1
# Copyright (C) 2005, 2006, 2007, 2008 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
from cStringIO import StringIO
56
from itertools import izip, chain
61
from bzrlib.lazy_import import lazy_import
62
lazy_import(globals(), """
82
from bzrlib.errors import (
90
RevisionAlreadyPresent,
93
from bzrlib.osutils import (
100
from bzrlib.versionedfile import (
101
AbsentContentFactory,
105
ChunkedContentFactory,
111
# TODO: Split out code specific to this format into an associated object.
113
# TODO: Can we put in some kind of value to check that the index and data
114
# files belong together?
116
# TODO: accommodate binaries, perhaps by storing a byte count
118
# TODO: function to check whole file
120
# TODO: atomically append data, then measure backwards from the cursor
121
# position after writing to work out where it was located. we may need to
122
# bypass python file buffering.
124
DATA_SUFFIX = '.knit'
125
INDEX_SUFFIX = '.kndx'
126
_STREAM_MIN_BUFFER_SIZE = 5*1024*1024
129
class KnitAdapter(object):
130
"""Base class for knit record adaption."""
132
def __init__(self, basis_vf):
133
"""Create an adapter which accesses full texts from basis_vf.
135
:param basis_vf: A versioned file to access basis texts of deltas from.
136
May be None for adapters that do not need to access basis texts.
138
self._data = KnitVersionedFiles(None, None)
139
self._annotate_factory = KnitAnnotateFactory()
140
self._plain_factory = KnitPlainFactory()
141
self._basis_vf = basis_vf
144
class FTAnnotatedToUnannotated(KnitAdapter):
145
"""An adapter from FT annotated knits to unannotated ones."""
147
def get_bytes(self, factory):
148
annotated_compressed_bytes = factory._raw_record
150
self._data._parse_record_unchecked(annotated_compressed_bytes)
151
content = self._annotate_factory.parse_fulltext(contents, rec[1])
152
size, bytes = self._data._record_to_data((rec[1],), rec[3], content.text())
156
class DeltaAnnotatedToUnannotated(KnitAdapter):
157
"""An adapter for deltas from annotated to unannotated."""
159
def get_bytes(self, factory):
160
annotated_compressed_bytes = factory._raw_record
162
self._data._parse_record_unchecked(annotated_compressed_bytes)
163
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
165
contents = self._plain_factory.lower_line_delta(delta)
166
size, bytes = self._data._record_to_data((rec[1],), rec[3], contents)
170
class FTAnnotatedToFullText(KnitAdapter):
171
"""An adapter from FT annotated knits to unannotated ones."""
173
def get_bytes(self, factory):
174
annotated_compressed_bytes = factory._raw_record
176
self._data._parse_record_unchecked(annotated_compressed_bytes)
177
content, delta = self._annotate_factory.parse_record(factory.key[-1],
178
contents, factory._build_details, None)
179
return ''.join(content.text())
182
class DeltaAnnotatedToFullText(KnitAdapter):
183
"""An adapter for deltas from annotated to unannotated."""
185
def get_bytes(self, factory):
186
annotated_compressed_bytes = factory._raw_record
188
self._data._parse_record_unchecked(annotated_compressed_bytes)
189
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
191
compression_parent = factory.parents[0]
192
basis_entry = self._basis_vf.get_record_stream(
193
[compression_parent], 'unordered', True).next()
194
if basis_entry.storage_kind == 'absent':
195
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
196
basis_chunks = basis_entry.get_bytes_as('chunked')
197
basis_lines = osutils.chunks_to_lines(basis_chunks)
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):
210
compressed_bytes = factory._raw_record
212
self._data._parse_record_unchecked(compressed_bytes)
213
content, delta = self._plain_factory.parse_record(factory.key[-1],
214
contents, factory._build_details, None)
215
return ''.join(content.text())
218
class DeltaPlainToFullText(KnitAdapter):
219
"""An adapter for deltas from annotated to unannotated."""
221
def get_bytes(self, factory):
222
compressed_bytes = factory._raw_record
224
self._data._parse_record_unchecked(compressed_bytes)
225
delta = self._plain_factory.parse_line_delta(contents, rec[1])
226
compression_parent = factory.parents[0]
227
# XXX: string splitting overhead.
228
basis_entry = self._basis_vf.get_record_stream(
229
[compression_parent], 'unordered', True).next()
230
if basis_entry.storage_kind == 'absent':
231
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
232
basis_chunks = basis_entry.get_bytes_as('chunked')
233
basis_lines = osutils.chunks_to_lines(basis_chunks)
234
basis_content = PlainKnitContent(basis_lines, compression_parent)
235
# Manually apply the delta because we have one annotated content and
237
content, _ = self._plain_factory.parse_record(rec[1], contents,
238
factory._build_details, basis_content)
239
return ''.join(content.text())
242
class KnitContentFactory(ContentFactory):
243
"""Content factory for streaming from knits.
245
:seealso ContentFactory:
248
def __init__(self, key, parents, build_details, sha1, raw_record,
249
annotated, knit=None, network_bytes=None):
250
"""Create a KnitContentFactory for key.
253
:param parents: The parents.
254
:param build_details: The build details as returned from
256
:param sha1: The sha1 expected from the full text of this object.
257
:param raw_record: The bytes of the knit data from disk.
258
:param annotated: True if the raw data is annotated.
259
:param network_bytes: None to calculate the network bytes on demand,
260
not-none if they are already known.
262
ContentFactory.__init__(self)
265
self.parents = parents
266
if build_details[0] == 'line-delta':
271
annotated_kind = 'annotated-'
274
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
275
self._raw_record = raw_record
276
self._network_bytes = network_bytes
277
self._build_details = build_details
280
def _create_network_bytes(self):
281
"""Create a fully serialised network version for transmission."""
282
# storage_kind, key, parents, Noeol, raw_record
283
key_bytes = '\x00'.join(self.key)
284
if self.parents is None:
285
parent_bytes = 'None:'
287
parent_bytes = '\t'.join('\x00'.join(key) for key in self.parents)
288
if self._build_details[1]:
292
network_bytes = "%s\n%s\n%s\n%s%s" % (self.storage_kind, key_bytes,
293
parent_bytes, noeol, self._raw_record)
294
self._network_bytes = network_bytes
296
def get_bytes_as(self, storage_kind):
297
if storage_kind == self.storage_kind:
298
if self._network_bytes is None:
299
self._create_network_bytes()
300
return self._network_bytes
301
if self._knit is not None:
302
if storage_kind == 'chunked':
303
return self._knit.get_lines(self.key[0])
304
elif storage_kind == 'fulltext':
305
return self._knit.get_text(self.key[0])
306
raise errors.UnavailableRepresentation(self.key, storage_kind,
310
class LazyKnitContentFactory(ContentFactory):
311
"""A ContentFactory which can either generate full text or a wire form.
313
:seealso ContentFactory:
316
def __init__(self, key, parents, generator, first):
317
"""Create a LazyKnitContentFactory.
319
:param key: The key of the record.
320
:param parents: The parents of the record.
321
:param generator: A _ContentMapGenerator containing the record for this
323
:param first: Is this the first content object returned from generator?
324
if it is, its storage kind is knit-delta-closure, otherwise it is
325
knit-delta-closure-ref
328
self.parents = parents
330
self._generator = generator
331
self.storage_kind = "knit-delta-closure"
333
self.storage_kind = self.storage_kind + "-ref"
336
def get_bytes_as(self, storage_kind):
337
if storage_kind == self.storage_kind:
339
return self._generator._wire_bytes()
341
# all the keys etc are contained in the bytes returned in the
344
if storage_kind in ('chunked', 'fulltext'):
345
chunks = self._generator._get_one_work(self.key).text()
346
if storage_kind == 'chunked':
349
return ''.join(chunks)
350
raise errors.UnavailableRepresentation(self.key, storage_kind,
354
def knit_delta_closure_to_records(storage_kind, bytes, line_end):
355
"""Convert a network record to a iterator over stream records.
357
:param storage_kind: The storage kind of the record.
358
Must be 'knit-delta-closure'.
359
:param bytes: The bytes of the record on the network.
361
generator = _NetworkContentMapGenerator(bytes, line_end)
362
return generator.get_record_stream()
365
def knit_network_to_record(storage_kind, bytes, line_end):
366
"""Convert a network record to a record object.
368
:param storage_kind: The storage kind of the record.
369
:param bytes: The bytes of the record on the network.
372
line_end = bytes.find('\n', start)
373
key = tuple(bytes[start:line_end].split('\x00'))
375
line_end = bytes.find('\n', start)
376
parent_line = bytes[start:line_end]
377
if parent_line == 'None:':
381
[tuple(segment.split('\x00')) for segment in parent_line.split('\t')
384
noeol = bytes[start] == 'N'
385
if 'ft' in storage_kind:
388
method = 'line-delta'
389
build_details = (method, noeol)
391
raw_record = bytes[start:]
392
annotated = 'annotated' in storage_kind
393
return [KnitContentFactory(key, parents, build_details, None, raw_record,
394
annotated, network_bytes=bytes)]
397
class KnitContent(object):
398
"""Content of a knit version to which deltas can be applied.
400
This is always stored in memory as a list of lines with \n at the end,
401
plus a flag saying if the final ending is really there or not, because that
402
corresponds to the on-disk knit representation.
406
self._should_strip_eol = False
408
def apply_delta(self, delta, new_version_id):
409
"""Apply delta to this object to become new_version_id."""
410
raise NotImplementedError(self.apply_delta)
412
def line_delta_iter(self, new_lines):
413
"""Generate line-based delta from this content to new_lines."""
414
new_texts = new_lines.text()
415
old_texts = self.text()
416
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
417
for tag, i1, i2, j1, j2 in s.get_opcodes():
420
# ofrom, oto, length, data
421
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
423
def line_delta(self, new_lines):
424
return list(self.line_delta_iter(new_lines))
427
def get_line_delta_blocks(knit_delta, source, target):
428
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
429
target_len = len(target)
432
for s_begin, s_end, t_len, new_text in knit_delta:
433
true_n = s_begin - s_pos
436
# knit deltas do not provide reliable info about whether the
437
# last line of a file matches, due to eol handling.
438
if source[s_pos + n -1] != target[t_pos + n -1]:
441
yield s_pos, t_pos, n
442
t_pos += t_len + true_n
444
n = target_len - t_pos
446
if source[s_pos + n -1] != target[t_pos + n -1]:
449
yield s_pos, t_pos, n
450
yield s_pos + (target_len - t_pos), target_len, 0
453
class AnnotatedKnitContent(KnitContent):
454
"""Annotated content."""
456
def __init__(self, lines):
457
KnitContent.__init__(self)
461
"""Return a list of (origin, text) for each content line."""
462
lines = self._lines[:]
463
if self._should_strip_eol:
464
origin, last_line = lines[-1]
465
lines[-1] = (origin, last_line.rstrip('\n'))
468
def apply_delta(self, delta, new_version_id):
469
"""Apply delta to this object to become new_version_id."""
472
for start, end, count, delta_lines in delta:
473
lines[offset+start:offset+end] = delta_lines
474
offset = offset + (start - end) + count
478
lines = [text for origin, text in self._lines]
479
except ValueError, e:
480
# most commonly (only?) caused by the internal form of the knit
481
# missing annotation information because of a bug - see thread
483
raise KnitCorrupt(self,
484
"line in annotated knit missing annotation information: %s"
486
if self._should_strip_eol:
487
lines[-1] = lines[-1].rstrip('\n')
491
return AnnotatedKnitContent(self._lines[:])
494
class PlainKnitContent(KnitContent):
495
"""Unannotated content.
497
When annotate[_iter] is called on this content, the same version is reported
498
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
502
def __init__(self, lines, version_id):
503
KnitContent.__init__(self)
505
self._version_id = version_id
508
"""Return a list of (origin, text) for each content line."""
509
return [(self._version_id, line) for line in self._lines]
511
def apply_delta(self, delta, new_version_id):
512
"""Apply delta to this object to become new_version_id."""
515
for start, end, count, delta_lines in delta:
516
lines[offset+start:offset+end] = delta_lines
517
offset = offset + (start - end) + count
518
self._version_id = new_version_id
521
return PlainKnitContent(self._lines[:], self._version_id)
525
if self._should_strip_eol:
527
lines[-1] = lines[-1].rstrip('\n')
531
class _KnitFactory(object):
532
"""Base class for common Factory functions."""
534
def parse_record(self, version_id, record, record_details,
535
base_content, copy_base_content=True):
536
"""Parse a record into a full content object.
538
:param version_id: The official version id for this content
539
:param record: The data returned by read_records_iter()
540
:param record_details: Details about the record returned by
542
:param base_content: If get_build_details returns a compression_parent,
543
you must return a base_content here, else use None
544
:param copy_base_content: When building from the base_content, decide
545
you can either copy it and return a new object, or modify it in
547
:return: (content, delta) A Content object and possibly a line-delta,
550
method, noeol = record_details
551
if method == 'line-delta':
552
if copy_base_content:
553
content = base_content.copy()
555
content = base_content
556
delta = self.parse_line_delta(record, version_id)
557
content.apply_delta(delta, version_id)
559
content = self.parse_fulltext(record, version_id)
561
content._should_strip_eol = noeol
562
return (content, delta)
565
class KnitAnnotateFactory(_KnitFactory):
566
"""Factory for creating annotated Content objects."""
