« back to all changes in this revision
Viewing changes to bzrlib/knit.py
- browse files at revision 1982.1.1
- compare with another revision
- download diff
- download tarball
- view history from revision 1982.1.1
- Committer: Alexander Belchenko
- Date: 2006-09-05 10:38:46 UTC
- mto: (1711.9.16 jam-integration)
- mto: This revision was merged to the branch mainline in revision 1984.
- Revision ID: bialix@ukr.net-20060905103846-648ced8671a8447b
Explicitly skip tests with file attribute manipulations instead of silently pass
- files added:
- NEWS.developers
- bzrlib/bundle/old
- bzrlib/tests/interversionedfile_implementations
- bzrlib/tests/revisionstore_implementations
- files removed:
- bzrlib/_btree_serializer_c.pyx
- bzrlib/filters
- bzrlib/help_topics
- bzrlib/help_topics/en
- bzrlib/plugins/netrc_credential_store
- bzrlib/plugins/netrc_credential_store/tests
- bzrlib/repofmt
- bzrlib/smart
- bzrlib/tests/commands
- bzrlib/tests/ftp_server
- bzrlib/tests/inventory_implementations
- bzrlib/tests/per_interbranch
- bzrlib/tests/per_lock
- bzrlib/tests/per_repository_reference
- bzrlib/tests/ssl_certs
- bzrlib/transport/ftp
- 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/packaging
- files renamed:
- doc/developers/HACKING.txt => HACKING
- bzrlib/deprecated_graph.py => bzrlib/graph.py
- bzrlib/_patiencediff_py.py => bzrlib/patiencediff.py
- bzrlib/tests/http_utils.py => bzrlib/tests/HTTPTestUtil.py
- bzrlib/tests/blackbox/test_send.py => bzrlib/tests/blackbox/test_bundle.py
- bzrlib/tests/per_repository/ => bzrlib/tests/repository_implementations/
- bzrlib/tests/test_commands.py => bzrlib/tests/test_command.py
- bzrlib/tests/test_deprecated_graph.py => bzrlib/tests/test_graph.py
- bzrlib/tests/test_revisionspec.py => bzrlib/tests/test_revisionnamespaces.py
- bzrlib/transport/ftp/__init__.py => bzrlib/transport/ftp.py
- bzrlib/win32utils.py => bzrlib/win32console.py
- bzrlib/xml8.py => bzrlib/xml5.py
- bzrlib/help_topics/en/configuration.txt => doc/configuration.txt
- doc/en/user-guide/index.txt => doc/index.txt
- doc/en/user-guide/plugins.txt => doc/plugins.txt
- doc/en/user-guide/setting_up_email.txt => doc/setting_up_email.txt
- doc/en/user-guide/specifying_revisions.txt => doc/specifying_revisions.txt
- doc/en/tutorials/tutorial.txt => doc/tutorial.txt
- doc/en/user-guide/using_aliases.txt => doc/using_aliases.txt
- files modified:
- .bzrignore
added
removed
bzrlib/knit.py
1
# Copyright (C) 2005, 2006, 2007, 2008 Canonical Ltd
1
# Copyright (C) 2005, 2006 by Canonical Ltd
2
# Written by Martin Pool.
3
# Modified by Johan Rydberg <jrydberg@gnu.org>
4
# Modified by Robert Collins <robert.collins@canonical.com>
5
# Modified by Aaron Bentley <aaron.bentley@utoronto.ca>
2
6
#
3
7
# This program is free software; you can redistribute it and/or modify
4
8
# it under the terms of the GNU General Public License as published by
12
16
#
13
17
# You should have received a copy of the GNU General Public License
14
18
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
16
20
17
21
"""Knit versionedfile implementation.
18
22
20
24
updates.
21
25
22
26
Knit file layout:
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
27
lifeless: the data file is made up of "delta records". each delta record has a delta header
28
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
29
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
30
end-marker; simply "end VERSION"
27
31
28
32
delta can be line or full contents.a
35
39
130,130,2
36
40
8 if elt.get('executable') == 'yes':
37
41
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
39
43
40
44
41
45
whats in an index:
51
55
52
56
"""
53
57
58
# TODOS:
59
# 10:16 < lifeless> make partial index writes safe
60
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
61
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
62
# always' approach.
63
# move sha1 out of the content so that join is faster at verifying parents
64
# record content length ?
65
54
66
67
from copy import copy
55
68
from cStringIO import StringIO
69
import difflib
56
70
from itertools import izip, chain
57
71
import operator
58
72
import os
59
73
import sys
74
import warnings
60
75
61
from bzrlib.lazy_import import lazy_import
62
lazy_import(globals(), """
63
from bzrlib import (
64
annotate,
65
debug,
66
diff,
67
graph as _mod_graph,
68
index as _mod_index,
69
lru_cache,
70
pack,
71
progress,
72
trace,
73
tsort,
74
tuned_gzip,
75
)
76
""")
77
from bzrlib import (
76
import bzrlib
77
from bzrlib import (
78
cache_utf8,
78
79
errors,
79
osutils,
80
patiencediff,
81
)
82
from bzrlib.errors import (
83
FileExists,
84
NoSuchFile,
85
KnitError,
86
InvalidRevisionId,
87
KnitCorrupt,
88
KnitHeaderError,
89
RevisionNotPresent,
90
RevisionAlreadyPresent,
91
SHA1KnitCorrupt,
92
)
93
from bzrlib.osutils import (
94
contains_whitespace,
95
contains_linebreaks,
96
sha_string,
97
sha_strings,
98
split_lines,
99
)
100
from bzrlib.versionedfile import (
101
AbsentContentFactory,
102
adapter_registry,
103
ConstantMapper,
104
ContentFactory,
105
ChunkedContentFactory,
106
sort_groupcompress,
107
VersionedFile,
108
VersionedFiles,
109
)
80
)
81
from bzrlib.errors import FileExists, NoSuchFile, KnitError, \
82
InvalidRevisionId, KnitCorrupt, KnitHeaderError, \
83
RevisionNotPresent, RevisionAlreadyPresent
84
from bzrlib.tuned_gzip import GzipFile
85
from bzrlib.trace import mutter
86
from bzrlib.osutils import contains_whitespace, contains_linebreaks, \
87
sha_strings
88
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
89
from bzrlib.tsort import topo_sort
90
import bzrlib.weave
91
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
110
92
111
93
112
94
# TODO: Split out code specific to this format into an associated object.
124
106
125
107
DATA_SUFFIX = '.knit'
126
108
INDEX_SUFFIX = '.kndx'
127
_STREAM_MIN_BUFFER_SIZE = 5*1024*1024
128
129
130
class KnitAdapter(object):
131
"""Base class for knit record adaption."""
132
133
def __init__(self, basis_vf):
134
"""Create an adapter which accesses full texts from basis_vf.
135
136
:param basis_vf: A versioned file to access basis texts of deltas from.
137
May be None for adapters that do not need to access basis texts.
138
"""
139
self._data = KnitVersionedFiles(None, None)
140
self._annotate_factory = KnitAnnotateFactory()
141
self._plain_factory = KnitPlainFactory()
142
self._basis_vf = basis_vf
143
144
145
class FTAnnotatedToUnannotated(KnitAdapter):
146
"""An adapter from FT annotated knits to unannotated ones."""
147
148
def get_bytes(self, factory):
149
annotated_compressed_bytes = factory._raw_record
150
rec, contents = \
151
self._data._parse_record_unchecked(annotated_compressed_bytes)
152
content = self._annotate_factory.parse_fulltext(contents, rec[1])
153
size, bytes = self._data._record_to_data((rec[1],), rec[3], content.text())
154
return bytes
155
156
157
class DeltaAnnotatedToUnannotated(KnitAdapter):
158
"""An adapter for deltas from annotated to unannotated."""
159
160
def get_bytes(self, factory):
161
annotated_compressed_bytes = factory._raw_record
162
rec, contents = \
163
self._data._parse_record_unchecked(annotated_compressed_bytes)
164
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
165
plain=True)
166
contents = self._plain_factory.lower_line_delta(delta)
167
size, bytes = self._data._record_to_data((rec[1],), rec[3], contents)
168
return bytes
169
170
171
class FTAnnotatedToFullText(KnitAdapter):
172
"""An adapter from FT annotated knits to unannotated ones."""
173
174
def get_bytes(self, factory):
175
annotated_compressed_bytes = factory._raw_record
176
rec, contents = \
177
self._data._parse_record_unchecked(annotated_compressed_bytes)
178
content, delta = self._annotate_factory.parse_record(factory.key[-1],
179
contents, factory._build_details, None)
180
return ''.join(content.text())
181
182
183
class DeltaAnnotatedToFullText(KnitAdapter):
184
"""An adapter for deltas from annotated to unannotated."""
185
186
def get_bytes(self, factory):
187
annotated_compressed_bytes = factory._raw_record
188
rec, contents = \
189
self._data._parse_record_unchecked(annotated_compressed_bytes)
190
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
191
plain=True)
192
compression_parent = factory.parents[0]
193
basis_entry = self._basis_vf.get_record_stream(
194
[compression_parent], 'unordered', True).next()
195
if basis_entry.storage_kind == 'absent':
196
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
197
basis_chunks = basis_entry.get_bytes_as('chunked')
198
basis_lines = osutils.chunks_to_lines(basis_chunks)
199
# Manually apply the delta because we have one annotated content and
200
# one plain.
201
basis_content = PlainKnitContent(basis_lines, compression_parent)
202
basis_content.apply_delta(delta, rec[1])
203
basis_content._should_strip_eol = factory._build_details[1]
204
return ''.join(basis_content.text())
205
206
207
class FTPlainToFullText(KnitAdapter):
208
"""An adapter from FT plain knits to unannotated ones."""
209
210
def get_bytes(self, factory):
211
compressed_bytes = factory._raw_record
212
rec, contents = \
213
self._data._parse_record_unchecked(compressed_bytes)
214
content, delta = self._plain_factory.parse_record(factory.key[-1],
215
contents, factory._build_details, None)
216
return ''.join(content.text())
217
218
219
class DeltaPlainToFullText(KnitAdapter):
220
"""An adapter for deltas from annotated to unannotated."""
221
222
def get_bytes(self, factory):
223
compressed_bytes = factory._raw_record
224
rec, contents = \
225
self._data._parse_record_unchecked(compressed_bytes)
226
delta = self._plain_factory.parse_line_delta(contents, rec[1])
227
compression_parent = factory.parents[0]
228
# XXX: string splitting overhead.
229
basis_entry = self._basis_vf.get_record_stream(
230
[compression_parent], 'unordered', True).next()
231
if basis_entry.storage_kind == 'absent':
232
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
233
basis_chunks = basis_entry.get_bytes_as('chunked')
234
basis_lines = osutils.chunks_to_lines(basis_chunks)
235
basis_content = PlainKnitContent(basis_lines, compression_parent)
236
# Manually apply the delta because we have one annotated content and
237
# one plain.
238
content, _ = self._plain_factory.parse_record(rec[1], contents,
239
factory._build_details, basis_content)
240
return ''.join(content.text())
241
242
243
class KnitContentFactory(ContentFactory):
244
"""Content factory for streaming from knits.
245
246
:seealso ContentFactory:
247
"""
248
249
def __init__(self, key, parents, build_details, sha1, raw_record,
250
annotated, knit=None, network_bytes=None):
251
"""Create a KnitContentFactory for key.
252
253
:param key: The key.
254
:param parents: The parents.
255
:param build_details: The build details as returned from
256
get_build_details.
257
:param sha1: The sha1 expected from the full text of this object.
258
:param raw_record: The bytes of the knit data from disk.
259
:param annotated: True if the raw data is annotated.
260
:param network_bytes: None to calculate the network bytes on demand,
261
not-none if they are already known.
262
"""
263
ContentFactory.__init__(self)
264
self.sha1 = sha1
265
self.key = key
266
self.parents = parents
267
if build_details[0] == 'line-delta':
268
kind = 'delta'
269
else:
270
kind = 'ft'
271
if annotated:
272
annotated_kind = 'annotated-'
273
else:
274
annotated_kind = ''
275
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
276
self._raw_record = raw_record
277
self._network_bytes = network_bytes
278
self._build_details = build_details
279
self._knit = knit
280
281
def _create_network_bytes(self):
282
"""Create a fully serialised network version for transmission."""
283
# storage_kind, key, parents, Noeol, raw_record
284
key_bytes = '\x00'.join(self.key)
285
if self.parents is None:
286
parent_bytes = 'None:'
287
else:
288
parent_bytes = '\t'.join('\x00'.join(key) for key in self.parents)
289
if self._build_details[1]:
290
noeol = 'N'
291
else:
292
noeol = ' '
293
network_bytes = "%s\n%s\n%s\n%s%s" % (self.storage_kind, key_bytes,
294
parent_bytes, noeol, self._raw_record)
295
self._network_bytes = network_bytes
296
297
def get_bytes_as(self, storage_kind):
298
if storage_kind == self.storage_kind:
299
if self._network_bytes is None:
300
self._create_network_bytes()
301
return self._network_bytes
302
if ('-ft-' in self.storage_kind and
303
storage_kind in ('chunked', 'fulltext')):
304
adapter_key = (self.storage_kind, 'fulltext')
305
adapter_factory = adapter_registry.get(adapter_key)
306
adapter = adapter_factory(None)
307
bytes = adapter.get_bytes(self)
308
if storage_kind == 'chunked':
309
return [bytes]
310
else:
311
return bytes
312
if self._knit is not None:
313
# Not redundant with direct conversion above - that only handles
314
# fulltext cases.
315
if storage_kind == 'chunked':
316
return self._knit.get_lines(self.key[0])
317
elif storage_kind == 'fulltext':
318
return self._knit.get_text(self.key[0])
319
raise errors.UnavailableRepresentation(self.key, storage_kind,
320
self.storage_kind)
321
322
323
class LazyKnitContentFactory(ContentFactory):
324
"""A ContentFactory which can either generate full text or a wire form.
325
326
:seealso ContentFactory:
327
"""
328
329
def __init__(self, key, parents, generator, first):
330
"""Create a LazyKnitContentFactory.
331
332
:param key: The key of the record.
333
:param parents: The parents of the record.
334
:param generator: A _ContentMapGenerator containing the record for this
335
key.
336
:param first: Is this the first content object returned from generator?
337
if it is, its storage kind is knit-delta-closure, otherwise it is
338
knit-delta-closure-ref
339
"""
340
self.key = key
341
self.parents = parents
342
self.sha1 = None
343
self._generator = generator
344
self.storage_kind = "knit-delta-closure"
345
if not first:
346
self.storage_kind = self.storage_kind + "-ref"
347
self._first = first
348
349
def get_bytes_as(self, storage_kind):
350
if storage_kind == self.storage_kind:
351
if self._first:
352
return self._generator._wire_bytes()
353
else:
354
# all the keys etc are contained in the bytes returned in the
355
# first record.
356
return ''
357
if storage_kind in ('chunked', 'fulltext'):
358
chunks = self._generator._get_one_work(self.key).text()
359
if storage_kind == 'chunked':
360
return chunks
361
else:
362
return ''.join(chunks)
363
raise errors.UnavailableRepresentation(self.key, storage_kind,
364
self.storage_kind)
365
366
367
def knit_delta_closure_to_records(storage_kind, bytes, line_end):
368
"""Convert a network record to a iterator over stream records.
369
370
:param storage_kind: The storage kind of the record.
371
Must be 'knit-delta-closure'.
372
:param bytes: The bytes of the record on the network.
373
"""
374
generator = _NetworkContentMapGenerator(bytes, line_end)
375
return generator.get_record_stream()
376
377
378
def knit_network_to_record(storage_kind, bytes, line_end):
379
"""Convert a network record to a record object.
380
381
:param storage_kind: The storage kind of the record.
382
:param bytes: The bytes of the record on the network.
383
"""
384
start = line_end
385
line_end = bytes.find('\n', start)
386
key = tuple(bytes[start:line_end].split('\x00'))
387
start = line_end + 1
388
line_end = bytes.find('\n', start)
389
parent_line = bytes[start:line_end]
390
if parent_line == 'None:':
391
parents = None
392
else:
393
parents = tuple(
394
[tuple(segment.split('\x00')) for segment in parent_line.split('\t')
395
if segment])
396
start = line_end + 1
397
noeol = bytes[start] == 'N'
398
if 'ft' in storage_kind:
399
method = 'fulltext'
400
else:
401
method = 'line-delta'
402
build_details = (method, noeol)
403
start = start + 1
404
raw_record = bytes[start:]
405
annotated = 'annotated' in storage_kind
406
return [KnitContentFactory(key, parents, build_details, None, raw_record,
407
annotated, network_bytes=bytes)]
408
109
409
110
410
111
class KnitContent(object):
411
"""Content of a knit version to which deltas can be applied.
412
413
This is always stored in memory as a list of lines with \n at the end,
414
plus a flag saying if the final ending is really there or not, because that
415
corresponds to the on-disk knit representation.
416
"""
417
418
def __init__(self):
419
self._should_strip_eol = False
420
421
def apply_delta(self, delta, new_version_id):
422
"""Apply delta to this object to become new_version_id."""
423
raise NotImplementedError(self.apply_delta)
112
"""Content of a knit version to which deltas can be applied."""
113
114
def __init__(self, lines):
115
self._lines = lines
116
117
def annotate_iter(self):
118
"""Yield tuples of (origin, text) for each content line."""
119
for origin, text in self._lines:
120
yield origin, text
121
122
def annotate(self):
123
"""Return a list of (origin, text) tuples."""
124
return list(self.annotate_iter())
424
125
425
126
def line_delta_iter(self, new_lines):
426
127
"""Generate line-based delta from this content to new_lines."""
427
new_texts = new_lines.text()
428
old_texts = self.text()
429
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
430
for tag, i1, i2, j1, j2 in s.get_opcodes():
431
if tag == 'equal':
128
new_texts = [text for origin, text in new_lines._lines]
129
old_texts = [text for origin, text in self._lines]
130
s = KnitSequenceMatcher(None, old_texts, new_texts)
131
for op in s.get_opcodes():
132
if op[0] == 'equal':
432
133
continue
433
# ofrom, oto, length, data
434
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
134
# ofrom oto length data
135
yield (op[1], op[2], op[4]-op[3], new_lines._lines[op[3]:op[4]])
435
136
436
137
def line_delta(self, new_lines):
437
138
return list(self.line_delta_iter(new_lines))
438
139
439
@staticmethod
440
def get_line_delta_blocks(knit_delta, source, target):
441
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
442
target_len = len(target)
443
s_pos = 0
444
t_pos = 0
445
for s_begin, s_end, t_len, new_text in knit_delta:
446
true_n = s_begin - s_pos
447
n = true_n
448
if n > 0:
449
# knit deltas do not provide reliable info about whether the
450
# last line of a file matches, due to eol handling.
451
if source[s_pos + n -1] != target[t_pos + n -1]:
452
n-=1
453
if n > 0:
454
yield s_pos, t_pos, n
455
t_pos += t_len + true_n
456
s_pos = s_end
457
n = target_len - t_pos
458
if n > 0:
459
if source[s_pos + n -1] != target[t_pos + n -1]:
460
n-=1
461
if n > 0:
462
yield s_pos, t_pos, n
463
yield s_pos + (target_len - t_pos), target_len, 0
464
465
466
class AnnotatedKnitContent(KnitContent):
467
"""Annotated content."""
468
469
def __init__(self, lines):
470
KnitContent.__init__(self)
471
self._lines = lines
472
473
def annotate(self):
474
"""Return a list of (origin, text) for each content line."""
475
lines = self._lines[:]
476
if self._should_strip_eol:
477
origin, last_line = lines[-1]
478
lines[-1] = (origin, last_line.rstrip('\n'))
479
return lines
480
481
def apply_delta(self, delta, new_version_id):
482
"""Apply delta to this object to become new_version_id."""
483
offset = 0
484
lines = self._lines
485
for start, end, count, delta_lines in delta:
486
lines[offset+start:offset+end] = delta_lines
487
offset = offset + (start - end) + count
488
489
def text(self):
490
try:
491
lines = [text for origin, text in self._lines]
492
except ValueError, e:
493
# most commonly (only?) caused by the internal form of the knit
494
# missing annotation information because of a bug - see thread
495
# around 20071015
496
raise KnitCorrupt(self,
497
"line in annotated knit missing annotation information: %s"
498
% (e,))
499
if self._should_strip_eol:
500
lines[-1] = lines[-1].rstrip('\n')
501
return lines
502
503
def copy(self):
504
return AnnotatedKnitContent(self._lines[:])
505
506
507
class PlainKnitContent(KnitContent):
508
"""Unannotated content.
509
510
When annotate[_iter] is called on this content, the same version is reported
511
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
512
objects.
513
"""
514
515
def __init__(self, lines, version_id):
516
KnitContent.__init__(self)
517
self._lines = lines
518
self._version_id = version_id
519
520
def annotate(self):
521
"""Return a list of (origin, text) for each content line."""
522
return [(self._version_id, line) for line in self._lines]
523
524
def apply_delta(self, delta, new_version_id):
525
"""Apply delta to this object to become new_version_id."""
526
offset = 0
527
lines = self._lines
528
for start, end, count, delta_lines in delta:
529
lines[offset+start:offset+end] = delta_lines
530
offset = offset + (start - end) + count
531
self._version_id = new_version_id
532
533
def copy(self):
534
return PlainKnitContent(self._lines[:], self._version_id)
535
536
def text(self):
537
lines = self._lines
538
if self._should_strip_eol:
539
lines = lines[:]
540
lines[-1] = lines[-1].rstrip('\n')
541
return lines
140
def text(self):
141
return [text for origin, text in self._lines]
142
143
def copy(self):
144
return KnitContent(self._lines[:])
542
145
543
146
544
147
class _KnitFactory(object):
545
"""Base class for common Factory functions."""
546
547
def parse_record(self, version_id, record, record_details,
548
base_content, copy_base_content=True):
549
"""Parse a record into a full content object.
550
551
:param version_id: The official version id for this content
552
:param record: The data returned by read_records_iter()
553
:param record_details: Details about the record returned by
554
get_build_details
555
:param base_content: If get_build_details returns a compression_parent,
556
you must return a base_content here, else use None
557
:param copy_base_content: When building from the base_content, decide
558
you can either copy it and return a new object, or modify it in
559
place.
