444
# by tracking each component and only doing the copy when the number of
1445
# children than need to apply delta's to it is > 1 or it is part of the
1446
# final output.
1447
version_ids = list(version_ids)
1448
multiple_versions = len(version_ids) != 1
1449
record_map = self._get_record_map(version_ids)
1450
1451
text_map = {}
1452
content_map = {}
1453
final_content = {}
1454
for version_id in version_ids:
1455
components = []
1456
cursor = version_id
1457
while cursor is not None:
1458
record, record_details, digest, next = record_map[cursor]
1459
components.append((cursor, record, record_details, digest))
1460
if cursor in content_map:
1461
break
1462
cursor = next
1463
1464
content = None
1465
for (component_id, record, record_details,
1466
digest) in reversed(components):
1467
if component_id in content_map:
1468
content = content_map[component_id]
1469
else:
1470
content, delta = self.factory.parse_record(version_id,
1471
record, record_details, content,
1472
copy_base_content=multiple_versions)
1473
if multiple_versions:
1474
content_map[component_id] = content
1475
1476
content.cleanup_eol(copy_on_mutate=multiple_versions)
1477
final_content[version_id] = content
1478
1479
# digest here is the digest from the last applied component.
1480
text = content.text()
1481
actual_sha = sha_strings(text)
1482
if actual_sha != digest:
1483
raise KnitCorrupt(self.filename,
1484
'\n sha-1 %s'
1485
'\n of reconstructed text does not match'
1486
'\n expected %s'
1487
'\n for version %s' %
1488
(actual_sha, digest, version_id))
1489
text_map[version_id] = text
1490
return text_map, final_content
1491
1492
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1493
pb=None):
728
1494
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
729
1495
if version_ids is None:
730
1496
version_ids = self.versions()
1497
if pb is None:
1498
pb = progress.DummyProgress()
731
1499
# we don't care about inclusions, the caller cares.
732
1500
# but we need to setup a list of records to visit.
733
1501
# we need version_id, position, length
734
1502
version_id_records = []
735
requested_versions = list(version_ids)
1503
requested_versions = set(version_ids)
736
1504
# filter for available versions
737
1505
for version_id in requested_versions:
738
1506
if not self.has_version(version_id):
739
1507
raise RevisionNotPresent(version_id, self.filename)
740
1508
# get a in-component-order queue:
741
version_ids = []
742
1509
for version_id in self.versions():
743
1510
if version_id in requested_versions:
744
version_ids.append(version_id)
745
data_pos, length = self._index.get_position(version_id)
746
version_id_records.append((version_id, data_pos, length))
1511
index_memo = self._index.get_position(version_id)
1512
version_id_records.append((version_id, index_memo))
747
1513
748
pb = bzrlib.ui.ui_factory.nested_progress_bar()
749
count = 0
750
1514
total = len(version_id_records)
751
try:
752
pb.update('Walking content.', count, total)
753
for version_id, data, sha_value in \
754
self._data.read_records_iter(version_id_records):
755
pb.update('Walking content.', count, total)
756
method = self._index.get_method(version_id)
757
version_idx = self._index.lookup(version_id)
758
assert method in ('fulltext', 'line-delta')
759
if method == 'fulltext':
760
content = self.factory.parse_fulltext(data, version_idx)
761
for line in content.text():
762
yield line
763
else:
764
delta = self.factory.parse_line_delta(data, version_idx)
765
for start, end, count, lines in delta:
766
for origin, line in lines:
767
yield line
768
count +=1
769
pb.update('Walking content.', total, total)
770
pb.finished()
771
except:
772
pb.update('Walking content.', total, total)
773
pb.finished()
774
raise
1515
for version_idx, (version_id, data, sha_value) in \
1516
enumerate(self._data.read_records_iter(version_id_records)):
1517
pb.update('Walking content.', version_idx, total)
1518
method = self._index.get_method(version_id)
1519
if method == 'fulltext':
1520
line_iterator = self.factory.get_fulltext_content(data)
1521
elif method == 'line-delta':
1522
line_iterator = self.factory.get_linedelta_content(data)
1523
else:
1524
raise ValueError('invalid method %r' % (method,))
1525
# XXX: It might be more efficient to yield (version_id,
1526
# line_iterator) in the future. However for now, this is a simpler
1527
# change to integrate into the rest of the codebase. RBC 20071110
1528
for line in line_iterator:
1529
yield line, version_id
1530
1531
pb.update('Walking content.', total, total)
775
1532
776
1533
def num_versions(self):
777
1534
"""See VersionedFile.num_versions()."""
779
1536
780
1537
__len__ = num_versions
781
1538
782
def annotate_iter(self, version_id):
783
"""See VersionedFile.annotate_iter."""
784
content = self._get_content(version_id)
785
for origin, text in content.annotate_iter():
786
yield origin, text
787
788
def get_parents(self, version_id):
789
"""See VersionedFile.get_parents."""
790
# perf notes:
791
# optimism counts!
792
# 52554 calls in 1264 872 internal down from 3674
793
try:
794
return self._index.get_parents(version_id)
795
except KeyError:
796
raise RevisionNotPresent(version_id, self.filename)
797
798
def get_parents_with_ghosts(self, version_id):
799
"""See VersionedFile.get_parents."""
800
try:
801
return self._index.get_parents_with_ghosts(version_id)
802
except KeyError:
803
raise RevisionNotPresent(version_id, self.filename)
804
805
def get_ancestry(self, versions):
1539
def annotate(self, version_id):
1540
"""See VersionedFile.annotate."""
1541
return self.factory.annotate(self, version_id)
1542
1543
def get_parent_map(self, version_ids):
1544
"""See VersionedFile.get_parent_map."""
1545
return self._index.get_parent_map(version_ids)
1546
1547
def get_ancestry(self, versions, topo_sorted=True):
806
1548
"""See VersionedFile.get_ancestry."""
807
1549
if isinstance(versions, basestring):
808
1550
versions = [versions]
809
1551
if not versions:
810
1552
return []
811
self._check_versions_present(versions)
812
return self._index.get_ancestry(versions)
1553
return self._index.get_ancestry(versions, topo_sorted)
813
1554
814
1555
def get_ancestry_with_ghosts(self, versions):
815
1556
"""See VersionedFile.get_ancestry_with_ghosts."""
817
1558
versions = [versions]
818
1559
if not versions:
819
1560
return []
820
self._check_versions_present(versions)
821
1561
return self._index.get_ancestry_with_ghosts(versions)
822
1562
823
#@deprecated_method(zero_eight)
824
def walk(self, version_ids):
825
"""See VersionedFile.walk."""
826
# We take the short path here, and extract all relevant texts
827
# and put them in a weave and let that do all the work. Far
828
# from optimal, but is much simpler.
829
# FIXME RB 20060228 this really is inefficient!
830
from bzrlib.weave import Weave
831
832
w = Weave(self.filename)
833
ancestry = self.get_ancestry(version_ids)
834
sorted_graph = topo_sort(self._index.get_graph())
835
version_list = [vid for vid in sorted_graph if vid in ancestry]
836
837
for version_id in version_list:
838
lines = self.get_lines(version_id)
839
w.add_lines(version_id, self.get_parents(version_id), lines)
840
841
for lineno, insert_id, dset, line in w.walk(version_ids):
842
yield lineno, insert_id, dset, line
843
844
1563
def plan_merge(self, ver_a, ver_b):
845
1564
"""See VersionedFile.plan_merge."""
846
ancestors_b = set(self.get_ancestry(ver_b))
847
def status_a(revision, text):
848
if revision in ancestors_b:
849
return 'killed-b', text
850
else:
851
return 'new-a', text
852
853
ancestors_a = set(self.get_ancestry(ver_a))
854
def status_b(revision, text):
855
if revision in ancestors_a:
856
return 'killed-a', text
857
else:
858
return 'new-b', text
859
1565
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1566
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
860
1567
annotated_a = self.annotate(ver_a)
861
1568
annotated_b = self.annotate(ver_b)
862
plain_a = [t for (a, t) in annotated_a]
863
plain_b = [t for (a, t) in annotated_b]
864
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
865
a_cur = 0
866
b_cur = 0
867
for ai, bi, l in blocks:
868
# process all mismatched sections
869
# (last mismatched section is handled because blocks always
870
# includes a 0-length last block)
871
for revision, text in annotated_a[a_cur:ai]:
872
yield status_a(revision, text)
873
for revision, text in annotated_b[b_cur:bi]:
874
yield status_b(revision, text)
875
876
# and now the matched section
877
a_cur = ai + l
878
b_cur = bi + l
879
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
880
assert text_a == text_b
881
yield "unchanged", text_a
1569
return merge._plan_annotate_merge(annotated_a, annotated_b,
1570
ancestors_a, ancestors_b)
882
1571
883
1572
884
1573
class _KnitComponentFile(object):
885
1574
"""One of the files used to implement a knit database"""
886
1575
887
def __init__(self, transport, filename, mode, file_mode=None):
1576
def __init__(self, transport, filename, mode, file_mode=None,
1577
create_parent_dir=False, dir_mode=None):
888
1578
self._transport = transport
889
1579
self._filename = filename
890
1580
self._mode = mode
891
self._file_mode=file_mode
1581
self._file_mode = file_mode
1582
self._dir_mode = dir_mode
1583
self._create_parent_dir = create_parent_dir
1584
self._need_to_create = False
892
1585
893
def write_header(self):
894
if self._transport.append(self._filename, StringIO(self.HEADER),
895
mode=self._file_mode):
896
raise KnitCorrupt(self._filename, 'misaligned after writing header')
1586
def _full_path(self):
1587
"""Return the full path to this file."""
1588
return self._transport.base + self._filename
897
1589
898
1590
def check_header(self, fp):
899
1591
line = fp.readline()
1592
if line == '':
1593
# An empty file can actually be treated as though the file doesn't
1594
# exist yet.
1595
raise errors.NoSuchFile(self._full_path())
900
1596
if line != self.HEADER:
901
raise KnitHeaderError(badline=line)
902
903
def commit(self):
904
"""Commit is a nop."""
1597
raise KnitHeaderError(badline=line,
1598
filename=self._transport.abspath(self._filename))
905
1599
906
1600
def __repr__(self):
907
1601
return '%s(%s)' % (self.__class__.__name__, self._filename)
949
1643
The ' :' marker is the end of record marker.
950
1644
951
1645
partial writes:
952
when a write is interrupted to the index file, it will result in a line that
953
does not end in ' :'. If the ' :' is not present at the end of a line, or at
954
the end of the file, then the record that is missing it will be ignored by
955
the parser.
1646
when a write is interrupted to the index file, it will result in a line
1647
that does not end in ' :'. If the ' :' is not present at the end of a line,
1648
or at the end of the file, then the record that is missing it will be
1649
ignored by the parser.
956
1650
957
1651
When writing new records to the index file, the data is preceded by '\n'
958
1652
to ensure that records always start on new lines even if the last write was
967
1661
968
1662
def _cache_version(self, version_id, options, pos, size, parents):
969
1663
"""Cache a version record in the history array and index cache.
970
971
This is inlined into __init__ for performance. KEEP IN SYNC.
1664
1665
This is inlined into _load_data for performance. KEEP IN SYNC.
972
1666
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
973
1667
indexes).
