« back to all changes in this revision
Viewing changes to bzrlib/knit.py
- browse files at revision 1594.2.4
- compare with another revision
- compare with revision 1185.72.12
- download diff
- download tarball
- view history from revision 1594.2.4
- Committer: Robert Collins
- Date: 2006-03-08 00:20:04 UTC
- mfrom: (1563.2.42 versioned-file-knit-stores)
- mto: (1596.2.3 integration) (1594.3.1 versioned-file-performance)
- mto: This revision was merged to the branch mainline in revision 1596.
- Revision ID: robertc@robertcollins.net-20060308002004-a9950b2fb75d14de
Merge in knit repository use of knits - still not a stable format, but can be experimented with.
- files added:
- bzrlib/inter.py
- bzrlib/store/revision
- bzrlib/store/versioned
- bzrlib/tests/interversionedfile_implementations
- bzrlib/tests/revisionstore_implementations
- files removed:
- bzrlib/tests/test_reweave.py
- files renamed:
- bzrlib/store/weave.py => bzrlib/store/versioned/__init__.py
- files modified:
- BRANCH.TODO
added
removed
bzrlib/knit.py
1
# Copyright (C) 2005, 2006 by Canonical Ltd
2
# Written by Martin Pool.
3
# Modified by Johan Rydberg <jrydberg@gnu.org>
4
# Modified by Robert Collins <robert.collins@canonical.com>
5
#
6
# This program is free software; you can redistribute it and/or modify
7
# it under the terms of the GNU General Public License as published by
8
# the Free Software Foundation; either version 2 of the License, or
9
# (at your option) any later version.
10
#
11
# This program is distributed in the hope that it will be useful,
12
# but WITHOUT ANY WARRANTY; without even the implied warranty of
13
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14
# GNU General Public License for more details.
15
#
16
# You should have received a copy of the GNU General Public License
17
# along with this program; if not, write to the Free Software
18
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19
20
"""Knit versionedfile implementation.
21
22
A knit is a versioned file implementation that supports efficient append only
23
updates.
24
25
Knit file layout:
26
lifeless: the data file is made up of "delta records". each delta record has a delta header
27
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
28
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
29
end-marker; simply "end VERSION"
30
31
delta can be line or full contents.a
32
... the 8's there are the index number of the annotation.
33
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
59,59,3
35
8
36
8 if ie.executable:
37
8 e.set('executable', 'yes')
38
130,130,2
39
8 if elt.get('executable') == 'yes':
40
8 ie.executable = True
41
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
43
44
whats in an index:
45
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
46
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
47
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
48
09:33 < lifeless> right
49
09:33 < jrydberg> lifeless: the position and size is the range in the data file
50
51
52
so the index sequence is the dictionary compressed sequence number used
53
in the deltas to provide line annotation
54
55
"""
56
57
# TODOS:
58
# 10:16 < lifeless> make partial index writes safe
59
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
60
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
61
# always' approach.
62
# move sha1 out of the content so that join is faster at verifying parents
63
# record content length ?
64
65
66
from cStringIO import StringIO
67
import difflib
68
from difflib import SequenceMatcher
69
from gzip import GzipFile
70
import os
71
72
import bzrlib.errors as errors
73
from bzrlib.errors import FileExists, NoSuchFile, KnitError, \
74
InvalidRevisionId, KnitCorrupt, KnitHeaderError, \
75
RevisionNotPresent, RevisionAlreadyPresent
76
from bzrlib.trace import mutter
77
from bzrlib.osutils import contains_whitespace, contains_linebreaks, \
78
sha_strings
79
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
80
from bzrlib.tsort import topo_sort
81
82
83
# TODO: Split out code specific to this format into an associated object.
84
85
# TODO: Can we put in some kind of value to check that the index and data
86
# files belong together?
87
88
# TODO: accomodate binaries, perhaps by storing a byte count
89
90
# TODO: function to check whole file
91
92
# TODO: atomically append data, then measure backwards from the cursor
93
# position after writing to work out where it was located. we may need to
94
# bypass python file buffering.
95
96
DATA_SUFFIX = '.knit'
97
INDEX_SUFFIX = '.kndx'
98
99
100
class KnitContent(object):
101
"""Content of a knit version to which deltas can be applied."""
