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 ?
63
from cStringIO import StringIO
64
from itertools import izip, chain
68
from bzrlib.lazy_import import lazy_import
69
lazy_import(globals(), """
89
from bzrlib.errors import (
97
RevisionAlreadyPresent,
99
from bzrlib.osutils import (
106
from bzrlib.versionedfile import (
107
AbsentContentFactory,
111
FulltextContentFactory,
117
# TODO: Split out code specific to this format into an associated object.
119
# TODO: Can we put in some kind of value to check that the index and data
120
# files belong together?
122
# TODO: accommodate binaries, perhaps by storing a byte count
124
# TODO: function to check whole file
126
# TODO: atomically append data, then measure backwards from the cursor
127
# position after writing to work out where it was located. we may need to
128
# bypass python file buffering.
130
DATA_SUFFIX = '.knit'
131
INDEX_SUFFIX = '.kndx'
134
class KnitAdapter(object):
135
"""Base class for knit record adaption."""
137
def __init__(self, basis_vf):
138
"""Create an adapter which accesses full texts from basis_vf.
140
:param basis_vf: A versioned file to access basis texts of deltas from.
141
May be None for adapters that do not need to access basis texts.
143
self._data = KnitVersionedFiles(None, None)
144
self._annotate_factory = KnitAnnotateFactory()
145
self._plain_factory = KnitPlainFactory()
146
self._basis_vf = basis_vf
149
class FTAnnotatedToUnannotated(KnitAdapter):
150
"""An adapter from FT annotated knits to unannotated ones."""
152
def get_bytes(self, factory, annotated_compressed_bytes):
154
self._data._parse_record_unchecked(annotated_compressed_bytes)
155
content = self._annotate_factory.parse_fulltext(contents, rec[1])
156
size, bytes = self._data._record_to_data((rec[1],), rec[3], content.text())
160
class DeltaAnnotatedToUnannotated(KnitAdapter):
161
"""An adapter for deltas from annotated to unannotated."""
163
def get_bytes(self, factory, annotated_compressed_bytes):
165
self._data._parse_record_unchecked(annotated_compressed_bytes)
166
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
168
contents = self._plain_factory.lower_line_delta(delta)
169
size, bytes = self._data._record_to_data((rec[1],), rec[3], contents)
173
class FTAnnotatedToFullText(KnitAdapter):
174
"""An adapter from FT annotated knits to unannotated ones."""
176
def get_bytes(self, factory, annotated_compressed_bytes):
178
self._data._parse_record_unchecked(annotated_compressed_bytes)
179
content, delta = self._annotate_factory.parse_record(factory.key[-1],
180
contents, factory._build_details, None)
181
return ''.join(content.text())
184
class DeltaAnnotatedToFullText(KnitAdapter):
185
"""An adapter for deltas from annotated to unannotated."""
187
def get_bytes(self, factory, annotated_compressed_bytes):
189
self._data._parse_record_unchecked(annotated_compressed_bytes)
190
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
192
compression_parent = factory.parents[0]
193
basis_entry = self._basis_vf.get_record_stream(
194
[compression_parent], 'unordered', True).next()
195
if basis_entry.storage_kind == 'absent':
196
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
197
basis_lines = split_lines(basis_entry.get_bytes_as('fulltext'))
198
# Manually apply the delta because we have one annotated content and
200
basis_content = PlainKnitContent(basis_lines, compression_parent)
201
basis_content.apply_delta(delta, rec[1])
202
basis_content._should_strip_eol = factory._build_details[1]
203
return ''.join(basis_content.text())
206
class FTPlainToFullText(KnitAdapter):
207
"""An adapter from FT plain knits to unannotated ones."""
209
def get_bytes(self, factory, compressed_bytes):
211
self._data._parse_record_unchecked(compressed_bytes)
212
content, delta = self._plain_factory.parse_record(factory.key[-1],
213
contents, factory._build_details, None)
214
return ''.join(content.text())
217
class DeltaPlainToFullText(KnitAdapter):
218
"""An adapter for deltas from annotated to unannotated."""
220
def get_bytes(self, factory, compressed_bytes):
222
self._data._parse_record_unchecked(compressed_bytes)
223
delta = self._plain_factory.parse_line_delta(contents, rec[1])
224
compression_parent = factory.parents[0]
225
# XXX: string splitting overhead.
226
basis_entry = self._basis_vf.get_record_stream(
227
[compression_parent], 'unordered', True).next()
228
if basis_entry.storage_kind == 'absent':
229
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
230
basis_lines = split_lines(basis_entry.get_bytes_as('fulltext'))
231
basis_content = PlainKnitContent(basis_lines, compression_parent)
232
# Manually apply the delta because we have one annotated content and
234
content, _ = self._plain_factory.parse_record(rec[1], contents,
235
factory._build_details, basis_content)
236
return ''.join(content.text())
239
class KnitContentFactory(ContentFactory):
240
"""Content factory for streaming from knits.
242
:seealso ContentFactory:
245
def __init__(self, key, parents, build_details, sha1, raw_record,
246
annotated, knit=None):
247
"""Create a KnitContentFactory for key.
250
:param parents: The parents.
251
:param build_details: The build details as returned from
253
:param sha1: The sha1 expected from the full text of this object.
254
:param raw_record: The bytes of the knit data from disk.
255
:param annotated: True if the raw data is annotated.
257
ContentFactory.__init__(self)
260
self.parents = parents
261
if build_details[0] == 'line-delta':
266
annotated_kind = 'annotated-'
269
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
270
self._raw_record = raw_record
271
self._build_details = build_details
274
def get_bytes_as(self, storage_kind):
275
if storage_kind == self.storage_kind:
276
return self._raw_record
277
if storage_kind == 'fulltext' and self._knit is not None:
278
return self._knit.get_text(self.key[0])
280
raise errors.UnavailableRepresentation(self.key, storage_kind,
284
class KnitContent(object):
285
"""Content of a knit version to which deltas can be applied.
287
This is always stored in memory as a list of lines with \n at the end,
288
plus a flag saying if the final ending is really there or not, because that
289
corresponds to the on-disk knit representation.
293
self._should_strip_eol = False
295
def apply_delta(self, delta, new_version_id):
296
"""Apply delta to this object to become new_version_id."""
297
raise NotImplementedError(self.apply_delta)
299
def line_delta_iter(self, new_lines):
300
"""Generate line-based delta from this content to new_lines."""
301
new_texts = new_lines.text()
302
old_texts = self.text()
303
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
304
for tag, i1, i2, j1, j2 in s.get_opcodes():
307
# ofrom, oto, length, data
308
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
310
def line_delta(self, new_lines):
311
return list(self.line_delta_iter(new_lines))
314
def get_line_delta_blocks(knit_delta, source, target):
315
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
316
target_len = len(target)
319
for s_begin, s_end, t_len, new_text in knit_delta:
320
true_n = s_begin - s_pos
323
# knit deltas do not provide reliable info about whether the
324
# last line of a file matches, due to eol handling.
325
if source[s_pos + n -1] != target[t_pos + n -1]:
328
yield s_pos, t_pos, n
329
t_pos += t_len + true_n
331
n = target_len - t_pos
333
if source[s_pos + n -1] != target[t_pos + n -1]:
336
yield s_pos, t_pos, n
337
yield s_pos + (target_len - t_pos), target_len, 0
340
class AnnotatedKnitContent(KnitContent):
341
"""Annotated content."""
343
def __init__(self, lines):
344
KnitContent.__init__(self)
348
"""Return a list of (origin, text) for each content line."""
349
lines = self._lines[:]
350
if self._should_strip_eol:
351
origin, last_line = lines[-1]
352
lines[-1] = (origin, last_line.rstrip('\n'))
355
def apply_delta(self, delta, new_version_id):
356
"""Apply delta to this object to become new_version_id."""
359
for start, end, count, delta_lines in delta:
360
lines[offset+start:offset+end] = delta_lines
361
offset = offset + (start - end) + count
365
lines = [text for origin, text in self._lines]
366
except ValueError, e:
367
# most commonly (only?) caused by the internal form of the knit
368
# missing annotation information because of a bug - see thread
370
raise KnitCorrupt(self,
371
"line in annotated knit missing annotation information: %s"
373
if self._should_strip_eol:
374
lines[-1] = lines[-1].rstrip('\n')
378
return AnnotatedKnitContent(self._lines[:])
381
class PlainKnitContent(KnitContent):
382
"""Unannotated content.
384
When annotate[_iter] is called on this content, the same version is reported
385
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
389
def __init__(self, lines, version_id):
390
KnitContent.__init__(self)
392
self._version_id = version_id
395
"""Return a list of (origin, text) for each content line."""
396
return [(self._version_id, line) for line in self._lines]
398
def apply_delta(self, delta, new_version_id):
399
"""Apply delta to this object to become new_version_id."""
402
for start, end, count, delta_lines in delta:
403
lines[offset+start:offset+end] = delta_lines
404
offset = offset + (start - end) + count
405
self._version_id = new_version_id
408
return PlainKnitContent(self._lines[:], self._version_id)
412
if self._should_strip_eol:
414
lines[-1] = lines[-1].rstrip('\n')
418
class _KnitFactory(object):
419
"""Base class for common Factory functions."""
421
def parse_record(self, version_id, record, record_details,
422
base_content, copy_base_content=True):
423
"""Parse a record into a full content object.
425
:param version_id: The official version id for this content
426
:param record: The data returned by read_records_iter()
427
:param record_details: Details about the record returned by
429
:param base_content: If get_build_details returns a compression_parent,
430
you must return a base_content here, else use None
431
:param copy_base_content: When building from the base_content, decide
432
you can either copy it and return a new object, or modify it in
434
:return: (content, delta) A Content object and possibly a line-delta,
437
method, noeol = record_details
438
if method == 'line-delta':
439
if copy_base_content:
440
content = base_content.copy()
442
content = base_content
443
delta = self.parse_line_delta(record, version_id)
444
content.apply_delta(delta, version_id)
446
content = self.parse_fulltext(record, version_id)
448
content._should_strip_eol = noeol
449
return (content, delta)
452
class KnitAnnotateFactory(_KnitFactory):
453
"""Factory for creating annotated Content objects."""