974
1668
"""
979
1673
self._history.append(version_id)
980
1674
else:
981
1675
index = self._cache[version_id][5]
982
self._cache[version_id] = (version_id,
1676
self._cache[version_id] = (version_id,
983
1677
options,
984
1678
pos,
985
1679
size,
986
1680
parents,
987
1681
index)
988
1682
989
def __init__(self, transport, filename, mode, create=False, file_mode=None):
990
_KnitComponentFile.__init__(self, transport, filename, mode, file_mode)
1683
def _check_write_ok(self):
1684
if self._get_scope() != self._scope:
1685
raise errors.OutSideTransaction()
1686
if self._mode != 'w':
1687
raise errors.ReadOnlyObjectDirtiedError(self)
1688
1689
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1690
create_parent_dir=False, delay_create=False, dir_mode=None,
1691
get_scope=None):
1692
_KnitComponentFile.__init__(self, transport, filename, mode,
1693
file_mode=file_mode,
1694
create_parent_dir=create_parent_dir,
1695
dir_mode=dir_mode)
991
1696
self._cache = {}
992
1697
# position in _history is the 'official' index for a revision
993
1698
# but the values may have come from a newer entry.
994
1699
# so - wc -l of a knit index is != the number of unique names
995
# in the weave.
1700
# in the knit.
996
1701
self._history = []
997
pb = bzrlib.ui.ui_factory.nested_progress_bar()
998
1702
try:
999
count = 0
1000
total = 1
1703
fp = self._transport.get(self._filename)
1001
1704
try:
1002
pb.update('read knit index', count, total)
1003
fp = self._transport.get(self._filename)
1004
self.check_header(fp)
1005
# readlines reads the whole file at once:
1006
# bad for transports like http, good for local disk
1007
# we save 60 ms doing this one change (
1008
# from calling readline each time to calling
1009
# readlines once.
1010
# probably what we want for nice behaviour on
1011
# http is a incremental readlines that yields, or
1012
# a check for local vs non local indexes,
1013
for l in fp.readlines():
1014
rec = l.split()
1015
if len(rec) < 5 or rec[-1] != ':':
1016
# corrupt line.
1017
# FIXME: in the future we should determine if its a
1018
# short write - and ignore it
1019
# or a different failure, and raise. RBC 20060407
1020
continue
1021
count += 1
1022
total += 1
1023
#pb.update('read knit index', count, total)
1024
# See self._parse_parents
1025
parents = []
1026
for value in rec[4:-1]:
1027
if '.' == value[0]:
1028
# uncompressed reference
1029
parents.append(value[1:])
1030
else:
1031
# this is 15/4000ms faster than isinstance,
1032
# (in lsprof)
1033
# this function is called thousands of times a
1034
# second so small variations add up.
1035
assert value.__class__ is str
1036
parents.append(self._history[int(value)])
1037
# end self._parse_parents
1038
# self._cache_version(rec[0],
1039
# rec[1].split(','),
1040
# int(rec[2]),
1041
# int(rec[3]),
1042
# parents)
1043
# --- self._cache_version
1044
# only want the _history index to reference the 1st
1045
# index entry for version_id
1046
version_id = rec[0]
1047
if version_id not in self._cache:
1048
index = len(self._history)
1049
self._history.append(version_id)
1050
else:
1051
index = self._cache[version_id][5]
1052
self._cache[version_id] = (version_id,
1053
rec[1].split(','),
1054
int(rec[2]),
1055
int(rec[3]),
1056
parents,
1057
index)
1058
# --- self._cache_version
1059
except NoSuchFile, e:
1060
if mode != 'w' or not create:
1061
raise
1062
self.write_header()
1063
finally:
1064
pb.update('read knit index', total, total)
1065
pb.finished()
1066
1067
def _parse_parents(self, compressed_parents):
1068
"""convert a list of string parent values into version ids.
1069
1070
ints are looked up in the index.
1071
.FOO values are ghosts and converted in to FOO.
1072
1073
NOTE: the function is retained here for clarity, and for possible
1074
use in partial index reads. However bulk processing now has
1075
it inlined in __init__ for inner-loop optimisation.
1076
"""
1077
result = []
1078
for value in compressed_parents:
1079
if value[-1] == '.':
1080
# uncompressed reference
1081
result.append(value[1:])
1705
# _load_data may raise NoSuchFile if the target knit is
1706
# completely empty.
1707
_load_data(self, fp)
1708
finally:
1709
fp.close()
1710
except NoSuchFile:
1711
if mode != 'w' or not create:
1712
raise
1713
elif delay_create:
1714
self._need_to_create = True
1082
1715
else:
1083
# this is 15/4000ms faster than isinstance,
1084
# this function is called thousands of times a
1085
# second so small variations add up.
1086
assert value.__class__ is str
1087
result.append(self._history[int(value)])
1088
return result
1089
1090
def get_graph(self):
1091
graph = []
1092
for version_id, index in self._cache.iteritems():
1093
graph.append((version_id, index[4]))
1094
return graph
1095
1096
def get_ancestry(self, versions):
1716
self._transport.put_bytes_non_atomic(
1717
self._filename, self.HEADER, mode=self._file_mode)
1718
self._scope = get_scope()
1719
self._get_scope = get_scope
1720
1721
def get_ancestry(self, versions, topo_sorted=True):
1097
1722
"""See VersionedFile.get_ancestry."""
1098
1723
# get a graph of all the mentioned versions:
1099
1724
graph = {}
1100
1725
pending = set(versions)
1101
while len(pending):
1726
cache = self._cache
1727
while pending:
1102
1728
version = pending.pop()
1103
parents = self._cache[version][4]
1104
# got the parents ok
1105
1729
# trim ghosts
1106
parents = [parent for parent in parents if parent in self._cache]
1107
for parent in parents:
1108
# if not completed and not a ghost
1109
if parent not in graph:
1110
pending.add(parent)
1730
try:
1731
parents = [p for p in cache[version][4] if p in cache]
1732
except KeyError:
1733
raise RevisionNotPresent(version, self._filename)
1734
# if not completed and not a ghost
1735
pending.update([p for p in parents if p not in graph])
1111
1736
graph[version] = parents
1737
if not topo_sorted:
1738
return graph.keys()
1112
1739
return topo_sort(graph.items())
1113
1740
1114
1741
def get_ancestry_with_ghosts(self, versions):
1115
1742
"""See VersionedFile.get_ancestry_with_ghosts."""
1116
1743
# get a graph of all the mentioned versions:
1744
self.check_versions_present(versions)
1745
cache = self._cache
1117
1746
graph = {}
1118
1747
pending = set(versions)
1119
while len(pending):
1748
while pending:
1120
1749
version = pending.pop()
1121
1750
try:
1122
parents = self._cache[version][4]
1751
parents = cache[version][4]
1123
1752
except KeyError:
1124
1753
# ghost, fake it
1125
1754
graph[version] = []
1126
pass
1127
1755
else:
1128
# got the parents ok
1129
for parent in parents:
1130
if parent not in graph:
1131
pending.add(parent)
1756
# if not completed
1757
pending.update([p for p in parents if p not in graph])
1132
1758
graph[version] = parents
1133
1759
return topo_sort(graph.items())
1134
1760
1761
def get_build_details(self, version_ids):
1762
"""Get the method, index_memo and compression parent for version_ids.
1763
1764
Ghosts are omitted from the result.
1765
1766
:param version_ids: An iterable of version_ids.
1767
:return: A dict of version_id:(index_memo, compression_parent,
1768
parents, record_details).
1769
index_memo
1770
opaque structure to pass to read_records to extract the raw
1771
data
1772
compression_parent
1773
Content that this record is built upon, may be None
1774
parents
1775
Logical parents of this node
1776
record_details
1777
extra information about the content which needs to be passed to
1778
Factory.parse_record
1779
"""
1780
result = {}
1781
for version_id in version_ids:
1782
if version_id not in self._cache:
1783
# ghosts are omitted
1784
continue
1785
method = self.get_method(version_id)
1786
parents = self.get_parents_with_ghosts(version_id)
1787
if method == 'fulltext':
1788
compression_parent = None
1789
else:
1790
compression_parent = parents[0]
1791
noeol = 'no-eol' in self.get_options(version_id)
1792
index_memo = self.get_position(version_id)
1793
result[version_id] = (index_memo, compression_parent,
1794
parents, (method, noeol))
1795
return result
1796
1135
1797
def num_versions(self):
1136
1798
return len(self._history)
1137
1799
1138
1800
__len__ = num_versions
1139
1801
1140
1802
def get_versions(self):
1803
"""Get all the versions in the file. not topologically sorted."""
1141
1804
return self._history
1142
1805
1143
def idx_to_name(self, idx):
1144
return self._history[idx]
1145
1146
def lookup(self, version_id):
1147
assert version_id in self._cache
1148
return self._cache[version_id][5]
1149
1150
1806
def _version_list_to_index(self, versions):
1151
1807
result_list = []
1808
cache = self._cache
1152
1809
for version in versions:
1153
if version in self._cache:
1810
if version in cache:
1154
1811
# -- inlined lookup() --
1155
result_list.append(str(self._cache[version][5]))
1812
result_list.append(str(cache[version][5]))
1156
1813
# -- end lookup () --
1157
1814
else:
1158
result_list.append('.' + version.encode('utf-8'))
1815
result_list.append('.' + version)
1159
1816
return ' '.join(result_list)
1160
1817
1161
def add_version(self, version_id, options, pos, size, parents):
1818
def add_version(self, version_id, options, index_memo, parents):
1162
1819
"""Add a version record to the index."""
1163
self.add_versions(((version_id, options, pos, size, parents),))
1820
self.add_versions(((version_id, options, index_memo, parents),))
1164
1821
1165
def add_versions(self, versions):
1822
def add_versions(self, versions, random_id=False):
1166
1823
"""Add multiple versions to the index.
1167
1824
1168
1825
:param versions: a list of tuples:
1169
1826
(version_id, options, pos, size, parents).
1827
:param random_id: If True the ids being added were randomly generated
1828
and no check for existence will be performed.
1170
1829
"""
1171
1830
lines = []
1172
for version_id, options, pos, size, parents in versions:
1173
line = "\n%s %s %s %s %s :" % (version_id.encode('utf-8'),
1174
','.join(options),
1175
pos,
1176
size,
1177
self._version_list_to_index(parents))
1178
assert isinstance(line, str), \
1179
'content must be utf-8 encoded: %r' % (line,)
1180
lines.append(line)
1181
self._transport.append(self._filename, StringIO(''.join(lines)))
1182
# cache after writing, so that a failed write leads to missing cache
1183
# entries not extra ones. XXX TODO: RBC 20060502 in the event of a
1184
# failure, reload the index or flush it or some such, to prevent
1185
# writing records that did complete twice.
1186
for version_id, options, pos, size, parents in versions:
1187
self._cache_version(version_id, options, pos, size, parents)
1188
1831
orig_history = self._history[:]
1832
orig_cache = self._cache.copy()
1833
1834
try:
1835
for version_id, options, (index, pos, size), parents in versions:
1836
line = "\n%s %s %s %s %s :" % (version_id,
1837
','.join(options),
1838
pos,
1839
size,
1840
self._version_list_to_index(parents))
1841
lines.append(line)
1842
self._cache_version(version_id, options, pos, size, tuple(parents))
1843
if not self._need_to_create:
1844
self._transport.append_bytes(self._filename, ''.join(lines))
1845
else:
1846
sio = StringIO()
1847
sio.write(self.HEADER)
1848
sio.writelines(lines)
1849
sio.seek(0)
1850
self._transport.put_file_non_atomic(self._filename, sio,
1851
create_parent_dir=self._create_parent_dir,
1852
mode=self._file_mode,
1853
dir_mode=self._dir_mode)
1854
self._need_to_create = False
1855
except:
1856
# If any problems happen, restore the original values and re-raise
1857
self._history = orig_history
1858
self._cache = orig_cache
1859
raise
1860
1189
1861
def has_version(self, version_id):
1190
1862
"""True if the version is in the index."""
