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,
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(VersionedFiles):
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 add_fallback_versioned_files(self, a_versioned_files):
891
"""Add a source of texts for texts not present in this knit.
893
:param a_versioned_files: A VersionedFiles object.
895
self._immediate_fallback_vfs.append(a_versioned_files)
897
def add_lines(self, key, parents, lines, parent_texts=None,
898
left_matching_blocks=None, nostore_sha=None, random_id=False,
900
"""See VersionedFiles.add_lines()."""
901
self._index._check_write_ok()
902
self._check_add(key, lines, random_id, check_content)
904
# The caller might pass None if there is no graph data, but kndx
905
# indexes can't directly store that, so we give them
906
# an empty tuple instead.
908
line_bytes = ''.join(lines)
909
return self._add(key, lines, parents,
910
parent_texts, left_matching_blocks, nostore_sha, random_id,
911
line_bytes=line_bytes)
913
def _add_text(self, key, parents, text, nostore_sha=None, random_id=False):
914
"""See VersionedFiles._add_text()."""
915
self._index._check_write_ok()
916
self._check_add(key, None, random_id, check_content=False)
917
if text.__class__ is not str:
918
raise errors.BzrBadParameterUnicode("text")
920
# The caller might pass None if there is no graph data, but kndx
921
# indexes can't directly store that, so we give them
922
# an empty tuple instead.
924
return self._add(key, None, parents,
925
None, None, nostore_sha, random_id,
928
def _add(self, key, lines, parents, parent_texts,
929
left_matching_blocks, nostore_sha, random_id,
931
"""Add a set of lines on top of version specified by parents.
933
Any versions not present will be converted into ghosts.
935
:param lines: A list of strings where each one is a single line (has a
936
single newline at the end of the string) This is now optional
937
(callers can pass None). It is left in its location for backwards
938
compatibility. It should ''.join(lines) must == line_bytes
939
:param line_bytes: A single string containing the content
941
We pass both lines and line_bytes because different routes bring the
942
values to this function. And for memory efficiency, we don't want to
943
have to split/join on-demand.
945
# first thing, if the content is something we don't need to store, find
947
digest = sha_string(line_bytes)
948
if nostore_sha == digest:
949
raise errors.ExistingContent
952
if parent_texts is None:
954
# Do a single query to ascertain parent presence; we only compress
955
# against parents in the same kvf.
956
present_parent_map = self._index.get_parent_map(parents)
957
for parent in parents:
958
if parent in present_parent_map:
959
present_parents.append(parent)
961
# Currently we can only compress against the left most present parent.
962
if (len(present_parents) == 0 or
963
present_parents[0] != parents[0]):
966
# To speed the extract of texts the delta chain is limited
967
# to a fixed number of deltas. This should minimize both
968
# I/O and the time spend applying deltas.
969
delta = self._check_should_delta(present_parents[0])
971
text_length = len(line_bytes)
974
# Note: line_bytes is not modified to add a newline, that is tracked
975
# via the no_eol flag. 'lines' *is* modified, because that is the
976
# general values needed by the Content code.
977
if line_bytes and line_bytes[-1] != '\n':
978
options.append('no-eol')
980
# Copy the existing list, or create a new one
982
lines = osutils.split_lines(line_bytes)
985
# Replace the last line with one that ends in a final newline
986
lines[-1] = lines[-1] + '\n'
988
lines = osutils.split_lines(line_bytes)
990
for element in key[:-1]:
991
if type(element) is not str:
992
raise TypeError("key contains non-strings: %r" % (key,))
994
key = key[:-1] + ('sha1:' + digest,)
995
elif type(key[-1]) is not str:
996
raise TypeError("key contains non-strings: %r" % (key,))
997
# Knit hunks are still last-element only
999
content = self._factory.make(lines, version_id)
1001
# Hint to the content object that its text() call should strip the
1003
content._should_strip_eol = True
1004
if delta or (self._factory.annotated and len(present_parents) > 0):
1005
# Merge annotations from parent texts if needed.
1006
delta_hunks = self._merge_annotations(content, present_parents,
1007
parent_texts, delta, self._factory.annotated,
1008
left_matching_blocks)
1011
options.append('line-delta')
1012
store_lines = self._factory.lower_line_delta(delta_hunks)
1013
size, bytes = self._record_to_data(key, digest,
1016
options.append('fulltext')
1017
# isinstance is slower and we have no hierarchy.
1018
if self._factory.__class__ is KnitPlainFactory:
1019
# Use the already joined bytes saving iteration time in
1021
dense_lines = [line_bytes]
1023
dense_lines.append('\n')
1024
size, bytes = self._record_to_data(key, digest,
1027
# get mixed annotation + content and feed it into the
1029
store_lines = self._factory.lower_fulltext(content)
1030
size, bytes = self._record_to_data(key, digest,
1033
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
1034
self._index.add_records(
1035
((key, options, access_memo, parents),),
1036
random_id=random_id)
1037
return digest, text_length, content
1039
def annotate(self, key):
1040
"""See VersionedFiles.annotate."""
1041
return self._factory.annotate(self, key)
1043
def get_annotator(self):
1044
return _KnitAnnotator(self)
1046
def check(self, progress_bar=None, keys=None):
1047
"""See VersionedFiles.check()."""
1049
return self._logical_check()
1051
# At the moment, check does not extra work over get_record_stream
1052
return self.get_record_stream(keys, 'unordered', True)
1054
def _logical_check(self):
1055
# This doesn't actually test extraction of everything, but that will
1056
# impact 'bzr check' substantially, and needs to be integrated with
1057
# care. However, it does check for the obvious problem of a delta with
1059
keys = self._index.keys()
1060
parent_map = self.get_parent_map(keys)
1062
if self._index.get_method(key) != 'fulltext':
1063
compression_parent = parent_map[key][0]
1064
if compression_parent not in parent_map:
1065
raise errors.KnitCorrupt(self,
1066
"Missing basis parent %s for %s" % (
1067
compression_parent, key))
1068
for fallback_vfs in self._immediate_fallback_vfs:
1069
fallback_vfs.check()
1071
def _check_add(self, key, lines, random_id, check_content):
1072
"""check that version_id and lines are safe to add."""
1073
version_id = key[-1]
1074
if version_id is not None:
1075
if contains_whitespace(version_id):
1076
raise InvalidRevisionId(version_id, self)
1077
self.check_not_reserved_id(version_id)
1078
# TODO: If random_id==False and the key is already present, we should
1079
# probably check that the existing content is identical to what is
1080
# being inserted, and otherwise raise an exception. This would make
1081
# the bundle code simpler.
1083
self._check_lines_not_unicode(lines)
1084
self._check_lines_are_lines(lines)
1086
def _check_header(self, key, line):
1087
rec = self._split_header(line)
1088
self._check_header_version(rec, key[-1])
1091
def _check_header_version(self, rec, version_id):
1092
"""Checks the header version on original format knit records.
1094
These have the last component of the key embedded in the record.
1096
if rec[1] != version_id:
1097
raise KnitCorrupt(self,
1098
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1100
def _check_should_delta(self, parent):
1101
"""Iterate back through the parent listing, looking for a fulltext.
1103
This is used when we want to decide whether to add a delta or a new
1104
fulltext. It searches for _max_delta_chain parents. When it finds a
1105
fulltext parent, it sees if the total size of the deltas leading up to
1106
it is large enough to indicate that we want a new full text anyway.
1108
Return True if we should create a new delta, False if we should use a
1112
fulltext_size = None
1113
for count in xrange(self._max_delta_chain):
1115
# Note that this only looks in the index of this particular
1116
# KnitVersionedFiles, not in the fallbacks. This ensures that
1117
# we won't store a delta spanning physical repository
1119
build_details = self._index.get_build_details([parent])
1120
parent_details = build_details[parent]
1121
except (RevisionNotPresent, KeyError), e:
1122
# Some basis is not locally present: always fulltext
1124
index_memo, compression_parent, _, _ = parent_details
1125
_, _, size = index_memo
1126
if compression_parent is None:
1127
fulltext_size = size
1130
# We don't explicitly check for presence because this is in an
1131
# inner loop, and if it's missing it'll fail anyhow.
1132
parent = compression_parent
1134
# We couldn't find a fulltext, so we must create a new one
1136
# Simple heuristic - if the total I/O wold be greater as a delta than
1137
# the originally installed fulltext, we create a new fulltext.
1138
return fulltext_size > delta_size
1140
def _build_details_to_components(self, build_details):
1141
"""Convert a build_details tuple to a position tuple."""