102
103
def __init__(self, lines):
104
self._lines = lines
105
106
def annotate_iter(self):
107
"""Yield tuples of (origin, text) for each content line."""
108
for origin, text in self._lines:
109
yield origin, text
110
111
def annotate(self):
112
"""Return a list of (origin, text) tuples."""
113
return list(self.annotate_iter())
114
115
def apply_delta(self, delta):
116
"""Apply delta to this content."""
117
offset = 0
118
for start, end, count, lines in delta:
119
self._lines[offset+start:offset+end] = lines
120
offset = offset + (start - end) + count
121
122
def line_delta_iter(self, new_lines):
123
"""Generate line-based delta from new_lines to this content."""
124
new_texts = [text for origin, text in new_lines._lines]
125
old_texts = [text for origin, text in self._lines]
126
s = difflib.SequenceMatcher(None, old_texts, new_texts)
127
for op in s.get_opcodes():
128
if op[0] == 'equal':
129
continue
130
yield (op[1], op[2], op[4]-op[3], new_lines._lines[op[3]:op[4]])
131
132
def line_delta(self, new_lines):
133
return list(self.line_delta_iter(new_lines))
134
135
def text(self):
136
return [text for origin, text in self._lines]
137
138
139
class _KnitFactory(object):
140
"""Base factory for creating content objects."""
141
142
def make(self, lines, version):
143
num_lines = len(lines)
144
return KnitContent(zip([version] * num_lines, lines))
145
146
147
class KnitAnnotateFactory(_KnitFactory):
148
"""Factory for creating annotated Content objects."""
149
150
annotated = True
151
152
def parse_fulltext(self, content, version):
153
lines = []
154
for line in content:
155
origin, text = line.split(' ', 1)
156
lines.append((int(origin), text))
157
return KnitContent(lines)
158
159
def parse_line_delta_iter(self, lines):
160
while lines:
161
header = lines.pop(0)
162
start, end, c = [int(n) for n in header.split(',')]
163
contents = []
164
for i in range(c):
165
origin, text = lines.pop(0).split(' ', 1)
166
contents.append((int(origin), text))
167
yield start, end, c, contents
168
169
def parse_line_delta(self, lines, version):
170
return list(self.parse_line_delta_iter(lines))
171
172
def lower_fulltext(self, content):
173
return ['%d %s' % (o, t) for o, t in content._lines]
174
175
def lower_line_delta(self, delta):
176
out = []
177
for start, end, c, lines in delta:
178
out.append('%d,%d,%d\n' % (start, end, c))
179
for origin, text in lines:
180
out.append('%d %s' % (origin, text))
181
return out
182
183
184
class KnitPlainFactory(_KnitFactory):
185
"""Factory for creating plain Content objects."""
186
187
annotated = False
188
189
def parse_fulltext(self, content, version):
190
return self.make(content, version)
191
192
def parse_line_delta_iter(self, lines, version):
193
while lines:
194
header = lines.pop(0)
195
start, end, c = [int(n) for n in header.split(',')]
196
yield start, end, c, zip([version] * c, lines[:c])
197
del lines[:c]
198
199
def parse_line_delta(self, lines, version):
200
return list(self.parse_line_delta_iter(lines, version))
201
202
def lower_fulltext(self, content):
203
return content.text()
204
205
def lower_line_delta(self, delta):
206
out = []
207
for start, end, c, lines in delta:
208
out.append('%d,%d,%d\n' % (start, end, c))
209
out.extend([text for origin, text in lines])
210
return out
211
212
213
def make_empty_knit(transport, relpath):
214
"""Construct a empty knit at the specified location."""
215
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
216
k._data._open_file()
217
218
219
class KnitVersionedFile(VersionedFile):
220
"""Weave-like structure with faster random access.
221
222
A knit stores a number of texts and a summary of the relationships
223
between them. Texts are identified by a string version-id. Texts
224
are normally stored and retrieved as a series of lines, but can
225
also be passed as single strings.
