« 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
- doc/en/developer-guide
- files removed:
- .testr.conf
- apport
- bzrlib/filters
- bzrlib/help_topics/es
- bzrlib/plugins/netrc_credential_store
- bzrlib/plugins/netrc_credential_store/tests
- bzrlib/plugins/news_merge
- bzrlib/plugins/news_merge/tests
- bzrlib/tests/ftp_server
- bzrlib/tests/per_bzrdir_colo
- bzrlib/tests/per_foreign_vcs
- bzrlib/tests/per_interbranch
- bzrlib/tests/per_repository_chk
- bzrlib/tests/per_repository_reference
- bzrlib/tests/per_uifactory
- bzrlib/tests/ssl_certs
- bzrlib/transport/ftp
- contrib/debian
- doc/developers/_static
- doc/developers/_templates
- doc/en/_static
- doc/en/_static/en
- doc/en/_templates
- doc/en/upgrade-guide
- doc/en/whats-new
- doc/es
- doc/es/_static
- doc/es/_static/es
- doc/es/_templates
- doc/es/mini-tutorial
- doc/es/quick-reference
- doc/es/user-guide
- doc/ja
- doc/ja/_static
- doc/ja/_templates
- doc/ja/mini-tutorial
- doc/ja/tutorials
- doc/ja/upgrade-guide
- doc/ja/user-guide
- doc/ja/user-guide/images
- doc/ja/user-reference
- doc/ru
- doc/ru/_static
- doc/ru/_static/ru
- doc/ru/_templates
- doc/ru/mini-tutorial
- doc/ru/quick-reference
- doc/ru/tutorials
- doc/ru/user-guide
- doc/ru/user-guide/images
- tools/packaging
- tools/win32/buildout-templates
- tools/win32/buildout-templates/bin
- files renamed:
- bzrlib/_dirstate_helpers_pyx.h => bzrlib/_dirstate_helpers_c.h
- bzrlib/_dirstate_helpers_pyx.pyx => bzrlib/_dirstate_helpers_c.pyx
- bzrlib/_knit_load_data_pyx.pyx => bzrlib/_knit_load_data_c.pyx
- bzrlib/help_topics/en/conflict-types.txt => bzrlib/help_topics/en/conflicts.txt
- bzrlib/tests/per_branch/ => bzrlib/tests/branch_implementations/
- bzrlib/tests/per_bzrdir/ => bzrlib/tests/bzrdir_implementations/
- bzrlib/tests/ftp_server/medusa_based.py => bzrlib/tests/ftp_server.py
- bzrlib/tests/per_interrepository/ => bzrlib/tests/interrepository_implementations/
- bzrlib/tests/per_intertree/ => bzrlib/tests/intertree_implementations/
- bzrlib/tests/per_inventory/ => bzrlib/tests/inventory_implementations/
- bzrlib/tests/per_repository/ => bzrlib/tests/repository_implementations/
- bzrlib/tests/test_revisionspec.py => bzrlib/tests/test_revisionnamespaces.py
- bzrlib/tests/per_transport.py => bzrlib/tests/test_transport_implementations.py
- bzrlib/tests/per_versionedfile.py => bzrlib/tests/test_versionedfile.py
- bzrlib/tests/per_tree/ => bzrlib/tests/tree_implementations/
- bzrlib/tests/per_workingtree/ => bzrlib/tests/workingtree_implementations/
- bzrlib/transport/ftp/__init__.py => bzrlib/transport/ftp.py
- bzrlib/util/_bencode_py.py => bzrlib/util/bencode.py
- bzrlib/xml8.py => bzrlib/xml5.py
- doc/en/_static/en/Makefile => doc/en/quick-reference/Makefile
- doc/en/_static/en/bzr-en-quick-reference.pdf => doc/en/quick-reference/quick-start-summary.pdf
- doc/en/_static/en/bzr-en-quick-reference.png => doc/en/quick-reference/quick-start-summary.png
- doc/en/_static/en/bzr-en-quick-reference.svg => doc/en/quick-reference/quick-start-summary.svg
- 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
- tools/generate_docs.py => generate_docs.py
- bzrlib/doc_generate/ => tools/doc_generate/
- files modified:
- .bzrignore
added
removed
bzrlib/knit.py
1
# Copyright (C) 2006-2010 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
from itertools import izip
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
static_tuple,
73
80
trace,
74
tsort,
75
tuned_gzip,
76
ui,
77
81
)
78
82
""")
79
83
from bzrlib import (
84
cache_utf8,
85
debug,
86
diff,
80
87
errors,
81
88
osutils,
82
89
patiencediff,
90
progress,
91
merge,
92
ui,
83
93
)
84
94
from bzrlib.errors import (
85
95
FileExists,
90
100
KnitHeaderError,
91
101
RevisionNotPresent,
92
102
RevisionAlreadyPresent,
93
SHA1KnitCorrupt,
94
103
)
104
from bzrlib.tuned_gzip import GzipFile, bytes_to_gzip
95
105
from bzrlib.osutils import (
96
106
contains_whitespace,
97
107
contains_linebreaks,
98
108
sha_string,
99
109
sha_strings,
100
split_lines,
101
)
102
from bzrlib.versionedfile import (
103
AbsentContentFactory,
104
adapter_registry,
105
ConstantMapper,
106
ContentFactory,
107
ChunkedContentFactory,
108
sort_groupcompress,
109
VersionedFile,
110
VersionedFiles,
111
)
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
112
116
113
117
114
118
# TODO: Split out code specific to this format into an associated object.
126
130
127
131
DATA_SUFFIX = '.knit'
128
132
INDEX_SUFFIX = '.kndx'
129
_STREAM_MIN_BUFFER_SIZE = 5*1024*1024
130
131
132
class KnitAdapter(object):
133
"""Base class for knit record adaption."""
134
135
def __init__(self, basis_vf):
136
"""Create an adapter which accesses full texts from basis_vf.
137
138
:param basis_vf: A versioned file to access basis texts of deltas from.
139
May be None for adapters that do not need to access basis texts.
140
"""
141
self._data = KnitVersionedFiles(None, None)
142
self._annotate_factory = KnitAnnotateFactory()
143
self._plain_factory = KnitPlainFactory()
144
self._basis_vf = basis_vf
145
146
147
class FTAnnotatedToUnannotated(KnitAdapter):
148
"""An adapter from FT annotated knits to unannotated ones."""
149
150
def get_bytes(self, factory):
151
annotated_compressed_bytes = factory._raw_record
152
rec, contents = \
153
self._data._parse_record_unchecked(annotated_compressed_bytes)
154
content = self._annotate_factory.parse_fulltext(contents, rec[1])
155
size, bytes = self._data._record_to_data((rec[1],), rec[3], content.text())
156
return bytes
157
158
159
class DeltaAnnotatedToUnannotated(KnitAdapter):
160
"""An adapter for deltas from annotated to unannotated."""
161
162
def get_bytes(self, factory):
163
annotated_compressed_bytes = factory._raw_record
164
rec, contents = \
165
self._data._parse_record_unchecked(annotated_compressed_bytes)
166
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
167
plain=True)
168
contents = self._plain_factory.lower_line_delta(delta)
169
size, bytes = self._data._record_to_data((rec[1],), rec[3], contents)
170
return bytes
171
172
173
class FTAnnotatedToFullText(KnitAdapter):
174
"""An adapter from FT annotated knits to unannotated ones."""
175
176
def get_bytes(self, factory):
177
annotated_compressed_bytes = factory._raw_record
178
rec, contents = \
179
self._data._parse_record_unchecked(annotated_compressed_bytes)
180
content, delta = self._annotate_factory.parse_record(factory.key[-1],
181
contents, factory._build_details, None)
182
return ''.join(content.text())
183
184
185
class DeltaAnnotatedToFullText(KnitAdapter):
186
"""An adapter for deltas from annotated to unannotated."""
187
188
def get_bytes(self, factory):
189
annotated_compressed_bytes = factory._raw_record
190
rec, contents = \
191
self._data._parse_record_unchecked(annotated_compressed_bytes)
192
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
193
plain=True)
194
compression_parent = factory.parents[0]
195
basis_entry = self._basis_vf.get_record_stream(
196
[compression_parent], 'unordered', True).next()
197
if basis_entry.storage_kind == 'absent':
198
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
199
basis_chunks = basis_entry.get_bytes_as('chunked')
200
basis_lines = osutils.chunks_to_lines(basis_chunks)
201
# Manually apply the delta because we have one annotated content and
202
# one plain.
203
basis_content = PlainKnitContent(basis_lines, compression_parent)
204
basis_content.apply_delta(delta, rec[1])
205
basis_content._should_strip_eol = factory._build_details[1]
206
return ''.join(basis_content.text())
207
208
209
class FTPlainToFullText(KnitAdapter):
210
"""An adapter from FT plain knits to unannotated ones."""
211
212
def get_bytes(self, factory):
213
compressed_bytes = factory._raw_record
214
rec, contents = \
215
self._data._parse_record_unchecked(compressed_bytes)
216
content, delta = self._plain_factory.parse_record(factory.key[-1],
217
contents, factory._build_details, None)
218
return ''.join(content.text())
219
220
221
class DeltaPlainToFullText(KnitAdapter):
222
"""An adapter for deltas from annotated to unannotated."""
223
224
def get_bytes(self, factory):
225
compressed_bytes = factory._raw_record
226
rec, contents = \
227
self._data._parse_record_unchecked(compressed_bytes)
228
delta = self._plain_factory.parse_line_delta(contents, rec[1])
229
compression_parent = factory.parents[0]
230
# XXX: string splitting overhead.
231
basis_entry = self._basis_vf.get_record_stream(
232
[compression_parent], 'unordered', True).next()
233
if basis_entry.storage_kind == 'absent':
234
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
235
basis_chunks = basis_entry.get_bytes_as('chunked')
236
basis_lines = osutils.chunks_to_lines(basis_chunks)
237
basis_content = PlainKnitContent(basis_lines, compression_parent)
238
# Manually apply the delta because we have one annotated content and
239
# one plain.
240
content, _ = self._plain_factory.parse_record(rec[1], contents,
241
factory._build_details, basis_content)
242
return ''.join(content.text())
243
244
245
class KnitContentFactory(ContentFactory):
246
"""Content factory for streaming from knits.
247
248
:seealso ContentFactory:
249
"""
250
251
def __init__(self, key, parents, build_details, sha1, raw_record,
252
annotated, knit=None, network_bytes=None):
253
"""Create a KnitContentFactory for key.
254
255
:param key: The key.
256
:param parents: The parents.
257
:param build_details: The build details as returned from
258
get_build_details.
259
:param sha1: The sha1 expected from the full text of this object.
260
:param raw_record: The bytes of the knit data from disk.
261
:param annotated: True if the raw data is annotated.
262
:param network_bytes: None to calculate the network bytes on demand,
263
not-none if they are already known.
264
"""
265
ContentFactory.__init__(self)
266
self.sha1 = sha1
267
self.key = key
268
self.parents = parents
269
if build_details[0] == 'line-delta':
270
kind = 'delta'
271
else:
272
kind = 'ft'
273
if annotated:
274
annotated_kind = 'annotated-'
275
else:
276
annotated_kind = ''
277
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
278
self._raw_record = raw_record
279
self._network_bytes = network_bytes
280
self._build_details = build_details
281
self._knit = knit
282
283
def _create_network_bytes(self):
284
"""Create a fully serialised network version for transmission."""
285
# storage_kind, key, parents, Noeol, raw_record
286
key_bytes = '\x00'.join(self.key)
287
if self.parents is None:
288
parent_bytes = 'None:'
289
else:
290
parent_bytes = '\t'.join('\x00'.join(key) for key in self.parents)
291
if self._build_details[1]:
292
noeol = 'N'
293
else:
294
noeol = ' '
295
network_bytes = "%s\n%s\n%s\n%s%s" % (self.storage_kind, key_bytes,
296
parent_bytes, noeol, self._raw_record)
297
self._network_bytes = network_bytes
298
299
def get_bytes_as(self, storage_kind):
300
if storage_kind == self.storage_kind:
301
if self._network_bytes is None:
302
self._create_network_bytes()
303
return self._network_bytes
304
if ('-ft-' in self.storage_kind and
305
storage_kind in ('chunked', 'fulltext')):
306
adapter_key = (self.storage_kind, 'fulltext')
307
adapter_factory = adapter_registry.get(adapter_key)
308
adapter = adapter_factory(None)
309
bytes = adapter.get_bytes(self)
310
if storage_kind == 'chunked':
311
return [bytes]
312
else:
313
return bytes
314
if self._knit is not None:
315
# Not redundant with direct conversion above - that only handles
316
# fulltext cases.
317
if storage_kind == 'chunked':
318
return self._knit.get_lines(self.key[0])
319
elif storage_kind == 'fulltext':
320
return self._knit.get_text(self.key[0])
321
raise errors.UnavailableRepresentation(self.key, storage_kind,
322
self.storage_kind)
323
324
325
class LazyKnitContentFactory(ContentFactory):
326
"""A ContentFactory which can either generate full text or a wire form.
327
328
:seealso ContentFactory:
329
"""
330
331
def __init__(self, key, parents, generator, first):
332
"""Create a LazyKnitContentFactory.
333
334
:param key: The key of the record.
335
:param parents: The parents of the record.
336
:param generator: A _ContentMapGenerator containing the record for this
337
key.
338
:param first: Is this the first content object returned from generator?
339
if it is, its storage kind is knit-delta-closure, otherwise it is
340
knit-delta-closure-ref
341
"""
342
self.key = key
343
self.parents = parents
344
self.sha1 = None
345
self._generator = generator
346
self.storage_kind = "knit-delta-closure"
347
if not first:
348
self.storage_kind = self.storage_kind + "-ref"
349
self._first = first
350
351
def get_bytes_as(self, storage_kind):
352
if storage_kind == self.storage_kind:
353
if self._first:
354
return self._generator._wire_bytes()
355
else:
356
# all the keys etc are contained in the bytes returned in the
357
# first record.
358
return ''
359
if storage_kind in ('chunked', 'fulltext'):
360
chunks = self._generator._get_one_work(self.key).text()
361
if storage_kind == 'chunked':
362
return chunks
363
else:
364
return ''.join(chunks)
365
raise errors.UnavailableRepresentation(self.key, storage_kind,
366
self.storage_kind)
367
368
369
def knit_delta_closure_to_records(storage_kind, bytes, line_end):
370
"""Convert a network record to a iterator over stream records.
371
372
:param storage_kind: The storage kind of the record.
373
Must be 'knit-delta-closure'.
374
:param bytes: The bytes of the record on the network.
375
"""
376
generator = _NetworkContentMapGenerator(bytes, line_end)
377
return generator.get_record_stream()
378
379
380
def knit_network_to_record(storage_kind, bytes, line_end):
381
"""Convert a network record to a record object.
382
383
:param storage_kind: The storage kind of the record.
384
:param bytes: The bytes of the record on the network.
385
"""
386
start = line_end
387
line_end = bytes.find('\n', start)
388
key = tuple(bytes[start:line_end].split('\x00'))
389
start = line_end + 1
390
line_end = bytes.find('\n', start)
391
parent_line = bytes[start:line_end]
392
if parent_line == 'None:':
393
parents = None
394
else:
395
parents = tuple(
396
[tuple(segment.split('\x00')) for segment in parent_line.split('\t')
397
if segment])
398
start = line_end + 1
399
noeol = bytes[start] == 'N'
400
if 'ft' in storage_kind:
401
method = 'fulltext'
402
else:
403
method = 'line-delta'
404
build_details = (method, noeol)
405
start = start + 1
406
raw_record = bytes[start:]
407
annotated = 'annotated' in storage_kind
408
return [KnitContentFactory(key, parents, build_details, None, raw_record,
409
annotated, network_bytes=bytes)]
410
133
411
134
412
135
class KnitContent(object):
413
"""Content of a knit version to which deltas can be applied.
414
415
This is always stored in memory as a list of lines with \n at the end,
416
plus a flag saying if the final ending is really there or not, because that
417
corresponds to the on-disk knit representation.
418
"""
136
"""Content of a knit version to which deltas can be applied."""
419
137
420
138
def __init__(self):
421
139
self._should_strip_eol = False
422
140
141
def annotate(self):
142
"""Return a list of (origin, text) tuples."""
143
return list(self.annotate_iter())
144
423
145
def apply_delta(self, delta, new_version_id):
424
146
"""Apply delta to this object to become new_version_id."""
425
147
raise NotImplementedError(self.apply_delta)
426
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
427
155
def line_delta_iter(self, new_lines):
428
156
"""Generate line-based delta from this content to new_lines."""
429
157
new_texts = new_lines.text()
472
200
KnitContent.__init__(self)
473
201
self._lines = lines
474
202
475
def annotate(self):
476
"""Return a list of (origin, text) for each content line."""
477
lines = self._lines[:]
478
if self._should_strip_eol:
479
origin, last_line = lines[-1]
480
lines[-1] = (origin, last_line.rstrip('\n'))
481
return lines
203
def annotate_iter(self):
204
"""Yield tuples of (origin, text) for each content line."""
205
return iter(self._lines)
482
206
483
207
def apply_delta(self, delta, new_version_id):
484
208
"""Apply delta to this object to become new_version_id."""
