« back to all changes in this revision
Viewing changes to bzrlib/knit.py
- browse files at revision 3286
- compare with another revision
- download diff
- download tarball
- view history from revision 3286
- Committer: Canonical.com Patch Queue Manager
- Date: 2008-03-16 14:01:20 UTC
- mfrom: (3280.2.5 integration)
- Revision ID: pqm@pqm.ubuntu.com-20080316140120-i3yq8yr1l66m11h7
Start 1.4 development
- files added:
- bzrlib/help_topics/en/hooks.txt
- bzrlib/tests/interversionedfile_implementations
- bzrlib/tests/revisionstore_implementations
- bzrlib/util/configobj/docs
- files removed:
- bzrlib/_btree_serializer_c.pyx
- bzrlib/filters
- bzrlib/plugins/netrc_credential_store
- bzrlib/plugins/netrc_credential_store/tests
- bzrlib/tests/ftp_server
- bzrlib/tests/per_interbranch
- bzrlib/tests/per_repository_chk
- bzrlib/tests/per_repository_reference
- bzrlib/tests/ssl_certs
- bzrlib/transport/ftp
- 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
- tools/packaging
- files renamed:
- bzrlib/tests/ftp_server/medusa_based.py => bzrlib/tests/ftp_server.py
- bzrlib/tests/per_repository/ => bzrlib/tests/repository_implementations/
- bzrlib/tests/test_revisionspec.py => bzrlib/tests/test_revisionnamespaces.py
- bzrlib/transport/ftp/__init__.py => bzrlib/transport/ftp.py
- bzrlib/xml8.py => bzrlib/xml5.py
- doc/en/user-guide/part2_intro.txt => doc/en/user-guide/best_practice_intro.txt
- doc/en/user-guide/zen.txt => doc/en/user-guide/revnos.txt
- files modified:
- .bzrignore
added
removed
bzrlib/knit.py
1
# Copyright (C) 2005, 2006, 2007, 2008 Canonical Ltd
1
# Copyright (C) 2005, 2006, 2007 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
56
65
from itertools import izip, chain
57
66
import operator
58
67
import os
59
68
import sys
69
import warnings
70
from zlib import Z_DEFAULT_COMPRESSION
60
71
72
import bzrlib
61
73
from bzrlib.lazy_import import lazy_import
62
74
lazy_import(globals(), """
63
75
from bzrlib import (
64
76
annotate,
65
debug,
66
diff,
67
77
graph as _mod_graph,
68
index as _mod_index,
69
78
lru_cache,
70
79
pack,
71
progress,
72
80
trace,
73
tsort,
74
tuned_gzip,
75
81
)
76
82
""")
77
83
from bzrlib import (
84
cache_utf8,
85
debug,
86
diff,
78
87
errors,
79
88
osutils,
80
89
patiencediff,
90
progress,
91
merge,
92
ui,
81
93
)
82
94
from bzrlib.errors import (
83
95
FileExists,
88
100
KnitHeaderError,
89
101
RevisionNotPresent,
90
102
RevisionAlreadyPresent,
91
SHA1KnitCorrupt,
92
103
)
104
from bzrlib.tuned_gzip import GzipFile, bytes_to_gzip
93
105
from bzrlib.osutils import (
94
106
contains_whitespace,
95
107
contains_linebreaks,
96
108
sha_string,
97
109
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
)
110
)
111
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
112
from bzrlib.tsort import topo_sort
113
import bzrlib.ui
114
import bzrlib.weave
115
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
110
116
111
117
112
118
# TODO: Split out code specific to this format into an associated object.
124
130
125
131
DATA_SUFFIX = '.knit'
126
132
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
133
409
134
410
135
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
"""
136
"""Content of a knit version to which deltas can be applied."""
417
137
418
138
def __init__(self):
419
139
self._should_strip_eol = False
420
140
141
def annotate(self):
142
"""Return a list of (origin, text) tuples."""
143
return list(self.annotate_iter())
144
421
145
def apply_delta(self, delta, new_version_id):
422
146
"""Apply delta to this object to become new_version_id."""
423
147
raise NotImplementedError(self.apply_delta)
424
148
149
def cleanup_eol(self, copy_on_mutate=True):
150
if self._should_strip_eol:
151
if copy_on_mutate:
152
self._lines = self._lines[:]
153
self.strip_last_line_newline()
154
425
155
def line_delta_iter(self, new_lines):
426
156
"""Generate line-based delta from this content to new_lines."""
427
157
new_texts = new_lines.text()
470
200
KnitContent.__init__(self)
471
201
self._lines = lines
472
202
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
203
def annotate_iter(self):
204
"""Yield tuples of (origin, text) for each content line."""
205
return iter(self._lines)
480
206
481
207
def apply_delta(self, delta, new_version_id):
482
208
"""Apply delta to this object to become new_version_id."""
486
212
lines[offset+start:offset+end] = delta_lines
487
213
offset = offset + (start - end) + count
488
214
215
def strip_last_line_newline(self):
216
line = self._lines[-1][1].rstrip('\n')
217
self._lines[-1] = (self._lines[-1][0], line)
218
self._should_strip_eol = False
219
489
220
def text(self):
490
221
try:
491
222
lines = [text for origin, text in self._lines]
496
227
raise KnitCorrupt(self,
497
228
"line in annotated knit missing annotation information: %s"
498
229
% (e,))
230
499
231
if self._should_strip_eol:
500
lines[-1] = lines[-1].rstrip('\n')
232
anno, line = lines[-1]
233
lines[-1] = (anno, line.rstrip('\n'))
501
234
return lines
502
235
503
236
def copy(self):
506
239
507
240
class PlainKnitContent(KnitContent):
508
241
"""Unannotated content.
509
242
510
243
When annotate[_iter] is called on this content, the same version is reported
511
244
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
512
245
objects.
517
250
self._lines = lines
518
251
self._version_id = version_id
519
252
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]
253
def annotate_iter(self):
254
"""Yield tuples of (origin, text) for each content line."""
255
for line in self._lines:
256
yield self._version_id, line
523
257
524
258
def apply_delta(self, delta, new_version_id):
525
259
"""Apply delta to this object to become new_version_id."""
533
267
def copy(self):
534
268
return PlainKnitContent(self._lines[:], self._version_id)
535
269
270
def strip_last_line_newline(self):
271
self._lines[-1] = self._lines[-1].rstrip('\n')
272
self._should_strip_eol = False
273
536
274
def text(self):
537
275
lines = self._lines
538
276
if self._should_strip_eol:
562
300
"""
563
301
method, noeol = record_details
564
302
if method == 'line-delta':
303
assert base_content is not None
565
304
if copy_base_content:
566
305
content = base_content.copy()
567
306
else:
682
421
for origin, text in lines)
683
422
return out
684
423
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()
424
def annotate_iter(self, knit, version_id):
425
content = knit._get_content(version_id)
426
return content.annotate_iter()
700
427
701
428
702
429
class KnitPlainFactory(_KnitFactory):
756
483
out.extend(lines)
757
484
return out
758
485
759
def annotate(self, knit, key):
486
def annotate_iter(self, knit, version_id):
760
487
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
488
return iter(annotator.annotate(version_id))
489
490
491
def make_empty_knit(transport, relpath):
492
"""Construct a empty knit at the specified location."""
493
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
494
495
496
class KnitVersionedFile(VersionedFile):
497
"""Weave-like structure with faster random access.
498
499
A knit stores a number of texts and a summary of the relationships
500
between them. Texts are identified by a string version-id. Texts
501
are normally stored and retrieved as a series of lines, but can
502
also be passed as single strings.
503
504
Lines are stored with the trailing newline (if any) included, to
505
avoid special cases for files with no final newline. Lines are
506
composed of 8-bit characters, not unicode. The combination of
507
these approaches should mean any 'binary' file can be safely
508
stored and retrieved.
509
"""
510
511
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
512
factory=None, delta=True, create=False, create_parent_dir=False,
513
delay_create=False, dir_mode=None, index=None, access_method=None):
514
"""Construct a knit at location specified by relpath.
515
516
:param create: If not True, only open an existing knit.
517
:param create_parent_dir: If True, create the parent directory if
518
creating the file fails. (This is used for stores with
519
hash-prefixes that may not exist yet)
520
:param delay_create: The calling code is aware that the knit won't
521
actually be created until the first data is stored.
522
:param index: An index to use for the knit.
523
"""
524
if access_mode is None:
525
access_mode = 'w'
526
super(KnitVersionedFile, self).__init__(access_mode)
527
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
528
self.transport = transport
529
self.filename = relpath
530
self.factory = factory or KnitAnnotateFactory()
531
self.writable = (access_mode == 'w')
532
self.delta = delta
533
534
self._max_delta_chain = 200
535
536
if index is None:
537
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
538
access_mode, create=create, file_mode=file_mode,
539
create_parent_dir=create_parent_dir, delay_create=delay_create,
540
dir_mode=dir_mode)
541
else:
542
self._index = index
543
if access_method is None:
544
_access = _KnitAccess(transport, relpath + DATA_SUFFIX, file_mode, dir_mode,
545
((create and not len(self)) and delay_create), create_parent_dir)
546
else:
547
_access = access_method
548
if create and not len(self) and not delay_create:
549
_access.create()
550
self._data = _KnitData(_access)
551
552
def __repr__(self):
553
return '%s(%s)' % (self.__class__.__name__,
554
self.transport.abspath(self.filename))
555
556
def _check_should_delta(self, first_parents):
557
"""Iterate back through the parent listing, looking for a fulltext.
558
559
This is used when we want to decide whether to add a delta or a new
560
fulltext. It searches for _max_delta_chain parents. When it finds a
561
fulltext parent, it sees if the total size of the deltas leading up to
562
it is large enough to indicate that we want a new full text anyway.
563
564
Return True if we should create a new delta, False if we should use a
565
full text.
566
"""
567
delta_size = 0
568
fulltext_size = None
569
delta_parents = first_parents
570
for count in xrange(self._max_delta_chain):
571
parent = delta_parents[0]
572
method = self._index.get_method(parent)
573
index, pos, size = self._index.get_position(parent)
574
if method == 'fulltext':
575
fulltext_size = size
576
break
577
delta_size += size
578
delta_parents = self._index.get_parents(parent)
579
else:
580
# We couldn't find a fulltext, so we must create a new one
581
return False
582
583
return fulltext_size > delta_size
584
585
def _add_raw_records(self, records, data):
586
"""Add all the records 'records' with data pre-joined in 'data'.
587
588
:param records: A list of tuples(version_id, options, parents, size).
589
:param data: The data for the records. When it is written, the records
590
are adjusted to have pos pointing into data by the sum of
591
the preceding records sizes.
592
"""
593
# write all the data
594
raw_record_sizes = [record[3] for record in records]
595
positions = self._data.add_raw_records(raw_record_sizes, data)
596
offset = 0
597
index_entries = []
598
for (version_id, options, parents, size), access_memo in zip(
599
records, positions):
600
index_entries.append((version_id, options, access_memo, parents))
601
if self._data._do_cache:
602
self._data._cache[version_id] = data[offset:offset+size]
603
offset += size
604
self._index.add_versions(index_entries)
605
606
def enable_cache(self):
607
"""Start caching data for this knit"""
608
self._data.enable_cache()
609
610
def clear_cache(self):
611
"""Clear the data cache only."""
612
self._data.clear_cache()
613
614
def copy_to(self, name, transport):
615
"""See VersionedFile.copy_to()."""
616
# copy the current index to a temp index to avoid racing with local
617
# writes
618
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
619
self.transport.get(self._index._filename))
620
# copy the data file
621
f = self._data._open_file()
622
try:
623
transport.put_file(name + DATA_SUFFIX, f)
624
finally:
625
f.close()
626
# move the copied index into place
627
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
628
629
def create_empty(self, name, transport, mode=None):
630
return KnitVersionedFile(name, transport, factory=self.factory,
631
delta=self.delta, create=True)
632
633
def get_data_stream(self, required_versions):
634
"""Get a data stream for the specified versions.
635
636
Versions may be returned in any order, not necessarily the order
637
specified. They are returned in a partial order by compression
638
parent, so that the deltas can be applied as the data stream is
639
inserted; however note that compression parents will not be sent
640
unless they were specifically requested, as the client may already
641
have them.
642
643
:param required_versions: The exact set of versions to be extracted.
644
Unlike some other knit methods, this is not used to generate a
645
transitive closure, rather it is used precisely as given.
646
647
:returns: format_signature, list of (version, options, length, parents),
648
reader_callable.