1142
# record_details, access_memo, compression_parent
1143
return build_details[3], build_details[0], build_details[1]
1145
def _get_components_positions(self, keys, allow_missing=False):
1146
"""Produce a map of position data for the components of keys.
1148
This data is intended to be used for retrieving the knit records.
1150
A dict of key to (record_details, index_memo, next, parents) is
1152
method is the way referenced data should be applied.
1153
index_memo is the handle to pass to the data access to actually get the
1155
next is the build-parent of the version, or None for fulltexts.
1156
parents is the version_ids of the parents of this version
1158
:param allow_missing: If True do not raise an error on a missing component,
1162
pending_components = keys
1163
while pending_components:
1164
build_details = self._index.get_build_details(pending_components)
1165
current_components = set(pending_components)
1166
pending_components = set()
1167
for key, details in build_details.iteritems():
1168
(index_memo, compression_parent, parents,
1169
record_details) = details
1170
method = record_details[0]
1171
if compression_parent is not None:
1172
pending_components.add(compression_parent)
1173
component_data[key] = self._build_details_to_components(details)
1174
missing = current_components.difference(build_details)
1175
if missing and not allow_missing:
1176
raise errors.RevisionNotPresent(missing.pop(), self)
1177
return component_data
1179
def _get_content(self, key, parent_texts={}):
1180
"""Returns a content object that makes up the specified
1182
cached_version = parent_texts.get(key, None)
1183
if cached_version is not None:
1184
# Ensure the cache dict is valid.
1185
if not self.get_parent_map([key]):
1186
raise RevisionNotPresent(key, self)
1187
return cached_version
1188
generator = _VFContentMapGenerator(self, [key])
1189
return generator._get_content(key)
1191
def get_known_graph_ancestry(self, keys):
1192
"""Get a KnownGraph instance with the ancestry of keys."""
1193
parent_map, missing_keys = self._index.find_ancestry(keys)
1194
for fallback in self._transitive_fallbacks():
1195
if not missing_keys:
1197
(f_parent_map, f_missing_keys) = fallback._index.find_ancestry(
1199
parent_map.update(f_parent_map)
1200
missing_keys = f_missing_keys
1201
kg = _mod_graph.KnownGraph(parent_map)
1204
def get_parent_map(self, keys):
1205
"""Get a map of the graph parents of keys.
1207
:param keys: The keys to look up parents for.
1208
:return: A mapping from keys to parents. Absent keys are absent from
1211
return self._get_parent_map_with_sources(keys)[0]
1213
def _get_parent_map_with_sources(self, keys):
1214
"""Get a map of the parents of keys.
1216
:param keys: The keys to look up parents for.
1217
:return: A tuple. The first element is a mapping from keys to parents.
1218
Absent keys are absent from the mapping. The second element is a
1219
list with the locations each key was found in. The first element
1220
is the in-this-knit parents, the second the first fallback source,
1224
sources = [self._index] + self._immediate_fallback_vfs
1227
for source in sources:
1230
new_result = source.get_parent_map(missing)
1231
source_results.append(new_result)
1232
result.update(new_result)
1233
missing.difference_update(set(new_result))
1234
return result, source_results
1236
def _get_record_map(self, keys, allow_missing=False):
1237
"""Produce a dictionary of knit records.
1239
:return: {key:(record, record_details, digest, next)}
1241
data returned from read_records (a KnitContentobject)
1243
opaque information to pass to parse_record
1245
SHA1 digest of the full text after all steps are done
1247
build-parent of the version, i.e. the leftmost ancestor.
1248
Will be None if the record is not a delta.
1249
:param keys: The keys to build a map for
1250
:param allow_missing: If some records are missing, rather than
1251
error, just return the data that could be generated.
1253
raw_map = self._get_record_map_unparsed(keys,
1254
allow_missing=allow_missing)
1255
return self._raw_map_to_record_map(raw_map)
1257
def _raw_map_to_record_map(self, raw_map):
1258
"""Parse the contents of _get_record_map_unparsed.
1260
:return: see _get_record_map.
1264
data, record_details, next = raw_map[key]
1265
content, digest = self._parse_record(key[-1], data)
1266
result[key] = content, record_details, digest, next
1269
def _get_record_map_unparsed(self, keys, allow_missing=False):
1270
"""Get the raw data for reconstructing keys without parsing it.
1272
:return: A dict suitable for parsing via _raw_map_to_record_map.
1273
key-> raw_bytes, (method, noeol), compression_parent
1275
# This retries the whole request if anything fails. Potentially we
1276
# could be a bit more selective. We could track the keys whose records
1277
# we have successfully found, and then only request the new records
1278
# from there. However, _get_components_positions grabs the whole build
1279
# chain, which means we'll likely try to grab the same records again
1280
# anyway. Also, can the build chains change as part of a pack
1281
# operation? We wouldn't want to end up with a broken chain.
1284
position_map = self._get_components_positions(keys,
1285
allow_missing=allow_missing)
1286
# key = component_id, r = record_details, i_m = index_memo,
1288
records = [(key, i_m) for key, (r, i_m, n)
1289
in position_map.iteritems()]
1290
# Sort by the index memo, so that we request records from the
1291
# same pack file together, and in forward-sorted order
1292
records.sort(key=operator.itemgetter(1))
1294
for key, data in self._read_records_iter_unchecked(records):
1295
(record_details, index_memo, next) = position_map[key]
1296
raw_record_map[key] = data, record_details, next
1297
return raw_record_map
1298
except errors.RetryWithNewPacks, e:
1299
self._access.reload_or_raise(e)
1302
def _split_by_prefix(cls, keys):
1303
"""For the given keys, split them up based on their prefix.
1305
To keep memory pressure somewhat under control, split the
1306
requests back into per-file-id requests, otherwise "bzr co"
1307
extracts the full tree into memory before writing it to disk.
1308
This should be revisited if _get_content_maps() can ever cross
1311
The keys for a given file_id are kept in the same relative order.
1312
Ordering between file_ids is not, though prefix_order will return the
1313
order that the key was first seen.
1315
:param keys: An iterable of key tuples
1316
:return: (split_map, prefix_order)
1317
split_map A dictionary mapping prefix => keys
1318
prefix_order The order that we saw the various prefixes
1320
split_by_prefix = {}
1328
if prefix in split_by_prefix:
1329
split_by_prefix[prefix].append(key)
1331
split_by_prefix[prefix] = [key]
1332
prefix_order.append(prefix)
1333
return split_by_prefix, prefix_order
1335
def _group_keys_for_io(self, keys, non_local_keys, positions,
1336
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1337
"""For the given keys, group them into 'best-sized' requests.
1339
The idea is to avoid making 1 request per file, but to never try to
1340
unpack an entire 1.5GB source tree in a single pass. Also when
1341
possible, we should try to group requests to the same pack file
1344
:return: list of (keys, non_local) tuples that indicate what keys
1345
should be fetched next.
1347
# TODO: Ideally we would group on 2 factors. We want to extract texts
1348
# from the same pack file together, and we want to extract all
1349
# the texts for a given build-chain together. Ultimately it
1350
# probably needs a better global view.
1351
total_keys = len(keys)
1352
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1353
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1355
cur_non_local = set()
1359
for prefix in prefix_order:
1360
keys = prefix_split_keys[prefix]
1361
non_local = prefix_split_non_local_keys.get(prefix, [])
1363
this_size = self._index._get_total_build_size(keys, positions)
1364
cur_size += this_size
1365
cur_keys.extend(keys)
1366
cur_non_local.update(non_local)
1367
if cur_size > _min_buffer_size:
1368
result.append((cur_keys, cur_non_local))
1369
sizes.append(cur_size)
1371
cur_non_local = set()
1374
result.append((cur_keys, cur_non_local))
1375
sizes.append(cur_size)
1378
def get_record_stream(self, keys, ordering, include_delta_closure):
1379
"""Get a stream of records for keys.
1381
:param keys: The keys to include.
1382
:param ordering: Either 'unordered' or 'topological'. A topologically
1383
sorted stream has compression parents strictly before their
1385
:param include_delta_closure: If True then the closure across any
1386
compression parents will be included (in the opaque data).
1387
:return: An iterator of ContentFactory objects, each of which is only
1388
valid until the iterator is advanced.
