1
# Copyright (C) 2006-2011 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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 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
60
from bzrlib.lazy_import import lazy_import
61
lazy_import(globals(), """
78
from bzrlib.repofmt import pack_repo
85
from bzrlib.errors import (
93
from bzrlib.osutils import (
99
from bzrlib.versionedfile import (
101
AbsentContentFactory,
106
VersionedFilesWithFallbacks,
110
# TODO: Split out code specific to this format into an associated object.
112
# TODO: Can we put in some kind of value to check that the index and data
113
# files belong together?
115
# TODO: accommodate binaries, perhaps by storing a byte count
117
# TODO: function to check whole file
119
# TODO: atomically append data, then measure backwards from the cursor
120
# position after writing to work out where it was located. we may need to
121
# bypass python file buffering.
123
DATA_SUFFIX = '.knit'
124
INDEX_SUFFIX = '.kndx'
125
_STREAM_MIN_BUFFER_SIZE = 5*1024*1024
128
class KnitAdapter(object):
129
"""Base class for knit record adaption."""
131
def __init__(self, basis_vf):
132
"""Create an adapter which accesses full texts from basis_vf.
134
:param basis_vf: A versioned file to access basis texts of deltas from.
135
May be None for adapters that do not need to access basis texts.
137
self._data = KnitVersionedFiles(None, None)
138
self._annotate_factory = KnitAnnotateFactory()
139
self._plain_factory = KnitPlainFactory()
140
self._basis_vf = basis_vf
143
class FTAnnotatedToUnannotated(KnitAdapter):
144
"""An adapter from FT annotated knits to unannotated ones."""
146
def get_bytes(self, factory):
147
annotated_compressed_bytes = factory._raw_record
149
self._data._parse_record_unchecked(annotated_compressed_bytes)
150
content = self._annotate_factory.parse_fulltext(contents, rec[1])
151
size, bytes = self._data._record_to_data((rec[1],), rec[3], content.text())
155
class DeltaAnnotatedToUnannotated(KnitAdapter):
156
"""An adapter for deltas from annotated to unannotated."""
158
def get_bytes(self, factory):
159
annotated_compressed_bytes = factory._raw_record
161
self._data._parse_record_unchecked(annotated_compressed_bytes)
162
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
164
contents = self._plain_factory.lower_line_delta(delta)
165
size, bytes = self._data._record_to_data((rec[1],), rec[3], contents)
169
class FTAnnotatedToFullText(KnitAdapter):
170
"""An adapter from FT annotated knits to unannotated ones."""
172
def get_bytes(self, factory):
173
annotated_compressed_bytes = factory._raw_record
175
self._data._parse_record_unchecked(annotated_compressed_bytes)
176
content, delta = self._annotate_factory.parse_record(factory.key[-1],
177
contents, factory._build_details, None)
178
return ''.join(content.text())
181
class DeltaAnnotatedToFullText(KnitAdapter):
182
"""An adapter for deltas from annotated to unannotated."""
184
def get_bytes(self, factory):
185
annotated_compressed_bytes = factory._raw_record
187
self._data._parse_record_unchecked(annotated_compressed_bytes)
188
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
190
compression_parent = factory.parents[0]
191
basis_entry = self._basis_vf.get_record_stream(
192
[compression_parent], 'unordered', True).next()
193
if basis_entry.storage_kind == 'absent':
194
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
195
basis_chunks = basis_entry.get_bytes_as('chunked')
196
basis_lines = osutils.chunks_to_lines(basis_chunks)
197
# Manually apply the delta because we have one annotated content and
199
basis_content = PlainKnitContent(basis_lines, compression_parent)
200
basis_content.apply_delta(delta, rec[1])
201
basis_content._should_strip_eol = factory._build_details[1]
202
return ''.join(basis_content.text())
205
class FTPlainToFullText(KnitAdapter):
206
"""An adapter from FT plain knits to unannotated ones."""
208
def get_bytes(self, factory):
209
compressed_bytes = factory._raw_record
211
self._data._parse_record_unchecked(compressed_bytes)
212
content, delta = self._plain_factory.parse_record(factory.key[-1],
213
contents, factory._build_details, None)
214
return ''.join(content.text())
217
class DeltaPlainToFullText(KnitAdapter):
218
"""An adapter for deltas from annotated to unannotated."""
220
def get_bytes(self, factory):
221
compressed_bytes = factory._raw_record
223
self._data._parse_record_unchecked(compressed_bytes)
224
delta = self._plain_factory.parse_line_delta(contents, rec[1])
225
compression_parent = factory.parents[0]
226
# XXX: string splitting overhead.
227
basis_entry = self._basis_vf.get_record_stream(
228
[compression_parent], 'unordered', True).next()
229
if basis_entry.storage_kind == 'absent':
230
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
231
basis_chunks = basis_entry.get_bytes_as('chunked')
232
basis_lines = osutils.chunks_to_lines(basis_chunks)
233
basis_content = PlainKnitContent(basis_lines, compression_parent)
234
# Manually apply the delta because we have one annotated content and
236
content, _ = self._plain_factory.parse_record(rec[1], contents,
237
factory._build_details, basis_content)
238
return ''.join(content.text())
241
class KnitContentFactory(ContentFactory):
242
"""Content factory for streaming from knits.
244
:seealso ContentFactory:
247
def __init__(self, key, parents, build_details, sha1, raw_record,
248
annotated, knit=None, network_bytes=None):
249
"""Create a KnitContentFactory for key.
252
:param parents: The parents.
253
:param build_details: The build details as returned from
255
:param sha1: The sha1 expected from the full text of this object.
256
:param raw_record: The bytes of the knit data from disk.
257
:param annotated: True if the raw data is annotated.
258
:param network_bytes: None to calculate the network bytes on demand,
259
not-none if they are already known.
261
ContentFactory.__init__(self)
264
self.parents = parents
265
if build_details[0] == 'line-delta':
270
annotated_kind = 'annotated-'
273
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
274
self._raw_record = raw_record
275
self._network_bytes = network_bytes
276
self._build_details = build_details
279
def _create_network_bytes(self):
280
"""Create a fully serialised network version for transmission."""
281
# storage_kind, key, parents, Noeol, raw_record
282
key_bytes = '\x00'.join(self.key)
283
if self.parents is None:
284
parent_bytes = 'None:'
286
parent_bytes = '\t'.join('\x00'.join(key) for key in self.parents)
287
if self._build_details[1]:
291
network_bytes = "%s\n%s\n%s\n%s%s" % (self.storage_kind, key_bytes,
292
parent_bytes, noeol, self._raw_record)
293
self._network_bytes = network_bytes
295
def get_bytes_as(self, storage_kind):
296
if storage_kind == self.storage_kind:
297
if self._network_bytes is None:
298
self._create_network_bytes()
299
return self._network_bytes
300
if ('-ft-' in self.storage_kind and
301
storage_kind in ('chunked', 'fulltext')):
302
adapter_key = (self.storage_kind, 'fulltext')
303
adapter_factory = adapter_registry.get(adapter_key)
304
adapter = adapter_factory(None)
305
bytes = adapter.get_bytes(self)
306
if storage_kind == 'chunked':
310
if self._knit is not None:
311
# Not redundant with direct conversion above - that only handles
313
if storage_kind == 'chunked':
314
return self._knit.get_lines(self.key[0])
315
elif storage_kind == 'fulltext':
316
return self._knit.get_text(self.key[0])
317
raise errors.UnavailableRepresentation(self.key, storage_kind,
321
class LazyKnitContentFactory(ContentFactory):
322
"""A ContentFactory which can either generate full text or a wire form.
324
:seealso ContentFactory:
327
def __init__(self, key, parents, generator, first):
328
"""Create a LazyKnitContentFactory.
330
:param key: The key of the record.
331
:param parents: The parents of the record.
332
:param generator: A _ContentMapGenerator containing the record for this
334
:param first: Is this the first content object returned from generator?
335
if it is, its storage kind is knit-delta-closure, otherwise it is
336
knit-delta-closure-ref
339
self.parents = parents
341
self._generator = generator
342
self.storage_kind = "knit-delta-closure"
344
self.storage_kind = self.storage_kind + "-ref"
347
def get_bytes_as(self, storage_kind):
348
if storage_kind == self.storage_kind:
350
return self._generator._wire_bytes()
352
# all the keys etc are contained in the bytes returned in the
355
if storage_kind in ('chunked', 'fulltext'):
356
chunks = self._generator._get_one_work(self.key).text()
357
if storage_kind == 'chunked':
360
return ''.join(chunks)
361
raise errors.UnavailableRepresentation(self.key, storage_kind,
365
def knit_delta_closure_to_records(storage_kind, bytes, line_end):
366
"""Convert a network record to a iterator over stream records.
368
:param storage_kind: The storage kind of the record.
369
Must be 'knit-delta-closure'.
370
:param bytes: The bytes of the record on the network.
372
generator = _NetworkContentMapGenerator(bytes, line_end)
373
return generator.get_record_stream()
376
def knit_network_to_record(storage_kind, bytes, line_end):
377
"""Convert a network record to a record object.
379
:param storage_kind: The storage kind of the record.
380
:param bytes: The bytes of the record on the network.
383
line_end = bytes.find('\n', start)
384
key = tuple(bytes[start:line_end].split('\x00'))
386
line_end = bytes.find('\n', start)
387
parent_line = bytes[start:line_end]
388
if parent_line == 'None:':
392
[tuple(segment.split('\x00')) for segment in parent_line.split('\t')
395
noeol = bytes[start] == 'N'
396
if 'ft' in storage_kind:
399
method = 'line-delta'
400
build_details = (method, noeol)
402
raw_record = bytes[start:]
403
annotated = 'annotated' in storage_kind
404
return [KnitContentFactory(key, parents, build_details, None, raw_record,
405
annotated, network_bytes=bytes)]
408
class KnitContent(object):
409
"""Content of a knit version to which deltas can be applied.
411
This is always stored in memory as a list of lines with \\n at the end,
412
plus a flag saying if the final ending is really there or not, because that
413
corresponds to the on-disk knit representation.
417
self._should_strip_eol = False
419
def apply_delta(self, delta, new_version_id):
420
"""Apply delta to this object to become new_version_id."""
421
raise NotImplementedError(self.apply_delta)
423
def line_delta_iter(self, new_lines):
424
"""Generate line-based delta from this content to new_lines."""
425
new_texts = new_lines.text()
426
old_texts = self.text()
427
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
428
for tag, i1, i2, j1, j2 in s.get_opcodes():
431
# ofrom, oto, length, data
432
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
434
def line_delta(self, new_lines):
435
return list(self.line_delta_iter(new_lines))
438
def get_line_delta_blocks(knit_delta, source, target):
439
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
440
target_len = len(target)
443
for s_begin, s_end, t_len, new_text in knit_delta:
444
true_n = s_begin - s_pos
447
# knit deltas do not provide reliable info about whether the
448
# last line of a file matches, due to eol handling.
449
if source[s_pos + n -1] != target[t_pos + n -1]:
452
yield s_pos, t_pos, n
453
t_pos += t_len + true_n
455
n = target_len - t_pos
457
if source[s_pos + n -1] != target[t_pos + n -1]:
460
yield s_pos, t_pos, n
461
yield s_pos + (target_len - t_pos), target_len, 0
464
class AnnotatedKnitContent(KnitContent):
465
"""Annotated content."""
467
def __init__(self, lines):
468
KnitContent.__init__(self)
472
"""Return a list of (origin, text) for each content line."""
473
lines = self._lines[:]
474
if self._should_strip_eol:
475
origin, last_line = lines[-1]
476
lines[-1] = (origin, last_line.rstrip('\n'))
479
def apply_delta(self, delta, new_version_id):
480
"""Apply delta to this object to become new_version_id."""
483
for start, end, count, delta_lines in delta:
484
lines[offset+start:offset+end] = delta_lines
485
offset = offset + (start - end) + count
489
lines = [text for origin, text in self._lines]
490
except ValueError, e:
491
# most commonly (only?) caused by the internal form of the knit
492
# missing annotation information because of a bug - see thread
494
raise KnitCorrupt(self,
495
"line in annotated knit missing annotation information: %s"
497
if self._should_strip_eol:
498
lines[-1] = lines[-1].rstrip('\n')
502
return AnnotatedKnitContent(self._lines[:])
505
class PlainKnitContent(KnitContent):
506
"""Unannotated content.
508
When annotate[_iter] is called on this content, the same version is reported
509
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
513
def __init__(self, lines, version_id):
514
KnitContent.__init__(self)
516
self._version_id = version_id
519
"""Return a list of (origin, text) for each content line."""
520
return [(self._version_id, line) for line in self._lines]
522
def apply_delta(self, delta, new_version_id):
523
"""Apply delta to this object to become new_version_id."""
526
for start, end, count, delta_lines in delta:
527
lines[offset+start:offset+end] = delta_lines
528
offset = offset + (start - end) + count
529
self._version_id = new_version_id
532
return PlainKnitContent(self._lines[:], self._version_id)
536
if self._should_strip_eol:
538
lines[-1] = lines[-1].rstrip('\n')
542
class _KnitFactory(object):
543
"""Base class for common Factory functions."""
545
def parse_record(self, version_id, record, record_details,
546
base_content, copy_base_content=True):
547
"""Parse a record into a full content object.
549
:param version_id: The official version id for this content
550
:param record: The data returned by read_records_iter()
551
:param record_details: Details about the record returned by
553
:param base_content: If get_build_details returns a compression_parent,
554
you must return a base_content here, else use None
555
:param copy_base_content: When building from the base_content, decide
556
you can either copy it and return a new object, or modify it in
558
:return: (content, delta) A Content object and possibly a line-delta,
561
method, noeol = record_details
562
if method == 'line-delta':
563
if copy_base_content:
564
content = base_content.copy()
566
content = base_content
567
delta = self.parse_line_delta(record, version_id)
568
content.apply_delta(delta, version_id)
570
content = self.parse_fulltext(record, version_id)
572
content._should_strip_eol = noeol
573
return (content, delta)
576
class KnitAnnotateFactory(_KnitFactory):
577
"""Factory for creating annotated Content objects."""
