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>
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.
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.
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
20
"""Knit versionedfile implementation.
22
A knit is a versioned file implementation that supports efficient append only
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"
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
37
8 e.set('executable', 'yes')
39
8 if elt.get('executable') == 'yes':
40
8 ie.executable = True
41
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
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
52
so the index sequence is the dictionary compressed sequence number used
53
in the deltas to provide line annotation
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
62
# move sha1 out of the content so that join is faster at verifying parents
63
# record content length ?
66
from cStringIO import StringIO
68
from difflib import SequenceMatcher
69
from gzip import GzipFile
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, \
79
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
80
from bzrlib.tsort import topo_sort
83
# TODO: Split out code specific to this format into an associated object.
85
# TODO: Can we put in some kind of value to check that the index and data
86
# files belong together?
88
# TODO: accomodate binaries, perhaps by storing a byte count
90
# TODO: function to check whole file
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.
97
INDEX_SUFFIX = '.kndx'
100
class KnitContent(object):
101
"""Content of a knit version to which deltas can be applied."""
103
def __init__(self, lines):
106
def annotate_iter(self):
107
"""Yield tuples of (origin, text) for each content line."""
108
for origin, text in self._lines:
112
"""Return a list of (origin, text) tuples."""
113
return list(self.annotate_iter())
115
def apply_delta(self, delta):
116
"""Apply delta to this content."""
118
for start, end, count, lines in delta:
119
self._lines[offset+start:offset+end] = lines
120
offset = offset + (start - end) + count
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():
130
yield (op[1], op[2], op[4]-op[3], new_lines._lines[op[3]:op[4]])
132
def line_delta(self, new_lines):
133
return list(self.line_delta_iter(new_lines))
136
return [text for origin, text in self._lines]
139
class _KnitFactory(object):
140
"""Base factory for creating content objects."""
142
def make(self, lines, version):
143
num_lines = len(lines)
144
return KnitContent(zip([version] * num_lines, lines))
147
class KnitAnnotateFactory(_KnitFactory):
148
"""Factory for creating annotated Content objects."""
152
def parse_fulltext(self, content, version):
155
origin, text = line.split(' ', 1)
156
lines.append((int(origin), text))
157
return KnitContent(lines)
159
def parse_line_delta_iter(self, lines):
161
header = lines.pop(0)
162
start, end, c = [int(n) for n in header.split(',')]
165
origin, text = lines.pop(0).split(' ', 1)
166
contents.append((int(origin), text))
167
yield start, end, c, contents
169
def parse_line_delta(self, lines, version):
170
return list(self.parse_line_delta_iter(lines))
172
def lower_fulltext(self, content):
173
return ['%d %s' % (o, t) for o, t in content._lines]
175
def lower_line_delta(self, delta):
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))
184
class KnitPlainFactory(_KnitFactory):
185
"""Factory for creating plain Content objects."""
189
def parse_fulltext(self, content, version):
190
return self.make(content, version)
192
def parse_line_delta_iter(self, lines, version):
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])
199
def parse_line_delta(self, lines, version):
200
return list(self.parse_line_delta_iter(lines, version))
202
def lower_fulltext(self, content):
203
return content.text()
205
def lower_line_delta(self, delta):
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])
213
def make_empty_knit(transport, relpath):
214
"""Construct a empty knit at the specified location."""
215
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
219
class KnitVersionedFile(VersionedFile):
220
"""Weave-like structure with faster random access.
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.
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.
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.
238
:param create: If not True, only open an existing knit.
240
if access_mode is None:
242
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
243
assert not basis_knit or isinstance(basis_knit, KnitVersionedFile), \
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')
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()))
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
262
transport.put(name + INDEX_SUFFIX + '.tmp', self.transport.get(self._index._filename))
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)
268
def create_empty(self, name, transport, mode=None):
269
return KnitVersionedFile(name, transport, factory=self.factory, delta=self.delta, create=True)
271
def fix_parents(self, version, new_parents):
272
"""Fix the parents list for version.
274
This is done by appending a new version to the index
275
with identical data except for the parents list.