457
def make(self, lines, version_id):
458
num_lines = len(lines)
459
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
461
def parse_fulltext(self, content, version_id):
462
"""Convert fulltext to internal representation
464
fulltext content is of the format
465
revid(utf8) plaintext\n
466
internal representation is of the format:
469
# TODO: jam 20070209 The tests expect this to be returned as tuples,
470
# but the code itself doesn't really depend on that.
471
# Figure out a way to not require the overhead of turning the
472
# list back into tuples.
473
lines = [tuple(line.split(' ', 1)) for line in content]
474
return AnnotatedKnitContent(lines)
476
def parse_line_delta_iter(self, lines):
477
return iter(self.parse_line_delta(lines))
479
def parse_line_delta(self, lines, version_id, plain=False):
480
"""Convert a line based delta into internal representation.
482
line delta is in the form of:
483
intstart intend intcount
485
revid(utf8) newline\n
486
internal representation is
487
(start, end, count, [1..count tuples (revid, newline)])
489
:param plain: If True, the lines are returned as a plain
490
list without annotations, not as a list of (origin, content) tuples, i.e.
491
(start, end, count, [1..count newline])
498
def cache_and_return(line):
499
origin, text = line.split(' ', 1)
500
return cache.setdefault(origin, origin), text
502
# walk through the lines parsing.
503
# Note that the plain test is explicitly pulled out of the
504
# loop to minimise any performance impact
507
start, end, count = [int(n) for n in header.split(',')]
508
contents = [next().split(' ', 1)[1] for i in xrange(count)]
509
result.append((start, end, count, contents))
512
start, end, count = [int(n) for n in header.split(',')]
513
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
514
result.append((start, end, count, contents))
517
def get_fulltext_content(self, lines):
518
"""Extract just the content lines from a fulltext."""
519
return (line.split(' ', 1)[1] for line in lines)
521
def get_linedelta_content(self, lines):
522
"""Extract just the content from a line delta.
524
This doesn't return all of the extra information stored in a delta.
525
Only the actual content lines.
530
header = header.split(',')
531
count = int(header[2])
532
for i in xrange(count):
533
origin, text = next().split(' ', 1)
536
def lower_fulltext(self, content):
537
"""convert a fulltext content record into a serializable form.
539
see parse_fulltext which this inverts.
541
# TODO: jam 20070209 We only do the caching thing to make sure that
542
# the origin is a valid utf-8 line, eventually we could remove it
543
return ['%s %s' % (o, t) for o, t in content._lines]
545
def lower_line_delta(self, delta):
546
"""convert a delta into a serializable form.
548
See parse_line_delta which this inverts.
550
# TODO: jam 20070209 We only do the caching thing to make sure that
551
# the origin is a valid utf-8 line, eventually we could remove it
553
for start, end, c, lines in delta:
554
out.append('%d,%d,%d\n' % (start, end, c))
555
out.extend(origin + ' ' + text
556
for origin, text in lines)
559
def annotate(self, knit, key):
560
content = knit._get_content(key)
561
# adjust for the fact that serialised annotations are only key suffixes
563
if type(key) == tuple:
565
origins = content.annotate()
567
for origin, line in origins:
568
result.append((prefix + (origin,), line))
571
# XXX: This smells a bit. Why would key ever be a non-tuple here?
572
# Aren't keys defined to be tuples? -- spiv 20080618
573
return content.annotate()
576
class KnitPlainFactory(_KnitFactory):
577
"""Factory for creating plain Content objects."""
581
def make(self, lines, version_id):
582
return PlainKnitContent(lines, version_id)
584
def parse_fulltext(self, content, version_id):
585
"""This parses an unannotated fulltext.
587
Note that this is not a noop - the internal representation
588
has (versionid, line) - its just a constant versionid.
590
return self.make(content, version_id)
592
def parse_line_delta_iter(self, lines, version_id):
594
num_lines = len(lines)
595
while cur < num_lines:
598
start, end, c = [int(n) for n in header.split(',')]
599
yield start, end, c, lines[cur:cur+c]
602
def parse_line_delta(self, lines, version_id):
603
return list(self.parse_line_delta_iter(lines, version_id))
605
def get_fulltext_content(self, lines):
606
"""Extract just the content lines from a fulltext."""
609
def get_linedelta_content(self, lines):
610
"""Extract just the content from a line delta.
612
This doesn't return all of the extra information stored in a delta.
613
Only the actual content lines.
618
header = header.split(',')
619
count = int(header[2])
620
for i in xrange(count):
623
def lower_fulltext(self, content):
624
return content.text()
626
def lower_line_delta(self, delta):
628
for start, end, c, lines in delta:
629
out.append('%d,%d,%d\n' % (start, end, c))
633
def annotate(self, knit, key):
634
annotator = _KnitAnnotator(knit)
635
return annotator.annotate(key)
639
def make_file_factory(annotated, mapper):
640
"""Create a factory for creating a file based KnitVersionedFiles.
642
This is only functional enough to run interface tests, it doesn't try to
643
provide a full pack environment.
645
:param annotated: knit annotations are wanted.
646
:param mapper: The mapper from keys to paths.
648
def factory(transport):
649
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
650
access = _KnitKeyAccess(transport, mapper)
651
return KnitVersionedFiles(index, access, annotated=annotated)
655
def make_pack_factory(graph, delta, keylength):
656
"""Create a factory for creating a pack based VersionedFiles.
658
This is only functional enough to run interface tests, it doesn't try to
659
provide a full pack environment.
661
:param graph: Store a graph.
662
:param delta: Delta compress contents.
663
:param keylength: How long should keys be.
665
def factory(transport):
666
parents = graph or delta
672
max_delta_chain = 200
675
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
676
key_elements=keylength)
677
stream = transport.open_write_stream('newpack')
678
writer = pack.ContainerWriter(stream.write)
680
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
681
deltas=delta, add_callback=graph_index.add_nodes)
682
access = _DirectPackAccess({})
683
access.set_writer(writer, graph_index, (transport, 'newpack'))
684
result = KnitVersionedFiles(index, access,
685
max_delta_chain=max_delta_chain)
686
result.stream = stream
687
result.writer = writer
692
def cleanup_pack_knit(versioned_files):
693
versioned_files.stream.close()
694
versioned_files.writer.end()
697
class KnitVersionedFiles(VersionedFiles):
698
"""Storage for many versioned files using knit compression.
700
Backend storage is managed by indices and data objects.
703
def __init__(self, index, data_access, max_delta_chain=200,
705
"""Create a KnitVersionedFiles with index and data_access.
707
:param index: The index for the knit data.
708
:param data_access: The access object to store and retrieve knit
710
:param max_delta_chain: The maximum number of deltas to permit during
711
insertion. Set to 0 to prohibit the use of deltas.
712
:param annotated: Set to True to cause annotations to be calculated and
713
stored during insertion.
716
self._access = data_access
717
self._max_delta_chain = max_delta_chain
719
self._factory = KnitAnnotateFactory()
721
self._factory = KnitPlainFactory()
722
self._fallback_vfs = []
724
def add_fallback_versioned_files(self, a_versioned_files):
725
"""Add a source of texts for texts not present in this knit.
727
:param a_versioned_files: A VersionedFiles object.
729
self._fallback_vfs.append(a_versioned_files)
731
def add_lines(self, key, parents, lines, parent_texts=None,
732
left_matching_blocks=None, nostore_sha=None, random_id=False,
734
"""See VersionedFiles.add_lines()."""
735
self._index._check_write_ok()
736
self._check_add(key, lines, random_id, check_content)
738
# The caller might pass None if there is no graph data, but kndx
739
# indexes can't directly store that, so we give them
740
# an empty tuple instead.
742
return self._add(key, lines, parents,
743
parent_texts, left_matching_blocks, nostore_sha, random_id)
745
def _add(self, key, lines, parents, parent_texts,
746
left_matching_blocks, nostore_sha, random_id):
747
"""Add a set of lines on top of version specified by parents.
749
Any versions not present will be converted into ghosts.
751
# first thing, if the content is something we don't need to store, find
753
line_bytes = ''.join(lines)
754
digest = sha_string(line_bytes)
755
if nostore_sha == digest:
756
raise errors.ExistingContent
759
if parent_texts is None:
761
# Do a single query to ascertain parent presence.
762
present_parent_map = self.get_parent_map(parents)
763
for parent in parents:
764
if parent in present_parent_map:
765
present_parents.append(parent)
767
# Currently we can only compress against the left most present parent.
768
if (len(present_parents) == 0 or
769
present_parents[0] != parents[0]):
772
# To speed the extract of texts the delta chain is limited
773
# to a fixed number of deltas. This should minimize both
774
# I/O and the time spend applying deltas.
775
delta = self._check_should_delta(present_parents[0])
777
text_length = len(line_bytes)
780
if lines[-1][-1] != '\n':
781
# copy the contents of lines.
783
options.append('no-eol')
784
lines[-1] = lines[-1] + '\n'
788
if type(element) != str:
789
raise TypeError("key contains non-strings: %r" % (key,))
790
# Knit hunks are still last-element only
792
content = self._factory.make(lines, version_id)
793
if 'no-eol' in options:
794
# Hint to the content object that its text() call should strip the
796
content._should_strip_eol = True
797
if delta or (self._factory.annotated and len(present_parents) > 0):
798
# Merge annotations from parent texts if needed.
799
delta_hunks = self._merge_annotations(content, present_parents,
800
parent_texts, delta, self._factory.annotated,
801
left_matching_blocks)
804
options.append('line-delta')
805
store_lines = self._factory.lower_line_delta(delta_hunks)
806
size, bytes = self._record_to_data(key, digest,
809
options.append('fulltext')
810
# isinstance is slower and we have no hierarchy.
811
if self._factory.__class__ == KnitPlainFactory:
812
# Use the already joined bytes saving iteration time in
814
size, bytes = self._record_to_data(key, digest,
817
# get mixed annotation + content and feed it into the
819
store_lines = self._factory.lower_fulltext(content)
820
size, bytes = self._record_to_data(key, digest,
823
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
824
self._index.add_records(
825
((key, options, access_memo, parents),),
827
return digest, text_length, content
829
def annotate(self, key):
830
"""See VersionedFiles.annotate."""