1191
return self._cache.has_key(version_id)
1863
return version_id in self._cache
1192
1864
1193
1865
def get_position(self, version_id):
1194
"""Return data position and size of specified version."""
1195
return (self._cache[version_id][2], \
1196
self._cache[version_id][3])
1866
"""Return details needed to access the version.
1867
1868
.kndx indices do not support split-out data, so return None for the
1869
index field.
1870
1871
:return: a tuple (None, data position, size) to hand to the access
1872
logic to get the record.
1873
"""
1874
entry = self._cache[version_id]
1875
return None, entry[2], entry[3]
1197
1876
1198
1877
def get_method(self, version_id):
1199
1878
"""Return compression method of specified version."""
1200
options = self._cache[version_id][1]
1879
try:
1880
options = self._cache[version_id][1]
1881
except KeyError:
1882
raise RevisionNotPresent(version_id, self._filename)
1201
1883
if 'fulltext' in options:
1202
1884
return 'fulltext'
1203
1885
else:
1204
assert 'line-delta' in options
1886
if 'line-delta' not in options:
1887
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1205
1888
return 'line-delta'
1206
1889
1207
1890
def get_options(self, version_id):
1891
"""Return a list representing options.
1892
1893
e.g. ['foo', 'bar']
1894
"""
1208
1895
return self._cache[version_id][1]
1209
1896
1210
def get_parents(self, version_id):
1211
"""Return parents of specified version ignoring ghosts."""
1212
return [parent for parent in self._cache[version_id][4]
1213
if parent in self._cache]
1214
1215
def get_parents_with_ghosts(self, version_id):
1216
"""Return parents of specified version with ghosts."""
1217
return self._cache[version_id][4]
1218
1219
def check_versions_present(self, version_ids):
1220
"""Check that all specified versions are present."""
1221
version_ids = set(version_ids)
1222
for version_id in list(version_ids):
1223
if version_id in self._cache:
1224
version_ids.remove(version_id)
1225
if version_ids:
1226
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1227
1228
1229
class _KnitData(_KnitComponentFile):
1230
"""Contents of the knit data file"""
1231
1232
HEADER = "# bzr knit data 8\n"
1233
1234
def __init__(self, transport, filename, mode, create=False, file_mode=None):
1235
_KnitComponentFile.__init__(self, transport, filename, mode)
1236
self._file = None
1897
def get_parent_map(self, version_ids):
1898
"""Passed through to by KnitVersionedFile.get_parent_map."""
1899
result = {}
1900
for version_id in version_ids:
1901
try:
1902
result[version_id] = tuple(self._cache[version_id][4])
1903
except KeyError:
1904
pass
1905
return result
1906
1907
def get_parents_with_ghosts(self, version_id):
1908
"""Return parents of specified version with ghosts."""
1909
try:
1910
return self.get_parent_map([version_id])[version_id]
1911
except KeyError:
1912
raise RevisionNotPresent(version_id, self)
1913
1914
def check_versions_present(self, version_ids):
1915
"""Check that all specified versions are present."""
1916
cache = self._cache
1917
for version_id in version_ids:
1918
if version_id not in cache:
1919
raise RevisionNotPresent(version_id, self._filename)
1920
1921
1922
class KnitGraphIndex(object):
1923
"""A knit index that builds on GraphIndex."""
1924
1925
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1926
"""Construct a KnitGraphIndex on a graph_index.
1927
1928
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1929
:param deltas: Allow delta-compressed records.
1930
:param add_callback: If not None, allow additions to the index and call
1931
this callback with a list of added GraphIndex nodes:
1932
[(node, value, node_refs), ...]
1933
:param parents: If True, record knits parents, if not do not record
1934
parents.
1935
"""
1936
self._graph_index = graph_index
1937
self._deltas = deltas
1938
self._add_callback = add_callback
1939
self._parents = parents
1940
if deltas and not parents:
1941
raise KnitCorrupt(self, "Cannot do delta compression without "
1942
"parent tracking.")
1943
1944
def _check_write_ok(self):
1945
pass
1946
1947
def _get_entries(self, keys, check_present=False):
1948
"""Get the entries for keys.
1949
1950
:param keys: An iterable of index keys, - 1-tuples.
1951
"""
1952
keys = set(keys)
1953
found_keys = set()
1954
if self._parents:
1955
for node in self._graph_index.iter_entries(keys):
1956
yield node
1957
found_keys.add(node[1])
1958
else:
1959
# adapt parentless index to the rest of the code.
1960
for node in self._graph_index.iter_entries(keys):
1961
yield node[0], node[1], node[2], ()
1962
found_keys.add(node[1])
1963
if check_present:
1964
missing_keys = keys.difference(found_keys)
1965
if missing_keys:
1966
raise RevisionNotPresent(missing_keys.pop(), self)
1967
1968
def _present_keys(self, version_ids):
1969
return set([
1970
node[1] for node in self._get_entries(version_ids)])
1971
1972
def _parentless_ancestry(self, versions):
1973
"""Honour the get_ancestry API for parentless knit indices."""
1974
wanted_keys = self._version_ids_to_keys(versions)
1975
present_keys = self._present_keys(wanted_keys)
1976
missing = set(wanted_keys).difference(present_keys)
1977
if missing:
1978
raise RevisionNotPresent(missing.pop(), self)
1979
return list(self._keys_to_version_ids(present_keys))
1980
1981
def get_ancestry(self, versions, topo_sorted=True):
1982
"""See VersionedFile.get_ancestry."""
1983
if not self._parents:
1984
return self._parentless_ancestry(versions)
1985
# XXX: This will do len(history) index calls - perhaps
1986
# it should be altered to be a index core feature?
1987
# get a graph of all the mentioned versions:
1988
graph = {}
1989
ghosts = set()
1990
versions = self._version_ids_to_keys(versions)
1991
pending = set(versions)
1992
while pending:
1993
# get all pending nodes
1994
this_iteration = pending
1995
new_nodes = self._get_entries(this_iteration)
1996
found = set()
1997
pending = set()
1998
for (index, key, value, node_refs) in new_nodes:
1999
# dont ask for ghosties - otherwise
2000
# we we can end up looping with pending
2001
# being entirely ghosted.
2002
graph[key] = [parent for parent in node_refs[0]
2003
if parent not in ghosts]
2004
# queue parents
2005
for parent in graph[key]:
2006
# dont examine known nodes again
2007
if parent in graph:
2008
continue
2009
pending.add(parent)
2010
found.add(key)
2011
ghosts.update(this_iteration.difference(found))
2012
if versions.difference(graph):
2013
raise RevisionNotPresent(versions.difference(graph).pop(), self)
2014
if topo_sorted:
2015
result_keys = topo_sort(graph.items())
2016
else:
2017
result_keys = graph.iterkeys()
2018
return [key[0] for key in result_keys]
2019
2020
def get_ancestry_with_ghosts(self, versions):
2021
"""See VersionedFile.get_ancestry."""
2022
if not self._parents:
2023
return self._parentless_ancestry(versions)
2024
# XXX: This will do len(history) index calls - perhaps
2025
# it should be altered to be a index core feature?
2026
# get a graph of all the mentioned versions:
2027
graph = {}
2028
versions = self._version_ids_to_keys(versions)
2029
pending = set(versions)
2030
while pending:
2031
# get all pending nodes
2032
this_iteration = pending
2033
new_nodes = self._get_entries(this_iteration)
2034
pending = set()
2035
for (index, key, value, node_refs) in new_nodes:
2036
graph[key] = node_refs[0]
2037
# queue parents
2038
for parent in graph[key]:
2039
# dont examine known nodes again
2040
if parent in graph:
2041
continue
2042
pending.add(parent)
2043
missing_versions = this_iteration.difference(graph)
2044
missing_needed = versions.intersection(missing_versions)
2045
if missing_needed:
2046
raise RevisionNotPresent(missing_needed.pop(), self)
2047
for missing_version in missing_versions:
2048
# add a key, no parents
2049
graph[missing_version] = []
2050
pending.discard(missing_version) # don't look for it
2051
result_keys = topo_sort(graph.items())
2052
return [key[0] for key in result_keys]
2053
2054
def get_build_details(self, version_ids):
2055
"""Get the method, index_memo and compression parent for version_ids.
2056
2057
Ghosts are omitted from the result.
2058
2059
:param version_ids: An iterable of version_ids.
2060
:return: A dict of version_id:(index_memo, compression_parent,
2061
parents, record_details).
2062
index_memo
2063
opaque structure to pass to read_records to extract the raw
2064
data
2065
compression_parent
2066
Content that this record is built upon, may be None
2067
parents
2068
Logical parents of this node
2069
record_details
2070
extra information about the content which needs to be passed to
2071
Factory.parse_record
2072
"""
2073
result = {}
2074
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
2075
for entry in entries:
2076
version_id = self._keys_to_version_ids((entry[1],))[0]
2077
if not self._parents:
2078
parents = ()
2079
else:
2080
parents = self._keys_to_version_ids(entry[3][0])
2081
if not self._deltas:
2082
compression_parent = None
2083
else:
2084
compression_parent_key = self._compression_parent(entry)
2085
if compression_parent_key:
2086
compression_parent = self._keys_to_version_ids(
2087
(compression_parent_key,))[0]
2088
else:
2089
compression_parent = None
2090
noeol = (entry[2][0] == 'N')
2091
if compression_parent:
2092
method = 'line-delta'
2093
else:
2094
method = 'fulltext'
2095
result[version_id] = (self._node_to_position(entry),
2096
compression_parent, parents,
2097
(method, noeol))
2098
return result
2099
2100
def _compression_parent(self, an_entry):
2101
# return the key that an_entry is compressed against, or None
2102
# Grab the second parent list (as deltas implies parents currently)
2103
compression_parents = an_entry[3][1]
2104
if not compression_parents:
2105
return None
2106
return compression_parents[0]
2107
2108
def _get_method(self, node):
2109
if not self._deltas:
2110
return 'fulltext'
2111
if self._compression_parent(node):
2112
return 'line-delta'
2113
else:
2114
return 'fulltext'
2115
2116
def num_versions(self):
2117
return len(list(self._graph_index.iter_all_entries()))
2118
2119
__len__ = num_versions
2120
2121
def get_versions(self):
2122
"""Get all the versions in the file. not topologically sorted."""
2123
return [node[1][0] for node in self._graph_index.iter_all_entries()]
2124
2125
def has_version(self, version_id):
2126
"""True if the version is in the index."""
2127
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
2128
2129
def _keys_to_version_ids(self, keys):
2130
return tuple(key[0] for key in keys)
2131
2132
def get_position(self, version_id):
2133
"""Return details needed to access the version.