581
def make(self, lines, version_id):
582
num_lines = len(lines)
583
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
585
def parse_fulltext(self, content, version_id):
586
"""Convert fulltext to internal representation
588
fulltext content is of the format
589
revid(utf8) plaintext\n
590
internal representation is of the format:
593
# TODO: jam 20070209 The tests expect this to be returned as tuples,
594
# but the code itself doesn't really depend on that.
595
# Figure out a way to not require the overhead of turning the
596
# list back into tuples.
597
lines = [tuple(line.split(' ', 1)) for line in content]
598
return AnnotatedKnitContent(lines)
600
def parse_line_delta_iter(self, lines):
601
return iter(self.parse_line_delta(lines))
603
def parse_line_delta(self, lines, version_id, plain=False):
604
"""Convert a line based delta into internal representation.
606
line delta is in the form of:
607
intstart intend intcount
609
revid(utf8) newline\n
610
internal representation is
611
(start, end, count, [1..count tuples (revid, newline)])
613
:param plain: If True, the lines are returned as a plain
614
list without annotations, not as a list of (origin, content) tuples, i.e.
615
(start, end, count, [1..count newline])
622
def cache_and_return(line):
623
origin, text = line.split(' ', 1)
624
return cache.setdefault(origin, origin), text
626
# walk through the lines parsing.
627
# Note that the plain test is explicitly pulled out of the
628
# loop to minimise any performance impact
631
start, end, count = [int(n) for n in header.split(',')]
632
contents = [next().split(' ', 1)[1] for i in xrange(count)]
633
result.append((start, end, count, contents))
636
start, end, count = [int(n) for n in header.split(',')]
637
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
638
result.append((start, end, count, contents))
641
def get_fulltext_content(self, lines):
642
"""Extract just the content lines from a fulltext."""
643
return (line.split(' ', 1)[1] for line in lines)
645
def get_linedelta_content(self, lines):
646
"""Extract just the content from a line delta.
648
This doesn't return all of the extra information stored in a delta.
649
Only the actual content lines.
654
header = header.split(',')
655
count = int(header[2])
656
for i in xrange(count):
657
origin, text = next().split(' ', 1)
660
def lower_fulltext(self, content):
661
"""convert a fulltext content record into a serializable form.
663
see parse_fulltext which this inverts.
665
return ['%s %s' % (o, t) for o, t in content._lines]
667
def lower_line_delta(self, delta):
668
"""convert a delta into a serializable form.
670
See parse_line_delta which this inverts.
672
# TODO: jam 20070209 We only do the caching thing to make sure that
673
# the origin is a valid utf-8 line, eventually we could remove it
675
for start, end, c, lines in delta:
676
out.append('%d,%d,%d\n' % (start, end, c))
677
out.extend(origin + ' ' + text
678
for origin, text in lines)
681
def annotate(self, knit, key):
682
content = knit._get_content(key)
683
# adjust for the fact that serialised annotations are only key suffixes
685
if type(key) is tuple:
687
origins = content.annotate()
689
for origin, line in origins:
690
result.append((prefix + (origin,), line))
693
# XXX: This smells a bit. Why would key ever be a non-tuple here?
694
# Aren't keys defined to be tuples? -- spiv 20080618
695
return content.annotate()
698
class KnitPlainFactory(_KnitFactory):
699
"""Factory for creating plain Content objects."""
703
def make(self, lines, version_id):
704
return PlainKnitContent(lines, version_id)
706
def parse_fulltext(self, content, version_id):
707
"""This parses an unannotated fulltext.
709
Note that this is not a noop - the internal representation
710
has (versionid, line) - its just a constant versionid.
712
return self.make(content, version_id)
714
def parse_line_delta_iter(self, lines, version_id):
716
num_lines = len(lines)
717
while cur < num_lines:
720
start, end, c = [int(n) for n in header.split(',')]
721
yield start, end, c, lines[cur:cur+c]
724
def parse_line_delta(self, lines, version_id):
725
return list(self.parse_line_delta_iter(lines, version_id))
727
def get_fulltext_content(self, lines):
728
"""Extract just the content lines from a fulltext."""
731
def get_linedelta_content(self, lines):
732
"""Extract just the content from a line delta.
734
This doesn't return all of the extra information stored in a delta.
735
Only the actual content lines.
740
header = header.split(',')
741
count = int(header[2])
742
for i in xrange(count):
745
def lower_fulltext(self, content):
746
return content.text()
748
def lower_line_delta(self, delta):
750
for start, end, c, lines in delta:
751
out.append('%d,%d,%d\n' % (start, end, c))
755
def annotate(self, knit, key):
756
annotator = _KnitAnnotator(knit)
757
return annotator.annotate_flat(key)
761
def make_file_factory(annotated, mapper):
762
"""Create a factory for creating a file based KnitVersionedFiles.
764
This is only functional enough to run interface tests, it doesn't try to
765
provide a full pack environment.
767
:param annotated: knit annotations are wanted.
768
:param mapper: The mapper from keys to paths.
770
def factory(transport):
771
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
772
access = _KnitKeyAccess(transport, mapper)
773
return KnitVersionedFiles(index, access, annotated=annotated)
777
def make_pack_factory(graph, delta, keylength):
778
"""Create a factory for creating a pack based VersionedFiles.
780
This is only functional enough to run interface tests, it doesn't try to
781
provide a full pack environment.
783
:param graph: Store a graph.
784
:param delta: Delta compress contents.
785
:param keylength: How long should keys be.
787
def factory(transport):
788
parents = graph or delta
794
max_delta_chain = 200
797
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
798
key_elements=keylength)
799
stream = transport.open_write_stream('newpack')
800
writer = pack.ContainerWriter(stream.write)
802
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
803
deltas=delta, add_callback=graph_index.add_nodes)
804
access = pack_repo._DirectPackAccess({})
805
access.set_writer(writer, graph_index, (transport, 'newpack'))
806
result = KnitVersionedFiles(index, access,
807
max_delta_chain=max_delta_chain)
808
result.stream = stream
809
result.writer = writer
814
def cleanup_pack_knit(versioned_files):
815
versioned_files.stream.close()
816
versioned_files.writer.end()
819
def _get_total_build_size(self, keys, positions):
820
"""Determine the total bytes to build these keys.
822
(helper function because _KnitGraphIndex and _KndxIndex work the same, but
823
don't inherit from a common base.)
825
:param keys: Keys that we want to build
826
:param positions: dict of {key, (info, index_memo, comp_parent)} (such
827
as returned by _get_components_positions)
828
:return: Number of bytes to build those keys
830
all_build_index_memos = {}
834
for key in build_keys:
835
# This is mostly for the 'stacked' case
836
# Where we will be getting the data from a fallback
837
if key not in positions:
839
_, index_memo, compression_parent = positions[key]
840
all_build_index_memos[key] = index_memo
841
if compression_parent not in all_build_index_memos:
842
next_keys.add(compression_parent)
843
build_keys = next_keys
844
return sum([index_memo[2] for index_memo
845
in all_build_index_memos.itervalues()])
848
class KnitVersionedFiles(VersionedFilesWithFallbacks):
849
"""Storage for many versioned files using knit compression.
851
Backend storage is managed by indices and data objects.
853
:ivar _index: A _KnitGraphIndex or similar that can describe the
854
parents, graph, compression and data location of entries in this
855
KnitVersionedFiles. Note that this is only the index for
856
*this* vfs; if there are fallbacks they must be queried separately.
859
def __init__(self, index, data_access, max_delta_chain=200,
860
annotated=False, reload_func=None):
861
"""Create a KnitVersionedFiles with index and data_access.
863
:param index: The index for the knit data.
864
:param data_access: The access object to store and retrieve knit
866
:param max_delta_chain: The maximum number of deltas to permit during
867
insertion. Set to 0 to prohibit the use of deltas.
868
:param annotated: Set to True to cause annotations to be calculated and
869
stored during insertion.
870
:param reload_func: An function that can be called if we think we need
871
to reload the pack listing and try again. See
872
'bzrlib.repofmt.pack_repo.AggregateIndex' for the signature.
875
self._access = data_access
876
self._max_delta_chain = max_delta_chain
878
self._factory = KnitAnnotateFactory()
880
self._factory = KnitPlainFactory()
881
self._immediate_fallback_vfs = []
882
self._reload_func = reload_func
885
return "%s(%r, %r)" % (
886
self.__class__.__name__,
890
def without_fallbacks(self):
891
"""Return a clone of this object without any fallbacks configured."""
892
return KnitVersionedFiles(self._index, self._access,
893
self._max_delta_chain, self._factory.annotated,
896
def add_fallback_versioned_files(self, a_versioned_files):
897
"""Add a source of texts for texts not present in this knit.
899
:param a_versioned_files: A VersionedFiles object.
901
self._immediate_fallback_vfs.append(a_versioned_files)
903
def add_lines(self, key, parents, lines, parent_texts=None,
904
left_matching_blocks=None, nostore_sha=None, random_id=False,
906
"""See VersionedFiles.add_lines()."""
907
self._index._check_write_ok()
908
self._check_add(key, lines, random_id, check_content)
910
# The caller might pass None if there is no graph data, but kndx
911
# indexes can't directly store that, so we give them
912
# an empty tuple instead.
914
line_bytes = ''.join(lines)
915
return self._add(key, lines, parents,
916
parent_texts, left_matching_blocks, nostore_sha, random_id,
917
line_bytes=line_bytes)
919
def _add_text(self, key, parents, text, nostore_sha=None, random_id=False):
920
"""See VersionedFiles._add_text()."""
921
self._index._check_write_ok()
922
self._check_add(key, None, random_id, check_content=False)
923
if text.__class__ is not str:
924
raise errors.BzrBadParameterUnicode("text")
926
# The caller might pass None if there is no graph data, but kndx
927
# indexes can't directly store that, so we give them
928
# an empty tuple instead.
930
return self._add(key, None, parents,
931
None, None, nostore_sha, random_id,
934
def _add(self, key, lines, parents, parent_texts,
935
left_matching_blocks, nostore_sha, random_id,
937
"""Add a set of lines on top of version specified by parents.
939
Any versions not present will be converted into ghosts.
941
:param lines: A list of strings where each one is a single line (has a
942
single newline at the end of the string) This is now optional
943
(callers can pass None). It is left in its location for backwards
944
compatibility. It should ''.join(lines) must == line_bytes
945
:param line_bytes: A single string containing the content
947
We pass both lines and line_bytes because different routes bring the
948
values to this function. And for memory efficiency, we don't want to
949
have to split/join on-demand.
951
# first thing, if the content is something we don't need to store, find
953
digest = sha_string(line_bytes)
954
if nostore_sha == digest:
955
raise errors.ExistingContent
958
if parent_texts is None:
960
# Do a single query to ascertain parent presence; we only compress
961
# against parents in the same kvf.
962
present_parent_map = self._index.get_parent_map(parents)
963
for parent in parents:
964
if parent in present_parent_map:
965
present_parents.append(parent)
967
# Currently we can only compress against the left most present parent.
968
if (len(present_parents) == 0 or
969
present_parents[0] != parents[0]):
972
# To speed the extract of texts the delta chain is limited
973
# to a fixed number of deltas. This should minimize both
974
# I/O and the time spend applying deltas.
975
delta = self._check_should_delta(present_parents[0])
977
text_length = len(line_bytes)
980
# Note: line_bytes is not modified to add a newline, that is tracked
981
# via the no_eol flag. 'lines' *is* modified, because that is the
982
# general values needed by the Content code.
983
if line_bytes and line_bytes[-1] != '\n':
984
options.append('no-eol')
986
# Copy the existing list, or create a new one
988
lines = osutils.split_lines(line_bytes)
991
# Replace the last line with one that ends in a final newline
992
lines[-1] = lines[-1] + '\n'
994
lines = osutils.split_lines(line_bytes)
996
for element in key[:-1]:
997
if type(element) is not str:
998
raise TypeError("key contains non-strings: %r" % (key,))
1000
key = key[:-1] + ('sha1:' + digest,)
1001
elif type(key[-1]) is not str:
1002
raise TypeError("key contains non-strings: %r" % (key,))
1003
# Knit hunks are still last-element only
1004
version_id = key[-1]
1005
content = self._factory.make(lines, version_id)
1007
# Hint to the content object that its text() call should strip the
1009
content._should_strip_eol = True
1010
if delta or (self._factory.annotated and len(present_parents) > 0):
1011
# Merge annotations from parent texts if needed.
1012
delta_hunks = self._merge_annotations(content, present_parents,
1013
parent_texts, delta, self._factory.annotated,
1014
left_matching_blocks)
1017
options.append('line-delta')
1018
store_lines = self._factory.lower_line_delta(delta_hunks)
1019
size, bytes = self._record_to_data(key, digest,
1022
options.append('fulltext')
1023
# isinstance is slower and we have no hierarchy.
1024
if self._factory.__class__ is KnitPlainFactory:
1025
# Use the already joined bytes saving iteration time in
1027
dense_lines = [line_bytes]
1029
dense_lines.append('\n')
1030
size, bytes = self._record_to_data(key, digest,
1033
# get mixed annotation + content and feed it into the
1035
store_lines = self._factory.lower_fulltext(content)
1036
size, bytes = self._record_to_data(key, digest,
1039
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
1040
self._index.add_records(
1041
((key, options, access_memo, parents),),
1042
random_id=random_id)
1043
return digest, text_length, content
1045
def annotate(self, key):
1046
"""See VersionedFiles.annotate."""