488
212
lines[offset+start:offset+end] = delta_lines
489
213
offset = offset + (start - end) + count
490
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
491
220
def text(self):
492
221
try:
493
222
lines = [text for origin, text in self._lines]
498
227
raise KnitCorrupt(self,
499
228
"line in annotated knit missing annotation information: %s"
500
229
% (e,))
230
501
231
if self._should_strip_eol:
502
lines[-1] = lines[-1].rstrip('\n')
232
anno, line = lines[-1]
233
lines[-1] = (anno, line.rstrip('\n'))
503
234
return lines
504
235
505
236
def copy(self):
508
239
509
240
class PlainKnitContent(KnitContent):
510
241
"""Unannotated content.
511
242
512
243
When annotate[_iter] is called on this content, the same version is reported
513
244
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
514
245
objects.
519
250
self._lines = lines
520
251
self._version_id = version_id
521
252
522
def annotate(self):
523
"""Return a list of (origin, text) for each content line."""
524
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
525
257
526
258
def apply_delta(self, delta, new_version_id):
527
259
"""Apply delta to this object to become new_version_id."""
535
267
def copy(self):
536
268
return PlainKnitContent(self._lines[:], self._version_id)
537
269
270
def strip_last_line_newline(self):
271
self._lines[-1] = self._lines[-1].rstrip('\n')
272
self._should_strip_eol = False
273
538
274
def text(self):
539
275
lines = self._lines
540
276
if self._should_strip_eol:
564
300
"""
565
301
method, noeol = record_details
566
302
if method == 'line-delta':
303
assert base_content is not None
567
304
if copy_base_content:
568
305
content = base_content.copy()
569
306
else:
666
403
667
404
see parse_fulltext which this inverts.
668
405
"""
406
# TODO: jam 20070209 We only do the caching thing to make sure that
407
# the origin is a valid utf-8 line, eventually we could remove it
669
408
return ['%s %s' % (o, t) for o, t in content._lines]
670
409
671
410
def lower_line_delta(self, delta):
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) is 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_flat(key)
762
763
764
765
def make_file_factory(annotated, mapper):
766
"""Create a factory for creating a file based KnitVersionedFiles.
767
768
This is only functional enough to run interface tests, it doesn't try to
769
provide a full pack environment.
770
771
:param annotated: knit annotations are wanted.
772
:param mapper: The mapper from keys to paths.
773
"""
774
def factory(transport):
775
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
776
access = _KnitKeyAccess(transport, mapper)
777
return KnitVersionedFiles(index, access, annotated=annotated)
778
return factory
779
780
781
def make_pack_factory(graph, delta, keylength):
782
"""Create a factory for creating a pack based VersionedFiles.
783
784
This is only functional enough to run interface tests, it doesn't try to
785
provide a full pack environment.
786
787
:param graph: Store a graph.
788
:param delta: Delta compress contents.
789
:param keylength: How long should keys be.
790
"""
791
def factory(transport):
792
parents = graph or delta
793
ref_length = 0
794
if graph:
795
ref_length += 1
796
if delta:
797
ref_length += 1
798
max_delta_chain = 200
799
else:
800
max_delta_chain = 0
801
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
802
key_elements=keylength)
803
stream = transport.open_write_stream('newpack')
804
writer = pack.ContainerWriter(stream.write)
805
writer.begin()
806
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
807
deltas=delta, add_callback=graph_index.add_nodes)
808
access = _DirectPackAccess({})
809
access.set_writer(writer, graph_index, (transport, 'newpack'))
810
result = KnitVersionedFiles(index, access,
811
max_delta_chain=max_delta_chain)
812
result.stream = stream
813
result.writer = writer
814
return result
815
return factory
816
817
818
def cleanup_pack_knit(versioned_files):
819
versioned_files.stream.close()
820
versioned_files.writer.end()
821
822
823
def _get_total_build_size(self, keys, positions):
824
"""Determine the total bytes to build these keys.
825
826
(helper function because _KnitGraphIndex and _KndxIndex work the same, but
827
don't inherit from a common base.)
828
829
:param keys: Keys that we want to build
830
:param positions: dict of {key, (info, index_memo, comp_parent)} (such
831
as returned by _get_components_positions)
832
:return: Number of bytes to build those keys
833
"""
834
all_build_index_memos = {}
835
build_keys = keys
836
while build_keys:
837
next_keys = set()
838
for key in build_keys:
839
# This is mostly for the 'stacked' case
840
# Where we will be getting the data from a fallback
841
if key not in positions:
842
continue
843
_, index_memo, compression_parent = positions[key]
844
all_build_index_memos[key] = index_memo
845
if compression_parent not in all_build_index_memos:
846
next_keys.add(compression_parent)
847
build_keys = next_keys
848
return sum([index_memo[2] for index_memo
849
in all_build_index_memos.itervalues()])
850
851
852
class KnitVersionedFiles(VersionedFiles):
853
"""Storage for many versioned files using knit compression.
854
855
Backend storage is managed by indices and data objects.
856
857
:ivar _index: A _KnitGraphIndex or similar that can describe the
858
parents, graph, compression and data location of entries in this
859
KnitVersionedFiles. Note that this is only the index for
860
*this* vfs; if there are fallbacks they must be queried separately.
861
"""
862
863
def __init__(self, index, data_access, max_delta_chain=200,
864
annotated=False, reload_func=None):
865
"""Create a KnitVersionedFiles with index and data_access.
866
867
:param index: The index for the knit data.
868
:param data_access: The access object to store and retrieve knit
869
records.
870
:param max_delta_chain: The maximum number of deltas to permit during
871
insertion. Set to 0 to prohibit the use of deltas.
872
:param annotated: Set to True to cause annotations to be calculated and
873
stored during insertion.
874
:param reload_func: An function that can be called if we think we need
875
to reload the pack listing and try again. See
876
'bzrlib.repofmt.pack_repo.AggregateIndex' for the signature.
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.
877
523
"""
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
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)
887
551
888
552
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
line_bytes = ''.join(lines)
913
return self._add(key, lines, parents,
914
parent_texts, left_matching_blocks, nostore_sha, random_id,
915
line_bytes=line_bytes)
916
917
def _add_text(self, key, parents, text, nostore_sha=None, random_id=False):
918
"""See VersionedFiles._add_text()."""
919
self._index._check_write_ok()
920
self._check_add(key, None, random_id, check_content=False)
921
if text.__class__ is not str:
922
raise errors.BzrBadParameterUnicode("text")
923
if parents is None:
924
# The caller might pass None if there is no graph data, but kndx
925
# indexes can't directly store that, so we give them
926
# an empty tuple instead.
927
parents = ()
928
return self._add(key, None, parents,
929
None, None, nostore_sha, random_id,
930
line_bytes=text)
931
932
def _add(self, key, lines, parents, parent_texts,
933
left_matching_blocks, nostore_sha, random_id,
934
line_bytes):
935
"""Add a set of lines on top of version specified by parents.
936
937
Any versions not present will be converted into ghosts.
938
939
:param lines: A list of strings where each one is a single line (has a
940
single newline at the end of the string) This is now optional
941
(callers can pass None). It is left in its location for backwards
942
compatibility. It should ''.join(lines) must == line_bytes
943
:param line_bytes: A single string containing the content
944
945
We pass both lines and line_bytes because different routes bring the
946
values to this function. And for memory efficiency, we don't want to
947
have to split/join on-demand.
948
"""
949
# first thing, if the content is something we don't need to store, find
950
# that out.
951
digest = sha_string(line_bytes)
952
if nostore_sha == digest:
953
raise errors.ExistingContent
954
955
present_parents = []
956
if parent_texts is None:
957
parent_texts = {}
958
# Do a single query to ascertain parent presence; we only compress
959
# against parents in the same kvf.
960
present_parent_map = self._index.get_parent_map(parents)
961
for parent in parents:
962
if parent in present_parent_map:
963
present_parents.append(parent)
964
965
# Currently we can only compress against the left most present parent.
966
if (len(present_parents) == 0 or
967
present_parents[0] != parents[0]):
968
delta = False
969
else:
970
# To speed the extract of texts the delta chain is limited
971
# to a fixed number of deltas. This should minimize both
972
# I/O and the time spend applying deltas.
973
delta = self._check_should_delta(present_parents[0])
974
975
text_length = len(line_bytes)
976
options = []
977
no_eol = False
978
# Note: line_bytes is not modified to add a newline, that is tracked
979
# via the no_eol flag. 'lines' *is* modified, because that is the
980
# general values needed by the Content code.
981
if line_bytes and line_bytes[-1] != '\n':
982
options.append('no-eol')
983
no_eol = True
984
# Copy the existing list, or create a new one
985
if lines is None:
986
lines = osutils.split_lines(line_bytes)
987
else:
988
lines = lines[:]
989
# Replace the last line with one that ends in a final newline
990
lines[-1] = lines[-1] + '\n'
991
if lines is None:
992
lines = osutils.split_lines(line_bytes)
993
994
for element in key[:-1]:
995
if type(element) is not str:
996
raise TypeError("key contains non-strings: %r" % (key,))
997
if key[-1] is None:
998
key = key[:-1] + ('sha1:' + digest,)
999
elif type(key[-1]) is not str:
1000
raise TypeError("key contains non-strings: %r" % (key,))
1001
# Knit hunks are still last-element only
1002
version_id = key[-1]
1003
content = self._factory.make(lines, version_id)
1004
if no_eol:
1005
# Hint to the content object that its text() call should strip the
1006
# EOL.
1007
content._should_strip_eol = True
1008
if delta or (self._factory.annotated and len(present_parents) > 0):
1009
# Merge annotations from parent texts if needed.
1010
delta_hunks = self._merge_annotations(content, present_parents,
1011
parent_texts, delta, self._factory.annotated,
1012
left_matching_blocks)
1013
1014
if delta:
1015
options.append('line-delta')
1016
store_lines = self._factory.lower_line_delta(delta_hunks)
1017
size, bytes = self._record_to_data(key, digest,
1018
store_lines)
1019
else:
1020
options.append('fulltext')
1021
# isinstance is slower and we have no hierarchy.
1022
if self._factory.__class__ is KnitPlainFactory:
1023
# Use the already joined bytes saving iteration time in
1024
# _record_to_data.
1025
dense_lines = [line_bytes]
1026
if no_eol:
1027
dense_lines.append('\n')
1028
size, bytes = self._record_to_data(key, digest,
1029
lines, dense_lines)
1030
else:
1031
# get mixed annotation + content and feed it into the
1032
# serialiser.
1033
store_lines = self._factory.lower_fulltext(content)
1034
size, bytes = self._record_to_data(key, digest,
1035
store_lines)
1036
1037
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
1038
self._index.add_records(
1039
((key, options, access_memo, parents),),
1040
random_id=random_id)
1041
return digest, text_length, content
1042
1043
def annotate(self, key):
1044
"""See VersionedFiles.annotate."""
1045
return self._factory.annotate(self, key)
1046
1047
def get_annotator(self):
1048
return _KnitAnnotator(self)
1049
1050
def check(self, progress_bar=None, keys=None):
1051
"""See VersionedFiles.check()."""
1052
if keys is None:
1053
return self._logical_check()
1054
else:
1055
# At the moment, check does not extra work over get_record_stream
1056
return self.get_record_stream(keys, 'unordered', True)
1057
1058
def _logical_check(self):
1059
# This doesn't actually test extraction of everything, but that will
1060
# impact 'bzr check' substantially, and needs to be integrated with
1061
# care. However, it does check for the obvious problem of a delta with
1062
# no basis.
1063
keys = self._index.keys()
1064
parent_map = self.get_parent_map(keys)
1065
for key in keys:
1066
if self._index.get_method(key) != 'fulltext':
1067
compression_parent = parent_map[key][0]
1068
if compression_parent not in parent_map:
1069
raise errors.KnitCorrupt(self,
1070
"Missing basis parent %s for %s" % (
1071
compression_parent, key))
1072
for fallback_vfs in self._fallback_vfs:
1073
fallback_vfs.check()
1074
1075
def _check_add(self, key, lines, random_id, check_content):
1076
"""check that version_id and lines are safe to add."""
1077
version_id = key[-1]
1078
if version_id is not None:
1079
if contains_whitespace(version_id):
1080
raise InvalidRevisionId(version_id, self)
1081
self.check_not_reserved_id(version_id)
1082
# TODO: If random_id==False and the key is already present, we should
1083
# probably check that the existing content is identical to what is
1084
# being inserted, and otherwise raise an exception. This would make
1085
# the bundle code simpler.
1086
if check_content:
1087
self._check_lines_not_unicode(lines)
1088
self._check_lines_are_lines(lines)
1089
1090
def _check_header(self, key, line):
1091
rec = self._split_header(line)
1092
self._check_header_version(rec, key[-1])
1093
return rec
1094
1095
def _check_header_version(self, rec, version_id):
1096
"""Checks the header version on original format knit records.
1097
1098
These have the last component of the key embedded in the record.
1099
"""
1100
if rec[1] != version_id:
1101
raise KnitCorrupt(self,
1102
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1103
1104
def _check_should_delta(self, parent):
553
return '%s(%s)' % (self.__class__.__name__,
554
self.transport.abspath(self.filename))
555
556
def _check_should_delta(self, first_parents):
1105
557
"""Iterate back through the parent listing, looking for a fulltext.
1106
558
1107
559
This is used when we want to decide whether to add a delta or a new
1114
566
"""
1115
567
delta_size = 0
1116
568
fulltext_size = None
569
delta_parents = first_parents
1117
570
for count in xrange(self._max_delta_chain):
1118
try:
1119
# Note that this only looks in the index of this particular
1120
# KnitVersionedFiles, not in the fallbacks. This ensures that
1121
# we won't store a delta spanning physical repository
1122
# boundaries.
1123
build_details = self._index.get_build_details([parent])
1124
parent_details = build_details[parent]
1125
except (RevisionNotPresent, KeyError), e:
1126
# Some basis is not locally present: always fulltext
1127
return False
1128
index_memo, compression_parent, _, _ = parent_details
1129
_, _, size = index_memo
1130
if compression_parent is None:
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':
1131
575
fulltext_size = size
1132
576
break
1133
577
delta_size += size
1134
# We don't explicitly check for presence because this is in an
1135
# inner loop, and if it's missing it'll fail anyhow.
1136
parent = compression_parent
578
delta_parents = self._index.get_parents(parent)
1137
579
else:
1138
580
# We couldn't find a fulltext, so we must create a new one
1139
581
return False
1140
# Simple heuristic - if the total I/O wold be greater as a delta than
1141
# the originally installed fulltext, we create a new fulltext.
582
1142
583
return fulltext_size > delta_size
1143
584
1144
def _build_details_to_components(self, build_details):
1145
"""Convert a build_details tuple to a position tuple."""
1146
# record_details, access_memo, compression_parent
1147
return build_details[3], build_details[0], build_details[1]
1148
1149
def _get_components_positions(self, keys, allow_missing=False):
1150
"""Produce a map of position data for the components of keys.
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]
913
if delta:
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':
927
continue
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.
1151
933
1152
934
This data is intended to be used for retrieving the knit records.
1153
935
1154
A dict of key to (record_details, index_memo, next, parents) is
936
A dict of version_id to (record_details, index_memo, next, parents) is
1155
937
returned.
1156
938
method is the way referenced data should be applied.
1157
939
index_memo is the handle to pass to the data access to actually get the
1158
940
data
1159
941
next is the build-parent of the version, or None for fulltexts.
1160
942
parents is the version_ids of the parents of this version
1161
1162
:param allow_missing: If True do not raise an error on a missing component,
1163
just ignore it.
1164
943
"""
1165
944
component_data = {}
1166
pending_components = keys
945
pending_components = version_ids
1167
946
while pending_components:
1168
947
build_details = self._index.get_build_details(pending_components)
1169
948
current_components = set(pending_components)
1170
949
pending_components = set()
1171
for key, details in build_details.iteritems():
950
for version_id, details in build_details.iteritems():
1172
951
(index_memo, compression_parent, parents,
1173
952
record_details) = details
1174
953
method = record_details[0]
1175
954
if compression_parent is not None:
1176
955
pending_components.add(compression_parent)
1177
component_data[key] = self._build_details_to_components(details)
956
component_data[version_id] = (record_details, index_memo,
957
compression_parent)
1178
958
missing = current_components.difference(build_details)
1179
if missing and not allow_missing:
1180
raise errors.RevisionNotPresent(missing.pop(), self)
959
if missing:
960
raise errors.RevisionNotPresent(missing.pop(), self.filename)
1181
961
return component_data
1182
1183
def _get_content(self, key, parent_texts={}):
962
963
def _get_content(self, version_id, parent_texts={}):
1184
964
"""Returns a content object that makes up the specified
1185
965
version."""
1186
cached_version = parent_texts.get(key, None)
966
cached_version = parent_texts.get(version_id, None)
1187
967
if cached_version is not None:
1188
# Ensure the cache dict is valid.
1189
if not self.get_parent_map([key]):
1190
raise RevisionNotPresent(key, self)
968
if not self.has_version(version_id):
969
raise RevisionNotPresent(version_id, self.filename)
1191
970
return cached_version
1192
generator = _VFContentMapGenerator(self, [key])
1193
return generator._get_content(key)
1194
1195
def get_known_graph_ancestry(self, keys):
1196
"""Get a KnownGraph instance with the ancestry of keys."""
1197
parent_map, missing_keys = self._index.find_ancestry(keys)
1198
for fallback in self._fallback_vfs:
1199
if not missing_keys:
1200
break
1201
(f_parent_map, f_missing_keys) = fallback._index.find_ancestry(
1202
missing_keys)
1203
parent_map.update(f_parent_map)
1204
missing_keys = f_missing_keys
1205
kg = _mod_graph.KnownGraph(parent_map)
1206
return kg
1207
1208
def get_parent_map(self, keys):
1209
"""Get a map of the graph parents of keys.
