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.
702
:ivar _index: A _KnitGraphIndex or similar that can describe the
703
parents, graph, compression and data location of entries in this
704
KnitVersionedFiles. Note that this is only the index for
705
*this* vfs; if there are fallbacks they must be queried separately.
708
def __init__(self, index, data_access, max_delta_chain=200,
710
"""Create a KnitVersionedFiles with index and data_access.
712
:param index: The index for the knit data.
713
:param data_access: The access object to store and retrieve knit
715
:param max_delta_chain: The maximum number of deltas to permit during
716
insertion. Set to 0 to prohibit the use of deltas.
717
:param annotated: Set to True to cause annotations to be calculated and
718
stored during insertion.
721
self._access = data_access
722
self._max_delta_chain = max_delta_chain
724
self._factory = KnitAnnotateFactory()
726
self._factory = KnitPlainFactory()
727
self._fallback_vfs = []
730
return "%s(%r, %r)" % (
731
self.__class__.__name__,
735
def add_fallback_versioned_files(self, a_versioned_files):
736
"""Add a source of texts for texts not present in this knit.
738
:param a_versioned_files: A VersionedFiles object.
740
self._fallback_vfs.append(a_versioned_files)
742
def add_lines(self, key, parents, lines, parent_texts=None,
743
left_matching_blocks=None, nostore_sha=None, random_id=False,
745
"""See VersionedFiles.add_lines()."""
746
self._index._check_write_ok()
747
self._check_add(key, lines, random_id, check_content)
749
# The caller might pass None if there is no graph data, but kndx
750
# indexes can't directly store that, so we give them
751
# an empty tuple instead.
753
return self._add(key, lines, parents,
754
parent_texts, left_matching_blocks, nostore_sha, random_id)
756
def _add(self, key, lines, parents, parent_texts,
757
left_matching_blocks, nostore_sha, random_id):
758
"""Add a set of lines on top of version specified by parents.
760
Any versions not present will be converted into ghosts.
762
# first thing, if the content is something we don't need to store, find
764
line_bytes = ''.join(lines)
765
digest = sha_string(line_bytes)
766
if nostore_sha == digest:
767
raise errors.ExistingContent
770
if parent_texts is None:
772
# Do a single query to ascertain parent presence.
773
present_parent_map = self.get_parent_map(parents)
774
for parent in parents:
775
if parent in present_parent_map:
776
present_parents.append(parent)
778
# Currently we can only compress against the left most present parent.
779
if (len(present_parents) == 0 or
780
present_parents[0] != parents[0]):
783
# To speed the extract of texts the delta chain is limited
784
# to a fixed number of deltas. This should minimize both
785
# I/O and the time spend applying deltas.
786
delta = self._check_should_delta(present_parents[0])
788
text_length = len(line_bytes)
791
if lines[-1][-1] != '\n':
792
# copy the contents of lines.
794
options.append('no-eol')
795
lines[-1] = lines[-1] + '\n'
799
if type(element) != str:
800
raise TypeError("key contains non-strings: %r" % (key,))
801
# Knit hunks are still last-element only
803
content = self._factory.make(lines, version_id)
804
if 'no-eol' in options:
805
# Hint to the content object that its text() call should strip the
807
content._should_strip_eol = True
808
if delta or (self._factory.annotated and len(present_parents) > 0):
809
# Merge annotations from parent texts if needed.
810
delta_hunks = self._merge_annotations(content, present_parents,
811
parent_texts, delta, self._factory.annotated,
812
left_matching_blocks)
815
options.append('line-delta')
816
store_lines = self._factory.lower_line_delta(delta_hunks)
817
size, bytes = self._record_to_data(key, digest,
820
options.append('fulltext')
821
# isinstance is slower and we have no hierarchy.
822
if self._factory.__class__ == KnitPlainFactory:
823
# Use the already joined bytes saving iteration time in
825
size, bytes = self._record_to_data(key, digest,
828
# get mixed annotation + content and feed it into the
830
store_lines = self._factory.lower_fulltext(content)
831
size, bytes = self._record_to_data(key, digest,
834
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
835
self._index.add_records(
836
((key, options, access_memo, parents),),
838
return digest, text_length, content
840
def annotate(self, key):
841
"""See VersionedFiles.annotate."""
842
return self._factory.annotate(self, key)
844
def check(self, progress_bar=None):
845
"""See VersionedFiles.check()."""
846
# This doesn't actually test extraction of everything, but that will
847
# impact 'bzr check' substantially, and needs to be integrated with
848
# care. However, it does check for the obvious problem of a delta with
850
keys = self._index.keys()
851
parent_map = self.get_parent_map(keys)
853
if self._index.get_method(key) != 'fulltext':
854
compression_parent = parent_map[key][0]
855
if compression_parent not in parent_map:
856
raise errors.KnitCorrupt(self,
857
"Missing basis parent %s for %s" % (
858
compression_parent, key))
859
for fallback_vfs in self._fallback_vfs:
862
def _check_add(self, key, lines, random_id, check_content):
863
"""check that version_id and lines are safe to add."""
865
if contains_whitespace(version_id):
866
raise InvalidRevisionId(version_id, self)
867
self.check_not_reserved_id(version_id)
868
# TODO: If random_id==False and the key is already present, we should
869
# probably check that the existing content is identical to what is
870
# being inserted, and otherwise raise an exception. This would make
871
# the bundle code simpler.
873
self._check_lines_not_unicode(lines)
874
self._check_lines_are_lines(lines)
876
def _check_header(self, key, line):
877
rec = self._split_header(line)
878
self._check_header_version(rec, key[-1])
881
def _check_header_version(self, rec, version_id):
882
"""Checks the header version on original format knit records.
884
These have the last component of the key embedded in the record.
886
if rec[1] != version_id:
887
raise KnitCorrupt(self,
888
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
890
def _check_should_delta(self, parent):
891
"""Iterate back through the parent listing, looking for a fulltext.
893
This is used when we want to decide whether to add a delta or a new
894
fulltext. It searches for _max_delta_chain parents. When it finds a
895
fulltext parent, it sees if the total size of the deltas leading up to
896
it is large enough to indicate that we want a new full text anyway.
898
Return True if we should create a new delta, False if we should use a
903
for count in xrange(self._max_delta_chain):
904
# XXX: Collapse these two queries:
906
# Note that this only looks in the index of this particular
907
# KnitVersionedFiles, not in the fallbacks. This ensures that
908
# we won't store a delta spanning physical repository
910
method = self._index.get_method(parent)
911
except RevisionNotPresent:
912
# Some basis is not locally present: always delta
914
index, pos, size = self._index.get_position(parent)
915
if method == 'fulltext':
919
# We don't explicitly check for presence because this is in an
920
# inner loop, and if it's missing it'll fail anyhow.
921
# TODO: This should be asking for compression parent, not graph
923
parent = self._index.get_parent_map([parent])[parent][0]
925
# We couldn't find a fulltext, so we must create a new one
927
# Simple heuristic - if the total I/O wold be greater as a delta than
928
# the originally installed fulltext, we create a new fulltext.
929
return fulltext_size > delta_size
931
def _build_details_to_components(self, build_details):
932
"""Convert a build_details tuple to a position tuple."""
933
# record_details, access_memo, compression_parent
934
return build_details[3], build_details[0], build_details[1]
936
def _get_components_positions(self, keys, allow_missing=False):
937
"""Produce a map of position data for the components of keys.
939
This data is intended to be used for retrieving the knit records.
941
A dict of key to (record_details, index_memo, next, parents) is
943
method is the way referenced data should be applied.
944
index_memo is the handle to pass to the data access to actually get the
946
next is the build-parent of the version, or None for fulltexts.
