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 their index must be
709
def __init__(self, index, data_access, max_delta_chain=200,
711
"""Create a KnitVersionedFiles with index and data_access.
713
:param index: The index for the knit data.
714
:param data_access: The access object to store and retrieve knit
716
:param max_delta_chain: The maximum number of deltas to permit during
717
insertion. Set to 0 to prohibit the use of deltas.
718
:param annotated: Set to True to cause annotations to be calculated and
719
stored during insertion.
722
self._access = data_access
723
self._max_delta_chain = max_delta_chain
725
self._factory = KnitAnnotateFactory()
727
self._factory = KnitPlainFactory()
728
self._fallback_vfs = []
730
def add_fallback_versioned_files(self, a_versioned_files):
731
"""Add a source of texts for texts not present in this knit.
733
:param a_versioned_files: A VersionedFiles object.
735
self._fallback_vfs.append(a_versioned_files)
737
def add_lines(self, key, parents, lines, parent_texts=None,
738
left_matching_blocks=None, nostore_sha=None, random_id=False,
740
"""See VersionedFiles.add_lines()."""
741
self._index._check_write_ok()
742
self._check_add(key, lines, random_id, check_content)
744
# The caller might pass None if there is no graph data, but kndx
745
# indexes can't directly store that, so we give them
746
# an empty tuple instead.
748
return self._add(key, lines, parents,
749
parent_texts, left_matching_blocks, nostore_sha, random_id)
751
def _add(self, key, lines, parents, parent_texts,
752
left_matching_blocks, nostore_sha, random_id):
753
"""Add a set of lines on top of version specified by parents.
755
Any versions not present will be converted into ghosts.
757
# first thing, if the content is something we don't need to store, find
759
line_bytes = ''.join(lines)
760
digest = sha_string(line_bytes)
761
if nostore_sha == digest:
762
raise errors.ExistingContent
765
if parent_texts is None:
767
# Do a single query to ascertain parent presence.
768
present_parent_map = self.get_parent_map(parents)
769
for parent in parents:
770
if parent in present_parent_map:
771
present_parents.append(parent)
773
# Currently we can only compress against the left most present parent.
774
if (len(present_parents) == 0 or
775
present_parents[0] != parents[0]):
778
# To speed the extract of texts the delta chain is limited
779
# to a fixed number of deltas. This should minimize both
780
# I/O and the time spend applying deltas.
781
delta = self._check_should_delta(present_parents[0])
783
text_length = len(line_bytes)
786
if lines[-1][-1] != '\n':
787
# copy the contents of lines.
789
options.append('no-eol')
790
lines[-1] = lines[-1] + '\n'
794
if type(element) != str:
795
raise TypeError("key contains non-strings: %r" % (key,))
796
# Knit hunks are still last-element only
798
content = self._factory.make(lines, version_id)
799
if 'no-eol' in options:
800
# Hint to the content object that its text() call should strip the
802
content._should_strip_eol = True
803
if delta or (self._factory.annotated and len(present_parents) > 0):
804
# Merge annotations from parent texts if needed.
805
delta_hunks = self._merge_annotations(content, present_parents,
806
parent_texts, delta, self._factory.annotated,
807
left_matching_blocks)
810
options.append('line-delta')
811
store_lines = self._factory.lower_line_delta(delta_hunks)
812
size, bytes = self._record_to_data(key, digest,
815
options.append('fulltext')
816
# isinstance is slower and we have no hierarchy.
817
if self._factory.__class__ == KnitPlainFactory:
818
# Use the already joined bytes saving iteration time in
820
size, bytes = self._record_to_data(key, digest,
823
# get mixed annotation + content and feed it into the
825
store_lines = self._factory.lower_fulltext(content)
826
size, bytes = self._record_to_data(key, digest,
829
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
830
self._index.add_records(
831
((key, options, access_memo, parents),),
833
return digest, text_length, content
835
def annotate(self, key):
836
"""See VersionedFiles.annotate."""
837
return self._factory.annotate(self, key)
839
def check(self, progress_bar=None):
840
"""See VersionedFiles.check()."""
841
# This doesn't actually test extraction of everything, but that will
842
# impact 'bzr check' substantially, and needs to be integrated with
843
# care. However, it does check for the obvious problem of a delta with
845
keys = self._index.keys()
846
parent_map = self.get_parent_map(keys)
848
if self._index.get_method(key) != 'fulltext':
849
compression_parent = parent_map[key][0]
850
if compression_parent not in parent_map:
851
raise errors.KnitCorrupt(self,
852
"Missing basis parent %s for %s" % (
853
compression_parent, key))
854
for fallback_vfs in self._fallback_vfs:
857
def _check_add(self, key, lines, random_id, check_content):
858
"""check that version_id and lines are safe to add."""
860
if contains_whitespace(version_id):
861
raise InvalidRevisionId(version_id, self)
862
self.check_not_reserved_id(version_id)
863
# TODO: If random_id==False and the key is already present, we should
864
# probably check that the existing content is identical to what is
865
# being inserted, and otherwise raise an exception. This would make
866
# the bundle code simpler.
868
self._check_lines_not_unicode(lines)
869
self._check_lines_are_lines(lines)
871
def _check_header(self, key, line):
872
rec = self._split_header(line)
873
self._check_header_version(rec, key[-1])
876
def _check_header_version(self, rec, version_id):
877
"""Checks the header version on original format knit records.
879
These have the last component of the key embedded in the record.
881
if rec[1] != version_id:
882
raise KnitCorrupt(self,
883
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
885
def _check_should_delta(self, parent):
886
"""Iterate back through the parent listing, looking for a fulltext.
888
This is used when we want to decide whether to add a delta or a new
889
fulltext. It searches for _max_delta_chain parents. When it finds a
890
fulltext parent, it sees if the total size of the deltas leading up to
891
it is large enough to indicate that we want a new full text anyway.
893
Return True if we should create a new delta, False if we should use a
898
for count in xrange(self._max_delta_chain):
899
# XXX: Collapse these two queries:
901
# Note that this only looks in the index of this particular
902
# KnitVersionedFiles, not in the fallbacks. This ensures that
903
# we won't store a delta spanning physical repository
905
method = self._index.get_method(parent)
906
except RevisionNotPresent:
907
# Some basis is not locally present: always delta
909
index, pos, size = self._index.get_position(parent)
910
if method == 'fulltext':
914
# We don't explicitly check for presence because this is in an
915
# inner loop, and if it's missing it'll fail anyhow.
916
# TODO: This should be asking for compression parent, not graph
918
parent = self._index.get_parent_map([parent])[parent][0]
920
# We couldn't find a fulltext, so we must create a new one
922
# Simple heuristic - if the total I/O wold be greater as a delta than
923
# the originally installed fulltext, we create a new fulltext.
924
return fulltext_size > delta_size
926
def _build_details_to_components(self, build_details):
927
"""Convert a build_details tuple to a position tuple."""
928
# record_details, access_memo, compression_parent
929
return build_details[3], build_details[0], build_details[1]
931
def _get_components_positions(self, keys, allow_missing=False):
932
"""Produce a map of position data for the components of keys.
934
This data is intended to be used for retrieving the knit records.
936
A dict of key to (record_details, index_memo, next, parents) is
938
method is the way referenced data should be applied.
939
index_memo is the handle to pass to the data access to actually get the
941
next is the build-parent of the version, or None for fulltexts.
