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 = []
729
def add_fallback_versioned_files(self, a_versioned_files):
730
"""Add a source of texts for texts not present in this knit.
732
:param a_versioned_files: A VersionedFiles object.
734
self._fallback_vfs.append(a_versioned_files)
736
def add_lines(self, key, parents, lines, parent_texts=None,
737
left_matching_blocks=None, nostore_sha=None, random_id=False,
739
"""See VersionedFiles.add_lines()."""
740
self._index._check_write_ok()
741
self._check_add(key, lines, random_id, check_content)
743
# The caller might pass None if there is no graph data, but kndx
744
# indexes can't directly store that, so we give them
745
# an empty tuple instead.
747
return self._add(key, lines, parents,
748
parent_texts, left_matching_blocks, nostore_sha, random_id)
750
def _add(self, key, lines, parents, parent_texts,
751
left_matching_blocks, nostore_sha, random_id):
752
"""Add a set of lines on top of version specified by parents.
754
Any versions not present will be converted into ghosts.
756
# first thing, if the content is something we don't need to store, find
758
line_bytes = ''.join(lines)
759
digest = sha_string(line_bytes)
760
if nostore_sha == digest:
761
raise errors.ExistingContent
764
if parent_texts is None:
766
# Do a single query to ascertain parent presence.
767
present_parent_map = self.get_parent_map(parents)
768
for parent in parents:
769
if parent in present_parent_map:
770
present_parents.append(parent)
772
# Currently we can only compress against the left most present parent.
773
if (len(present_parents) == 0 or
774
present_parents[0] != parents[0]):
777
# To speed the extract of texts the delta chain is limited
778
# to a fixed number of deltas. This should minimize both
779
# I/O and the time spend applying deltas.
780
delta = self._check_should_delta(present_parents[0])
782
text_length = len(line_bytes)
785
if lines[-1][-1] != '\n':
786
# copy the contents of lines.
788
options.append('no-eol')
789
lines[-1] = lines[-1] + '\n'
793
if type(element) != str:
794
raise TypeError("key contains non-strings: %r" % (key,))
795
# Knit hunks are still last-element only
797
content = self._factory.make(lines, version_id)
798
if 'no-eol' in options:
799
# Hint to the content object that its text() call should strip the
801
content._should_strip_eol = True
802
if delta or (self._factory.annotated and len(present_parents) > 0):
803
# Merge annotations from parent texts if needed.
804
delta_hunks = self._merge_annotations(content, present_parents,
805
parent_texts, delta, self._factory.annotated,
806
left_matching_blocks)
809
options.append('line-delta')
810
store_lines = self._factory.lower_line_delta(delta_hunks)
811
size, bytes = self._record_to_data(key, digest,
814
options.append('fulltext')
815
# isinstance is slower and we have no hierarchy.
816
if self._factory.__class__ == KnitPlainFactory:
817
# Use the already joined bytes saving iteration time in
819
size, bytes = self._record_to_data(key, digest,
822
# get mixed annotation + content and feed it into the
824
store_lines = self._factory.lower_fulltext(content)
825
size, bytes = self._record_to_data(key, digest,
828
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
829
self._index.add_records(
830
((key, options, access_memo, parents),),
832
return digest, text_length, content
834
def annotate(self, key):
835
"""See VersionedFiles.annotate."""
836
return self._factory.annotate(self, key)
838
def check(self, progress_bar=None):
839
"""See VersionedFiles.check()."""
840
# This doesn't actually test extraction of everything, but that will
841
# impact 'bzr check' substantially, and needs to be integrated with
842
# care. However, it does check for the obvious problem of a delta with
844
keys = self._index.keys()
845
parent_map = self.get_parent_map(keys)
847
if self._index.get_method(key) != 'fulltext':
848
compression_parent = parent_map[key][0]
849
if compression_parent not in parent_map:
850
raise errors.KnitCorrupt(self,
851
"Missing basis parent %s for %s" % (
852
compression_parent, key))
853
for fallback_vfs in self._fallback_vfs:
856
def _check_add(self, key, lines, random_id, check_content):
857
"""check that version_id and lines are safe to add."""
859
if contains_whitespace(version_id):
860
raise InvalidRevisionId(version_id, self)
861
self.check_not_reserved_id(version_id)
862
# TODO: If random_id==False and the key is already present, we should
863
# probably check that the existing content is identical to what is
864
# being inserted, and otherwise raise an exception. This would make
865
# the bundle code simpler.
867
self._check_lines_not_unicode(lines)
868
self._check_lines_are_lines(lines)
870
def _check_header(self, key, line):
871
rec = self._split_header(line)
872
self._check_header_version(rec, key[-1])
875
def _check_header_version(self, rec, version_id):
876
"""Checks the header version on original format knit records.
878
These have the last component of the key embedded in the record.
880
if rec[1] != version_id:
881
raise KnitCorrupt(self,
882
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
884
def _check_should_delta(self, parent):
885
"""Iterate back through the parent listing, looking for a fulltext.
887
This is used when we want to decide whether to add a delta or a new
888
fulltext. It searches for _max_delta_chain parents. When it finds a
889
fulltext parent, it sees if the total size of the deltas leading up to
890
it is large enough to indicate that we want a new full text anyway.
892
Return True if we should create a new delta, False if we should use a
897
for count in xrange(self._max_delta_chain):
898
# XXX: Collapse these two queries:
900
# Note that this only looks in the index of this particular
901
# KnitVersionedFiles, not in the fallbacks. This ensures that
902
# we won't store a delta spanning physical repository
904
method = self._index.get_method(parent)
905
except RevisionNotPresent:
906
# Some basis is not locally present: always delta
908
index, pos, size = self._index.get_position(parent)
909
if method == 'fulltext':
913
# We don't explicitly check for presence because this is in an
914
# inner loop, and if it's missing it'll fail anyhow.
915
# TODO: This should be asking for compression parent, not graph
917
parent = self._index.get_parent_map([parent])[parent][0]
919
# We couldn't find a fulltext, so we must create a new one
921
# Simple heuristic - if the total I/O wold be greater as a delta than
922
# the originally installed fulltext, we create a new fulltext.
923
return fulltext_size > delta_size
925
def _build_details_to_components(self, build_details):
926
"""Convert a build_details tuple to a position tuple."""
927
# record_details, access_memo, compression_parent
928
return build_details[3], build_details[0], build_details[1]
930
def _get_components_positions(self, keys, allow_missing=False):
931
"""Produce a map of position data for the components of keys.
933
This data is intended to be used for retrieving the knit records.
935
A dict of key to (record_details, index_memo, next, parents) is
937
method is the way referenced data should be applied.
938
index_memo is the handle to pass to the data access to actually get the
940
next is the build-parent of the version, or None for fulltexts.
941
parents is the version_ids of the parents of this version
943
:param allow_missing: If True do not raise an error on a missing component,
947
pending_components = keys
948
while pending_components:
949
build_details = self._index.get_build_details(pending_components)
950
current_components = set(pending_components)
951
pending_components = set()
952
for key, details in build_details.iteritems():
953
(index_memo, compression_parent, parents,
954
record_details) = details
955
method = record_details[0]
956
if compression_parent is not None:
957
pending_components.add(compression_parent)
958
component_data[key] = self._build_details_to_components(details)
959
missing = current_components.difference(build_details)
960
if missing and not allow_missing:
961
raise errors.RevisionNotPresent(missing.pop(), self)
962
return component_data
964
def _get_content(self, key, parent_texts={}):
965
"""Returns a content object that makes up the specified
967
cached_version = parent_texts.get(key, None)
968
if cached_version is not None:
969
# Ensure the cache dict is valid.
970
if not self.get_parent_map([key]):
971
raise RevisionNotPresent(key, self)
972
return cached_version
973
text_map, contents_map = self._get_content_maps([key])
974
return contents_map[key]
976
def _get_content_maps(self, keys, nonlocal_keys=None):
977
"""Produce maps of text and KnitContents
979
:param keys: The keys to produce content maps for.