226
227
Lines are stored with the trailing newline (if any) included, to
228
avoid special cases for files with no final newline. Lines are
229
composed of 8-bit characters, not unicode. The combination of
230
these approaches should mean any 'binary' file can be safely
231
stored and retrieved.
232
"""
233
234
def __init__(self, relpath, transport, file_mode=None, access_mode=None, factory=None,
235
basis_knit=None, delta=True, create=False):
236
"""Construct a knit at location specified by relpath.
237
238
:param create: If not True, only open an existing knit.
239
"""
240
if access_mode is None:
241
access_mode = 'w'
242
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
243
assert not basis_knit or isinstance(basis_knit, KnitVersionedFile), \
244
type(basis_knit)
245
246
self.transport = transport
247
self.filename = relpath
248
self.basis_knit = basis_knit
249
self.factory = factory or KnitAnnotateFactory()
250
self.writable = (access_mode == 'w')
251
self.delta = delta
252
253
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
254
access_mode, create=create)
255
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
256
access_mode, create=not len(self.versions()))
257
258
def copy_to(self, name, transport):
259
"""See VersionedFile.copy_to()."""
260
# copy the current index to a temp index to avoid racing with local
261
# writes
262
transport.put(name + INDEX_SUFFIX + '.tmp', self.transport.get(self._index._filename))
263
# copy the data file
264
transport.put(name + DATA_SUFFIX, self._data._open_file())
265
# rename the copied index into place
266
transport.rename(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
267
268
def create_empty(self, name, transport, mode=None):
269
return KnitVersionedFile(name, transport, factory=self.factory, delta=self.delta, create=True)
270
271
@staticmethod
272
def get_suffixes():
273
"""See VersionedFile.get_suffixes()."""
274
return [DATA_SUFFIX, INDEX_SUFFIX]
275
276
def versions(self):
277
"""See VersionedFile.versions."""
278
return self._index.get_versions()
279
280
def has_version(self, version_id):
281
"""See VersionedFile.has_version."""
282
return self._index.has_version(version_id)
283
284
__contains__ = has_version
285
286
def _merge_annotations(self, content, parents):
287
"""Merge annotations for content. This is done by comparing
288
the annotations based on changed to the text."""
289
for parent_id in parents:
290
merge_content = self._get_content(parent_id)
291
seq = SequenceMatcher(None, merge_content.text(), content.text())
292
for i, j, n in seq.get_matching_blocks():
293
if n == 0:
294
continue
295
content._lines[j:j+n] = merge_content._lines[i:i+n]
296
297
def _get_components(self, version_id):
298
"""Return a list of (version_id, method, data) tuples that
299
makes up version specified by version_id of the knit.
300
301
The components should be applied in the order of the returned
302
list.
303
304
The basis knit will be used to the largest extent possible
305
since it is assumed that accesses to it is faster.
306
"""
307
# needed_revisions holds a list of (method, version_id) of
308
# versions that is needed to be fetched to construct the final
309
# version of the file.
310
#
311
# basis_revisions is a list of versions that needs to be
312
# fetched but exists in the basis knit.
313
314
basis = self.basis_knit
315
needed_versions = []
316
basis_versions = []
317
cursor = version_id
318
319
while 1:
320
picked_knit = self
321
if basis and basis._index.has_version(cursor):
322
picked_knit = basis
323
basis_versions.append(cursor)
324
method = picked_knit._index.get_method(cursor)
325
needed_versions.append((method, cursor))
326
if method == 'fulltext':
327
break
328
cursor = picked_knit.get_parents(cursor)[0]
329
330
components = {}
331
if basis_versions:
332
records = []
333
for comp_id in basis_versions:
334
data_pos, data_size = basis._index.get_data_position(comp_id)
335
records.append((piece_id, data_pos, data_size))
336
components.update(basis._data.read_records(records))
337
338
records = []
339
for comp_id in [vid for method, vid in needed_versions
340
if vid not in basis_versions]:
341
data_pos, data_size = self._index.get_position(comp_id)
342
records.append((comp_id, data_pos, data_size))
343
components.update(self._data.read_records(records))
344
345
# get_data_records returns a mapping with the version id as
346
# index and the value as data. The order the components need
347
# to be applied is held by needed_versions (reversed).
348
out = []
349
for method, comp_id in reversed(needed_versions):
350
out.append((comp_id, method, components[comp_id]))
351
352
return out
353
354
def _get_content(self, version_id):
355
"""Returns a content object that makes up the specified
356
version."""