947
parents is the version_ids of the parents of this version
949
:param allow_missing: If True do not raise an error on a missing component,
953
pending_components = keys
954
while pending_components:
955
build_details = self._index.get_build_details(pending_components)
956
current_components = set(pending_components)
957
pending_components = set()
958
for key, details in build_details.iteritems():
959
(index_memo, compression_parent, parents,
960
record_details) = details
961
method = record_details[0]
962
if compression_parent is not None:
963
pending_components.add(compression_parent)
964
component_data[key] = self._build_details_to_components(details)
965
missing = current_components.difference(build_details)
966
if missing and not allow_missing:
967
raise errors.RevisionNotPresent(missing.pop(), self)
968
return component_data
970
def _get_content(self, key, parent_texts={}):
971
"""Returns a content object that makes up the specified
973
cached_version = parent_texts.get(key, None)
974
if cached_version is not None:
975
# Ensure the cache dict is valid.
976
if not self.get_parent_map([key]):
977
raise RevisionNotPresent(key, self)
978
return cached_version
979
text_map, contents_map = self._get_content_maps([key])
980
return contents_map[key]
982
def _get_content_maps(self, keys, nonlocal_keys=None):
983
"""Produce maps of text and KnitContents
985
:param keys: The keys to produce content maps for.
986
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
987
which are known to not be in this knit, but rather in one of the
989
:return: (text_map, content_map) where text_map contains the texts for
990
the requested versions and content_map contains the KnitContents.
992
# FUTURE: This function could be improved for the 'extract many' case
993
# by tracking each component and only doing the copy when the number of
994
# children than need to apply delta's to it is > 1 or it is part of the
997
multiple_versions = len(keys) != 1
998
record_map = self._get_record_map(keys, allow_missing=True)
1003
if nonlocal_keys is None:
1004
nonlocal_keys = set()
1006
nonlocal_keys = frozenset(nonlocal_keys)
1007
missing_keys = set(nonlocal_keys)
1008
for source in self._fallback_vfs:
1009
if not missing_keys:
1011
for record in source.get_record_stream(missing_keys,
1013
if record.storage_kind == 'absent':
1015
missing_keys.remove(record.key)
1016
lines = split_lines(record.get_bytes_as('fulltext'))
1017
text_map[record.key] = lines
1018
content_map[record.key] = PlainKnitContent(lines, record.key)
1019
if record.key in keys:
1020
final_content[record.key] = content_map[record.key]
1022
if key in nonlocal_keys:
1027
while cursor is not None:
1029
record, record_details, digest, next = record_map[cursor]
1031
raise RevisionNotPresent(cursor, self)
1032
components.append((cursor, record, record_details, digest))
1034
if cursor in content_map:
1035
# no need to plan further back
1036
components.append((cursor, None, None, None))
1040
for (component_id, record, record_details,
1041
digest) in reversed(components):
1042
if component_id in content_map:
1043
content = content_map[component_id]
1045
content, delta = self._factory.parse_record(key[-1],
1046
record, record_details, content,
1047
copy_base_content=multiple_versions)
1048
if multiple_versions:
1049
content_map[component_id] = content
1051
final_content[key] = content
1053
# digest here is the digest from the last applied component.
1054
text = content.text()
1055
actual_sha = sha_strings(text)
1056
if actual_sha != digest:
1057
raise KnitCorrupt(self,
1059
'\n of reconstructed text does not match'
1061
'\n for version %s' %
1062
(actual_sha, digest, key))
1063
text_map[key] = text
1064
return text_map, final_content
1066
def get_parent_map(self, keys):
1067
"""Get a map of the graph parents of keys.
1069
:param keys: The keys to look up parents for.
1070
:return: A mapping from keys to parents. Absent keys are absent from
1073
return self._get_parent_map_with_sources(keys)[0]
1075
def _get_parent_map_with_sources(self, keys):
1076
"""Get a map of the parents of keys.
1078
:param keys: The keys to look up parents for.
1079
:return: A tuple. The first element is a mapping from keys to parents.
1080
Absent keys are absent from the mapping. The second element is a
1081
list with the locations each key was found in. The first element
1082
is the in-this-knit parents, the second the first fallback source,
1086
sources = [self._index] + self._fallback_vfs
1089
for source in sources:
1092
new_result = source.get_parent_map(missing)
1093
source_results.append(new_result)
1094
result.update(new_result)
1095
missing.difference_update(set(new_result))
1096
return result, source_results
1098
def _get_record_map(self, keys, allow_missing=False):
1099
"""Produce a dictionary of knit records.
1101
:return: {key:(record, record_details, digest, next)}
1103
data returned from read_records
1105
opaque information to pass to parse_record
1107
SHA1 digest of the full text after all steps are done
1109
build-parent of the version, i.e. the leftmost ancestor.
1110
Will be None if the record is not a delta.
1111
:param keys: The keys to build a map for
1112
:param allow_missing: If some records are missing, rather than
1113
error, just return the data that could be generated.
1115
position_map = self._get_components_positions(keys,
1116
allow_missing=allow_missing)
1117
# key = component_id, r = record_details, i_m = index_memo, n = next
1118
records = [(key, i_m) for key, (r, i_m, n)
1119
in position_map.iteritems()]
1121
for key, record, digest in \
1122
self._read_records_iter(records):
1123
(record_details, index_memo, next) = position_map[key]
1124
record_map[key] = record, record_details, digest, next
1127
def get_record_stream(self, keys, ordering, include_delta_closure):
1128
"""Get a stream of records for keys.
1130
:param keys: The keys to include.
1131
:param ordering: Either 'unordered' or 'topological'. A topologically
1132
sorted stream has compression parents strictly before their
1134
:param include_delta_closure: If True then the closure across any
1135
compression parents will be included (in the opaque data).
1136
:return: An iterator of ContentFactory objects, each of which is only
1137
valid until the iterator is advanced.
1139
# keys might be a generator
1143
if not self._index.has_graph:
1144
# Cannot topological order when no graph has been stored.
1145
ordering = 'unordered'
1146
if include_delta_closure:
1147
positions = self._get_components_positions(keys, allow_missing=True)
1149
build_details = self._index.get_build_details(keys)
1151
# (record_details, access_memo, compression_parent_key)
1152
positions = dict((key, self._build_details_to_components(details))
1153
for key, details in build_details.iteritems())
1154
absent_keys = keys.difference(set(positions))
1155
# There may be more absent keys : if we're missing the basis component
1156
# and are trying to include the delta closure.
1157
if include_delta_closure:
1158
needed_from_fallback = set()
1159
# Build up reconstructable_keys dict. key:True in this dict means
1160
# the key can be reconstructed.
1161
reconstructable_keys = {}
1165
chain = [key, positions[key][2]]
1167
needed_from_fallback.add(key)
1170
while chain[-1] is not None:
1171
if chain[-1] in reconstructable_keys:
1172
result = reconstructable_keys[chain[-1]]
1176
chain.append(positions[chain[-1]][2])
1178
# missing basis component
1179
needed_from_fallback.add(chain[-1])
1182
for chain_key in chain[:-1]:
1183
reconstructable_keys[chain_key] = result
1185
needed_from_fallback.add(key)
1186
# Double index lookups here : need a unified api ?
1187
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1188
if ordering == 'topological':
1189
# Global topological sort
1190
present_keys = tsort.topo_sort(global_map)
1191
# Now group by source:
1193
current_source = None
1194
for key in present_keys:
1195
for parent_map in parent_maps:
1196
if key in parent_map:
1197
key_source = parent_map
1199
if current_source is not key_source:
1200
source_keys.append((key_source, []))
1201
current_source = key_source
1202
source_keys[-1][1].append(key)
1204
if ordering != 'unordered':
1205
raise AssertionError('valid values for ordering are:'
1206
' "unordered" or "topological" not: %r'
1208
# Just group by source; remote sources first.
1211
for parent_map in reversed(parent_maps):
1212
source_keys.append((parent_map, []))
1213
for key in parent_map:
1214
present_keys.append(key)
1215
source_keys[-1][1].append(key)
1216
absent_keys = keys - set(global_map)
1217
for key in absent_keys:
1218
yield AbsentContentFactory(key)
1219
# restrict our view to the keys we can answer.
1220
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1221
# XXX: At that point we need to consider the impact of double reads by
1222
# utilising components multiple times.
1223
if include_delta_closure:
1224
# XXX: get_content_maps performs its own index queries; allow state
1226
text_map, _ = self._get_content_maps(present_keys,
1227
needed_from_fallback - absent_keys)
1228
for key in present_keys:
1229
yield FulltextContentFactory(key, global_map[key], None,
1230
''.join(text_map[key]))
1232
for source, keys in source_keys:
1233
if source is parent_maps[0]:
1234
# this KnitVersionedFiles
1235
records = [(key, positions[key][1]) for key in keys]
1236
for key, raw_data, sha1 in self._read_records_iter_raw(records):
1237
(record_details, index_memo, _) = positions[key]
1238
yield KnitContentFactory(key, global_map[key],
1239
record_details, sha1, raw_data, self._factory.annotated, None)
1241
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1242
for record in vf.get_record_stream(keys, ordering,
1243
include_delta_closure):
1246
def get_sha1s(self, keys):
1247
"""See VersionedFiles.get_sha1s()."""