980
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
981
which are known to not be in this knit, but rather in one of the
983
:return: (text_map, content_map) where text_map contains the texts for
984
the requested versions and content_map contains the KnitContents.
986
# FUTURE: This function could be improved for the 'extract many' case
987
# by tracking each component and only doing the copy when the number of
988
# children than need to apply delta's to it is > 1 or it is part of the
991
multiple_versions = len(keys) != 1
992
record_map = self._get_record_map(keys, allow_missing=True)
997
if nonlocal_keys is None:
998
nonlocal_keys = set()
1000
nonlocal_keys = frozenset(nonlocal_keys)
1001
missing_keys = set(nonlocal_keys)
1002
for source in self._fallback_vfs:
1003
if not missing_keys:
1005
for record in source.get_record_stream(missing_keys,
1007
if record.storage_kind == 'absent':
1009
missing_keys.remove(record.key)
1010
lines = split_lines(record.get_bytes_as('fulltext'))
1011
text_map[record.key] = lines
1012
content_map[record.key] = PlainKnitContent(lines, record.key)
1013
if record.key in keys:
1014
final_content[record.key] = content_map[record.key]
1016
if key in nonlocal_keys:
1021
while cursor is not None:
1023
record, record_details, digest, next = record_map[cursor]
1025
raise RevisionNotPresent(cursor, self)
1026
components.append((cursor, record, record_details, digest))
1028
if cursor in content_map:
1029
# no need to plan further back
1030
components.append((cursor, None, None, None))
1034
for (component_id, record, record_details,
1035
digest) in reversed(components):
1036
if component_id in content_map:
1037
content = content_map[component_id]
1039
content, delta = self._factory.parse_record(key[-1],
1040
record, record_details, content,
1041
copy_base_content=multiple_versions)
1042
if multiple_versions:
1043
content_map[component_id] = content
1045
final_content[key] = content
1047
# digest here is the digest from the last applied component.
1048
text = content.text()
1049
actual_sha = sha_strings(text)
1050
if actual_sha != digest:
1051
raise KnitCorrupt(self,
1053
'\n of reconstructed text does not match'
1055
'\n for version %s' %
1056
(actual_sha, digest, key))
1057
text_map[key] = text
1058
return text_map, final_content
1060
def get_parent_map(self, keys):
1061
"""Get a map of the graph parents of keys.
1063
:param keys: The keys to look up parents for.
1064
:return: A mapping from keys to parents. Absent keys are absent from
1067
return self._get_parent_map_with_sources(keys)[0]
1069
def _get_parent_map_with_sources(self, keys):
1070
"""Get a map of the parents of keys.
1072
:param keys: The keys to look up parents for.
1073
:return: A tuple. The first element is a mapping from keys to parents.
1074
Absent keys are absent from the mapping. The second element is a
1075
list with the locations each key was found in. The first element
1076
is the in-this-knit parents, the second the first fallback source,
1080
sources = [self._index] + self._fallback_vfs
1083
for source in sources:
1086
new_result = source.get_parent_map(missing)
1087
source_results.append(new_result)
1088
result.update(new_result)
1089
missing.difference_update(set(new_result))
1090
return result, source_results
1092
def _get_record_map(self, keys, allow_missing=False):
1093
"""Produce a dictionary of knit records.
1095
:return: {key:(record, record_details, digest, next)}
1097
data returned from read_records
1099
opaque information to pass to parse_record
1101
SHA1 digest of the full text after all steps are done
1103
build-parent of the version, i.e. the leftmost ancestor.
1104
Will be None if the record is not a delta.
1105
:param keys: The keys to build a map for
1106
:param allow_missing: If some records are missing, rather than
1107
error, just return the data that could be generated.
1109
position_map = self._get_components_positions(keys,
1110
allow_missing=allow_missing)
1111
# key = component_id, r = record_details, i_m = index_memo, n = next
1112
records = [(key, i_m) for key, (r, i_m, n)
1113
in position_map.iteritems()]
1115
for key, record, digest in \
1116
self._read_records_iter(records):
1117
(record_details, index_memo, next) = position_map[key]
1118
record_map[key] = record, record_details, digest, next
1121
def get_record_stream(self, keys, ordering, include_delta_closure):
1122
"""Get a stream of records for keys.
1124
:param keys: The keys to include.
1125
:param ordering: Either 'unordered' or 'topological'. A topologically
1126
sorted stream has compression parents strictly before their
1128
:param include_delta_closure: If True then the closure across any
1129
compression parents will be included (in the opaque data).
1130
:return: An iterator of ContentFactory objects, each of which is only
1131
valid until the iterator is advanced.
1133
# keys might be a generator
1137
if not self._index.has_graph:
1138
# Cannot topological order when no graph has been stored.
1139
ordering = 'unordered'
1140
if include_delta_closure:
1141
positions = self._get_components_positions(keys, allow_missing=True)
1143
build_details = self._index.get_build_details(keys)
1145
# (record_details, access_memo, compression_parent_key)
1146
positions = dict((key, self._build_details_to_components(details))
1147
for key, details in build_details.iteritems())
1148
absent_keys = keys.difference(set(positions))
1149
# There may be more absent keys : if we're missing the basis component
1150
# and are trying to include the delta closure.
1151
if include_delta_closure:
1152
needed_from_fallback = set()
1153
# Build up reconstructable_keys dict. key:True in this dict means
1154
# the key can be reconstructed.
1155
reconstructable_keys = {}
1159
chain = [key, positions[key][2]]
1161
needed_from_fallback.add(key)
1164
while chain[-1] is not None:
1165
if chain[-1] in reconstructable_keys:
1166
result = reconstructable_keys[chain[-1]]
1170
chain.append(positions[chain[-1]][2])
1172
# missing basis component
1173
needed_from_fallback.add(chain[-1])
1176
for chain_key in chain[:-1]:
1177
reconstructable_keys[chain_key] = result
1179
needed_from_fallback.add(key)
1180
# Double index lookups here : need a unified api ?
1181
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1182
if ordering == 'topological':
1183
# Global topological sort
1184
present_keys = tsort.topo_sort(global_map)
1185
# Now group by source:
1187
current_source = None
1188
for key in present_keys:
1189
for parent_map in parent_maps:
1190
if key in parent_map:
1191
key_source = parent_map
1193
if current_source is not key_source:
1194
source_keys.append((key_source, []))
1195
current_source = key_source
1196
source_keys[-1][1].append(key)
1198
# Just group by source; remote sources first.
1201
for parent_map in reversed(parent_maps):
1202
source_keys.append((parent_map, []))
1203
for key in parent_map:
1204
present_keys.append(key)
1205
source_keys[-1][1].append(key)
1206
absent_keys = keys - set(global_map)
1207
for key in absent_keys:
1208
yield AbsentContentFactory(key)
1209
# restrict our view to the keys we can answer.
1210
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1211
# XXX: At that point we need to consider the impact of double reads by
1212
# utilising components multiple times.
1213
if include_delta_closure:
1214
# XXX: get_content_maps performs its own index queries; allow state
1216
text_map, _ = self._get_content_maps(present_keys,
1217
needed_from_fallback - absent_keys)
1218
for key in present_keys:
1219
yield FulltextContentFactory(key, global_map[key], None,
1220
''.join(text_map[key]))
1222
for source, keys in source_keys:
1223
if source is parent_maps[0]:
1224
# this KnitVersionedFiles
1225
records = [(key, positions[key][1]) for key in keys]
1226
for key, raw_data, sha1 in self._read_records_iter_raw(records):
1227
(record_details, index_memo, _) = positions[key]
1228
yield KnitContentFactory(key, global_map[key],
1229
record_details, sha1, raw_data, self._factory.annotated, None)
1231
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1232
for record in vf.get_record_stream(keys, ordering,
1233
include_delta_closure):
1236
def get_sha1s(self, keys):
1237
"""See VersionedFiles.get_sha1s()."""