649
"""
650
required_version_set = frozenset(required_versions)
651
version_index = {}
652
# list of revisions that can just be sent without waiting for their
653
# compression parent
654
ready_to_send = []
655
# map from revision to the children based on it
656
deferred = {}
657
# first, read all relevant index data, enough to sort into the right
658
# order to return
659
for version_id in required_versions:
660
options = self._index.get_options(version_id)
661
parents = self._index.get_parents_with_ghosts(version_id)
662
index_memo = self._index.get_position(version_id)
663
version_index[version_id] = (index_memo, options, parents)
664
if ('line-delta' in options
665
and parents[0] in required_version_set):
666
# must wait until the parent has been sent
667
deferred.setdefault(parents[0], []). \
668
append(version_id)
669
else:
670
# either a fulltext, or a delta whose parent the client did
671
# not ask for and presumably already has
672
ready_to_send.append(version_id)
673
# build a list of results to return, plus instructions for data to
674
# read from the file
675
copy_queue_records = []
676
temp_version_list = []
677
while ready_to_send:
678
# XXX: pushing and popping lists may be a bit inefficient
679
version_id = ready_to_send.pop(0)
680
(index_memo, options, parents) = version_index[version_id]
681
copy_queue_records.append((version_id, index_memo))
682
none, data_pos, data_size = index_memo
683
temp_version_list.append((version_id, options, data_size,
684
parents))
685
if version_id in deferred:
686
# now we can send all the children of this revision - we could
687
# put them in anywhere, but we hope that sending them soon
688
# after the fulltext will give good locality in the receiver
689
ready_to_send[:0] = deferred.pop(version_id)
690
assert len(deferred) == 0, \
691
"Still have compressed child versions waiting to be sent"
692
# XXX: The stream format is such that we cannot stream it - we have to
693
# know the length of all the data a-priori.
694
raw_datum = []
695
result_version_list = []
696
for (version_id, raw_data), \
697
(version_id2, options, _, parents) in \
698
izip(self._data.read_records_iter_raw(copy_queue_records),
699
temp_version_list):
700
assert version_id == version_id2, \
701
'logic error, inconsistent results'
702
raw_datum.append(raw_data)
703
result_version_list.append(
704
(version_id, options, len(raw_data), parents))
705
# provide a callback to get data incrementally.
706
pseudo_file = StringIO(''.join(raw_datum))
707
def read(length):
708
if length is None:
709
return pseudo_file.read()
710
else:
711
return pseudo_file.read(length)
712
return (self.get_format_signature(), result_version_list, read)
713
714
def _extract_blocks(self, version_id, source, target):
715
if self._index.get_method(version_id) != 'line-delta':
716
return None
717
parent, sha1, noeol, delta = self.get_delta(version_id)
718
return KnitContent.get_line_delta_blocks(delta, source, target)
719
720
def get_delta(self, version_id):
721
"""Get a delta for constructing version from some other version."""
722
self.check_not_reserved_id(version_id)
723
parents = self.get_parents(version_id)
724
if len(parents):
725
parent = parents[0]
726
else:
727
parent = None
728
index_memo = self._index.get_position(version_id)
729
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
730
noeol = 'no-eol' in self._index.get_options(version_id)
731
if 'fulltext' == self._index.get_method(version_id):
732
new_content = self.factory.parse_fulltext(data, version_id)
733
if parent is not None:
734
reference_content = self._get_content(parent)
735
old_texts = reference_content.text()
736
else:
737
old_texts = []
738
new_texts = new_content.text()
739
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
740
new_texts)
741
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
742
else:
743
delta = self.factory.parse_line_delta(data, version_id)
744
return parent, sha1, noeol, delta
745
746
def get_format_signature(self):
747
"""See VersionedFile.get_format_signature()."""
748
if self.factory.annotated:
749
annotated_part = "annotated"
750
else:
751
annotated_part = "plain"
752
return "knit-%s" % (annotated_part,)
753
754
def get_graph_with_ghosts(self):
755
"""See VersionedFile.get_graph_with_ghosts()."""
756
graph_items = self._index.get_graph()
757
return dict(graph_items)
758
759
def get_sha1(self, version_id):
760
return self.get_sha1s([version_id])[0]
761
762
def get_sha1s(self, version_ids):
763
"""See VersionedFile.get_sha1()."""
764
record_map = self._get_record_map(version_ids)
765
# record entry 2 is the 'digest'.
766
return [record_map[v][2] for v in version_ids]
767
768
@staticmethod
769
def get_suffixes():
770
"""See VersionedFile.get_suffixes()."""
771
return [DATA_SUFFIX, INDEX_SUFFIX]
772
773
def has_ghost(self, version_id):
774
"""True if there is a ghost reference in the file to version_id."""
775
# maybe we have it
776
if self.has_version(version_id):
777
return False
778
# optimisable if needed by memoising the _ghosts set.
779
items = self._index.get_graph()
780
for node, parents in items:
781
for parent in parents:
782
if parent not in self._index._cache:
783
if parent == version_id:
784
return True
785
return False
786
787
def insert_data_stream(self, (format, data_list, reader_callable)):
788
"""Insert knit records from a data stream into this knit.
789
790
If a version in the stream is already present in this knit, it will not
791
be inserted a second time. It will be checked for consistency with the
792
stored version however, and may cause a KnitCorrupt error to be raised
793
if the data in the stream disagrees with the already stored data.
794
795
:seealso: get_data_stream
796
"""
797
if format != self.get_format_signature():
798
if 'knit' in debug.debug_flags:
799
trace.mutter(
800
'incompatible format signature inserting to %r', self)
801
source = self._knit_from_datastream(
802
(format, data_list, reader_callable))
803
self.join(source)
804
return
805
806
for version_id, options, length, parents in data_list:
807
if self.has_version(version_id):
808
# First check: the list of parents.
809
my_parents = self.get_parents_with_ghosts(version_id)
810
if tuple(my_parents) != tuple(parents):
811
# XXX: KnitCorrupt is not quite the right exception here.
812
raise KnitCorrupt(
813
self.filename,
814
'parents list %r from data stream does not match '
815
'already recorded parents %r for %s'
816
% (parents, my_parents, version_id))
817
818
# Also check the SHA-1 of the fulltext this content will
819
# produce.
820
raw_data = reader_callable(length)
821
my_fulltext_sha1 = self.get_sha1(version_id)
822
df, rec = self._data._parse_record_header(version_id, raw_data)
823
stream_fulltext_sha1 = rec[3]
824
if my_fulltext_sha1 != stream_fulltext_sha1:
825
# Actually, we don't know if it's this knit that's corrupt,
826
# or the data stream we're trying to insert.
827
raise KnitCorrupt(
828
self.filename, 'sha-1 does not match %s' % version_id)
829
else:
830
if 'line-delta' in options:
831
# Make sure that this knit record is actually useful: a
832
# line-delta is no use unless we have its parent.
833
# Fetching from a broken repository with this problem
834
# shouldn't break the target repository.
835
#
836
# See https://bugs.launchpad.net/bzr/+bug/164443
837
if not self._index.has_version(parents[0]):
838
raise KnitCorrupt(
839
self.filename,
840
'line-delta from stream '
841
'for version %s '
842
'references '
843
'missing parent %s\n'
844
'Try running "bzr check" '
845
'on the source repository, and "bzr reconcile" '
846
'if necessary.' %
847
(version_id, parents[0]))
848
self._add_raw_records(
849
[(version_id, options, parents, length)],
850
reader_callable(length))
851
852
def _knit_from_datastream(self, (format, data_list, reader_callable)):
853
"""Create a knit object from a data stream.
854
855
This method exists to allow conversion of data streams that do not
856
match the signature of this knit. Generally it will be slower and use
857
more memory to use this method to insert data, but it will work.
858
859
:seealso: get_data_stream for details on datastreams.
860
:return: A knit versioned file which can be used to join the datastream
861
into self.
862
"""
863
if format == "knit-plain":
864
factory = KnitPlainFactory()
865
elif format == "knit-annotated":
866
factory = KnitAnnotateFactory()
867
else:
868
raise errors.KnitDataStreamUnknown(format)
869
index = _StreamIndex(data_list, self._index)
870
access = _StreamAccess(reader_callable, index, self, factory)
871
return KnitVersionedFile(self.filename, self.transport,
872
factory=factory, index=index, access_method=access)
873
874
def versions(self):
875
"""See VersionedFile.versions."""
876
if 'evil' in debug.debug_flags:
877
trace.mutter_callsite(2, "versions scales with size of history")
878
return self._index.get_versions()
879
880
def has_version(self, version_id):
881
"""See VersionedFile.has_version."""
882
if 'evil' in debug.debug_flags:
883
trace.mutter_callsite(2, "has_version is a LBYL scenario")
884
return self._index.has_version(version_id)
885
886
__contains__ = has_version
887
888
def _merge_annotations(self, content, parents, parent_texts={},
889
delta=None, annotated=None,
890
left_matching_blocks=None):
891
"""Merge annotations for content. This is done by comparing
892
the annotations based on changed to the text.
893
"""
894
if left_matching_blocks is not None:
895
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
896
else:
897
delta_seq = None
898
if annotated:
899
for parent_id in parents:
900
merge_content = self._get_content(parent_id, parent_texts)
901
if (parent_id == parents[0] and delta_seq is not None):
902
seq = delta_seq
903
else:
904
seq = patiencediff.PatienceSequenceMatcher(
905
None, merge_content.text(), content.text())
906
for i, j, n in seq.get_matching_blocks():
907
if n == 0:
908
continue
909
# this appears to copy (origin, text) pairs across to the
910
# new content for any line that matches the last-checked
911
# parent.
912
content._lines[j:j+n] = merge_content._lines[i:i+n]
796
913
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:
914
if delta_seq is None:
915
reference_content = self._get_content(parents[0], parent_texts)
916
new_texts = content.text()
917
old_texts = reference_content.text()
918
delta_seq = patiencediff.PatienceSequenceMatcher(
919
None, old_texts, new_texts)
920
return self._make_line_delta(delta_seq, content)
921
922
def _make_line_delta(self, delta_seq, new_content):
923
"""Generate a line delta from delta_seq and new_content."""
924
diff_hunks = []
925
for op in delta_seq.get_opcodes():
926
if op[0] == 'equal':
842
927
continue
843
_, index_memo, compression_parent = positions[key]
844
all_build_index_memos[key] = index_memo
845
if compression_parent not in all_build_index_memos:
846
next_keys.add(compression_parent)
847
build_keys = next_keys
848
return sum([index_memo[2] for index_memo
849
in all_build_index_memos.itervalues()])
850
851
852
class KnitVersionedFiles(VersionedFiles):
853
"""Storage for many versioned files using knit compression.
854
855
Backend storage is managed by indices and data objects.
856
857
:ivar _index: A _KnitGraphIndex or similar that can describe the
858
parents, graph, compression and data location of entries in this
859
KnitVersionedFiles. Note that this is only the index for
860
*this* vfs; if there are fallbacks they must be queried separately.
861
"""
862
863
def __init__(self, index, data_access, max_delta_chain=200,
864
annotated=False, reload_func=None):
865
"""Create a KnitVersionedFiles with index and data_access.
866
867
:param index: The index for the knit data.
868
:param data_access: The access object to store and retrieve knit
869
records.
870
:param max_delta_chain: The maximum number of deltas to permit during
871
insertion. Set to 0 to prohibit the use of deltas.
872
:param annotated: Set to True to cause annotations to be calculated and
873
stored during insertion.
874
:param reload_func: An function that can be called if we think we need
875
to reload the pack listing and try again. See
876
'bzrlib.repofmt.pack_repo.AggregateIndex' for the signature.
877
"""
878
self._index = index
879
self._access = data_access
880
self._max_delta_chain = max_delta_chain
881
if annotated:
882
self._factory = KnitAnnotateFactory()
883
else:
884
self._factory = KnitPlainFactory()
885
self._fallback_vfs = []
886
self._reload_func = reload_func
887
888
def __repr__(self):
889
return "%s(%r, %r)" % (
890
self.__class__.__name__,
891
self._index,
892
self._access)
893
894
def add_fallback_versioned_files(self, a_versioned_files):
895
"""Add a source of texts for texts not present in this knit.
896
897
:param a_versioned_files: A VersionedFiles object.
898
"""
899
self._fallback_vfs.append(a_versioned_files)
900
901
def add_lines(self, key, parents, lines, parent_texts=None,
902
left_matching_blocks=None, nostore_sha=None, random_id=False,
903
check_content=True):
904
"""See VersionedFiles.add_lines()."""
905
self._index._check_write_ok()
906
self._check_add(key, lines, random_id, check_content)
907
if parents is None:
908
# The caller might pass None if there is no graph data, but kndx
909
# indexes can't directly store that, so we give them
910
# an empty tuple instead.
911
parents = ()
912
return self._add(key, lines, parents,
928
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
929
return diff_hunks
930
931
def _get_components_positions(self, version_ids):
932
"""Produce a map of position data for the components of versions.
933
934
This data is intended to be used for retrieving the knit records.
935
936
A dict of version_id to (record_details, index_memo, next, parents) is
937
returned.
938
method is the way referenced data should be applied.
939
index_memo is the handle to pass to the data access to actually get the
940
data
941
next is the build-parent of the version, or None for fulltexts.
942
parents is the version_ids of the parents of this version
943
"""
944
component_data = {}
945
pending_components = version_ids
946
while pending_components:
947
build_details = self._index.get_build_details(pending_components)
948
current_components = set(pending_components)
949
pending_components = set()
950
for version_id, details in build_details.iteritems():
951
(index_memo, compression_parent, parents,
952
record_details) = details
953
method = record_details[0]
954
if compression_parent is not None:
955
pending_components.add(compression_parent)
956
component_data[version_id] = (record_details, index_memo,
957
compression_parent)
958
missing = current_components.difference(build_details)
959
if missing:
960
raise errors.RevisionNotPresent(missing.pop(), self.filename)
961
return component_data
962
963
def _get_content(self, version_id, parent_texts={}):
964
"""Returns a content object that makes up the specified
965
version."""
966
cached_version = parent_texts.get(version_id, None)
967
if cached_version is not None:
968
if not self.has_version(version_id):
969
raise RevisionNotPresent(version_id, self.filename)
970
return cached_version
971
972
text_map, contents_map = self._get_content_maps([version_id])
973
return contents_map[version_id]
974
975
def _check_versions_present(self, version_ids):
976
"""Check that all specified versions are present."""