1390
# keys might be a generator
1394
if not self._index.has_graph:
1395
# Cannot sort when no graph has been stored.
1396
ordering = 'unordered'
1398
remaining_keys = keys
1401
keys = set(remaining_keys)
1402
for content_factory in self._get_remaining_record_stream(keys,
1403
ordering, include_delta_closure):
1404
remaining_keys.discard(content_factory.key)
1405
yield content_factory
1407
except errors.RetryWithNewPacks, e:
1408
self._access.reload_or_raise(e)
1410
def _get_remaining_record_stream(self, keys, ordering,
1411
include_delta_closure):
1412
"""This function is the 'retry' portion for get_record_stream."""
1413
if include_delta_closure:
1414
positions = self._get_components_positions(keys, allow_missing=True)
1416
build_details = self._index.get_build_details(keys)
1418
# (record_details, access_memo, compression_parent_key)
1419
positions = dict((key, self._build_details_to_components(details))
1420
for key, details in build_details.iteritems())
1421
absent_keys = keys.difference(set(positions))
1422
# There may be more absent keys : if we're missing the basis component
1423
# and are trying to include the delta closure.
1424
# XXX: We should not ever need to examine remote sources because we do
1425
# not permit deltas across versioned files boundaries.
1426
if include_delta_closure:
1427
needed_from_fallback = set()
1428
# Build up reconstructable_keys dict. key:True in this dict means
1429
# the key can be reconstructed.
1430
reconstructable_keys = {}
1434
chain = [key, positions[key][2]]
1436
needed_from_fallback.add(key)
1439
while chain[-1] is not None:
1440
if chain[-1] in reconstructable_keys:
1441
result = reconstructable_keys[chain[-1]]
1445
chain.append(positions[chain[-1]][2])
1447
# missing basis component
1448
needed_from_fallback.add(chain[-1])
1451
for chain_key in chain[:-1]:
1452
reconstructable_keys[chain_key] = result
1454
needed_from_fallback.add(key)
1455
# Double index lookups here : need a unified api ?
1456
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1457
if ordering in ('topological', 'groupcompress'):
1458
if ordering == 'topological':
1459
# Global topological sort
1460
present_keys = tsort.topo_sort(global_map)
1462
present_keys = sort_groupcompress(global_map)
1463
# Now group by source:
1465
current_source = None
1466
for key in present_keys:
1467
for parent_map in parent_maps:
1468
if key in parent_map:
1469
key_source = parent_map
1471
if current_source is not key_source:
1472
source_keys.append((key_source, []))
1473
current_source = key_source
1474
source_keys[-1][1].append(key)
1476
if ordering != 'unordered':
1477
raise AssertionError('valid values for ordering are:'
1478
' "unordered", "groupcompress" or "topological" not: %r'
1480
# Just group by source; remote sources first.
1483
for parent_map in reversed(parent_maps):
1484
source_keys.append((parent_map, []))
1485
for key in parent_map:
1486
present_keys.append(key)
1487
source_keys[-1][1].append(key)
1488
# We have been requested to return these records in an order that
1489
# suits us. So we ask the index to give us an optimally sorted
1491
for source, sub_keys in source_keys:
1492
if source is parent_maps[0]:
1493
# Only sort the keys for this VF
1494
self._index._sort_keys_by_io(sub_keys, positions)
1495
absent_keys = keys - set(global_map)
1496
for key in absent_keys:
1497
yield AbsentContentFactory(key)
1498
# restrict our view to the keys we can answer.
1499
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1500
# XXX: At that point we need to consider the impact of double reads by
1501
# utilising components multiple times.
1502
if include_delta_closure:
1503
# XXX: get_content_maps performs its own index queries; allow state
1505
non_local_keys = needed_from_fallback - absent_keys
1506
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1509
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1512
for record in generator.get_record_stream():
1515
for source, keys in source_keys:
1516
if source is parent_maps[0]:
1517
# this KnitVersionedFiles
1518
records = [(key, positions[key][1]) for key in keys]
1519
for key, raw_data in self._read_records_iter_unchecked(records):
1520
(record_details, index_memo, _) = positions[key]
1521
yield KnitContentFactory(key, global_map[key],
1522
record_details, None, raw_data, self._factory.annotated, None)
1524
vf = self._immediate_fallback_vfs[parent_maps.index(source) - 1]
1525
for record in vf.get_record_stream(keys, ordering,
1526
include_delta_closure):
1529
def get_sha1s(self, keys):
1530
"""See VersionedFiles.get_sha1s()."""
1532
record_map = self._get_record_map(missing, allow_missing=True)
1534
for key, details in record_map.iteritems():
1535
if key not in missing:
1537
# record entry 2 is the 'digest'.
1538
result[key] = details[2]
1539
missing.difference_update(set(result))
1540
for source in self._immediate_fallback_vfs:
1543
new_result = source.get_sha1s(missing)
1544
result.update(new_result)
1545
missing.difference_update(set(new_result))
1548
def insert_record_stream(self, stream):
1549
"""Insert a record stream into this container.
1551
:param stream: A stream of records to insert.
1553
:seealso VersionedFiles.get_record_stream:
1555
def get_adapter(adapter_key):
1557
return adapters[adapter_key]
1559
adapter_factory = adapter_registry.get(adapter_key)
1560
adapter = adapter_factory(self)
1561
adapters[adapter_key] = adapter
1564
if self._factory.annotated:
1565
# self is annotated, we need annotated knits to use directly.
1566
annotated = "annotated-"
1569
# self is not annotated, but we can strip annotations cheaply.
1571
convertibles = set(["knit-annotated-ft-gz"])
1572
if self._max_delta_chain:
1573
delta_types.add("knit-annotated-delta-gz")
1574
convertibles.add("knit-annotated-delta-gz")
1575
# The set of types we can cheaply adapt without needing basis texts.
1576
native_types = set()
1577
if self._max_delta_chain:
1578
native_types.add("knit-%sdelta-gz" % annotated)
1579
delta_types.add("knit-%sdelta-gz" % annotated)
1580
native_types.add("knit-%sft-gz" % annotated)
1581
knit_types = native_types.union(convertibles)
1583
# Buffer all index entries that we can't add immediately because their
1584
# basis parent is missing. We don't buffer all because generating
1585
# annotations may require access to some of the new records. However we
1586
# can't generate annotations from new deltas until their basis parent
1587
# is present anyway, so we get away with not needing an index that
1588
# includes the new keys.
1590
# See <http://launchpad.net/bugs/300177> about ordering of compression
1591
# parents in the records - to be conservative, we insist that all
1592
# parents must be present to avoid expanding to a fulltext.
1594
# key = basis_parent, value = index entry to add
1595
buffered_index_entries = {}
1596
for record in stream:
1597
kind = record.storage_kind
1598
if kind.startswith('knit-') and kind.endswith('-gz'):
1599
# Check that the ID in the header of the raw knit bytes matches
1600
# the record metadata.
1601
raw_data = record._raw_record
1602
df, rec = self._parse_record_header(record.key, raw_data)
1605
parents = record.parents
1606
if record.storage_kind in delta_types:
1607
# TODO: eventually the record itself should track
1608
# compression_parent
1609
compression_parent = parents[0]
1611
compression_parent = None
1612
# Raise an error when a record is missing.
1613
if record.storage_kind == 'absent':
1614
raise RevisionNotPresent([record.key], self)
1615
elif ((record.storage_kind in knit_types)
1616
and (compression_parent is None
1617
or not self._immediate_fallback_vfs
1618
or self._index.has_key(compression_parent)
1619
or not self.has_key(compression_parent))):
1620
# we can insert the knit record literally if either it has no
1621
# compression parent OR we already have its basis in this kvf
1622
# OR the basis is not present even in the fallbacks. In the
1623
# last case it will either turn up later in the stream and all
1624
# will be well, or it won't turn up at all and we'll raise an
1627
# TODO: self.has_key is somewhat redundant with
1628
# self._index.has_key; we really want something that directly
1629
# asks if it's only present in the fallbacks. -- mbp 20081119
1630
if record.storage_kind not in native_types:
1632
adapter_key = (record.storage_kind, "knit-delta-gz")
1633
adapter = get_adapter(adapter_key)
1635
adapter_key = (record.storage_kind, "knit-ft-gz")
1636
adapter = get_adapter(adapter_key)
1637
bytes = adapter.get_bytes(record)
