« back to all changes in this revision
Viewing changes to bzrlib/knit.py
- browse files at revision 1540.3.26
- compare with another revision
- compare with revision 331
- download diff
- download tarball
- view history from revision 1540.3.26
- Committer: Martin Pool
- Date: 2006-03-10 06:29:53 UTC
- mfrom: (1608 +trunk)
- mto: This revision was merged to the branch mainline in revision 1611.
- Revision ID: mbp@sourcefrog.net-20060310062953-bc1c7ade75c89a7a
[merge] bzr.dev; pycurl not updated for readv yet
- 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 copy import copy
67
from cStringIO import StringIO
68
import difflib
69
from difflib import SequenceMatcher
70
from gzip import GzipFile
71
from itertools import izip
72
import os
73
74
75
import bzrlib
76
import bzrlib.errors as errors
77
from bzrlib.errors import FileExists, NoSuchFile, KnitError, \
78
InvalidRevisionId, KnitCorrupt, KnitHeaderError, \
79
RevisionNotPresent, RevisionAlreadyPresent
80
from bzrlib.trace import mutter
81
from bzrlib.osutils import contains_whitespace, contains_linebreaks, \
82
sha_strings
83
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
84
from bzrlib.tsort import topo_sort
85
86
87
# TODO: Split out code specific to this format into an associated object.
88
89
# TODO: Can we put in some kind of value to check that the index and data
90
# files belong together?
91
92
# TODO: accomodate binaries, perhaps by storing a byte count
93
94
# TODO: function to check whole file
95
96
# TODO: atomically append data, then measure backwards from the cursor
97
# position after writing to work out where it was located. we may need to
98
# bypass python file buffering.
99
100
DATA_SUFFIX = '.knit'
101
INDEX_SUFFIX = '.kndx'
102
103
104
class KnitContent(object):
105
"""Content of a knit version to which deltas can be applied."""
106
107
def __init__(self, lines):
108
self._lines = lines
109
110
def annotate_iter(self):
111
"""Yield tuples of (origin, text) for each content line."""
112
for origin, text in self._lines:
113
yield origin, text
114
115
def annotate(self):
116
"""Return a list of (origin, text) tuples."""
117
return list(self.annotate_iter())
118
119
def apply_delta(self, delta):
120
"""Apply delta to this content."""
121
offset = 0
122
for start, end, count, lines in delta:
123
self._lines[offset+start:offset+end] = lines
124
offset = offset + (start - end) + count
125
126
def line_delta_iter(self, new_lines):
127
"""Generate line-based delta from new_lines to this content."""
128
new_texts = [text for origin, text in new_lines._lines]
129
old_texts = [text for origin, text in self._lines]
130
s = difflib.SequenceMatcher(None, old_texts, new_texts)
131
for op in s.get_opcodes():
132
if op[0] == 'equal':
133
continue
134
yield (op[1], op[2], op[4]-op[3], new_lines._lines[op[3]:op[4]])
135
136
def line_delta(self, new_lines):
137
return list(self.line_delta_iter(new_lines))
138
139
def text(self):
140
return [text for origin, text in self._lines]
141
142
143
class _KnitFactory(object):
144
"""Base factory for creating content objects."""
145
146
def make(self, lines, version):
147
num_lines = len(lines)
148
return KnitContent(zip([version] * num_lines, lines))
149
150
151
class KnitAnnotateFactory(_KnitFactory):
152
"""Factory for creating annotated Content objects."""
153
154
annotated = True
155
156
def parse_fulltext(self, content, version):
157
"""Convert fulltext to internal representation
158
159
fulltext content is of the format
160
revid(utf8) plaintext\n
161
internal representation is of the format:
162
(revid, plaintext)
163
"""
164
lines = []
165
for line in content:
166
origin, text = line.split(' ', 1)
167
lines.append((origin.decode('utf-8'), text))
168
return KnitContent(lines)
169
170
def parse_line_delta_iter(self, lines):
171
"""Convert a line based delta into internal representation.
172
173
line delta is in the form of:
174
intstart intend intcount
175
1..count lines:
176
revid(utf8) newline\n
177
internal represnetation is
178
(start, end, count, [1..count tuples (revid, newline)])
179
"""
180
while lines:
181
header = lines.pop(0)
182
start, end, c = [int(n) for n in header.split(',')]
183
contents = []
184
for i in range(c):
185
origin, text = lines.pop(0).split(' ', 1)
186
contents.append((origin.decode('utf-8'), text))
187
yield start, end, c, contents
188
189
def parse_line_delta(self, lines, version):
190
return list(self.parse_line_delta_iter(lines))
191
192
def lower_fulltext(self, content):
193
"""convert a fulltext content record into a serializable form.
194
195
see parse_fulltext which this inverts.
