« back to all changes in this revision
Viewing changes to bzrlib/knit.py
- browse files at revision 2382.2.4
- compare with another revision
- download diff
- download tarball
- view history from revision 2382.2.4
- Committer: Martin Pool
- Date: 2007-04-04 01:22:11 UTC
- mfrom: (2393.1.1 bzr.docs)
- mto: This revision was merged to the branch mainline in revision 2397.
- Revision ID: mbp@sourcefrog.net-20070404012211-sq269me6bai9m6xk
merge trunk and doc fix from elliot
- files added:
- build-api
- bzrlib/bundle/old
- bzrlib/tests/interversionedfile_implementations
- bzrlib/tests/revisionstore_implementations
- bzrlib/util/configobj/docs
- 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/smart
- bzrlib/tests/commands
- bzrlib/tests/ftp_server
- bzrlib/tests/inventory_implementations
- bzrlib/tests/per_interbranch
- bzrlib/tests/per_repository_chk
- bzrlib/tests/per_repository_reference
- bzrlib/tests/ssl_certs
- bzrlib/transport/ftp
- bzrlib/util/tests
- doc/developers
- doc/en
- doc/en/admin-guide
- doc/en/developer-guide
- doc/en/mini-tutorial
- doc/en/quick-reference
- doc/en/release-notes
- doc/en/tutorials
- doc/en/user-guide
- doc/en/user-guide/images
- doc/en/user-reference
- 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/help_topics/__init__.py => bzrlib/help_topics.py
- bzrlib/_patiencediff_py.py => bzrlib/patiencediff.py
- bzrlib/plugins/launchpad/lp_directory.py => bzrlib/plugins/launchpad/lp_indirect.py
- bzrlib/plugins/launchpad/test_lp_directory.py => bzrlib/plugins/launchpad/test_lp_indirect.py
- bzrlib/tests/http_utils.py => bzrlib/tests/HTTPTestUtil.py
- bzrlib/tests/http_server.py => bzrlib/tests/HttpServer.py
- bzrlib/tests/blackbox/test_send.py => bzrlib/tests/blackbox/test_bundle.py
- bzrlib/tests/per_repository/ => bzrlib/tests/repository_implementations/
- 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/transport/remote.py => bzrlib/transport/smart.py
- bzrlib/xml8.py => bzrlib/xml5.py
- doc/en/tutorials/centralized_workflow.txt => doc/centralized_workflow.txt
- bzrlib/help_topics/en/configuration.txt => doc/configuration.txt
- doc/en/user-guide/http_smart_server.txt => doc/http_smart_server.txt
- doc/en/user-guide/index.txt => doc/index.txt
- doc/en/user-guide/plugins.txt => doc/plugins.txt
- doc/en/user-guide/server.txt => doc/server.txt
- doc/en/user-guide/setting_up_email.txt => doc/setting_up_email.txt
- doc/en/user-guide/specifying_revisions.txt => doc/specifying_revisions.txt
- doc/en/tutorials/tutorial.txt => doc/tutorial.txt
- doc/en/user-guide/using_aliases.txt => doc/using_aliases.txt
- doc/en/user-guide/version_info.txt => doc/version_info.txt
- files modified:
- .bzrignore
- setup.py *
added
removed
bzrlib/knit.py
1
# Copyright (C) 2005, 2006, 2007, 2008 Canonical Ltd
1
# Copyright (C) 2005, 2006 Canonical Ltd
2
2
#
3
3
# This program is free software; you can redistribute it and/or modify
4
4
# it under the terms of the GNU General Public License as published by
12
12
#
13
13
# You should have received a copy of the GNU General Public License
14
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
16
16
17
17
"""Knit versionedfile implementation.
18
18
20
20
updates.
21
21
22
22
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
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
26
26
end-marker; simply "end VERSION"
27
27
28
28
delta can be line or full contents.a
35
35
130,130,2
36
36
8 if elt.get('executable') == 'yes':
37
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
39
39
40
40
41
41
whats in an index:
51
51
52
52
"""
53
53
54
# TODOS:
55
# 10:16 < lifeless> make partial index writes safe
56
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
57
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
58
# always' approach.
59
# move sha1 out of the content so that join is faster at verifying parents
60
# record content length ?
61
54
62
63
from copy import copy
55
64
from cStringIO import StringIO
65
import difflib
56
66
from itertools import izip, chain
57
67
import operator
58
68
import os
59
69
import sys
70
import warnings
60
71
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 (
72
import bzrlib
73
from bzrlib import (
74
cache_utf8,
78
75
errors,
79
76
osutils,
80
77
patiencediff,
78
progress,
79
ui,
81
80
)
82
81
from bzrlib.errors import (
83
82
FileExists,
88
87
KnitHeaderError,
89
88
RevisionNotPresent,
90
89
RevisionAlreadyPresent,
91
SHA1KnitCorrupt,
92
90
)
91
from bzrlib.tuned_gzip import GzipFile
92
from bzrlib.trace import mutter
93
93
from bzrlib.osutils import (
94
94
contains_whitespace,
95
95
contains_linebreaks,
96
sha_string,
97
96
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
)
97
)
98
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
99
from bzrlib.tsort import topo_sort
100
import bzrlib.ui
101
import bzrlib.weave
102
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
110
103
111
104
112
105
# TODO: Split out code specific to this format into an associated object.
124
117
125
118
DATA_SUFFIX = '.knit'
126
119
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
120
409
121
410
122
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)
123
"""Content of a knit version to which deltas can be applied."""
124
125
def __init__(self, lines):
126
self._lines = lines
127
128
def annotate_iter(self):
129
"""Yield tuples of (origin, text) for each content line."""
130
return iter(self._lines)
131
132
def annotate(self):
133
"""Return a list of (origin, text) tuples."""
134
return list(self.annotate_iter())
424
135
425
136
def line_delta_iter(self, new_lines):
426
137
"""Generate line-based delta from this content to new_lines."""
427
138
new_texts = new_lines.text()
428
139
old_texts = self.text()
429
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
140
s = KnitSequenceMatcher(None, old_texts, new_texts)
430
141
for tag, i1, i2, j1, j2 in s.get_opcodes():
431
142
if tag == 'equal':
432
143
continue
436
147
def line_delta(self, new_lines):
437
148
return list(self.line_delta_iter(new_lines))
438
149
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
150
def text(self):
151
return [text for origin, text in self._lines]
152
153
def copy(self):
154
return KnitContent(self._lines[:])
542
155
543
156
544
157
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)
158
"""Base factory for creating content objects."""
159
160
def make(self, lines, version_id):
161
num_lines = len(lines)
162
return KnitContent(zip([version_id] * num_lines, lines))
576
163
577
164
578
165
class KnitAnnotateFactory(_KnitFactory):
580
167
581
168
annotated = True
582
169
583
def make(self, lines, version_id):
584
num_lines = len(lines)
585
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
586
587
170
def parse_fulltext(self, content, version_id):
588
171
"""Convert fulltext to internal representation
589
172
597
180
# Figure out a way to not require the overhead of turning the
598
181
# list back into tuples.
599
182
lines = [tuple(line.split(' ', 1)) for line in content]
600
return AnnotatedKnitContent(lines)
183
return KnitContent(lines)
601
184
602
185
def parse_line_delta_iter(self, lines):
603
186
return iter(self.parse_line_delta(lines))
604
187
605
def parse_line_delta(self, lines, version_id, plain=False):
188
def parse_line_delta(self, lines, version_id):
606
189
"""Convert a line based delta into internal representation.
607
190
608
191
line delta is in the form of:
611
194
revid(utf8) newline\n
612
195
internal representation is
613
196
(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
197
"""
619
198
result = []
620
199
lines = iter(lines)
626
205
return cache.setdefault(origin, origin), text
627
206
628
207
# 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))
208
for header in lines:
209
start, end, count = [int(n) for n in header.split(',')]
210
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
211
result.append((start, end, count, contents))
641
212
return result
642
213
643
214
def get_fulltext_content(self, lines):
682
253
for origin, text in lines)
683
254
return out
684
255
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
256
702
257
class KnitPlainFactory(_KnitFactory):
703
258
"""Factory for creating plain Content objects."""
704
259
705
260
annotated = False
706
261
707
def make(self, lines, version_id):
708
return PlainKnitContent(lines, version_id)
709
710
262
def parse_fulltext(self, content, version_id):
711
263
"""This parses an unannotated fulltext.
712
264
722
274
header = lines[cur]
723
275
cur += 1
724
276
start, end, c = [int(n) for n in header.split(',')]
725
yield start, end, c, lines[cur:cur+c]
277
yield start, end, c, zip([version_id] * c, lines[cur:cur+c])
726
278
cur += c
727
279
728
280
def parse_line_delta(self, lines, version_id):
753
305
out = []
754
306
for start, end, c, lines in delta:
755
307
out.append('%d,%d,%d\n' % (start, end, c))
756
out.extend(lines)
308
out.extend([text for origin, text in lines])
757
309
return out
758
310
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
796
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:
842
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.
311
312
def make_empty_knit(transport, relpath):
313
"""Construct a empty knit at the specified location."""
314
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
315
k._data._open_file()
316
317
318
class KnitVersionedFile(VersionedFile):
319
"""Weave-like structure with faster random access.
320
321
A knit stores a number of texts and a summary of the relationships
322
between them. Texts are identified by a string version-id. Texts
323
are normally stored and retrieved as a series of lines, but can
324
also be passed as single strings.
325
326
Lines are stored with the trailing newline (if any) included, to
327
avoid special cases for files with no final newline. Lines are
328
composed of 8-bit characters, not unicode. The combination of
329
these approaches should mean any 'binary' file can be safely
330
stored and retrieved.
331
"""
332
333
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
334
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
335
create=False, create_parent_dir=False, delay_create=False,
336
dir_mode=None):
337
"""Construct a knit at location specified by relpath.
338
339
:param create: If not True, only open an existing knit.
340
:param create_parent_dir: If True, create the parent directory if
341
creating the file fails. (This is used for stores with
342
hash-prefixes that may not exist yet)
343
:param delay_create: The calling code is aware that the knit won't
344
actually be created until the first data is stored.
877
345
"""
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
346
if deprecated_passed(basis_knit):
347
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
348
" deprecated as of bzr 0.9.",
349
DeprecationWarning, stacklevel=2)
350
if access_mode is None:
351
access_mode = 'w'
352
super(KnitVersionedFile, self).__init__(access_mode)
353
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
354
self.transport = transport
355
self.filename = relpath
356
self.factory = factory or KnitAnnotateFactory()
357
self.writable = (access_mode == 'w')
358
self.delta = delta
359
360
self._max_delta_chain = 200
361
362
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
363
access_mode, create=create, file_mode=file_mode,
364
create_parent_dir=create_parent_dir, delay_create=delay_create,
365
dir_mode=dir_mode)
366
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
367
access_mode, create=create and not len(self), file_mode=file_mode,
368
create_parent_dir=create_parent_dir, delay_create=delay_create,
369
dir_mode=dir_mode)
887
370
888
371
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):
917
"""Add a set of lines on top of version specified by parents.
918
919
Any versions not present will be converted into ghosts.
920
"""
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
927
928
present_parents = []
929
if parent_texts is None:
930
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
for parent in parents:
935
if parent in present_parent_map:
936
present_parents.append(parent)
937
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]):
941
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
948
text_length = len(line_bytes)
949
options = []
950
if lines:
951
if lines[-1][-1] != '\n':
952
# copy the contents of lines.
953
lines = lines[:]
954
options.append('no-eol')
955
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)
977
978
if delta:
979
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)
983
else:
984
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)
1007
1008
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):
372
return '%s(%s)' % (self.__class__.__name__,
373
self.transport.abspath(self.filename))
374
375
def _check_should_delta(self, first_parents):
1056
376
"""Iterate back through the parent listing, looking for a fulltext.