357
if not self.has_version(version_id):
358
raise RevisionNotPresent(version_id, self.filename)
359
360
if self.basis_knit and version_id in self.basis_knit:
361
return self.basis_knit._get_content(version_id)
362
363
content = None
364
components = self._get_components(version_id)
365
for component_id, method, (data, digest) in components:
366
version_idx = self._index.lookup(component_id)
367
if method == 'fulltext':
368
assert content is None
369
content = self.factory.parse_fulltext(data, version_idx)
370
elif method == 'line-delta':
371
delta = self.factory.parse_line_delta(data, version_idx)
372
content.apply_delta(delta)
373
374
if 'no-eol' in self._index.get_options(version_id):
375
line = content._lines[-1][1].rstrip('\n')
376
content._lines[-1] = (content._lines[-1][0], line)
377
378
if sha_strings(content.text()) != digest:
379
raise KnitCorrupt(self.filename, 'sha-1 does not match')
380
381
return content
382
383
def _check_versions_present(self, version_ids):
384
"""Check that all specified versions are present."""
385
version_ids = set(version_ids)
386
for r in list(version_ids):
387
if self._index.has_version(r):
388
version_ids.remove(r)
389
if version_ids:
390
raise RevisionNotPresent(list(version_ids)[0], self.filename)
391
392
def add_lines(self, version_id, parents, lines):
393
"""See VersionedFile.add_lines."""
394
assert self.writable, "knit is not opened for write"
395
### FIXME escape. RBC 20060228
396
if contains_whitespace(version_id):
397
raise InvalidRevisionId(version_id)
398
if self.has_version(version_id):
399
raise RevisionAlreadyPresent(version_id, self.filename)
400
401
if True or __debug__:
402
for l in lines:
403
assert '\n' not in l[:-1]
404
405
self._check_versions_present(parents)
406
return self._add(version_id, lines[:], parents, self.delta)
407
408
def _add(self, version_id, lines, parents, delta):
409
"""Add a set of lines on top of version specified by parents.
410
411
If delta is true, compress the text as a line-delta against
412
the first parent.
413
"""
414
if delta and not parents:
415
delta = False
416
417
digest = sha_strings(lines)
418
options = []
419
if lines:
420
if lines[-1][-1] != '\n':
421
options.append('no-eol')
422
lines[-1] = lines[-1] + '\n'
423
424
lines = self.factory.make(lines, len(self._index))
425
if self.factory.annotated and len(parents) > 0:
426
# Merge annotations from parent texts if so is needed.
427
self._merge_annotations(lines, parents)
428
429
if parents and delta:
430
# To speed the extract of texts the delta chain is limited
431
# to a fixed number of deltas. This should minimize both
432
# I/O and the time spend applying deltas.
433
count = 0
434
delta_parents = parents
435
while count < 25:
436
parent = delta_parents[0]
437
method = self._index.get_method(parent)
438
if method == 'fulltext':
439
break
440
delta_parents = self._index.get_parents(parent)
441
count = count + 1
442
if method == 'line-delta':
443
delta = False
444
445
if delta:
446
options.append('line-delta')
447
content = self._get_content(parents[0])
448
delta_hunks = content.line_delta(lines)
449
store_lines = self.factory.lower_line_delta(delta_hunks)
450
else:
451
options.append('fulltext')
452
store_lines = self.factory.lower_fulltext(lines)
453
454
where, size = self._data.add_record(version_id, digest, store_lines)
455
self._index.add_version(version_id, options, where, size, parents)
456
457
def check(self, progress_bar=None):
458
"""See VersionedFile.check()."""