942
parents is the version_ids of the parents of this version
944
:param allow_missing: If True do not raise an error on a missing component,
948
pending_components = keys
949
while pending_components:
950
build_details = self._index.get_build_details(pending_components)
951
current_components = set(pending_components)
952
pending_components = set()
953
for key, details in build_details.iteritems():
954
(index_memo, compression_parent, parents,
955
record_details) = details
956
method = record_details[0]
957
if compression_parent is not None:
958
pending_components.add(compression_parent)
959
component_data[key] = self._build_details_to_components(details)
960
missing = current_components.difference(build_details)
961
if missing and not allow_missing:
962
raise errors.RevisionNotPresent(missing.pop(), self)
963
return component_data
965
def _get_content(self, key, parent_texts={}):
966
"""Returns a content object that makes up the specified
968
cached_version = parent_texts.get(key, None)
969
if cached_version is not None:
970
# Ensure the cache dict is valid.
971
if not self.get_parent_map([key]):
972
raise RevisionNotPresent(key, self)
973
return cached_version
974
text_map, contents_map = self._get_content_maps([key])
975
return contents_map[key]
977
def _get_content_maps(self, keys, nonlocal_keys=None):
978
"""Produce maps of text and KnitContents
980
:param keys: The keys to produce content maps for.
981
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
982
which are known to not be in this knit, but rather in one of the
984
:return: (text_map, content_map) where text_map contains the texts for
985
the requested versions and content_map contains the KnitContents.
987
# FUTURE: This function could be improved for the 'extract many' case
988
# by tracking each component and only doing the copy when the number of
989
# children than need to apply delta's to it is > 1 or it is part of the
992
multiple_versions = len(keys) != 1
993
record_map = self._get_record_map(keys, allow_missing=True)
998
if nonlocal_keys is None:
999
nonlocal_keys = set()
1001
nonlocal_keys = frozenset(nonlocal_keys)
1002
missing_keys = set(nonlocal_keys)
1003
for source in self._fallback_vfs:
1004
if not missing_keys:
1006
for record in source.get_record_stream(missing_keys,
1008
if record.storage_kind == 'absent':
1010
missing_keys.remove(record.key)
1011
lines = split_lines(record.get_bytes_as('fulltext'))
1012
text_map[record.key] = lines
1013
content_map[record.key] = PlainKnitContent(lines, record.key)
1014
if record.key in keys:
1015
final_content[record.key] = content_map[record.key]
1017
if key in nonlocal_keys:
1022
while cursor is not None:
1024
record, record_details, digest, next = record_map[cursor]
1026
raise RevisionNotPresent(cursor, self)
1027
components.append((cursor, record, record_details, digest))
1029
if cursor in content_map:
1030
# no need to plan further back
1031
components.append((cursor, None, None, None))
1035
for (component_id, record, record_details,
1036
digest) in reversed(components):
1037
if component_id in content_map:
1038
content = content_map[component_id]
1040
content, delta = self._factory.parse_record(key[-1],
1041
record, record_details, content,
1042
copy_base_content=multiple_versions)
1043
if multiple_versions:
1044
content_map[component_id] = content
1046
final_content[key] = content
1048
# digest here is the digest from the last applied component.
1049
text = content.text()
1050
actual_sha = sha_strings(text)
1051
if actual_sha != digest:
1052
raise KnitCorrupt(self,
1054
'\n of reconstructed text does not match'
1056
'\n for version %s' %
1057
(actual_sha, digest, key))
1058
text_map[key] = text
1059
return text_map, final_content
1061
def get_parent_map(self, keys):
1062
"""Get a map of the graph parents of keys.
1064
:param keys: The keys to look up parents for.
1065
:return: A mapping from keys to parents. Absent keys are absent from
1068
return self._get_parent_map_with_sources(keys)[0]
1070
def _get_parent_map_with_sources(self, keys):
1071
"""Get a map of the parents of keys.
1073
:param keys: The keys to look up parents for.
1074
:return: A tuple. The first element is a mapping from keys to parents.
1075
Absent keys are absent from the mapping. The second element is a
1076
list with the locations each key was found in. The first element
1077
is the in-this-knit parents, the second the first fallback source,
1081
sources = [self._index] + self._fallback_vfs
1084
for source in sources:
1087
new_result = source.get_parent_map(missing)
1088
source_results.append(new_result)
1089
result.update(new_result)
1090
missing.difference_update(set(new_result))
1091
return result, source_results
1093
def _get_record_map(self, keys, allow_missing=False):
1094
"""Produce a dictionary of knit records.
1096
:return: {key:(record, record_details, digest, next)}
1098
data returned from read_records
1100
opaque information to pass to parse_record
1102
SHA1 digest of the full text after all steps are done
1104
build-parent of the version, i.e. the leftmost ancestor.
1105
Will be None if the record is not a delta.
1106
:param keys: The keys to build a map for
1107
:param allow_missing: If some records are missing, rather than
1108
error, just return the data that could be generated.
1110
position_map = self._get_components_positions(keys,
1111
allow_missing=allow_missing)
1112
# key = component_id, r = record_details, i_m = index_memo, n = next
1113
records = [(key, i_m) for key, (r, i_m, n)
1114
in position_map.iteritems()]
1116
for key, record, digest in \
1117
self._read_records_iter(records):
1118
(record_details, index_memo, next) = position_map[key]
1119
record_map[key] = record, record_details, digest, next
1122
def get_record_stream(self, keys, ordering, include_delta_closure):
1123
"""Get a stream of records for keys.
1125
:param keys: The keys to include.
1126
:param ordering: Either 'unordered' or 'topological'. A topologically
1127
sorted stream has compression parents strictly before their
1129
:param include_delta_closure: If True then the closure across any
1130
compression parents will be included (in the opaque data).
1131
:return: An iterator of ContentFactory objects, each of which is only
1132
valid until the iterator is advanced.
1134
# keys might be a generator
1138
if not self._index.has_graph:
1139
# Cannot topological order when no graph has been stored.
1140
ordering = 'unordered'
1141
if include_delta_closure:
1142
positions = self._get_components_positions(keys, allow_missing=True)
1144
build_details = self._index.get_build_details(keys)
1146
# (record_details, access_memo, compression_parent_key)
1147
positions = dict((key, self._build_details_to_components(details))
1148
for key, details in build_details.iteritems())
1149
absent_keys = keys.difference(set(positions))
1150
# There may be more absent keys : if we're missing the basis component
1151
# and are trying to include the delta closure.
1152
if include_delta_closure:
1153
needed_from_fallback = set()
1154
# Build up reconstructable_keys dict. key:True in this dict means
1155
# the key can be reconstructed.
1156
reconstructable_keys = {}
1160
chain = [key, positions[key][2]]
1162
needed_from_fallback.add(key)
1165
while chain[-1] is not None:
1166
if chain[-1] in reconstructable_keys:
1167
result = reconstructable_keys[chain[-1]]
1171
chain.append(positions[chain[-1]][2])
1173
# missing basis component
1174
needed_from_fallback.add(chain[-1])
1177
for chain_key in chain[:-1]:
1178
reconstructable_keys[chain_key] = result
1180
needed_from_fallback.add(key)
1181
# Double index lookups here : need a unified api ?
1182
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1183
if ordering == 'topological':
1184
# Global topological sort
1185
present_keys = tsort.topo_sort(global_map)
1186
# Now group by source:
1188
current_source = None
1189
for key in present_keys:
1190
for parent_map in parent_maps:
1191
if key in parent_map:
1192
key_source = parent_map
1194
if current_source is not key_source:
1195
source_keys.append((key_source, []))
1196
current_source = key_source
1197
source_keys[-1][1].append(key)
1199
# Just group by source; remote sources first.
1202
for parent_map in reversed(parent_maps):
1203
source_keys.append((parent_map, []))
1204
for key in parent_map:
1205
present_keys.append(key)
1206
source_keys[-1][1].append(key)
1207
absent_keys = keys - set(global_map)
1208
for key in absent_keys:
1209
yield AbsentContentFactory(key)
1210
# restrict our view to the keys we can answer.
1211
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1212
# XXX: At that point we need to consider the impact of double reads by
1213
# utilising components multiple times.
1214
if include_delta_closure:
1215
# XXX: get_content_maps performs its own index queries; allow state
1217
text_map, _ = self._get_content_maps(present_keys,
1218
needed_from_fallback - absent_keys)
1219
for key in present_keys:
1220
yield FulltextContentFactory(key, global_map[key], None,
1221
''.join(text_map[key]))
1223
for source, keys in source_keys:
1224
if source is parent_maps[0]:
1225
# this KnitVersionedFiles
1226
records = [(key, positions[key][1]) for key in keys]
1227
for key, raw_data, sha1 in self._read_records_iter_raw(records):
1228
(record_details, index_memo, _) = positions[key]
1229
yield KnitContentFactory(key, global_map[key],
1230
record_details, sha1, raw_data, self._factory.annotated, None)
1232
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1233
for record in vf.get_record_stream(keys, ordering,
1234
include_delta_closure):
1237
def get_sha1s(self, keys):
1238
"""See VersionedFiles.get_sha1s()."""