1057
377
1058
378
This is used when we want to decide whether to add a delta or a new
1065
385
"""
1066
386
delta_size = 0
1067
387
fulltext_size = None
388
delta_parents = first_parents
1068
389
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:
390
parent = delta_parents[0]
391
method = self._index.get_method(parent)
392
pos, size = self._index.get_position(parent)
393
if method == 'fulltext':
1082
394
fulltext_size = size
1083
395
break
1084
396
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
397
delta_parents = self._index.get_parents(parent)
1088
398
else:
1089
399
# We couldn't find a fulltext, so we must create a new one
1090
400
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.
401
1093
402
return fulltext_size > delta_size
1094
403
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.
404
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
405
"""See VersionedFile._add_delta()."""
406
self._check_add(version_id, []) # should we check the lines ?
407
self._check_versions_present(parents)
408
present_parents = []
409
ghosts = []
410
parent_texts = {}
411
for parent in parents:
412
if not self.has_version(parent):
413
ghosts.append(parent)
414
else:
415
present_parents.append(parent)
416
417
if delta_parent is None:
418
# reconstitute as full text.
419
assert len(delta) == 1 or len(delta) == 0
420
if len(delta):
421
assert delta[0][0] == 0
422
assert delta[0][1] == 0, delta[0][1]
423
return super(KnitVersionedFile, self)._add_delta(version_id,
424
parents,
425
delta_parent,
426
sha1,
427
noeol,
428
delta)
429
430
digest = sha1
431
432
options = []
433
if noeol:
434
options.append('no-eol')
435
436
if delta_parent is not None:
437
# determine the current delta chain length.
438
# To speed the extract of texts the delta chain is limited
439
# to a fixed number of deltas. This should minimize both
440
# I/O and the time spend applying deltas.
441
# The window was changed to a maximum of 200 deltas, but also added
442
# was a check that the total compressed size of the deltas is
443
# smaller than the compressed size of the fulltext.
444
if not self._check_should_delta([delta_parent]):
445
# We don't want a delta here, just do a normal insertion.
446
return super(KnitVersionedFile, self)._add_delta(version_id,
447
parents,
448
delta_parent,
449
sha1,
450
noeol,
451
delta)
452
453
options.append('line-delta')
454
store_lines = self.factory.lower_line_delta(delta)
455
456
where, size = self._data.add_record(version_id, digest, store_lines)
457
self._index.add_version(version_id, options, where, size, parents)
458
459
def _add_raw_records(self, records, data):
460
"""Add all the records 'records' with data pre-joined in 'data'.
461
462
:param records: A list of tuples(version_id, options, parents, size).
463
:param data: The data for the records. When it is written, the records
464
are adjusted to have pos pointing into data by the sum of
465
the preceding records sizes.
466
"""
467
# write all the data
468
pos = self._data.add_raw_record(data)
469
offset = 0
470
index_entries = []
471
for (version_id, options, parents, size) in records:
472
index_entries.append((version_id, options, pos+offset,
473
size, parents))
474
if self._data._do_cache:
475
self._data._cache[version_id] = data[offset:offset+size]
476
offset += size
477
self._index.add_versions(index_entries)
478
479
def enable_cache(self):
480
"""Start caching data for this knit"""
481
self._data.enable_cache()
482
483
def clear_cache(self):
484
"""Clear the data cache only."""
485
self._data.clear_cache()
486
487
def copy_to(self, name, transport):
488
"""See VersionedFile.copy_to()."""
489
# copy the current index to a temp index to avoid racing with local
490
# writes
491
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
492
self.transport.get(self._index._filename))
493
# copy the data file
494
f = self._data._open_file()
495
try:
496
transport.put_file(name + DATA_SUFFIX, f)
497
finally:
498
f.close()
499
# move the copied index into place
500
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
501
502
def create_empty(self, name, transport, mode=None):
503
return KnitVersionedFile(name, transport, factory=self.factory,
504
delta=self.delta, create=True)
505
506
def _fix_parents(self, version_id, new_parents):
507
"""Fix the parents list for version.
508
509
This is done by appending a new version to the index
510
with identical data except for the parents list.
511
the parents list must be a superset of the current
512
list.
513
"""
514
current_values = self._index._cache[version_id]
515
assert set(current_values[4]).difference(set(new_parents)) == set()
516
self._index.add_version(version_id,
517
current_values[1],
518
current_values[2],
519
current_values[3],
520
new_parents)
521
522
def get_delta(self, version_id):
523
"""Get a delta for constructing version from some other version."""
524
version_id = osutils.safe_revision_id(version_id)
525
self.check_not_reserved_id(version_id)
526
if not self.has_version(version_id):
527
raise RevisionNotPresent(version_id, self.filename)
528
529
parents = self.get_parents(version_id)
530
if len(parents):
531
parent = parents[0]
532
else:
533
parent = None
534
data_pos, data_size = self._index.get_position(version_id)
535
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
536
noeol = 'no-eol' in self._index.get_options(version_id)
537
if 'fulltext' == self._index.get_method(version_id):
538
new_content = self.factory.parse_fulltext(data, version_id)
539
if parent is not None:
540
reference_content = self._get_content(parent)
541
old_texts = reference_content.text()
542
else:
543
old_texts = []
544
new_texts = new_content.text()
545
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
546
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
547
else:
548
delta = self.factory.parse_line_delta(data, version_id)
549
return parent, sha1, noeol, delta
550
551
def get_graph_with_ghosts(self):
552
"""See VersionedFile.get_graph_with_ghosts()."""
553
graph_items = self._index.get_graph()
554
return dict(graph_items)
555
556
def get_sha1(self, version_id):
557
"""See VersionedFile.get_sha1()."""
558
version_id = osutils.safe_revision_id(version_id)
559
record_map = self._get_record_map([version_id])
560
method, content, digest, next = record_map[version_id]
561
return digest
562
563
@staticmethod
564
def get_suffixes():
565
"""See VersionedFile.get_suffixes()."""
566
return [DATA_SUFFIX, INDEX_SUFFIX]
567
568
def has_ghost(self, version_id):
569
"""True if there is a ghost reference in the file to version_id."""
570
version_id = osutils.safe_revision_id(version_id)
571
# maybe we have it
572
if self.has_version(version_id):
573
return False
574
# optimisable if needed by memoising the _ghosts set.
575
items = self._index.get_graph()
576
for node, parents in items:
577
for parent in parents:
578
if parent not in self._index._cache:
579
if parent == version_id:
580
return True
581
return False
582
583
def versions(self):
584
"""See VersionedFile.versions."""
585
return self._index.get_versions()
586
587
def has_version(self, version_id):
588
"""See VersionedFile.has_version."""
589
version_id = osutils.safe_revision_id(version_id)
590
return self._index.has_version(version_id)
591
592
__contains__ = has_version
593
594
def _merge_annotations(self, content, parents, parent_texts={},
595
delta=None, annotated=None):
596
"""Merge annotations for content. This is done by comparing
597
the annotations based on changed to the text.
598
"""
599
if annotated:
600
delta_seq = None
601
for parent_id in parents:
602
merge_content = self._get_content(parent_id, parent_texts)
603
seq = patiencediff.PatienceSequenceMatcher(
604
None, merge_content.text(), content.text())
605
if delta_seq is None:
606
# setup a delta seq to reuse.
607
delta_seq = seq
608
for i, j, n in seq.get_matching_blocks():
609
if n == 0:
610
continue
611
# this appears to copy (origin, text) pairs across to the new
612
# content for any line that matches the last-checked parent.
613
# FIXME: save the sequence control data for delta compression
614
# against the most relevant parent rather than rediffing.
615
content._lines[j:j+n] = merge_content._lines[i:i+n]
616
if delta:
617
if not annotated:
618
reference_content = self._get_content(parents[0], parent_texts)
619
new_texts = content.text()
620
old_texts = reference_content.text()
621
delta_seq = patiencediff.PatienceSequenceMatcher(
622
None, old_texts, new_texts)
623
return self._make_line_delta(delta_seq, content)
624
625
def _make_line_delta(self, delta_seq, new_content):
626
"""Generate a line delta from delta_seq and new_content."""
627
diff_hunks = []
628
for op in delta_seq.get_opcodes():
629
if op[0] == 'equal':
630
continue
631
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
632
return diff_hunks
633
634
def _get_components_positions(self, version_ids):
635
"""Produce a map of position data for the components of versions.
1102
636
1103
637
This data is intended to be used for retrieving the knit records.
1104
638
1105
A dict of key to (record_details, index_memo, next, parents) is
639
A dict of version_id to (method, data_pos, data_size, next) is
1106
640
returned.
1107
641
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
642
data_pos is the position of the data in the knit.
643
data_size is the size of the data in the knit.
1110
644
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
645
"""
1116
646
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)
647
for version_id in version_ids:
648
cursor = version_id
649
650
while cursor is not None and cursor not in component_data:
651
method = self._index.get_method(cursor)
652
if method == 'fulltext':
653
next = None
654
else:
655
next = self.get_parents(cursor)[0]
656
data_pos, data_size = self._index.get_position(cursor)
657
component_data[cursor] = (method, data_pos, data_size, next)
658
cursor = next
1132
659
return component_data
1133
1134
def _get_content(self, key, parent_texts={}):
660
661
def _get_content(self, version_id, parent_texts={}):
1135
662
"""Returns a content object that makes up the specified
1136
663
version."""
1137
cached_version = parent_texts.get(key, None)
664
if not self.has_version(version_id):
665
raise RevisionNotPresent(version_id, self.filename)
666
667
cached_version = parent_texts.get(version_id, None)
1138
668
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
669
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):
670
671
text_map, contents_map = self._get_content_maps([version_id])
672
return contents_map[version_id]
673
674
def _check_versions_present(self, version_ids):
675
"""Check that all specified versions are present."""
676
self._index.check_versions_present(version_ids)
677
678
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
679
"""See VersionedFile.add_lines_with_ghosts()."""
680
self._check_add(version_id, lines)
681
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
682
683
def _add_lines(self, version_id, parents, lines, parent_texts):
684
"""See VersionedFile.add_lines."""
685
self._check_add(version_id, lines)
686
self._check_versions_present(parents)
687
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
688
689
def _check_add(self, version_id, lines):
690
"""check that version_id and lines are safe to add."""
691
assert self.writable, "knit is not opened for write"
692
### FIXME escape. RBC 20060228
693
if contains_whitespace(version_id):
694
raise InvalidRevisionId(version_id, self.filename)
695
self.check_not_reserved_id(version_id)
696
if self.has_version(version_id):
697
raise RevisionAlreadyPresent(version_id, self.filename)
698
self._check_lines_not_unicode(lines)
699
self._check_lines_are_lines(lines)
700
701
def _add(self, version_id, lines, parents, delta, parent_texts):
702
"""Add a set of lines on top of version specified by parents.
703
704
If delta is true, compress the text as a line-delta against
705
the first parent.
706
707
Any versions not present will be converted into ghosts.
708
"""
709
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
710
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
711
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
712
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
713
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
714
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
715
# +1383 0 8.0370 8.0370 +<len>
716
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
717
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
718
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
719
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
720
721
present_parents = []
722
ghosts = []
723
if parent_texts is None:
724
parent_texts = {}
725
for parent in parents:
726
if not self.has_version(parent):
727
ghosts.append(parent)
728
else:
729
present_parents.append(parent)
730
731
if delta and not len(present_parents):
732
delta = False
733
734
digest = sha_strings(lines)
735
options = []
736
if lines:
737
if lines[-1][-1] != '\n':
738
options.append('no-eol')
739
lines[-1] = lines[-1] + '\n'
740
741
if len(present_parents) and delta:
742
# To speed the extract of texts the delta chain is limited
743
# to a fixed number of deltas. This should minimize both
744
# I/O and the time spend applying deltas.
745
delta = self._check_should_delta(present_parents)
746
747
assert isinstance(version_id, str)
748
lines = self.factory.make(lines, version_id)
749
if delta or (self.factory.annotated and len(present_parents) > 0):
750
# Merge annotations from parent texts if so is needed.
751
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
752
delta, self.factory.annotated)
753
754
if delta:
755
options.append('line-delta')
756
store_lines = self.factory.lower_line_delta(delta_hunks)
757
else:
758
options.append('fulltext')
759
store_lines = self.factory.lower_fulltext(lines)
760
761
where, size = self._data.add_record(version_id, digest, store_lines)
762
self._index.add_version(version_id, options, where, size, parents)
763
return lines
764
765
def check(self, progress_bar=None):
766
"""See VersionedFile.check()."""
