1
# Copyright (C) 2005, 2006, 2007 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""Knit versionedfile implementation.
19
A knit is a versioned file implementation that supports efficient append only
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
end-marker; simply "end VERSION"
28
delta can be line or full contents.a
29
... the 8's there are the index number of the annotation.
30
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
8 e.set('executable', 'yes')
36
8 if elt.get('executable') == 'yes':
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
43
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
44
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
45
09:33 < lifeless> right
46
09:33 < jrydberg> lifeless: the position and size is the range in the data file
49
so the index sequence is the dictionary compressed sequence number used
50
in the deltas to provide line annotation
55
# 10:16 < lifeless> make partial index writes safe
56
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
57
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
59
# move sha1 out of the content so that join is faster at verifying parents
60
# record content length ?
64
from cStringIO import StringIO
66
from itertools import izip, chain
73
from bzrlib.lazy_import import lazy_import
74
lazy_import(globals(), """
88
from bzrlib.errors import (
96
RevisionAlreadyPresent,
98
from bzrlib.tuned_gzip import GzipFile
99
from bzrlib.trace import mutter
100
from bzrlib.osutils import (
105
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
106
from bzrlib.tsort import topo_sort
109
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
112
# TODO: Split out code specific to this format into an associated object.
114
# TODO: Can we put in some kind of value to check that the index and data
115
# files belong together?
117
# TODO: accommodate binaries, perhaps by storing a byte count
119
# TODO: function to check whole file
121
# TODO: atomically append data, then measure backwards from the cursor
122
# position after writing to work out where it was located. we may need to
123
# bypass python file buffering.
125
DATA_SUFFIX = '.knit'
126
INDEX_SUFFIX = '.kndx'
129
class KnitContent(object):
130
"""Content of a knit version to which deltas can be applied."""
132
def __init__(self, lines):
135
def annotate_iter(self):
136
"""Yield tuples of (origin, text) for each content line."""
137
return iter(self._lines)
140
"""Return a list of (origin, text) tuples."""
141
return list(self.annotate_iter())
143
def line_delta_iter(self, new_lines):
144
"""Generate line-based delta from this content to new_lines."""
145
new_texts = new_lines.text()
146
old_texts = self.text()
147
s = KnitSequenceMatcher(None, old_texts, new_texts)
148
for tag, i1, i2, j1, j2 in s.get_opcodes():
151
# ofrom, oto, length, data
152
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
154
def line_delta(self, new_lines):
155
return list(self.line_delta_iter(new_lines))
158
return [text for origin, text in self._lines]
161
return KnitContent(self._lines[:])
164
def get_line_delta_blocks(knit_delta, source, target):
165
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
166
target_len = len(target)
169
for s_begin, s_end, t_len, new_text in knit_delta:
170
true_n = s_begin - s_pos
173
# knit deltas do not provide reliable info about whether the
174
# last line of a file matches, due to eol handling.
175
if source[s_pos + n -1] != target[t_pos + n -1]:
178
yield s_pos, t_pos, n
179
t_pos += t_len + true_n
181
n = target_len - t_pos
183
if source[s_pos + n -1] != target[t_pos + n -1]:
186
yield s_pos, t_pos, n
187
yield s_pos + (target_len - t_pos), target_len, 0
190
class _KnitFactory(object):
191
"""Base factory for creating content objects."""
193
def make(self, lines, version_id):
194
num_lines = len(lines)
195
return KnitContent(zip([version_id] * num_lines, lines))
198
class KnitAnnotateFactory(_KnitFactory):
199
"""Factory for creating annotated Content objects."""
203
def parse_fulltext(self, content, version_id):
204
"""Convert fulltext to internal representation
206
fulltext content is of the format
207
revid(utf8) plaintext\n
208
internal representation is of the format:
211
# TODO: jam 20070209 The tests expect this to be returned as tuples,
212
# but the code itself doesn't really depend on that.
213
# Figure out a way to not require the overhead of turning the
214
# list back into tuples.
215
lines = [tuple(line.split(' ', 1)) for line in content]
216
return KnitContent(lines)
218
def parse_line_delta_iter(self, lines):
219
return iter(self.parse_line_delta(lines))
221
def parse_line_delta(self, lines, version_id):
222
"""Convert a line based delta into internal representation.
224
line delta is in the form of:
225
intstart intend intcount
227
revid(utf8) newline\n
228
internal representation is
229
(start, end, count, [1..count tuples (revid, newline)])
236
def cache_and_return(line):
237
origin, text = line.split(' ', 1)
238
return cache.setdefault(origin, origin), text
240
# walk through the lines parsing.
242
start, end, count = [int(n) for n in header.split(',')]
243
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
244
result.append((start, end, count, contents))
247
def get_fulltext_content(self, lines):
248
"""Extract just the content lines from a fulltext."""
249
return (line.split(' ', 1)[1] for line in lines)
251
def get_linedelta_content(self, lines):
252
"""Extract just the content from a line delta.
254
This doesn't return all of the extra information stored in a delta.
255
Only the actual content lines.
260
header = header.split(',')
261
count = int(header[2])
262
for i in xrange(count):
263
origin, text = next().split(' ', 1)
266
def lower_fulltext(self, content):
267
"""convert a fulltext content record into a serializable form.
269
see parse_fulltext which this inverts.
271
# TODO: jam 20070209 We only do the caching thing to make sure that
272
# the origin is a valid utf-8 line, eventually we could remove it
273
return ['%s %s' % (o, t) for o, t in content._lines]
275
def lower_line_delta(self, delta):
276
"""convert a delta into a serializable form.
278
See parse_line_delta which this inverts.
280
# TODO: jam 20070209 We only do the caching thing to make sure that
281
# the origin is a valid utf-8 line, eventually we could remove it
283
for start, end, c, lines in delta:
284
out.append('%d,%d,%d\n' % (start, end, c))
285
out.extend(origin + ' ' + text
286
for origin, text in lines)
290
class KnitPlainFactory(_KnitFactory):
291
"""Factory for creating plain Content objects."""
295
def parse_fulltext(self, content, version_id):
296
"""This parses an unannotated fulltext.
298
Note that this is not a noop - the internal representation
299
has (versionid, line) - its just a constant versionid.
301
return self.make(content, version_id)
303
def parse_line_delta_iter(self, lines, version_id):
305
num_lines = len(lines)
306
while cur < num_lines:
309
start, end, c = [int(n) for n in header.split(',')]
310
yield start, end, c, zip([version_id] * c, lines[cur:cur+c])
313
def parse_line_delta(self, lines, version_id):
314
return list(self.parse_line_delta_iter(lines, version_id))
316
def get_fulltext_content(self, lines):
317
"""Extract just the content lines from a fulltext."""
320
def get_linedelta_content(self, lines):
321
"""Extract just the content from a line delta.
323
This doesn't return all of the extra information stored in a delta.
324
Only the actual content lines.
329
header = header.split(',')
330
count = int(header[2])
331
for i in xrange(count):
334
def lower_fulltext(self, content):
335
return content.text()
337
def lower_line_delta(self, delta):
339
for start, end, c, lines in delta:
340
out.append('%d,%d,%d\n' % (start, end, c))
341
out.extend([text for origin, text in lines])
345
def make_empty_knit(transport, relpath):
346
"""Construct a empty knit at the specified location."""
347
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
350
class KnitVersionedFile(VersionedFile):
351
"""Weave-like structure with faster random access.
353
A knit stores a number of texts and a summary of the relationships
354
between them. Texts are identified by a string version-id. Texts
355
are normally stored and retrieved as a series of lines, but can
356
also be passed as single strings.
358
Lines are stored with the trailing newline (if any) included, to
359
avoid special cases for files with no final newline. Lines are
360
composed of 8-bit characters, not unicode. The combination of
361
these approaches should mean any 'binary' file can be safely
362
stored and retrieved.
365
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
366
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
367
create=False, create_parent_dir=False, delay_create=False,
368
dir_mode=None, index=None, access_method=None):
369
"""Construct a knit at location specified by relpath.
371
:param create: If not True, only open an existing knit.
372
:param create_parent_dir: If True, create the parent directory if
373
creating the file fails. (This is used for stores with
374
hash-prefixes that may not exist yet)
375
:param delay_create: The calling code is aware that the knit won't
376
actually be created until the first data is stored.
377
:param index: An index to use for the knit.