1240
record_map = self._get_record_map(missing, allow_missing=True)
1242
for key, details in record_map.iteritems():
1243
if key not in missing:
1245
# record entry 2 is the 'digest'.
1246
result[key] = details[2]
1247
missing.difference_update(set(result))
1248
for source in self._fallback_vfs:
1251
new_result = source.get_sha1s(missing)
1252
result.update(new_result)
1253
missing.difference_update(set(new_result))
1256
def insert_record_stream(self, stream):
1257
"""Insert a record stream into this container.
1259
:param stream: A stream of records to insert.
1261
:seealso VersionedFiles.get_record_stream:
1263
def get_adapter(adapter_key):
1265
return adapters[adapter_key]
1267
adapter_factory = adapter_registry.get(adapter_key)
1268
adapter = adapter_factory(self)
1269
adapters[adapter_key] = adapter
1271
if self._factory.annotated:
1272
# self is annotated, we need annotated knits to use directly.
1273
annotated = "annotated-"
1276
# self is not annotated, but we can strip annotations cheaply.
1278
convertibles = set(["knit-annotated-ft-gz"])
1279
if self._max_delta_chain:
1280
convertibles.add("knit-annotated-delta-gz")
1281
# The set of types we can cheaply adapt without needing basis texts.
1282
native_types = set()
1283
if self._max_delta_chain:
1284
native_types.add("knit-%sdelta-gz" % annotated)
1285
native_types.add("knit-%sft-gz" % annotated)
1286
knit_types = native_types.union(convertibles)
1288
# Buffer all index entries that we can't add immediately because their
1289
# basis parent is missing. We don't buffer all because generating
1290
# annotations may require access to some of the new records. However we
1291
# can't generate annotations from new deltas until their basis parent
1292
# is present anyway, so we get away with not needing an index that
1293
# includes the new keys.
1294
# key = basis_parent, value = index entry to add
1295
buffered_index_entries = {}
1296
for record in stream:
1297
parents = record.parents
1298
# Raise an error when a record is missing.
1299
if record.storage_kind == 'absent':
1300
raise RevisionNotPresent([record.key], self)
1301
if record.storage_kind in knit_types:
1302
if record.storage_kind not in native_types:
1304
adapter_key = (record.storage_kind, "knit-delta-gz")
1305
adapter = get_adapter(adapter_key)
1307
adapter_key = (record.storage_kind, "knit-ft-gz")
1308
adapter = get_adapter(adapter_key)
1309
bytes = adapter.get_bytes(
1310
record, record.get_bytes_as(record.storage_kind))
1312
bytes = record.get_bytes_as(record.storage_kind)
1313
options = [record._build_details[0]]
1314
if record._build_details[1]:
1315
options.append('no-eol')
1316
# Just blat it across.
1317
# Note: This does end up adding data on duplicate keys. As
1318
# modern repositories use atomic insertions this should not
1319
# lead to excessive growth in the event of interrupted fetches.
1320
# 'knit' repositories may suffer excessive growth, but as a
1321
# deprecated format this is tolerable. It can be fixed if
1322
# needed by in the kndx index support raising on a duplicate
1323
# add with identical parents and options.
1324
access_memo = self._access.add_raw_records(
1325
[(record.key, len(bytes))], bytes)[0]
1326
index_entry = (record.key, options, access_memo, parents)
1328
if 'fulltext' not in options:
1329
basis_parent = parents[0]
1330
# Note that pack backed knits don't need to buffer here
1331
# because they buffer all writes to the transaction level,
1332
# but we don't expose that difference at the index level. If
1333
# the query here has sufficient cost to show up in
1334
# profiling we should do that.
1335
if basis_parent not in self.get_parent_map([basis_parent]):
1336
pending = buffered_index_entries.setdefault(
1338
pending.append(index_entry)
1341
self._index.add_records([index_entry])
1342
elif record.storage_kind == 'fulltext':
1343
self.add_lines(record.key, parents,
1344
split_lines(record.get_bytes_as('fulltext')))
1346
adapter_key = record.storage_kind, 'fulltext'
1347
adapter = get_adapter(adapter_key)
1348
lines = split_lines(adapter.get_bytes(
1349
record, record.get_bytes_as(record.storage_kind)))
1351
self.add_lines(record.key, parents, lines)
1352
except errors.RevisionAlreadyPresent:
1354
# Add any records whose basis parent is now available.
1355
added_keys = [record.key]
1357
key = added_keys.pop(0)
1358
if key in buffered_index_entries:
1359
index_entries = buffered_index_entries[key]
1360
self._index.add_records(index_entries)
1362
[index_entry[0] for index_entry in index_entries])
1363
del buffered_index_entries[key]
1364
# If there were any deltas which had a missing basis parent, error.
1365
if buffered_index_entries:
1366
raise errors.RevisionNotPresent(buffered_index_entries.keys()[0],
1369
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1370
"""Iterate over the lines in the versioned files from keys.
1372
This may return lines from other keys. Each item the returned
1373
iterator yields is a tuple of a line and a text version that that line
1374
is present in (not introduced in).
1376
Ordering of results is in whatever order is most suitable for the
1377
underlying storage format.
1379
If a progress bar is supplied, it may be used to indicate progress.
1380
The caller is responsible for cleaning up progress bars (because this
1384
* Lines are normalised by the underlying store: they will all have \n
1386
* Lines are returned in arbitrary order.
1388
:return: An iterator over (line, key).
1391
pb = progress.DummyProgress()
1394
# we don't care about inclusions, the caller cares.
1395
# but we need to setup a list of records to visit.
1396
# we need key, position, length
1398
build_details = self._index.get_build_details(keys)
1399
for key, details in build_details.iteritems():
1401
key_records.append((key, details[0]))
1403
records_iter = enumerate(self._read_records_iter(key_records))
1404
for (key_idx, (key, data, sha_value)) in records_iter:
1405
pb.update('Walking content.', key_idx, total)
1406
compression_parent = build_details[key][1]
1407
if compression_parent is None:
1409
line_iterator = self._factory.get_fulltext_content(data)
1412
line_iterator = self._factory.get_linedelta_content(data)
1413
# XXX: It might be more efficient to yield (key,
1414
# line_iterator) in the future. However for now, this is a simpler
1415
# change to integrate into the rest of the codebase. RBC 20071110
1416
for line in line_iterator:
1418
for source in self._fallback_vfs:
1422
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1423
source_keys.add(key)
1425
keys.difference_update(source_keys)
1427
raise RevisionNotPresent(keys, self.filename)
1428
pb.update('Walking content.', total, total)
1430
def _make_line_delta(self, delta_seq, new_content):
1431
"""Generate a line delta from delta_seq and new_content."""
1433
for op in delta_seq.get_opcodes():
1434
if op[0] == 'equal':
1436
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1439
def _merge_annotations(self, content, parents, parent_texts={},
1440
delta=None, annotated=None,
1441
left_matching_blocks=None):
1442
"""Merge annotations for content and generate deltas.