767
768
def _clone_text(self, new_version_id, old_version_id, parents):
769
"""See VersionedFile.clone_text()."""
770
# FIXME RBC 20060228 make fast by only inserting an index with null
771
# delta.
772
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
773
774
def get_lines(self, version_id):
775
"""See VersionedFile.get_lines()."""
776
return self.get_line_list([version_id])[0]
777
778
def _get_record_map(self, version_ids):
1179
779
"""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
780
781
The keys are version_ids, the values are tuples of (method, content,
782
digest, next).
783
method is the way the content should be applied.
784
content is a KnitContent object.
785
digest is the SHA1 digest of this version id after all steps are done
786
next is the build-parent of the version, i.e. the leftmost ancestor.
787
If the method is fulltext, next will be None.
788
"""
789
position_map = self._get_components_positions(version_ids)
790
# c = component_id, m = method, p = position, s = size, n = next
791
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
792
record_map = {}
793
for component_id, content, digest in \
794
self._data.read_records_iter(records):
795
method, position, size, next = position_map[component_id]
796
record_map[component_id] = method, content, digest, next
797
798
return record_map
799
800
def get_text(self, version_id):
801
"""See VersionedFile.get_text"""
802
return self.get_texts([version_id])[0]
803
804
def get_texts(self, version_ids):
805
return [''.join(l) for l in self.get_line_list(version_ids)]
806
807
def get_line_list(self, version_ids):
808
"""Return the texts of listed versions as a list of strings."""
809
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
810
for version_id in version_ids:
811
self.check_not_reserved_id(version_id)
812
text_map, content_map = self._get_content_maps(version_ids)
813
return [text_map[v] for v in version_ids]
814
815
def _get_content_maps(self, version_ids):
816
"""Produce maps of text and KnitContents
817
1982
818
:return: (text_map, content_map) where text_map contains the texts for
1983
the requested versions and content_map contains the KnitContents.
819
the requested versions and content_map contains the KnitContents.
820
Both dicts take version_ids as their keys.
1984
821
"""
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
822
for version_id in version_ids:
823
if not self.has_version(version_id):
824
raise RevisionNotPresent(version_id, self.filename)
825
record_map = self._get_record_map(version_ids)
826
827
text_map = {}
828
content_map = {}
829
final_content = {}
830
for version_id in version_ids:
2039
831
components = []
2040
cursor = key
832
cursor = version_id
2041
833
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))
2047
cursor = next
2048
if cursor in self._contents_map:
2049
# no need to plan further back
2050
components.append((cursor, None, None, None))
834
method, data, digest, next = record_map[cursor]
835
components.append((cursor, method, data, digest))
836
if cursor in content_map:
2051
837
break
838
cursor = next
2052
839
2053
840
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]
841
for component_id, method, data, digest in reversed(components):
842
if component_id in content_map:
843
content = content_map[component_id]
2058
844
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
845
if method == 'fulltext':
846
assert content is None
847
content = self.factory.parse_fulltext(data, version_id)
848
elif method == 'line-delta':
849
delta = self.factory.parse_line_delta(data, version_id)
850
content = content.copy()
851
content._lines = self._apply_delta(content._lines,
852
delta)
853
content_map[component_id] = content
854
855
if 'no-eol' in self._index.get_options(version_id):
856
content = content.copy()
857
line = content._lines[-1][1].rstrip('\n')
858
content._lines[-1] = (content._lines[-1][0], line)
859
final_content[version_id] = content
2064
860
2065
861
# digest here is the digest from the last applied component.
2066
862
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
863
if sha_strings(text) != digest:
864
raise KnitCorrupt(self.filename,
865
'sha-1 does not match %s' % version_id)
866
867
text_map[version_id] = text
868
return text_map, final_content
869
870
def iter_lines_added_or_present_in_versions(self, version_ids=None,
871
pb=None):
872
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
873
if version_ids is None:
874
version_ids = self.versions()
875
else:
876
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
877
if pb is None:
878
pb = progress.DummyProgress()
879
# we don't care about inclusions, the caller cares.
880
# but we need to setup a list of records to visit.
881
# we need version_id, position, length
882
version_id_records = []
883
requested_versions = set(version_ids)
884
# filter for available versions
885
for version_id in requested_versions:
886
if not self.has_version(version_id):
887
raise RevisionNotPresent(version_id, self.filename)
888
# get a in-component-order queue:
889
for version_id in self.versions():
890
if version_id in requested_versions:
891
data_pos, length = self._index.get_position(version_id)
892
version_id_records.append((version_id, data_pos, length))
893
894
total = len(version_id_records)
895
for version_idx, (version_id, data, sha_value) in \
896
enumerate(self._data.read_records_iter(version_id_records)):
897
pb.update('Walking content.', version_idx, total)
898
method = self._index.get_method(version_id)
899
900
assert method in ('fulltext', 'line-delta')
901
if method == 'fulltext':
902
line_iterator = self.factory.get_fulltext_content(data)
903
else:
904
line_iterator = self.factory.get_linedelta_content(data)
905
for line in line_iterator:
906
yield line
907
908
pb.update('Walking content.', total, total)
909
910
def num_versions(self):
911
"""See VersionedFile.num_versions()."""
912
return self._index.num_versions()
913
914
__len__ = num_versions
915
916
def annotate_iter(self, version_id):
917
"""See VersionedFile.annotate_iter."""
918
version_id = osutils.safe_revision_id(version_id)
919
content = self._get_content(version_id)
920
for origin, text in content.annotate_iter():
921
yield origin, text
922
923
def get_parents(self, version_id):
924
"""See VersionedFile.get_parents."""
925
# perf notes:
926
# optimism counts!
927
# 52554 calls in 1264 872 internal down from 3674
928
version_id = osutils.safe_revision_id(version_id)
929
try:
930
return self._index.get_parents(version_id)
931
except KeyError:
932
raise RevisionNotPresent(version_id, self.filename)
933
934
def get_parents_with_ghosts(self, version_id):
935
"""See VersionedFile.get_parents."""
936
version_id = osutils.safe_revision_id(version_id)
937
try:
938
return self._index.get_parents_with_ghosts(version_id)
939
except KeyError:
940
raise RevisionNotPresent(version_id, self.filename)
941
942
def get_ancestry(self, versions):
943
"""See VersionedFile.get_ancestry."""
944
if isinstance(versions, basestring):
945
versions = [versions]
946
if not versions:
947
return []
948
versions = [osutils.safe_revision_id(v) for v in versions]
949
return self._index.get_ancestry(versions)
950
951
def get_ancestry_with_ghosts(self, versions):
952
"""See VersionedFile.get_ancestry_with_ghosts."""
953
if isinstance(versions, basestring):
954
versions = [versions]
955
if not versions:
956
return []
957
versions = [osutils.safe_revision_id(v) for v in versions]
958
return self._index.get_ancestry_with_ghosts(versions)
959
960
#@deprecated_method(zero_eight)
961
def walk(self, version_ids):
962
"""See VersionedFile.walk."""
963
# We take the short path here, and extract all relevant texts
964
# and put them in a weave and let that do all the work. Far
965
# from optimal, but is much simpler.
966
# FIXME RB 20060228 this really is inefficient!
967
from bzrlib.weave import Weave
968
969
w = Weave(self.filename)
970
ancestry = self.get_ancestry(version_ids)
971
sorted_graph = topo_sort(self._index.get_graph())
972
version_list = [vid for vid in sorted_graph if vid in ancestry]
973
974
for version_id in version_list:
975
lines = self.get_lines(version_id)
976
w.add_lines(version_id, self.get_parents(version_id), lines)
977
978
for lineno, insert_id, dset, line in w.walk(version_ids):
979
yield lineno, insert_id, dset, line
980
981
def plan_merge(self, ver_a, ver_b):
982
"""See VersionedFile.plan_merge."""
983
ver_a = osutils.safe_revision_id(ver_a)
984
ver_b = osutils.safe_revision_id(ver_b)
985
ancestors_b = set(self.get_ancestry(ver_b))
986
def status_a(revision, text):
987
if revision in ancestors_b:
988
return 'killed-b', text
989
else:
990
return 'new-a', text
991
992
ancestors_a = set(self.get_ancestry(ver_a))
993
def status_b(revision, text):
994
if revision in ancestors_a:
995
return 'killed-a', text
996
else:
997
return 'new-b', text
998
999
annotated_a = self.annotate(ver_a)
1000
annotated_b = self.annotate(ver_b)
1001
plain_a = [t for (a, t) in annotated_a]
1002
plain_b = [t for (a, t) in annotated_b]
1003
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
1004
a_cur = 0
1005
b_cur = 0
1006
for ai, bi, l in blocks:
1007
# process all mismatched sections
1008
# (last mismatched section is handled because blocks always
1009
# includes a 0-length last block)
1010
for revision, text in annotated_a[a_cur:ai]:
1011
yield status_a(revision, text)
1012
for revision, text in annotated_b[b_cur:bi]:
1013
yield status_b(revision, text)
1014
1015
# and now the matched section
1016
a_cur = ai + l
1017
b_cur = bi + l
1018
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
1019
assert text_a == text_b
1020
yield "unchanged", text_a
1021
1022
1023
class _KnitComponentFile(object):
1024
"""One of the files used to implement a knit database"""
1025
1026
def __init__(self, transport, filename, mode, file_mode=None,
1027
create_parent_dir=False, dir_mode=None):
1028
self._transport = transport
1029
self._filename = filename
1030
self._mode = mode
1031
self._file_mode = file_mode
1032
self._dir_mode = dir_mode
1033
self._create_parent_dir = create_parent_dir
1034
self._need_to_create = False
1035
1036
def _full_path(self):
1037
"""Return the full path to this file."""
1038
return self._transport.base + self._filename
1039
1040
def check_header(self, fp):
1041
line = fp.readline()
1042
if line == '':
1043
# An empty file can actually be treated as though the file doesn't
1044
# exist yet.
1045
raise errors.NoSuchFile(self._full_path())
1046
if line != self.HEADER:
1047
raise KnitHeaderError(badline=line,
1048
filename=self._transport.abspath(self._filename))
1049
1050
def commit(self):
1051
"""Commit is a nop."""
1052
1053
def __repr__(self):
1054
return '%s(%s)' % (self.__class__.__name__, self._filename)
1055
1056
1057
class _KnitIndex(_KnitComponentFile):
1058
"""Manages knit index file.
1059
1060
The index is already kept in memory and read on startup, to enable
2272
1061
fast lookups of revision information. The cursor of the index
2273
1062
file is always pointing to the end, making it easy to append
2274
1063
entries.
2284
1073
2285
1074
Duplicate entries may be written to the index for a single version id
2286
1075
if this is done then the latter one completely replaces the former:
2287
this allows updates to correct version and parent information.
1076
this allows updates to correct version and parent information.
2288
1077
Note that the two entries may share the delta, and that successive
2289
1078
annotations and references MUST point to the first entry.
2290
1079
2291
1080
The index file on disc contains a header, followed by one line per knit
2292
1081
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
1082
The first occurrence gets assigned a sequence number starting from 0.
1083
2295
1084
The format of a single line is
2296
1085
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
2297
1086
REVISION_ID is a utf8-encoded revision id
2298
FLAGS is a comma separated list of flags about the record. Values include
1087
FLAGS is a comma separated list of flags about the record. Values include
2299
1088
no-eol, line-delta, fulltext.
2300
1089
BYTE_OFFSET is the ascii representation of the byte offset in the data file
2301
1090
that the the compressed data starts at.
2305
1094
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
2306
1095
revision id already in the knit that is a parent of REVISION_ID.
2307
1096
The ' :' marker is the end of record marker.