379
if deprecated_passed(basis_knit):
380
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
381
" deprecated as of bzr 0.9.",
382
DeprecationWarning, stacklevel=2)
383
if access_mode is None:
385
super(KnitVersionedFile, self).__init__(access_mode)
386
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
387
self.transport = transport
388
self.filename = relpath
389
self.factory = factory or KnitAnnotateFactory()
390
self.writable = (access_mode == 'w')
393
self._max_delta_chain = 200
396
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
397
access_mode, create=create, file_mode=file_mode,
398
create_parent_dir=create_parent_dir, delay_create=delay_create,
402
if access_method is None:
403
_access = _KnitAccess(transport, relpath + DATA_SUFFIX, file_mode, dir_mode,
404
((create and not len(self)) and delay_create), create_parent_dir)
406
_access = access_method
407
if create and not len(self) and not delay_create:
409
self._data = _KnitData(_access)
412
return '%s(%s)' % (self.__class__.__name__,
413
self.transport.abspath(self.filename))
415
def _check_should_delta(self, first_parents):
416
"""Iterate back through the parent listing, looking for a fulltext.
418
This is used when we want to decide whether to add a delta or a new
419
fulltext. It searches for _max_delta_chain parents. When it finds a
420
fulltext parent, it sees if the total size of the deltas leading up to
421
it is large enough to indicate that we want a new full text anyway.
423
Return True if we should create a new delta, False if we should use a
428
delta_parents = first_parents
429
for count in xrange(self._max_delta_chain):
430
parent = delta_parents[0]
431
method = self._index.get_method(parent)
432
index, pos, size = self._index.get_position(parent)
433
if method == 'fulltext':
437
delta_parents = self._index.get_parents(parent)
439
# We couldn't find a fulltext, so we must create a new one
442
return fulltext_size > delta_size
444
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
445
"""See VersionedFile._add_delta()."""
446
self._check_add(version_id, []) # should we check the lines ?
447
self._check_versions_present(parents)
451
for parent in parents:
452
if not self.has_version(parent):
453
ghosts.append(parent)
455
present_parents.append(parent)
457
if delta_parent is None:
458
# reconstitute as full text.
459
assert len(delta) == 1 or len(delta) == 0
461
assert delta[0][0] == 0
462
assert delta[0][1] == 0, delta[0][1]
463
return super(KnitVersionedFile, self)._add_delta(version_id,
474
options.append('no-eol')
476
if delta_parent is not None:
477
# determine the current delta chain length.
478
# To speed the extract of texts the delta chain is limited
479
# to a fixed number of deltas. This should minimize both
480
# I/O and the time spend applying deltas.
481
# The window was changed to a maximum of 200 deltas, but also added
482
# was a check that the total compressed size of the deltas is
483
# smaller than the compressed size of the fulltext.
484
if not self._check_should_delta([delta_parent]):
485
# We don't want a delta here, just do a normal insertion.
486
return super(KnitVersionedFile, self)._add_delta(version_id,
493
options.append('line-delta')
494
store_lines = self.factory.lower_line_delta(delta)
496
access_memo = self._data.add_record(version_id, digest, store_lines)
497
self._index.add_version(version_id, options, access_memo, parents)
499
def _add_raw_records(self, records, data):
500
"""Add all the records 'records' with data pre-joined in 'data'.
502
:param records: A list of tuples(version_id, options, parents, size).
503
:param data: The data for the records. When it is written, the records
504
are adjusted to have pos pointing into data by the sum of
505
the preceding records sizes.
508
raw_record_sizes = [record[3] for record in records]
509
positions = self._data.add_raw_records(raw_record_sizes, data)
512
for (version_id, options, parents, size), access_memo in zip(
514
index_entries.append((version_id, options, access_memo, parents))
515
if self._data._do_cache:
516
self._data._cache[version_id] = data[offset:offset+size]
518
self._index.add_versions(index_entries)
520
def enable_cache(self):
521
"""Start caching data for this knit"""
522
self._data.enable_cache()
524
def clear_cache(self):
525
"""Clear the data cache only."""
526
self._data.clear_cache()
528
def copy_to(self, name, transport):
529
"""See VersionedFile.copy_to()."""
530
# copy the current index to a temp index to avoid racing with local
532
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
533
self.transport.get(self._index._filename))
535
f = self._data._open_file()
537
transport.put_file(name + DATA_SUFFIX, f)
540
# move the copied index into place
541
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
543
def create_empty(self, name, transport, mode=None):
544
return KnitVersionedFile(name, transport, factory=self.factory,
545
delta=self.delta, create=True)
547
def _fix_parents(self, version_id, new_parents):
548
"""Fix the parents list for version.
550
This is done by appending a new version to the index
551
with identical data except for the parents list.
552
the parents list must be a superset of the current
555
current_values = self._index._cache[version_id]
556
assert set(current_values[4]).difference(set(new_parents)) == set()
557
self._index.add_version(version_id,
559
(None, current_values[2], current_values[3]),
562
def _extract_blocks(self, version_id, source, target):
563
if self._index.get_method(version_id) != 'line-delta':
565
parent, sha1, noeol, delta = self.get_delta(version_id)
566
return KnitContent.get_line_delta_blocks(delta, source, target)
568
def get_delta(self, version_id):
569
"""Get a delta for constructing version from some other version."""
570
version_id = osutils.safe_revision_id(version_id)
571
self.check_not_reserved_id(version_id)
572
if not self.has_version(version_id):
573
raise RevisionNotPresent(version_id, self.filename)
575
parents = self.get_parents(version_id)
580
index_memo = self._index.get_position(version_id)
581
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
582
noeol = 'no-eol' in self._index.get_options(version_id)
583
if 'fulltext' == self._index.get_method(version_id):
584
new_content = self.factory.parse_fulltext(data, version_id)
585
if parent is not None:
586
reference_content = self._get_content(parent)
587
old_texts = reference_content.text()
590
new_texts = new_content.text()
591
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
592
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
594
delta = self.factory.parse_line_delta(data, version_id)
595
return parent, sha1, noeol, delta
597
def get_graph_with_ghosts(self):
598
"""See VersionedFile.get_graph_with_ghosts()."""
599
graph_items = self._index.get_graph()
600
return dict(graph_items)
602
def get_sha1(self, version_id):
603
return self.get_sha1s([version_id])[0]
605
def get_sha1s(self, version_ids):
606
"""See VersionedFile.get_sha1()."""
607
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
608
record_map = self._get_record_map(version_ids)
609
# record entry 2 is the 'digest'.
610
return [record_map[v][2] for v in version_ids]
614
"""See VersionedFile.get_suffixes()."""
615
return [DATA_SUFFIX, INDEX_SUFFIX]
617
def has_ghost(self, version_id):
618
"""True if there is a ghost reference in the file to version_id."""
619
version_id = osutils.safe_revision_id(version_id)
621
if self.has_version(version_id):
623
# optimisable if needed by memoising the _ghosts set.
624
items = self._index.get_graph()
625
for node, parents in items:
626
for parent in parents:
627
if parent not in self._index._cache:
628
if parent == version_id:
633
"""See VersionedFile.versions."""
634
return self._index.get_versions()
636
def has_version(self, version_id):
637
"""See VersionedFile.has_version."""
638
version_id = osutils.safe_revision_id(version_id)
639
return self._index.has_version(version_id)
641
__contains__ = has_version
643
def _merge_annotations(self, content, parents, parent_texts={},
644
delta=None, annotated=None):
645
"""Merge annotations for content. This is done by comparing
646
the annotations based on changed to the text.
650
for parent_id in parents:
651
merge_content = self._get_content(parent_id, parent_texts)
652
seq = patiencediff.PatienceSequenceMatcher(
653
None, merge_content.text(), content.text())
654
if delta_seq is None:
655
# setup a delta seq to reuse.
657
for i, j, n in seq.get_matching_blocks():
660
# this appears to copy (origin, text) pairs across to the new
661
# content for any line that matches the last-checked parent.
662
# FIXME: save the sequence control data for delta compression
663
# against the most relevant parent rather than rediffing.
664
content._lines[j:j+n] = merge_content._lines[i:i+n]
667
reference_content = self._get_content(parents[0], parent_texts)
668
new_texts = content.text()
669
old_texts = reference_content.text()
670
delta_seq = patiencediff.PatienceSequenceMatcher(
671
None, old_texts, new_texts)
672
return self._make_line_delta(delta_seq, content)
674
def _make_line_delta(self, delta_seq, new_content):
675
"""Generate a line delta from delta_seq and new_content."""