1639
# It's a knit record, it has a _raw_record field (even if
1640
# it was reconstituted from a network stream).
1641
bytes = record._raw_record
1642
options = [record._build_details[0]]
1643
if record._build_details[1]:
1644
options.append('no-eol')
1645
# Just blat it across.
1646
# Note: This does end up adding data on duplicate keys. As
1647
# modern repositories use atomic insertions this should not
1648
# lead to excessive growth in the event of interrupted fetches.
1649
# 'knit' repositories may suffer excessive growth, but as a
1650
# deprecated format this is tolerable. It can be fixed if
1651
# needed by in the kndx index support raising on a duplicate
1652
# add with identical parents and options.
1653
access_memo = self._access.add_raw_records(
1654
[(record.key, len(bytes))], bytes)[0]
1655
index_entry = (record.key, options, access_memo, parents)
1656
if 'fulltext' not in options:
1657
# Not a fulltext, so we need to make sure the compression
1658
# parent will also be present.
1659
# Note that pack backed knits don't need to buffer here
1660
# because they buffer all writes to the transaction level,
1661
# but we don't expose that difference at the index level. If
1662
# the query here has sufficient cost to show up in
1663
# profiling we should do that.
1665
# They're required to be physically in this
1666
# KnitVersionedFiles, not in a fallback.
1667
if not self._index.has_key(compression_parent):
1668
pending = buffered_index_entries.setdefault(
1669
compression_parent, [])
1670
pending.append(index_entry)
1673
self._index.add_records([index_entry])
1674
elif record.storage_kind == 'chunked':
1675
self.add_lines(record.key, parents,
1676
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1678
# Not suitable for direct insertion as a
1679
# delta, either because it's not the right format, or this
1680
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1681
# 0) or because it depends on a base only present in the
1683
self._access.flush()
1685
# Try getting a fulltext directly from the record.
1686
bytes = record.get_bytes_as('fulltext')
1687
except errors.UnavailableRepresentation:
1688
adapter_key = record.storage_kind, 'fulltext'
1689
adapter = get_adapter(adapter_key)
1690
bytes = adapter.get_bytes(record)
1691
lines = split_lines(bytes)
1693
self.add_lines(record.key, parents, lines)
1694
except errors.RevisionAlreadyPresent:
1696
# Add any records whose basis parent is now available.
1698
added_keys = [record.key]
1700
key = added_keys.pop(0)
1701
if key in buffered_index_entries:
1702
index_entries = buffered_index_entries[key]
1703
self._index.add_records(index_entries)
1705
[index_entry[0] for index_entry in index_entries])
1706
del buffered_index_entries[key]
1707
if buffered_index_entries:
1708
# There were index entries buffered at the end of the stream,
1709
# So these need to be added (if the index supports holding such
1710
# entries for later insertion)
1712
for key in buffered_index_entries:
1713
index_entries = buffered_index_entries[key]
1714
all_entries.extend(index_entries)
1715
self._index.add_records(
1716
all_entries, missing_compression_parents=True)
1718
def get_missing_compression_parent_keys(self):
1719
"""Return an iterable of keys of missing compression parents.
1721
Check this after calling insert_record_stream to find out if there are
1722
any missing compression parents. If there are, the records that
1723
depend on them are not able to be inserted safely. For atomic
1724
KnitVersionedFiles built on packs, the transaction should be aborted or
1725
suspended - commit will fail at this point. Nonatomic knits will error
1726
earlier because they have no staging area to put pending entries into.
1728
return self._index.get_missing_compression_parents()
1730
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1731
"""Iterate over the lines in the versioned files from keys.
1733
This may return lines from other keys. Each item the returned
1734
iterator yields is a tuple of a line and a text version that that line
1735
is present in (not introduced in).
1737
Ordering of results is in whatever order is most suitable for the
1738
underlying storage format.
1740
If a progress bar is supplied, it may be used to indicate progress.
1741
The caller is responsible for cleaning up progress bars (because this
1745
* Lines are normalised by the underlying store: they will all have \\n
1747
* Lines are returned in arbitrary order.
1748
* If a requested key did not change any lines (or didn't have any
1749
lines), it may not be mentioned at all in the result.
1751
:param pb: Progress bar supplied by caller.
1752
:return: An iterator over (line, key).
1755
pb = ui.ui_factory.nested_progress_bar()
1761
# we don't care about inclusions, the caller cares.
1762
# but we need to setup a list of records to visit.
1763
# we need key, position, length
1765
build_details = self._index.get_build_details(keys)
1766
for key, details in build_details.iteritems():
1768
key_records.append((key, details[0]))
1769
records_iter = enumerate(self._read_records_iter(key_records))
1770
for (key_idx, (key, data, sha_value)) in records_iter:
1771
pb.update('Walking content', key_idx, total)
1772
compression_parent = build_details[key][1]
1773
if compression_parent is None:
1775
line_iterator = self._factory.get_fulltext_content(data)
1778
line_iterator = self._factory.get_linedelta_content(data)
1779
# Now that we are yielding the data for this key, remove it
1782
# XXX: It might be more efficient to yield (key,
1783
# line_iterator) in the future. However for now, this is a
1784
# simpler change to integrate into the rest of the
1785
# codebase. RBC 20071110
1786
for line in line_iterator:
1789
except errors.RetryWithNewPacks, e:
1790
self._access.reload_or_raise(e)
1791
# If there are still keys we've not yet found, we look in the fallback
1792
# vfs, and hope to find them there. Note that if the keys are found
1793
# but had no changes or no content, the fallback may not return
1795
if keys and not self._immediate_fallback_vfs:
1796
# XXX: strictly the second parameter is meant to be the file id
1797
# but it's not easily accessible here.
1798
raise RevisionNotPresent(keys, repr(self))
1799
for source in self._immediate_fallback_vfs:
1803
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1804
source_keys.add(key)
1806
keys.difference_update(source_keys)
1807
pb.update('Walking content', total, total)
1809
def _make_line_delta(self, delta_seq, new_content):
1810
"""Generate a line delta from delta_seq and new_content."""
1812
for op in delta_seq.get_opcodes():
1813
if op[0] == 'equal':
1815
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1818
def _merge_annotations(self, content, parents, parent_texts={},
1819
delta=None, annotated=None,
1820
left_matching_blocks=None):
1821
"""Merge annotations for content and generate deltas.
1823
This is done by comparing the annotations based on changes to the text
1824
and generating a delta on the resulting full texts. If annotations are
1825
not being created then a simple delta is created.
1827
if left_matching_blocks is not None:
1828
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1832
for parent_key in parents:
1833
merge_content = self._get_content(parent_key, parent_texts)
1834
if (parent_key == parents[0] and delta_seq is not None):
1837
seq = patiencediff.PatienceSequenceMatcher(
1838
None, merge_content.text(), content.text())
1839
for i, j, n in seq.get_matching_blocks():
1842
# this copies (origin, text) pairs across to the new
1843
# content for any line that matches the last-checked
1845
content._lines[j:j+n] = merge_content._lines[i:i+n]
1846
# XXX: Robert says the following block is a workaround for a
1847
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1848
if content._lines and content._lines[-1][1][-1] != '\n':
1849
# The copied annotation was from a line without a trailing EOL,
1850
# reinstate one for the content object, to ensure correct
1852
line = content._lines[-1][1] + '\n'
1853
content._lines[-1] = (content._lines[-1][0], line)
1855
if delta_seq is None:
1856
reference_content = self._get_content(parents[0], parent_texts)
1857
new_texts = content.text()
1858
old_texts = reference_content.text()
1859
delta_seq = patiencediff.PatienceSequenceMatcher(
1860
None, old_texts, new_texts)
1861
return self._make_line_delta(delta_seq, content)
1863
def _parse_record(self, version_id, data):
1864
"""Parse an original format knit record.
1866
These have the last element of the key only present in the stored data.
1868
rec, record_contents = self._parse_record_unchecked(data)
1869
self._check_header_version(rec, version_id)
1870
return record_contents, rec[3]
1872
def _parse_record_header(self, key, raw_data):
1873
"""Parse a record header for consistency.
1875
:return: the header and the decompressor stream.