196
"""
197
return ['%s %s' % (o.encode('utf-8'), t) for o, t in content._lines]
198
199
def lower_line_delta(self, delta):
200
"""convert a delta into a serializable form.
201
202
See parse_line_delta_iter which this inverts.
203
"""
204
out = []
205
for start, end, c, lines in delta:
206
out.append('%d,%d,%d\n' % (start, end, c))
207
for origin, text in lines:
208
out.append('%s %s' % (origin.encode('utf-8'), text))
209
return out
210
211
212
class KnitPlainFactory(_KnitFactory):
213
"""Factory for creating plain Content objects."""
214
215
annotated = False
216
217
def parse_fulltext(self, content, version):
218
"""This parses an unannotated fulltext.
219
220
Note that this is not a noop - the internal representation
221
has (versionid, line) - its just a constant versionid.
222
"""
223
return self.make(content, version)
224
225
def parse_line_delta_iter(self, lines, version):
226
while lines:
227
header = lines.pop(0)
228
start, end, c = [int(n) for n in header.split(',')]
229
yield start, end, c, zip([version] * c, lines[:c])
230
del lines[:c]
231
232
def parse_line_delta(self, lines, version):
233
return list(self.parse_line_delta_iter(lines, version))
234
235
def lower_fulltext(self, content):
236
return content.text()
237
238
def lower_line_delta(self, delta):
239
out = []
240
for start, end, c, lines in delta:
241
out.append('%d,%d,%d\n' % (start, end, c))
242
out.extend([text for origin, text in lines])
243
return out
244
245
246
def make_empty_knit(transport, relpath):
247
"""Construct a empty knit at the specified location."""
248
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
249
k._data._open_file()
250
251
252
class KnitVersionedFile(VersionedFile):
253
"""Weave-like structure with faster random access.
254
255
A knit stores a number of texts and a summary of the relationships
256
between them. Texts are identified by a string version-id. Texts
257
are normally stored and retrieved as a series of lines, but can
258
also be passed as single strings.
259
260
Lines are stored with the trailing newline (if any) included, to
261
avoid special cases for files with no final newline. Lines are
262
composed of 8-bit characters, not unicode. The combination of
263
these approaches should mean any 'binary' file can be safely
264
stored and retrieved.
265
"""
266
267
def __init__(self, relpath, transport, file_mode=None, access_mode=None, factory=None,
268
basis_knit=None, delta=True, create=False):
269
"""Construct a knit at location specified by relpath.
270
271
:param create: If not True, only open an existing knit.
272
"""
273
if access_mode is None:
274
access_mode = 'w'
275
super(KnitVersionedFile, self).__init__(access_mode)
276
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
277
assert not basis_knit or isinstance(basis_knit, KnitVersionedFile), \
278
type(basis_knit)
279
280
self.transport = transport
281
self.filename = relpath
282
self.basis_knit = basis_knit
283
self.factory = factory or KnitAnnotateFactory()
284
self.writable = (access_mode == 'w')
285
self.delta = delta
286
287
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
288
access_mode, create=create)
289
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
290
access_mode, create=not len(self.versions()))
291
292
def clear_cache(self):
293
"""Clear the data cache only."""
294
self._data.clear_cache()
295
296
def copy_to(self, name, transport):
297
"""See VersionedFile.copy_to()."""
298
# copy the current index to a temp index to avoid racing with local
299
# writes
300
transport.put(name + INDEX_SUFFIX + '.tmp', self.transport.get(self._index._filename))
301
# copy the data file
302
transport.put(name + DATA_SUFFIX, self._data._open_file())
303
# rename the copied index into place
304
transport.rename(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
305
306
def create_empty(self, name, transport, mode=None):
307
return KnitVersionedFile(name, transport, factory=self.factory, delta=self.delta, create=True)
308
309
def _fix_parents(self, version, new_parents):
310
"""Fix the parents list for version.
311
312
This is done by appending a new version to the index
313
with identical data except for the parents list.
314
the parents list must be a superset of the current
315
list.
316
"""
317
current_values = self._index._cache[version]
318
assert set(current_values[4]).difference(set(new_parents)) == set()
319
self._index.add_version(version,
320
current_values[1],
321
current_values[2],
322
current_values[3],
323
new_parents)
324
325
def get_graph_with_ghosts(self):
326
"""See VersionedFile.get_graph_with_ghosts()."""
327
graph_items = self._index.get_graph()
328
return dict(graph_items)
329
330
@staticmethod
331
def get_suffixes():
332
"""See VersionedFile.get_suffixes()."""
333
return [DATA_SUFFIX, INDEX_SUFFIX]
334
335
def has_ghost(self, version_id):
336
"""True if there is a ghost reference in the file to version_id."""