2134
2135
:return: a tuple (index, data position, size) to hand to the access
2136
logic to get the record.
2137
"""
2138
node = self._get_node(version_id)
2139
return self._node_to_position(node)
2140
2141
def _node_to_position(self, node):
2142
"""Convert an index value to position details."""
2143
bits = node[2][1:].split(' ')
2144
return node[0], int(bits[0]), int(bits[1])
2145
2146
def get_method(self, version_id):
2147
"""Return compression method of specified version."""
2148
return self._get_method(self._get_node(version_id))
2149
2150
def _get_node(self, version_id):
2151
try:
2152
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
2153
except IndexError:
2154
raise RevisionNotPresent(version_id, self)
2155
2156
def get_options(self, version_id):
2157
"""Return a list representing options.
2158
2159
e.g. ['foo', 'bar']
2160
"""
2161
node = self._get_node(version_id)
2162
options = [self._get_method(node)]
2163
if node[2][0] == 'N':
2164
options.append('no-eol')
2165
return options
2166
2167
def get_parent_map(self, version_ids):
2168
"""Passed through to by KnitVersionedFile.get_parent_map."""
2169
nodes = self._get_entries(self._version_ids_to_keys(version_ids))
2170
result = {}
2171
if self._parents:
2172
for node in nodes:
2173
result[node[1][0]] = self._keys_to_version_ids(node[3][0])
2174
else:
2175
for node in nodes:
2176
result[node[1][0]] = ()
2177
return result
2178
2179
def get_parents_with_ghosts(self, version_id):
2180
"""Return parents of specified version with ghosts."""
2181
try:
2182
return self.get_parent_map([version_id])[version_id]
2183
except KeyError:
2184
raise RevisionNotPresent(version_id, self)
2185
2186
def check_versions_present(self, version_ids):
2187
"""Check that all specified versions are present."""
2188
keys = self._version_ids_to_keys(version_ids)
2189
present = self._present_keys(keys)
2190
missing = keys.difference(present)
2191
if missing:
2192
raise RevisionNotPresent(missing.pop(), self)
2193
2194
def add_version(self, version_id, options, access_memo, parents):
2195
"""Add a version record to the index."""
2196
return self.add_versions(((version_id, options, access_memo, parents),))
2197
2198
def add_versions(self, versions, random_id=False):
2199
"""Add multiple versions to the index.
2200
2201
This function does not insert data into the Immutable GraphIndex
2202
backing the KnitGraphIndex, instead it prepares data for insertion by
2203
the caller and checks that it is safe to insert then calls
2204
self._add_callback with the prepared GraphIndex nodes.
2205
2206
:param versions: a list of tuples:
2207
(version_id, options, pos, size, parents).
2208
:param random_id: If True the ids being added were randomly generated
2209
and no check for existence will be performed.
2210
"""
2211
if not self._add_callback:
2212
raise errors.ReadOnlyError(self)
2213
# we hope there are no repositories with inconsistent parentage
2214
# anymore.
2215
# check for dups
2216
2217
keys = {}
2218
for (version_id, options, access_memo, parents) in versions:
2219
index, pos, size = access_memo
2220
key = (version_id, )
2221
parents = tuple((parent, ) for parent in parents)
2222
if 'no-eol' in options:
2223
value = 'N'
2224
else:
2225
value = ' '
2226
value += "%d %d" % (pos, size)
2227
if not self._deltas:
2228
if 'line-delta' in options:
2229
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2230
if self._parents:
2231
if self._deltas:
2232
if 'line-delta' in options:
2233
node_refs = (parents, (parents[0],))
2234
else:
2235
node_refs = (parents, ())
2236
else:
2237
node_refs = (parents, )
2238
else:
2239
if parents:
2240
raise KnitCorrupt(self, "attempt to add node with parents "
2241
"in parentless index.")
2242
node_refs = ()
2243
keys[key] = (value, node_refs)
2244
if not random_id:
2245
present_nodes = self._get_entries(keys)
2246
for (index, key, value, node_refs) in present_nodes:
2247
if (value, node_refs) != keys[key]:
2248
raise KnitCorrupt(self, "inconsistent details in add_versions"
2249
": %s %s" % ((value, node_refs), keys[key]))
2250
del keys[key]
2251
result = []
2252
if self._parents:
2253
for key, (value, node_refs) in keys.iteritems():
2254
result.append((key, value, node_refs))
2255
else:
2256
for key, (value, node_refs) in keys.iteritems():
2257
result.append((key, value))
2258
self._add_callback(result)
2259
2260
def _version_ids_to_keys(self, version_ids):
2261
return set((version_id, ) for version_id in version_ids)
2262
2263
2264
class _KnitAccess(object):
2265
"""Access to knit records in a .knit file."""
2266
2267
def __init__(self, transport, filename, _file_mode, _dir_mode,
2268
_need_to_create, _create_parent_dir):
2269
"""Create a _KnitAccess for accessing and inserting data.
2270
2271
:param transport: The transport the .knit is located on.
2272
:param filename: The filename of the .knit.
2273
"""
2274
self._transport = transport
2275
self._filename = filename
2276
self._file_mode = _file_mode
2277
self._dir_mode = _dir_mode
2278
self._need_to_create = _need_to_create
2279
self._create_parent_dir = _create_parent_dir
2280
2281
def add_raw_records(self, sizes, raw_data):
2282
"""Add raw knit bytes to a storage area.
2283
2284
The data is spooled to whereever the access method is storing data.
2285
2286
:param sizes: An iterable containing the size of each raw data segment.
2287
:param raw_data: A bytestring containing the data.
2288
:return: A list of memos to retrieve the record later. Each memo is a
2289
tuple - (index, pos, length), where the index field is always None
2290
for the .knit access method.
2291
"""
2292
if not self._need_to_create:
2293
base = self._transport.append_bytes(self._filename, raw_data)
2294
else:
2295
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2296
create_parent_dir=self._create_parent_dir,
2297
mode=self._file_mode,
2298
dir_mode=self._dir_mode)
2299
self._need_to_create = False
2300
base = 0
2301
result = []
2302
for size in sizes:
2303
result.append((None, base, size))
2304
base += size
2305
return result
2306
2307
def create(self):
2308
"""IFF this data access has its own storage area, initialise it.
2309
2310
:return: None.
2311
"""
2312
self._transport.put_bytes_non_atomic(self._filename, '',
2313
mode=self._file_mode)
2314
2315
def open_file(self):
2316
"""IFF this data access can be represented as a single file, open it.
2317
2318
For knits that are not mapped to a single file on disk this will
2319
always return None.
2320
2321
:return: None or a file handle.
2322
"""
2323
try:
2324
return self._transport.get(self._filename)
2325
except NoSuchFile:
2326
pass
2327
return None
2328
2329
def get_raw_records(self, memos_for_retrieval):
2330
"""Get the raw bytes for a records.
2331
2332
:param memos_for_retrieval: An iterable containing the (index, pos,
2333
length) memo for retrieving the bytes. The .knit method ignores
2334
the index as there is always only a single file.
2335
:return: An iterator over the bytes of the records.
2336
"""
2337
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2338
for pos, data in self._transport.readv(self._filename, read_vector):
2339
yield data
2340
2341
2342
class _PackAccess(object):
2343
"""Access to knit records via a collection of packs."""
2344
2345
def __init__(self, index_to_packs, writer=None):
2346
"""Create a _PackAccess object.
2347
2348
:param index_to_packs: A dict mapping index objects to the transport
2349
and file names for obtaining data.
2350
:param writer: A tuple (pack.ContainerWriter, write_index) which
2351
contains the pack to write, and the index that reads from it will
2352
be associated with.
2353
"""
2354
if writer:
2355
self.container_writer = writer[0]
2356
self.write_index = writer[1]
2357
else:
2358
self.container_writer = None
2359
self.write_index = None
2360
self.indices = index_to_packs
2361
2362
def add_raw_records(self, sizes, raw_data):
2363
"""Add raw knit bytes to a storage area.
2364
2365
The data is spooled to the container writer in one bytes-record per
2366
raw data item.
2367
2368
:param sizes: An iterable containing the size of each raw data segment.
2369
:param raw_data: A bytestring containing the data.
2370
:return: A list of memos to retrieve the record later. Each memo is a
2371
tuple - (index, pos, length), where the index field is the
2372
write_index object supplied to the PackAccess object.
2373
"""
2374
result = []
2375
offset = 0
2376
for size in sizes:
2377
p_offset, p_length = self.container_writer.add_bytes_record(
2378
raw_data[offset:offset+size], [])
2379
offset += size
2380
result.append((self.write_index, p_offset, p_length))
2381
return result
2382
2383
def create(self):
2384
"""Pack based knits do not get individually created."""
2385
2386
def get_raw_records(self, memos_for_retrieval):
2387
"""Get the raw bytes for a records.
2388
2389
:param memos_for_retrieval: An iterable containing the (index, pos,
2390
length) memo for retrieving the bytes. The Pack access method
2391
looks up the pack to use for a given record in its index_to_pack
2392
map.
2393
:return: An iterator over the bytes of the records.
2394
"""
2395
# first pass, group into same-index requests
2396
request_lists = []
2397
current_index = None
2398
for (index, offset, length) in memos_for_retrieval:
2399
if current_index == index:
2400
current_list.append((offset, length))
2401
else:
2402
if current_index is not None:
2403
request_lists.append((current_index, current_list))
2404
current_index = index
2405
current_list = [(offset, length)]
2406
# handle the last entry
2407
if current_index is not None:
2408
request_lists.append((current_index, current_list))
2409
for index, offsets in request_lists:
2410
transport, path = self.indices[index]
2411
reader = pack.make_readv_reader(transport, path, offsets)
2412
for names, read_func in reader.iter_records():
2413
yield read_func(None)
2414
2415
def open_file(self):
2416
"""Pack based knits have no single file."""
2417
return None
2418
2419
def set_writer(self, writer, index, (transport, packname)):
2420
"""Set a writer to use for adding data."""
2421
if index is not None:
2422
self.indices[index] = (transport, packname)
2423
self.container_writer = writer
2424
self.write_index = index
2425
2426
2427
class _StreamAccess(object):
2428
"""A Knit Access object that provides data from a datastream.
2429
2430
It also provides a fallback to present as unannotated data, annotated data
2431
from a *backing* access object.
2432
2433
This is triggered by a index_memo which is pointing to a different index
2434
than this was constructed with, and is used to allow extracting full
2435
unannotated texts for insertion into annotated knits.
2436
"""
2437
2438
def __init__(self, reader_callable, stream_index, backing_knit,
2439
orig_factory):
2440
"""Create a _StreamAccess object.
2441
2442
:param reader_callable: The reader_callable from the datastream.
2443
This is called to buffer all the data immediately, for
2444
random access.
2445
:param stream_index: The index the data stream this provides access to
2446
which will be present in native index_memo's.
2447
:param backing_knit: The knit object that will provide access to
2448
annotated texts which are not available in the stream, so as to
2449
create unannotated texts.
2450
:param orig_factory: The original content factory used to generate the
2451
stream. This is used for checking whether the thunk code for
2452
supporting _copy_texts will generate the correct form of data.
2453
"""
2454
self.data = reader_callable(None)
2455
self.stream_index = stream_index
2456
self.backing_knit = backing_knit
2457
self.orig_factory = orig_factory
2458
2459
def get_raw_records(self, memos_for_retrieval):
2460
"""Get the raw bytes for a records.