1047
return self._factory.annotate(self, key)
1049
def get_annotator(self):
1050
return _KnitAnnotator(self)
1052
def check(self, progress_bar=None, keys=None):
1053
"""See VersionedFiles.check()."""
1055
return self._logical_check()
1057
# At the moment, check does not extra work over get_record_stream
1058
return self.get_record_stream(keys, 'unordered', True)
1060
def _logical_check(self):
1061
# This doesn't actually test extraction of everything, but that will
1062
# impact 'bzr check' substantially, and needs to be integrated with
1063
# care. However, it does check for the obvious problem of a delta with
1065
keys = self._index.keys()
1066
parent_map = self.get_parent_map(keys)
1068
if self._index.get_method(key) != 'fulltext':
1069
compression_parent = parent_map[key][0]
1070
if compression_parent not in parent_map:
1071
raise errors.KnitCorrupt(self,
1072
"Missing basis parent %s for %s" % (
1073
compression_parent, key))
1074
for fallback_vfs in self._immediate_fallback_vfs:
1075
fallback_vfs.check()
1077
def _check_add(self, key, lines, random_id, check_content):
1078
"""check that version_id and lines are safe to add."""
1079
version_id = key[-1]
1080
if version_id is not None:
1081
if contains_whitespace(version_id):
1082
raise InvalidRevisionId(version_id, self)
1083
self.check_not_reserved_id(version_id)
1084
# TODO: If random_id==False and the key is already present, we should
1085
# probably check that the existing content is identical to what is
1086
# being inserted, and otherwise raise an exception. This would make
1087
# the bundle code simpler.
1089
self._check_lines_not_unicode(lines)
1090
self._check_lines_are_lines(lines)
1092
def _check_header(self, key, line):
1093
rec = self._split_header(line)
1094
self._check_header_version(rec, key[-1])
1097
def _check_header_version(self, rec, version_id):
1098
"""Checks the header version on original format knit records.
1100
These have the last component of the key embedded in the record.
1102
if rec[1] != version_id:
1103
raise KnitCorrupt(self,
1104
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1106
def _check_should_delta(self, parent):
1107
"""Iterate back through the parent listing, looking for a fulltext.
1109
This is used when we want to decide whether to add a delta or a new
1110
fulltext. It searches for _max_delta_chain parents. When it finds a
1111
fulltext parent, it sees if the total size of the deltas leading up to
1112
it is large enough to indicate that we want a new full text anyway.
1114
Return True if we should create a new delta, False if we should use a
1118
fulltext_size = None
1119
for count in xrange(self._max_delta_chain):
1121
# Note that this only looks in the index of this particular
1122
# KnitVersionedFiles, not in the fallbacks. This ensures that
1123
# we won't store a delta spanning physical repository
1125
build_details = self._index.get_build_details([parent])
1126
parent_details = build_details[parent]
1127
except (RevisionNotPresent, KeyError), e:
1128
# Some basis is not locally present: always fulltext
1130
index_memo, compression_parent, _, _ = parent_details
1131
_, _, size = index_memo
1132
if compression_parent is None:
1133
fulltext_size = size
1136
# We don't explicitly check for presence because this is in an
1137
# inner loop, and if it's missing it'll fail anyhow.
1138
parent = compression_parent
1140
# We couldn't find a fulltext, so we must create a new one
1142
# Simple heuristic - if the total I/O wold be greater as a delta than
1143
# the originally installed fulltext, we create a new fulltext.
1144
return fulltext_size > delta_size
1146
def _build_details_to_components(self, build_details):
1147
"""Convert a build_details tuple to a position tuple."""
1148
# record_details, access_memo, compression_parent
1149
return build_details[3], build_details[0], build_details[1]
1151
def _get_components_positions(self, keys, allow_missing=False):
1152
"""Produce a map of position data for the components of keys.
1154
This data is intended to be used for retrieving the knit records.
1156
A dict of key to (record_details, index_memo, next, parents) is
1159
* method is the way referenced data should be applied.
1160
* index_memo is the handle to pass to the data access to actually get
1162
* next is the build-parent of the version, or None for fulltexts.
1163
* parents is the version_ids of the parents of this version
1165
:param allow_missing: If True do not raise an error on a missing
1166
component, just ignore it.
1169
pending_components = keys
1170
while pending_components:
1171
build_details = self._index.get_build_details(pending_components)
1172
current_components = set(pending_components)
1173
pending_components = set()
1174
for key, details in build_details.iteritems():
1175
(index_memo, compression_parent, parents,
1176
record_details) = details
1177
method = record_details[0]
1178
if compression_parent is not None:
1179
pending_components.add(compression_parent)
1180
component_data[key] = self._build_details_to_components(details)
1181
missing = current_components.difference(build_details)
1182
if missing and not allow_missing:
1183
raise errors.RevisionNotPresent(missing.pop(), self)
1184
return component_data
1186
def _get_content(self, key, parent_texts={}):
1187
"""Returns a content object that makes up the specified
1189
cached_version = parent_texts.get(key, None)
1190
if cached_version is not None:
1191
# Ensure the cache dict is valid.
1192
if not self.get_parent_map([key]):
1193
raise RevisionNotPresent(key, self)
1194
return cached_version
1195
generator = _VFContentMapGenerator(self, [key])
1196
return generator._get_content(key)
1198
def get_parent_map(self, keys):
1199
"""Get a map of the graph parents of keys.
1201
:param keys: The keys to look up parents for.
1202
:return: A mapping from keys to parents. Absent keys are absent from
1205
return self._get_parent_map_with_sources(keys)[0]
1207
def _get_parent_map_with_sources(self, keys):
1208
"""Get a map of the parents of keys.
1210
:param keys: The keys to look up parents for.
1211
:return: A tuple. The first element is a mapping from keys to parents.
1212
Absent keys are absent from the mapping. The second element is a
1213
list with the locations each key was found in. The first element
1214
is the in-this-knit parents, the second the first fallback source,
1218
sources = [self._index] + self._immediate_fallback_vfs
1221
for source in sources:
1224
new_result = source.get_parent_map(missing)
1225
source_results.append(new_result)
1226
result.update(new_result)
1227
missing.difference_update(set(new_result))
1228
return result, source_results
1230
def _get_record_map(self, keys, allow_missing=False):
1231
"""Produce a dictionary of knit records.
1233
:return: {key:(record, record_details, digest, next)}
1235
* record: data returned from read_records (a KnitContentobject)
1236
* record_details: opaque information to pass to parse_record
1237
* digest: SHA1 digest of the full text after all steps are done
1238
* next: build-parent of the version, i.e. the leftmost ancestor.
1239
Will be None if the record is not a delta.
1241
:param keys: The keys to build a map for
1242
:param allow_missing: If some records are missing, rather than
1243
error, just return the data that could be generated.
1245
raw_map = self._get_record_map_unparsed(keys,
1246
allow_missing=allow_missing)
1247
return self._raw_map_to_record_map(raw_map)
1249
def _raw_map_to_record_map(self, raw_map):
1250
"""Parse the contents of _get_record_map_unparsed.
1252
:return: see _get_record_map.
1256
data, record_details, next = raw_map[key]
1257
content, digest = self._parse_record(key[-1], data)
1258
result[key] = content, record_details, digest, next
1261
def _get_record_map_unparsed(self, keys, allow_missing=False):
1262
"""Get the raw data for reconstructing keys without parsing it.
1264
:return: A dict suitable for parsing via _raw_map_to_record_map.
1265
key-> raw_bytes, (method, noeol), compression_parent
1267
# This retries the whole request if anything fails. Potentially we
1268
# could be a bit more selective. We could track the keys whose records
1269
# we have successfully found, and then only request the new records
1270
# from there. However, _get_components_positions grabs the whole build
1271
# chain, which means we'll likely try to grab the same records again
1272
# anyway. Also, can the build chains change as part of a pack
1273
# operation? We wouldn't want to end up with a broken chain.
1276
position_map = self._get_components_positions(keys,
1277
allow_missing=allow_missing)
1278
# key = component_id, r = record_details, i_m = index_memo,
1280
records = [(key, i_m) for key, (r, i_m, n)
1281
in position_map.iteritems()]
1282
# Sort by the index memo, so that we request records from the
1283
# same pack file together, and in forward-sorted order
1284
records.sort(key=operator.itemgetter(1))
1286
for key, data in self._read_records_iter_unchecked(records):
1287
(record_details, index_memo, next) = position_map[key]
1288
raw_record_map[key] = data, record_details, next
1289
return raw_record_map
1290
except errors.RetryWithNewPacks, e:
1291
self._access.reload_or_raise(e)
1294
def _split_by_prefix(cls, keys):
1295
"""For the given keys, split them up based on their prefix.
1297
To keep memory pressure somewhat under control, split the
1298
requests back into per-file-id requests, otherwise "bzr co"
1299
extracts the full tree into memory before writing it to disk.
1300
This should be revisited if _get_content_maps() can ever cross
1303
The keys for a given file_id are kept in the same relative order.
1304
Ordering between file_ids is not, though prefix_order will return the
1305
order that the key was first seen.
1307
:param keys: An iterable of key tuples
1308
:return: (split_map, prefix_order)
1309
split_map A dictionary mapping prefix => keys
1310
prefix_order The order that we saw the various prefixes
1312
split_by_prefix = {}
1320
if prefix in split_by_prefix:
1321
split_by_prefix[prefix].append(key)
1323
split_by_prefix[prefix] = [key]
1324
prefix_order.append(prefix)
1325
return split_by_prefix, prefix_order
1327
def _group_keys_for_io(self, keys, non_local_keys, positions,
1328
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1329
"""For the given keys, group them into 'best-sized' requests.
1331
The idea is to avoid making 1 request per file, but to never try to
1332
unpack an entire 1.5GB source tree in a single pass. Also when
1333
possible, we should try to group requests to the same pack file
1336
:return: list of (keys, non_local) tuples that indicate what keys
1337
should be fetched next.
1339
# TODO: Ideally we would group on 2 factors. We want to extract texts
1340
# from the same pack file together, and we want to extract all
1341
# the texts for a given build-chain together. Ultimately it
1342
# probably needs a better global view.
1343
total_keys = len(keys)
1344
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1345
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1347
cur_non_local = set()
1351
for prefix in prefix_order:
1352
keys = prefix_split_keys[prefix]
1353
non_local = prefix_split_non_local_keys.get(prefix, [])
1355
this_size = self._index._get_total_build_size(keys, positions)
1356
cur_size += this_size
1357
cur_keys.extend(keys)
1358
cur_non_local.update(non_local)
1359
if cur_size > _min_buffer_size:
1360
result.append((cur_keys, cur_non_local))
1361
sizes.append(cur_size)
1363
cur_non_local = set()
1366
result.append((cur_keys, cur_non_local))
1367
sizes.append(cur_size)
1370
def get_record_stream(self, keys, ordering, include_delta_closure):
1371
"""Get a stream of records for keys.
1373
:param keys: The keys to include.
1374
:param ordering: Either 'unordered' or 'topological'. A topologically
1375
sorted stream has compression parents strictly before their
1377
:param include_delta_closure: If True then the closure across any
1378
compression parents will be included (in the opaque data).
1379
:return: An iterator of ContentFactory objects, each of which is only
1380
valid until the iterator is advanced.
1382
# keys might be a generator
1386
if not self._index.has_graph:
1387
# Cannot sort when no graph has been stored.
1388
ordering = 'unordered'
1390
remaining_keys = keys
1393
keys = set(remaining_keys)
1394
for content_factory in self._get_remaining_record_stream(keys,
1395
ordering, include_delta_closure):
1396
remaining_keys.discard(content_factory.key)
1397
yield content_factory
1399
except errors.RetryWithNewPacks, e:
1400
self._access.reload_or_raise(e)
1402
def _get_remaining_record_stream(self, keys, ordering,
1403
include_delta_closure):
1404
"""This function is the 'retry' portion for get_record_stream."""