1239
record_map = self._get_record_map(missing, allow_missing=True)
1241
for key, details in record_map.iteritems():
1242
if key not in missing:
1244
# record entry 2 is the 'digest'.
1245
result[key] = details[2]
1246
missing.difference_update(set(result))
1247
for source in self._fallback_vfs:
1250
new_result = source.get_sha1s(missing)
1251
result.update(new_result)
1252
missing.difference_update(set(new_result))
1255
def insert_record_stream(self, stream):
1256
"""Insert a record stream into this container.
1258
:param stream: A stream of records to insert.
1260
:seealso VersionedFiles.get_record_stream:
1262
def get_adapter(adapter_key):
1264
return adapters[adapter_key]
1266
adapter_factory = adapter_registry.get(adapter_key)
1267
adapter = adapter_factory(self)
1268
adapters[adapter_key] = adapter
1270
if self._factory.annotated:
1271
# self is annotated, we need annotated knits to use directly.
1272
annotated = "annotated-"
1275
# self is not annotated, but we can strip annotations cheaply.
1277
convertibles = set(["knit-annotated-ft-gz"])
1278
if self._max_delta_chain:
1279
convertibles.add("knit-annotated-delta-gz")
1280
# The set of types we can cheaply adapt without needing basis texts.
1281
native_types = set()
1282
if self._max_delta_chain:
1283
native_types.add("knit-%sdelta-gz" % annotated)
1284
native_types.add("knit-%sft-gz" % annotated)
1285
knit_types = native_types.union(convertibles)
1287
# Buffer all index entries that we can't add immediately because their
1288
# basis parent is missing. We don't buffer all because generating
1289
# annotations may require access to some of the new records. However we
1290
# can't generate annotations from new deltas until their basis parent
1291
# is present anyway, so we get away with not needing an index that
1292
# includes the new keys.
1293
# key = basis_parent, value = index entry to add
1294
buffered_index_entries = {}
1295
for record in stream:
1296
parents = record.parents
1297
# Raise an error when a record is missing.
1298
if record.storage_kind == 'absent':
1299
raise RevisionNotPresent([record.key], self)
1300
if record.storage_kind in knit_types:
1301
if record.storage_kind not in native_types:
1303
adapter_key = (record.storage_kind, "knit-delta-gz")
1304
adapter = get_adapter(adapter_key)
1306
adapter_key = (record.storage_kind, "knit-ft-gz")
1307
adapter = get_adapter(adapter_key)
1308
bytes = adapter.get_bytes(
1309
record, record.get_bytes_as(record.storage_kind))
1311
bytes = record.get_bytes_as(record.storage_kind)
1312
options = [record._build_details[0]]
1313
if record._build_details[1]:
1314
options.append('no-eol')
1315
# Just blat it across.
1316
# Note: This does end up adding data on duplicate keys. As
1317
# modern repositories use atomic insertions this should not
1318
# lead to excessive growth in the event of interrupted fetches.
1319
# 'knit' repositories may suffer excessive growth, but as a
1320
# deprecated format this is tolerable. It can be fixed if
1321
# needed by in the kndx index support raising on a duplicate
1322
# add with identical parents and options.
1323
access_memo = self._access.add_raw_records(
1324
[(record.key, len(bytes))], bytes)[0]
1325
index_entry = (record.key, options, access_memo, parents)
1327
if 'fulltext' not in options:
1328
basis_parent = parents[0]
1329
# Note that pack backed knits don't need to buffer here
1330
# because they buffer all writes to the transaction level,
1331
# but we don't expose that difference at the index level. If
1332
# the query here has sufficient cost to show up in
1333
# profiling we should do that.
1334
if basis_parent not in self.get_parent_map([basis_parent]):
1335
pending = buffered_index_entries.setdefault(
1337
pending.append(index_entry)
1340
self._index.add_records([index_entry])
1341
elif record.storage_kind == 'fulltext':
1342
self.add_lines(record.key, parents,
1343
split_lines(record.get_bytes_as('fulltext')))
1345
adapter_key = record.storage_kind, 'fulltext'
1346
adapter = get_adapter(adapter_key)
1347
lines = split_lines(adapter.get_bytes(
1348
record, record.get_bytes_as(record.storage_kind)))
1350
self.add_lines(record.key, parents, lines)
1351
except errors.RevisionAlreadyPresent:
1353
# Add any records whose basis parent is now available.
1354
added_keys = [record.key]
1356
key = added_keys.pop(0)
1357
if key in buffered_index_entries:
1358
index_entries = buffered_index_entries[key]
1359
self._index.add_records(index_entries)
1361
[index_entry[0] for index_entry in index_entries])
1362
del buffered_index_entries[key]
1363
# If there were any deltas which had a missing basis parent, error.
1364
if buffered_index_entries:
1365
raise errors.RevisionNotPresent(buffered_index_entries.keys()[0],
1368
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1369
"""Iterate over the lines in the versioned files from keys.
1371
This may return lines from other keys. Each item the returned
1372
iterator yields is a tuple of a line and a text version that that line
1373
is present in (not introduced in).
1375
Ordering of results is in whatever order is most suitable for the
1376
underlying storage format.
1378
If a progress bar is supplied, it may be used to indicate progress.
1379
The caller is responsible for cleaning up progress bars (because this
1383
* Lines are normalised by the underlying store: they will all have \n
1385
* Lines are returned in arbitrary order.
1387
:return: An iterator over (line, key).
1390
pb = progress.DummyProgress()
1393
# we don't care about inclusions, the caller cares.
1394
# but we need to setup a list of records to visit.
1395
# we need key, position, length
1397
build_details = self._index.get_build_details(keys)
1398
for key, details in build_details.iteritems():
1400
key_records.append((key, details[0]))
1402
records_iter = enumerate(self._read_records_iter(key_records))
1403
for (key_idx, (key, data, sha_value)) in records_iter:
1404
pb.update('Walking content.', key_idx, total)
1405
compression_parent = build_details[key][1]
1406
if compression_parent is None:
1408
line_iterator = self._factory.get_fulltext_content(data)
1411
line_iterator = self._factory.get_linedelta_content(data)
1412
# XXX: It might be more efficient to yield (key,
1413
# line_iterator) in the future. However for now, this is a simpler
1414
# change to integrate into the rest of the codebase. RBC 20071110
1415
for line in line_iterator:
1417
for source in self._fallback_vfs:
1421
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1422
source_keys.add(key)
1424
keys.difference_update(source_keys)
1426
raise RevisionNotPresent(keys, self.filename)
1427
pb.update('Walking content.', total, total)
1429
def _make_line_delta(self, delta_seq, new_content):
1430
"""Generate a line delta from delta_seq and new_content."""
1432
for op in delta_seq.get_opcodes():
1433
if op[0] == 'equal':
1435
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1438
def _merge_annotations(self, content, parents, parent_texts={},
1439
delta=None, annotated=None,
1440
left_matching_blocks=None):
1441
"""Merge annotations for content and generate deltas.
1443
This is done by comparing the annotations based on changes to the text
1444
and generating a delta on the resulting full texts. If annotations are
1445
not being created then a simple delta is created.