1210
1211
:param keys: The keys to look up parents for.
1212
:return: A mapping from keys to parents. Absent keys are absent from
1213
the mapping.
1214
"""
1215
return self._get_parent_map_with_sources(keys)[0]
1216
1217
def _get_parent_map_with_sources(self, keys):
1218
"""Get a map of the parents of keys.
1219
1220
:param keys: The keys to look up parents for.
1221
:return: A tuple. The first element is a mapping from keys to parents.
1222
Absent keys are absent from the mapping. The second element is a
1223
list with the locations each key was found in. The first element
1224
is the in-this-knit parents, the second the first fallback source,
1225
and so on.
1226
"""
1227
result = {}
1228
sources = [self._index] + self._fallback_vfs
1229
source_results = []
1230
missing = set(keys)
1231
for source in sources:
1232
if not missing:
1233
break
1234
new_result = source.get_parent_map(missing)
1235
source_results.append(new_result)
1236
result.update(new_result)
1237
missing.difference_update(set(new_result))
1238
return result, source_results
1239
1240
def _get_record_map(self, keys, allow_missing=False):
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,
992
parent_texts, left_matching_blocks, nostore_sha, random_id)
993
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,
1010
left_matching_blocks, nostore_sha, random_id):
1011
"""Add a set of lines on top of version specified by parents.
1012
1013
If delta is true, compress the text as a line-delta against
1014
the first parent.
1015
1016
Any versions not present will be converted into ghosts.
1017
"""
1018
# first thing, if the content is something we don't need to store, find
1019
# that out.
1020
line_bytes = ''.join(lines)
1021
digest = sha_string(line_bytes)
1022
if nostore_sha == digest:
1023
raise errors.ExistingContent
1024
1025
present_parents = []
1026
if parent_texts is None:
1027
parent_texts = {}
1028
for parent in parents:
1029
if self.has_version(parent):
1030
present_parents.append(parent)
1031
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])):
1036
delta = False
1037
1038
text_length = len(line_bytes)
1039
options = []
1040
if lines:
1041
if lines[-1][-1] != '\n':
1042
# copy the contents of lines.
1043
lines = lines[:]
1044
options.append('no-eol')
1045
lines[-1] = lines[-1] + '\n'
1046
line_bytes += '\n'
1047
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):
1057
# Merge annotations from parent texts if needed.
1058
delta_hunks = self._merge_annotations(content, present_parents,
1059
parent_texts, delta, self.factory.annotated,
1060
left_matching_blocks)
1061
1062
if delta:
1063
options.append('line-delta')
1064
store_lines = self.factory.lower_line_delta(delta_hunks)
1065
size, bytes = self._data._record_to_data(version_id, digest,
1066
store_lines)
1067
else:
1068
options.append('fulltext')
1069
# isinstance is slower and we have no hierarchy.
1070
if self.factory.__class__ == KnitPlainFactory:
1071
# Use the already joined bytes saving iteration time in
1072
# _record_to_data.
1073
size, bytes = self._data._record_to_data(version_id, digest,
1074
lines, [line_bytes])
1075
else:
1076
# get mixed annotation + content and feed it into the
1077
# serialiser.
1078
store_lines = self.factory.lower_fulltext(content)
1079
size, bytes = self._data._record_to_data(version_id, digest,
1080
store_lines)
1081
1082
access_memo = self._data.add_raw_records([size], bytes)[0]
1083
self._index.add_versions(
1084
((version_id, options, access_memo, parents),),
1085
random_id=random_id)
1086
return digest, text_length, content
1087
1088
def check(self, progress_bar=None):
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):
1241
1102
"""Produce a dictionary of knit records.
1242
1243
:return: {key:(record, record_details, digest, next)}
1103
1104
:return: {version_id:(record, record_details, digest, next)}
1244
1105
record
1245
data returned from read_records (a KnitContentobject)
1106
data returned from read_records
1246
1107
record_details
1247
1108
opaque information to pass to parse_record
1248
1109
digest
1250
1111
next
1251
1112
build-parent of the version, i.e. the leftmost ancestor.
1252
1113
Will be None if the record is not a delta.
1253
:param keys: The keys to build a map for
1254
:param allow_missing: If some records are missing, rather than
1255
error, just return the data that could be generated.
1256
"""
1257
raw_map = self._get_record_map_unparsed(keys,
1258
allow_missing=allow_missing)
1259
return self._raw_map_to_record_map(raw_map)
1260
1261
def _raw_map_to_record_map(self, raw_map):
1262
"""Parse the contents of _get_record_map_unparsed.
1263
1264
:return: see _get_record_map.
1265
"""
1266
result = {}
1267
for key in raw_map:
1268
data, record_details, next = raw_map[key]
1269
content, digest = self._parse_record(key[-1], data)
1270
result[key] = content, record_details, digest, next
1271
return result
1272
1273
def _get_record_map_unparsed(self, keys, allow_missing=False):
1274
"""Get the raw data for reconstructing keys without parsing it.
1275
1276
:return: A dict suitable for parsing via _raw_map_to_record_map.
1277
key-> raw_bytes, (method, noeol), compression_parent
1278
"""
1279
# This retries the whole request if anything fails. Potentially we
1280
# could be a bit more selective. We could track the keys whose records
1281
# we have successfully found, and then only request the new records
1282
# from there. However, _get_components_positions grabs the whole build
1283
# chain, which means we'll likely try to grab the same records again
1284
# anyway. Also, can the build chains change as part of a pack
1285
# operation? We wouldn't want to end up with a broken chain.
1286
while True:
1287
try:
1288
position_map = self._get_components_positions(keys,
1289
allow_missing=allow_missing)
1290
# key = component_id, r = record_details, i_m = index_memo,
1291
# n = next
1292
records = [(key, i_m) for key, (r, i_m, n)
1293
in position_map.iteritems()]
1294
# Sort by the index memo, so that we request records from the
1295
# same pack file together, and in forward-sorted order
1296
records.sort(key=operator.itemgetter(1))
1297
raw_record_map = {}
1298
for key, data in self._read_records_iter_unchecked(records):
1299
(record_details, index_memo, next) = position_map[key]
1300
raw_record_map[key] = data, record_details, next
1301
return raw_record_map
1302
except errors.RetryWithNewPacks, e:
1303
self._access.reload_or_raise(e)
1304
1305
@classmethod
1306
def _split_by_prefix(cls, keys):
1307
"""For the given keys, split them up based on their prefix.
1308
1309
To keep memory pressure somewhat under control, split the
1310
requests back into per-file-id requests, otherwise "bzr co"
1311
extracts the full tree into memory before writing it to disk.
1312
This should be revisited if _get_content_maps() can ever cross
1313
file-id boundaries.
1314
1315
The keys for a given file_id are kept in the same relative order.
1316
Ordering between file_ids is not, though prefix_order will return the
1317
order that the key was first seen.
1318
1319
:param keys: An iterable of key tuples
1320
:return: (split_map, prefix_order)
1321
split_map A dictionary mapping prefix => keys
1322
prefix_order The order that we saw the various prefixes
1323
"""
1324
split_by_prefix = {}
1325
prefix_order = []
1326
for key in keys:
1327
if len(key) == 1:
1328
prefix = ''
1329
else:
1330
prefix = key[0]
1331
1332
if prefix in split_by_prefix:
1333
split_by_prefix[prefix].append(key)
1334
else:
1335
split_by_prefix[prefix] = [key]
1336
prefix_order.append(prefix)
1337
return split_by_prefix, prefix_order
1338
1339
def _group_keys_for_io(self, keys, non_local_keys, positions,
1340
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1341
"""For the given keys, group them into 'best-sized' requests.
1342
1343
The idea is to avoid making 1 request per file, but to never try to
1344
unpack an entire 1.5GB source tree in a single pass. Also when
1345
possible, we should try to group requests to the same pack file
1346
together.
1347
1348
:return: list of (keys, non_local) tuples that indicate what keys
1349
should be fetched next.
1350
"""
1351
# TODO: Ideally we would group on 2 factors. We want to extract texts
1352
# from the same pack file together, and we want to extract all
1353
# the texts for a given build-chain together. Ultimately it
1354
# probably needs a better global view.
1355
total_keys = len(keys)
1356
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1357
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1358
cur_keys = []
1359
cur_non_local = set()
1360
cur_size = 0
1361
result = []
1362
sizes = []
1363
for prefix in prefix_order:
1364
keys = prefix_split_keys[prefix]
1365
non_local = prefix_split_non_local_keys.get(prefix, [])
1366
1367
this_size = self._index._get_total_build_size(keys, positions)
1368
cur_size += this_size
1369
cur_keys.extend(keys)
1370
cur_non_local.update(non_local)
1371
if cur_size > _min_buffer_size:
1372
result.append((cur_keys, cur_non_local))
1373
sizes.append(cur_size)
1374
cur_keys = []
1375
cur_non_local = set()
1376
cur_size = 0
1377
if cur_keys:
1378
result.append((cur_keys, cur_non_local))
1379
sizes.append(cur_size)
1380
return result
1381
1382
def get_record_stream(self, keys, ordering, include_delta_closure):
1383
"""Get a stream of records for keys.
1384
1385
:param keys: The keys to include.
1386
:param ordering: Either 'unordered' or 'topological'. A topologically
1387
sorted stream has compression parents strictly before their
1388
children.
1389
:param include_delta_closure: If True then the closure across any
1390
compression parents will be included (in the opaque data).
1391
:return: An iterator of ContentFactory objects, each of which is only
1392
valid until the iterator is advanced.
1393
"""
1394
# keys might be a generator
1395
keys = set(keys)
1396
if not keys:
1397
return
1398
if not self._index.has_graph:
1399
# Cannot sort when no graph has been stored.
1400
ordering = 'unordered'
1401
1402
remaining_keys = keys
1403
while True:
1404
try:
1405
keys = set(remaining_keys)
1406
for content_factory in self._get_remaining_record_stream(keys,
1407
ordering, include_delta_closure):
1408
remaining_keys.discard(content_factory.key)
1409
yield content_factory
1410
return
1411
except errors.RetryWithNewPacks, e:
1412
self._access.reload_or_raise(e)
1413
1414
def _get_remaining_record_stream(self, keys, ordering,
1415
include_delta_closure):
1416
"""This function is the 'retry' portion for get_record_stream."""
1417
if include_delta_closure:
1418
positions = self._get_components_positions(keys, allow_missing=True)
1419
else:
1420
build_details = self._index.get_build_details(keys)
1421
# map from key to
1422
# (record_details, access_memo, compression_parent_key)
1423
positions = dict((key, self._build_details_to_components(details))
1424
for key, details in build_details.iteritems())
1425
absent_keys = keys.difference(set(positions))
1426
# There may be more absent keys : if we're missing the basis component
1427
# and are trying to include the delta closure.
1428
# XXX: We should not ever need to examine remote sources because we do
1429
# not permit deltas across versioned files boundaries.
1430
if include_delta_closure:
1431
needed_from_fallback = set()
1432
# Build up reconstructable_keys dict. key:True in this dict means
1433
# the key can be reconstructed.
1434
reconstructable_keys = {}
1435
for key in keys:
1436
# the delta chain
1437
try:
1438
chain = [key, positions[key][2]]
1439
except KeyError:
1440
needed_from_fallback.add(key)
1441
continue
1442
result = True
1443
while chain[-1] is not None:
1444
if chain[-1] in reconstructable_keys:
1445
result = reconstructable_keys[chain[-1]]
1446
break
1447
else:
1448
try:
1449
chain.append(positions[chain[-1]][2])
1450
except KeyError:
1451
# missing basis component
1452
needed_from_fallback.add(chain[-1])
1453
result = True
1454
break
1455
for chain_key in chain[:-1]:
1456
reconstructable_keys[chain_key] = result
1457
if not result:
1458
needed_from_fallback.add(key)
1459
# Double index lookups here : need a unified api ?
1460
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1461
if ordering in ('topological', 'groupcompress'):
1462
if ordering == 'topological':
1463
# Global topological sort
1464
present_keys = tsort.topo_sort(global_map)
1465
else:
1466
present_keys = sort_groupcompress(global_map)
1467
# Now group by source:
1468
source_keys = []
1469
current_source = None
1470
for key in present_keys:
1471
for parent_map in parent_maps:
1472
if key in parent_map:
1473
key_source = parent_map
1474
break
1475
if current_source is not key_source:
1476
source_keys.append((key_source, []))
1477
current_source = key_source
1478
source_keys[-1][1].append(key)
1479
else:
1480
if ordering != 'unordered':
1481
raise AssertionError('valid values for ordering are:'
1482
' "unordered", "groupcompress" or "topological" not: %r'
1483
% (ordering,))
1484
# Just group by source; remote sources first.
1485
present_keys = []
1486
source_keys = []
1487
for parent_map in reversed(parent_maps):
1488
source_keys.append((parent_map, []))
1489
for key in parent_map:
1490
present_keys.append(key)
1491
source_keys[-1][1].append(key)
1492
# We have been requested to return these records in an order that
1493
# suits us. So we ask the index to give us an optimally sorted
1494
# order.
1495
for source, sub_keys in source_keys:
1496
if source is parent_maps[0]:
1497
# Only sort the keys for this VF
1498
self._index._sort_keys_by_io(sub_keys, positions)
1499
absent_keys = keys - set(global_map)
1500
for key in absent_keys:
1501
yield AbsentContentFactory(key)
1502
# restrict our view to the keys we can answer.
1503
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1504
# XXX: At that point we need to consider the impact of double reads by
1505
# utilising components multiple times.
1506
if include_delta_closure:
1507
# XXX: get_content_maps performs its own index queries; allow state
1508
# to be passed in.
1509
non_local_keys = needed_from_fallback - absent_keys
1510
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1511
non_local_keys,
1512
positions):
1513
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1514
global_map,
1515
ordering=ordering)
1516
for record in generator.get_record_stream():
1517
yield record
1518
else:
1519
for source, keys in source_keys:
1520
if source is parent_maps[0]:
1521
# this KnitVersionedFiles
1522
records = [(key, positions[key][1]) for key in keys]
1523
for key, raw_data in self._read_records_iter_unchecked(records):
1524
(record_details, index_memo, _) = positions[key]
1525
yield KnitContentFactory(key, global_map[key],
1526
record_details, None, raw_data, self._factory.annotated, None)
1527
else:
1528
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1529
for record in vf.get_record_stream(keys, ordering,
1530
include_delta_closure):
1531
yield record
1532
1533
def get_sha1s(self, keys):
1534
"""See VersionedFiles.get_sha1s()."""
1535
missing = set(keys)
1536
record_map = self._get_record_map(missing, allow_missing=True)
1537
result = {}
1538
for key, details in record_map.iteritems():
1539
if key not in missing:
1540
continue
1541
# record entry 2 is the 'digest'.
1542
result[key] = details[2]
1543
missing.difference_update(set(result))
1544
for source in self._fallback_vfs:
1545
if not missing:
1546
break
1547
new_result = source.get_sha1s(missing)
1548
result.update(new_result)
1549
missing.difference_update(set(new_result))
1550
return result
1551
1552
def insert_record_stream(self, stream):
1553
"""Insert a record stream into this container.
1554
1555
:param stream: A stream of records to insert.