1405
if include_delta_closure:
1406
positions = self._get_components_positions(keys, allow_missing=True)
1408
build_details = self._index.get_build_details(keys)
1410
# (record_details, access_memo, compression_parent_key)
1411
positions = dict((key, self._build_details_to_components(details))
1412
for key, details in build_details.iteritems())
1413
absent_keys = keys.difference(set(positions))
1414
# There may be more absent keys : if we're missing the basis component
1415
# and are trying to include the delta closure.
1416
# XXX: We should not ever need to examine remote sources because we do
1417
# not permit deltas across versioned files boundaries.
1418
if include_delta_closure:
1419
needed_from_fallback = set()
1420
# Build up reconstructable_keys dict. key:True in this dict means
1421
# the key can be reconstructed.
1422
reconstructable_keys = {}
1426
chain = [key, positions[key][2]]
1428
needed_from_fallback.add(key)
1431
while chain[-1] is not None:
1432
if chain[-1] in reconstructable_keys:
1433
result = reconstructable_keys[chain[-1]]
1437
chain.append(positions[chain[-1]][2])
1439
# missing basis component
1440
needed_from_fallback.add(chain[-1])
1443
for chain_key in chain[:-1]:
1444
reconstructable_keys[chain_key] = result
1446
needed_from_fallback.add(key)
1447
# Double index lookups here : need a unified api ?
1448
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1449
if ordering in ('topological', 'groupcompress'):
1450
if ordering == 'topological':
1451
# Global topological sort
1452
present_keys = tsort.topo_sort(global_map)
1454
present_keys = sort_groupcompress(global_map)
1455
# Now group by source:
1457
current_source = None
1458
for key in present_keys:
1459
for parent_map in parent_maps:
1460
if key in parent_map:
1461
key_source = parent_map
1463
if current_source is not key_source:
1464
source_keys.append((key_source, []))
1465
current_source = key_source
1466
source_keys[-1][1].append(key)
1468
if ordering != 'unordered':
1469
raise AssertionError('valid values for ordering are:'
1470
' "unordered", "groupcompress" or "topological" not: %r'
1472
# Just group by source; remote sources first.
1475
for parent_map in reversed(parent_maps):
1476
source_keys.append((parent_map, []))
1477
for key in parent_map:
1478
present_keys.append(key)
1479
source_keys[-1][1].append(key)
1480
# We have been requested to return these records in an order that
1481
# suits us. So we ask the index to give us an optimally sorted
1483
for source, sub_keys in source_keys:
1484
if source is parent_maps[0]:
1485
# Only sort the keys for this VF
1486
self._index._sort_keys_by_io(sub_keys, positions)
1487
absent_keys = keys - set(global_map)
1488
for key in absent_keys:
1489
yield AbsentContentFactory(key)
1490
# restrict our view to the keys we can answer.
1491
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1492
# XXX: At that point we need to consider the impact of double reads by
1493
# utilising components multiple times.
1494
if include_delta_closure:
1495
# XXX: get_content_maps performs its own index queries; allow state
1497
non_local_keys = needed_from_fallback - absent_keys
1498
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1501
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1504
for record in generator.get_record_stream():
1507
for source, keys in source_keys:
1508
if source is parent_maps[0]:
1509
# this KnitVersionedFiles
1510
records = [(key, positions[key][1]) for key in keys]
1511
for key, raw_data in self._read_records_iter_unchecked(records):
1512
(record_details, index_memo, _) = positions[key]
1513
yield KnitContentFactory(key, global_map[key],
1514
record_details, None, raw_data, self._factory.annotated, None)
1516
vf = self._immediate_fallback_vfs[parent_maps.index(source) - 1]
1517
for record in vf.get_record_stream(keys, ordering,
1518
include_delta_closure):
1521
def get_sha1s(self, keys):
1522
"""See VersionedFiles.get_sha1s()."""
1524
record_map = self._get_record_map(missing, allow_missing=True)
1526
for key, details in record_map.iteritems():
1527
if key not in missing:
1529
# record entry 2 is the 'digest'.
1530
result[key] = details[2]
1531
missing.difference_update(set(result))
1532
for source in self._immediate_fallback_vfs:
1535
new_result = source.get_sha1s(missing)
1536
result.update(new_result)
1537
missing.difference_update(set(new_result))
1540
def insert_record_stream(self, stream):
1541
"""Insert a record stream into this container.
1543
:param stream: A stream of records to insert.
1545
:seealso VersionedFiles.get_record_stream:
1547
def get_adapter(adapter_key):
1549
return adapters[adapter_key]
1551
adapter_factory = adapter_registry.get(adapter_key)
1552
adapter = adapter_factory(self)
1553
adapters[adapter_key] = adapter
1556
if self._factory.annotated:
1557
# self is annotated, we need annotated knits to use directly.
1558
annotated = "annotated-"
1561
# self is not annotated, but we can strip annotations cheaply.
1563
convertibles = set(["knit-annotated-ft-gz"])
1564
if self._max_delta_chain:
1565
delta_types.add("knit-annotated-delta-gz")
1566
convertibles.add("knit-annotated-delta-gz")
1567
# The set of types we can cheaply adapt without needing basis texts.
1568
native_types = set()
1569
if self._max_delta_chain:
1570
native_types.add("knit-%sdelta-gz" % annotated)
1571
delta_types.add("knit-%sdelta-gz" % annotated)
1572
native_types.add("knit-%sft-gz" % annotated)
1573
knit_types = native_types.union(convertibles)
1575
# Buffer all index entries that we can't add immediately because their
1576
# basis parent is missing. We don't buffer all because generating
1577
# annotations may require access to some of the new records. However we
1578
# can't generate annotations from new deltas until their basis parent
1579
# is present anyway, so we get away with not needing an index that
1580
# includes the new keys.
1582
# See <http://launchpad.net/bugs/300177> about ordering of compression
1583
# parents in the records - to be conservative, we insist that all
1584
# parents must be present to avoid expanding to a fulltext.
1586
# key = basis_parent, value = index entry to add
1587
buffered_index_entries = {}
1588
for record in stream:
1589
kind = record.storage_kind
1590
if kind.startswith('knit-') and kind.endswith('-gz'):
1591
# Check that the ID in the header of the raw knit bytes matches
1592
# the record metadata.
1593
raw_data = record._raw_record
1594
df, rec = self._parse_record_header(record.key, raw_data)
1597
parents = record.parents
1598
if record.storage_kind in delta_types:
1599
# TODO: eventually the record itself should track
1600
# compression_parent
1601
compression_parent = parents[0]
1603
compression_parent = None
1604
# Raise an error when a record is missing.
1605
if record.storage_kind == 'absent':
1606
raise RevisionNotPresent([record.key], self)
1607
elif ((record.storage_kind in knit_types)
1608
and (compression_parent is None
1609
or not self._immediate_fallback_vfs
1610
or self._index.has_key(compression_parent)
1611
or not self.has_key(compression_parent))):
1612
# we can insert the knit record literally if either it has no
1613
# compression parent OR we already have its basis in this kvf
1614
# OR the basis is not present even in the fallbacks. In the
1615
# last case it will either turn up later in the stream and all
1616
# will be well, or it won't turn up at all and we'll raise an
1619
# TODO: self.has_key is somewhat redundant with
1620
# self._index.has_key; we really want something that directly
1621
# asks if it's only present in the fallbacks. -- mbp 20081119
1622
if record.storage_kind not in native_types:
1624
adapter_key = (record.storage_kind, "knit-delta-gz")
1625
adapter = get_adapter(adapter_key)
1627
adapter_key = (record.storage_kind, "knit-ft-gz")
1628
adapter = get_adapter(adapter_key)
1629
bytes = adapter.get_bytes(record)
1631
# It's a knit record, it has a _raw_record field (even if
1632
# it was reconstituted from a network stream).
1633
bytes = record._raw_record
1634
options = [record._build_details[0]]
1635
if record._build_details[1]:
1636
options.append('no-eol')
1637
# Just blat it across.
1638
# Note: This does end up adding data on duplicate keys. As
1639
# modern repositories use atomic insertions this should not
1640
# lead to excessive growth in the event of interrupted fetches.
1641
# 'knit' repositories may suffer excessive growth, but as a
1642
# deprecated format this is tolerable. It can be fixed if
1643
# needed by in the kndx index support raising on a duplicate
1644
# add with identical parents and options.
1645
access_memo = self._access.add_raw_records(
1646
[(record.key, len(bytes))], bytes)[0]
1647
index_entry = (record.key, options, access_memo, parents)
1648
if 'fulltext' not in options:
1649
# Not a fulltext, so we need to make sure the compression
1650
# parent will also be present.
1651
# Note that pack backed knits don't need to buffer here
1652
# because they buffer all writes to the transaction level,
1653
# but we don't expose that difference at the index level. If
1654
# the query here has sufficient cost to show up in
1655
# profiling we should do that.
1657
# They're required to be physically in this
1658
# KnitVersionedFiles, not in a fallback.
1659
if not self._index.has_key(compression_parent):
1660
pending = buffered_index_entries.setdefault(
1661
compression_parent, [])
1662
pending.append(index_entry)
1665
self._index.add_records([index_entry])
1666
elif record.storage_kind == 'chunked':
1667
self.add_lines(record.key, parents,
1668
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1670
# Not suitable for direct insertion as a
1671
# delta, either because it's not the right format, or this
1672
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1673
# 0) or because it depends on a base only present in the
1675
self._access.flush()
1677
# Try getting a fulltext directly from the record.
1678
bytes = record.get_bytes_as('fulltext')
1679
except errors.UnavailableRepresentation:
1680
adapter_key = record.storage_kind, 'fulltext'
1681
adapter = get_adapter(adapter_key)
1682
bytes = adapter.get_bytes(record)
1683
lines = split_lines(bytes)
1685
self.add_lines(record.key, parents, lines)
1686
except errors.RevisionAlreadyPresent:
1688
# Add any records whose basis parent is now available.
1690
added_keys = [record.key]
1692
key = added_keys.pop(0)
1693
if key in buffered_index_entries:
1694
index_entries = buffered_index_entries[key]
1695
self._index.add_records(index_entries)
1697
[index_entry[0] for index_entry in index_entries])
1698
del buffered_index_entries[key]
1699
if buffered_index_entries:
1700
# There were index entries buffered at the end of the stream,
1701
# So these need to be added (if the index supports holding such
1702
# entries for later insertion)
1704
for key in buffered_index_entries:
1705
index_entries = buffered_index_entries[key]
1706
all_entries.extend(index_entries)
1707
self._index.add_records(
1708
all_entries, missing_compression_parents=True)
1710
def get_missing_compression_parent_keys(self):
1711
"""Return an iterable of keys of missing compression parents.
1713
Check this after calling insert_record_stream to find out if there are
1714
any missing compression parents. If there are, the records that
1715
depend on them are not able to be inserted safely. For atomic
1716
KnitVersionedFiles built on packs, the transaction should be aborted or
1717
suspended - commit will fail at this point. Nonatomic knits will error
1718
earlier because they have no staging area to put pending entries into.
1720
return self._index.get_missing_compression_parents()
1722
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1723
"""Iterate over the lines in the versioned files from keys.
1725
This may return lines from other keys. Each item the returned
1726
iterator yields is a tuple of a line and a text version that that line
1727
is present in (not introduced in).
1729
Ordering of results is in whatever order is most suitable for the
1730
underlying storage format.
1732
If a progress bar is supplied, it may be used to indicate progress.
1733
The caller is responsible for cleaning up progress bars (because this
1737
* Lines are normalised by the underlying store: they will all have \\n
1739
* Lines are returned in arbitrary order.
1740
* If a requested key did not change any lines (or didn't have any
1741
lines), it may not be mentioned at all in the result.
1743
:param pb: Progress bar supplied by caller.
1744
:return: An iterator over (line, key).
1747
pb = ui.ui_factory.nested_progress_bar()
1753
# we don't care about inclusions, the caller cares.
1754
# but we need to setup a list of records to visit.
1755
# we need key, position, length
1757
build_details = self._index.get_build_details(keys)
1758
for key, details in build_details.iteritems():
1760
key_records.append((key, details[0]))
1761
records_iter = enumerate(self._read_records_iter(key_records))
1762
for (key_idx, (key, data, sha_value)) in records_iter:
1763
pb.update('Walking content', key_idx, total)
1764
compression_parent = build_details[key][1]
1765
if compression_parent is None:
1767
line_iterator = self._factory.get_fulltext_content(data)
1770
line_iterator = self._factory.get_linedelta_content(data)
1771
# Now that we are yielding the data for this key, remove it
1774
# XXX: It might be more efficient to yield (key,
1775
# line_iterator) in the future. However for now, this is a
1776
# simpler change to integrate into the rest of the
1777
# codebase. RBC 20071110
1778
for line in line_iterator:
1781
except errors.RetryWithNewPacks, e:
1782
self._access.reload_or_raise(e)
1783
# If there are still keys we've not yet found, we look in the fallback
1784
# vfs, and hope to find them there. Note that if the keys are found
1785
# but had no changes or no content, the fallback may not return
1787
if keys and not self._immediate_fallback_vfs:
1788
# XXX: strictly the second parameter is meant to be the file id
1789
# but it's not easily accessible here.
1790
raise RevisionNotPresent(keys, repr(self))
1791
for source in self._immediate_fallback_vfs:
1795
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1796
source_keys.add(key)
1798
keys.difference_update(source_keys)
1799
pb.update('Walking content', total, total)
1801
def _make_line_delta(self, delta_seq, new_content):
1802
"""Generate a line delta from delta_seq and new_content."""
1804
for op in delta_seq.get_opcodes():
1805
if op[0] == 'equal':
1807
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1810
def _merge_annotations(self, content, parents, parent_texts={},
1811
delta=None, annotated=None,
1812
left_matching_blocks=None):
1813
"""Merge annotations for content and generate deltas.
1815
This is done by comparing the annotations based on changes to the text
1816
and generating a delta on the resulting full texts. If annotations are
1817
not being created then a simple delta is created.
1819
if left_matching_blocks is not None:
1820
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1824
for parent_key in parents:
1825
merge_content = self._get_content(parent_key, parent_texts)
1826
if (parent_key == parents[0] and delta_seq is not None):
1829
seq = patiencediff.PatienceSequenceMatcher(
1830
None, merge_content.text(), content.text())
1831
for i, j, n in seq.get_matching_blocks():
1834
# this copies (origin, text) pairs across to the new
1835
# content for any line that matches the last-checked
1837
content._lines[j:j+n] = merge_content._lines[i:i+n]
1838
# XXX: Robert says the following block is a workaround for a
1839
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1840
if content._lines and content._lines[-1][1][-1] != '\n':
1841
# The copied annotation was from a line without a trailing EOL,
1842
# reinstate one for the content object, to ensure correct
1844
line = content._lines[-1][1] + '\n'
1845
content._lines[-1] = (content._lines[-1][0], line)
1847
if delta_seq is None:
1848
reference_content = self._get_content(parents[0], parent_texts)
1849
new_texts = content.text()
1850
old_texts = reference_content.text()
1851
delta_seq = patiencediff.PatienceSequenceMatcher(
1852
None, old_texts, new_texts)
1853
return self._make_line_delta(delta_seq, content)
1855
def _parse_record(self, version_id, data):
1856
"""Parse an original format knit record.
1858
These have the last element of the key only present in the stored data.