1444
This is done by comparing the annotations based on changes to the text
1445
and generating a delta on the resulting full texts. If annotations are
1446
not being created then a simple delta is created.
1448
if left_matching_blocks is not None:
1449
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1453
for parent_key in parents:
1454
merge_content = self._get_content(parent_key, parent_texts)
1455
if (parent_key == parents[0] and delta_seq is not None):
1458
seq = patiencediff.PatienceSequenceMatcher(
1459
None, merge_content.text(), content.text())
1460
for i, j, n in seq.get_matching_blocks():
1463
# this copies (origin, text) pairs across to the new
1464
# content for any line that matches the last-checked
1466
content._lines[j:j+n] = merge_content._lines[i:i+n]
1467
# XXX: Robert says the following block is a workaround for a
1468
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1469
if content._lines and content._lines[-1][1][-1] != '\n':
1470
# The copied annotation was from a line without a trailing EOL,
1471
# reinstate one for the content object, to ensure correct
1473
line = content._lines[-1][1] + '\n'
1474
content._lines[-1] = (content._lines[-1][0], line)
1476
if delta_seq is None:
1477
reference_content = self._get_content(parents[0], parent_texts)
1478
new_texts = content.text()
1479
old_texts = reference_content.text()
1480
delta_seq = patiencediff.PatienceSequenceMatcher(
1481
None, old_texts, new_texts)
1482
return self._make_line_delta(delta_seq, content)
1484
def _parse_record(self, version_id, data):
1485
"""Parse an original format knit record.
1487
These have the last element of the key only present in the stored data.
1489
rec, record_contents = self._parse_record_unchecked(data)
1490
self._check_header_version(rec, version_id)
1491
return record_contents, rec[3]
1493
def _parse_record_header(self, key, raw_data):
1494
"""Parse a record header for consistency.
1496
:return: the header and the decompressor stream.
1497
as (stream, header_record)
1499
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1502
rec = self._check_header(key, df.readline())
1503
except Exception, e:
1504
raise KnitCorrupt(self,
1505
"While reading {%s} got %s(%s)"
1506
% (key, e.__class__.__name__, str(e)))
1509
def _parse_record_unchecked(self, data):
1511
# 4168 calls in 2880 217 internal
1512
# 4168 calls to _parse_record_header in 2121
1513
# 4168 calls to readlines in 330
1514
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1516
record_contents = df.readlines()
1517
except Exception, e:
1518
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1519
(data, e.__class__.__name__, str(e)))
1520
header = record_contents.pop(0)
1521
rec = self._split_header(header)
1522
last_line = record_contents.pop()
1523
if len(record_contents) != int(rec[2]):
1524
raise KnitCorrupt(self,
1525
'incorrect number of lines %s != %s'
1526
' for version {%s} %s'
1527
% (len(record_contents), int(rec[2]),
1528
rec[1], record_contents))
1529
if last_line != 'end %s\n' % rec[1]:
1530
raise KnitCorrupt(self,
1531
'unexpected version end line %r, wanted %r'
1532
% (last_line, rec[1]))
1534
return rec, record_contents
1536
def _read_records_iter(self, records):
1537
"""Read text records from data file and yield result.
1539
The result will be returned in whatever is the fastest to read.
1540
Not by the order requested. Also, multiple requests for the same
1541
record will only yield 1 response.
1542
:param records: A list of (key, access_memo) entries
1543
:return: Yields (key, contents, digest) in the order
1544
read, not the order requested
1549
# XXX: This smells wrong, IO may not be getting ordered right.
1550
needed_records = sorted(set(records), key=operator.itemgetter(1))
1551
if not needed_records:
1554
# The transport optimizes the fetching as well
1555
# (ie, reads continuous ranges.)
1556
raw_data = self._access.get_raw_records(
1557
[index_memo for key, index_memo in needed_records])
1559
for (key, index_memo), data in \
1560
izip(iter(needed_records), raw_data):
1561
content, digest = self._parse_record(key[-1], data)
1562
yield key, content, digest
1564
def _read_records_iter_raw(self, records):
1565
"""Read text records from data file and yield raw data.
1567
This unpacks enough of the text record to validate the id is
1568
as expected but thats all.
1570
Each item the iterator yields is (key, bytes, sha1_of_full_text).
1572
# setup an iterator of the external records:
1573
# uses readv so nice and fast we hope.
1575
# grab the disk data needed.
1576
needed_offsets = [index_memo for key, index_memo
1578
raw_records = self._access.get_raw_records(needed_offsets)
1580
for key, index_memo in records:
1581
data = raw_records.next()
1582
# validate the header (note that we can only use the suffix in
1583
# current knit records).
1584
df, rec = self._parse_record_header(key, data)
1586
yield key, data, rec[3]
1588
def _record_to_data(self, key, digest, lines, dense_lines=None):
1589
"""Convert key, digest, lines into a raw data block.
1591
:param key: The key of the record. Currently keys are always serialised
1592
using just the trailing component.
1593
:param dense_lines: The bytes of lines but in a denser form. For
1594
instance, if lines is a list of 1000 bytestrings each ending in \n,
1595
dense_lines may be a list with one line in it, containing all the
1596
1000's lines and their \n's. Using dense_lines if it is already
1597
known is a win because the string join to create bytes in this
1598
function spends less time resizing the final string.
1599
:return: (len, a StringIO instance with the raw data ready to read.)
1601
# Note: using a string copy here increases memory pressure with e.g.
1602
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
1603
# when doing the initial commit of a mozilla tree. RBC 20070921
1604
bytes = ''.join(chain(
1605
["version %s %d %s\n" % (key[-1],
1608
dense_lines or lines,
1609
["end %s\n" % key[-1]]))
1610
if type(bytes) != str:
1611
raise AssertionError(
1612
'data must be plain bytes was %s' % type(bytes))
1613
if lines and lines[-1][-1] != '\n':
1614
raise ValueError('corrupt lines value %r' % lines)
1615
compressed_bytes = tuned_gzip.bytes_to_gzip(bytes)
1616
return len(compressed_bytes), compressed_bytes
1618
def _split_header(self, line):
1621
raise KnitCorrupt(self,
1622
'unexpected number of elements in record header')
1626
"""See VersionedFiles.keys."""
1627
if 'evil' in debug.debug_flags:
1628
trace.mutter_callsite(2, "keys scales with size of history")
1629
sources = [self._index] + self._fallback_vfs
1631
for source in sources:
1632
result.update(source.keys())
1637
class _KndxIndex(object):
1638
"""Manages knit index files
1640
The index is kept in memory and read on startup, to enable
1641
fast lookups of revision information. The cursor of the index
1642
file is always pointing to the end, making it easy to append
1645
_cache is a cache for fast mapping from version id to a Index
1648
_history is a cache for fast mapping from indexes to version ids.
1650
The index data format is dictionary compressed when it comes to
1651
parent references; a index entry may only have parents that with a
1652
lover index number. As a result, the index is topological sorted.
1654
Duplicate entries may be written to the index for a single version id
1655
if this is done then the latter one completely replaces the former:
1656
this allows updates to correct version and parent information.
1657
Note that the two entries may share the delta, and that successive
1658
annotations and references MUST point to the first entry.
1660
The index file on disc contains a header, followed by one line per knit
1661
record. The same revision can be present in an index file more than once.
1662
The first occurrence gets assigned a sequence number starting from 0.
1664
The format of a single line is
1665
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1666
REVISION_ID is a utf8-encoded revision id
1667
FLAGS is a comma separated list of flags about the record. Values include
1668
no-eol, line-delta, fulltext.
1669
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1670
that the the compressed data starts at.
1671
LENGTH is the ascii representation of the length of the data file.
1672
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1674
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1675
revision id already in the knit that is a parent of REVISION_ID.
1676
The ' :' marker is the end of record marker.