977
self._index.check_versions_present(version_ids)
978
979
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
980
nostore_sha, random_id, check_content):
981
"""See VersionedFile.add_lines_with_ghosts()."""
982
self._check_add(version_id, lines, random_id, check_content)
983
return self._add(version_id, lines, parents, self.delta,
984
parent_texts, None, nostore_sha, random_id)
985
986
def _add_lines(self, version_id, parents, lines, parent_texts,
987
left_matching_blocks, nostore_sha, random_id, check_content):
988
"""See VersionedFile.add_lines."""
989
self._check_add(version_id, lines, random_id, check_content)
990
self._check_versions_present(parents)
991
return self._add(version_id, lines[:], parents, self.delta,
913
992
parent_texts, left_matching_blocks, nostore_sha, random_id)
914
993
915
def _add(self, key, lines, parents, parent_texts,
994
def _check_add(self, version_id, lines, random_id, check_content):
995
"""check that version_id and lines are safe to add."""
996
if contains_whitespace(version_id):
997
raise InvalidRevisionId(version_id, self.filename)
998
self.check_not_reserved_id(version_id)
999
# Technically this could be avoided if we are happy to allow duplicate
1000
# id insertion when other things than bzr core insert texts, but it
1001
# seems useful for folk using the knit api directly to have some safety
1002
# blanket that we can disable.
1003
if not random_id and self.has_version(version_id):
1004
raise RevisionAlreadyPresent(version_id, self.filename)
1005
if check_content:
1006
self._check_lines_not_unicode(lines)
1007
self._check_lines_are_lines(lines)
1008
1009
def _add(self, version_id, lines, parents, delta, parent_texts,
916
1010
left_matching_blocks, nostore_sha, random_id):
917
1011
"""Add a set of lines on top of version specified by parents.
918
1012
1013
If delta is true, compress the text as a line-delta against
1014
the first parent.
1015
919
1016
Any versions not present will be converted into ghosts.
920
1017
"""
921
1018
# first thing, if the content is something we don't need to store, find
928
1025
present_parents = []
929
1026
if parent_texts is None:
930
1027
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
1028
for parent in parents:
935
if parent in present_parent_map:
1029
if self.has_version(parent):
936
1030
present_parents.append(parent)
937
1031
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]):
1032
# can only compress against the left most present parent.
1033
if (delta and
1034
(len(present_parents) == 0 or
1035
present_parents[0] != parents[0])):
941
1036
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
1037
948
1038
text_length = len(line_bytes)
949
1039
options = []
955
1045
lines[-1] = lines[-1] + '\n'
956
1046
line_bytes += '\n'
957
1047
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):
1048
if delta:
1049
# To speed the extract of texts the delta chain is limited
1050
# to a fixed number of deltas. This should minimize both
1051
# I/O and the time spend applying deltas.
1052
delta = self._check_should_delta(present_parents)
1053
1054
assert isinstance(version_id, str)
1055
content = self.factory.make(lines, version_id)
1056
if delta or (self.factory.annotated and len(present_parents) > 0):
973
1057
# Merge annotations from parent texts if needed.
974
1058
delta_hunks = self._merge_annotations(content, present_parents,
975
parent_texts, delta, self._factory.annotated,
1059
parent_texts, delta, self.factory.annotated,
976
1060
left_matching_blocks)
977
1061
978
1062
if delta:
979
1063
options.append('line-delta')
980
store_lines = self._factory.lower_line_delta(delta_hunks)
981
size, bytes = self._record_to_data(key, digest,
1064
store_lines = self.factory.lower_line_delta(delta_hunks)
1065
size, bytes = self._data._record_to_data(version_id, digest,
982
1066
store_lines)
983
1067
else:
984
1068
options.append('fulltext')
985
1069
# isinstance is slower and we have no hierarchy.
986
if self._factory.__class__ is KnitPlainFactory:
1070
if self.factory.__class__ == KnitPlainFactory:
987
1071
# Use the already joined bytes saving iteration time in
988
1072
# _record_to_data.
989
size, bytes = self._record_to_data(key, digest,
1073
size, bytes = self._data._record_to_data(version_id, digest,
990
1074
lines, [line_bytes])
991
1075
else:
992
1076
# get mixed annotation + content and feed it into the
993
1077
# serialiser.
994
store_lines = self._factory.lower_fulltext(content)
995
size, bytes = self._record_to_data(key, digest,
1078
store_lines = self.factory.lower_fulltext(content)
1079
size, bytes = self._data._record_to_data(version_id, digest,
996
1080
store_lines)
997
1081
998
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
999
self._index.add_records(
1000
((key, options, access_memo, parents),),
1082
access_memo = self._data.add_raw_records([size], bytes)[0]
1083
self._index.add_versions(
1084
((version_id, options, access_memo, parents),),
1001
1085
random_id=random_id)
1002
1086
return digest, text_length, content
1003
1087
1004
def annotate(self, key):
1005
"""See VersionedFiles.annotate."""
1006
return self._factory.annotate(self, key)
1007
1008
1088
def check(self, progress_bar=None):
1009
"""See VersionedFiles.check()."""
1010
# This doesn't actually test extraction of everything, but that will
1011
# impact 'bzr check' substantially, and needs to be integrated with
1012
# care. However, it does check for the obvious problem of a delta with
1013
# no basis.
1014
keys = self._index.keys()
1015
parent_map = self.get_parent_map(keys)
1016
for key in keys:
1017
if self._index.get_method(key) != 'fulltext':
1018
compression_parent = parent_map[key][0]
1019
if compression_parent not in parent_map:
1020
raise errors.KnitCorrupt(self,
1021
"Missing basis parent %s for %s" % (
1022
compression_parent, key))
1023
for fallback_vfs in self._fallback_vfs:
1024
fallback_vfs.check()
1025
1026
def _check_add(self, key, lines, random_id, check_content):
1027
"""check that version_id and lines are safe to add."""
1028
version_id = key[-1]
1029
if version_id is not None:
1030
if contains_whitespace(version_id):
1031
raise InvalidRevisionId(version_id, self)
1032
self.check_not_reserved_id(version_id)
1033
# TODO: If random_id==False and the key is already present, we should
1034
# probably check that the existing content is identical to what is
1035
# being inserted, and otherwise raise an exception. This would make
1036
# the bundle code simpler.
1037
if check_content:
1038
self._check_lines_not_unicode(lines)
1039
self._check_lines_are_lines(lines)
1040
1041
def _check_header(self, key, line):
1042
rec = self._split_header(line)
1043
self._check_header_version(rec, key[-1])
1044
return rec
1045
1046
def _check_header_version(self, rec, version_id):
1047
"""Checks the header version on original format knit records.
1048
1049
These have the last component of the key embedded in the record.
1050
"""
1051
if rec[1] != version_id:
1052
raise KnitCorrupt(self,
1053
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1054
1055
def _check_should_delta(self, parent):
1056
"""Iterate back through the parent listing, looking for a fulltext.
1057
1058
This is used when we want to decide whether to add a delta or a new
1059
fulltext. It searches for _max_delta_chain parents. When it finds a
1060
fulltext parent, it sees if the total size of the deltas leading up to
1061
it is large enough to indicate that we want a new full text anyway.
1062
1063
Return True if we should create a new delta, False if we should use a
1064
full text.
1065
"""
1066
delta_size = 0
1067
fulltext_size = None
1068
for count in xrange(self._max_delta_chain):
1069
try:
1070
# Note that this only looks in the index of this particular
1071
# KnitVersionedFiles, not in the fallbacks. This ensures that
1072
# we won't store a delta spanning physical repository
1073
# boundaries.
1074
build_details = self._index.get_build_details([parent])
1075
parent_details = build_details[parent]
1076
except (RevisionNotPresent, KeyError), e:
1077
# Some basis is not locally present: always fulltext
1078
return False
1079
index_memo, compression_parent, _, _ = parent_details
1080
_, _, size = index_memo
1081
if compression_parent is None:
1082
fulltext_size = size
1083
break
1084
delta_size += size
1085
# We don't explicitly check for presence because this is in an
1086
# inner loop, and if it's missing it'll fail anyhow.
1087
parent = compression_parent
1088
else:
1089
# We couldn't find a fulltext, so we must create a new one
1090
return False
1091
# Simple heuristic - if the total I/O wold be greater as a delta than
1092
# the originally installed fulltext, we create a new fulltext.
1093
return fulltext_size > delta_size
1094
1095
def _build_details_to_components(self, build_details):
1096
"""Convert a build_details tuple to a position tuple."""
1097
# record_details, access_memo, compression_parent
1098
return build_details[3], build_details[0], build_details[1]
1099
1100
def _get_components_positions(self, keys, allow_missing=False):
1101
"""Produce a map of position data for the components of keys.
1102
1103
This data is intended to be used for retrieving the knit records.
1104
1105
A dict of key to (record_details, index_memo, next, parents) is
1106
returned.
1107
method is the way referenced data should be applied.
1108
index_memo is the handle to pass to the data access to actually get the
1109
data
1110
next is the build-parent of the version, or None for fulltexts.
1111
parents is the version_ids of the parents of this version
1112
1113
:param allow_missing: If True do not raise an error on a missing component,
1114
just ignore it.
1115
"""
1116
component_data = {}
1117
pending_components = keys
1118
while pending_components:
1119
build_details = self._index.get_build_details(pending_components)
1120
current_components = set(pending_components)
1121
pending_components = set()
1122
for key, details in build_details.iteritems():
1123
(index_memo, compression_parent, parents,
1124
record_details) = details
1125
method = record_details[0]
1126
if compression_parent is not None:
1127
pending_components.add(compression_parent)
1128
component_data[key] = self._build_details_to_components(details)
1129
missing = current_components.difference(build_details)
1130
if missing and not allow_missing:
1131
raise errors.RevisionNotPresent(missing.pop(), self)
1132
return component_data
1133
1134
def _get_content(self, key, parent_texts={}):
1135
"""Returns a content object that makes up the specified
1136
version."""
1137
cached_version = parent_texts.get(key, None)
1138
if cached_version is not None:
1139
# Ensure the cache dict is valid.
1140
if not self.get_parent_map([key]):
1141
raise RevisionNotPresent(key, self)
1142
return cached_version
1143
generator = _VFContentMapGenerator(self, [key])
1144
return generator._get_content(key)
1145
1146
def get_parent_map(self, keys):
1147
"""Get a map of the graph parents of keys.
1148
1149
:param keys: The keys to look up parents for.
1150
:return: A mapping from keys to parents. Absent keys are absent from
1151
the mapping.
1152
"""
1153
return self._get_parent_map_with_sources(keys)[0]
1154
1155
def _get_parent_map_with_sources(self, keys):
1156
"""Get a map of the parents of keys.
1157
1158
:param keys: The keys to look up parents for.
1159
:return: A tuple. The first element is a mapping from keys to parents.
1160
Absent keys are absent from the mapping. The second element is a
1161
list with the locations each key was found in. The first element
1162
is the in-this-knit parents, the second the first fallback source,
1163
and so on.
1164
"""
1165
result = {}
1166
sources = [self._index] + self._fallback_vfs
1167
source_results = []
1168
missing = set(keys)
1169
for source in sources:
1170
if not missing:
1171
break
1172
new_result = source.get_parent_map(missing)
1173
source_results.append(new_result)
1174
result.update(new_result)
1175
missing.difference_update(set(new_result))
1176
return result, source_results
1177
1178
def _get_record_map(self, keys, allow_missing=False):
1089
"""See VersionedFile.check()."""
1090
1091
def _clone_text(self, new_version_id, old_version_id, parents):
1092
"""See VersionedFile.clone_text()."""
1093
# FIXME RBC 20060228 make fast by only inserting an index with null
1094
# delta.
1095
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
1096
1097
def get_lines(self, version_id):
1098
"""See VersionedFile.get_lines()."""
1099
return self.get_line_list([version_id])[0]
1100
1101
def _get_record_map(self, version_ids):
1179
1102
"""Produce a dictionary of knit records.
1180
1181
:return: {key:(record, record_details, digest, next)}
1103
1104
:return: {version_id:(record, record_details, digest, next)}
1182
1105
record
1183
data returned from read_records (a KnitContentobject)
1106
data returned from read_records
1184
1107
record_details
1185
1108
opaque information to pass to parse_record
1186
1109
digest
1188
1111
next
1189
1112
build-parent of the version, i.e. the leftmost ancestor.
1190
1113
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
1114
"""
1115
position_map = self._get_components_positions(version_ids)
1116
# c = component_id, r = record_details, i_m = index_memo, n = next
1117
records = [(c, i_m) for c, (r, i_m, n)
1118
in position_map.iteritems()]
1119
record_map = {}
1120
for component_id, record, digest in \
1121
self._data.read_records_iter(records):
1122
(record_details, index_memo, next) = position_map[component_id]
1123
record_map[component_id] = record, record_details, digest, next
1124
1125
return record_map
1126
1127
def get_text(self, version_id):
1128
"""See VersionedFile.get_text"""
1129
return self.get_texts([version_id])[0]
1130
1131
def get_texts(self, version_ids):
1132
return [''.join(l) for l in self.get_line_list(version_ids)]
1133
1134
def get_line_list(self, version_ids):
1135
"""Return the texts of listed versions as a list of strings."""
1136
for version_id in version_ids:
1137
self.check_not_reserved_id(version_id)
1138
text_map, content_map = self._get_content_maps(version_ids)
1139
return [text_map[v] for v in version_ids]
1140
1141
_get_lf_split_line_list = get_line_list
1142
1143
def _get_content_maps(self, version_ids):
1144
"""Produce maps of text and KnitContents
1145
1982
1146
:return: (text_map, content_map) where text_map contains the texts for
1983
the requested versions and content_map contains the KnitContents.