677
for op in delta_seq.get_opcodes():
680
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
683
def _get_components_positions(self, version_ids):
684
"""Produce a map of position data for the components of versions.
686
This data is intended to be used for retrieving the knit records.
688
A dict of version_id to (method, data_pos, data_size, next) is
690
method is the way referenced data should be applied.
691
data_pos is the position of the data in the knit.
692
data_size is the size of the data in the knit.
693
next is the build-parent of the version, or None for fulltexts.
696
for version_id in version_ids:
699
while cursor is not None and cursor not in component_data:
700
method = self._index.get_method(cursor)
701
if method == 'fulltext':
704
next = self.get_parents(cursor)[0]
705
index_memo = self._index.get_position(cursor)
706
component_data[cursor] = (method, index_memo, next)
708
return component_data
710
def _get_content(self, version_id, parent_texts={}):
711
"""Returns a content object that makes up the specified
713
if not self.has_version(version_id):
714
raise RevisionNotPresent(version_id, self.filename)
716
cached_version = parent_texts.get(version_id, None)
717
if cached_version is not None:
718
return cached_version
720
text_map, contents_map = self._get_content_maps([version_id])
721
return contents_map[version_id]
723
def _check_versions_present(self, version_ids):
724
"""Check that all specified versions are present."""
725
self._index.check_versions_present(version_ids)
727
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
728
"""See VersionedFile.add_lines_with_ghosts()."""
729
self._check_add(version_id, lines)
730
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
732
def _add_lines(self, version_id, parents, lines, parent_texts):
733
"""See VersionedFile.add_lines."""
734
self._check_add(version_id, lines)
735
self._check_versions_present(parents)
736
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
738
def _check_add(self, version_id, lines):
739
"""check that version_id and lines are safe to add."""
740
assert self.writable, "knit is not opened for write"
741
### FIXME escape. RBC 20060228
742
if contains_whitespace(version_id):
743
raise InvalidRevisionId(version_id, self.filename)
744
self.check_not_reserved_id(version_id)
745
if self.has_version(version_id):
746
raise RevisionAlreadyPresent(version_id, self.filename)
747
self._check_lines_not_unicode(lines)
748
self._check_lines_are_lines(lines)
750
def _add(self, version_id, lines, parents, delta, parent_texts):
751
"""Add a set of lines on top of version specified by parents.
753
If delta is true, compress the text as a line-delta against
756
Any versions not present will be converted into ghosts.
758
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
759
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
760
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
761
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
762
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
763
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
764
# +1383 0 8.0370 8.0370 +<len>
765
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
766
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
767
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
768
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
772
if parent_texts is None:
774
for parent in parents:
775
if not self.has_version(parent):
776
ghosts.append(parent)
778
present_parents.append(parent)
780
if delta and not len(present_parents):
783
digest = sha_strings(lines)
786
if lines[-1][-1] != '\n':
787
options.append('no-eol')
788
lines[-1] = lines[-1] + '\n'
790
if len(present_parents) and delta:
791
# To speed the extract of texts the delta chain is limited
792
# to a fixed number of deltas. This should minimize both
793
# I/O and the time spend applying deltas.
794
delta = self._check_should_delta(present_parents)
796
assert isinstance(version_id, str)
797
lines = self.factory.make(lines, version_id)
798
if delta or (self.factory.annotated and len(present_parents) > 0):
799
# Merge annotations from parent texts if so is needed.
800
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
801
delta, self.factory.annotated)
804
options.append('line-delta')
805
store_lines = self.factory.lower_line_delta(delta_hunks)
807
options.append('fulltext')
808
store_lines = self.factory.lower_fulltext(lines)
810
access_memo = self._data.add_record(version_id, digest, store_lines)
811
self._index.add_version(version_id, options, access_memo, parents)
814
def check(self, progress_bar=None):
815
"""See VersionedFile.check()."""
817
def _clone_text(self, new_version_id, old_version_id, parents):
818
"""See VersionedFile.clone_text()."""
819
# FIXME RBC 20060228 make fast by only inserting an index with null
821
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
823
def get_lines(self, version_id):
824
"""See VersionedFile.get_lines()."""
825
return self.get_line_list([version_id])[0]
827
def _get_record_map(self, version_ids):
828
"""Produce a dictionary of knit records.
830
The keys are version_ids, the values are tuples of (method, content,
832
method is the way the content should be applied.
833
content is a KnitContent object.
834
digest is the SHA1 digest of this version id after all steps are done
835
next is the build-parent of the version, i.e. the leftmost ancestor.
836
If the method is fulltext, next will be None.
838
position_map = self._get_components_positions(version_ids)
839
# c = component_id, m = method, i_m = index_memo, n = next
840
records = [(c, i_m) for c, (m, i_m, n) in position_map.iteritems()]
842
for component_id, content, digest in \
843
self._data.read_records_iter(records):
844
method, index_memo, next = position_map[component_id]
845
record_map[component_id] = method, content, digest, next
849
def get_text(self, version_id):
850
"""See VersionedFile.get_text"""
851
return self.get_texts([version_id])[0]
853
def get_texts(self, version_ids):
854
return [''.join(l) for l in self.get_line_list(version_ids)]
856
def get_line_list(self, version_ids):
857
"""Return the texts of listed versions as a list of strings."""
858
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
859
for version_id in version_ids:
860
self.check_not_reserved_id(version_id)
861
text_map, content_map = self._get_content_maps(version_ids)
862
return [text_map[v] for v in version_ids]
864
_get_lf_split_line_list = get_line_list
866
def _get_content_maps(self, version_ids):
867
"""Produce maps of text and KnitContents
869
:return: (text_map, content_map) where text_map contains the texts for
870
the requested versions and content_map contains the KnitContents.
871
Both dicts take version_ids as their keys.
873
for version_id in version_ids:
874
if not self.has_version(version_id):
875
raise RevisionNotPresent(version_id, self.filename)
876
record_map = self._get_record_map(version_ids)
881
for version_id in version_ids:
884
while cursor is not None:
885
method, data, digest, next = record_map[cursor]
886
components.append((cursor, method, data, digest))
887
if cursor in content_map:
892
for component_id, method, data, digest in reversed(components):
893
if component_id in content_map:
894
content = content_map[component_id]
896
if method == 'fulltext':
897
assert content is None
898
content = self.factory.parse_fulltext(data, version_id)
899
elif method == 'line-delta':
900
delta = self.factory.parse_line_delta(data, version_id)
901
content = content.copy()
902
content._lines = self._apply_delta(content._lines,
904
content_map[component_id] = content
906
if 'no-eol' in self._index.get_options(version_id):
907
content = content.copy()
908
line = content._lines[-1][1].rstrip('\n')
909
content._lines[-1] = (content._lines[-1][0], line)
910
final_content[version_id] = content
912
# digest here is the digest from the last applied component.
913
text = content.text()
914
if sha_strings(text) != digest:
915
raise KnitCorrupt(self.filename,
916
'sha-1 does not match %s' % version_id)
918
text_map[version_id] = text
919
return text_map, final_content
921
def iter_lines_added_or_present_in_versions(self, version_ids=None,
923
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
924
if version_ids is None:
925
version_ids = self.versions()
927
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
929
pb = progress.DummyProgress()
930
# we don't care about inclusions, the caller cares.
931
# but we need to setup a list of records to visit.
932
# we need version_id, position, length
933
version_id_records = []
934
requested_versions = set(version_ids)
935
# filter for available versions
936
for version_id in requested_versions:
937
if not self.has_version(version_id):
938
raise RevisionNotPresent(version_id, self.filename)
939
# get a in-component-order queue:
940
for version_id in self.versions():
941
if version_id in requested_versions:
942
index_memo = self._index.get_position(version_id)
943
version_id_records.append((version_id, index_memo))
945
total = len(version_id_records)
946
for version_idx, (version_id, data, sha_value) in \
947
enumerate(self._data.read_records_iter(version_id_records)):
948
pb.update('Walking content.', version_idx, total)
949
method = self._index.get_method(version_id)
951
assert method in ('fulltext', 'line-delta')
952
if method == 'fulltext':
953
line_iterator = self.factory.get_fulltext_content(data)
955
line_iterator = self.factory.get_linedelta_content(data)
956
for line in line_iterator:
959
pb.update('Walking content.', total, total)
961
def iter_parents(self, version_ids):
962
"""Iterate through the parents for many version ids.