1679
when a write is interrupted to the index file, it will result in a line
1680
that does not end in ' :'. If the ' :' is not present at the end of a line,
1681
or at the end of the file, then the record that is missing it will be
1682
ignored by the parser.
1684
When writing new records to the index file, the data is preceded by '\n'
1685
to ensure that records always start on new lines even if the last write was
1686
interrupted. As a result its normal for the last line in the index to be
1687
missing a trailing newline. One can be added with no harmful effects.
1689
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
1690
where prefix is e.g. the (fileid,) for .texts instances or () for
1691
constant-mapped things like .revisions, and the old state is
1692
tuple(history_vector, cache_dict). This is used to prevent having an
1693
ABI change with the C extension that reads .kndx files.
1696
HEADER = "# bzr knit index 8\n"
1698
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
1699
"""Create a _KndxIndex on transport using mapper."""
1700
self._transport = transport
1701
self._mapper = mapper
1702
self._get_scope = get_scope
1703
self._allow_writes = allow_writes
1704
self._is_locked = is_locked
1706
self.has_graph = True
1708
def add_records(self, records, random_id=False):
1709
"""Add multiple records to the index.
1711
:param records: a list of tuples:
1712
(key, options, access_memo, parents).
1713
:param random_id: If True the ids being added were randomly generated
1714
and no check for existence will be performed.
1717
for record in records:
1720
path = self._mapper.map(key) + '.kndx'
1721
path_keys = paths.setdefault(path, (prefix, []))
1722
path_keys[1].append(record)
1723
for path in sorted(paths):
1724
prefix, path_keys = paths[path]
1725
self._load_prefixes([prefix])
1727
orig_history = self._kndx_cache[prefix][1][:]
1728
orig_cache = self._kndx_cache[prefix][0].copy()
1731
for key, options, (_, pos, size), parents in path_keys:
1733
# kndx indices cannot be parentless.
1735
line = "\n%s %s %s %s %s :" % (
1736
key[-1], ','.join(options), pos, size,
1737
self._dictionary_compress(parents))
1738
if type(line) != str:
1739
raise AssertionError(
1740
'data must be utf8 was %s' % type(line))
1742
self._cache_key(key, options, pos, size, parents)
1743
if len(orig_history):
1744
self._transport.append_bytes(path, ''.join(lines))
1746
self._init_index(path, lines)
1748
# If any problems happen, restore the original values and re-raise
1749
self._kndx_cache[prefix] = (orig_cache, orig_history)
1752
def _cache_key(self, key, options, pos, size, parent_keys):
1753
"""Cache a version record in the history array and index cache.
1755
This is inlined into _load_data for performance. KEEP IN SYNC.
1756
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1760
version_id = key[-1]
1761
# last-element only for compatibilty with the C load_data.
1762
parents = tuple(parent[-1] for parent in parent_keys)
1763
for parent in parent_keys:
1764
if parent[:-1] != prefix:
1765
raise ValueError("mismatched prefixes for %r, %r" % (
1767
cache, history = self._kndx_cache[prefix]
1768
# only want the _history index to reference the 1st index entry
1770
if version_id not in cache:
1771
index = len(history)
1772
history.append(version_id)
1774
index = cache[version_id][5]
1775
cache[version_id] = (version_id,
1782
def check_header(self, fp):
1783
line = fp.readline()
1785
# An empty file can actually be treated as though the file doesn't
1787
raise errors.NoSuchFile(self)
1788
if line != self.HEADER:
1789
raise KnitHeaderError(badline=line, filename=self)
1791
def _check_read(self):
1792
if not self._is_locked():
1793
raise errors.ObjectNotLocked(self)
1794
if self._get_scope() != self._scope:
1797
def _check_write_ok(self):
1798
"""Assert if not writes are permitted."""
1799
if not self._is_locked():
1800
raise errors.ObjectNotLocked(self)
1801
if self._get_scope() != self._scope:
1803
if self._mode != 'w':
1804
raise errors.ReadOnlyObjectDirtiedError(self)
1806
def get_build_details(self, keys):
1807
"""Get the method, index_memo and compression parent for keys.
1809
Ghosts are omitted from the result.
1811
:param keys: An iterable of keys.
1812
:return: A dict of key:(index_memo, compression_parent, parents,
1815
opaque structure to pass to read_records to extract the raw
1818
Content that this record is built upon, may be None
1820
Logical parents of this node
1822
extra information about the content which needs to be passed to
1823
Factory.parse_record
1825
prefixes = self._partition_keys(keys)
1826
parent_map = self.get_parent_map(keys)
1829
if key not in parent_map:
1831
method = self.get_method(key)
1832
parents = parent_map[key]
1833
if method == 'fulltext':
1834
compression_parent = None
1836
compression_parent = parents[0]
1837
noeol = 'no-eol' in self.get_options(key)
1838
index_memo = self.get_position(key)
1839
result[key] = (index_memo, compression_parent,
1840
parents, (method, noeol))
1843
def get_method(self, key):
1844
"""Return compression method of specified key."""
1845
options = self.get_options(key)
1846
if 'fulltext' in options:
1848
elif 'line-delta' in options:
1851
raise errors.KnitIndexUnknownMethod(self, options)
1853
def get_options(self, key):
1854
"""Return a list representing options.
1858
prefix, suffix = self._split_key(key)
1859
self._load_prefixes([prefix])
1861
return self._kndx_cache[prefix][0][suffix][1]
1863
raise RevisionNotPresent(key, self)
1865
def get_parent_map(self, keys):
1866
"""Get a map of the parents of keys.
1868
:param keys: The keys to look up parents for.
1869
:return: A mapping from keys to parents. Absent keys are absent from
1872
# Parse what we need to up front, this potentially trades off I/O
1873
# locality (.kndx and .knit in the same block group for the same file
1874
# id) for less checking in inner loops.
1875
prefixes = set(key[:-1] for key in keys)
1876
self._load_prefixes(prefixes)
1881
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
1885
result[key] = tuple(prefix + (suffix,) for
1886
suffix in suffix_parents)
1889
def get_position(self, key):
1890
"""Return details needed to access the version.
1892
:return: a tuple (key, data position, size) to hand to the access
1893
logic to get the record.
1895
prefix, suffix = self._split_key(key)
1896
self._load_prefixes([prefix])
1897
entry = self._kndx_cache[prefix][0][suffix]
1898
return key, entry[2], entry[3]
1900
def _init_index(self, path, extra_lines=[]):
1901
"""Initialize an index."""
1903
sio.write(self.HEADER)
1904
sio.writelines(extra_lines)
1906
self._transport.put_file_non_atomic(path, sio,
1907
create_parent_dir=True)
1908
# self._create_parent_dir)
1909
# mode=self._file_mode,
1910
# dir_mode=self._dir_mode)
1913
"""Get all the keys in the collection.
1915
The keys are not ordered.
1918
# Identify all key prefixes.
1919
# XXX: A bit hacky, needs polish.
1920
if type(self._mapper) == ConstantMapper:
1924
for quoted_relpath in self._transport.iter_files_recursive():
1925
path, ext = os.path.splitext(quoted_relpath)
1927
prefixes = [self._mapper.unmap(path) for path in relpaths]
1928
self._load_prefixes(prefixes)
1929
for prefix in prefixes:
1930
for suffix in self._kndx_cache[prefix][1]:
1931
result.add(prefix + (suffix,))
1934
def _load_prefixes(self, prefixes):
1935
"""Load the indices for prefixes."""
1937
for prefix in prefixes:
1938
if prefix not in self._kndx_cache:
1939
# the load_data interface writes to these variables.
1942
self._filename = prefix
1944
path = self._mapper.map(prefix) + '.kndx'
1945
fp = self._transport.get(path)
1947
# _load_data may raise NoSuchFile if the target knit is
1949
_load_data(self, fp)
1952
self._kndx_cache[prefix] = (self._cache, self._history)
1957
self._kndx_cache[prefix] = ({}, [])
1958
if type(self._mapper) == ConstantMapper:
1959
# preserve behaviour for revisions.kndx etc.
1960
self._init_index(path)
1965
def _partition_keys(self, keys):
1966
"""Turn keys into a dict of prefix:suffix_list."""