560
:return: (content, delta) A Content object and possibly a line-delta,
561
delta may be None
562
"""
563
method, noeol = record_details
564
if method == 'line-delta':
565
if copy_base_content:
566
content = base_content.copy()
567
else:
568
content = base_content
569
delta = self.parse_line_delta(record, version_id)
570
content.apply_delta(delta, version_id)
571
else:
572
content = self.parse_fulltext(record, version_id)
573
delta = None
574
content._should_strip_eol = noeol
575
return (content, delta)
148
"""Base factory for creating content objects."""
149
150
def make(self, lines, version):
151
num_lines = len(lines)
152
return KnitContent(zip([version] * num_lines, lines))
576
153
577
154
578
155
class KnitAnnotateFactory(_KnitFactory):
580
157
581
158
annotated = True
582
159
583
def make(self, lines, version_id):
584
num_lines = len(lines)
585
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
586
587
def parse_fulltext(self, content, version_id):
160
def parse_fulltext(self, content, version):
588
161
"""Convert fulltext to internal representation
589
162
590
163
fulltext content is of the format
592
165
internal representation is of the format:
593
166
(revid, plaintext)
594
167
"""
595
# TODO: jam 20070209 The tests expect this to be returned as tuples,
596
# but the code itself doesn't really depend on that.
597
# Figure out a way to not require the overhead of turning the
598
# list back into tuples.
599
lines = [tuple(line.split(' ', 1)) for line in content]
600
return AnnotatedKnitContent(lines)
168
decode_utf8 = cache_utf8.decode
169
lines = []
170
for line in content:
171
origin, text = line.split(' ', 1)
172
lines.append((decode_utf8(origin), text))
173
return KnitContent(lines)
601
174
602
175
def parse_line_delta_iter(self, lines):
603
return iter(self.parse_line_delta(lines))
176
for result_item in self.parse_line_delta[lines]:
177
yield result_item
604
178
605
def parse_line_delta(self, lines, version_id, plain=False):
179
def parse_line_delta(self, lines, version):
606
180
"""Convert a line based delta into internal representation.
607
181
608
182
line delta is in the form of:
611
185
revid(utf8) newline\n
612
186
internal representation is
613
187
(start, end, count, [1..count tuples (revid, newline)])
614
615
:param plain: If True, the lines are returned as a plain
616
list without annotations, not as a list of (origin, content) tuples, i.e.
617
(start, end, count, [1..count newline])
618
188
"""
189
decode_utf8 = cache_utf8.decode
619
190
result = []
620
191
lines = iter(lines)
621
192
next = lines.next
622
623
cache = {}
624
def cache_and_return(line):
625
origin, text = line.split(' ', 1)
626
return cache.setdefault(origin, origin), text
627
628
193
# walk through the lines parsing.
629
# Note that the plain test is explicitly pulled out of the
630
# loop to minimise any performance impact
631
if plain:
632
for header in lines:
633
start, end, count = [int(n) for n in header.split(',')]
634
contents = [next().split(' ', 1)[1] for i in xrange(count)]
635
result.append((start, end, count, contents))
636
else:
637
for header in lines:
638
start, end, count = [int(n) for n in header.split(',')]
639
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
640
result.append((start, end, count, contents))
641
return result
642
643
def get_fulltext_content(self, lines):
644
"""Extract just the content lines from a fulltext."""
645
return (line.split(' ', 1)[1] for line in lines)
646
647
def get_linedelta_content(self, lines):
648
"""Extract just the content from a line delta.
649
650
This doesn't return all of the extra information stored in a delta.
651
Only the actual content lines.
652
"""
653
lines = iter(lines)
654
next = lines.next
655
194
for header in lines:
656
header = header.split(',')
657
count = int(header[2])
658
for i in xrange(count):
195
start, end, count = [int(n) for n in header.split(',')]
196
contents = []
197
remaining = count
198
while remaining:
659
199
origin, text = next().split(' ', 1)
660
yield text
200
remaining -= 1
201
contents.append((decode_utf8(origin), text))
202
result.append((start, end, count, contents))
203
return result
661
204
662
205
def lower_fulltext(self, content):
663
206
"""convert a fulltext content record into a serializable form.
664
207
665
208
see parse_fulltext which this inverts.
666
209
"""
667
# TODO: jam 20070209 We only do the caching thing to make sure that
668
# the origin is a valid utf-8 line, eventually we could remove it
669
return ['%s %s' % (o, t) for o, t in content._lines]
210
encode_utf8 = cache_utf8.encode
211
return ['%s %s' % (encode_utf8(o), t) for o, t in content._lines]
670
212
671
213
def lower_line_delta(self, delta):
672
214
"""convert a delta into a serializable form.
673
215
674
216
See parse_line_delta which this inverts.
675
217
"""
676
# TODO: jam 20070209 We only do the caching thing to make sure that
677
# the origin is a valid utf-8 line, eventually we could remove it
218
encode_utf8 = cache_utf8.encode
678
219
out = []
679
220
for start, end, c, lines in delta:
680
221
out.append('%d,%d,%d\n' % (start, end, c))
681
out.extend(origin + ' ' + text
222
out.extend(encode_utf8(origin) + ' ' + text
682
223
for origin, text in lines)
683
224
return out
684
225
685
def annotate(self, knit, key):
686
content = knit._get_content(key)
687
# adjust for the fact that serialised annotations are only key suffixes
688
# for this factory.
689
if type(key) == tuple:
690
prefix = key[:-1]
691
origins = content.annotate()
692
result = []
693
for origin, line in origins:
694
result.append((prefix + (origin,), line))
695
return result
696
else:
697
# XXX: This smells a bit. Why would key ever be a non-tuple here?
698
# Aren't keys defined to be tuples? -- spiv 20080618
699
return content.annotate()
700
701
226
702
227
class KnitPlainFactory(_KnitFactory):
703
228
"""Factory for creating plain Content objects."""
704
229
705
230
annotated = False
706
231
707
def make(self, lines, version_id):
708
return PlainKnitContent(lines, version_id)
709
710
def parse_fulltext(self, content, version_id):
232
def parse_fulltext(self, content, version):
711
233
"""This parses an unannotated fulltext.
712
234
713
235
Note that this is not a noop - the internal representation
714
236
has (versionid, line) - its just a constant versionid.
715
237
"""
716
return self.make(content, version_id)
238
return self.make(content, version)
717
239
718
def parse_line_delta_iter(self, lines, version_id):
719
cur = 0
720
num_lines = len(lines)
721
while cur < num_lines:
722
header = lines[cur]
723
cur += 1
240
def parse_line_delta_iter(self, lines, version):
241
while lines:
242
header = lines.pop(0)
724
243
start, end, c = [int(n) for n in header.split(',')]
725
yield start, end, c, lines[cur:cur+c]
726
cur += c
727
728
def parse_line_delta(self, lines, version_id):
729
return list(self.parse_line_delta_iter(lines, version_id))
730
731
def get_fulltext_content(self, lines):
732
"""Extract just the content lines from a fulltext."""
733
return iter(lines)
734
735
def get_linedelta_content(self, lines):
736
"""Extract just the content from a line delta.
737
738
This doesn't return all of the extra information stored in a delta.
739
Only the actual content lines.
740
"""
741
lines = iter(lines)
742
next = lines.next
743
for header in lines:
744
header = header.split(',')
745
count = int(header[2])
746
for i in xrange(count):
747
yield next()
748
244
yield start, end, c, zip([version] * c, lines[:c])
245
del lines[:c]
246
247
def parse_line_delta(self, lines, version):
248
return list(self.parse_line_delta_iter(lines, version))
249
749
250
def lower_fulltext(self, content):
750
251
return content.text()
751
252
753
254
out = []
754
255
for start, end, c, lines in delta:
755
256
out.append('%d,%d,%d\n' % (start, end, c))
756
out.extend(lines)
257
out.extend([text for origin, text in lines])
757
258
return out
758
259
759
def annotate(self, knit, key):
760
annotator = _KnitAnnotator(knit)
761
return annotator.annotate(key)
762
763
764
765
def make_file_factory(annotated, mapper):
766
"""Create a factory for creating a file based KnitVersionedFiles.
767
768
This is only functional enough to run interface tests, it doesn't try to
769
provide a full pack environment.
770
771
:param annotated: knit annotations are wanted.
772
:param mapper: The mapper from keys to paths.
773
"""
774
def factory(transport):
775
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
776
access = _KnitKeyAccess(transport, mapper)
777
return KnitVersionedFiles(index, access, annotated=annotated)
778
return factory
779
780
781
def make_pack_factory(graph, delta, keylength):
782
"""Create a factory for creating a pack based VersionedFiles.
783
784
This is only functional enough to run interface tests, it doesn't try to
785
provide a full pack environment.
786
787
:param graph: Store a graph.
788
:param delta: Delta compress contents.
789
:param keylength: How long should keys be.
790
"""
791
def factory(transport):
792
parents = graph or delta
793
ref_length = 0
794
if graph:
795
ref_length += 1
260
261
def make_empty_knit(transport, relpath):
262
"""Construct a empty knit at the specified location."""
263
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
264
k._data._open_file()
265
266
267
class KnitVersionedFile(VersionedFile):
268
"""Weave-like structure with faster random access.
269
270
A knit stores a number of texts and a summary of the relationships
271
between them. Texts are identified by a string version-id. Texts
272
are normally stored and retrieved as a series of lines, but can
273
also be passed as single strings.
274
275
Lines are stored with the trailing newline (if any) included, to
276
avoid special cases for files with no final newline. Lines are
277
composed of 8-bit characters, not unicode. The combination of
278
these approaches should mean any 'binary' file can be safely
279
stored and retrieved.
280
"""
281
282
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
283
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
284
create=False):
285
"""Construct a knit at location specified by relpath.
286
287
:param create: If not True, only open an existing knit.
288
"""
289
if deprecated_passed(basis_knit):
290
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
291
" deprecated as of bzr 0.9.",
292
DeprecationWarning, stacklevel=2)
293
if access_mode is None:
294
access_mode = 'w'
295
super(KnitVersionedFile, self).__init__(access_mode)
296
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
297
self.transport = transport
298
self.filename = relpath
299
self.factory = factory or KnitAnnotateFactory()
300
self.writable = (access_mode == 'w')
301
self.delta = delta
302
303
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
304
access_mode, create=create, file_mode=file_mode)
305
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
306
access_mode, create=create and not len(self), file_mode=file_mode)
307
308
def __repr__(self):
309
return '%s(%s)' % (self.__class__.__name__,
310
self.transport.abspath(self.filename))
311
312
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
313
"""See VersionedFile._add_delta()."""
314
self._check_add(version_id, []) # should we check the lines ?
315
self._check_versions_present(parents)
316
present_parents = []
317
ghosts = []
318
parent_texts = {}
319
for parent in parents:
320
if not self.has_version(parent):
321
ghosts.append(parent)
322
else:
323
present_parents.append(parent)
324
325
if delta_parent is None:
326
# reconstitute as full text.
327
assert len(delta) == 1 or len(delta) == 0
328
if len(delta):
329
assert delta[0][0] == 0
330
assert delta[0][1] == 0, delta[0][1]
331
return super(KnitVersionedFile, self)._add_delta(version_id,
332
parents,
333
delta_parent,
334
sha1,
335
noeol,
336
delta)
337
338
digest = sha1
339
340
options = []
341
if noeol:
342
options.append('no-eol')
343
344
if delta_parent is not None:
345
# determine the current delta chain length.
346
# To speed the extract of texts the delta chain is limited
347
# to a fixed number of deltas. This should minimize both
348
# I/O and the time spend applying deltas.
349
count = 0
350
delta_parents = [delta_parent]
351
while count < 25:
352
parent = delta_parents[0]
353
method = self._index.get_method(parent)
354
if method == 'fulltext':
355
break
356
delta_parents = self._index.get_parents(parent)
357
count = count + 1
358
if method == 'line-delta':
359
# did not find a fulltext in the delta limit.
360
# just do a normal insertion.
361
return super(KnitVersionedFile, self)._add_delta(version_id,
362
parents,
363
delta_parent,
364
sha1,
365
noeol,
366
delta)
367
368
options.append('line-delta')
369
store_lines = self.factory.lower_line_delta(delta)
370
371
where, size = self._data.add_record(version_id, digest, store_lines)
372
self._index.add_version(version_id, options, where, size, parents)
373
374
def _add_raw_records(self, records, data):
375
"""Add all the records 'records' with data pre-joined in 'data'.
376
377
:param records: A list of tuples(version_id, options, parents, size).
378
:param data: The data for the records. When it is written, the records
379
are adjusted to have pos pointing into data by the sum of
380
the preceding records sizes.
381
"""
382
# write all the data
383
pos = self._data.add_raw_record(data)
384
offset = 0
385
index_entries = []
386
for (version_id, options, parents, size) in records:
387
index_entries.append((version_id, options, pos+offset,
388
size, parents))
389
if self._data._do_cache:
390
self._data._cache[version_id] = data[offset:offset+size]
391
offset += size
392
self._index.add_versions(index_entries)
393
394
def enable_cache(self):
395
"""Start caching data for this knit"""
396
self._data.enable_cache()
397
398
def clear_cache(self):
399
"""Clear the data cache only."""
400
self._data.clear_cache()
401
402
def copy_to(self, name, transport):
403
"""See VersionedFile.copy_to()."""
404
# copy the current index to a temp index to avoid racing with local
405
# writes
406
transport.put(name + INDEX_SUFFIX + '.tmp', self.transport.get(self._index._filename),)
407
# copy the data file
408
f = self._data._open_file()
409
try:
410
transport.put(name + DATA_SUFFIX, f)
411
finally:
412
f.close()
413
# move the copied index into place
414
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
415
416
def create_empty(self, name, transport, mode=None):
417
return KnitVersionedFile(name, transport, factory=self.factory, delta=self.delta, create=True)
418
419
def _fix_parents(self, version, new_parents):
420
"""Fix the parents list for version.
421
422
This is done by appending a new version to the index
423
with identical data except for the parents list.
424
the parents list must be a superset of the current
425
list.
426
"""
427
current_values = self._index._cache[version]
428
assert set(current_values[4]).difference(set(new_parents)) == set()
429
self._index.add_version(version,
430
current_values[1],
431
current_values[2],
432
current_values[3],
433
new_parents)
434
435
def get_delta(self, version_id):
436
"""Get a delta for constructing version from some other version."""
437
if not self.has_version(version_id):
438
raise RevisionNotPresent(version_id, self.filename)
439
440
parents = self.get_parents(version_id)
441
if len(parents):
442
parent = parents[0]
443
else:
444
parent = None
445
data_pos, data_size = self._index.get_position(version_id)
446
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
447
version_idx = self._index.lookup(version_id)
448
noeol = 'no-eol' in self._index.get_options(version_id)
449
if 'fulltext' == self._index.get_method(version_id):
450
new_content = self.factory.parse_fulltext(data, version_idx)
451
if parent is not None:
452
reference_content = self._get_content(parent)
453
old_texts = reference_content.text()
454
else:
455
old_texts = []
456
new_texts = new_content.text()
457
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
458
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
459
else:
460
delta = self.factory.parse_line_delta(data, version_idx)
461
return parent, sha1, noeol, delta
462
463
def get_graph_with_ghosts(self):
464
"""See VersionedFile.get_graph_with_ghosts()."""
465
graph_items = self._index.get_graph()
466
return dict(graph_items)
467
468
def get_sha1(self, version_id):
469
"""See VersionedFile.get_sha1()."""
470
record_map = self._get_record_map([version_id])
471
method, content, digest, next = record_map[version_id]
472
return digest
473
474
@staticmethod
475
def get_suffixes():
476
"""See VersionedFile.get_suffixes()."""
477
return [DATA_SUFFIX, INDEX_SUFFIX]
478
479
def has_ghost(self, version_id):
480
"""True if there is a ghost reference in the file to version_id."""
481
# maybe we have it
482
if self.has_version(version_id):
483
return False
484
# optimisable if needed by memoising the _ghosts set.
485
items = self._index.get_graph()
486
for node, parents in items:
487
for parent in parents:
488
if parent not in self._index._cache:
489
if parent == version_id:
490
return True
491
return False
492
493
def versions(self):
494
"""See VersionedFile.versions."""
495
return self._index.get_versions()
496
497
def has_version(self, version_id):
498
"""See VersionedFile.has_version."""
499
return self._index.has_version(version_id)
500
501
__contains__ = has_version
502
503
def _merge_annotations(self, content, parents, parent_texts={},
504
delta=None, annotated=None):
505
"""Merge annotations for content. This is done by comparing
506
the annotations based on changed to the text.
507
"""
508
if annotated:
509
delta_seq = None
510
for parent_id in parents:
511
merge_content = self._get_content(parent_id, parent_texts)
512
seq = KnitSequenceMatcher(None, merge_content.text(), content.text())
513
if delta_seq is None:
514
# setup a delta seq to reuse.
515
delta_seq = seq
516
for i, j, n in seq.get_matching_blocks():
517
if n == 0:
518
continue
519
# this appears to copy (origin, text) pairs across to the new
520
# content for any line that matches the last-checked parent.
521
# FIXME: save the sequence control data for delta compression
522
# against the most relevant parent rather than rediffing.
523
content._lines[j:j+n] = merge_content._lines[i:i+n]
796
524
if delta:
797
ref_length += 1
798
max_delta_chain = 200
799
else:
800
max_delta_chain = 0
801
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
802
key_elements=keylength)
803
stream = transport.open_write_stream('newpack')
804
writer = pack.ContainerWriter(stream.write)
805
writer.begin()
806
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
807
deltas=delta, add_callback=graph_index.add_nodes)
808
access = _DirectPackAccess({})
809
access.set_writer(writer, graph_index, (transport, 'newpack'))
810
result = KnitVersionedFiles(index, access,
811
max_delta_chain=max_delta_chain)
812
result.stream = stream
813
result.writer = writer
814
return result
815
return factory
816
817
818
def cleanup_pack_knit(versioned_files):
819
versioned_files.stream.close()
820
versioned_files.writer.end()
821
822
823
def _get_total_build_size(self, keys, positions):
824
"""Determine the total bytes to build these keys.
825
826
(helper function because _KnitGraphIndex and _KndxIndex work the same, but
827
don't inherit from a common base.)
828
829
:param keys: Keys that we want to build
830
:param positions: dict of {key, (info, index_memo, comp_parent)} (such
831
as returned by _get_components_positions)
832
:return: Number of bytes to build those keys
833
"""
834
all_build_index_memos = {}
835
build_keys = keys
836
while build_keys:
837
next_keys = set()
838
for key in build_keys:
839
# This is mostly for the 'stacked' case
840
# Where we will be getting the data from a fallback
841
if key not in positions:
525
if not annotated:
526
reference_content = self._get_content(parents[0], parent_texts)
527
new_texts = content.text()
528
old_texts = reference_content.text()
529
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
530
return self._make_line_delta(delta_seq, content)
531
532
def _make_line_delta(self, delta_seq, new_content):
533
"""Generate a line delta from delta_seq and new_content."""
534
diff_hunks = []
535
for op in delta_seq.get_opcodes():
536
if op[0] == 'equal':
842
537
continue
843
_, index_memo, compression_parent = positions[key]
844
all_build_index_memos[key] = index_memo
845
if compression_parent not in all_build_index_memos:
846
next_keys.add(compression_parent)
847
build_keys = next_keys
848
return sum([index_memo[2] for index_memo
849
in all_build_index_memos.itervalues()])
850
851
852
class KnitVersionedFiles(VersionedFiles):
853
"""Storage for many versioned files using knit compression.
854
855
Backend storage is managed by indices and data objects.
856
857
:ivar _index: A _KnitGraphIndex or similar that can describe the
858
parents, graph, compression and data location of entries in this
859
KnitVersionedFiles. Note that this is only the index for
860
*this* vfs; if there are fallbacks they must be queried separately.
861
"""
862
863
def __init__(self, index, data_access, max_delta_chain=200,
864
annotated=False, reload_func=None):
865
"""Create a KnitVersionedFiles with index and data_access.
866
867
:param index: The index for the knit data.
868
:param data_access: The access object to store and retrieve knit
869
records.
870
:param max_delta_chain: The maximum number of deltas to permit during
871
insertion. Set to 0 to prohibit the use of deltas.
872
:param annotated: Set to True to cause annotations to be calculated and
873
stored during insertion.
874
:param reload_func: An function that can be called if we think we need
875
to reload the pack listing and try again. See
876
'bzrlib.repofmt.pack_repo.AggregateIndex' for the signature.
877
"""
878
self._index = index
879
self._access = data_access
880
self._max_delta_chain = max_delta_chain
881
if annotated:
882
self._factory = KnitAnnotateFactory()
883
else:
884
self._factory = KnitPlainFactory()
885
self._fallback_vfs = []
886
self._reload_func = reload_func
887
888
def __repr__(self):
889
return "%s(%r, %r)" % (
890
self.__class__.__name__,
891
self._index,
892
self._access)
893
894
def add_fallback_versioned_files(self, a_versioned_files):
895
"""Add a source of texts for texts not present in this knit.
896
897
:param a_versioned_files: A VersionedFiles object.
898
"""
899
self._fallback_vfs.append(a_versioned_files)
900
901
def add_lines(self, key, parents, lines, parent_texts=None,
902
left_matching_blocks=None, nostore_sha=None, random_id=False,
903
check_content=True):
904
"""See VersionedFiles.add_lines()."""
905
self._index._check_write_ok()
906
self._check_add(key, lines, random_id, check_content)
907
if parents is None:
908
# The caller might pass None if there is no graph data, but kndx
909
# indexes can't directly store that, so we give them
910
# an empty tuple instead.
911
parents = ()
912
return self._add(key, lines, parents,
913
parent_texts, left_matching_blocks, nostore_sha, random_id)
914
915
def _add(self, key, lines, parents, parent_texts,
916
left_matching_blocks, nostore_sha, random_id):
538
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
539
return diff_hunks
540
541
def _get_components_positions(self, version_ids):
542
"""Produce a map of position data for the components of versions.
543
544
This data is intended to be used for retrieving the knit records.
545
546
A dict of version_id to (method, data_pos, data_size, next) is
547
returned.
548
method is the way referenced data should be applied.
549
data_pos is the position of the data in the knit.
550
data_size is the size of the data in the knit.
551
next is the build-parent of the version, or None for fulltexts.
552
"""
553
component_data = {}
554
for version_id in version_ids:
555
cursor = version_id
556
557
while cursor is not None and cursor not in component_data:
558
method = self._index.get_method(cursor)
559
if method == 'fulltext':
560
next = None
561
else:
562
next = self.get_parents(cursor)[0]
563
data_pos, data_size = self._index.get_position(cursor)
564
component_data[cursor] = (method, data_pos, data_size, next)
565
cursor = next
566
return component_data
567
568
def _get_content(self, version_id, parent_texts={}):
569
"""Returns a content object that makes up the specified
570
version."""