964
:param version_ids: An iterable yielding version_ids.
965
:return: An iterator that yields (version_id, parents). Requested
966
version_ids not present in the versioned file are simply skipped.
967
The order is undefined, allowing for different optimisations in
968
the underlying implementation.
970
version_ids = [osutils.safe_revision_id(version_id) for
971
version_id in version_ids]
972
return self._index.iter_parents(version_ids)
974
def num_versions(self):
975
"""See VersionedFile.num_versions()."""
976
return self._index.num_versions()
978
__len__ = num_versions
980
def annotate_iter(self, version_id):
981
"""See VersionedFile.annotate_iter."""
982
version_id = osutils.safe_revision_id(version_id)
983
content = self._get_content(version_id)
984
for origin, text in content.annotate_iter():
987
def get_parents(self, version_id):
988
"""See VersionedFile.get_parents."""
991
# 52554 calls in 1264 872 internal down from 3674
992
version_id = osutils.safe_revision_id(version_id)
994
return self._index.get_parents(version_id)
996
raise RevisionNotPresent(version_id, self.filename)
998
def get_parents_with_ghosts(self, version_id):
999
"""See VersionedFile.get_parents."""
1000
version_id = osutils.safe_revision_id(version_id)
1002
return self._index.get_parents_with_ghosts(version_id)
1004
raise RevisionNotPresent(version_id, self.filename)
1006
def get_ancestry(self, versions, topo_sorted=True):
1007
"""See VersionedFile.get_ancestry."""
1008
if isinstance(versions, basestring):
1009
versions = [versions]
1012
versions = [osutils.safe_revision_id(v) for v in versions]
1013
return self._index.get_ancestry(versions, topo_sorted)
1015
def get_ancestry_with_ghosts(self, versions):
1016
"""See VersionedFile.get_ancestry_with_ghosts."""
1017
if isinstance(versions, basestring):
1018
versions = [versions]
1021
versions = [osutils.safe_revision_id(v) for v in versions]
1022
return self._index.get_ancestry_with_ghosts(versions)
1024
def plan_merge(self, ver_a, ver_b):
1025
"""See VersionedFile.plan_merge."""
1026
ver_a = osutils.safe_revision_id(ver_a)
1027
ver_b = osutils.safe_revision_id(ver_b)
1028
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1030
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1031
annotated_a = self.annotate(ver_a)
1032
annotated_b = self.annotate(ver_b)
1033
return merge._plan_annotate_merge(annotated_a, annotated_b,
1034
ancestors_a, ancestors_b)
1037
class _KnitComponentFile(object):
1038
"""One of the files used to implement a knit database"""
1040
def __init__(self, transport, filename, mode, file_mode=None,
1041
create_parent_dir=False, dir_mode=None):
1042
self._transport = transport
1043
self._filename = filename
1045
self._file_mode = file_mode
1046
self._dir_mode = dir_mode
1047
self._create_parent_dir = create_parent_dir
1048
self._need_to_create = False
1050
def _full_path(self):
1051
"""Return the full path to this file."""
1052
return self._transport.base + self._filename
1054
def check_header(self, fp):
1055
line = fp.readline()
1057
# An empty file can actually be treated as though the file doesn't
1059
raise errors.NoSuchFile(self._full_path())
1060
if line != self.HEADER:
1061
raise KnitHeaderError(badline=line,
1062
filename=self._transport.abspath(self._filename))
1065
return '%s(%s)' % (self.__class__.__name__, self._filename)
1068
class _KnitIndex(_KnitComponentFile):
1069
"""Manages knit index file.
1071
The index is already kept in memory and read on startup, to enable
1072
fast lookups of revision information. The cursor of the index
1073
file is always pointing to the end, making it easy to append
1076
_cache is a cache for fast mapping from version id to a Index
1079
_history is a cache for fast mapping from indexes to version ids.
1081
The index data format is dictionary compressed when it comes to
1082
parent references; a index entry may only have parents that with a
1083
lover index number. As a result, the index is topological sorted.
1085
Duplicate entries may be written to the index for a single version id
1086
if this is done then the latter one completely replaces the former:
1087
this allows updates to correct version and parent information.
1088
Note that the two entries may share the delta, and that successive
1089
annotations and references MUST point to the first entry.
1091
The index file on disc contains a header, followed by one line per knit
1092
record. The same revision can be present in an index file more than once.
1093
The first occurrence gets assigned a sequence number starting from 0.
1095
The format of a single line is
1096
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1097
REVISION_ID is a utf8-encoded revision id
1098
FLAGS is a comma separated list of flags about the record. Values include
1099
no-eol, line-delta, fulltext.
1100
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1101
that the the compressed data starts at.
1102
LENGTH is the ascii representation of the length of the data file.
1103
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1105
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1106
revision id already in the knit that is a parent of REVISION_ID.
1107
The ' :' marker is the end of record marker.
1110
when a write is interrupted to the index file, it will result in a line
1111
that does not end in ' :'. If the ' :' is not present at the end of a line,
1112
or at the end of the file, then the record that is missing it will be
1113
ignored by the parser.
1115
When writing new records to the index file, the data is preceded by '\n'
1116
to ensure that records always start on new lines even if the last write was
1117
interrupted. As a result its normal for the last line in the index to be
1118
missing a trailing newline. One can be added with no harmful effects.
1121
HEADER = "# bzr knit index 8\n"
1123
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1124
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1126
def _cache_version(self, version_id, options, pos, size, parents):
1127
"""Cache a version record in the history array and index cache.
1129
This is inlined into _load_data for performance. KEEP IN SYNC.
1130
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1133
# only want the _history index to reference the 1st index entry
1135
if version_id not in self._cache:
1136
index = len(self._history)
1137
self._history.append(version_id)
1139
index = self._cache[version_id][5]
1140
self._cache[version_id] = (version_id,
1147
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1148
create_parent_dir=False, delay_create=False, dir_mode=None):
1149
_KnitComponentFile.__init__(self, transport, filename, mode,
1150
file_mode=file_mode,
1151
create_parent_dir=create_parent_dir,
1154
# position in _history is the 'official' index for a revision
1155
# but the values may have come from a newer entry.
1156
# so - wc -l of a knit index is != the number of unique names
1160
fp = self._transport.get(self._filename)
1162
# _load_data may raise NoSuchFile if the target knit is
1164
_load_data(self, fp)
1168
if mode != 'w' or not create:
1171
self._need_to_create = True
1173
self._transport.put_bytes_non_atomic(
1174
self._filename, self.HEADER, mode=self._file_mode)
1176
def get_graph(self):
1177
"""Return a list of the node:parents lists from this knit index."""
1178
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1180
def get_ancestry(self, versions, topo_sorted=True):
1181
"""See VersionedFile.get_ancestry."""
1182
# get a graph of all the mentioned versions:
1184
pending = set(versions)
1187
version = pending.pop()
1190
parents = [p for p in cache[version][4] if p in cache]
1192
raise RevisionNotPresent(version, self._filename)
1193
# if not completed and not a ghost
1194
pending.update([p for p in parents if p not in graph])
1195
graph[version] = parents
1198
return topo_sort(graph.items())
1200
def get_ancestry_with_ghosts(self, versions):
1201
"""See VersionedFile.get_ancestry_with_ghosts."""
1202
# get a graph of all the mentioned versions:
1203
self.check_versions_present(versions)
1206
pending = set(versions)
1208
version = pending.pop()
1210
parents = cache[version][4]
1216
pending.update([p for p in parents if p not in graph])
1217
graph[version] = parents
1218
return topo_sort(graph.items())
1220
def iter_parents(self, version_ids):
1221
"""Iterate through the parents for many version ids.
1223
:param version_ids: An iterable yielding version_ids.
1224
:return: An iterator that yields (version_id, parents). Requested
1225
version_ids not present in the versioned file are simply skipped.
1226
The order is undefined, allowing for different optimisations in
1227
the underlying implementation.
1229
for version_id in version_ids:
1231
yield version_id, tuple(self.get_parents(version_id))
1235
def num_versions(self):
1236
return len(self._history)
1238
__len__ = num_versions
1240
def get_versions(self):
1241
"""Get all the versions in the file. not topologically sorted."""