459
460
def clone_text(self, new_version_id, old_version_id, parents):
461
"""See VersionedFile.clone_text()."""
462
# FIXME RBC 20060228 make fast by only inserting an index with null delta.
463
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
464
465
def get_lines(self, version_id):
466
"""See VersionedFile.get_lines()."""
467
return self._get_content(version_id).text()
468
469
def num_versions(self):
470
"""See VersionedFile.num_versions()."""
471
return self._index.num_versions()
472
473
__len__ = num_versions
474
475
def annotate_iter(self, version_id):
476
"""See VersionedFile.annotate_iter."""
477
content = self._get_content(version_id)
478
for origin, text in content.annotate_iter():
479
yield self._index.idx_to_name(origin), text
480
481
def get_parents(self, version_id):
482
"""See VersionedFile.get_parents."""
483
self._check_versions_present([version_id])
484
return list(self._index.get_parents(version_id))
485
486
def get_ancestry(self, versions):
487
"""See VersionedFile.get_ancestry."""
488
if isinstance(versions, basestring):
489
versions = [versions]
490
if not versions:
491
return []
492
self._check_versions_present(versions)
493
return self._index.get_ancestry(versions)
494
495
def _reannotate_line_delta(self, other, lines, new_version_id,
496
new_version_idx):
497
"""Re-annotate line-delta and return new delta."""
498
new_delta = []
499
for start, end, count, contents \
500
in self.factory.parse_line_delta_iter(lines):
501
new_lines = []
502
for origin, line in contents:
503
old_version_id = other._index.idx_to_name(origin)
504
if old_version_id == new_version_id:
505
idx = new_version_idx
506
else:
507
idx = self._index.lookup(old_version_id)
508
new_lines.append((idx, line))
509
new_delta.append((start, end, count, new_lines))
510
511
return self.factory.lower_line_delta(new_delta)
512
513
def _reannotate_fulltext(self, other, lines, new_version_id,
514
new_version_idx):
515
"""Re-annotate fulltext and return new version."""
516
content = self.factory.parse_fulltext(lines, new_version_idx)
517
new_lines = []
518
for origin, line in content.annotate_iter():
519
old_version_id = other._index.idx_to_name(origin)
520
if old_version_id == new_version_id:
521
idx = new_version_idx
522
else:
523
idx = self._index.lookup(old_version_id)
524
new_lines.append((idx, line))
525
526
return self.factory.lower_fulltext(KnitContent(new_lines))
527
528
def walk(self, version_ids):
529
"""See VersionedFile.walk."""
530
# We take the short path here, and extract all relevant texts
531
# and put them in a weave and let that do all the work. Far
532
# from optimal, but is much simpler.
533
# FIXME RB 20060228 this really is inefficient!
534
from bzrlib.weave import Weave
535
536
w = Weave(self.filename)
537
ancestry = self.get_ancestry(version_ids)
538
sorted_graph = topo_sort(self._index.get_graph())
539
version_list = [vid for vid in sorted_graph if vid in ancestry]
540
541
for version_id in version_list:
542
lines = self.get_lines(version_id)
543
w.add_lines(version_id, self.get_parents(version_id), lines)
544
545
for lineno, insert_id, dset, line in w.walk(version_ids):
546
yield lineno, insert_id, dset, line
547
548
549
class _KnitComponentFile(object):
550
"""One of the files used to implement a knit database"""
551
552
def __init__(self, transport, filename, mode):
553
self._transport = transport
554
self._filename = filename
555
self._mode = mode
556
557
def write_header(self):
558
old_len = self._transport.append(self._filename, StringIO(self.HEADER))
559
if old_len != 0:
560
raise KnitCorrupt(self._filename, 'misaligned after writing header')
561
562
def check_header(self, fp):
563
line = fp.read(len(self.HEADER))
564
if line != self.HEADER:
565
raise KnitHeaderError(badline=line)
566
567
def commit(self):
568
"""Commit is a nop."""
569
570
def __repr__(self):
571
return '%s(%s)' % (self.__class__.__name__, self._filename)
572
573
574
class _KnitIndex(_KnitComponentFile):
575
"""Manages knit index file.