1876
as (stream, header_record)
1878
df = gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1881
rec = self._check_header(key, df.readline())
1882
except Exception, e:
1883
raise KnitCorrupt(self,
1884
"While reading {%s} got %s(%s)"
1885
% (key, e.__class__.__name__, str(e)))
1888
def _parse_record_unchecked(self, data):
1890
# 4168 calls in 2880 217 internal
1891
# 4168 calls to _parse_record_header in 2121
1892
# 4168 calls to readlines in 330
1893
df = gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1895
record_contents = df.readlines()
1896
except Exception, e:
1897
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1898
(data, e.__class__.__name__, str(e)))
1899
header = record_contents.pop(0)
1900
rec = self._split_header(header)
1901
last_line = record_contents.pop()
1902
if len(record_contents) != int(rec[2]):
1903
raise KnitCorrupt(self,
1904
'incorrect number of lines %s != %s'
1905
' for version {%s} %s'
1906
% (len(record_contents), int(rec[2]),
1907
rec[1], record_contents))
1908
if last_line != 'end %s\n' % rec[1]:
1909
raise KnitCorrupt(self,
1910
'unexpected version end line %r, wanted %r'
1911
% (last_line, rec[1]))
1913
return rec, record_contents
1915
def _read_records_iter(self, records):
1916
"""Read text records from data file and yield result.
1918
The result will be returned in whatever is the fastest to read.
1919
Not by the order requested. Also, multiple requests for the same
1920
record will only yield 1 response.
1921
:param records: A list of (key, access_memo) entries
1922
:return: Yields (key, contents, digest) in the order
1923
read, not the order requested
1928
# XXX: This smells wrong, IO may not be getting ordered right.
1929
needed_records = sorted(set(records), key=operator.itemgetter(1))
1930
if not needed_records:
1933
# The transport optimizes the fetching as well
1934
# (ie, reads continuous ranges.)
1935
raw_data = self._access.get_raw_records(
1936
[index_memo for key, index_memo in needed_records])
1938
for (key, index_memo), data in \
1939
izip(iter(needed_records), raw_data):
1940
content, digest = self._parse_record(key[-1], data)
1941
yield key, content, digest
1943
def _read_records_iter_raw(self, records):
1944
"""Read text records from data file and yield raw data.
1946
This unpacks enough of the text record to validate the id is
1947
as expected but thats all.
1949
Each item the iterator yields is (key, bytes,
1950
expected_sha1_of_full_text).
1952
for key, data in self._read_records_iter_unchecked(records):
1953
# validate the header (note that we can only use the suffix in
1954
# current knit records).
1955
df, rec = self._parse_record_header(key, data)
1957
yield key, data, rec[3]
1959
def _read_records_iter_unchecked(self, records):
1960
"""Read text records from data file and yield raw data.
1962
No validation is done.
1964
Yields tuples of (key, data).
1966
# setup an iterator of the external records:
1967
# uses readv so nice and fast we hope.
1969
# grab the disk data needed.
1970
needed_offsets = [index_memo for key, index_memo
1972
raw_records = self._access.get_raw_records(needed_offsets)
1974
for key, index_memo in records:
1975
data = raw_records.next()
1978
def _record_to_data(self, key, digest, lines, dense_lines=None):
1979
"""Convert key, digest, lines into a raw data block.
1981
:param key: The key of the record. Currently keys are always serialised
1982
using just the trailing component.
1983
:param dense_lines: The bytes of lines but in a denser form. For
1984
instance, if lines is a list of 1000 bytestrings each ending in \n,
1985
dense_lines may be a list with one line in it, containing all the
1986
1000's lines and their \n's. Using dense_lines if it is already
1987
known is a win because the string join to create bytes in this
1988
function spends less time resizing the final string.
1989
:return: (len, a StringIO instance with the raw data ready to read.)
1991
chunks = ["version %s %d %s\n" % (key[-1], len(lines), digest)]
1992
chunks.extend(dense_lines or lines)
1993
chunks.append("end %s\n" % key[-1])
1994
for chunk in chunks:
1995
if type(chunk) is not str:
1996
raise AssertionError(
1997
'data must be plain bytes was %s' % type(chunk))
1998
if lines and lines[-1][-1] != '\n':
1999
raise ValueError('corrupt lines value %r' % lines)
2000
compressed_bytes = tuned_gzip.chunks_to_gzip(chunks)
2001
return len(compressed_bytes), compressed_bytes
2003
def _split_header(self, line):
2006
raise KnitCorrupt(self,
2007
'unexpected number of elements in record header')
2011
"""See VersionedFiles.keys."""
2012
if 'evil' in debug.debug_flags:
2013
trace.mutter_callsite(2, "keys scales with size of history")
2014
sources = [self._index] + self._immediate_fallback_vfs
2016
for source in sources:
2017
result.update(source.keys())
2021
class _ContentMapGenerator(object):
2022
"""Generate texts or expose raw deltas for a set of texts."""
2024
def __init__(self, ordering='unordered'):
2025
self._ordering = ordering
2027
def _get_content(self, key):
2028
"""Get the content object for key."""
2029
# Note that _get_content is only called when the _ContentMapGenerator
2030
# has been constructed with just one key requested for reconstruction.
2031
if key in self.nonlocal_keys:
2032
record = self.get_record_stream().next()
2033
# Create a content object on the fly
2034
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
2035
return PlainKnitContent(lines, record.key)
2037
# local keys we can ask for directly
2038
return self._get_one_work(key)
2040
def get_record_stream(self):
2041
"""Get a record stream for the keys requested during __init__."""
2042
for record in self._work():
2046
"""Produce maps of text and KnitContents as dicts.
2048
:return: (text_map, content_map) where text_map contains the texts for
2049
the requested versions and content_map contains the KnitContents.
2051
# NB: By definition we never need to read remote sources unless texts
2052
# are requested from them: we don't delta across stores - and we
2053
# explicitly do not want to to prevent data loss situations.
2054
if self.global_map is None:
2055
self.global_map = self.vf.get_parent_map(self.keys)
2056
nonlocal_keys = self.nonlocal_keys
2058
missing_keys = set(nonlocal_keys)
2059
# Read from remote versioned file instances and provide to our caller.
2060
for source in self.vf._immediate_fallback_vfs:
2061
if not missing_keys:
2063
# Loop over fallback repositories asking them for texts - ignore
2064
# any missing from a particular fallback.
2065
for record in source.get_record_stream(missing_keys,
2066
self._ordering, True):
2067
if record.storage_kind == 'absent':
2068
# Not in thie particular stream, may be in one of the
2069
# other fallback vfs objects.
2071
missing_keys.remove(record.key)
2074
if self._raw_record_map is None:
2075
raise AssertionError('_raw_record_map should have been filled')
2077
for key in self.keys:
2078
if key in self.nonlocal_keys:
2080
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2083
def _get_one_work(self, requested_key):
2084
# Now, if we have calculated everything already, just return the
2086
if requested_key in self._contents_map:
2087
return self._contents_map[requested_key]
2088
# To simplify things, parse everything at once - code that wants one text
2089
# probably wants them all.
2090
# FUTURE: This function could be improved for the 'extract many' case
2091
# by tracking each component and only doing the copy when the number of
2092
# children than need to apply delta's to it is > 1 or it is part of the
2094
multiple_versions = len(self.keys) != 1
2095
if self._record_map is None:
2096
self._record_map = self.vf._raw_map_to_record_map(
2097
self._raw_record_map)
2098
record_map = self._record_map
2099
# raw_record_map is key:
2100
# Have read and parsed records at this point.
2101
for key in self.keys:
2102
if key in self.nonlocal_keys:
2107
while cursor is not None:
2109
record, record_details, digest, next = record_map[cursor]
2111
raise RevisionNotPresent(cursor, self)
2112
components.append((cursor, record, record_details, digest))
2114
if cursor in self._contents_map:
2115
# no need to plan further back
2116
components.append((cursor, None, None, None))
2120
for (component_id, record, record_details,
2121
digest) in reversed(components):
2122
if component_id in self._contents_map:
2123
content = self._contents_map[component_id]
2125
content, delta = self._factory.parse_record(key[-1],
2126
record, record_details, content,
2127
copy_base_content=multiple_versions)
2128
if multiple_versions:
2129
self._contents_map[component_id] = content
2131
# digest here is the digest from the last applied component.
2132
text = content.text()
2133
actual_sha = sha_strings(text)
2134
if actual_sha != digest:
2135
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2136
if multiple_versions:
2137
return self._contents_map[requested_key]
2141
def _wire_bytes(self):
2142
"""Get the bytes to put on the wire for 'key'.
2144
The first collection of bytes asked for returns the serialised
2145
raw_record_map and the additional details (key, parent) for key.
2146
Subsequent calls return just the additional details (key, parent).
2147
The wire storage_kind given for the first key is 'knit-delta-closure',
2148
For subsequent keys it is 'knit-delta-closure-ref'.