570
def make(self, lines, version_id):
571
num_lines = len(lines)
572
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
574
def parse_fulltext(self, content, version_id):
575
"""Convert fulltext to internal representation
577
fulltext content is of the format
578
revid(utf8) plaintext\n
579
internal representation is of the format:
582
# TODO: jam 20070209 The tests expect this to be returned as tuples,
583
# but the code itself doesn't really depend on that.
584
# Figure out a way to not require the overhead of turning the
585
# list back into tuples.
586
lines = [tuple(line.split(' ', 1)) for line in content]
587
return AnnotatedKnitContent(lines)
589
def parse_line_delta_iter(self, lines):
590
return iter(self.parse_line_delta(lines))
592
def parse_line_delta(self, lines, version_id, plain=False):
593
"""Convert a line based delta into internal representation.
595
line delta is in the form of:
596
intstart intend intcount
598
revid(utf8) newline\n
599
internal representation is
600
(start, end, count, [1..count tuples (revid, newline)])
602
:param plain: If True, the lines are returned as a plain
603
list without annotations, not as a list of (origin, content) tuples, i.e.
604
(start, end, count, [1..count newline])
611
def cache_and_return(line):
612
origin, text = line.split(' ', 1)
613
return cache.setdefault(origin, origin), text
615
# walk through the lines parsing.
616
# Note that the plain test is explicitly pulled out of the
617
# loop to minimise any performance impact
620
start, end, count = [int(n) for n in header.split(',')]
621
contents = [next().split(' ', 1)[1] for i in xrange(count)]
622
result.append((start, end, count, contents))
625
start, end, count = [int(n) for n in header.split(',')]
626
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
627
result.append((start, end, count, contents))
630
def get_fulltext_content(self, lines):
631
"""Extract just the content lines from a fulltext."""
632
return (line.split(' ', 1)[1] for line in lines)
634
def get_linedelta_content(self, lines):
635
"""Extract just the content from a line delta.
637
This doesn't return all of the extra information stored in a delta.
638
Only the actual content lines.
643
header = header.split(',')
644
count = int(header[2])
645
for i in xrange(count):
646
origin, text = next().split(' ', 1)
649
def lower_fulltext(self, content):
650
"""convert a fulltext content record into a serializable form.
652
see parse_fulltext which this inverts.
654
# TODO: jam 20070209 We only do the caching thing to make sure that
655
# the origin is a valid utf-8 line, eventually we could remove it
656
return ['%s %s' % (o, t) for o, t in content._lines]
658
def lower_line_delta(self, delta):
659
"""convert a delta into a serializable form.
661
See parse_line_delta which this inverts.
663
# TODO: jam 20070209 We only do the caching thing to make sure that
664
# the origin is a valid utf-8 line, eventually we could remove it
666
for start, end, c, lines in delta:
667
out.append('%d,%d,%d\n' % (start, end, c))
668
out.extend(origin + ' ' + text
669
for origin, text in lines)
672
def annotate(self, knit, key):
673
content = knit._get_content(key)
674
# adjust for the fact that serialised annotations are only key suffixes
676
if type(key) == tuple:
678
origins = content.annotate()
680
for origin, line in origins:
681
result.append((prefix + (origin,), line))
684
# XXX: This smells a bit. Why would key ever be a non-tuple here?
685
# Aren't keys defined to be tuples? -- spiv 20080618
686
return content.annotate()
689
class KnitPlainFactory(_KnitFactory):
690
"""Factory for creating plain Content objects."""
694
def make(self, lines, version_id):
695
return PlainKnitContent(lines, version_id)
697
def parse_fulltext(self, content, version_id):
698
"""This parses an unannotated fulltext.
700
Note that this is not a noop - the internal representation
701
has (versionid, line) - its just a constant versionid.
703
return self.make(content, version_id)
705
def parse_line_delta_iter(self, lines, version_id):
707
num_lines = len(lines)
708
while cur < num_lines:
711
start, end, c = [int(n) for n in header.split(',')]
712
yield start, end, c, lines[cur:cur+c]
715
def parse_line_delta(self, lines, version_id):
716
return list(self.parse_line_delta_iter(lines, version_id))
718
def get_fulltext_content(self, lines):
719
"""Extract just the content lines from a fulltext."""
722
def get_linedelta_content(self, lines):
723
"""Extract just the content from a line delta.
725
This doesn't return all of the extra information stored in a delta.
726
Only the actual content lines.
731
header = header.split(',')
732
count = int(header[2])
733
for i in xrange(count):
736
def lower_fulltext(self, content):
737
return content.text()
739
def lower_line_delta(self, delta):
741
for start, end, c, lines in delta:
742
out.append('%d,%d,%d\n' % (start, end, c))
746
def annotate(self, knit, key):
747
annotator = _KnitAnnotator(knit)
748
return annotator.annotate(key)
752
def make_file_factory(annotated, mapper):
753
"""Create a factory for creating a file based KnitVersionedFiles.
755
This is only functional enough to run interface tests, it doesn't try to
756
provide a full pack environment.
758
:param annotated: knit annotations are wanted.
759
:param mapper: The mapper from keys to paths.
761
def factory(transport):
762
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
763
access = _KnitKeyAccess(transport, mapper)
764
return KnitVersionedFiles(index, access, annotated=annotated)
768
def make_pack_factory(graph, delta, keylength):
769
"""Create a factory for creating a pack based VersionedFiles.
771
This is only functional enough to run interface tests, it doesn't try to
772
provide a full pack environment.
774
:param graph: Store a graph.
775
:param delta: Delta compress contents.
776
:param keylength: How long should keys be.
778
def factory(transport):
779
parents = graph or delta
785
max_delta_chain = 200
788
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
789
key_elements=keylength)
790
stream = transport.open_write_stream('newpack')
791
writer = pack.ContainerWriter(stream.write)
793
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
794
deltas=delta, add_callback=graph_index.add_nodes)
795
access = _DirectPackAccess({})
796
access.set_writer(writer, graph_index, (transport, 'newpack'))
797
result = KnitVersionedFiles(index, access,
798
max_delta_chain=max_delta_chain)
799
result.stream = stream
800
result.writer = writer
805
def cleanup_pack_knit(versioned_files):
806
versioned_files.stream.close()
807
versioned_files.writer.end()
810
def _get_total_build_size(self, keys, positions):
811
"""Determine the total bytes to build these keys.
813
(helper function because _KnitGraphIndex and _KndxIndex work the same, but
814
don't inherit from a common base.)
816
:param keys: Keys that we want to build
817
:param positions: dict of {key, (info, index_memo, comp_parent)} (such
818
as returned by _get_components_positions)
819
:return: Number of bytes to build those keys
821
all_build_index_memos = {}
825
for key in build_keys:
826
# This is mostly for the 'stacked' case
827
# Where we will be getting the data from a fallback
828
if key not in positions:
830
_, index_memo, compression_parent = positions[key]
831
all_build_index_memos[key] = index_memo
832
if compression_parent not in all_build_index_memos:
833
next_keys.add(compression_parent)
834
build_keys = next_keys
835
return sum([index_memo[2] for index_memo
836
in all_build_index_memos.itervalues()])
839
class KnitVersionedFiles(VersionedFiles):
840
"""Storage for many versioned files using knit compression.
842
Backend storage is managed by indices and data objects.
844
:ivar _index: A _KnitGraphIndex or similar that can describe the
845
parents, graph, compression and data location of entries in this
846
KnitVersionedFiles. Note that this is only the index for
847
*this* vfs; if there are fallbacks they must be queried separately.
850
def __init__(self, index, data_access, max_delta_chain=200,
851
annotated=False, reload_func=None):
852
"""Create a KnitVersionedFiles with index and data_access.
854
:param index: The index for the knit data.
855
:param data_access: The access object to store and retrieve knit
857
:param max_delta_chain: The maximum number of deltas to permit during
858
insertion. Set to 0 to prohibit the use of deltas.
859
:param annotated: Set to True to cause annotations to be calculated and
860
stored during insertion.
861
:param reload_func: An function that can be called if we think we need
862
to reload the pack listing and try again. See
863
'bzrlib.repofmt.pack_repo.AggregateIndex' for the signature.
866
self._access = data_access
867
self._max_delta_chain = max_delta_chain
869
self._factory = KnitAnnotateFactory()
871
self._factory = KnitPlainFactory()
872
self._fallback_vfs = []
873
self._reload_func = reload_func
876
return "%s(%r, %r)" % (
877
self.__class__.__name__,
881
def add_fallback_versioned_files(self, a_versioned_files):
882
"""Add a source of texts for texts not present in this knit.
884
:param a_versioned_files: A VersionedFiles object.
886
self._fallback_vfs.append(a_versioned_files)
888
def add_lines(self, key, parents, lines, parent_texts=None,
889
left_matching_blocks=None, nostore_sha=None, random_id=False,
891
"""See VersionedFiles.add_lines()."""
892
self._index._check_write_ok()
893
self._check_add(key, lines, random_id, check_content)
895
# The caller might pass None if there is no graph data, but kndx
896
# indexes can't directly store that, so we give them
897
# an empty tuple instead.
899
return self._add(key, lines, parents,
900
parent_texts, left_matching_blocks, nostore_sha, random_id)
902
def _add(self, key, lines, parents, parent_texts,
903
left_matching_blocks, nostore_sha, random_id):
904
"""Add a set of lines on top of version specified by parents.
906
Any versions not present will be converted into ghosts.
908
# first thing, if the content is something we don't need to store, find
910
line_bytes = ''.join(lines)
911
digest = sha_string(line_bytes)
912
if nostore_sha == digest:
913
raise errors.ExistingContent
916
if parent_texts is None:
918
# Do a single query to ascertain parent presence; we only compress
919
# against parents in the same kvf.
920
present_parent_map = self._index.get_parent_map(parents)
921
for parent in parents:
922
if parent in present_parent_map:
923
present_parents.append(parent)
925
# Currently we can only compress against the left most present parent.
926
if (len(present_parents) == 0 or
927
present_parents[0] != parents[0]):
930
# To speed the extract of texts the delta chain is limited
931
# to a fixed number of deltas. This should minimize both
932
# I/O and the time spend applying deltas.
933
delta = self._check_should_delta(present_parents[0])
935
text_length = len(line_bytes)
938
if lines[-1][-1] != '\n':
939
# copy the contents of lines.
941
options.append('no-eol')
942
lines[-1] = lines[-1] + '\n'
946
if type(element) != str:
947
raise TypeError("key contains non-strings: %r" % (key,))
948
# Knit hunks are still last-element only
950
content = self._factory.make(lines, version_id)
951
if 'no-eol' in options:
952
# Hint to the content object that its text() call should strip the
954
content._should_strip_eol = True
955
if delta or (self._factory.annotated and len(present_parents) > 0):
956
# Merge annotations from parent texts if needed.
957
delta_hunks = self._merge_annotations(content, present_parents,
958
parent_texts, delta, self._factory.annotated,
959
left_matching_blocks)
962
options.append('line-delta')
963
store_lines = self._factory.lower_line_delta(delta_hunks)
964
size, bytes = self._record_to_data(key, digest,
967
options.append('fulltext')
968
# isinstance is slower and we have no hierarchy.
969
if self._factory.__class__ == KnitPlainFactory:
970
# Use the already joined bytes saving iteration time in
972
size, bytes = self._record_to_data(key, digest,
975
# get mixed annotation + content and feed it into the
977
store_lines = self._factory.lower_fulltext(content)
978
size, bytes = self._record_to_data(key, digest,
981
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
982
self._index.add_records(
983
((key, options, access_memo, parents),),
985
return digest, text_length, content
987
def annotate(self, key):
988
"""See VersionedFiles.annotate."""
989
return self._factory.annotate(self, key)
991
def check(self, progress_bar=None):
992
"""See VersionedFiles.check()."""
993
# This doesn't actually test extraction of everything, but that will
994
# impact 'bzr check' substantially, and needs to be integrated with
995
# care. However, it does check for the obvious problem of a delta with
997
keys = self._index.keys()
998
parent_map = self.get_parent_map(keys)
1000
if self._index.get_method(key) != 'fulltext':
1001
compression_parent = parent_map[key][0]
1002
if compression_parent not in parent_map:
1003
raise errors.KnitCorrupt(self,
1004
"Missing basis parent %s for %s" % (
1005
compression_parent, key))
1006
for fallback_vfs in self._fallback_vfs:
1007
fallback_vfs.check()
1009
def _check_add(self, key, lines, random_id, check_content):
1010
"""check that version_id and lines are safe to add."""