1147
the requested versions and content_map contains the KnitContents.
1148
Both dicts take version_ids as their keys.
1984
1149
"""
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
1150
# FUTURE: This function could be improved for the 'extract many' case
2025
1151
# by tracking each component and only doing the copy when the number of
2026
1152
# children than need to apply delta's to it is > 1 or it is part of the
2027
1153
# 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
1154
version_ids = list(version_ids)
1155
multiple_versions = len(version_ids) != 1
1156
record_map = self._get_record_map(version_ids)
1157
1158
text_map = {}
1159
content_map = {}
1160
final_content = {}
1161
for version_id in version_ids:
2039
1162
components = []
2040
cursor = key
1163
cursor = version_id
2041
1164
while cursor is not None:
2042
try:
2043
record, record_details, digest, next = record_map[cursor]
2044
except KeyError:
2045
raise RevisionNotPresent(cursor, self)
1165
record, record_details, digest, next = record_map[cursor]
2046
1166
components.append((cursor, record, record_details, digest))
1167
if cursor in content_map:
1168
break
2047
1169
cursor = next
2048
if cursor in self._contents_map:
2049
# no need to plan further back
2050
components.append((cursor, None, None, None))
2051
break
2052
1170
2053
1171
content = None
2054
1172
for (component_id, record, record_details,
2055
1173
digest) in reversed(components):
2056
if component_id in self._contents_map:
2057
content = self._contents_map[component_id]
1174
if component_id in content_map:
1175
content = content_map[component_id]
2058
1176
else:
2059
content, delta = self._factory.parse_record(key[-1],
1177
content, delta = self.factory.parse_record(version_id,
2060
1178
record, record_details, content,
2061
1179
copy_base_content=multiple_versions)
2062
1180
if multiple_versions:
2063
self._contents_map[component_id] = content
1181
content_map[component_id] = content
1182
1183
content.cleanup_eol(copy_on_mutate=multiple_versions)
1184
final_content[version_id] = content
2064
1185
2065
1186
# digest here is the digest from the last applied component.
2066
1187
text = content.text()
2067
1188
actual_sha = sha_strings(text)
2068
1189
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
1190
raise KnitCorrupt(self.filename,
1191
'\n sha-1 %s'
1192
'\n of reconstructed text does not match'
1193
'\n expected %s'
1194
'\n for version %s' %
1195
(actual_sha, digest, version_id))
1196
text_map[version_id] = text
1197
return text_map, final_content
1198
1199
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1200
pb=None):
1201
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1202
if version_ids is None:
1203
version_ids = self.versions()
1204
if pb is None:
1205
pb = progress.DummyProgress()
1206
# we don't care about inclusions, the caller cares.
1207
# but we need to setup a list of records to visit.
1208
# we need version_id, position, length
1209
version_id_records = []
1210
requested_versions = set(version_ids)
1211
# filter for available versions
1212
for version_id in requested_versions:
1213
if not self.has_version(version_id):
1214
raise RevisionNotPresent(version_id, self.filename)
1215
# get a in-component-order queue:
1216
for version_id in self.versions():
1217
if version_id in requested_versions:
1218
index_memo = self._index.get_position(version_id)
1219
version_id_records.append((version_id, index_memo))
1220
1221
total = len(version_id_records)
1222
for version_idx, (version_id, data, sha_value) in \
1223
enumerate(self._data.read_records_iter(version_id_records)):
1224
pb.update('Walking content.', version_idx, total)
1225
method = self._index.get_method(version_id)
1226
1227
assert method in ('fulltext', 'line-delta')
1228
if method == 'fulltext':
1229
line_iterator = self.factory.get_fulltext_content(data)
1230
else:
1231
line_iterator = self.factory.get_linedelta_content(data)
1232
# XXX: It might be more efficient to yield (version_id,
1233
# line_iterator) in the future. However for now, this is a simpler
1234
# change to integrate into the rest of the codebase. RBC 20071110
1235
for line in line_iterator:
1236
yield line, version_id
1237
1238
pb.update('Walking content.', total, total)
1239
1240
def iter_parents(self, version_ids):
1241
"""Iterate through the parents for many version ids.
1242
1243
:param version_ids: An iterable yielding version_ids.
1244
:return: An iterator that yields (version_id, parents). Requested
1245
version_ids not present in the versioned file are simply skipped.
1246
The order is undefined, allowing for different optimisations in
1247
the underlying implementation.
1248
"""
1249
return self._index.iter_parents(version_ids)
1250
1251
def num_versions(self):
1252
"""See VersionedFile.num_versions()."""
1253
return self._index.num_versions()
1254
1255
__len__ = num_versions
1256
1257
def annotate_iter(self, version_id):
1258
"""See VersionedFile.annotate_iter."""
1259
return self.factory.annotate_iter(self, version_id)
1260
1261
def get_parents(self, version_id):
1262
"""See VersionedFile.get_parents."""
1263
# perf notes:
1264
# optimism counts!
1265
# 52554 calls in 1264 872 internal down from 3674
1266
try:
1267
return self._index.get_parents(version_id)
1268
except KeyError:
1269
raise RevisionNotPresent(version_id, self.filename)
1270
1271
def get_parents_with_ghosts(self, version_id):
1272
"""See VersionedFile.get_parents."""
1273
try:
1274
return self._index.get_parents_with_ghosts(version_id)
1275
except KeyError:
1276
raise RevisionNotPresent(version_id, self.filename)
1277
1278
def get_ancestry(self, versions, topo_sorted=True):
1279
"""See VersionedFile.get_ancestry."""
1280
if isinstance(versions, basestring):
1281
versions = [versions]
1282
if not versions:
1283
return []
1284
return self._index.get_ancestry(versions, topo_sorted)
1285
1286
def get_ancestry_with_ghosts(self, versions):
1287
"""See VersionedFile.get_ancestry_with_ghosts."""
1288
if isinstance(versions, basestring):
1289
versions = [versions]
1290
if not versions:
1291
return []
1292
return self._index.get_ancestry_with_ghosts(versions)
1293
1294
def plan_merge(self, ver_a, ver_b):
1295
"""See VersionedFile.plan_merge."""
1296
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1297
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1298
annotated_a = self.annotate(ver_a)
1299
annotated_b = self.annotate(ver_b)
1300
return merge._plan_annotate_merge(annotated_a, annotated_b,
1301
ancestors_a, ancestors_b)
1302
1303
1304
class _KnitComponentFile(object):
1305
"""One of the files used to implement a knit database"""
1306
1307
def __init__(self, transport, filename, mode, file_mode=None,
1308
create_parent_dir=False, dir_mode=None):
1309
self._transport = transport
1310
self._filename = filename
1311
self._mode = mode
1312
self._file_mode = file_mode
1313
self._dir_mode = dir_mode
1314
self._create_parent_dir = create_parent_dir
1315
self._need_to_create = False
1316
1317
def _full_path(self):
1318
"""Return the full path to this file."""
1319
return self._transport.base + self._filename
1320
1321
def check_header(self, fp):
1322
line = fp.readline()
1323
if line == '':
1324
# An empty file can actually be treated as though the file doesn't
1325
# exist yet.
1326
raise errors.NoSuchFile(self._full_path())
1327
if line != self.HEADER:
1328
raise KnitHeaderError(badline=line,
1329
filename=self._transport.abspath(self._filename))
1330
1331
def __repr__(self):
1332
return '%s(%s)' % (self.__class__.__name__, self._filename)
1333
1334
1335
class _KnitIndex(_KnitComponentFile):
1336
"""Manages knit index file.
1337
1338
The index is already kept in memory and read on startup, to enable
2272
1339
fast lookups of revision information. The cursor of the index
2273
1340
file is always pointing to the end, making it easy to append
2274
1341
entries.
2284
1351
2285
1352
Duplicate entries may be written to the index for a single version id
2286
1353
if this is done then the latter one completely replaces the former:
2287
this allows updates to correct version and parent information.
1354
this allows updates to correct version and parent information.
2288
1355
Note that the two entries may share the delta, and that successive
2289
1356
annotations and references MUST point to the first entry.
2290
1357
2291
1358
The index file on disc contains a header, followed by one line per knit
2292
1359
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
1360
The first occurrence gets assigned a sequence number starting from 0.
1361
2295
1362
The format of a single line is
2296
1363
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
2297
1364
REVISION_ID is a utf8-encoded revision id
2298
FLAGS is a comma separated list of flags about the record. Values include
1365
FLAGS is a comma separated list of flags about the record. Values include
2299
1366
no-eol, line-delta, fulltext.
2300
1367
BYTE_OFFSET is the ascii representation of the byte offset in the data file
2301
1368
that the the compressed data starts at.
2305
1372
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
2306
1373
revision id already in the knit that is a parent of REVISION_ID.
2307
1374
The ' :' marker is the end of record marker.
2308
1375
2309
1376
partial writes:
2310
1377
when a write is interrupted to the index file, it will result in a line
2311
1378
that does not end in ' :'. If the ' :' is not present at the end of a line,
2316
1383
to ensure that records always start on new lines even if the last write was
2317
1384
interrupted. As a result its normal for the last line in the index to be
2318
1385
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
1386
"""
2326
1387
2327
1388
HEADER = "# bzr knit index 8\n"
2328
1389
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):
1390
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1391
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1392
1393
def _cache_version(self, version_id, options, pos, size, parents):
2405
1394
"""Cache a version record in the history array and index cache.
2406
1395
2407
1396
This is inlined into _load_data for performance. KEEP IN SYNC.
2408
1397
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2409
1398
indexes).
2410
1399
"""
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
1400
# only want the _history index to reference the 1st index entry
2421
1401
# for version_id
2422
if version_id not in cache:
2423
index = len(history)
2424
history.append(version_id)
1402
if version_id not in self._cache:
1403
index = len(self._history)
1404
self._history.append(version_id)
2425
1405
else:
2426
index = cache[version_id][5]
2427
cache[version_id] = (version_id,
1406
index = self._cache[version_id][5]
1407
self._cache[version_id] = (version_id,
2428
1408
options,
2429
1409
pos,
2430
1410
size,
2431
1411
parents,
2432
1412
index)
2433
1413
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.
1414
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1415
create_parent_dir=False, delay_create=False, dir_mode=None):
1416
_KnitComponentFile.__init__(self, transport, filename, mode,
1417
file_mode=file_mode,
1418
create_parent_dir=create_parent_dir,
1419
dir_mode=dir_mode)
1420
self._cache = {}
1421
# position in _history is the 'official' index for a revision
1422
# but the values may have come from a newer entry.
1423
# so - wc -l of a knit index is != the number of unique names
1424
# in the knit.
1425
self._history = []
1426
try:
1427
fp = self._transport.get(self._filename)
1428
try:
1429
# _load_data may raise NoSuchFile if the target knit is
1430
# completely empty.
1431
_load_data(self, fp)
1432
finally:
1433
fp.close()
1434
except NoSuchFile:
1435
if mode != 'w' or not create:
1436
raise
1437
elif delay_create:
1438
self._need_to_create = True
1439
else:
1440
self._transport.put_bytes_non_atomic(
1441
self._filename, self.HEADER, mode=self._file_mode)
1442
1443
def get_graph(self):
1444
"""Return a list of the node:parents lists from this knit index."""
1445
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1446
1447
def get_ancestry(self, versions, topo_sorted=True):
1448
"""See VersionedFile.get_ancestry."""
1449
# get a graph of all the mentioned versions:
1450
graph = {}
1451
pending = set(versions)
1452
cache = self._cache
1453
while pending:
1454
version = pending.pop()
1455
# trim ghosts
1456
try:
1457
parents = [p for p in cache[version][4] if p in cache]
1458
except KeyError:
1459
raise RevisionNotPresent(version, self._filename)
1460
# if not completed and not a ghost
1461
pending.update([p for p in parents if p not in graph])
1462
graph[version] = parents
1463
if not topo_sorted:
1464
return graph.keys()
1465
return topo_sort(graph.items())
1466
1467
def get_ancestry_with_ghosts(self, versions):
1468
"""See VersionedFile.get_ancestry_with_ghosts."""
1469
# get a graph of all the mentioned versions:
1470
self.check_versions_present(versions)
1471
cache = self._cache
1472
graph = {}
1473
pending = set(versions)
1474
while pending:
1475
version = pending.pop()
1476
try:
1477
parents = cache[version][4]
1478
except KeyError:
1479
# ghost, fake it
1480
graph[version] = []
1481
else:
1482
# if not completed
1483
pending.update([p for p in parents if p not in graph])
1484
graph[version] = parents
1485
return topo_sort(graph.items())
1486
1487
def get_build_details(self, version_ids):
1488
"""Get the method, index_memo and compression parent for version_ids.
2460
1489
2461
1490
Ghosts are omitted from the result.
2462
1491
2463
:param keys: An iterable of keys.
2464
:return: A dict of key:(index_memo, compression_parent, parents,
2465
record_details).
1492
:param version_ids: An iterable of version_ids.
1493
:return: A dict of version_id:(index_memo, compression_parent,
1494
parents, record_details).