2150
:param key: A key from the content generator.
2151
:return: Bytes to put on the wire.
2154
# kind marker for dispatch on the far side,
2155
lines.append('knit-delta-closure')
2157
if self.vf._factory.annotated:
2158
lines.append('annotated')
2161
# then the list of keys
2162
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2163
if key not in self.nonlocal_keys]))
2164
# then the _raw_record_map in serialised form:
2166
# for each item in the map:
2168
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2169
# one line with method
2170
# one line with noeol
2171
# one line with next ('' for None)
2172
# one line with byte count of the record bytes
2174
for key, (record_bytes, (method, noeol), next) in \
2175
self._raw_record_map.iteritems():
2176
key_bytes = '\x00'.join(key)
2177
parents = self.global_map.get(key, None)
2179
parent_bytes = 'None:'
2181
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2182
method_bytes = method
2188
next_bytes = '\x00'.join(next)
2191
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2192
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2193
len(record_bytes), record_bytes))
2194
map_bytes = ''.join(map_byte_list)
2195
lines.append(map_bytes)
2196
bytes = '\n'.join(lines)
2200
class _VFContentMapGenerator(_ContentMapGenerator):
2201
"""Content map generator reading from a VersionedFiles object."""
2203
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2204
global_map=None, raw_record_map=None, ordering='unordered'):
2205
"""Create a _ContentMapGenerator.
2207
:param versioned_files: The versioned files that the texts are being
2209
:param keys: The keys to produce content maps for.
2210
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2211
which are known to not be in this knit, but rather in one of the
2213
:param global_map: The result of get_parent_map(keys) (or a supermap).
2214
This is required if get_record_stream() is to be used.
2215
:param raw_record_map: A unparsed raw record map to use for answering
2218
_ContentMapGenerator.__init__(self, ordering=ordering)
2219
# The vf to source data from
2220
self.vf = versioned_files
2222
self.keys = list(keys)
2223
# Keys known to be in fallback vfs objects
2224
if nonlocal_keys is None:
2225
self.nonlocal_keys = set()
2227
self.nonlocal_keys = frozenset(nonlocal_keys)
2228
# Parents data for keys to be returned in get_record_stream
2229
self.global_map = global_map
2230
# The chunked lists for self.keys in text form
2232
# A cache of KnitContent objects used in extracting texts.
2233
self._contents_map = {}
2234
# All the knit records needed to assemble the requested keys as full
2236
self._record_map = None
2237
if raw_record_map is None:
2238
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2241
self._raw_record_map = raw_record_map
2242
# the factory for parsing records
2243
self._factory = self.vf._factory
2246
class _NetworkContentMapGenerator(_ContentMapGenerator):
2247
"""Content map generator sourced from a network stream."""
2249
def __init__(self, bytes, line_end):
2250
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2252
self.global_map = {}
2253
self._raw_record_map = {}
2254
self._contents_map = {}
2255
self._record_map = None
2256
self.nonlocal_keys = []
2257
# Get access to record parsing facilities
2258
self.vf = KnitVersionedFiles(None, None)
2261
line_end = bytes.find('\n', start)
2262
line = bytes[start:line_end]
2263
start = line_end + 1
2264
if line == 'annotated':
2265
self._factory = KnitAnnotateFactory()
2267
self._factory = KnitPlainFactory()
2268
# list of keys to emit in get_record_stream
2269
line_end = bytes.find('\n', start)
2270
line = bytes[start:line_end]
2271
start = line_end + 1
2273
tuple(segment.split('\x00')) for segment in line.split('\t')
2275
# now a loop until the end. XXX: It would be nice if this was just a
2276
# bunch of the same records as get_record_stream(..., False) gives, but
2277
# there is a decent sized gap stopping that at the moment.
2281
line_end = bytes.find('\n', start)
2282
key = tuple(bytes[start:line_end].split('\x00'))
2283
start = line_end + 1
2284
# 1 line with parents (None: for None, '' for ())
2285
line_end = bytes.find('\n', start)
2286
line = bytes[start:line_end]
2291
[tuple(segment.split('\x00')) for segment in line.split('\t')
2293
self.global_map[key] = parents
2294
start = line_end + 1
2295
# one line with method
2296
line_end = bytes.find('\n', start)
2297
line = bytes[start:line_end]
2299
start = line_end + 1
2300
# one line with noeol
2301
line_end = bytes.find('\n', start)
2302
line = bytes[start:line_end]
2304
start = line_end + 1
2305
# one line with next ('' for None)
2306
line_end = bytes.find('\n', start)
2307
line = bytes[start:line_end]
2311
next = tuple(bytes[start:line_end].split('\x00'))
2312
start = line_end + 1
2313
# one line with byte count of the record bytes
2314
line_end = bytes.find('\n', start)
2315
line = bytes[start:line_end]
2317
start = line_end + 1
2319
record_bytes = bytes[start:start+count]
2320
start = start + count
2322
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2324
def get_record_stream(self):
2325
"""Get a record stream for for keys requested by the bytestream."""
2327
for key in self.keys:
2328
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2331
def _wire_bytes(self):
2335
class _KndxIndex(object):
2336
"""Manages knit index files
2338
The index is kept in memory and read on startup, to enable
2339
fast lookups of revision information. The cursor of the index
2340
file is always pointing to the end, making it easy to append
2343
_cache is a cache for fast mapping from version id to a Index
2346
_history is a cache for fast mapping from indexes to version ids.
2348
The index data format is dictionary compressed when it comes to
2349
parent references; a index entry may only have parents that with a
2350
lover index number. As a result, the index is topological sorted.
2352
Duplicate entries may be written to the index for a single version id
2353
if this is done then the latter one completely replaces the former:
2354
this allows updates to correct version and parent information.
2355
Note that the two entries may share the delta, and that successive
2356
annotations and references MUST point to the first entry.
2358
The index file on disc contains a header, followed by one line per knit
2359
record. The same revision can be present in an index file more than once.
2360
The first occurrence gets assigned a sequence number starting from 0.
2362
The format of a single line is
2363
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
2364
REVISION_ID is a utf8-encoded revision id
2365
FLAGS is a comma separated list of flags about the record. Values include
2366
no-eol, line-delta, fulltext.
2367
BYTE_OFFSET is the ascii representation of the byte offset in the data file
2368
that the compressed data starts at.
2369
LENGTH is the ascii representation of the length of the data file.
2370
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
2372
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
2373
revision id already in the knit that is a parent of REVISION_ID.
2374
The ' :' marker is the end of record marker.
2377
when a write is interrupted to the index file, it will result in a line
2378
that does not end in ' :'. If the ' :' is not present at the end of a line,
2379
or at the end of the file, then the record that is missing it will be
2380
ignored by the parser.
2382
When writing new records to the index file, the data is preceded by '\n'
2383
to ensure that records always start on new lines even if the last write was
2384
interrupted. As a result its normal for the last line in the index to be
2385
missing a trailing newline. One can be added with no harmful effects.
2387
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2388
where prefix is e.g. the (fileid,) for .texts instances or () for
2389
constant-mapped things like .revisions, and the old state is
2390
tuple(history_vector, cache_dict). This is used to prevent having an
2391
ABI change with the C extension that reads .kndx files.
2394
HEADER = "# bzr knit index 8\n"
2396
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2397
"""Create a _KndxIndex on transport using mapper."""
2398
self._transport = transport
2399
self._mapper = mapper
2400
self._get_scope = get_scope
2401
self._allow_writes = allow_writes
2402
self._is_locked = is_locked
2404
self.has_graph = True
2406
def add_records(self, records, random_id=False, missing_compression_parents=False):
2407
"""Add multiple records to the index.
2409
:param records: a list of tuples:
2410
(key, options, access_memo, parents).
2411
:param random_id: If True the ids being added were randomly generated
2412
and no check for existence will be performed.
2413
:param missing_compression_parents: If True the records being added are
2414
only compressed against texts already in the index (or inside
2415
records). If False the records all refer to unavailable texts (or
2416
texts inside records) as compression parents.