571
if not self.has_version(version_id):
572
raise RevisionNotPresent(version_id, self.filename)
573
574
cached_version = parent_texts.get(version_id, None)
575
if cached_version is not None:
576
return cached_version
577
578
text_map, contents_map = self._get_content_maps([version_id])
579
return contents_map[version_id]
580
581
def _check_versions_present(self, version_ids):
582
"""Check that all specified versions are present."""
583
version_ids = set(version_ids)
584
for r in list(version_ids):
585
if self._index.has_version(r):
586
version_ids.remove(r)
587
if version_ids:
588
raise RevisionNotPresent(list(version_ids)[0], self.filename)
589
590
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
591
"""See VersionedFile.add_lines_with_ghosts()."""
592
self._check_add(version_id, lines)
593
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
594
595
def _add_lines(self, version_id, parents, lines, parent_texts):
596
"""See VersionedFile.add_lines."""
597
self._check_add(version_id, lines)
598
self._check_versions_present(parents)
599
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
600
601
def _check_add(self, version_id, lines):
602
"""check that version_id and lines are safe to add."""
603
assert self.writable, "knit is not opened for write"
604
### FIXME escape. RBC 20060228
605
if contains_whitespace(version_id):
606
raise InvalidRevisionId(version_id, self.filename)
607
if self.has_version(version_id):
608
raise RevisionAlreadyPresent(version_id, self.filename)
609
self._check_lines_not_unicode(lines)
610
self._check_lines_are_lines(lines)
611
612
def _add(self, version_id, lines, parents, delta, parent_texts):
917
613
"""Add a set of lines on top of version specified by parents.
918
614
615
If delta is true, compress the text as a line-delta against
616
the first parent.
617
919
618
Any versions not present will be converted into ghosts.
920
619
"""
921
# first thing, if the content is something we don't need to store, find
922
# that out.
923
line_bytes = ''.join(lines)
924
digest = sha_string(line_bytes)
925
if nostore_sha == digest:
926
raise errors.ExistingContent
620
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
621
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
622
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
623
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
624
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
625
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
626
# +1383 0 8.0370 8.0370 +<len>
627
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
628
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
629
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
630
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
927
631
928
632
present_parents = []
633
ghosts = []
929
634
if parent_texts is None:
930
635
parent_texts = {}
931
# Do a single query to ascertain parent presence; we only compress
932
# against parents in the same kvf.
933
present_parent_map = self._index.get_parent_map(parents)
934
636
for parent in parents:
935
if parent in present_parent_map:
637
if not self.has_version(parent):
638
ghosts.append(parent)
639
else:
936
640
present_parents.append(parent)
937
641
938
# Currently we can only compress against the left most present parent.
939
if (len(present_parents) == 0 or
940
present_parents[0] != parents[0]):
642
if delta and not len(present_parents):
941
643
delta = False
942
else:
943
# To speed the extract of texts the delta chain is limited
944
# to a fixed number of deltas. This should minimize both
945
# I/O and the time spend applying deltas.
946
delta = self._check_should_delta(present_parents[0])
947
644
948
text_length = len(line_bytes)
645
digest = sha_strings(lines)
949
646
options = []
950
647
if lines:
951
648
if lines[-1][-1] != '\n':
952
# copy the contents of lines.
953
lines = lines[:]
954
649
options.append('no-eol')
955
650
lines[-1] = lines[-1] + '\n'
956
line_bytes += '\n'
957
958
for element in key[:-1]:
959
if type(element) != str:
960
raise TypeError("key contains non-strings: %r" % (key,))
961
if key[-1] is None:
962
key = key[:-1] + ('sha1:' + digest,)
963
elif type(key[-1]) != str:
964
raise TypeError("key contains non-strings: %r" % (key,))
965
# Knit hunks are still last-element only
966
version_id = key[-1]
967
content = self._factory.make(lines, version_id)
968
if 'no-eol' in options:
969
# Hint to the content object that its text() call should strip the
970
# EOL.
971
content._should_strip_eol = True
972
if delta or (self._factory.annotated and len(present_parents) > 0):
973
# Merge annotations from parent texts if needed.
974
delta_hunks = self._merge_annotations(content, present_parents,
975
parent_texts, delta, self._factory.annotated,
976
left_matching_blocks)
651
652
if len(present_parents) and delta:
653
# To speed the extract of texts the delta chain is limited
654
# to a fixed number of deltas. This should minimize both
655
# I/O and the time spend applying deltas.
656
count = 0
657
delta_parents = present_parents
658
while count < 25:
659
parent = delta_parents[0]
660
method = self._index.get_method(parent)
661
if method == 'fulltext':
662
break
663
delta_parents = self._index.get_parents(parent)
664
count = count + 1
665
if method == 'line-delta':
666
delta = False
667
668
lines = self.factory.make(lines, version_id)
669
if delta or (self.factory.annotated and len(present_parents) > 0):
670
# Merge annotations from parent texts if so is needed.
671
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
672
delta, self.factory.annotated)
977
673
978
674
if delta:
979
675
options.append('line-delta')
980
store_lines = self._factory.lower_line_delta(delta_hunks)
981
size, bytes = self._record_to_data(key, digest,
982
store_lines)
676
store_lines = self.factory.lower_line_delta(delta_hunks)
983
677
else:
984
678
options.append('fulltext')
985
# isinstance is slower and we have no hierarchy.
986
if self._factory.__class__ is KnitPlainFactory:
987
# Use the already joined bytes saving iteration time in
988
# _record_to_data.
989
size, bytes = self._record_to_data(key, digest,
990
lines, [line_bytes])
991
else:
992
# get mixed annotation + content and feed it into the
993
# serialiser.
994
store_lines = self._factory.lower_fulltext(content)
995
size, bytes = self._record_to_data(key, digest,
996
store_lines)
997
998
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
999
self._index.add_records(
1000
((key, options, access_memo, parents),),
1001
random_id=random_id)
1002
return digest, text_length, content
1003
1004
def annotate(self, key):
1005
"""See VersionedFiles.annotate."""
1006
return self._factory.annotate(self, key)
679
store_lines = self.factory.lower_fulltext(lines)
680
681
where, size = self._data.add_record(version_id, digest, store_lines)
682
self._index.add_version(version_id, options, where, size, parents)
683
return lines
1007
684
1008
685
def check(self, progress_bar=None):
1009
"""See VersionedFiles.check()."""
1010
# This doesn't actually test extraction of everything, but that will
1011
# impact 'bzr check' substantially, and needs to be integrated with
1012
# care. However, it does check for the obvious problem of a delta with
1013
# no basis.
1014
keys = self._index.keys()
1015
parent_map = self.get_parent_map(keys)
1016
for key in keys:
1017
if self._index.get_method(key) != 'fulltext':
1018
compression_parent = parent_map[key][0]
1019
if compression_parent not in parent_map:
1020
raise errors.KnitCorrupt(self,
1021
"Missing basis parent %s for %s" % (
1022
compression_parent, key))
1023
for fallback_vfs in self._fallback_vfs:
1024
fallback_vfs.check()
1025
1026
def _check_add(self, key, lines, random_id, check_content):
1027
"""check that version_id and lines are safe to add."""
1028
version_id = key[-1]
1029
if version_id is not None:
1030
if contains_whitespace(version_id):
1031
raise InvalidRevisionId(version_id, self)
1032
self.check_not_reserved_id(version_id)
1033
# TODO: If random_id==False and the key is already present, we should
1034
# probably check that the existing content is identical to what is
1035
# being inserted, and otherwise raise an exception. This would make
1036
# the bundle code simpler.
1037
if check_content:
1038
self._check_lines_not_unicode(lines)
1039
self._check_lines_are_lines(lines)
1040
1041
def _check_header(self, key, line):
1042
rec = self._split_header(line)
1043
self._check_header_version(rec, key[-1])
1044
return rec
1045
1046
def _check_header_version(self, rec, version_id):
1047
"""Checks the header version on original format knit records.
1048
1049
These have the last component of the key embedded in the record.
1050
"""
1051
if rec[1] != version_id:
1052
raise KnitCorrupt(self,
1053
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1054
1055
def _check_should_delta(self, parent):
1056
"""Iterate back through the parent listing, looking for a fulltext.
1057
1058
This is used when we want to decide whether to add a delta or a new
1059
fulltext. It searches for _max_delta_chain parents. When it finds a
1060
fulltext parent, it sees if the total size of the deltas leading up to
1061
it is large enough to indicate that we want a new full text anyway.
1062
1063
Return True if we should create a new delta, False if we should use a
1064
full text.
1065
"""
1066
delta_size = 0
1067
fulltext_size = None
1068
for count in xrange(self._max_delta_chain):
1069
try:
1070
# Note that this only looks in the index of this particular
1071
# KnitVersionedFiles, not in the fallbacks. This ensures that
1072
# we won't store a delta spanning physical repository
1073
# boundaries.
1074
build_details = self._index.get_build_details([parent])
1075
parent_details = build_details[parent]
1076
except (RevisionNotPresent, KeyError), e:
1077
# Some basis is not locally present: always fulltext
1078
return False
1079
index_memo, compression_parent, _, _ = parent_details
1080
_, _, size = index_memo
1081
if compression_parent is None:
1082
fulltext_size = size
1083
break
1084
delta_size += size
1085
# We don't explicitly check for presence because this is in an
1086
# inner loop, and if it's missing it'll fail anyhow.
1087
parent = compression_parent
1088
else:
1089
# We couldn't find a fulltext, so we must create a new one
1090
return False
1091
# Simple heuristic - if the total I/O wold be greater as a delta than
1092
# the originally installed fulltext, we create a new fulltext.
1093
return fulltext_size > delta_size
1094
1095
def _build_details_to_components(self, build_details):
1096
"""Convert a build_details tuple to a position tuple."""
1097
# record_details, access_memo, compression_parent
1098
return build_details[3], build_details[0], build_details[1]
1099
1100
def _get_components_positions(self, keys, allow_missing=False):
1101
"""Produce a map of position data for the components of keys.
1102
1103
This data is intended to be used for retrieving the knit records.
1104
1105
A dict of key to (record_details, index_memo, next, parents) is
1106
returned.
1107
method is the way referenced data should be applied.
1108
index_memo is the handle to pass to the data access to actually get the
1109
data
1110
next is the build-parent of the version, or None for fulltexts.
1111
parents is the version_ids of the parents of this version
1112
1113
:param allow_missing: If True do not raise an error on a missing component,
1114
just ignore it.
1115
"""
1116
component_data = {}
1117
pending_components = keys
1118
while pending_components:
1119
build_details = self._index.get_build_details(pending_components)
1120
current_components = set(pending_components)
1121
pending_components = set()
1122
for key, details in build_details.iteritems():
1123
(index_memo, compression_parent, parents,
1124
record_details) = details
1125
method = record_details[0]
1126
if compression_parent is not None:
1127
pending_components.add(compression_parent)
1128
component_data[key] = self._build_details_to_components(details)
1129
missing = current_components.difference(build_details)
1130
if missing and not allow_missing:
1131
raise errors.RevisionNotPresent(missing.pop(), self)
1132
return component_data
1133
1134
def _get_content(self, key, parent_texts={}):
1135
"""Returns a content object that makes up the specified
1136
version."""
1137
cached_version = parent_texts.get(key, None)
1138
if cached_version is not None:
1139
# Ensure the cache dict is valid.
1140
if not self.get_parent_map([key]):
1141
raise RevisionNotPresent(key, self)
1142
return cached_version
1143
generator = _VFContentMapGenerator(self, [key])
1144
return generator._get_content(key)
1145
1146
def get_parent_map(self, keys):
1147
"""Get a map of the graph parents of keys.
1148
1149
:param keys: The keys to look up parents for.
1150
:return: A mapping from keys to parents. Absent keys are absent from
1151
the mapping.
1152
"""
1153
return self._get_parent_map_with_sources(keys)[0]
1154
1155
def _get_parent_map_with_sources(self, keys):
1156
"""Get a map of the parents of keys.
1157
1158
:param keys: The keys to look up parents for.
1159
:return: A tuple. The first element is a mapping from keys to parents.
1160
Absent keys are absent from the mapping. The second element is a
1161
list with the locations each key was found in. The first element
1162
is the in-this-knit parents, the second the first fallback source,
1163
and so on.
1164
"""
1165
result = {}
1166
sources = [self._index] + self._fallback_vfs
1167
source_results = []
1168
missing = set(keys)
1169
for source in sources:
1170
if not missing:
1171
break
1172
new_result = source.get_parent_map(missing)
1173
source_results.append(new_result)
1174
result.update(new_result)
1175
missing.difference_update(set(new_result))
1176
return result, source_results
1177
1178
def _get_record_map(self, keys, allow_missing=False):
686
"""See VersionedFile.check()."""
687
688
def _clone_text(self, new_version_id, old_version_id, parents):
689
"""See VersionedFile.clone_text()."""
690
# FIXME RBC 20060228 make fast by only inserting an index with null
691
# delta.
692
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
693
694
def get_lines(self, version_id):
695
"""See VersionedFile.get_lines()."""
696
return self.get_line_list([version_id])[0]
697
698
def _get_record_map(self, version_ids):
1179
699
"""Produce a dictionary of knit records.
1180
1181
:return: {key:(record, record_details, digest, next)}
1182
record
1183
data returned from read_records (a KnitContentobject)
1184
record_details
1185
opaque information to pass to parse_record
1186
digest
1187
SHA1 digest of the full text after all steps are done
1188
next
1189
build-parent of the version, i.e. the leftmost ancestor.
1190
Will be None if the record is not a delta.
1191
:param keys: The keys to build a map for
1192
:param allow_missing: If some records are missing, rather than
1193
error, just return the data that could be generated.
1194
"""
1195
raw_map = self._get_record_map_unparsed(keys,
1196
allow_missing=allow_missing)
1197
return self._raw_map_to_record_map(raw_map)
1198
1199
def _raw_map_to_record_map(self, raw_map):
1200
"""Parse the contents of _get_record_map_unparsed.
1201
1202
:return: see _get_record_map.
1203
"""
1204
result = {}
1205
for key in raw_map:
1206
data, record_details, next = raw_map[key]
1207
content, digest = self._parse_record(key[-1], data)
1208
result[key] = content, record_details, digest, next
1209
return result
1210
1211
def _get_record_map_unparsed(self, keys, allow_missing=False):
1212
"""Get the raw data for reconstructing keys without parsing it.
1213
1214
:return: A dict suitable for parsing via _raw_map_to_record_map.
1215
key-> raw_bytes, (method, noeol), compression_parent
1216
"""
1217
# This retries the whole request if anything fails. Potentially we
1218
# could be a bit more selective. We could track the keys whose records
1219
# we have successfully found, and then only request the new records
1220
# from there. However, _get_components_positions grabs the whole build
1221
# chain, which means we'll likely try to grab the same records again
1222
# anyway. Also, can the build chains change as part of a pack
1223
# operation? We wouldn't want to end up with a broken chain.
1224
while True:
1225
try:
1226
position_map = self._get_components_positions(keys,
1227
allow_missing=allow_missing)
1228
# key = component_id, r = record_details, i_m = index_memo,
1229
# n = next
1230
records = [(key, i_m) for key, (r, i_m, n)
1231
in position_map.iteritems()]
1232
# Sort by the index memo, so that we request records from the
1233
# same pack file together, and in forward-sorted order
1234
records.sort(key=operator.itemgetter(1))
1235
raw_record_map = {}
1236
for key, data in self._read_records_iter_unchecked(records):
1237
(record_details, index_memo, next) = position_map[key]
1238
raw_record_map[key] = data, record_details, next
1239
return raw_record_map
1240
except errors.RetryWithNewPacks, e:
1241
self._access.reload_or_raise(e)
1242
1243
@classmethod
1244
def _split_by_prefix(cls, keys):
1245
"""For the given keys, split them up based on their prefix.
1246
1247
To keep memory pressure somewhat under control, split the
1248
requests back into per-file-id requests, otherwise "bzr co"
1249
extracts the full tree into memory before writing it to disk.
1250
This should be revisited if _get_content_maps() can ever cross
1251
file-id boundaries.
1252
1253
The keys for a given file_id are kept in the same relative order.
1254
Ordering between file_ids is not, though prefix_order will return the
1255
order that the key was first seen.
1256
1257
:param keys: An iterable of key tuples
1258
:return: (split_map, prefix_order)
1259
split_map A dictionary mapping prefix => keys
1260
prefix_order The order that we saw the various prefixes
1261
"""
1262
split_by_prefix = {}
1263
prefix_order = []
1264
for key in keys:
1265
if len(key) == 1:
1266
prefix = ''
1267
else:
1268
prefix = key[0]
1269
1270
if prefix in split_by_prefix:
1271
split_by_prefix[prefix].append(key)
1272
else:
1273
split_by_prefix[prefix] = [key]
1274
prefix_order.append(prefix)
1275
return split_by_prefix, prefix_order
1276
1277
def _group_keys_for_io(self, keys, non_local_keys, positions,
1278
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1279
"""For the given keys, group them into 'best-sized' requests.
1280
1281
The idea is to avoid making 1 request per file, but to never try to
1282
unpack an entire 1.5GB source tree in a single pass. Also when
1283
possible, we should try to group requests to the same pack file
1284
together.
1285
1286
:return: list of (keys, non_local) tuples that indicate what keys
1287
should be fetched next.
1288
"""
1289
# TODO: Ideally we would group on 2 factors. We want to extract texts
1290
# from the same pack file together, and we want to extract all
1291
# the texts for a given build-chain together. Ultimately it
1292
# probably needs a better global view.
1293
total_keys = len(keys)
1294
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1295
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1296
cur_keys = []
1297
cur_non_local = set()
1298
cur_size = 0
1299
result = []
1300
sizes = []
1301
for prefix in prefix_order:
1302
keys = prefix_split_keys[prefix]
1303
non_local = prefix_split_non_local_keys.get(prefix, [])
1304
1305
this_size = self._index._get_total_build_size(keys, positions)
1306
cur_size += this_size
1307
cur_keys.extend(keys)
1308
cur_non_local.update(non_local)
1309
if cur_size > _min_buffer_size:
1310
result.append((cur_keys, cur_non_local))
1311
sizes.append(cur_size)
1312
cur_keys = []
1313
cur_non_local = set()
1314
cur_size = 0
1315
if cur_keys:
1316
result.append((cur_keys, cur_non_local))
1317
sizes.append(cur_size)
1318
return result
1319
1320
def get_record_stream(self, keys, ordering, include_delta_closure):
1321
"""Get a stream of records for keys.
1322
1323
:param keys: The keys to include.
1324
:param ordering: Either 'unordered' or 'topological'. A topologically
1325
sorted stream has compression parents strictly before their
1326
children.
1327
:param include_delta_closure: If True then the closure across any
1328
compression parents will be included (in the opaque data).
1329
:return: An iterator of ContentFactory objects, each of which is only
1330
valid until the iterator is advanced.
1331
"""
1332
# keys might be a generator
1333
keys = set(keys)
1334
if not keys:
1335
return
1336
if not self._index.has_graph:
1337
# Cannot sort when no graph has been stored.
1338
ordering = 'unordered'
1339
1340
remaining_keys = keys
1341
while True:
1342
try:
1343
keys = set(remaining_keys)
1344
for content_factory in self._get_remaining_record_stream(keys,
1345
ordering, include_delta_closure):
1346
remaining_keys.discard(content_factory.key)
1347
yield content_factory
1348
return
1349
except errors.RetryWithNewPacks, e:
1350
self._access.reload_or_raise(e)
1351
1352
def _get_remaining_record_stream(self, keys, ordering,
1353
include_delta_closure):
1354
"""This function is the 'retry' portion for get_record_stream."""
1355
if include_delta_closure:
1356
positions = self._get_components_positions(keys, allow_missing=True)
1357
else:
1358
build_details = self._index.get_build_details(keys)
1359
# map from key to
1360
# (record_details, access_memo, compression_parent_key)
1361
positions = dict((key, self._build_details_to_components(details))
1362
for key, details in build_details.iteritems())
1363
absent_keys = keys.difference(set(positions))
1364
# There may be more absent keys : if we're missing the basis component
1365
# and are trying to include the delta closure.
1366
# XXX: We should not ever need to examine remote sources because we do
1367
# not permit deltas across versioned files boundaries.
1368
if include_delta_closure:
1369
needed_from_fallback = set()
1370
# Build up reconstructable_keys dict. key:True in this dict means
1371
# the key can be reconstructed.
1372
reconstructable_keys = {}
1373
for key in keys:
1374
# the delta chain
1375
try:
1376
chain = [key, positions[key][2]]
1377
except KeyError:
1378
needed_from_fallback.add(key)
1379
continue
1380
result = True
1381
while chain[-1] is not None:
1382
if chain[-1] in reconstructable_keys:
1383
result = reconstructable_keys[chain[-1]]
1384
break
1385
else:
1386
try:
1387
chain.append(positions[chain[-1]][2])
1388
except KeyError:
1389
# missing basis component
1390
needed_from_fallback.add(chain[-1])
1391
result = True
1392
break
1393
for chain_key in chain[:-1]:
1394
reconstructable_keys[chain_key] = result
1395
if not result:
1396
needed_from_fallback.add(key)
1397
# Double index lookups here : need a unified api ?
1398
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1399
if ordering in ('topological', 'groupcompress'):
1400
if ordering == 'topological':
1401
# Global topological sort
1402
present_keys = tsort.topo_sort(global_map)
1403
else:
1404
present_keys = sort_groupcompress(global_map)
1405
# Now group by source:
1406
source_keys = []
1407
current_source = None
1408
for key in present_keys:
1409
for parent_map in parent_maps:
1410
if key in parent_map:
1411
key_source = parent_map
1412
break
1413
if current_source is not key_source:
1414
source_keys.append((key_source, []))
1415
current_source = key_source
1416
source_keys[-1][1].append(key)
1417
else:
1418
if ordering != 'unordered':
1419
raise AssertionError('valid values for ordering are:'
1420
' "unordered", "groupcompress" or "topological" not: %r'
1421
% (ordering,))
1422
# Just group by source; remote sources first.
1423
present_keys = []
1424
source_keys = []
1425
for parent_map in reversed(parent_maps):
1426
source_keys.append((parent_map, []))
1427
for key in parent_map:
1428
present_keys.append(key)
1429
source_keys[-1][1].append(key)
1430
# We have been requested to return these records in an order that
1431
# suits us. So we ask the index to give us an optimally sorted
1432
# order.
1433
for source, sub_keys in source_keys:
1434
if source is parent_maps[0]:
1435
# Only sort the keys for this VF
1436
self._index._sort_keys_by_io(sub_keys, positions)
1437
absent_keys = keys - set(global_map)
1438
for key in absent_keys:
1439
yield AbsentContentFactory(key)
1440
# restrict our view to the keys we can answer.
1441
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1442
# XXX: At that point we need to consider the impact of double reads by
1443
# utilising components multiple times.
1444
if include_delta_closure:
1445
# XXX: get_content_maps performs its own index queries; allow state
1446
# to be passed in.
1447
non_local_keys = needed_from_fallback - absent_keys
1448
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1449
non_local_keys,
1450
positions):
1451
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1452
global_map)
1453
for record in generator.get_record_stream():
1454
yield record
1455
else:
1456
for source, keys in source_keys:
1457
if source is parent_maps[0]:
1458
# this KnitVersionedFiles
1459
records = [(key, positions[key][1]) for key in keys]
1460
for key, raw_data, sha1 in self._read_records_iter_raw(records):
1461
(record_details, index_memo, _) = positions[key]
1462
yield KnitContentFactory(key, global_map[key],
1463
record_details, sha1, raw_data, self._factory.annotated, None)
1464
else:
1465
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1466
for record in vf.get_record_stream(keys, ordering,
1467
include_delta_closure):
1468
yield record
1469
1470
def get_sha1s(self, keys):
1471
"""See VersionedFiles.get_sha1s()."""