2466
1495
index_memo
2467
1496
opaque structure to pass to read_records to extract the raw
2468
1497
data
2474
1503
extra information about the content which needs to be passed to
2475
1504
Factory.parse_record
2476
1505
"""
2477
parent_map = self.get_parent_map(keys)
2478
1506
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]
1507
for version_id in version_ids:
1508
if version_id not in self._cache:
1509
# ghosts are omitted
1510
continue
1511
method = self.get_method(version_id)
1512
parents = self.get_parents_with_ghosts(version_id)
2484
1513
if method == 'fulltext':
2485
1514
compression_parent = None
2486
1515
else:
2487
1516
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,
1517
noeol = 'no-eol' in self.get_options(version_id)
1518
index_memo = self.get_position(version_id)
1519
result[version_id] = (index_memo, compression_parent,
2491
1520
parents, (method, noeol))
2492
1521
return result
2493
1522
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])
2511
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]
1523
def iter_parents(self, version_ids):
1524
"""Iterate through the parents for many version ids.
1525
1526
:param version_ids: An iterable yielding version_ids.
1527
:return: An iterator that yields (version_id, parents). Requested
1528
version_ids not present in the versioned file are simply skipped.
1529
The order is undefined, allowing for different optimisations in
1530
the underlying implementation.
1531
"""
1532
for version_id in version_ids:
2531
1533
try:
2532
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
1534
yield version_id, tuple(self.get_parents(version_id))
2533
1535
except KeyError:
2534
1536
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 ''
1537
1538
def num_versions(self):
1539
return len(self._history)
1540
1541
__len__ = num_versions
1542
1543
def get_versions(self):
1544
"""Get all the versions in the file. not topologically sorted."""
1545
return self._history
1546
1547
def _version_list_to_index(self, versions):
2638
1548
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:
1549
cache = self._cache
1550
for version in versions:
1551
if version in cache:
2646
1552
# -- inlined lookup() --
2647
result_list.append(str(cache[key[-1]][5]))
1553
result_list.append(str(cache[version][5]))
2648
1554
# -- end lookup () --
2649
1555
else:
2650
result_list.append('.' + key[-1])
1556
result_list.append('.' + version)
2651
1557
return ' '.join(result_list)
2652
1558
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'
1559
def add_version(self, version_id, options, index_memo, parents):
1560
"""Add a version record to the index."""
1561
self.add_versions(((version_id, options, index_memo, parents),))
1562
1563
def add_versions(self, versions, random_id=False):
1564
"""Add multiple versions to the index.
1565
1566
:param versions: a list of tuples:
1567
(version_id, options, pos, size, parents).
1568
:param random_id: If True the ids being added were randomly generated
1569
and no check for existence will be performed.
1570
"""
1571
lines = []
1572
orig_history = self._history[:]
1573
orig_cache = self._cache.copy()
1574
1575
try:
1576
for version_id, options, (index, pos, size), parents in versions:
1577
line = "\n%s %s %s %s %s :" % (version_id,
1578
','.join(options),
1579
pos,
1580
size,
1581
self._version_list_to_index(parents))
1582
assert isinstance(line, str), \
1583
'content must be utf-8 encoded: %r' % (line,)
1584
lines.append(line)
1585
self._cache_version(version_id, options, pos, size, parents)
1586
if not self._need_to_create:
1587
self._transport.append_bytes(self._filename, ''.join(lines))
1588
else:
1589
sio = StringIO()
1590
sio.write(self.HEADER)
1591
sio.writelines(lines)
1592
sio.seek(0)
1593
self._transport.put_file_non_atomic(self._filename, sio,
1594
create_parent_dir=self._create_parent_dir,
1595
mode=self._file_mode,
1596
dir_mode=self._dir_mode)
1597
self._need_to_create = False
1598
except:
1599
# If any problems happen, restore the original values and re-raise
1600
self._history = orig_history
1601
self._cache = orig_cache
1602
raise
1603
1604
def has_version(self, version_id):
1605
"""True if the version is in the index."""
1606
return version_id in self._cache
1607
1608
def get_position(self, version_id):
1609
"""Return details needed to access the version.
1610
1611
.kndx indices do not support split-out data, so return None for the
1612
index field.
1613
1614
:return: a tuple (None, data position, size) to hand to the access
1615
logic to get the record.
1616
"""
1617
entry = self._cache[version_id]
1618
return None, entry[2], entry[3]
1619
1620
def get_method(self, version_id):
1621
"""Return compression method of specified version."""
1622
try:
1623
options = self._cache[version_id][1]
1624
except KeyError:
1625
raise RevisionNotPresent(version_id, self._filename)
1626
if 'fulltext' in options:
1627
return 'fulltext'
2661
1628
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
1629
if 'line-delta' not in options:
1630
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1631
return 'line-delta'
1632
1633
def get_options(self, version_id):
1634
"""Return a list representing options.
1635
1636
e.g. ['foo', 'bar']
2674
1637
"""
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):
1638
return self._cache[version_id][1]
1639
1640
def get_parents(self, version_id):
1641
"""Return parents of specified version ignoring ghosts."""
1642
return [parent for parent in self._cache[version_id][4]
1643
if parent in self._cache]
1644
1645
def get_parents_with_ghosts(self, version_id):
1646
"""Return parents of specified version with ghosts."""
1647
return self._cache[version_id][4]
1648
1649
def check_versions_present(self, version_ids):
1650
"""Check that all specified versions are present."""
1651
cache = self._cache
1652
for version_id in version_ids:
1653
if version_id not in cache:
1654
raise RevisionNotPresent(version_id, self._filename)
1655
1656
1657
class KnitGraphIndex(object):
1658
"""A knit index that builds on GraphIndex."""
1659
1660
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
2696
1661
"""Construct a KnitGraphIndex on a graph_index.
2697
1662
2698
1663
: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
1664
:param deltas: Allow delta-compressed records.
2702
:param parents: If True, record knits parents, if not do not record
2703
parents.
2704
1665
:param add_callback: If not None, allow additions to the index and call
2705
1666
this callback with a list of added GraphIndex nodes:
2706
1667
[(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().
1668
:param parents: If True, record knits parents, if not do not record
1669
parents.
2712
1670
"""
2713
self._add_callback = add_callback
2714
1671
self._graph_index = graph_index
2715
1672
self._deltas = deltas
1673
self._add_callback = add_callback
2716
1674
self._parents = parents
2717
1675
if deltas and not parents:
2718
# XXX: TODO: Delta tree and parent graph should be conceptually
2719
# separate.
2720
1676
raise KnitCorrupt(self, "Cannot do delta compression without "
2721
1677
"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
1678
1679
def _get_entries(self, keys, check_present=False):
1680
"""Get the entries for keys.
1681
1682
:param keys: An iterable of index keys, - 1-tuples.
1683
"""
1684
keys = set(keys)
1685
found_keys = set()
1686
if self._parents:
1687
for node in self._graph_index.iter_entries(keys):
1688
yield node
1689
found_keys.add(node[1])
1690
else:
1691
# adapt parentless index to the rest of the code.
1692
for node in self._graph_index.iter_entries(keys):
1693
yield node[0], node[1], node[2], ()
1694
found_keys.add(node[1])
1695
if check_present:
1696
missing_keys = keys.difference(found_keys)
1697
if missing_keys:
1698
raise RevisionNotPresent(missing_keys.pop(), self)
1699
1700
def _present_keys(self, version_ids):
1701
return set([
1702
node[1] for node in self._get_entries(version_ids)])
1703
1704
def _parentless_ancestry(self, versions):
1705
"""Honour the get_ancestry API for parentless knit indices."""
1706
wanted_keys = self._version_ids_to_keys(versions)
1707
present_keys = self._present_keys(wanted_keys)
1708
missing = set(wanted_keys).difference(present_keys)
1709
if missing:
1710
raise RevisionNotPresent(missing.pop(), self)
1711
return list(self._keys_to_version_ids(present_keys))
1712
1713
def get_ancestry(self, versions, topo_sorted=True):
1714
"""See VersionedFile.get_ancestry."""
1715
if not self._parents:
1716
return self._parentless_ancestry(versions)
1717
# XXX: This will do len(history) index calls - perhaps
1718
# it should be altered to be a index core feature?
1719
# get a graph of all the mentioned versions:
1720
graph = {}
1721
ghosts = set()
1722
versions = self._version_ids_to_keys(versions)
1723
pending = set(versions)
1724
while pending:
1725
# get all pending nodes
1726
this_iteration = pending
1727
new_nodes = self._get_entries(this_iteration)
1728
found = set()
1729
pending = set()
1730
for (index, key, value, node_refs) in new_nodes:
1731
# dont ask for ghosties - otherwise
1732
# we we can end up looping with pending
1733
# being entirely ghosted.
1734
graph[key] = [parent for parent in node_refs[0]
1735
if parent not in ghosts]
1736
# queue parents
1737
for parent in graph[key]:
1738
# dont examine known nodes again
1739
if parent in graph:
1740
continue
1741
pending.add(parent)
1742
found.add(key)
1743
ghosts.update(this_iteration.difference(found))
1744
if versions.difference(graph):
1745
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1746
if topo_sorted:
1747
result_keys = topo_sort(graph.items())
1748
else:
1749
result_keys = graph.iterkeys()
1750
return [key[0] for key in result_keys]
1751
1752
def get_ancestry_with_ghosts(self, versions):
1753
"""See VersionedFile.get_ancestry."""
1754
if not self._parents:
1755
return self._parentless_ancestry(versions)
1756
# XXX: This will do len(history) index calls - perhaps
1757
# it should be altered to be a index core feature?
1758
# get a graph of all the mentioned versions:
1759
graph = {}
1760
versions = self._version_ids_to_keys(versions)
1761
pending = set(versions)
1762
while pending:
1763
# get all pending nodes
1764
this_iteration = pending
1765
new_nodes = self._get_entries(this_iteration)
1766
pending = set()
1767
for (index, key, value, node_refs) in new_nodes:
1768
graph[key] = node_refs[0]
1769
# queue parents
1770
for parent in graph[key]:
1771
# dont examine known nodes again
1772
if parent in graph:
1773
continue
1774
pending.add(parent)
1775
missing_versions = this_iteration.difference(graph)
1776
missing_needed = versions.intersection(missing_versions)
1777
if missing_needed:
1778
raise RevisionNotPresent(missing_needed.pop(), self)
1779
for missing_version in missing_versions:
1780
# add a key, no parents
1781
graph[missing_version] = []
1782
pending.discard(missing_version) # don't look for it
1783
result_keys = topo_sort(graph.items())
1784
return [key[0] for key in result_keys]
1785
1786
def get_build_details(self, version_ids):
1787
"""Get the method, index_memo and compression parent for version_ids.
1788
1789
Ghosts are omitted from the result.
1790
1791
:param version_ids: An iterable of version_ids.
1792
:return: A dict of version_id:(index_memo, compression_parent,
1793
parents, record_details).
1794
index_memo
1795
opaque structure to pass to read_records to extract the raw
1796
data
1797
compression_parent
1798
Content that this record is built upon, may be None
1799
parents
1800
Logical parents of this node
1801
record_details
1802
extra information about the content which needs to be passed to
1803
Factory.parse_record
1804
"""
1805
result = {}
1806
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
1807
for entry in entries:
1808
version_id = self._keys_to_version_ids((entry[1],))[0]
1809
if not self._parents:
1810
parents = ()
1811
else:
1812
parents = self._keys_to_version_ids(entry[3][0])
1813
if not self._deltas:
1814
compression_parent = None
1815
else:
1816
compression_parent_key = self._compression_parent(entry)
1817
if compression_parent_key:
1818
compression_parent = self._keys_to_version_ids(
1819
(compression_parent_key,))[0]
1820
else:
1821
compression_parent = None
1822
noeol = (entry[2][0] == 'N')
1823
if compression_parent:
1824
method = 'line-delta'
1825
else:
1826
method = 'fulltext'
1827
result[version_id] = (self._node_to_position(entry),
1828
compression_parent, parents,
1829
(method, noeol))
1830
return result
1831
1832
def _compression_parent(self, an_entry):
1833
# return the key that an_entry is compressed against, or None
1834
# Grab the second parent list (as deltas implies parents currently)
1835
compression_parents = an_entry[3][1]
1836
if not compression_parents:
1837
return None
1838
assert len(compression_parents) == 1
1839
return compression_parents[0]
1840
1841
def _get_method(self, node):
1842
if not self._deltas:
1843
return 'fulltext'
1844
if self._compression_parent(node):
1845
return 'line-delta'
1846
else:
1847
return 'fulltext'
1848
1849
def get_graph(self):
1850
"""Return a list of the node:parents lists from this knit index."""
1851
if not self._parents:
1852
return [(key, ()) for key in self.get_versions()]
1853
result = []
1854
for index, key, value, refs in self._graph_index.iter_all_entries():
1855
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1856
return result
1857
1858
def iter_parents(self, version_ids):
1859
"""Iterate through the parents for many version ids.
1860
1861
:param version_ids: An iterable yielding version_ids.
1862
:return: An iterator that yields (version_id, parents). Requested
1863
version_ids not present in the versioned file are simply skipped.
1864
The order is undefined, allowing for different optimisations in
1865
the underlying implementation.
1866
"""
1867
if self._parents:
1868
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1869
all_parents = set()
1870
present_parents = set()
1871
for node in all_nodes:
1872
all_parents.update(node[3][0])
1873
# any node we are querying must be present
1874
present_parents.add(node[1])
1875
unknown_parents = all_parents.difference(present_parents)
1876
present_parents.update(self._present_keys(unknown_parents))
1877
for node in all_nodes:
1878
parents = []
1879
for parent in node[3][0]:
1880
if parent in present_parents:
1881
parents.append(parent[0])
1882
yield node[1][0], tuple(parents)
1883
else:
1884
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1885
yield node[1][0], ()
1886
1887
def num_versions(self):
1888
return len(list(self._graph_index.iter_all_entries()))
1889
1890
__len__ = num_versions
1891
1892
def get_versions(self):
1893
"""Get all the versions in the file. not topologically sorted."""