831
return self._factory.annotate(self, key)
833
def check(self, progress_bar=None):
834
"""See VersionedFiles.check()."""
835
# This doesn't actually test extraction of everything, but that will
836
# impact 'bzr check' substantially, and needs to be integrated with
837
# care. However, it does check for the obvious problem of a delta with
839
keys = self._index.keys()
840
parent_map = self.get_parent_map(keys)
842
if self._index.get_method(key) != 'fulltext':
843
compression_parent = parent_map[key][0]
844
if compression_parent not in parent_map:
845
raise errors.KnitCorrupt(self,
846
"Missing basis parent %s for %s" % (
847
compression_parent, key))
848
for fallback_vfs in self._fallback_vfs:
851
def _check_add(self, key, lines, random_id, check_content):
852
"""check that version_id and lines are safe to add."""
854
if contains_whitespace(version_id):
855
raise InvalidRevisionId(version_id, self)
856
self.check_not_reserved_id(version_id)
857
# TODO: If random_id==False and the key is already present, we should
858
# probably check that the existing content is identical to what is
859
# being inserted, and otherwise raise an exception. This would make
860
# the bundle code simpler.
862
self._check_lines_not_unicode(lines)
863
self._check_lines_are_lines(lines)
865
def _check_header(self, key, line):
866
rec = self._split_header(line)
867
self._check_header_version(rec, key[-1])
870
def _check_header_version(self, rec, version_id):
871
"""Checks the header version on original format knit records.
873
These have the last component of the key embedded in the record.
875
if rec[1] != version_id:
876
raise KnitCorrupt(self,
877
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
879
def _check_should_delta(self, parent):
880
"""Iterate back through the parent listing, looking for a fulltext.
882
This is used when we want to decide whether to add a delta or a new
883
fulltext. It searches for _max_delta_chain parents. When it finds a
884
fulltext parent, it sees if the total size of the deltas leading up to
885
it is large enough to indicate that we want a new full text anyway.
887
Return True if we should create a new delta, False if we should use a
892
for count in xrange(self._max_delta_chain):
893
# XXX: Collapse these two queries:
895
method = self._index.get_method(parent)
896
except RevisionNotPresent:
897
# Some basis is not locally present: always delta
899
index, pos, size = self._index.get_position(parent)
900
if method == 'fulltext':
904
# We don't explicitly check for presence because this is in an
905
# inner loop, and if it's missing it'll fail anyhow.
906
# TODO: This should be asking for compression parent, not graph
908
parent = self._index.get_parent_map([parent])[parent][0]
910
# We couldn't find a fulltext, so we must create a new one
912
# Simple heuristic - if the total I/O wold be greater as a delta than
913
# the originally installed fulltext, we create a new fulltext.
914
return fulltext_size > delta_size
916
def _build_details_to_components(self, build_details):
917
"""Convert a build_details tuple to a position tuple."""
918
# record_details, access_memo, compression_parent
919
return build_details[3], build_details[0], build_details[1]
921
def _get_components_positions(self, keys, allow_missing=False):
922
"""Produce a map of position data for the components of keys.
924
This data is intended to be used for retrieving the knit records.
926
A dict of key to (record_details, index_memo, next, parents) is
928
method is the way referenced data should be applied.
929
index_memo is the handle to pass to the data access to actually get the
931
next is the build-parent of the version, or None for fulltexts.
932
parents is the version_ids of the parents of this version
934
:param allow_missing: If True do not raise an error on a missing component,
938
pending_components = keys
939
while pending_components:
940
build_details = self._index.get_build_details(pending_components)
941
current_components = set(pending_components)
942
pending_components = set()
943
for key, details in build_details.iteritems():
944
(index_memo, compression_parent, parents,
945
record_details) = details
946
method = record_details[0]
947
if compression_parent is not None:
948
pending_components.add(compression_parent)
949
component_data[key] = self._build_details_to_components(details)
950
missing = current_components.difference(build_details)
951
if missing and not allow_missing:
952
raise errors.RevisionNotPresent(missing.pop(), self)
953
return component_data
955
def _get_content(self, key, parent_texts={}):
956
"""Returns a content object that makes up the specified
958
cached_version = parent_texts.get(key, None)
959
if cached_version is not None:
960
# Ensure the cache dict is valid.
961
if not self.get_parent_map([key]):
962
raise RevisionNotPresent(key, self)
963
return cached_version
964
text_map, contents_map = self._get_content_maps([key])
965
return contents_map[key]
967
def _get_content_maps(self, keys, nonlocal_keys=None):
968
"""Produce maps of text and KnitContents
970
:param keys: The keys to produce content maps for.
971
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
972
which are known to not be in this knit, but rather in one of the
974
:return: (text_map, content_map) where text_map contains the texts for
975
the requested versions and content_map contains the KnitContents.
977
# FUTURE: This function could be improved for the 'extract many' case
978
# by tracking each component and only doing the copy when the number of
979
# children than need to apply delta's to it is > 1 or it is part of the
982
multiple_versions = len(keys) != 1
983
record_map = self._get_record_map(keys, allow_missing=True)
988
if nonlocal_keys is None:
989
nonlocal_keys = set()
991
nonlocal_keys = frozenset(nonlocal_keys)
992
missing_keys = set(nonlocal_keys)
993
for source in self._fallback_vfs:
996
for record in source.get_record_stream(missing_keys,
998
if record.storage_kind == 'absent':
1000
missing_keys.remove(record.key)
1001
lines = split_lines(record.get_bytes_as('fulltext'))
1002
text_map[record.key] = lines
1003
content_map[record.key] = PlainKnitContent(lines, record.key)
1004
if record.key in keys:
1005
final_content[record.key] = content_map[record.key]
1007
if key in nonlocal_keys:
1012
while cursor is not None:
1014
record, record_details, digest, next = record_map[cursor]
1016
raise RevisionNotPresent(cursor, self)
1017
components.append((cursor, record, record_details, digest))
1019
if cursor in content_map:
1020
# no need to plan further back
1021
components.append((cursor, None, None, None))
1025
for (component_id, record, record_details,
1026
digest) in reversed(components):
1027
if component_id in content_map:
1028
content = content_map[component_id]
1030
content, delta = self._factory.parse_record(key[-1],
1031
record, record_details, content,
1032
copy_base_content=multiple_versions)
1033
if multiple_versions:
1034
content_map[component_id] = content
1036
final_content[key] = content
1038
# digest here is the digest from the last applied component.
1039
text = content.text()
1040
actual_sha = sha_strings(text)
1041
if actual_sha != digest:
1042
raise KnitCorrupt(self,
1044
'\n of reconstructed text does not match'
1046
'\n for version %s' %
1047
(actual_sha, digest, key))
1048
text_map[key] = text
1049
return text_map, final_content
1051
def get_parent_map(self, keys):
1052
"""Get a map of the graph parents of keys.
1054
:param keys: The keys to look up parents for.
1055
:return: A mapping from keys to parents. Absent keys are absent from
1058
return self._get_parent_map_with_sources(keys)[0]
1060
def _get_parent_map_with_sources(self, keys):
1061
"""Get a map of the parents of keys.
1063
:param keys: The keys to look up parents for.
1064
:return: A tuple. The first element is a mapping from keys to parents.
1065
Absent keys are absent from the mapping. The second element is a
1066
list with the locations each key was found in. The first element
1067
is the in-this-knit parents, the second the first fallback source,
1071
sources = [self._index] + self._fallback_vfs
1074
for source in sources:
1077
new_result = source.get_parent_map(missing)
1078
source_results.append(new_result)
1079
result.update(new_result)
1080
missing.difference_update(set(new_result))
1081
return result, source_results
1083
def _get_record_map(self, keys, allow_missing=False):
1084
"""Produce a dictionary of knit records.
1086
:return: {key:(record, record_details, digest, next)}
1088
data returned from read_records
1090
opaque information to pass to parse_record
1092
SHA1 digest of the full text after all steps are done
1094
build-parent of the version, i.e. the leftmost ancestor.
1095
Will be None if the record is not a delta.
1096
:param keys: The keys to build a map for
1097
:param allow_missing: If some records are missing, rather than
1098
error, just return the data that could be generated.
1100
position_map = self._get_components_positions(keys,
1101
allow_missing=allow_missing)
1102
# key = component_id, r = record_details, i_m = index_memo, n = next
1103
records = [(key, i_m) for key, (r, i_m, n)
1104
in position_map.iteritems()]
1106
for key, record, digest in \
1107
self._read_records_iter(records):
1108
(record_details, index_memo, next) = position_map[key]
1109
record_map[key] = record, record_details, digest, next
1112
def get_record_stream(self, keys, ordering, include_delta_closure):
1113
"""Get a stream of records for keys.
1115
:param keys: The keys to include.
1116
:param ordering: Either 'unordered' or 'topological'. A topologically
1117
sorted stream has compression parents strictly before their
1119
:param include_delta_closure: If True then the closure across any
1120
compression parents will be included (in the opaque data).
1121
:return: An iterator of ContentFactory objects, each of which is only
1122
valid until the iterator is advanced.