2308
1097
2309
1098
partial writes:
2310
1099
when a write is interrupted to the index file, it will result in a line
2311
1100
that does not end in ' :'. If the ' :' is not present at the end of a line,
2316
1105
to ensure that records always start on new lines even if the last write was
2317
1106
interrupted. As a result its normal for the last line in the index to be
2318
1107
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
1108
"""
2326
1109
2327
1110
HEADER = "# bzr knit index 8\n"
2328
1111
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):
1112
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1113
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1114
1115
def _cache_version(self, version_id, options, pos, size, parents):
2405
1116
"""Cache a version record in the history array and index cache.
2406
1117
2407
1118
This is inlined into _load_data for performance. KEEP IN SYNC.
2408
1119
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2409
1120
indexes).
2410
1121
"""
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
1122
# only want the _history index to reference the 1st index entry
2421
1123
# for version_id
2422
if version_id not in cache:
2423
index = len(history)
2424
history.append(version_id)
1124
if version_id not in self._cache:
1125
index = len(self._history)
1126
self._history.append(version_id)
2425
1127
else:
2426
index = cache[version_id][5]
2427
cache[version_id] = (version_id,
1128
index = self._cache[version_id][5]
1129
self._cache[version_id] = (version_id,
2428
1130
options,
2429
1131
pos,
2430
1132
size,
2431
1133
parents,
2432
1134
index)
2433
1135
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])
1136
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1137
create_parent_dir=False, delay_create=False, dir_mode=None):
1138
_KnitComponentFile.__init__(self, transport, filename, mode,
1139
file_mode=file_mode,
1140
create_parent_dir=create_parent_dir,
1141
dir_mode=dir_mode)
1142
self._cache = {}
1143
# position in _history is the 'official' index for a revision
1144
# but the values may have come from a newer entry.
1145
# so - wc -l of a knit index is != the number of unique names
1146
# in the knit.
1147
self._history = []
2511
1148
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.
2522
"""
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]
2531
try:
2532
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2533
except KeyError:
2534
pass
2535
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 ''
1149
fp = self._transport.get(self._filename)
1150
try:
1151
# _load_data may raise NoSuchFile if the target knit is
1152
# completely empty.
1153
self._load_data(fp)
1154
finally:
1155
fp.close()
1156
except NoSuchFile:
1157
if mode != 'w' or not create:
1158
raise
1159
elif delay_create:
1160
self._need_to_create = True
1161
else:
1162
self._transport.put_bytes_non_atomic(
1163
self._filename, self.HEADER, mode=self._file_mode)
1164
1165
def _load_data(self, fp):
1166
cache = self._cache
1167
history = self._history
1168
1169
self.check_header(fp)
1170
# readlines reads the whole file at once:
1171
# bad for transports like http, good for local disk
1172
# we save 60 ms doing this one change (
1173
# from calling readline each time to calling
1174
# readlines once.
1175
# probably what we want for nice behaviour on
1176
# http is a incremental readlines that yields, or
1177
# a check for local vs non local indexes,
1178
history_top = len(history) - 1
1179
for line in fp.readlines():
1180
rec = line.split()
1181
if len(rec) < 5 or rec[-1] != ':':
1182
# corrupt line.
1183
# FIXME: in the future we should determine if its a
1184
# short write - and ignore it
1185
# or a different failure, and raise. RBC 20060407
1186
continue
1187
1188
parents = []
1189
for value in rec[4:-1]:
1190
if value[0] == '.':
1191
# uncompressed reference
1192
parent_id = value[1:]
1193
else:
1194
parent_id = history[int(value)]
1195
parents.append(parent_id)
1196
1197
version_id, options, pos, size = rec[:4]
1198
version_id = version_id
1199
1200
# See self._cache_version
1201
# only want the _history index to reference the 1st
1202
# index entry for version_id
1203
if version_id not in cache:
1204
history_top += 1
1205
index = history_top
1206
history.append(version_id)
1207
else:
1208
index = cache[version_id][5]
1209
cache[version_id] = (version_id,
1210
options.split(','),
1211
int(pos),
1212
int(size),
1213
parents,
1214
index)
1215
# end self._cache_version
1216
1217
def get_graph(self):
1218
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1219
1220
def get_ancestry(self, versions):
1221
"""See VersionedFile.get_ancestry."""
1222
# get a graph of all the mentioned versions:
1223
graph = {}
1224
pending = set(versions)
1225
cache = self._cache
1226
while pending:
1227
version = pending.pop()
1228
# trim ghosts
1229
try:
1230
parents = [p for p in cache[version][4] if p in cache]
1231
except KeyError:
1232
raise RevisionNotPresent(version, self._filename)
1233
# if not completed and not a ghost
1234
pending.update([p for p in parents if p not in graph])
1235
graph[version] = parents
1236
return topo_sort(graph.items())
1237
1238
def get_ancestry_with_ghosts(self, versions):
1239
"""See VersionedFile.get_ancestry_with_ghosts."""
1240
# get a graph of all the mentioned versions:
1241
self.check_versions_present(versions)
1242
cache = self._cache
1243
graph = {}
1244
pending = set(versions)
1245
while pending:
1246
version = pending.pop()
1247
try:
1248
parents = cache[version][4]
1249
except KeyError:
1250
# ghost, fake it
1251
graph[version] = []
1252
else:
1253
# if not completed
1254
pending.update([p for p in parents if p not in graph])
1255
graph[version] = parents
1256
return topo_sort(graph.items())
1257
1258
def num_versions(self):
1259
return len(self._history)
1260
1261
__len__ = num_versions
1262
1263
def get_versions(self):
1264
return self._history
1265
1266
def idx_to_name(self, idx):
1267
return self._history[idx]
1268
1269
def lookup(self, version_id):
1270
assert version_id in self._cache
1271
return self._cache[version_id][5]
1272
1273
def _version_list_to_index(self, versions):
2638
1274
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:
1275
cache = self._cache
1276
for version in versions:
1277
if version in cache:
2646
1278
# -- inlined lookup() --
2647
result_list.append(str(cache[key[-1]][5]))
1279
result_list.append(str(cache[version][5]))
2648
1280
# -- end lookup () --
2649
1281
else:
2650
result_list.append('.' + key[-1])
1282
result_list.append('.' + version)
2651
1283
return ' '.join(result_list)
2652
1284
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'
2661
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, track_external_parent_refs=False):
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
:param track_external_parent_refs: If True, record all external parent
2710
references parents from added records. These can be retrieved
2711
later by calling get_missing_parents().
2712
"""
2713
self._add_callback = add_callback
2714
self._graph_index = graph_index
2715
self._deltas = deltas
2716
self._parents = parents
2717
if deltas and not parents:
2718
# XXX: TODO: Delta tree and parent graph should be conceptually
2719
# separate.
2720
raise KnitCorrupt(self, "Cannot do delta compression without "
2721
"parent tracking.")
2722
self.has_graph = parents
2723
self._is_locked = is_locked
2724
self._missing_compression_parents = set()
2725
if track_external_parent_refs:
2726
self._external_parent_refs = set()
2727
else:
2728
self._external_parent_refs = None
2729
2730
def __repr__(self):
2731
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2732
2733
def add_records(self, records, random_id=False,
2734
missing_compression_parents=False):
2735
"""Add multiple records to the index.
2736
2737
This function does not insert data into the Immutable GraphIndex
2738
backing the KnitGraphIndex, instead it prepares data for insertion by
2739
the caller and checks that it is safe to insert then calls
2740
self._add_callback with the prepared GraphIndex nodes.
2741
2742
:param records: a list of tuples:
2743
(key, options, access_memo, parents).
2744
:param random_id: If True the ids being added were randomly generated
2745
and no check for existence will be performed.
2746
:param missing_compression_parents: If True the records being added are
2747
only compressed against texts already in the index (or inside
2748
records). If False the records all refer to unavailable texts (or
2749
texts inside records) as compression parents.
2750
"""
2751
if not self._add_callback:
2752
raise errors.ReadOnlyError(self)
2753
# we hope there are no repositories with inconsistent parentage
2754
# anymore.
2755
2756
keys = {}
2757
compression_parents = set()
2758
parent_refs = self._external_parent_refs
2759
for (key, options, access_memo, parents) in records:
2760
if self._parents:
2761
parents = tuple(parents)
2762
if parent_refs is not None:
2763
parent_refs.update(parents)
2764
parent_refs.discard(key)
2765
index, pos, size = access_memo
2766
if 'no-eol' in options:
2767
value = 'N'
2768
else:
2769
value = ' '
2770
value += "%d %d" % (pos, size)
2771
if not self._deltas:
2772
if 'line-delta' in options:
2773
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2774
if self._parents:
2775
if self._deltas:
2776
if 'line-delta' in options:
2777
node_refs = (parents, (parents[0],))
2778
if missing_compression_parents:
2779
compression_parents.add(parents[0])
2780
else:
2781
node_refs = (parents, ())
1285
def add_version(self, version_id, options, pos, size, parents):
1286
"""Add a version record to the index."""
1287
self.add_versions(((version_id, options, pos, size, parents),))
1288
1289
def add_versions(self, versions):
1290
"""Add multiple versions to the index.
1291
1292
:param versions: a list of tuples:
1293
(version_id, options, pos, size, parents).
1294
"""
1295
lines = []
1296
orig_history = self._history[:]
1297
orig_cache = self._cache.copy()
1298
1299
try:
1300
for version_id, options, pos, size, parents in versions:
1301
line = "\n%s %s %s %s %s :" % (version_id,
1302
','.join(options),
1303
pos,
1304
size,
1305
self._version_list_to_index(parents))
1306
assert isinstance(line, str), \
1307
'content must be utf-8 encoded: %r' % (line,)
1308
lines.append(line)
1309
self._cache_version(version_id, options, pos, size, parents)
1310
if not self._need_to_create:
1311
self._transport.append_bytes(self._filename, ''.join(lines))
1312
else:
1313
sio = StringIO()
1314
sio.write(self.HEADER)
1315
sio.writelines(lines)
1316
sio.seek(0)
1317
self._transport.put_file_non_atomic(self._filename, sio,
1318
create_parent_dir=self._create_parent_dir,
1319
mode=self._file_mode,
1320
dir_mode=self._dir_mode)
1321
self._need_to_create = False
1322
except:
1323
# If any problems happen, restore the original values and re-raise
1324
self._history = orig_history
1325
self._cache = orig_cache
1326
raise
1327
1328
def has_version(self, version_id):
1329
"""True if the version is in the index."""
1330
return version_id in self._cache
1331
1332
def get_position(self, version_id):
1333
"""Return data position and size of specified version."""
1334
entry = self._cache[version_id]
1335
return entry[2], entry[3]
1336
1337
def get_method(self, version_id):
1338
"""Return compression method of specified version."""
1339
options = self._cache[version_id][1]
1340
if 'fulltext' in options:
1341
return 'fulltext'
1342
else:
1343
if 'line-delta' not in options:
1344
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1345
return 'line-delta'
1346
1347
def get_options(self, version_id):
1348
return self._cache[version_id][1]
1349
1350
def get_parents(self, version_id):
1351
"""Return parents of specified version ignoring ghosts."""
1352
return [parent for parent in self._cache[version_id][4]
1353
if parent in self._cache]
1354
1355
def get_parents_with_ghosts(self, version_id):
1356
"""Return parents of specified version with ghosts."""
1357
return self._cache[version_id][4]
1358
1359
def check_versions_present(self, version_ids):
1360
"""Check that all specified versions are present."""
1361
cache = self._cache
1362
for version_id in version_ids:
1363
if version_id not in cache:
1364
raise RevisionNotPresent(version_id, self._filename)
1365
1366
1367
class _KnitData(_KnitComponentFile):
1368
"""Contents of the knit data file"""
1369
1370
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1371
create_parent_dir=False, delay_create=False,
1372
dir_mode=None):
1373
_KnitComponentFile.__init__(self, transport, filename, mode,
1374
file_mode=file_mode,
1375
create_parent_dir=create_parent_dir,
1376
dir_mode=dir_mode)
1377
self._checked = False
1378
# TODO: jam 20060713 conceptually, this could spill to disk
1379
# if the cached size gets larger than a certain amount
1380
# but it complicates the model a bit, so for now just use
1381
# a simple dictionary
1382
self._cache = {}
1383
self._do_cache = False
1384
if create:
1385
if delay_create:
1386
self._need_to_create = create
1387
else:
1388
self._transport.put_bytes_non_atomic(self._filename, '',
1389
mode=self._file_mode)
1390
1391
def enable_cache(self):
1392
"""Enable caching of reads."""