1447
if left_matching_blocks is not None:
1448
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1452
for parent_key in parents:
1453
merge_content = self._get_content(parent_key, parent_texts)
1454
if (parent_key == parents[0] and delta_seq is not None):
1457
seq = patiencediff.PatienceSequenceMatcher(
1458
None, merge_content.text(), content.text())
1459
for i, j, n in seq.get_matching_blocks():
1462
# this copies (origin, text) pairs across to the new
1463
# content for any line that matches the last-checked
1465
content._lines[j:j+n] = merge_content._lines[i:i+n]
1466
# XXX: Robert says the following block is a workaround for a
1467
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1468
if content._lines and content._lines[-1][1][-1] != '\n':
1469
# The copied annotation was from a line without a trailing EOL,
1470
# reinstate one for the content object, to ensure correct
1472
line = content._lines[-1][1] + '\n'
1473
content._lines[-1] = (content._lines[-1][0], line)
1475
if delta_seq is None:
1476
reference_content = self._get_content(parents[0], parent_texts)
1477
new_texts = content.text()
1478
old_texts = reference_content.text()
1479
delta_seq = patiencediff.PatienceSequenceMatcher(
1480
None, old_texts, new_texts)
1481
return self._make_line_delta(delta_seq, content)
1483
def _parse_record(self, version_id, data):
1484
"""Parse an original format knit record.
1486
These have the last element of the key only present in the stored data.
1488
rec, record_contents = self._parse_record_unchecked(data)
1489
self._check_header_version(rec, version_id)
1490
return record_contents, rec[3]
1492
def _parse_record_header(self, key, raw_data):
1493
"""Parse a record header for consistency.
1495
:return: the header and the decompressor stream.
1496
as (stream, header_record)
1498
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1501
rec = self._check_header(key, df.readline())
1502
except Exception, e:
1503
raise KnitCorrupt(self,
1504
"While reading {%s} got %s(%s)"
1505
% (key, e.__class__.__name__, str(e)))
1508
def _parse_record_unchecked(self, data):
1510
# 4168 calls in 2880 217 internal
1511
# 4168 calls to _parse_record_header in 2121
1512
# 4168 calls to readlines in 330
1513
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1515
record_contents = df.readlines()
1516
except Exception, e:
1517
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1518
(data, e.__class__.__name__, str(e)))
1519
header = record_contents.pop(0)
1520
rec = self._split_header(header)
1521
last_line = record_contents.pop()
1522
if len(record_contents) != int(rec[2]):
1523
raise KnitCorrupt(self,
1524
'incorrect number of lines %s != %s'
1525
' for version {%s} %s'
1526
% (len(record_contents), int(rec[2]),
1527
rec[1], record_contents))
1528
if last_line != 'end %s\n' % rec[1]:
1529
raise KnitCorrupt(self,
1530
'unexpected version end line %r, wanted %r'
1531
% (last_line, rec[1]))
1533
return rec, record_contents
1535
def _read_records_iter(self, records):
1536
"""Read text records from data file and yield result.
1538
The result will be returned in whatever is the fastest to read.
1539
Not by the order requested. Also, multiple requests for the same
1540
record will only yield 1 response.
1541
:param records: A list of (key, access_memo) entries
1542
:return: Yields (key, contents, digest) in the order
1543
read, not the order requested
1548
# XXX: This smells wrong, IO may not be getting ordered right.
1549
needed_records = sorted(set(records), key=operator.itemgetter(1))
1550
if not needed_records:
1553
# The transport optimizes the fetching as well
1554
# (ie, reads continuous ranges.)
1555
raw_data = self._access.get_raw_records(
1556
[index_memo for key, index_memo in needed_records])
1558
for (key, index_memo), data in \
1559
izip(iter(needed_records), raw_data):
1560
content, digest = self._parse_record(key[-1], data)
1561
yield key, content, digest
1563
def _read_records_iter_raw(self, records):
1564
"""Read text records from data file and yield raw data.
1566
This unpacks enough of the text record to validate the id is
1567
as expected but thats all.
1569
Each item the iterator yields is (key, bytes, sha1_of_full_text).
1571
# setup an iterator of the external records:
1572
# uses readv so nice and fast we hope.
1574
# grab the disk data needed.
1575
needed_offsets = [index_memo for key, index_memo
1577
raw_records = self._access.get_raw_records(needed_offsets)
1579
for key, index_memo in records:
1580
data = raw_records.next()
1581
# validate the header (note that we can only use the suffix in
1582
# current knit records).
1583
df, rec = self._parse_record_header(key, data)
1585
yield key, data, rec[3]
1587
def _record_to_data(self, key, digest, lines, dense_lines=None):
1588
"""Convert key, digest, lines into a raw data block.
1590
:param key: The key of the record. Currently keys are always serialised
1591
using just the trailing component.
1592
:param dense_lines: The bytes of lines but in a denser form. For
1593
instance, if lines is a list of 1000 bytestrings each ending in \n,
1594
dense_lines may be a list with one line in it, containing all the
1595
1000's lines and their \n's. Using dense_lines if it is already
1596
known is a win because the string join to create bytes in this
1597
function spends less time resizing the final string.
1598
:return: (len, a StringIO instance with the raw data ready to read.)
1600
# Note: using a string copy here increases memory pressure with e.g.
1601
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
1602
# when doing the initial commit of a mozilla tree. RBC 20070921
1603
bytes = ''.join(chain(
1604
["version %s %d %s\n" % (key[-1],
1607
dense_lines or lines,
1608
["end %s\n" % key[-1]]))
1609
if type(bytes) != str:
1610
raise AssertionError(
1611
'data must be plain bytes was %s' % type(bytes))
1612
if lines and lines[-1][-1] != '\n':
1613
raise ValueError('corrupt lines value %r' % lines)
1614
compressed_bytes = tuned_gzip.bytes_to_gzip(bytes)
1615
return len(compressed_bytes), compressed_bytes
1617
def _split_header(self, line):
1620
raise KnitCorrupt(self,
1621
'unexpected number of elements in record header')
1625
"""See VersionedFiles.keys."""
1626
if 'evil' in debug.debug_flags:
1627
trace.mutter_callsite(2, "keys scales with size of history")
1628
sources = [self._index] + self._fallback_vfs
1630
for source in sources:
1631
result.update(source.keys())
1636
class _KndxIndex(object):
1637
"""Manages knit index files
1639
The index is kept in memory and read on startup, to enable
1640
fast lookups of revision information. The cursor of the index
1641
file is always pointing to the end, making it easy to append
1644
_cache is a cache for fast mapping from version id to a Index
1647
_history is a cache for fast mapping from indexes to version ids.
1649
The index data format is dictionary compressed when it comes to
1650
parent references; a index entry may only have parents that with a
1651
lover index number. As a result, the index is topological sorted.
1653
Duplicate entries may be written to the index for a single version id
1654
if this is done then the latter one completely replaces the former:
1655
this allows updates to correct version and parent information.
1656
Note that the two entries may share the delta, and that successive
1657
annotations and references MUST point to the first entry.
1659
The index file on disc contains a header, followed by one line per knit
1660
record. The same revision can be present in an index file more than once.
1661
The first occurrence gets assigned a sequence number starting from 0.
1663
The format of a single line is
1664
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1665
REVISION_ID is a utf8-encoded revision id
1666
FLAGS is a comma separated list of flags about the record. Values include
1667
no-eol, line-delta, fulltext.
1668
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1669
that the the compressed data starts at.
1670
LENGTH is the ascii representation of the length of the data file.
1671
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1673
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1674
revision id already in the knit that is a parent of REVISION_ID.
1675
The ' :' marker is the end of record marker.
1678
when a write is interrupted to the index file, it will result in a line
1679
that does not end in ' :'. If the ' :' is not present at the end of a line,
1680
or at the end of the file, then the record that is missing it will be
1681
ignored by the parser.
1683
When writing new records to the index file, the data is preceded by '\n'
1684
to ensure that records always start on new lines even if the last write was
1685
interrupted. As a result its normal for the last line in the index to be
1686
missing a trailing newline. One can be added with no harmful effects.