1556
:return: None
1557
:seealso VersionedFiles.get_record_stream:
1558
"""
1559
def get_adapter(adapter_key):
1560
try:
1561
return adapters[adapter_key]
1562
except KeyError:
1563
adapter_factory = adapter_registry.get(adapter_key)
1564
adapter = adapter_factory(self)
1565
adapters[adapter_key] = adapter
1566
return adapter
1567
delta_types = set()
1568
if self._factory.annotated:
1569
# self is annotated, we need annotated knits to use directly.
1570
annotated = "annotated-"
1571
convertibles = []
1572
else:
1573
# self is not annotated, but we can strip annotations cheaply.
1574
annotated = ""
1575
convertibles = set(["knit-annotated-ft-gz"])
1576
if self._max_delta_chain:
1577
delta_types.add("knit-annotated-delta-gz")
1578
convertibles.add("knit-annotated-delta-gz")
1579
# The set of types we can cheaply adapt without needing basis texts.
1580
native_types = set()
1581
if self._max_delta_chain:
1582
native_types.add("knit-%sdelta-gz" % annotated)
1583
delta_types.add("knit-%sdelta-gz" % annotated)
1584
native_types.add("knit-%sft-gz" % annotated)
1585
knit_types = native_types.union(convertibles)
1586
adapters = {}
1587
# Buffer all index entries that we can't add immediately because their
1588
# basis parent is missing. We don't buffer all because generating
1589
# annotations may require access to some of the new records. However we
1590
# can't generate annotations from new deltas until their basis parent
1591
# is present anyway, so we get away with not needing an index that
1592
# includes the new keys.
1593
#
1594
# See <http://launchpad.net/bugs/300177> about ordering of compression
1595
# parents in the records - to be conservative, we insist that all
1596
# parents must be present to avoid expanding to a fulltext.
1597
#
1598
# key = basis_parent, value = index entry to add
1599
buffered_index_entries = {}
1600
for record in stream:
1601
kind = record.storage_kind
1602
if kind.startswith('knit-') and kind.endswith('-gz'):
1603
# Check that the ID in the header of the raw knit bytes matches
1604
# the record metadata.
1605
raw_data = record._raw_record
1606
df, rec = self._parse_record_header(record.key, raw_data)
1607
df.close()
1608
buffered = False
1609
parents = record.parents
1610
if record.storage_kind in delta_types:
1611
# TODO: eventually the record itself should track
1612
# compression_parent
1613
compression_parent = parents[0]
1614
else:
1615
compression_parent = None
1616
# Raise an error when a record is missing.
1617
if record.storage_kind == 'absent':
1618
raise RevisionNotPresent([record.key], self)
1619
elif ((record.storage_kind in knit_types)
1620
and (compression_parent is None
1621
or not self._fallback_vfs
1622
or self._index.has_key(compression_parent)
1623
or not self.has_key(compression_parent))):
1624
# we can insert the knit record literally if either it has no
1625
# compression parent OR we already have its basis in this kvf
1626
# OR the basis is not present even in the fallbacks. In the
1627
# last case it will either turn up later in the stream and all
1628
# will be well, or it won't turn up at all and we'll raise an
1629
# error at the end.
1630
#
1631
# TODO: self.has_key is somewhat redundant with
1632
# self._index.has_key; we really want something that directly
1633
# asks if it's only present in the fallbacks. -- mbp 20081119
1634
if record.storage_kind not in native_types:
1635
try:
1636
adapter_key = (record.storage_kind, "knit-delta-gz")
1637
adapter = get_adapter(adapter_key)
1638
except KeyError:
1639
adapter_key = (record.storage_kind, "knit-ft-gz")
1640
adapter = get_adapter(adapter_key)
1641
bytes = adapter.get_bytes(record)
1642
else:
1643
# It's a knit record, it has a _raw_record field (even if
1644
# it was reconstituted from a network stream).
1645
bytes = record._raw_record
1646
options = [record._build_details[0]]
1647
if record._build_details[1]:
1648
options.append('no-eol')
1649
# Just blat it across.
1650
# Note: This does end up adding data on duplicate keys. As
1651
# modern repositories use atomic insertions this should not
1652
# lead to excessive growth in the event of interrupted fetches.
1653
# 'knit' repositories may suffer excessive growth, but as a
1654
# deprecated format this is tolerable. It can be fixed if
1655
# needed by in the kndx index support raising on a duplicate
1656
# add with identical parents and options.
1657
access_memo = self._access.add_raw_records(
1658
[(record.key, len(bytes))], bytes)[0]
1659
index_entry = (record.key, options, access_memo, parents)
1660
if 'fulltext' not in options:
1661
# Not a fulltext, so we need to make sure the compression
1662
# parent will also be present.
1663
# Note that pack backed knits don't need to buffer here
1664
# because they buffer all writes to the transaction level,
1665
# but we don't expose that difference at the index level. If
1666
# the query here has sufficient cost to show up in
1667
# profiling we should do that.
1668
#
1669
# They're required to be physically in this
1670
# KnitVersionedFiles, not in a fallback.
1671
if not self._index.has_key(compression_parent):
1672
pending = buffered_index_entries.setdefault(
1673
compression_parent, [])
1674
pending.append(index_entry)
1675
buffered = True
1676
if not buffered:
1677
self._index.add_records([index_entry])
1678
elif record.storage_kind == 'chunked':
1679
self.add_lines(record.key, parents,
1680
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1681
else:
1682
# Not suitable for direct insertion as a
1683
# delta, either because it's not the right format, or this
1684
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1685
# 0) or because it depends on a base only present in the
1686
# fallback kvfs.
1687
self._access.flush()
1688
try:
1689
# Try getting a fulltext directly from the record.
1690
bytes = record.get_bytes_as('fulltext')
1691
except errors.UnavailableRepresentation:
1692
adapter_key = record.storage_kind, 'fulltext'
1693
adapter = get_adapter(adapter_key)
1694
bytes = adapter.get_bytes(record)
1695
lines = split_lines(bytes)
1696
try:
1697
self.add_lines(record.key, parents, lines)
1698
except errors.RevisionAlreadyPresent:
1699
pass
1700
# Add any records whose basis parent is now available.
1701
if not buffered:
1702
added_keys = [record.key]
1703
while added_keys:
1704
key = added_keys.pop(0)
1705
if key in buffered_index_entries:
1706
index_entries = buffered_index_entries[key]
1707
self._index.add_records(index_entries)
1708
added_keys.extend(
1709
[index_entry[0] for index_entry in index_entries])
1710
del buffered_index_entries[key]
1711
if buffered_index_entries:
1712
# There were index entries buffered at the end of the stream,
1713
# So these need to be added (if the index supports holding such
1714
# entries for later insertion)
1715
all_entries = []
1716
for key in buffered_index_entries:
1717
index_entries = buffered_index_entries[key]
1718
all_entries.extend(index_entries)
1719
self._index.add_records(
1720
all_entries, missing_compression_parents=True)
1721
1722
def get_missing_compression_parent_keys(self):
1723
"""Return an iterable of keys of missing compression parents.
1724
1725
Check this after calling insert_record_stream to find out if there are
1726
any missing compression parents. If there are, the records that
1727
depend on them are not able to be inserted safely. For atomic
1728
KnitVersionedFiles built on packs, the transaction should be aborted or
1729
suspended - commit will fail at this point. Nonatomic knits will error
1730
earlier because they have no staging area to put pending entries into.
1731
"""
1732
return self._index.get_missing_compression_parents()
1733
1734
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1735
"""Iterate over the lines in the versioned files from keys.
1736
1737
This may return lines from other keys. Each item the returned
1738
iterator yields is a tuple of a line and a text version that that line
1739
is present in (not introduced in).
1740
1741
Ordering of results is in whatever order is most suitable for the
1742
underlying storage format.
1743
1744
If a progress bar is supplied, it may be used to indicate progress.
1745
The caller is responsible for cleaning up progress bars (because this
1746
is an iterator).
1747
1748
NOTES:
1749
* Lines are normalised by the underlying store: they will all have \\n
1750
terminators.
1751
* Lines are returned in arbitrary order.
1752
* If a requested key did not change any lines (or didn't have any
1753
lines), it may not be mentioned at all in the result.
1754
1755
:param pb: Progress bar supplied by caller.
1756
:return: An iterator over (line, key).
1757
"""
1758
if pb is None:
1759
pb = ui.ui_factory.nested_progress_bar()
1760
keys = set(keys)
1761
total = len(keys)
1762
done = False
1763
while not done:
1764
try:
1765
# we don't care about inclusions, the caller cares.
1766
# but we need to setup a list of records to visit.
1767
# we need key, position, length
1768
key_records = []
1769
build_details = self._index.get_build_details(keys)
1770
for key, details in build_details.iteritems():
1771
if key in keys:
1772
key_records.append((key, details[0]))
1773
records_iter = enumerate(self._read_records_iter(key_records))
1774
for (key_idx, (key, data, sha_value)) in records_iter:
1775
pb.update('Walking content', key_idx, total)
1776
compression_parent = build_details[key][1]
1777
if compression_parent is None:
1778
# fulltext
1779
line_iterator = self._factory.get_fulltext_content(data)
1780
else:
1781
# Delta
1782
line_iterator = self._factory.get_linedelta_content(data)
1783
# Now that we are yielding the data for this key, remove it
1784
# from the list
1785
keys.remove(key)
1786
# XXX: It might be more efficient to yield (key,
1787
# line_iterator) in the future. However for now, this is a
1788
# simpler change to integrate into the rest of the
1789
# codebase. RBC 20071110
1790
for line in line_iterator:
1791
yield line, key
1792
done = True
1793
except errors.RetryWithNewPacks, e:
1794
self._access.reload_or_raise(e)
1795
# If there are still keys we've not yet found, we look in the fallback
1796
# vfs, and hope to find them there. Note that if the keys are found
1797
# but had no changes or no content, the fallback may not return
1798
# anything.
1799
if keys and not self._fallback_vfs:
1800
# XXX: strictly the second parameter is meant to be the file id
1801
# but it's not easily accessible here.
1802
raise RevisionNotPresent(keys, repr(self))
1803
for source in self._fallback_vfs:
1804
if not keys:
1805
break
1806
source_keys = set()
1807
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1808
source_keys.add(key)
1809
yield line, key
1810
keys.difference_update(source_keys)
1811
pb.update('Walking content', total, total)
1812
1813
def _make_line_delta(self, delta_seq, new_content):
1814
"""Generate a line delta from delta_seq and new_content."""
1815
diff_hunks = []
1816
for op in delta_seq.get_opcodes():
1817
if op[0] == 'equal':
1818
continue
1819
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1820
return diff_hunks
1821
1822
def _merge_annotations(self, content, parents, parent_texts={},
1823
delta=None, annotated=None,
1824
left_matching_blocks=None):
1825
"""Merge annotations for content and generate deltas.
1826
1827
This is done by comparing the annotations based on changes to the text
1828
and generating a delta on the resulting full texts. If annotations are
1829
not being created then a simple delta is created.
1830
"""
1831
if left_matching_blocks is not None:
1832
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1833
else:
1834
delta_seq = None
1835
if annotated:
1836
for parent_key in parents:
1837
merge_content = self._get_content(parent_key, parent_texts)
1838
if (parent_key == parents[0] and delta_seq is not None):
1839
seq = delta_seq
1840
else:
1841
seq = patiencediff.PatienceSequenceMatcher(
1842
None, merge_content.text(), content.text())
1843
for i, j, n in seq.get_matching_blocks():
1844
if n == 0:
1845
continue
1846
# this copies (origin, text) pairs across to the new
1847
# content for any line that matches the last-checked
1848
# parent.
1849
content._lines[j:j+n] = merge_content._lines[i:i+n]
1850
# XXX: Robert says the following block is a workaround for a
1851
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1852
if content._lines and content._lines[-1][1][-1] != '\n':
1853
# The copied annotation was from a line without a trailing EOL,
1854
# reinstate one for the content object, to ensure correct
1855
# serialization.
1856
line = content._lines[-1][1] + '\n'
1857
content._lines[-1] = (content._lines[-1][0], line)
1858
if delta:
1859
if delta_seq is None:
1860
reference_content = self._get_content(parents[0], parent_texts)
1861
new_texts = content.text()
1862
old_texts = reference_content.text()
1863
delta_seq = patiencediff.PatienceSequenceMatcher(
1864
None, old_texts, new_texts)
1865
return self._make_line_delta(delta_seq, content)
1866
1867
def _parse_record(self, version_id, data):
1868
"""Parse an original format knit record.
1869
1870
These have the last element of the key only present in the stored data.
1871
"""
1872
rec, record_contents = self._parse_record_unchecked(data)
1873
self._check_header_version(rec, version_id)
1874
return record_contents, rec[3]
1875
1876
def _parse_record_header(self, key, raw_data):
1877
"""Parse a record header for consistency.
1878
1879
:return: the header and the decompressor stream.
1880
as (stream, header_record)
1881
"""
1882
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1883
try:
1884
# Current serialise
1885
rec = self._check_header(key, df.readline())
1886
except Exception, e:
1887
raise KnitCorrupt(self,
1888
"While reading {%s} got %s(%s)"
1889
% (key, e.__class__.__name__, str(e)))
1890
return df, rec
1891
1892
def _parse_record_unchecked(self, data):
1893
# profiling notes:
1894
# 4168 calls in 2880 217 internal
1895
# 4168 calls to _parse_record_header in 2121
1896
# 4168 calls to readlines in 330
1897
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1898
try:
1899
record_contents = df.readlines()
1900
except Exception, e:
1901
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1902
(data, e.__class__.__name__, str(e)))
1903
header = record_contents.pop(0)
1904
rec = self._split_header(header)
1905
last_line = record_contents.pop()
1906
if len(record_contents) != int(rec[2]):
1907
raise KnitCorrupt(self,
1908
'incorrect number of lines %s != %s'
1909
' for version {%s} %s'
1910
% (len(record_contents), int(rec[2]),
1911
rec[1], record_contents))
1912
if last_line != 'end %s\n' % rec[1]:
1913
raise KnitCorrupt(self,
1914
'unexpected version end line %r, wanted %r'
1915
% (last_line, rec[1]))
1916
df.close()
1917
return rec, record_contents
1918
1919
def _read_records_iter(self, records):
1920
"""Read text records from data file and yield result.
1921
1922
The result will be returned in whatever is the fastest to read.
1923
Not by the order requested. Also, multiple requests for the same
1924
record will only yield 1 response.
1925
:param records: A list of (key, access_memo) entries
1926
:return: Yields (key, contents, digest) in the order
1927
read, not the order requested
1928
"""
1929
if not records:
1930
return
1931
1932
# XXX: This smells wrong, IO may not be getting ordered right.
1933
needed_records = sorted(set(records), key=operator.itemgetter(1))
1934
if not needed_records:
1935
return
1936
1937
# The transport optimizes the fetching as well
1938
# (ie, reads continuous ranges.)
1939
raw_data = self._access.get_raw_records(
1940
[index_memo for key, index_memo in needed_records])
1941
1942
for (key, index_memo), data in \
1943
izip(iter(needed_records), raw_data):
1944
content, digest = self._parse_record(key[-1], data)
1945
yield key, content, digest
1946
1947
def _read_records_iter_raw(self, records):
1948
"""Read text records from data file and yield raw data.
1949
1950
This unpacks enough of the text record to validate the id is
1951
as expected but thats all.
1952
1953
Each item the iterator yields is (key, bytes,
1954
expected_sha1_of_full_text).
1955
"""
1956
for key, data in self._read_records_iter_unchecked(records):
1957
# validate the header (note that we can only use the suffix in
1958
# current knit records).
1959
df, rec = self._parse_record_header(key, data)
1960
df.close()
1961
yield key, data, rec[3]
1962
1963
def _read_records_iter_unchecked(self, records):
1964
"""Read text records from data file and yield raw data.
1965
1966
No validation is done.
1967
1968
Yields tuples of (key, data).
1969
"""
1970
# setup an iterator of the external records:
1971
# uses readv so nice and fast we hope.
1972
if len(records):
1973
# grab the disk data needed.
1974
needed_offsets = [index_memo for key, index_memo
1975
in records]
1976
raw_records = self._access.get_raw_records(needed_offsets)
1977
1978
for key, index_memo in records:
1979
data = raw_records.next()
1980
yield key, data
1981
1982
def _record_to_data(self, key, digest, lines, dense_lines=None):
1983
"""Convert key, digest, lines into a raw data block.
1984
1985
:param key: The key of the record. Currently keys are always serialised
1986
using just the trailing component.
1987
:param dense_lines: The bytes of lines but in a denser form. For
1988
instance, if lines is a list of 1000 bytestrings each ending in \n,
1989
dense_lines may be a list with one line in it, containing all the
1990
1000's lines and their \n's. Using dense_lines if it is already
1991
known is a win because the string join to create bytes in this
1992
function spends less time resizing the final string.
1993
:return: (len, a StringIO instance with the raw data ready to read.)
1994
"""
1995
chunks = ["version %s %d %s\n" % (key[-1], len(lines), digest)]
1996
chunks.extend(dense_lines or lines)
1997
chunks.append("end %s\n" % key[-1])
1998
for chunk in chunks:
1999
if type(chunk) is not str:
2000
raise AssertionError(
2001
'data must be plain bytes was %s' % type(chunk))
2002
if lines and lines[-1][-1] != '\n':
2003
raise ValueError('corrupt lines value %r' % lines)
2004
compressed_bytes = tuned_gzip.chunks_to_gzip(chunks)
2005
return len(compressed_bytes), compressed_bytes
2006
2007
def _split_header(self, line):
2008
rec = line.split()
2009
if len(rec) != 4:
2010
raise KnitCorrupt(self,
2011
'unexpected number of elements in record header')
2012
return rec
2013
2014
def keys(self):
2015
"""See VersionedFiles.keys."""
2016
if 'evil' in debug.debug_flags:
2017
trace.mutter_callsite(2, "keys scales with size of history")
2018
sources = [self._index] + self._fallback_vfs
2019
result = set()
2020
for source in sources:
2021
result.update(source.keys())
2022
return result
2023
2024
2025
class _ContentMapGenerator(object):
2026
"""Generate texts or expose raw deltas for a set of texts."""
2027
2028
def __init__(self, ordering='unordered'):
2029
self._ordering = ordering
2030
2031
def _get_content(self, key):
2032
"""Get the content object for key."""
2033
# Note that _get_content is only called when the _ContentMapGenerator
2034
# has been constructed with just one key requested for reconstruction.
2035
if key in self.nonlocal_keys:
2036
record = self.get_record_stream().next()
2037
# Create a content object on the fly
2038
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
2039
return PlainKnitContent(lines, record.key)
2040
else:
2041
# local keys we can ask for directly
2042
return self._get_one_work(key)
2043
2044
def get_record_stream(self):
2045
"""Get a record stream for the keys requested during __init__."""
2046
for record in self._work():
2047
yield record
2048
2049
def _work(self):
2050
"""Produce maps of text and KnitContents as dicts.
2051
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
2052
1146
:return: (text_map, content_map) where text_map contains the texts for
2053
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.
