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
"""For the given keys, group them into 'best-sized' requests.
1262
The idea is to avoid making 1 request per file, but to never try to
1263
unpack an entire 1.5GB source tree in a single pass. Also when
1264
possible, we should try to group requests to the same pack file
1267
:return: yield (keys, non_local) tuples that indicate what keys should
1270
# TODO: Ideally we would group on 2 factors. We want to extract texts
1271
# from the same pack file together, and we want to extract all
1272
# the texts for a given build-chain together. Ultimately it
1273
# probably needs a better global view.
1274
total_keys = len(keys)
1275
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1276
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1278
cur_non_local = set()
1282
for prefix in prefix_order:
1283
keys = prefix_split_keys[prefix]
1284
non_local = prefix_split_non_local_keys.get(prefix, [])
1286
this_size = self._index._get_total_build_size(keys, positions)
1287
cur_size += this_size
1288
cur_keys.extend(keys)
1289
cur_non_local.update(non_local)
1290
if cur_size > _STREAM_MIN_BUFFER_SIZE:
1291
result.append((cur_keys, cur_non_local))
1292
sizes.append(cur_size)
1294
cur_non_local = set()
1297
result.append((cur_keys, cur_non_local))
1298
sizes.append(cur_size)
1299
trace.mutter('Collapsed %d keys into %d requests w/ %d file_ids'
1300
' w/ sizes: %s', total_keys, len(result),
1301
len(prefix_split_keys), sizes)
1304
def get_record_stream(self, keys, ordering, include_delta_closure):
1305
"""Get a stream of records for keys.
1307
:param keys: The keys to include.
1308
:param ordering: Either 'unordered' or 'topological'. A topologically
1309
sorted stream has compression parents strictly before their
1311
:param include_delta_closure: If True then the closure across any
1312
compression parents will be included (in the opaque data).
1313
:return: An iterator of ContentFactory objects, each of which is only
1314
valid until the iterator is advanced.
1316
# keys might be a generator
1320
if not self._index.has_graph:
1321
# Cannot topological order when no graph has been stored.
1322
ordering = 'unordered'
1324
remaining_keys = keys
1327
keys = set(remaining_keys)
1328
for content_factory in self._get_remaining_record_stream(keys,
1329
ordering, include_delta_closure):
1330
remaining_keys.discard(content_factory.key)
1331
yield content_factory
1333
except errors.RetryWithNewPacks, e:
1334
self._access.reload_or_raise(e)
1336
def _get_remaining_record_stream(self, keys, ordering,
1337
include_delta_closure):
1338
"""This function is the 'retry' portion for get_record_stream."""
1339
if include_delta_closure:
1340
positions = self._get_components_positions(keys, allow_missing=True)
1342
build_details = self._index.get_build_details(keys)
1344
# (record_details, access_memo, compression_parent_key)
1345
positions = dict((key, self._build_details_to_components(details))
1346
for key, details in build_details.iteritems())
1347
absent_keys = keys.difference(set(positions))
1348
# There may be more absent keys : if we're missing the basis component
1349
# and are trying to include the delta closure.
1350
# XXX: We should not ever need to examine remote sources because we do
1351
# not permit deltas across versioned files boundaries.
1352
if include_delta_closure:
1353
needed_from_fallback = set()
1354
# Build up reconstructable_keys dict. key:True in this dict means
1355
# the key can be reconstructed.
1356
reconstructable_keys = {}
1360
chain = [key, positions[key][2]]
1362
needed_from_fallback.add(key)
1365
while chain[-1] is not None:
1366
if chain[-1] in reconstructable_keys:
1367
result = reconstructable_keys[chain[-1]]
1371
chain.append(positions[chain[-1]][2])
1373
# missing basis component
1374
needed_from_fallback.add(chain[-1])
1377
for chain_key in chain[:-1]:
1378
reconstructable_keys[chain_key] = result
1380
needed_from_fallback.add(key)
1381
# Double index lookups here : need a unified api ?
1382
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1383
if ordering == 'topological':
1384
# Global topological sort
1385
present_keys = tsort.topo_sort(global_map)
1386
# Now group by source:
1388
current_source = None
1389
for key in present_keys:
1390
for parent_map in parent_maps:
1391
if key in parent_map:
1392
key_source = parent_map
1394
if current_source is not key_source:
1395
source_keys.append((key_source, []))
1396
current_source = key_source
1397
source_keys[-1][1].append(key)
1399
if ordering != 'unordered':
1400
raise AssertionError('valid values for ordering are:'
1401
' "unordered" or "topological" not: %r'
1403
# Just group by source; remote sources first.
1406
for parent_map in reversed(parent_maps):
1407
source_keys.append((parent_map, []))
1408
for key in parent_map:
1409
present_keys.append(key)
1410
source_keys[-1][1].append(key)
1411
# We have been requested to return these records in an order that
1412
# suits us. So we ask the index to give us an optimally sorted
1414
for source, sub_keys in source_keys:
1415
if source is parent_maps[0]:
1416
# Only sort the keys for this VF
1417
self._index._sort_keys_by_io(sub_keys, positions)
1418
absent_keys = keys - set(global_map)
1419
for key in absent_keys:
1420
yield AbsentContentFactory(key)
1421
# restrict our view to the keys we can answer.
1422
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1423
# XXX: At that point we need to consider the impact of double reads by
1424
# utilising components multiple times.
1425
if include_delta_closure:
1426
# XXX: get_content_maps performs its own index queries; allow state
1428
non_local_keys = needed_from_fallback - absent_keys
1429
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1432
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1434
for record in generator.get_record_stream():
1437
for source, keys in source_keys:
1438
if source is parent_maps[0]:
1439
# this KnitVersionedFiles
1440
records = [(key, positions[key][1]) for key in keys]
1441
for key, raw_data, sha1 in self._read_records_iter_raw(records):
1442
(record_details, index_memo, _) = positions[key]
1443
yield KnitContentFactory(key, global_map[key],
1444
record_details, sha1, raw_data, self._factory.annotated, None)
1446
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1447
for record in vf.get_record_stream(keys, ordering,
1448
include_delta_closure):
1451
def get_sha1s(self, keys):
1452
"""See VersionedFiles.get_sha1s()."""
1454
record_map = self._get_record_map(missing, allow_missing=True)
1456
for key, details in record_map.iteritems():
1457
if key not in missing:
1459
# record entry 2 is the 'digest'.
1460
result[key] = details[2]
1461
missing.difference_update(set(result))
1462
for source in self._fallback_vfs:
1465
new_result = source.get_sha1s(missing)
1466
result.update(new_result)
1467
missing.difference_update(set(new_result))
1470
def insert_record_stream(self, stream):
1471
"""Insert a record stream into this container.
1473
:param stream: A stream of records to insert.
1475
:seealso VersionedFiles.get_record_stream:
1477
def get_adapter(adapter_key):
1479
return adapters[adapter_key]
1481
adapter_factory = adapter_registry.get(adapter_key)
1482
adapter = adapter_factory(self)
1483
adapters[adapter_key] = adapter
1486
if self._factory.annotated:
1487
# self is annotated, we need annotated knits to use directly.
1488
annotated = "annotated-"
1491
# self is not annotated, but we can strip annotations cheaply.
1493
convertibles = set(["knit-annotated-ft-gz"])
1494
if self._max_delta_chain:
1495
delta_types.add("knit-annotated-delta-gz")
1496
convertibles.add("knit-annotated-delta-gz")
1497
# The set of types we can cheaply adapt without needing basis texts.
1498
native_types = set()
1499
if self._max_delta_chain:
1500
native_types.add("knit-%sdelta-gz" % annotated)
1501
delta_types.add("knit-%sdelta-gz" % annotated)
1502
native_types.add("knit-%sft-gz" % annotated)
1503
knit_types = native_types.union(convertibles)
1505
# Buffer all index entries that we can't add immediately because their
1506
# basis parent is missing. We don't buffer all because generating
1507
# annotations may require access to some of the new records. However we
1508
# can't generate annotations from new deltas until their basis parent
1509
# is present anyway, so we get away with not needing an index that
1510
# includes the new keys.
1512
# See <http://launchpad.net/bugs/300177> about ordering of compression
1513
# parents in the records - to be conservative, we insist that all
1514
# parents must be present to avoid expanding to a fulltext.
1516
# key = basis_parent, value = index entry to add
1517
buffered_index_entries = {}
1518
for record in stream:
1519
parents = record.parents
1520
if record.storage_kind in delta_types:
1521
# TODO: eventually the record itself should track
1522
# compression_parent
1523
compression_parent = parents[0]