337
# maybe we have it
338
if self.has_version(version_id):
339
return False
340
# optimisable if needed by memoising the _ghosts set.
341
items = self._index.get_graph()
342
for node, parents in items:
343
for parent in parents:
344
if parent not in self._index._cache:
345
if parent == version_id:
346
return True
347
return False
348
349
def versions(self):
350
"""See VersionedFile.versions."""
351
return self._index.get_versions()
352
353
def has_version(self, version_id):
354
"""See VersionedFile.has_version."""
355
return self._index.has_version(version_id)
356
357
__contains__ = has_version
358
359
def _merge_annotations(self, content, parents):
360
"""Merge annotations for content. This is done by comparing
361
the annotations based on changed to the text."""
362
for parent_id in parents:
363
merge_content = self._get_content(parent_id)
364
seq = SequenceMatcher(None, merge_content.text(), content.text())
365
for i, j, n in seq.get_matching_blocks():
366
if n == 0:
367
continue
368
content._lines[j:j+n] = merge_content._lines[i:i+n]
369
370
def _get_components(self, version_id):
371
"""Return a list of (version_id, method, data) tuples that
372
makes up version specified by version_id of the knit.
373
374
The components should be applied in the order of the returned
375
list.
376
377
The basis knit will be used to the largest extent possible
378
since it is assumed that accesses to it is faster.
379
"""
380
# needed_revisions holds a list of (method, version_id) of
381
# versions that is needed to be fetched to construct the final
382
# version of the file.
383
#
384
# basis_revisions is a list of versions that needs to be
385
# fetched but exists in the basis knit.
386
387
basis = self.basis_knit
388
needed_versions = []
389
basis_versions = []
390
cursor = version_id
391
392
while 1:
393
picked_knit = self
394
if basis and basis._index.has_version(cursor):
395
picked_knit = basis
396
basis_versions.append(cursor)
397
method = picked_knit._index.get_method(cursor)
398
needed_versions.append((method, cursor))
399
if method == 'fulltext':
400
break
401
cursor = picked_knit.get_parents(cursor)[0]
402
403
components = {}
404
if basis_versions:
405
records = []
406
for comp_id in basis_versions:
407
data_pos, data_size = basis._index.get_data_position(comp_id)
408
records.append((piece_id, data_pos, data_size))
409
components.update(basis._data.read_records(records))
410
411
records = []
412
for comp_id in [vid for method, vid in needed_versions
413
if vid not in basis_versions]:
414
data_pos, data_size = self._index.get_position(comp_id)
415
records.append((comp_id, data_pos, data_size))
416
components.update(self._data.read_records(records))
417
418
# get_data_records returns a mapping with the version id as
419
# index and the value as data. The order the components need
420
# to be applied is held by needed_versions (reversed).
421
out = []
422
for method, comp_id in reversed(needed_versions):
423
out.append((comp_id, method, components[comp_id]))
424
425
return out
426
427
def _get_content(self, version_id):
428
"""Returns a content object that makes up the specified
429
version."""
430
if not self.has_version(version_id):
431
raise RevisionNotPresent(version_id, self.filename)
432
433
if self.basis_knit and version_id in self.basis_knit:
434
return self.basis_knit._get_content(version_id)
435
436
content = None
437
components = self._get_components(version_id)
438
for component_id, method, (data, digest) in components:
439
version_idx = self._index.lookup(component_id)
440
if method == 'fulltext':
441
assert content is None
442
content = self.factory.parse_fulltext(data, version_idx)
443
elif method == 'line-delta':
444
delta = self.factory.parse_line_delta(data, version_idx)
445
content.apply_delta(delta)
446
447
if 'no-eol' in self._index.get_options(version_id):
448
line = content._lines[-1][1].rstrip('\n')
449
content._lines[-1] = (content._lines[-1][0], line)
450
451
if sha_strings(content.text()) != digest:
452
raise KnitCorrupt(self.filename, 'sha-1 does not match')
453
454
return content
455
456
def _check_versions_present(self, version_ids):
457
"""Check that all specified versions are present."""
458
version_ids = set(version_ids)
459
for r in list(version_ids):
460
if self._index.has_version(r):
461
version_ids.remove(r)
462
if version_ids:
463
raise RevisionNotPresent(list(version_ids)[0], self.filename)
464
465
def _add_lines_with_ghosts(self, version_id, parents, lines):
466
"""See VersionedFile.add_lines_with_ghosts()."""
467
self._check_add(version_id, lines)
468
return self._add(version_id, lines[:], parents, self.delta)
469
470
def _add_lines(self, version_id, parents, lines):
471
"""See VersionedFile.add_lines."""
472
self._check_add(version_id, lines)
473
self._check_versions_present(parents)
474
return self._add(version_id, lines[:], parents, self.delta)
475
476
def _check_add(self, version_id, lines):
477
"""check that version_id and lines are safe to add."""