1124
# keys might be a generator
1128
if not self._index.has_graph:
1129
# Cannot topological order when no graph has been stored.
1130
ordering = 'unordered'
1131
if include_delta_closure:
1132
positions = self._get_components_positions(keys, allow_missing=True)
1134
build_details = self._index.get_build_details(keys)
1136
# (record_details, access_memo, compression_parent_key)
1137
positions = dict((key, self._build_details_to_components(details))
1138
for key, details in build_details.iteritems())
1139
absent_keys = keys.difference(set(positions))
1140
# There may be more absent keys : if we're missing the basis component
1141
# and are trying to include the delta closure.
1142
if include_delta_closure:
1143
needed_from_fallback = set()
1144
# Build up reconstructable_keys dict. key:True in this dict means
1145
# the key can be reconstructed.
1146
reconstructable_keys = {}
1150
chain = [key, positions[key][2]]
1152
needed_from_fallback.add(key)
1155
while chain[-1] is not None:
1156
if chain[-1] in reconstructable_keys:
1157
result = reconstructable_keys[chain[-1]]
1161
chain.append(positions[chain[-1]][2])
1163
# missing basis component
1164
needed_from_fallback.add(chain[-1])
1167
for chain_key in chain[:-1]:
1168
reconstructable_keys[chain_key] = result
1170
needed_from_fallback.add(key)
1171
# Double index lookups here : need a unified api ?
1172
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1173
if ordering == 'topological':
1174
# Global topological sort
1175
present_keys = tsort.topo_sort(global_map)
1176
# Now group by source:
1178
current_source = None
1179
for key in present_keys:
1180
for parent_map in parent_maps:
1181
if key in parent_map:
1182
key_source = parent_map
1184
if current_source is not key_source:
1185
source_keys.append((key_source, []))
1186
current_source = key_source
1187
source_keys[-1][1].append(key)
1189
# Just group by source; remote sources first.
1192
for parent_map in reversed(parent_maps):
1193
source_keys.append((parent_map, []))
1194
for key in parent_map:
1195
present_keys.append(key)
1196
source_keys[-1][1].append(key)
1197
absent_keys = keys - set(global_map)
1198
for key in absent_keys:
1199
yield AbsentContentFactory(key)
1200
# restrict our view to the keys we can answer.
1201
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1202
# XXX: At that point we need to consider the impact of double reads by
1203
# utilising components multiple times.
1204
if include_delta_closure:
1205
# XXX: get_content_maps performs its own index queries; allow state
1207
text_map, _ = self._get_content_maps(present_keys,
1208
needed_from_fallback - absent_keys)
1209
for key in present_keys:
1210
yield FulltextContentFactory(key, global_map[key], None,
1211
''.join(text_map[key]))
1213
for source, keys in source_keys:
1214
if source is parent_maps[0]:
1215
# this KnitVersionedFiles
1216
records = [(key, positions[key][1]) for key in keys]
1217
for key, raw_data, sha1 in self._read_records_iter_raw(records):
1218
(record_details, index_memo, _) = positions[key]
1219
yield KnitContentFactory(key, global_map[key],
1220
record_details, sha1, raw_data, self._factory.annotated, None)
1222
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1223
for record in vf.get_record_stream(keys, ordering,
1224
include_delta_closure):
1227
def get_sha1s(self, keys):
1228
"""See VersionedFiles.get_sha1s()."""
1230
record_map = self._get_record_map(missing, allow_missing=True)
1232
for key, details in record_map.iteritems():
1233
if key not in missing:
1235
# record entry 2 is the 'digest'.
1236
result[key] = details[2]
1237
missing.difference_update(set(result))
1238
for source in self._fallback_vfs:
1241
new_result = source.get_sha1s(missing)
1242
result.update(new_result)
1243
missing.difference_update(set(new_result))
1246
def insert_record_stream(self, stream):
1247
"""Insert a record stream into this container.
1249
:param stream: A stream of records to insert.
1251
:seealso VersionedFiles.get_record_stream:
1253
def get_adapter(adapter_key):
1255
return adapters[adapter_key]
1257
adapter_factory = adapter_registry.get(adapter_key)
1258
adapter = adapter_factory(self)
1259
adapters[adapter_key] = adapter
1261
if self._factory.annotated:
1262
# self is annotated, we need annotated knits to use directly.
1263
annotated = "annotated-"
1266
# self is not annotated, but we can strip annotations cheaply.
1268
convertibles = set(["knit-annotated-ft-gz"])
1269
if self._max_delta_chain:
1270
convertibles.add("knit-annotated-delta-gz")
1271
# The set of types we can cheaply adapt without needing basis texts.
1272
native_types = set()
1273
if self._max_delta_chain:
1274
native_types.add("knit-%sdelta-gz" % annotated)
1275
native_types.add("knit-%sft-gz" % annotated)
1276
knit_types = native_types.union(convertibles)
1278
# Buffer all index entries that we can't add immediately because their
1279
# basis parent is missing. We don't buffer all because generating
1280
# annotations may require access to some of the new records. However we
1281
# can't generate annotations from new deltas until their basis parent
1282
# is present anyway, so we get away with not needing an index that
1283
# includes the new keys.
1284
# key = basis_parent, value = index entry to add
1285
buffered_index_entries = {}
1286
for record in stream:
1287
parents = record.parents
1288
# Raise an error when a record is missing.
1289
if record.storage_kind == 'absent':
1290
raise RevisionNotPresent([record.key], self)
1291
if record.storage_kind in knit_types:
1292
if record.storage_kind not in native_types:
1294
adapter_key = (record.storage_kind, "knit-delta-gz")
1295
adapter = get_adapter(adapter_key)
1297
adapter_key = (record.storage_kind, "knit-ft-gz")
1298
adapter = get_adapter(adapter_key)
1299
bytes = adapter.get_bytes(
1300
record, record.get_bytes_as(record.storage_kind))
1302
bytes = record.get_bytes_as(record.storage_kind)
1303
options = [record._build_details[0]]
1304
if record._build_details[1]:
1305
options.append('no-eol')
1306
# Just blat it across.
1307
# Note: This does end up adding data on duplicate keys. As
1308
# modern repositories use atomic insertions this should not
1309
# lead to excessive growth in the event of interrupted fetches.
1310
# 'knit' repositories may suffer excessive growth, but as a
1311
# deprecated format this is tolerable. It can be fixed if
1312
# needed by in the kndx index support raising on a duplicate
1313
# add with identical parents and options.
1314
access_memo = self._access.add_raw_records(
1315
[(record.key, len(bytes))], bytes)[0]
1316
index_entry = (record.key, options, access_memo, parents)
1318
if 'fulltext' not in options:
1319
basis_parent = parents[0]
1320
# Note that pack backed knits don't need to buffer here
1321
# because they buffer all writes to the transaction level,
1322
# but we don't expose that difference at the index level. If
1323
# the query here has sufficient cost to show up in
1324
# profiling we should do that.
1325
if basis_parent not in self.get_parent_map([basis_parent]):
1326
pending = buffered_index_entries.setdefault(
1328
pending.append(index_entry)
1331
self._index.add_records([index_entry])
1332
elif record.storage_kind == 'fulltext':
1333
self.add_lines(record.key, parents,
1334
split_lines(record.get_bytes_as('fulltext')))
1336
adapter_key = record.storage_kind, 'fulltext'
1337
adapter = get_adapter(adapter_key)
1338
lines = split_lines(adapter.get_bytes(
1339
record, record.get_bytes_as(record.storage_kind)))
1341
self.add_lines(record.key, parents, lines)
1342
except errors.RevisionAlreadyPresent:
1344
# Add any records whose basis parent is now available.
1345
added_keys = [record.key]
1347
key = added_keys.pop(0)
1348
if key in buffered_index_entries:
1349
index_entries = buffered_index_entries[key]
1350
self._index.add_records(index_entries)
1352
[index_entry[0] for index_entry in index_entries])
1353
del buffered_index_entries[key]
1354
# If there were any deltas which had a missing basis parent, error.
1355
if buffered_index_entries:
1356
raise errors.RevisionNotPresent(buffered_index_entries.keys()[0],
1359
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1360
"""Iterate over the lines in the versioned files from keys.
1362
This may return lines from other keys. Each item the returned
1363
iterator yields is a tuple of a line and a text version that that line
1364
is present in (not introduced in).
1366
Ordering of results is in whatever order is most suitable for the
1367
underlying storage format.
1369
If a progress bar is supplied, it may be used to indicate progress.
1370
The caller is responsible for cleaning up progress bars (because this
1374
* Lines are normalised by the underlying store: they will all have \n
1376
* Lines are returned in arbitrary order.
1378
:return: An iterator over (line, key).
1381
pb = progress.DummyProgress()
1384
# we don't care about inclusions, the caller cares.
1385
# but we need to setup a list of records to visit.
1386
# we need key, position, length
1388
build_details = self._index.get_build_details(keys)
1389
for key, details in build_details.iteritems():
1391
key_records.append((key, details[0]))
1393
records_iter = enumerate(self._read_records_iter(key_records))
1394
for (key_idx, (key, data, sha_value)) in records_iter:
1395
pb.update('Walking content.', key_idx, total)
1396
compression_parent = build_details[key][1]
1397
if compression_parent is None:
1399
line_iterator = self._factory.get_fulltext_content(data)
1402
line_iterator = self._factory.get_linedelta_content(data)
1403
# XXX: It might be more efficient to yield (key,
1404
# line_iterator) in the future. However for now, this is a simpler
1405
# change to integrate into the rest of the codebase. RBC 20071110
1406
for line in line_iterator:
1408
for source in self._fallback_vfs:
1412
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1413
source_keys.add(key)
1415
keys.difference_update(source_keys)
1417
raise RevisionNotPresent(keys, self.filename)
1418
pb.update('Walking content.', total, total)
1420
def _make_line_delta(self, delta_seq, new_content):
1421
"""Generate a line delta from delta_seq and new_content."""