1525
compression_parent = None
1526
# Raise an error when a record is missing.
1527
if record.storage_kind == 'absent':
1528
raise RevisionNotPresent([record.key], self)
1529
elif ((record.storage_kind in knit_types)
1530
and (compression_parent is None
1531
or not self._fallback_vfs
1532
or self._index.has_key(compression_parent)
1533
or not self.has_key(compression_parent))):
1534
# we can insert the knit record literally if either it has no
1535
# compression parent OR we already have its basis in this kvf
1536
# OR the basis is not present even in the fallbacks. In the
1537
# last case it will either turn up later in the stream and all
1538
# will be well, or it won't turn up at all and we'll raise an
1541
# TODO: self.has_key is somewhat redundant with
1542
# self._index.has_key; we really want something that directly
1543
# asks if it's only present in the fallbacks. -- mbp 20081119
1544
if record.storage_kind not in native_types:
1546
adapter_key = (record.storage_kind, "knit-delta-gz")
1547
adapter = get_adapter(adapter_key)
1549
adapter_key = (record.storage_kind, "knit-ft-gz")
1550
adapter = get_adapter(adapter_key)
1551
bytes = adapter.get_bytes(record)
1553
# It's a knit record, it has a _raw_record field (even if
1554
# it was reconstituted from a network stream).
1555
bytes = record._raw_record
1556
options = [record._build_details[0]]
1557
if record._build_details[1]:
1558
options.append('no-eol')
1559
# Just blat it across.
1560
# Note: This does end up adding data on duplicate keys. As
1561
# modern repositories use atomic insertions this should not
1562
# lead to excessive growth in the event of interrupted fetches.
1563
# 'knit' repositories may suffer excessive growth, but as a
1564
# deprecated format this is tolerable. It can be fixed if
1565
# needed by in the kndx index support raising on a duplicate
1566
# add with identical parents and options.
1567
access_memo = self._access.add_raw_records(
1568
[(record.key, len(bytes))], bytes)[0]
1569
index_entry = (record.key, options, access_memo, parents)
1571
if 'fulltext' not in options:
1572
# Not a fulltext, so we need to make sure the compression
1573
# parent will also be present.
1574
# Note that pack backed knits don't need to buffer here
1575
# because they buffer all writes to the transaction level,
1576
# but we don't expose that difference at the index level. If
1577
# the query here has sufficient cost to show up in
1578
# profiling we should do that.
1580
# They're required to be physically in this
1581
# KnitVersionedFiles, not in a fallback.
1582
if not self._index.has_key(compression_parent):
1583
pending = buffered_index_entries.setdefault(
1584
compression_parent, [])
1585
pending.append(index_entry)
1588
self._index.add_records([index_entry])
1589
elif record.storage_kind == 'chunked':
1590
self.add_lines(record.key, parents,
1591
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1593
# Not suitable for direct insertion as a
1594
# delta, either because it's not the right format, or this
1595
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1596
# 0) or because it depends on a base only present in the
1599
# Try getting a fulltext directly from the record.
1600
bytes = record.get_bytes_as('fulltext')
1601
except errors.UnavailableRepresentation:
1602
adapter_key = record.storage_kind, 'fulltext'
1603
adapter = get_adapter(adapter_key)
1604
bytes = adapter.get_bytes(record)
1605
lines = split_lines(bytes)
1607
self.add_lines(record.key, parents, lines)
1608
except errors.RevisionAlreadyPresent:
1610
# Add any records whose basis parent is now available.
1611
added_keys = [record.key]
1613
key = added_keys.pop(0)
1614
if key in buffered_index_entries:
1615
index_entries = buffered_index_entries[key]
1616
self._index.add_records(index_entries)
1618
[index_entry[0] for index_entry in index_entries])
1619
del buffered_index_entries[key]
1620
if buffered_index_entries:
1621
# There were index entries buffered at the end of the stream,
1622
# So these need to be added (if the index supports holding such
1623
# entries for later insertion)
1624
for key in buffered_index_entries:
1625
index_entries = buffered_index_entries[key]
1626
self._index.add_records(index_entries,
1627
missing_compression_parents=True)
1629
def get_missing_compression_parent_keys(self):
1630
"""Return an iterable of keys of missing compression parents.
1632
Check this after calling insert_record_stream to find out if there are
1633
any missing compression parents. If there are, the records that
1634
depend on them are not able to be inserted safely. For atomic
1635
KnitVersionedFiles built on packs, the transaction should be aborted or
1636
suspended - commit will fail at this point. Nonatomic knits will error
1637
earlier because they have no staging area to put pending entries into.
1639
return self._index.get_missing_compression_parents()
1641
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1642
"""Iterate over the lines in the versioned files from keys.
1644
This may return lines from other keys. Each item the returned
1645
iterator yields is a tuple of a line and a text version that that line
1646
is present in (not introduced in).
1648
Ordering of results is in whatever order is most suitable for the
1649
underlying storage format.
1651
If a progress bar is supplied, it may be used to indicate progress.
1652
The caller is responsible for cleaning up progress bars (because this
1656
* Lines are normalised by the underlying store: they will all have \\n
1658
* Lines are returned in arbitrary order.
1659
* If a requested key did not change any lines (or didn't have any
1660
lines), it may not be mentioned at all in the result.
1662
:return: An iterator over (line, key).
1665
pb = progress.DummyProgress()
1671
# we don't care about inclusions, the caller cares.
1672
# but we need to setup a list of records to visit.
1673
# we need key, position, length
1675
build_details = self._index.get_build_details(keys)
1676
for key, details in build_details.iteritems():
1678
key_records.append((key, details[0]))
1679
records_iter = enumerate(self._read_records_iter(key_records))
1680
for (key_idx, (key, data, sha_value)) in records_iter:
1681
pb.update('Walking content.', key_idx, total)
1682
compression_parent = build_details[key][1]
1683
if compression_parent is None:
1685
line_iterator = self._factory.get_fulltext_content(data)
1688
line_iterator = self._factory.get_linedelta_content(data)
1689
# Now that we are yielding the data for this key, remove it
1692
# XXX: It might be more efficient to yield (key,
1693
# line_iterator) in the future. However for now, this is a
1694
# simpler change to integrate into the rest of the
1695
# codebase. RBC 20071110
1696
for line in line_iterator:
1699
except errors.RetryWithNewPacks, e:
1700
self._access.reload_or_raise(e)
1701
# If there are still keys we've not yet found, we look in the fallback
1702
# vfs, and hope to find them there. Note that if the keys are found
1703
# but had no changes or no content, the fallback may not return
1705
if keys and not self._fallback_vfs:
1706
# XXX: strictly the second parameter is meant to be the file id
1707
# but it's not easily accessible here.
1708
raise RevisionNotPresent(keys, repr(self))
1709
for source in self._fallback_vfs:
1713
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1714
source_keys.add(key)
1716
keys.difference_update(source_keys)
1717
pb.update('Walking content.', total, total)
1719
def _make_line_delta(self, delta_seq, new_content):
1720
"""Generate a line delta from delta_seq and new_content."""
1722
for op in delta_seq.get_opcodes():
1723
if op[0] == 'equal':
1725
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1728
def _merge_annotations(self, content, parents, parent_texts={},
1729
delta=None, annotated=None,
1730
left_matching_blocks=None):
1731
"""Merge annotations for content and generate deltas.
1733
This is done by comparing the annotations based on changes to the text
1734
and generating a delta on the resulting full texts. If annotations are
1735
not being created then a simple delta is created.
1737
if left_matching_blocks is not None:
1738
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1742
for parent_key in parents:
1743
merge_content = self._get_content(parent_key, parent_texts)
1744
if (parent_key == parents[0] and delta_seq is not None):
1747
seq = patiencediff.PatienceSequenceMatcher(
1748
None, merge_content.text(), content.text())
1749
for i, j, n in seq.get_matching_blocks():
1752
# this copies (origin, text) pairs across to the new
1753
# content for any line that matches the last-checked
1755
content._lines[j:j+n] = merge_content._lines[i:i+n]
1756
# XXX: Robert says the following block is a workaround for a
1757
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1758
if content._lines and content._lines[-1][1][-1] != '\n':
1759
# The copied annotation was from a line without a trailing EOL,
1760
# reinstate one for the content object, to ensure correct
1762
line = content._lines[-1][1] + '\n'
1763
content._lines[-1] = (content._lines[-1][0], line)
1765
if delta_seq is None:
1766
reference_content = self._get_content(parents[0], parent_texts)
1767
new_texts = content.text()
1768
old_texts = reference_content.text()
1769
delta_seq = patiencediff.PatienceSequenceMatcher(
1770
None, old_texts, new_texts)
1771
return self._make_line_delta(delta_seq, content)
1773
def _parse_record(self, version_id, data):
1774
"""Parse an original format knit record.
1776
These have the last element of the key only present in the stored data.
1778
rec, record_contents = self._parse_record_unchecked(data)
1779
self._check_header_version(rec, version_id)
1780
return record_contents, rec[3]
1782
def _parse_record_header(self, key, raw_data):
1783
"""Parse a record header for consistency.