1249
record_map = self._get_record_map(missing, allow_missing=True)
1251
for key, details in record_map.iteritems():
1252
if key not in missing:
1254
# record entry 2 is the 'digest'.
1255
result[key] = details[2]
1256
missing.difference_update(set(result))
1257
for source in self._fallback_vfs:
1260
new_result = source.get_sha1s(missing)
1261
result.update(new_result)
1262
missing.difference_update(set(new_result))
1265
def insert_record_stream(self, stream):
1266
"""Insert a record stream into this container.
1268
:param stream: A stream of records to insert.
1270
:seealso VersionedFiles.get_record_stream:
1272
def get_adapter(adapter_key):
1274
return adapters[adapter_key]
1276
adapter_factory = adapter_registry.get(adapter_key)
1277
adapter = adapter_factory(self)
1278
adapters[adapter_key] = adapter
1280
if self._factory.annotated:
1281
# self is annotated, we need annotated knits to use directly.
1282
annotated = "annotated-"
1285
# self is not annotated, but we can strip annotations cheaply.
1287
convertibles = set(["knit-annotated-ft-gz"])
1288
if self._max_delta_chain:
1289
convertibles.add("knit-annotated-delta-gz")
1290
# The set of types we can cheaply adapt without needing basis texts.
1291
native_types = set()
1292
if self._max_delta_chain:
1293
native_types.add("knit-%sdelta-gz" % annotated)
1294
native_types.add("knit-%sft-gz" % annotated)
1295
knit_types = native_types.union(convertibles)
1297
# Buffer all index entries that we can't add immediately because their
1298
# basis parent is missing. We don't buffer all because generating
1299
# annotations may require access to some of the new records. However we
1300
# can't generate annotations from new deltas until their basis parent
1301
# is present anyway, so we get away with not needing an index that
1302
# includes the new keys.
1303
# key = basis_parent, value = index entry to add
1304
buffered_index_entries = {}
1305
for record in stream:
1306
parents = record.parents
1307
# Raise an error when a record is missing.
1308
if record.storage_kind == 'absent':
1309
raise RevisionNotPresent([record.key], self)
1310
if record.storage_kind in knit_types:
1311
if record.storage_kind not in native_types:
1313
adapter_key = (record.storage_kind, "knit-delta-gz")
1314
adapter = get_adapter(adapter_key)
1316
adapter_key = (record.storage_kind, "knit-ft-gz")
1317
adapter = get_adapter(adapter_key)
1318
bytes = adapter.get_bytes(
1319
record, record.get_bytes_as(record.storage_kind))
1321
bytes = record.get_bytes_as(record.storage_kind)
1322
options = [record._build_details[0]]
1323
if record._build_details[1]:
1324
options.append('no-eol')
1325
# Just blat it across.
1326
# Note: This does end up adding data on duplicate keys. As
1327
# modern repositories use atomic insertions this should not
1328
# lead to excessive growth in the event of interrupted fetches.
1329
# 'knit' repositories may suffer excessive growth, but as a
1330
# deprecated format this is tolerable. It can be fixed if
1331
# needed by in the kndx index support raising on a duplicate
1332
# add with identical parents and options.
1333
access_memo = self._access.add_raw_records(
1334
[(record.key, len(bytes))], bytes)[0]
1335
index_entry = (record.key, options, access_memo, parents)
1337
if 'fulltext' not in options:
1338
basis_parent = parents[0]
1339
# Note that pack backed knits don't need to buffer here
1340
# because they buffer all writes to the transaction level,
1341
# but we don't expose that difference at the index level. If
1342
# the query here has sufficient cost to show up in
1343
# profiling we should do that.
1344
if basis_parent not in self.get_parent_map([basis_parent]):
1345
pending = buffered_index_entries.setdefault(
1347
pending.append(index_entry)
1350
self._index.add_records([index_entry])
1351
elif record.storage_kind == 'fulltext':
1352
self.add_lines(record.key, parents,
1353
split_lines(record.get_bytes_as('fulltext')))
1355
adapter_key = record.storage_kind, 'fulltext'
1356
adapter = get_adapter(adapter_key)
1357
lines = split_lines(adapter.get_bytes(
1358
record, record.get_bytes_as(record.storage_kind)))
1360
self.add_lines(record.key, parents, lines)
1361
except errors.RevisionAlreadyPresent:
1363
# Add any records whose basis parent is now available.
1364
added_keys = [record.key]
1366
key = added_keys.pop(0)
1367
if key in buffered_index_entries:
1368
index_entries = buffered_index_entries[key]
1369
self._index.add_records(index_entries)
1371
[index_entry[0] for index_entry in index_entries])
1372
del buffered_index_entries[key]
1373
# If there were any deltas which had a missing basis parent, error.
1374
if buffered_index_entries:
1375
raise errors.RevisionNotPresent(buffered_index_entries.keys()[0],
1378
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1379
"""Iterate over the lines in the versioned files from keys.
1381
This may return lines from other keys. Each item the returned
1382
iterator yields is a tuple of a line and a text version that that line
1383
is present in (not introduced in).
1385
Ordering of results is in whatever order is most suitable for the
1386
underlying storage format.
1388
If a progress bar is supplied, it may be used to indicate progress.
1389
The caller is responsible for cleaning up progress bars (because this
1393
* Lines are normalised by the underlying store: they will all have \n
1395
* Lines are returned in arbitrary order.
1397
:return: An iterator over (line, key).
1400
pb = progress.DummyProgress()
1403
# we don't care about inclusions, the caller cares.
1404
# but we need to setup a list of records to visit.
1405
# we need key, position, length
1407
build_details = self._index.get_build_details(keys)
1408
for key, details in build_details.iteritems():
1410
key_records.append((key, details[0]))
1412
records_iter = enumerate(self._read_records_iter(key_records))
1413
for (key_idx, (key, data, sha_value)) in records_iter:
1414
pb.update('Walking content.', key_idx, total)
1415
compression_parent = build_details[key][1]
1416
if compression_parent is None:
1418
line_iterator = self._factory.get_fulltext_content(data)
1421
line_iterator = self._factory.get_linedelta_content(data)
1422
# XXX: It might be more efficient to yield (key,
1423
# line_iterator) in the future. However for now, this is a simpler
1424
# change to integrate into the rest of the codebase. RBC 20071110
1425
for line in line_iterator:
1427
for source in self._fallback_vfs:
1431
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1432
source_keys.add(key)
1434
keys.difference_update(source_keys)
1436
# XXX: strictly the second parameter is meant to be the file id
1437
# but it's not easily accessible here.
1438
raise RevisionNotPresent(keys, repr(self))
1439
pb.update('Walking content.', total, total)
1441
def _make_line_delta(self, delta_seq, new_content):
1442
"""Generate a line delta from delta_seq and new_content."""