1242
return self._history
1244
def _version_list_to_index(self, versions):
1247
for version in versions:
1248
if version in cache:
1249
# -- inlined lookup() --
1250
result_list.append(str(cache[version][5]))
1251
# -- end lookup () --
1253
result_list.append('.' + version)
1254
return ' '.join(result_list)
1256
def add_version(self, version_id, options, index_memo, parents):
1257
"""Add a version record to the index."""
1258
self.add_versions(((version_id, options, index_memo, parents),))
1260
def add_versions(self, versions):
1261
"""Add multiple versions to the index.
1263
:param versions: a list of tuples:
1264
(version_id, options, pos, size, parents).
1267
orig_history = self._history[:]
1268
orig_cache = self._cache.copy()
1271
for version_id, options, (index, pos, size), parents in versions:
1272
line = "\n%s %s %s %s %s :" % (version_id,
1276
self._version_list_to_index(parents))
1277
assert isinstance(line, str), \
1278
'content must be utf-8 encoded: %r' % (line,)
1280
self._cache_version(version_id, options, pos, size, parents)
1281
if not self._need_to_create:
1282
self._transport.append_bytes(self._filename, ''.join(lines))
1285
sio.write(self.HEADER)
1286
sio.writelines(lines)
1288
self._transport.put_file_non_atomic(self._filename, sio,
1289
create_parent_dir=self._create_parent_dir,
1290
mode=self._file_mode,
1291
dir_mode=self._dir_mode)
1292
self._need_to_create = False
1294
# If any problems happen, restore the original values and re-raise
1295
self._history = orig_history
1296
self._cache = orig_cache
1299
def has_version(self, version_id):
1300
"""True if the version is in the index."""
1301
return version_id in self._cache
1303
def get_position(self, version_id):
1304
"""Return details needed to access the version.
1306
.kndx indices do not support split-out data, so return None for the
1309
:return: a tuple (None, data position, size) to hand to the access
1310
logic to get the record.
1312
entry = self._cache[version_id]
1313
return None, entry[2], entry[3]
1315
def get_method(self, version_id):
1316
"""Return compression method of specified version."""
1317
options = self._cache[version_id][1]
1318
if 'fulltext' in options:
1321
if 'line-delta' not in options:
1322
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1325
def get_options(self, version_id):
1326
"""Return a string represention options.
1330
return self._cache[version_id][1]
1332
def get_parents(self, version_id):
1333
"""Return parents of specified version ignoring ghosts."""
1334
return [parent for parent in self._cache[version_id][4]
1335
if parent in self._cache]
1337
def get_parents_with_ghosts(self, version_id):
1338
"""Return parents of specified version with ghosts."""
1339
return self._cache[version_id][4]
1341
def check_versions_present(self, version_ids):
1342
"""Check that all specified versions are present."""
1344
for version_id in version_ids:
1345
if version_id not in cache:
1346
raise RevisionNotPresent(version_id, self._filename)
1349
class KnitGraphIndex(object):
1350
"""A knit index that builds on GraphIndex."""
1352
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1353
"""Construct a KnitGraphIndex on a graph_index.
1355
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1356
:param deltas: Allow delta-compressed records.
1357
:param add_callback: If not None, allow additions to the index and call
1358
this callback with a list of added GraphIndex nodes:
1359
[(node, value, node_refs), ...]
1360
:param parents: If True, record knits parents, if not do not record
1363
self._graph_index = graph_index
1364
self._deltas = deltas
1365
self._add_callback = add_callback
1366
self._parents = parents
1367
if deltas and not parents:
1368
raise KnitCorrupt(self, "Cannot do delta compression without "
1371
def _get_entries(self, keys, check_present=False):
1372
"""Get the entries for keys.
1374
:param keys: An iterable of index keys, - 1-tuples.
1379
for node in self._graph_index.iter_entries(keys):
1381
found_keys.add(node[1])
1383
# adapt parentless index to the rest of the code.
1384
for node in self._graph_index.iter_entries(keys):
1385
yield node[0], node[1], node[2], ()
1386
found_keys.add(node[1])
1388
missing_keys = keys.difference(found_keys)
1390
raise RevisionNotPresent(missing_keys.pop(), self)
1392
def _present_keys(self, version_ids):
1394
node[1] for node in self._get_entries(version_ids)])
1396
def _parentless_ancestry(self, versions):
1397
"""Honour the get_ancestry API for parentless knit indices."""
1398
wanted_keys = self._version_ids_to_keys(versions)
1399
present_keys = self._present_keys(wanted_keys)
1400
missing = set(wanted_keys).difference(present_keys)
1402
raise RevisionNotPresent(missing.pop(), self)
1403
return list(self._keys_to_version_ids(present_keys))
1405
def get_ancestry(self, versions, topo_sorted=True):
1406
"""See VersionedFile.get_ancestry."""
1407
if not self._parents:
1408
return self._parentless_ancestry(versions)
1409
# XXX: This will do len(history) index calls - perhaps
1410
# it should be altered to be a index core feature?
1411
# get a graph of all the mentioned versions:
1414
versions = self._version_ids_to_keys(versions)
1415
pending = set(versions)
1417
# get all pending nodes
1418
this_iteration = pending
1419
new_nodes = self._get_entries(this_iteration)
1422
for (index, key, value, node_refs) in new_nodes:
1423
# dont ask for ghosties - otherwise
1424
# we we can end up looping with pending
1425
# being entirely ghosted.
1426
graph[key] = [parent for parent in node_refs[0]
1427
if parent not in ghosts]
1429
for parent in graph[key]:
1430
# dont examine known nodes again
1435
ghosts.update(this_iteration.difference(found))
1436
if versions.difference(graph):
1437
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1439
result_keys = topo_sort(graph.items())
1441
result_keys = graph.iterkeys()
1442
return [key[0] for key in result_keys]
1444
def get_ancestry_with_ghosts(self, versions):
1445
"""See VersionedFile.get_ancestry."""
1446
if not self._parents:
1447
return self._parentless_ancestry(versions)
1448
# XXX: This will do len(history) index calls - perhaps
1449
# it should be altered to be a index core feature?
1450
# get a graph of all the mentioned versions:
1452
versions = self._version_ids_to_keys(versions)
1453
pending = set(versions)
1455
# get all pending nodes
1456
this_iteration = pending
1457
new_nodes = self._get_entries(this_iteration)
1459
for (index, key, value, node_refs) in new_nodes:
1460
graph[key] = node_refs[0]
1462
for parent in graph[key]:
1463
# dont examine known nodes again
1467
missing_versions = this_iteration.difference(graph)
1468
missing_needed = versions.intersection(missing_versions)
1470
raise RevisionNotPresent(missing_needed.pop(), self)
1471
for missing_version in missing_versions:
1472
# add a key, no parents
1473
graph[missing_version] = []
1474
pending.discard(missing_version) # don't look for it
1475
result_keys = topo_sort(graph.items())
1476
return [key[0] for key in result_keys]
1478
def get_graph(self):
1479
"""Return a list of the node:parents lists from this knit index."""
1480
if not self._parents:
1481
return [(key, ()) for key in self.get_versions()]
1483
for index, key, value, refs in self._graph_index.iter_all_entries():
1484
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1487
def iter_parents(self, version_ids):
1488
"""Iterate through the parents for many version ids.
1490
:param version_ids: An iterable yielding version_ids.
1491
:return: An iterator that yields (version_id, parents). Requested
1492
version_ids not present in the versioned file are simply skipped.
1493
The order is undefined, allowing for different optimisations in
1494
the underlying implementation.
1497
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1499
present_parents = set()
1500
for node in all_nodes:
1501
all_parents.update(node[3][0])
1502
# any node we are querying must be present
1503
present_parents.add(node[1])
1504
unknown_parents = all_parents.difference(present_parents)
1505
present_parents.update(self._present_keys(unknown_parents))
1506
for node in all_nodes:
1508
for parent in node[3][0]:
1509
if parent in present_parents:
1510
parents.append(parent[0])
1511
yield node[1][0], tuple(parents)
1513
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1514
yield node[1][0], ()
1516
def num_versions(self):
1517
return len(list(self._graph_index.iter_all_entries()))
1519
__len__ = num_versions
1521
def get_versions(self):
1522
"""Get all the versions in the file. not topologically sorted."""
1523
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1525
def has_version(self, version_id):
1526
"""True if the version is in the index."""
1527
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1529
def _keys_to_version_ids(self, keys):
1530
return tuple(key[0] for key in keys)
1532
def get_position(self, version_id):
1533
"""Return details needed to access the version.