576
577
The index is already kept in memory and read on startup, to enable
578
fast lookups of revision information. The cursor of the index
579
file is always pointing to the end, making it easy to append
580
entries.
581
582
_cache is a cache for fast mapping from version id to a Index
583
object.
584
585
_history is a cache for fast mapping from indexes to version ids.
586
587
The index data format is dictionary compressed when it comes to
588
parent references; a index entry may only have parents that with a
589
lover index number. As a result, the index is topological sorted.
590
591
Duplicate entries may be written to the index for a single version id
592
if this is done then the latter one completely replaces the former:
593
this allows updates to correct version and parent information.
594
Note that the two entries may share the delta, and that successive
595
annotations and references MUST point to the first entry.
596
"""
597
598
HEADER = "# bzr knit index 7\n"
599
600
def _cache_version(self, version_id, options, pos, size, parents):
601
val = (version_id, options, pos, size, parents)
602
self._cache[version_id] = val
603
if not version_id in self._history:
604
self._history.append(version_id)
605
606
def _iter_index(self, fp):
607
lines = fp.read()
608
for l in lines.splitlines(False):
609
yield l.split()
610
611
def __init__(self, transport, filename, mode, create=False):
612
_KnitComponentFile.__init__(self, transport, filename, mode)
613
self._cache = {}
614
# position in _history is the 'official' index for a revision
615
# but the values may have come from a newer entry.
616
# so - wc -l of a knit index is != the number of uniqe names
617
# in the weave.
618
self._history = []
619
try:
620
fp = self._transport.get(self._filename)
621
self.check_header(fp)
622
for rec in self._iter_index(fp):
623
self._cache_version(rec[0], rec[1].split(','), int(rec[2]), int(rec[3]),
624
[self._history[int(i)] for i in rec[4:]])
625
except NoSuchFile, e:
626
if mode != 'w' or not create:
627
raise
628
self.write_header()
629
630
def get_graph(self):
631
graph = []
632
for version_id, index in self._cache.iteritems():
633
graph.append((version_id, index[4]))
634
return graph
635
636
def get_ancestry(self, versions):
637
"""See VersionedFile.get_ancestry."""
638
# get a graph of all the mentioned versions:
639
graph = {}
640
pending = set(versions)
641
while len(pending):
642
version = pending.pop()
643
parents = self._cache[version][4]
644
for parent in parents:
645
if parent not in graph:
646
pending.add(parent)
647
graph[version] = parents
648
return topo_sort(graph.items())
649
650
def num_versions(self):
651
return len(self._history)
652
653
__len__ = num_versions
654
655
def get_versions(self):
656
return self._history
657
658
def idx_to_name(self, idx):
659
return self._history[idx]
660
661
def lookup(self, version_id):
662
assert version_id in self._cache
663
return self._history.index(version_id)
664
665
def add_version(self, version_id, options, pos, size, parents):
666
"""Add a version record to the index."""
667
self._cache_version(version_id, options, pos, size, parents)
668
669
content = "%s %s %s %s %s\n" % (version_id,
670
','.join(options),
671
pos,
672
size,
673
' '.join([str(self.lookup(vid)) for
674
vid in parents]))
675
self._transport.append(self._filename, StringIO(content))
676
677
def has_version(self, version_id):
678
"""True if the version is in the index."""
679
return self._cache.has_key(version_id)
680
681
def get_position(self, version_id):
682
"""Return data position and size of specified version."""
683
return (self._cache[version_id][2], \
684
self._cache[version_id][3])
685
686
def get_method(self, version_id):
687
"""Return compression method of specified version."""
688
options = self._cache[version_id][1]
689
if 'fulltext' in options:
690
return 'fulltext'
691
else:
692
assert 'line-delta' in options
693
return 'line-delta'
694
695
def get_options(self, version_id):
696
return self._cache[version_id][1]
697
698
def get_parents(self, version_id):
699
"""Return parents of specified version."""