1011
version_id = key[-1]
1012
if contains_whitespace(version_id):
1013
raise InvalidRevisionId(version_id, self)
1014
self.check_not_reserved_id(version_id)
1015
# TODO: If random_id==False and the key is already present, we should
1016
# probably check that the existing content is identical to what is
1017
# being inserted, and otherwise raise an exception. This would make
1018
# the bundle code simpler.
1020
self._check_lines_not_unicode(lines)
1021
self._check_lines_are_lines(lines)
1023
def _check_header(self, key, line):
1024
rec = self._split_header(line)
1025
self._check_header_version(rec, key[-1])
1028
def _check_header_version(self, rec, version_id):
1029
"""Checks the header version on original format knit records.
1031
These have the last component of the key embedded in the record.
1033
if rec[1] != version_id:
1034
raise KnitCorrupt(self,
1035
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1037
def _check_should_delta(self, parent):
1038
"""Iterate back through the parent listing, looking for a fulltext.
1040
This is used when we want to decide whether to add a delta or a new
1041
fulltext. It searches for _max_delta_chain parents. When it finds a
1042
fulltext parent, it sees if the total size of the deltas leading up to
1043
it is large enough to indicate that we want a new full text anyway.
1045
Return True if we should create a new delta, False if we should use a
1049
fulltext_size = None
1050
for count in xrange(self._max_delta_chain):
1052
# Note that this only looks in the index of this particular
1053
# KnitVersionedFiles, not in the fallbacks. This ensures that
1054
# we won't store a delta spanning physical repository
1056
build_details = self._index.get_build_details([parent])
1057
parent_details = build_details[parent]
1058
except (RevisionNotPresent, KeyError), e:
1059
# Some basis is not locally present: always fulltext
1061
index_memo, compression_parent, _, _ = parent_details
1062
_, _, size = index_memo
1063
if compression_parent is None:
1064
fulltext_size = size
1067
# We don't explicitly check for presence because this is in an
1068
# inner loop, and if it's missing it'll fail anyhow.
1069
parent = compression_parent
1071
# We couldn't find a fulltext, so we must create a new one
1073
# Simple heuristic - if the total I/O wold be greater as a delta than
1074
# the originally installed fulltext, we create a new fulltext.
1075
return fulltext_size > delta_size
1077
def _build_details_to_components(self, build_details):
1078
"""Convert a build_details tuple to a position tuple."""
1079
# record_details, access_memo, compression_parent
1080
return build_details[3], build_details[0], build_details[1]
1082
def _get_components_positions(self, keys, allow_missing=False):
1083
"""Produce a map of position data for the components of keys.
1085
This data is intended to be used for retrieving the knit records.
1087
A dict of key to (record_details, index_memo, next, parents) is
1089
method is the way referenced data should be applied.
1090
index_memo is the handle to pass to the data access to actually get the
1092
next is the build-parent of the version, or None for fulltexts.
1093
parents is the version_ids of the parents of this version
1095
:param allow_missing: If True do not raise an error on a missing component,
1099
pending_components = keys
1100
while pending_components:
1101
build_details = self._index.get_build_details(pending_components)
1102
current_components = set(pending_components)
1103
pending_components = set()
1104
for key, details in build_details.iteritems():
1105
(index_memo, compression_parent, parents,
1106
record_details) = details
1107
method = record_details[0]
1108
if compression_parent is not None:
1109
pending_components.add(compression_parent)
1110
component_data[key] = self._build_details_to_components(details)
1111
missing = current_components.difference(build_details)
1112
if missing and not allow_missing:
1113
raise errors.RevisionNotPresent(missing.pop(), self)
1114
return component_data
1116
def _get_content(self, key, parent_texts={}):
1117
"""Returns a content object that makes up the specified
1119
cached_version = parent_texts.get(key, None)
1120
if cached_version is not None:
1121
# Ensure the cache dict is valid.
1122
if not self.get_parent_map([key]):
1123
raise RevisionNotPresent(key, self)
1124
return cached_version
1125
generator = _VFContentMapGenerator(self, [key])
1126
return generator._get_content(key)
1128
def get_parent_map(self, keys):
1129
"""Get a map of the graph parents of keys.
1131
:param keys: The keys to look up parents for.
1132
:return: A mapping from keys to parents. Absent keys are absent from
1135
return self._get_parent_map_with_sources(keys)[0]
1137
def _get_parent_map_with_sources(self, keys):
1138
"""Get a map of the parents of keys.
1140
:param keys: The keys to look up parents for.
1141
:return: A tuple. The first element is a mapping from keys to parents.
1142
Absent keys are absent from the mapping. The second element is a
1143
list with the locations each key was found in. The first element
1144
is the in-this-knit parents, the second the first fallback source,
1148
sources = [self._index] + self._fallback_vfs
1151
for source in sources:
1154
new_result = source.get_parent_map(missing)
1155
source_results.append(new_result)
1156
result.update(new_result)
1157
missing.difference_update(set(new_result))
1158
return result, source_results
1160
def _get_record_map(self, keys, allow_missing=False):
1161
"""Produce a dictionary of knit records.
1163
:return: {key:(record, record_details, digest, next)}
1165
data returned from read_records (a KnitContentobject)
1167
opaque information to pass to parse_record
1169
SHA1 digest of the full text after all steps are done
1171
build-parent of the version, i.e. the leftmost ancestor.
1172
Will be None if the record is not a delta.
1173
:param keys: The keys to build a map for
1174
:param allow_missing: If some records are missing, rather than
1175
error, just return the data that could be generated.
1177
raw_map = self._get_record_map_unparsed(keys,
1178
allow_missing=allow_missing)
1179
return self._raw_map_to_record_map(raw_map)
1181
def _raw_map_to_record_map(self, raw_map):
1182
"""Parse the contents of _get_record_map_unparsed.
1184
:return: see _get_record_map.
1188
data, record_details, next = raw_map[key]
1189
content, digest = self._parse_record(key[-1], data)
1190
result[key] = content, record_details, digest, next
1193
def _get_record_map_unparsed(self, keys, allow_missing=False):
1194
"""Get the raw data for reconstructing keys without parsing it.
1196
:return: A dict suitable for parsing via _raw_map_to_record_map.
1197
key-> raw_bytes, (method, noeol), compression_parent
1199
# This retries the whole request if anything fails. Potentially we
1200
# could be a bit more selective. We could track the keys whose records
1201
# we have successfully found, and then only request the new records
1202
# from there. However, _get_components_positions grabs the whole build
1203
# chain, which means we'll likely try to grab the same records again
1204
# anyway. Also, can the build chains change as part of a pack
1205
# operation? We wouldn't want to end up with a broken chain.
1208
position_map = self._get_components_positions(keys,
1209
allow_missing=allow_missing)
1210
# key = component_id, r = record_details, i_m = index_memo,
1212
records = [(key, i_m) for key, (r, i_m, n)
1213
in position_map.iteritems()]
1214
# Sort by the index memo, so that we request records from the
1215
# same pack file together, and in forward-sorted order
1216
records.sort(key=operator.itemgetter(1))
1218
for key, data in self._read_records_iter_unchecked(records):
1219
(record_details, index_memo, next) = position_map[key]
1220
raw_record_map[key] = data, record_details, next
1221
return raw_record_map
1222
except errors.RetryWithNewPacks, e:
1223
self._access.reload_or_raise(e)
1226
def _split_by_prefix(cls, keys):
1227
"""For the given keys, split them up based on their prefix.
1229
To keep memory pressure somewhat under control, split the
1230
requests back into per-file-id requests, otherwise "bzr co"
1231
extracts the full tree into memory before writing it to disk.
1232
This should be revisited if _get_content_maps() can ever cross
1235
The keys for a given file_id are kept in the same relative order.
1236
Ordering between file_ids is not, though prefix_order will return the
1237
order that the key was first seen.
1239
:param keys: An iterable of key tuples
1240
:return: (split_map, prefix_order)
1241
split_map A dictionary mapping prefix => keys
1242
prefix_order The order that we saw the various prefixes
1244
split_by_prefix = {}
1252
if prefix in split_by_prefix:
1253
split_by_prefix[prefix].append(key)
1255
split_by_prefix[prefix] = [key]
1256
prefix_order.append(prefix)
1257
return split_by_prefix, prefix_order
1259
def _group_keys_for_io(self, keys, non_local_keys, positions,
1260
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1261
"""For the given keys, group them into 'best-sized' requests.
1263
The idea is to avoid making 1 request per file, but to never try to
1264
unpack an entire 1.5GB source tree in a single pass. Also when
1265
possible, we should try to group requests to the same pack file
1268
:return: list of (keys, non_local) tuples that indicate what keys
1269
should be fetched next.
1271
# TODO: Ideally we would group on 2 factors. We want to extract texts
1272
# from the same pack file together, and we want to extract all
1273
# the texts for a given build-chain together. Ultimately it
1274
# probably needs a better global view.
1275
total_keys = len(keys)
1276
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1277
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1279
cur_non_local = set()
1283
for prefix in prefix_order:
1284
keys = prefix_split_keys[prefix]
1285
non_local = prefix_split_non_local_keys.get(prefix, [])
1287
this_size = self._index._get_total_build_size(keys, positions)
1288
cur_size += this_size
1289
cur_keys.extend(keys)
1290
cur_non_local.update(non_local)
1291
if cur_size > _min_buffer_size:
1292
result.append((cur_keys, cur_non_local))
1293
sizes.append(cur_size)
1295
cur_non_local = set()
1298
result.append((cur_keys, cur_non_local))
1299
sizes.append(cur_size)
1300
trace.mutter('Collapsed %d keys into %d requests w/ %d file_ids'
1301
' w/ sizes: %s', total_keys, len(result),
1302
len(prefix_split_keys), sizes)
1305
def get_record_stream(self, keys, ordering, include_delta_closure):
1306
"""Get a stream of records for keys.
1308
:param keys: The keys to include.
1309
:param ordering: Either 'unordered' or 'topological'. A topologically
1310
sorted stream has compression parents strictly before their
1312
:param include_delta_closure: If True then the closure across any
1313
compression parents will be included (in the opaque data).
1314
:return: An iterator of ContentFactory objects, each of which is only
1315
valid until the iterator is advanced.
1317
# keys might be a generator
1321
if not self._index.has_graph:
1322
# Cannot topological order when no graph has been stored.
1323
ordering = 'unordered'
1325
remaining_keys = keys
1328
keys = set(remaining_keys)
1329
for content_factory in self._get_remaining_record_stream(keys,
1330
ordering, include_delta_closure):
1331
remaining_keys.discard(content_factory.key)
1332
yield content_factory
1334
except errors.RetryWithNewPacks, e:
1335
self._access.reload_or_raise(e)
1337
def _get_remaining_record_stream(self, keys, ordering,
1338
include_delta_closure):
1339
"""This function is the 'retry' portion for get_record_stream."""
1340
if include_delta_closure:
1341
positions = self._get_components_positions(keys, allow_missing=True)
1343
build_details = self._index.get_build_details(keys)
1345
# (record_details, access_memo, compression_parent_key)
1346
positions = dict((key, self._build_details_to_components(details))
1347
for key, details in build_details.iteritems())
1348
absent_keys = keys.difference(set(positions))
1349
# There may be more absent keys : if we're missing the basis component
1350
# and are trying to include the delta closure.
1351
# XXX: We should not ever need to examine remote sources because we do
1352
# not permit deltas across versioned files boundaries.
1353
if include_delta_closure:
1354
needed_from_fallback = set()
1355
# Build up reconstructable_keys dict. key:True in this dict means
1356
# the key can be reconstructed.
1357
reconstructable_keys = {}
1361
chain = [key, positions[key][2]]
1363
needed_from_fallback.add(key)
1366
while chain[-1] is not None:
1367
if chain[-1] in reconstructable_keys:
1368
result = reconstructable_keys[chain[-1]]
1372
chain.append(positions[chain[-1]][2])
1374
# missing basis component
1375
needed_from_fallback.add(chain[-1])
1378
for chain_key in chain[:-1]:
1379
reconstructable_keys[chain_key] = result
1381
needed_from_fallback.add(key)
1382
# Double index lookups here : need a unified api ?
1383
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1384
if ordering == 'topological':
1385
# Global topological sort
1386
present_keys = tsort.topo_sort(global_map)
1387
# Now group by source:
1389
current_source = None
1390
for key in present_keys:
1391
for parent_map in parent_maps:
1392
if key in parent_map:
1393
key_source = parent_map
1395
if current_source is not key_source:
1396
source_keys.append((key_source, []))
1397
current_source = key_source
1398
source_keys[-1][1].append(key)
1400
if ordering != 'unordered':
1401
raise AssertionError('valid values for ordering are:'
1402
' "unordered" or "topological" not: %r'
1404
# Just group by source; remote sources first.
1407
for parent_map in reversed(parent_maps):
1408
source_keys.append((parent_map, []))
1409
for key in parent_map:
1410
present_keys.append(key)
1411
source_keys[-1][1].append(key)
1412
# We have been requested to return these records in an order that
1413
# suits us. So we ask the index to give us an optimally sorted
1415
for source, sub_keys in source_keys:
1416
if source is parent_maps[0]:
1417
# Only sort the keys for this VF
1418
self._index._sort_keys_by_io(sub_keys, positions)
1419
absent_keys = keys - set(global_map)
1420
for key in absent_keys:
1421
yield AbsentContentFactory(key)
1422
# restrict our view to the keys we can answer.
1423
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1424
# XXX: At that point we need to consider the impact of double reads by
1425
# utilising components multiple times.
1426
if include_delta_closure:
1427
# XXX: get_content_maps performs its own index queries; allow state
1429
non_local_keys = needed_from_fallback - absent_keys
1430
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1433
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1435
for record in generator.get_record_stream():
1438
for source, keys in source_keys:
1439
if source is parent_maps[0]:
1440
# this KnitVersionedFiles
1441
records = [(key, positions[key][1]) for key in keys]
1442
for key, raw_data, sha1 in self._read_records_iter_raw(records):
1443
(record_details, index_memo, _) = positions[key]
1444
yield KnitContentFactory(key, global_map[key],
1445
record_details, sha1, raw_data, self._factory.annotated, None)
1447
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1448
for record in vf.get_record_stream(keys, ordering,
1449
include_delta_closure):
1452
def get_sha1s(self, keys):
1453
"""See VersionedFiles.get_sha1s()."""