1860
rec, record_contents = self._parse_record_unchecked(data)
1861
self._check_header_version(rec, version_id)
1862
return record_contents, rec[3]
1864
def _parse_record_header(self, key, raw_data):
1865
"""Parse a record header for consistency.
1867
:return: the header and the decompressor stream.
1868
as (stream, header_record)
1870
df = gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1873
rec = self._check_header(key, df.readline())
1874
except Exception, e:
1875
raise KnitCorrupt(self,
1876
"While reading {%s} got %s(%s)"
1877
% (key, e.__class__.__name__, str(e)))
1880
def _parse_record_unchecked(self, data):
1882
# 4168 calls in 2880 217 internal
1883
# 4168 calls to _parse_record_header in 2121
1884
# 4168 calls to readlines in 330
1885
df = gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1887
record_contents = df.readlines()
1888
except Exception, e:
1889
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1890
(data, e.__class__.__name__, str(e)))
1891
header = record_contents.pop(0)
1892
rec = self._split_header(header)
1893
last_line = record_contents.pop()
1894
if len(record_contents) != int(rec[2]):
1895
raise KnitCorrupt(self,
1896
'incorrect number of lines %s != %s'
1897
' for version {%s} %s'
1898
% (len(record_contents), int(rec[2]),
1899
rec[1], record_contents))
1900
if last_line != 'end %s\n' % rec[1]:
1901
raise KnitCorrupt(self,
1902
'unexpected version end line %r, wanted %r'
1903
% (last_line, rec[1]))
1905
return rec, record_contents
1907
def _read_records_iter(self, records):
1908
"""Read text records from data file and yield result.
1910
The result will be returned in whatever is the fastest to read.
1911
Not by the order requested. Also, multiple requests for the same
1912
record will only yield 1 response.
1914
:param records: A list of (key, access_memo) entries
1915
:return: Yields (key, contents, digest) in the order
1916
read, not the order requested
1921
# XXX: This smells wrong, IO may not be getting ordered right.
1922
needed_records = sorted(set(records), key=operator.itemgetter(1))
1923
if not needed_records:
1926
# The transport optimizes the fetching as well
1927
# (ie, reads continuous ranges.)
1928
raw_data = self._access.get_raw_records(
1929
[index_memo for key, index_memo in needed_records])
1931
for (key, index_memo), data in \
1932
izip(iter(needed_records), raw_data):
1933
content, digest = self._parse_record(key[-1], data)
1934
yield key, content, digest
1936
def _read_records_iter_raw(self, records):
1937
"""Read text records from data file and yield raw data.
1939
This unpacks enough of the text record to validate the id is
1940
as expected but thats all.
1942
Each item the iterator yields is (key, bytes,
1943
expected_sha1_of_full_text).
1945
for key, data in self._read_records_iter_unchecked(records):
1946
# validate the header (note that we can only use the suffix in
1947
# current knit records).
1948
df, rec = self._parse_record_header(key, data)
1950
yield key, data, rec[3]
1952
def _read_records_iter_unchecked(self, records):
1953
"""Read text records from data file and yield raw data.
1955
No validation is done.
1957
Yields tuples of (key, data).
1959
# setup an iterator of the external records:
1960
# uses readv so nice and fast we hope.
1962
# grab the disk data needed.
1963
needed_offsets = [index_memo for key, index_memo
1965
raw_records = self._access.get_raw_records(needed_offsets)
1967
for key, index_memo in records:
1968
data = raw_records.next()
1971
def _record_to_data(self, key, digest, lines, dense_lines=None):
1972
"""Convert key, digest, lines into a raw data block.
1974
:param key: The key of the record. Currently keys are always serialised
1975
using just the trailing component.
1976
:param dense_lines: The bytes of lines but in a denser form. For
1977
instance, if lines is a list of 1000 bytestrings each ending in
1978
\\n, dense_lines may be a list with one line in it, containing all
1979
the 1000's lines and their \\n's. Using dense_lines if it is
1980
already known is a win because the string join to create bytes in
1981
this function spends less time resizing the final string.
1982
:return: (len, a StringIO instance with the raw data ready to read.)
1984
chunks = ["version %s %d %s\n" % (key[-1], len(lines), digest)]
1985
chunks.extend(dense_lines or lines)
1986
chunks.append("end %s\n" % key[-1])
1987
for chunk in chunks:
1988
if type(chunk) is not str:
1989
raise AssertionError(
1990
'data must be plain bytes was %s' % type(chunk))
1991
if lines and lines[-1][-1] != '\n':
1992
raise ValueError('corrupt lines value %r' % lines)
1993
compressed_bytes = tuned_gzip.chunks_to_gzip(chunks)
1994
return len(compressed_bytes), compressed_bytes
1996
def _split_header(self, line):
1999
raise KnitCorrupt(self,
2000
'unexpected number of elements in record header')
2004
"""See VersionedFiles.keys."""
2005
if 'evil' in debug.debug_flags:
2006
trace.mutter_callsite(2, "keys scales with size of history")
2007
sources = [self._index] + self._immediate_fallback_vfs
2009
for source in sources:
2010
result.update(source.keys())
2014
class _ContentMapGenerator(object):
2015
"""Generate texts or expose raw deltas for a set of texts."""
2017
def __init__(self, ordering='unordered'):
2018
self._ordering = ordering
2020
def _get_content(self, key):
2021
"""Get the content object for key."""
2022
# Note that _get_content is only called when the _ContentMapGenerator
2023
# has been constructed with just one key requested for reconstruction.
2024
if key in self.nonlocal_keys:
2025
record = self.get_record_stream().next()
2026
# Create a content object on the fly
2027
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
2028
return PlainKnitContent(lines, record.key)
2030
# local keys we can ask for directly
2031
return self._get_one_work(key)
2033
def get_record_stream(self):
2034
"""Get a record stream for the keys requested during __init__."""
2035
for record in self._work():
2039
"""Produce maps of text and KnitContents as dicts.
2041
:return: (text_map, content_map) where text_map contains the texts for
2042
the requested versions and content_map contains the KnitContents.
2044
# NB: By definition we never need to read remote sources unless texts
2045
# are requested from them: we don't delta across stores - and we
2046
# explicitly do not want to to prevent data loss situations.
2047
if self.global_map is None:
2048
self.global_map = self.vf.get_parent_map(self.keys)
2049
nonlocal_keys = self.nonlocal_keys
2051
missing_keys = set(nonlocal_keys)
2052
# Read from remote versioned file instances and provide to our caller.
2053
for source in self.vf._immediate_fallback_vfs:
2054
if not missing_keys:
2056
# Loop over fallback repositories asking them for texts - ignore
2057
# any missing from a particular fallback.
2058
for record in source.get_record_stream(missing_keys,
2059
self._ordering, True):
2060
if record.storage_kind == 'absent':
2061
# Not in thie particular stream, may be in one of the
2062
# other fallback vfs objects.
2064
missing_keys.remove(record.key)
2067
if self._raw_record_map is None:
2068
raise AssertionError('_raw_record_map should have been filled')
2070
for key in self.keys:
2071
if key in self.nonlocal_keys:
2073
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2076
def _get_one_work(self, requested_key):
2077
# Now, if we have calculated everything already, just return the
2079
if requested_key in self._contents_map:
2080
return self._contents_map[requested_key]
2081
# To simplify things, parse everything at once - code that wants one text
2082
# probably wants them all.
2083
# FUTURE: This function could be improved for the 'extract many' case
2084
# by tracking each component and only doing the copy when the number of
2085
# children than need to apply delta's to it is > 1 or it is part of the
2087
multiple_versions = len(self.keys) != 1
2088
if self._record_map is None:
2089
self._record_map = self.vf._raw_map_to_record_map(
2090
self._raw_record_map)
2091
record_map = self._record_map
2092
# raw_record_map is key:
2093
# Have read and parsed records at this point.
2094
for key in self.keys:
2095
if key in self.nonlocal_keys:
2100
while cursor is not None:
2102
record, record_details, digest, next = record_map[cursor]
2104
raise RevisionNotPresent(cursor, self)
2105
components.append((cursor, record, record_details, digest))
2107
if cursor in self._contents_map:
2108
# no need to plan further back
2109
components.append((cursor, None, None, None))
2113
for (component_id, record, record_details,
2114
digest) in reversed(components):
2115
if component_id in self._contents_map:
2116
content = self._contents_map[component_id]
2118
content, delta = self._factory.parse_record(key[-1],
2119
record, record_details, content,
2120
copy_base_content=multiple_versions)
2121
if multiple_versions:
2122
self._contents_map[component_id] = content
2124
# digest here is the digest from the last applied component.
2125
text = content.text()
2126
actual_sha = sha_strings(text)
2127
if actual_sha != digest:
2128
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2129
if multiple_versions:
2130
return self._contents_map[requested_key]
2134
def _wire_bytes(self):
2135
"""Get the bytes to put on the wire for 'key'.
2137
The first collection of bytes asked for returns the serialised
2138
raw_record_map and the additional details (key, parent) for key.
2139
Subsequent calls return just the additional details (key, parent).
2140
The wire storage_kind given for the first key is 'knit-delta-closure',
2141
For subsequent keys it is 'knit-delta-closure-ref'.
2143
:param key: A key from the content generator.
2144
:return: Bytes to put on the wire.
2147
# kind marker for dispatch on the far side,
2148
lines.append('knit-delta-closure')
2150
if self.vf._factory.annotated:
2151
lines.append('annotated')
2154
# then the list of keys
2155
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2156
if key not in self.nonlocal_keys]))
2157
# then the _raw_record_map in serialised form:
2159
# for each item in the map:
2161
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2162
# one line with method
2163
# one line with noeol
2164
# one line with next ('' for None)
2165
# one line with byte count of the record bytes
2167
for key, (record_bytes, (method, noeol), next) in \
2168
self._raw_record_map.iteritems():
2169
key_bytes = '\x00'.join(key)
2170
parents = self.global_map.get(key, None)
2172
parent_bytes = 'None:'
2174
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2175
method_bytes = method
2181
next_bytes = '\x00'.join(next)
2184
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2185
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2186
len(record_bytes), record_bytes))
2187
map_bytes = ''.join(map_byte_list)
2188
lines.append(map_bytes)
2189
bytes = '\n'.join(lines)
2193
class _VFContentMapGenerator(_ContentMapGenerator):
2194
"""Content map generator reading from a VersionedFiles object."""
2196
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2197
global_map=None, raw_record_map=None, ordering='unordered'):
2198
"""Create a _ContentMapGenerator.
2200
:param versioned_files: The versioned files that the texts are being
2202
:param keys: The keys to produce content maps for.
2203
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2204
which are known to not be in this knit, but rather in one of the
2206
:param global_map: The result of get_parent_map(keys) (or a supermap).
2207
This is required if get_record_stream() is to be used.
2208
:param raw_record_map: A unparsed raw record map to use for answering
2211
_ContentMapGenerator.__init__(self, ordering=ordering)
2212
# The vf to source data from
2213
self.vf = versioned_files
2215
self.keys = list(keys)
2216
# Keys known to be in fallback vfs objects
2217
if nonlocal_keys is None:
2218
self.nonlocal_keys = set()
2220
self.nonlocal_keys = frozenset(nonlocal_keys)
2221
# Parents data for keys to be returned in get_record_stream
2222
self.global_map = global_map
2223
# The chunked lists for self.keys in text form
2225
# A cache of KnitContent objects used in extracting texts.
2226
self._contents_map = {}
2227
# All the knit records needed to assemble the requested keys as full
2229
self._record_map = None
2230
if raw_record_map is None:
2231
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2234
self._raw_record_map = raw_record_map
2235
# the factory for parsing records
2236
self._factory = self.vf._factory
2239
class _NetworkContentMapGenerator(_ContentMapGenerator):
2240
"""Content map generator sourced from a network stream."""
2242
def __init__(self, bytes, line_end):
2243
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2245
self.global_map = {}
2246
self._raw_record_map = {}
2247
self._contents_map = {}
2248
self._record_map = None
2249
self.nonlocal_keys = []
2250
# Get access to record parsing facilities
2251
self.vf = KnitVersionedFiles(None, None)
2254
line_end = bytes.find('\n', start)
2255
line = bytes[start:line_end]
2256
start = line_end + 1
2257
if line == 'annotated':
2258
self._factory = KnitAnnotateFactory()
2260
self._factory = KnitPlainFactory()
2261
# list of keys to emit in get_record_stream
2262
line_end = bytes.find('\n', start)
2263
line = bytes[start:line_end]
2264
start = line_end + 1
2266
tuple(segment.split('\x00')) for segment in line.split('\t')
2268
# now a loop until the end. XXX: It would be nice if this was just a
2269
# bunch of the same records as get_record_stream(..., False) gives, but
2270
# there is a decent sized gap stopping that at the moment.
2274
line_end = bytes.find('\n', start)
2275
key = tuple(bytes[start:line_end].split('\x00'))
2276
start = line_end + 1
2277
# 1 line with parents (None: for None, '' for ())
2278
line_end = bytes.find('\n', start)
2279
line = bytes[start:line_end]
2284
[tuple(segment.split('\x00')) for segment in line.split('\t')
2286
self.global_map[key] = parents
2287
start = line_end + 1
2288
# one line with method
2289
line_end = bytes.find('\n', start)
2290
line = bytes[start:line_end]
2292
start = line_end + 1
2293
# one line with noeol
2294
line_end = bytes.find('\n', start)
2295
line = bytes[start:line_end]
2297
start = line_end + 1
2298
# one line with next ('' for None)
2299
line_end = bytes.find('\n', start)
2300
line = bytes[start:line_end]
2304
next = tuple(bytes[start:line_end].split('\x00'))
2305
start = line_end + 1
2306
# one line with byte count of the record bytes
2307
line_end = bytes.find('\n', start)
2308
line = bytes[start:line_end]
2310
start = line_end + 1
2312
record_bytes = bytes[start:start+count]
2313
start = start + count
2315
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2317
def get_record_stream(self):
2318
"""Get a record stream for for keys requested by the bytestream."""