2054
1149
"""
2055
# NB: By definition we never need to read remote sources unless texts
2056
# are requested from them: we don't delta across stores - and we
2057
# explicitly do not want to to prevent data loss situations.
2058
if self.global_map is None:
2059
self.global_map = self.vf.get_parent_map(self.keys)
2060
nonlocal_keys = self.nonlocal_keys
2061
2062
missing_keys = set(nonlocal_keys)
2063
# Read from remote versioned file instances and provide to our caller.
2064
for source in self.vf._fallback_vfs:
2065
if not missing_keys:
2066
break
2067
# Loop over fallback repositories asking them for texts - ignore
2068
# any missing from a particular fallback.
2069
for record in source.get_record_stream(missing_keys,
2070
self._ordering, True):
2071
if record.storage_kind == 'absent':
2072
# Not in thie particular stream, may be in one of the
2073
# other fallback vfs objects.
2074
continue
2075
missing_keys.remove(record.key)
2076
yield record
2077
2078
if self._raw_record_map is None:
2079
raise AssertionError('_raw_record_map should have been filled')
2080
first = True
2081
for key in self.keys:
2082
if key in self.nonlocal_keys:
2083
continue
2084
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2085
first = False
2086
2087
def _get_one_work(self, requested_key):
2088
# Now, if we have calculated everything already, just return the
2089
# desired text.
2090
if requested_key in self._contents_map:
2091
return self._contents_map[requested_key]
2092
# To simplify things, parse everything at once - code that wants one text
2093
# probably wants them all.
2094
1150
# FUTURE: This function could be improved for the 'extract many' case
2095
1151
# by tracking each component and only doing the copy when the number of
2096
1152
# children than need to apply delta's to it is > 1 or it is part of the
2097
1153
# final output.
2098
multiple_versions = len(self.keys) != 1
2099
if self._record_map is None:
2100
self._record_map = self.vf._raw_map_to_record_map(
2101
self._raw_record_map)
2102
record_map = self._record_map
2103
# raw_record_map is key:
2104
# Have read and parsed records at this point.
2105
for key in self.keys:
2106
if key in self.nonlocal_keys:
2107
# already handled
2108
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:
2109
1162
components = []
2110
cursor = key
1163
cursor = version_id
2111
1164
while cursor is not None:
2112
try:
2113
record, record_details, digest, next = record_map[cursor]
2114
except KeyError:
2115
raise RevisionNotPresent(cursor, self)
1165
record, record_details, digest, next = record_map[cursor]
2116
1166
components.append((cursor, record, record_details, digest))
1167
if cursor in content_map:
1168
break
2117
1169
cursor = next
2118
if cursor in self._contents_map:
2119
# no need to plan further back
2120
components.append((cursor, None, None, None))
2121
break
2122
1170
2123
1171
content = None
2124
1172
for (component_id, record, record_details,
2125
1173
digest) in reversed(components):
2126
if component_id in self._contents_map:
2127
content = self._contents_map[component_id]
1174
if component_id in content_map:
1175
content = content_map[component_id]
2128
1176
else:
2129
content, delta = self._factory.parse_record(key[-1],
1177
content, delta = self.factory.parse_record(version_id,
2130
1178
record, record_details, content,
2131
1179
copy_base_content=multiple_versions)
2132
1180
if multiple_versions:
2133
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
2134
1185
2135
1186
# digest here is the digest from the last applied component.
2136
1187
text = content.text()
2137
1188
actual_sha = sha_strings(text)
2138
1189
if actual_sha != digest:
2139
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2140
if multiple_versions:
2141
return self._contents_map[requested_key]
2142
else:
2143
return content
2144
2145
def _wire_bytes(self):
2146
"""Get the bytes to put on the wire for 'key'.
2147
2148
The first collection of bytes asked for returns the serialised
2149
raw_record_map and the additional details (key, parent) for key.
2150
Subsequent calls return just the additional details (key, parent).
2151
The wire storage_kind given for the first key is 'knit-delta-closure',
2152
For subsequent keys it is 'knit-delta-closure-ref'.
2153
2154
:param key: A key from the content generator.
2155
:return: Bytes to put on the wire.
2156
"""
2157
lines = []
2158
# kind marker for dispatch on the far side,
2159
lines.append('knit-delta-closure')
2160
# Annotated or not
2161
if self.vf._factory.annotated:
2162
lines.append('annotated')
2163
else:
2164
lines.append('')
2165
# then the list of keys
2166
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2167
if key not in self.nonlocal_keys]))
2168
# then the _raw_record_map in serialised form:
2169
map_byte_list = []
2170
# for each item in the map:
2171
# 1 line with key
2172
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2173
# one line with method
2174
# one line with noeol
2175
# one line with next ('' for None)
2176
# one line with byte count of the record bytes
2177
# the record bytes
2178
for key, (record_bytes, (method, noeol), next) in \
2179
self._raw_record_map.iteritems():
2180
key_bytes = '\x00'.join(key)
2181
parents = self.global_map.get(key, None)
2182
if parents is None:
2183
parent_bytes = 'None:'
2184
else:
2185
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2186
method_bytes = method
2187
if noeol:
2188
noeol_bytes = "T"
2189
else:
2190
noeol_bytes = "F"
2191
if next:
2192
next_bytes = '\x00'.join(next)
2193
else:
2194
next_bytes = ''
2195
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2196
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2197
len(record_bytes), record_bytes))
2198
map_bytes = ''.join(map_byte_list)
2199
lines.append(map_bytes)
2200
bytes = '\n'.join(lines)
2201
return bytes
2202
2203
2204
class _VFContentMapGenerator(_ContentMapGenerator):
2205
"""Content map generator reading from a VersionedFiles object."""
2206
2207
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2208
global_map=None, raw_record_map=None, ordering='unordered'):
2209
"""Create a _ContentMapGenerator.
2210
2211
:param versioned_files: The versioned files that the texts are being
2212
extracted from.
2213
:param keys: The keys to produce content maps for.
2214
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2215
which are known to not be in this knit, but rather in one of the
2216
fallback knits.
2217
:param global_map: The result of get_parent_map(keys) (or a supermap).
2218
This is required if get_record_stream() is to be used.
2219
:param raw_record_map: A unparsed raw record map to use for answering
2220
contents.
2221
"""
2222
_ContentMapGenerator.__init__(self, ordering=ordering)
2223
# The vf to source data from
2224
self.vf = versioned_files
2225
# The keys desired
2226
self.keys = list(keys)
2227
# Keys known to be in fallback vfs objects
2228
if nonlocal_keys is None:
2229
self.nonlocal_keys = set()
2230
else:
2231
self.nonlocal_keys = frozenset(nonlocal_keys)
2232
# Parents data for keys to be returned in get_record_stream
2233
self.global_map = global_map
2234
# The chunked lists for self.keys in text form
2235
self._text_map = {}
2236
# A cache of KnitContent objects used in extracting texts.
2237
self._contents_map = {}
2238
# All the knit records needed to assemble the requested keys as full
2239
# texts.
2240
self._record_map = None
2241
if raw_record_map is None:
2242
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2243
allow_missing=True)
2244
else:
2245
self._raw_record_map = raw_record_map
2246
# the factory for parsing records
2247
self._factory = self.vf._factory
2248
2249
2250
class _NetworkContentMapGenerator(_ContentMapGenerator):
2251
"""Content map generator sourced from a network stream."""
2252
2253
def __init__(self, bytes, line_end):
2254
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2255
self._bytes = bytes
2256
self.global_map = {}
2257
self._raw_record_map = {}
2258
self._contents_map = {}
2259
self._record_map = None
2260
self.nonlocal_keys = []
2261
# Get access to record parsing facilities
2262
self.vf = KnitVersionedFiles(None, None)
2263
start = line_end
2264
# Annotated or not
2265
line_end = bytes.find('\n', start)
2266
line = bytes[start:line_end]
2267
start = line_end + 1
2268
if line == 'annotated':
2269
self._factory = KnitAnnotateFactory()
2270
else:
2271
self._factory = KnitPlainFactory()
2272
# list of keys to emit in get_record_stream
2273
line_end = bytes.find('\n', start)
2274
line = bytes[start:line_end]
2275
start = line_end + 1
2276
self.keys = [
2277
tuple(segment.split('\x00')) for segment in line.split('\t')
2278
if segment]
2279
# now a loop until the end. XXX: It would be nice if this was just a
2280
# bunch of the same records as get_record_stream(..., False) gives, but
2281
# there is a decent sized gap stopping that at the moment.
2282
end = len(bytes)
2283
while start < end:
2284
# 1 line with key
2285
line_end = bytes.find('\n', start)
2286
key = tuple(bytes[start:line_end].split('\x00'))
2287
start = line_end + 1
2288
# 1 line with parents (None: for None, '' for ())
2289
line_end = bytes.find('\n', start)
2290
line = bytes[start:line_end]
2291
if line == 'None:':
2292
parents = None
2293
else:
2294
parents = tuple(
2295
[tuple(segment.split('\x00')) for segment in line.split('\t')
2296
if segment])
2297
self.global_map[key] = parents
2298
start = line_end + 1
2299
# one line with method
2300
line_end = bytes.find('\n', start)
2301
line = bytes[start:line_end]
2302
method = line
2303
start = line_end + 1
2304
# one line with noeol
2305
line_end = bytes.find('\n', start)
2306
line = bytes[start:line_end]
2307
noeol = line == "T"
2308
start = line_end + 1
2309
# one line with next ('' for None)
2310
line_end = bytes.find('\n', start)
2311
line = bytes[start:line_end]
2312
if not line:
2313
next = None
2314
else:
2315
next = tuple(bytes[start:line_end].split('\x00'))
2316
start = line_end + 1
2317
# one line with byte count of the record bytes
2318
line_end = bytes.find('\n', start)
2319
line = bytes[start:line_end]
2320
count = int(line)
2321
start = line_end + 1
2322
# the record bytes
2323
record_bytes = bytes[start:start+count]
2324
start = start + count
2325
# put it in the map
2326
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2327
2328
def get_record_stream(self):
2329
"""Get a record stream for for keys requested by the bytestream."""
2330
first = True
2331
for key in self.keys:
2332
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2333
first = False
2334
2335
def _wire_bytes(self):
2336
return self._bytes
2337
2338
2339
class _KndxIndex(object):
2340
"""Manages knit index files
2341
2342
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
2343
1339
fast lookups of revision information. The cursor of the index
2344
1340
file is always pointing to the end, making it easy to append
2345
1341
entries.
2355
1351
2356
1352
Duplicate entries may be written to the index for a single version id
2357
1353
if this is done then the latter one completely replaces the former:
2358
this allows updates to correct version and parent information.
1354
this allows updates to correct version and parent information.
2359
1355
Note that the two entries may share the delta, and that successive
2360
1356
annotations and references MUST point to the first entry.
2361
1357
2362
1358
The index file on disc contains a header, followed by one line per knit
2363
1359
record. The same revision can be present in an index file more than once.
2364
The first occurrence gets assigned a sequence number starting from 0.
2365
1360
The first occurrence gets assigned a sequence number starting from 0.
1361
2366
1362
The format of a single line is
2367
1363
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
2368
1364
REVISION_ID is a utf8-encoded revision id
2369
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
2370
1366
no-eol, line-delta, fulltext.
2371
1367
BYTE_OFFSET is the ascii representation of the byte offset in the data file
2372
that the compressed data starts at.
1368
that the the compressed data starts at.
2373
1369
LENGTH is the ascii representation of the length of the data file.
2374
1370
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
2375
1371
REVISION_ID.
2376
1372
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
2377
1373
revision id already in the knit that is a parent of REVISION_ID.
2378
1374
The ' :' marker is the end of record marker.
2379
1375
2380
1376
partial writes:
2381
1377
when a write is interrupted to the index file, it will result in a line
2382
1378
that does not end in ' :'. If the ' :' is not present at the end of a line,
2387
1383
to ensure that records always start on new lines even if the last write was
2388
1384
interrupted. As a result its normal for the last line in the index to be
2389
1385
missing a trailing newline. One can be added with no harmful effects.
2390
2391
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2392
where prefix is e.g. the (fileid,) for .texts instances or () for
2393
constant-mapped things like .revisions, and the old state is
2394
tuple(history_vector, cache_dict). This is used to prevent having an
2395
ABI change with the C extension that reads .kndx files.
2396
1386
"""
2397
1387
2398
1388
HEADER = "# bzr knit index 8\n"
2399
1389
2400
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2401
"""Create a _KndxIndex on transport using mapper."""
2402
self._transport = transport
2403
self._mapper = mapper
2404
self._get_scope = get_scope
2405
self._allow_writes = allow_writes
2406
self._is_locked = is_locked
2407
self._reset_cache()
2408
self.has_graph = True
2409
2410
def add_records(self, records, random_id=False, missing_compression_parents=False):
2411
"""Add multiple records to the index.
2412
2413
:param records: a list of tuples:
2414
(key, options, access_memo, parents).
2415
:param random_id: If True the ids being added were randomly generated
2416
and no check for existence will be performed.
2417
:param missing_compression_parents: If True the records being added are
2418
only compressed against texts already in the index (or inside
2419
records). If False the records all refer to unavailable texts (or
2420
texts inside records) as compression parents.
2421
"""
2422
if missing_compression_parents:
2423
# It might be nice to get the edge of the records. But keys isn't
2424
# _wrong_.
2425
keys = sorted(record[0] for record in records)
2426
raise errors.RevisionNotPresent(keys, self)
2427
paths = {}
2428
for record in records:
2429
key = record[0]
2430
prefix = key[:-1]
2431
path = self._mapper.map(key) + '.kndx'
2432
path_keys = paths.setdefault(path, (prefix, []))
2433
path_keys[1].append(record)
2434
for path in sorted(paths):
2435
prefix, path_keys = paths[path]
2436
self._load_prefixes([prefix])
2437
lines = []
2438
orig_history = self._kndx_cache[prefix][1][:]
2439
orig_cache = self._kndx_cache[prefix][0].copy()
2440
2441
try:
2442
for key, options, (_, pos, size), parents in path_keys:
2443
if parents is None:
2444
# kndx indices cannot be parentless.
2445
parents = ()
2446
line = "\n%s %s %s %s %s :" % (
2447
key[-1], ','.join(options), pos, size,
2448
self._dictionary_compress(parents))
2449
if type(line) is not str:
2450
raise AssertionError(
2451
'data must be utf8 was %s' % type(line))
2452
lines.append(line)
2453
self._cache_key(key, options, pos, size, parents)
2454
if len(orig_history):
2455
self._transport.append_bytes(path, ''.join(lines))
2456
else:
2457
self._init_index(path, lines)
2458
except:
2459
# If any problems happen, restore the original values and re-raise
2460
self._kndx_cache[prefix] = (orig_cache, orig_history)
2461
raise
2462
2463
def scan_unvalidated_index(self, graph_index):
2464
"""See _KnitGraphIndex.scan_unvalidated_index."""
2465
# Because kndx files do not support atomic insertion via separate index
2466
# files, they do not support this method.
2467
raise NotImplementedError(self.scan_unvalidated_index)
2468
2469
def get_missing_compression_parents(self):
2470
"""See _KnitGraphIndex.get_missing_compression_parents."""
2471
# Because kndx files do not support atomic insertion via separate index
2472
# files, they do not support this method.
2473
raise NotImplementedError(self.get_missing_compression_parents)
2474
2475
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):
2476
1394
"""Cache a version record in the history array and index cache.
2477
1395
2478
1396
This is inlined into _load_data for performance. KEEP IN SYNC.
2479
1397
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2480
1398
indexes).
2481
1399
"""
2482
prefix = key[:-1]
2483
version_id = key[-1]
2484
# last-element only for compatibilty with the C load_data.
2485
parents = tuple(parent[-1] for parent in parent_keys)
2486
for parent in parent_keys:
2487
if parent[:-1] != prefix:
2488
raise ValueError("mismatched prefixes for %r, %r" % (
2489
key, parent_keys))
2490
cache, history = self._kndx_cache[prefix]
2491
1400
# only want the _history index to reference the 1st index entry
2492
1401
# for version_id
2493
if version_id not in cache:
2494
index = len(history)
2495
history.append(version_id)
1402
if version_id not in self._cache:
1403
index = len(self._history)
1404
self._history.append(version_id)
2496
1405
else:
2497
index = cache[version_id][5]
2498
cache[version_id] = (version_id,
1406
index = self._cache[version_id][5]
1407
self._cache[version_id] = (version_id,
2499
1408
options,
2500
1409
pos,
2501
1410
size,
2502
1411
parents,
2503
1412
index)
2504
1413
2505
def check_header(self, fp):
2506
line = fp.readline()
2507
if line == '':
2508
# An empty file can actually be treated as though the file doesn't
2509
# exist yet.
2510
raise errors.NoSuchFile(self)
2511
if line != self.HEADER:
2512
raise KnitHeaderError(badline=line, filename=self)
2513
2514
def _check_read(self):
2515
if not self._is_locked():
2516
raise errors.ObjectNotLocked(self)
2517
if self._get_scope() != self._scope:
2518
self._reset_cache()
2519
2520
def _check_write_ok(self):
2521
"""Assert if not writes are permitted."""
2522
if not self._is_locked():
2523
raise errors.ObjectNotLocked(self)
2524
if self._get_scope() != self._scope:
2525
self._reset_cache()
2526
if self._mode != 'w':
2527
raise errors.ReadOnlyObjectDirtiedError(self)
2528
2529
def get_build_details(self, keys):
2530
"""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.
2531
1489
2532
1490
Ghosts are omitted from the result.
2533
1491
2534
:param keys: An iterable of keys.
2535
:return: A dict of key:(index_memo, compression_parent, parents,
2536
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).