1423
for op in delta_seq.get_opcodes():
1424
if op[0] == 'equal':
1426
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1429
def _merge_annotations(self, content, parents, parent_texts={},
1430
delta=None, annotated=None,
1431
left_matching_blocks=None):
1432
"""Merge annotations for content and generate deltas.
1434
This is done by comparing the annotations based on changes to the text
1435
and generating a delta on the resulting full texts. If annotations are
1436
not being created then a simple delta is created.
1438
if left_matching_blocks is not None:
1439
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1443
for parent_key in parents:
1444
merge_content = self._get_content(parent_key, parent_texts)
1445
if (parent_key == parents[0] and delta_seq is not None):
1448
seq = patiencediff.PatienceSequenceMatcher(
1449
None, merge_content.text(), content.text())
1450
for i, j, n in seq.get_matching_blocks():
1453
# this copies (origin, text) pairs across to the new
1454
# content for any line that matches the last-checked
1456
content._lines[j:j+n] = merge_content._lines[i:i+n]
1457
# XXX: Robert says the following block is a workaround for a
1458
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1459
if content._lines and content._lines[-1][1][-1] != '\n':
1460
# The copied annotation was from a line without a trailing EOL,
1461
# reinstate one for the content object, to ensure correct
1463
line = content._lines[-1][1] + '\n'
1464
content._lines[-1] = (content._lines[-1][0], line)
1466
if delta_seq is None:
1467
reference_content = self._get_content(parents[0], parent_texts)
1468
new_texts = content.text()
1469
old_texts = reference_content.text()
1470
delta_seq = patiencediff.PatienceSequenceMatcher(
1471
None, old_texts, new_texts)
1472
return self._make_line_delta(delta_seq, content)
1474
def _parse_record(self, version_id, data):
1475
"""Parse an original format knit record.
1477
These have the last element of the key only present in the stored data.
1479
rec, record_contents = self._parse_record_unchecked(data)
1480
self._check_header_version(rec, version_id)
1481
return record_contents, rec[3]
1483
def _parse_record_header(self, key, raw_data):
1484
"""Parse a record header for consistency.
1486
:return: the header and the decompressor stream.
1487
as (stream, header_record)
1489
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1492
rec = self._check_header(key, df.readline())
1493
except Exception, e:
1494
raise KnitCorrupt(self,
1495
"While reading {%s} got %s(%s)"
1496
% (key, e.__class__.__name__, str(e)))
1499
def _parse_record_unchecked(self, data):
1501
# 4168 calls in 2880 217 internal
1502
# 4168 calls to _parse_record_header in 2121
1503
# 4168 calls to readlines in 330
1504
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1506
record_contents = df.readlines()
1507
except Exception, e:
1508
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1509
(data, e.__class__.__name__, str(e)))
1510
header = record_contents.pop(0)
1511
rec = self._split_header(header)
1512
last_line = record_contents.pop()
1513
if len(record_contents) != int(rec[2]):
1514
raise KnitCorrupt(self,
1515
'incorrect number of lines %s != %s'
1516
' for version {%s} %s'
1517
% (len(record_contents), int(rec[2]),
1518
rec[1], record_contents))
1519
if last_line != 'end %s\n' % rec[1]:
1520
raise KnitCorrupt(self,
1521
'unexpected version end line %r, wanted %r'
1522
% (last_line, rec[1]))
1524
return rec, record_contents
1526
def _read_records_iter(self, records):
1527
"""Read text records from data file and yield result.
1529
The result will be returned in whatever is the fastest to read.
1530
Not by the order requested. Also, multiple requests for the same
1531
record will only yield 1 response.
1532
:param records: A list of (key, access_memo) entries
1533
:return: Yields (key, contents, digest) in the order
1534
read, not the order requested
1539
# XXX: This smells wrong, IO may not be getting ordered right.
1540
needed_records = sorted(set(records), key=operator.itemgetter(1))
1541
if not needed_records:
1544
# The transport optimizes the fetching as well
1545
# (ie, reads continuous ranges.)
1546
raw_data = self._access.get_raw_records(
1547
[index_memo for key, index_memo in needed_records])
1549
for (key, index_memo), data in \
1550
izip(iter(needed_records), raw_data):
1551
content, digest = self._parse_record(key[-1], data)
1552
yield key, content, digest
1554
def _read_records_iter_raw(self, records):
1555
"""Read text records from data file and yield raw data.
1557
This unpacks enough of the text record to validate the id is
1558
as expected but thats all.
1560
Each item the iterator yields is (key, bytes, sha1_of_full_text).
1562
# setup an iterator of the external records:
1563
# uses readv so nice and fast we hope.
1565
# grab the disk data needed.
1566
needed_offsets = [index_memo for key, index_memo
1568
raw_records = self._access.get_raw_records(needed_offsets)
1570
for key, index_memo in records:
1571
data = raw_records.next()
1572
# validate the header (note that we can only use the suffix in
1573
# current knit records).
1574
df, rec = self._parse_record_header(key, data)
1576
yield key, data, rec[3]
1578
def _record_to_data(self, key, digest, lines, dense_lines=None):
1579
"""Convert key, digest, lines into a raw data block.
1581
:param key: The key of the record. Currently keys are always serialised
1582
using just the trailing component.
1583
:param dense_lines: The bytes of lines but in a denser form. For
1584
instance, if lines is a list of 1000 bytestrings each ending in \n,
1585
dense_lines may be a list with one line in it, containing all the
1586
1000's lines and their \n's. Using dense_lines if it is already
1587
known is a win because the string join to create bytes in this
1588
function spends less time resizing the final string.
1589
:return: (len, a StringIO instance with the raw data ready to read.)
1591
# Note: using a string copy here increases memory pressure with e.g.
1592
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
1593
# when doing the initial commit of a mozilla tree. RBC 20070921
1594
bytes = ''.join(chain(
1595
["version %s %d %s\n" % (key[-1],
1598
dense_lines or lines,
1599
["end %s\n" % key[-1]]))
1600
if type(bytes) != str:
1601
raise AssertionError(
1602
'data must be plain bytes was %s' % type(bytes))
1603
if lines and lines[-1][-1] != '\n':
1604
raise ValueError('corrupt lines value %r' % lines)
1605
compressed_bytes = tuned_gzip.bytes_to_gzip(bytes)
1606
return len(compressed_bytes), compressed_bytes
1608
def _split_header(self, line):
1611
raise KnitCorrupt(self,
1612
'unexpected number of elements in record header')
1616
"""See VersionedFiles.keys."""
1617
if 'evil' in debug.debug_flags:
1618
trace.mutter_callsite(2, "keys scales with size of history")
1619
sources = [self._index] + self._fallback_vfs
1621
for source in sources:
1622
result.update(source.keys())
1627
class _KndxIndex(object):
1628
"""Manages knit index files
1630
The index is kept in memory and read on startup, to enable
1631
fast lookups of revision information. The cursor of the index
1632
file is always pointing to the end, making it easy to append
1635
_cache is a cache for fast mapping from version id to a Index
1638
_history is a cache for fast mapping from indexes to version ids.
1640
The index data format is dictionary compressed when it comes to
1641
parent references; a index entry may only have parents that with a
1642
lover index number. As a result, the index is topological sorted.
1644
Duplicate entries may be written to the index for a single version id
1645
if this is done then the latter one completely replaces the former:
1646
this allows updates to correct version and parent information.
1647
Note that the two entries may share the delta, and that successive
1648
annotations and references MUST point to the first entry.
1650
The index file on disc contains a header, followed by one line per knit
1651
record. The same revision can be present in an index file more than once.
1652
The first occurrence gets assigned a sequence number starting from 0.
1654
The format of a single line is
1655
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1656
REVISION_ID is a utf8-encoded revision id
1657
FLAGS is a comma separated list of flags about the record. Values include
1658
no-eol, line-delta, fulltext.
1659
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1660
that the the compressed data starts at.
1661
LENGTH is the ascii representation of the length of the data file.
1662
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1664
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1665
revision id already in the knit that is a parent of REVISION_ID.
1666
The ' :' marker is the end of record marker.
1669
when a write is interrupted to the index file, it will result in a line
1670
that does not end in ' :'. If the ' :' is not present at the end of a line,
1671
or at the end of the file, then the record that is missing it will be
1672
ignored by the parser.
1674
When writing new records to the index file, the data is preceded by '\n'
1675
to ensure that records always start on new lines even if the last write was
1676
interrupted. As a result its normal for the last line in the index to be
1677
missing a trailing newline. One can be added with no harmful effects.
1679
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
1680
where prefix is e.g. the (fileid,) for .texts instances or () for
1681
constant-mapped things like .revisions, and the old state is
1682
tuple(history_vector, cache_dict). This is used to prevent having an
1683
ABI change with the C extension that reads .kndx files.
1686
HEADER = "# bzr knit index 8\n"
1688
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
1689
"""Create a _KndxIndex on transport using mapper."""
1690
self._transport = transport
1691
self._mapper = mapper
1692
self._get_scope = get_scope
1693
self._allow_writes = allow_writes
1694
self._is_locked = is_locked
1696
self.has_graph = True
1698
def add_records(self, records, random_id=False):
1699
"""Add multiple records to the index.
1701
:param records: a list of tuples:
1702
(key, options, access_memo, parents).
1703
:param random_id: If True the ids being added were randomly generated
1704
and no check for existence will be performed.