1444
for op in delta_seq.get_opcodes():
1445
if op[0] == 'equal':
1447
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1450
def _merge_annotations(self, content, parents, parent_texts={},
1451
delta=None, annotated=None,
1452
left_matching_blocks=None):
1453
"""Merge annotations for content and generate deltas.
1455
This is done by comparing the annotations based on changes to the text
1456
and generating a delta on the resulting full texts. If annotations are
1457
not being created then a simple delta is created.
1459
if left_matching_blocks is not None:
1460
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1464
for parent_key in parents:
1465
merge_content = self._get_content(parent_key, parent_texts)
1466
if (parent_key == parents[0] and delta_seq is not None):
1469
seq = patiencediff.PatienceSequenceMatcher(
1470
None, merge_content.text(), content.text())
1471
for i, j, n in seq.get_matching_blocks():
1474
# this copies (origin, text) pairs across to the new
1475
# content for any line that matches the last-checked
1477
content._lines[j:j+n] = merge_content._lines[i:i+n]
1478
# XXX: Robert says the following block is a workaround for a
1479
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1480
if content._lines and content._lines[-1][1][-1] != '\n':
1481
# The copied annotation was from a line without a trailing EOL,
1482
# reinstate one for the content object, to ensure correct
1484
line = content._lines[-1][1] + '\n'
1485
content._lines[-1] = (content._lines[-1][0], line)
1487
if delta_seq is None:
1488
reference_content = self._get_content(parents[0], parent_texts)
1489
new_texts = content.text()
1490
old_texts = reference_content.text()
1491
delta_seq = patiencediff.PatienceSequenceMatcher(
1492
None, old_texts, new_texts)
1493
return self._make_line_delta(delta_seq, content)
1495
def _parse_record(self, version_id, data):
1496
"""Parse an original format knit record.
1498
These have the last element of the key only present in the stored data.
1500
rec, record_contents = self._parse_record_unchecked(data)
1501
self._check_header_version(rec, version_id)
1502
return record_contents, rec[3]
1504
def _parse_record_header(self, key, raw_data):
1505
"""Parse a record header for consistency.
1507
:return: the header and the decompressor stream.
1508
as (stream, header_record)
1510
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1513
rec = self._check_header(key, df.readline())
1514
except Exception, e:
1515
raise KnitCorrupt(self,
1516
"While reading {%s} got %s(%s)"
1517
% (key, e.__class__.__name__, str(e)))
1520
def _parse_record_unchecked(self, data):
1522
# 4168 calls in 2880 217 internal
1523
# 4168 calls to _parse_record_header in 2121
1524
# 4168 calls to readlines in 330
1525
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1527
record_contents = df.readlines()
1528
except Exception, e:
1529
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1530
(data, e.__class__.__name__, str(e)))
1531
header = record_contents.pop(0)
1532
rec = self._split_header(header)
1533
last_line = record_contents.pop()
1534
if len(record_contents) != int(rec[2]):
1535
raise KnitCorrupt(self,
1536
'incorrect number of lines %s != %s'
1537
' for version {%s} %s'
1538
% (len(record_contents), int(rec[2]),
1539
rec[1], record_contents))
1540
if last_line != 'end %s\n' % rec[1]:
1541
raise KnitCorrupt(self,
1542
'unexpected version end line %r, wanted %r'
1543
% (last_line, rec[1]))
1545
return rec, record_contents
1547
def _read_records_iter(self, records):
1548
"""Read text records from data file and yield result.
1550
The result will be returned in whatever is the fastest to read.
1551
Not by the order requested. Also, multiple requests for the same
1552
record will only yield 1 response.
1553
:param records: A list of (key, access_memo) entries
1554
:return: Yields (key, contents, digest) in the order
1555
read, not the order requested
1560
# XXX: This smells wrong, IO may not be getting ordered right.
1561
needed_records = sorted(set(records), key=operator.itemgetter(1))
1562
if not needed_records:
1565
# The transport optimizes the fetching as well
1566
# (ie, reads continuous ranges.)
1567
raw_data = self._access.get_raw_records(
1568
[index_memo for key, index_memo in needed_records])
1570
for (key, index_memo), data in \
1571
izip(iter(needed_records), raw_data):
1572
content, digest = self._parse_record(key[-1], data)
1573
yield key, content, digest
1575
def _read_records_iter_raw(self, records):
1576
"""Read text records from data file and yield raw data.
1578
This unpacks enough of the text record to validate the id is
1579
as expected but thats all.
1581
Each item the iterator yields is (key, bytes, sha1_of_full_text).
1583
# setup an iterator of the external records:
1584
# uses readv so nice and fast we hope.
1586
# grab the disk data needed.
1587
needed_offsets = [index_memo for key, index_memo
1589
raw_records = self._access.get_raw_records(needed_offsets)
1591
for key, index_memo in records:
1592
data = raw_records.next()
1593
# validate the header (note that we can only use the suffix in
1594
# current knit records).
1595
df, rec = self._parse_record_header(key, data)
1597
yield key, data, rec[3]
1599
def _record_to_data(self, key, digest, lines, dense_lines=None):
1600
"""Convert key, digest, lines into a raw data block.
1602
:param key: The key of the record. Currently keys are always serialised
1603
using just the trailing component.
1604
:param dense_lines: The bytes of lines but in a denser form. For
1605
instance, if lines is a list of 1000 bytestrings each ending in \n,
1606
dense_lines may be a list with one line in it, containing all the
1607
1000's lines and their \n's. Using dense_lines if it is already
1608
known is a win because the string join to create bytes in this
1609
function spends less time resizing the final string.
1610
:return: (len, a StringIO instance with the raw data ready to read.)
1612
# Note: using a string copy here increases memory pressure with e.g.
1613
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
1614
# when doing the initial commit of a mozilla tree. RBC 20070921
1615
bytes = ''.join(chain(
1616
["version %s %d %s\n" % (key[-1],
1619
dense_lines or lines,
1620
["end %s\n" % key[-1]]))
1621
if type(bytes) != str:
1622
raise AssertionError(
1623
'data must be plain bytes was %s' % type(bytes))
1624
if lines and lines[-1][-1] != '\n':
1625
raise ValueError('corrupt lines value %r' % lines)
1626
compressed_bytes = tuned_gzip.bytes_to_gzip(bytes)
1627
return len(compressed_bytes), compressed_bytes
1629
def _split_header(self, line):
1632
raise KnitCorrupt(self,
1633
'unexpected number of elements in record header')
1637
"""See VersionedFiles.keys."""
1638
if 'evil' in debug.debug_flags:
1639
trace.mutter_callsite(2, "keys scales with size of history")
1640
sources = [self._index] + self._fallback_vfs
1642
for source in sources:
1643
result.update(source.keys())
1648
class _KndxIndex(object):
1649
"""Manages knit index files
1651
The index is kept in memory and read on startup, to enable
1652
fast lookups of revision information. The cursor of the index
1653
file is always pointing to the end, making it easy to append
1656
_cache is a cache for fast mapping from version id to a Index
1659
_history is a cache for fast mapping from indexes to version ids.
1661
The index data format is dictionary compressed when it comes to
1662
parent references; a index entry may only have parents that with a
1663
lover index number. As a result, the index is topological sorted.
1665
Duplicate entries may be written to the index for a single version id
1666
if this is done then the latter one completely replaces the former:
1667
this allows updates to correct version and parent information.
1668
Note that the two entries may share the delta, and that successive
1669
annotations and references MUST point to the first entry.
1671
The index file on disc contains a header, followed by one line per knit
1672
record. The same revision can be present in an index file more than once.
1673
The first occurrence gets assigned a sequence number starting from 0.
1675
The format of a single line is
1676
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1677
REVISION_ID is a utf8-encoded revision id
1678
FLAGS is a comma separated list of flags about the record. Values include
1679
no-eol, line-delta, fulltext.
1680
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1681
that the the compressed data starts at.