1393
self._do_cache = True
1394
1395
def clear_cache(self):
1396
"""Clear the record cache."""
1397
self._do_cache = False
1398
self._cache = {}
1399
1400
def _open_file(self):
1401
try:
1402
return self._transport.get(self._filename)
1403
except NoSuchFile:
1404
pass
1405
return None
1406
1407
def _record_to_data(self, version_id, digest, lines):
1408
"""Convert version_id, digest, lines into a raw data block.
1409
1410
:return: (len, a StringIO instance with the raw data ready to read.)
1411
"""
1412
sio = StringIO()
1413
data_file = GzipFile(None, mode='wb', fileobj=sio)
1414
1415
assert isinstance(version_id, str)
1416
data_file.writelines(chain(
1417
["version %s %d %s\n" % (version_id,
1418
len(lines),
1419
digest)],
1420
lines,
1421
["end %s\n" % version_id]))
1422
data_file.close()
1423
length= sio.tell()
1424
1425
sio.seek(0)
1426
return length, sio
1427
1428
def add_raw_record(self, raw_data):
1429
"""Append a prepared record to the data file.
1430
1431
:return: the offset in the data file raw_data was written.
1432
"""
1433
assert isinstance(raw_data, str), 'data must be plain bytes'
1434
if not self._need_to_create:
1435
return self._transport.append_bytes(self._filename, raw_data)
1436
else:
1437
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1438
create_parent_dir=self._create_parent_dir,
1439
mode=self._file_mode,
1440
dir_mode=self._dir_mode)
1441
self._need_to_create = False
1442
return 0
1443
1444
def add_record(self, version_id, digest, lines):
1445
"""Write new text record to disk. Returns the position in the
1446
file where it was written."""
1447
size, sio = self._record_to_data(version_id, digest, lines)
1448
# write to disk
1449
if not self._need_to_create:
1450
start_pos = self._transport.append_file(self._filename, sio)
1451
else:
1452
self._transport.put_file_non_atomic(self._filename, sio,
1453
create_parent_dir=self._create_parent_dir,
1454
mode=self._file_mode,
1455
dir_mode=self._dir_mode)
1456
self._need_to_create = False
1457
start_pos = 0
1458
if self._do_cache:
1459
self._cache[version_id] = sio.getvalue()
1460
return start_pos, size
1461
1462
def _parse_record_header(self, version_id, raw_data):
1463
"""Parse a record header for consistency.
1464
1465
:return: the header and the decompressor stream.
1466
as (stream, header_record)
1467
"""
1468
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1469
try:
1470
rec = self._check_header(version_id, df.readline())
1471
except Exception, e:
1472
raise KnitCorrupt(self._filename,
1473
"While reading {%s} got %s(%s)"
1474
% (version_id, e.__class__.__name__, str(e)))
1475
return df, rec
1476
1477
def _check_header(self, version_id, line):
1478
rec = line.split()
1479
if len(rec) != 4:
1480
raise KnitCorrupt(self._filename,
1481
'unexpected number of elements in record header')
1482
if rec[1] != version_id:
1483
raise KnitCorrupt(self._filename,
1484
'unexpected version, wanted %r, got %r'
1485
% (version_id, rec[1]))
1486
return rec
1487
1488
def _parse_record(self, version_id, data):
1489
# profiling notes:
1490
# 4168 calls in 2880 217 internal
1491
# 4168 calls to _parse_record_header in 2121
1492
# 4168 calls to readlines in 330
1493
df = GzipFile(mode='rb', fileobj=StringIO(data))
1494
1495
try:
1496
record_contents = df.readlines()
1497
except Exception, e:
1498
raise KnitCorrupt(self._filename,
1499
"While reading {%s} got %s(%s)"
1500
% (version_id, e.__class__.__name__, str(e)))
1501
header = record_contents.pop(0)
1502
rec = self._check_header(version_id, header)
1503
1504
last_line = record_contents.pop()
1505
if len(record_contents) != int(rec[2]):
1506
raise KnitCorrupt(self._filename,
1507
'incorrect number of lines %s != %s'
1508
' for version {%s}'
1509
% (len(record_contents), int(rec[2]),
1510
version_id))
1511
if last_line != 'end %s\n' % rec[1]:
1512
raise KnitCorrupt(self._filename,
1513
'unexpected version end line %r, wanted %r'
1514
% (last_line, version_id))
1515
df.close()
1516
return record_contents, rec[3]
1517
1518
def read_records_iter_raw(self, records):
1519
"""Read text records from data file and yield raw data.
1520
1521
This unpacks enough of the text record to validate the id is
1522
as expected but thats all.
1523
"""
1524
# setup an iterator of the external records:
1525
# uses readv so nice and fast we hope.
1526
if len(records):
1527
# grab the disk data needed.
1528
if self._cache:
1529
# Don't check _cache if it is empty
1530
needed_offsets = [(pos, size) for version_id, pos, size
1531
in records
1532
if version_id not in self._cache]
1533
else:
1534
needed_offsets = [(pos, size) for version_id, pos, size
1535
in records]
1536
1537
raw_records = self._transport.readv(self._filename, needed_offsets)
1538
1539
for version_id, pos, size in records:
1540
if version_id in self._cache:
1541
# This data has already been validated
1542
data = self._cache[version_id]
1543
else:
1544
pos, data = raw_records.next()
1545
if self._do_cache:
1546
self._cache[version_id] = data
1547
1548
# validate the header
1549
df, rec = self._parse_record_header(version_id, data)
1550
df.close()
1551
yield version_id, data
1552
1553
def read_records_iter(self, records):
1554
"""Read text records from data file and yield result.
1555
1556
The result will be returned in whatever is the fastest to read.
1557
Not by the order requested. Also, multiple requests for the same
1558
record will only yield 1 response.
1559
:param records: A list of (version_id, pos, len) entries
1560
:return: Yields (version_id, contents, digest) in the order
1561
read, not the order requested
1562
"""
1563
if not records:
1564
return
1565
1566
if self._cache:
1567
# Skip records we have alread seen
1568
yielded_records = set()
1569
needed_records = set()
1570
for record in records:
1571
if record[0] in self._cache:
1572
if record[0] in yielded_records:
1573
continue
1574
yielded_records.add(record[0])
1575
data = self._cache[record[0]]
1576
content, digest = self._parse_record(record[0], data)
1577
yield (record[0], content, digest)
2782
1578
else:
2783
node_refs = (parents, )
2784
else:
2785
if parents:
2786
raise KnitCorrupt(self, "attempt to add node with parents "
2787
"in parentless index.")
2788
node_refs = ()
2789
keys[key] = (value, node_refs)
2790
# check for dups
2791
if not random_id:
2792
present_nodes = self._get_entries(keys)
2793
for (index, key, value, node_refs) in present_nodes:
2794
if (value[0] != keys[key][0][0] or
2795
node_refs[:1] != keys[key][1][:1]):
2796
raise KnitCorrupt(self, "inconsistent details in add_records"
2797
": %s %s" % ((value, node_refs), keys[key]))
2798
del keys[key]
2799
result = []
2800
if self._parents:
2801
for key, (value, node_refs) in keys.iteritems():
2802
result.append((key, value, node_refs))
2803
else:
2804
for key, (value, node_refs) in keys.iteritems():
2805
result.append((key, value))
2806
self._add_callback(result)
2807
if missing_compression_parents:
2808
# This may appear to be incorrect (it does not check for
2809
# compression parents that are in the existing graph index),
2810
# but such records won't have been buffered, so this is
2811
# actually correct: every entry when
2812
# missing_compression_parents==True either has a missing parent, or
2813
# a parent that is one of the keys in records.
2814
compression_parents.difference_update(keys)
2815
self._missing_compression_parents.update(compression_parents)
2816
# Adding records may have satisfied missing compression parents.
2817
self._missing_compression_parents.difference_update(keys)
2818
2819
def scan_unvalidated_index(self, graph_index):
2820
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2821
2822
This allows this _KnitGraphIndex to keep track of any missing
2823
compression parents we may want to have filled in to make those
2824
indices valid.
2825
2826
:param graph_index: A GraphIndex
2827
"""
2828
if self._deltas:
2829
new_missing = graph_index.external_references(ref_list_num=1)
2830
new_missing.difference_update(self.get_parent_map(new_missing))
2831
self._missing_compression_parents.update(new_missing)
2832
if self._external_parent_refs is not None:
2833
# Add parent refs from graph_index (and discard parent refs that
2834
# the graph_index has).
2835
for node in graph_index.iter_all_entries():
2836
self._external_parent_refs.update(node[3][0])
2837
self._external_parent_refs.discard(node[1])
2838
2839
def get_missing_compression_parents(self):
2840
"""Return the keys of missing compression parents.
2841
2842
Missing compression parents occur when a record stream was missing
2843
basis texts, or a index was scanned that had missing basis texts.
2844
"""
2845
return frozenset(self._missing_compression_parents)
2846
2847
def get_missing_parents(self):
2848
"""Return the keys of missing parents."""
2849
# We may have false positives, so filter those out.
2850
self._external_parent_refs.difference_update(
2851
self.get_parent_map(self._external_parent_refs))
2852
return frozenset(self._external_parent_refs)
2853
2854
def _check_read(self):
2855
"""raise if reads are not permitted."""
2856
if not self._is_locked():
2857
raise errors.ObjectNotLocked(self)
2858
2859
def _check_write_ok(self):
2860
"""Assert if writes are not permitted."""
2861
if not self._is_locked():
2862
raise errors.ObjectNotLocked(self)
2863
2864
def _compression_parent(self, an_entry):
2865
# return the key that an_entry is compressed against, or None
2866
# Grab the second parent list (as deltas implies parents currently)
2867
compression_parents = an_entry[3][1]
2868
if not compression_parents:
2869
return None
2870
if len(compression_parents) != 1:
2871
raise AssertionError(
2872
"Too many compression parents: %r" % compression_parents)
2873
return compression_parents[0]
2874
2875
def get_build_details(self, keys):
2876
"""Get the method, index_memo and compression parent for version_ids.
2877
2878
Ghosts are omitted from the result.
2879
2880
:param keys: An iterable of keys.
2881
:return: A dict of key:
2882
(index_memo, compression_parent, parents, record_details).
2883
index_memo
2884
opaque structure to pass to read_records to extract the raw
2885
data
2886
compression_parent
2887
Content that this record is built upon, may be None
2888
parents
2889
Logical parents of this node
2890
record_details
2891
extra information about the content which needs to be passed to
2892
Factory.parse_record
2893
"""
2894
self._check_read()
2895
result = {}
2896
entries = self._get_entries(keys, False)
2897
for entry in entries:
2898
key = entry[1]
2899
if not self._parents:
2900
parents = ()
2901
else:
2902
parents = entry[3][0]
2903
if not self._deltas:
2904
compression_parent_key = None
2905
else:
2906
compression_parent_key = self._compression_parent(entry)
2907
noeol = (entry[2][0] == 'N')
2908
if compression_parent_key:
2909
method = 'line-delta'
2910
else:
2911
method = 'fulltext'
2912
result[key] = (self._node_to_position(entry),
2913
compression_parent_key, parents,
2914
(method, noeol))
2915
return result
2916
2917
def _get_entries(self, keys, check_present=False):
2918
"""Get the entries for keys.
2919
2920
:param keys: An iterable of index key tuples.
2921
"""
2922
keys = set(keys)
2923
found_keys = set()
2924
if self._parents:
2925
for node in self._graph_index.iter_entries(keys):
2926
yield node
2927
found_keys.add(node[1])
2928
else:
2929
# adapt parentless index to the rest of the code.
2930
for node in self._graph_index.iter_entries(keys):
2931
yield node[0], node[1], node[2], ()
2932
found_keys.add(node[1])
2933
if check_present:
2934
missing_keys = keys.difference(found_keys)
2935
if missing_keys:
2936
raise RevisionNotPresent(missing_keys.pop(), self)
2937
2938
def get_method(self, key):
2939
"""Return compression method of specified key."""