1472
missing = set(keys)
1473
record_map = self._get_record_map(missing, allow_missing=True)
1474
result = {}
1475
for key, details in record_map.iteritems():
1476
if key not in missing:
1477
continue
1478
# record entry 2 is the 'digest'.
1479
result[key] = details[2]
1480
missing.difference_update(set(result))
1481
for source in self._fallback_vfs:
1482
if not missing:
1483
break
1484
new_result = source.get_sha1s(missing)
1485
result.update(new_result)
1486
missing.difference_update(set(new_result))
1487
return result
1488
1489
def insert_record_stream(self, stream):
1490
"""Insert a record stream into this container.
1491
1492
:param stream: A stream of records to insert.
1493
:return: None
1494
:seealso VersionedFiles.get_record_stream:
1495
"""
1496
def get_adapter(adapter_key):
1497
try:
1498
return adapters[adapter_key]
1499
except KeyError:
1500
adapter_factory = adapter_registry.get(adapter_key)
1501
adapter = adapter_factory(self)
1502
adapters[adapter_key] = adapter
1503
return adapter
1504
delta_types = set()
1505
if self._factory.annotated:
1506
# self is annotated, we need annotated knits to use directly.
1507
annotated = "annotated-"
1508
convertibles = []
1509
else:
1510
# self is not annotated, but we can strip annotations cheaply.
1511
annotated = ""
1512
convertibles = set(["knit-annotated-ft-gz"])
1513
if self._max_delta_chain:
1514
delta_types.add("knit-annotated-delta-gz")
1515
convertibles.add("knit-annotated-delta-gz")
1516
# The set of types we can cheaply adapt without needing basis texts.
1517
native_types = set()
1518
if self._max_delta_chain:
1519
native_types.add("knit-%sdelta-gz" % annotated)
1520
delta_types.add("knit-%sdelta-gz" % annotated)
1521
native_types.add("knit-%sft-gz" % annotated)
1522
knit_types = native_types.union(convertibles)
1523
adapters = {}
1524
# Buffer all index entries that we can't add immediately because their
1525
# basis parent is missing. We don't buffer all because generating
1526
# annotations may require access to some of the new records. However we
1527
# can't generate annotations from new deltas until their basis parent
1528
# is present anyway, so we get away with not needing an index that
1529
# includes the new keys.
1530
#
1531
# See <http://launchpad.net/bugs/300177> about ordering of compression
1532
# parents in the records - to be conservative, we insist that all
1533
# parents must be present to avoid expanding to a fulltext.
1534
#
1535
# key = basis_parent, value = index entry to add
1536
buffered_index_entries = {}
1537
for record in stream:
1538
buffered = False
1539
parents = record.parents
1540
if record.storage_kind in delta_types:
1541
# TODO: eventually the record itself should track
1542
# compression_parent
1543
compression_parent = parents[0]
1544
else:
1545
compression_parent = None
1546
# Raise an error when a record is missing.
1547
if record.storage_kind == 'absent':
1548
raise RevisionNotPresent([record.key], self)
1549
elif ((record.storage_kind in knit_types)
1550
and (compression_parent is None
1551
or not self._fallback_vfs
1552
or self._index.has_key(compression_parent)
1553
or not self.has_key(compression_parent))):
1554
# we can insert the knit record literally if either it has no
1555
# compression parent OR we already have its basis in this kvf
1556
# OR the basis is not present even in the fallbacks. In the
1557
# last case it will either turn up later in the stream and all
1558
# will be well, or it won't turn up at all and we'll raise an
1559
# error at the end.
1560
#
1561
# TODO: self.has_key is somewhat redundant with
1562
# self._index.has_key; we really want something that directly
1563
# asks if it's only present in the fallbacks. -- mbp 20081119
1564
if record.storage_kind not in native_types:
1565
try:
1566
adapter_key = (record.storage_kind, "knit-delta-gz")
1567
adapter = get_adapter(adapter_key)
1568
except KeyError:
1569
adapter_key = (record.storage_kind, "knit-ft-gz")
1570
adapter = get_adapter(adapter_key)
1571
bytes = adapter.get_bytes(record)
1572
else:
1573
# It's a knit record, it has a _raw_record field (even if
1574
# it was reconstituted from a network stream).
1575
bytes = record._raw_record
1576
options = [record._build_details[0]]
1577
if record._build_details[1]:
1578
options.append('no-eol')
1579
# Just blat it across.
1580
# Note: This does end up adding data on duplicate keys. As
1581
# modern repositories use atomic insertions this should not
1582
# lead to excessive growth in the event of interrupted fetches.
1583
# 'knit' repositories may suffer excessive growth, but as a
1584
# deprecated format this is tolerable. It can be fixed if
1585
# needed by in the kndx index support raising on a duplicate
1586
# add with identical parents and options.
1587
access_memo = self._access.add_raw_records(
1588
[(record.key, len(bytes))], bytes)[0]
1589
index_entry = (record.key, options, access_memo, parents)
1590
if 'fulltext' not in options:
1591
# Not a fulltext, so we need to make sure the compression
1592
# parent will also be present.
1593
# Note that pack backed knits don't need to buffer here
1594
# because they buffer all writes to the transaction level,
1595
# but we don't expose that difference at the index level. If
1596
# the query here has sufficient cost to show up in
1597
# profiling we should do that.
1598
#
1599
# They're required to be physically in this
1600
# KnitVersionedFiles, not in a fallback.
1601
if not self._index.has_key(compression_parent):
1602
pending = buffered_index_entries.setdefault(
1603
compression_parent, [])
1604
pending.append(index_entry)
1605
buffered = True
1606
if not buffered:
1607
self._index.add_records([index_entry])
1608
elif record.storage_kind == 'chunked':
1609
self.add_lines(record.key, parents,
1610
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1611
else:
1612
# Not suitable for direct insertion as a
1613
# delta, either because it's not the right format, or this
1614
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1615
# 0) or because it depends on a base only present in the
1616
# fallback kvfs.
1617
self._access.flush()
1618
try:
1619
# Try getting a fulltext directly from the record.
1620
bytes = record.get_bytes_as('fulltext')
1621
except errors.UnavailableRepresentation:
1622
adapter_key = record.storage_kind, 'fulltext'
1623
adapter = get_adapter(adapter_key)
1624
bytes = adapter.get_bytes(record)
1625
lines = split_lines(bytes)
1626
try:
1627
self.add_lines(record.key, parents, lines)
1628
except errors.RevisionAlreadyPresent:
1629
pass
1630
# Add any records whose basis parent is now available.
1631
if not buffered:
1632
added_keys = [record.key]
1633
while added_keys:
1634
key = added_keys.pop(0)
1635
if key in buffered_index_entries:
1636
index_entries = buffered_index_entries[key]
1637
self._index.add_records(index_entries)
1638
added_keys.extend(
1639
[index_entry[0] for index_entry in index_entries])
1640
del buffered_index_entries[key]
1641
if buffered_index_entries:
1642
# There were index entries buffered at the end of the stream,
1643
# So these need to be added (if the index supports holding such
1644
# entries for later insertion)
1645
for key in buffered_index_entries:
1646
index_entries = buffered_index_entries[key]
1647
self._index.add_records(index_entries,
1648
missing_compression_parents=True)
1649
1650
def get_missing_compression_parent_keys(self):
1651
"""Return an iterable of keys of missing compression parents.
1652
1653
Check this after calling insert_record_stream to find out if there are
1654
any missing compression parents. If there are, the records that
1655
depend on them are not able to be inserted safely. For atomic
1656
KnitVersionedFiles built on packs, the transaction should be aborted or
1657
suspended - commit will fail at this point. Nonatomic knits will error
1658
earlier because they have no staging area to put pending entries into.
1659
"""
1660
return self._index.get_missing_compression_parents()
1661
1662
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1663
"""Iterate over the lines in the versioned files from keys.
1664
1665
This may return lines from other keys. Each item the returned
1666
iterator yields is a tuple of a line and a text version that that line
1667
is present in (not introduced in).
1668
1669
Ordering of results is in whatever order is most suitable for the
1670
underlying storage format.
1671
1672
If a progress bar is supplied, it may be used to indicate progress.
1673
The caller is responsible for cleaning up progress bars (because this
1674
is an iterator).
1675
1676
NOTES:
1677
* Lines are normalised by the underlying store: they will all have \\n
1678
terminators.
1679
* Lines are returned in arbitrary order.
1680
* If a requested key did not change any lines (or didn't have any
1681
lines), it may not be mentioned at all in the result.
1682
1683
:param pb: Progress bar supplied by caller.
1684
:return: An iterator over (line, key).
1685
"""
1686
if pb is None:
1687
pb = progress.DummyProgress()
1688
keys = set(keys)
1689
total = len(keys)
1690
done = False
1691
while not done:
1692
try:
1693
# we don't care about inclusions, the caller cares.
1694
# but we need to setup a list of records to visit.
1695
# we need key, position, length
1696
key_records = []
1697
build_details = self._index.get_build_details(keys)
1698
for key, details in build_details.iteritems():
1699
if key in keys:
1700
key_records.append((key, details[0]))
1701
records_iter = enumerate(self._read_records_iter(key_records))
1702
for (key_idx, (key, data, sha_value)) in records_iter:
1703
pb.update('Walking content', key_idx, total)
1704
compression_parent = build_details[key][1]
1705
if compression_parent is None:
1706
# fulltext
1707
line_iterator = self._factory.get_fulltext_content(data)
1708
else:
1709
# Delta
1710
line_iterator = self._factory.get_linedelta_content(data)
1711
# Now that we are yielding the data for this key, remove it
1712
# from the list
1713
keys.remove(key)
1714
# XXX: It might be more efficient to yield (key,
1715
# line_iterator) in the future. However for now, this is a
1716
# simpler change to integrate into the rest of the
1717
# codebase. RBC 20071110
1718
for line in line_iterator:
1719
yield line, key
1720
done = True
1721
except errors.RetryWithNewPacks, e:
1722
self._access.reload_or_raise(e)
1723
# If there are still keys we've not yet found, we look in the fallback
1724
# vfs, and hope to find them there. Note that if the keys are found
1725
# but had no changes or no content, the fallback may not return
1726
# anything.
1727
if keys and not self._fallback_vfs:
1728
# XXX: strictly the second parameter is meant to be the file id
1729
# but it's not easily accessible here.
1730
raise RevisionNotPresent(keys, repr(self))
1731
for source in self._fallback_vfs:
1732
if not keys:
1733
break
1734
source_keys = set()
1735
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1736
source_keys.add(key)
1737
yield line, key
1738
keys.difference_update(source_keys)
1739
pb.update('Walking content', total, total)
1740
1741
def _make_line_delta(self, delta_seq, new_content):
1742
"""Generate a line delta from delta_seq and new_content."""
1743
diff_hunks = []
1744
for op in delta_seq.get_opcodes():
1745
if op[0] == 'equal':
1746
continue
1747
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1748
return diff_hunks
1749
1750
def _merge_annotations(self, content, parents, parent_texts={},
1751
delta=None, annotated=None,
1752
left_matching_blocks=None):
1753
"""Merge annotations for content and generate deltas.
1754
1755
This is done by comparing the annotations based on changes to the text
1756
and generating a delta on the resulting full texts. If annotations are
1757
not being created then a simple delta is created.
1758
"""
1759
if left_matching_blocks is not None:
1760
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1761
else:
1762
delta_seq = None
1763
if annotated:
1764
for parent_key in parents:
1765
merge_content = self._get_content(parent_key, parent_texts)
1766
if (parent_key == parents[0] and delta_seq is not None):
1767
seq = delta_seq
1768
else:
1769
seq = patiencediff.PatienceSequenceMatcher(
1770
None, merge_content.text(), content.text())
1771
for i, j, n in seq.get_matching_blocks():
1772
if n == 0:
1773
continue
1774
# this copies (origin, text) pairs across to the new
1775
# content for any line that matches the last-checked
1776
# parent.
1777
content._lines[j:j+n] = merge_content._lines[i:i+n]
1778
# XXX: Robert says the following block is a workaround for a
1779
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1780
if content._lines and content._lines[-1][1][-1] != '\n':
1781
# The copied annotation was from a line without a trailing EOL,
1782
# reinstate one for the content object, to ensure correct
1783
# serialization.
1784
line = content._lines[-1][1] + '\n'
1785
content._lines[-1] = (content._lines[-1][0], line)
1786
if delta:
1787
if delta_seq is None:
1788
reference_content = self._get_content(parents[0], parent_texts)
1789
new_texts = content.text()
1790
old_texts = reference_content.text()
1791
delta_seq = patiencediff.PatienceSequenceMatcher(
1792
None, old_texts, new_texts)
1793
return self._make_line_delta(delta_seq, content)
1794
1795
def _parse_record(self, version_id, data):
1796
"""Parse an original format knit record.
1797
1798
These have the last element of the key only present in the stored data.
1799
"""
1800
rec, record_contents = self._parse_record_unchecked(data)
1801
self._check_header_version(rec, version_id)
1802
return record_contents, rec[3]
1803
1804
def _parse_record_header(self, key, raw_data):
1805
"""Parse a record header for consistency.
1806
1807
:return: the header and the decompressor stream.
1808
as (stream, header_record)
1809
"""
1810
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1811
try:
1812
# Current serialise
1813
rec = self._check_header(key, df.readline())
1814
except Exception, e:
1815
raise KnitCorrupt(self,
1816
"While reading {%s} got %s(%s)"
1817
% (key, e.__class__.__name__, str(e)))
1818
return df, rec
1819
1820
def _parse_record_unchecked(self, data):
1821
# profiling notes:
1822
# 4168 calls in 2880 217 internal
1823
# 4168 calls to _parse_record_header in 2121
1824
# 4168 calls to readlines in 330
1825
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1826
try:
1827
record_contents = df.readlines()
1828
except Exception, e:
1829
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1830
(data, e.__class__.__name__, str(e)))
1831
header = record_contents.pop(0)
1832
rec = self._split_header(header)
1833
last_line = record_contents.pop()
1834
if len(record_contents) != int(rec[2]):
1835
raise KnitCorrupt(self,
1836
'incorrect number of lines %s != %s'
1837
' for version {%s} %s'
1838
% (len(record_contents), int(rec[2]),
1839
rec[1], record_contents))
1840
if last_line != 'end %s\n' % rec[1]:
1841
raise KnitCorrupt(self,
1842
'unexpected version end line %r, wanted %r'
1843
% (last_line, rec[1]))
1844
df.close()
1845
return rec, record_contents
1846
1847
def _read_records_iter(self, records):
1848
"""Read text records from data file and yield result.
1849
1850
The result will be returned in whatever is the fastest to read.
1851
Not by the order requested. Also, multiple requests for the same
1852
record will only yield 1 response.
1853
:param records: A list of (key, access_memo) entries
1854
:return: Yields (key, contents, digest) in the order
1855
read, not the order requested
1856
"""
1857
if not records:
1858
return
1859
1860
# XXX: This smells wrong, IO may not be getting ordered right.
1861
needed_records = sorted(set(records), key=operator.itemgetter(1))
1862
if not needed_records:
1863
return
1864
1865
# The transport optimizes the fetching as well
1866
# (ie, reads continuous ranges.)
1867
raw_data = self._access.get_raw_records(
1868
[index_memo for key, index_memo in needed_records])
1869
1870
for (key, index_memo), data in \
1871
izip(iter(needed_records), raw_data):
1872
content, digest = self._parse_record(key[-1], data)
1873
yield key, content, digest
1874
1875
def _read_records_iter_raw(self, records):
1876
"""Read text records from data file and yield raw data.
1877
1878
This unpacks enough of the text record to validate the id is
1879
as expected but thats all.
1880
1881
Each item the iterator yields is (key, bytes,
1882
expected_sha1_of_full_text).
1883
"""
1884
for key, data in self._read_records_iter_unchecked(records):
1885
# validate the header (note that we can only use the suffix in
1886
# current knit records).
1887
df, rec = self._parse_record_header(key, data)
1888
df.close()
1889
yield key, data, rec[3]
1890
1891
def _read_records_iter_unchecked(self, records):
1892
"""Read text records from data file and yield raw data.
1893
1894
No validation is done.
1895
1896
Yields tuples of (key, data).
1897
"""
1898
# setup an iterator of the external records:
1899
# uses readv so nice and fast we hope.
1900
if len(records):
1901
# grab the disk data needed.
1902
needed_offsets = [index_memo for key, index_memo
1903
in records]
1904
raw_records = self._access.get_raw_records(needed_offsets)
1905
1906
for key, index_memo in records:
1907
data = raw_records.next()
1908
yield key, data
1909
1910
def _record_to_data(self, key, digest, lines, dense_lines=None):
1911
"""Convert key, digest, lines into a raw data block.
1912
1913
:param key: The key of the record. Currently keys are always serialised
1914
using just the trailing component.
1915
:param dense_lines: The bytes of lines but in a denser form. For
1916
instance, if lines is a list of 1000 bytestrings each ending in \n,
1917
dense_lines may be a list with one line in it, containing all the
1918
1000's lines and their \n's. Using dense_lines if it is already
1919
known is a win because the string join to create bytes in this
1920
function spends less time resizing the final string.
1921
:return: (len, a StringIO instance with the raw data ready to read.)
1922
"""
1923
# Note: using a string copy here increases memory pressure with e.g.
1924
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
1925
# when doing the initial commit of a mozilla tree. RBC 20070921
1926
bytes = ''.join(chain(
1927
["version %s %d %s\n" % (key[-1],
1928
len(lines),
1929
digest)],
1930
dense_lines or lines,
1931
["end %s\n" % key[-1]]))
1932
if type(bytes) != str:
1933
raise AssertionError(
1934
'data must be plain bytes was %s' % type(bytes))
1935
if lines and lines[-1][-1] != '\n':
1936
raise ValueError('corrupt lines value %r' % lines)
1937
compressed_bytes = tuned_gzip.bytes_to_gzip(bytes)
1938
return len(compressed_bytes), compressed_bytes
1939
1940
def _split_header(self, line):
1941
rec = line.split()
1942
if len(rec) != 4:
1943
raise KnitCorrupt(self,
1944
'unexpected number of elements in record header')
1945
return rec
1946
1947
def keys(self):
1948
"""See VersionedFiles.keys."""
1949
if 'evil' in debug.debug_flags:
1950
trace.mutter_callsite(2, "keys scales with size of history")
1951
sources = [self._index] + self._fallback_vfs
1952
result = set()
1953
for source in sources:
1954
result.update(source.keys())
1955
return result
1956
1957
1958
class _ContentMapGenerator(object):
1959
"""Generate texts or expose raw deltas for a set of texts."""
1960
1961
def _get_content(self, key):
1962
"""Get the content object for key."""
1963
# Note that _get_content is only called when the _ContentMapGenerator
1964
# has been constructed with just one key requested for reconstruction.
1965
if key in self.nonlocal_keys:
1966
record = self.get_record_stream().next()
1967
# Create a content object on the fly
1968
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
1969
return PlainKnitContent(lines, record.key)
1970
else:
1971
# local keys we can ask for directly
1972
return self._get_one_work(key)
1973
1974
def get_record_stream(self):
1975
"""Get a record stream for the keys requested during __init__."""
1976
for record in self._work():
1977
yield record
1978
1979
def _work(self):
1980
"""Produce maps of text and KnitContents as dicts.
1981
700
701
The keys are version_ids, the values are tuples of (method, content,
702
digest, next).
703
method is the way the content should be applied.
704
content is a KnitContent object.
705
digest is the SHA1 digest of this version id after all steps are done
706
next is the build-parent of the version, i.e. the leftmost ancestor.
707
If the method is fulltext, next will be None.
708
"""
709
position_map = self._get_components_positions(version_ids)
710
# c = component_id, m = method, p = position, s = size, n = next
711
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
712
record_map = {}
713
for component_id, content, digest in \
714
self._data.read_records_iter(records):
715
method, position, size, next = position_map[component_id]
716
record_map[component_id] = method, content, digest, next
717
718
return record_map
719
720
def get_text(self, version_id):
721
"""See VersionedFile.get_text"""
722
return self.get_texts([version_id])[0]
723
724
def get_texts(self, version_ids):
725
return [''.join(l) for l in self.get_line_list(version_ids)]
726
727
def get_line_list(self, version_ids):
728
"""Return the texts of listed versions as a list of strings."""
729
text_map, content_map = self._get_content_maps(version_ids)
730
return [text_map[v] for v in version_ids]
731
732
def _get_content_maps(self, version_ids):
733
"""Produce maps of text and KnitContents
734
1982
735
:return: (text_map, content_map) where text_map contains the texts for
1983
the requested versions and content_map contains the KnitContents.
736
the requested versions and content_map contains the KnitContents.
737
Both dicts take version_ids as their keys.