1455
record_map = self._get_record_map(missing, allow_missing=True)
1457
for key, details in record_map.iteritems():
1458
if key not in missing:
1460
# record entry 2 is the 'digest'.
1461
result[key] = details[2]
1462
missing.difference_update(set(result))
1463
for source in self._fallback_vfs:
1466
new_result = source.get_sha1s(missing)
1467
result.update(new_result)
1468
missing.difference_update(set(new_result))
1471
def insert_record_stream(self, stream):
1472
"""Insert a record stream into this container.
1474
:param stream: A stream of records to insert.
1476
:seealso VersionedFiles.get_record_stream:
1478
def get_adapter(adapter_key):
1480
return adapters[adapter_key]
1482
adapter_factory = adapter_registry.get(adapter_key)
1483
adapter = adapter_factory(self)
1484
adapters[adapter_key] = adapter
1487
if self._factory.annotated:
1488
# self is annotated, we need annotated knits to use directly.
1489
annotated = "annotated-"
1492
# self is not annotated, but we can strip annotations cheaply.
1494
convertibles = set(["knit-annotated-ft-gz"])
1495
if self._max_delta_chain:
1496
delta_types.add("knit-annotated-delta-gz")
1497
convertibles.add("knit-annotated-delta-gz")
1498
# The set of types we can cheaply adapt without needing basis texts.
1499
native_types = set()
1500
if self._max_delta_chain:
1501
native_types.add("knit-%sdelta-gz" % annotated)
1502
delta_types.add("knit-%sdelta-gz" % annotated)
1503
native_types.add("knit-%sft-gz" % annotated)
1504
knit_types = native_types.union(convertibles)
1506
# Buffer all index entries that we can't add immediately because their
1507
# basis parent is missing. We don't buffer all because generating
1508
# annotations may require access to some of the new records. However we
1509
# can't generate annotations from new deltas until their basis parent
1510
# is present anyway, so we get away with not needing an index that
1511
# includes the new keys.
1513
# See <http://launchpad.net/bugs/300177> about ordering of compression
1514
# parents in the records - to be conservative, we insist that all
1515
# parents must be present to avoid expanding to a fulltext.
1517
# key = basis_parent, value = index entry to add
1518
buffered_index_entries = {}
1519
for record in stream:
1520
parents = record.parents
1521
if record.storage_kind in delta_types:
1522
# TODO: eventually the record itself should track
1523
# compression_parent
1524
compression_parent = parents[0]
1526
compression_parent = None
1527
# Raise an error when a record is missing.
1528
if record.storage_kind == 'absent':
1529
raise RevisionNotPresent([record.key], self)
1530
elif ((record.storage_kind in knit_types)
1531
and (compression_parent is None
1532
or not self._fallback_vfs
1533
or self._index.has_key(compression_parent)
1534
or not self.has_key(compression_parent))):
1535
# we can insert the knit record literally if either it has no
1536
# compression parent OR we already have its basis in this kvf
1537
# OR the basis is not present even in the fallbacks. In the
1538
# last case it will either turn up later in the stream and all
1539
# will be well, or it won't turn up at all and we'll raise an
1542
# TODO: self.has_key is somewhat redundant with
1543
# self._index.has_key; we really want something that directly
1544
# asks if it's only present in the fallbacks. -- mbp 20081119
1545
if record.storage_kind not in native_types:
1547
adapter_key = (record.storage_kind, "knit-delta-gz")
1548
adapter = get_adapter(adapter_key)
1550
adapter_key = (record.storage_kind, "knit-ft-gz")
1551
adapter = get_adapter(adapter_key)
1552
bytes = adapter.get_bytes(record)
1554
# It's a knit record, it has a _raw_record field (even if
1555
# it was reconstituted from a network stream).
1556
bytes = record._raw_record
1557
options = [record._build_details[0]]
1558
if record._build_details[1]:
1559
options.append('no-eol')
1560
# Just blat it across.
1561
# Note: This does end up adding data on duplicate keys. As
1562
# modern repositories use atomic insertions this should not
1563
# lead to excessive growth in the event of interrupted fetches.
1564
# 'knit' repositories may suffer excessive growth, but as a
1565
# deprecated format this is tolerable. It can be fixed if
1566
# needed by in the kndx index support raising on a duplicate
1567
# add with identical parents and options.
1568
access_memo = self._access.add_raw_records(
1569
[(record.key, len(bytes))], bytes)[0]
1570
index_entry = (record.key, options, access_memo, parents)
1572
if 'fulltext' not in options:
1573
# Not a fulltext, so we need to make sure the compression
1574
# parent will also be present.
1575
# Note that pack backed knits don't need to buffer here
1576
# because they buffer all writes to the transaction level,
1577
# but we don't expose that difference at the index level. If
1578
# the query here has sufficient cost to show up in
1579
# profiling we should do that.
1581
# They're required to be physically in this
1582
# KnitVersionedFiles, not in a fallback.
1583
if not self._index.has_key(compression_parent):
1584
pending = buffered_index_entries.setdefault(
1585
compression_parent, [])
1586
pending.append(index_entry)
1589
self._index.add_records([index_entry])
1590
elif record.storage_kind == 'chunked':
1591
self.add_lines(record.key, parents,
1592
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1594
# Not suitable for direct insertion as a
1595
# delta, either because it's not the right format, or this
1596
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1597
# 0) or because it depends on a base only present in the
1600
# Try getting a fulltext directly from the record.
1601
bytes = record.get_bytes_as('fulltext')
1602
except errors.UnavailableRepresentation:
1603
adapter_key = record.storage_kind, 'fulltext'
1604
adapter = get_adapter(adapter_key)
1605
bytes = adapter.get_bytes(record)
1606
lines = split_lines(bytes)
1608
self.add_lines(record.key, parents, lines)
1609
except errors.RevisionAlreadyPresent:
1611
# Add any records whose basis parent is now available.
1612
added_keys = [record.key]
1614
key = added_keys.pop(0)
1615
if key in buffered_index_entries:
1616
index_entries = buffered_index_entries[key]
1617
self._index.add_records(index_entries)
1619
[index_entry[0] for index_entry in index_entries])
1620
del buffered_index_entries[key]
1621
if buffered_index_entries:
1622
# There were index entries buffered at the end of the stream,
1623
# So these need to be added (if the index supports holding such
1624
# entries for later insertion)
1625
for key in buffered_index_entries:
1626
index_entries = buffered_index_entries[key]
1627
self._index.add_records(index_entries,
1628
missing_compression_parents=True)
1630
def get_missing_compression_parent_keys(self):
1631
"""Return an iterable of keys of missing compression parents.
1633
Check this after calling insert_record_stream to find out if there are
1634
any missing compression parents. If there are, the records that
1635
depend on them are not able to be inserted safely. For atomic
1636
KnitVersionedFiles built on packs, the transaction should be aborted or
1637
suspended - commit will fail at this point. Nonatomic knits will error
1638
earlier because they have no staging area to put pending entries into.
1640
return self._index.get_missing_compression_parents()
1642
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1643
"""Iterate over the lines in the versioned files from keys.
1645
This may return lines from other keys. Each item the returned
1646
iterator yields is a tuple of a line and a text version that that line
1647
is present in (not introduced in).
1649
Ordering of results is in whatever order is most suitable for the
1650
underlying storage format.
1652
If a progress bar is supplied, it may be used to indicate progress.
1653
The caller is responsible for cleaning up progress bars (because this
1657
* Lines are normalised by the underlying store: they will all have \\n
1659
* Lines are returned in arbitrary order.
1660
* If a requested key did not change any lines (or didn't have any
1661
lines), it may not be mentioned at all in the result.
1663
:return: An iterator over (line, key).
1666
pb = progress.DummyProgress()
1672
# we don't care about inclusions, the caller cares.
1673
# but we need to setup a list of records to visit.
1674
# we need key, position, length
1676
build_details = self._index.get_build_details(keys)
1677
for key, details in build_details.iteritems():
1679
key_records.append((key, details[0]))
1680
records_iter = enumerate(self._read_records_iter(key_records))
1681
for (key_idx, (key, data, sha_value)) in records_iter:
1682
pb.update('Walking content.', key_idx, total)
1683
compression_parent = build_details[key][1]
1684
if compression_parent is None:
1686
line_iterator = self._factory.get_fulltext_content(data)
1689
line_iterator = self._factory.get_linedelta_content(data)
1690
# Now that we are yielding the data for this key, remove it
1693
# XXX: It might be more efficient to yield (key,
1694
# line_iterator) in the future. However for now, this is a
1695
# simpler change to integrate into the rest of the
1696
# codebase. RBC 20071110
1697
for line in line_iterator:
1700
except errors.RetryWithNewPacks, e:
1701
self._access.reload_or_raise(e)
1702
# If there are still keys we've not yet found, we look in the fallback
1703
# vfs, and hope to find them there. Note that if the keys are found
1704
# but had no changes or no content, the fallback may not return
1706
if keys and not self._fallback_vfs:
1707
# XXX: strictly the second parameter is meant to be the file id
1708
# but it's not easily accessible here.
1709
raise RevisionNotPresent(keys, repr(self))
1710
for source in self._fallback_vfs:
1714
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1715
source_keys.add(key)
1717
keys.difference_update(source_keys)
1718
pb.update('Walking content.', total, total)
1720
def _make_line_delta(self, delta_seq, new_content):
1721
"""Generate a line delta from delta_seq and new_content."""
1723
for op in delta_seq.get_opcodes():
1724
if op[0] == 'equal':
1726
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1729
def _merge_annotations(self, content, parents, parent_texts={},
1730
delta=None, annotated=None,
1731
left_matching_blocks=None):
1732
"""Merge annotations for content and generate deltas.
1734
This is done by comparing the annotations based on changes to the text
1735
and generating a delta on the resulting full texts. If annotations are
1736
not being created then a simple delta is created.
1738
if left_matching_blocks is not None:
1739
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1743
for parent_key in parents:
1744
merge_content = self._get_content(parent_key, parent_texts)
1745
if (parent_key == parents[0] and delta_seq is not None):
1748
seq = patiencediff.PatienceSequenceMatcher(
1749
None, merge_content.text(), content.text())
1750
for i, j, n in seq.get_matching_blocks():
1753
# this copies (origin, text) pairs across to the new
1754
# content for any line that matches the last-checked
1756
content._lines[j:j+n] = merge_content._lines[i:i+n]
1757
# XXX: Robert says the following block is a workaround for a
1758
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1759
if content._lines and content._lines[-1][1][-1] != '\n':
1760
# The copied annotation was from a line without a trailing EOL,
1761
# reinstate one for the content object, to ensure correct
1763
line = content._lines[-1][1] + '\n'
1764
content._lines[-1] = (content._lines[-1][0], line)
1766
if delta_seq is None:
1767
reference_content = self._get_content(parents[0], parent_texts)
1768
new_texts = content.text()
1769
old_texts = reference_content.text()
1770
delta_seq = patiencediff.PatienceSequenceMatcher(
1771
None, old_texts, new_texts)
1772
return self._make_line_delta(delta_seq, content)
1774
def _parse_record(self, version_id, data):
1775
"""Parse an original format knit record.
1777
These have the last element of the key only present in the stored data.
1779
rec, record_contents = self._parse_record_unchecked(data)
1780
self._check_header_version(rec, version_id)
1781
return record_contents, rec[3]
1783
def _parse_record_header(self, key, raw_data):
1784
"""Parse a record header for consistency.
1786
:return: the header and the decompressor stream.