2320
for key in self.keys:
2321
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2324
def _wire_bytes(self):
2328
class _KndxIndex(object):
2329
"""Manages knit index files
2331
The index is kept in memory and read on startup, to enable
2332
fast lookups of revision information. The cursor of the index
2333
file is always pointing to the end, making it easy to append
2336
_cache is a cache for fast mapping from version id to a Index
2339
_history is a cache for fast mapping from indexes to version ids.
2341
The index data format is dictionary compressed when it comes to
2342
parent references; a index entry may only have parents that with a
2343
lover index number. As a result, the index is topological sorted.
2345
Duplicate entries may be written to the index for a single version id
2346
if this is done then the latter one completely replaces the former:
2347
this allows updates to correct version and parent information.
2348
Note that the two entries may share the delta, and that successive
2349
annotations and references MUST point to the first entry.
2351
The index file on disc contains a header, followed by one line per knit
2352
record. The same revision can be present in an index file more than once.
2353
The first occurrence gets assigned a sequence number starting from 0.
2355
The format of a single line is
2356
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
2357
REVISION_ID is a utf8-encoded revision id
2358
FLAGS is a comma separated list of flags about the record. Values include
2359
no-eol, line-delta, fulltext.
2360
BYTE_OFFSET is the ascii representation of the byte offset in the data file
2361
that the compressed data starts at.
2362
LENGTH is the ascii representation of the length of the data file.
2363
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
2365
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
2366
revision id already in the knit that is a parent of REVISION_ID.
2367
The ' :' marker is the end of record marker.
2370
when a write is interrupted to the index file, it will result in a line
2371
that does not end in ' :'. If the ' :' is not present at the end of a line,
2372
or at the end of the file, then the record that is missing it will be
2373
ignored by the parser.
2375
When writing new records to the index file, the data is preceded by '\n'
2376
to ensure that records always start on new lines even if the last write was
2377
interrupted. As a result its normal for the last line in the index to be
2378
missing a trailing newline. One can be added with no harmful effects.
2380
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2381
where prefix is e.g. the (fileid,) for .texts instances or () for
2382
constant-mapped things like .revisions, and the old state is
2383
tuple(history_vector, cache_dict). This is used to prevent having an
2384
ABI change with the C extension that reads .kndx files.
2387
HEADER = "# bzr knit index 8\n"
2389
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2390
"""Create a _KndxIndex on transport using mapper."""
2391
self._transport = transport
2392
self._mapper = mapper
2393
self._get_scope = get_scope
2394
self._allow_writes = allow_writes
2395
self._is_locked = is_locked
2397
self.has_graph = True
2399
def add_records(self, records, random_id=False, missing_compression_parents=False):
2400
"""Add multiple records to the index.
2402
:param records: a list of tuples:
2403
(key, options, access_memo, parents).
2404
:param random_id: If True the ids being added were randomly generated
2405
and no check for existence will be performed.
2406
:param missing_compression_parents: If True the records being added are
2407
only compressed against texts already in the index (or inside
2408
records). If False the records all refer to unavailable texts (or
2409
texts inside records) as compression parents.
2411
if missing_compression_parents:
2412
# It might be nice to get the edge of the records. But keys isn't
2414
keys = sorted(record[0] for record in records)
2415
raise errors.RevisionNotPresent(keys, self)
2417
for record in records:
2420
path = self._mapper.map(key) + '.kndx'
2421
path_keys = paths.setdefault(path, (prefix, []))
2422
path_keys[1].append(record)
2423
for path in sorted(paths):
2424
prefix, path_keys = paths[path]
2425
self._load_prefixes([prefix])
2427
orig_history = self._kndx_cache[prefix][1][:]
2428
orig_cache = self._kndx_cache[prefix][0].copy()
2431
for key, options, (_, pos, size), parents in path_keys:
2433
# kndx indices cannot be parentless.
2435
line = "\n%s %s %s %s %s :" % (
2436
key[-1], ','.join(options), pos, size,
2437
self._dictionary_compress(parents))
2438
if type(line) is not str:
2439
raise AssertionError(
2440
'data must be utf8 was %s' % type(line))
2442
self._cache_key(key, options, pos, size, parents)
2443
if len(orig_history):
2444
self._transport.append_bytes(path, ''.join(lines))
2446
self._init_index(path, lines)
2448
# If any problems happen, restore the original values and re-raise
2449
self._kndx_cache[prefix] = (orig_cache, orig_history)
2452
def scan_unvalidated_index(self, graph_index):
2453
"""See _KnitGraphIndex.scan_unvalidated_index."""
2454
# Because kndx files do not support atomic insertion via separate index
2455
# files, they do not support this method.
2456
raise NotImplementedError(self.scan_unvalidated_index)
2458
def get_missing_compression_parents(self):
2459
"""See _KnitGraphIndex.get_missing_compression_parents."""
2460
# Because kndx files do not support atomic insertion via separate index
2461
# files, they do not support this method.
2462
raise NotImplementedError(self.get_missing_compression_parents)
2464
def _cache_key(self, key, options, pos, size, parent_keys):
2465
"""Cache a version record in the history array and index cache.
2467
This is inlined into _load_data for performance. KEEP IN SYNC.
2468
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2472
version_id = key[-1]
2473
# last-element only for compatibilty with the C load_data.
2474
parents = tuple(parent[-1] for parent in parent_keys)
2475
for parent in parent_keys:
2476
if parent[:-1] != prefix:
2477
raise ValueError("mismatched prefixes for %r, %r" % (
2479
cache, history = self._kndx_cache[prefix]
2480
# only want the _history index to reference the 1st index entry
2482
if version_id not in cache:
2483
index = len(history)
2484
history.append(version_id)
2486
index = cache[version_id][5]
2487
cache[version_id] = (version_id,
2494
def check_header(self, fp):
2495
line = fp.readline()
2497
# An empty file can actually be treated as though the file doesn't
2499
raise errors.NoSuchFile(self)
2500
if line != self.HEADER:
2501
raise KnitHeaderError(badline=line, filename=self)
2503
def _check_read(self):
2504
if not self._is_locked():
2505
raise errors.ObjectNotLocked(self)
2506
if self._get_scope() != self._scope:
2509
def _check_write_ok(self):
2510
"""Assert if not writes are permitted."""
2511
if not self._is_locked():
2512
raise errors.ObjectNotLocked(self)
2513
if self._get_scope() != self._scope:
2515
if self._mode != 'w':
2516
raise errors.ReadOnlyObjectDirtiedError(self)
2518
def get_build_details(self, keys):
2519
"""Get the method, index_memo and compression parent for keys.
2521
Ghosts are omitted from the result.
2523
:param keys: An iterable of keys.
2524
:return: A dict of key:(index_memo, compression_parent, parents,
2527
opaque structure to pass to read_records to extract the raw
2530
Content that this record is built upon, may be None
2532
Logical parents of this node
2534
extra information about the content which needs to be passed to
2535
Factory.parse_record
2537
parent_map = self.get_parent_map(keys)
2540
if key not in parent_map:
2542
method = self.get_method(key)
2543
parents = parent_map[key]
2544
if method == 'fulltext':
2545
compression_parent = None
2547
compression_parent = parents[0]
2548
noeol = 'no-eol' in self.get_options(key)
2549
index_memo = self.get_position(key)
2550
result[key] = (index_memo, compression_parent,
2551
parents, (method, noeol))
2554
def get_method(self, key):
2555
"""Return compression method of specified key."""
2556
options = self.get_options(key)
2557
if 'fulltext' in options:
2559
elif 'line-delta' in options:
2562
raise errors.KnitIndexUnknownMethod(self, options)
2564
def get_options(self, key):
2565
"""Return a list representing options.
2569
prefix, suffix = self._split_key(key)
2570
self._load_prefixes([prefix])
2572
return self._kndx_cache[prefix][0][suffix][1]
2574
raise RevisionNotPresent(key, self)
2576
def find_ancestry(self, keys):
2577
"""See CombinedGraphIndex.find_ancestry()"""
2578
prefixes = set(key[:-1] for key in keys)
2579
self._load_prefixes(prefixes)
2582
missing_keys = set()
2583
pending_keys = list(keys)
2584
# This assumes that keys will not reference parents in a different
2585
# prefix, which is accurate so far.
2587
key = pending_keys.pop()
2588
if key in parent_map:
2592
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2594
missing_keys.add(key)
2596
parent_keys = tuple([prefix + (suffix,)
2597
for suffix in suffix_parents])
2598
parent_map[key] = parent_keys
2599
pending_keys.extend([p for p in parent_keys
2600
if p not in parent_map])
2601
return parent_map, missing_keys
2603
def get_parent_map(self, keys):
2604
"""Get a map of the parents of keys.
2606
:param keys: The keys to look up parents for.
2607
:return: A mapping from keys to parents. Absent keys are absent from
2610
# Parse what we need to up front, this potentially trades off I/O
2611
# locality (.kndx and .knit in the same block group for the same file
2612
# id) for less checking in inner loops.
2613
prefixes = set(key[:-1] for key in keys)
2614
self._load_prefixes(prefixes)
2619
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2623
result[key] = tuple(prefix + (suffix,) for
2624
suffix in suffix_parents)
2627
def get_position(self, key):
2628
"""Return details needed to access the version.
2630
:return: a tuple (key, data position, size) to hand to the access
2631
logic to get the record.
2633
prefix, suffix = self._split_key(key)
2634
self._load_prefixes([prefix])
2635
entry = self._kndx_cache[prefix][0][suffix]
2636
return key, entry[2], entry[3]
2638
has_key = _mod_index._has_key_from_parent_map
2640
def _init_index(self, path, extra_lines=[]):
2641
"""Initialize an index."""
2643
sio.write(self.HEADER)
2644
sio.writelines(extra_lines)
2646
self._transport.put_file_non_atomic(path, sio,
2647
create_parent_dir=True)
2648
# self._create_parent_dir)
2649
# mode=self._file_mode,
2650
# dir_mode=self._dir_mode)
2653
"""Get all the keys in the collection.
2655
The keys are not ordered.
2658
# Identify all key prefixes.
2659
# XXX: A bit hacky, needs polish.
2660
if type(self._mapper) is ConstantMapper:
2664
for quoted_relpath in self._transport.iter_files_recursive():
2665
path, ext = os.path.splitext(quoted_relpath)
2667
prefixes = [self._mapper.unmap(path) for path in relpaths]
2668
self._load_prefixes(prefixes)
2669
for prefix in prefixes:
2670
for suffix in self._kndx_cache[prefix][1]:
2671
result.add(prefix + (suffix,))
2674
def _load_prefixes(self, prefixes):
2675
"""Load the indices for prefixes."""
2677
for prefix in prefixes:
2678
if prefix not in self._kndx_cache:
2679
# the load_data interface writes to these variables.
2682
self._filename = prefix
2684
path = self._mapper.map(prefix) + '.kndx'
2685
fp = self._transport.get(path)
2687
# _load_data may raise NoSuchFile if the target knit is
2689
_load_data(self, fp)
2692
self._kndx_cache[prefix] = (self._cache, self._history)
2697
self._kndx_cache[prefix] = ({}, [])
2698
if type(self._mapper) is ConstantMapper:
2699
# preserve behaviour for revisions.kndx etc.
2700
self._init_index(path)
2705
missing_keys = _mod_index._missing_keys_from_parent_map
2707
def _partition_keys(self, keys):
2708
"""Turn keys into a dict of prefix:suffix_list."""
2711
prefix_keys = result.setdefault(key[:-1], [])
2712
prefix_keys.append(key[-1])
2715
def _dictionary_compress(self, keys):
2716
"""Dictionary compress keys.
2718
:param keys: The keys to generate references to.
2719
:return: A string representation of keys. keys which are present are
2720
dictionary compressed, and others are emitted as fulltext with a
2726
prefix = keys[0][:-1]
2727
cache = self._kndx_cache[prefix][0]
2729
if key[:-1] != prefix:
2730
# kndx indices cannot refer across partitioned storage.
2731
raise ValueError("mismatched prefixes for %r" % keys)
2732
if key[-1] in cache:
2733
# -- inlined lookup() --
2734
result_list.append(str(cache[key[-1]][5]))
2735
# -- end lookup () --
2737
result_list.append('.' + key[-1])
2738
return ' '.join(result_list)
2740
def _reset_cache(self):
2741
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2742
# (cache_dict, history_vector) for parsed kndx files.
2743
self._kndx_cache = {}
2744
self._scope = self._get_scope()
2745
allow_writes = self._allow_writes()
2751
def _sort_keys_by_io(self, keys, positions):
2752
"""Figure out an optimal order to read the records for the given keys.
2754
Sort keys, grouped by index and sorted by position.
2756
:param keys: A list of keys whose records we want to read. This will be
2758
:param positions: A dict, such as the one returned by
2759
_get_components_positions()
2762
def get_sort_key(key):
2763
index_memo = positions[key][1]
2764
# Group by prefix and position. index_memo[0] is the key, so it is
2765
# (file_id, revision_id) and we don't want to sort on revision_id,
2766
# index_memo[1] is the position, and index_memo[2] is the size,
2767
# which doesn't matter for the sort
2768
return index_memo[0][:-1], index_memo[1]
2769
return keys.sort(key=get_sort_key)
2771
_get_total_build_size = _get_total_build_size
2773
def _split_key(self, key):
2774
"""Split key into a prefix and suffix."""
2775
return key[:-1], key[-1]
2778
class _KnitGraphIndex(object):
2779
"""A KnitVersionedFiles index layered on GraphIndex."""