2940
return self._get_method(self._get_node(key))
2941
2942
def _get_method(self, node):
2943
if not self._deltas:
2944
return 'fulltext'
2945
if self._compression_parent(node):
2946
return 'line-delta'
2947
else:
2948
return 'fulltext'
2949
2950
def _get_node(self, key):
2951
try:
2952
return list(self._get_entries([key]))[0]
2953
except IndexError:
2954
raise RevisionNotPresent(key, self)
2955
2956
def get_options(self, key):
2957
"""Return a list representing options.
2958
2959
e.g. ['foo', 'bar']
2960
"""
2961
node = self._get_node(key)
2962
options = [self._get_method(node)]
2963
if node[2][0] == 'N':
2964
options.append('no-eol')
2965
return options
2966
2967
def get_parent_map(self, keys):
2968
"""Get a map of the parents of keys.
2969
2970
:param keys: The keys to look up parents for.
2971
:return: A mapping from keys to parents. Absent keys are absent from
2972
the mapping.
2973
"""
2974
self._check_read()
2975
nodes = self._get_entries(keys)
2976
result = {}
2977
if self._parents:
2978
for node in nodes:
2979
result[node[1]] = node[3][0]
2980
else:
2981
for node in nodes:
2982
result[node[1]] = None
2983
return result
2984
2985
def get_position(self, key):
2986
"""Return details needed to access the version.
2987
2988
:return: a tuple (index, data position, size) to hand to the access
2989
logic to get the record.
2990
"""
2991
node = self._get_node(key)
2992
return self._node_to_position(node)
2993
2994
has_key = _mod_index._has_key_from_parent_map
2995
2996
def keys(self):
2997
"""Get all the keys in the collection.
2998
2999
The keys are not ordered.
3000
"""
3001
self._check_read()
3002
return [node[1] for node in self._graph_index.iter_all_entries()]
3003
3004
missing_keys = _mod_index._missing_keys_from_parent_map
3005
3006
def _node_to_position(self, node):
3007
"""Convert an index value to position details."""
3008
bits = node[2][1:].split(' ')
3009
return node[0], int(bits[0]), int(bits[1])
3010
3011
def _sort_keys_by_io(self, keys, positions):
3012
"""Figure out an optimal order to read the records for the given keys.
3013
3014
Sort keys, grouped by index and sorted by position.
3015
3016
:param keys: A list of keys whose records we want to read. This will be
3017
sorted 'in-place'.
3018
:param positions: A dict, such as the one returned by
3019
_get_components_positions()
3020
:return: None
3021
"""
3022
def get_index_memo(key):
3023
# index_memo is at offset [1]. It is made up of (GraphIndex,
3024
# position, size). GI is an object, which will be unique for each
3025
# pack file. This causes us to group by pack file, then sort by
3026
# position. Size doesn't matter, but it isn't worth breaking up the
3027
# tuple.
3028
return positions[key][1]
3029
return keys.sort(key=get_index_memo)
3030
3031
_get_total_build_size = _get_total_build_size
3032
3033
3034
class _KnitKeyAccess(object):
3035
"""Access to records in .knit files."""
3036
3037
def __init__(self, transport, mapper):
3038
"""Create a _KnitKeyAccess with transport and mapper.
3039
3040
:param transport: The transport the access object is rooted at.
3041
:param mapper: The mapper used to map keys to .knit files.
3042
"""
3043
self._transport = transport
3044
self._mapper = mapper
3045
3046
def add_raw_records(self, key_sizes, raw_data):
3047
"""Add raw knit bytes to a storage area.
3048
3049
The data is spooled to the container writer in one bytes-record per
3050
raw data item.
3051
3052
:param sizes: An iterable of tuples containing the key and size of each
3053
raw data segment.
3054
:param raw_data: A bytestring containing the data.
3055
:return: A list of memos to retrieve the record later. Each memo is an
3056
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3057
length), where the key is the record key.
3058
"""
3059
if type(raw_data) != str:
3060
raise AssertionError(
3061
'data must be plain bytes was %s' % type(raw_data))
3062
result = []
3063
offset = 0
3064
# TODO: This can be tuned for writing to sftp and other servers where
3065
# append() is relatively expensive by grouping the writes to each key
3066
# prefix.
3067
for key, size in key_sizes:
3068
path = self._mapper.map(key)
3069
try:
3070
base = self._transport.append_bytes(path + '.knit',
3071
raw_data[offset:offset+size])
3072
except errors.NoSuchFile:
3073
self._transport.mkdir(osutils.dirname(path))
3074
base = self._transport.append_bytes(path + '.knit',
3075
raw_data[offset:offset+size])
3076
# if base == 0:
3077
# chmod.
3078
offset += size
3079
result.append((key, base, size))
3080
return result
3081
3082
def flush(self):
3083
"""Flush pending writes on this access object.
3084
3085
For .knit files this is a no-op.
3086
"""
3087
pass
3088
3089
def get_raw_records(self, memos_for_retrieval):
3090
"""Get the raw bytes for a records.
3091
3092
:param memos_for_retrieval: An iterable containing the access memo for
3093
retrieving the bytes.
3094
:return: An iterator over the bytes of the records.
3095
"""
3096
# first pass, group into same-index request to minimise readv's issued.
3097
request_lists = []
3098
current_prefix = None
3099
for (key, offset, length) in memos_for_retrieval:
3100
if current_prefix == key[:-1]:
3101
current_list.append((offset, length))
3102
else:
3103
if current_prefix is not None:
3104
request_lists.append((current_prefix, current_list))
3105
current_prefix = key[:-1]
3106
current_list = [(offset, length)]
3107
# handle the last entry
3108
if current_prefix is not None:
3109
request_lists.append((current_prefix, current_list))
3110
for prefix, read_vector in request_lists:
3111
path = self._mapper.map(prefix) + '.knit'
3112
for pos, data in self._transport.readv(path, read_vector):
3113
yield data
3114
3115
3116
class _DirectPackAccess(object):
3117
"""Access to data in one or more packs with less translation."""
3118
3119
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
3120
"""Create a _DirectPackAccess object.
3121
3122
:param index_to_packs: A dict mapping index objects to the transport
3123
and file names for obtaining data.
3124
:param reload_func: A function to call if we determine that the pack
3125
files have moved and we need to reload our caches. See
3126
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
3127
"""
3128
self._container_writer = None
3129
self._write_index = None
3130
self._indices = index_to_packs
3131
self._reload_func = reload_func
3132
self._flush_func = flush_func
3133
3134
def add_raw_records(self, key_sizes, raw_data):
3135
"""Add raw knit bytes to a storage area.
3136
3137
The data is spooled to the container writer in one bytes-record per
3138
raw data item.
3139
3140
:param sizes: An iterable of tuples containing the key and size of each
3141
raw data segment.
3142
:param raw_data: A bytestring containing the data.
3143
:return: A list of memos to retrieve the record later. Each memo is an
3144
opaque index memo. For _DirectPackAccess the memo is (index, pos,
3145
length), where the index field is the write_index object supplied
3146
to the PackAccess object.
3147
"""
3148
if type(raw_data) != str:
3149
raise AssertionError(
3150
'data must be plain bytes was %s' % type(raw_data))
3151
result = []
3152
offset = 0
3153
for key, size in key_sizes:
3154
p_offset, p_length = self._container_writer.add_bytes_record(
3155
raw_data[offset:offset+size], [])
3156
offset += size
3157
result.append((self._write_index, p_offset, p_length))
3158
return result
3159
3160
def flush(self):
3161
"""Flush pending writes on this access object.
3162
3163
This will flush any buffered writes to a NewPack.
3164
"""
3165
if self._flush_func is not None:
3166
self._flush_func()
1579
needed_records.add(record)
1580
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1581
else:
1582
needed_records = sorted(set(records), key=operator.itemgetter(1))
1583
1584
if not needed_records:
1585
return
1586
1587
# The transport optimizes the fetching as well
1588
# (ie, reads continuous ranges.)
1589
readv_response = self._transport.readv(self._filename,
1590
[(pos, size) for version_id, pos, size in needed_records])
1591
1592
for (version_id, pos, size), (pos, data) in \
1593
izip(iter(needed_records), readv_response):
1594
content, digest = self._parse_record(version_id, data)
1595
if self._do_cache:
1596
self._cache[version_id] = data
1597
yield version_id, content, digest
1598
1599
def read_records(self, records):
1600
"""Read records into a dictionary."""
1601
components = {}
1602
for record_id, content, digest in \
1603
self.read_records_iter(records):
1604
components[record_id] = (content, digest)
1605
return components
1606
1607
1608
class InterKnit(InterVersionedFile):
1609
"""Optimised code paths for knit to knit operations."""
1610
1611
_matching_file_from_factory = KnitVersionedFile
1612
_matching_file_to_factory = KnitVersionedFile
1613
1614
@staticmethod
1615
def is_compatible(source, target):
1616
"""Be compatible with knits. """
1617
try:
1618
return (isinstance(source, KnitVersionedFile) and
1619
isinstance(target, KnitVersionedFile))
1620
except AttributeError:
1621
return False
1622
1623
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1624
"""See InterVersionedFile.join."""
1625
assert isinstance(self.source, KnitVersionedFile)
1626
assert isinstance(self.target, KnitVersionedFile)
1627
1628
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1629
1630
if not version_ids:
1631
return 0
1632
1633
pb = ui.ui_factory.nested_progress_bar()
1634
try:
1635
version_ids = list(version_ids)
1636
if None in version_ids:
1637
version_ids.remove(None)
1638
1639
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1640
this_versions = set(self.target._index.get_versions())
1641
needed_versions = self.source_ancestry - this_versions
1642
cross_check_versions = self.source_ancestry.intersection(this_versions)
1643
mismatched_versions = set()
1644
for version in cross_check_versions:
1645
# scan to include needed parents.
1646
n1 = set(self.target.get_parents_with_ghosts(version))
1647
n2 = set(self.source.get_parents_with_ghosts(version))
1648
if n1 != n2:
1649
# FIXME TEST this check for cycles being introduced works
1650
# the logic is we have a cycle if in our graph we are an
1651
# ancestor of any of the n2 revisions.
1652
for parent in n2:
1653
if parent in n1:
1654
# safe
1655
continue
1656
else:
1657
parent_ancestors = self.source.get_ancestry(parent)
1658
if version in parent_ancestors:
1659
raise errors.GraphCycleError([parent, version])
1660
# ensure this parent will be available later.
1661
new_parents = n2.difference(n1)
1662
needed_versions.update(new_parents.difference(this_versions))
1663
mismatched_versions.add(version)
1664
1665
if not needed_versions and not mismatched_versions:
1666
return 0
1667
full_list = topo_sort(self.source.get_graph())
1668
1669
version_list = [i for i in full_list if (not self.target.has_version(i)
1670
and i in needed_versions)]
1671
1672
# plan the join:
1673
copy_queue = []
1674
copy_queue_records = []
1675
copy_set = set()
1676
for version_id in version_list:
1677
options = self.source._index.get_options(version_id)
1678
parents = self.source._index.get_parents_with_ghosts(version_id)
1679
# check that its will be a consistent copy:
1680
for parent in parents:
1681
# if source has the parent, we must :
1682
# * already have it or
1683
# * have it scheduled already
1684
# otherwise we don't care
1685
assert (self.target.has_version(parent) or
1686
parent in copy_set or
1687
not self.source.has_version(parent))
1688
data_pos, data_size = self.source._index.get_position(version_id)
1689
copy_queue_records.append((version_id, data_pos, data_size))
1690
copy_queue.append((version_id, options, parents))
1691
copy_set.add(version_id)
1692
1693
# data suck the join:
1694
count = 0
1695
total = len(version_list)
1696
raw_datum = []
1697
raw_records = []
1698
for (version_id, raw_data), \
1699
(version_id2, options, parents) in \
1700
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1701
copy_queue):
1702
assert version_id == version_id2, 'logic error, inconsistent results'
1703
count = count + 1
1704
pb.update("Joining knit", count, total)
1705
raw_records.append((version_id, options, parents, len(raw_data)))
1706
raw_datum.append(raw_data)
1707
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1708
1709
for version in mismatched_versions:
1710
# FIXME RBC 20060309 is this needed?
1711
n1 = set(self.target.get_parents_with_ghosts(version))
1712
n2 = set(self.source.get_parents_with_ghosts(version))
1713
# write a combined record to our history preserving the current
1714
# parents as first in the list
1715
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1716
self.target.fix_parents(version, new_parents)
1717
return count
1718
finally:
1719
pb.finished()
1720
1721
1722
InterVersionedFile.register_optimiser(InterKnit)
1723
1724
1725
class WeaveToKnit(InterVersionedFile):
1726
"""Optimised code paths for weave to knit operations."""