478
assert self.writable, "knit is not opened for write"
479
### FIXME escape. RBC 20060228
480
if contains_whitespace(version_id):
481
raise InvalidRevisionId(version_id)
482
if self.has_version(version_id):
483
raise RevisionAlreadyPresent(version_id, self.filename)
484
485
if False or __debug__:
486
for l in lines:
487
assert '\n' not in l[:-1]
488
489
def _add(self, version_id, lines, parents, delta):
490
"""Add a set of lines on top of version specified by parents.
491
492
If delta is true, compress the text as a line-delta against
493
the first parent.
494
495
Any versions not present will be converted into ghosts.
496
"""
497
present_parents = []
498
ghosts = []
499
for parent in parents:
500
if not self.has_version(parent):
501
ghosts.append(parent)
502
else:
503
present_parents.append(parent)
504
505
if delta and not len(present_parents):
506
delta = False
507
508
digest = sha_strings(lines)
509
options = []
510
if lines:
511
if lines[-1][-1] != '\n':
512
options.append('no-eol')
513
lines[-1] = lines[-1] + '\n'
514
515
lines = self.factory.make(lines, version_id)
516
if self.factory.annotated and len(present_parents) > 0:
517
# Merge annotations from parent texts if so is needed.
518
self._merge_annotations(lines, present_parents)
519
520
if len(present_parents) and delta:
521
# To speed the extract of texts the delta chain is limited
522
# to a fixed number of deltas. This should minimize both
523
# I/O and the time spend applying deltas.
524
count = 0
525
delta_parents = present_parents
526
while count < 25:
527
parent = delta_parents[0]
528
method = self._index.get_method(parent)
529
if method == 'fulltext':
530
break
531
delta_parents = self._index.get_parents(parent)
532
count = count + 1
533
if method == 'line-delta':
534
delta = False
535
536
if delta:
537
options.append('line-delta')
538
content = self._get_content(present_parents[0])
539
delta_hunks = content.line_delta(lines)
540
store_lines = self.factory.lower_line_delta(delta_hunks)
541
else:
542
options.append('fulltext')
543
store_lines = self.factory.lower_fulltext(lines)
544
545
where, size = self._data.add_record(version_id, digest, store_lines)
546
self._index.add_version(version_id, options, where, size, parents)
547
548
def check(self, progress_bar=None):
549
"""See VersionedFile.check()."""
550
551
def _clone_text(self, new_version_id, old_version_id, parents):
552
"""See VersionedFile.clone_text()."""
553
# FIXME RBC 20060228 make fast by only inserting an index with null delta.
554
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
555
556
def get_lines(self, version_id):
557
"""See VersionedFile.get_lines()."""
558
return self._get_content(version_id).text()
559
560
def iter_lines_added_or_present_in_versions(self, version_ids=None):
561
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
562
if version_ids is None:
563
version_ids = self.versions()
564
# we dont care about inclusions, the caller cares.
565
# but we need to setup a list of records to visit.
566
# we need version_id, position, length
567
version_id_records = []
568
requested_versions = list(version_ids)
569
# filter for available versions
570
for version_id in requested_versions:
571
if not self.has_version(version_id):
572
raise RevisionNotPresent(version_id, self.filename)
573
# get a in-component-order queue:
574
version_ids = []
575
for version_id in self.versions():
576
if version_id in requested_versions:
577
version_ids.append(version_id)
578
data_pos, length = self._index.get_position(version_id)
579
version_id_records.append((version_id, data_pos, length))
580
581
pb = bzrlib.ui.ui_factory.nested_progress_bar()
582
count = 0
583
total = len(version_id_records)
584
try:
585
pb.update('Walking content.', count, total)
586
for version_id, data, sha_value in \
587
self._data.read_records_iter(version_id_records):
588
pb.update('Walking content.', count, total)
589
method = self._index.get_method(version_id)
590
version_idx = self._index.lookup(version_id)
591
assert method in ('fulltext', 'line-delta')
592
if method == 'fulltext':
593
content = self.factory.parse_fulltext(data, version_idx)
594
for line in content.text():
595
yield line
596
else:
597
delta = self.factory.parse_line_delta(data, version_idx)
598
for start, end, count, lines in delta:
599
for origin, line in lines:
600
yield line
601
count +=1
602
pb.update('Walking content.', total, total)
603
pb.finished()
604
except:
605
pb.update('Walking content.', total, total)
606
pb.finished()
607
raise
608
609
def num_versions(self):
610
"""See VersionedFile.num_versions()."""
611
return self._index.num_versions()
612
613
__len__ = num_versions
614
615
def annotate_iter(self, version_id):
616
"""See VersionedFile.annotate_iter."""
617
content = self._get_content(version_id)
618
for origin, text in content.annotate_iter():
619
yield origin, text
620
621
def get_parents(self, version_id):
622
"""See VersionedFile.get_parents."""