2461
2462
:param memos_for_retrieval: An iterable of memos from the
2463
_StreamIndex object identifying bytes to read; for these classes
2464
they are (from_backing_knit, index, start, end) and can point to
2465
either the backing knit or streamed data.
2466
:return: An iterator yielding a byte string for each record in
2467
memos_for_retrieval.
2468
"""
2469
# use a generator for memory friendliness
2470
for from_backing_knit, version_id, start, end in memos_for_retrieval:
2471
if not from_backing_knit:
2472
if version_id is not self.stream_index:
2473
raise AssertionError()
2474
yield self.data[start:end]
2475
continue
2476
# we have been asked to thunk. This thunking only occurs when
2477
# we are obtaining plain texts from an annotated backing knit
2478
# so that _copy_texts will work.
2479
# We could improve performance here by scanning for where we need
2480
# to do this and using get_line_list, then interleaving the output
2481
# as desired. However, for now, this is sufficient.
2482
if self.orig_factory.__class__ != KnitPlainFactory:
2483
raise errors.KnitCorrupt(
2484
self, 'Bad thunk request %r cannot be backed by %r' %
2485
(version_id, self.orig_factory))
2486
lines = self.backing_knit.get_lines(version_id)
2487
line_bytes = ''.join(lines)
2488
digest = sha_string(line_bytes)
2489
# the packed form of the fulltext always has a trailing newline,
2490
# even if the actual text does not, unless the file is empty. the
2491
# record options including the noeol flag are passed through by
2492
# _StreamIndex, so this is safe.
2493
if lines:
2494
if lines[-1][-1] != '\n':
2495
lines[-1] = lines[-1] + '\n'
2496
line_bytes += '\n'
2497
# We want plain data, because we expect to thunk only to allow text
2498
# extraction.
2499
size, bytes = self.backing_knit._data._record_to_data(version_id,
2500
digest, lines, line_bytes)
2501
yield bytes
2502
2503
2504
class _StreamIndex(object):
2505
"""A Knit Index object that uses the data map from a datastream."""
2506
2507
def __init__(self, data_list, backing_index):
2508
"""Create a _StreamIndex object.
2509
2510
:param data_list: The data_list from the datastream.
2511
:param backing_index: The index which will supply values for nodes
2512
referenced outside of this stream.
2513
"""
2514
self.data_list = data_list
2515
self.backing_index = backing_index
2516
self._by_version = {}
2517
pos = 0
2518
for key, options, length, parents in data_list:
2519
self._by_version[key] = options, (pos, pos + length), parents
2520
pos += length
2521
2522
def get_ancestry(self, versions, topo_sorted):
2523
"""Get an ancestry list for versions."""
2524
if topo_sorted:
2525
# Not needed for basic joins
2526
raise NotImplementedError(self.get_ancestry)
2527
# get a graph of all the mentioned versions:
2528
# Little ugly - basically copied from KnitIndex, but don't want to
2529
# accidentally incorporate too much of that index's code.
2530
ancestry = set()
2531
pending = set(versions)
2532
cache = self._by_version
2533
while pending:
2534
version = pending.pop()
2535
# trim ghosts
2536
try:
2537
parents = [p for p in cache[version][2] if p in cache]
2538
except KeyError:
2539
raise RevisionNotPresent(version, self)
2540
# if not completed and not a ghost
2541
pending.update([p for p in parents if p not in ancestry])
2542
ancestry.add(version)
2543
return list(ancestry)
2544
2545
def get_build_details(self, version_ids):
2546
"""Get the method, index_memo and compression parent for version_ids.
2547
2548
Ghosts are omitted from the result.
2549
2550
:param version_ids: An iterable of version_ids.
2551
:return: A dict of version_id:(index_memo, compression_parent,
2552
parents, record_details).
2553
index_memo
2554
opaque memo that can be passed to _StreamAccess.read_records
2555
to extract the raw data; for these classes it is
2556
(from_backing_knit, index, start, end)
2557
compression_parent
2558
Content that this record is built upon, may be None
2559
parents
2560
Logical parents of this node
2561
record_details
2562
extra information about the content which needs to be passed to
2563
Factory.parse_record
2564
"""
2565
result = {}
2566
for version_id in version_ids:
2567
try:
2568
method = self.get_method(version_id)
2569
except errors.RevisionNotPresent:
2570
# ghosts are omitted
2571
continue
2572
parent_ids = self.get_parents_with_ghosts(version_id)
2573
noeol = ('no-eol' in self.get_options(version_id))
2574
index_memo = self.get_position(version_id)
2575
from_backing_knit = index_memo[0]
2576
if from_backing_knit:
2577
# texts retrieved from the backing knit are always full texts
2578
method = 'fulltext'
2579
if method == 'fulltext':
2580
compression_parent = None
2581
else:
2582
compression_parent = parent_ids[0]
2583
result[version_id] = (index_memo, compression_parent,
2584
parent_ids, (method, noeol))
2585
return result
2586
2587
def get_method(self, version_id):
2588
"""Return compression method of specified version."""
2589
options = self.get_options(version_id)
2590
if 'fulltext' in options:
2591
return 'fulltext'
2592
elif 'line-delta' in options:
2593
return 'line-delta'
2594
else:
2595
raise errors.KnitIndexUnknownMethod(self, options)
2596
2597
def get_options(self, version_id):
2598
"""Return a list representing options.
2599
2600
e.g. ['foo', 'bar']
2601
"""
2602
try:
2603
return self._by_version[version_id][0]
2604
except KeyError:
2605
options = list(self.backing_index.get_options(version_id))
2606
if 'fulltext' in options:
2607
pass
2608
elif 'line-delta' in options:
2609
# Texts from the backing knit are always returned from the stream
2610
# as full texts
2611
options.remove('line-delta')
2612
options.append('fulltext')
2613
else:
2614
raise errors.KnitIndexUnknownMethod(self, options)
2615
return tuple(options)
2616
2617
def get_parent_map(self, version_ids):
2618
"""Passed through to by KnitVersionedFile.get_parent_map."""
2619
result = {}
2620
pending_ids = set()
2621
for version_id in version_ids:
2622
try:
2623
result[version_id] = self._by_version[version_id][2]
2624
except KeyError:
2625
pending_ids.add(version_id)
2626
result.update(self.backing_index.get_parent_map(pending_ids))
2627
return result
2628
2629
def get_parents_with_ghosts(self, version_id):
2630
"""Return parents of specified version with ghosts."""
2631
try:
2632
return self.get_parent_map([version_id])[version_id]
2633
except KeyError:
2634
raise RevisionNotPresent(version_id, self)
2635
2636
def get_position(self, version_id):
2637
"""Return details needed to access the version.
2638
2639
_StreamAccess has the data as a big array, so we return slice
2640
coordinates into that (as index_memo's are opaque outside the
2641
index and matching access class).
2642
2643
:return: a tuple (from_backing_knit, index, start, end) that can
2644
be passed e.g. to get_raw_records.
2645
If from_backing_knit is False, index will be self, otherwise it
2646
will be a version id.
2647
"""
2648
try:
2649
start, end = self._by_version[version_id][1]
2650
return False, self, start, end
2651
except KeyError:
2652
# Signal to the access object to handle this from the backing knit.
2653
return (True, version_id, None, None)
2654
2655
def get_versions(self):
2656
"""Get all the versions in the stream."""
2657
return self._by_version.keys()
2658
2659
2660
class _KnitData(object):
2661
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2662
2663
The KnitData class provides the logic for parsing and using knit records,
2664
making use of an access method for the low level read and write operations.
2665
"""
2666
2667
def __init__(self, access):
2668
"""Create a KnitData object.
2669
2670
:param access: The access method to use. Access methods such as
2671
_KnitAccess manage the insertion of raw records and the subsequent
2672
retrieval of the same.
2673
"""
2674
self._access = access
1237
2675
self._checked = False
1238
if create:
1239
self._transport.put(self._filename, StringIO(''), mode=file_mode)
1240
self._records = {}
1241
1242
def clear_cache(self):
1243
"""Clear the record cache."""
1244
self._records = {}
1245
2676
1246
2677
def _open_file(self):
1247
if self._file is None:
1248
try:
1249
self._file = self._transport.get(self._filename)
1250
except NoSuchFile:
1251
pass
1252
return self._file
2678
return self._access.open_file()
1253
2679
1254
def _record_to_data(self, version_id, digest, lines):
2680
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
1255
2681
"""Convert version_id, digest, lines into a raw data block.
1256
2682
2683
:param dense_lines: The bytes of lines but in a denser form. For
2684
instance, if lines is a list of 1000 bytestrings each ending in \n,
2685
dense_lines may be a list with one line in it, containing all the
2686
1000's lines and their \n's. Using dense_lines if it is already
2687
known is a win because the string join to create bytes in this
2688
function spends less time resizing the final string.
1257
2689
:return: (len, a StringIO instance with the raw data ready to read.)
1258
2690
"""
1259
sio = StringIO()
1260
data_file = GzipFile(None, mode='wb', fileobj=sio)
1261
data_file.writelines(chain(
1262
["version %s %d %s\n" % (version_id.encode('utf-8'),
2691
# Note: using a string copy here increases memory pressure with e.g.
2692
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2693
# when doing the initial commit of a mozilla tree. RBC 20070921
2694
bytes = ''.join(chain(
2695
["version %s %d %s\n" % (version_id,
1263
2696
len(lines),
1264
2697
digest)],
1265
lines,
1266
["end %s\n" % version_id.encode('utf-8')]))
1267
data_file.close()
1268
length= sio.tell()
1269
1270
sio.seek(0)
1271
return length, sio
1272
1273
def add_raw_record(self, raw_data):
2698
dense_lines or lines,
2699
["end %s\n" % version_id]))
2700
compressed_bytes = bytes_to_gzip(bytes)
2701
return len(compressed_bytes), compressed_bytes
2702
2703
def add_raw_records(self, sizes, raw_data):
1274
2704
"""Append a prepared record to the data file.
1275
2705
1276
:return: the offset in the data file raw_data was written.
2706
:param sizes: An iterable containing the size of each raw data segment.
2707
:param raw_data: A bytestring containing the data.
2708
:return: a list of index data for the way the data was stored.
2709
See the access method add_raw_records documentation for more
2710
details.
1277
2711
"""
1278
assert isinstance(raw_data, str), 'data must be plain bytes'
1279
return self._transport.append(self._filename, StringIO(raw_data))
2712
return self._access.add_raw_records(sizes, raw_data)
1280
2713
1281
def add_record(self, version_id, digest, lines):
1282
"""Write new text record to disk. Returns the position in the
1283
file where it was written."""
1284
size, sio = self._record_to_data(version_id, digest, lines)
1285
# cache
1286
self._records[version_id] = (digest, lines)
1287
# write to disk
1288
start_pos = self._transport.append(self._filename, sio)
1289
return start_pos, size
1290
1291
2714
def _parse_record_header(self, version_id, raw_data):
1292
2715
"""Parse a record header for consistency.