1785
:return: the header and the decompressor stream.
1786
as (stream, header_record)
1788
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1791
rec = self._check_header(key, df.readline())
1792
except Exception, e:
1793
raise KnitCorrupt(self,
1794
"While reading {%s} got %s(%s)"
1795
% (key, e.__class__.__name__, str(e)))
1798
def _parse_record_unchecked(self, data):
1800
# 4168 calls in 2880 217 internal
1801
# 4168 calls to _parse_record_header in 2121
1802
# 4168 calls to readlines in 330
1803
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1805
record_contents = df.readlines()
1806
except Exception, e:
1807
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1808
(data, e.__class__.__name__, str(e)))
1809
header = record_contents.pop(0)
1810
rec = self._split_header(header)
1811
last_line = record_contents.pop()
1812
if len(record_contents) != int(rec[2]):
1813
raise KnitCorrupt(self,
1814
'incorrect number of lines %s != %s'
1815
' for version {%s} %s'
1816
% (len(record_contents), int(rec[2]),
1817
rec[1], record_contents))
1818
if last_line != 'end %s\n' % rec[1]:
1819
raise KnitCorrupt(self,
1820
'unexpected version end line %r, wanted %r'
1821
% (last_line, rec[1]))
1823
return rec, record_contents
1825
def _read_records_iter(self, records):
1826
"""Read text records from data file and yield result.
1828
The result will be returned in whatever is the fastest to read.
1829
Not by the order requested. Also, multiple requests for the same
1830
record will only yield 1 response.
1831
:param records: A list of (key, access_memo) entries
1832
:return: Yields (key, contents, digest) in the order
1833
read, not the order requested
1838
# XXX: This smells wrong, IO may not be getting ordered right.
1839
needed_records = sorted(set(records), key=operator.itemgetter(1))
1840
if not needed_records:
1843
# The transport optimizes the fetching as well
1844
# (ie, reads continuous ranges.)
1845
raw_data = self._access.get_raw_records(
1846
[index_memo for key, index_memo in needed_records])
1848
for (key, index_memo), data in \
1849
izip(iter(needed_records), raw_data):
1850
content, digest = self._parse_record(key[-1], data)
1851
yield key, content, digest
1853
def _read_records_iter_raw(self, records):
1854
"""Read text records from data file and yield raw data.
1856
This unpacks enough of the text record to validate the id is
1857
as expected but thats all.
1859
Each item the iterator yields is (key, bytes,
1860
expected_sha1_of_full_text).
1862
for key, data in self._read_records_iter_unchecked(records):
1863
# validate the header (note that we can only use the suffix in
1864
# current knit records).
1865
df, rec = self._parse_record_header(key, data)
1867
yield key, data, rec[3]
1869
def _read_records_iter_unchecked(self, records):
1870
"""Read text records from data file and yield raw data.
1872
No validation is done.
1874
Yields tuples of (key, data).
1876
# setup an iterator of the external records:
1877
# uses readv so nice and fast we hope.
1879
# grab the disk data needed.
1880
needed_offsets = [index_memo for key, index_memo
1882
raw_records = self._access.get_raw_records(needed_offsets)
1884
for key, index_memo in records:
1885
data = raw_records.next()
1888
def _record_to_data(self, key, digest, lines, dense_lines=None):
1889
"""Convert key, digest, lines into a raw data block.
1891
:param key: The key of the record. Currently keys are always serialised
1892
using just the trailing component.
1893
:param dense_lines: The bytes of lines but in a denser form. For
1894
instance, if lines is a list of 1000 bytestrings each ending in \n,
1895
dense_lines may be a list with one line in it, containing all the
1896
1000's lines and their \n's. Using dense_lines if it is already
1897
known is a win because the string join to create bytes in this
1898
function spends less time resizing the final string.
1899
:return: (len, a StringIO instance with the raw data ready to read.)
1901
# Note: using a string copy here increases memory pressure with e.g.
1902
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
1903
# when doing the initial commit of a mozilla tree. RBC 20070921
1904
bytes = ''.join(chain(
1905
["version %s %d %s\n" % (key[-1],
1908
dense_lines or lines,
1909
["end %s\n" % key[-1]]))
1910
if type(bytes) != str:
1911
raise AssertionError(
1912
'data must be plain bytes was %s' % type(bytes))
1913
if lines and lines[-1][-1] != '\n':
1914
raise ValueError('corrupt lines value %r' % lines)
1915
compressed_bytes = tuned_gzip.bytes_to_gzip(bytes)
1916
return len(compressed_bytes), compressed_bytes
1918
def _split_header(self, line):
1921
raise KnitCorrupt(self,
1922
'unexpected number of elements in record header')
1926
"""See VersionedFiles.keys."""
1927
if 'evil' in debug.debug_flags:
1928
trace.mutter_callsite(2, "keys scales with size of history")
1929
sources = [self._index] + self._fallback_vfs
1931
for source in sources:
1932
result.update(source.keys())
1936
class _ContentMapGenerator(object):
1937
"""Generate texts or expose raw deltas for a set of texts."""
1939
def _get_content(self, key):
1940
"""Get the content object for key."""
1941
# Note that _get_content is only called when the _ContentMapGenerator
1942
# has been constructed with just one key requested for reconstruction.
1943
if key in self.nonlocal_keys:
1944
record = self.get_record_stream().next()
1945
# Create a content object on the fly
1946
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
1947
return PlainKnitContent(lines, record.key)
1949
# local keys we can ask for directly
1950
return self._get_one_work(key)
1952
def get_record_stream(self):
1953
"""Get a record stream for the keys requested during __init__."""
1954
for record in self._work():
1958
"""Produce maps of text and KnitContents as dicts.
1960
:return: (text_map, content_map) where text_map contains the texts for
1961
the requested versions and content_map contains the KnitContents.
1963
# NB: By definition we never need to read remote sources unless texts
1964
# are requested from them: we don't delta across stores - and we
1965
# explicitly do not want to to prevent data loss situations.
1966
if self.global_map is None:
1967
self.global_map = self.vf.get_parent_map(self.keys)
1968
nonlocal_keys = self.nonlocal_keys
1970
missing_keys = set(nonlocal_keys)
1971
# Read from remote versioned file instances and provide to our caller.
1972
for source in self.vf._fallback_vfs:
1973
if not missing_keys:
1975
# Loop over fallback repositories asking them for texts - ignore
1976
# any missing from a particular fallback.
1977
for record in source.get_record_stream(missing_keys,
1979
if record.storage_kind == 'absent':
1980
# Not in thie particular stream, may be in one of the
1981
# other fallback vfs objects.
1983
missing_keys.remove(record.key)
1986
self._raw_record_map = self.vf._get_record_map_unparsed(self.keys,
1989
for key in self.keys:
1990
if key in self.nonlocal_keys:
1992
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
1995
def _get_one_work(self, requested_key):
1996
# Now, if we have calculated everything already, just return the
1998
if requested_key in self._contents_map:
1999
return self._contents_map[requested_key]
2000
# To simplify things, parse everything at once - code that wants one text
2001
# probably wants them all.
2002
# FUTURE: This function could be improved for the 'extract many' case
2003
# by tracking each component and only doing the copy when the number of
2004
# children than need to apply delta's to it is > 1 or it is part of the
2006
multiple_versions = len(self.keys) != 1
2007
if self._record_map is None:
2008
self._record_map = self.vf._raw_map_to_record_map(
2009
self._raw_record_map)
2010
record_map = self._record_map
2011
# raw_record_map is key:
2012
# Have read and parsed records at this point.
2013
for key in self.keys:
2014
if key in self.nonlocal_keys:
2019
while cursor is not None:
2021
record, record_details, digest, next = record_map[cursor]
2023
raise RevisionNotPresent(cursor, self)
2024
components.append((cursor, record, record_details, digest))
2026
if cursor in self._contents_map:
2027
# no need to plan further back
2028
components.append((cursor, None, None, None))
2032
for (component_id, record, record_details,
2033
digest) in reversed(components):
2034
if component_id in self._contents_map:
2035
content = self._contents_map[component_id]
2037
content, delta = self._factory.parse_record(key[-1],
2038
record, record_details, content,
2039
copy_base_content=multiple_versions)
2040
if multiple_versions:
2041
self._contents_map[component_id] = content
2043
# digest here is the digest from the last applied component.
2044
text = content.text()
2045
actual_sha = sha_strings(text)
2046
if actual_sha != digest:
2047
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2048
if multiple_versions:
2049
return self._contents_map[requested_key]
2053
def _wire_bytes(self):
2054
"""Get the bytes to put on the wire for 'key'.
2056
The first collection of bytes asked for returns the serialised
2057
raw_record_map and the additional details (key, parent) for key.
2058
Subsequent calls return just the additional details (key, parent).