2781
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2782
add_callback=None, track_external_parent_refs=False):
2783
"""Construct a KnitGraphIndex on a graph_index.
2785
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2786
:param is_locked: A callback to check whether the object should answer
2788
:param deltas: Allow delta-compressed records.
2789
:param parents: If True, record knits parents, if not do not record
2791
:param add_callback: If not None, allow additions to the index and call
2792
this callback with a list of added GraphIndex nodes:
2793
[(node, value, node_refs), ...]
2794
:param is_locked: A callback, returns True if the index is locked and
2796
:param track_external_parent_refs: If True, record all external parent
2797
references parents from added records. These can be retrieved
2798
later by calling get_missing_parents().
2800
self._add_callback = add_callback
2801
self._graph_index = graph_index
2802
self._deltas = deltas
2803
self._parents = parents
2804
if deltas and not parents:
2805
# XXX: TODO: Delta tree and parent graph should be conceptually
2807
raise KnitCorrupt(self, "Cannot do delta compression without "
2809
self.has_graph = parents
2810
self._is_locked = is_locked
2811
self._missing_compression_parents = set()
2812
if track_external_parent_refs:
2813
self._key_dependencies = _KeyRefs()
2815
self._key_dependencies = None
2818
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2820
def add_records(self, records, random_id=False,
2821
missing_compression_parents=False):
2822
"""Add multiple records to the index.
2824
This function does not insert data into the Immutable GraphIndex
2825
backing the KnitGraphIndex, instead it prepares data for insertion by
2826
the caller and checks that it is safe to insert then calls
2827
self._add_callback with the prepared GraphIndex nodes.
2829
:param records: a list of tuples:
2830
(key, options, access_memo, parents).
2831
:param random_id: If True the ids being added were randomly generated
2832
and no check for existence will be performed.
2833
:param missing_compression_parents: If True the records being added are
2834
only compressed against texts already in the index (or inside
2835
records). If False the records all refer to unavailable texts (or
2836
texts inside records) as compression parents.
2838
if not self._add_callback:
2839
raise errors.ReadOnlyError(self)
2840
# we hope there are no repositories with inconsistent parentage
2844
compression_parents = set()
2845
key_dependencies = self._key_dependencies
2846
for (key, options, access_memo, parents) in records:
2848
parents = tuple(parents)
2849
if key_dependencies is not None:
2850
key_dependencies.add_references(key, parents)
2851
index, pos, size = access_memo
2852
if 'no-eol' in options:
2856
value += "%d %d" % (pos, size)
2857
if not self._deltas:
2858
if 'line-delta' in options:
2859
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2862
if 'line-delta' in options:
2863
node_refs = (parents, (parents[0],))
2864
if missing_compression_parents:
2865
compression_parents.add(parents[0])
2867
node_refs = (parents, ())
2869
node_refs = (parents, )
2872
raise KnitCorrupt(self, "attempt to add node with parents "
2873
"in parentless index.")
2875
keys[key] = (value, node_refs)
2878
present_nodes = self._get_entries(keys)
2879
for (index, key, value, node_refs) in present_nodes:
2880
parents = node_refs[:1]
2881
# Sometimes these are passed as a list rather than a tuple
2882
passed = static_tuple.as_tuples(keys[key])
2883
passed_parents = passed[1][:1]
2884
if (value[0] != keys[key][0][0] or
2885
parents != passed_parents):
2886
node_refs = static_tuple.as_tuples(node_refs)
2887
raise KnitCorrupt(self, "inconsistent details in add_records"
2888
": %s %s" % ((value, node_refs), passed))
2892
for key, (value, node_refs) in keys.iteritems():
2893
result.append((key, value, node_refs))
2895
for key, (value, node_refs) in keys.iteritems():
2896
result.append((key, value))
2897
self._add_callback(result)
2898
if missing_compression_parents:
2899
# This may appear to be incorrect (it does not check for
2900
# compression parents that are in the existing graph index),
2901
# but such records won't have been buffered, so this is
2902
# actually correct: every entry when
2903
# missing_compression_parents==True either has a missing parent, or
2904
# a parent that is one of the keys in records.
2905
compression_parents.difference_update(keys)
2906
self._missing_compression_parents.update(compression_parents)
2907
# Adding records may have satisfied missing compression parents.
2908
self._missing_compression_parents.difference_update(keys)
2910
def scan_unvalidated_index(self, graph_index):
2911
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2913
This allows this _KnitGraphIndex to keep track of any missing
2914
compression parents we may want to have filled in to make those
2917
:param graph_index: A GraphIndex
2920
new_missing = graph_index.external_references(ref_list_num=1)
2921
new_missing.difference_update(self.get_parent_map(new_missing))
2922
self._missing_compression_parents.update(new_missing)
2923
if self._key_dependencies is not None:
2924
# Add parent refs from graph_index (and discard parent refs that
2925
# the graph_index has).
2926
for node in graph_index.iter_all_entries():
2927
self._key_dependencies.add_references(node[1], node[3][0])
2929
def get_missing_compression_parents(self):
2930
"""Return the keys of missing compression parents.
2932
Missing compression parents occur when a record stream was missing
2933
basis texts, or a index was scanned that had missing basis texts.
2935
return frozenset(self._missing_compression_parents)
2937
def get_missing_parents(self):
2938
"""Return the keys of missing parents."""
2939
# If updating this, you should also update
2940
# groupcompress._GCGraphIndex.get_missing_parents
2941
# We may have false positives, so filter those out.
2942
self._key_dependencies.satisfy_refs_for_keys(
2943
self.get_parent_map(self._key_dependencies.get_unsatisfied_refs()))
2944
return frozenset(self._key_dependencies.get_unsatisfied_refs())
2946
def _check_read(self):
2947
"""raise if reads are not permitted."""
2948
if not self._is_locked():
2949
raise errors.ObjectNotLocked(self)
2951
def _check_write_ok(self):
2952
"""Assert if writes are not permitted."""
2953
if not self._is_locked():
2954
raise errors.ObjectNotLocked(self)
2956
def _compression_parent(self, an_entry):
2957
# return the key that an_entry is compressed against, or None
2958
# Grab the second parent list (as deltas implies parents currently)
2959
compression_parents = an_entry[3][1]
2960
if not compression_parents:
2962
if len(compression_parents) != 1:
2963
raise AssertionError(
2964
"Too many compression parents: %r" % compression_parents)
2965
return compression_parents[0]
2967
def get_build_details(self, keys):
2968
"""Get the method, index_memo and compression parent for version_ids.
2970
Ghosts are omitted from the result.
2972
:param keys: An iterable of keys.
2973
:return: A dict of key:
2974
(index_memo, compression_parent, parents, record_details).
2976
opaque structure to pass to read_records to extract the raw
2979
Content that this record is built upon, may be None
2981
Logical parents of this node
2983
extra information about the content which needs to be passed to
2984
Factory.parse_record
2988
entries = self._get_entries(keys, False)
2989
for entry in entries:
2991
if not self._parents:
2994
parents = entry[3][0]
2995
if not self._deltas:
2996
compression_parent_key = None
2998
compression_parent_key = self._compression_parent(entry)
2999
noeol = (entry[2][0] == 'N')
3000
if compression_parent_key:
3001
method = 'line-delta'
3004
result[key] = (self._node_to_position(entry),
3005
compression_parent_key, parents,
3009
def _get_entries(self, keys, check_present=False):
3010
"""Get the entries for keys.
3012
:param keys: An iterable of index key tuples.
3017
for node in self._graph_index.iter_entries(keys):
3019
found_keys.add(node[1])
3021
# adapt parentless index to the rest of the code.
3022
for node in self._graph_index.iter_entries(keys):
3023
yield node[0], node[1], node[2], ()
3024
found_keys.add(node[1])
3026
missing_keys = keys.difference(found_keys)
3028
raise RevisionNotPresent(missing_keys.pop(), self)
3030
def get_method(self, key):
3031
"""Return compression method of specified key."""
3032
return self._get_method(self._get_node(key))
3034
def _get_method(self, node):
3035
if not self._deltas:
3037
if self._compression_parent(node):
3042
def _get_node(self, key):
3044
return list(self._get_entries([key]))[0]
3046
raise RevisionNotPresent(key, self)
3048
def get_options(self, key):
3049
"""Return a list representing options.
3053
node = self._get_node(key)
3054
options = [self._get_method(node)]
3055
if node[2][0] == 'N':
3056
options.append('no-eol')
3059
def find_ancestry(self, keys):
3060
"""See CombinedGraphIndex.find_ancestry()"""
3061
return self._graph_index.find_ancestry(keys, 0)
3063
def get_parent_map(self, keys):
3064
"""Get a map of the parents of keys.
3066
:param keys: The keys to look up parents for.
3067
:return: A mapping from keys to parents. Absent keys are absent from
3071
nodes = self._get_entries(keys)
3075
result[node[1]] = node[3][0]
3078
result[node[1]] = None
3081
def get_position(self, key):
3082
"""Return details needed to access the version.
3084
:return: a tuple (index, data position, size) to hand to the access
3085
logic to get the record.
3087
node = self._get_node(key)
3088
return self._node_to_position(node)
3090
has_key = _mod_index._has_key_from_parent_map
3093
"""Get all the keys in the collection.
3095
The keys are not ordered.
3098
return [node[1] for node in self._graph_index.iter_all_entries()]
3100
missing_keys = _mod_index._missing_keys_from_parent_map
3102
def _node_to_position(self, node):
3103
"""Convert an index value to position details."""
3104
bits = node[2][1:].split(' ')
3105
return node[0], int(bits[0]), int(bits[1])
3107
def _sort_keys_by_io(self, keys, positions):
3108
"""Figure out an optimal order to read the records for the given keys.
3110
Sort keys, grouped by index and sorted by position.
3112
:param keys: A list of keys whose records we want to read. This will be
3114
:param positions: A dict, such as the one returned by
3115
_get_components_positions()
3118
def get_index_memo(key):
3119
# index_memo is at offset [1]. It is made up of (GraphIndex,
3120
# position, size). GI is an object, which will be unique for each
3121
# pack file. This causes us to group by pack file, then sort by
3122
# position. Size doesn't matter, but it isn't worth breaking up the
3124
return positions[key][1]
3125
return keys.sort(key=get_index_memo)
3127
_get_total_build_size = _get_total_build_size
3130
class _KnitKeyAccess(object):
3131
"""Access to records in .knit files."""
3133
def __init__(self, transport, mapper):
3134
"""Create a _KnitKeyAccess with transport and mapper.
3136
:param transport: The transport the access object is rooted at.
3137
:param mapper: The mapper used to map keys to .knit files.
3139
self._transport = transport
3140
self._mapper = mapper
3142
def add_raw_records(self, key_sizes, raw_data):
3143
"""Add raw knit bytes to a storage area.
3145
The data is spooled to the container writer in one bytes-record per
3148
:param sizes: An iterable of tuples containing the key and size of each
3150
:param raw_data: A bytestring containing the data.
3151
:return: A list of memos to retrieve the record later. Each memo is an
3152
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3153
length), where the key is the record key.
3155
if type(raw_data) is not str:
3156
raise AssertionError(
3157
'data must be plain bytes was %s' % type(raw_data))
3160
# TODO: This can be tuned for writing to sftp and other servers where
3161
# append() is relatively expensive by grouping the writes to each key
3163
for key, size in key_sizes:
3164
path = self._mapper.map(key)
3166
base = self._transport.append_bytes(path + '.knit',
3167
raw_data[offset:offset+size])
3168
except errors.NoSuchFile:
3169
self._transport.mkdir(osutils.dirname(path))
3170
base = self._transport.append_bytes(path + '.knit',
3171
raw_data[offset:offset+size])
3175
result.append((key, base, size))
3179
"""Flush pending writes on this access object.
3181
For .knit files this is a no-op.
3185
def get_raw_records(self, memos_for_retrieval):
3186
"""Get the raw bytes for a records.
3188
:param memos_for_retrieval: An iterable containing the access memo for
3189
retrieving the bytes.
3190
:return: An iterator over the bytes of the records.
3192
# first pass, group into same-index request to minimise readv's issued.
3194
current_prefix = None
3195
for (key, offset, length) in memos_for_retrieval:
3196
if current_prefix == key[:-1]:
3197
current_list.append((offset, length))
3199
if current_prefix is not None:
3200
request_lists.append((current_prefix, current_list))
3201
current_prefix = key[:-1]
3202
current_list = [(offset, length)]
3203
# handle the last entry
3204
if current_prefix is not None:
3205
request_lists.append((current_prefix, current_list))
3206
for prefix, read_vector in request_lists:
3207
path = self._mapper.map(prefix) + '.knit'
3208
for pos, data in self._transport.readv(path, read_vector):
3212
class _DirectPackAccess(object):
3213
"""Access to data in one or more packs with less translation."""
3215
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
3216
"""Create a _DirectPackAccess object.
3218
:param index_to_packs: A dict mapping index objects to the transport
3219
and file names for obtaining data.
3220
:param reload_func: A function to call if we determine that the pack
3221
files have moved and we need to reload our caches. See
3222
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
3224
self._container_writer = None
3225
self._write_index = None
3226
self._indices = index_to_packs
3227
self._reload_func = reload_func
3228
self._flush_func = flush_func
3230
def add_raw_records(self, key_sizes, raw_data):
3231
"""Add raw knit bytes to a storage area.
3233
The data is spooled to the container writer in one bytes-record per
3236
:param sizes: An iterable of tuples containing the key and size of each
3238
:param raw_data: A bytestring containing the data.
3239
:return: A list of memos to retrieve the record later. Each memo is an
3240
opaque index memo. For _DirectPackAccess the memo is (index, pos,
3241
length), where the index field is the write_index object supplied
3242
to the PackAccess object.