1682
LENGTH is the ascii representation of the length of the data file.
1683
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1685
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1686
revision id already in the knit that is a parent of REVISION_ID.
1687
The ' :' marker is the end of record marker.
1690
when a write is interrupted to the index file, it will result in a line
1691
that does not end in ' :'. If the ' :' is not present at the end of a line,
1692
or at the end of the file, then the record that is missing it will be
1693
ignored by the parser.
1695
When writing new records to the index file, the data is preceded by '\n'
1696
to ensure that records always start on new lines even if the last write was
1697
interrupted. As a result its normal for the last line in the index to be
1698
missing a trailing newline. One can be added with no harmful effects.
1700
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
1701
where prefix is e.g. the (fileid,) for .texts instances or () for
1702
constant-mapped things like .revisions, and the old state is
1703
tuple(history_vector, cache_dict). This is used to prevent having an
1704
ABI change with the C extension that reads .kndx files.
1707
HEADER = "# bzr knit index 8\n"
1709
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
1710
"""Create a _KndxIndex on transport using mapper."""
1711
self._transport = transport
1712
self._mapper = mapper
1713
self._get_scope = get_scope
1714
self._allow_writes = allow_writes
1715
self._is_locked = is_locked
1717
self.has_graph = True
1719
def add_records(self, records, random_id=False):
1720
"""Add multiple records to the index.
1722
:param records: a list of tuples:
1723
(key, options, access_memo, parents).
1724
:param random_id: If True the ids being added were randomly generated
1725
and no check for existence will be performed.
1728
for record in records:
1731
path = self._mapper.map(key) + '.kndx'
1732
path_keys = paths.setdefault(path, (prefix, []))
1733
path_keys[1].append(record)
1734
for path in sorted(paths):
1735
prefix, path_keys = paths[path]
1736
self._load_prefixes([prefix])
1738
orig_history = self._kndx_cache[prefix][1][:]
1739
orig_cache = self._kndx_cache[prefix][0].copy()
1742
for key, options, (_, pos, size), parents in path_keys:
1744
# kndx indices cannot be parentless.
1746
line = "\n%s %s %s %s %s :" % (
1747
key[-1], ','.join(options), pos, size,
1748
self._dictionary_compress(parents))
1749
if type(line) != str:
1750
raise AssertionError(
1751
'data must be utf8 was %s' % type(line))
1753
self._cache_key(key, options, pos, size, parents)
1754
if len(orig_history):
1755
self._transport.append_bytes(path, ''.join(lines))
1757
self._init_index(path, lines)
1759
# If any problems happen, restore the original values and re-raise
1760
self._kndx_cache[prefix] = (orig_cache, orig_history)
1763
def _cache_key(self, key, options, pos, size, parent_keys):
1764
"""Cache a version record in the history array and index cache.
1766
This is inlined into _load_data for performance. KEEP IN SYNC.
1767
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1771
version_id = key[-1]
1772
# last-element only for compatibilty with the C load_data.
1773
parents = tuple(parent[-1] for parent in parent_keys)
1774
for parent in parent_keys:
1775
if parent[:-1] != prefix:
1776
raise ValueError("mismatched prefixes for %r, %r" % (
1778
cache, history = self._kndx_cache[prefix]
1779
# only want the _history index to reference the 1st index entry
1781
if version_id not in cache:
1782
index = len(history)
1783
history.append(version_id)
1785
index = cache[version_id][5]
1786
cache[version_id] = (version_id,
1793
def check_header(self, fp):
1794
line = fp.readline()
1796
# An empty file can actually be treated as though the file doesn't
1798
raise errors.NoSuchFile(self)
1799
if line != self.HEADER:
1800
raise KnitHeaderError(badline=line, filename=self)
1802
def _check_read(self):
1803
if not self._is_locked():
1804
raise errors.ObjectNotLocked(self)
1805
if self._get_scope() != self._scope:
1808
def _check_write_ok(self):
1809
"""Assert if not writes are permitted."""
1810
if not self._is_locked():
1811
raise errors.ObjectNotLocked(self)
1812
if self._get_scope() != self._scope:
1814
if self._mode != 'w':
1815
raise errors.ReadOnlyObjectDirtiedError(self)
1817
def get_build_details(self, keys):
1818
"""Get the method, index_memo and compression parent for keys.
1820
Ghosts are omitted from the result.
1822
:param keys: An iterable of keys.
1823
:return: A dict of key:(index_memo, compression_parent, parents,
1826
opaque structure to pass to read_records to extract the raw
1829
Content that this record is built upon, may be None
1831
Logical parents of this node
1833
extra information about the content which needs to be passed to
1834
Factory.parse_record
1836
prefixes = self._partition_keys(keys)
1837
parent_map = self.get_parent_map(keys)
1840
if key not in parent_map:
1842
method = self.get_method(key)
1843
parents = parent_map[key]
1844
if method == 'fulltext':
1845
compression_parent = None
1847
compression_parent = parents[0]
1848
noeol = 'no-eol' in self.get_options(key)
1849
index_memo = self.get_position(key)
1850
result[key] = (index_memo, compression_parent,
1851
parents, (method, noeol))
1854
def get_method(self, key):
1855
"""Return compression method of specified key."""
1856
options = self.get_options(key)
1857
if 'fulltext' in options:
1859
elif 'line-delta' in options:
1862
raise errors.KnitIndexUnknownMethod(self, options)
1864
def get_options(self, key):
1865
"""Return a list representing options.
1869
prefix, suffix = self._split_key(key)
1870
self._load_prefixes([prefix])
1872
return self._kndx_cache[prefix][0][suffix][1]
1874
raise RevisionNotPresent(key, self)
1876
def get_parent_map(self, keys):
1877
"""Get a map of the parents of keys.
1879
:param keys: The keys to look up parents for.
1880
:return: A mapping from keys to parents. Absent keys are absent from
1883
# Parse what we need to up front, this potentially trades off I/O
1884
# locality (.kndx and .knit in the same block group for the same file
1885
# id) for less checking in inner loops.
1886
prefixes = set(key[:-1] for key in keys)
1887
self._load_prefixes(prefixes)
1892
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
1896
result[key] = tuple(prefix + (suffix,) for
1897
suffix in suffix_parents)
1900
def get_position(self, key):
1901
"""Return details needed to access the version.
1903
:return: a tuple (key, data position, size) to hand to the access
1904
logic to get the record.
1906
prefix, suffix = self._split_key(key)
1907
self._load_prefixes([prefix])
1908
entry = self._kndx_cache[prefix][0][suffix]
1909
return key, entry[2], entry[3]
1911
def _init_index(self, path, extra_lines=[]):
1912
"""Initialize an index."""
1914
sio.write(self.HEADER)
1915
sio.writelines(extra_lines)
1917
self._transport.put_file_non_atomic(path, sio,
1918
create_parent_dir=True)
1919
# self._create_parent_dir)
1920
# mode=self._file_mode,
1921
# dir_mode=self._dir_mode)
1924
"""Get all the keys in the collection.
1926
The keys are not ordered.
1929
# Identify all key prefixes.
1930
# XXX: A bit hacky, needs polish.
1931
if type(self._mapper) == ConstantMapper:
1935
for quoted_relpath in self._transport.iter_files_recursive():
1936
path, ext = os.path.splitext(quoted_relpath)
1938
prefixes = [self._mapper.unmap(path) for path in relpaths]
1939
self._load_prefixes(prefixes)
1940
for prefix in prefixes:
1941
for suffix in self._kndx_cache[prefix][1]:
1942
result.add(prefix + (suffix,))
1945
def _load_prefixes(self, prefixes):
1946
"""Load the indices for prefixes."""