2059
The wire storage_kind given for the first key is 'knit-delta-closure',
2060
For subsequent keys it is 'knit-delta-closure-ref'.
2062
:param key: A key from the content generator.
2063
:return: Bytes to put on the wire.
2066
# kind marker for dispatch on the far side,
2067
lines.append('knit-delta-closure')
2069
if self.vf._factory.annotated:
2070
lines.append('annotated')
2073
# then the list of keys
2074
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2075
if key not in self.nonlocal_keys]))
2076
# then the _raw_record_map in serialised form:
2078
# for each item in the map:
2080
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2081
# one line with method
2082
# one line with noeol
2083
# one line with next ('' for None)
2084
# one line with byte count of the record bytes
2086
for key, (record_bytes, (method, noeol), next) in \
2087
self._raw_record_map.iteritems():
2088
key_bytes = '\x00'.join(key)
2089
parents = self.global_map.get(key, None)
2091
parent_bytes = 'None:'
2093
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2094
method_bytes = method
2100
next_bytes = '\x00'.join(next)
2103
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2104
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2105
len(record_bytes), record_bytes))
2106
map_bytes = ''.join(map_byte_list)
2107
lines.append(map_bytes)
2108
bytes = '\n'.join(lines)
2112
class _VFContentMapGenerator(_ContentMapGenerator):
2113
"""Content map generator reading from a VersionedFiles object."""
2115
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2116
global_map=None, raw_record_map=None):
2117
"""Create a _ContentMapGenerator.
2119
:param versioned_files: The versioned files that the texts are being
2121
:param keys: The keys to produce content maps for.
2122
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2123
which are known to not be in this knit, but rather in one of the
2125
:param global_map: The result of get_parent_map(keys) (or a supermap).
2126
This is required if get_record_stream() is to be used.
2127
:param raw_record_map: A unparsed raw record map to use for answering
2130
# The vf to source data from
2131
self.vf = versioned_files
2133
self.keys = list(keys)
2134
# Keys known to be in fallback vfs objects
2135
if nonlocal_keys is None:
2136
self.nonlocal_keys = set()
2138
self.nonlocal_keys = frozenset(nonlocal_keys)
2139
# Parents data for keys to be returned in get_record_stream
2140
self.global_map = global_map
2141
# The chunked lists for self.keys in text form
2143
# A cache of KnitContent objects used in extracting texts.
2144
self._contents_map = {}
2145
# All the knit records needed to assemble the requested keys as full
2147
self._record_map = None
2148
if raw_record_map is None:
2149
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2152
self._raw_record_map = raw_record_map
2153
# the factory for parsing records
2154
self._factory = self.vf._factory
2157
class _NetworkContentMapGenerator(_ContentMapGenerator):
2158
"""Content map generator sourced from a network stream."""
2160
def __init__(self, bytes, line_end):
2161
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2163
self.global_map = {}
2164
self._raw_record_map = {}
2165
self._contents_map = {}
2166
self._record_map = None
2167
self.nonlocal_keys = []
2168
# Get access to record parsing facilities
2169
self.vf = KnitVersionedFiles(None, None)
2172
line_end = bytes.find('\n', start)
2173
line = bytes[start:line_end]
2174
start = line_end + 1
2175
if line == 'annotated':
2176
self._factory = KnitAnnotateFactory()
2178
self._factory = KnitPlainFactory()
2179
# list of keys to emit in get_record_stream
2180
line_end = bytes.find('\n', start)
2181
line = bytes[start:line_end]
2182
start = line_end + 1
2184
tuple(segment.split('\x00')) for segment in line.split('\t')
2186
# now a loop until the end. XXX: It would be nice if this was just a
2187
# bunch of the same records as get_record_stream(..., False) gives, but
2188
# there is a decent sized gap stopping that at the moment.
2192
line_end = bytes.find('\n', start)
2193
key = tuple(bytes[start:line_end].split('\x00'))
2194
start = line_end + 1
2195
# 1 line with parents (None: for None, '' for ())
2196
line_end = bytes.find('\n', start)
2197
line = bytes[start:line_end]
2202
[tuple(segment.split('\x00')) for segment in line.split('\t')
2204
self.global_map[key] = parents
2205
start = line_end + 1
2206
# one line with method
2207
line_end = bytes.find('\n', start)
2208
line = bytes[start:line_end]
2210
start = line_end + 1
2211
# one line with noeol
2212
line_end = bytes.find('\n', start)
2213
line = bytes[start:line_end]
2215
start = line_end + 1
2216
# one line with next ('' for None)
2217
line_end = bytes.find('\n', start)
2218
line = bytes[start:line_end]
2222
next = tuple(bytes[start:line_end].split('\x00'))
2223
start = line_end + 1
2224
# one line with byte count of the record bytes
2225
line_end = bytes.find('\n', start)
2226
line = bytes[start:line_end]
2228
start = line_end + 1
2230
record_bytes = bytes[start:start+count]
2231
start = start + count
2233
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2235
def get_record_stream(self):
2236
"""Get a record stream for for keys requested by the bytestream."""
2238
for key in self.keys:
2239
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2242
def _wire_bytes(self):
2246
class _KndxIndex(object):
2247
"""Manages knit index files
2249
The index is kept in memory and read on startup, to enable
2250
fast lookups of revision information. The cursor of the index
2251
file is always pointing to the end, making it easy to append
2254
_cache is a cache for fast mapping from version id to a Index
2257
_history is a cache for fast mapping from indexes to version ids.
2259
The index data format is dictionary compressed when it comes to
2260
parent references; a index entry may only have parents that with a
2261
lover index number. As a result, the index is topological sorted.
2263
Duplicate entries may be written to the index for a single version id
2264
if this is done then the latter one completely replaces the former:
2265
this allows updates to correct version and parent information.
2266
Note that the two entries may share the delta, and that successive
2267
annotations and references MUST point to the first entry.
2269
The index file on disc contains a header, followed by one line per knit
2270
record. The same revision can be present in an index file more than once.
2271
The first occurrence gets assigned a sequence number starting from 0.
2273
The format of a single line is
2274
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
2275
REVISION_ID is a utf8-encoded revision id
2276
FLAGS is a comma separated list of flags about the record. Values include
2277
no-eol, line-delta, fulltext.
2278
BYTE_OFFSET is the ascii representation of the byte offset in the data file
2279
that the the compressed data starts at.
2280
LENGTH is the ascii representation of the length of the data file.
2281
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
2283
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
2284
revision id already in the knit that is a parent of REVISION_ID.
2285
The ' :' marker is the end of record marker.
2288
when a write is interrupted to the index file, it will result in a line
2289
that does not end in ' :'. If the ' :' is not present at the end of a line,
2290
or at the end of the file, then the record that is missing it will be
2291
ignored by the parser.
2293
When writing new records to the index file, the data is preceded by '\n'
2294
to ensure that records always start on new lines even if the last write was
2295
interrupted. As a result its normal for the last line in the index to be
2296
missing a trailing newline. One can be added with no harmful effects.
2298
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2299
where prefix is e.g. the (fileid,) for .texts instances or () for
2300
constant-mapped things like .revisions, and the old state is
2301
tuple(history_vector, cache_dict). This is used to prevent having an
2302
ABI change with the C extension that reads .kndx files.
2305
HEADER = "# bzr knit index 8\n"
2307
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2308
"""Create a _KndxIndex on transport using mapper."""
2309
self._transport = transport
2310
self._mapper = mapper
2311
self._get_scope = get_scope
2312
self._allow_writes = allow_writes
2313
self._is_locked = is_locked
2315
self.has_graph = True
2317
def add_records(self, records, random_id=False, missing_compression_parents=False):
2318
"""Add multiple records to the index.
2320
:param records: a list of tuples:
2321
(key, options, access_memo, parents).
2322
:param random_id: If True the ids being added were randomly generated
2323
and no check for existence will be performed.
2324
:param missing_compression_parents: If True the records being added are
2325
only compressed against texts already in the index (or inside
2326
records). If False the records all refer to unavailable texts (or
2327
texts inside records) as compression parents.
2329
if missing_compression_parents:
2330
# It might be nice to get the edge of the records. But keys isn't
2332
keys = sorted(record[0] for record in records)
2333
raise errors.RevisionNotPresent(keys, self)
2335
for record in records:
2338
path = self._mapper.map(key) + '.kndx'
2339
path_keys = paths.setdefault(path, (prefix, []))
2340
path_keys[1].append(record)
2341
for path in sorted(paths):
2342
prefix, path_keys = paths[path]
2343
self._load_prefixes([prefix])
2345
orig_history = self._kndx_cache[prefix][1][:]
2346
orig_cache = self._kndx_cache[prefix][0].copy()
2349
for key, options, (_, pos, size), parents in path_keys:
2351
# kndx indices cannot be parentless.
2353
line = "\n%s %s %s %s %s :" % (
2354
key[-1], ','.join(options), pos, size,
2355
self._dictionary_compress(parents))
2356
if type(line) != str:
2357
raise AssertionError(
2358
'data must be utf8 was %s' % type(line))
2360
self._cache_key(key, options, pos, size, parents)
2361
if len(orig_history):
2362
self._transport.append_bytes(path, ''.join(lines))
2364
self._init_index(path, lines)
2366
# If any problems happen, restore the original values and re-raise
2367
self._kndx_cache[prefix] = (orig_cache, orig_history)
2370
def scan_unvalidated_index(self, graph_index):
2371
"""See _KnitGraphIndex.scan_unvalidated_index."""