1984
738
"""
1985
# NB: By definition we never need to read remote sources unless texts
1986
# are requested from them: we don't delta across stores - and we
1987
# explicitly do not want to to prevent data loss situations.
1988
if self.global_map is None:
1989
self.global_map = self.vf.get_parent_map(self.keys)
1990
nonlocal_keys = self.nonlocal_keys
1991
1992
missing_keys = set(nonlocal_keys)
1993
# Read from remote versioned file instances and provide to our caller.
1994
for source in self.vf._fallback_vfs:
1995
if not missing_keys:
1996
break
1997
# Loop over fallback repositories asking them for texts - ignore
1998
# any missing from a particular fallback.
1999
for record in source.get_record_stream(missing_keys,
2000
'unordered', True):
2001
if record.storage_kind == 'absent':
2002
# Not in thie particular stream, may be in one of the
2003
# other fallback vfs objects.
2004
continue
2005
missing_keys.remove(record.key)
2006
yield record
2007
2008
self._raw_record_map = self.vf._get_record_map_unparsed(self.keys,
2009
allow_missing=True)
2010
first = True
2011
for key in self.keys:
2012
if key in self.nonlocal_keys:
2013
continue
2014
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2015
first = False
2016
2017
def _get_one_work(self, requested_key):
2018
# Now, if we have calculated everything already, just return the
2019
# desired text.
2020
if requested_key in self._contents_map:
2021
return self._contents_map[requested_key]
2022
# To simplify things, parse everything at once - code that wants one text
2023
# probably wants them all.
2024
# FUTURE: This function could be improved for the 'extract many' case
2025
# by tracking each component and only doing the copy when the number of
2026
# children than need to apply delta's to it is > 1 or it is part of the
2027
# final output.
2028
multiple_versions = len(self.keys) != 1
2029
if self._record_map is None:
2030
self._record_map = self.vf._raw_map_to_record_map(
2031
self._raw_record_map)
2032
record_map = self._record_map
2033
# raw_record_map is key:
2034
# Have read and parsed records at this point.
2035
for key in self.keys:
2036
if key in self.nonlocal_keys:
2037
# already handled
2038
continue
739
for version_id in version_ids:
740
if not self.has_version(version_id):
741
raise RevisionNotPresent(version_id, self.filename)
742
record_map = self._get_record_map(version_ids)
743
744
text_map = {}
745
content_map = {}
746
final_content = {}
747
for version_id in version_ids:
2039
748
components = []
2040
cursor = key
749
cursor = version_id
2041
750
while cursor is not None:
2042
try:
2043
record, record_details, digest, next = record_map[cursor]
2044
except KeyError:
2045
raise RevisionNotPresent(cursor, self)
2046
components.append((cursor, record, record_details, digest))
751
method, data, digest, next = record_map[cursor]
752
components.append((cursor, method, data, digest))
753
if cursor in content_map:
754
break
2047
755
cursor = next
2048
if cursor in self._contents_map:
2049
# no need to plan further back
2050
components.append((cursor, None, None, None))
2051
break
2052
756
2053
757
content = None
2054
for (component_id, record, record_details,
2055
digest) in reversed(components):
2056
if component_id in self._contents_map:
2057
content = self._contents_map[component_id]
758
for component_id, method, data, digest in reversed(components):
759
if component_id in content_map:
760
content = content_map[component_id]
2058
761
else:
2059
content, delta = self._factory.parse_record(key[-1],
2060
record, record_details, content,
2061
copy_base_content=multiple_versions)
2062
if multiple_versions:
2063
self._contents_map[component_id] = content
762
version_idx = self._index.lookup(component_id)
763
if method == 'fulltext':
764
assert content is None
765
content = self.factory.parse_fulltext(data, version_idx)
766
elif method == 'line-delta':
767
delta = self.factory.parse_line_delta(data[:],
768
version_idx)
769
content = content.copy()
770
content._lines = self._apply_delta(content._lines,
771
delta)
772
content_map[component_id] = content
773
774
if 'no-eol' in self._index.get_options(version_id):
775
content = content.copy()
776
line = content._lines[-1][1].rstrip('\n')
777
content._lines[-1] = (content._lines[-1][0], line)
778
final_content[version_id] = content
2064
779
2065
780
# digest here is the digest from the last applied component.
2066
781
text = content.text()
2067
actual_sha = sha_strings(text)
2068
if actual_sha != digest:
2069
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2070
if multiple_versions:
2071
return self._contents_map[requested_key]
2072
else:
2073
return content
2074
2075
def _wire_bytes(self):
2076
"""Get the bytes to put on the wire for 'key'.
2077
2078
The first collection of bytes asked for returns the serialised
2079
raw_record_map and the additional details (key, parent) for key.
2080
Subsequent calls return just the additional details (key, parent).
2081
The wire storage_kind given for the first key is 'knit-delta-closure',
2082
For subsequent keys it is 'knit-delta-closure-ref'.
2083
2084
:param key: A key from the content generator.
2085
:return: Bytes to put on the wire.
2086
"""
2087
lines = []
2088
# kind marker for dispatch on the far side,
2089
lines.append('knit-delta-closure')
2090
# Annotated or not
2091
if self.vf._factory.annotated:
2092
lines.append('annotated')
2093
else:
2094
lines.append('')
2095
# then the list of keys
2096
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2097
if key not in self.nonlocal_keys]))
2098
# then the _raw_record_map in serialised form:
2099
map_byte_list = []
2100
# for each item in the map:
2101
# 1 line with key
2102
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2103
# one line with method
2104
# one line with noeol
2105
# one line with next ('' for None)
2106
# one line with byte count of the record bytes
2107
# the record bytes
2108
for key, (record_bytes, (method, noeol), next) in \
2109
self._raw_record_map.iteritems():
2110
key_bytes = '\x00'.join(key)
2111
parents = self.global_map.get(key, None)
2112
if parents is None:
2113
parent_bytes = 'None:'
2114
else:
2115
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2116
method_bytes = method
2117
if noeol:
2118
noeol_bytes = "T"
2119
else:
2120
noeol_bytes = "F"
2121
if next:
2122
next_bytes = '\x00'.join(next)
2123
else:
2124
next_bytes = ''
2125
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2126
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2127
len(record_bytes), record_bytes))
2128
map_bytes = ''.join(map_byte_list)
2129
lines.append(map_bytes)
2130
bytes = '\n'.join(lines)
2131
return bytes
2132
2133
2134
class _VFContentMapGenerator(_ContentMapGenerator):
2135
"""Content map generator reading from a VersionedFiles object."""
2136
2137
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2138
global_map=None, raw_record_map=None):
2139
"""Create a _ContentMapGenerator.
2140
2141
:param versioned_files: The versioned files that the texts are being
2142
extracted from.
2143
:param keys: The keys to produce content maps for.
2144
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2145
which are known to not be in this knit, but rather in one of the
2146
fallback knits.
2147
:param global_map: The result of get_parent_map(keys) (or a supermap).
2148
This is required if get_record_stream() is to be used.
2149
:param raw_record_map: A unparsed raw record map to use for answering
2150
contents.
2151
"""
2152
# The vf to source data from
2153
self.vf = versioned_files
2154
# The keys desired
2155
self.keys = list(keys)
2156
# Keys known to be in fallback vfs objects
2157
if nonlocal_keys is None:
2158
self.nonlocal_keys = set()
2159
else:
2160
self.nonlocal_keys = frozenset(nonlocal_keys)
2161
# Parents data for keys to be returned in get_record_stream
2162
self.global_map = global_map
2163
# The chunked lists for self.keys in text form
2164
self._text_map = {}
2165
# A cache of KnitContent objects used in extracting texts.
2166
self._contents_map = {}
2167
# All the knit records needed to assemble the requested keys as full
2168
# texts.
2169
self._record_map = None
2170
if raw_record_map is None:
2171
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2172
allow_missing=True)
2173
else:
2174
self._raw_record_map = raw_record_map
2175
# the factory for parsing records
2176
self._factory = self.vf._factory
2177
2178
2179
class _NetworkContentMapGenerator(_ContentMapGenerator):
2180
"""Content map generator sourced from a network stream."""
2181
2182
def __init__(self, bytes, line_end):
2183
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2184
self._bytes = bytes
2185
self.global_map = {}
2186
self._raw_record_map = {}
2187
self._contents_map = {}
2188
self._record_map = None
2189
self.nonlocal_keys = []
2190
# Get access to record parsing facilities
2191
self.vf = KnitVersionedFiles(None, None)
2192
start = line_end
2193
# Annotated or not
2194
line_end = bytes.find('\n', start)
2195
line = bytes[start:line_end]
2196
start = line_end + 1
2197
if line == 'annotated':
2198
self._factory = KnitAnnotateFactory()
2199
else:
2200
self._factory = KnitPlainFactory()
2201
# list of keys to emit in get_record_stream
2202
line_end = bytes.find('\n', start)
2203
line = bytes[start:line_end]
2204
start = line_end + 1
2205
self.keys = [
2206
tuple(segment.split('\x00')) for segment in line.split('\t')
2207
if segment]
2208
# now a loop until the end. XXX: It would be nice if this was just a
2209
# bunch of the same records as get_record_stream(..., False) gives, but
2210
# there is a decent sized gap stopping that at the moment.
2211
end = len(bytes)
2212
while start < end:
2213
# 1 line with key
2214
line_end = bytes.find('\n', start)
2215
key = tuple(bytes[start:line_end].split('\x00'))
2216
start = line_end + 1
2217
# 1 line with parents (None: for None, '' for ())
2218
line_end = bytes.find('\n', start)
2219
line = bytes[start:line_end]
2220
if line == 'None:':
2221
parents = None
2222
else:
2223
parents = tuple(
2224
[tuple(segment.split('\x00')) for segment in line.split('\t')
2225
if segment])
2226
self.global_map[key] = parents
2227
start = line_end + 1
2228
# one line with method
2229
line_end = bytes.find('\n', start)
2230
line = bytes[start:line_end]
2231
method = line
2232
start = line_end + 1
2233
# one line with noeol
2234
line_end = bytes.find('\n', start)
2235
line = bytes[start:line_end]
2236
noeol = line == "T"
2237
start = line_end + 1
2238
# one line with next ('' for None)
2239
line_end = bytes.find('\n', start)
2240
line = bytes[start:line_end]
2241
if not line:
2242
next = None
2243
else:
2244
next = tuple(bytes[start:line_end].split('\x00'))
2245
start = line_end + 1
2246
# one line with byte count of the record bytes
2247
line_end = bytes.find('\n', start)
2248
line = bytes[start:line_end]
2249
count = int(line)
2250
start = line_end + 1
2251
# the record bytes
2252
record_bytes = bytes[start:start+count]
2253
start = start + count
2254
# put it in the map
2255
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2256
2257
def get_record_stream(self):
2258
"""Get a record stream for for keys requested by the bytestream."""
2259
first = True
2260
for key in self.keys:
2261
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2262
first = False
2263
2264
def _wire_bytes(self):
2265
return self._bytes
2266
2267
2268
class _KndxIndex(object):
2269
"""Manages knit index files
2270
2271
The index is kept in memory and read on startup, to enable
782
if sha_strings(text) != digest:
783
raise KnitCorrupt(self.filename,
784
'sha-1 does not match %s' % version_id)
785
786
text_map[version_id] = text
787
return text_map, final_content
788
789
def iter_lines_added_or_present_in_versions(self, version_ids=None):
790
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
791
if version_ids is None:
792
version_ids = self.versions()
793
# we don't care about inclusions, the caller cares.
794
# but we need to setup a list of records to visit.
795
# we need version_id, position, length
796
version_id_records = []
797
requested_versions = list(version_ids)
798
# filter for available versions
799
for version_id in requested_versions:
800
if not self.has_version(version_id):
801
raise RevisionNotPresent(version_id, self.filename)
802
# get a in-component-order queue:
803
version_ids = []
804
for version_id in self.versions():
805
if version_id in requested_versions:
806
version_ids.append(version_id)
807
data_pos, length = self._index.get_position(version_id)
808
version_id_records.append((version_id, data_pos, length))
809
810
pb = bzrlib.ui.ui_factory.nested_progress_bar()
811
count = 0
812
total = len(version_id_records)
813
try:
814
pb.update('Walking content.', count, total)
815
for version_id, data, sha_value in \
816
self._data.read_records_iter(version_id_records):
817
pb.update('Walking content.', count, total)
818
method = self._index.get_method(version_id)
819
version_idx = self._index.lookup(version_id)
820
assert method in ('fulltext', 'line-delta')
821
if method == 'fulltext':
822
content = self.factory.parse_fulltext(data, version_idx)
823
for line in content.text():
824
yield line
825
else:
826
delta = self.factory.parse_line_delta(data, version_idx)
827
for start, end, count, lines in delta:
828
for origin, line in lines:
829
yield line
830
count +=1
831
pb.update('Walking content.', total, total)
832
pb.finished()
833
except:
834
pb.update('Walking content.', total, total)
835
pb.finished()
836
raise
837
838
def num_versions(self):
839
"""See VersionedFile.num_versions()."""
840
return self._index.num_versions()
841
842
__len__ = num_versions
843
844
def annotate_iter(self, version_id):
845
"""See VersionedFile.annotate_iter."""
846
content = self._get_content(version_id)
847
for origin, text in content.annotate_iter():
848
yield origin, text
849
850
def get_parents(self, version_id):
851
"""See VersionedFile.get_parents."""
852
# perf notes:
853
# optimism counts!
854
# 52554 calls in 1264 872 internal down from 3674
855
try:
856
return self._index.get_parents(version_id)
857
except KeyError:
858
raise RevisionNotPresent(version_id, self.filename)
859
860
def get_parents_with_ghosts(self, version_id):
861
"""See VersionedFile.get_parents."""
862
try:
863
return self._index.get_parents_with_ghosts(version_id)
864
except KeyError:
865
raise RevisionNotPresent(version_id, self.filename)
866
867
def get_ancestry(self, versions):
868
"""See VersionedFile.get_ancestry."""
869
if isinstance(versions, basestring):
870
versions = [versions]
871
if not versions:
872
return []
873
self._check_versions_present(versions)
874
return self._index.get_ancestry(versions)
875
876
def get_ancestry_with_ghosts(self, versions):
877
"""See VersionedFile.get_ancestry_with_ghosts."""
878
if isinstance(versions, basestring):
879
versions = [versions]
880
if not versions:
881
return []
882
self._check_versions_present(versions)
883
return self._index.get_ancestry_with_ghosts(versions)
884
885
#@deprecated_method(zero_eight)
886
def walk(self, version_ids):
887
"""See VersionedFile.walk."""
888
# We take the short path here, and extract all relevant texts
889
# and put them in a weave and let that do all the work. Far
890
# from optimal, but is much simpler.
891
# FIXME RB 20060228 this really is inefficient!
892
from bzrlib.weave import Weave
893
894
w = Weave(self.filename)
895
ancestry = self.get_ancestry(version_ids)
896
sorted_graph = topo_sort(self._index.get_graph())
897
version_list = [vid for vid in sorted_graph if vid in ancestry]
898
899
for version_id in version_list:
900
lines = self.get_lines(version_id)
901
w.add_lines(version_id, self.get_parents(version_id), lines)
902
903
for lineno, insert_id, dset, line in w.walk(version_ids):
904
yield lineno, insert_id, dset, line
905
906
def plan_merge(self, ver_a, ver_b):
907
"""See VersionedFile.plan_merge."""
908
ancestors_b = set(self.get_ancestry(ver_b))
909
def status_a(revision, text):
910
if revision in ancestors_b:
911
return 'killed-b', text
912
else:
913
return 'new-a', text
914
915
ancestors_a = set(self.get_ancestry(ver_a))
916
def status_b(revision, text):
917
if revision in ancestors_a:
918
return 'killed-a', text
919
else:
920
return 'new-b', text
921
922
annotated_a = self.annotate(ver_a)
923
annotated_b = self.annotate(ver_b)
924
plain_a = [t for (a, t) in annotated_a]
925
plain_b = [t for (a, t) in annotated_b]
926
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
927
a_cur = 0
928
b_cur = 0
929
for ai, bi, l in blocks:
930
# process all mismatched sections
931
# (last mismatched section is handled because blocks always
932
# includes a 0-length last block)
933
for revision, text in annotated_a[a_cur:ai]:
934
yield status_a(revision, text)
935
for revision, text in annotated_b[b_cur:bi]:
936
yield status_b(revision, text)
937
938
# and now the matched section
939
a_cur = ai + l
940
b_cur = bi + l
941
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
942
assert text_a == text_b
943
yield "unchanged", text_a
944
945
946
class _KnitComponentFile(object):
947
"""One of the files used to implement a knit database"""
948
949
def __init__(self, transport, filename, mode, file_mode=None):
950
self._transport = transport
951
self._filename = filename
952
self._mode = mode
953
self._file_mode=file_mode
954
955
def write_header(self):
956
if self._transport.append(self._filename, StringIO(self.HEADER),
957
mode=self._file_mode):
958
raise KnitCorrupt(self._filename, 'misaligned after writing header')
959
960
def check_header(self, fp):
961
line = fp.readline()
962
if line != self.HEADER:
963
raise KnitHeaderError(badline=line)
964
965
def commit(self):
966
"""Commit is a nop."""
967
968
def __repr__(self):
969
return '%s(%s)' % (self.__class__.__name__, self._filename)
970
971
972
class _KnitIndex(_KnitComponentFile):
973
"""Manages knit index file.
974
975
The index is already kept in memory and read on startup, to enable
2272
976
fast lookups of revision information. The cursor of the index
2273
977
file is always pointing to the end, making it easy to append
2274
978
entries.
2284
988
2285
989
Duplicate entries may be written to the index for a single version id
2286
990
if this is done then the latter one completely replaces the former:
2287
this allows updates to correct version and parent information.
991
this allows updates to correct version and parent information.
2288
992
Note that the two entries may share the delta, and that successive
2289
993
annotations and references MUST point to the first entry.
2290
994
2291
995
The index file on disc contains a header, followed by one line per knit
2292
996
record. The same revision can be present in an index file more than once.
2293
The first occurrence gets assigned a sequence number starting from 0.
2294
997
The first occurrence gets assigned a sequence number starting from 0.
998
2295
999
The format of a single line is
2296
1000
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
2297
1001
REVISION_ID is a utf8-encoded revision id
2298
FLAGS is a comma separated list of flags about the record. Values include
1002
FLAGS is a comma separated list of flags about the record. Values include
2299
1003
no-eol, line-delta, fulltext.
2300
1004
BYTE_OFFSET is the ascii representation of the byte offset in the data file
2301
1005
that the the compressed data starts at.
2305
1009
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
2306
1010
revision id already in the knit that is a parent of REVISION_ID.
2307
1011
The ' :' marker is the end of record marker.
2308
1012
2309
1013
partial writes:
2310
when a write is interrupted to the index file, it will result in a line
2311
that does not end in ' :'. If the ' :' is not present at the end of a line,
2312
or at the end of the file, then the record that is missing it will be
2313
ignored by the parser.
1014
when a write is interrupted to the index file, it will result in a line that
1015
does not end in ' :'. If the ' :' is not present at the end of a line, or at
1016
the end of the file, then the record that is missing it will be ignored by
1017
the parser.
2314
1018
2315
1019
When writing new records to the index file, the data is preceded by '\n'
2316
1020
to ensure that records always start on new lines even if the last write was
2317
1021
interrupted. As a result its normal for the last line in the index to be
2318
1022
missing a trailing newline. One can be added with no harmful effects.
2319
2320
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2321
where prefix is e.g. the (fileid,) for .texts instances or () for
2322
constant-mapped things like .revisions, and the old state is
2323
tuple(history_vector, cache_dict). This is used to prevent having an
2324
ABI change with the C extension that reads .kndx files.
2325
1023
"""
2326
1024
2327
1025
HEADER = "# bzr knit index 8\n"
2328
1026
2329
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2330
"""Create a _KndxIndex on transport using mapper."""
2331
self._transport = transport
2332
self._mapper = mapper
2333
self._get_scope = get_scope
2334
self._allow_writes = allow_writes
2335
self._is_locked = is_locked
2336
self._reset_cache()
2337
self.has_graph = True
2338
2339
def add_records(self, records, random_id=False, missing_compression_parents=False):
2340
"""Add multiple records to the index.
2341
2342
:param records: a list of tuples:
2343
(key, options, access_memo, parents).
2344
:param random_id: If True the ids being added were randomly generated
2345
and no check for existence will be performed.
2346
:param missing_compression_parents: If True the records being added are
2347
only compressed against texts already in the index (or inside
2348
records). If False the records all refer to unavailable texts (or
2349
texts inside records) as compression parents.
2350
"""
2351
if missing_compression_parents:
2352
# It might be nice to get the edge of the records. But keys isn't
2353
# _wrong_.
2354
keys = sorted(record[0] for record in records)
2355
raise errors.RevisionNotPresent(keys, self)
2356
paths = {}
2357
for record in records:
2358
key = record[0]
2359
prefix = key[:-1]
2360
path = self._mapper.map(key) + '.kndx'
2361
path_keys = paths.setdefault(path, (prefix, []))
2362
path_keys[1].append(record)
2363
for path in sorted(paths):
2364
prefix, path_keys = paths[path]
2365
self._load_prefixes([prefix])
2366
lines = []
2367
orig_history = self._kndx_cache[prefix][1][:]
2368
orig_cache = self._kndx_cache[prefix][0].copy()
2369
2370
try:
2371
for key, options, (_, pos, size), parents in path_keys:
2372
if parents is None:
2373
# kndx indices cannot be parentless.
2374
parents = ()
2375
line = "\n%s %s %s %s %s :" % (
2376
key[-1], ','.join(options), pos, size,
2377
self._dictionary_compress(parents))
2378
if type(line) != str:
2379
raise AssertionError(
2380
'data must be utf8 was %s' % type(line))
2381
lines.append(line)
2382
self._cache_key(key, options, pos, size, parents)
2383
if len(orig_history):
2384
self._transport.append_bytes(path, ''.join(lines))
2385
else:
2386
self._init_index(path, lines)
2387
except:
2388
# If any problems happen, restore the original values and re-raise
2389
self._kndx_cache[prefix] = (orig_cache, orig_history)
2390
raise
2391
2392
def scan_unvalidated_index(self, graph_index):
2393
"""See _KnitGraphIndex.scan_unvalidated_index."""
2394
# Because kndx files do not support atomic insertion via separate index
2395
# files, they do not support this method.
2396
raise NotImplementedError(self.scan_unvalidated_index)
2397
2398
def get_missing_compression_parents(self):
2399
"""See _KnitGraphIndex.get_missing_compression_parents."""