623
self._check_versions_present([version_id])
624
return list(self._index.get_parents(version_id))
625
626
def get_parents_with_ghosts(self, version_id):
627
"""See VersionedFile.get_parents."""
628
self._check_versions_present([version_id])
629
return list(self._index.get_parents_with_ghosts(version_id))
630
631
def get_ancestry(self, versions):
632
"""See VersionedFile.get_ancestry."""
633
if isinstance(versions, basestring):
634
versions = [versions]
635
if not versions:
636
return []
637
self._check_versions_present(versions)
638
return self._index.get_ancestry(versions)
639
640
def get_ancestry_with_ghosts(self, versions):
641
"""See VersionedFile.get_ancestry_with_ghosts."""
642
if isinstance(versions, basestring):
643
versions = [versions]
644
if not versions:
645
return []
646
self._check_versions_present(versions)
647
return self._index.get_ancestry_with_ghosts(versions)
648
649
#@deprecated_method(zero_eight)
650
def walk(self, version_ids):
651
"""See VersionedFile.walk."""
652
# We take the short path here, and extract all relevant texts
653
# and put them in a weave and let that do all the work. Far
654
# from optimal, but is much simpler.
655
# FIXME RB 20060228 this really is inefficient!
656
from bzrlib.weave import Weave
657
658
w = Weave(self.filename)
659
ancestry = self.get_ancestry(version_ids)
660
sorted_graph = topo_sort(self._index.get_graph())
661
version_list = [vid for vid in sorted_graph if vid in ancestry]
662
663
for version_id in version_list:
664
lines = self.get_lines(version_id)
665
w.add_lines(version_id, self.get_parents(version_id), lines)
666
667
for lineno, insert_id, dset, line in w.walk(version_ids):
668
yield lineno, insert_id, dset, line
669
670
671
class _KnitComponentFile(object):
672
"""One of the files used to implement a knit database"""
673
674
def __init__(self, transport, filename, mode):
675
self._transport = transport
676
self._filename = filename
677
self._mode = mode
678
679
def write_header(self):
680
old_len = self._transport.append(self._filename, StringIO(self.HEADER))
681
if old_len != 0:
682
raise KnitCorrupt(self._filename, 'misaligned after writing header')
683
684
def check_header(self, fp):
685
line = fp.read(len(self.HEADER))
686
if line != self.HEADER:
687
raise KnitHeaderError(badline=line)
688
689
def commit(self):
690
"""Commit is a nop."""
691
692
def __repr__(self):
693
return '%s(%s)' % (self.__class__.__name__, self._filename)
694
695
696
class _KnitIndex(_KnitComponentFile):
697
"""Manages knit index file.
698
699
The index is already kept in memory and read on startup, to enable
700
fast lookups of revision information. The cursor of the index
701
file is always pointing to the end, making it easy to append
702
entries.
703
704
_cache is a cache for fast mapping from version id to a Index
705
object.
706
707
_history is a cache for fast mapping from indexes to version ids.
708
709
The index data format is dictionary compressed when it comes to
710
parent references; a index entry may only have parents that with a
711
lover index number. As a result, the index is topological sorted.
712
713
Duplicate entries may be written to the index for a single version id
714
if this is done then the latter one completely replaces the former:
715
this allows updates to correct version and parent information.
716
Note that the two entries may share the delta, and that successive
717
annotations and references MUST point to the first entry.
718
"""
719
720
HEADER = "# bzr knit index 7\n"
721
722
def _cache_version(self, version_id, options, pos, size, parents):
723
val = (version_id, options, pos, size, parents)
724
self._cache[version_id] = val
725
if not version_id in self._history:
726
self._history.append(version_id)
727
728
def _iter_index(self, fp):
729
l = fp.readline()
730
while l != '':
731
yield l.split()
732
l = fp.readline()
733
#lines = fp.read()
734
#for l in lines.splitlines(False):
735
# yield l.split()
736
737
def __init__(self, transport, filename, mode, create=False):
738
_KnitComponentFile.__init__(self, transport, filename, mode)
739
self._cache = {}
740
# position in _history is the 'official' index for a revision
741
# but the values may have come from a newer entry.
742
# so - wc -l of a knit index is != the number of uniqe names
743
# in the weave.
744
self._history = []
745
pb = bzrlib.ui.ui_factory.nested_progress_bar()
746
try:
747
count = 0
748
total = 1
749
try:
750
pb.update('read knit index', count, total)
751
fp = self._transport.get(self._filename)
752
self.check_header(fp)
753
for rec in self._iter_index(fp):
754
count += 1
755
total += 1
756
pb.update('read knit index', count, total)
757
parents = self._parse_parents(rec[4:])
758
self._cache_version(rec[0], rec[1].split(','), int(rec[2]), int(rec[3]),
759
parents)
760
except NoSuchFile, e:
761
if mode != 'w' or not create:
762
raise
763
self.write_header()
764
finally:
765
pb.update('read knit index', total, total)
766
pb.finished()
767
768
def _parse_parents(self, compressed_parents):
769
"""convert a list of string parent values into version ids.