2372
# Because kndx files do not support atomic insertion via separate index
2373
# files, they do not support this method.
2374
raise NotImplementedError(self.scan_unvalidated_index)
2376
def get_missing_compression_parents(self):
2377
"""See _KnitGraphIndex.get_missing_compression_parents."""
2378
# Because kndx files do not support atomic insertion via separate index
2379
# files, they do not support this method.
2380
raise NotImplementedError(self.get_missing_compression_parents)
2382
def _cache_key(self, key, options, pos, size, parent_keys):
2383
"""Cache a version record in the history array and index cache.
2385
This is inlined into _load_data for performance. KEEP IN SYNC.
2386
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2390
version_id = key[-1]
2391
# last-element only for compatibilty with the C load_data.
2392
parents = tuple(parent[-1] for parent in parent_keys)
2393
for parent in parent_keys:
2394
if parent[:-1] != prefix:
2395
raise ValueError("mismatched prefixes for %r, %r" % (
2397
cache, history = self._kndx_cache[prefix]
2398
# only want the _history index to reference the 1st index entry
2400
if version_id not in cache:
2401
index = len(history)
2402
history.append(version_id)
2404
index = cache[version_id][5]
2405
cache[version_id] = (version_id,
2412
def check_header(self, fp):
2413
line = fp.readline()
2415
# An empty file can actually be treated as though the file doesn't
2417
raise errors.NoSuchFile(self)
2418
if line != self.HEADER:
2419
raise KnitHeaderError(badline=line, filename=self)
2421
def _check_read(self):
2422
if not self._is_locked():
2423
raise errors.ObjectNotLocked(self)
2424
if self._get_scope() != self._scope:
2427
def _check_write_ok(self):
2428
"""Assert if not writes are permitted."""
2429
if not self._is_locked():
2430
raise errors.ObjectNotLocked(self)
2431
if self._get_scope() != self._scope:
2433
if self._mode != 'w':
2434
raise errors.ReadOnlyObjectDirtiedError(self)
2436
def get_build_details(self, keys):
2437
"""Get the method, index_memo and compression parent for keys.
2439
Ghosts are omitted from the result.
2441
:param keys: An iterable of keys.
2442
:return: A dict of key:(index_memo, compression_parent, parents,
2445
opaque structure to pass to read_records to extract the raw
2448
Content that this record is built upon, may be None
2450
Logical parents of this node
2452
extra information about the content which needs to be passed to
2453
Factory.parse_record
2455
parent_map = self.get_parent_map(keys)
2458
if key not in parent_map:
2460
method = self.get_method(key)
2461
parents = parent_map[key]
2462
if method == 'fulltext':
2463
compression_parent = None
2465
compression_parent = parents[0]
2466
noeol = 'no-eol' in self.get_options(key)
2467
index_memo = self.get_position(key)
2468
result[key] = (index_memo, compression_parent,
2469
parents, (method, noeol))
2472
def get_method(self, key):
2473
"""Return compression method of specified key."""
2474
options = self.get_options(key)
2475
if 'fulltext' in options:
2477
elif 'line-delta' in options:
2480
raise errors.KnitIndexUnknownMethod(self, options)
2482
def get_options(self, key):
2483
"""Return a list representing options.
2487
prefix, suffix = self._split_key(key)
2488
self._load_prefixes([prefix])
2490
return self._kndx_cache[prefix][0][suffix][1]
2492
raise RevisionNotPresent(key, self)
2494
def get_parent_map(self, keys):
2495
"""Get a map of the parents of keys.
2497
:param keys: The keys to look up parents for.
2498
:return: A mapping from keys to parents. Absent keys are absent from
2501
# Parse what we need to up front, this potentially trades off I/O
2502
# locality (.kndx and .knit in the same block group for the same file
2503
# id) for less checking in inner loops.
2504
prefixes = set(key[:-1] for key in keys)
2505
self._load_prefixes(prefixes)
2510
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2514
result[key] = tuple(prefix + (suffix,) for
2515
suffix in suffix_parents)
2518
def get_position(self, key):
2519
"""Return details needed to access the version.
2521
:return: a tuple (key, data position, size) to hand to the access
2522
logic to get the record.
2524
prefix, suffix = self._split_key(key)
2525
self._load_prefixes([prefix])
2526
entry = self._kndx_cache[prefix][0][suffix]
2527
return key, entry[2], entry[3]
2529
has_key = _mod_index._has_key_from_parent_map
2531
def _init_index(self, path, extra_lines=[]):
2532
"""Initialize an index."""
2534
sio.write(self.HEADER)
2535
sio.writelines(extra_lines)
2537
self._transport.put_file_non_atomic(path, sio,
2538
create_parent_dir=True)
2539
# self._create_parent_dir)
2540
# mode=self._file_mode,
2541
# dir_mode=self._dir_mode)
2544
"""Get all the keys in the collection.
2546
The keys are not ordered.
2549
# Identify all key prefixes.
2550
# XXX: A bit hacky, needs polish.
2551
if type(self._mapper) == ConstantMapper:
2555
for quoted_relpath in self._transport.iter_files_recursive():
2556
path, ext = os.path.splitext(quoted_relpath)
2558
prefixes = [self._mapper.unmap(path) for path in relpaths]
2559
self._load_prefixes(prefixes)
2560
for prefix in prefixes:
2561
for suffix in self._kndx_cache[prefix][1]:
2562
result.add(prefix + (suffix,))
2565
def _load_prefixes(self, prefixes):
2566
"""Load the indices for prefixes."""
2568
for prefix in prefixes:
2569
if prefix not in self._kndx_cache:
2570
# the load_data interface writes to these variables.
2573
self._filename = prefix
2575
path = self._mapper.map(prefix) + '.kndx'
2576
fp = self._transport.get(path)
2578
# _load_data may raise NoSuchFile if the target knit is
2580
_load_data(self, fp)
2583
self._kndx_cache[prefix] = (self._cache, self._history)
2588
self._kndx_cache[prefix] = ({}, [])
2589
if type(self._mapper) == ConstantMapper:
2590
# preserve behaviour for revisions.kndx etc.
2591
self._init_index(path)
2596
missing_keys = _mod_index._missing_keys_from_parent_map
2598
def _partition_keys(self, keys):
2599
"""Turn keys into a dict of prefix:suffix_list."""
2602
prefix_keys = result.setdefault(key[:-1], [])
2603
prefix_keys.append(key[-1])
2606
def _dictionary_compress(self, keys):
2607
"""Dictionary compress keys.
2609
:param keys: The keys to generate references to.
2610
:return: A string representation of keys. keys which are present are
2611
dictionary compressed, and others are emitted as fulltext with a
2617
prefix = keys[0][:-1]
2618
cache = self._kndx_cache[prefix][0]
2620
if key[:-1] != prefix:
2621
# kndx indices cannot refer across partitioned storage.
2622
raise ValueError("mismatched prefixes for %r" % keys)
2623
if key[-1] in cache:
2624
# -- inlined lookup() --
2625
result_list.append(str(cache[key[-1]][5]))
2626
# -- end lookup () --
2628
result_list.append('.' + key[-1])
2629
return ' '.join(result_list)
2631
def _reset_cache(self):
2632
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2633
# (cache_dict, history_vector) for parsed kndx files.
2634
self._kndx_cache = {}
2635
self._scope = self._get_scope()
2636
allow_writes = self._allow_writes()
2642
def _sort_keys_by_io(self, keys, positions):
2643
"""Figure out an optimal order to read the records for the given keys.
2645
Sort keys, grouped by index and sorted by position.
2647
:param keys: A list of keys whose records we want to read. This will be
2649
:param positions: A dict, such as the one returned by
2650
_get_components_positions()
2653
def get_sort_key(key):
2654
index_memo = positions[key][1]
2655
# Group by prefix and position. index_memo[0] is the key, so it is
2656
# (file_id, revision_id) and we don't want to sort on revision_id,
2657
# index_memo[1] is the position, and index_memo[2] is the size,
2658
# which doesn't matter for the sort
2659
return index_memo[0][:-1], index_memo[1]
2660
return keys.sort(key=get_sort_key)
2662
_get_total_build_size = _get_total_build_size
2664
def _split_key(self, key):
2665
"""Split key into a prefix and suffix."""