1535
:return: a tuple (index, data position, size) to hand to the access
1536
logic to get the record.
1538
node = self._get_node(version_id)
1539
bits = node[2][1:].split(' ')
1540
return node[0], int(bits[0]), int(bits[1])
1542
def get_method(self, version_id):
1543
"""Return compression method of specified version."""
1544
if not self._deltas:
1546
return self._parent_compression(self._get_node(version_id)[3][1])
1548
def _parent_compression(self, reference_list):
1549
# use the second reference list to decide if this is delta'd or not.
1550
if len(reference_list):
1555
def _get_node(self, version_id):
1556
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1558
def get_options(self, version_id):
1559
"""Return a string represention options.
1563
node = self._get_node(version_id)
1564
if not self._deltas:
1565
options = ['fulltext']
1567
options = [self._parent_compression(node[3][1])]
1568
if node[2][0] == 'N':
1569
options.append('no-eol')
1572
def get_parents(self, version_id):
1573
"""Return parents of specified version ignoring ghosts."""
1574
parents = list(self.iter_parents([version_id]))
1577
raise errors.RevisionNotPresent(version_id, self)
1578
return parents[0][1]
1580
def get_parents_with_ghosts(self, version_id):
1581
"""Return parents of specified version with ghosts."""
1582
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1583
check_present=True))
1584
if not self._parents:
1586
return self._keys_to_version_ids(nodes[0][3][0])
1588
def check_versions_present(self, version_ids):
1589
"""Check that all specified versions are present."""
1590
keys = self._version_ids_to_keys(version_ids)
1591
present = self._present_keys(keys)
1592
missing = keys.difference(present)
1594
raise RevisionNotPresent(missing.pop(), self)
1596
def add_version(self, version_id, options, access_memo, parents):
1597
"""Add a version record to the index."""
1598
return self.add_versions(((version_id, options, access_memo, parents),))
1600
def add_versions(self, versions):
1601
"""Add multiple versions to the index.
1603
This function does not insert data into the Immutable GraphIndex
1604
backing the KnitGraphIndex, instead it prepares data for insertion by
1605
the caller and checks that it is safe to insert then calls
1606
self._add_callback with the prepared GraphIndex nodes.
1608
:param versions: a list of tuples:
1609
(version_id, options, pos, size, parents).
1611
if not self._add_callback:
1612
raise errors.ReadOnlyError(self)
1613
# we hope there are no repositories with inconsistent parentage
1618
for (version_id, options, access_memo, parents) in versions:
1619
index, pos, size = access_memo
1620
key = (version_id, )
1621
parents = tuple((parent, ) for parent in parents)
1622
if 'no-eol' in options:
1626
value += "%d %d" % (pos, size)
1627
if not self._deltas:
1628
if 'line-delta' in options:
1629
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1632
if 'line-delta' in options:
1633
node_refs = (parents, (parents[0],))
1635
node_refs = (parents, ())
1637
node_refs = (parents, )
1640
raise KnitCorrupt(self, "attempt to add node with parents "
1641
"in parentless index.")
1643
keys[key] = (value, node_refs)
1644
present_nodes = self._get_entries(keys)
1645
for (index, key, value, node_refs) in present_nodes:
1646
if (value, node_refs) != keys[key]:
1647
raise KnitCorrupt(self, "inconsistent details in add_versions"
1648
": %s %s" % ((value, node_refs), keys[key]))
1652
for key, (value, node_refs) in keys.iteritems():
1653
result.append((key, value, node_refs))
1655
for key, (value, node_refs) in keys.iteritems():
1656
result.append((key, value))
1657
self._add_callback(result)
1659
def _version_ids_to_keys(self, version_ids):
1660
return set((version_id, ) for version_id in version_ids)
1663
class _KnitAccess(object):
1664
"""Access to knit records in a .knit file."""
1666
def __init__(self, transport, filename, _file_mode, _dir_mode,
1667
_need_to_create, _create_parent_dir):
1668
"""Create a _KnitAccess for accessing and inserting data.
1670
:param transport: The transport the .knit is located on.
1671
:param filename: The filename of the .knit.
1673
self._transport = transport
1674
self._filename = filename
1675
self._file_mode = _file_mode
1676
self._dir_mode = _dir_mode
1677
self._need_to_create = _need_to_create
1678
self._create_parent_dir = _create_parent_dir
1680
def add_raw_records(self, sizes, raw_data):
1681
"""Add raw knit bytes to a storage area.
1683
The data is spooled to whereever the access method is storing data.
1685
:param sizes: An iterable containing the size of each raw data segment.
1686
:param raw_data: A bytestring containing the data.
1687
:return: A list of memos to retrieve the record later. Each memo is a
1688
tuple - (index, pos, length), where the index field is always None
1689
for the .knit access method.
1691
assert type(raw_data) == str, \
1692
'data must be plain bytes was %s' % type(raw_data)
1693
if not self._need_to_create:
1694
base = self._transport.append_bytes(self._filename, raw_data)
1696
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1697
create_parent_dir=self._create_parent_dir,
1698
mode=self._file_mode,
1699
dir_mode=self._dir_mode)
1700
self._need_to_create = False
1704
result.append((None, base, size))
1709
"""IFF this data access has its own storage area, initialise it.
1713
self._transport.put_bytes_non_atomic(self._filename, '',
1714
mode=self._file_mode)
1716
def open_file(self):
1717
"""IFF this data access can be represented as a single file, open it.
1719
For knits that are not mapped to a single file on disk this will
1722
:return: None or a file handle.
1725
return self._transport.get(self._filename)
1730
def get_raw_records(self, memos_for_retrieval):
1731
"""Get the raw bytes for a records.
1733
:param memos_for_retrieval: An iterable containing the (index, pos,
1734
length) memo for retrieving the bytes. The .knit method ignores
1735
the index as there is always only a single file.
1736
:return: An iterator over the bytes of the records.
1738
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
1739
for pos, data in self._transport.readv(self._filename, read_vector):
1743
class _PackAccess(object):
1744
"""Access to knit records via a collection of packs."""
1746
def __init__(self, index_to_packs, writer=None):
1747
"""Create a _PackAccess object.
1749
:param index_to_packs: A dict mapping index objects to the transport
1750
and file names for obtaining data.
1751
:param writer: A tuple (pack.ContainerWriter, write_index) which
1752
contains the pack to write, and the index that reads from it will
1756
self.container_writer = writer[0]
1757
self.write_index = writer[1]
1759
self.container_writer = None
1760
self.write_index = None
1761
self.indices = index_to_packs
1763
def add_raw_records(self, sizes, raw_data):
1764
"""Add raw knit bytes to a storage area.
1766
The data is spooled to the container writer in one bytes-record per
1769
:param sizes: An iterable containing the size of each raw data segment.
1770
:param raw_data: A bytestring containing the data.
1771
:return: A list of memos to retrieve the record later. Each memo is a
1772
tuple - (index, pos, length), where the index field is the
1773
write_index object supplied to the PackAccess object.
1775
assert type(raw_data) == str, \
1776
'data must be plain bytes was %s' % type(raw_data)
1780
p_offset, p_length = self.container_writer.add_bytes_record(
1781
raw_data[offset:offset+size], [])
1783
result.append((self.write_index, p_offset, p_length))
1787
"""Pack based knits do not get individually created."""
1789
def get_raw_records(self, memos_for_retrieval):
1790
"""Get the raw bytes for a records.
1792
:param memos_for_retrieval: An iterable containing the (index, pos,
1793
length) memo for retrieving the bytes. The Pack access method
1794
looks up the pack to use for a given record in its index_to_pack
1796
:return: An iterator over the bytes of the records.
1798
# first pass, group into same-index requests
1800
current_index = None
1801
for (index, offset, length) in memos_for_retrieval:
1802
if current_index == index:
1803
current_list.append((offset, length))
1805
if current_index is not None:
1806
request_lists.append((current_index, current_list))
1807
current_index = index
1808
current_list = [(offset, length)]
1809
# handle the last entry
1810
if current_index is not None:
1811
request_lists.append((current_index, current_list))
1812
for index, offsets in request_lists:
1813
transport, path = self.indices[index]
1814
reader = pack.make_readv_reader(transport, path, offsets)
1815
for names, read_func in reader.iter_records():
1816
yield read_func(None)
1818
def open_file(self):
1819
"""Pack based knits have no single file."""