1948
for prefix in prefixes:
1949
if prefix not in self._kndx_cache:
1950
# the load_data interface writes to these variables.
1953
self._filename = prefix
1955
path = self._mapper.map(prefix) + '.kndx'
1956
fp = self._transport.get(path)
1958
# _load_data may raise NoSuchFile if the target knit is
1960
_load_data(self, fp)
1963
self._kndx_cache[prefix] = (self._cache, self._history)
1968
self._kndx_cache[prefix] = ({}, [])
1969
if type(self._mapper) == ConstantMapper:
1970
# preserve behaviour for revisions.kndx etc.
1971
self._init_index(path)
1976
def _partition_keys(self, keys):
1977
"""Turn keys into a dict of prefix:suffix_list."""
1980
prefix_keys = result.setdefault(key[:-1], [])
1981
prefix_keys.append(key[-1])
1984
def _dictionary_compress(self, keys):
1985
"""Dictionary compress keys.
1987
:param keys: The keys to generate references to.
1988
:return: A string representation of keys. keys which are present are
1989
dictionary compressed, and others are emitted as fulltext with a
1995
prefix = keys[0][:-1]
1996
cache = self._kndx_cache[prefix][0]
1998
if key[:-1] != prefix:
1999
# kndx indices cannot refer across partitioned storage.
2000
raise ValueError("mismatched prefixes for %r" % keys)
2001
if key[-1] in cache:
2002
# -- inlined lookup() --
2003
result_list.append(str(cache[key[-1]][5]))
2004
# -- end lookup () --
2006
result_list.append('.' + key[-1])
2007
return ' '.join(result_list)
2009
def _reset_cache(self):
2010
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2011
# (cache_dict, history_vector) for parsed kndx files.
2012
self._kndx_cache = {}
2013
self._scope = self._get_scope()
2014
allow_writes = self._allow_writes()
2020
def _split_key(self, key):
2021
"""Split key into a prefix and suffix."""
2022
return key[:-1], key[-1]
2025
class _KnitGraphIndex(object):
2026
"""A KnitVersionedFiles index layered on GraphIndex."""
2028
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2030
"""Construct a KnitGraphIndex on a graph_index.
2032
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2033
:param is_locked: A callback to check whether the object should answer
2035
:param deltas: Allow delta-compressed records.
2036
:param parents: If True, record knits parents, if not do not record
2038
:param add_callback: If not None, allow additions to the index and call
2039
this callback with a list of added GraphIndex nodes:
2040
[(node, value, node_refs), ...]
2041
:param is_locked: A callback, returns True if the index is locked and
2044
self._add_callback = add_callback
2045
self._graph_index = graph_index
2046
self._deltas = deltas
2047
self._parents = parents
2048
if deltas and not parents:
2049
# XXX: TODO: Delta tree and parent graph should be conceptually
2051
raise KnitCorrupt(self, "Cannot do delta compression without "
2053
self.has_graph = parents
2054
self._is_locked = is_locked
2057
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2059
def add_records(self, records, random_id=False):
2060
"""Add multiple records to the index.
2062
This function does not insert data into the Immutable GraphIndex
2063
backing the KnitGraphIndex, instead it prepares data for insertion by
2064
the caller and checks that it is safe to insert then calls
2065
self._add_callback with the prepared GraphIndex nodes.
2067
:param records: a list of tuples:
2068
(key, options, access_memo, parents).
2069
:param random_id: If True the ids being added were randomly generated
2070
and no check for existence will be performed.
2072
if not self._add_callback:
2073
raise errors.ReadOnlyError(self)
2074
# we hope there are no repositories with inconsistent parentage
2078
for (key, options, access_memo, parents) in records:
2080
parents = tuple(parents)
2081
index, pos, size = access_memo
2082
if 'no-eol' in options:
2086
value += "%d %d" % (pos, size)
2087
if not self._deltas:
2088
if 'line-delta' in options:
2089
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2092
if 'line-delta' in options:
2093
node_refs = (parents, (parents[0],))
2095
node_refs = (parents, ())
2097
node_refs = (parents, )
2100
raise KnitCorrupt(self, "attempt to add node with parents "
2101
"in parentless index.")
2103
keys[key] = (value, node_refs)
2106
present_nodes = self._get_entries(keys)
2107
for (index, key, value, node_refs) in present_nodes:
2108
if (value[0] != keys[key][0][0] or
2109
node_refs != keys[key][1]):
2110
raise KnitCorrupt(self, "inconsistent details in add_records"
2111
": %s %s" % ((value, node_refs), keys[key]))
2115
for key, (value, node_refs) in keys.iteritems():
2116
result.append((key, value, node_refs))
2118
for key, (value, node_refs) in keys.iteritems():
2119
result.append((key, value))
2120
self._add_callback(result)
2122
def _check_read(self):
2123
"""raise if reads are not permitted."""
2124
if not self._is_locked():
2125
raise errors.ObjectNotLocked(self)
2127
def _check_write_ok(self):
2128
"""Assert if writes are not permitted."""
2129
if not self._is_locked():
2130
raise errors.ObjectNotLocked(self)
2132
def _compression_parent(self, an_entry):
2133
# return the key that an_entry is compressed against, or None
2134
# Grab the second parent list (as deltas implies parents currently)
2135
compression_parents = an_entry[3][1]
2136
if not compression_parents:
2138
if len(compression_parents) != 1:
2139
raise AssertionError(
2140
"Too many compression parents: %r" % compression_parents)
2141
return compression_parents[0]
2143
def get_build_details(self, keys):
2144
"""Get the method, index_memo and compression parent for version_ids.
2146
Ghosts are omitted from the result.
2148
:param keys: An iterable of keys.
2149
:return: A dict of key:
2150
(index_memo, compression_parent, parents, record_details).
2152
opaque structure to pass to read_records to extract the raw
2155
Content that this record is built upon, may be None
2157
Logical parents of this node
2159
extra information about the content which needs to be passed to
2160
Factory.parse_record
2164
entries = self._get_entries(keys, False)
2165
for entry in entries:
2167
if not self._parents:
2170
parents = entry[3][0]
2171
if not self._deltas:
2172
compression_parent_key = None
2174
compression_parent_key = self._compression_parent(entry)
2175
noeol = (entry[2][0] == 'N')
2176
if compression_parent_key:
2177
method = 'line-delta'
2180
result[key] = (self._node_to_position(entry),
2181
compression_parent_key, parents,
2185
def _get_entries(self, keys, check_present=False):
2186
"""Get the entries for keys.
2188
:param keys: An iterable of index key tuples.
2193
for node in self._graph_index.iter_entries(keys):
2195
found_keys.add(node[1])
2197
# adapt parentless index to the rest of the code.
2198
for node in self._graph_index.iter_entries(keys):
2199
yield node[0], node[1], node[2], ()
2200
found_keys.add(node[1])
2202
missing_keys = keys.difference(found_keys)
2204
raise RevisionNotPresent(missing_keys.pop(), self)
2206
def get_method(self, key):
2207
"""Return compression method of specified key."""
2208
return self._get_method(self._get_node(key))
2210
def _get_method(self, node):
2211
if not self._deltas:
2213
if self._compression_parent(node):
2218
def _get_node(self, key):
2220
return list(self._get_entries([key]))[0]
2222
raise RevisionNotPresent(key, self)
2224
def get_options(self, key):
2225
"""Return a list representing options.
2229
node = self._get_node(key)
2230
options = [self._get_method(node)]
2231
if node[2][0] == 'N':
2232
options.append('no-eol')
2235
def get_parent_map(self, keys):
2236
"""Get a map of the parents of keys.
2238
:param keys: The keys to look up parents for.
2239
:return: A mapping from keys to parents. Absent keys are absent from
2243
nodes = self._get_entries(keys)
2247
result[node[1]] = node[3][0]
2250
result[node[1]] = None
2253
def get_position(self, key):
2254
"""Return details needed to access the version.
2256
:return: a tuple (index, data position, size) to hand to the access
2257
logic to get the record.
2259
node = self._get_node(key)
2260
return self._node_to_position(node)
2263
"""Get all the keys in the collection.
2265
The keys are not ordered.