2537
1495
index_memo
2538
1496
opaque structure to pass to read_records to extract the raw
2539
1497
data
2545
1503
extra information about the content which needs to be passed to
2546
1504
Factory.parse_record
2547
1505
"""
2548
parent_map = self.get_parent_map(keys)
2549
1506
result = {}
2550
for key in keys:
2551
if key not in parent_map:
2552
continue # Ghost
2553
method = self.get_method(key)
2554
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)
2555
1513
if method == 'fulltext':
2556
1514
compression_parent = None
2557
1515
else:
2558
1516
compression_parent = parents[0]
2559
noeol = 'no-eol' in self.get_options(key)
2560
index_memo = self.get_position(key)
2561
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,
2562
1520
parents, (method, noeol))
2563
1521
return result
2564
1522
2565
def get_method(self, key):
2566
"""Return compression method of specified key."""
2567
options = self.get_options(key)
2568
if 'fulltext' in options:
2569
return 'fulltext'
2570
elif 'line-delta' in options:
2571
return 'line-delta'
2572
else:
2573
raise errors.KnitIndexUnknownMethod(self, options)
2574
2575
def get_options(self, key):
2576
"""Return a list representing options.
2577
2578
e.g. ['foo', 'bar']
2579
"""
2580
prefix, suffix = self._split_key(key)
2581
self._load_prefixes([prefix])
2582
try:
2583
return self._kndx_cache[prefix][0][suffix][1]
2584
except KeyError:
2585
raise RevisionNotPresent(key, self)
2586
2587
def find_ancestry(self, keys):
2588
"""See CombinedGraphIndex.find_ancestry()"""
2589
prefixes = set(key[:-1] for key in keys)
2590
self._load_prefixes(prefixes)
2591
result = {}
2592
parent_map = {}
2593
missing_keys = set()
2594
pending_keys = list(keys)
2595
# This assumes that keys will not reference parents in a different
2596
# prefix, which is accurate so far.
2597
while pending_keys:
2598
key = pending_keys.pop()
2599
if key in parent_map:
2600
continue
2601
prefix = key[:-1]
2602
try:
2603
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2604
except KeyError:
2605
missing_keys.add(key)
2606
else:
2607
parent_keys = tuple([prefix + (suffix,)
2608
for suffix in suffix_parents])
2609
parent_map[key] = parent_keys
2610
pending_keys.extend([p for p in parent_keys
2611
if p not in parent_map])
2612
return parent_map, missing_keys
2613
2614
def get_parent_map(self, keys):
2615
"""Get a map of the parents of keys.
2616
2617
:param keys: The keys to look up parents for.
2618
:return: A mapping from keys to parents. Absent keys are absent from
2619
the mapping.
2620
"""
2621
# Parse what we need to up front, this potentially trades off I/O
2622
# locality (.kndx and .knit in the same block group for the same file
2623
# id) for less checking in inner loops.
2624
prefixes = set(key[:-1] for key in keys)
2625
self._load_prefixes(prefixes)
2626
result = {}
2627
for key in keys:
2628
prefix = key[:-1]
2629
try:
2630
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
1523
def iter_parents(self, version_ids):
1524
"""Iterate through the parents for many version ids.
1525
1526
:param version_ids: An iterable yielding version_ids.
1527
:return: An iterator that yields (version_id, parents). Requested
1528
version_ids not present in the versioned file are simply skipped.
1529
The order is undefined, allowing for different optimisations in
1530
the underlying implementation.
1531
"""
1532
for version_id in version_ids:
1533
try:
1534
yield version_id, tuple(self.get_parents(version_id))
2631
1535
except KeyError:
2632
1536
pass
2633
else:
2634
result[key] = tuple(prefix + (suffix,) for
2635
suffix in suffix_parents)
2636
return result
2637
2638
def get_position(self, key):
2639
"""Return details needed to access the version.
2640
2641
:return: a tuple (key, data position, size) to hand to the access
2642
logic to get the record.
2643
"""
2644
prefix, suffix = self._split_key(key)
2645
self._load_prefixes([prefix])
2646
entry = self._kndx_cache[prefix][0][suffix]
2647
return key, entry[2], entry[3]
2648
2649
has_key = _mod_index._has_key_from_parent_map
2650
2651
def _init_index(self, path, extra_lines=[]):
2652
"""Initialize an index."""
2653
sio = StringIO()
2654
sio.write(self.HEADER)
2655
sio.writelines(extra_lines)
2656
sio.seek(0)
2657
self._transport.put_file_non_atomic(path, sio,
2658
create_parent_dir=True)
2659
# self._create_parent_dir)
2660
# mode=self._file_mode,
2661
# dir_mode=self._dir_mode)
2662
2663
def keys(self):
2664
"""Get all the keys in the collection.
2665
2666
The keys are not ordered.
2667
"""
2668
result = set()
2669
# Identify all key prefixes.
2670
# XXX: A bit hacky, needs polish.
2671
if type(self._mapper) is ConstantMapper:
2672
prefixes = [()]
2673
else:
2674
relpaths = set()
2675
for quoted_relpath in self._transport.iter_files_recursive():
2676
path, ext = os.path.splitext(quoted_relpath)
2677
relpaths.add(path)
2678
prefixes = [self._mapper.unmap(path) for path in relpaths]
2679
self._load_prefixes(prefixes)
2680
for prefix in prefixes:
2681
for suffix in self._kndx_cache[prefix][1]:
2682
result.add(prefix + (suffix,))
2683
return result
2684
2685
def _load_prefixes(self, prefixes):
2686
"""Load the indices for prefixes."""
2687
self._check_read()
2688
for prefix in prefixes:
2689
if prefix not in self._kndx_cache:
2690
# the load_data interface writes to these variables.
2691
self._cache = {}
2692
self._history = []
2693
self._filename = prefix
2694
try:
2695
path = self._mapper.map(prefix) + '.kndx'
2696
fp = self._transport.get(path)
2697
try:
2698
# _load_data may raise NoSuchFile if the target knit is
2699
# completely empty.
2700
_load_data(self, fp)
2701
finally:
2702
fp.close()
2703
self._kndx_cache[prefix] = (self._cache, self._history)
2704
del self._cache
2705
del self._filename
2706
del self._history
2707
except NoSuchFile:
2708
self._kndx_cache[prefix] = ({}, [])
2709
if type(self._mapper) is ConstantMapper:
2710
# preserve behaviour for revisions.kndx etc.
2711
self._init_index(path)
2712
del self._cache
2713
del self._filename
2714
del self._history
2715
2716
missing_keys = _mod_index._missing_keys_from_parent_map
2717
2718
def _partition_keys(self, keys):
2719
"""Turn keys into a dict of prefix:suffix_list."""
2720
result = {}
2721
for key in keys:
2722
prefix_keys = result.setdefault(key[:-1], [])
2723
prefix_keys.append(key[-1])
2724
return result
2725
2726
def _dictionary_compress(self, keys):
2727
"""Dictionary compress keys.
2728
2729
:param keys: The keys to generate references to.
2730
:return: A string representation of keys. keys which are present are
2731
dictionary compressed, and others are emitted as fulltext with a
2732
'.' prefix.
2733
"""
2734
if not keys:
2735
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):
2736
1548
result_list = []
2737
prefix = keys[0][:-1]
2738
cache = self._kndx_cache[prefix][0]
2739
for key in keys:
2740
if key[:-1] != prefix:
2741
# kndx indices cannot refer across partitioned storage.
2742
raise ValueError("mismatched prefixes for %r" % keys)
2743
if key[-1] in cache:
1549
cache = self._cache
1550
for version in versions:
1551
if version in cache:
2744
1552
# -- inlined lookup() --
2745
result_list.append(str(cache[key[-1]][5]))
1553
result_list.append(str(cache[version][5]))
2746
1554
# -- end lookup () --
2747
1555
else:
2748
result_list.append('.' + key[-1])
1556
result_list.append('.' + version)
2749
1557
return ' '.join(result_list)
2750
1558
2751
def _reset_cache(self):
2752
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2753
# (cache_dict, history_vector) for parsed kndx files.
2754
self._kndx_cache = {}
2755
self._scope = self._get_scope()
2756
allow_writes = self._allow_writes()
2757
if allow_writes:
2758
self._mode = 'w'
2759
else:
2760
self._mode = 'r'
2761
2762
def _sort_keys_by_io(self, keys, positions):
2763
"""Figure out an optimal order to read the records for the given keys.
2764
2765
Sort keys, grouped by index and sorted by position.
2766
2767
:param keys: A list of keys whose records we want to read. This will be
2768
sorted 'in-place'.
2769
:param positions: A dict, such as the one returned by
2770
_get_components_positions()
2771
:return: None
2772
"""
2773
def get_sort_key(key):
2774
index_memo = positions[key][1]
2775
# Group by prefix and position. index_memo[0] is the key, so it is
2776
# (file_id, revision_id) and we don't want to sort on revision_id,
2777
# index_memo[1] is the position, and index_memo[2] is the size,
2778
# which doesn't matter for the sort
2779
return index_memo[0][:-1], index_memo[1]
2780
return keys.sort(key=get_sort_key)
2781
2782
_get_total_build_size = _get_total_build_size
2783
2784
def _split_key(self, key):
2785
"""Split key into a prefix and suffix."""
2786
return key[:-1], key[-1]
2787
2788
2789
class _KeyRefs(object):
2790
2791
def __init__(self, track_new_keys=False):
2792
# dict mapping 'key' to 'set of keys referring to that key'
2793
self.refs = {}
2794
if track_new_keys:
2795
# set remembering all new keys
2796
self.new_keys = set()
2797
else:
2798
self.new_keys = None
2799
2800
def clear(self):
2801
if self.refs:
2802
self.refs.clear()
2803
if self.new_keys:
2804
self.new_keys.clear()
2805
2806
def add_references(self, key, refs):
2807
# Record the new references
2808
for referenced in refs:
2809
try:
2810
needed_by = self.refs[referenced]
2811
except KeyError:
2812
needed_by = self.refs[referenced] = set()
2813
needed_by.add(key)
2814
# Discard references satisfied by the new key
2815
self.add_key(key)
2816
2817
def get_new_keys(self):
2818
return self.new_keys
2819
2820
def get_unsatisfied_refs(self):
2821
return self.refs.iterkeys()
2822
2823
def _satisfy_refs_for_key(self, key):
2824
try:
2825
del self.refs[key]
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]
2826
1624
except KeyError:
2827
# No keys depended on this key. That's ok.
2828
pass
2829
2830
def add_key(self, key):
2831
# satisfy refs for key, and remember that we've seen this key.
2832
self._satisfy_refs_for_key(key)
2833
if self.new_keys is not None:
2834
self.new_keys.add(key)
2835
2836
def satisfy_refs_for_keys(self, keys):
2837
for key in keys:
2838
self._satisfy_refs_for_key(key)
2839
2840
def get_referrers(self):
2841
result = set()
2842
for referrers in self.refs.itervalues():
2843
result.update(referrers)
2844
return result
2845
2846
2847
class _KnitGraphIndex(object):
2848
"""A KnitVersionedFiles index layered on GraphIndex."""
2849
2850
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2851
add_callback=None, track_external_parent_refs=False):
1625
raise RevisionNotPresent(version_id, self._filename)
1626
if 'fulltext' in options:
1627
return 'fulltext'
1628
else:
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']
1637
"""
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):
2852
1661
"""Construct a KnitGraphIndex on a graph_index.
2853
1662
2854
1663
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2855
:param is_locked: A callback to check whether the object should answer
2856
queries.
2857
1664
:param deltas: Allow delta-compressed records.
2858
:param parents: If True, record knits parents, if not do not record
2859
parents.
2860
1665
:param add_callback: If not None, allow additions to the index and call
2861
1666
this callback with a list of added GraphIndex nodes:
2862
1667
[(node, value, node_refs), ...]
2863
:param is_locked: A callback, returns True if the index is locked and
2864
thus usable.
2865
:param track_external_parent_refs: If True, record all external parent
2866
references parents from added records. These can be retrieved
2867
later by calling get_missing_parents().
1668
:param parents: If True, record knits parents, if not do not record
1669
parents.
2868
1670
"""
2869
self._add_callback = add_callback
2870
1671
self._graph_index = graph_index
2871
1672
self._deltas = deltas
1673
self._add_callback = add_callback
2872
1674
self._parents = parents
2873
1675
if deltas and not parents:
2874
# XXX: TODO: Delta tree and parent graph should be conceptually
2875
# separate.
2876
1676
raise KnitCorrupt(self, "Cannot do delta compression without "
2877
1677
"parent tracking.")
2878
self.has_graph = parents
2879
self._is_locked = is_locked
2880
self._missing_compression_parents = set()
2881
if track_external_parent_refs:
2882
self._key_dependencies = _KeyRefs()
2883
else:
2884
self._key_dependencies = None
2885
2886
def __repr__(self):
2887
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2888
2889
def add_records(self, records, random_id=False,
2890
missing_compression_parents=False):
2891
"""Add multiple records to the index.
2892
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
2893
1969
This function does not insert data into the Immutable GraphIndex
2894
1970
backing the KnitGraphIndex, instead it prepares data for insertion by
2895
1971
the caller and checks that it is safe to insert then calls
2896
1972
self._add_callback with the prepared GraphIndex nodes.
2897
1973
2898
:param records: a list of tuples:
2899
(key, options, access_memo, parents).
1974
:param versions: a list of tuples:
1975
(version_id, options, pos, size, parents).
2900
1976
:param random_id: If True the ids being added were randomly generated
2901
1977
and no check for existence will be performed.
2902
:param missing_compression_parents: If True the records being added are
2903
only compressed against texts already in the index (or inside
2904
records). If False the records all refer to unavailable texts (or
2905
texts inside records) as compression parents.
2906
1978
"""
2907
1979
if not self._add_callback:
2908
1980
raise errors.ReadOnlyError(self)
2909
1981
# we hope there are no repositories with inconsistent parentage
2910
1982
# anymore.
1983
# check for dups
2911
1984
2912
1985
keys = {}
2913
compression_parents = set()
2914
key_dependencies = self._key_dependencies
2915
for (key, options, access_memo, parents) in records:
2916
if self._parents:
2917
parents = tuple(parents)
2918
if key_dependencies is not None:
2919
key_dependencies.add_references(key, parents)
1986
for (version_id, options, access_memo, parents) in versions:
2920
1987
index, pos, size = access_memo
1988
key = (version_id, )
1989
parents = tuple((parent, ) for parent in parents)
2921
1990
if 'no-eol' in options:
2922
1991
value = 'N'
2923
1992
else:
2930
1999
if self._deltas:
2931
2000
if 'line-delta' in options:
2932
2001
node_refs = (parents, (parents[0],))
2933
if missing_compression_parents:
2934
compression_parents.add(parents[0])
2935
2002
else:
2936
2003
node_refs = (parents, ())
2937
2004
else:
2942
2009
"in parentless index.")
2943
2010
node_refs = ()
2944
2011
keys[key] = (value, node_refs)
2945
# check for dups
2946
2012
if not random_id:
2947
2013
present_nodes = self._get_entries(keys)
2948
2014
for (index, key, value, node_refs) in present_nodes:
2949
parents = node_refs[:1]
2950
# Sometimes these are passed as a list rather than a tuple
2951
passed = static_tuple.as_tuples(keys[key])
2952
passed_parents = passed[1][:1]
2953
if (value[0] != keys[key][0][0] or
2954
parents != passed_parents):
2955
node_refs = static_tuple.as_tuples(node_refs)
2956
raise KnitCorrupt(self, "inconsistent details in add_records"
2957
": %s %s" % ((value, node_refs), passed))
2015
if (value, node_refs) != keys[key]:
2016
raise KnitCorrupt(self, "inconsistent details in add_versions"
2017
": %s %s" % ((value, node_refs), keys[key]))
2958
2018
del keys[key]
2959
2019
result = []
2960
2020
if self._parents:
2964
2024
for key, (value, node_refs) in keys.iteritems():
2965
2025
result.append((key, value))
2966
2026
self._add_callback(result)
2967
if missing_compression_parents:
2968
# This may appear to be incorrect (it does not check for
2969
# compression parents that are in the existing graph index),
2970
# but such records won't have been buffered, so this is
2971
# actually correct: every entry when
2972
# missing_compression_parents==True either has a missing parent, or
2973
# a parent that is one of the keys in records.
2974
compression_parents.difference_update(keys)
2975
self._missing_compression_parents.update(compression_parents)
2976
# Adding records may have satisfied missing compression parents.
2977
self._missing_compression_parents.difference_update(keys)
2978
2979
def scan_unvalidated_index(self, graph_index):
2980
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2981
2982
This allows this _KnitGraphIndex to keep track of any missing
2983
compression parents we may want to have filled in to make those
2984
indices valid.
2985
2986
:param graph_index: A GraphIndex
2987
"""
2988
if self._deltas:
2989
new_missing = graph_index.external_references(ref_list_num=1)
2990
new_missing.difference_update(self.get_parent_map(new_missing))
2991
self._missing_compression_parents.update(new_missing)
2992
if self._key_dependencies is not None:
2993
# Add parent refs from graph_index (and discard parent refs that
2994
# the graph_index has).
2995
for node in graph_index.iter_all_entries():
2996
self._key_dependencies.add_references(node[1], node[3][0])
2997
2998
def get_missing_compression_parents(self):
2999
"""Return the keys of missing compression parents.
3000
3001
Missing compression parents occur when a record stream was missing
3002
basis texts, or a index was scanned that had missing basis texts.
3003
"""
3004
return frozenset(self._missing_compression_parents)
3005
3006
def get_missing_parents(self):
3007
"""Return the keys of missing parents."""