2666
return key[:-1], key[-1]
2669
class _KnitGraphIndex(object):
2670
"""A KnitVersionedFiles index layered on GraphIndex."""
2672
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2674
"""Construct a KnitGraphIndex on a graph_index.
2676
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2677
:param is_locked: A callback to check whether the object should answer
2679
:param deltas: Allow delta-compressed records.
2680
:param parents: If True, record knits parents, if not do not record
2682
:param add_callback: If not None, allow additions to the index and call
2683
this callback with a list of added GraphIndex nodes:
2684
[(node, value, node_refs), ...]
2685
:param is_locked: A callback, returns True if the index is locked and
2688
self._add_callback = add_callback
2689
self._graph_index = graph_index
2690
self._deltas = deltas
2691
self._parents = parents
2692
if deltas and not parents:
2693
# XXX: TODO: Delta tree and parent graph should be conceptually
2695
raise KnitCorrupt(self, "Cannot do delta compression without "
2697
self.has_graph = parents
2698
self._is_locked = is_locked
2699
self._missing_compression_parents = set()
2702
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2704
def add_records(self, records, random_id=False,
2705
missing_compression_parents=False):
2706
"""Add multiple records to the index.
2708
This function does not insert data into the Immutable GraphIndex
2709
backing the KnitGraphIndex, instead it prepares data for insertion by
2710
the caller and checks that it is safe to insert then calls
2711
self._add_callback with the prepared GraphIndex nodes.
2713
:param records: a list of tuples:
2714
(key, options, access_memo, parents).
2715
:param random_id: If True the ids being added were randomly generated
2716
and no check for existence will be performed.
2717
:param missing_compression_parents: If True the records being added are
2718
only compressed against texts already in the index (or inside
2719
records). If False the records all refer to unavailable texts (or
2720
texts inside records) as compression parents.
2722
if not self._add_callback:
2723
raise errors.ReadOnlyError(self)
2724
# we hope there are no repositories with inconsistent parentage
2728
compression_parents = set()
2729
for (key, options, access_memo, parents) in records:
2731
parents = tuple(parents)
2732
index, pos, size = access_memo
2733
if 'no-eol' in options:
2737
value += "%d %d" % (pos, size)
2738
if not self._deltas:
2739
if 'line-delta' in options:
2740
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2743
if 'line-delta' in options:
2744
node_refs = (parents, (parents[0],))
2745
if missing_compression_parents:
2746
compression_parents.add(parents[0])
2748
node_refs = (parents, ())
2750
node_refs = (parents, )
2753
raise KnitCorrupt(self, "attempt to add node with parents "
2754
"in parentless index.")
2756
keys[key] = (value, node_refs)
2759
present_nodes = self._get_entries(keys)
2760
for (index, key, value, node_refs) in present_nodes:
2761
if (value[0] != keys[key][0][0] or
2762
node_refs[:1] != keys[key][1][:1]):
2763
raise KnitCorrupt(self, "inconsistent details in add_records"
2764
": %s %s" % ((value, node_refs), keys[key]))
2768
for key, (value, node_refs) in keys.iteritems():
2769
result.append((key, value, node_refs))
2771
for key, (value, node_refs) in keys.iteritems():
2772
result.append((key, value))
2773
self._add_callback(result)
2774
if missing_compression_parents:
2775
# This may appear to be incorrect (it does not check for
2776
# compression parents that are in the existing graph index),
2777
# but such records won't have been buffered, so this is
2778
# actually correct: every entry when
2779
# missing_compression_parents==True either has a missing parent, or
2780
# a parent that is one of the keys in records.
2781
compression_parents.difference_update(keys)
2782
self._missing_compression_parents.update(compression_parents)
2783
# Adding records may have satisfied missing compression parents.
2784
self._missing_compression_parents.difference_update(keys)
2786
def scan_unvalidated_index(self, graph_index):
2787
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2789
This allows this _KnitGraphIndex to keep track of any missing
2790
compression parents we may want to have filled in to make those
2793
:param graph_index: A GraphIndex
2796
new_missing = graph_index.external_references(ref_list_num=1)
2797
new_missing.difference_update(self.get_parent_map(new_missing))
2798
self._missing_compression_parents.update(new_missing)
2800
def get_missing_compression_parents(self):
2801
"""Return the keys of missing compression parents.
2803
Missing compression parents occur when a record stream was missing
2804
basis texts, or a index was scanned that had missing basis texts.
2806
return frozenset(self._missing_compression_parents)
2808
def _check_read(self):
2809
"""raise if reads are not permitted."""
2810
if not self._is_locked():
2811
raise errors.ObjectNotLocked(self)
2813
def _check_write_ok(self):
2814
"""Assert if writes are not permitted."""
2815
if not self._is_locked():
2816
raise errors.ObjectNotLocked(self)
2818
def _compression_parent(self, an_entry):
2819
# return the key that an_entry is compressed against, or None
2820
# Grab the second parent list (as deltas implies parents currently)
2821
compression_parents = an_entry[3][1]
2822
if not compression_parents:
2824
if len(compression_parents) != 1:
2825
raise AssertionError(
2826
"Too many compression parents: %r" % compression_parents)
2827
return compression_parents[0]
2829
def get_build_details(self, keys):
2830
"""Get the method, index_memo and compression parent for version_ids.
2832
Ghosts are omitted from the result.
2834
:param keys: An iterable of keys.
2835
:return: A dict of key:
2836
(index_memo, compression_parent, parents, record_details).
2838
opaque structure to pass to read_records to extract the raw
2841
Content that this record is built upon, may be None
2843
Logical parents of this node
2845
extra information about the content which needs to be passed to
2846
Factory.parse_record
2850
entries = self._get_entries(keys, False)
2851
for entry in entries:
2853
if not self._parents:
2856
parents = entry[3][0]
2857
if not self._deltas:
2858
compression_parent_key = None
2860
compression_parent_key = self._compression_parent(entry)
2861
noeol = (entry[2][0] == 'N')
2862
if compression_parent_key:
2863
method = 'line-delta'
2866
result[key] = (self._node_to_position(entry),
2867
compression_parent_key, parents,
2871
def _get_entries(self, keys, check_present=False):
2872
"""Get the entries for keys.
2874
:param keys: An iterable of index key tuples.
2879
for node in self._graph_index.iter_entries(keys):
2881
found_keys.add(node[1])
2883
# adapt parentless index to the rest of the code.
2884
for node in self._graph_index.iter_entries(keys):
2885
yield node[0], node[1], node[2], ()
2886
found_keys.add(node[1])
2888
missing_keys = keys.difference(found_keys)
2890
raise RevisionNotPresent(missing_keys.pop(), self)
2892
def get_method(self, key):
2893
"""Return compression method of specified key."""
2894
return self._get_method(self._get_node(key))
2896
def _get_method(self, node):
2897
if not self._deltas:
2899
if self._compression_parent(node):
2904
def _get_node(self, key):
2906
return list(self._get_entries([key]))[0]
2908
raise RevisionNotPresent(key, self)
2910
def get_options(self, key):
2911
"""Return a list representing options.
2915
node = self._get_node(key)
2916
options = [self._get_method(node)]
2917
if node[2][0] == 'N':
2918
options.append('no-eol')
2921
def get_parent_map(self, keys):
2922
"""Get a map of the parents of keys.
2924
:param keys: The keys to look up parents for.
2925
:return: A mapping from keys to parents. Absent keys are absent from
2929
nodes = self._get_entries(keys)
2933
result[node[1]] = node[3][0]
2936
result[node[1]] = None
2939
def get_position(self, key):
2940
"""Return details needed to access the version.
2942
:return: a tuple (index, data position, size) to hand to the access
2943
logic to get the record.
2945
node = self._get_node(key)
2946
return self._node_to_position(node)
2948
has_key = _mod_index._has_key_from_parent_map
2951
"""Get all the keys in the collection.
2953
The keys are not ordered.
2956
return [node[1] for node in self._graph_index.iter_all_entries()]
2958
missing_keys = _mod_index._missing_keys_from_parent_map
2960
def _node_to_position(self, node):
2961
"""Convert an index value to position details."""
2962
bits = node[2][1:].split(' ')
2963
return node[0], int(bits[0]), int(bits[1])
2965
def _sort_keys_by_io(self, keys, positions):
2966
"""Figure out an optimal order to read the records for the given keys.
2968
Sort keys, grouped by index and sorted by position.
2970
:param keys: A list of keys whose records we want to read. This will be
2972
:param positions: A dict, such as the one returned by
2973
_get_components_positions()
2976
def get_index_memo(key):
2977
# index_memo is at offset [1]. It is made up of (GraphIndex,
2978
# position, size). GI is an object, which will be unique for each
2979
# pack file. This causes us to group by pack file, then sort by
2980
# position. Size doesn't matter, but it isn't worth breaking up the
2982
return positions[key][1]
2983
return keys.sort(key=get_index_memo)
2985
_get_total_build_size = _get_total_build_size
2988
class _KnitKeyAccess(object):
2989
"""Access to records in .knit files."""