276
the parents list must be a superset of the current
279
current_values = self._index._cache[version]
280
assert set(current_values[4]).difference(set(new_parents)) == set()
281
self._index.add_version(version,
289
"""See VersionedFile.get_suffixes()."""
290
return [DATA_SUFFIX, INDEX_SUFFIX]
293
"""See VersionedFile.versions."""
294
return self._index.get_versions()
296
def has_version(self, version_id):
297
"""See VersionedFile.has_version."""
298
return self._index.has_version(version_id)
300
__contains__ = has_version
302
def _merge_annotations(self, content, parents):
303
"""Merge annotations for content. This is done by comparing
304
the annotations based on changed to the text."""
305
for parent_id in parents:
306
merge_content = self._get_content(parent_id)
307
seq = SequenceMatcher(None, merge_content.text(), content.text())
308
for i, j, n in seq.get_matching_blocks():
311
content._lines[j:j+n] = merge_content._lines[i:i+n]
313
def _get_components(self, version_id):
314
"""Return a list of (version_id, method, data) tuples that
315
makes up version specified by version_id of the knit.
317
The components should be applied in the order of the returned
320
The basis knit will be used to the largest extent possible
321
since it is assumed that accesses to it is faster.
323
# needed_revisions holds a list of (method, version_id) of
324
# versions that is needed to be fetched to construct the final
325
# version of the file.
327
# basis_revisions is a list of versions that needs to be
328
# fetched but exists in the basis knit.
330
basis = self.basis_knit
337
if basis and basis._index.has_version(cursor):
339
basis_versions.append(cursor)
340
method = picked_knit._index.get_method(cursor)
341
needed_versions.append((method, cursor))
342
if method == 'fulltext':
344
cursor = picked_knit.get_parents(cursor)[0]
349
for comp_id in basis_versions:
350
data_pos, data_size = basis._index.get_data_position(comp_id)
351
records.append((piece_id, data_pos, data_size))
352
components.update(basis._data.read_records(records))
355
for comp_id in [vid for method, vid in needed_versions
356
if vid not in basis_versions]:
357
data_pos, data_size = self._index.get_position(comp_id)
358
records.append((comp_id, data_pos, data_size))
359
components.update(self._data.read_records(records))
361
# get_data_records returns a mapping with the version id as
362
# index and the value as data. The order the components need
363
# to be applied is held by needed_versions (reversed).
365
for method, comp_id in reversed(needed_versions):
366
out.append((comp_id, method, components[comp_id]))
370
def _get_content(self, version_id):
371
"""Returns a content object that makes up the specified
373
if not self.has_version(version_id):
374
raise RevisionNotPresent(version_id, self.filename)
376
if self.basis_knit and version_id in self.basis_knit:
377
return self.basis_knit._get_content(version_id)
380
components = self._get_components(version_id)
381
for component_id, method, (data, digest) in components:
382
version_idx = self._index.lookup(component_id)
383
if method == 'fulltext':
384
assert content is None
385
content = self.factory.parse_fulltext(data, version_idx)
386
elif method == 'line-delta':
387
delta = self.factory.parse_line_delta(data, version_idx)
388
content.apply_delta(delta)
390
if 'no-eol' in self._index.get_options(version_id):
391
line = content._lines[-1][1].rstrip('\n')
392
content._lines[-1] = (content._lines[-1][0], line)
394
if sha_strings(content.text()) != digest:
395
raise KnitCorrupt(self.filename, 'sha-1 does not match')
399
def _check_versions_present(self, version_ids):
400
"""Check that all specified versions are present."""
401
version_ids = set(version_ids)
402
for r in list(version_ids):
403
if self._index.has_version(r):
404
version_ids.remove(r)
406
raise RevisionNotPresent(list(version_ids)[0], self.filename)
408
def add_lines(self, version_id, parents, lines):
409
"""See VersionedFile.add_lines."""
410
assert self.writable, "knit is not opened for write"
411
### FIXME escape. RBC 20060228
412
if contains_whitespace(version_id):
413
raise InvalidRevisionId(version_id)
414
if self.has_version(version_id):
415
raise RevisionAlreadyPresent(version_id, self.filename)
417
if False or __debug__:
419
assert '\n' not in l[:-1]
421
self._check_versions_present(parents)
422
return self._add(version_id, lines[:], parents, self.delta)
424
def _add(self, version_id, lines, parents, delta):
425
"""Add a set of lines on top of version specified by parents.