1787
as (stream, header_record)
1789
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1792
rec = self._check_header(key, df.readline())
1793
except Exception, e:
1794
raise KnitCorrupt(self,
1795
"While reading {%s} got %s(%s)"
1796
% (key, e.__class__.__name__, str(e)))
1799
def _parse_record_unchecked(self, data):
1801
# 4168 calls in 2880 217 internal
1802
# 4168 calls to _parse_record_header in 2121
1803
# 4168 calls to readlines in 330
1804
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1806
record_contents = df.readlines()
1807
except Exception, e:
1808
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1809
(data, e.__class__.__name__, str(e)))
1810
header = record_contents.pop(0)
1811
rec = self._split_header(header)
1812
last_line = record_contents.pop()
1813
if len(record_contents) != int(rec[2]):
1814
raise KnitCorrupt(self,
1815
'incorrect number of lines %s != %s'
1816
' for version {%s} %s'
1817
% (len(record_contents), int(rec[2]),
1818
rec[1], record_contents))
1819
if last_line != 'end %s\n' % rec[1]:
1820
raise KnitCorrupt(self,
1821
'unexpected version end line %r, wanted %r'
1822
% (last_line, rec[1]))
1824
return rec, record_contents
1826
def _read_records_iter(self, records):
1827
"""Read text records from data file and yield result.
1829
The result will be returned in whatever is the fastest to read.
1830
Not by the order requested. Also, multiple requests for the same
1831
record will only yield 1 response.
1832
:param records: A list of (key, access_memo) entries
1833
:return: Yields (key, contents, digest) in the order
1834
read, not the order requested
1839
# XXX: This smells wrong, IO may not be getting ordered right.
1840
needed_records = sorted(set(records), key=operator.itemgetter(1))
1841
if not needed_records:
1844
# The transport optimizes the fetching as well
1845
# (ie, reads continuous ranges.)
1846
raw_data = self._access.get_raw_records(
1847
[index_memo for key, index_memo in needed_records])
1849
for (key, index_memo), data in \
1850
izip(iter(needed_records), raw_data):
1851
content, digest = self._parse_record(key[-1], data)
1852
yield key, content, digest
1854
def _read_records_iter_raw(self, records):
1855
"""Read text records from data file and yield raw data.
1857
This unpacks enough of the text record to validate the id is
1858
as expected but thats all.
1860
Each item the iterator yields is (key, bytes,
1861
expected_sha1_of_full_text).
1863
for key, data in self._read_records_iter_unchecked(records):
1864
# validate the header (note that we can only use the suffix in
1865
# current knit records).
1866
df, rec = self._parse_record_header(key, data)
1868
yield key, data, rec[3]
1870
def _read_records_iter_unchecked(self, records):
1871
"""Read text records from data file and yield raw data.
1873
No validation is done.
1875
Yields tuples of (key, data).
1877
# setup an iterator of the external records:
1878
# uses readv so nice and fast we hope.
1880
# grab the disk data needed.
1881
needed_offsets = [index_memo for key, index_memo
1883
raw_records = self._access.get_raw_records(needed_offsets)
1885
for key, index_memo in records:
1886
data = raw_records.next()
1889
def _record_to_data(self, key, digest, lines, dense_lines=None):
1890
"""Convert key, digest, lines into a raw data block.
1892
:param key: The key of the record. Currently keys are always serialised
1893
using just the trailing component.
1894
:param dense_lines: The bytes of lines but in a denser form. For
1895
instance, if lines is a list of 1000 bytestrings each ending in \n,
1896
dense_lines may be a list with one line in it, containing all the
1897
1000's lines and their \n's. Using dense_lines if it is already
1898
known is a win because the string join to create bytes in this
1899
function spends less time resizing the final string.
1900
:return: (len, a StringIO instance with the raw data ready to read.)
1902
# Note: using a string copy here increases memory pressure with e.g.
1903
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
1904
# when doing the initial commit of a mozilla tree. RBC 20070921
1905
bytes = ''.join(chain(
1906
["version %s %d %s\n" % (key[-1],
1909
dense_lines or lines,
1910
["end %s\n" % key[-1]]))
1911
if type(bytes) != str:
1912
raise AssertionError(
1913
'data must be plain bytes was %s' % type(bytes))
1914
if lines and lines[-1][-1] != '\n':
1915
raise ValueError('corrupt lines value %r' % lines)
1916
compressed_bytes = tuned_gzip.bytes_to_gzip(bytes)
1917
return len(compressed_bytes), compressed_bytes
1919
def _split_header(self, line):
1922
raise KnitCorrupt(self,
1923
'unexpected number of elements in record header')
1927
"""See VersionedFiles.keys."""
1928
if 'evil' in debug.debug_flags:
1929
trace.mutter_callsite(2, "keys scales with size of history")
1930
sources = [self._index] + self._fallback_vfs
1932
for source in sources:
1933
result.update(source.keys())
1937
class _ContentMapGenerator(object):
1938
"""Generate texts or expose raw deltas for a set of texts."""
1940
def _get_content(self, key):
1941
"""Get the content object for key."""
1942
# Note that _get_content is only called when the _ContentMapGenerator
1943
# has been constructed with just one key requested for reconstruction.
1944
if key in self.nonlocal_keys:
1945
record = self.get_record_stream().next()
1946
# Create a content object on the fly
1947
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
1948
return PlainKnitContent(lines, record.key)
1950
# local keys we can ask for directly
1951
return self._get_one_work(key)
1953
def get_record_stream(self):
1954
"""Get a record stream for the keys requested during __init__."""
1955
for record in self._work():
1959
"""Produce maps of text and KnitContents as dicts.
1961
:return: (text_map, content_map) where text_map contains the texts for
1962
the requested versions and content_map contains the KnitContents.
1964
# NB: By definition we never need to read remote sources unless texts
1965
# are requested from them: we don't delta across stores - and we
1966
# explicitly do not want to to prevent data loss situations.
1967
if self.global_map is None:
1968
self.global_map = self.vf.get_parent_map(self.keys)
1969
nonlocal_keys = self.nonlocal_keys
1971
missing_keys = set(nonlocal_keys)
1972
# Read from remote versioned file instances and provide to our caller.
1973
for source in self.vf._fallback_vfs:
1974
if not missing_keys:
1976
# Loop over fallback repositories asking them for texts - ignore
1977
# any missing from a particular fallback.
1978
for record in source.get_record_stream(missing_keys,
1980
if record.storage_kind == 'absent':
1981
# Not in thie particular stream, may be in one of the
1982
# other fallback vfs objects.
1984
missing_keys.remove(record.key)
1987
self._raw_record_map = self.vf._get_record_map_unparsed(self.keys,
1990
for key in self.keys:
1991
if key in self.nonlocal_keys:
1993
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
1996
def _get_one_work(self, requested_key):
1997
# Now, if we have calculated everything already, just return the
1999
if requested_key in self._contents_map:
2000
return self._contents_map[requested_key]
2001
# To simplify things, parse everything at once - code that wants one text
2002
# probably wants them all.
2003
# FUTURE: This function could be improved for the 'extract many' case
2004
# by tracking each component and only doing the copy when the number of
2005
# children than need to apply delta's to it is > 1 or it is part of the
2007
multiple_versions = len(self.keys) != 1
2008
if self._record_map is None:
2009
self._record_map = self.vf._raw_map_to_record_map(
2010
self._raw_record_map)
2011
record_map = self._record_map
2012
# raw_record_map is key:
2013
# Have read and parsed records at this point.
2014
for key in self.keys:
2015
if key in self.nonlocal_keys:
2020
while cursor is not None:
2022
record, record_details, digest, next = record_map[cursor]
2024
raise RevisionNotPresent(cursor, self)
2025
components.append((cursor, record, record_details, digest))
2027
if cursor in self._contents_map:
2028
# no need to plan further back
2029
components.append((cursor, None, None, None))
2033
for (component_id, record, record_details,
2034
digest) in reversed(components):
2035
if component_id in self._contents_map:
2036
content = self._contents_map[component_id]
2038
content, delta = self._factory.parse_record(key[-1],
2039
record, record_details, content,
2040
copy_base_content=multiple_versions)
2041
if multiple_versions:
2042
self._contents_map[component_id] = content
2044
# digest here is the digest from the last applied component.
2045
text = content.text()
2046
actual_sha = sha_strings(text)
2047
if actual_sha != digest:
2048
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2049
if multiple_versions:
2050
return self._contents_map[requested_key]
2054
def _wire_bytes(self):
2055
"""Get the bytes to put on the wire for 'key'.
2057
The first collection of bytes asked for returns the serialised
2058
raw_record_map and the additional details (key, parent) for key.
2059
Subsequent calls return just the additional details (key, parent).
2060
The wire storage_kind given for the first key is 'knit-delta-closure',
2061
For subsequent keys it is 'knit-delta-closure-ref'.
2063
:param key: A key from the content generator.
2064
:return: Bytes to put on the wire.
2067
# kind marker for dispatch on the far side,
2068
lines.append('knit-delta-closure')
2070
if self.vf._factory.annotated:
2071
lines.append('annotated')
2074
# then the list of keys
2075
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2076
if key not in self.nonlocal_keys]))
2077
# then the _raw_record_map in serialised form:
2079
# for each item in the map:
2081
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2082
# one line with method
2083
# one line with noeol
2084
# one line with next ('' for None)
2085
# one line with byte count of the record bytes
2087
for key, (record_bytes, (method, noeol), next) in \
2088
self._raw_record_map.iteritems():
2089
key_bytes = '\x00'.join(key)
2090
parents = self.global_map.get(key, None)
2092
parent_bytes = 'None:'
2094
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2095
method_bytes = method
2101
next_bytes = '\x00'.join(next)
2104
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2105
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2106
len(record_bytes), record_bytes))
2107
map_bytes = ''.join(map_byte_list)
2108
lines.append(map_bytes)
2109
bytes = '\n'.join(lines)
2113
class _VFContentMapGenerator(_ContentMapGenerator):
2114
"""Content map generator reading from a VersionedFiles object."""
2116
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2117
global_map=None, raw_record_map=None):
2118
"""Create a _ContentMapGenerator.
2120
:param versioned_files: The versioned files that the texts are being
2122
:param keys: The keys to produce content maps for.
2123
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2124
which are known to not be in this knit, but rather in one of the
2126
:param global_map: The result of get_parent_map(keys) (or a supermap).
2127
This is required if get_record_stream() is to be used.
2128
:param raw_record_map: A unparsed raw record map to use for answering
2131
# The vf to source data from
2132
self.vf = versioned_files
2134
self.keys = list(keys)
2135
# Keys known to be in fallback vfs objects
2136
if nonlocal_keys is None:
2137
self.nonlocal_keys = set()
2139
self.nonlocal_keys = frozenset(nonlocal_keys)
2140
# Parents data for keys to be returned in get_record_stream
2141
self.global_map = global_map
2142
# The chunked lists for self.keys in text form
2144
# A cache of KnitContent objects used in extracting texts.
2145
self._contents_map = {}
2146
# All the knit records needed to assemble the requested keys as full
2148
self._record_map = None
2149
if raw_record_map is None:
2150
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2153
self._raw_record_map = raw_record_map
2154
# the factory for parsing records
2155
self._factory = self.vf._factory
2158
class _NetworkContentMapGenerator(_ContentMapGenerator):
2159
"""Content map generator sourced from a network stream."""
2161
def __init__(self, bytes, line_end):
2162
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2164
self.global_map = {}
2165
self._raw_record_map = {}
2166
self._contents_map = {}
2167
self._record_map = None
2168
self.nonlocal_keys = []
2169
# Get access to record parsing facilities
2170
self.vf = KnitVersionedFiles(None, None)
2173
line_end = bytes.find('\n', start)
2174
line = bytes[start:line_end]
2175
start = line_end + 1
2176
if line == 'annotated':
2177
self._factory = KnitAnnotateFactory()
2179
self._factory = KnitPlainFactory()
2180
# list of keys to emit in get_record_stream
2181
line_end = bytes.find('\n', start)
2182
line = bytes[start:line_end]
2183
start = line_end + 1
2185
tuple(segment.split('\x00')) for segment in line.split('\t')
2187
# now a loop until the end. XXX: It would be nice if this was just a
2188
# bunch of the same records as get_record_stream(..., False) gives, but
2189
# there is a decent sized gap stopping that at the moment.
2193
line_end = bytes.find('\n', start)
2194
key = tuple(bytes[start:line_end].split('\x00'))
2195
start = line_end + 1
2196
# 1 line with parents (None: for None, '' for ())
2197
line_end = bytes.find('\n', start)
2198
line = bytes[start:line_end]
2203
[tuple(segment.split('\x00')) for segment in line.split('\t')
2205
self.global_map[key] = parents
2206
start = line_end + 1
2207
# one line with method
2208
line_end = bytes.find('\n', start)
2209
line = bytes[start:line_end]
2211
start = line_end + 1
2212
# one line with noeol
2213
line_end = bytes.find('\n', start)
2214
line = bytes[start:line_end]
2216
start = line_end + 1
2217
# one line with next ('' for None)
2218
line_end = bytes.find('\n', start)
2219
line = bytes[start:line_end]
2223
next = tuple(bytes[start:line_end].split('\x00'))
2224
start = line_end + 1
2225
# one line with byte count of the record bytes
2226
line_end = bytes.find('\n', start)
2227
line = bytes[start:line_end]
2229
start = line_end + 1
2231
record_bytes = bytes[start:start+count]
2232
start = start + count
2234
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2236
def get_record_stream(self):
2237
"""Get a record stream for for keys requested by the bytestream."""