1293
2716
1295
2718
as (stream, header_record)
1296
2719
"""
1297
2720
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1298
rec = df.readline().split()
2721
try:
2722
rec = self._check_header(version_id, df.readline())
2723
except Exception, e:
2724
raise KnitCorrupt(self._access,
2725
"While reading {%s} got %s(%s)"
2726
% (version_id, e.__class__.__name__, str(e)))
2727
return df, rec
2728
2729
def _split_header(self, line):
2730
rec = line.split()
1299
2731
if len(rec) != 4:
1300
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
1301
if rec[1].decode('utf-8')!= version_id:
1302
raise KnitCorrupt(self._filename,
1303
'unexpected version, wanted %r, got %r' % (
1304
version_id, rec[1]))
1305
return df, rec
1306
1307
def _parse_record(self, version_id, data):
2732
raise KnitCorrupt(self._access,
2733
'unexpected number of elements in record header')
2734
return rec
2735
2736
def _check_header_version(self, rec, version_id):
2737
if rec[1] != version_id:
2738
raise KnitCorrupt(self._access,
2739
'unexpected version, wanted %r, got %r'
2740
% (version_id, rec[1]))
2741
2742
def _check_header(self, version_id, line):
2743
rec = self._split_header(line)
2744
self._check_header_version(rec, version_id)
2745
return rec
2746
2747
def _parse_record_unchecked(self, data):
1308
2748
# profiling notes:
1309
2749
# 4168 calls in 2880 217 internal
1310
2750
# 4168 calls to _parse_record_header in 2121
1311
2751
# 4168 calls to readlines in 330
1312
df, rec = self._parse_record_header(version_id, data)
1313
record_contents = df.readlines()
1314
l = record_contents.pop()
1315
assert len(record_contents) == int(rec[2])
1316
if l.decode('utf-8') != 'end %s\n' % version_id:
1317
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
1318
% (l, version_id))
2752
df = GzipFile(mode='rb', fileobj=StringIO(data))
2753
try:
2754
record_contents = df.readlines()
2755
except Exception, e:
2756
raise KnitCorrupt(self._access, "Corrupt compressed record %r, got %s(%s)" %
2757
(data, e.__class__.__name__, str(e)))
2758
header = record_contents.pop(0)
2759
rec = self._split_header(header)
2760
last_line = record_contents.pop()
2761
if len(record_contents) != int(rec[2]):
2762
raise KnitCorrupt(self._access,
2763
'incorrect number of lines %s != %s'
2764
' for version {%s}'
2765
% (len(record_contents), int(rec[2]),
2766
rec[1]))
2767
if last_line != 'end %s\n' % rec[1]:
2768
raise KnitCorrupt(self._access,
2769
'unexpected version end line %r, wanted %r'
2770
% (last_line, rec[1]))
1319
2771
df.close()
2772
return rec, record_contents
2773
2774
def _parse_record(self, version_id, data):
2775
rec, record_contents = self._parse_record_unchecked(data)
2776
self._check_header_version(rec, version_id)
1320
2777
return record_contents, rec[3]
1321
2778
1322
2779
def read_records_iter_raw(self, records):
1325
2782
This unpacks enough of the text record to validate the id is
1326
2783
as expected but thats all.
1327
2784
1328
It will actively recompress currently cached records on the
1329
basis that that is cheaper than I/O activity.
2785
Each item the iterator yields is (version_id, bytes,
2786
sha1_of_full_text).
1330
2787
"""
1331
needed_records = []
1332
for version_id, pos, size in records:
1333
if version_id not in self._records:
1334
needed_records.append((version_id, pos, size))
1335
1336
2788
# setup an iterator of the external records:
1337
2789
# uses readv so nice and fast we hope.
1338
if len(needed_records):
2790
if len(records):
1339
2791
# grab the disk data needed.
1340
raw_records = self._transport.readv(self._filename,
1341
[(pos, size) for version_id, pos, size in needed_records])
1342
1343
for version_id, pos, size in records:
1344
if version_id in self._records:
1345
# compress a new version
1346
size, sio = self._record_to_data(version_id,
1347
self._records[version_id][0],
1348
self._records[version_id][1])
1349
yield version_id, sio.getvalue()
1350
else:
1351
pos, data = raw_records.next()
1352
# validate the header
1353
df, rec = self._parse_record_header(version_id, data)
1354
df.close()
1355
yield version_id, data
1356
2792
needed_offsets = [index_memo for version_id, index_memo
2793
in records]
2794
raw_records = self._access.get_raw_records(needed_offsets)
2795
2796
for version_id, index_memo in records:
2797
data = raw_records.next()
2798
# validate the header
2799
df, rec = self._parse_record_header(version_id, data)
2800
df.close()
2801
yield version_id, data, rec[3]
1357
2802
1358
2803
def read_records_iter(self, records):
1359
2804
"""Read text records from data file and yield result.
1360
2805
1361
Each passed record is a tuple of (version_id, pos, len) and
1362
will be read in the given order. Yields (version_id,
1363
contents, digest).
2806
The result will be returned in whatever is the fastest to read.
2807
Not by the order requested. Also, multiple requests for the same
2808
record will only yield 1 response.
2809
:param records: A list of (version_id, pos, len) entries
2810
:return: Yields (version_id, contents, digest) in the order
2811
read, not the order requested
1364
2812
"""
1365
# profiling notes:
1366
# 60890 calls for 4168 extractions in 5045, 683 internal.
1367
# 4168 calls to readv in 1411
1368
# 4168 calls to parse_record in 2880
1369
1370
needed_records = []
1371
for version_id, pos, size in records:
1372
if version_id not in self._records:
1373
needed_records.append((version_id, pos, size))
1374
1375
if len(needed_records):
1376
# We take it that the transport optimizes the fetching as good
1377
# as possible (ie, reads continuous ranges.)
1378
response = self._transport.readv(self._filename,
1379
[(pos, size) for version_id, pos, size in needed_records])
1380
1381
for (record_id, pos, size), (pos, data) in izip(iter(needed_records), response):
1382
content, digest = self._parse_record(record_id, data)
1383
self._records[record_id] = (digest, content)
1384
1385
for version_id, pos, size in records:
1386
yield version_id, list(self._records[version_id][1]), self._records[version_id][0]
2813
if not records:
2814
return
2815
2816
needed_records = sorted(set(records), key=operator.itemgetter(1))
2817
if not needed_records:
2818
return
2819
2820
# The transport optimizes the fetching as well
2821
# (ie, reads continuous ranges.)
2822
raw_data = self._access.get_raw_records(
2823
[index_memo for version_id, index_memo in needed_records])
2824
2825
for (version_id, index_memo), data in \
2826
izip(iter(needed_records), raw_data):
2827
content, digest = self._parse_record(version_id, data)
2828
yield version_id, content, digest
1387
2829
1388
2830
def read_records(self, records):
1389
2831
"""Read records into a dictionary."""
1390
2832
components = {}
1391
for record_id, content, digest in self.read_records_iter(records):
2833
for record_id, content, digest in \
2834
self.read_records_iter(records):
1392
2835
components[record_id] = (content, digest)
1393
2836
return components
1394
2837
1396
2839
class InterKnit(InterVersionedFile):
1397
2840
"""Optimised code paths for knit to knit operations."""
1398
2841
1399
_matching_file_from_factory = KnitVersionedFile
1400
_matching_file_to_factory = KnitVersionedFile
2842
_matching_file_from_factory = staticmethod(make_file_knit)
2843
_matching_file_to_factory = staticmethod(make_file_knit)
1401
2844
1402
2845
@staticmethod
1403
2846
def is_compatible(source, target):
1408
2851
except AttributeError:
1409
2852
return False
1410
2853
2854
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2855
"""Copy texts to the target by extracting and adding them one by one.
2856
2857
see join() for the parameter definitions.
2858
"""
2859
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2860
# --- the below is factorable out with VersionedFile.join, but wait for
2861
# VersionedFiles, it may all be simpler then.
2862
graph = Graph(self.source)
2863
search = graph._make_breadth_first_searcher(version_ids)
2864
transitive_ids = set()
2865
map(transitive_ids.update, list(search))
2866
parent_map = self.source.get_parent_map(transitive_ids)
2867
order = topo_sort(parent_map.items())
2868
2869
def size_of_content(content):
2870
return sum(len(line) for line in content.text())
2871
# Cache at most 10MB of parent texts
2872
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2873
compute_size=size_of_content)
2874
# TODO: jam 20071116 It would be nice to have a streaming interface to
2875
# get multiple texts from a source. The source could be smarter
2876
# about how it handled intermediate stages.
2877
# get_line_list() or make_mpdiffs() seem like a possibility, but
2878
# at the moment they extract all full texts into memory, which
2879
# causes us to store more than our 3x fulltext goal.
2880
# Repository.iter_files_bytes() may be another possibility
2881
to_process = [version for version in order
2882
if version not in self.target]
2883
total = len(to_process)
2884
pb = ui.ui_factory.nested_progress_bar()
2885
try:
2886
for index, version in enumerate(to_process):
2887
pb.update('Converting versioned data', index, total)
2888
sha1, num_bytes, parent_text = self.target.add_lines(version,
2889
self.source.get_parents_with_ghosts(version),
2890
self.source.get_lines(version),
2891
parent_texts=parent_cache)
2892
parent_cache[version] = parent_text
2893
finally:
2894
pb.finished()
2895
return total
2896
1411
2897
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1412
2898
"""See InterVersionedFile.join."""
1413
assert isinstance(self.source, KnitVersionedFile)
1414
assert isinstance(self.target, KnitVersionedFile)
2899
# If the source and target are mismatched w.r.t. annotations vs
2900
# plain, the data needs to be converted accordingly
2901
if self.source.factory.annotated == self.target.factory.annotated:
2902
converter = None
2903
elif self.source.factory.annotated:
2904
converter = self._anno_to_plain_converter
2905
else:
2906
# We're converting from a plain to an annotated knit. Copy them
2907
# across by full texts.
2908
return self._copy_texts(pb, msg, version_ids, ignore_missing)
1415
2909
1416
2910
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1417
1418
2911
if not version_ids:
1419
2912
return 0
1420
2913
1421
pb = bzrlib.ui.ui_factory.nested_progress_bar()
2914
pb = ui.ui_factory.nested_progress_bar()
1422
2915
try:
1423
2916
version_ids = list(version_ids)
1424
2917
if None in version_ids:
1425
2918
version_ids.remove(None)
1426
2919
1427
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2920
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2921
topo_sorted=False))
1428
2922
this_versions = set(self.target._index.get_versions())
2923
# XXX: For efficiency we should not look at the whole index,
2924
# we only need to consider the referenced revisions - they
2925
# must all be present, or the method must be full-text.
2926
# TODO, RBC 20070919
1429
2927
needed_versions = self.source_ancestry - this_versions
1430
cross_check_versions = self.source_ancestry.intersection(this_versions)
1431
mismatched_versions = set()
1432
for version in cross_check_versions:
1433
# scan to include needed parents.
1434
n1 = set(self.target.get_parents_with_ghosts(version))
1435
n2 = set(self.source.get_parents_with_ghosts(version))
1436
if n1 != n2:
1437
# FIXME TEST this check for cycles being introduced works
1438
# the logic is we have a cycle if in our graph we are an
1439
# ancestor of any of the n2 revisions.
1440
for parent in n2:
1441
if parent in n1:
1442
# safe
1443
continue
1444
else:
1445
parent_ancestors = self.source.get_ancestry(parent)
1446
if version in parent_ancestors:
1447
raise errors.GraphCycleError([parent, version])