3008
# If updating this, you should also update
3009
# groupcompress._GCGraphIndex.get_missing_parents
3010
# We may have false positives, so filter those out.
3011
self._key_dependencies.satisfy_refs_for_keys(
3012
self.get_parent_map(self._key_dependencies.get_unsatisfied_refs()))
3013
return frozenset(self._key_dependencies.get_unsatisfied_refs())
3014
3015
def _check_read(self):
3016
"""raise if reads are not permitted."""
3017
if not self._is_locked():
3018
raise errors.ObjectNotLocked(self)
3019
3020
def _check_write_ok(self):
3021
"""Assert if writes are not permitted."""
3022
if not self._is_locked():
3023
raise errors.ObjectNotLocked(self)
3024
3025
def _compression_parent(self, an_entry):
3026
# return the key that an_entry is compressed against, or None
3027
# Grab the second parent list (as deltas implies parents currently)
3028
compression_parents = an_entry[3][1]
3029
if not compression_parents:
3030
return None
3031
if len(compression_parents) != 1:
3032
raise AssertionError(
3033
"Too many compression parents: %r" % compression_parents)
3034
return compression_parents[0]
3035
3036
def get_build_details(self, keys):
3037
"""Get the method, index_memo and compression parent for version_ids.
3038
3039
Ghosts are omitted from the result.
3040
3041
:param keys: An iterable of keys.
3042
:return: A dict of key:
3043
(index_memo, compression_parent, parents, record_details).
3044
index_memo
3045
opaque structure to pass to read_records to extract the raw
3046
data
3047
compression_parent
3048
Content that this record is built upon, may be None
3049
parents
3050
Logical parents of this node
3051
record_details
3052
extra information about the content which needs to be passed to
3053
Factory.parse_record
3054
"""
3055
self._check_read()
3056
result = {}
3057
entries = self._get_entries(keys, False)
3058
for entry in entries:
3059
key = entry[1]
3060
if not self._parents:
3061
parents = ()
3062
else:
3063
parents = entry[3][0]
3064
if not self._deltas:
3065
compression_parent_key = None
3066
else:
3067
compression_parent_key = self._compression_parent(entry)
3068
noeol = (entry[2][0] == 'N')
3069
if compression_parent_key:
3070
method = 'line-delta'
3071
else:
3072
method = 'fulltext'
3073
result[key] = (self._node_to_position(entry),
3074
compression_parent_key, parents,
3075
(method, noeol))
3076
return result
3077
3078
def _get_entries(self, keys, check_present=False):
3079
"""Get the entries for keys.
3080
3081
:param keys: An iterable of index key tuples.
3082
"""
3083
keys = set(keys)
3084
found_keys = set()
3085
if self._parents:
3086
for node in self._graph_index.iter_entries(keys):
3087
yield node
3088
found_keys.add(node[1])
3089
else:
3090
# adapt parentless index to the rest of the code.
3091
for node in self._graph_index.iter_entries(keys):
3092
yield node[0], node[1], node[2], ()
3093
found_keys.add(node[1])
3094
if check_present:
3095
missing_keys = keys.difference(found_keys)
3096
if missing_keys:
3097
raise RevisionNotPresent(missing_keys.pop(), self)
3098
3099
def get_method(self, key):
3100
"""Return compression method of specified key."""
3101
return self._get_method(self._get_node(key))
3102
3103
def _get_method(self, node):
3104
if not self._deltas:
3105
return 'fulltext'
3106
if self._compression_parent(node):
3107
return 'line-delta'
3108
else:
3109
return 'fulltext'
3110
3111
def _get_node(self, key):
3112
try:
3113
return list(self._get_entries([key]))[0]
3114
except IndexError:
3115
raise RevisionNotPresent(key, self)
3116
3117
def get_options(self, key):
3118
"""Return a list representing options.
3119
3120
e.g. ['foo', 'bar']
3121
"""
3122
node = self._get_node(key)
3123
options = [self._get_method(node)]
3124
if node[2][0] == 'N':
3125
options.append('no-eol')
3126
return options
3127
3128
def find_ancestry(self, keys):
3129
"""See CombinedGraphIndex.find_ancestry()"""
3130
return self._graph_index.find_ancestry(keys, 0)
3131
3132
def get_parent_map(self, keys):
3133
"""Get a map of the parents of keys.
3134
3135
:param keys: The keys to look up parents for.
3136
:return: A mapping from keys to parents. Absent keys are absent from
3137
the mapping.
3138
"""
3139
self._check_read()
3140
nodes = self._get_entries(keys)
3141
result = {}
3142
if self._parents:
3143
for node in nodes:
3144
result[node[1]] = node[3][0]
3145
else:
3146
for node in nodes:
3147
result[node[1]] = None
3148
return result
3149
3150
def get_position(self, key):
3151
"""Return details needed to access the version.
3152
3153
:return: a tuple (index, data position, size) to hand to the access
3154
logic to get the record.
3155
"""
3156
node = self._get_node(key)
3157
return self._node_to_position(node)
3158
3159
has_key = _mod_index._has_key_from_parent_map
3160
3161
def keys(self):
3162
"""Get all the keys in the collection.
3163
3164
The keys are not ordered.
3165
"""
3166
self._check_read()
3167
return [node[1] for node in self._graph_index.iter_all_entries()]
3168
3169
missing_keys = _mod_index._missing_keys_from_parent_map
3170
3171
def _node_to_position(self, node):
3172
"""Convert an index value to position details."""
3173
bits = node[2][1:].split(' ')
3174
return node[0], int(bits[0]), int(bits[1])
3175
3176
def _sort_keys_by_io(self, keys, positions):
3177
"""Figure out an optimal order to read the records for the given keys.
3178
3179
Sort keys, grouped by index and sorted by position.
3180
3181
:param keys: A list of keys whose records we want to read. This will be
3182
sorted 'in-place'.
3183
:param positions: A dict, such as the one returned by
3184
_get_components_positions()
3185
:return: None
3186
"""
3187
def get_index_memo(key):
3188
# index_memo is at offset [1]. It is made up of (GraphIndex,
3189
# position, size). GI is an object, which will be unique for each
3190
# pack file. This causes us to group by pack file, then sort by
3191
# position. Size doesn't matter, but it isn't worth breaking up the
3192
# tuple.
3193
return positions[key][1]
3194
return keys.sort(key=get_index_memo)
3195
3196
_get_total_build_size = _get_total_build_size
3197
3198
3199
class _KnitKeyAccess(object):
3200
"""Access to records in .knit files."""
3201
3202
def __init__(self, transport, mapper):
3203
"""Create a _KnitKeyAccess with transport and mapper.
3204
3205
:param transport: The transport the access object is rooted at.
3206
: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.
3207
2041
"""
3208
2042
self._transport = transport
3209
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
3210
2048
3211
def add_raw_records(self, key_sizes, raw_data):
2049
def add_raw_records(self, sizes, raw_data):
3212
2050
"""Add raw knit bytes to a storage area.
3213
2051
3214
The data is spooled to the container writer in one bytes-record per
3215
raw data item.
2052
The data is spooled to whereever the access method is storing data.
3216
2053
3217
:param sizes: An iterable of tuples containing the key and size of each
3218
raw data segment.
2054
:param sizes: An iterable containing the size of each raw data segment.
3219
2055
:param raw_data: A bytestring containing the data.
3220
:return: A list of memos to retrieve the record later. Each memo is an
3221
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3222
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.
3223
2059
"""
3224
if type(raw_data) is not str:
3225
raise AssertionError(
3226
'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
3227
2071
result = []
3228
offset = 0
3229
# TODO: This can be tuned for writing to sftp and other servers where
3230
# append() is relatively expensive by grouping the writes to each key
3231
# prefix.
3232
for key, size in key_sizes:
3233
path = self._mapper.map(key)
3234
try:
3235
base = self._transport.append_bytes(path + '.knit',
3236
raw_data[offset:offset+size])
3237
except errors.NoSuchFile:
3238
self._transport.mkdir(osutils.dirname(path))
3239
base = self._transport.append_bytes(path + '.knit',
3240
raw_data[offset:offset+size])
3241
# if base == 0:
3242
# chmod.
3243
offset += size
3244
result.append((key, base, size))
2072
for size in sizes:
2073
result.append((None, base, size))
2074
base += size
3245
2075
return result
3246
2076
3247
def flush(self):
3248
"""Flush pending writes on this access object.
3249
3250
For .knit files this is a no-op.
3251
"""
3252
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
3253
2098
3254
2099
def get_raw_records(self, memos_for_retrieval):
3255
2100
"""Get the raw bytes for a records.
3256
2101
3257
:param memos_for_retrieval: An iterable containing the access memo for
3258
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.
3259
2105
:return: An iterator over the bytes of the records.
3260
2106
"""
3261
# first pass, group into same-index request to minimise readv's issued.
3262
request_lists = []
3263
current_prefix = None
3264
for (key, offset, length) in memos_for_retrieval:
3265
if current_prefix == key[:-1]:
3266
current_list.append((offset, length))
3267
else:
3268
if current_prefix is not None:
3269
request_lists.append((current_prefix, current_list))
3270
current_prefix = key[:-1]
3271
current_list = [(offset, length)]
3272
# handle the last entry
3273
if current_prefix is not None:
3274
request_lists.append((current_prefix, current_list))
3275
for prefix, read_vector in request_lists:
3276
path = self._mapper.map(prefix) + '.knit'
3277
for pos, data in self._transport.readv(path, read_vector):
3278
yield data
3279
3280
3281
class _DirectPackAccess(object):
3282
"""Access to data in one or more packs with less translation."""
3283
3284
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
3285
"""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.
3286
2117
3287
2118
:param index_to_packs: A dict mapping index objects to the transport
3288
2119
and file names for obtaining data.
3289
:param reload_func: A function to call if we determine that the pack
3290
files have moved and we need to reload our caches. See
3291
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.
3292
2123
"""
3293
self._container_writer = None
3294
self._write_index = None
3295
self._indices = index_to_packs
3296
self._reload_func = reload_func
3297
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
3298
2131
3299
def add_raw_records(self, key_sizes, raw_data):
2132
def add_raw_records(self, sizes, raw_data):
3300
2133
"""Add raw knit bytes to a storage area.
3301
2134
3302
2135
The data is spooled to the container writer in one bytes-record per
3303
2136
raw data item.
3304
2137
3305
:param sizes: An iterable of tuples containing the key and size of each
3306
raw data segment.
2138
:param sizes: An iterable containing the size of each raw data segment.
3307
2139
:param raw_data: A bytestring containing the data.
3308
:return: A list of memos to retrieve the record later. Each memo is an
3309
opaque index memo. For _DirectPackAccess the memo is (index, pos,
3310
length), where the index field is the write_index object supplied
3311
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.
3312
2143
"""
3313
if type(raw_data) is not str:
3314
raise AssertionError(
3315
'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)
3316
2146
result = []
3317
2147
offset = 0
3318
for key, size in key_sizes:
3319
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(
3320
2150
raw_data[offset:offset+size], [])
3321
2151
offset += size
3322
result.append((self._write_index, p_offset, p_length))
2152
result.append((self.write_index, p_offset, p_length))
3323
2153
return result
3324
2154
3325
def flush(self):
3326
"""Flush pending writes on this access object.
2155
def create(self):
2156
"""Pack based knits do not get individually created."""
3327
2157
3328
This will flush any buffered writes to a NewPack.
3329
"""
3330
if self._flush_func is not None:
3331
self._flush_func()
3332
3333
2158
def get_raw_records(self, memos_for_retrieval):
3334
2159
"""Get the raw bytes for a records.
3335
2160
3336
:param memos_for_retrieval: An iterable containing the (index, pos,
2161
:param memos_for_retrieval: An iterable containing the (index, pos,
3337
2162
length) memo for retrieving the bytes. The Pack access method
3338
2163
looks up the pack to use for a given record in its index_to_pack
3339
2164
map.
3354
2179
if current_index is not None:
3355
2180
request_lists.append((current_index, current_list))
3356
2181
for index, offsets in request_lists:
3357
try:
3358
transport, path = self._indices[index]
3359
except KeyError:
3360
# A KeyError here indicates that someone has triggered an index
3361
# reload, and this index has gone missing, we need to start
3362
# over.
3363
if self._reload_func is None:
3364
# If we don't have a _reload_func there is nothing that can
3365
# be done
3366
raise
3367
raise errors.RetryWithNewPacks(index,
3368
reload_occurred=True,
3369
exc_info=sys.exc_info())
3370
try:
3371
reader = pack.make_readv_reader(transport, path, offsets)
3372
for names, read_func in reader.iter_records():
3373
yield read_func(None)
3374
except errors.NoSuchFile:
3375
# A NoSuchFile error indicates that a pack file has gone
3376
# missing on disk, we need to trigger a reload, and start over.
3377
if self._reload_func is None:
3378
raise
3379
raise errors.RetryWithNewPacks(transport.abspath(path),
3380
reload_occurred=False,
3381
exc_info=sys.exc_info())
3382
3383
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)):
3384
2192
"""Set a writer to use for adding data."""
3385
2193
if index is not None:
3386
self._indices[index] = transport_packname
3387
self._container_writer = writer
3388
self._write_index = index
3389
3390
def reload_or_raise(self, retry_exc):
3391
"""Try calling the reload function, or re-raise the original exception.
3392
3393
This should be called after _DirectPackAccess raises a
3394
RetryWithNewPacks exception. This function will handle the common logic
3395
of determining when the error is fatal versus being temporary.
3396
It will also make sure that the original exception is raised, rather
3397
than the RetryWithNewPacks exception.
3398
3399
If this function returns, then the calling function should retry
3400
whatever operation was being performed. Otherwise an exception will
3401
be raised.
3402
3403
:param retry_exc: A RetryWithNewPacks exception.
3404
"""
3405
is_error = False
3406
if self._reload_func is None:
3407
is_error = True
3408
elif not self._reload_func():
3409
# The reload claimed that nothing changed
3410
if not retry_exc.reload_occurred:
3411
# If there wasn't an earlier reload, then we really were
3412
# expecting to find changes. We didn't find them, so this is a
3413
# hard error
3414
is_error = True
3415
if is_error:
3416
exc_class, exc_value, exc_traceback = retry_exc.exc_info
3417
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)
3418
2853
3419
2854
3420
2855
# Deprecated, use PatienceSequenceMatcher instead
3429
2864
recommended.
3430
2865
"""
3431
2866
annotator = _KnitAnnotator(knit)
3432
return iter(annotator.annotate_flat(revision_id))
3433
3434
3435
class _KnitAnnotator(annotate.Annotator):
2867
return iter(annotator.annotate(revision_id))
2868
2869
2870
class _KnitAnnotator(object):
3436
2871
"""Build up the annotations for a text."""
3437
2872
3438
def __init__(self, vf):
3439
annotate.Annotator.__init__(self, vf)
3440
3441
# TODO: handle Nodes which cannot be extracted
3442
# self._ghosts = set()
3443
3444
# Map from (key, parent_key) => matching_blocks, should be 'use once'
3445
self._matching_blocks = {}
3446
3447
# KnitContent objects
3448
self._content_objects = {}
3449
# The number of children that depend on this fulltext content object
3450
self._num_compression_children = {}
3451
# Delta records that need their compression parent before they can be
3452
# expanded
3453
self._pending_deltas = {}
3454
# Fulltext records that are waiting for their parents fulltexts before
3455
# they can be yielded for annotation
3456
self._pending_annotation = {}
2873
def __init__(self, knit):
2874
self._knit = knit
2875
2876
# Content objects, differs from fulltexts because of how final newlines
2877
# are treated by knits. the content objects here will always have a
2878
# final newline
2879
self._fulltext_contents = {}
2880
2881
# Annotated lines of specific revisions
2882
self._annotated_lines = {}
2883
2884
# Track the raw data for nodes that we could not process yet.
2885
# This maps the revision_id of the base to a list of children that will
2886
# annotated from it.
2887
self._pending_children = {}
2888
2889
# Nodes which cannot be extracted
2890
self._ghosts = set()
2891
2892
# Track how many children this node has, so we know if we need to keep
2893
# it
2894
self._annotate_children = {}
2895
self._compression_children = {}
3457
2896
3458
2897
self._all_build_details = {}
3459
3460
def _get_build_graph(self, key):
2898
# The children => parent revision_id graph
2899
self._revision_id_graph = {}
2900
2901
self._heads_provider = None
2902
2903
self._nodes_to_keep_annotations = set()
2904
self._generations_until_keep = 100
2905
2906
def set_generations_until_keep(self, value):
2907
"""Set the number of generations before caching a node.
2908
2909
Setting this to -1 will cache every merge node, setting this higher
2910
will cache fewer nodes.
2911
"""
2912
self._generations_until_keep = value
2913
2914
def _add_fulltext_content(self, revision_id, content_obj):
2915
self._fulltext_contents[revision_id] = content_obj
2916
# TODO: jam 20080305 It might be good to check the sha1digest here
2917
return content_obj.text()
2918
2919
def _check_parents(self, child, nodes_to_annotate):
2920
"""Check if all parents have been processed.
2921
2922
:param child: A tuple of (rev_id, parents, raw_content)
2923
:param nodes_to_annotate: If child is ready, add it to
2924
nodes_to_annotate, otherwise put it back in self._pending_children
2925
"""
2926
for parent_id in child[1]:
2927
if (parent_id not in self._annotated_lines):
2928
# This parent is present, but another parent is missing
2929
self._pending_children.setdefault(parent_id,
2930
[]).append(child)
2931
break
2932
else:
2933
# This one is ready to be processed
2934
nodes_to_annotate.append(child)
2935
2936
def _add_annotation(self, revision_id, fulltext, parent_ids,
2937
left_matching_blocks=None):
2938
"""Add an annotation entry.