2239
for key in self.keys:
2240
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2243
def _wire_bytes(self):
2247
class _KndxIndex(object):
2248
"""Manages knit index files
2250
The index is kept in memory and read on startup, to enable
2251
fast lookups of revision information. The cursor of the index
2252
file is always pointing to the end, making it easy to append
2255
_cache is a cache for fast mapping from version id to a Index
2258
_history is a cache for fast mapping from indexes to version ids.
2260
The index data format is dictionary compressed when it comes to
2261
parent references; a index entry may only have parents that with a
2262
lover index number. As a result, the index is topological sorted.
2264
Duplicate entries may be written to the index for a single version id
2265
if this is done then the latter one completely replaces the former:
2266
this allows updates to correct version and parent information.
2267
Note that the two entries may share the delta, and that successive
2268
annotations and references MUST point to the first entry.
2270
The index file on disc contains a header, followed by one line per knit
2271
record. The same revision can be present in an index file more than once.
2272
The first occurrence gets assigned a sequence number starting from 0.
2274
The format of a single line is
2275
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
2276
REVISION_ID is a utf8-encoded revision id
2277
FLAGS is a comma separated list of flags about the record. Values include
2278
no-eol, line-delta, fulltext.
2279
BYTE_OFFSET is the ascii representation of the byte offset in the data file
2280
that the the compressed data starts at.
2281
LENGTH is the ascii representation of the length of the data file.
2282
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
2284
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
2285
revision id already in the knit that is a parent of REVISION_ID.
2286
The ' :' marker is the end of record marker.
2289
when a write is interrupted to the index file, it will result in a line
2290
that does not end in ' :'. If the ' :' is not present at the end of a line,
2291
or at the end of the file, then the record that is missing it will be
2292
ignored by the parser.
2294
When writing new records to the index file, the data is preceded by '\n'
2295
to ensure that records always start on new lines even if the last write was
2296
interrupted. As a result its normal for the last line in the index to be
2297
missing a trailing newline. One can be added with no harmful effects.
2299
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2300
where prefix is e.g. the (fileid,) for .texts instances or () for
2301
constant-mapped things like .revisions, and the old state is
2302
tuple(history_vector, cache_dict). This is used to prevent having an
2303
ABI change with the C extension that reads .kndx files.
2306
HEADER = "# bzr knit index 8\n"
2308
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2309
"""Create a _KndxIndex on transport using mapper."""
2310
self._transport = transport
2311
self._mapper = mapper
2312
self._get_scope = get_scope
2313
self._allow_writes = allow_writes
2314
self._is_locked = is_locked
2316
self.has_graph = True
2318
def add_records(self, records, random_id=False, missing_compression_parents=False):
2319
"""Add multiple records to the index.
2321
:param records: a list of tuples:
2322
(key, options, access_memo, parents).
2323
:param random_id: If True the ids being added were randomly generated
2324
and no check for existence will be performed.
2325
:param missing_compression_parents: If True the records being added are
2326
only compressed against texts already in the index (or inside
2327
records). If False the records all refer to unavailable texts (or
2328
texts inside records) as compression parents.
2330
if missing_compression_parents:
2331
# It might be nice to get the edge of the records. But keys isn't
2333
keys = sorted(record[0] for record in records)
2334
raise errors.RevisionNotPresent(keys, self)
2336
for record in records:
2339
path = self._mapper.map(key) + '.kndx'
2340
path_keys = paths.setdefault(path, (prefix, []))
2341
path_keys[1].append(record)
2342
for path in sorted(paths):
2343
prefix, path_keys = paths[path]
2344
self._load_prefixes([prefix])
2346
orig_history = self._kndx_cache[prefix][1][:]
2347
orig_cache = self._kndx_cache[prefix][0].copy()
2350
for key, options, (_, pos, size), parents in path_keys:
2352
# kndx indices cannot be parentless.
2354
line = "\n%s %s %s %s %s :" % (
2355
key[-1], ','.join(options), pos, size,
2356
self._dictionary_compress(parents))
2357
if type(line) != str:
2358
raise AssertionError(
2359
'data must be utf8 was %s' % type(line))
2361
self._cache_key(key, options, pos, size, parents)
2362
if len(orig_history):
2363
self._transport.append_bytes(path, ''.join(lines))
2365
self._init_index(path, lines)
2367
# If any problems happen, restore the original values and re-raise
2368
self._kndx_cache[prefix] = (orig_cache, orig_history)
2371
def scan_unvalidated_index(self, graph_index):
2372
"""See _KnitGraphIndex.scan_unvalidated_index."""
2373
# Because kndx files do not support atomic insertion via separate index
2374
# files, they do not support this method.
2375
raise NotImplementedError(self.scan_unvalidated_index)
2377
def get_missing_compression_parents(self):
2378
"""See _KnitGraphIndex.get_missing_compression_parents."""
2379
# Because kndx files do not support atomic insertion via separate index
2380
# files, they do not support this method.
2381
raise NotImplementedError(self.get_missing_compression_parents)
2383
def _cache_key(self, key, options, pos, size, parent_keys):
2384
"""Cache a version record in the history array and index cache.
2386
This is inlined into _load_data for performance. KEEP IN SYNC.
2387
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2391
version_id = key[-1]
2392
# last-element only for compatibilty with the C load_data.
2393
parents = tuple(parent[-1] for parent in parent_keys)
2394
for parent in parent_keys:
2395
if parent[:-1] != prefix:
2396
raise ValueError("mismatched prefixes for %r, %r" % (
2398
cache, history = self._kndx_cache[prefix]
2399
# only want the _history index to reference the 1st index entry
2401
if version_id not in cache:
2402
index = len(history)
2403
history.append(version_id)
2405
index = cache[version_id][5]
2406
cache[version_id] = (version_id,
2413
def check_header(self, fp):
2414
line = fp.readline()
2416
# An empty file can actually be treated as though the file doesn't
2418
raise errors.NoSuchFile(self)
2419
if line != self.HEADER:
2420
raise KnitHeaderError(badline=line, filename=self)
2422
def _check_read(self):
2423
if not self._is_locked():
2424
raise errors.ObjectNotLocked(self)
2425
if self._get_scope() != self._scope:
2428
def _check_write_ok(self):
2429
"""Assert if not writes are permitted."""
2430
if not self._is_locked():
2431
raise errors.ObjectNotLocked(self)
2432
if self._get_scope() != self._scope:
2434
if self._mode != 'w':
2435
raise errors.ReadOnlyObjectDirtiedError(self)
2437
def get_build_details(self, keys):
2438
"""Get the method, index_memo and compression parent for keys.
2440
Ghosts are omitted from the result.
2442
:param keys: An iterable of keys.
2443
:return: A dict of key:(index_memo, compression_parent, parents,
2446
opaque structure to pass to read_records to extract the raw
2449
Content that this record is built upon, may be None
2451
Logical parents of this node
2453
extra information about the content which needs to be passed to
2454
Factory.parse_record
2456
parent_map = self.get_parent_map(keys)
2459
if key not in parent_map:
2461
method = self.get_method(key)
2462
parents = parent_map[key]
2463
if method == 'fulltext':
2464
compression_parent = None
2466
compression_parent = parents[0]
2467
noeol = 'no-eol' in self.get_options(key)
2468
index_memo = self.get_position(key)
2469
result[key] = (index_memo, compression_parent,
2470
parents, (method, noeol))
2473
def get_method(self, key):
2474
"""Return compression method of specified key."""
2475
options = self.get_options(key)
2476
if 'fulltext' in options:
2478
elif 'line-delta' in options:
2481
raise errors.KnitIndexUnknownMethod(self, options)
2483
def get_options(self, key):
2484
"""Return a list representing options.
2488
prefix, suffix = self._split_key(key)
2489
self._load_prefixes([prefix])
2491
return self._kndx_cache[prefix][0][suffix][1]
2493
raise RevisionNotPresent(key, self)
2495
def get_parent_map(self, keys):
2496
"""Get a map of the parents of keys.
2498
:param keys: The keys to look up parents for.
2499
:return: A mapping from keys to parents. Absent keys are absent from
2502
# Parse what we need to up front, this potentially trades off I/O
2503
# locality (.kndx and .knit in the same block group for the same file
2504
# id) for less checking in inner loops.
2505
prefixes = set(key[:-1] for key in keys)
2506
self._load_prefixes(prefixes)
2511
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2515
result[key] = tuple(prefix + (suffix,) for
2516
suffix in suffix_parents)
2519
def get_position(self, key):
2520
"""Return details needed to access the version.
2522
:return: a tuple (key, data position, size) to hand to the access
2523
logic to get the record.
2525
prefix, suffix = self._split_key(key)
2526
self._load_prefixes([prefix])
2527
entry = self._kndx_cache[prefix][0][suffix]
2528
return key, entry[2], entry[3]
2530
has_key = _mod_index._has_key_from_parent_map
2532
def _init_index(self, path, extra_lines=[]):
2533
"""Initialize an index."""
2535
sio.write(self.HEADER)
2536
sio.writelines(extra_lines)
2538
self._transport.put_file_non_atomic(path, sio,
2539
create_parent_dir=True)
2540
# self._create_parent_dir)
2541
# mode=self._file_mode,
2542
# dir_mode=self._dir_mode)
2545
"""Get all the keys in the collection.
2547
The keys are not ordered.
2550
# Identify all key prefixes.
2551
# XXX: A bit hacky, needs polish.
2552
if type(self._mapper) == ConstantMapper:
2556
for quoted_relpath in self._transport.iter_files_recursive():
2557
path, ext = os.path.splitext(quoted_relpath)
2559
prefixes = [self._mapper.unmap(path) for path in relpaths]
2560
self._load_prefixes(prefixes)
2561
for prefix in prefixes:
2562
for suffix in self._kndx_cache[prefix][1]:
2563
result.add(prefix + (suffix,))
2566
def _load_prefixes(self, prefixes):
2567
"""Load the indices for prefixes."""
2569
for prefix in prefixes:
2570
if prefix not in self._kndx_cache:
2571
# the load_data interface writes to these variables.
2574
self._filename = prefix
2576
path = self._mapper.map(prefix) + '.kndx'
2577
fp = self._transport.get(path)
2579
# _load_data may raise NoSuchFile if the target knit is
2581
_load_data(self, fp)
2584
self._kndx_cache[prefix] = (self._cache, self._history)
2589
self._kndx_cache[prefix] = ({}, [])
2590
if type(self._mapper) == ConstantMapper:
2591
# preserve behaviour for revisions.kndx etc.
2592
self._init_index(path)
2597
missing_keys = _mod_index._missing_keys_from_parent_map
2599
def _partition_keys(self, keys):
2600
"""Turn keys into a dict of prefix:suffix_list."""
2603
prefix_keys = result.setdefault(key[:-1], [])
2604
prefix_keys.append(key[-1])
2607
def _dictionary_compress(self, keys):
2608
"""Dictionary compress keys.
2610
:param keys: The keys to generate references to.
2611
:return: A string representation of keys. keys which are present are
2612
dictionary compressed, and others are emitted as fulltext with a
2618
prefix = keys[0][:-1]
2619
cache = self._kndx_cache[prefix][0]
2621
if key[:-1] != prefix:
2622
# kndx indices cannot refer across partitioned storage.
2623
raise ValueError("mismatched prefixes for %r" % keys)
2624
if key[-1] in cache:
2625
# -- inlined lookup() --
2626
result_list.append(str(cache[key[-1]][5]))
2627
# -- end lookup () --
2629
result_list.append('.' + key[-1])
2630
return ' '.join(result_list)
2632
def _reset_cache(self):
2633
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2634
# (cache_dict, history_vector) for parsed kndx files.
2635
self._kndx_cache = {}
2636
self._scope = self._get_scope()
2637
allow_writes = self._allow_writes()
2643
def _sort_keys_by_io(self, keys, positions):
2644
"""Figure out an optimal order to read the records for the given keys.
2646
Sort keys, grouped by index and sorted by position.
2648
:param keys: A list of keys whose records we want to read. This will be
2650
:param positions: A dict, such as the one returned by
2651
_get_components_positions()
2654
def get_sort_key(key):
2655
index_memo = positions[key][1]
2656
# Group by prefix and position. index_memo[0] is the key, so it is
2657
# (file_id, revision_id) and we don't want to sort on revision_id,
2658
# index_memo[1] is the position, and index_memo[2] is the size,
2659
# which doesn't matter for the sort
2660
return index_memo[0][:-1], index_memo[1]
2661
return keys.sort(key=get_sort_key)
2663
_get_total_build_size = _get_total_build_size
2665
def _split_key(self, key):
2666
"""Split key into a prefix and suffix."""
2667
return key[:-1], key[-1]
2670
class _KnitGraphIndex(object):
2671
"""A KnitVersionedFiles index layered on GraphIndex."""