2418
if missing_compression_parents:
2419
# It might be nice to get the edge of the records. But keys isn't
2421
keys = sorted(record[0] for record in records)
2422
raise errors.RevisionNotPresent(keys, self)
2424
for record in records:
2427
path = self._mapper.map(key) + '.kndx'
2428
path_keys = paths.setdefault(path, (prefix, []))
2429
path_keys[1].append(record)
2430
for path in sorted(paths):
2431
prefix, path_keys = paths[path]
2432
self._load_prefixes([prefix])
2434
orig_history = self._kndx_cache[prefix][1][:]
2435
orig_cache = self._kndx_cache[prefix][0].copy()
2438
for key, options, (_, pos, size), parents in path_keys:
2440
# kndx indices cannot be parentless.
2442
line = "\n%s %s %s %s %s :" % (
2443
key[-1], ','.join(options), pos, size,
2444
self._dictionary_compress(parents))
2445
if type(line) is not str:
2446
raise AssertionError(
2447
'data must be utf8 was %s' % type(line))
2449
self._cache_key(key, options, pos, size, parents)
2450
if len(orig_history):
2451
self._transport.append_bytes(path, ''.join(lines))
2453
self._init_index(path, lines)
2455
# If any problems happen, restore the original values and re-raise
2456
self._kndx_cache[prefix] = (orig_cache, orig_history)
2459
def scan_unvalidated_index(self, graph_index):
2460
"""See _KnitGraphIndex.scan_unvalidated_index."""
2461
# Because kndx files do not support atomic insertion via separate index
2462
# files, they do not support this method.
2463
raise NotImplementedError(self.scan_unvalidated_index)
2465
def get_missing_compression_parents(self):
2466
"""See _KnitGraphIndex.get_missing_compression_parents."""
2467
# Because kndx files do not support atomic insertion via separate index
2468
# files, they do not support this method.
2469
raise NotImplementedError(self.get_missing_compression_parents)
2471
def _cache_key(self, key, options, pos, size, parent_keys):
2472
"""Cache a version record in the history array and index cache.
2474
This is inlined into _load_data for performance. KEEP IN SYNC.
2475
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2479
version_id = key[-1]
2480
# last-element only for compatibilty with the C load_data.
2481
parents = tuple(parent[-1] for parent in parent_keys)
2482
for parent in parent_keys:
2483
if parent[:-1] != prefix:
2484
raise ValueError("mismatched prefixes for %r, %r" % (
2486
cache, history = self._kndx_cache[prefix]
2487
# only want the _history index to reference the 1st index entry
2489
if version_id not in cache:
2490
index = len(history)
2491
history.append(version_id)
2493
index = cache[version_id][5]
2494
cache[version_id] = (version_id,
2501
def check_header(self, fp):
2502
line = fp.readline()
2504
# An empty file can actually be treated as though the file doesn't
2506
raise errors.NoSuchFile(self)
2507
if line != self.HEADER:
2508
raise KnitHeaderError(badline=line, filename=self)
2510
def _check_read(self):
2511
if not self._is_locked():
2512
raise errors.ObjectNotLocked(self)
2513
if self._get_scope() != self._scope:
2516
def _check_write_ok(self):
2517
"""Assert if not writes are permitted."""
2518
if not self._is_locked():
2519
raise errors.ObjectNotLocked(self)
2520
if self._get_scope() != self._scope:
2522
if self._mode != 'w':
2523
raise errors.ReadOnlyObjectDirtiedError(self)
2525
def get_build_details(self, keys):
2526
"""Get the method, index_memo and compression parent for keys.
2528
Ghosts are omitted from the result.
2530
:param keys: An iterable of keys.
2531
:return: A dict of key:(index_memo, compression_parent, parents,
2534
opaque structure to pass to read_records to extract the raw
2537
Content that this record is built upon, may be None
2539
Logical parents of this node
2541
extra information about the content which needs to be passed to
2542
Factory.parse_record
2544
parent_map = self.get_parent_map(keys)
2547
if key not in parent_map:
2549
method = self.get_method(key)
2550
parents = parent_map[key]
2551
if method == 'fulltext':
2552
compression_parent = None
2554
compression_parent = parents[0]
2555
noeol = 'no-eol' in self.get_options(key)
2556
index_memo = self.get_position(key)
2557
result[key] = (index_memo, compression_parent,
2558
parents, (method, noeol))
2561
def get_method(self, key):
2562
"""Return compression method of specified key."""
2563
options = self.get_options(key)
2564
if 'fulltext' in options:
2566
elif 'line-delta' in options:
2569
raise errors.KnitIndexUnknownMethod(self, options)
2571
def get_options(self, key):
2572
"""Return a list representing options.
2576
prefix, suffix = self._split_key(key)
2577
self._load_prefixes([prefix])
2579
return self._kndx_cache[prefix][0][suffix][1]
2581
raise RevisionNotPresent(key, self)
2583
def find_ancestry(self, keys):
2584
"""See CombinedGraphIndex.find_ancestry()"""
2585
prefixes = set(key[:-1] for key in keys)
2586
self._load_prefixes(prefixes)
2589
missing_keys = set()
2590
pending_keys = list(keys)
2591
# This assumes that keys will not reference parents in a different
2592
# prefix, which is accurate so far.
2594
key = pending_keys.pop()
2595
if key in parent_map:
2599
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2601
missing_keys.add(key)
2603
parent_keys = tuple([prefix + (suffix,)
2604
for suffix in suffix_parents])
2605
parent_map[key] = parent_keys
2606
pending_keys.extend([p for p in parent_keys
2607
if p not in parent_map])
2608
return parent_map, missing_keys
2610
def get_parent_map(self, keys):
2611
"""Get a map of the parents of keys.
2613
:param keys: The keys to look up parents for.
2614
:return: A mapping from keys to parents. Absent keys are absent from
2617
# Parse what we need to up front, this potentially trades off I/O
2618
# locality (.kndx and .knit in the same block group for the same file
2619
# id) for less checking in inner loops.
2620
prefixes = set(key[:-1] for key in keys)
2621
self._load_prefixes(prefixes)
2626
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2630
result[key] = tuple(prefix + (suffix,) for
2631
suffix in suffix_parents)
2634
def get_position(self, key):
2635
"""Return details needed to access the version.
2637
:return: a tuple (key, data position, size) to hand to the access
2638
logic to get the record.
2640
prefix, suffix = self._split_key(key)
2641
self._load_prefixes([prefix])
2642
entry = self._kndx_cache[prefix][0][suffix]
2643
return key, entry[2], entry[3]
2645
has_key = _mod_index._has_key_from_parent_map
2647
def _init_index(self, path, extra_lines=[]):
2648
"""Initialize an index."""
2650
sio.write(self.HEADER)
2651
sio.writelines(extra_lines)
2653
self._transport.put_file_non_atomic(path, sio,
2654
create_parent_dir=True)
2655
# self._create_parent_dir)
2656
# mode=self._file_mode,
2657
# dir_mode=self._dir_mode)
2660
"""Get all the keys in the collection.
2662
The keys are not ordered.
2665
# Identify all key prefixes.
2666
# XXX: A bit hacky, needs polish.
2667
if type(self._mapper) is ConstantMapper:
2671
for quoted_relpath in self._transport.iter_files_recursive():
2672
path, ext = os.path.splitext(quoted_relpath)
2674
prefixes = [self._mapper.unmap(path) for path in relpaths]
2675
self._load_prefixes(prefixes)
2676
for prefix in prefixes:
2677
for suffix in self._kndx_cache[prefix][1]:
2678
result.add(prefix + (suffix,))
2681
def _load_prefixes(self, prefixes):
2682
"""Load the indices for prefixes."""
2684
for prefix in prefixes:
2685
if prefix not in self._kndx_cache:
2686
# the load_data interface writes to these variables.
2689
self._filename = prefix
2691
path = self._mapper.map(prefix) + '.kndx'
2692
fp = self._transport.get(path)
2694
# _load_data may raise NoSuchFile if the target knit is
2696
_load_data(self, fp)
2699
self._kndx_cache[prefix] = (self._cache, self._history)
2704
self._kndx_cache[prefix] = ({}, [])
2705
if type(self._mapper) is ConstantMapper:
2706
# preserve behaviour for revisions.kndx etc.
2707
self._init_index(path)
2712
missing_keys = _mod_index._missing_keys_from_parent_map
2714
def _partition_keys(self, keys):
2715
"""Turn keys into a dict of prefix:suffix_list."""
2718
prefix_keys = result.setdefault(key[:-1], [])
2719
prefix_keys.append(key[-1])
2722
def _dictionary_compress(self, keys):
2723
"""Dictionary compress keys.
2725
:param keys: The keys to generate references to.