2991
def __init__(self, transport, mapper):
2992
"""Create a _KnitKeyAccess with transport and mapper.
2994
:param transport: The transport the access object is rooted at.
2995
:param mapper: The mapper used to map keys to .knit files.
2997
self._transport = transport
2998
self._mapper = mapper
3000
def add_raw_records(self, key_sizes, raw_data):
3001
"""Add raw knit bytes to a storage area.
3003
The data is spooled to the container writer in one bytes-record per
3006
:param sizes: An iterable of tuples containing the key and size of each
3008
:param raw_data: A bytestring containing the data.
3009
:return: A list of memos to retrieve the record later. Each memo is an
3010
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3011
length), where the key is the record key.
3013
if type(raw_data) != str:
3014
raise AssertionError(
3015
'data must be plain bytes was %s' % type(raw_data))
3018
# TODO: This can be tuned for writing to sftp and other servers where
3019
# append() is relatively expensive by grouping the writes to each key
3021
for key, size in key_sizes:
3022
path = self._mapper.map(key)
3024
base = self._transport.append_bytes(path + '.knit',
3025
raw_data[offset:offset+size])
3026
except errors.NoSuchFile:
3027
self._transport.mkdir(osutils.dirname(path))
3028
base = self._transport.append_bytes(path + '.knit',
3029
raw_data[offset:offset+size])
3033
result.append((key, base, size))
3036
def get_raw_records(self, memos_for_retrieval):
3037
"""Get the raw bytes for a records.
3039
:param memos_for_retrieval: An iterable containing the access memo for
3040
retrieving the bytes.
3041
:return: An iterator over the bytes of the records.
3043
# first pass, group into same-index request to minimise readv's issued.
3045
current_prefix = None
3046
for (key, offset, length) in memos_for_retrieval:
3047
if current_prefix == key[:-1]:
3048
current_list.append((offset, length))
3050
if current_prefix is not None:
3051
request_lists.append((current_prefix, current_list))
3052
current_prefix = key[:-1]
3053
current_list = [(offset, length)]
3054
# handle the last entry
3055
if current_prefix is not None:
3056
request_lists.append((current_prefix, current_list))
3057
for prefix, read_vector in request_lists:
3058
path = self._mapper.map(prefix) + '.knit'
3059
for pos, data in self._transport.readv(path, read_vector):
3063
class _DirectPackAccess(object):
3064
"""Access to data in one or more packs with less translation."""
3066
def __init__(self, index_to_packs, reload_func=None):
3067
"""Create a _DirectPackAccess object.
3069
:param index_to_packs: A dict mapping index objects to the transport
3070
and file names for obtaining data.
3071
:param reload_func: A function to call if we determine that the pack
3072
files have moved and we need to reload our caches. See
3073
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
3075
self._container_writer = None
3076
self._write_index = None
3077
self._indices = index_to_packs
3078
self._reload_func = reload_func
3080
def add_raw_records(self, key_sizes, raw_data):
3081
"""Add raw knit bytes to a storage area.
3083
The data is spooled to the container writer in one bytes-record per
3086
:param sizes: An iterable of tuples containing the key and size of each
3088
:param raw_data: A bytestring containing the data.
3089
:return: A list of memos to retrieve the record later. Each memo is an
3090
opaque index memo. For _DirectPackAccess the memo is (index, pos,
3091
length), where the index field is the write_index object supplied
3092
to the PackAccess object.
3094
if type(raw_data) != str:
3095
raise AssertionError(
3096
'data must be plain bytes was %s' % type(raw_data))
3099
for key, size in key_sizes:
3100
p_offset, p_length = self._container_writer.add_bytes_record(
3101
raw_data[offset:offset+size], [])
3103
result.append((self._write_index, p_offset, p_length))
3106
def get_raw_records(self, memos_for_retrieval):
3107
"""Get the raw bytes for a records.
3109
:param memos_for_retrieval: An iterable containing the (index, pos,
3110
length) memo for retrieving the bytes. The Pack access method
3111
looks up the pack to use for a given record in its index_to_pack
3113
:return: An iterator over the bytes of the records.
3115
# first pass, group into same-index requests
3117
current_index = None
3118
for (index, offset, length) in memos_for_retrieval:
3119
if current_index == index:
3120
current_list.append((offset, length))
3122
if current_index is not None:
3123
request_lists.append((current_index, current_list))
3124
current_index = index
3125
current_list = [(offset, length)]
3126
# handle the last entry
3127
if current_index is not None:
3128
request_lists.append((current_index, current_list))
3129
for index, offsets in request_lists:
3131
transport, path = self._indices[index]
3133
# A KeyError here indicates that someone has triggered an index
3134
# reload, and this index has gone missing, we need to start
3136
if self._reload_func is None:
3137
# If we don't have a _reload_func there is nothing that can
3140
raise errors.RetryWithNewPacks(index,
3141
reload_occurred=True,
3142
exc_info=sys.exc_info())
3144
reader = pack.make_readv_reader(transport, path, offsets)
3145
for names, read_func in reader.iter_records():
3146
yield read_func(None)
3147
except errors.NoSuchFile:
3148
# A NoSuchFile error indicates that a pack file has gone
3149
# missing on disk, we need to trigger a reload, and start over.
3150
if self._reload_func is None:
3152
raise errors.RetryWithNewPacks(transport.abspath(path),
3153
reload_occurred=False,
3154
exc_info=sys.exc_info())
3156
def set_writer(self, writer, index, transport_packname):
3157
"""Set a writer to use for adding data."""
3158
if index is not None:
3159
self._indices[index] = transport_packname
3160
self._container_writer = writer
3161
self._write_index = index
3163
def reload_or_raise(self, retry_exc):
3164
"""Try calling the reload function, or re-raise the original exception.
3166
This should be called after _DirectPackAccess raises a
3167
RetryWithNewPacks exception. This function will handle the common logic
3168
of determining when the error is fatal versus being temporary.
3169
It will also make sure that the original exception is raised, rather
3170
than the RetryWithNewPacks exception.
3172
If this function returns, then the calling function should retry
3173
whatever operation was being performed. Otherwise an exception will
3176
:param retry_exc: A RetryWithNewPacks exception.
3179
if self._reload_func is None:
3181
elif not self._reload_func():
3182
# The reload claimed that nothing changed
3183
if not retry_exc.reload_occurred:
3184
# If there wasn't an earlier reload, then we really were
3185
# expecting to find changes. We didn't find them, so this is a
3189
exc_class, exc_value, exc_traceback = retry_exc.exc_info
3190
raise exc_class, exc_value, exc_traceback
3193
# Deprecated, use PatienceSequenceMatcher instead
3194
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
3197
def annotate_knit(knit, revision_id):
3198
"""Annotate a knit with no cached annotations.
3200
This implementation is for knits with no cached annotations.
3201
It will work for knits with cached annotations, but this is not
3204
annotator = _KnitAnnotator(knit)
3205
return iter(annotator.annotate(revision_id))
3208
class _KnitAnnotator(object):
3209
"""Build up the annotations for a text."""
3211
def __init__(self, knit):
3214
# Content objects, differs from fulltexts because of how final newlines
3215
# are treated by knits. the content objects here will always have a
3217
self._fulltext_contents = {}
3219
# Annotated lines of specific revisions
3220
self._annotated_lines = {}
3222
# Track the raw data for nodes that we could not process yet.
3223
# This maps the revision_id of the base to a list of children that will
3224
# annotated from it.
3225
self._pending_children = {}
3227
# Nodes which cannot be extracted
3228
self._ghosts = set()
3230
# Track how many children this node has, so we know if we need to keep
3232
self._annotate_children = {}
3233
self._compression_children = {}
3235
self._all_build_details = {}
3236
# The children => parent revision_id graph
3237
self._revision_id_graph = {}
3239
self._heads_provider = None
3241
self._nodes_to_keep_annotations = set()
3242
self._generations_until_keep = 100
3244
def set_generations_until_keep(self, value):
3245
"""Set the number of generations before caching a node.
3247
Setting this to -1 will cache every merge node, setting this higher
3248
will cache fewer nodes.
3250
self._generations_until_keep = value
3252
def _add_fulltext_content(self, revision_id, content_obj):
3253
self._fulltext_contents[revision_id] = content_obj
3254
# TODO: jam 20080305 It might be good to check the sha1digest here
3255
return content_obj.text()
3257
def _check_parents(self, child, nodes_to_annotate):
3258
"""Check if all parents have been processed.