427
If delta is true, compress the text as a line-delta against
430
if delta and not parents:
433
digest = sha_strings(lines)
436
if lines[-1][-1] != '\n':
437
options.append('no-eol')
438
lines[-1] = lines[-1] + '\n'
440
lines = self.factory.make(lines, len(self._index))
441
if self.factory.annotated and len(parents) > 0:
442
# Merge annotations from parent texts if so is needed.
443
self._merge_annotations(lines, parents)
445
if parents and delta:
446
# To speed the extract of texts the delta chain is limited
447
# to a fixed number of deltas. This should minimize both
448
# I/O and the time spend applying deltas.
450
delta_parents = parents
452
parent = delta_parents[0]
453
method = self._index.get_method(parent)
454
if method == 'fulltext':
456
delta_parents = self._index.get_parents(parent)
458
if method == 'line-delta':
462
options.append('line-delta')
463
content = self._get_content(parents[0])
464
delta_hunks = content.line_delta(lines)
465
store_lines = self.factory.lower_line_delta(delta_hunks)
467
options.append('fulltext')
468
store_lines = self.factory.lower_fulltext(lines)
470
where, size = self._data.add_record(version_id, digest, store_lines)
471
self._index.add_version(version_id, options, where, size, parents)
473
def check(self, progress_bar=None):
474
"""See VersionedFile.check()."""
476
def clone_text(self, new_version_id, old_version_id, parents):
477
"""See VersionedFile.clone_text()."""
478
# FIXME RBC 20060228 make fast by only inserting an index with null delta.
479
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
481
def get_lines(self, version_id):
482
"""See VersionedFile.get_lines()."""
483
return self._get_content(version_id).text()
485
def iter_lines_added_or_present_in_versions(self, version_ids=None):
486
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
487
if version_ids is None:
488
version_ids = self.versions()
489
# we dont care about inclusions, the caller cares.
490
# but we need to setup a list of records to visit.
491
# we need version_id, position, length
492
version_id_records = []
493
for version_id in version_ids:
494
if not self.has_version(version_id):
495
raise RevisionNotPresent(version_id, self.filename)
496
data_pos, length = self._index.get_position(version_id)
497
version_id_records.append((version_id, data_pos, length))
498
for version_id, data, sha_value in \
499
self._data.read_records_iter(version_id_records):
500
method = self._index.get_method(version_id)
501
version_idx = self._index.lookup(version_id)
502
assert method in ('fulltext', 'line-delta')
503
if method == 'fulltext':
504
content = self.factory.parse_fulltext(data, version_idx)
505
for line in content.text():
508
delta = self.factory.parse_line_delta(data, version_idx)
509
for start, end, count, lines in delta:
510
for origin, line in lines:
513
def num_versions(self):
514
"""See VersionedFile.num_versions()."""
515
return self._index.num_versions()
517
__len__ = num_versions
519
def annotate_iter(self, version_id):
520
"""See VersionedFile.annotate_iter."""
521
content = self._get_content(version_id)
522
for origin, text in content.annotate_iter():
523
yield self._index.idx_to_name(origin), text
525
def get_parents(self, version_id):
526
"""See VersionedFile.get_parents."""
527
self._check_versions_present([version_id])
528
return list(self._index.get_parents(version_id))
530
def get_ancestry(self, versions):
531
"""See VersionedFile.get_ancestry."""
532
if isinstance(versions, basestring):
533
versions = [versions]
536
self._check_versions_present(versions)
537
return self._index.get_ancestry(versions)
539
def _reannotate_line_delta(self, other, lines, new_version_id,
541
"""Re-annotate line-delta and return new delta."""