2673
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2675
"""Construct a KnitGraphIndex on a graph_index.
2677
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2678
:param is_locked: A callback to check whether the object should answer
2680
:param deltas: Allow delta-compressed records.
2681
:param parents: If True, record knits parents, if not do not record
2683
:param add_callback: If not None, allow additions to the index and call
2684
this callback with a list of added GraphIndex nodes:
2685
[(node, value, node_refs), ...]
2686
:param is_locked: A callback, returns True if the index is locked and
2689
self._add_callback = add_callback
2690
self._graph_index = graph_index
2691
self._deltas = deltas
2692
self._parents = parents
2693
if deltas and not parents:
2694
# XXX: TODO: Delta tree and parent graph should be conceptually
2696
raise KnitCorrupt(self, "Cannot do delta compression without "
2698
self.has_graph = parents
2699
self._is_locked = is_locked
2700
self._missing_compression_parents = set()
2703
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2705
def add_records(self, records, random_id=False,
2706
missing_compression_parents=False):
2707
"""Add multiple records to the index.
2709
This function does not insert data into the Immutable GraphIndex
2710
backing the KnitGraphIndex, instead it prepares data for insertion by
2711
the caller and checks that it is safe to insert then calls
2712
self._add_callback with the prepared GraphIndex nodes.
2714
:param records: a list of tuples:
2715
(key, options, access_memo, parents).
2716
:param random_id: If True the ids being added were randomly generated
2717
and no check for existence will be performed.
2718
:param missing_compression_parents: If True the records being added are
2719
only compressed against texts already in the index (or inside
2720
records). If False the records all refer to unavailable texts (or
2721
texts inside records) as compression parents.
2723
if not self._add_callback:
2724
raise errors.ReadOnlyError(self)
2725
# we hope there are no repositories with inconsistent parentage
2729
compression_parents = set()
2730
for (key, options, access_memo, parents) in records:
2732
parents = tuple(parents)
2733
index, pos, size = access_memo
2734
if 'no-eol' in options:
2738
value += "%d %d" % (pos, size)
2739
if not self._deltas:
2740
if 'line-delta' in options:
2741
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2744
if 'line-delta' in options:
2745
node_refs = (parents, (parents[0],))
2746
if missing_compression_parents:
2747
compression_parents.add(parents[0])
2749
node_refs = (parents, ())
2751
node_refs = (parents, )
2754
raise KnitCorrupt(self, "attempt to add node with parents "
2755
"in parentless index.")
2757
keys[key] = (value, node_refs)
2760
present_nodes = self._get_entries(keys)
2761
for (index, key, value, node_refs) in present_nodes:
2762
if (value[0] != keys[key][0][0] or
2763
node_refs[:1] != keys[key][1][:1]):
2764
raise KnitCorrupt(self, "inconsistent details in add_records"
2765
": %s %s" % ((value, node_refs), keys[key]))
2769
for key, (value, node_refs) in keys.iteritems():
2770
result.append((key, value, node_refs))
2772
for key, (value, node_refs) in keys.iteritems():
2773
result.append((key, value))
2774
self._add_callback(result)
2775
if missing_compression_parents:
2776
# This may appear to be incorrect (it does not check for
2777
# compression parents that are in the existing graph index),
2778
# but such records won't have been buffered, so this is
2779
# actually correct: every entry when
2780
# missing_compression_parents==True either has a missing parent, or
2781
# a parent that is one of the keys in records.
2782
compression_parents.difference_update(keys)
2783
self._missing_compression_parents.update(compression_parents)
2784
# Adding records may have satisfied missing compression parents.
2785
self._missing_compression_parents.difference_update(keys)
2787
def scan_unvalidated_index(self, graph_index):
2788
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2790
This allows this _KnitGraphIndex to keep track of any missing
2791
compression parents we may want to have filled in to make those
2794
:param graph_index: A GraphIndex
2797
new_missing = graph_index.external_references(ref_list_num=1)
2798
new_missing.difference_update(self.get_parent_map(new_missing))
2799
self._missing_compression_parents.update(new_missing)
2801
def get_missing_compression_parents(self):
2802
"""Return the keys of missing compression parents.
2804
Missing compression parents occur when a record stream was missing
2805
basis texts, or a index was scanned that had missing basis texts.
2807
return frozenset(self._missing_compression_parents)
2809
def _check_read(self):
2810
"""raise if reads are not permitted."""
2811
if not self._is_locked():
2812
raise errors.ObjectNotLocked(self)
2814
def _check_write_ok(self):
2815
"""Assert if writes are not permitted."""
2816
if not self._is_locked():
2817
raise errors.ObjectNotLocked(self)
2819
def _compression_parent(self, an_entry):
2820
# return the key that an_entry is compressed against, or None
2821
# Grab the second parent list (as deltas implies parents currently)
2822
compression_parents = an_entry[3][1]
2823
if not compression_parents:
2825
if len(compression_parents) != 1:
2826
raise AssertionError(
2827
"Too many compression parents: %r" % compression_parents)
2828
return compression_parents[0]
2830
def get_build_details(self, keys):
2831
"""Get the method, index_memo and compression parent for version_ids.
2833
Ghosts are omitted from the result.
2835
:param keys: An iterable of keys.
2836
:return: A dict of key:
2837
(index_memo, compression_parent, parents, record_details).
2839
opaque structure to pass to read_records to extract the raw
2842
Content that this record is built upon, may be None
2844
Logical parents of this node
2846
extra information about the content which needs to be passed to
2847
Factory.parse_record
2851
entries = self._get_entries(keys, False)
2852
for entry in entries:
2854
if not self._parents:
2857
parents = entry[3][0]
2858
if not self._deltas:
2859
compression_parent_key = None
2861
compression_parent_key = self._compression_parent(entry)
2862
noeol = (entry[2][0] == 'N')
2863
if compression_parent_key:
2864
method = 'line-delta'
2867
result[key] = (self._node_to_position(entry),
2868
compression_parent_key, parents,
2872
def _get_entries(self, keys, check_present=False):
2873
"""Get the entries for keys.
2875
:param keys: An iterable of index key tuples.
2880
for node in self._graph_index.iter_entries(keys):
2882
found_keys.add(node[1])
2884
# adapt parentless index to the rest of the code.
2885
for node in self._graph_index.iter_entries(keys):
2886
yield node[0], node[1], node[2], ()
2887
found_keys.add(node[1])
2889
missing_keys = keys.difference(found_keys)
2891
raise RevisionNotPresent(missing_keys.pop(), self)
2893
def get_method(self, key):
2894
"""Return compression method of specified key."""
2895
return self._get_method(self._get_node(key))
2897
def _get_method(self, node):
2898
if not self._deltas:
2900
if self._compression_parent(node):
2905
def _get_node(self, key):
2907
return list(self._get_entries([key]))[0]
2909
raise RevisionNotPresent(key, self)
2911
def get_options(self, key):
2912
"""Return a list representing options.
2916
node = self._get_node(key)
2917
options = [self._get_method(node)]
2918
if node[2][0] == 'N':
2919
options.append('no-eol')
2922
def get_parent_map(self, keys):
2923
"""Get a map of the parents of keys.
2925
:param keys: The keys to look up parents for.
2926
:return: A mapping from keys to parents. Absent keys are absent from
2930
nodes = self._get_entries(keys)
2934
result[node[1]] = node[3][0]
2937
result[node[1]] = None
2940
def get_position(self, key):
2941
"""Return details needed to access the version.
2943
:return: a tuple (index, data position, size) to hand to the access
2944
logic to get the record.
2946
node = self._get_node(key)
2947
return self._node_to_position(node)
2949
has_key = _mod_index._has_key_from_parent_map
2952
"""Get all the keys in the collection.
2954
The keys are not ordered.
2957
return [node[1] for node in self._graph_index.iter_all_entries()]
2959
missing_keys = _mod_index._missing_keys_from_parent_map
2961
def _node_to_position(self, node):
2962
"""Convert an index value to position details."""
2963
bits = node[2][1:].split(' ')
2964
return node[0], int(bits[0]), int(bits[1])
2966
def _sort_keys_by_io(self, keys, positions):
2967
"""Figure out an optimal order to read the records for the given keys.
2969
Sort keys, grouped by index and sorted by position.
2971
:param keys: A list of keys whose records we want to read. This will be
2973
:param positions: A dict, such as the one returned by
2974
_get_components_positions()
2977
def get_index_memo(key):
2978
# index_memo is at offset [1]. It is made up of (GraphIndex,
2979
# position, size). GI is an object, which will be unique for each
2980
# pack file. This causes us to group by pack file, then sort by
2981
# position. Size doesn't matter, but it isn't worth breaking up the
2983
return positions[key][1]
2984
return keys.sort(key=get_index_memo)
2986
_get_total_build_size = _get_total_build_size
2989
class _KnitKeyAccess(object):
2990
"""Access to records in .knit files."""
2992
def __init__(self, transport, mapper):
2993
"""Create a _KnitKeyAccess with transport and mapper.
2995
:param transport: The transport the access object is rooted at.
2996
:param mapper: The mapper used to map keys to .knit files.
2998
self._transport = transport
2999
self._mapper = mapper
3001
def add_raw_records(self, key_sizes, raw_data):
3002
"""Add raw knit bytes to a storage area.
3004
The data is spooled to the container writer in one bytes-record per
3007
:param sizes: An iterable of tuples containing the key and size of each
3009
:param raw_data: A bytestring containing the data.
3010
:return: A list of memos to retrieve the record later. Each memo is an
3011
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3012
length), where the key is the record key.
3014
if type(raw_data) != str:
3015
raise AssertionError(
3016
'data must be plain bytes was %s' % type(raw_data))
3019
# TODO: This can be tuned for writing to sftp and other servers where
3020
# append() is relatively expensive by grouping the writes to each key
3022
for key, size in key_sizes:
3023
path = self._mapper.map(key)
3025
base = self._transport.append_bytes(path + '.knit',
3026
raw_data[offset:offset+size])
3027
except errors.NoSuchFile:
3028
self._transport.mkdir(osutils.dirname(path))
3029
base = self._transport.append_bytes(path + '.knit',
3030
raw_data[offset:offset+size])
3034
result.append((key, base, size))
3037
def get_raw_records(self, memos_for_retrieval):
3038
"""Get the raw bytes for a records.
3040
:param memos_for_retrieval: An iterable containing the access memo for
3041
retrieving the bytes.
3042
:return: An iterator over the bytes of the records.
3044
# first pass, group into same-index request to minimise readv's issued.
3046
current_prefix = None
3047
for (key, offset, length) in memos_for_retrieval:
3048
if current_prefix == key[:-1]:
3049
current_list.append((offset, length))
3051
if current_prefix is not None:
3052
request_lists.append((current_prefix, current_list))
3053
current_prefix = key[:-1]
3054
current_list = [(offset, length)]
3055
# handle the last entry
3056
if current_prefix is not None:
3057
request_lists.append((current_prefix, current_list))
3058
for prefix, read_vector in request_lists:
3059
path = self._mapper.map(prefix) + '.knit'
3060
for pos, data in self._transport.readv(path, read_vector):
3064
class _DirectPackAccess(object):
3065
"""Access to data in one or more packs with less translation."""
3067
def __init__(self, index_to_packs, reload_func=None):
3068
"""Create a _DirectPackAccess object.
3070
:param index_to_packs: A dict mapping index objects to the transport
3071
and file names for obtaining data.
3072
:param reload_func: A function to call if we determine that the pack
3073
files have moved and we need to reload our caches. See
3074
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
3076
self._container_writer = None
3077
self._write_index = None
3078
self._indices = index_to_packs
3079
self._reload_func = reload_func
3081
def add_raw_records(self, key_sizes, raw_data):
3082
"""Add raw knit bytes to a storage area.
3084
The data is spooled to the container writer in one bytes-record per
3087
:param sizes: An iterable of tuples containing the key and size of each
3089
:param raw_data: A bytestring containing the data.
3090
:return: A list of memos to retrieve the record later. Each memo is an
3091
opaque index memo. For _DirectPackAccess the memo is (index, pos,
3092
length), where the index field is the write_index object supplied
3093
to the PackAccess object.
3095
if type(raw_data) != str:
3096
raise AssertionError(
3097
'data must be plain bytes was %s' % type(raw_data))
3100
for key, size in key_sizes:
3101
p_offset, p_length = self._container_writer.add_bytes_record(
3102
raw_data[offset:offset+size], [])
3104
result.append((self._write_index, p_offset, p_length))
3107
def get_raw_records(self, memos_for_retrieval):
3108
"""Get the raw bytes for a records.
3110
:param memos_for_retrieval: An iterable containing the (index, pos,
3111
length) memo for retrieving the bytes. The Pack access method
3112
looks up the pack to use for a given record in its index_to_pack
3114
:return: An iterator over the bytes of the records.