1688
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
1689
where prefix is e.g. the (fileid,) for .texts instances or () for
1690
constant-mapped things like .revisions, and the old state is
1691
tuple(history_vector, cache_dict). This is used to prevent having an
1692
ABI change with the C extension that reads .kndx files.
1695
HEADER = "# bzr knit index 8\n"
1697
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
1698
"""Create a _KndxIndex on transport using mapper."""
1699
self._transport = transport
1700
self._mapper = mapper
1701
self._get_scope = get_scope
1702
self._allow_writes = allow_writes
1703
self._is_locked = is_locked
1705
self.has_graph = True
1707
def add_records(self, records, random_id=False):
1708
"""Add multiple records to the index.
1710
:param records: a list of tuples:
1711
(key, options, access_memo, parents).
1712
:param random_id: If True the ids being added were randomly generated
1713
and no check for existence will be performed.
1716
for record in records:
1719
path = self._mapper.map(key) + '.kndx'
1720
path_keys = paths.setdefault(path, (prefix, []))
1721
path_keys[1].append(record)
1722
for path in sorted(paths):
1723
prefix, path_keys = paths[path]
1724
self._load_prefixes([prefix])
1726
orig_history = self._kndx_cache[prefix][1][:]
1727
orig_cache = self._kndx_cache[prefix][0].copy()
1730
for key, options, (_, pos, size), parents in path_keys:
1732
# kndx indices cannot be parentless.
1734
line = "\n%s %s %s %s %s :" % (
1735
key[-1], ','.join(options), pos, size,
1736
self._dictionary_compress(parents))
1737
if type(line) != str:
1738
raise AssertionError(
1739
'data must be utf8 was %s' % type(line))
1741
self._cache_key(key, options, pos, size, parents)
1742
if len(orig_history):
1743
self._transport.append_bytes(path, ''.join(lines))
1745
self._init_index(path, lines)
1747
# If any problems happen, restore the original values and re-raise
1748
self._kndx_cache[prefix] = (orig_cache, orig_history)
1751
def _cache_key(self, key, options, pos, size, parent_keys):
1752
"""Cache a version record in the history array and index cache.
1754
This is inlined into _load_data for performance. KEEP IN SYNC.
1755
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1759
version_id = key[-1]
1760
# last-element only for compatibilty with the C load_data.
1761
parents = tuple(parent[-1] for parent in parent_keys)
1762
for parent in parent_keys:
1763
if parent[:-1] != prefix:
1764
raise ValueError("mismatched prefixes for %r, %r" % (
1766
cache, history = self._kndx_cache[prefix]
1767
# only want the _history index to reference the 1st index entry
1769
if version_id not in cache:
1770
index = len(history)
1771
history.append(version_id)
1773
index = cache[version_id][5]
1774
cache[version_id] = (version_id,
1781
def check_header(self, fp):
1782
line = fp.readline()
1784
# An empty file can actually be treated as though the file doesn't
1786
raise errors.NoSuchFile(self)
1787
if line != self.HEADER:
1788
raise KnitHeaderError(badline=line, filename=self)
1790
def _check_read(self):
1791
if not self._is_locked():
1792
raise errors.ObjectNotLocked(self)
1793
if self._get_scope() != self._scope:
1796
def _check_write_ok(self):
1797
"""Assert if not writes are permitted."""
1798
if not self._is_locked():
1799
raise errors.ObjectNotLocked(self)
1800
if self._get_scope() != self._scope:
1802
if self._mode != 'w':
1803
raise errors.ReadOnlyObjectDirtiedError(self)
1805
def get_build_details(self, keys):
1806
"""Get the method, index_memo and compression parent for keys.
1808
Ghosts are omitted from the result.
1810
:param keys: An iterable of keys.
1811
:return: A dict of key:(index_memo, compression_parent, parents,
1814
opaque structure to pass to read_records to extract the raw
1817
Content that this record is built upon, may be None
1819
Logical parents of this node
1821
extra information about the content which needs to be passed to
1822
Factory.parse_record
1824
prefixes = self._partition_keys(keys)
1825
parent_map = self.get_parent_map(keys)
1828
if key not in parent_map:
1830
method = self.get_method(key)
1831
parents = parent_map[key]
1832
if method == 'fulltext':
1833
compression_parent = None
1835
compression_parent = parents[0]
1836
noeol = 'no-eol' in self.get_options(key)
1837
index_memo = self.get_position(key)
1838
result[key] = (index_memo, compression_parent,
1839
parents, (method, noeol))
1842
def get_method(self, key):
1843
"""Return compression method of specified key."""
1844
options = self.get_options(key)
1845
if 'fulltext' in options:
1847
elif 'line-delta' in options:
1850
raise errors.KnitIndexUnknownMethod(self, options)
1852
def get_options(self, key):
1853
"""Return a list representing options.
1857
prefix, suffix = self._split_key(key)
1858
self._load_prefixes([prefix])
1860
return self._kndx_cache[prefix][0][suffix][1]
1862
raise RevisionNotPresent(key, self)
1864
def get_parent_map(self, keys):
1865
"""Get a map of the parents of keys.
1867
:param keys: The keys to look up parents for.
1868
:return: A mapping from keys to parents. Absent keys are absent from
1871
# Parse what we need to up front, this potentially trades off I/O
1872
# locality (.kndx and .knit in the same block group for the same file
1873
# id) for less checking in inner loops.
1874
prefixes = set(key[:-1] for key in keys)
1875
self._load_prefixes(prefixes)
1880
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
1884
result[key] = tuple(prefix + (suffix,) for
1885
suffix in suffix_parents)
1888
def get_position(self, key):
1889
"""Return details needed to access the version.
1891
:return: a tuple (key, data position, size) to hand to the access
1892
logic to get the record.
1894
prefix, suffix = self._split_key(key)
1895
self._load_prefixes([prefix])
1896
entry = self._kndx_cache[prefix][0][suffix]
1897
return key, entry[2], entry[3]
1899
def _init_index(self, path, extra_lines=[]):
1900
"""Initialize an index."""
1902
sio.write(self.HEADER)
1903
sio.writelines(extra_lines)
1905
self._transport.put_file_non_atomic(path, sio,
1906
create_parent_dir=True)
1907
# self._create_parent_dir)
1908
# mode=self._file_mode,
1909
# dir_mode=self._dir_mode)
1912
"""Get all the keys in the collection.
1914
The keys are not ordered.
1917
# Identify all key prefixes.
1918
# XXX: A bit hacky, needs polish.
1919
if type(self._mapper) == ConstantMapper:
1923
for quoted_relpath in self._transport.iter_files_recursive():
1924
path, ext = os.path.splitext(quoted_relpath)
1926
prefixes = [self._mapper.unmap(path) for path in relpaths]
1927
self._load_prefixes(prefixes)
1928
for prefix in prefixes:
1929
for suffix in self._kndx_cache[prefix][1]:
1930
result.add(prefix + (suffix,))
1933
def _load_prefixes(self, prefixes):
1934
"""Load the indices for prefixes."""
1936
for prefix in prefixes:
1937
if prefix not in self._kndx_cache:
1938
# the load_data interface writes to these variables.
1941
self._filename = prefix
1943
path = self._mapper.map(prefix) + '.kndx'
1944
fp = self._transport.get(path)
1946
# _load_data may raise NoSuchFile if the target knit is
1948
_load_data(self, fp)
1951
self._kndx_cache[prefix] = (self._cache, self._history)
1956
self._kndx_cache[prefix] = ({}, [])
1957
if type(self._mapper) == ConstantMapper:
1958
# preserve behaviour for revisions.kndx etc.
1959
self._init_index(path)
1964
def _partition_keys(self, keys):
1965
"""Turn keys into a dict of prefix:suffix_list."""