543
for start, end, count, contents \
544
in self.factory.parse_line_delta_iter(lines):
546
for origin, line in contents:
547
old_version_id = other._index.idx_to_name(origin)
548
if old_version_id == new_version_id:
549
idx = new_version_idx
551
idx = self._index.lookup(old_version_id)
552
new_lines.append((idx, line))
553
new_delta.append((start, end, count, new_lines))
555
return self.factory.lower_line_delta(new_delta)
557
def _reannotate_fulltext(self, other, lines, new_version_id,
559
"""Re-annotate fulltext and return new version."""
560
content = self.factory.parse_fulltext(lines, new_version_idx)
562
for origin, line in content.annotate_iter():
563
old_version_id = other._index.idx_to_name(origin)
564
if old_version_id == new_version_id:
565
idx = new_version_idx
567
idx = self._index.lookup(old_version_id)
568
new_lines.append((idx, line))
570
return self.factory.lower_fulltext(KnitContent(new_lines))
572
#@deprecated_method(zero_eight)
573
def walk(self, version_ids):
574
"""See VersionedFile.walk."""
575
# We take the short path here, and extract all relevant texts
576
# and put them in a weave and let that do all the work. Far
577
# from optimal, but is much simpler.
578
# FIXME RB 20060228 this really is inefficient!
579
from bzrlib.weave import Weave
581
w = Weave(self.filename)
582
ancestry = self.get_ancestry(version_ids)
583
sorted_graph = topo_sort(self._index.get_graph())
584
version_list = [vid for vid in sorted_graph if vid in ancestry]
586
for version_id in version_list:
587
lines = self.get_lines(version_id)
588
w.add_lines(version_id, self.get_parents(version_id), lines)
590
for lineno, insert_id, dset, line in w.walk(version_ids):
591
yield lineno, insert_id, dset, line
594
class _KnitComponentFile(object):
595
"""One of the files used to implement a knit database"""
597
def __init__(self, transport, filename, mode):
598
self._transport = transport
599
self._filename = filename
602
def write_header(self):
603
old_len = self._transport.append(self._filename, StringIO(self.HEADER))
605
raise KnitCorrupt(self._filename, 'misaligned after writing header')
607
def check_header(self, fp):
608
line = fp.read(len(self.HEADER))
609
if line != self.HEADER:
610
raise KnitHeaderError(badline=line)
613
"""Commit is a nop."""
616
return '%s(%s)' % (self.__class__.__name__, self._filename)
619
class _KnitIndex(_KnitComponentFile):
620
"""Manages knit index file.
622
The index is already kept in memory and read on startup, to enable
623
fast lookups of revision information. The cursor of the index
624
file is always pointing to the end, making it easy to append
627
_cache is a cache for fast mapping from version id to a Index
630
_history is a cache for fast mapping from indexes to version ids.
632
The index data format is dictionary compressed when it comes to
633
parent references; a index entry may only have parents that with a
634
lover index number. As a result, the index is topological sorted.
636
Duplicate entries may be written to the index for a single version id
637
if this is done then the latter one completely replaces the former:
638
this allows updates to correct version and parent information.
639
Note that the two entries may share the delta, and that successive
640
annotations and references MUST point to the first entry.
643
HEADER = "# bzr knit index 7\n"
645
def _cache_version(self, version_id, options, pos, size, parents):
646
val = (version_id, options, pos, size, parents)
647
self._cache[version_id] = val
648
if not version_id in self._history:
649
self._history.append(version_id)
651
def _iter_index(self, fp):
653
for l in lines.splitlines(False):
656
def __init__(self, transport, filename, mode, create=False):
657
_KnitComponentFile.__init__(self, transport, filename, mode)
659
# position in _history is the 'official' index for a revision
660
# but the values may have come from a newer entry.
661
# so - wc -l of a knit index is != the number of uniqe names
665
fp = self._transport.get(self._filename)
666
self.check_header(fp)
667
for rec in self._iter_index(fp):
668
self._cache_version(rec[0], rec[1].split(','), int(rec[2]), int(rec[3]),
669
[self._history[int(i)] for i in rec[4:]])
670
except NoSuchFile, e:
671
if mode != 'w' or not create:
677
for version_id, index in self._cache.iteritems():
678
graph.append((version_id, index[4]))
681
def get_ancestry(self, versions):
682
"""See VersionedFile.get_ancestry."""