1969
prefix_keys = result.setdefault(key[:-1], [])
1970
prefix_keys.append(key[-1])
1973
def _dictionary_compress(self, keys):
1974
"""Dictionary compress keys.
1976
:param keys: The keys to generate references to.
1977
:return: A string representation of keys. keys which are present are
1978
dictionary compressed, and others are emitted as fulltext with a
1984
prefix = keys[0][:-1]
1985
cache = self._kndx_cache[prefix][0]
1987
if key[:-1] != prefix:
1988
# kndx indices cannot refer across partitioned storage.
1989
raise ValueError("mismatched prefixes for %r" % keys)
1990
if key[-1] in cache:
1991
# -- inlined lookup() --
1992
result_list.append(str(cache[key[-1]][5]))
1993
# -- end lookup () --
1995
result_list.append('.' + key[-1])
1996
return ' '.join(result_list)
1998
def _reset_cache(self):
1999
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2000
# (cache_dict, history_vector) for parsed kndx files.
2001
self._kndx_cache = {}
2002
self._scope = self._get_scope()
2003
allow_writes = self._allow_writes()
2009
def _split_key(self, key):
2010
"""Split key into a prefix and suffix."""
2011
return key[:-1], key[-1]
2014
class _KnitGraphIndex(object):
2015
"""A KnitVersionedFiles index layered on GraphIndex."""
2017
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2019
"""Construct a KnitGraphIndex on a graph_index.
2021
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2022
:param is_locked: A callback to check whether the object should answer
2024
:param deltas: Allow delta-compressed records.
2025
:param parents: If True, record knits parents, if not do not record
2027
:param add_callback: If not None, allow additions to the index and call
2028
this callback with a list of added GraphIndex nodes:
2029
[(node, value, node_refs), ...]
2030
:param is_locked: A callback, returns True if the index is locked and
2033
self._add_callback = add_callback
2034
self._graph_index = graph_index
2035
self._deltas = deltas
2036
self._parents = parents
2037
if deltas and not parents:
2038
# XXX: TODO: Delta tree and parent graph should be conceptually
2040
raise KnitCorrupt(self, "Cannot do delta compression without "
2042
self.has_graph = parents
2043
self._is_locked = is_locked
2046
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2048
def add_records(self, records, random_id=False):
2049
"""Add multiple records to the index.
2051
This function does not insert data into the Immutable GraphIndex
2052
backing the KnitGraphIndex, instead it prepares data for insertion by
2053
the caller and checks that it is safe to insert then calls
2054
self._add_callback with the prepared GraphIndex nodes.
2056
:param records: a list of tuples:
2057
(key, options, access_memo, parents).
2058
:param random_id: If True the ids being added were randomly generated
2059
and no check for existence will be performed.
2061
if not self._add_callback:
2062
raise errors.ReadOnlyError(self)
2063
# we hope there are no repositories with inconsistent parentage
2067
for (key, options, access_memo, parents) in records:
2069
parents = tuple(parents)
2070
index, pos, size = access_memo
2071
if 'no-eol' in options:
2075
value += "%d %d" % (pos, size)
2076
if not self._deltas:
2077
if 'line-delta' in options:
2078
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2081
if 'line-delta' in options:
2082
node_refs = (parents, (parents[0],))
2084
node_refs = (parents, ())
2086
node_refs = (parents, )
2089
raise KnitCorrupt(self, "attempt to add node with parents "
2090
"in parentless index.")
2092
keys[key] = (value, node_refs)
2095
present_nodes = self._get_entries(keys)
2096
for (index, key, value, node_refs) in present_nodes:
2097
if (value[0] != keys[key][0][0] or
2098
node_refs != keys[key][1]):
2099
raise KnitCorrupt(self, "inconsistent details in add_records"
2100
": %s %s" % ((value, node_refs), keys[key]))
2104
for key, (value, node_refs) in keys.iteritems():
2105
result.append((key, value, node_refs))
2107
for key, (value, node_refs) in keys.iteritems():
2108
result.append((key, value))
2109
self._add_callback(result)
2111
def _check_read(self):
2112
"""raise if reads are not permitted."""
2113
if not self._is_locked():
2114
raise errors.ObjectNotLocked(self)
2116
def _check_write_ok(self):
2117
"""Assert if writes are not permitted."""
2118
if not self._is_locked():
2119
raise errors.ObjectNotLocked(self)
2121
def _compression_parent(self, an_entry):
2122
# return the key that an_entry is compressed against, or None
2123
# Grab the second parent list (as deltas implies parents currently)
2124
compression_parents = an_entry[3][1]
2125
if not compression_parents:
2127
if len(compression_parents) != 1:
2128
raise AssertionError(
2129
"Too many compression parents: %r" % compression_parents)
2130
return compression_parents[0]
2132
def get_build_details(self, keys):
2133
"""Get the method, index_memo and compression parent for version_ids.
2135
Ghosts are omitted from the result.
2137
:param keys: An iterable of keys.
2138
:return: A dict of key:
2139
(index_memo, compression_parent, parents, record_details).
2141
opaque structure to pass to read_records to extract the raw
2144
Content that this record is built upon, may be None
2146
Logical parents of this node
2148
extra information about the content which needs to be passed to
2149
Factory.parse_record
2153
entries = self._get_entries(keys, False)
2154
for entry in entries:
2156
if not self._parents:
2159
parents = entry[3][0]
2160
if not self._deltas:
2161
compression_parent_key = None
2163
compression_parent_key = self._compression_parent(entry)
2164
noeol = (entry[2][0] == 'N')
2165
if compression_parent_key:
2166
method = 'line-delta'
2169
result[key] = (self._node_to_position(entry),
2170
compression_parent_key, parents,
2174
def _get_entries(self, keys, check_present=False):
2175
"""Get the entries for keys.
2177
:param keys: An iterable of index key tuples.
2182
for node in self._graph_index.iter_entries(keys):
2184
found_keys.add(node[1])
2186
# adapt parentless index to the rest of the code.
2187
for node in self._graph_index.iter_entries(keys):
2188
yield node[0], node[1], node[2], ()
2189
found_keys.add(node[1])
2191
missing_keys = keys.difference(found_keys)
2193
raise RevisionNotPresent(missing_keys.pop(), self)
2195
def get_method(self, key):
2196
"""Return compression method of specified key."""
2197
return self._get_method(self._get_node(key))
2199
def _get_method(self, node):
2200
if not self._deltas:
2202
if self._compression_parent(node):
2207
def _get_node(self, key):
2209
return list(self._get_entries([key]))[0]
2211
raise RevisionNotPresent(key, self)
2213
def get_options(self, key):
2214
"""Return a list representing options.
2218
node = self._get_node(key)
2219
options = [self._get_method(node)]
2220
if node[2][0] == 'N':
2221
options.append('no-eol')
2224
def get_parent_map(self, keys):
2225
"""Get a map of the parents of keys.
2227
:param keys: The keys to look up parents for.
2228
:return: A mapping from keys to parents. Absent keys are absent from
2232
nodes = self._get_entries(keys)
2236
result[node[1]] = node[3][0]
2239
result[node[1]] = None
2242
def get_position(self, key):
2243
"""Return details needed to access the version.
2245
:return: a tuple (index, data position, size) to hand to the access
2246
logic to get the record.
2248
node = self._get_node(key)
2249
return self._node_to_position(node)
2252
"""Get all the keys in the collection.
2254
The keys are not ordered.
2257
return [node[1] for node in self._graph_index.iter_all_entries()]
2259
def _node_to_position(self, node):
2260
"""Convert an index value to position details."""