1448
# ensure this parent will be available later.
1449
new_parents = n2.difference(n1)
1450
needed_versions.update(new_parents.difference(this_versions))
1451
mismatched_versions.add(version)
1452
2928
1453
if not needed_versions and not mismatched_versions:
2929
if not needed_versions:
1454
2930
return 0
1455
full_list = topo_sort(self.source.get_graph())
2931
full_list = topo_sort(
2932
self.source.get_parent_map(self.source.versions()))
1456
2933
1457
2934
version_list = [i for i in full_list if (not self.target.has_version(i)
1458
2935
and i in needed_versions)]
1470
2947
# * already have it or
1471
2948
# * have it scheduled already
1472
2949
# otherwise we don't care
1473
assert (self.target.has_version(parent) or
2950
if not (self.target.has_version(parent) or
1474
2951
parent in copy_set or
1475
not self.source.has_version(parent))
1476
data_pos, data_size = self.source._index.get_position(version_id)
1477
copy_queue_records.append((version_id, data_pos, data_size))
2952
not self.source.has_version(parent)):
2953
raise AssertionError("problem joining parent %r "
2954
"from %r to %r"
2955
% (parent, self.source, self.target))
2956
index_memo = self.source._index.get_position(version_id)
2957
copy_queue_records.append((version_id, index_memo))
1478
2958
copy_queue.append((version_id, options, parents))
1479
2959
copy_set.add(version_id)
1480
2960
1483
2963
total = len(version_list)
1484
2964
raw_datum = []
1485
2965
raw_records = []
1486
for (version_id, raw_data), \
2966
for (version_id, raw_data, _), \
1487
2967
(version_id2, options, parents) in \
1488
2968
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1489
2969
copy_queue):
1490
assert version_id == version_id2, 'logic error, inconsistent results'
2970
if not (version_id == version_id2):
2971
raise AssertionError('logic error, inconsistent results')
1491
2972
count = count + 1
1492
2973
pb.update("Joining knit", count, total)
1493
raw_records.append((version_id, options, parents, len(raw_data)))
2974
if converter:
2975
size, raw_data = converter(raw_data, version_id, options,
2976
parents)
2977
else:
2978
size = len(raw_data)
2979
raw_records.append((version_id, options, parents, size))
1494
2980
raw_datum.append(raw_data)
1495
2981
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1496
1497
for version in mismatched_versions:
1498
# FIXME RBC 20060309 is this needed?
1499
n1 = set(self.target.get_parents_with_ghosts(version))
1500
n2 = set(self.source.get_parents_with_ghosts(version))
1501
# write a combined record to our history preserving the current
1502
# parents as first in the list
1503
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1504
self.target.fix_parents(version, new_parents)
1505
2982
return count
1506
2983
finally:
1507
2984
pb.finished()
1508
2985
2986
def _anno_to_plain_converter(self, raw_data, version_id, options,
2987
parents):
2988
"""Convert annotated content to plain content."""
2989
data, digest = self.source._data._parse_record(version_id, raw_data)
2990
if 'fulltext' in options:
2991
content = self.source.factory.parse_fulltext(data, version_id)
2992
lines = self.target.factory.lower_fulltext(content)
2993
else:
2994
delta = self.source.factory.parse_line_delta(data, version_id,
2995
plain=True)
2996
lines = self.target.factory.lower_line_delta(delta)
2997
return self.target._data._record_to_data(version_id, digest, lines)
2998
1509
2999
1510
3000
InterVersionedFile.register_optimiser(InterKnit)
1511
3001
1514
3004
"""Optimised code paths for weave to knit operations."""
1515
3005
1516
3006
_matching_file_from_factory = bzrlib.weave.WeaveFile
1517
_matching_file_to_factory = KnitVersionedFile
3007
_matching_file_to_factory = staticmethod(make_file_knit)
1518
3008
1519
3009
@staticmethod
1520
3010
def is_compatible(source, target):
1527
3017
1528
3018
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1529
3019
"""See InterVersionedFile.join."""
1530
assert isinstance(self.source, bzrlib.weave.Weave)
1531
assert isinstance(self.target, KnitVersionedFile)
1532
1533
3020
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1534
3021
1535
3022
if not version_ids:
1536
3023
return 0
1537
3024
1538
pb = bzrlib.ui.ui_factory.nested_progress_bar()
3025
pb = ui.ui_factory.nested_progress_bar()
1539
3026
try:
1540
3027
version_ids = list(version_ids)
1541
3028
1542
3029
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1543
3030
this_versions = set(self.target._index.get_versions())
1544
3031
needed_versions = self.source_ancestry - this_versions
1545
cross_check_versions = self.source_ancestry.intersection(this_versions)
1546
mismatched_versions = set()
1547
for version in cross_check_versions:
1548
# scan to include needed parents.
1549
n1 = set(self.target.get_parents_with_ghosts(version))
1550
n2 = set(self.source.get_parents(version))
1551
# if all of n2's parents are in n1, then its fine.
1552
if n2.difference(n1):
1553
# FIXME TEST this check for cycles being introduced works
1554
# the logic is we have a cycle if in our graph we are an
1555
# ancestor of any of the n2 revisions.
1556
for parent in n2:
1557
if parent in n1:
1558
# safe
1559
continue
1560
else:
1561
parent_ancestors = self.source.get_ancestry(parent)
1562
if version in parent_ancestors:
1563
raise errors.GraphCycleError([parent, version])
1564
# ensure this parent will be available later.
1565
new_parents = n2.difference(n1)
1566
needed_versions.update(new_parents.difference(this_versions))
1567
mismatched_versions.add(version)
1568
3032
1569
if not needed_versions and not mismatched_versions:
3033
if not needed_versions:
1570
3034
return 0
1571
full_list = topo_sort(self.source.get_graph())
3035
full_list = topo_sort(
3036
self.source.get_parent_map(self.source.versions()))
1572
3037
1573
3038
version_list = [i for i in full_list if (not self.target.has_version(i)
1574
3039
and i in needed_versions)]
1576
3041
# do the join:
1577
3042
count = 0
1578
3043
total = len(version_list)
3044
parent_map = self.source.get_parent_map(version_list)
1579
3045
for version_id in version_list:
1580
3046
pb.update("Converting to knit", count, total)
1581
parents = self.source.get_parents(version_id)
3047
parents = parent_map[version_id]
1582
3048
# check that its will be a consistent copy:
1583
3049
for parent in parents:
1584
3050
# if source has the parent, we must already have it
1585
assert (self.target.has_version(parent))
3051
if not self.target.has_version(parent):
3052
raise AssertionError("%r does not have parent %r"
3053
% (self.target, parent))
1586
3054
self.target.add_lines(
1587
3055
version_id, parents, self.source.get_lines(version_id))
1588
3056
count = count + 1
1589
1590
for version in mismatched_versions:
1591
# FIXME RBC 20060309 is this needed?
1592
n1 = set(self.target.get_parents_with_ghosts(version))
1593
n2 = set(self.source.get_parents(version))
1594
# write a combined record to our history preserving the current
1595
# parents as first in the list
1596
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1597
self.target.fix_parents(version, new_parents)
1598
3057
return count
1599
3058
finally:
1600
3059
pb.finished()
1603
3062
InterVersionedFile.register_optimiser(WeaveToKnit)
1604
3063
1605
3064
1606
class KnitSequenceMatcher(difflib.SequenceMatcher):
1607
"""Knit tuned sequence matcher.
1608
1609
This is based on profiling of difflib which indicated some improvements
1610
for our usage pattern.
3065
# Deprecated, use PatienceSequenceMatcher instead
3066
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
3067
3068
3069
def annotate_knit(knit, revision_id):
3070
"""Annotate a knit with no cached annotations.
3071
3072
This implementation is for knits with no cached annotations.
3073
It will work for knits with cached annotations, but this is not
3074
recommended.
1611
3075
"""
1612
1613
def find_longest_match(self, alo, ahi, blo, bhi):
1614
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1615
1616
If isjunk is not defined:
1617
1618
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1619
alo <= i <= i+k <= ahi
1620
blo <= j <= j+k <= bhi
1621
and for all (i',j',k') meeting those conditions,
1622
k >= k'
1623
i <= i'
1624
and if i == i', j <= j'
1625
1626
In other words, of all maximal matching blocks, return one that
1627
starts earliest in a, and of all those maximal matching blocks that
1628
start earliest in a, return the one that starts earliest in b.
1629
1630
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1631
>>> s.find_longest_match(0, 5, 0, 9)
1632
(0, 4, 5)
1633
1634
If isjunk is defined, first the longest matching block is
1635
determined as above, but with the additional restriction that no
1636
junk element appears in the block. Then that block is extended as
1637
far as possible by matching (only) junk elements on both sides. So
1638
the resulting block never matches on junk except as identical junk
1639
happens to be adjacent to an "interesting" match.
1640
1641
Here's the same example as before, but considering blanks to be
1642
junk. That prevents " abcd" from matching the " abcd" at the tail
1643
end of the second sequence directly. Instead only the "abcd" can
1644
match, and matches the leftmost "abcd" in the second sequence:
1645
1646
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1647
>>> s.find_longest_match(0, 5, 0, 9)
1648
(1, 0, 4)
1649
1650
If no blocks match, return (alo, blo, 0).
1651
1652
>>> s = SequenceMatcher(None, "ab", "c")
1653
>>> s.find_longest_match(0, 2, 0, 1)
1654
(0, 0, 0)
1655
"""
1656
1657
# CAUTION: stripping common prefix or suffix would be incorrect.
1658
# E.g.,
1659
# ab
1660
# acab
1661
# Longest matching block is "ab", but if common prefix is
1662
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1663
# strip, so ends up claiming that ab is changed to acab by
1664
# inserting "ca" in the middle. That's minimal but unintuitive:
1665
# "it's obvious" that someone inserted "ac" at the front.
1666
# Windiff ends up at the same place as diff, but by pairing up
1667
# the unique 'b's and then matching the first two 'a's.
1668
1669
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1670
besti, bestj, bestsize = alo, blo, 0
1671
# find longest junk-free match
1672
# during an iteration of the loop, j2len[j] = length of longest
1673
# junk-free match ending with a[i-1] and b[j]
1674
j2len = {}
1675
# nothing = []
1676
b2jget = b2j.get
1677
for i in xrange(alo, ahi):
1678
# look at all instances of a[i] in b; note that because
1679
# b2j has no junk keys, the loop is skipped if a[i] is junk
1680
j2lenget = j2len.get
1681
newj2len = {}
1682
1683
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1684
# following improvement
1685
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1686
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1687
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1688
# to
1689
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1690
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1691
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
1692
1693
try:
1694
js = b2j[a[i]]
1695
except KeyError:
1696
pass
3076
annotator = _KnitAnnotator(knit)
3077
return iter(annotator.annotate(revision_id))
3078
3079
3080
class _KnitAnnotator(object):
3081
"""Build up the annotations for a text."""