1968
prefix_keys = result.setdefault(key[:-1], [])
1969
prefix_keys.append(key[-1])
1972
def _dictionary_compress(self, keys):
1973
"""Dictionary compress keys.
1975
:param keys: The keys to generate references to.
1976
:return: A string representation of keys. keys which are present are
1977
dictionary compressed, and others are emitted as fulltext with a
1983
prefix = keys[0][:-1]
1984
cache = self._kndx_cache[prefix][0]
1986
if key[:-1] != prefix:
1987
# kndx indices cannot refer across partitioned storage.
1988
raise ValueError("mismatched prefixes for %r" % keys)
1989
if key[-1] in cache:
1990
# -- inlined lookup() --
1991
result_list.append(str(cache[key[-1]][5]))
1992
# -- end lookup () --
1994
result_list.append('.' + key[-1])
1995
return ' '.join(result_list)
1997
def _reset_cache(self):
1998
# Possibly this should be a LRU cache. A dictionary from key_prefix to
1999
# (cache_dict, history_vector) for parsed kndx files.
2000
self._kndx_cache = {}
2001
self._scope = self._get_scope()
2002
allow_writes = self._allow_writes()
2008
def _split_key(self, key):
2009
"""Split key into a prefix and suffix."""
2010
return key[:-1], key[-1]
2013
class _KnitGraphIndex(object):
2014
"""A KnitVersionedFiles index layered on GraphIndex."""
2016
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2018
"""Construct a KnitGraphIndex on a graph_index.
2020
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2021
:param is_locked: A callback to check whether the object should answer
2023
:param deltas: Allow delta-compressed records.
2024
:param parents: If True, record knits parents, if not do not record
2026
:param add_callback: If not None, allow additions to the index and call
2027
this callback with a list of added GraphIndex nodes:
2028
[(node, value, node_refs), ...]
2029
:param is_locked: A callback, returns True if the index is locked and
2032
self._add_callback = add_callback
2033
self._graph_index = graph_index
2034
self._deltas = deltas
2035
self._parents = parents
2036
if deltas and not parents:
2037
# XXX: TODO: Delta tree and parent graph should be conceptually
2039
raise KnitCorrupt(self, "Cannot do delta compression without "
2041
self.has_graph = parents
2042
self._is_locked = is_locked
2045
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2047
def add_records(self, records, random_id=False):
2048
"""Add multiple records to the index.
2050
This function does not insert data into the Immutable GraphIndex
2051
backing the KnitGraphIndex, instead it prepares data for insertion by
2052
the caller and checks that it is safe to insert then calls
2053
self._add_callback with the prepared GraphIndex nodes.
2055
:param records: a list of tuples:
2056
(key, options, access_memo, parents).
2057
:param random_id: If True the ids being added were randomly generated
2058
and no check for existence will be performed.
2060
if not self._add_callback:
2061
raise errors.ReadOnlyError(self)
2062
# we hope there are no repositories with inconsistent parentage
2066
for (key, options, access_memo, parents) in records:
2068
parents = tuple(parents)
2069
index, pos, size = access_memo
2070
if 'no-eol' in options:
2074
value += "%d %d" % (pos, size)
2075
if not self._deltas:
2076
if 'line-delta' in options:
2077
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2080
if 'line-delta' in options:
2081
node_refs = (parents, (parents[0],))
2083
node_refs = (parents, ())
2085
node_refs = (parents, )
2088
raise KnitCorrupt(self, "attempt to add node with parents "
2089
"in parentless index.")
2091
keys[key] = (value, node_refs)
2094
present_nodes = self._get_entries(keys)
2095
for (index, key, value, node_refs) in present_nodes:
2096
if (value[0] != keys[key][0][0] or
2097
node_refs != keys[key][1]):
2098
raise KnitCorrupt(self, "inconsistent details in add_records"
2099
": %s %s" % ((value, node_refs), keys[key]))
2103
for key, (value, node_refs) in keys.iteritems():
2104
result.append((key, value, node_refs))
2106
for key, (value, node_refs) in keys.iteritems():
2107
result.append((key, value))
2108
self._add_callback(result)
2110
def _check_read(self):
2111
"""raise if reads are not permitted."""
2112
if not self._is_locked():
2113
raise errors.ObjectNotLocked(self)
2115
def _check_write_ok(self):
2116
"""Assert if writes are not permitted."""
2117
if not self._is_locked():
2118
raise errors.ObjectNotLocked(self)
2120
def _compression_parent(self, an_entry):
2121
# return the key that an_entry is compressed against, or None
2122
# Grab the second parent list (as deltas implies parents currently)
2123
compression_parents = an_entry[3][1]
2124
if not compression_parents:
2126
if len(compression_parents) != 1:
2127
raise AssertionError(
2128
"Too many compression parents: %r" % compression_parents)
2129
return compression_parents[0]
2131
def get_build_details(self, keys):
2132
"""Get the method, index_memo and compression parent for version_ids.
2134
Ghosts are omitted from the result.
2136
:param keys: An iterable of keys.
2137
:return: A dict of key:
2138
(index_memo, compression_parent, parents, record_details).
2140
opaque structure to pass to read_records to extract the raw
2143
Content that this record is built upon, may be None
2145
Logical parents of this node
2147
extra information about the content which needs to be passed to
2148
Factory.parse_record
2152
entries = self._get_entries(keys, False)
2153
for entry in entries:
2155
if not self._parents:
2158
parents = entry[3][0]
2159
if not self._deltas:
2160
compression_parent_key = None
2162
compression_parent_key = self._compression_parent(entry)
2163
noeol = (entry[2][0] == 'N')
2164
if compression_parent_key:
2165
method = 'line-delta'
2168
result[key] = (self._node_to_position(entry),
2169
compression_parent_key, parents,
2173
def _get_entries(self, keys, check_present=False):
2174
"""Get the entries for keys.
2176
:param keys: An iterable of index key tuples.
2181
for node in self._graph_index.iter_entries(keys):
2183
found_keys.add(node[1])
2185
# adapt parentless index to the rest of the code.
2186
for node in self._graph_index.iter_entries(keys):
2187
yield node[0], node[1], node[2], ()
2188
found_keys.add(node[1])
2190
missing_keys = keys.difference(found_keys)
2192
raise RevisionNotPresent(missing_keys.pop(), self)
2194
def get_method(self, key):
2195
"""Return compression method of specified key."""
2196
return self._get_method(self._get_node(key))
2198
def _get_method(self, node):
2199
if not self._deltas:
2201
if self._compression_parent(node):
2206
def _get_node(self, key):
2208
return list(self._get_entries([key]))[0]
2210
raise RevisionNotPresent(key, self)
2212
def get_options(self, key):
2213
"""Return a list representing options.
2217
node = self._get_node(key)
2218
options = [self._get_method(node)]
2219
if node[2][0] == 'N':
2220
options.append('no-eol')
2223
def get_parent_map(self, keys):
2224
"""Get a map of the parents of keys.
2226
:param keys: The keys to look up parents for.
2227
:return: A mapping from keys to parents. Absent keys are absent from
2231
nodes = self._get_entries(keys)
2235
result[node[1]] = node[3][0]
2238
result[node[1]] = None
2241
def get_position(self, key):
2242
"""Return details needed to access the version.
2244
:return: a tuple (index, data position, size) to hand to the access
2245
logic to get the record.
2247
node = self._get_node(key)
2248
return self._node_to_position(node)
2251
"""Get all the keys in the collection.
2253
The keys are not ordered.
2256
return [node[1] for node in self._graph_index.iter_all_entries()]
2258
def _node_to_position(self, node):
2259
"""Convert an index value to position details."""
2260
bits = node[2][1:].split(' ')
2261
return node[0], int(bits[0]), int(bits[1])
2264
class _KnitKeyAccess(object):
2265
"""Access to records in .knit files."""