2261
bits = node[2][1:].split(' ')
2262
return node[0], int(bits[0]), int(bits[1])
2265
class _KnitKeyAccess(object):
2266
"""Access to records in .knit files."""
2268
def __init__(self, transport, mapper):
2269
"""Create a _KnitKeyAccess with transport and mapper.
2271
:param transport: The transport the access object is rooted at.
2272
:param mapper: The mapper used to map keys to .knit files.
2274
self._transport = transport
2275
self._mapper = mapper
2277
def add_raw_records(self, key_sizes, raw_data):
2278
"""Add raw knit bytes to a storage area.
2280
The data is spooled to the container writer in one bytes-record per
2283
:param sizes: An iterable of tuples containing the key and size of each
2285
:param raw_data: A bytestring containing the data.
2286
:return: A list of memos to retrieve the record later. Each memo is an
2287
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
2288
length), where the key is the record key.
2290
if type(raw_data) != str:
2291
raise AssertionError(
2292
'data must be plain bytes was %s' % type(raw_data))
2295
# TODO: This can be tuned for writing to sftp and other servers where
2296
# append() is relatively expensive by grouping the writes to each key
2298
for key, size in key_sizes:
2299
path = self._mapper.map(key)
2301
base = self._transport.append_bytes(path + '.knit',
2302
raw_data[offset:offset+size])
2303
except errors.NoSuchFile:
2304
self._transport.mkdir(osutils.dirname(path))
2305
base = self._transport.append_bytes(path + '.knit',
2306
raw_data[offset:offset+size])
2310
result.append((key, base, size))
2313
def get_raw_records(self, memos_for_retrieval):
2314
"""Get the raw bytes for a records.
2316
:param memos_for_retrieval: An iterable containing the access memo for
2317
retrieving the bytes.
2318
:return: An iterator over the bytes of the records.
2320
# first pass, group into same-index request to minimise readv's issued.
2322
current_prefix = None
2323
for (key, offset, length) in memos_for_retrieval:
2324
if current_prefix == key[:-1]:
2325
current_list.append((offset, length))
2327
if current_prefix is not None:
2328
request_lists.append((current_prefix, current_list))
2329
current_prefix = key[:-1]
2330
current_list = [(offset, length)]
2331
# handle the last entry
2332
if current_prefix is not None:
2333
request_lists.append((current_prefix, current_list))
2334
for prefix, read_vector in request_lists:
2335
path = self._mapper.map(prefix) + '.knit'
2336
for pos, data in self._transport.readv(path, read_vector):
2340
class _DirectPackAccess(object):
2341
"""Access to data in one or more packs with less translation."""
2343
def __init__(self, index_to_packs):
2344
"""Create a _DirectPackAccess object.
2346
:param index_to_packs: A dict mapping index objects to the transport
2347
and file names for obtaining data.
2349
self._container_writer = None
2350
self._write_index = None
2351
self._indices = index_to_packs
2353
def add_raw_records(self, key_sizes, raw_data):
2354
"""Add raw knit bytes to a storage area.
2356
The data is spooled to the container writer in one bytes-record per
2359
:param sizes: An iterable of tuples containing the key and size of each
2361
:param raw_data: A bytestring containing the data.
2362
:return: A list of memos to retrieve the record later. Each memo is an
2363
opaque index memo. For _DirectPackAccess the memo is (index, pos,
2364
length), where the index field is the write_index object supplied
2365
to the PackAccess object.
2367
if type(raw_data) != str:
2368
raise AssertionError(
2369
'data must be plain bytes was %s' % type(raw_data))
2372
for key, size in key_sizes:
2373
p_offset, p_length = self._container_writer.add_bytes_record(
2374
raw_data[offset:offset+size], [])
2376
result.append((self._write_index, p_offset, p_length))
2379
def get_raw_records(self, memos_for_retrieval):
2380
"""Get the raw bytes for a records.
2382
:param memos_for_retrieval: An iterable containing the (index, pos,
2383
length) memo for retrieving the bytes. The Pack access method
2384
looks up the pack to use for a given record in its index_to_pack
2386
:return: An iterator over the bytes of the records.
2388
# first pass, group into same-index requests
2390
current_index = None
2391
for (index, offset, length) in memos_for_retrieval:
2392
if current_index == index:
2393
current_list.append((offset, length))
2395
if current_index is not None:
2396
request_lists.append((current_index, current_list))
2397
current_index = index
2398
current_list = [(offset, length)]
2399
# handle the last entry
2400
if current_index is not None:
2401
request_lists.append((current_index, current_list))
2402
for index, offsets in request_lists:
2403
transport, path = self._indices[index]
2404
reader = pack.make_readv_reader(transport, path, offsets)
2405
for names, read_func in reader.iter_records():
2406
yield read_func(None)
2408
def set_writer(self, writer, index, transport_packname):
2409
"""Set a writer to use for adding data."""
2410
if index is not None:
2411
self._indices[index] = transport_packname
2412
self._container_writer = writer
2413
self._write_index = index
2416
# Deprecated, use PatienceSequenceMatcher instead
2417
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2420
def annotate_knit(knit, revision_id):
2421
"""Annotate a knit with no cached annotations.
2423
This implementation is for knits with no cached annotations.
2424
It will work for knits with cached annotations, but this is not
2427
annotator = _KnitAnnotator(knit)
2428
return iter(annotator.annotate(revision_id))
2431
class _KnitAnnotator(object):
2432
"""Build up the annotations for a text."""
2434
def __init__(self, knit):
2437
# Content objects, differs from fulltexts because of how final newlines
2438
# are treated by knits. the content objects here will always have a
2440
self._fulltext_contents = {}
2442
# Annotated lines of specific revisions
2443
self._annotated_lines = {}
2445
# Track the raw data for nodes that we could not process yet.
2446
# This maps the revision_id of the base to a list of children that will
2447
# annotated from it.
2448
self._pending_children = {}
2450
# Nodes which cannot be extracted
2451
self._ghosts = set()
2453
# Track how many children this node has, so we know if we need to keep
2455
self._annotate_children = {}
2456
self._compression_children = {}
2458
self._all_build_details = {}
2459
# The children => parent revision_id graph
2460
self._revision_id_graph = {}
2462
self._heads_provider = None
2464
self._nodes_to_keep_annotations = set()
2465
self._generations_until_keep = 100
2467
def set_generations_until_keep(self, value):
2468
"""Set the number of generations before caching a node.
2470
Setting this to -1 will cache every merge node, setting this higher
2471
will cache fewer nodes.
2473
self._generations_until_keep = value
2475
def _add_fulltext_content(self, revision_id, content_obj):
2476
self._fulltext_contents[revision_id] = content_obj
2477
# TODO: jam 20080305 It might be good to check the sha1digest here
2478
return content_obj.text()
2480
def _check_parents(self, child, nodes_to_annotate):
2481
"""Check if all parents have been processed.
2483
:param child: A tuple of (rev_id, parents, raw_content)
2484
:param nodes_to_annotate: If child is ready, add it to
2485
nodes_to_annotate, otherwise put it back in self._pending_children
2487
for parent_id in child[1]:
2488
if (parent_id not in self._annotated_lines):
2489
# This parent is present, but another parent is missing
2490
self._pending_children.setdefault(parent_id,
2494
# This one is ready to be processed
2495
nodes_to_annotate.append(child)
2497
def _add_annotation(self, revision_id, fulltext, parent_ids,
2498
left_matching_blocks=None):
2499
"""Add an annotation entry.
2501
All parents should already have been annotated.
2502
:return: A list of children that now have their parents satisfied.