2400
# Because kndx files do not support atomic insertion via separate index
2401
# files, they do not support this method.
2402
raise NotImplementedError(self.get_missing_compression_parents)
2403
2404
def _cache_key(self, key, options, pos, size, parent_keys):
1027
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1028
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1029
1030
def _cache_version(self, version_id, options, pos, size, parents):
2405
1031
"""Cache a version record in the history array and index cache.
2406
2407
This is inlined into _load_data for performance. KEEP IN SYNC.
1032
1033
This is inlined into __init__ for performance. KEEP IN SYNC.
2408
1034
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2409
1035
indexes).
2410
1036
"""
2411
prefix = key[:-1]
2412
version_id = key[-1]
2413
# last-element only for compatibilty with the C load_data.
2414
parents = tuple(parent[-1] for parent in parent_keys)
2415
for parent in parent_keys:
2416
if parent[:-1] != prefix:
2417
raise ValueError("mismatched prefixes for %r, %r" % (
2418
key, parent_keys))
2419
cache, history = self._kndx_cache[prefix]
2420
1037
# only want the _history index to reference the 1st index entry
2421
1038
# for version_id
2422
if version_id not in cache:
2423
index = len(history)
2424
history.append(version_id)
1039
if version_id not in self._cache:
1040
index = len(self._history)
1041
self._history.append(version_id)
2425
1042
else:
2426
index = cache[version_id][5]
2427
cache[version_id] = (version_id,
1043
index = self._cache[version_id][5]
1044
self._cache[version_id] = (version_id,
2428
1045
options,
2429
1046
pos,
2430
1047
size,
2431
1048
parents,
2432
1049
index)
2433
1050
2434
def check_header(self, fp):
2435
line = fp.readline()
2436
if line == '':
2437
# An empty file can actually be treated as though the file doesn't
2438
# exist yet.
2439
raise errors.NoSuchFile(self)
2440
if line != self.HEADER:
2441
raise KnitHeaderError(badline=line, filename=self)
2442
2443
def _check_read(self):
2444
if not self._is_locked():
2445
raise errors.ObjectNotLocked(self)
2446
if self._get_scope() != self._scope:
2447
self._reset_cache()
2448
2449
def _check_write_ok(self):
2450
"""Assert if not writes are permitted."""
2451
if not self._is_locked():
2452
raise errors.ObjectNotLocked(self)
2453
if self._get_scope() != self._scope:
2454
self._reset_cache()
2455
if self._mode != 'w':
2456
raise errors.ReadOnlyObjectDirtiedError(self)
2457
2458
def get_build_details(self, keys):
2459
"""Get the method, index_memo and compression parent for keys.
2460
2461
Ghosts are omitted from the result.
2462
2463
:param keys: An iterable of keys.
2464
:return: A dict of key:(index_memo, compression_parent, parents,
2465
record_details).
2466
index_memo
2467
opaque structure to pass to read_records to extract the raw
2468
data
2469
compression_parent
2470
Content that this record is built upon, may be None
2471
parents
2472
Logical parents of this node
2473
record_details
2474
extra information about the content which needs to be passed to
2475
Factory.parse_record
2476
"""
2477
parent_map = self.get_parent_map(keys)
2478
result = {}
2479
for key in keys:
2480
if key not in parent_map:
2481
continue # Ghost
2482
method = self.get_method(key)
2483
parents = parent_map[key]
2484
if method == 'fulltext':
2485
compression_parent = None
2486
else:
2487
compression_parent = parents[0]
2488
noeol = 'no-eol' in self.get_options(key)
2489
index_memo = self.get_position(key)
2490
result[key] = (index_memo, compression_parent,
2491
parents, (method, noeol))
2492
return result
2493
2494
def get_method(self, key):
2495
"""Return compression method of specified key."""
2496
options = self.get_options(key)
2497
if 'fulltext' in options:
2498
return 'fulltext'
2499
elif 'line-delta' in options:
2500
return 'line-delta'
2501
else:
2502
raise errors.KnitIndexUnknownMethod(self, options)
2503
2504
def get_options(self, key):
2505
"""Return a list representing options.
2506
2507
e.g. ['foo', 'bar']
2508
"""
2509
prefix, suffix = self._split_key(key)
2510
self._load_prefixes([prefix])
1051
def __init__(self, transport, filename, mode, create=False, file_mode=None):
1052
_KnitComponentFile.__init__(self, transport, filename, mode, file_mode)
1053
self._cache = {}
1054
# position in _history is the 'official' index for a revision
1055
# but the values may have come from a newer entry.
1056
# so - wc -l of a knit index is != the number of unique names
1057
# in the knit.
1058
self._history = []
1059
pb = bzrlib.ui.ui_factory.nested_progress_bar()
2511
1060
try:
2512
return self._kndx_cache[prefix][0][suffix][1]
2513
except KeyError:
2514
raise RevisionNotPresent(key, self)
2515
2516
def get_parent_map(self, keys):
2517
"""Get a map of the parents of keys.
2518
2519
:param keys: The keys to look up parents for.
2520
:return: A mapping from keys to parents. Absent keys are absent from
2521
the mapping.
1061
count = 0
1062
total = 1
1063
try:
1064
pb.update('read knit index', count, total)
1065
fp = self._transport.get(self._filename)
1066
try:
1067
self.check_header(fp)
1068
# readlines reads the whole file at once:
1069
# bad for transports like http, good for local disk
1070
# we save 60 ms doing this one change (
1071
# from calling readline each time to calling
1072
# readlines once.
1073
# probably what we want for nice behaviour on
1074
# http is a incremental readlines that yields, or
1075
# a check for local vs non local indexes,
1076
for l in fp.readlines():
1077
rec = l.split()
1078
if len(rec) < 5 or rec[-1] != ':':
1079
# corrupt line.
1080
# FIXME: in the future we should determine if its a
1081
# short write - and ignore it
1082
# or a different failure, and raise. RBC 20060407
1083
continue
1084
count += 1
1085
total += 1
1086
#pb.update('read knit index', count, total)
1087
# See self._parse_parents
1088
parents = []
1089
for value in rec[4:-1]:
1090
if '.' == value[0]:
1091
# uncompressed reference
1092
parents.append(value[1:])
1093
else:
1094
# this is 15/4000ms faster than isinstance,
1095
# (in lsprof)
1096
# this function is called thousands of times a
1097
# second so small variations add up.
1098
assert value.__class__ is str
1099
parents.append(self._history[int(value)])
1100
# end self._parse_parents
1101
# self._cache_version(rec[0],
1102
# rec[1].split(','),
1103
# int(rec[2]),
1104
# int(rec[3]),
1105
# parents)
1106
# --- self._cache_version
1107
# only want the _history index to reference the 1st
1108
# index entry for version_id
1109
version_id = rec[0]
1110
if version_id not in self._cache:
1111
index = len(self._history)
1112
self._history.append(version_id)
1113
else:
1114
index = self._cache[version_id][5]
1115
self._cache[version_id] = (version_id,
1116
rec[1].split(','),
1117
int(rec[2]),
1118
int(rec[3]),
1119
parents,
1120
index)
1121
# --- self._cache_version
1122
finally:
1123
fp.close()
1124
except NoSuchFile, e:
1125
if mode != 'w' or not create:
1126
raise
1127
self.write_header()
1128
finally:
1129
pb.update('read knit index', total, total)
1130
pb.finished()
1131
1132
def _parse_parents(self, compressed_parents):
1133
"""convert a list of string parent values into version ids.
1134
1135
ints are looked up in the index.
1136
.FOO values are ghosts and converted in to FOO.
1137
1138
NOTE: the function is retained here for clarity, and for possible
1139
use in partial index reads. However bulk processing now has
1140
it inlined in __init__ for inner-loop optimisation.
2522
1141
"""
2523
# Parse what we need to up front, this potentially trades off I/O
2524
# locality (.kndx and .knit in the same block group for the same file
2525
# id) for less checking in inner loops.
2526
prefixes = set(key[:-1] for key in keys)
2527
self._load_prefixes(prefixes)
2528
result = {}
2529
for key in keys:
2530
prefix = key[:-1]
1142
result = []
1143
for value in compressed_parents:
1144
if value[-1] == '.':
1145
# uncompressed reference
1146
result.append(value[1:])
1147
else:
1148
# this is 15/4000ms faster than isinstance,
1149
# this function is called thousands of times a
1150
# second so small variations add up.
1151
assert value.__class__ is str
1152
result.append(self._history[int(value)])
1153
return result
1154
1155
def get_graph(self):
1156
graph = []
1157
for version_id, index in self._cache.iteritems():
1158
graph.append((version_id, index[4]))
1159
return graph
1160
1161
def get_ancestry(self, versions):
1162
"""See VersionedFile.get_ancestry."""
1163
# get a graph of all the mentioned versions:
1164
graph = {}
1165
pending = set(versions)
1166
while len(pending):
1167
version = pending.pop()
1168
parents = self._cache[version][4]
1169
# got the parents ok
1170
# trim ghosts
1171
parents = [parent for parent in parents if parent in self._cache]
1172
for parent in parents:
1173
# if not completed and not a ghost
1174
if parent not in graph:
1175
pending.add(parent)
1176
graph[version] = parents
1177
return topo_sort(graph.items())
1178
1179
def get_ancestry_with_ghosts(self, versions):
1180
"""See VersionedFile.get_ancestry_with_ghosts."""
1181
# get a graph of all the mentioned versions:
1182
graph = {}
1183
pending = set(versions)
1184
while len(pending):
1185
version = pending.pop()
2531
1186
try:
2532
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
1187
parents = self._cache[version][4]
2533
1188
except KeyError:
1189
# ghost, fake it
1190
graph[version] = []
2534
1191
pass
2535
1192
else:
2536
result[key] = tuple(prefix + (suffix,) for
2537
suffix in suffix_parents)
2538
return result
2539
2540
def get_position(self, key):
2541
"""Return details needed to access the version.
2542
2543
:return: a tuple (key, data position, size) to hand to the access
2544
logic to get the record.
2545
"""
2546
prefix, suffix = self._split_key(key)
2547
self._load_prefixes([prefix])
2548
entry = self._kndx_cache[prefix][0][suffix]
2549
return key, entry[2], entry[3]
2550
2551
has_key = _mod_index._has_key_from_parent_map
2552
2553
def _init_index(self, path, extra_lines=[]):
2554
"""Initialize an index."""
2555
sio = StringIO()
2556
sio.write(self.HEADER)
2557
sio.writelines(extra_lines)
2558
sio.seek(0)
2559
self._transport.put_file_non_atomic(path, sio,
2560
create_parent_dir=True)
2561
# self._create_parent_dir)
2562
# mode=self._file_mode,
2563
# dir_mode=self._dir_mode)
2564
2565
def keys(self):
2566
"""Get all the keys in the collection.
2567
2568
The keys are not ordered.
2569
"""
2570
result = set()
2571
# Identify all key prefixes.
2572
# XXX: A bit hacky, needs polish.
2573
if type(self._mapper) == ConstantMapper:
2574
prefixes = [()]
2575
else:
2576
relpaths = set()
2577
for quoted_relpath in self._transport.iter_files_recursive():
2578
path, ext = os.path.splitext(quoted_relpath)
2579
relpaths.add(path)
2580
prefixes = [self._mapper.unmap(path) for path in relpaths]
2581
self._load_prefixes(prefixes)
2582
for prefix in prefixes:
2583
for suffix in self._kndx_cache[prefix][1]:
2584
result.add(prefix + (suffix,))
2585
return result
2586
2587
def _load_prefixes(self, prefixes):
2588
"""Load the indices for prefixes."""
2589
self._check_read()
2590
for prefix in prefixes:
2591
if prefix not in self._kndx_cache:
2592
# the load_data interface writes to these variables.
2593
self._cache = {}
2594
self._history = []
2595
self._filename = prefix
2596
try:
2597
path = self._mapper.map(prefix) + '.kndx'
2598
fp = self._transport.get(path)
2599
try:
2600
# _load_data may raise NoSuchFile if the target knit is
2601
# completely empty.
2602
_load_data(self, fp)
2603
finally:
2604
fp.close()
2605
self._kndx_cache[prefix] = (self._cache, self._history)
2606
del self._cache
2607
del self._filename
2608
del self._history
2609
except NoSuchFile:
2610
self._kndx_cache[prefix] = ({}, [])
2611
if type(self._mapper) == ConstantMapper:
2612
# preserve behaviour for revisions.kndx etc.
2613
self._init_index(path)
2614
del self._cache
2615
del self._filename
2616
del self._history
2617
2618
missing_keys = _mod_index._missing_keys_from_parent_map
2619
2620
def _partition_keys(self, keys):
2621
"""Turn keys into a dict of prefix:suffix_list."""
2622
result = {}
2623
for key in keys:
2624
prefix_keys = result.setdefault(key[:-1], [])
2625
prefix_keys.append(key[-1])
2626
return result
2627
2628
def _dictionary_compress(self, keys):
2629
"""Dictionary compress keys.
2630
2631
:param keys: The keys to generate references to.
2632
:return: A string representation of keys. keys which are present are
2633
dictionary compressed, and others are emitted as fulltext with a
2634
'.' prefix.
2635
"""
2636
if not keys:
2637
return ''
1193
# got the parents ok
1194
for parent in parents:
1195
if parent not in graph:
1196
pending.add(parent)
1197
graph[version] = parents
1198
return topo_sort(graph.items())
1199
1200
def num_versions(self):
1201
return len(self._history)
1202
1203
__len__ = num_versions
1204
1205
def get_versions(self):
1206
return self._history
1207
1208
def idx_to_name(self, idx):
1209
return self._history[idx]
1210
1211
def lookup(self, version_id):
1212
assert version_id in self._cache
1213
return self._cache[version_id][5]
1214
1215
def _version_list_to_index(self, versions):
1216
encode_utf8 = cache_utf8.encode
2638
1217
result_list = []
2639
prefix = keys[0][:-1]
2640
cache = self._kndx_cache[prefix][0]
2641
for key in keys:
2642
if key[:-1] != prefix:
2643
# kndx indices cannot refer across partitioned storage.
2644
raise ValueError("mismatched prefixes for %r" % keys)
2645
if key[-1] in cache:
1218
for version in versions:
1219
if version in self._cache:
2646
1220
# -- inlined lookup() --
2647
result_list.append(str(cache[key[-1]][5]))
1221
result_list.append(str(self._cache[version][5]))
2648
1222
# -- end lookup () --
2649
1223
else:
2650
result_list.append('.' + key[-1])
1224
result_list.append('.' + encode_utf8(version))
2651
1225
return ' '.join(result_list)
2652
1226
2653
def _reset_cache(self):
2654
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2655
# (cache_dict, history_vector) for parsed kndx files.
2656
self._kndx_cache = {}
2657
self._scope = self._get_scope()
2658
allow_writes = self._allow_writes()
2659
if allow_writes:
2660
self._mode = 'w'
1227
def add_version(self, version_id, options, pos, size, parents):
1228
"""Add a version record to the index."""
1229
self.add_versions(((version_id, options, pos, size, parents),))
1230
1231
def add_versions(self, versions):
1232
"""Add multiple versions to the index.
1233
1234
:param versions: a list of tuples:
1235
(version_id, options, pos, size, parents).
1236
"""
1237
lines = []
1238
encode_utf8 = cache_utf8.encode
1239
for version_id, options, pos, size, parents in versions:
1240
line = "\n%s %s %s %s %s :" % (encode_utf8(version_id),
1241
','.join(options),
1242
pos,
1243
size,
1244
self._version_list_to_index(parents))
1245
assert isinstance(line, str), \
1246
'content must be utf-8 encoded: %r' % (line,)
1247
lines.append(line)
1248
self._transport.append(self._filename, StringIO(''.join(lines)))
1249
# cache after writing, so that a failed write leads to missing cache
1250
# entries not extra ones. XXX TODO: RBC 20060502 in the event of a
1251
# failure, reload the index or flush it or some such, to prevent
1252
# writing records that did complete twice.
1253
for version_id, options, pos, size, parents in versions:
1254
self._cache_version(version_id, options, pos, size, parents)
1255
1256
def has_version(self, version_id):
1257
"""True if the version is in the index."""
1258
return (version_id in self._cache)
1259
1260
def get_position(self, version_id):
1261
"""Return data position and size of specified version."""
1262
return (self._cache[version_id][2], \
1263
self._cache[version_id][3])
1264
1265
def get_method(self, version_id):
1266
"""Return compression method of specified version."""
1267
options = self._cache[version_id][1]
1268
if 'fulltext' in options:
1269
return 'fulltext'
2661
1270
else:
2662
self._mode = 'r'
2663
2664
def _sort_keys_by_io(self, keys, positions):
2665
"""Figure out an optimal order to read the records for the given keys.
2666
2667
Sort keys, grouped by index and sorted by position.
2668
2669
:param keys: A list of keys whose records we want to read. This will be
2670
sorted 'in-place'.
2671
:param positions: A dict, such as the one returned by
2672
_get_components_positions()
2673
:return: None
2674
"""
2675
def get_sort_key(key):
2676
index_memo = positions[key][1]
2677
# Group by prefix and position. index_memo[0] is the key, so it is
2678
# (file_id, revision_id) and we don't want to sort on revision_id,
2679
# index_memo[1] is the position, and index_memo[2] is the size,
2680
# which doesn't matter for the sort
2681
return index_memo[0][:-1], index_memo[1]
2682
return keys.sort(key=get_sort_key)
2683
2684
_get_total_build_size = _get_total_build_size
2685
2686
def _split_key(self, key):
2687
"""Split key into a prefix and suffix."""
2688
return key[:-1], key[-1]
2689
2690
2691
class _KnitGraphIndex(object):
2692
"""A KnitVersionedFiles index layered on GraphIndex."""
2693
2694
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2695
add_callback=None):
2696
"""Construct a KnitGraphIndex on a graph_index.
2697
2698
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2699
:param is_locked: A callback to check whether the object should answer
2700
queries.
2701
:param deltas: Allow delta-compressed records.
2702
:param parents: If True, record knits parents, if not do not record
2703
parents.
2704
:param add_callback: If not None, allow additions to the index and call
2705
this callback with a list of added GraphIndex nodes:
2706
[(node, value, node_refs), ...]
2707
:param is_locked: A callback, returns True if the index is locked and
2708
thus usable.
2709
"""
2710
self._add_callback = add_callback
2711
self._graph_index = graph_index
2712
self._deltas = deltas
2713
self._parents = parents
2714
if deltas and not parents:
2715
# XXX: TODO: Delta tree and parent graph should be conceptually
2716
# separate.
2717
raise KnitCorrupt(self, "Cannot do delta compression without "
2718
"parent tracking.")
2719
self.has_graph = parents
2720
self._is_locked = is_locked
2721
self._missing_compression_parents = set()
2722
2723
def __repr__(self):
2724
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2725
2726
def add_records(self, records, random_id=False,
2727
missing_compression_parents=False):
2728
"""Add multiple records to the index.
2729
2730
This function does not insert data into the Immutable GraphIndex
2731
backing the KnitGraphIndex, instead it prepares data for insertion by
2732
the caller and checks that it is safe to insert then calls
2733
self._add_callback with the prepared GraphIndex nodes.
2734
2735
:param records: a list of tuples:
2736
(key, options, access_memo, parents).
2737
:param random_id: If True the ids being added were randomly generated
2738
and no check for existence will be performed.
2739
:param missing_compression_parents: If True the records being added are
2740
only compressed against texts already in the index (or inside
2741
records). If False the records all refer to unavailable texts (or
2742
texts inside records) as compression parents.
2743
"""
2744
if not self._add_callback:
2745
raise errors.ReadOnlyError(self)
2746
# we hope there are no repositories with inconsistent parentage
2747
# anymore.
2748
2749
keys = {}
2750
compression_parents = set()
2751
for (key, options, access_memo, parents) in records:
2752
if self._parents:
2753
parents = tuple(parents)
2754
index, pos, size = access_memo
2755
if 'no-eol' in options:
2756
value = 'N'
2757
else:
2758
value = ' '
2759
value += "%d %d" % (pos, size)
2760
if not self._deltas:
2761
if 'line-delta' in options:
2762
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2763
if self._parents:
2764
if self._deltas:
2765
if 'line-delta' in options:
2766
node_refs = (parents, (parents[0],))
2767
if missing_compression_parents:
2768
compression_parents.add(parents[0])
2769
else:
2770
node_refs = (parents, ())
1271
assert 'line-delta' in options
1272
return 'line-delta'
1273
1274
def get_options(self, version_id):
1275
return self._cache[version_id][1]
1276
1277
def get_parents(self, version_id):
1278
"""Return parents of specified version ignoring ghosts."""
1279
return [parent for parent in self._cache[version_id][4]
1280
if parent in self._cache]
1281
1282
def get_parents_with_ghosts(self, version_id):
1283
"""Return parents of specified version with ghosts."""
1284
return self._cache[version_id][4]
1285
1286
def check_versions_present(self, version_ids):
1287
"""Check that all specified versions are present."""
1288
version_ids = set(version_ids)
1289
for version_id in list(version_ids):
1290
if version_id in self._cache:
1291
version_ids.remove(version_id)
1292
if version_ids:
1293
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1294
1295
1296
class _KnitData(_KnitComponentFile):
1297
"""Contents of the knit data file"""
1298
1299
HEADER = "# bzr knit data 8\n"
1300
1301
def __init__(self, transport, filename, mode, create=False, file_mode=None):
1302
_KnitComponentFile.__init__(self, transport, filename, mode)
1303
self._checked = False
1304
# TODO: jam 20060713 conceptually, this could spill to disk
1305
# if the cached size gets larger than a certain amount
1306
# but it complicates the model a bit, so for now just use
1307
# a simple dictionary
1308
self._cache = {}
1309
self._do_cache = False
1310
if create:
1311
self._transport.put(self._filename, StringIO(''), mode=file_mode)
1312
1313
def enable_cache(self):
1314
"""Enable caching of reads."""
1315
self._do_cache = True
1316
1317
def clear_cache(self):
1318
"""Clear the record cache."""
1319
self._do_cache = False
1320
self._cache = {}
1321
1322
def _open_file(self):
1323
try:
1324
return self._transport.get(self._filename)
1325
except NoSuchFile:
1326
pass
1327
return None
1328
1329
def _record_to_data(self, version_id, digest, lines):
1330
"""Convert version_id, digest, lines into a raw data block.
1331
1332
:return: (len, a StringIO instance with the raw data ready to read.)