2267
def __init__(self, transport, mapper):
2268
"""Create a _KnitKeyAccess with transport and mapper.
2270
:param transport: The transport the access object is rooted at.
2271
:param mapper: The mapper used to map keys to .knit files.
2273
self._transport = transport
2274
self._mapper = mapper
2276
def add_raw_records(self, key_sizes, raw_data):
2277
"""Add raw knit bytes to a storage area.
2279
The data is spooled to the container writer in one bytes-record per
2282
:param sizes: An iterable of tuples containing the key and size of each
2284
:param raw_data: A bytestring containing the data.
2285
:return: A list of memos to retrieve the record later. Each memo is an
2286
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
2287
length), where the key is the record key.
2289
if type(raw_data) != str:
2290
raise AssertionError(
2291
'data must be plain bytes was %s' % type(raw_data))
2294
# TODO: This can be tuned for writing to sftp and other servers where
2295
# append() is relatively expensive by grouping the writes to each key
2297
for key, size in key_sizes:
2298
path = self._mapper.map(key)
2300
base = self._transport.append_bytes(path + '.knit',
2301
raw_data[offset:offset+size])
2302
except errors.NoSuchFile:
2303
self._transport.mkdir(osutils.dirname(path))
2304
base = self._transport.append_bytes(path + '.knit',
2305
raw_data[offset:offset+size])
2309
result.append((key, base, size))
2312
def get_raw_records(self, memos_for_retrieval):
2313
"""Get the raw bytes for a records.
2315
:param memos_for_retrieval: An iterable containing the access memo for
2316
retrieving the bytes.
2317
:return: An iterator over the bytes of the records.
2319
# first pass, group into same-index request to minimise readv's issued.
2321
current_prefix = None
2322
for (key, offset, length) in memos_for_retrieval:
2323
if current_prefix == key[:-1]:
2324
current_list.append((offset, length))
2326
if current_prefix is not None:
2327
request_lists.append((current_prefix, current_list))
2328
current_prefix = key[:-1]
2329
current_list = [(offset, length)]
2330
# handle the last entry
2331
if current_prefix is not None:
2332
request_lists.append((current_prefix, current_list))
2333
for prefix, read_vector in request_lists:
2334
path = self._mapper.map(prefix) + '.knit'
2335
for pos, data in self._transport.readv(path, read_vector):
2339
class _DirectPackAccess(object):
2340
"""Access to data in one or more packs with less translation."""
2342
def __init__(self, index_to_packs):
2343
"""Create a _DirectPackAccess object.
2345
:param index_to_packs: A dict mapping index objects to the transport
2346
and file names for obtaining data.
2348
self._container_writer = None
2349
self._write_index = None
2350
self._indices = index_to_packs
2352
def add_raw_records(self, key_sizes, raw_data):
2353
"""Add raw knit bytes to a storage area.
2355
The data is spooled to the container writer in one bytes-record per
2358
:param sizes: An iterable of tuples containing the key and size of each
2360
:param raw_data: A bytestring containing the data.
2361
:return: A list of memos to retrieve the record later. Each memo is an
2362
opaque index memo. For _DirectPackAccess the memo is (index, pos,
2363
length), where the index field is the write_index object supplied
2364
to the PackAccess object.
2366
if type(raw_data) != str:
2367
raise AssertionError(
2368
'data must be plain bytes was %s' % type(raw_data))
2371
for key, size in key_sizes:
2372
p_offset, p_length = self._container_writer.add_bytes_record(
2373
raw_data[offset:offset+size], [])
2375
result.append((self._write_index, p_offset, p_length))
2378
def get_raw_records(self, memos_for_retrieval):
2379
"""Get the raw bytes for a records.
2381
:param memos_for_retrieval: An iterable containing the (index, pos,
2382
length) memo for retrieving the bytes. The Pack access method
2383
looks up the pack to use for a given record in its index_to_pack
2385
:return: An iterator over the bytes of the records.
2387
# first pass, group into same-index requests
2389
current_index = None
2390
for (index, offset, length) in memos_for_retrieval:
2391
if current_index == index:
2392
current_list.append((offset, length))
2394
if current_index is not None:
2395
request_lists.append((current_index, current_list))
2396
current_index = index
2397
current_list = [(offset, length)]
2398
# handle the last entry
2399
if current_index is not None:
2400
request_lists.append((current_index, current_list))
2401
for index, offsets in request_lists:
2402
transport, path = self._indices[index]
2403
reader = pack.make_readv_reader(transport, path, offsets)
2404
for names, read_func in reader.iter_records():
2405
yield read_func(None)
2407
def set_writer(self, writer, index, transport_packname):
2408
"""Set a writer to use for adding data."""
2409
if index is not None:
2410
self._indices[index] = transport_packname
2411
self._container_writer = writer
2412
self._write_index = index
2415
# Deprecated, use PatienceSequenceMatcher instead
2416
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2419
def annotate_knit(knit, revision_id):
2420
"""Annotate a knit with no cached annotations.
2422
This implementation is for knits with no cached annotations.
2423
It will work for knits with cached annotations, but this is not
2426
annotator = _KnitAnnotator(knit)
2427
return iter(annotator.annotate(revision_id))
2430
class _KnitAnnotator(object):
2431
"""Build up the annotations for a text."""
2433
def __init__(self, knit):
2436
# Content objects, differs from fulltexts because of how final newlines
2437
# are treated by knits. the content objects here will always have a
2439
self._fulltext_contents = {}
2441
# Annotated lines of specific revisions
2442
self._annotated_lines = {}
2444
# Track the raw data for nodes that we could not process yet.
2445
# This maps the revision_id of the base to a list of children that will
2446
# annotated from it.
2447
self._pending_children = {}
2449
# Nodes which cannot be extracted
2450
self._ghosts = set()
2452
# Track how many children this node has, so we know if we need to keep
2454
self._annotate_children = {}
2455
self._compression_children = {}
2457
self._all_build_details = {}
2458
# The children => parent revision_id graph
2459
self._revision_id_graph = {}
2461
self._heads_provider = None
2463
self._nodes_to_keep_annotations = set()
2464
self._generations_until_keep = 100
2466
def set_generations_until_keep(self, value):
2467
"""Set the number of generations before caching a node.
2469
Setting this to -1 will cache every merge node, setting this higher
2470
will cache fewer nodes.
2472
self._generations_until_keep = value
2474
def _add_fulltext_content(self, revision_id, content_obj):
2475
self._fulltext_contents[revision_id] = content_obj
2476
# TODO: jam 20080305 It might be good to check the sha1digest here
2477
return content_obj.text()
2479
def _check_parents(self, child, nodes_to_annotate):
2480
"""Check if all parents have been processed.
2482
:param child: A tuple of (rev_id, parents, raw_content)
2483
:param nodes_to_annotate: If child is ready, add it to
2484
nodes_to_annotate, otherwise put it back in self._pending_children
2486
for parent_id in child[1]:
2487
if (parent_id not in self._annotated_lines):
2488
# This parent is present, but another parent is missing
2489
self._pending_children.setdefault(parent_id,
2493
# This one is ready to be processed
2494
nodes_to_annotate.append(child)
2496
def _add_annotation(self, revision_id, fulltext, parent_ids,
2497
left_matching_blocks=None):
2498
"""Add an annotation entry.
2500
All parents should already have been annotated.
2501
:return: A list of children that now have their parents satisfied.