2504
a = self._annotated_lines
2505
annotated_parent_lines = [a[p] for p in parent_ids]
2506
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2507
fulltext, revision_id, left_matching_blocks,
2508
heads_provider=self._get_heads_provider()))
2509
self._annotated_lines[revision_id] = annotated_lines
2510
for p in parent_ids:
2511
ann_children = self._annotate_children[p]
2512
ann_children.remove(revision_id)
2513
if (not ann_children
2514
and p not in self._nodes_to_keep_annotations):
2515
del self._annotated_lines[p]
2516
del self._all_build_details[p]
2517
if p in self._fulltext_contents:
2518
del self._fulltext_contents[p]
2519
# Now that we've added this one, see if there are any pending
2520
# deltas to be done, certainly this parent is finished
2521
nodes_to_annotate = []
2522
for child in self._pending_children.pop(revision_id, []):
2523
self._check_parents(child, nodes_to_annotate)
2524
return nodes_to_annotate
2526
def _get_build_graph(self, key):
2527
"""Get the graphs for building texts and annotations.
2529
The data you need for creating a full text may be different than the
2530
data you need to annotate that text. (At a minimum, you need both
2531
parents to create an annotation, but only need 1 parent to generate the
2534
:return: A list of (key, index_memo) records, suitable for
2535
passing to read_records_iter to start reading in the raw data fro/
2538
if key in self._annotated_lines:
2541
pending = set([key])
2546
# get all pending nodes
2548
this_iteration = pending
2549
build_details = self._knit._index.get_build_details(this_iteration)
2550
self._all_build_details.update(build_details)
2551
# new_nodes = self._knit._index._get_entries(this_iteration)
2553
for key, details in build_details.iteritems():
2554
(index_memo, compression_parent, parents,
2555
record_details) = details
2556
self._revision_id_graph[key] = parents
2557
records.append((key, index_memo))
2558
# Do we actually need to check _annotated_lines?
2559
pending.update(p for p in parents
2560
if p not in self._all_build_details)
2561
if compression_parent:
2562
self._compression_children.setdefault(compression_parent,
2565
for parent in parents:
2566
self._annotate_children.setdefault(parent,
2568
num_gens = generation - kept_generation
2569
if ((num_gens >= self._generations_until_keep)
2570
and len(parents) > 1):
2571
kept_generation = generation
2572
self._nodes_to_keep_annotations.add(key)
2574
missing_versions = this_iteration.difference(build_details.keys())
2575
self._ghosts.update(missing_versions)
2576
for missing_version in missing_versions:
2577
# add a key, no parents
2578
self._revision_id_graph[missing_version] = ()
2579
pending.discard(missing_version) # don't look for it
2580
if self._ghosts.intersection(self._compression_children):
2582
"We cannot have nodes which have a ghost compression parent:\n"
2584
"compression children: %r"
2585
% (self._ghosts, self._compression_children))
2586
# Cleanout anything that depends on a ghost so that we don't wait for
2587
# the ghost to show up
2588
for node in self._ghosts:
2589
if node in self._annotate_children:
2590
# We won't be building this node
2591
del self._annotate_children[node]
2592
# Generally we will want to read the records in reverse order, because
2593
# we find the parent nodes after the children
2597
def _annotate_records(self, records):
2598
"""Build the annotations for the listed records."""
2599
# We iterate in the order read, rather than a strict order requested
2600
# However, process what we can, and put off to the side things that
2601
# still need parents, cleaning them up when those parents are
2603
for (rev_id, record,
2604
digest) in self._knit._read_records_iter(records):
2605
if rev_id in self._annotated_lines:
2607
parent_ids = self._revision_id_graph[rev_id]
2608
parent_ids = [p for p in parent_ids if p not in self._ghosts]
2609
details = self._all_build_details[rev_id]
2610
(index_memo, compression_parent, parents,
2611
record_details) = details
2612
nodes_to_annotate = []
2613
# TODO: Remove the punning between compression parents, and
2614
# parent_ids, we should be able to do this without assuming
2616
if len(parent_ids) == 0:
2617
# There are no parents for this node, so just add it
2618
# TODO: This probably needs to be decoupled
2619
fulltext_content, delta = self._knit._factory.parse_record(
2620
rev_id, record, record_details, None)
2621
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
2622
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
2623
parent_ids, left_matching_blocks=None))
2625
child = (rev_id, parent_ids, record)
2626
# Check if all the parents are present
2627
self._check_parents(child, nodes_to_annotate)
2628
while nodes_to_annotate:
2629
# Should we use a queue here instead of a stack?
2630
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
2631
(index_memo, compression_parent, parents,
2632
record_details) = self._all_build_details[rev_id]
2633
if compression_parent is not None:
2634
comp_children = self._compression_children[compression_parent]
2635
if rev_id not in comp_children:
2636
raise AssertionError("%r not in compression children %r"
2637
% (rev_id, comp_children))
2638
# If there is only 1 child, it is safe to reuse this
2640
reuse_content = (len(comp_children) == 1
2641
and compression_parent not in
2642
self._nodes_to_keep_annotations)
2644
# Remove it from the cache since it will be changing
2645
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
2646
# Make sure to copy the fulltext since it might be
2648
parent_fulltext = list(parent_fulltext_content.text())
2650
parent_fulltext_content = self._fulltext_contents[compression_parent]
2651
parent_fulltext = parent_fulltext_content.text()
2652
comp_children.remove(rev_id)
2653
fulltext_content, delta = self._knit._factory.parse_record(
2654
rev_id, record, record_details,
2655
parent_fulltext_content,
2656
copy_base_content=(not reuse_content))
2657
fulltext = self._add_fulltext_content(rev_id,
2659
blocks = KnitContent.get_line_delta_blocks(delta,
2660
parent_fulltext, fulltext)
2662
fulltext_content = self._knit._factory.parse_fulltext(
2664
fulltext = self._add_fulltext_content(rev_id,
2667
nodes_to_annotate.extend(
2668
self._add_annotation(rev_id, fulltext, parent_ids,
2669
left_matching_blocks=blocks))
2671
def _get_heads_provider(self):
2672
"""Create a heads provider for resolving ancestry issues."""
2673
if self._heads_provider is not None:
2674
return self._heads_provider
2675
parent_provider = _mod_graph.DictParentsProvider(
2676
self._revision_id_graph)
2677
graph_obj = _mod_graph.Graph(parent_provider)
2678
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
2679
self._heads_provider = head_cache
2682
def annotate(self, key):
2683
"""Return the annotated fulltext at the given key.
2685
:param key: The key to annotate.
2687
if True or len(self._knit._fallback_vfs) > 0:
2688
# stacked knits can't use the fast path at present.
2689
return self._simple_annotate(key)
2690
records = self._get_build_graph(key)
2691
if key in self._ghosts:
2692
raise errors.RevisionNotPresent(key, self._knit)
2693
self._annotate_records(records)
2694
return self._annotated_lines[key]
2696
def _simple_annotate(self, key):
2697
"""Return annotated fulltext, rediffing from the full texts.
2699
This is slow but makes no assumptions about the repository
2700
being able to produce line deltas.
2702
# TODO: this code generates a parent maps of present ancestors; it
2703
# could be split out into a separate method, and probably should use
2704
# iter_ancestry instead. -- mbp and robertc 20080704
2705
graph = _mod_graph.Graph(self._knit)
2706
head_cache = _mod_graph.FrozenHeadsCache(graph)
2707
search = graph._make_breadth_first_searcher([key])
2711
present, ghosts = search.next_with_ghosts()
2712
except StopIteration:
2714
keys.update(present)
2715
parent_map = self._knit.get_parent_map(keys)
2717
reannotate = annotate.reannotate
2718
for record in self._knit.get_record_stream(keys, 'topological', True):
2720
fulltext = split_lines(record.get_bytes_as('fulltext'))
2721
parents = parent_map[key]
2722
if parents is not None:
2723
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
2726
parent_cache[key] = list(
2727
reannotate(parent_lines, fulltext, key, None, head_cache))
2729
return parent_cache[key]
2731
raise errors.RevisionNotPresent(key, self._knit)
2735
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2737
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