1822
def set_writer(self, writer, index, (transport, packname)):
1823
"""Set a writer to use for adding data."""
1824
self.indices[index] = (transport, packname)
1825
self.container_writer = writer
1826
self.write_index = index
1829
class _KnitData(object):
1830
"""Manage extraction of data from a KnitAccess, caching and decompressing.
1832
The KnitData class provides the logic for parsing and using knit records,
1833
making use of an access method for the low level read and write operations.
1836
def __init__(self, access):
1837
"""Create a KnitData object.
1839
:param access: The access method to use. Access methods such as
1840
_KnitAccess manage the insertion of raw records and the subsequent
1841
retrieval of the same.
1843
self._access = access
1844
self._checked = False
1845
# TODO: jam 20060713 conceptually, this could spill to disk
1846
# if the cached size gets larger than a certain amount
1847
# but it complicates the model a bit, so for now just use
1848
# a simple dictionary
1850
self._do_cache = False
1852
def enable_cache(self):
1853
"""Enable caching of reads."""
1854
self._do_cache = True
1856
def clear_cache(self):
1857
"""Clear the record cache."""
1858
self._do_cache = False
1861
def _open_file(self):
1862
return self._access.open_file()
1864
def _record_to_data(self, version_id, digest, lines):
1865
"""Convert version_id, digest, lines into a raw data block.
1867
:return: (len, a StringIO instance with the raw data ready to read.)
1870
data_file = GzipFile(None, mode='wb', fileobj=sio)
1872
assert isinstance(version_id, str)
1873
data_file.writelines(chain(
1874
["version %s %d %s\n" % (version_id,
1878
["end %s\n" % version_id]))
1885
def add_raw_records(self, sizes, raw_data):
1886
"""Append a prepared record to the data file.
1888
:param sizes: An iterable containing the size of each raw data segment.
1889
:param raw_data: A bytestring containing the data.
1890
:return: a list of index data for the way the data was stored.
1891
See the access method add_raw_records documentation for more
1894
return self._access.add_raw_records(sizes, raw_data)
1896
def add_record(self, version_id, digest, lines):
1897
"""Write new text record to disk.
1899
Returns index data for retrieving it later, as per add_raw_records.
1901
size, sio = self._record_to_data(version_id, digest, lines)
1902
result = self.add_raw_records([size], sio.getvalue())
1904
self._cache[version_id] = sio.getvalue()
1907
def _parse_record_header(self, version_id, raw_data):
1908
"""Parse a record header for consistency.
1910
:return: the header and the decompressor stream.
1911
as (stream, header_record)
1913
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1915
rec = self._check_header(version_id, df.readline())
1916
except Exception, e:
1917
raise KnitCorrupt(self._access,
1918
"While reading {%s} got %s(%s)"
1919
% (version_id, e.__class__.__name__, str(e)))
1922
def _check_header(self, version_id, line):
1925
raise KnitCorrupt(self._access,
1926
'unexpected number of elements in record header')
1927
if rec[1] != version_id:
1928
raise KnitCorrupt(self._access,
1929
'unexpected version, wanted %r, got %r'
1930
% (version_id, rec[1]))
1933
def _parse_record(self, version_id, data):
1935
# 4168 calls in 2880 217 internal
1936
# 4168 calls to _parse_record_header in 2121
1937
# 4168 calls to readlines in 330
1938
df = GzipFile(mode='rb', fileobj=StringIO(data))
1941
record_contents = df.readlines()
1942
except Exception, e:
1943
raise KnitCorrupt(self._access,
1944
"While reading {%s} got %s(%s)"
1945
% (version_id, e.__class__.__name__, str(e)))
1946
header = record_contents.pop(0)
1947
rec = self._check_header(version_id, header)
1949
last_line = record_contents.pop()
1950
if len(record_contents) != int(rec[2]):
1951
raise KnitCorrupt(self._access,
1952
'incorrect number of lines %s != %s'
1954
% (len(record_contents), int(rec[2]),
1956
if last_line != 'end %s\n' % rec[1]:
1957
raise KnitCorrupt(self._access,
1958
'unexpected version end line %r, wanted %r'
1959
% (last_line, version_id))
1961
return record_contents, rec[3]
1963
def read_records_iter_raw(self, records):
1964
"""Read text records from data file and yield raw data.
1966
This unpacks enough of the text record to validate the id is
1967
as expected but thats all.
1969
# setup an iterator of the external records:
1970
# uses readv so nice and fast we hope.
1972
# grab the disk data needed.
1974
# Don't check _cache if it is empty
1975
needed_offsets = [index_memo for version_id, index_memo
1977
if version_id not in self._cache]
1979
needed_offsets = [index_memo for version_id, index_memo
1982
raw_records = self._access.get_raw_records(needed_offsets)
1984
for version_id, index_memo in records:
1985
if version_id in self._cache:
1986
# This data has already been validated
1987
data = self._cache[version_id]
1989
data = raw_records.next()
1991
self._cache[version_id] = data
1993
# validate the header
1994
df, rec = self._parse_record_header(version_id, data)
1996
yield version_id, data
1998
def read_records_iter(self, records):
1999
"""Read text records from data file and yield result.
2001
The result will be returned in whatever is the fastest to read.
2002
Not by the order requested. Also, multiple requests for the same
2003
record will only yield 1 response.
2004
:param records: A list of (version_id, pos, len) entries
2005
:return: Yields (version_id, contents, digest) in the order
2006
read, not the order requested
2012
# Skip records we have alread seen
2013
yielded_records = set()
2014
needed_records = set()
2015
for record in records:
2016
if record[0] in self._cache:
2017
if record[0] in yielded_records:
2019
yielded_records.add(record[0])
2020
data = self._cache[record[0]]
2021
content, digest = self._parse_record(record[0], data)
2022
yield (record[0], content, digest)
2024
needed_records.add(record)
2025
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2027
needed_records = sorted(set(records), key=operator.itemgetter(1))
2029
if not needed_records:
2032
# The transport optimizes the fetching as well
2033
# (ie, reads continuous ranges.)
2034
raw_data = self._access.get_raw_records(
2035
[index_memo for version_id, index_memo in needed_records])
2037
for (version_id, index_memo), data in \
2038
izip(iter(needed_records), raw_data):
2039
content, digest = self._parse_record(version_id, data)
2041
self._cache[version_id] = data
2042
yield version_id, content, digest
2044
def read_records(self, records):
2045
"""Read records into a dictionary."""
2047
for record_id, content, digest in \
2048
self.read_records_iter(records):
2049
components[record_id] = (content, digest)
2053
class InterKnit(InterVersionedFile):
2054
"""Optimised code paths for knit to knit operations."""
2056
_matching_file_from_factory = KnitVersionedFile
2057
_matching_file_to_factory = KnitVersionedFile
2060
def is_compatible(source, target):
2061
"""Be compatible with knits. """
2063
return (isinstance(source, KnitVersionedFile) and
2064
isinstance(target, KnitVersionedFile))
2065
except AttributeError:
2068
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2069
"""See InterVersionedFile.join."""
2070
assert isinstance(self.source, KnitVersionedFile)
2071
assert isinstance(self.target, KnitVersionedFile)
2073
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2078
pb = ui.ui_factory.nested_progress_bar()
2080
version_ids = list(version_ids)
2081
if None in version_ids:
2082
version_ids.remove(None)
2084
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2085
this_versions = set(self.target._index.get_versions())
2086
needed_versions = self.source_ancestry - this_versions
2087
cross_check_versions = self.source_ancestry.intersection(this_versions)
2088
mismatched_versions = set()
2089
for version in cross_check_versions:
2090
# scan to include needed parents.
2091
n1 = set(self.target.get_parents_with_ghosts(version))
2092
n2 = set(self.source.get_parents_with_ghosts(version))
2094
# FIXME TEST this check for cycles being introduced works
2095
# the logic is we have a cycle if in our graph we are an
2096
# ancestor of any of the n2 revisions.
2102
parent_ancestors = self.source.get_ancestry(parent)
2103
if version in parent_ancestors:
2104
raise errors.GraphCycleError([parent, version])