1707
for record in records:
1710
path = self._mapper.map(key) + '.kndx'
1711
path_keys = paths.setdefault(path, (prefix, []))
1712
path_keys[1].append(record)
1713
for path in sorted(paths):
1714
prefix, path_keys = paths[path]
1715
self._load_prefixes([prefix])
1717
orig_history = self._kndx_cache[prefix][1][:]
1718
orig_cache = self._kndx_cache[prefix][0].copy()
1721
for key, options, (_, pos, size), parents in path_keys:
1723
# kndx indices cannot be parentless.
1725
line = "\n%s %s %s %s %s :" % (
1726
key[-1], ','.join(options), pos, size,
1727
self._dictionary_compress(parents))
1728
if type(line) != str:
1729
raise AssertionError(
1730
'data must be utf8 was %s' % type(line))
1732
self._cache_key(key, options, pos, size, parents)
1733
if len(orig_history):
1734
self._transport.append_bytes(path, ''.join(lines))
1736
self._init_index(path, lines)
1738
# If any problems happen, restore the original values and re-raise
1739
self._kndx_cache[prefix] = (orig_cache, orig_history)
1742
def _cache_key(self, key, options, pos, size, parent_keys):
1743
"""Cache a version record in the history array and index cache.
1745
This is inlined into _load_data for performance. KEEP IN SYNC.
1746
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1750
version_id = key[-1]
1751
# last-element only for compatibilty with the C load_data.
1752
parents = tuple(parent[-1] for parent in parent_keys)
1753
for parent in parent_keys:
1754
if parent[:-1] != prefix:
1755
raise ValueError("mismatched prefixes for %r, %r" % (
1757
cache, history = self._kndx_cache[prefix]
1758
# only want the _history index to reference the 1st index entry
1760
if version_id not in cache:
1761
index = len(history)
1762
history.append(version_id)
1764
index = cache[version_id][5]
1765
cache[version_id] = (version_id,
1772
def check_header(self, fp):
1773
line = fp.readline()
1775
# An empty file can actually be treated as though the file doesn't
1777
raise errors.NoSuchFile(self)
1778
if line != self.HEADER:
1779
raise KnitHeaderError(badline=line, filename=self)
1781
def _check_read(self):
1782
if not self._is_locked():
1783
raise errors.ObjectNotLocked(self)
1784
if self._get_scope() != self._scope:
1787
def _check_write_ok(self):
1788
"""Assert if not writes are permitted."""
1789
if not self._is_locked():
1790
raise errors.ObjectNotLocked(self)
1791
if self._get_scope() != self._scope:
1793
if self._mode != 'w':
1794
raise errors.ReadOnlyObjectDirtiedError(self)
1796
def get_build_details(self, keys):
1797
"""Get the method, index_memo and compression parent for keys.
1799
Ghosts are omitted from the result.
1801
:param keys: An iterable of keys.
1802
:return: A dict of key:(index_memo, compression_parent, parents,
1805
opaque structure to pass to read_records to extract the raw
1808
Content that this record is built upon, may be None
1810
Logical parents of this node
1812
extra information about the content which needs to be passed to
1813
Factory.parse_record
1815
prefixes = self._partition_keys(keys)
1816
parent_map = self.get_parent_map(keys)
1819
if key not in parent_map:
1821
method = self.get_method(key)
1822
parents = parent_map[key]
1823
if method == 'fulltext':
1824
compression_parent = None
1826
compression_parent = parents[0]
1827
noeol = 'no-eol' in self.get_options(key)
1828
index_memo = self.get_position(key)
1829
result[key] = (index_memo, compression_parent,
1830
parents, (method, noeol))
1833
def get_method(self, key):
1834
"""Return compression method of specified key."""
1835
options = self.get_options(key)
1836
if 'fulltext' in options:
1838
elif 'line-delta' in options:
1841
raise errors.KnitIndexUnknownMethod(self, options)
1843
def get_options(self, key):
1844
"""Return a list representing options.
1848
prefix, suffix = self._split_key(key)
1849
self._load_prefixes([prefix])
1851
return self._kndx_cache[prefix][0][suffix][1]
1853
raise RevisionNotPresent(key, self)
1855
def get_parent_map(self, keys):
1856
"""Get a map of the parents of keys.
1858
:param keys: The keys to look up parents for.
1859
:return: A mapping from keys to parents. Absent keys are absent from
1862
# Parse what we need to up front, this potentially trades off I/O
1863
# locality (.kndx and .knit in the same block group for the same file
1864
# id) for less checking in inner loops.
1865
prefixes = set(key[:-1] for key in keys)
1866
self._load_prefixes(prefixes)
1871
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
1875
result[key] = tuple(prefix + (suffix,) for
1876
suffix in suffix_parents)
1879
def get_position(self, key):
1880
"""Return details needed to access the version.
1882
:return: a tuple (key, data position, size) to hand to the access
1883
logic to get the record.
1885
prefix, suffix = self._split_key(key)
1886
self._load_prefixes([prefix])
1887
entry = self._kndx_cache[prefix][0][suffix]
1888
return key, entry[2], entry[3]
1890
def _init_index(self, path, extra_lines=[]):
1891
"""Initialize an index."""
1893
sio.write(self.HEADER)
1894
sio.writelines(extra_lines)
1896
self._transport.put_file_non_atomic(path, sio,
1897
create_parent_dir=True)
1898
# self._create_parent_dir)
1899
# mode=self._file_mode,
1900
# dir_mode=self._dir_mode)
1903
"""Get all the keys in the collection.
1905
The keys are not ordered.
1908
# Identify all key prefixes.
1909
# XXX: A bit hacky, needs polish.
1910
if type(self._mapper) == ConstantMapper:
1914
for quoted_relpath in self._transport.iter_files_recursive():
1915
path, ext = os.path.splitext(quoted_relpath)
1917
prefixes = [self._mapper.unmap(path) for path in relpaths]
1918
self._load_prefixes(prefixes)
1919
for prefix in prefixes:
1920
for suffix in self._kndx_cache[prefix][1]:
1921
result.add(prefix + (suffix,))
1924
def _load_prefixes(self, prefixes):
1925
"""Load the indices for prefixes."""
1927
for prefix in prefixes:
1928
if prefix not in self._kndx_cache:
1929
# the load_data interface writes to these variables.
1932
self._filename = prefix
1934
path = self._mapper.map(prefix) + '.kndx'
1935
fp = self._transport.get(path)
1937
# _load_data may raise NoSuchFile if the target knit is
1939
_load_data(self, fp)
1942
self._kndx_cache[prefix] = (self._cache, self._history)
1947
self._kndx_cache[prefix] = ({}, [])
1948
if type(self._mapper) == ConstantMapper:
1949
# preserve behaviour for revisions.kndx etc.
1950
self._init_index(path)
1955
def _partition_keys(self, keys):
1956
"""Turn keys into a dict of prefix:suffix_list."""
1959
prefix_keys = result.setdefault(key[:-1], [])
1960
prefix_keys.append(key[-1])
1963
def _dictionary_compress(self, keys):
1964
"""Dictionary compress keys.
1966
:param keys: The keys to generate references to.
1967
:return: A string representation of keys. keys which are present are
1968
dictionary compressed, and others are emitted as fulltext with a
1974
prefix = keys[0][:-1]
1975
cache = self._kndx_cache[prefix][0]
1977
if key[:-1] != prefix:
1978
# kndx indices cannot refer across partitioned storage.
1979
raise ValueError("mismatched prefixes for %r" % keys)
1980
if key[-1] in cache:
1981
# -- inlined lookup() --
1982
result_list.append(str(cache[key[-1]][5]))
1983
# -- end lookup () --
1985
result_list.append('.' + key[-1])
1986
return ' '.join(result_list)
1988
def _reset_cache(self):
1989
# Possibly this should be a LRU cache. A dictionary from key_prefix to
1990
# (cache_dict, history_vector) for parsed kndx files.
1991
self._kndx_cache = {}
1992
self._scope = self._get_scope()
1993
allow_writes = self._allow_writes()
1999
def _split_key(self, key):
2000
"""Split key into a prefix and suffix."""
2001
return key[:-1], key[-1]
2004
class _KnitGraphIndex(object):
2005
"""A KnitVersionedFiles index layered on GraphIndex."""
2007
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2009
"""Construct a KnitGraphIndex on a graph_index.
2011
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2012
:param is_locked: A callback to check whether the object should answer
2014
:param deltas: Allow delta-compressed records.
2015
:param parents: If True, record knits parents, if not do not record
2017
:param add_callback: If not None, allow additions to the index and call
2018
this callback with a list of added GraphIndex nodes:
2019
[(node, value, node_refs), ...]
2020
:param is_locked: A callback, returns True if the index is locked and
2023
self._add_callback = add_callback
2024
self._graph_index = graph_index
2025
self._deltas = deltas
2026
self._parents = parents
2027
if deltas and not parents:
2028
# XXX: TODO: Delta tree and parent graph should be conceptually
2030
raise KnitCorrupt(self, "Cannot do delta compression without "
2032
self.has_graph = parents
2033
self._is_locked = is_locked
2036
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2038
def add_records(self, records, random_id=False):
2039
"""Add multiple records to the index.
2041
This function does not insert data into the Immutable GraphIndex
2042
backing the KnitGraphIndex, instead it prepares data for insertion by
2043
the caller and checks that it is safe to insert then calls
2044
self._add_callback with the prepared GraphIndex nodes.
2046
:param records: a list of tuples:
2047
(key, options, access_memo, parents).
2048
:param random_id: If True the ids being added were randomly generated
2049
and no check for existence will be performed.
2051
if not self._add_callback:
2052
raise errors.ReadOnlyError(self)
2053
# we hope there are no repositories with inconsistent parentage
2057
for (key, options, access_memo, parents) in records:
2059
parents = tuple(parents)
2060
index, pos, size = access_memo
2061
if 'no-eol' in options:
2065
value += "%d %d" % (pos, size)
2066
if not self._deltas:
2067
if 'line-delta' in options:
2068
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2071
if 'line-delta' in options:
2072
node_refs = (parents, (parents[0],))
2074
node_refs = (parents, ())
2076
node_refs = (parents, )
2079
raise KnitCorrupt(self, "attempt to add node with parents "
2080
"in parentless index.")