1727
1728
_matching_file_from_factory = bzrlib.weave.WeaveFile
1729
_matching_file_to_factory = KnitVersionedFile
1730
1731
@staticmethod
1732
def is_compatible(source, target):
1733
"""Be compatible with weaves to knits."""
1734
try:
1735
return (isinstance(source, bzrlib.weave.Weave) and
1736
isinstance(target, KnitVersionedFile))
1737
except AttributeError:
1738
return False
1739
1740
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1741
"""See InterVersionedFile.join."""
1742
assert isinstance(self.source, bzrlib.weave.Weave)
1743
assert isinstance(self.target, KnitVersionedFile)
1744
1745
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1746
1747
if not version_ids:
1748
return 0
1749
1750
pb = ui.ui_factory.nested_progress_bar()
1751
try:
1752
version_ids = list(version_ids)
1753
1754
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1755
this_versions = set(self.target._index.get_versions())
1756
needed_versions = self.source_ancestry - this_versions
1757
cross_check_versions = self.source_ancestry.intersection(this_versions)
1758
mismatched_versions = set()
1759
for version in cross_check_versions:
1760
# scan to include needed parents.
1761
n1 = set(self.target.get_parents_with_ghosts(version))
1762
n2 = set(self.source.get_parents(version))
1763
# if all of n2's parents are in n1, then its fine.
1764
if n2.difference(n1):
1765
# FIXME TEST this check for cycles being introduced works
1766
# the logic is we have a cycle if in our graph we are an
1767
# ancestor of any of the n2 revisions.
1768
for parent in n2:
1769
if parent in n1:
1770
# safe
1771
continue
1772
else:
1773
parent_ancestors = self.source.get_ancestry(parent)
1774
if version in parent_ancestors:
1775
raise errors.GraphCycleError([parent, version])
1776
# ensure this parent will be available later.
1777
new_parents = n2.difference(n1)
1778
needed_versions.update(new_parents.difference(this_versions))
1779
mismatched_versions.add(version)
1780
1781
if not needed_versions and not mismatched_versions:
1782
return 0
1783
full_list = topo_sort(self.source.get_graph())
1784
1785
version_list = [i for i in full_list if (not self.target.has_version(i)
1786
and i in needed_versions)]
1787
1788
# do the join:
1789
count = 0
1790
total = len(version_list)
1791
for version_id in version_list:
1792
pb.update("Converting to knit", count, total)
1793
parents = self.source.get_parents(version_id)
1794
# check that its will be a consistent copy:
1795
for parent in parents:
1796
# if source has the parent, we must already have it
1797
assert (self.target.has_version(parent))
1798
self.target.add_lines(
1799
version_id, parents, self.source.get_lines(version_id))
1800
count = count + 1
1801
1802
for version in mismatched_versions:
1803
# FIXME RBC 20060309 is this needed?
1804
n1 = set(self.target.get_parents_with_ghosts(version))
1805
n2 = set(self.source.get_parents(version))
1806
# write a combined record to our history preserving the current
1807
# parents as first in the list
1808
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1809
self.target.fix_parents(version, new_parents)
1810
return count
1811
finally:
1812
pb.finished()
1813
1814
1815
InterVersionedFile.register_optimiser(WeaveToKnit)
1816
1817
1818
class KnitSequenceMatcher(difflib.SequenceMatcher):
1819
"""Knit tuned sequence matcher.
1820
1821
This is based on profiling of difflib which indicated some improvements
1822
for our usage pattern.
1823
"""
1824
1825
def find_longest_match(self, alo, ahi, blo, bhi):
1826
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1827
1828
If isjunk is not defined:
1829
1830
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1831
alo <= i <= i+k <= ahi
1832
blo <= j <= j+k <= bhi
1833
and for all (i',j',k') meeting those conditions,
1834
k >= k'
1835
i <= i'
1836
and if i == i', j <= j'
1837
1838
In other words, of all maximal matching blocks, return one that
1839
starts earliest in a, and of all those maximal matching blocks that
1840
start earliest in a, return the one that starts earliest in b.
1841
1842
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1843
>>> s.find_longest_match(0, 5, 0, 9)
1844
(0, 4, 5)
1845
1846
If isjunk is defined, first the longest matching block is
1847
determined as above, but with the additional restriction that no
1848
junk element appears in the block. Then that block is extended as
1849
far as possible by matching (only) junk elements on both sides. So
1850
the resulting block never matches on junk except as identical junk
1851
happens to be adjacent to an "interesting" match.
1852
1853
Here's the same example as before, but considering blanks to be
1854
junk. That prevents " abcd" from matching the " abcd" at the tail
1855
end of the second sequence directly. Instead only the "abcd" can
1856
match, and matches the leftmost "abcd" in the second sequence:
1857
1858
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1859
>>> s.find_longest_match(0, 5, 0, 9)
1860
(1, 0, 4)
1861
1862
If no blocks match, return (alo, blo, 0).
1863
1864
>>> s = SequenceMatcher(None, "ab", "c")
1865
>>> s.find_longest_match(0, 2, 0, 1)
1866
(0, 0, 0)
1867
"""
1868
1869
# CAUTION: stripping common prefix or suffix would be incorrect.
1870
# E.g.,
1871
# ab
1872
# acab
1873
# Longest matching block is "ab", but if common prefix is
1874
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1875
# strip, so ends up claiming that ab is changed to acab by
1876
# inserting "ca" in the middle. That's minimal but unintuitive:
1877
# "it's obvious" that someone inserted "ac" at the front.
1878
# Windiff ends up at the same place as diff, but by pairing up
1879
# the unique 'b's and then matching the first two 'a's.
1880
1881
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1882
besti, bestj, bestsize = alo, blo, 0
1883
# find longest junk-free match
1884
# during an iteration of the loop, j2len[j] = length of longest
1885
# junk-free match ending with a[i-1] and b[j]
1886
j2len = {}
1887
# nothing = []
1888
b2jget = b2j.get
1889
for i in xrange(alo, ahi):
1890
# look at all instances of a[i] in b; note that because
1891
# b2j has no junk keys, the loop is skipped if a[i] is junk
1892
j2lenget = j2len.get
1893
newj2len = {}
3167
1894
3168
def get_raw_records(self, memos_for_retrieval):
3169
"""Get the raw bytes for a records.
1895
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1896
# following improvement
1897
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1898
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1899
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1900
# to
1901
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1902
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1903
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
3170
1904
3171
:param memos_for_retrieval: An iterable containing the (index, pos,
3172
length) memo for retrieving the bytes. The Pack access method
3173
looks up the pack to use for a given record in its index_to_pack
3174
map.
3175
:return: An iterator over the bytes of the records.
3176
"""
3177
# first pass, group into same-index requests
3178
request_lists = []
3179
current_index = None
3180
for (index, offset, length) in memos_for_retrieval:
3181
if current_index == index:
3182
current_list.append((offset, length))
3183
else:
3184
if current_index is not None:
3185
request_lists.append((current_index, current_list))
3186
current_index = index
3187
current_list = [(offset, length)]
3188
# handle the last entry
3189
if current_index is not None:
3190
request_lists.append((current_index, current_list))
3191
for index, offsets in request_lists:
3192
1905
try:
3193
transport, path = self._indices[index]
1906
js = b2j[a[i]]
3194
1907
except KeyError:
3195
# A KeyError here indicates that someone has triggered an index
3196
# reload, and this index has gone missing, we need to start
3197
# over.
3198
if self._reload_func is None:
3199
# If we don't have a _reload_func there is nothing that can
3200
# be done
3201
raise
3202
raise errors.RetryWithNewPacks(index,
3203
reload_occurred=True,
3204
exc_info=sys.exc_info())
3205
try:
3206
reader = pack.make_readv_reader(transport, path, offsets)
3207
for names, read_func in reader.iter_records():
3208
yield read_func(None)
3209
except errors.NoSuchFile:
3210
# A NoSuchFile error indicates that a pack file has gone
3211
# missing on disk, we need to trigger a reload, and start over.
3212
if self._reload_func is None:
3213
raise
3214
raise errors.RetryWithNewPacks(transport.abspath(path),
3215
reload_occurred=False,
3216
exc_info=sys.exc_info())
3217
3218
def set_writer(self, writer, index, transport_packname):
3219
"""Set a writer to use for adding data."""
3220
if index is not None:
3221
self._indices[index] = transport_packname
3222
self._container_writer = writer
3223
self._write_index = index
3224
3225
def reload_or_raise(self, retry_exc):
3226
"""Try calling the reload function, or re-raise the original exception.
3227
3228
This should be called after _DirectPackAccess raises a
3229
RetryWithNewPacks exception. This function will handle the common logic
3230
of determining when the error is fatal versus being temporary.
3231
It will also make sure that the original exception is raised, rather
3232
than the RetryWithNewPacks exception.
3233
3234
If this function returns, then the calling function should retry
3235
whatever operation was being performed. Otherwise an exception will
3236
be raised.
3237
3238
:param retry_exc: A RetryWithNewPacks exception.
3239
"""
3240
is_error = False
3241
if self._reload_func is None:
3242
is_error = True
3243
elif not self._reload_func():
3244
# The reload claimed that nothing changed
3245
if not retry_exc.reload_occurred:
3246
# If there wasn't an earlier reload, then we really were
3247
# expecting to find changes. We didn't find them, so this is a
3248
# hard error
3249
is_error = True
3250
if is_error:
3251
exc_class, exc_value, exc_traceback = retry_exc.exc_info
3252
raise exc_class, exc_value, exc_traceback
3253
3254
3255
# Deprecated, use PatienceSequenceMatcher instead
3256
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
3257
3258
3259
def annotate_knit(knit, revision_id):
3260
"""Annotate a knit with no cached annotations.
3261
3262
This implementation is for knits with no cached annotations.
3263
It will work for knits with cached annotations, but this is not
3264
recommended.
3265
"""
3266
annotator = _KnitAnnotator(knit)
3267
return iter(annotator.annotate(revision_id))
3268
3269
3270
class _KnitAnnotator(object):
3271
"""Build up the annotations for a text."""
3272
3273
def __init__(self, knit):
3274
self._knit = knit
3275
3276
# Content objects, differs from fulltexts because of how final newlines
3277
# are treated by knits. the content objects here will always have a
3278
# final newline
3279
self._fulltext_contents = {}
3280
3281
# Annotated lines of specific revisions
3282
self._annotated_lines = {}
3283
3284
# Track the raw data for nodes that we could not process yet.
3285
# This maps the revision_id of the base to a list of children that will
3286
# annotated from it.
3287
self._pending_children = {}
3288
3289
# Nodes which cannot be extracted
3290
self._ghosts = set()
3291
3292
# Track how many children this node has, so we know if we need to keep
3293
# it
3294
self._annotate_children = {}
3295
self._compression_children = {}
3296
3297
self._all_build_details = {}
3298
# The children => parent revision_id graph
3299
self._revision_id_graph = {}
3300
3301
self._heads_provider = None
3302
3303
self._nodes_to_keep_annotations = set()
3304
self._generations_until_keep = 100
3305
3306
def set_generations_until_keep(self, value):
3307
"""Set the number of generations before caching a node.
3308
3309
Setting this to -1 will cache every merge node, setting this higher
3310
will cache fewer nodes.