770
771
ints are looked up in the index.
772
.FOO values are ghosts and converted in to FOO.
773
"""
774
result = []
775
for value in compressed_parents:
776
if value.startswith('.'):
777
result.append(value[1:])
778
else:
779
assert isinstance(value, str)
780
result.append(self._history[int(value)])
781
return result
782
783
def get_graph(self):
784
graph = []
785
for version_id, index in self._cache.iteritems():
786
graph.append((version_id, index[4]))
787
return graph
788
789
def get_ancestry(self, versions):
790
"""See VersionedFile.get_ancestry."""
791
# get a graph of all the mentioned versions:
792
graph = {}
793
pending = set(versions)
794
while len(pending):
795
version = pending.pop()
796
parents = self._cache[version][4]
797
# got the parents ok
798
# trim ghosts
799
parents = [parent for parent in parents if parent in self._cache]
800
for parent in parents:
801
# if not completed and not a ghost
802
if parent not in graph:
803
pending.add(parent)
804
graph[version] = parents
805
return topo_sort(graph.items())
806
807
def get_ancestry_with_ghosts(self, versions):
808
"""See VersionedFile.get_ancestry_with_ghosts."""
809
# get a graph of all the mentioned versions:
810
graph = {}
811
pending = set(versions)
812
while len(pending):
813
version = pending.pop()
814
try:
815
parents = self._cache[version][4]
816
except KeyError:
817
# ghost, fake it
818
graph[version] = []
819
pass
820
else:
821
# got the parents ok
822
for parent in parents:
823
if parent not in graph:
824
pending.add(parent)
825
graph[version] = parents
826
return topo_sort(graph.items())
827
828
def num_versions(self):
829
return len(self._history)
830
831
__len__ = num_versions
832
833
def get_versions(self):
834
return self._history
835
836
def idx_to_name(self, idx):
837
return self._history[idx]
838
839
def lookup(self, version_id):
840
assert version_id in self._cache
841
return self._history.index(version_id)
842
843
def _version_list_to_index(self, versions):
844
result_list = []
845
for version in versions:
846
if version in self._cache:
847
result_list.append(str(self._history.index(version)))
848
else:
849
result_list.append('.' + version.encode('utf-8'))
850
return ' '.join(result_list)
851
852
def add_version(self, version_id, options, pos, size, parents):
853
"""Add a version record to the index."""
854
self._cache_version(version_id, options, pos, size, parents)
855
856
content = "%s %s %s %s %s\n" % (version_id.encode('utf-8'),
857
','.join(options),
858
pos,
859
size,
860
self._version_list_to_index(parents))
861
assert isinstance(content, str), 'content must be utf-8 encoded'
862
self._transport.append(self._filename, StringIO(content))
863
864
def has_version(self, version_id):
865
"""True if the version is in the index."""
866
return self._cache.has_key(version_id)
867
868
def get_position(self, version_id):
869
"""Return data position and size of specified version."""
870
return (self._cache[version_id][2], \
871
self._cache[version_id][3])
872
873
def get_method(self, version_id):
874
"""Return compression method of specified version."""
875
options = self._cache[version_id][1]
876
if 'fulltext' in options:
877
return 'fulltext'
878
else:
879
assert 'line-delta' in options
880
return 'line-delta'
881
882
def get_options(self, version_id):
883
return self._cache[version_id][1]
884
885
def get_parents(self, version_id):
886
"""Return parents of specified version ignoring ghosts."""
887
return [parent for parent in self._cache[version_id][4]
888
if parent in self._cache]
889
890
def get_parents_with_ghosts(self, version_id):
891
"""Return parents of specified version wth ghosts."""
892
return self._cache[version_id][4]
893
894
def check_versions_present(self, version_ids):
895
"""Check that all specified versions are present."""
896
version_ids = set(version_ids)
897
for version_id in list(version_ids):
898
if version_id in self._cache:
899
version_ids.remove(version_id)
900
if version_ids:
901
raise RevisionNotPresent(list(version_ids)[0], self.filename)
902
903
904
class _KnitData(_KnitComponentFile):
905
"""Contents of the knit data file"""
906
907
HEADER = "# bzr knit data 7\n"
908
909
def __init__(self, transport, filename, mode, create=False):
910
_KnitComponentFile.__init__(self, transport, filename, mode)
911
self._file = None
912
self._checked = False
913
if create:
914
self._transport.put(self._filename, StringIO(''))
915
self._records = {}
916
917
def clear_cache(self):
918
"""Clear the record cache."""