3082
3083
def __init__(self, knit):
3084
self._knit = knit
3085
3086
# Content objects, differs from fulltexts because of how final newlines
3087
# are treated by knits. the content objects here will always have a
3088
# final newline
3089
self._fulltext_contents = {}
3090
3091
# Annotated lines of specific revisions
3092
self._annotated_lines = {}
3093
3094
# Track the raw data for nodes that we could not process yet.
3095
# This maps the revision_id of the base to a list of children that will
3096
# annotated from it.
3097
self._pending_children = {}
3098
3099
# Nodes which cannot be extracted
3100
self._ghosts = set()
3101
3102
# Track how many children this node has, so we know if we need to keep
3103
# it
3104
self._annotate_children = {}
3105
self._compression_children = {}
3106
3107
self._all_build_details = {}
3108
# The children => parent revision_id graph
3109
self._revision_id_graph = {}
3110
3111
self._heads_provider = None
3112
3113
self._nodes_to_keep_annotations = set()
3114
self._generations_until_keep = 100
3115
3116
def set_generations_until_keep(self, value):
3117
"""Set the number of generations before caching a node.
3118
3119
Setting this to -1 will cache every merge node, setting this higher
3120
will cache fewer nodes.
3121
"""
3122
self._generations_until_keep = value
3123
3124
def _add_fulltext_content(self, revision_id, content_obj):
3125
self._fulltext_contents[revision_id] = content_obj
3126
# TODO: jam 20080305 It might be good to check the sha1digest here
3127
return content_obj.text()
3128
3129
def _check_parents(self, child, nodes_to_annotate):
3130
"""Check if all parents have been processed.
3131
3132
:param child: A tuple of (rev_id, parents, raw_content)
3133
:param nodes_to_annotate: If child is ready, add it to
3134
nodes_to_annotate, otherwise put it back in self._pending_children
3135
"""
3136
for parent_id in child[1]:
3137
if (parent_id not in self._annotated_lines):
3138
# This parent is present, but another parent is missing
3139
self._pending_children.setdefault(parent_id,
3140
[]).append(child)
3141
break
3142
else:
3143
# This one is ready to be processed
3144
nodes_to_annotate.append(child)
3145
3146
def _add_annotation(self, revision_id, fulltext, parent_ids,
3147
left_matching_blocks=None):
3148
"""Add an annotation entry.
3149
3150
All parents should already have been annotated.
3151
:return: A list of children that now have their parents satisfied.
3152
"""
3153
a = self._annotated_lines
3154
annotated_parent_lines = [a[p] for p in parent_ids]
3155
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
3156
fulltext, revision_id, left_matching_blocks,
3157
heads_provider=self._get_heads_provider()))
3158
self._annotated_lines[revision_id] = annotated_lines
3159
for p in parent_ids:
3160
ann_children = self._annotate_children[p]
3161
ann_children.remove(revision_id)
3162
if (not ann_children
3163
and p not in self._nodes_to_keep_annotations):
3164
del self._annotated_lines[p]
3165
del self._all_build_details[p]
3166
if p in self._fulltext_contents:
3167
del self._fulltext_contents[p]
3168
# Now that we've added this one, see if there are any pending
3169
# deltas to be done, certainly this parent is finished
3170
nodes_to_annotate = []
3171
for child in self._pending_children.pop(revision_id, []):
3172
self._check_parents(child, nodes_to_annotate)
3173
return nodes_to_annotate
3174
3175
def _get_build_graph(self, revision_id):
3176
"""Get the graphs for building texts and annotations.
3177
3178
The data you need for creating a full text may be different than the
3179
data you need to annotate that text. (At a minimum, you need both
3180
parents to create an annotation, but only need 1 parent to generate the
3181
fulltext.)
3182
3183
:return: A list of (revision_id, index_memo) records, suitable for
3184
passing to read_records_iter to start reading in the raw data fro/
3185
the pack file.
3186
"""
3187
if revision_id in self._annotated_lines:
3188
# Nothing to do
3189
return []
3190
pending = set([revision_id])
3191
records = []
3192
generation = 0
3193
kept_generation = 0
3194
while pending:
3195
# get all pending nodes
3196
generation += 1
3197
this_iteration = pending
3198
build_details = self._knit._index.get_build_details(this_iteration)
3199
self._all_build_details.update(build_details)
3200
# new_nodes = self._knit._index._get_entries(this_iteration)
3201
pending = set()
3202
for rev_id, details in build_details.iteritems():
3203
(index_memo, compression_parent, parents,
3204
record_details) = details
3205
self._revision_id_graph[rev_id] = parents
3206
records.append((rev_id, index_memo))
3207
# Do we actually need to check _annotated_lines?
3208
pending.update(p for p in parents
3209
if p not in self._all_build_details)
3210
if compression_parent:
3211
self._compression_children.setdefault(compression_parent,
3212
[]).append(rev_id)
3213
if parents:
3214
for parent in parents:
3215
self._annotate_children.setdefault(parent,
3216
[]).append(rev_id)
3217
num_gens = generation - kept_generation
3218
if ((num_gens >= self._generations_until_keep)
3219
and len(parents) > 1):
3220
kept_generation = generation
3221
self._nodes_to_keep_annotations.add(rev_id)
3222
3223
missing_versions = this_iteration.difference(build_details.keys())
3224
self._ghosts.update(missing_versions)
3225
for missing_version in missing_versions:
3226
# add a key, no parents
3227
self._revision_id_graph[missing_version] = ()
3228
pending.discard(missing_version) # don't look for it
3229
if self._ghosts.intersection(self._compression_children):
3230
raise KnitCorrupt(
3231
"We cannot have nodes which have a ghost compression parent:\n"
3232
"ghosts: %r\n"
3233
"compression children: %r"
3234
% (self._ghosts, self._compression_children))
3235
# Cleanout anything that depends on a ghost so that we don't wait for
3236
# the ghost to show up
3237
for node in self._ghosts:
3238
if node in self._annotate_children:
3239
# We won't be building this node
3240
del self._annotate_children[node]
3241
# Generally we will want to read the records in reverse order, because
3242
# we find the parent nodes after the children
3243
records.reverse()
3244
return records
3245
3246
def _annotate_records(self, records):
3247
"""Build the annotations for the listed records."""
3248
# We iterate in the order read, rather than a strict order requested
3249
# However, process what we can, and put off to the side things that
3250
# still need parents, cleaning them up when those parents are
3251
# processed.
3252
for (rev_id, record,
3253
digest) in self._knit._data.read_records_iter(records):
3254
if rev_id in self._annotated_lines:
3255
continue
3256
parent_ids = self._revision_id_graph[rev_id]
3257
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3258
details = self._all_build_details[rev_id]
3259
(index_memo, compression_parent, parents,
3260
record_details) = details
3261
nodes_to_annotate = []
3262
# TODO: Remove the punning between compression parents, and
3263
# parent_ids, we should be able to do this without assuming
3264
# the build order
3265
if len(parent_ids) == 0:
3266
# There are no parents for this node, so just add it
3267
# TODO: This probably needs to be decoupled
3268
fulltext_content, delta = self._knit.factory.parse_record(
3269
rev_id, record, record_details, None)
3270
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3271
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3272
parent_ids, left_matching_blocks=None))
1697
3273
else:
1698
for j in js:
1699
# a[i] matches b[j]
1700
if j >= blo:
1701
if j >= bhi:
1702
break
1703
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1704
if k > bestsize:
1705
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1706
j2len = newj2len
1707
1708
# Extend the best by non-junk elements on each end. In particular,
1709
# "popular" non-junk elements aren't in b2j, which greatly speeds
1710
# the inner loop above, but also means "the best" match so far
1711
# doesn't contain any junk *or* popular non-junk elements.
1712
while besti > alo and bestj > blo and \
1713
not isbjunk(b[bestj-1]) and \
1714
a[besti-1] == b[bestj-1]:
1715
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1716
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1717
not isbjunk(b[bestj+bestsize]) and \
1718
a[besti+bestsize] == b[bestj+bestsize]:
1719
bestsize += 1
1720
1721
# Now that we have a wholly interesting match (albeit possibly
1722
# empty!), we may as well suck up the matching junk on each
1723
# side of it too. Can't think of a good reason not to, and it
1724
# saves post-processing the (possibly considerable) expense of
1725
# figuring out what to do with it. In the case of an empty
1726
# interesting match, this is clearly the right thing to do,
1727
# because no other kind of match is possible in the regions.
1728
while besti > alo and bestj > blo and \
1729
isbjunk(b[bestj-1]) and \
1730
a[besti-1] == b[bestj-1]:
1731
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1732
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1733
isbjunk(b[bestj+bestsize]) and \
1734
a[besti+bestsize] == b[bestj+bestsize]:
1735
bestsize = bestsize + 1
1736
1737
return besti, bestj, bestsize
1738
3274
child = (rev_id, parent_ids, record)
3275
# Check if all the parents are present
3276
self._check_parents(child, nodes_to_annotate)
3277
while nodes_to_annotate:
3278
# Should we use a queue here instead of a stack?
3279
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3280
(index_memo, compression_parent, parents,
3281
record_details) = self._all_build_details[rev_id]
3282
if compression_parent is not None:
3283
comp_children = self._compression_children[compression_parent]
3284
if rev_id not in comp_children:
3285
raise AssertionError("%r not in compression children %r"
3286
% (rev_id, comp_children))
3287
# If there is only 1 child, it is safe to reuse this
3288
# content
3289
reuse_content = (len(comp_children) == 1
3290
and compression_parent not in
3291
self._nodes_to_keep_annotations)
3292
if reuse_content:
3293
# Remove it from the cache since it will be changing
3294
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3295
# Make sure to copy the fulltext since it might be
3296
# modified
3297
parent_fulltext = list(parent_fulltext_content.text())
3298
else:
3299
parent_fulltext_content = self._fulltext_contents[compression_parent]
3300
parent_fulltext = parent_fulltext_content.text()
3301
comp_children.remove(rev_id)
3302
fulltext_content, delta = self._knit.factory.parse_record(
3303
rev_id, record, record_details,
3304
parent_fulltext_content,
3305
copy_base_content=(not reuse_content))
3306
fulltext = self._add_fulltext_content(rev_id,
3307
fulltext_content)
3308
blocks = KnitContent.get_line_delta_blocks(delta,
3309
parent_fulltext, fulltext)
3310
else:
3311
fulltext_content = self._knit.factory.parse_fulltext(
3312
record, rev_id)
3313
fulltext = self._add_fulltext_content(rev_id,
3314
fulltext_content)
3315
blocks = None
3316
nodes_to_annotate.extend(
3317
self._add_annotation(rev_id, fulltext, parent_ids,
3318
left_matching_blocks=blocks))
3319
3320
def _get_heads_provider(self):
3321
"""Create a heads provider for resolving ancestry issues."""
3322
if self._heads_provider is not None:
3323
return self._heads_provider
3324
parent_provider = _mod_graph.DictParentsProvider(
3325
self._revision_id_graph)
3326
graph_obj = _mod_graph.Graph(parent_provider)
3327
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3328
self._heads_provider = head_cache
3329
return head_cache
3330
3331
def annotate(self, revision_id):
3332
"""Return the annotated fulltext at the given revision.
3333
3334
:param revision_id: The revision id for this file
3335
"""
3336
records = self._get_build_graph(revision_id)
3337
if revision_id in self._ghosts:
3338
raise errors.RevisionNotPresent(revision_id, self._knit)
3339
self._annotate_records(records)
3340
return self._annotated_lines[revision_id]
3341
3342
3343
try:
3344
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3345
except ImportError:
3346
from bzrlib._knit_load_data_py import _load_data_py as _load_data
Loggerhead is a web-based interface for Breezy
Version: 2.0.1