1333
"""
1334
sio = StringIO()
1335
data_file = GzipFile(None, mode='wb', fileobj=sio)
1336
1337
version_id_utf8 = cache_utf8.encode(version_id)
1338
data_file.writelines(chain(
1339
["version %s %d %s\n" % (version_id_utf8,
1340
len(lines),
1341
digest)],
1342
lines,
1343
["end %s\n" % version_id_utf8]))
1344
data_file.close()
1345
length= sio.tell()
1346
1347
sio.seek(0)
1348
return length, sio
1349
1350
def add_raw_record(self, raw_data):
1351
"""Append a prepared record to the data file.
1352
1353
:return: the offset in the data file raw_data was written.
1354
"""
1355
assert isinstance(raw_data, str), 'data must be plain bytes'
1356
return self._transport.append(self._filename, StringIO(raw_data))
1357
1358
def add_record(self, version_id, digest, lines):
1359
"""Write new text record to disk. Returns the position in the
1360
file where it was written."""
1361
size, sio = self._record_to_data(version_id, digest, lines)
1362
# write to disk
1363
start_pos = self._transport.append(self._filename, sio)
1364
if self._do_cache:
1365
self._cache[version_id] = sio.getvalue()
1366
return start_pos, size
1367
1368
def _parse_record_header(self, version_id, raw_data):
1369
"""Parse a record header for consistency.
1370
1371
:return: the header and the decompressor stream.
1372
as (stream, header_record)
1373
"""
1374
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1375
rec = df.readline().split()
1376
if len(rec) != 4:
1377
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
1378
if cache_utf8.decode(rec[1]) != version_id:
1379
raise KnitCorrupt(self._filename,
1380
'unexpected version, wanted %r, got %r' % (
1381
version_id, rec[1]))
1382
return df, rec
1383
1384
def _parse_record(self, version_id, data):
1385
# profiling notes:
1386
# 4168 calls in 2880 217 internal
1387
# 4168 calls to _parse_record_header in 2121
1388
# 4168 calls to readlines in 330
1389
df, rec = self._parse_record_header(version_id, data)
1390
record_contents = df.readlines()
1391
l = record_contents.pop()
1392
assert len(record_contents) == int(rec[2])
1393
if l != 'end %s\n' % cache_utf8.encode(version_id):
1394
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
1395
% (l, version_id))
1396
df.close()
1397
return record_contents, rec[3]
1398
1399
def read_records_iter_raw(self, records):
1400
"""Read text records from data file and yield raw data.
1401
1402
This unpacks enough of the text record to validate the id is
1403
as expected but thats all.
1404
"""
1405
# setup an iterator of the external records:
1406
# uses readv so nice and fast we hope.
1407
if len(records):
1408
# grab the disk data needed.
1409
if self._cache:
1410
# Don't check _cache if it is empty
1411
needed_offsets = [(pos, size) for version_id, pos, size
1412
in records
1413
if version_id not in self._cache]
1414
else:
1415
needed_offsets = [(pos, size) for version_id, pos, size
1416
in records]
1417
1418
raw_records = self._transport.readv(self._filename, needed_offsets)
1419
1420
1421
for version_id, pos, size in records:
1422
if version_id in self._cache:
1423
# This data has already been validated
1424
data = self._cache[version_id]
1425
else:
1426
pos, data = raw_records.next()
1427
if self._do_cache:
1428
self._cache[version_id] = data
1429
1430
# validate the header
1431
df, rec = self._parse_record_header(version_id, data)
1432
df.close()
1433
yield version_id, data
1434
1435
def read_records_iter(self, records):
1436
"""Read text records from data file and yield result.
1437
1438
The result will be returned in whatever is the fastest to read.
1439
Not by the order requested. Also, multiple requests for the same
1440
record will only yield 1 response.
1441
:param records: A list of (version_id, pos, len) entries
1442
:return: Yields (version_id, contents, digest) in the order
1443
read, not the order requested
1444
"""
1445
if not records:
1446
return
1447
1448
if self._cache:
1449
# Skip records we have alread seen
1450
yielded_records = set()
1451
needed_records = set()
1452
for record in records:
1453
if record[0] in self._cache:
1454
if record[0] in yielded_records:
1455
continue
1456
yielded_records.add(record[0])
1457
data = self._cache[record[0]]
1458
content, digest = self._parse_record(record[0], data)
1459
yield (record[0], content, digest)
2771
1460
else:
2772
node_refs = (parents, )
2773
else:
2774
if parents:
2775
raise KnitCorrupt(self, "attempt to add node with parents "
2776
"in parentless index.")
2777
node_refs = ()
2778
keys[key] = (value, node_refs)
2779
# check for dups
2780
if not random_id:
2781
present_nodes = self._get_entries(keys)
2782
for (index, key, value, node_refs) in present_nodes:
2783
if (value[0] != keys[key][0][0] or
2784
node_refs[:1] != keys[key][1][:1]):
2785
raise KnitCorrupt(self, "inconsistent details in add_records"
2786
": %s %s" % ((value, node_refs), keys[key]))
2787
del keys[key]
2788
result = []
2789
if self._parents:
2790
for key, (value, node_refs) in keys.iteritems():
2791
result.append((key, value, node_refs))
2792
else:
2793
for key, (value, node_refs) in keys.iteritems():
2794
result.append((key, value))
2795
self._add_callback(result)
2796
if missing_compression_parents:
2797
# This may appear to be incorrect (it does not check for
2798
# compression parents that are in the existing graph index),
2799
# but such records won't have been buffered, so this is
2800
# actually correct: every entry when
2801
# missing_compression_parents==True either has a missing parent, or
2802
# a parent that is one of the keys in records.
2803
compression_parents.difference_update(keys)
2804
self._missing_compression_parents.update(compression_parents)
2805
# Adding records may have satisfied missing compression parents.
2806
self._missing_compression_parents.difference_update(keys)
2807
2808
def scan_unvalidated_index(self, graph_index):
2809
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2810
2811
This allows this _KnitGraphIndex to keep track of any missing
2812
compression parents we may want to have filled in to make those
2813
indices valid.
2814
2815
:param graph_index: A GraphIndex
2816
"""
2817
if self._deltas:
2818
new_missing = graph_index.external_references(ref_list_num=1)
2819
new_missing.difference_update(self.get_parent_map(new_missing))
2820
self._missing_compression_parents.update(new_missing)
2821
2822
def get_missing_compression_parents(self):
2823
"""Return the keys of missing compression parents.
2824
2825
Missing compression parents occur when a record stream was missing
2826
basis texts, or a index was scanned that had missing basis texts.
2827
"""
2828
return frozenset(self._missing_compression_parents)
2829
2830
def _check_read(self):
2831
"""raise if reads are not permitted."""
2832
if not self._is_locked():
2833
raise errors.ObjectNotLocked(self)
2834
2835
def _check_write_ok(self):
2836
"""Assert if writes are not permitted."""
2837
if not self._is_locked():
2838
raise errors.ObjectNotLocked(self)
2839
2840
def _compression_parent(self, an_entry):
2841
# return the key that an_entry is compressed against, or None
2842
# Grab the second parent list (as deltas implies parents currently)
2843
compression_parents = an_entry[3][1]
2844
if not compression_parents:
2845
return None
2846
if len(compression_parents) != 1:
2847
raise AssertionError(
2848
"Too many compression parents: %r" % compression_parents)
2849
return compression_parents[0]
2850
2851
def get_build_details(self, keys):
2852
"""Get the method, index_memo and compression parent for version_ids.
2853
2854
Ghosts are omitted from the result.
2855
2856
:param keys: An iterable of keys.
2857
:return: A dict of key:
2858
(index_memo, compression_parent, parents, record_details).
2859
index_memo
2860
opaque structure to pass to read_records to extract the raw
2861
data
2862
compression_parent
2863
Content that this record is built upon, may be None
2864
parents
2865
Logical parents of this node
2866
record_details
2867
extra information about the content which needs to be passed to
2868
Factory.parse_record
2869
"""
2870
self._check_read()
2871
result = {}
2872
entries = self._get_entries(keys, False)
2873
for entry in entries:
2874
key = entry[1]
2875
if not self._parents:
2876
parents = ()
2877
else:
2878
parents = entry[3][0]
2879
if not self._deltas:
2880
compression_parent_key = None
2881
else:
2882
compression_parent_key = self._compression_parent(entry)
2883
noeol = (entry[2][0] == 'N')
2884
if compression_parent_key:
2885
method = 'line-delta'
2886
else:
2887
method = 'fulltext'
2888
result[key] = (self._node_to_position(entry),
2889
compression_parent_key, parents,
2890
(method, noeol))
2891
return result
2892
2893
def _get_entries(self, keys, check_present=False):
2894
"""Get the entries for keys.
2895
2896
:param keys: An iterable of index key tuples.
2897
"""
2898
keys = set(keys)
2899
found_keys = set()
2900
if self._parents:
2901
for node in self._graph_index.iter_entries(keys):
2902
yield node
2903
found_keys.add(node[1])
2904
else:
2905
# adapt parentless index to the rest of the code.
2906
for node in self._graph_index.iter_entries(keys):
2907
yield node[0], node[1], node[2], ()
2908
found_keys.add(node[1])
2909
if check_present:
2910
missing_keys = keys.difference(found_keys)
2911
if missing_keys:
2912
raise RevisionNotPresent(missing_keys.pop(), self)
2913
2914
def get_method(self, key):
2915
"""Return compression method of specified key."""
2916
return self._get_method(self._get_node(key))
2917
2918
def _get_method(self, node):
2919
if not self._deltas:
2920
return 'fulltext'
2921
if self._compression_parent(node):
2922
return 'line-delta'
2923
else:
2924
return 'fulltext'
2925
2926
def _get_node(self, key):
2927
try:
2928
return list(self._get_entries([key]))[0]
2929
except IndexError:
2930
raise RevisionNotPresent(key, self)
2931
2932
def get_options(self, key):
2933
"""Return a list representing options.
2934
2935
e.g. ['foo', 'bar']
2936
"""
2937
node = self._get_node(key)
2938
options = [self._get_method(node)]
2939
if node[2][0] == 'N':
2940
options.append('no-eol')
2941
return options
2942
2943
def get_parent_map(self, keys):
2944
"""Get a map of the parents of keys.
2945
2946
:param keys: The keys to look up parents for.
2947
:return: A mapping from keys to parents. Absent keys are absent from
2948
the mapping.
2949
"""
2950
self._check_read()
2951
nodes = self._get_entries(keys)
2952
result = {}
2953
if self._parents:
2954
for node in nodes:
2955
result[node[1]] = node[3][0]
2956
else:
2957
for node in nodes:
2958
result[node[1]] = None
2959
return result
2960
2961
def get_position(self, key):
2962
"""Return details needed to access the version.
2963
2964
:return: a tuple (index, data position, size) to hand to the access
2965
logic to get the record.
2966
"""
2967
node = self._get_node(key)
2968
return self._node_to_position(node)
2969
2970
has_key = _mod_index._has_key_from_parent_map
2971
2972
def keys(self):
2973
"""Get all the keys in the collection.
2974
2975
The keys are not ordered.
2976
"""
2977
self._check_read()
2978
return [node[1] for node in self._graph_index.iter_all_entries()]
2979
2980
missing_keys = _mod_index._missing_keys_from_parent_map
2981
2982
def _node_to_position(self, node):
2983
"""Convert an index value to position details."""
2984
bits = node[2][1:].split(' ')
2985
return node[0], int(bits[0]), int(bits[1])
2986
2987
def _sort_keys_by_io(self, keys, positions):
2988
"""Figure out an optimal order to read the records for the given keys.
2989
2990
Sort keys, grouped by index and sorted by position.
2991
2992
:param keys: A list of keys whose records we want to read. This will be
2993
sorted 'in-place'.
2994
:param positions: A dict, such as the one returned by
2995
_get_components_positions()
2996
:return: None
2997
"""
2998
def get_index_memo(key):
2999
# index_memo is at offset [1]. It is made up of (GraphIndex,
3000
# position, size). GI is an object, which will be unique for each
3001
# pack file. This causes us to group by pack file, then sort by
3002
# position. Size doesn't matter, but it isn't worth breaking up the
3003
# tuple.
3004
return positions[key][1]
3005
return keys.sort(key=get_index_memo)
3006
3007
_get_total_build_size = _get_total_build_size
3008
3009
3010
class _KnitKeyAccess(object):
3011
"""Access to records in .knit files."""
3012
3013
def __init__(self, transport, mapper):
3014
"""Create a _KnitKeyAccess with transport and mapper.
3015
3016
:param transport: The transport the access object is rooted at.
3017
:param mapper: The mapper used to map keys to .knit files.
3018
"""
3019
self._transport = transport
3020
self._mapper = mapper
3021
3022
def add_raw_records(self, key_sizes, raw_data):
3023
"""Add raw knit bytes to a storage area.
3024
3025
The data is spooled to the container writer in one bytes-record per
3026
raw data item.
3027
3028
:param sizes: An iterable of tuples containing the key and size of each
3029
raw data segment.
3030
:param raw_data: A bytestring containing the data.
3031
:return: A list of memos to retrieve the record later. Each memo is an
3032
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3033
length), where the key is the record key.
3034
"""
3035
if type(raw_data) != str:
3036
raise AssertionError(
3037
'data must be plain bytes was %s' % type(raw_data))
3038
result = []
3039
offset = 0
3040
# TODO: This can be tuned for writing to sftp and other servers where
3041
# append() is relatively expensive by grouping the writes to each key
3042
# prefix.
3043
for key, size in key_sizes:
3044
path = self._mapper.map(key)
3045
try:
3046
base = self._transport.append_bytes(path + '.knit',
3047
raw_data[offset:offset+size])
3048
except errors.NoSuchFile:
3049
self._transport.mkdir(osutils.dirname(path))
3050
base = self._transport.append_bytes(path + '.knit',
3051
raw_data[offset:offset+size])
3052
# if base == 0:
3053
# chmod.
3054
offset += size
3055
result.append((key, base, size))
3056
return result
3057
3058
def flush(self):
3059
"""Flush pending writes on this access object.
3060
3061
For .knit files this is a no-op.
3062
"""
3063
pass
3064
3065
def get_raw_records(self, memos_for_retrieval):
3066
"""Get the raw bytes for a records.
3067
3068
:param memos_for_retrieval: An iterable containing the access memo for
3069
retrieving the bytes.
3070
:return: An iterator over the bytes of the records.
3071
"""
3072
# first pass, group into same-index request to minimise readv's issued.
3073
request_lists = []
3074
current_prefix = None
3075
for (key, offset, length) in memos_for_retrieval:
3076
if current_prefix == key[:-1]:
3077
current_list.append((offset, length))
3078
else:
3079
if current_prefix is not None:
3080
request_lists.append((current_prefix, current_list))
3081
current_prefix = key[:-1]
3082
current_list = [(offset, length)]
3083
# handle the last entry
3084
if current_prefix is not None:
3085
request_lists.append((current_prefix, current_list))
3086
for prefix, read_vector in request_lists:
3087
path = self._mapper.map(prefix) + '.knit'
3088
for pos, data in self._transport.readv(path, read_vector):
3089
yield data
3090
3091
3092
class _DirectPackAccess(object):
3093
"""Access to data in one or more packs with less translation."""
3094
3095
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
3096
"""Create a _DirectPackAccess object.
3097
3098
:param index_to_packs: A dict mapping index objects to the transport
3099
and file names for obtaining data.
3100
:param reload_func: A function to call if we determine that the pack
3101
files have moved and we need to reload our caches. See
3102
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
3103
"""
3104
self._container_writer = None
3105
self._write_index = None
3106
self._indices = index_to_packs
3107
self._reload_func = reload_func
3108
self._flush_func = flush_func
3109
3110
def add_raw_records(self, key_sizes, raw_data):
3111
"""Add raw knit bytes to a storage area.
3112
3113
The data is spooled to the container writer in one bytes-record per
3114
raw data item.
3115
3116
:param sizes: An iterable of tuples containing the key and size of each
3117
raw data segment.
3118
:param raw_data: A bytestring containing the data.
3119
:return: A list of memos to retrieve the record later. Each memo is an
3120
opaque index memo. For _DirectPackAccess the memo is (index, pos,
3121
length), where the index field is the write_index object supplied
3122
to the PackAccess object.
3123
"""
3124
if type(raw_data) != str:
3125
raise AssertionError(
3126
'data must be plain bytes was %s' % type(raw_data))
3127
result = []
3128
offset = 0
3129
for key, size in key_sizes:
3130
p_offset, p_length = self._container_writer.add_bytes_record(
3131
raw_data[offset:offset+size], [])
3132
offset += size
3133
result.append((self._write_index, p_offset, p_length))
3134
return result
3135
3136
def flush(self):
3137
"""Flush pending writes on this access object.
3138
3139
This will flush any buffered writes to a NewPack.
3140
"""
3141
if self._flush_func is not None:
3142
self._flush_func()
1461
needed_records.add(record)
1462
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1463
else:
1464
needed_records = sorted(set(records), key=operator.itemgetter(1))
1465
1466
if not needed_records:
1467
return
1468
1469
# The transport optimizes the fetching as well
1470
# (ie, reads continuous ranges.)
1471
readv_response = self._transport.readv(self._filename,
1472
[(pos, size) for version_id, pos, size in needed_records])
1473
1474
for (version_id, pos, size), (pos, data) in \
1475
izip(iter(needed_records), readv_response):
1476
content, digest = self._parse_record(version_id, data)
1477
if self._do_cache:
1478
self._cache[version_id] = data
1479
yield version_id, content, digest
1480
1481
def read_records(self, records):
1482
"""Read records into a dictionary."""
1483
components = {}
1484
for record_id, content, digest in \
1485
self.read_records_iter(records):
1486
components[record_id] = (content, digest)
1487
return components
1488
1489
1490
class InterKnit(InterVersionedFile):
1491
"""Optimised code paths for knit to knit operations."""
1492
1493
_matching_file_from_factory = KnitVersionedFile
1494
_matching_file_to_factory = KnitVersionedFile
1495
1496
@staticmethod
1497
def is_compatible(source, target):
1498
"""Be compatible with knits. """
1499
try:
1500
return (isinstance(source, KnitVersionedFile) and
1501
isinstance(target, KnitVersionedFile))
1502
except AttributeError:
1503
return False
1504
1505
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1506
"""See InterVersionedFile.join."""
1507
assert isinstance(self.source, KnitVersionedFile)
1508
assert isinstance(self.target, KnitVersionedFile)
1509
1510
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1511
1512
if not version_ids:
1513
return 0
1514
1515
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1516
try:
1517
version_ids = list(version_ids)
1518
if None in version_ids:
1519
version_ids.remove(None)
1520
1521
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1522
this_versions = set(self.target._index.get_versions())
1523
needed_versions = self.source_ancestry - this_versions
1524
cross_check_versions = self.source_ancestry.intersection(this_versions)
1525
mismatched_versions = set()
1526
for version in cross_check_versions:
1527
# scan to include needed parents.
1528
n1 = set(self.target.get_parents_with_ghosts(version))
1529
n2 = set(self.source.get_parents_with_ghosts(version))
1530
if n1 != n2:
1531
# FIXME TEST this check for cycles being introduced works
1532
# the logic is we have a cycle if in our graph we are an
1533
# ancestor of any of the n2 revisions.
1534
for parent in n2:
1535
if parent in n1:
1536
# safe
1537
continue
1538
else:
1539
parent_ancestors = self.source.get_ancestry(parent)
1540
if version in parent_ancestors:
1541
raise errors.GraphCycleError([parent, version])
1542
# ensure this parent will be available later.
1543
new_parents = n2.difference(n1)
1544
needed_versions.update(new_parents.difference(this_versions))
1545
mismatched_versions.add(version)
1546
1547
if not needed_versions and not mismatched_versions:
1548
return 0
1549
full_list = topo_sort(self.source.get_graph())
1550
1551
version_list = [i for i in full_list if (not self.target.has_version(i)
1552
and i in needed_versions)]
1553
1554
# plan the join:
1555
copy_queue = []
1556
copy_queue_records = []
1557
copy_set = set()
1558
for version_id in version_list:
1559
options = self.source._index.get_options(version_id)
1560
parents = self.source._index.get_parents_with_ghosts(version_id)
1561
# check that its will be a consistent copy:
1562
for parent in parents:
1563
# if source has the parent, we must :
1564
# * already have it or
1565
# * have it scheduled already
1566
# otherwise we don't care
1567
assert (self.target.has_version(parent) or
1568
parent in copy_set or
1569
not self.source.has_version(parent))
1570
data_pos, data_size = self.source._index.get_position(version_id)
1571
copy_queue_records.append((version_id, data_pos, data_size))
1572
copy_queue.append((version_id, options, parents))
1573
copy_set.add(version_id)
1574
1575
# data suck the join:
1576
count = 0
1577
total = len(version_list)
1578
raw_datum = []
1579
raw_records = []
1580
for (version_id, raw_data), \
1581
(version_id2, options, parents) in \
1582
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1583
copy_queue):
1584
assert version_id == version_id2, 'logic error, inconsistent results'
1585
count = count + 1
1586
pb.update("Joining knit", count, total)
1587
raw_records.append((version_id, options, parents, len(raw_data)))
1588
raw_datum.append(raw_data)
1589
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1590
1591
for version in mismatched_versions:
1592
# FIXME RBC 20060309 is this needed?
1593
n1 = set(self.target.get_parents_with_ghosts(version))
1594
n2 = set(self.source.get_parents_with_ghosts(version))
1595
# write a combined record to our history preserving the current
1596
# parents as first in the list
1597
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1598
self.target.fix_parents(version, new_parents)
1599
return count
1600
finally:
1601
pb.finished()
1602
1603
1604
InterVersionedFile.register_optimiser(InterKnit)
1605
1606
1607
class WeaveToKnit(InterVersionedFile):
1608
"""Optimised code paths for weave to knit operations."""
1609
1610
_matching_file_from_factory = bzrlib.weave.WeaveFile
1611
_matching_file_to_factory = KnitVersionedFile
1612
1613
@staticmethod
1614
def is_compatible(source, target):
1615
"""Be compatible with weaves to knits."""
1616
try:
1617
return (isinstance(source, bzrlib.weave.Weave) and
1618
isinstance(target, KnitVersionedFile))
1619
except AttributeError:
1620
return False
1621
1622
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1623
"""See InterVersionedFile.join."""