683
# get a graph of all the mentioned versions:
685
pending = set(versions)
687
version = pending.pop()
688
parents = self._cache[version][4]
689
for parent in parents:
690
if parent not in graph:
692
graph[version] = parents
693
return topo_sort(graph.items())
695
def num_versions(self):
696
return len(self._history)
698
__len__ = num_versions
700
def get_versions(self):
703
def idx_to_name(self, idx):
704
return self._history[idx]
706
def lookup(self, version_id):
707
assert version_id in self._cache
708
return self._history.index(version_id)
710
def add_version(self, version_id, options, pos, size, parents):
711
"""Add a version record to the index."""
712
self._cache_version(version_id, options, pos, size, parents)
714
content = "%s %s %s %s %s\n" % (version_id,
718
' '.join([str(self.lookup(vid)) for
720
self._transport.append(self._filename, StringIO(content))
722
def has_version(self, version_id):
723
"""True if the version is in the index."""
724
return self._cache.has_key(version_id)
726
def get_position(self, version_id):
727
"""Return data position and size of specified version."""
728
return (self._cache[version_id][2], \
729
self._cache[version_id][3])
731
def get_method(self, version_id):
732
"""Return compression method of specified version."""
733
options = self._cache[version_id][1]
734
if 'fulltext' in options:
737
assert 'line-delta' in options
740
def get_options(self, version_id):
741
return self._cache[version_id][1]
743
def get_parents(self, version_id):
744
"""Return parents of specified version."""
745
return self._cache[version_id][4]
747
def check_versions_present(self, version_ids):
748
"""Check that all specified versions are present."""
749
version_ids = set(version_ids)
750
for version_id in list(version_ids):
751
if version_id in self._cache:
752
version_ids.remove(version_id)
754
raise RevisionNotPresent(list(version_ids)[0], self.filename)
757
class _KnitData(_KnitComponentFile):
758
"""Contents of the knit data file"""
760
HEADER = "# bzr knit data 7\n"
762
def __init__(self, transport, filename, mode, create=False):
763
_KnitComponentFile.__init__(self, transport, filename, mode)
765
self._checked = False
767
self._transport.put(self._filename, StringIO(''))
769
def _open_file(self):
770
if self._file is None:
772
self._file = self._transport.get(self._filename)
777
def add_record(self, version_id, digest, lines):
778
"""Write new text record to disk. Returns the position in the
779
file where it was written."""
781
data_file = GzipFile(None, mode='wb', fileobj=sio)
782
print >>data_file, "version %s %d %s" % (version_id, len(lines), digest)
783
data_file.writelines(lines)
784
print >>data_file, "end %s\n" % version_id
787
content = sio.getvalue()
788
start_pos = self._transport.append(self._filename, StringIO(content))
789
return start_pos, len(content)
791
def _parse_record(self, version_id, data):
792
df = GzipFile(mode='rb', fileobj=StringIO(data))
793
rec = df.readline().split()
795
raise KnitCorrupt(self._filename, 'unexpected number of records')
796
if rec[1] != version_id:
797
raise KnitCorrupt(self.file.name,
798
'unexpected version, wanted %r' % version_id)
800
record_contents = self._read_record_contents(df, lines)
802
if l != 'end %s\n' % version_id:
803
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
805
return record_contents, rec[3]
807
def _read_record_contents(self, df, record_lines):
808
"""Read and return n lines from datafile."""
810
for i in range(record_lines):
811
r.append(df.readline())
814
def read_records_iter(self, records):
815
"""Read text records from data file and yield result.
817
Each passed record is a tuple of (version_id, pos, len) and
818
will be read in the given order. Yields (version_id,
822
class ContinuousRange:
823
def __init__(self, rec_id, pos, size):
825
self.end_pos = pos + size
826
self.versions = [(rec_id, pos, size)]
828
def add(self, rec_id, pos, size):
829
if self.end_pos != pos:
831
self.end_pos = pos + size
832
self.versions.append((rec_id, pos, size))
836
for rec_id, pos, size in self.versions:
837
yield rec_id, fp.read(size)
839
# We take it that the transport optimizes the fetching as good
840
# as possible (ie, reads continous ranges.)
841
response = self._transport.readv(self._filename,
842
[(pos, size) for version_id, pos, size in records])
844
for (record_id, pos, size), (pos, data) in zip(records, response):
845
content, digest = self._parse_record(record_id, data)
846
yield record_id, content, digest
848
def read_records(self, records):
849
"""Read records into a dictionary."""