700
return self._cache[version_id][4]
701
702
def check_versions_present(self, version_ids):
703
"""Check that all specified versions are present."""
704
version_ids = set(version_ids)
705
for version_id in list(version_ids):
706
if version_id in self._cache:
707
version_ids.remove(version_id)
708
if version_ids:
709
raise RevisionNotPresent(list(version_ids)[0], self.filename)
710
711
712
class _KnitData(_KnitComponentFile):
713
"""Contents of the knit data file"""
714
715
HEADER = "# bzr knit data 7\n"
716
717
def __init__(self, transport, filename, mode, create=False):
718
_KnitComponentFile.__init__(self, transport, filename, mode)
719
self._file = None
720
self._checked = False
721
if create:
722
self._transport.put(self._filename, StringIO(''))
723
724
def _open_file(self):
725
if self._file is None:
726
try:
727
self._file = self._transport.get(self._filename)
728
except NoSuchFile:
729
pass
730
return self._file
731
732
def add_record(self, version_id, digest, lines):
733
"""Write new text record to disk. Returns the position in the
734
file where it was written."""
735
sio = StringIO()
736
data_file = GzipFile(None, mode='wb', fileobj=sio)
737
print >>data_file, "version %s %d %s" % (version_id, len(lines), digest)
738
data_file.writelines(lines)
739
print >>data_file, "end %s\n" % version_id
740
data_file.close()
741
742
content = sio.getvalue()
743
start_pos = self._transport.append(self._filename, StringIO(content))
744
return start_pos, len(content)
745
746
def _parse_record(self, version_id, data):
747
df = GzipFile(mode='rb', fileobj=StringIO(data))
748
rec = df.readline().split()
749
if len(rec) != 4:
750
raise KnitCorrupt(self._filename, 'unexpected number of records')
751
if rec[1] != version_id:
752
raise KnitCorrupt(self.file.name,
753
'unexpected version, wanted %r' % version_id)
754
lines = int(rec[2])
755
record_contents = self._read_record_contents(df, lines)
756
l = df.readline()
757
if l != 'end %s\n' % version_id:
758
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
759
% (l, version_id))
760
return record_contents, rec[3]
761
762
def _read_record_contents(self, df, record_lines):
763
"""Read and return n lines from datafile."""
764
r = []
765
for i in range(record_lines):
766
r.append(df.readline())
767
return r
768
769
def read_records_iter(self, records):
770
"""Read text records from data file and yield result.
771
772
Each passed record is a tuple of (version_id, pos, len) and
773
will be read in the given order. Yields (version_id,
774
contents, digest).
775
"""
776
777
class ContinuousRange:
778
def __init__(self, rec_id, pos, size):
779
self.start_pos = pos
780
self.end_pos = pos + size
781
self.versions = [(rec_id, pos, size)]
782
783
def add(self, rec_id, pos, size):
784
if self.end_pos != pos:
785
return False
786
self.end_pos = pos + size
787
self.versions.append((rec_id, pos, size))
788
return True
789
790
def split(self, fp):
791
for rec_id, pos, size in self.versions:
792
yield rec_id, fp.read(size)
793
794
fp = self._open_file()
795
796
# Loop through all records and try to collect as large
797
# continuous region as possible to read.
798
while records:
799
record_id, pos, size = records.pop(0)
800
continuous_range = ContinuousRange(record_id, pos, size)
801
while records:
802
record_id, pos, size = records[0]
803
if continuous_range.add(record_id, pos, size):
804
del records[0]
805
else:
806
break
807
fp.seek(continuous_range.start_pos, 0)
808
for record_id, data in continuous_range.split(fp):
809
content, digest = self._parse_record(record_id, data)
810
yield record_id, content, digest
811
812
self._file = None
813
814
def read_records(self, records):
815
"""Read records into a dictionary."""
816
components = {}
817
for record_id, content, digest in self.read_records_iter(records):
818
components[record_id] = (content, digest)
819
return components
820
821
822
class InterKnit(InterVersionedFile):
823
"""Optimised code paths for knit to knit operations."""