919
self._records = {}
920
921
def _open_file(self):
922
if self._file is None:
923
try:
924
self._file = self._transport.get(self._filename)
925
except NoSuchFile:
926
pass
927
return self._file
928
929
def _record_to_data(self, version_id, digest, lines):
930
"""Convert version_id, digest, lines into a raw data block.
931
932
:return: (len, a StringIO instance with the raw data ready to read.)
933
"""
934
sio = StringIO()
935
data_file = GzipFile(None, mode='wb', fileobj=sio)
936
print >>data_file, "version %s %d %s" % (version_id.encode('utf-8'), len(lines), digest)
937
data_file.writelines(lines)
938
print >>data_file, "end %s\n" % version_id.encode('utf-8')
939
data_file.close()
940
length= sio.tell()
941
sio.seek(0)
942
return length, sio
943
944
def add_raw_record(self, raw_data):
945
"""Append a prepared record to the data file."""
946
assert isinstance(raw_data, str), 'data must be plain bytes'
947
start_pos = self._transport.append(self._filename, StringIO(raw_data))
948
return start_pos, len(raw_data)
949
950
def add_record(self, version_id, digest, lines):
951
"""Write new text record to disk. Returns the position in the
952
file where it was written."""
953
size, sio = self._record_to_data(version_id, digest, lines)
954
# cache
955
self._records[version_id] = (digest, lines)
956
# write to disk
957
start_pos = self._transport.append(self._filename, sio)
958
return start_pos, size
959
960
def _parse_record_header(self, version_id, raw_data):
961
"""Parse a record header for consistency.
962
963
:return: the header and the decompressor stream.
964
as (stream, header_record)
965
"""
966
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
967
rec = df.readline().split()
968
if len(rec) != 4:
969
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
970
if rec[1].decode('utf-8')!= version_id:
971
raise KnitCorrupt(self._filename,
972
'unexpected version, wanted %r, got %r' % (
973
version_id, rec[1]))
974
return df, rec
975
976
def _parse_record(self, version_id, data):
977
df, rec = self._parse_record_header(version_id, data)
978
lines = int(rec[2])
979
record_contents = self._read_record_contents(df, lines)
980
l = df.readline()
981
if l.decode('utf-8') != 'end %s\n' % version_id:
982
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
983
% (l, version_id))
984
df.close()
985
return record_contents, rec[3]
986
987
def _read_record_contents(self, df, record_lines):
988
"""Read and return n lines from datafile."""
989
r = []
990
for i in range(record_lines):
991
r.append(df.readline())
992
return r
993
994
def read_records_iter_raw(self, records):
995
"""Read text records from data file and yield raw data.
996
997
This unpacks enough of the text record to validate the id is
998
as expected but thats all.
999
1000
It will actively recompress currently cached records on the
1001
basis that that is cheaper than I/O activity.
1002
"""
1003
needed_records = []
1004
for version_id, pos, size in records:
1005
if version_id not in self._records:
1006
needed_records.append((version_id, pos, size))
1007
1008
# setup an iterator of the external records:
1009
# uses readv so nice and fast we hope.
1010
if len(needed_records):
1011
# grab the disk data needed.
1012
raw_records = self._transport.readv(self._filename,
1013
[(pos, size) for version_id, pos, size in needed_records])
1014
1015
for version_id, pos, size in records:
1016
if version_id in self._records:
1017
# compress a new version
1018
size, sio = self._record_to_data(version_id,
1019
self._records[version_id][0],
1020
self._records[version_id][1])
1021
yield version_id, sio.getvalue()
1022
else:
1023
pos, data = raw_records.next()
1024
# validate the header
1025
df, rec = self._parse_record_header(version_id, data)
1026
df.close()
1027
yield version_id, data
1028
1029
1030
def read_records_iter(self, records):
1031
"""Read text records from data file and yield result.
1032
1033
Each passed record is a tuple of (version_id, pos, len) and
1034
will be read in the given order. Yields (version_id,
1035
contents, digest).
1036
"""
1037
1038
needed_records = []
1039
for version_id, pos, size in records:
1040
if version_id not in self._records:
1041
needed_records.append((version_id, pos, size))
1042
1043
if len(needed_records):
1044
# We take it that the transport optimizes the fetching as good
1045
# as possible (ie, reads continous ranges.)
1046
response = self._transport.readv(self._filename,
1047
[(pos, size) for version_id, pos, size in needed_records])
1048
1049
for (record_id, pos, size), (pos, data) in izip(iter(needed_records), response):
1050
content, digest = self._parse_record(record_id, data)
1051
self._records[record_id] = (digest, content)
1052
1053
for version_id, pos, size in records:
1054
yield version_id, copy(self._records[version_id][1]), copy(self._records[version_id][0])
1055
1056
def read_records(self, records):
1057
"""Read records into a dictionary."""