3311
"""
3312
self._generations_until_keep = value
3313
3314
def _add_fulltext_content(self, revision_id, content_obj):
3315
self._fulltext_contents[revision_id] = content_obj
3316
# TODO: jam 20080305 It might be good to check the sha1digest here
3317
return content_obj.text()
3318
3319
def _check_parents(self, child, nodes_to_annotate):
3320
"""Check if all parents have been processed.
3321
3322
:param child: A tuple of (rev_id, parents, raw_content)
3323
:param nodes_to_annotate: If child is ready, add it to
3324
nodes_to_annotate, otherwise put it back in self._pending_children
3325
"""
3326
for parent_id in child[1]:
3327
if (parent_id not in self._annotated_lines):
3328
# This parent is present, but another parent is missing
3329
self._pending_children.setdefault(parent_id,
3330
[]).append(child)
3331
break
3332
else:
3333
# This one is ready to be processed
3334
nodes_to_annotate.append(child)
3335
3336
def _add_annotation(self, revision_id, fulltext, parent_ids,
3337
left_matching_blocks=None):
3338
"""Add an annotation entry.
3339
3340
All parents should already have been annotated.
3341
:return: A list of children that now have their parents satisfied.
3342
"""
3343
a = self._annotated_lines
3344
annotated_parent_lines = [a[p] for p in parent_ids]
3345
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
3346
fulltext, revision_id, left_matching_blocks,
3347
heads_provider=self._get_heads_provider()))
3348
self._annotated_lines[revision_id] = annotated_lines
3349
for p in parent_ids:
3350
ann_children = self._annotate_children[p]
3351
ann_children.remove(revision_id)
3352
if (not ann_children
3353
and p not in self._nodes_to_keep_annotations):
3354
del self._annotated_lines[p]
3355
del self._all_build_details[p]
3356
if p in self._fulltext_contents:
3357
del self._fulltext_contents[p]
3358
# Now that we've added this one, see if there are any pending
3359
# deltas to be done, certainly this parent is finished
3360
nodes_to_annotate = []
3361
for child in self._pending_children.pop(revision_id, []):
3362
self._check_parents(child, nodes_to_annotate)
3363
return nodes_to_annotate
3364
3365
def _get_build_graph(self, key):
3366
"""Get the graphs for building texts and annotations.
3367
3368
The data you need for creating a full text may be different than the
3369
data you need to annotate that text. (At a minimum, you need both
3370
parents to create an annotation, but only need 1 parent to generate the
3371
fulltext.)
3372
3373
:return: A list of (key, index_memo) records, suitable for
3374
passing to read_records_iter to start reading in the raw data fro/
3375
the pack file.
3376
"""
3377
if key in self._annotated_lines:
3378
# Nothing to do
3379
return []
3380
pending = set([key])
3381
records = []
3382
generation = 0
3383
kept_generation = 0
3384
while pending:
3385
# get all pending nodes
3386
generation += 1
3387
this_iteration = pending
3388
build_details = self._knit._index.get_build_details(this_iteration)
3389
self._all_build_details.update(build_details)
3390
# new_nodes = self._knit._index._get_entries(this_iteration)
3391
pending = set()
3392
for key, details in build_details.iteritems():
3393
(index_memo, compression_parent, parents,
3394
record_details) = details
3395
self._revision_id_graph[key] = parents
3396
records.append((key, index_memo))
3397
# Do we actually need to check _annotated_lines?
3398
pending.update(p for p in parents
3399
if p not in self._all_build_details)
3400
if compression_parent:
3401
self._compression_children.setdefault(compression_parent,
3402
[]).append(key)
3403
if parents:
3404
for parent in parents:
3405
self._annotate_children.setdefault(parent,
3406
[]).append(key)
3407
num_gens = generation - kept_generation
3408
if ((num_gens >= self._generations_until_keep)
3409
and len(parents) > 1):
3410
kept_generation = generation
3411
self._nodes_to_keep_annotations.add(key)
3412
3413
missing_versions = this_iteration.difference(build_details.keys())
3414
self._ghosts.update(missing_versions)
3415
for missing_version in missing_versions:
3416
# add a key, no parents
3417
self._revision_id_graph[missing_version] = ()
3418
pending.discard(missing_version) # don't look for it
3419
if self._ghosts.intersection(self._compression_children):
3420
raise KnitCorrupt(
3421
"We cannot have nodes which have a ghost compression parent:\n"
3422
"ghosts: %r\n"
3423
"compression children: %r"
3424
% (self._ghosts, self._compression_children))
3425
# Cleanout anything that depends on a ghost so that we don't wait for
3426
# the ghost to show up
3427
for node in self._ghosts:
3428
if node in self._annotate_children:
3429
# We won't be building this node
3430
del self._annotate_children[node]
3431
# Generally we will want to read the records in reverse order, because
3432
# we find the parent nodes after the children
3433
records.reverse()
3434
return records
3435
3436
def _annotate_records(self, records):
3437
"""Build the annotations for the listed records."""
3438
# We iterate in the order read, rather than a strict order requested
3439
# However, process what we can, and put off to the side things that
3440
# still need parents, cleaning them up when those parents are
3441
# processed.
3442
for (rev_id, record,
3443
digest) in self._knit._read_records_iter(records):
3444
if rev_id in self._annotated_lines:
3445
continue
3446
parent_ids = self._revision_id_graph[rev_id]
3447
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3448
details = self._all_build_details[rev_id]
3449
(index_memo, compression_parent, parents,
3450
record_details) = details
3451
nodes_to_annotate = []
3452
# TODO: Remove the punning between compression parents, and
3453
# parent_ids, we should be able to do this without assuming
3454
# the build order
3455
if len(parent_ids) == 0:
3456
# There are no parents for this node, so just add it
3457
# TODO: This probably needs to be decoupled
3458
fulltext_content, delta = self._knit._factory.parse_record(
3459
rev_id, record, record_details, None)
3460
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3461
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3462
parent_ids, left_matching_blocks=None))
3463
else:
3464
child = (rev_id, parent_ids, record)
3465
# Check if all the parents are present
3466
self._check_parents(child, nodes_to_annotate)
3467
while nodes_to_annotate:
3468
# Should we use a queue here instead of a stack?
3469
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3470
(index_memo, compression_parent, parents,
3471
record_details) = self._all_build_details[rev_id]
3472
blocks = None
3473
if compression_parent is not None:
3474
comp_children = self._compression_children[compression_parent]
3475
if rev_id not in comp_children:
3476
raise AssertionError("%r not in compression children %r"
3477
% (rev_id, comp_children))
3478
# If there is only 1 child, it is safe to reuse this
3479
# content
3480
reuse_content = (len(comp_children) == 1
3481
and compression_parent not in
3482
self._nodes_to_keep_annotations)
3483
if reuse_content:
3484
# Remove it from the cache since it will be changing
3485
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3486
# Make sure to copy the fulltext since it might be
3487
# modified
3488
parent_fulltext = list(parent_fulltext_content.text())
3489
else:
3490
parent_fulltext_content = self._fulltext_contents[compression_parent]
3491
parent_fulltext = parent_fulltext_content.text()
3492
comp_children.remove(rev_id)
3493
fulltext_content, delta = self._knit._factory.parse_record(
3494
rev_id, record, record_details,
3495
parent_fulltext_content,
3496
copy_base_content=(not reuse_content))
3497
fulltext = self._add_fulltext_content(rev_id,
3498
fulltext_content)
3499
if compression_parent == parent_ids[0]:
3500
# the compression_parent is the left parent, so we can
3501
# re-use the delta
3502
blocks = KnitContent.get_line_delta_blocks(delta,
3503
parent_fulltext, fulltext)
3504
else:
3505
fulltext_content = self._knit._factory.parse_fulltext(
3506
record, rev_id)
3507
fulltext = self._add_fulltext_content(rev_id,
3508
fulltext_content)
3509
nodes_to_annotate.extend(
3510
self._add_annotation(rev_id, fulltext, parent_ids,
3511
left_matching_blocks=blocks))
3512
3513
def _get_heads_provider(self):
3514
"""Create a heads provider for resolving ancestry issues."""
3515
if self._heads_provider is not None:
3516
return self._heads_provider
3517
parent_provider = _mod_graph.DictParentsProvider(
3518
self._revision_id_graph)
3519
graph_obj = _mod_graph.Graph(parent_provider)
3520
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3521
self._heads_provider = head_cache
3522
return head_cache
3523
3524
def annotate(self, key):
3525
"""Return the annotated fulltext at the given key.
3526
3527
:param key: The key to annotate.
3528
"""
3529
if len(self._knit._fallback_vfs) > 0:
3530
# stacked knits can't use the fast path at present.
3531
return self._simple_annotate(key)
3532
while True:
3533
try:
3534
records = self._get_build_graph(key)
3535
if key in self._ghosts:
3536
raise errors.RevisionNotPresent(key, self._knit)
3537
self._annotate_records(records)
3538
return self._annotated_lines[key]
3539
except errors.RetryWithNewPacks, e:
3540
self._knit._access.reload_or_raise(e)
3541
# The cached build_details are no longer valid
3542
self._all_build_details.clear()
3543
3544
def _simple_annotate(self, key):
3545
"""Return annotated fulltext, rediffing from the full texts.
3546
3547
This is slow but makes no assumptions about the repository
3548
being able to produce line deltas.
3549
"""
3550
# TODO: this code generates a parent maps of present ancestors; it
3551
# could be split out into a separate method, and probably should use
3552
# iter_ancestry instead. -- mbp and robertc 20080704
3553
graph = _mod_graph.Graph(self._knit)
3554
head_cache = _mod_graph.FrozenHeadsCache(graph)
3555
search = graph._make_breadth_first_searcher([key])
3556
keys = set()
3557
while True:
3558
try:
3559
present, ghosts = search.next_with_ghosts()
3560
except StopIteration:
3561
break
3562
keys.update(present)
3563
parent_map = self._knit.get_parent_map(keys)
3564
parent_cache = {}
3565
reannotate = annotate.reannotate
3566
for record in self._knit.get_record_stream(keys, 'topological', True):
3567
key = record.key
3568
fulltext = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
3569
parents = parent_map[key]
3570
if parents is not None:
3571
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
3572
else:
3573
parent_lines = []
3574
parent_cache[key] = list(
3575
reannotate(parent_lines, fulltext, key, None, head_cache))
3576
try:
3577
return parent_cache[key]
3578
except KeyError, e:
3579
raise errors.RevisionNotPresent(key, self._knit)
3580
3581
3582
try:
3583
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3584
except ImportError:
3585
from bzrlib._knit_load_data_py import _load_data_py as _load_data
1908
pass
1909
else:
1910
for j in js:
1911
# a[i] matches b[j]
1912
if j >= blo:
1913
if j >= bhi:
1914
break
1915
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1916
if k > bestsize:
1917
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1918
j2len = newj2len
1919
1920
# Extend the best by non-junk elements on each end. In particular,
1921
# "popular" non-junk elements aren't in b2j, which greatly speeds
1922
# the inner loop above, but also means "the best" match so far
1923
# doesn't contain any junk *or* popular non-junk elements.
1924
while besti > alo and bestj > blo and \
1925
not isbjunk(b[bestj-1]) and \
1926
a[besti-1] == b[bestj-1]:
1927
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1928
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1929
not isbjunk(b[bestj+bestsize]) and \
1930
a[besti+bestsize] == b[bestj+bestsize]:
1931
bestsize += 1
1932
1933
# Now that we have a wholly interesting match (albeit possibly
1934
# empty!), we may as well suck up the matching junk on each
1935
# side of it too. Can't think of a good reason not to, and it
1936
# saves post-processing the (possibly considerable) expense of
1937
# figuring out what to do with it. In the case of an empty
1938
# interesting match, this is clearly the right thing to do,
1939
# because no other kind of match is possible in the regions.
1940
while besti > alo and bestj > blo and \
1941
isbjunk(b[bestj-1]) and \
1942
a[besti-1] == b[bestj-1]:
1943
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1944
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1945
isbjunk(b[bestj+bestsize]) and \
1946
a[besti+bestsize] == b[bestj+bestsize]:
1947
bestsize = bestsize + 1
1948
1949
return besti, bestj, bestsize
Loggerhead is a web-based interface for Breezy
Version: 2.0.1