2268
return [node[1] for node in self._graph_index.iter_all_entries()]
2270
def _node_to_position(self, node):
2271
"""Convert an index value to position details."""
2272
bits = node[2][1:].split(' ')
2273
return node[0], int(bits[0]), int(bits[1])
2276
class _KnitKeyAccess(object):
2277
"""Access to records in .knit files."""
2279
def __init__(self, transport, mapper):
2280
"""Create a _KnitKeyAccess with transport and mapper.
2282
:param transport: The transport the access object is rooted at.
2283
:param mapper: The mapper used to map keys to .knit files.
2285
self._transport = transport
2286
self._mapper = mapper
2288
def add_raw_records(self, key_sizes, raw_data):
2289
"""Add raw knit bytes to a storage area.
2291
The data is spooled to the container writer in one bytes-record per
2294
:param sizes: An iterable of tuples containing the key and size of each
2296
:param raw_data: A bytestring containing the data.
2297
:return: A list of memos to retrieve the record later. Each memo is an
2298
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
2299
length), where the key is the record key.
2301
if type(raw_data) != str:
2302
raise AssertionError(
2303
'data must be plain bytes was %s' % type(raw_data))
2306
# TODO: This can be tuned for writing to sftp and other servers where
2307
# append() is relatively expensive by grouping the writes to each key
2309
for key, size in key_sizes:
2310
path = self._mapper.map(key)
2312
base = self._transport.append_bytes(path + '.knit',
2313
raw_data[offset:offset+size])
2314
except errors.NoSuchFile:
2315
self._transport.mkdir(osutils.dirname(path))
2316
base = self._transport.append_bytes(path + '.knit',
2317
raw_data[offset:offset+size])
2321
result.append((key, base, size))
2324
def get_raw_records(self, memos_for_retrieval):
2325
"""Get the raw bytes for a records.
2327
:param memos_for_retrieval: An iterable containing the access memo for
2328
retrieving the bytes.
2329
:return: An iterator over the bytes of the records.
2331
# first pass, group into same-index request to minimise readv's issued.
2333
current_prefix = None
2334
for (key, offset, length) in memos_for_retrieval:
2335
if current_prefix == key[:-1]:
2336
current_list.append((offset, length))
2338
if current_prefix is not None:
2339
request_lists.append((current_prefix, current_list))
2340
current_prefix = key[:-1]
2341
current_list = [(offset, length)]
2342
# handle the last entry
2343
if current_prefix is not None:
2344
request_lists.append((current_prefix, current_list))
2345
for prefix, read_vector in request_lists:
2346
path = self._mapper.map(prefix) + '.knit'
2347
for pos, data in self._transport.readv(path, read_vector):
2351
class _DirectPackAccess(object):
2352
"""Access to data in one or more packs with less translation."""
2354
def __init__(self, index_to_packs):
2355
"""Create a _DirectPackAccess object.
2357
:param index_to_packs: A dict mapping index objects to the transport
2358
and file names for obtaining data.
2360
self._container_writer = None
2361
self._write_index = None
2362
self._indices = index_to_packs
2364
def add_raw_records(self, key_sizes, raw_data):
2365
"""Add raw knit bytes to a storage area.
2367
The data is spooled to the container writer in one bytes-record per
2370
:param sizes: An iterable of tuples containing the key and size of each
2372
:param raw_data: A bytestring containing the data.
2373
:return: A list of memos to retrieve the record later. Each memo is an
2374
opaque index memo. For _DirectPackAccess the memo is (index, pos,
2375
length), where the index field is the write_index object supplied
2376
to the PackAccess object.
2378
if type(raw_data) != str:
2379
raise AssertionError(
2380
'data must be plain bytes was %s' % type(raw_data))
2383
for key, size in key_sizes:
2384
p_offset, p_length = self._container_writer.add_bytes_record(
2385
raw_data[offset:offset+size], [])
2387
result.append((self._write_index, p_offset, p_length))
2390
def get_raw_records(self, memos_for_retrieval):
2391
"""Get the raw bytes for a records.
2393
:param memos_for_retrieval: An iterable containing the (index, pos,
2394
length) memo for retrieving the bytes. The Pack access method
2395
looks up the pack to use for a given record in its index_to_pack
2397
:return: An iterator over the bytes of the records.
2399
# first pass, group into same-index requests
2401
current_index = None
2402
for (index, offset, length) in memos_for_retrieval:
2403
if current_index == index:
2404
current_list.append((offset, length))
2406
if current_index is not None:
2407
request_lists.append((current_index, current_list))
2408
current_index = index
2409
current_list = [(offset, length)]
2410
# handle the last entry
2411
if current_index is not None:
2412
request_lists.append((current_index, current_list))
2413
for index, offsets in request_lists:
2414
transport, path = self._indices[index]
2415
reader = pack.make_readv_reader(transport, path, offsets)
2416
for names, read_func in reader.iter_records():
2417
yield read_func(None)
2419
def set_writer(self, writer, index, transport_packname):
2420
"""Set a writer to use for adding data."""
2421
if index is not None:
2422
self._indices[index] = transport_packname
2423
self._container_writer = writer
2424
self._write_index = index
2427
# Deprecated, use PatienceSequenceMatcher instead
2428
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2431
def annotate_knit(knit, revision_id):
2432
"""Annotate a knit with no cached annotations.
2434
This implementation is for knits with no cached annotations.
2435
It will work for knits with cached annotations, but this is not
2438
annotator = _KnitAnnotator(knit)
2439
return iter(annotator.annotate(revision_id))
2442
class _KnitAnnotator(object):
2443
"""Build up the annotations for a text."""
2445
def __init__(self, knit):
2448
# Content objects, differs from fulltexts because of how final newlines
2449
# are treated by knits. the content objects here will always have a
2451
self._fulltext_contents = {}
2453
# Annotated lines of specific revisions
2454
self._annotated_lines = {}
2456
# Track the raw data for nodes that we could not process yet.
2457
# This maps the revision_id of the base to a list of children that will
2458
# annotated from it.
2459
self._pending_children = {}
2461
# Nodes which cannot be extracted
2462
self._ghosts = set()
2464
# Track how many children this node has, so we know if we need to keep
2466
self._annotate_children = {}
2467
self._compression_children = {}
2469
self._all_build_details = {}
2470
# The children => parent revision_id graph
2471
self._revision_id_graph = {}
2473
self._heads_provider = None
2475
self._nodes_to_keep_annotations = set()
2476
self._generations_until_keep = 100
2478
def set_generations_until_keep(self, value):
2479
"""Set the number of generations before caching a node.
2481
Setting this to -1 will cache every merge node, setting this higher
2482
will cache fewer nodes.
2484
self._generations_until_keep = value
2486
def _add_fulltext_content(self, revision_id, content_obj):
2487
self._fulltext_contents[revision_id] = content_obj
2488
# TODO: jam 20080305 It might be good to check the sha1digest here
2489
return content_obj.text()
2491
def _check_parents(self, child, nodes_to_annotate):
2492
"""Check if all parents have been processed.
2494
:param child: A tuple of (rev_id, parents, raw_content)
2495
:param nodes_to_annotate: If child is ready, add it to
2496
nodes_to_annotate, otherwise put it back in self._pending_children
2498
for parent_id in child[1]:
2499
if (parent_id not in self._annotated_lines):
2500
# This parent is present, but another parent is missing
2501
self._pending_children.setdefault(parent_id,
2505
# This one is ready to be processed
2506
nodes_to_annotate.append(child)
2508
def _add_annotation(self, revision_id, fulltext, parent_ids,
2509
left_matching_blocks=None):
2510
"""Add an annotation entry.
2512
All parents should already have been annotated.
2513
:return: A list of children that now have their parents satisfied.