3244
if type(raw_data) is not str:
3245
raise AssertionError(
3246
'data must be plain bytes was %s' % type(raw_data))
3249
for key, size in key_sizes:
3250
p_offset, p_length = self._container_writer.add_bytes_record(
3251
raw_data[offset:offset+size], [])
3253
result.append((self._write_index, p_offset, p_length))
3257
"""Flush pending writes on this access object.
3259
This will flush any buffered writes to a NewPack.
3261
if self._flush_func is not None:
3264
def get_raw_records(self, memos_for_retrieval):
3265
"""Get the raw bytes for a records.
3267
:param memos_for_retrieval: An iterable containing the (index, pos,
3268
length) memo for retrieving the bytes. The Pack access method
3269
looks up the pack to use for a given record in its index_to_pack
3271
:return: An iterator over the bytes of the records.
3273
# first pass, group into same-index requests
3275
current_index = None
3276
for (index, offset, length) in memos_for_retrieval:
3277
if current_index == index:
3278
current_list.append((offset, length))
3280
if current_index is not None:
3281
request_lists.append((current_index, current_list))
3282
current_index = index
3283
current_list = [(offset, length)]
3284
# handle the last entry
3285
if current_index is not None:
3286
request_lists.append((current_index, current_list))
3287
for index, offsets in request_lists:
3289
transport, path = self._indices[index]
3291
# A KeyError here indicates that someone has triggered an index
3292
# reload, and this index has gone missing, we need to start
3294
if self._reload_func is None:
3295
# If we don't have a _reload_func there is nothing that can
3298
raise errors.RetryWithNewPacks(index,
3299
reload_occurred=True,
3300
exc_info=sys.exc_info())
3302
reader = pack.make_readv_reader(transport, path, offsets)
3303
for names, read_func in reader.iter_records():
3304
yield read_func(None)
3305
except errors.NoSuchFile:
3306
# A NoSuchFile error indicates that a pack file has gone
3307
# missing on disk, we need to trigger a reload, and start over.
3308
if self._reload_func is None:
3310
raise errors.RetryWithNewPacks(transport.abspath(path),
3311
reload_occurred=False,
3312
exc_info=sys.exc_info())
3314
def set_writer(self, writer, index, transport_packname):
3315
"""Set a writer to use for adding data."""
3316
if index is not None:
3317
self._indices[index] = transport_packname
3318
self._container_writer = writer
3319
self._write_index = index
3321
def reload_or_raise(self, retry_exc):
3322
"""Try calling the reload function, or re-raise the original exception.
3324
This should be called after _DirectPackAccess raises a
3325
RetryWithNewPacks exception. This function will handle the common logic
3326
of determining when the error is fatal versus being temporary.
3327
It will also make sure that the original exception is raised, rather
3328
than the RetryWithNewPacks exception.
3330
If this function returns, then the calling function should retry
3331
whatever operation was being performed. Otherwise an exception will
3334
:param retry_exc: A RetryWithNewPacks exception.
3337
if self._reload_func is None:
3339
elif not self._reload_func():
3340
# The reload claimed that nothing changed
3341
if not retry_exc.reload_occurred:
3342
# If there wasn't an earlier reload, then we really were
3343
# expecting to find changes. We didn't find them, so this is a
3347
exc_class, exc_value, exc_traceback = retry_exc.exc_info
3348
raise exc_class, exc_value, exc_traceback
3351
def annotate_knit(knit, revision_id):
3352
"""Annotate a knit with no cached annotations.
3354
This implementation is for knits with no cached annotations.
3355
It will work for knits with cached annotations, but this is not
3358
annotator = _KnitAnnotator(knit)
3359
return iter(annotator.annotate_flat(revision_id))
3362
class _KnitAnnotator(annotate.Annotator):
3363
"""Build up the annotations for a text."""
3365
def __init__(self, vf):
3366
annotate.Annotator.__init__(self, vf)
3368
# TODO: handle Nodes which cannot be extracted
3369
# self._ghosts = set()
3371
# Map from (key, parent_key) => matching_blocks, should be 'use once'
3372
self._matching_blocks = {}
3374
# KnitContent objects
3375
self._content_objects = {}
3376
# The number of children that depend on this fulltext content object
3377
self._num_compression_children = {}
3378
# Delta records that need their compression parent before they can be
3380
self._pending_deltas = {}
3381
# Fulltext records that are waiting for their parents fulltexts before
3382
# they can be yielded for annotation
3383
self._pending_annotation = {}
3385
self._all_build_details = {}
3387
def _get_build_graph(self, key):
3388
"""Get the graphs for building texts and annotations.
3390
The data you need for creating a full text may be different than the
3391
data you need to annotate that text. (At a minimum, you need both
3392
parents to create an annotation, but only need 1 parent to generate the
3395
:return: A list of (key, index_memo) records, suitable for
3396
passing to read_records_iter to start reading in the raw data from
3399
pending = set([key])
3402
self._num_needed_children[key] = 1
3404
# get all pending nodes
3405
this_iteration = pending
3406
build_details = self._vf._index.get_build_details(this_iteration)
3407
self._all_build_details.update(build_details)
3408
# new_nodes = self._vf._index._get_entries(this_iteration)
3410
for key, details in build_details.iteritems():
3411
(index_memo, compression_parent, parent_keys,
3412
record_details) = details
3413
self._parent_map[key] = parent_keys
3414
self._heads_provider = None
3415
records.append((key, index_memo))
3416
# Do we actually need to check _annotated_lines?
3417
pending.update([p for p in parent_keys
3418
if p not in self._all_build_details])
3420
for parent_key in parent_keys:
3421
if parent_key in self._num_needed_children:
3422
self._num_needed_children[parent_key] += 1
3424
self._num_needed_children[parent_key] = 1
3425
if compression_parent:
3426
if compression_parent in self._num_compression_children:
3427
self._num_compression_children[compression_parent] += 1
3429
self._num_compression_children[compression_parent] = 1
3431
missing_versions = this_iteration.difference(build_details.keys())
3432
if missing_versions:
3433
for key in missing_versions:
3434
if key in self._parent_map and key in self._text_cache:
3435
# We already have this text ready, we just need to
3436
# yield it later so we get it annotated
3438
parent_keys = self._parent_map[key]
3439
for parent_key in parent_keys:
3440
if parent_key in self._num_needed_children:
3441
self._num_needed_children[parent_key] += 1
3443
self._num_needed_children[parent_key] = 1
3444
pending.update([p for p in parent_keys
3445
if p not in self._all_build_details])
3447
raise errors.RevisionNotPresent(key, self._vf)
3448
# Generally we will want to read the records in reverse order, because
3449
# we find the parent nodes after the children
3451
return records, ann_keys
3453
def _get_needed_texts(self, key, pb=None):
3454
# if True or len(self._vf._immediate_fallback_vfs) > 0:
3455
if len(self._vf._immediate_fallback_vfs) > 0:
3456
# If we have fallbacks, go to the generic path
3457
for v in annotate.Annotator._get_needed_texts(self, key, pb=pb):
3462
records, ann_keys = self._get_build_graph(key)
3463
for idx, (sub_key, text, num_lines) in enumerate(
3464
self._extract_texts(records)):
3466
pb.update('annotating', idx, len(records))
3467
yield sub_key, text, num_lines
3468
for sub_key in ann_keys:
3469
text = self._text_cache[sub_key]
3470
num_lines = len(text) # bad assumption
3471
yield sub_key, text, num_lines
3473
except errors.RetryWithNewPacks, e:
3474
self._vf._access.reload_or_raise(e)
3475
# The cached build_details are no longer valid
3476
self._all_build_details.clear()
3478
def _cache_delta_blocks(self, key, compression_parent, delta, lines):
3479
parent_lines = self._text_cache[compression_parent]
3480
blocks = list(KnitContent.get_line_delta_blocks(delta, parent_lines, lines))
3481
self._matching_blocks[(key, compression_parent)] = blocks
3483
def _expand_record(self, key, parent_keys, compression_parent, record,
3486
if compression_parent:
3487
if compression_parent not in self._content_objects:
3488
# Waiting for the parent
3489
self._pending_deltas.setdefault(compression_parent, []).append(
3490
(key, parent_keys, record, record_details))
3492
# We have the basis parent, so expand the delta
3493
num = self._num_compression_children[compression_parent]
3496
base_content = self._content_objects.pop(compression_parent)
3497
self._num_compression_children.pop(compression_parent)
3499
self._num_compression_children[compression_parent] = num
3500
base_content = self._content_objects[compression_parent]
3501
# It is tempting to want to copy_base_content=False for the last
3502
# child object. However, whenever noeol=False,
3503
# self._text_cache[parent_key] is content._lines. So mutating it
3504
# gives very bad results.
3505
# The alternative is to copy the lines into text cache, but then we
3506
# are copying anyway, so just do it here.
3507
content, delta = self._vf._factory.parse_record(
3508
key, record, record_details, base_content,
3509
copy_base_content=True)
3512
content, _ = self._vf._factory.parse_record(
3513
key, record, record_details, None)
3514
if self._num_compression_children.get(key, 0) > 0:
3515
self._content_objects[key] = content
3516
lines = content.text()
3517
self._text_cache[key] = lines
3518
if delta is not None:
3519
self._cache_delta_blocks(key, compression_parent, delta, lines)
3522
def _get_parent_annotations_and_matches(self, key, text, parent_key):
3523
"""Get the list of annotations for the parent, and the matching lines.
3525
:param text: The opaque value given by _get_needed_texts
3526
:param parent_key: The key for the parent text
3527
:return: (parent_annotations, matching_blocks)
3528
parent_annotations is a list as long as the number of lines in
3530
matching_blocks is a list of (parent_idx, text_idx, len) tuples
3531
indicating which lines match between the two texts
3533
block_key = (key, parent_key)
3534
if block_key in self._matching_blocks:
3535
blocks = self._matching_blocks.pop(block_key)
3536
parent_annotations = self._annotations_cache[parent_key]
3537
return parent_annotations, blocks
3538
return annotate.Annotator._get_parent_annotations_and_matches(self,
3539
key, text, parent_key)
3541
def _process_pending(self, key):
3542
"""The content for 'key' was just processed.
3544
Determine if there is any more pending work to be processed.
3547
if key in self._pending_deltas:
3548
compression_parent = key
3549
children = self._pending_deltas.pop(key)
3550
for child_key, parent_keys, record, record_details in children:
3551
lines = self._expand_record(child_key, parent_keys,
3553
record, record_details)
3554
if self._check_ready_for_annotations(child_key, parent_keys):
3555
to_return.append(child_key)
3556
# Also check any children that are waiting for this parent to be
3558
if key in self._pending_annotation:
3559
children = self._pending_annotation.pop(key)
3560
to_return.extend([c for c, p_keys in children
3561
if self._check_ready_for_annotations(c, p_keys)])
3564
def _check_ready_for_annotations(self, key, parent_keys):
3565
"""return true if this text is ready to be yielded.
3567
Otherwise, this will return False, and queue the text into
3568
self._pending_annotation
3570
for parent_key in parent_keys:
3571
if parent_key not in self._annotations_cache:
3572
# still waiting on at least one parent text, so queue it up
3573
# Note that if there are multiple parents, we need to wait
3575
self._pending_annotation.setdefault(parent_key,
3576
[]).append((key, parent_keys))
3580
def _extract_texts(self, records):
3581
"""Extract the various texts needed based on records"""
3582
# We iterate in the order read, rather than a strict order requested
3583
# However, process what we can, and put off to the side things that
3584
# still need parents, cleaning them up when those parents are
3587
# 1) As 'records' are read, see if we can expand these records into
3588
# Content objects (and thus lines)
3589
# 2) If a given line-delta is waiting on its compression parent, it
3590
# gets queued up into self._pending_deltas, otherwise we expand
3591
# it, and put it into self._text_cache and self._content_objects
3592
# 3) If we expanded the text, we will then check to see if all
3593
# parents have also been processed. If so, this text gets yielded,
3594
# else this record gets set aside into pending_annotation
3595
# 4) Further, if we expanded the text in (2), we will then check to
3596
# see if there are any children in self._pending_deltas waiting to
3597
# also be processed. If so, we go back to (2) for those
3598
# 5) Further again, if we yielded the text, we can then check if that
3599
# 'unlocks' any of the texts in pending_annotations, which should
3600
# then get yielded as well
3601
# Note that both steps 4 and 5 are 'recursive' in that unlocking one
3602
# compression child could unlock yet another, and yielding a fulltext
3603
# will also 'unlock' the children that are waiting on that annotation.
3604
# (Though also, unlocking 1 parent's fulltext, does not unlock a child
3605
# if other parents are also waiting.)
3606
# We want to yield content before expanding child content objects, so
3607
# that we know when we can re-use the content lines, and the annotation
3608
# code can know when it can stop caching fulltexts, as well.
3610
# Children that are missing their compression parent
3612
for (key, record, digest) in self._vf._read_records_iter(records):
3614
details = self._all_build_details[key]
3615
(_, compression_parent, parent_keys, record_details) = details
3616
lines = self._expand_record(key, parent_keys, compression_parent,
3617
record, record_details)
3619
# Pending delta should be queued up
3621
# At this point, we may be able to yield this content, if all
3622
# parents are also finished
3623
yield_this_text = self._check_ready_for_annotations(key,
3626
# All parents present
3627
yield key, lines, len(lines)
3628
to_process = self._process_pending(key)
3630
this_process = to_process
3632
for key in this_process:
3633
lines = self._text_cache[key]
3634
yield key, lines, len(lines)
3635
to_process.extend(self._process_pending(key))
3638
from bzrlib._knit_load_data_pyx import _load_data_c as _load_data
3639
except ImportError, e:
3640
osutils.failed_to_load_extension(e)
3641
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