3260
:param child: A tuple of (rev_id, parents, raw_content)
3261
:param nodes_to_annotate: If child is ready, add it to
3262
nodes_to_annotate, otherwise put it back in self._pending_children
3264
for parent_id in child[1]:
3265
if (parent_id not in self._annotated_lines):
3266
# This parent is present, but another parent is missing
3267
self._pending_children.setdefault(parent_id,
3271
# This one is ready to be processed
3272
nodes_to_annotate.append(child)
3274
def _add_annotation(self, revision_id, fulltext, parent_ids,
3275
left_matching_blocks=None):
3276
"""Add an annotation entry.
3278
All parents should already have been annotated.
3279
:return: A list of children that now have their parents satisfied.
3281
a = self._annotated_lines
3282
annotated_parent_lines = [a[p] for p in parent_ids]
3283
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
3284
fulltext, revision_id, left_matching_blocks,
3285
heads_provider=self._get_heads_provider()))
3286
self._annotated_lines[revision_id] = annotated_lines
3287
for p in parent_ids:
3288
ann_children = self._annotate_children[p]
3289
ann_children.remove(revision_id)
3290
if (not ann_children
3291
and p not in self._nodes_to_keep_annotations):
3292
del self._annotated_lines[p]
3293
del self._all_build_details[p]
3294
if p in self._fulltext_contents:
3295
del self._fulltext_contents[p]
3296
# Now that we've added this one, see if there are any pending
3297
# deltas to be done, certainly this parent is finished
3298
nodes_to_annotate = []
3299
for child in self._pending_children.pop(revision_id, []):
3300
self._check_parents(child, nodes_to_annotate)
3301
return nodes_to_annotate
3303
def _get_build_graph(self, key):
3304
"""Get the graphs for building texts and annotations.
3306
The data you need for creating a full text may be different than the
3307
data you need to annotate that text. (At a minimum, you need both
3308
parents to create an annotation, but only need 1 parent to generate the
3311
:return: A list of (key, index_memo) records, suitable for
3312
passing to read_records_iter to start reading in the raw data fro/
3315
if key in self._annotated_lines:
3318
pending = set([key])
3323
# get all pending nodes
3325
this_iteration = pending
3326
build_details = self._knit._index.get_build_details(this_iteration)
3327
self._all_build_details.update(build_details)
3328
# new_nodes = self._knit._index._get_entries(this_iteration)
3330
for key, details in build_details.iteritems():
3331
(index_memo, compression_parent, parents,
3332
record_details) = details
3333
self._revision_id_graph[key] = parents
3334
records.append((key, index_memo))
3335
# Do we actually need to check _annotated_lines?
3336
pending.update(p for p in parents
3337
if p not in self._all_build_details)
3338
if compression_parent:
3339
self._compression_children.setdefault(compression_parent,
3342
for parent in parents:
3343
self._annotate_children.setdefault(parent,
3345
num_gens = generation - kept_generation
3346
if ((num_gens >= self._generations_until_keep)
3347
and len(parents) > 1):
3348
kept_generation = generation
3349
self._nodes_to_keep_annotations.add(key)
3351
missing_versions = this_iteration.difference(build_details.keys())
3352
self._ghosts.update(missing_versions)
3353
for missing_version in missing_versions:
3354
# add a key, no parents
3355
self._revision_id_graph[missing_version] = ()
3356
pending.discard(missing_version) # don't look for it
3357
if self._ghosts.intersection(self._compression_children):
3359
"We cannot have nodes which have a ghost compression parent:\n"
3361
"compression children: %r"
3362
% (self._ghosts, self._compression_children))
3363
# Cleanout anything that depends on a ghost so that we don't wait for
3364
# the ghost to show up
3365
for node in self._ghosts:
3366
if node in self._annotate_children:
3367
# We won't be building this node
3368
del self._annotate_children[node]
3369
# Generally we will want to read the records in reverse order, because
3370
# we find the parent nodes after the children
3374
def _annotate_records(self, records):
3375
"""Build the annotations for the listed records."""
3376
# We iterate in the order read, rather than a strict order requested
3377
# However, process what we can, and put off to the side things that
3378
# still need parents, cleaning them up when those parents are
3380
for (rev_id, record,
3381
digest) in self._knit._read_records_iter(records):
3382
if rev_id in self._annotated_lines:
3384
parent_ids = self._revision_id_graph[rev_id]
3385
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3386
details = self._all_build_details[rev_id]
3387
(index_memo, compression_parent, parents,
3388
record_details) = details
3389
nodes_to_annotate = []
3390
# TODO: Remove the punning between compression parents, and
3391
# parent_ids, we should be able to do this without assuming
3393
if len(parent_ids) == 0:
3394
# There are no parents for this node, so just add it
3395
# TODO: This probably needs to be decoupled
3396
fulltext_content, delta = self._knit._factory.parse_record(
3397
rev_id, record, record_details, None)
3398
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3399
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3400
parent_ids, left_matching_blocks=None))
3402
child = (rev_id, parent_ids, record)
3403
# Check if all the parents are present
3404
self._check_parents(child, nodes_to_annotate)
3405
while nodes_to_annotate:
3406
# Should we use a queue here instead of a stack?
3407
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3408
(index_memo, compression_parent, parents,
3409
record_details) = self._all_build_details[rev_id]
3411
if compression_parent is not None:
3412
comp_children = self._compression_children[compression_parent]
3413
if rev_id not in comp_children:
3414
raise AssertionError("%r not in compression children %r"
3415
% (rev_id, comp_children))
3416
# If there is only 1 child, it is safe to reuse this
3418
reuse_content = (len(comp_children) == 1
3419
and compression_parent not in
3420
self._nodes_to_keep_annotations)
3422
# Remove it from the cache since it will be changing
3423
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3424
# Make sure to copy the fulltext since it might be
3426
parent_fulltext = list(parent_fulltext_content.text())
3428
parent_fulltext_content = self._fulltext_contents[compression_parent]
3429
parent_fulltext = parent_fulltext_content.text()
3430
comp_children.remove(rev_id)
3431
fulltext_content, delta = self._knit._factory.parse_record(
3432
rev_id, record, record_details,
3433
parent_fulltext_content,
3434
copy_base_content=(not reuse_content))
3435
fulltext = self._add_fulltext_content(rev_id,
3437
if compression_parent == parent_ids[0]:
3438
# the compression_parent is the left parent, so we can
3440
blocks = KnitContent.get_line_delta_blocks(delta,
3441
parent_fulltext, fulltext)
3443
fulltext_content = self._knit._factory.parse_fulltext(
3445
fulltext = self._add_fulltext_content(rev_id,
3447
nodes_to_annotate.extend(
3448
self._add_annotation(rev_id, fulltext, parent_ids,
3449
left_matching_blocks=blocks))
3451
def _get_heads_provider(self):
3452
"""Create a heads provider for resolving ancestry issues."""
3453
if self._heads_provider is not None:
3454
return self._heads_provider
3455
parent_provider = _mod_graph.DictParentsProvider(
3456
self._revision_id_graph)
3457
graph_obj = _mod_graph.Graph(parent_provider)
3458
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3459
self._heads_provider = head_cache
3462
def annotate(self, key):
3463
"""Return the annotated fulltext at the given key.
3465
:param key: The key to annotate.
3467
if len(self._knit._fallback_vfs) > 0:
3468
# stacked knits can't use the fast path at present.
3469
return self._simple_annotate(key)
3472
records = self._get_build_graph(key)
3473
if key in self._ghosts:
3474
raise errors.RevisionNotPresent(key, self._knit)
3475
self._annotate_records(records)
3476
return self._annotated_lines[key]
3477
except errors.RetryWithNewPacks, e:
3478
self._knit._access.reload_or_raise(e)
3479
# The cached build_details are no longer valid
3480
self._all_build_details.clear()
3482
def _simple_annotate(self, key):
3483
"""Return annotated fulltext, rediffing from the full texts.
3485
This is slow but makes no assumptions about the repository
3486
being able to produce line deltas.
3488
# TODO: this code generates a parent maps of present ancestors; it
3489
# could be split out into a separate method, and probably should use
3490
# iter_ancestry instead. -- mbp and robertc 20080704
3491
graph = _mod_graph.Graph(self._knit)
3492
head_cache = _mod_graph.FrozenHeadsCache(graph)
3493
search = graph._make_breadth_first_searcher([key])
3497
present, ghosts = search.next_with_ghosts()
3498
except StopIteration:
3500
keys.update(present)
3501
parent_map = self._knit.get_parent_map(keys)
3503
reannotate = annotate.reannotate
3504
for record in self._knit.get_record_stream(keys, 'topological', True):
3506
fulltext = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
3507
parents = parent_map[key]
3508
if parents is not None:
3509
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
3512
parent_cache[key] = list(
3513
reannotate(parent_lines, fulltext, key, None, head_cache))
3515
return parent_cache[key]
3517
raise errors.RevisionNotPresent(key, self._knit)
3521
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3523
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