2515
a = self._annotated_lines
2516
annotated_parent_lines = [a[p] for p in parent_ids]
2517
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2518
fulltext, revision_id, left_matching_blocks,
2519
heads_provider=self._get_heads_provider()))
2520
self._annotated_lines[revision_id] = annotated_lines
2521
for p in parent_ids:
2522
ann_children = self._annotate_children[p]
2523
ann_children.remove(revision_id)
2524
if (not ann_children
2525
and p not in self._nodes_to_keep_annotations):
2526
del self._annotated_lines[p]
2527
del self._all_build_details[p]
2528
if p in self._fulltext_contents:
2529
del self._fulltext_contents[p]
2530
# Now that we've added this one, see if there are any pending
2531
# deltas to be done, certainly this parent is finished
2532
nodes_to_annotate = []
2533
for child in self._pending_children.pop(revision_id, []):
2534
self._check_parents(child, nodes_to_annotate)
2535
return nodes_to_annotate
2537
def _get_build_graph(self, key):
2538
"""Get the graphs for building texts and annotations.
2540
The data you need for creating a full text may be different than the
2541
data you need to annotate that text. (At a minimum, you need both
2542
parents to create an annotation, but only need 1 parent to generate the
2545
:return: A list of (key, index_memo) records, suitable for
2546
passing to read_records_iter to start reading in the raw data fro/
2549
if key in self._annotated_lines:
2552
pending = set([key])
2557
# get all pending nodes
2559
this_iteration = pending
2560
build_details = self._knit._index.get_build_details(this_iteration)
2561
self._all_build_details.update(build_details)
2562
# new_nodes = self._knit._index._get_entries(this_iteration)
2564
for key, details in build_details.iteritems():
2565
(index_memo, compression_parent, parents,
2566
record_details) = details
2567
self._revision_id_graph[key] = parents
2568
records.append((key, index_memo))
2569
# Do we actually need to check _annotated_lines?
2570
pending.update(p for p in parents
2571
if p not in self._all_build_details)
2572
if compression_parent:
2573
self._compression_children.setdefault(compression_parent,
2576
for parent in parents:
2577
self._annotate_children.setdefault(parent,
2579
num_gens = generation - kept_generation
2580
if ((num_gens >= self._generations_until_keep)
2581
and len(parents) > 1):
2582
kept_generation = generation
2583
self._nodes_to_keep_annotations.add(key)
2585
missing_versions = this_iteration.difference(build_details.keys())
2586
self._ghosts.update(missing_versions)
2587
for missing_version in missing_versions:
2588
# add a key, no parents
2589
self._revision_id_graph[missing_version] = ()
2590
pending.discard(missing_version) # don't look for it
2591
if self._ghosts.intersection(self._compression_children):
2593
"We cannot have nodes which have a ghost compression parent:\n"
2595
"compression children: %r"
2596
% (self._ghosts, self._compression_children))
2597
# Cleanout anything that depends on a ghost so that we don't wait for
2598
# the ghost to show up
2599
for node in self._ghosts:
2600
if node in self._annotate_children:
2601
# We won't be building this node
2602
del self._annotate_children[node]
2603
# Generally we will want to read the records in reverse order, because
2604
# we find the parent nodes after the children
2608
def _annotate_records(self, records):
2609
"""Build the annotations for the listed records."""
2610
# We iterate in the order read, rather than a strict order requested
2611
# However, process what we can, and put off to the side things that
2612
# still need parents, cleaning them up when those parents are
2614
for (rev_id, record,
2615
digest) in self._knit._read_records_iter(records):
2616
if rev_id in self._annotated_lines:
2618
parent_ids = self._revision_id_graph[rev_id]
2619
parent_ids = [p for p in parent_ids if p not in self._ghosts]
2620
details = self._all_build_details[rev_id]
2621
(index_memo, compression_parent, parents,
2622
record_details) = details
2623
nodes_to_annotate = []
2624
# TODO: Remove the punning between compression parents, and
2625
# parent_ids, we should be able to do this without assuming
2627
if len(parent_ids) == 0:
2628
# There are no parents for this node, so just add it
2629
# TODO: This probably needs to be decoupled
2630
fulltext_content, delta = self._knit._factory.parse_record(
2631
rev_id, record, record_details, None)
2632
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
2633
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
2634
parent_ids, left_matching_blocks=None))
2636
child = (rev_id, parent_ids, record)
2637
# Check if all the parents are present
2638
self._check_parents(child, nodes_to_annotate)
2639
while nodes_to_annotate:
2640
# Should we use a queue here instead of a stack?
2641
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
2642
(index_memo, compression_parent, parents,
2643
record_details) = self._all_build_details[rev_id]
2644
if compression_parent is not None:
2645
comp_children = self._compression_children[compression_parent]
2646
if rev_id not in comp_children:
2647
raise AssertionError("%r not in compression children %r"
2648
% (rev_id, comp_children))
2649
# If there is only 1 child, it is safe to reuse this
2651
reuse_content = (len(comp_children) == 1
2652
and compression_parent not in
2653
self._nodes_to_keep_annotations)
2655
# Remove it from the cache since it will be changing
2656
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
2657
# Make sure to copy the fulltext since it might be
2659
parent_fulltext = list(parent_fulltext_content.text())
2661
parent_fulltext_content = self._fulltext_contents[compression_parent]
2662
parent_fulltext = parent_fulltext_content.text()
2663
comp_children.remove(rev_id)
2664
fulltext_content, delta = self._knit._factory.parse_record(
2665
rev_id, record, record_details,
2666
parent_fulltext_content,
2667
copy_base_content=(not reuse_content))
2668
fulltext = self._add_fulltext_content(rev_id,
2670
blocks = KnitContent.get_line_delta_blocks(delta,
2671
parent_fulltext, fulltext)
2673
fulltext_content = self._knit._factory.parse_fulltext(
2675
fulltext = self._add_fulltext_content(rev_id,
2678
nodes_to_annotate.extend(
2679
self._add_annotation(rev_id, fulltext, parent_ids,
2680
left_matching_blocks=blocks))
2682
def _get_heads_provider(self):
2683
"""Create a heads provider for resolving ancestry issues."""
2684
if self._heads_provider is not None:
2685
return self._heads_provider
2686
parent_provider = _mod_graph.DictParentsProvider(
2687
self._revision_id_graph)
2688
graph_obj = _mod_graph.Graph(parent_provider)
2689
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
2690
self._heads_provider = head_cache
2693
def annotate(self, key):
2694
"""Return the annotated fulltext at the given key.
2696
:param key: The key to annotate.
2698
if True or len(self._knit._fallback_vfs) > 0:
2699
# stacked knits can't use the fast path at present.
2700
return self._simple_annotate(key)
2701
records = self._get_build_graph(key)
2702
if key in self._ghosts:
2703
raise errors.RevisionNotPresent(key, self._knit)
2704
self._annotate_records(records)
2705
return self._annotated_lines[key]
2707
def _simple_annotate(self, key):
2708
"""Return annotated fulltext, rediffing from the full texts.
2710
This is slow but makes no assumptions about the repository
2711
being able to produce line deltas.
2713
# TODO: this code generates a parent maps of present ancestors; it
2714
# could be split out into a separate method, and probably should use
2715
# iter_ancestry instead. -- mbp and robertc 20080704
2716
graph = _mod_graph.Graph(self._knit)
2717
head_cache = _mod_graph.FrozenHeadsCache(graph)
2718
search = graph._make_breadth_first_searcher([key])
2722
present, ghosts = search.next_with_ghosts()
2723
except StopIteration:
2725
keys.update(present)
2726
parent_map = self._knit.get_parent_map(keys)
2728
reannotate = annotate.reannotate
2729
for record in self._knit.get_record_stream(keys, 'topological', True):
2731
fulltext = split_lines(record.get_bytes_as('fulltext'))
2732
parents = parent_map[key]
2733
if parents is not None:
2734
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
2737
parent_cache[key] = list(
2738
reannotate(parent_lines, fulltext, key, None, head_cache))
2740
return parent_cache[key]
2742
raise errors.RevisionNotPresent(key, self._knit)
2746
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2748
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