3116
# first pass, group into same-index requests
3118
current_index = None
3119
for (index, offset, length) in memos_for_retrieval:
3120
if current_index == index:
3121
current_list.append((offset, length))
3123
if current_index is not None:
3124
request_lists.append((current_index, current_list))
3125
current_index = index
3126
current_list = [(offset, length)]
3127
# handle the last entry
3128
if current_index is not None:
3129
request_lists.append((current_index, current_list))
3130
for index, offsets in request_lists:
3132
transport, path = self._indices[index]
3134
# A KeyError here indicates that someone has triggered an index
3135
# reload, and this index has gone missing, we need to start
3137
if self._reload_func is None:
3138
# If we don't have a _reload_func there is nothing that can
3141
raise errors.RetryWithNewPacks(index,
3142
reload_occurred=True,
3143
exc_info=sys.exc_info())
3145
reader = pack.make_readv_reader(transport, path, offsets)
3146
for names, read_func in reader.iter_records():
3147
yield read_func(None)
3148
except errors.NoSuchFile:
3149
# A NoSuchFile error indicates that a pack file has gone
3150
# missing on disk, we need to trigger a reload, and start over.
3151
if self._reload_func is None:
3153
raise errors.RetryWithNewPacks(transport.abspath(path),
3154
reload_occurred=False,
3155
exc_info=sys.exc_info())
3157
def set_writer(self, writer, index, transport_packname):
3158
"""Set a writer to use for adding data."""
3159
if index is not None:
3160
self._indices[index] = transport_packname
3161
self._container_writer = writer
3162
self._write_index = index
3164
def reload_or_raise(self, retry_exc):
3165
"""Try calling the reload function, or re-raise the original exception.
3167
This should be called after _DirectPackAccess raises a
3168
RetryWithNewPacks exception. This function will handle the common logic
3169
of determining when the error is fatal versus being temporary.
3170
It will also make sure that the original exception is raised, rather
3171
than the RetryWithNewPacks exception.
3173
If this function returns, then the calling function should retry
3174
whatever operation was being performed. Otherwise an exception will
3177
:param retry_exc: A RetryWithNewPacks exception.
3180
if self._reload_func is None:
3182
elif not self._reload_func():
3183
# The reload claimed that nothing changed
3184
if not retry_exc.reload_occurred:
3185
# If there wasn't an earlier reload, then we really were
3186
# expecting to find changes. We didn't find them, so this is a
3190
exc_class, exc_value, exc_traceback = retry_exc.exc_info
3191
raise exc_class, exc_value, exc_traceback
3194
# Deprecated, use PatienceSequenceMatcher instead
3195
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
3198
def annotate_knit(knit, revision_id):
3199
"""Annotate a knit with no cached annotations.
3201
This implementation is for knits with no cached annotations.
3202
It will work for knits with cached annotations, but this is not
3205
annotator = _KnitAnnotator(knit)
3206
return iter(annotator.annotate(revision_id))
3209
class _KnitAnnotator(object):
3210
"""Build up the annotations for a text."""
3212
def __init__(self, knit):
3215
# Content objects, differs from fulltexts because of how final newlines
3216
# are treated by knits. the content objects here will always have a
3218
self._fulltext_contents = {}
3220
# Annotated lines of specific revisions
3221
self._annotated_lines = {}
3223
# Track the raw data for nodes that we could not process yet.
3224
# This maps the revision_id of the base to a list of children that will
3225
# annotated from it.
3226
self._pending_children = {}
3228
# Nodes which cannot be extracted
3229
self._ghosts = set()
3231
# Track how many children this node has, so we know if we need to keep
3233
self._annotate_children = {}
3234
self._compression_children = {}
3236
self._all_build_details = {}
3237
# The children => parent revision_id graph
3238
self._revision_id_graph = {}
3240
self._heads_provider = None
3242
self._nodes_to_keep_annotations = set()
3243
self._generations_until_keep = 100
3245
def set_generations_until_keep(self, value):
3246
"""Set the number of generations before caching a node.
3248
Setting this to -1 will cache every merge node, setting this higher
3249
will cache fewer nodes.
3251
self._generations_until_keep = value
3253
def _add_fulltext_content(self, revision_id, content_obj):
3254
self._fulltext_contents[revision_id] = content_obj
3255
# TODO: jam 20080305 It might be good to check the sha1digest here
3256
return content_obj.text()
3258
def _check_parents(self, child, nodes_to_annotate):
3259
"""Check if all parents have been processed.
3261
:param child: A tuple of (rev_id, parents, raw_content)
3262
:param nodes_to_annotate: If child is ready, add it to
3263
nodes_to_annotate, otherwise put it back in self._pending_children
3265
for parent_id in child[1]:
3266
if (parent_id not in self._annotated_lines):
3267
# This parent is present, but another parent is missing
3268
self._pending_children.setdefault(parent_id,
3272
# This one is ready to be processed
3273
nodes_to_annotate.append(child)
3275
def _add_annotation(self, revision_id, fulltext, parent_ids,
3276
left_matching_blocks=None):
3277
"""Add an annotation entry.
3279
All parents should already have been annotated.
3280
:return: A list of children that now have their parents satisfied.
3282
a = self._annotated_lines
3283
annotated_parent_lines = [a[p] for p in parent_ids]
3284
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
3285
fulltext, revision_id, left_matching_blocks,
3286
heads_provider=self._get_heads_provider()))
3287
self._annotated_lines[revision_id] = annotated_lines
3288
for p in parent_ids:
3289
ann_children = self._annotate_children[p]
3290
ann_children.remove(revision_id)
3291
if (not ann_children
3292
and p not in self._nodes_to_keep_annotations):
3293
del self._annotated_lines[p]
3294
del self._all_build_details[p]
3295
if p in self._fulltext_contents:
3296
del self._fulltext_contents[p]
3297
# Now that we've added this one, see if there are any pending
3298
# deltas to be done, certainly this parent is finished
3299
nodes_to_annotate = []
3300
for child in self._pending_children.pop(revision_id, []):
3301
self._check_parents(child, nodes_to_annotate)
3302
return nodes_to_annotate
3304
def _get_build_graph(self, key):
3305
"""Get the graphs for building texts and annotations.
3307
The data you need for creating a full text may be different than the
3308
data you need to annotate that text. (At a minimum, you need both
3309
parents to create an annotation, but only need 1 parent to generate the
3312
:return: A list of (key, index_memo) records, suitable for
3313
passing to read_records_iter to start reading in the raw data fro/
3316
if key in self._annotated_lines:
3319
pending = set([key])
3324
# get all pending nodes
3326
this_iteration = pending
3327
build_details = self._knit._index.get_build_details(this_iteration)
3328
self._all_build_details.update(build_details)
3329
# new_nodes = self._knit._index._get_entries(this_iteration)
3331
for key, details in build_details.iteritems():
3332
(index_memo, compression_parent, parents,
3333
record_details) = details
3334
self._revision_id_graph[key] = parents
3335
records.append((key, index_memo))
3336
# Do we actually need to check _annotated_lines?
3337
pending.update(p for p in parents
3338
if p not in self._all_build_details)
3339
if compression_parent:
3340
self._compression_children.setdefault(compression_parent,
3343
for parent in parents:
3344
self._annotate_children.setdefault(parent,
3346
num_gens = generation - kept_generation
3347
if ((num_gens >= self._generations_until_keep)
3348
and len(parents) > 1):
3349
kept_generation = generation
3350
self._nodes_to_keep_annotations.add(key)
3352
missing_versions = this_iteration.difference(build_details.keys())
3353
self._ghosts.update(missing_versions)
3354
for missing_version in missing_versions:
3355
# add a key, no parents
3356
self._revision_id_graph[missing_version] = ()
3357
pending.discard(missing_version) # don't look for it
3358
if self._ghosts.intersection(self._compression_children):
3360
"We cannot have nodes which have a ghost compression parent:\n"
3362
"compression children: %r"
3363
% (self._ghosts, self._compression_children))
3364
# Cleanout anything that depends on a ghost so that we don't wait for
3365
# the ghost to show up
3366
for node in self._ghosts:
3367
if node in self._annotate_children:
3368
# We won't be building this node
3369
del self._annotate_children[node]
3370
# Generally we will want to read the records in reverse order, because
3371
# we find the parent nodes after the children
3375
def _annotate_records(self, records):
3376
"""Build the annotations for the listed records."""
3377
# We iterate in the order read, rather than a strict order requested
3378
# However, process what we can, and put off to the side things that
3379
# still need parents, cleaning them up when those parents are
3381
for (rev_id, record,
3382
digest) in self._knit._read_records_iter(records):
3383
if rev_id in self._annotated_lines:
3385
parent_ids = self._revision_id_graph[rev_id]
3386
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3387
details = self._all_build_details[rev_id]
3388
(index_memo, compression_parent, parents,
3389
record_details) = details
3390
nodes_to_annotate = []
3391
# TODO: Remove the punning between compression parents, and
3392
# parent_ids, we should be able to do this without assuming
3394
if len(parent_ids) == 0:
3395
# There are no parents for this node, so just add it
3396
# TODO: This probably needs to be decoupled
3397
fulltext_content, delta = self._knit._factory.parse_record(
3398
rev_id, record, record_details, None)
3399
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3400
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3401
parent_ids, left_matching_blocks=None))
3403
child = (rev_id, parent_ids, record)
3404
# Check if all the parents are present
3405
self._check_parents(child, nodes_to_annotate)
3406
while nodes_to_annotate:
3407
# Should we use a queue here instead of a stack?
3408
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3409
(index_memo, compression_parent, parents,
3410
record_details) = self._all_build_details[rev_id]
3412
if compression_parent is not None:
3413
comp_children = self._compression_children[compression_parent]
3414
if rev_id not in comp_children:
3415
raise AssertionError("%r not in compression children %r"
3416
% (rev_id, comp_children))
3417
# If there is only 1 child, it is safe to reuse this
3419
reuse_content = (len(comp_children) == 1
3420
and compression_parent not in
3421
self._nodes_to_keep_annotations)
3423
# Remove it from the cache since it will be changing
3424
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3425
# Make sure to copy the fulltext since it might be
3427
parent_fulltext = list(parent_fulltext_content.text())
3429
parent_fulltext_content = self._fulltext_contents[compression_parent]
3430
parent_fulltext = parent_fulltext_content.text()
3431
comp_children.remove(rev_id)
3432
fulltext_content, delta = self._knit._factory.parse_record(
3433
rev_id, record, record_details,
3434
parent_fulltext_content,
3435
copy_base_content=(not reuse_content))
3436
fulltext = self._add_fulltext_content(rev_id,
3438
if compression_parent == parent_ids[0]:
3439
# the compression_parent is the left parent, so we can
3441
blocks = KnitContent.get_line_delta_blocks(delta,
3442
parent_fulltext, fulltext)
3444
fulltext_content = self._knit._factory.parse_fulltext(
3446
fulltext = self._add_fulltext_content(rev_id,
3448
nodes_to_annotate.extend(
3449
self._add_annotation(rev_id, fulltext, parent_ids,
3450
left_matching_blocks=blocks))
3452
def _get_heads_provider(self):
3453
"""Create a heads provider for resolving ancestry issues."""
3454
if self._heads_provider is not None:
3455
return self._heads_provider
3456
parent_provider = _mod_graph.DictParentsProvider(
3457
self._revision_id_graph)
3458
graph_obj = _mod_graph.Graph(parent_provider)
3459
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3460
self._heads_provider = head_cache
3463
def annotate(self, key):
3464
"""Return the annotated fulltext at the given key.
3466
:param key: The key to annotate.
3468
if len(self._knit._fallback_vfs) > 0:
3469
# stacked knits can't use the fast path at present.
3470
return self._simple_annotate(key)
3473
records = self._get_build_graph(key)
3474
if key in self._ghosts:
3475
raise errors.RevisionNotPresent(key, self._knit)
3476
self._annotate_records(records)
3477
return self._annotated_lines[key]
3478
except errors.RetryWithNewPacks, e:
3479
self._knit._access.reload_or_raise(e)
3480
# The cached build_details are no longer valid
3481
self._all_build_details.clear()
3483
def _simple_annotate(self, key):
3484
"""Return annotated fulltext, rediffing from the full texts.
3486
This is slow but makes no assumptions about the repository
3487
being able to produce line deltas.
3489
# TODO: this code generates a parent maps of present ancestors; it
3490
# could be split out into a separate method, and probably should use
3491
# iter_ancestry instead. -- mbp and robertc 20080704
3492
graph = _mod_graph.Graph(self._knit)
3493
head_cache = _mod_graph.FrozenHeadsCache(graph)
3494
search = graph._make_breadth_first_searcher([key])
3498
present, ghosts = search.next_with_ghosts()
3499
except StopIteration:
3501
keys.update(present)
3502
parent_map = self._knit.get_parent_map(keys)
3504
reannotate = annotate.reannotate
3505
for record in self._knit.get_record_stream(keys, 'topological', True):
3507
fulltext = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
3508
parents = parent_map[key]
3509
if parents is not None:
3510
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
3513
parent_cache[key] = list(
3514
reannotate(parent_lines, fulltext, key, None, head_cache))
3516
return parent_cache[key]
3518
raise errors.RevisionNotPresent(key, self._knit)
3522
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3524
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