1058
components = {}
1059
for record_id, content, digest in self.read_records_iter(records):
1060
components[record_id] = (content, digest)
1061
return components
1062
1063
1064
class InterKnit(InterVersionedFile):
1065
"""Optimised code paths for knit to knit operations."""
1066
1067
_matching_file_factory = KnitVersionedFile
1068
1069
@staticmethod
1070
def is_compatible(source, target):
1071
"""Be compatible with knits. """
1072
try:
1073
return (isinstance(source, KnitVersionedFile) and
1074
isinstance(target, KnitVersionedFile))
1075
except AttributeError:
1076
return False
1077
1078
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1079
"""See InterVersionedFile.join."""
1080
assert isinstance(self.source, KnitVersionedFile)
1081
assert isinstance(self.target, KnitVersionedFile)
1082
1083
if version_ids is None:
1084
version_ids = self.source.versions()
1085
else:
1086
if not ignore_missing:
1087
self.source._check_versions_present(version_ids)
1088
else:
1089
version_ids = set(self.source.versions()).intersection(
1090
set(version_ids))
1091
1092
if not version_ids:
1093
return 0
1094
1095
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1096
try:
1097
version_ids = list(version_ids)
1098
if None in version_ids:
1099
version_ids.remove(None)
1100
1101
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1102
this_versions = set(self.target._index.get_versions())
1103
needed_versions = self.source_ancestry - this_versions
1104
cross_check_versions = self.source_ancestry.intersection(this_versions)
1105
mismatched_versions = set()
1106
for version in cross_check_versions:
1107
# scan to include needed parents.
1108
n1 = set(self.target.get_parents_with_ghosts(version))
1109
n2 = set(self.source.get_parents_with_ghosts(version))
1110
if n1 != n2:
1111
# FIXME TEST this check for cycles being introduced works
1112
# the logic is we have a cycle if in our graph we are an
1113
# ancestor of any of the n2 revisions.
1114
for parent in n2:
1115
if parent in n1:
1116
# safe
1117
continue
1118
else:
1119
parent_ancestors = self.source.get_ancestry(parent)
1120
if version in parent_ancestors:
1121
raise errors.GraphCycleError([parent, version])
1122
# ensure this parent will be available later.
1123
new_parents = n2.difference(n1)
1124
needed_versions.update(new_parents.difference(this_versions))
1125
mismatched_versions.add(version)
1126
1127
if not needed_versions and not cross_check_versions:
1128
return 0
1129
full_list = topo_sort(self.source.get_graph())
1130
1131
version_list = [i for i in full_list if (not self.target.has_version(i)
1132
and i in needed_versions)]
1133
1134
# plan the join:
1135
copy_queue = []
1136
copy_queue_records = []
1137
copy_set = set()
1138
for version_id in version_list:
1139
options = self.source._index.get_options(version_id)
1140
parents = self.source._index.get_parents_with_ghosts(version_id)
1141
# check that its will be a consistent copy:
1142
for parent in parents:
1143
# if source has the parent, we must :
1144
# * already have it or
1145
# * have it scheduled already
1146
# otherwise we dont care
1147
assert (self.target.has_version(parent) or
1148
parent in copy_set or
1149
not self.source.has_version(parent))
1150
data_pos, data_size = self.source._index.get_position(version_id)
1151
copy_queue_records.append((version_id, data_pos, data_size))
1152
copy_queue.append((version_id, options, parents))
1153
copy_set.add(version_id)
1154
1155
# data suck the join:
1156
count = 0
1157
total = len(version_list)
1158
# we want the raw gzip for bulk copying, but the record validated
1159
# just enough to be sure its the right one.
1160
# TODO: consider writev or write combining to reduce
1161
# death of a thousand cuts feeling.
1162
for (version_id, raw_data), \
1163
(version_id2, options, parents) in \
1164
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1165
copy_queue):
1166
assert version_id == version_id2, 'logic error, inconsistent results'
1167
count = count + 1
1168
pb.update("Joining knit", count, total)
1169
pos, size = self.target._data.add_raw_record(raw_data)
1170
self.target._index.add_version(version_id, options, pos, size, parents)
1171
1172
for version in mismatched_versions:
1173
# FIXME RBC 20060309 is this needed?
1174
n1 = set(self.target.get_parents_with_ghosts(version))
1175
n2 = set(self.source.get_parents_with_ghosts(version))
1176
# write a combined record to our history preserving the current
1177
# parents as first in the list
1178
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1179
self.target.fix_parents(version, new_parents)
1180
return count
1181
finally:
1182
pb.clear()
1183
pb.finished()
1184
1185
1186
InterVersionedFile.register_optimiser(InterKnit)
Loggerhead is a web-based interface for Breezy
Version: 2.0.1