1894
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1895
1896
def has_version(self, version_id):
1897
"""True if the version is in the index."""
1898
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1899
1900
def _keys_to_version_ids(self, keys):
1901
return tuple(key[0] for key in keys)
1902
1903
def get_position(self, version_id):
1904
"""Return details needed to access the version.
1905
1906
:return: a tuple (index, data position, size) to hand to the access
1907
logic to get the record.
1908
"""
1909
node = self._get_node(version_id)
1910
return self._node_to_position(node)
1911
1912
def _node_to_position(self, node):
1913
"""Convert an index value to position details."""
1914
bits = node[2][1:].split(' ')
1915
return node[0], int(bits[0]), int(bits[1])
1916
1917
def get_method(self, version_id):
1918
"""Return compression method of specified version."""
1919
return self._get_method(self._get_node(version_id))
1920
1921
def _get_node(self, version_id):
1922
try:
1923
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1924
except IndexError:
1925
raise RevisionNotPresent(version_id, self)
1926
1927
def get_options(self, version_id):
1928
"""Return a list representing options.
1929
1930
e.g. ['foo', 'bar']
1931
"""
1932
node = self._get_node(version_id)
1933
options = [self._get_method(node)]
1934
if node[2][0] == 'N':
1935
options.append('no-eol')
1936
return options
1937
1938
def get_parents(self, version_id):
1939
"""Return parents of specified version ignoring ghosts."""
1940
parents = list(self.iter_parents([version_id]))
1941
if not parents:
1942
# missing key
1943
raise errors.RevisionNotPresent(version_id, self)
1944
return parents[0][1]
1945
1946
def get_parents_with_ghosts(self, version_id):
1947
"""Return parents of specified version with ghosts."""
1948
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1949
check_present=True))
1950
if not self._parents:
1951
return ()
1952
return self._keys_to_version_ids(nodes[0][3][0])
1953
1954
def check_versions_present(self, version_ids):
1955
"""Check that all specified versions are present."""
1956
keys = self._version_ids_to_keys(version_ids)
1957
present = self._present_keys(keys)
1958
missing = keys.difference(present)
1959
if missing:
1960
raise RevisionNotPresent(missing.pop(), self)
1961
1962
def add_version(self, version_id, options, access_memo, parents):
1963
"""Add a version record to the index."""
1964
return self.add_versions(((version_id, options, access_memo, parents),))
1965
1966
def add_versions(self, versions, random_id=False):
1967
"""Add multiple versions to the index.
1968
2737
1969
This function does not insert data into the Immutable GraphIndex
2738
1970
backing the KnitGraphIndex, instead it prepares data for insertion by
2739
1971
the caller and checks that it is safe to insert then calls
2740
1972
self._add_callback with the prepared GraphIndex nodes.
2741
1973
2742
:param records: a list of tuples:
2743
(key, options, access_memo, parents).
1974
:param versions: a list of tuples:
1975
(version_id, options, pos, size, parents).
2744
1976
:param random_id: If True the ids being added were randomly generated
2745
1977
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
1978
"""
2751
1979
if not self._add_callback:
2752
1980
raise errors.ReadOnlyError(self)
2753
1981
# we hope there are no repositories with inconsistent parentage
2754
1982
# anymore.
1983
# check for dups
2755
1984
2756
1985
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)
1986
for (version_id, options, access_memo, parents) in versions:
2765
1987
index, pos, size = access_memo
1988
key = (version_id, )
1989
parents = tuple((parent, ) for parent in parents)
2766
1990
if 'no-eol' in options:
2767
1991
value = 'N'
2768
1992
else:
2775
1999
if self._deltas:
2776
2000
if 'line-delta' in options:
2777
2001
node_refs = (parents, (parents[0],))
2778
if missing_compression_parents:
2779
compression_parents.add(parents[0])
2780
2002
else:
2781
2003
node_refs = (parents, ())
2782
2004
else:
2787
2009
"in parentless index.")
2788
2010
node_refs = ()
2789
2011
keys[key] = (value, node_refs)
2790
# check for dups
2791
2012
if not random_id:
2792
2013
present_nodes = self._get_entries(keys)
2793
2014
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"
2015
if (value, node_refs) != keys[key]:
2016
raise KnitCorrupt(self, "inconsistent details in add_versions"
2797
2017
": %s %s" % ((value, node_refs), keys[key]))
2798
2018
del keys[key]
2799
2019
result = []
2804
2024
for key, (value, node_refs) in keys.iteritems():
2805
2025
result.append((key, value))
2806
2026
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.
2027
2028
def _version_ids_to_keys(self, version_ids):
2029
return set((version_id, ) for version_id in version_ids)
2030
2031
2032
class _KnitAccess(object):
2033
"""Access to knit records in a .knit file."""
2034
2035
def __init__(self, transport, filename, _file_mode, _dir_mode,
2036
_need_to_create, _create_parent_dir):
2037
"""Create a _KnitAccess for accessing and inserting data.
2038
2039
:param transport: The transport the .knit is located on.
2040
:param filename: The filename of the .knit.
3042
2041
"""
3043
2042
self._transport = transport
3044
self._mapper = mapper
2043
self._filename = filename
2044
self._file_mode = _file_mode
2045
self._dir_mode = _dir_mode
2046
self._need_to_create = _need_to_create
2047
self._create_parent_dir = _create_parent_dir
3045
2048
3046
def add_raw_records(self, key_sizes, raw_data):
2049
def add_raw_records(self, sizes, raw_data):
3047
2050
"""Add raw knit bytes to a storage area.
3048
2051
3049
The data is spooled to the container writer in one bytes-record per
3050
raw data item.
2052
The data is spooled to whereever the access method is storing data.
3051
2053
3052
:param sizes: An iterable of tuples containing the key and size of each
3053
raw data segment.
2054
:param sizes: An iterable containing the size of each raw data segment.
3054
2055
: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.
2056
:return: A list of memos to retrieve the record later. Each memo is a
2057
tuple - (index, pos, length), where the index field is always None
2058
for the .knit access method.
3058
2059
"""
3059
if type(raw_data) != str:
3060
raise AssertionError(
3061
'data must be plain bytes was %s' % type(raw_data))
2060
assert type(raw_data) == str, \
2061
'data must be plain bytes was %s' % type(raw_data)
2062
if not self._need_to_create:
2063
base = self._transport.append_bytes(self._filename, raw_data)
2064
else:
2065
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2066
create_parent_dir=self._create_parent_dir,
2067
mode=self._file_mode,
2068
dir_mode=self._dir_mode)
2069
self._need_to_create = False
2070
base = 0
3062
2071
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))
2072
for size in sizes:
2073
result.append((None, base, size))
2074
base += size
3080
2075
return result
3081
2076
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
2077
def create(self):
2078
"""IFF this data access has its own storage area, initialise it.
2079
2080
:return: None.
2081
"""
2082
self._transport.put_bytes_non_atomic(self._filename, '',
2083
mode=self._file_mode)
2084
2085
def open_file(self):
2086
"""IFF this data access can be represented as a single file, open it.
2087
2088
For knits that are not mapped to a single file on disk this will
2089
always return None.
2090
2091
:return: None or a file handle.
2092
"""
2093
try:
2094
return self._transport.get(self._filename)
2095
except NoSuchFile:
2096
pass
2097
return None
3088
2098
3089
2099
def get_raw_records(self, memos_for_retrieval):
3090
2100
"""Get the raw bytes for a records.
3091
2101
3092
:param memos_for_retrieval: An iterable containing the access memo for
3093
retrieving the bytes.
2102
:param memos_for_retrieval: An iterable containing the (index, pos,
2103
length) memo for retrieving the bytes. The .knit method ignores
2104
the index as there is always only a single file.
3094
2105
:return: An iterator over the bytes of the records.
3095
2106
"""
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.
2107
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2108
for pos, data in self._transport.readv(self._filename, read_vector):
2109
yield data
2110
2111
2112
class _PackAccess(object):
2113
"""Access to knit records via a collection of packs."""
2114
2115
def __init__(self, index_to_packs, writer=None):
2116
"""Create a _PackAccess object.
3121
2117
3122
2118
:param index_to_packs: A dict mapping index objects to the transport
3123
2119
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.
2120
:param writer: A tuple (pack.ContainerWriter, write_index) which
2121
contains the pack to write, and the index that reads from it will
2122
be associated with.
3127
2123
"""
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
2124
if writer:
2125
self.container_writer = writer[0]
2126
self.write_index = writer[1]
2127
else:
2128
self.container_writer = None
2129
self.write_index = None
2130
self.indices = index_to_packs
3133
2131
3134
def add_raw_records(self, key_sizes, raw_data):
2132
def add_raw_records(self, sizes, raw_data):
3135
2133
"""Add raw knit bytes to a storage area.
3136
2134
3137
2135
The data is spooled to the container writer in one bytes-record per
3138
2136
raw data item.
3139
2137
3140
:param sizes: An iterable of tuples containing the key and size of each
3141
raw data segment.
2138
:param sizes: An iterable containing the size of each raw data segment.
3142
2139
: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.
2140
:return: A list of memos to retrieve the record later. Each memo is a
2141
tuple - (index, pos, length), where the index field is the
2142
write_index object supplied to the PackAccess object.
3147
2143
"""
3148
if type(raw_data) != str:
3149
raise AssertionError(
3150
'data must be plain bytes was %s' % type(raw_data))
2144
assert type(raw_data) == str, \
2145
'data must be plain bytes was %s' % type(raw_data)
3151
2146
result = []
3152
2147
offset = 0
3153
for key, size in key_sizes:
3154
p_offset, p_length = self._container_writer.add_bytes_record(
2148
for size in sizes:
2149
p_offset, p_length = self.container_writer.add_bytes_record(
3155
2150
raw_data[offset:offset+size], [])
3156
2151
offset += size
3157
result.append((self._write_index, p_offset, p_length))
2152
result.append((self.write_index, p_offset, p_length))
3158
2153
return result
3159
2154
3160
def flush(self):
3161
"""Flush pending writes on this access object.
2155
def create(self):
2156
"""Pack based knits do not get individually created."""
3162
2157
3163
This will flush any buffered writes to a NewPack.
3164
"""
3165
if self._flush_func is not None:
3166
self._flush_func()
3167
3168
2158
def get_raw_records(self, memos_for_retrieval):
3169
2159
"""Get the raw bytes for a records.
3170
2160
3171
:param memos_for_retrieval: An iterable containing the (index, pos,
2161
:param memos_for_retrieval: An iterable containing the (index, pos,
3172
2162
length) memo for retrieving the bytes. The Pack access method
3173
2163
looks up the pack to use for a given record in its index_to_pack
3174
2164
map.
3189
2179
if current_index is not None:
3190
2180
request_lists.append((current_index, current_list))
3191
2181
for index, offsets in request_lists:
3192
try:
3193
transport, path = self._indices[index]
3194
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):
2182
transport, path = self.indices[index]
2183
reader = pack.make_readv_reader(transport, path, offsets)
2184
for names, read_func in reader.iter_records():
2185
yield read_func(None)
2186
2187
def open_file(self):
2188
"""Pack based knits have no single file."""
2189
return None
2190
2191
def set_writer(self, writer, index, (transport, packname)):
3219
2192
"""Set a writer to use for adding data."""
3220
2193
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
2194
self.indices[index] = (transport, packname)
2195
self.container_writer = writer
2196
self.write_index = index
2197
2198
2199
class _StreamAccess(object):
2200
"""A Knit Access object that provides data from a datastream.
2201
2202
It also provides a fallback to present as unannotated data, annotated data
2203
from a *backing* access object.
2204
2205
This is triggered by a index_memo which is pointing to a different index
2206
than this was constructed with, and is used to allow extracting full
2207
unannotated texts for insertion into annotated knits.
2208
"""
2209
2210
def __init__(self, reader_callable, stream_index, backing_knit,
2211
orig_factory):
2212
"""Create a _StreamAccess object.
2213
2214
:param reader_callable: The reader_callable from the datastream.
2215
This is called to buffer all the data immediately, for
2216
random access.
2217
:param stream_index: The index the data stream this provides access to
2218
which will be present in native index_memo's.
2219
:param backing_knit: The knit object that will provide access to
2220
annotated texts which are not available in the stream, so as to
2221
create unannotated texts.
2222
:param orig_factory: The original content factory used to generate the
2223
stream. This is used for checking whether the thunk code for
2224
supporting _copy_texts will generate the correct form of data.
2225
"""
2226
self.data = reader_callable(None)
2227
self.stream_index = stream_index
2228
self.backing_knit = backing_knit
2229
self.orig_factory = orig_factory
2230
2231
def get_raw_records(self, memos_for_retrieval):
2232
"""Get the raw bytes for a records.
2233
2234
:param memos_for_retrieval: An iterable containing the (thunk_flag,
2235
index, start, end) memo for retrieving the bytes.
2236
:return: An iterator over the bytes of the records.