851
for record_id, content, digest in self.read_records_iter(records):
852
components[record_id] = (content, digest)
856
class InterKnit(InterVersionedFile):
857
"""Optimised code paths for knit to knit operations."""
859
_matching_file_factory = KnitVersionedFile
862
def is_compatible(source, target):
863
"""Be compatible with knits. """
865
return (isinstance(source, KnitVersionedFile) and
866
isinstance(target, KnitVersionedFile))
867
except AttributeError:
870
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
871
"""See InterVersionedFile.join."""
872
assert isinstance(self.source, KnitVersionedFile)
873
assert isinstance(self.target, KnitVersionedFile)
875
if version_ids is None:
876
version_ids = self.source.versions()
878
if not ignore_missing:
879
self.source._check_versions_present(version_ids)
881
version_ids = set(self.source.versions()).intersection(
888
from bzrlib.progress import DummyProgress
891
version_ids = list(version_ids)
892
if None in version_ids:
893
version_ids.remove(None)
895
self.source_ancestry = set(self.source.get_ancestry(version_ids))
896
this_versions = set(self.target._index.get_versions())
897
needed_versions = self.source_ancestry - this_versions
898
cross_check_versions = self.source_ancestry.intersection(this_versions)
899
mismatched_versions = set()
900
for version in cross_check_versions:
901
# scan to include needed parents.
902
n1 = set(self.target.get_parents(version))
903
n2 = set(self.source.get_parents(version))
905
# FIXME TEST this check for cycles being introduced works
906
# the logic is we have a cycle if in our graph we are an
907
# ancestor of any of the n2 revisions.
913
parent_ancestors = self.source.get_ancestry(parent)
914
if version in parent_ancestors:
915
raise errors.GraphCycleError([parent, version])
916
# ensure this parent will be available later.
917
new_parents = n2.difference(n1)
918
needed_versions.update(new_parents.difference(this_versions))
919
mismatched_versions.add(version)
921
if not needed_versions and not cross_check_versions:
923
full_list = topo_sort(self.source._index.get_graph())
925
version_list = [i for i in full_list if (not self.target.has_version(i)
926
and i in needed_versions)]
929
for version_id in version_list:
930
data_pos, data_size = self.source._index.get_position(version_id)
931
records.append((version_id, data_pos, data_size))
934
for version_id, lines, digest \
935
in self.source._data.read_records_iter(records):
936
options = self.source._index.get_options(version_id)
937
parents = self.source._index.get_parents(version_id)
939
for parent in parents:
940
assert self.target.has_version(parent)
942
if self.target.factory.annotated:
943
# FIXME jrydberg: it should be possible to skip
944
# re-annotating components if we know that we are
945
# going to pull all revisions in the same order.
946
new_version_id = version_id
947
new_version_idx = self.target._index.num_versions()
948
if 'fulltext' in options:
949
lines = self.target._reannotate_fulltext(self.source, lines,
950
new_version_id, new_version_idx)
951
elif 'line-delta' in options:
952
lines = self.target._reannotate_line_delta(self.source, lines,
953
new_version_id, new_version_idx)
956
pb.update("Joining knit", count, len(version_list))
958
pos, size = self.target._data.add_record(version_id, digest, lines)
959
self.target._index.add_version(version_id, options, pos, size, parents)
961
for version in mismatched_versions:
962
n1 = set(self.target.get_parents(version))
963
n2 = set(self.source.get_parents(version))
964
# write a combined record to our history preserving the current
965
# parents as first in the list
966
new_parents = self.target.get_parents(version) + list(n2.difference(n1))
967
self.target.fix_parents(version, new_parents)
972
InterVersionedFile.register_optimiser(InterKnit)
added
removed
bzrlib/knit.py