1624
assert isinstance(self.source, bzrlib.weave.Weave)
1625
assert isinstance(self.target, KnitVersionedFile)
1626
1627
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1628
1629
if not version_ids:
1630
return 0
1631
1632
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1633
try:
1634
version_ids = list(version_ids)
1635
1636
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1637
this_versions = set(self.target._index.get_versions())
1638
needed_versions = self.source_ancestry - this_versions
1639
cross_check_versions = self.source_ancestry.intersection(this_versions)
1640
mismatched_versions = set()
1641
for version in cross_check_versions:
1642
# scan to include needed parents.
1643
n1 = set(self.target.get_parents_with_ghosts(version))
1644
n2 = set(self.source.get_parents(version))
1645
# if all of n2's parents are in n1, then its fine.
1646
if n2.difference(n1):
1647
# FIXME TEST this check for cycles being introduced works
1648
# the logic is we have a cycle if in our graph we are an
1649
# ancestor of any of the n2 revisions.
1650
for parent in n2:
1651
if parent in n1:
1652
# safe
1653
continue
1654
else:
1655
parent_ancestors = self.source.get_ancestry(parent)
1656
if version in parent_ancestors:
1657
raise errors.GraphCycleError([parent, version])
1658
# ensure this parent will be available later.
1659
new_parents = n2.difference(n1)
1660
needed_versions.update(new_parents.difference(this_versions))
1661
mismatched_versions.add(version)
1662
1663
if not needed_versions and not mismatched_versions:
1664
return 0
1665
full_list = topo_sort(self.source.get_graph())
1666
1667
version_list = [i for i in full_list if (not self.target.has_version(i)
1668
and i in needed_versions)]
1669
1670
# do the join:
1671
count = 0
1672
total = len(version_list)
1673
for version_id in version_list:
1674
pb.update("Converting to knit", count, total)
1675
parents = self.source.get_parents(version_id)
1676
# check that its will be a consistent copy:
1677
for parent in parents:
1678
# if source has the parent, we must already have it
1679
assert (self.target.has_version(parent))
1680
self.target.add_lines(
1681
version_id, parents, self.source.get_lines(version_id))
1682
count = count + 1
1683
1684
for version in mismatched_versions:
1685
# FIXME RBC 20060309 is this needed?
1686
n1 = set(self.target.get_parents_with_ghosts(version))
1687
n2 = set(self.source.get_parents(version))
1688
# write a combined record to our history preserving the current
1689
# parents as first in the list
1690
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1691
self.target.fix_parents(version, new_parents)
1692
return count
1693
finally:
1694
pb.finished()
1695
1696
1697
InterVersionedFile.register_optimiser(WeaveToKnit)
1698
1699
1700
class KnitSequenceMatcher(difflib.SequenceMatcher):
1701
"""Knit tuned sequence matcher.
1702
1703
This is based on profiling of difflib which indicated some improvements
1704
for our usage pattern.
1705
"""
1706
1707
def find_longest_match(self, alo, ahi, blo, bhi):
1708
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1709
1710
If isjunk is not defined:
1711
1712
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1713
alo <= i <= i+k <= ahi
1714
blo <= j <= j+k <= bhi
1715
and for all (i',j',k') meeting those conditions,
1716
k >= k'
1717
i <= i'
1718
and if i == i', j <= j'
1719
1720
In other words, of all maximal matching blocks, return one that
1721
starts earliest in a, and of all those maximal matching blocks that
1722
start earliest in a, return the one that starts earliest in b.
1723
1724
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1725
>>> s.find_longest_match(0, 5, 0, 9)
1726
(0, 4, 5)
1727
1728
If isjunk is defined, first the longest matching block is
1729
determined as above, but with the additional restriction that no
1730
junk element appears in the block. Then that block is extended as
1731
far as possible by matching (only) junk elements on both sides. So
1732
the resulting block never matches on junk except as identical junk
1733
happens to be adjacent to an "interesting" match.
1734
1735
Here's the same example as before, but considering blanks to be
1736
junk. That prevents " abcd" from matching the " abcd" at the tail
1737
end of the second sequence directly. Instead only the "abcd" can
1738
match, and matches the leftmost "abcd" in the second sequence:
1739
1740
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1741
>>> s.find_longest_match(0, 5, 0, 9)
1742
(1, 0, 4)
1743
1744
If no blocks match, return (alo, blo, 0).
1745
1746
>>> s = SequenceMatcher(None, "ab", "c")
1747
>>> s.find_longest_match(0, 2, 0, 1)
1748
(0, 0, 0)
1749
"""
1750
1751
# CAUTION: stripping common prefix or suffix would be incorrect.
1752
# E.g.,
1753
# ab
1754
# acab
1755
# Longest matching block is "ab", but if common prefix is
1756
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1757
# strip, so ends up claiming that ab is changed to acab by
1758
# inserting "ca" in the middle. That's minimal but unintuitive:
1759
# "it's obvious" that someone inserted "ac" at the front.
1760
# Windiff ends up at the same place as diff, but by pairing up
1761
# the unique 'b's and then matching the first two 'a's.
1762
1763
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1764
besti, bestj, bestsize = alo, blo, 0
1765
# find longest junk-free match
1766
# during an iteration of the loop, j2len[j] = length of longest
1767
# junk-free match ending with a[i-1] and b[j]
1768
j2len = {}
1769
# nothing = []
1770
b2jget = b2j.get
1771
for i in xrange(alo, ahi):
1772
# look at all instances of a[i] in b; note that because
1773
# b2j has no junk keys, the loop is skipped if a[i] is junk
1774
j2lenget = j2len.get
1775
newj2len = {}
3143
1776
3144
def get_raw_records(self, memos_for_retrieval):
3145
"""Get the raw bytes for a records.
1777
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1778
# following improvement
1779
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1780
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1781
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1782
# to
1783
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1784
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1785
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
3146
1786
3147
:param memos_for_retrieval: An iterable containing the (index, pos,
3148
length) memo for retrieving the bytes. The Pack access method
3149
looks up the pack to use for a given record in its index_to_pack
3150
map.
3151
:return: An iterator over the bytes of the records.
3152
"""
3153
# first pass, group into same-index requests
3154
request_lists = []
3155
current_index = None
3156
for (index, offset, length) in memos_for_retrieval:
3157
if current_index == index:
3158
current_list.append((offset, length))
3159
else:
3160
if current_index is not None:
3161
request_lists.append((current_index, current_list))
3162
current_index = index
3163
current_list = [(offset, length)]
3164
# handle the last entry
3165
if current_index is not None:
3166
request_lists.append((current_index, current_list))
3167
for index, offsets in request_lists:
3168
1787
try:
3169
transport, path = self._indices[index]
1788
js = b2j[a[i]]
3170
1789
except KeyError:
3171
# A KeyError here indicates that someone has triggered an index
3172
# reload, and this index has gone missing, we need to start
3173
# over.
3174
if self._reload_func is None:
3175
# If we don't have a _reload_func there is nothing that can
3176
# be done
3177
raise
3178
raise errors.RetryWithNewPacks(index,
3179
reload_occurred=True,
3180
exc_info=sys.exc_info())
3181
try:
3182
reader = pack.make_readv_reader(transport, path, offsets)
3183
for names, read_func in reader.iter_records():
3184
yield read_func(None)
3185
except errors.NoSuchFile:
3186
# A NoSuchFile error indicates that a pack file has gone
3187
# missing on disk, we need to trigger a reload, and start over.
3188
if self._reload_func is None:
3189
raise
3190
raise errors.RetryWithNewPacks(transport.abspath(path),
3191
reload_occurred=False,
3192
exc_info=sys.exc_info())
3193
3194
def set_writer(self, writer, index, transport_packname):
3195
"""Set a writer to use for adding data."""
3196
if index is not None:
3197
self._indices[index] = transport_packname
3198
self._container_writer = writer
3199
self._write_index = index
3200
3201
def reload_or_raise(self, retry_exc):
3202
"""Try calling the reload function, or re-raise the original exception.
3203
3204
This should be called after _DirectPackAccess raises a
3205
RetryWithNewPacks exception. This function will handle the common logic
3206
of determining when the error is fatal versus being temporary.
3207
It will also make sure that the original exception is raised, rather
3208
than the RetryWithNewPacks exception.
3209
3210
If this function returns, then the calling function should retry
3211
whatever operation was being performed. Otherwise an exception will
3212
be raised.
3213
3214
:param retry_exc: A RetryWithNewPacks exception.
3215
"""
3216
is_error = False
3217
if self._reload_func is None:
3218
is_error = True
3219
elif not self._reload_func():
3220
# The reload claimed that nothing changed
3221
if not retry_exc.reload_occurred:
3222
# If there wasn't an earlier reload, then we really were
3223
# expecting to find changes. We didn't find them, so this is a
3224
# hard error
3225
is_error = True
3226
if is_error:
3227
exc_class, exc_value, exc_traceback = retry_exc.exc_info
3228
raise exc_class, exc_value, exc_traceback
3229
3230
3231
# Deprecated, use PatienceSequenceMatcher instead
3232
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
3233
3234
3235
def annotate_knit(knit, revision_id):
3236
"""Annotate a knit with no cached annotations.
3237
3238
This implementation is for knits with no cached annotations.
3239
It will work for knits with cached annotations, but this is not
3240
recommended.
3241
"""
3242
annotator = _KnitAnnotator(knit)
3243
return iter(annotator.annotate(revision_id))
3244
3245
3246
class _KnitAnnotator(object):
3247
"""Build up the annotations for a text."""
3248
3249
def __init__(self, knit):
3250
self._knit = knit
3251
3252
# Content objects, differs from fulltexts because of how final newlines
3253
# are treated by knits. the content objects here will always have a
3254
# final newline
3255
self._fulltext_contents = {}
3256
3257
# Annotated lines of specific revisions
3258
self._annotated_lines = {}
3259
3260
# Track the raw data for nodes that we could not process yet.
3261
# This maps the revision_id of the base to a list of children that will
3262
# annotated from it.
3263
self._pending_children = {}
3264
3265
# Nodes which cannot be extracted
3266
self._ghosts = set()
3267
3268
# Track how many children this node has, so we know if we need to keep
3269
# it
3270
self._annotate_children = {}
3271
self._compression_children = {}
3272
3273
self._all_build_details = {}
3274
# The children => parent revision_id graph
3275
self._revision_id_graph = {}
3276
3277
self._heads_provider = None
3278
3279
self._nodes_to_keep_annotations = set()
3280
self._generations_until_keep = 100
3281
3282
def set_generations_until_keep(self, value):
3283
"""Set the number of generations before caching a node.
3284
3285
Setting this to -1 will cache every merge node, setting this higher
3286
will cache fewer nodes.
3287
"""
3288
self._generations_until_keep = value
3289
3290
def _add_fulltext_content(self, revision_id, content_obj):
3291
self._fulltext_contents[revision_id] = content_obj
3292
# TODO: jam 20080305 It might be good to check the sha1digest here
3293
return content_obj.text()
3294
3295
def _check_parents(self, child, nodes_to_annotate):
3296
"""Check if all parents have been processed.
3297
3298
:param child: A tuple of (rev_id, parents, raw_content)
3299
:param nodes_to_annotate: If child is ready, add it to
3300
nodes_to_annotate, otherwise put it back in self._pending_children
3301
"""
3302
for parent_id in child[1]:
3303
if (parent_id not in self._annotated_lines):
3304
# This parent is present, but another parent is missing
3305
self._pending_children.setdefault(parent_id,
3306
[]).append(child)
3307
break
3308
else:
3309
# This one is ready to be processed
3310
nodes_to_annotate.append(child)
3311
3312
def _add_annotation(self, revision_id, fulltext, parent_ids,
3313
left_matching_blocks=None):
3314
"""Add an annotation entry.
3315
3316
All parents should already have been annotated.
3317
:return: A list of children that now have their parents satisfied.
3318
"""
3319
a = self._annotated_lines
3320
annotated_parent_lines = [a[p] for p in parent_ids]
3321
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
3322
fulltext, revision_id, left_matching_blocks,
3323
heads_provider=self._get_heads_provider()))
3324
self._annotated_lines[revision_id] = annotated_lines
3325
for p in parent_ids:
3326
ann_children = self._annotate_children[p]
3327
ann_children.remove(revision_id)
3328
if (not ann_children
3329
and p not in self._nodes_to_keep_annotations):
3330
del self._annotated_lines[p]
3331
del self._all_build_details[p]
3332
if p in self._fulltext_contents:
3333
del self._fulltext_contents[p]
3334
# Now that we've added this one, see if there are any pending
3335
# deltas to be done, certainly this parent is finished
3336
nodes_to_annotate = []
3337
for child in self._pending_children.pop(revision_id, []):
3338
self._check_parents(child, nodes_to_annotate)
3339
return nodes_to_annotate
3340
3341
def _get_build_graph(self, key):
3342
"""Get the graphs for building texts and annotations.
3343
3344
The data you need for creating a full text may be different than the
3345
data you need to annotate that text. (At a minimum, you need both
3346
parents to create an annotation, but only need 1 parent to generate the
3347
fulltext.)
3348
3349
:return: A list of (key, index_memo) records, suitable for
3350
passing to read_records_iter to start reading in the raw data fro/
3351
the pack file.
3352
"""
3353
if key in self._annotated_lines:
3354
# Nothing to do
3355
return []
3356
pending = set([key])
3357
records = []
3358
generation = 0
3359
kept_generation = 0
3360
while pending:
3361
# get all pending nodes
3362
generation += 1
3363
this_iteration = pending
3364
build_details = self._knit._index.get_build_details(this_iteration)
3365
self._all_build_details.update(build_details)
3366
# new_nodes = self._knit._index._get_entries(this_iteration)
3367
pending = set()
3368
for key, details in build_details.iteritems():
3369
(index_memo, compression_parent, parents,
3370
record_details) = details
3371
self._revision_id_graph[key] = parents
3372
records.append((key, index_memo))
3373
# Do we actually need to check _annotated_lines?
3374
pending.update(p for p in parents
3375
if p not in self._all_build_details)
3376
if compression_parent:
3377
self._compression_children.setdefault(compression_parent,
3378
[]).append(key)
3379
if parents:
3380
for parent in parents:
3381
self._annotate_children.setdefault(parent,
3382
[]).append(key)
3383
num_gens = generation - kept_generation
3384
if ((num_gens >= self._generations_until_keep)
3385
and len(parents) > 1):
3386
kept_generation = generation
3387
self._nodes_to_keep_annotations.add(key)
3388
3389
missing_versions = this_iteration.difference(build_details.keys())
3390
self._ghosts.update(missing_versions)
3391
for missing_version in missing_versions:
3392
# add a key, no parents
3393
self._revision_id_graph[missing_version] = ()
3394
pending.discard(missing_version) # don't look for it
3395
if self._ghosts.intersection(self._compression_children):
3396
raise KnitCorrupt(
3397
"We cannot have nodes which have a ghost compression parent:\n"
3398
"ghosts: %r\n"
3399
"compression children: %r"
3400
% (self._ghosts, self._compression_children))
3401
# Cleanout anything that depends on a ghost so that we don't wait for
3402
# the ghost to show up
3403
for node in self._ghosts:
3404
if node in self._annotate_children:
3405
# We won't be building this node
3406
del self._annotate_children[node]
3407
# Generally we will want to read the records in reverse order, because
3408
# we find the parent nodes after the children
3409
records.reverse()
3410
return records
3411
3412
def _annotate_records(self, records):
3413
"""Build the annotations for the listed records."""
3414
# We iterate in the order read, rather than a strict order requested
3415
# However, process what we can, and put off to the side things that
3416
# still need parents, cleaning them up when those parents are
3417
# processed.
3418
for (rev_id, record,
3419
digest) in self._knit._read_records_iter(records):
3420
if rev_id in self._annotated_lines:
3421
continue
3422
parent_ids = self._revision_id_graph[rev_id]
3423
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3424
details = self._all_build_details[rev_id]
3425
(index_memo, compression_parent, parents,
3426
record_details) = details
3427
nodes_to_annotate = []
3428
# TODO: Remove the punning between compression parents, and
3429
# parent_ids, we should be able to do this without assuming
3430
# the build order
3431
if len(parent_ids) == 0:
3432
# There are no parents for this node, so just add it
3433
# TODO: This probably needs to be decoupled
3434
fulltext_content, delta = self._knit._factory.parse_record(
3435
rev_id, record, record_details, None)
3436
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3437
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3438
parent_ids, left_matching_blocks=None))
3439
else:
3440
child = (rev_id, parent_ids, record)
3441
# Check if all the parents are present
3442
self._check_parents(child, nodes_to_annotate)
3443
while nodes_to_annotate:
3444
# Should we use a queue here instead of a stack?
3445
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3446
(index_memo, compression_parent, parents,
3447
record_details) = self._all_build_details[rev_id]
3448
blocks = None
3449
if compression_parent is not None:
3450
comp_children = self._compression_children[compression_parent]
3451
if rev_id not in comp_children:
3452
raise AssertionError("%r not in compression children %r"
3453
% (rev_id, comp_children))
3454
# If there is only 1 child, it is safe to reuse this
3455
# content
3456
reuse_content = (len(comp_children) == 1
3457
and compression_parent not in
3458
self._nodes_to_keep_annotations)
3459
if reuse_content:
3460
# Remove it from the cache since it will be changing
3461
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3462
# Make sure to copy the fulltext since it might be
3463
# modified
3464
parent_fulltext = list(parent_fulltext_content.text())
3465
else:
3466
parent_fulltext_content = self._fulltext_contents[compression_parent]
3467
parent_fulltext = parent_fulltext_content.text()
3468
comp_children.remove(rev_id)
3469
fulltext_content, delta = self._knit._factory.parse_record(
3470
rev_id, record, record_details,
3471
parent_fulltext_content,
3472
copy_base_content=(not reuse_content))
3473
fulltext = self._add_fulltext_content(rev_id,
3474
fulltext_content)
3475
if compression_parent == parent_ids[0]:
3476
# the compression_parent is the left parent, so we can
3477
# re-use the delta
3478
blocks = KnitContent.get_line_delta_blocks(delta,
3479
parent_fulltext, fulltext)
3480
else:
3481
fulltext_content = self._knit._factory.parse_fulltext(
3482
record, rev_id)
3483
fulltext = self._add_fulltext_content(rev_id,
3484
fulltext_content)
3485
nodes_to_annotate.extend(
3486
self._add_annotation(rev_id, fulltext, parent_ids,
3487
left_matching_blocks=blocks))
3488
3489
def _get_heads_provider(self):
3490
"""Create a heads provider for resolving ancestry issues."""
3491
if self._heads_provider is not None:
3492
return self._heads_provider
3493
parent_provider = _mod_graph.DictParentsProvider(
3494
self._revision_id_graph)
3495
graph_obj = _mod_graph.Graph(parent_provider)
3496
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3497
self._heads_provider = head_cache
3498
return head_cache
3499
3500
def annotate(self, key):
3501
"""Return the annotated fulltext at the given key.
3502
3503
:param key: The key to annotate.
3504
"""
3505
if len(self._knit._fallback_vfs) > 0:
3506
# stacked knits can't use the fast path at present.
3507
return self._simple_annotate(key)
3508
while True:
3509
try:
3510
records = self._get_build_graph(key)
3511
if key in self._ghosts:
3512
raise errors.RevisionNotPresent(key, self._knit)
3513
self._annotate_records(records)
3514
return self._annotated_lines[key]
3515
except errors.RetryWithNewPacks, e:
3516
self._knit._access.reload_or_raise(e)
3517
# The cached build_details are no longer valid
3518
self._all_build_details.clear()
3519
3520
def _simple_annotate(self, key):
3521
"""Return annotated fulltext, rediffing from the full texts.
3522
3523
This is slow but makes no assumptions about the repository
3524
being able to produce line deltas.
3525
"""
3526
# TODO: this code generates a parent maps of present ancestors; it
3527
# could be split out into a separate method, and probably should use
3528
# iter_ancestry instead. -- mbp and robertc 20080704
3529
graph = _mod_graph.Graph(self._knit)
3530
head_cache = _mod_graph.FrozenHeadsCache(graph)
3531
search = graph._make_breadth_first_searcher([key])
3532
keys = set()
3533
while True:
3534
try:
3535
present, ghosts = search.next_with_ghosts()
3536
except StopIteration:
3537
break
3538
keys.update(present)
3539
parent_map = self._knit.get_parent_map(keys)
3540
parent_cache = {}
3541
reannotate = annotate.reannotate
3542
for record in self._knit.get_record_stream(keys, 'topological', True):
3543
key = record.key
3544
fulltext = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
3545
parents = parent_map[key]
3546
if parents is not None:
3547
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
3548
else:
3549
parent_lines = []
3550
parent_cache[key] = list(
3551
reannotate(parent_lines, fulltext, key, None, head_cache))
3552
try:
3553
return parent_cache[key]
3554
except KeyError, e:
3555
raise errors.RevisionNotPresent(key, self._knit)
3556
3557
3558
try:
3559
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3560
except ImportError:
3561
from bzrlib._knit_load_data_py import _load_data_py as _load_data
1790
pass
1791
else:
1792
for j in js:
1793
# a[i] matches b[j]
1794
if j >= blo:
1795
if j >= bhi:
1796
break
1797
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1798
if k > bestsize:
1799
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1800
j2len = newj2len
1801
1802
# Extend the best by non-junk elements on each end. In particular,
1803
# "popular" non-junk elements aren't in b2j, which greatly speeds
1804
# the inner loop above, but also means "the best" match so far
1805
# doesn't contain any junk *or* popular non-junk elements.
1806
while besti > alo and bestj > blo and \
1807
not isbjunk(b[bestj-1]) and \
1808
a[besti-1] == b[bestj-1]:
1809
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1810
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1811
not isbjunk(b[bestj+bestsize]) and \
1812
a[besti+bestsize] == b[bestj+bestsize]:
1813
bestsize += 1
1814
1815
# Now that we have a wholly interesting match (albeit possibly
1816
# empty!), we may as well suck up the matching junk on each
1817
# side of it too. Can't think of a good reason not to, and it
1818
# saves post-processing the (possibly considerable) expense of
1819
# figuring out what to do with it. In the case of an empty
1820
# interesting match, this is clearly the right thing to do,
1821
# because no other kind of match is possible in the regions.
1822
while besti > alo and bestj > blo and \
1823
isbjunk(b[bestj-1]) and \
1824
a[besti-1] == b[bestj-1]:
1825
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1826
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1827
isbjunk(b[bestj+bestsize]) and \
1828
a[besti+bestsize] == b[bestj+bestsize]:
1829
bestsize = bestsize + 1
1830
1831
return besti, bestj, bestsize
1832
Loggerhead is a web-based interface for Breezy
Version: 2.0.1