2237
"""
2238
# use a generator for memory friendliness
2239
for thunk_flag, version_id, start, end in memos_for_retrieval:
2240
if version_id is self.stream_index:
2241
yield self.data[start:end]
2242
continue
2243
# we have been asked to thunk. This thunking only occurs when
2244
# we are obtaining plain texts from an annotated backing knit
2245
# so that _copy_texts will work.
2246
# We could improve performance here by scanning for where we need
2247
# to do this and using get_line_list, then interleaving the output
2248
# as desired. However, for now, this is sufficient.
2249
if self.orig_factory.__class__ != KnitPlainFactory:
2250
raise errors.KnitCorrupt(
2251
self, 'Bad thunk request %r' % version_id)
2252
lines = self.backing_knit.get_lines(version_id)
2253
line_bytes = ''.join(lines)
2254
digest = sha_string(line_bytes)
2255
if lines:
2256
if lines[-1][-1] != '\n':
2257
lines[-1] = lines[-1] + '\n'
2258
line_bytes += '\n'
2259
orig_options = list(self.backing_knit._index.get_options(version_id))
2260
if 'fulltext' not in orig_options:
2261
if 'line-delta' not in orig_options:
2262
raise errors.KnitCorrupt(self,
2263
'Unknown compression method %r' % orig_options)
2264
orig_options.remove('line-delta')
2265
orig_options.append('fulltext')
2266
# We want plain data, because we expect to thunk only to allow text
2267
# extraction.
2268
size, bytes = self.backing_knit._data._record_to_data(version_id,
2269
digest, lines, line_bytes)
2270
yield bytes
2271
2272
2273
class _StreamIndex(object):
2274
"""A Knit Index object that uses the data map from a datastream."""
2275
2276
def __init__(self, data_list, backing_index):
2277
"""Create a _StreamIndex object.
2278
2279
:param data_list: The data_list from the datastream.
2280
:param backing_index: The index which will supply values for nodes
2281
referenced outside of this stream.
2282
"""
2283
self.data_list = data_list
2284
self.backing_index = backing_index
2285
self._by_version = {}
2286
pos = 0
2287
for key, options, length, parents in data_list:
2288
self._by_version[key] = options, (pos, pos + length), parents
2289
pos += length
2290
2291
def get_ancestry(self, versions, topo_sorted):
2292
"""Get an ancestry list for versions."""
2293
if topo_sorted:
2294
# Not needed for basic joins
2295
raise NotImplementedError(self.get_ancestry)
2296
# get a graph of all the mentioned versions:
2297
# Little ugly - basically copied from KnitIndex, but don't want to
2298
# accidentally incorporate too much of that index's code.
2299
ancestry = set()
2300
pending = set(versions)
2301
cache = self._by_version
2302
while pending:
2303
version = pending.pop()
2304
# trim ghosts
2305
try:
2306
parents = [p for p in cache[version][2] if p in cache]
2307
except KeyError:
2308
raise RevisionNotPresent(version, self)
2309
# if not completed and not a ghost
2310
pending.update([p for p in parents if p not in ancestry])
2311
ancestry.add(version)
2312
return list(ancestry)
2313
2314
def get_build_details(self, version_ids):
2315
"""Get the method, index_memo and compression parent for version_ids.
2316
2317
Ghosts are omitted from the result.
2318
2319
:param version_ids: An iterable of version_ids.
2320
:return: A dict of version_id:(index_memo, compression_parent,
2321
parents, record_details).
2322
index_memo
2323
opaque structure to pass to read_records to extract the raw
2324
data
2325
compression_parent
2326
Content that this record is built upon, may be None
2327
parents
2328
Logical parents of this node
2329
record_details
2330
extra information about the content which needs to be passed to
2331
Factory.parse_record
2332
"""
2333
result = {}
2334
for version_id in version_ids:
2335
try:
2336
method = self.get_method(version_id)
2337
except errors.RevisionNotPresent:
2338
# ghosts are omitted
2339
continue
2340
parent_ids = self.get_parents_with_ghosts(version_id)
2341
noeol = ('no-eol' in self.get_options(version_id))
2342
if method == 'fulltext':
2343
compression_parent = None
2344
else:
2345
compression_parent = parent_ids[0]
2346
index_memo = self.get_position(version_id)
2347
result[version_id] = (index_memo, compression_parent,
2348
parent_ids, (method, noeol))
2349
return result
2350
2351
def get_method(self, version_id):
2352
"""Return compression method of specified version."""
2353
try:
2354
options = self._by_version[version_id][0]
2355
except KeyError:
2356
# Strictly speaking this should check in the backing knit, but
2357
# until we have a test to discriminate, this will do.
2358
return self.backing_index.get_method(version_id)
2359
if 'fulltext' in options:
2360
return 'fulltext'
2361
elif 'line-delta' in options:
2362
return 'line-delta'
2363
else:
2364
raise errors.KnitIndexUnknownMethod(self, options)
2365
2366
def get_options(self, version_id):
2367
"""Return a list representing options.
2368
2369
e.g. ['foo', 'bar']
2370
"""
2371
try:
2372
return self._by_version[version_id][0]
2373
except KeyError:
2374
return self.backing_index.get_options(version_id)
2375
2376
def get_parents_with_ghosts(self, version_id):
2377
"""Return parents of specified version with ghosts."""
2378
try:
2379
return self._by_version[version_id][2]
2380
except KeyError:
2381
return self.backing_index.get_parents_with_ghosts(version_id)
2382
2383
def get_position(self, version_id):
2384
"""Return details needed to access the version.
2385
2386
_StreamAccess has the data as a big array, so we return slice
2387
coordinates into that (as index_memo's are opaque outside the
2388
index and matching access class).
2389
2390
:return: a tuple (thunk_flag, index, start, end). If thunk_flag is
2391
False, index will be self, otherwise it will be a version id.
2392
"""
2393
try:
2394
start, end = self._by_version[version_id][1]
2395
return False, self, start, end
2396
except KeyError:
2397
# Signal to the access object to handle this from the backing knit.
2398
return (True, version_id, None, None)
2399
2400
def get_versions(self):
2401
"""Get all the versions in the stream."""
2402
return self._by_version.keys()
2403
2404
def iter_parents(self, version_ids):
2405
"""Iterate through the parents for many version ids.
2406
2407
:param version_ids: An iterable yielding version_ids.
2408
:return: An iterator that yields (version_id, parents). Requested
2409
version_ids not present in the versioned file are simply skipped.
2410
The order is undefined, allowing for different optimisations in
2411
the underlying implementation.
2412
"""
2413
result = []
2414
for version in version_ids:
2415
try:
2416
result.append((version, self._by_version[version][2]))
2417
except KeyError:
2418
pass
2419
return result
2420
2421
2422
class _KnitData(object):
2423
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2424
2425
The KnitData class provides the logic for parsing and using knit records,
2426
making use of an access method for the low level read and write operations.
2427
"""
2428
2429
def __init__(self, access):
2430
"""Create a KnitData object.
2431
2432
:param access: The access method to use. Access methods such as
2433
_KnitAccess manage the insertion of raw records and the subsequent
2434
retrieval of the same.
2435
"""
2436
self._access = access
2437
self._checked = False
2438
# TODO: jam 20060713 conceptually, this could spill to disk
2439
# if the cached size gets larger than a certain amount
2440
# but it complicates the model a bit, so for now just use
2441
# a simple dictionary
2442
self._cache = {}
2443
self._do_cache = False
2444
2445
def enable_cache(self):
2446
"""Enable caching of reads."""
2447
self._do_cache = True
2448
2449
def clear_cache(self):
2450
"""Clear the record cache."""
2451
self._do_cache = False
2452
self._cache = {}
2453
2454
def _open_file(self):
2455
return self._access.open_file()
2456
2457
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2458
"""Convert version_id, digest, lines into a raw data block.
2459
2460
:param dense_lines: The bytes of lines but in a denser form. For
2461
instance, if lines is a list of 1000 bytestrings each ending in \n,
2462
dense_lines may be a list with one line in it, containing all the
2463
1000's lines and their \n's. Using dense_lines if it is already
2464
known is a win because the string join to create bytes in this
2465
function spends less time resizing the final string.
2466
:return: (len, a StringIO instance with the raw data ready to read.)
2467
"""
2468
# Note: using a string copy here increases memory pressure with e.g.
2469
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2470
# when doing the initial commit of a mozilla tree. RBC 20070921
2471
bytes = ''.join(chain(
2472
["version %s %d %s\n" % (version_id,
2473
len(lines),
2474
digest)],
2475
dense_lines or lines,
2476
["end %s\n" % version_id]))
2477
assert bytes.__class__ == str
2478
compressed_bytes = bytes_to_gzip(bytes)
2479
return len(compressed_bytes), compressed_bytes
2480
2481
def add_raw_records(self, sizes, raw_data):
2482
"""Append a prepared record to the data file.
2483
2484
:param sizes: An iterable containing the size of each raw data segment.
2485
:param raw_data: A bytestring containing the data.
2486
:return: a list of index data for the way the data was stored.
2487
See the access method add_raw_records documentation for more
2488
details.
2489
"""
2490
return self._access.add_raw_records(sizes, raw_data)
2491
2492
def _parse_record_header(self, version_id, raw_data):
2493
"""Parse a record header for consistency.
2494
2495
:return: the header and the decompressor stream.
2496
as (stream, header_record)
2497
"""
2498
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2499
try:
2500
rec = self._check_header(version_id, df.readline())
2501
except Exception, e:
2502
raise KnitCorrupt(self._access,
2503
"While reading {%s} got %s(%s)"
2504
% (version_id, e.__class__.__name__, str(e)))
2505
return df, rec
2506
2507
def _check_header(self, version_id, line):
2508
rec = line.split()
2509
if len(rec) != 4:
2510
raise KnitCorrupt(self._access,
2511
'unexpected number of elements in record header')
2512
if rec[1] != version_id:
2513
raise KnitCorrupt(self._access,
2514
'unexpected version, wanted %r, got %r'
2515
% (version_id, rec[1]))
2516
return rec
2517
2518
def _parse_record(self, version_id, data):
2519
# profiling notes:
2520
# 4168 calls in 2880 217 internal
2521
# 4168 calls to _parse_record_header in 2121
2522
# 4168 calls to readlines in 330
2523
df = GzipFile(mode='rb', fileobj=StringIO(data))
2524
2525
try:
2526
record_contents = df.readlines()
2527
except Exception, e:
2528
raise KnitCorrupt(self._access,
2529
"While reading {%s} got %s(%s)"
2530
% (version_id, e.__class__.__name__, str(e)))
2531
header = record_contents.pop(0)
2532
rec = self._check_header(version_id, header)
2533
2534
last_line = record_contents.pop()
2535
if len(record_contents) != int(rec[2]):
2536
raise KnitCorrupt(self._access,
2537
'incorrect number of lines %s != %s'
2538
' for version {%s}'
2539
% (len(record_contents), int(rec[2]),
2540
version_id))
2541
if last_line != 'end %s\n' % rec[1]:
2542
raise KnitCorrupt(self._access,
2543
'unexpected version end line %r, wanted %r'
2544
% (last_line, version_id))
2545
df.close()
2546
return record_contents, rec[3]
2547
2548
def read_records_iter_raw(self, records):
2549
"""Read text records from data file and yield raw data.
2550
2551
This unpacks enough of the text record to validate the id is
2552
as expected but thats all.
2553
"""
2554
# setup an iterator of the external records:
2555
# uses readv so nice and fast we hope.
2556
if len(records):
2557
# grab the disk data needed.
2558
if self._cache:
2559
# Don't check _cache if it is empty
2560
needed_offsets = [index_memo for version_id, index_memo
2561
in records
2562
if version_id not in self._cache]
2563
else:
2564
needed_offsets = [index_memo for version_id, index_memo
2565
in records]
2566
2567
raw_records = self._access.get_raw_records(needed_offsets)
2568
2569
for version_id, index_memo in records:
2570
if version_id in self._cache:
2571
# This data has already been validated
2572
data = self._cache[version_id]
2573
else:
2574
data = raw_records.next()
2575
if self._do_cache:
2576
self._cache[version_id] = data
2577
2578
# validate the header
2579
df, rec = self._parse_record_header(version_id, data)
2580
df.close()
2581
yield version_id, data
2582
2583
def read_records_iter(self, records):
2584
"""Read text records from data file and yield result.
2585
2586
The result will be returned in whatever is the fastest to read.
2587
Not by the order requested. Also, multiple requests for the same
2588
record will only yield 1 response.
2589
:param records: A list of (version_id, pos, len) entries
2590
:return: Yields (version_id, contents, digest) in the order
2591
read, not the order requested
2592
"""
2593
if not records:
2594
return
2595
2596
if self._cache:
2597
# Skip records we have alread seen
2598
yielded_records = set()
2599
needed_records = set()
2600
for record in records:
2601
if record[0] in self._cache:
2602
if record[0] in yielded_records:
2603
continue
2604
yielded_records.add(record[0])
2605
data = self._cache[record[0]]
2606
content, digest = self._parse_record(record[0], data)
2607
yield (record[0], content, digest)
2608
else:
2609
needed_records.add(record)
2610
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2611
else:
2612
needed_records = sorted(set(records), key=operator.itemgetter(1))
2613
2614
if not needed_records:
2615
return
2616
2617
# The transport optimizes the fetching as well
2618
# (ie, reads continuous ranges.)
2619
raw_data = self._access.get_raw_records(
2620
[index_memo for version_id, index_memo in needed_records])
2621
2622
for (version_id, index_memo), data in \
2623
izip(iter(needed_records), raw_data):
2624
content, digest = self._parse_record(version_id, data)
2625
if self._do_cache:
2626
self._cache[version_id] = data
2627
yield version_id, content, digest
2628
2629
def read_records(self, records):
2630
"""Read records into a dictionary."""