2082
keys[key] = (value, node_refs)
2085
present_nodes = self._get_entries(keys)
2086
for (index, key, value, node_refs) in present_nodes:
2087
if (value[0] != keys[key][0][0] or
2088
node_refs != keys[key][1]):
2089
raise KnitCorrupt(self, "inconsistent details in add_records"
2090
": %s %s" % ((value, node_refs), keys[key]))
2094
for key, (value, node_refs) in keys.iteritems():
2095
result.append((key, value, node_refs))
2097
for key, (value, node_refs) in keys.iteritems():
2098
result.append((key, value))
2099
self._add_callback(result)
2101
def _check_read(self):
2102
"""raise if reads are not permitted."""
2103
if not self._is_locked():
2104
raise errors.ObjectNotLocked(self)
2106
def _check_write_ok(self):
2107
"""Assert if writes are not permitted."""
2108
if not self._is_locked():
2109
raise errors.ObjectNotLocked(self)
2111
def _compression_parent(self, an_entry):
2112
# return the key that an_entry is compressed against, or None
2113
# Grab the second parent list (as deltas implies parents currently)
2114
compression_parents = an_entry[3][1]
2115
if not compression_parents:
2117
if len(compression_parents) != 1:
2118
raise AssertionError(
2119
"Too many compression parents: %r" % compression_parents)
2120
return compression_parents[0]
2122
def get_build_details(self, keys):
2123
"""Get the method, index_memo and compression parent for version_ids.
2125
Ghosts are omitted from the result.
2127
:param keys: An iterable of keys.
2128
:return: A dict of key:
2129
(index_memo, compression_parent, parents, record_details).
2131
opaque structure to pass to read_records to extract the raw
2134
Content that this record is built upon, may be None
2136
Logical parents of this node
2138
extra information about the content which needs to be passed to
2139
Factory.parse_record
2143
entries = self._get_entries(keys, False)
2144
for entry in entries:
2146
if not self._parents:
2149
parents = entry[3][0]
2150
if not self._deltas:
2151
compression_parent_key = None
2153
compression_parent_key = self._compression_parent(entry)
2154
noeol = (entry[2][0] == 'N')
2155
if compression_parent_key:
2156
method = 'line-delta'
2159
result[key] = (self._node_to_position(entry),
2160
compression_parent_key, parents,
2164
def _get_entries(self, keys, check_present=False):
2165
"""Get the entries for keys.
2167
:param keys: An iterable of index key tuples.
2172
for node in self._graph_index.iter_entries(keys):
2174
found_keys.add(node[1])
2176
# adapt parentless index to the rest of the code.
2177
for node in self._graph_index.iter_entries(keys):
2178
yield node[0], node[1], node[2], ()
2179
found_keys.add(node[1])
2181
missing_keys = keys.difference(found_keys)
2183
raise RevisionNotPresent(missing_keys.pop(), self)
2185
def get_method(self, key):
2186
"""Return compression method of specified key."""
2187
return self._get_method(self._get_node(key))
2189
def _get_method(self, node):
2190
if not self._deltas:
2192
if self._compression_parent(node):
2197
def _get_node(self, key):
2199
return list(self._get_entries([key]))[0]
2201
raise RevisionNotPresent(key, self)
2203
def get_options(self, key):
2204
"""Return a list representing options.
2208
node = self._get_node(key)
2209
options = [self._get_method(node)]
2210
if node[2][0] == 'N':
2211
options.append('no-eol')
2214
def get_parent_map(self, keys):
2215
"""Get a map of the parents of keys.
2217
:param keys: The keys to look up parents for.
2218
:return: A mapping from keys to parents. Absent keys are absent from
2222
nodes = self._get_entries(keys)
2226
result[node[1]] = node[3][0]
2229
result[node[1]] = None
2232
def get_position(self, key):
2233
"""Return details needed to access the version.
2235
:return: a tuple (index, data position, size) to hand to the access
2236
logic to get the record.
2238
node = self._get_node(key)
2239
return self._node_to_position(node)
2242
"""Get all the keys in the collection.
2244
The keys are not ordered.
2247
return [node[1] for node in self._graph_index.iter_all_entries()]
2249
def _node_to_position(self, node):
2250
"""Convert an index value to position details."""
2251
bits = node[2][1:].split(' ')
2252
return node[0], int(bits[0]), int(bits[1])
2255
class _KnitKeyAccess(object):
2256
"""Access to records in .knit files."""
2258
def __init__(self, transport, mapper):
2259
"""Create a _KnitKeyAccess with transport and mapper.
2261
:param transport: The transport the access object is rooted at.
2262
:param mapper: The mapper used to map keys to .knit files.
2264
self._transport = transport
2265
self._mapper = mapper
2267
def add_raw_records(self, key_sizes, raw_data):
2268
"""Add raw knit bytes to a storage area.
2270
The data is spooled to the container writer in one bytes-record per
2273
:param sizes: An iterable of tuples containing the key and size of each
2275
:param raw_data: A bytestring containing the data.
2276
:return: A list of memos to retrieve the record later. Each memo is an
2277
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
2278
length), where the key is the record key.
2280
if type(raw_data) != str:
2281
raise AssertionError(
2282
'data must be plain bytes was %s' % type(raw_data))
2285
# TODO: This can be tuned for writing to sftp and other servers where
2286
# append() is relatively expensive by grouping the writes to each key
2288
for key, size in key_sizes:
2289
path = self._mapper.map(key)
2291
base = self._transport.append_bytes(path + '.knit',
2292
raw_data[offset:offset+size])
2293
except errors.NoSuchFile:
2294
self._transport.mkdir(osutils.dirname(path))
2295
base = self._transport.append_bytes(path + '.knit',
2296
raw_data[offset:offset+size])
2300
result.append((key, base, size))
2303
def get_raw_records(self, memos_for_retrieval):
2304
"""Get the raw bytes for a records.
2306
:param memos_for_retrieval: An iterable containing the access memo for
2307
retrieving the bytes.
2308
:return: An iterator over the bytes of the records.
2310
# first pass, group into same-index request to minimise readv's issued.
2312
current_prefix = None
2313
for (key, offset, length) in memos_for_retrieval:
2314
if current_prefix == key[:-1]:
2315
current_list.append((offset, length))
2317
if current_prefix is not None:
2318
request_lists.append((current_prefix, current_list))
2319
current_prefix = key[:-1]
2320
current_list = [(offset, length)]
2321
# handle the last entry
2322
if current_prefix is not None:
2323
request_lists.append((current_prefix, current_list))
2324
for prefix, read_vector in request_lists:
2325
path = self._mapper.map(prefix) + '.knit'
2326
for pos, data in self._transport.readv(path, read_vector):
2330
class _DirectPackAccess(object):
2331
"""Access to data in one or more packs with less translation."""
2333
def __init__(self, index_to_packs):
2334
"""Create a _DirectPackAccess object.
2336
:param index_to_packs: A dict mapping index objects to the transport
2337
and file names for obtaining data.
2339
self._container_writer = None
2340
self._write_index = None
2341
self._indices = index_to_packs
2343
def add_raw_records(self, key_sizes, raw_data):
2344
"""Add raw knit bytes to a storage area.
2346
The data is spooled to the container writer in one bytes-record per
2349
:param sizes: An iterable of tuples containing the key and size of each
2351
:param raw_data: A bytestring containing the data.
2352
:return: A list of memos to retrieve the record later. Each memo is an
2353
opaque index memo. For _DirectPackAccess the memo is (index, pos,
2354
length), where the index field is the write_index object supplied
2355
to the PackAccess object.
2357
if type(raw_data) != str:
2358
raise AssertionError(
2359
'data must be plain bytes was %s' % type(raw_data))
2362
for key, size in key_sizes:
2363
p_offset, p_length = self._container_writer.add_bytes_record(
2364
raw_data[offset:offset+size], [])
2366
result.append((self._write_index, p_offset, p_length))
2369
def get_raw_records(self, memos_for_retrieval):
2370
"""Get the raw bytes for a records.
2372
:param memos_for_retrieval: An iterable containing the (index, pos,
2373
length) memo for retrieving the bytes. The Pack access method
2374
looks up the pack to use for a given record in its index_to_pack
2376
:return: An iterator over the bytes of the records.
2378
# first pass, group into same-index requests
2380
current_index = None
2381
for (index, offset, length) in memos_for_retrieval:
2382
if current_index == index:
2383
current_list.append((offset, length))
2385
if current_index is not None:
2386
request_lists.append((current_index, current_list))
2387
current_index = index
2388
current_list = [(offset, length)]
2389
# handle the last entry
2390
if current_index is not None:
2391
request_lists.append((current_index, current_list))
2392
for index, offsets in request_lists:
2393
transport, path = self._indices[index]
2394
reader = pack.make_readv_reader(transport, path, offsets)
2395
for names, read_func in reader.iter_records():
2396
yield read_func(None)
2398
def set_writer(self, writer, index, transport_packname):
2399
"""Set a writer to use for adding data."""
2400
if index is not None:
2401
self._indices[index] = transport_packname
2402
self._container_writer = writer
2403
self._write_index = index
2406
# Deprecated, use PatienceSequenceMatcher instead
2407
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2410
def annotate_knit(knit, revision_id):
2411
"""Annotate a knit with no cached annotations.
2413
This implementation is for knits with no cached annotations.
2414
It will work for knits with cached annotations, but this is not
2417
annotator = _KnitAnnotator(knit)
2418
return iter(annotator.annotate(revision_id))
2421
class _KnitAnnotator(object):
2422
"""Build up the annotations for a text."""