824
825
_matching_file_factory = KnitVersionedFile
826
827
@staticmethod
828
def is_compatible(source, target):
829
"""Be compatible with knits. """
830
try:
831
return (isinstance(source, KnitVersionedFile) and
832
isinstance(target, KnitVersionedFile))
833
except AttributeError:
834
return False
835
836
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
837
"""See InterVersionedFile.join."""
838
assert isinstance(self.source, KnitVersionedFile)
839
assert isinstance(self.target, KnitVersionedFile)
840
841
if version_ids is None:
842
version_ids = self.source.versions()
843
else:
844
if not ignore_missing:
845
self.source._check_versions_present(version_ids)
846
else:
847
version_ids = set(self.source.versions()).intersection(
848
set(version_ids))
849
850
if not version_ids:
851
return 0
852
853
if pb is None:
854
from bzrlib.progress import DummyProgress
855
pb = DummyProgress()
856
857
version_ids = list(version_ids)
858
if None in version_ids:
859
version_ids.remove(None)
860
861
self.source_ancestry = set(self.source.get_ancestry(version_ids))
862
this_versions = set(self.target._index.get_versions())
863
needed_versions = self.source_ancestry - this_versions
864
cross_check_versions = self.source_ancestry.intersection(this_versions)
865
mismatched_versions = set()
866
for version in cross_check_versions:
867
# scan to include needed parents.
868
n1 = set(self.target.get_parents(version))
869
n2 = set(self.source.get_parents(version))
870
if n1 != n2:
871
# FIXME TEST this check for cycles being introduced works
872
# the logic is we have a cycle if in our graph we are an
873
# ancestor of any of the n2 revisions.
874
for parent in n2:
875
if parent in n1:
876
# safe
877
continue
878
else:
879
parent_ancestors = self.source.get_ancestry(parent)
880
if version in parent_ancestors:
881
raise errors.GraphCycleError([parent, version])
882
# ensure this parent will be available later.
883
new_parents = n2.difference(n1)
884
needed_versions.update(new_parents.difference(this_versions))
885
mismatched_versions.add(version)
886
887
if not needed_versions and not cross_check_versions:
888
return 0
889
full_list = topo_sort(self.source._index.get_graph())
890
891
version_list = [i for i in full_list if (not self.target.has_version(i)
892
and i in needed_versions)]
893
894
records = []
895
for version_id in version_list:
896
data_pos, data_size = self.source._index.get_position(version_id)
897
records.append((version_id, data_pos, data_size))
898
899
count = 0
900
for version_id, lines, digest \
901
in self.source._data.read_records_iter(records):
902
options = self.source._index.get_options(version_id)
903
parents = self.source._index.get_parents(version_id)
904
905
for parent in parents:
906
assert self.target.has_version(parent)
907
908
if self.target.factory.annotated:
909
# FIXME jrydberg: it should be possible to skip
910
# re-annotating components if we know that we are
911
# going to pull all revisions in the same order.
912
new_version_id = version_id
913
new_version_idx = self.target._index.num_versions()
914
if 'fulltext' in options:
915
lines = self.target._reannotate_fulltext(self.source, lines,
916
new_version_id, new_version_idx)
917
elif 'line-delta' in options:
918
lines = self.target._reannotate_line_delta(self.source, lines,
919
new_version_id, new_version_idx)
920
921
count = count + 1
922
pb.update("Joining knit", count, len(version_list))
923
924
pos, size = self.target._data.add_record(version_id, digest, lines)
925
self.target._index.add_version(version_id, options, pos, size, parents)
926
927
for version in mismatched_versions:
928
n1 = set(self.target.get_parents(version))
929
n2 = set(self.source.get_parents(version))
930
# write a combined record to our history.
931
new_parents = self.target.get_parents(version) + list(n2.difference(n1))
932
current_values = self.target._index._cache[version]
933
self.target._index.add_version(version,
934
current_values[1],
935
current_values[2],
936
current_values[3],
937
new_parents)
938
pb.clear()
939
return count
940
941
942
InterVersionedFile.register_optimiser(InterKnit)
Loggerhead is a web-based interface for Breezy
Version: 2.0.1