2503
a = self._annotated_lines
2504
annotated_parent_lines = [a[p] for p in parent_ids]
2505
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2506
fulltext, revision_id, left_matching_blocks,
2507
heads_provider=self._get_heads_provider()))
2508
self._annotated_lines[revision_id] = annotated_lines
2509
for p in parent_ids:
2510
ann_children = self._annotate_children[p]
2511
ann_children.remove(revision_id)
2512
if (not ann_children
2513
and p not in self._nodes_to_keep_annotations):
2514
del self._annotated_lines[p]
2515
del self._all_build_details[p]
2516
if p in self._fulltext_contents:
2517
del self._fulltext_contents[p]
2518
# Now that we've added this one, see if there are any pending
2519
# deltas to be done, certainly this parent is finished
2520
nodes_to_annotate = []
2521
for child in self._pending_children.pop(revision_id, []):
2522
self._check_parents(child, nodes_to_annotate)
2523
return nodes_to_annotate
2525
def _get_build_graph(self, key):
2526
"""Get the graphs for building texts and annotations.
2528
The data you need for creating a full text may be different than the
2529
data you need to annotate that text. (At a minimum, you need both
2530
parents to create an annotation, but only need 1 parent to generate the
2533
:return: A list of (key, index_memo) records, suitable for
2534
passing to read_records_iter to start reading in the raw data fro/
2537
if key in self._annotated_lines:
2540
pending = set([key])
2545
# get all pending nodes
2547
this_iteration = pending
2548
build_details = self._knit._index.get_build_details(this_iteration)
2549
self._all_build_details.update(build_details)
2550
# new_nodes = self._knit._index._get_entries(this_iteration)
2552
for key, details in build_details.iteritems():
2553
(index_memo, compression_parent, parents,
2554
record_details) = details
2555
self._revision_id_graph[key] = parents
2556
records.append((key, index_memo))
2557
# Do we actually need to check _annotated_lines?
2558
pending.update(p for p in parents
2559
if p not in self._all_build_details)
2560
if compression_parent:
2561
self._compression_children.setdefault(compression_parent,
2564
for parent in parents:
2565
self._annotate_children.setdefault(parent,
2567
num_gens = generation - kept_generation
2568
if ((num_gens >= self._generations_until_keep)
2569
and len(parents) > 1):
2570
kept_generation = generation
2571
self._nodes_to_keep_annotations.add(key)
2573
missing_versions = this_iteration.difference(build_details.keys())
2574
self._ghosts.update(missing_versions)
2575
for missing_version in missing_versions:
2576
# add a key, no parents
2577
self._revision_id_graph[missing_version] = ()
2578
pending.discard(missing_version) # don't look for it
2579
if self._ghosts.intersection(self._compression_children):
2581
"We cannot have nodes which have a ghost compression parent:\n"
2583
"compression children: %r"
2584
% (self._ghosts, self._compression_children))
2585
# Cleanout anything that depends on a ghost so that we don't wait for
2586
# the ghost to show up
2587
for node in self._ghosts:
2588
if node in self._annotate_children:
2589
# We won't be building this node
2590
del self._annotate_children[node]
2591
# Generally we will want to read the records in reverse order, because
2592
# we find the parent nodes after the children
2596
def _annotate_records(self, records):
2597
"""Build the annotations for the listed records."""
2598
# We iterate in the order read, rather than a strict order requested
2599
# However, process what we can, and put off to the side things that
2600
# still need parents, cleaning them up when those parents are
2602
for (rev_id, record,
2603
digest) in self._knit._read_records_iter(records):
2604
if rev_id in self._annotated_lines:
2606
parent_ids = self._revision_id_graph[rev_id]
2607
parent_ids = [p for p in parent_ids if p not in self._ghosts]
2608
details = self._all_build_details[rev_id]
2609
(index_memo, compression_parent, parents,
2610
record_details) = details
2611
nodes_to_annotate = []
2612
# TODO: Remove the punning between compression parents, and
2613
# parent_ids, we should be able to do this without assuming
2615
if len(parent_ids) == 0:
2616
# There are no parents for this node, so just add it
2617
# TODO: This probably needs to be decoupled
2618
fulltext_content, delta = self._knit._factory.parse_record(
2619
rev_id, record, record_details, None)
2620
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
2621
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
2622
parent_ids, left_matching_blocks=None))
2624
child = (rev_id, parent_ids, record)
2625
# Check if all the parents are present
2626
self._check_parents(child, nodes_to_annotate)
2627
while nodes_to_annotate:
2628
# Should we use a queue here instead of a stack?
2629
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
2630
(index_memo, compression_parent, parents,
2631
record_details) = self._all_build_details[rev_id]
2632
if compression_parent is not None:
2633
comp_children = self._compression_children[compression_parent]
2634
if rev_id not in comp_children:
2635
raise AssertionError("%r not in compression children %r"
2636
% (rev_id, comp_children))
2637
# If there is only 1 child, it is safe to reuse this
2639
reuse_content = (len(comp_children) == 1
2640
and compression_parent not in
2641
self._nodes_to_keep_annotations)
2643
# Remove it from the cache since it will be changing
2644
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
2645
# Make sure to copy the fulltext since it might be
2647
parent_fulltext = list(parent_fulltext_content.text())
2649
parent_fulltext_content = self._fulltext_contents[compression_parent]
2650
parent_fulltext = parent_fulltext_content.text()
2651
comp_children.remove(rev_id)
2652
fulltext_content, delta = self._knit._factory.parse_record(
2653
rev_id, record, record_details,
2654
parent_fulltext_content,
2655
copy_base_content=(not reuse_content))
2656
fulltext = self._add_fulltext_content(rev_id,
2658
blocks = KnitContent.get_line_delta_blocks(delta,
2659
parent_fulltext, fulltext)
2661
fulltext_content = self._knit._factory.parse_fulltext(
2663
fulltext = self._add_fulltext_content(rev_id,
2666
nodes_to_annotate.extend(
2667
self._add_annotation(rev_id, fulltext, parent_ids,
2668
left_matching_blocks=blocks))
2670
def _get_heads_provider(self):
2671
"""Create a heads provider for resolving ancestry issues."""
2672
if self._heads_provider is not None:
2673
return self._heads_provider
2674
parent_provider = _mod_graph.DictParentsProvider(
2675
self._revision_id_graph)
2676
graph_obj = _mod_graph.Graph(parent_provider)
2677
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
2678
self._heads_provider = head_cache
2681
def annotate(self, key):
2682
"""Return the annotated fulltext at the given key.
2684
:param key: The key to annotate.
2686
if True or len(self._knit._fallback_vfs) > 0:
2687
# stacked knits can't use the fast path at present.
2688
return self._simple_annotate(key)
2689
records = self._get_build_graph(key)
2690
if key in self._ghosts:
2691
raise errors.RevisionNotPresent(key, self._knit)
2692
self._annotate_records(records)
2693
return self._annotated_lines[key]
2695
def _simple_annotate(self, key):
2696
"""Return annotated fulltext, rediffing from the full texts.
2698
This is slow but makes no assumptions about the repository
2699
being able to produce line deltas.
2701
# TODO: this code generates a parent maps of present ancestors; it
2702
# could be split out into a separate method, and probably should use
2703
# iter_ancestry instead. -- mbp and robertc 20080704
2704
graph = _mod_graph.Graph(self._knit)
2705
head_cache = _mod_graph.FrozenHeadsCache(graph)
2706
search = graph._make_breadth_first_searcher([key])
2710
present, ghosts = search.next_with_ghosts()
2711
except StopIteration:
2713
keys.update(present)
2714
parent_map = self._knit.get_parent_map(keys)
2716
reannotate = annotate.reannotate
2717
for record in self._knit.get_record_stream(keys, 'topological', True):
2719
fulltext = split_lines(record.get_bytes_as('fulltext'))
2720
parents = parent_map[key]
2721
if parents is not None:
2722
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
2725
parent_cache[key] = list(
2726
reannotate(parent_lines, fulltext, key, None, head_cache))
2728
return parent_cache[key]
2730
raise errors.RevisionNotPresent(key, self._knit)
2734
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2736
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