2424
def __init__(self, knit):
2427
# Content objects, differs from fulltexts because of how final newlines
2428
# are treated by knits. the content objects here will always have a
2430
self._fulltext_contents = {}
2432
# Annotated lines of specific revisions
2433
self._annotated_lines = {}
2435
# Track the raw data for nodes that we could not process yet.
2436
# This maps the revision_id of the base to a list of children that will
2437
# annotated from it.
2438
self._pending_children = {}
2440
# Nodes which cannot be extracted
2441
self._ghosts = set()
2443
# Track how many children this node has, so we know if we need to keep
2445
self._annotate_children = {}
2446
self._compression_children = {}
2448
self._all_build_details = {}
2449
# The children => parent revision_id graph
2450
self._revision_id_graph = {}
2452
self._heads_provider = None
2454
self._nodes_to_keep_annotations = set()
2455
self._generations_until_keep = 100
2457
def set_generations_until_keep(self, value):
2458
"""Set the number of generations before caching a node.
2460
Setting this to -1 will cache every merge node, setting this higher
2461
will cache fewer nodes.
2463
self._generations_until_keep = value
2465
def _add_fulltext_content(self, revision_id, content_obj):
2466
self._fulltext_contents[revision_id] = content_obj
2467
# TODO: jam 20080305 It might be good to check the sha1digest here
2468
return content_obj.text()
2470
def _check_parents(self, child, nodes_to_annotate):
2471
"""Check if all parents have been processed.
2473
:param child: A tuple of (rev_id, parents, raw_content)
2474
:param nodes_to_annotate: If child is ready, add it to
2475
nodes_to_annotate, otherwise put it back in self._pending_children
2477
for parent_id in child[1]:
2478
if (parent_id not in self._annotated_lines):
2479
# This parent is present, but another parent is missing
2480
self._pending_children.setdefault(parent_id,
2484
# This one is ready to be processed
2485
nodes_to_annotate.append(child)
2487
def _add_annotation(self, revision_id, fulltext, parent_ids,
2488
left_matching_blocks=None):
2489
"""Add an annotation entry.
2491
All parents should already have been annotated.
2492
:return: A list of children that now have their parents satisfied.
2494
a = self._annotated_lines
2495
annotated_parent_lines = [a[p] for p in parent_ids]
2496
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2497
fulltext, revision_id, left_matching_blocks,
2498
heads_provider=self._get_heads_provider()))
2499
self._annotated_lines[revision_id] = annotated_lines
2500
for p in parent_ids:
2501
ann_children = self._annotate_children[p]
2502
ann_children.remove(revision_id)
2503
if (not ann_children
2504
and p not in self._nodes_to_keep_annotations):
2505
del self._annotated_lines[p]
2506
del self._all_build_details[p]
2507
if p in self._fulltext_contents:
2508
del self._fulltext_contents[p]
2509
# Now that we've added this one, see if there are any pending
2510
# deltas to be done, certainly this parent is finished
2511
nodes_to_annotate = []
2512
for child in self._pending_children.pop(revision_id, []):
2513
self._check_parents(child, nodes_to_annotate)
2514
return nodes_to_annotate
2516
def _get_build_graph(self, key):
2517
"""Get the graphs for building texts and annotations.
2519
The data you need for creating a full text may be different than the
2520
data you need to annotate that text. (At a minimum, you need both
2521
parents to create an annotation, but only need 1 parent to generate the
2524
:return: A list of (key, index_memo) records, suitable for
2525
passing to read_records_iter to start reading in the raw data fro/
2528
if key in self._annotated_lines:
2531
pending = set([key])
2536
# get all pending nodes
2538
this_iteration = pending
2539
build_details = self._knit._index.get_build_details(this_iteration)
2540
self._all_build_details.update(build_details)
2541
# new_nodes = self._knit._index._get_entries(this_iteration)
2543
for key, details in build_details.iteritems():
2544
(index_memo, compression_parent, parents,
2545
record_details) = details
2546
self._revision_id_graph[key] = parents
2547
records.append((key, index_memo))
2548
# Do we actually need to check _annotated_lines?
2549
pending.update(p for p in parents
2550
if p not in self._all_build_details)
2551
if compression_parent:
2552
self._compression_children.setdefault(compression_parent,
2555
for parent in parents:
2556
self._annotate_children.setdefault(parent,
2558
num_gens = generation - kept_generation
2559
if ((num_gens >= self._generations_until_keep)
2560
and len(parents) > 1):
2561
kept_generation = generation
2562
self._nodes_to_keep_annotations.add(key)
2564
missing_versions = this_iteration.difference(build_details.keys())
2565
self._ghosts.update(missing_versions)
2566
for missing_version in missing_versions:
2567
# add a key, no parents
2568
self._revision_id_graph[missing_version] = ()
2569
pending.discard(missing_version) # don't look for it
2570
if self._ghosts.intersection(self._compression_children):
2572
"We cannot have nodes which have a ghost compression parent:\n"
2574
"compression children: %r"
2575
% (self._ghosts, self._compression_children))
2576
# Cleanout anything that depends on a ghost so that we don't wait for
2577
# the ghost to show up
2578
for node in self._ghosts:
2579
if node in self._annotate_children:
2580
# We won't be building this node
2581
del self._annotate_children[node]
2582
# Generally we will want to read the records in reverse order, because
2583
# we find the parent nodes after the children
2587
def _annotate_records(self, records):
2588
"""Build the annotations for the listed records."""
2589
# We iterate in the order read, rather than a strict order requested
2590
# However, process what we can, and put off to the side things that
2591
# still need parents, cleaning them up when those parents are
2593
for (rev_id, record,
2594
digest) in self._knit._read_records_iter(records):
2595
if rev_id in self._annotated_lines:
2597
parent_ids = self._revision_id_graph[rev_id]
2598
parent_ids = [p for p in parent_ids if p not in self._ghosts]
2599
details = self._all_build_details[rev_id]
2600
(index_memo, compression_parent, parents,
2601
record_details) = details
2602
nodes_to_annotate = []
2603
# TODO: Remove the punning between compression parents, and
2604
# parent_ids, we should be able to do this without assuming
2606
if len(parent_ids) == 0:
2607
# There are no parents for this node, so just add it
2608
# TODO: This probably needs to be decoupled
2609
fulltext_content, delta = self._knit._factory.parse_record(
2610
rev_id, record, record_details, None)
2611
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
2612
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
2613
parent_ids, left_matching_blocks=None))
2615
child = (rev_id, parent_ids, record)
2616
# Check if all the parents are present
2617
self._check_parents(child, nodes_to_annotate)
2618
while nodes_to_annotate:
2619
# Should we use a queue here instead of a stack?
2620
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
2621
(index_memo, compression_parent, parents,
2622
record_details) = self._all_build_details[rev_id]
2623
if compression_parent is not None:
2624
comp_children = self._compression_children[compression_parent]
2625
if rev_id not in comp_children:
2626
raise AssertionError("%r not in compression children %r"
2627
% (rev_id, comp_children))
2628
# If there is only 1 child, it is safe to reuse this
2630
reuse_content = (len(comp_children) == 1
2631
and compression_parent not in
2632
self._nodes_to_keep_annotations)
2634
# Remove it from the cache since it will be changing
2635
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
2636
# Make sure to copy the fulltext since it might be
2638
parent_fulltext = list(parent_fulltext_content.text())
2640
parent_fulltext_content = self._fulltext_contents[compression_parent]
2641
parent_fulltext = parent_fulltext_content.text()
2642
comp_children.remove(rev_id)
2643
fulltext_content, delta = self._knit._factory.parse_record(
2644
rev_id, record, record_details,
2645
parent_fulltext_content,
2646
copy_base_content=(not reuse_content))
2647
fulltext = self._add_fulltext_content(rev_id,
2649
blocks = KnitContent.get_line_delta_blocks(delta,
2650
parent_fulltext, fulltext)
2652
fulltext_content = self._knit._factory.parse_fulltext(
2654
fulltext = self._add_fulltext_content(rev_id,
2657
nodes_to_annotate.extend(
2658
self._add_annotation(rev_id, fulltext, parent_ids,
2659
left_matching_blocks=blocks))
2661
def _get_heads_provider(self):
2662
"""Create a heads provider for resolving ancestry issues."""
2663
if self._heads_provider is not None:
2664
return self._heads_provider
2665
parent_provider = _mod_graph.DictParentsProvider(
2666
self._revision_id_graph)
2667
graph_obj = _mod_graph.Graph(parent_provider)
2668
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
2669
self._heads_provider = head_cache
2672
def annotate(self, key):
2673
"""Return the annotated fulltext at the given key.
2675
:param key: The key to annotate.
2677
if True or len(self._knit._fallback_vfs) > 0:
2678
# stacked knits can't use the fast path at present.
2679
return self._simple_annotate(key)
2680
records = self._get_build_graph(key)
2681
if key in self._ghosts:
2682
raise errors.RevisionNotPresent(key, self._knit)
2683
self._annotate_records(records)
2684
return self._annotated_lines[key]
2686
def _simple_annotate(self, key):
2687
"""Return annotated fulltext, rediffing from the full texts.
2689
This is slow but makes no assumptions about the repository
2690
being able to produce line deltas.
2692
# TODO: this code generates a parent maps of present ancestors; it
2693
# could be split out into a separate method, and probably should use
2694
# iter_ancestry instead. -- mbp and robertc 20080704
2695
graph = _mod_graph.Graph(self._knit)
2696
head_cache = _mod_graph.FrozenHeadsCache(graph)
2697
search = graph._make_breadth_first_searcher([key])
2701
present, ghosts = search.next_with_ghosts()
2702
except StopIteration:
2704
keys.update(present)
2705
parent_map = self._knit.get_parent_map(keys)
2707
reannotate = annotate.reannotate
2708
for record in self._knit.get_record_stream(keys, 'topological', True):
2710
fulltext = split_lines(record.get_bytes_as('fulltext'))
2711
parents = parent_map[key]
2712
if parents is not None:
2713
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
2716
parent_cache[key] = list(
2717
reannotate(parent_lines, fulltext, key, None, head_cache))
2719
return parent_cache[key]
2721
raise errors.RevisionNotPresent(key, self._knit)
2725
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2727
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