2631
components = {}
2632
for record_id, content, digest in \
2633
self.read_records_iter(records):
2634
components[record_id] = (content, digest)
2635
return components
2636
2637
2638
class InterKnit(InterVersionedFile):
2639
"""Optimised code paths for knit to knit operations."""
2640
2641
_matching_file_from_factory = KnitVersionedFile
2642
_matching_file_to_factory = KnitVersionedFile
2643
2644
@staticmethod
2645
def is_compatible(source, target):
2646
"""Be compatible with knits. """
2647
try:
2648
return (isinstance(source, KnitVersionedFile) and
2649
isinstance(target, KnitVersionedFile))
2650
except AttributeError:
2651
return False
2652
2653
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2654
"""Copy texts to the target by extracting and adding them one by one.
2655
2656
see join() for the parameter definitions.
2657
"""
2658
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2659
graph = self.source.get_graph(version_ids)
2660
order = topo_sort(graph.items())
2661
2662
def size_of_content(content):
2663
return sum(len(line) for line in content.text())
2664
# Cache at most 10MB of parent texts
2665
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2666
compute_size=size_of_content)
2667
# TODO: jam 20071116 It would be nice to have a streaming interface to
2668
# get multiple texts from a source. The source could be smarter
2669
# about how it handled intermediate stages.
2670
# get_line_list() or make_mpdiffs() seem like a possibility, but
2671
# at the moment they extract all full texts into memory, which
2672
# causes us to store more than our 3x fulltext goal.
2673
# Repository.iter_files_bytes() may be another possibility
2674
to_process = [version for version in order
2675
if version not in self.target]
2676
total = len(to_process)
2677
pb = ui.ui_factory.nested_progress_bar()
2678
try:
2679
for index, version in enumerate(to_process):
2680
pb.update('Converting versioned data', index, total)
2681
sha1, num_bytes, parent_text = self.target.add_lines(version,
2682
self.source.get_parents_with_ghosts(version),
2683
self.source.get_lines(version),
2684
parent_texts=parent_cache)
2685
parent_cache[version] = parent_text
2686
finally:
2687
pb.finished()
2688
return total
2689
2690
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2691
"""See InterVersionedFile.join."""
2692
assert isinstance(self.source, KnitVersionedFile)
2693
assert isinstance(self.target, KnitVersionedFile)
2694
2695
# If the source and target are mismatched w.r.t. annotations vs
2696
# plain, the data needs to be converted accordingly
2697
if self.source.factory.annotated == self.target.factory.annotated:
2698
converter = None
2699
elif self.source.factory.annotated:
2700
converter = self._anno_to_plain_converter
2701
else:
2702
# We're converting from a plain to an annotated knit. Copy them
2703
# across by full texts.
2704
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2705
2706
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2707
if not version_ids:
2708
return 0
2709
2710
pb = ui.ui_factory.nested_progress_bar()
2711
try:
2712
version_ids = list(version_ids)
2713
if None in version_ids:
2714
version_ids.remove(None)
2715
2716
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2717
topo_sorted=False))
2718
this_versions = set(self.target._index.get_versions())
2719
# XXX: For efficiency we should not look at the whole index,
2720
# we only need to consider the referenced revisions - they
2721
# must all be present, or the method must be full-text.
2722
# TODO, RBC 20070919
2723
needed_versions = self.source_ancestry - this_versions
2724
2725
if not needed_versions:
2726
return 0
2727
full_list = topo_sort(self.source.get_graph())
2728
2729
version_list = [i for i in full_list if (not self.target.has_version(i)
2730
and i in needed_versions)]
2731
2732
# plan the join:
2733
copy_queue = []
2734
copy_queue_records = []
2735
copy_set = set()
2736
for version_id in version_list:
2737
options = self.source._index.get_options(version_id)
2738
parents = self.source._index.get_parents_with_ghosts(version_id)
2739
# check that its will be a consistent copy:
2740
for parent in parents:
2741
# if source has the parent, we must :
2742
# * already have it or
2743
# * have it scheduled already
2744
# otherwise we don't care
2745
assert (self.target.has_version(parent) or
2746
parent in copy_set or
2747
not self.source.has_version(parent))
2748
index_memo = self.source._index.get_position(version_id)
2749
copy_queue_records.append((version_id, index_memo))
2750
copy_queue.append((version_id, options, parents))
2751
copy_set.add(version_id)
2752
2753
# data suck the join:
2754
count = 0
2755
total = len(version_list)
2756
raw_datum = []
2757
raw_records = []
2758
for (version_id, raw_data), \
2759
(version_id2, options, parents) in \
2760
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2761
copy_queue):
2762
assert version_id == version_id2, 'logic error, inconsistent results'
2763
count = count + 1
2764
pb.update("Joining knit", count, total)
2765
if converter:
2766
size, raw_data = converter(raw_data, version_id, options,
2767
parents)
2768
else:
2769
size = len(raw_data)
2770
raw_records.append((version_id, options, parents, size))
2771
raw_datum.append(raw_data)
2772
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2773
return count
2774
finally:
2775
pb.finished()
2776
2777
def _anno_to_plain_converter(self, raw_data, version_id, options,
2778
parents):
2779
"""Convert annotated content to plain content."""
2780
data, digest = self.source._data._parse_record(version_id, raw_data)
2781
if 'fulltext' in options:
2782
content = self.source.factory.parse_fulltext(data, version_id)
2783
lines = self.target.factory.lower_fulltext(content)
2784
else:
2785
delta = self.source.factory.parse_line_delta(data, version_id,
2786
plain=True)
2787
lines = self.target.factory.lower_line_delta(delta)
2788
return self.target._data._record_to_data(version_id, digest, lines)
2789
2790
2791
InterVersionedFile.register_optimiser(InterKnit)
2792
2793
2794
class WeaveToKnit(InterVersionedFile):
2795
"""Optimised code paths for weave to knit operations."""
2796
2797
_matching_file_from_factory = bzrlib.weave.WeaveFile
2798
_matching_file_to_factory = KnitVersionedFile
2799
2800
@staticmethod
2801
def is_compatible(source, target):
2802
"""Be compatible with weaves to knits."""
2803
try:
2804
return (isinstance(source, bzrlib.weave.Weave) and
2805
isinstance(target, KnitVersionedFile))
2806
except AttributeError:
2807
return False
2808
2809
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2810
"""See InterVersionedFile.join."""
2811
assert isinstance(self.source, bzrlib.weave.Weave)
2812
assert isinstance(self.target, KnitVersionedFile)
2813
2814
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2815
2816
if not version_ids:
2817
return 0
2818
2819
pb = ui.ui_factory.nested_progress_bar()
2820
try:
2821
version_ids = list(version_ids)
2822
2823
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2824
this_versions = set(self.target._index.get_versions())
2825
needed_versions = self.source_ancestry - this_versions
2826
2827
if not needed_versions:
2828
return 0
2829
full_list = topo_sort(self.source.get_graph())
2830
2831
version_list = [i for i in full_list if (not self.target.has_version(i)
2832
and i in needed_versions)]
2833
2834
# do the join:
2835
count = 0
2836
total = len(version_list)
2837
for version_id in version_list:
2838
pb.update("Converting to knit", count, total)
2839
parents = self.source.get_parents(version_id)
2840
# check that its will be a consistent copy:
2841
for parent in parents:
2842
# if source has the parent, we must already have it
2843
assert (self.target.has_version(parent))
2844
self.target.add_lines(
2845
version_id, parents, self.source.get_lines(version_id))
2846
count = count + 1
2847
return count
2848
finally:
2849
pb.finished()
2850
2851
2852
InterVersionedFile.register_optimiser(WeaveToKnit)
3253
2853
3254
2854
3255
2855
# Deprecated, use PatienceSequenceMatcher instead
3362
2962
self._check_parents(child, nodes_to_annotate)
3363
2963
return nodes_to_annotate
3364
2964
3365
def _get_build_graph(self, key):
2965
def _get_build_graph(self, revision_id):
3366
2966
"""Get the graphs for building texts and annotations.
3367
2967
3368
2968
The data you need for creating a full text may be different than the
3370
2970
parents to create an annotation, but only need 1 parent to generate the
3371
2971
fulltext.)
3372
2972
3373
:return: A list of (key, index_memo) records, suitable for
2973
:return: A list of (revision_id, index_memo) records, suitable for
3374
2974
passing to read_records_iter to start reading in the raw data fro/
3375
2975
the pack file.
3376
2976
"""
3377
if key in self._annotated_lines:
2977
if revision_id in self._annotated_lines:
3378
2978
# Nothing to do
3379
2979
return []
3380
pending = set([key])
2980
pending = set([revision_id])
3381
2981
records = []
3382
2982
generation = 0
3383
2983
kept_generation = 0
3389
2989
self._all_build_details.update(build_details)
3390
2990
# new_nodes = self._knit._index._get_entries(this_iteration)
3391
2991
pending = set()
3392
for key, details in build_details.iteritems():
2992
for rev_id, details in build_details.iteritems():
3393
2993
(index_memo, compression_parent, parents,
3394
2994
record_details) = details
3395
self._revision_id_graph[key] = parents
3396
records.append((key, index_memo))
2995
self._revision_id_graph[rev_id] = parents
2996
records.append((rev_id, index_memo))
3397
2997
# Do we actually need to check _annotated_lines?
3398
2998
pending.update(p for p in parents
3399
2999
if p not in self._all_build_details)
3400
3000
if compression_parent:
3401
3001
self._compression_children.setdefault(compression_parent,
3402
[]).append(key)
3002
[]).append(rev_id)
3403
3003
if parents:
3404
3004
for parent in parents:
3405
3005
self._annotate_children.setdefault(parent,
3406
[]).append(key)
3006
[]).append(rev_id)
3407
3007
num_gens = generation - kept_generation
3408
3008
if ((num_gens >= self._generations_until_keep)
3409
3009
and len(parents) > 1):
3410
3010
kept_generation = generation
3411
self._nodes_to_keep_annotations.add(key)
3011
self._nodes_to_keep_annotations.add(rev_id)
3412
3012
3413
3013
missing_versions = this_iteration.difference(build_details.keys())
3414
3014
self._ghosts.update(missing_versions)
3416
3016
# add a key, no parents
3417
3017
self._revision_id_graph[missing_version] = ()
3418
3018
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))
3019
# XXX: This should probably be a real exception, as it is a data
3020
# inconsistency
3021
assert not self._ghosts.intersection(self._compression_children), \
3022
"We cannot have nodes which have a compression parent of a ghost."
3425
3023
# Cleanout anything that depends on a ghost so that we don't wait for
3426
3024
# the ghost to show up
3427
3025
for node in self._ghosts:
3440
3038
# still need parents, cleaning them up when those parents are
3441
3039
# processed.
3442
3040
for (rev_id, record,
3443
digest) in self._knit._read_records_iter(records):
3041
digest) in self._knit._data.read_records_iter(records):
3444
3042
if rev_id in self._annotated_lines:
3445
3043
continue
3446
3044
parent_ids = self._revision_id_graph[rev_id]
3455
3053
if len(parent_ids) == 0:
3456
3054
# There are no parents for this node, so just add it
3457
3055
# TODO: This probably needs to be decoupled
3458
fulltext_content, delta = self._knit._factory.parse_record(
3056
assert compression_parent is None
3057
fulltext_content, delta = self._knit.factory.parse_record(
3459
3058
rev_id, record, record_details, None)
3460
3059
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3461
3060
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3469
3068
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3470
3069
(index_memo, compression_parent, parents,
3471
3070
record_details) = self._all_build_details[rev_id]
3472
blocks = None
3473
3071
if compression_parent is not None:
3474
3072
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))
3073
assert rev_id in comp_children
3478
3074
# If there is only 1 child, it is safe to reuse this
3479
3075
# content
3480
3076
reuse_content = (len(comp_children) == 1
3490
3086
parent_fulltext_content = self._fulltext_contents[compression_parent]
3491
3087
parent_fulltext = parent_fulltext_content.text()
3492
3088
comp_children.remove(rev_id)
3493
fulltext_content, delta = self._knit._factory.parse_record(
3089
fulltext_content, delta = self._knit.factory.parse_record(
3494
3090
rev_id, record, record_details,
3495
3091
parent_fulltext_content,
3496
3092
copy_base_content=(not reuse_content))
3497
3093
fulltext = self._add_fulltext_content(rev_id,
3498
3094
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)
3095
blocks = KnitContent.get_line_delta_blocks(delta,
3096
parent_fulltext, fulltext)
3504
3097
else:
3505
fulltext_content = self._knit._factory.parse_fulltext(
3098
fulltext_content = self._knit.factory.parse_fulltext(
3506
3099
record, rev_id)
3507
3100
fulltext = self._add_fulltext_content(rev_id,
3508
3101
fulltext_content)
3102
blocks = None
3509
3103
nodes_to_annotate.extend(
3510
3104
self._add_annotation(rev_id, fulltext, parent_ids,
3511
3105
left_matching_blocks=blocks))
3521
3115
self._heads_provider = head_cache
3522
3116
return head_cache
3523
3117
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)
3118
def annotate(self, revision_id):
3119
"""Return the annotated fulltext at the given revision.
3120
3121
:param revision_id: The revision id for this file
3122
"""
3123
records = self._get_build_graph(revision_id)
3124
if revision_id in self._ghosts:
3125
raise errors.RevisionNotPresent(revision_id, self._knit)
3126
self._annotate_records(records)
3127
return self._annotated_lines[revision_id]
3580
3128
3581
3129
3582
3130
try:
Loggerhead is a web-based interface for Breezy
Version: 2.0.1