2939
2940
All parents should already have been annotated.
2941
:return: A list of children that now have their parents satisfied.
2942
"""
2943
a = self._annotated_lines
2944
annotated_parent_lines = [a[p] for p in parent_ids]
2945
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2946
fulltext, revision_id, left_matching_blocks,
2947
heads_provider=self._get_heads_provider()))
2948
self._annotated_lines[revision_id] = annotated_lines
2949
for p in parent_ids:
2950
ann_children = self._annotate_children[p]
2951
ann_children.remove(revision_id)
2952
if (not ann_children
2953
and p not in self._nodes_to_keep_annotations):
2954
del self._annotated_lines[p]
2955
del self._all_build_details[p]
2956
if p in self._fulltext_contents:
2957
del self._fulltext_contents[p]
2958
# Now that we've added this one, see if there are any pending
2959
# deltas to be done, certainly this parent is finished
2960
nodes_to_annotate = []
2961
for child in self._pending_children.pop(revision_id, []):
2962
self._check_parents(child, nodes_to_annotate)
2963
return nodes_to_annotate
2964
2965
def _get_build_graph(self, revision_id):
3461
2966
"""Get the graphs for building texts and annotations.
3462
2967
3463
2968
The data you need for creating a full text may be different than the
3465
2970
parents to create an annotation, but only need 1 parent to generate the
3466
2971
fulltext.)
3467
2972
3468
:return: A list of (key, index_memo) records, suitable for
3469
passing to read_records_iter to start reading in the raw data from
2973
:return: A list of (revision_id, index_memo) records, suitable for
2974
passing to read_records_iter to start reading in the raw data fro/
3470
2975
the pack file.
3471
2976
"""
3472
pending = set([key])
2977
if revision_id in self._annotated_lines:
2978
# Nothing to do
2979
return []
2980
pending = set([revision_id])
3473
2981
records = []
3474
ann_keys = set()
3475
self._num_needed_children[key] = 1
2982
generation = 0
2983
kept_generation = 0
3476
2984
while pending:
3477
2985
# get all pending nodes
2986
generation += 1
3478
2987
this_iteration = pending
3479
build_details = self._vf._index.get_build_details(this_iteration)
2988
build_details = self._knit._index.get_build_details(this_iteration)
3480
2989
self._all_build_details.update(build_details)
3481
# new_nodes = self._vf._index._get_entries(this_iteration)
2990
# new_nodes = self._knit._index._get_entries(this_iteration)
3482
2991
pending = set()
3483
for key, details in build_details.iteritems():
3484
(index_memo, compression_parent, parent_keys,
2992
for rev_id, details in build_details.iteritems():
2993
(index_memo, compression_parent, parents,
3485
2994
record_details) = details
3486
self._parent_map[key] = parent_keys
3487
self._heads_provider = None
3488
records.append((key, index_memo))
2995
self._revision_id_graph[rev_id] = parents
2996
records.append((rev_id, index_memo))
3489
2997
# Do we actually need to check _annotated_lines?
3490
pending.update([p for p in parent_keys
3491
if p not in self._all_build_details])
3492
if parent_keys:
3493
for parent_key in parent_keys:
3494
if parent_key in self._num_needed_children:
3495
self._num_needed_children[parent_key] += 1
3496
else:
3497
self._num_needed_children[parent_key] = 1
2998
pending.update(p for p in parents
2999
if p not in self._all_build_details)
3498
3000
if compression_parent:
3499
if compression_parent in self._num_compression_children:
3500
self._num_compression_children[compression_parent] += 1
3501
else:
3502
self._num_compression_children[compression_parent] = 1
3001
self._compression_children.setdefault(compression_parent,
3002
[]).append(rev_id)
3003
if parents:
3004
for parent in parents:
3005
self._annotate_children.setdefault(parent,
3006
[]).append(rev_id)
3007
num_gens = generation - kept_generation
3008
if ((num_gens >= self._generations_until_keep)
3009
and len(parents) > 1):
3010
kept_generation = generation
3011
self._nodes_to_keep_annotations.add(rev_id)
3503
3012
3504
3013
missing_versions = this_iteration.difference(build_details.keys())
3505
if missing_versions:
3506
for key in missing_versions:
3507
if key in self._parent_map and key in self._text_cache:
3508
# We already have this text ready, we just need to
3509
# yield it later so we get it annotated
3510
ann_keys.add(key)
3511
parent_keys = self._parent_map[key]
3512
for parent_key in parent_keys:
3513
if parent_key in self._num_needed_children:
3514
self._num_needed_children[parent_key] += 1
3515
else:
3516
self._num_needed_children[parent_key] = 1
3517
pending.update([p for p in parent_keys
3518
if p not in self._all_build_details])
3519
else:
3520
raise errors.RevisionNotPresent(key, self._vf)
3014
self._ghosts.update(missing_versions)
3015
for missing_version in missing_versions:
3016
# add a key, no parents
3017
self._revision_id_graph[missing_version] = ()
3018
pending.discard(missing_version) # don't look for it
3019
# XXX: This should probably be a real exception, as it is a data
3020
# inconsistency
3021
assert not self._ghosts.intersection(self._compression_children), \
3022
"We cannot have nodes which have a compression parent of a ghost."
3023
# Cleanout anything that depends on a ghost so that we don't wait for
3024
# the ghost to show up
3025
for node in self._ghosts:
3026
if node in self._annotate_children:
3027
# We won't be building this node
3028
del self._annotate_children[node]
3521
3029
# Generally we will want to read the records in reverse order, because
3522
3030
# we find the parent nodes after the children
3523
3031
records.reverse()
3524
return records, ann_keys
3525
3526
def _get_needed_texts(self, key, pb=None):
3527
# if True or len(self._vf._fallback_vfs) > 0:
3528
if len(self._vf._fallback_vfs) > 0:
3529
# If we have fallbacks, go to the generic path
3530
for v in annotate.Annotator._get_needed_texts(self, key, pb=pb):
3531
yield v
3532
return
3533
while True:
3534
try:
3535
records, ann_keys = self._get_build_graph(key)
3536
for idx, (sub_key, text, num_lines) in enumerate(
3537
self._extract_texts(records)):
3538
if pb is not None:
3539
pb.update('annotating', idx, len(records))
3540
yield sub_key, text, num_lines
3541
for sub_key in ann_keys:
3542
text = self._text_cache[sub_key]
3543
num_lines = len(text) # bad assumption
3544
yield sub_key, text, num_lines
3545
return
3546
except errors.RetryWithNewPacks, e:
3547
self._vf._access.reload_or_raise(e)
3548
# The cached build_details are no longer valid
3549
self._all_build_details.clear()
3550
3551
def _cache_delta_blocks(self, key, compression_parent, delta, lines):
3552
parent_lines = self._text_cache[compression_parent]
3553
blocks = list(KnitContent.get_line_delta_blocks(delta, parent_lines, lines))
3554
self._matching_blocks[(key, compression_parent)] = blocks
3555
3556
def _expand_record(self, key, parent_keys, compression_parent, record,
3557
record_details):
3558
delta = None
3559
if compression_parent:
3560
if compression_parent not in self._content_objects:
3561
# Waiting for the parent
3562
self._pending_deltas.setdefault(compression_parent, []).append(
3563
(key, parent_keys, record, record_details))
3564
return None
3565
# We have the basis parent, so expand the delta
3566
num = self._num_compression_children[compression_parent]
3567
num -= 1
3568
if num == 0:
3569
base_content = self._content_objects.pop(compression_parent)
3570
self._num_compression_children.pop(compression_parent)
3571
else:
3572
self._num_compression_children[compression_parent] = num
3573
base_content = self._content_objects[compression_parent]
3574
# It is tempting to want to copy_base_content=False for the last
3575
# child object. However, whenever noeol=False,
3576
# self._text_cache[parent_key] is content._lines. So mutating it
3577
# gives very bad results.
3578
# The alternative is to copy the lines into text cache, but then we
3579
# are copying anyway, so just do it here.
3580
content, delta = self._vf._factory.parse_record(
3581
key, record, record_details, base_content,
3582
copy_base_content=True)
3583
else:
3584
# Fulltext record
3585
content, _ = self._vf._factory.parse_record(
3586
key, record, record_details, None)
3587
if self._num_compression_children.get(key, 0) > 0:
3588
self._content_objects[key] = content
3589
lines = content.text()
3590
self._text_cache[key] = lines
3591
if delta is not None:
3592
self._cache_delta_blocks(key, compression_parent, delta, lines)
3593
return lines
3594
3595
def _get_parent_annotations_and_matches(self, key, text, parent_key):
3596
"""Get the list of annotations for the parent, and the matching lines.
3597
3598
:param text: The opaque value given by _get_needed_texts
3599
:param parent_key: The key for the parent text
3600
:return: (parent_annotations, matching_blocks)
3601
parent_annotations is a list as long as the number of lines in
3602
parent
3603
matching_blocks is a list of (parent_idx, text_idx, len) tuples
3604
indicating which lines match between the two texts
3605
"""
3606
block_key = (key, parent_key)
3607
if block_key in self._matching_blocks:
3608
blocks = self._matching_blocks.pop(block_key)
3609
parent_annotations = self._annotations_cache[parent_key]
3610
return parent_annotations, blocks
3611
return annotate.Annotator._get_parent_annotations_and_matches(self,
3612
key, text, parent_key)
3613
3614
def _process_pending(self, key):
3615
"""The content for 'key' was just processed.
3616
3617
Determine if there is any more pending work to be processed.
3618
"""
3619
to_return = []
3620
if key in self._pending_deltas:
3621
compression_parent = key
3622
children = self._pending_deltas.pop(key)
3623
for child_key, parent_keys, record, record_details in children:
3624
lines = self._expand_record(child_key, parent_keys,
3625
compression_parent,
3626
record, record_details)
3627
if self._check_ready_for_annotations(child_key, parent_keys):
3628
to_return.append(child_key)
3629
# Also check any children that are waiting for this parent to be
3630
# annotation ready
3631
if key in self._pending_annotation:
3632
children = self._pending_annotation.pop(key)
3633
to_return.extend([c for c, p_keys in children
3634
if self._check_ready_for_annotations(c, p_keys)])
3635
return to_return
3636
3637
def _check_ready_for_annotations(self, key, parent_keys):
3638
"""return true if this text is ready to be yielded.
3639
3640
Otherwise, this will return False, and queue the text into
3641
self._pending_annotation
3642
"""
3643
for parent_key in parent_keys:
3644
if parent_key not in self._annotations_cache:
3645
# still waiting on at least one parent text, so queue it up
3646
# Note that if there are multiple parents, we need to wait
3647
# for all of them.
3648
self._pending_annotation.setdefault(parent_key,
3649
[]).append((key, parent_keys))
3650
return False
3651
return True
3652
3653
def _extract_texts(self, records):
3654
"""Extract the various texts needed based on records"""
3032
return records
3033
3034
def _annotate_records(self, records):
3035
"""Build the annotations for the listed records."""
3655
3036
# We iterate in the order read, rather than a strict order requested
3656
3037
# However, process what we can, and put off to the side things that
3657
3038
# still need parents, cleaning them up when those parents are
3658
3039
# processed.
3659
# Basic data flow:
3660
# 1) As 'records' are read, see if we can expand these records into
3661
# Content objects (and thus lines)
3662
# 2) If a given line-delta is waiting on its compression parent, it
3663
# gets queued up into self._pending_deltas, otherwise we expand
3664
# it, and put it into self._text_cache and self._content_objects
3665
# 3) If we expanded the text, we will then check to see if all
3666
# parents have also been processed. If so, this text gets yielded,
3667
# else this record gets set aside into pending_annotation
3668
# 4) Further, if we expanded the text in (2), we will then check to
3669
# see if there are any children in self._pending_deltas waiting to
3670
# also be processed. If so, we go back to (2) for those
3671
# 5) Further again, if we yielded the text, we can then check if that
3672
# 'unlocks' any of the texts in pending_annotations, which should
3673
# then get yielded as well
3674
# Note that both steps 4 and 5 are 'recursive' in that unlocking one
3675
# compression child could unlock yet another, and yielding a fulltext
3676
# will also 'unlock' the children that are waiting on that annotation.
3677
# (Though also, unlocking 1 parent's fulltext, does not unlock a child
3678
# if other parents are also waiting.)
3679
# We want to yield content before expanding child content objects, so
3680
# that we know when we can re-use the content lines, and the annotation
3681
# code can know when it can stop caching fulltexts, as well.
3682
3683
# Children that are missing their compression parent
3684
pending_deltas = {}
3685
for (key, record, digest) in self._vf._read_records_iter(records):
3686
# ghosts?
3687
details = self._all_build_details[key]
3688
(_, compression_parent, parent_keys, record_details) = details
3689
lines = self._expand_record(key, parent_keys, compression_parent,
3690
record, record_details)
3691
if lines is None:
3692
# Pending delta should be queued up
3040
for (rev_id, record,
3041
digest) in self._knit._data.read_records_iter(records):
3042
if rev_id in self._annotated_lines:
3693
3043
continue
3694
# At this point, we may be able to yield this content, if all
3695
# parents are also finished
3696
yield_this_text = self._check_ready_for_annotations(key,
3697
parent_keys)
3698
if yield_this_text:
3699
# All parents present
3700
yield key, lines, len(lines)
3701
to_process = self._process_pending(key)
3702
while to_process:
3703
this_process = to_process
3704
to_process = []
3705
for key in this_process:
3706
lines = self._text_cache[key]
3707
yield key, lines, len(lines)
3708
to_process.extend(self._process_pending(key))
3044
parent_ids = self._revision_id_graph[rev_id]
3045
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3046
details = self._all_build_details[rev_id]
3047
(index_memo, compression_parent, parents,
3048
record_details) = details
3049
nodes_to_annotate = []
3050
# TODO: Remove the punning between compression parents, and
3051
# parent_ids, we should be able to do this without assuming
3052
# the build order
3053
if len(parent_ids) == 0:
3054
# There are no parents for this node, so just add it
3055
# TODO: This probably needs to be decoupled
3056
assert compression_parent is None
3057
fulltext_content, delta = self._knit.factory.parse_record(
3058
rev_id, record, record_details, None)
3059
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3060
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3061
parent_ids, left_matching_blocks=None))
3062
else:
3063
child = (rev_id, parent_ids, record)
3064
# Check if all the parents are present
3065
self._check_parents(child, nodes_to_annotate)
3066
while nodes_to_annotate:
3067
# Should we use a queue here instead of a stack?
3068
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3069
(index_memo, compression_parent, parents,
3070
record_details) = self._all_build_details[rev_id]
3071
if compression_parent is not None:
3072
comp_children = self._compression_children[compression_parent]
3073
assert rev_id in comp_children
3074
# If there is only 1 child, it is safe to reuse this
3075
# content
3076
reuse_content = (len(comp_children) == 1
3077
and compression_parent not in
3078
self._nodes_to_keep_annotations)
3079
if reuse_content:
3080
# Remove it from the cache since it will be changing
3081
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3082
# Make sure to copy the fulltext since it might be
3083
# modified
3084
parent_fulltext = list(parent_fulltext_content.text())
3085
else:
3086
parent_fulltext_content = self._fulltext_contents[compression_parent]
3087
parent_fulltext = parent_fulltext_content.text()
3088
comp_children.remove(rev_id)
3089
fulltext_content, delta = self._knit.factory.parse_record(
3090
rev_id, record, record_details,
3091
parent_fulltext_content,
3092
copy_base_content=(not reuse_content))
3093
fulltext = self._add_fulltext_content(rev_id,
3094
fulltext_content)
3095
blocks = KnitContent.get_line_delta_blocks(delta,
3096
parent_fulltext, fulltext)
3097
else:
3098
fulltext_content = self._knit.factory.parse_fulltext(
3099
record, rev_id)
3100
fulltext = self._add_fulltext_content(rev_id,
3101
fulltext_content)
3102
blocks = None
3103
nodes_to_annotate.extend(
3104
self._add_annotation(rev_id, fulltext, parent_ids,
3105
left_matching_blocks=blocks))
3106
3107
def _get_heads_provider(self):
3108
"""Create a heads provider for resolving ancestry issues."""
3109
if self._heads_provider is not None:
3110
return self._heads_provider
3111
parent_provider = _mod_graph.DictParentsProvider(
3112
self._revision_id_graph)
3113
graph_obj = _mod_graph.Graph(parent_provider)
3114
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3115
self._heads_provider = head_cache
3116
return head_cache
3117
3118
def annotate(self, revision_id):
3119
"""Return the annotated fulltext at the given revision.
3120
3121
:param revision_id: The revision id for this file
3122
"""
3123
records = self._get_build_graph(revision_id)
3124
if revision_id in self._ghosts:
3125
raise errors.RevisionNotPresent(revision_id, self._knit)
3126
self._annotate_records(records)
3127
return self._annotated_lines[revision_id]
3128
3709
3129
3710
3130
try:
3711
from bzrlib._knit_load_data_pyx import _load_data_c as _load_data
3712
except ImportError, e:
3713
osutils.failed_to_load_extension(e)
3131
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3132
except ImportError:
3714
3133
from bzrlib._knit_load_data_py import _load_data_py as _load_data
Loggerhead is a web-based interface for Breezy
Version: 2.0.1