2105
# ensure this parent will be available later.
2106
new_parents = n2.difference(n1)
2107
needed_versions.update(new_parents.difference(this_versions))
2108
mismatched_versions.add(version)
2110
if not needed_versions and not mismatched_versions:
2112
full_list = topo_sort(self.source.get_graph())
2114
version_list = [i for i in full_list if (not self.target.has_version(i)
2115
and i in needed_versions)]
2119
copy_queue_records = []
2121
for version_id in version_list:
2122
options = self.source._index.get_options(version_id)
2123
parents = self.source._index.get_parents_with_ghosts(version_id)
2124
# check that its will be a consistent copy:
2125
for parent in parents:
2126
# if source has the parent, we must :
2127
# * already have it or
2128
# * have it scheduled already
2129
# otherwise we don't care
2130
assert (self.target.has_version(parent) or
2131
parent in copy_set or
2132
not self.source.has_version(parent))
2133
index_memo = self.source._index.get_position(version_id)
2134
copy_queue_records.append((version_id, index_memo))
2135
copy_queue.append((version_id, options, parents))
2136
copy_set.add(version_id)
2138
# data suck the join:
2140
total = len(version_list)
2143
for (version_id, raw_data), \
2144
(version_id2, options, parents) in \
2145
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2147
assert version_id == version_id2, 'logic error, inconsistent results'
2149
pb.update("Joining knit", count, total)
2150
raw_records.append((version_id, options, parents, len(raw_data)))
2151
raw_datum.append(raw_data)
2152
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2154
for version in mismatched_versions:
2155
# FIXME RBC 20060309 is this needed?
2156
n1 = set(self.target.get_parents_with_ghosts(version))
2157
n2 = set(self.source.get_parents_with_ghosts(version))
2158
# write a combined record to our history preserving the current
2159
# parents as first in the list
2160
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2161
self.target.fix_parents(version, new_parents)
2167
InterVersionedFile.register_optimiser(InterKnit)
2170
class WeaveToKnit(InterVersionedFile):
2171
"""Optimised code paths for weave to knit operations."""
2173
_matching_file_from_factory = bzrlib.weave.WeaveFile
2174
_matching_file_to_factory = KnitVersionedFile
2177
def is_compatible(source, target):
2178
"""Be compatible with weaves to knits."""
2180
return (isinstance(source, bzrlib.weave.Weave) and
2181
isinstance(target, KnitVersionedFile))
2182
except AttributeError:
2185
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2186
"""See InterVersionedFile.join."""
2187
assert isinstance(self.source, bzrlib.weave.Weave)
2188
assert isinstance(self.target, KnitVersionedFile)
2190
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2195
pb = ui.ui_factory.nested_progress_bar()
2197
version_ids = list(version_ids)
2199
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2200
this_versions = set(self.target._index.get_versions())
2201
needed_versions = self.source_ancestry - this_versions
2202
cross_check_versions = self.source_ancestry.intersection(this_versions)
2203
mismatched_versions = set()
2204
for version in cross_check_versions:
2205
# scan to include needed parents.
2206
n1 = set(self.target.get_parents_with_ghosts(version))
2207
n2 = set(self.source.get_parents(version))
2208
# if all of n2's parents are in n1, then its fine.
2209
if n2.difference(n1):
2210
# FIXME TEST this check for cycles being introduced works
2211
# the logic is we have a cycle if in our graph we are an
2212
# ancestor of any of the n2 revisions.
2218
parent_ancestors = self.source.get_ancestry(parent)
2219
if version in parent_ancestors:
2220
raise errors.GraphCycleError([parent, version])
2221
# ensure this parent will be available later.
2222
new_parents = n2.difference(n1)
2223
needed_versions.update(new_parents.difference(this_versions))
2224
mismatched_versions.add(version)
2226
if not needed_versions and not mismatched_versions:
2228
full_list = topo_sort(self.source.get_graph())
2230
version_list = [i for i in full_list if (not self.target.has_version(i)
2231
and i in needed_versions)]
2235
total = len(version_list)
2236
for version_id in version_list:
2237
pb.update("Converting to knit", count, total)
2238
parents = self.source.get_parents(version_id)
2239
# check that its will be a consistent copy:
2240
for parent in parents:
2241
# if source has the parent, we must already have it
2242
assert (self.target.has_version(parent))
2243
self.target.add_lines(
2244
version_id, parents, self.source.get_lines(version_id))
2247
for version in mismatched_versions:
2248
# FIXME RBC 20060309 is this needed?
2249
n1 = set(self.target.get_parents_with_ghosts(version))
2250
n2 = set(self.source.get_parents(version))
2251
# write a combined record to our history preserving the current
2252
# parents as first in the list
2253
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2254
self.target.fix_parents(version, new_parents)
2260
InterVersionedFile.register_optimiser(WeaveToKnit)
2263
class KnitSequenceMatcher(difflib.SequenceMatcher):
2264
"""Knit tuned sequence matcher.
2266
This is based on profiling of difflib which indicated some improvements
2267
for our usage pattern.
2270
def find_longest_match(self, alo, ahi, blo, bhi):
2271
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
2273
If isjunk is not defined:
2275
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
2276
alo <= i <= i+k <= ahi
2277
blo <= j <= j+k <= bhi
2278
and for all (i',j',k') meeting those conditions,
2281
and if i == i', j <= j'
2283
In other words, of all maximal matching blocks, return one that
2284
starts earliest in a, and of all those maximal matching blocks that
2285
start earliest in a, return the one that starts earliest in b.
2287
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
2288
>>> s.find_longest_match(0, 5, 0, 9)
2291
If isjunk is defined, first the longest matching block is
2292
determined as above, but with the additional restriction that no
2293
junk element appears in the block. Then that block is extended as
2294
far as possible by matching (only) junk elements on both sides. So
2295
the resulting block never matches on junk except as identical junk
2296
happens to be adjacent to an "interesting" match.
2298
Here's the same example as before, but considering blanks to be
2299
junk. That prevents " abcd" from matching the " abcd" at the tail
2300
end of the second sequence directly. Instead only the "abcd" can
2301
match, and matches the leftmost "abcd" in the second sequence:
2303
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
2304
>>> s.find_longest_match(0, 5, 0, 9)
2307
If no blocks match, return (alo, blo, 0).
2309
>>> s = SequenceMatcher(None, "ab", "c")
2310
>>> s.find_longest_match(0, 2, 0, 1)
2314
# CAUTION: stripping common prefix or suffix would be incorrect.
2318
# Longest matching block is "ab", but if common prefix is
2319
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
2320
# strip, so ends up claiming that ab is changed to acab by
2321
# inserting "ca" in the middle. That's minimal but unintuitive:
2322
# "it's obvious" that someone inserted "ac" at the front.
2323
# Windiff ends up at the same place as diff, but by pairing up
2324
# the unique 'b's and then matching the first two 'a's.
2326
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
2327
besti, bestj, bestsize = alo, blo, 0
2328
# find longest junk-free match
2329
# during an iteration of the loop, j2len[j] = length of longest
2330
# junk-free match ending with a[i-1] and b[j]
2334
for i in xrange(alo, ahi):
2335
# look at all instances of a[i] in b; note that because
2336
# b2j has no junk keys, the loop is skipped if a[i] is junk
2337
j2lenget = j2len.get
2340
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
2341
# following improvement
2342
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
2343
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
2344
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
2346
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
2347
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
2348
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
2360
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
2362
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
2365
# Extend the best by non-junk elements on each end. In particular,
2366
# "popular" non-junk elements aren't in b2j, which greatly speeds
2367
# the inner loop above, but also means "the best" match so far
2368
# doesn't contain any junk *or* popular non-junk elements.
2369
while besti > alo and bestj > blo and \
2370
not isbjunk(b[bestj-1]) and \
2371
a[besti-1] == b[bestj-1]:
2372
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2373
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2374
not isbjunk(b[bestj+bestsize]) and \
2375
a[besti+bestsize] == b[bestj+bestsize]:
2378
# Now that we have a wholly interesting match (albeit possibly
2379
# empty!), we may as well suck up the matching junk on each
2380
# side of it too. Can't think of a good reason not to, and it
2381
# saves post-processing the (possibly considerable) expense of
2382
# figuring out what to do with it. In the case of an empty
2383
# interesting match, this is clearly the right thing to do,
2384
# because no other kind of match is possible in the regions.
2385
while besti > alo and bestj > blo and \
2386
isbjunk(b[bestj-1]) and \
2387
a[besti-1] == b[bestj-1]:
2388
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2389
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2390
isbjunk(b[bestj+bestsize]) and \
2391
a[besti+bestsize] == b[bestj+bestsize]:
2392
bestsize = bestsize + 1
2394
return besti, bestj, bestsize
2398
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2400
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