2726
:return: A string representation of keys. keys which are present are
2727
dictionary compressed, and others are emitted as fulltext with a
2733
prefix = keys[0][:-1]
2734
cache = self._kndx_cache[prefix][0]
2736
if key[:-1] != prefix:
2737
# kndx indices cannot refer across partitioned storage.
2738
raise ValueError("mismatched prefixes for %r" % keys)
2739
if key[-1] in cache:
2740
# -- inlined lookup() --
2741
result_list.append(str(cache[key[-1]][5]))
2742
# -- end lookup () --
2744
result_list.append('.' + key[-1])
2745
return ' '.join(result_list)
2747
def _reset_cache(self):
2748
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2749
# (cache_dict, history_vector) for parsed kndx files.
2750
self._kndx_cache = {}
2751
self._scope = self._get_scope()
2752
allow_writes = self._allow_writes()
2758
def _sort_keys_by_io(self, keys, positions):
2759
"""Figure out an optimal order to read the records for the given keys.
2761
Sort keys, grouped by index and sorted by position.
2763
:param keys: A list of keys whose records we want to read. This will be
2765
:param positions: A dict, such as the one returned by
2766
_get_components_positions()
2769
def get_sort_key(key):
2770
index_memo = positions[key][1]
2771
# Group by prefix and position. index_memo[0] is the key, so it is
2772
# (file_id, revision_id) and we don't want to sort on revision_id,
2773
# index_memo[1] is the position, and index_memo[2] is the size,
2774
# which doesn't matter for the sort
2775
return index_memo[0][:-1], index_memo[1]
2776
return keys.sort(key=get_sort_key)
2778
_get_total_build_size = _get_total_build_size
2780
def _split_key(self, key):
2781
"""Split key into a prefix and suffix."""
2782
return key[:-1], key[-1]
2785
class _KnitGraphIndex(object):
2786
"""A KnitVersionedFiles index layered on GraphIndex."""
2788
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2789
add_callback=None, track_external_parent_refs=False):
2790
"""Construct a KnitGraphIndex on a graph_index.
2792
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2793
:param is_locked: A callback to check whether the object should answer
2795
:param deltas: Allow delta-compressed records.
2796
:param parents: If True, record knits parents, if not do not record
2798
:param add_callback: If not None, allow additions to the index and call
2799
this callback with a list of added GraphIndex nodes:
2800
[(node, value, node_refs), ...]
2801
:param is_locked: A callback, returns True if the index is locked and
2803
:param track_external_parent_refs: If True, record all external parent
2804
references parents from added records. These can be retrieved
2805
later by calling get_missing_parents().
2807
self._add_callback = add_callback
2808
self._graph_index = graph_index
2809
self._deltas = deltas
2810
self._parents = parents
2811
if deltas and not parents:
2812
# XXX: TODO: Delta tree and parent graph should be conceptually
2814
raise KnitCorrupt(self, "Cannot do delta compression without "
2816
self.has_graph = parents
2817
self._is_locked = is_locked
2818
self._missing_compression_parents = set()
2819
if track_external_parent_refs:
2820
self._key_dependencies = _KeyRefs()
2822
self._key_dependencies = None
2825
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2827
def add_records(self, records, random_id=False,
2828
missing_compression_parents=False):
2829
"""Add multiple records to the index.
2831
This function does not insert data into the Immutable GraphIndex
2832
backing the KnitGraphIndex, instead it prepares data for insertion by
2833
the caller and checks that it is safe to insert then calls
2834
self._add_callback with the prepared GraphIndex nodes.
2836
:param records: a list of tuples:
2837
(key, options, access_memo, parents).
2838
:param random_id: If True the ids being added were randomly generated
2839
and no check for existence will be performed.
2840
:param missing_compression_parents: If True the records being added are
2841
only compressed against texts already in the index (or inside
2842
records). If False the records all refer to unavailable texts (or
2843
texts inside records) as compression parents.
2845
if not self._add_callback:
2846
raise errors.ReadOnlyError(self)
2847
# we hope there are no repositories with inconsistent parentage
2851
compression_parents = set()
2852
key_dependencies = self._key_dependencies
2853
for (key, options, access_memo, parents) in records:
2855
parents = tuple(parents)
2856
if key_dependencies is not None:
2857
key_dependencies.add_references(key, parents)
2858
index, pos, size = access_memo
2859
if 'no-eol' in options:
2863
value += "%d %d" % (pos, size)
2864
if not self._deltas:
2865
if 'line-delta' in options:
2866
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2869
if 'line-delta' in options:
2870
node_refs = (parents, (parents[0],))
2871
if missing_compression_parents:
2872
compression_parents.add(parents[0])
2874
node_refs = (parents, ())
2876
node_refs = (parents, )
2879
raise KnitCorrupt(self, "attempt to add node with parents "
2880
"in parentless index.")
2882
keys[key] = (value, node_refs)
2885
present_nodes = self._get_entries(keys)
2886
for (index, key, value, node_refs) in present_nodes:
2887
parents = node_refs[:1]
2888
# Sometimes these are passed as a list rather than a tuple
2889
passed = static_tuple.as_tuples(keys[key])
2890
passed_parents = passed[1][:1]
2891
if (value[0] != keys[key][0][0] or
2892
parents != passed_parents):
2893
node_refs = static_tuple.as_tuples(node_refs)
2894
raise KnitCorrupt(self, "inconsistent details in add_records"
2895
": %s %s" % ((value, node_refs), passed))
2899
for key, (value, node_refs) in keys.iteritems():
2900
result.append((key, value, node_refs))
2902
for key, (value, node_refs) in keys.iteritems():
2903
result.append((key, value))
2904
self._add_callback(result)
2905
if missing_compression_parents:
2906
# This may appear to be incorrect (it does not check for
2907
# compression parents that are in the existing graph index),
2908
# but such records won't have been buffered, so this is
2909
# actually correct: every entry when
2910
# missing_compression_parents==True either has a missing parent, or
2911
# a parent that is one of the keys in records.
2912
compression_parents.difference_update(keys)
2913
self._missing_compression_parents.update(compression_parents)
2914
# Adding records may have satisfied missing compression parents.
2915
self._missing_compression_parents.difference_update(keys)
2917
def scan_unvalidated_index(self, graph_index):
2918
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2920
This allows this _KnitGraphIndex to keep track of any missing
2921
compression parents we may want to have filled in to make those
2924
:param graph_index: A GraphIndex
2927
new_missing = graph_index.external_references(ref_list_num=1)
2928
new_missing.difference_update(self.get_parent_map(new_missing))
2929
self._missing_compression_parents.update(new_missing)
2930
if self._key_dependencies is not None:
2931
# Add parent refs from graph_index (and discard parent refs that
2932
# the graph_index has).
2933
for node in graph_index.iter_all_entries():
2934
self._key_dependencies.add_references(node[1], node[3][0])
2936
def get_missing_compression_parents(self):
2937
"""Return the keys of missing compression parents.
2939
Missing compression parents occur when a record stream was missing
2940
basis texts, or a index was scanned that had missing basis texts.
2942
return frozenset(self._missing_compression_parents)
2944
def get_missing_parents(self):
2945
"""Return the keys of missing parents."""
2946
# If updating this, you should also update
2947
# groupcompress._GCGraphIndex.get_missing_parents
2948
# We may have false positives, so filter those out.
2949
self._key_dependencies.satisfy_refs_for_keys(
2950
self.get_parent_map(self._key_dependencies.get_unsatisfied_refs()))
2951
return frozenset(self._key_dependencies.get_unsatisfied_refs())
2953
def _check_read(self):
2954
"""raise if reads are not permitted."""
2955
if not self._is_locked():
2956
raise errors.ObjectNotLocked(self)
2958
def _check_write_ok(self):
2959
"""Assert if writes are not permitted."""
2960
if not self._is_locked():
2961
raise errors.ObjectNotLocked(self)
2963
def _compression_parent(self, an_entry):
2964
# return the key that an_entry is compressed against, or None
2965
# Grab the second parent list (as deltas implies parents currently)
2966
compression_parents = an_entry[3][1]
2967
if not compression_parents:
2969
if len(compression_parents) != 1:
2970
raise AssertionError(
2971
"Too many compression parents: %r" % compression_parents)
2972
return compression_parents[0]
2974
def get_build_details(self, keys):
2975
"""Get the method, index_memo and compression parent for version_ids.
2977
Ghosts are omitted from the result.
2979
:param keys: An iterable of keys.
2980
:return: A dict of key:
2981
(index_memo, compression_parent, parents, record_details).
2983
opaque structure to pass to read_records to extract the raw
2986
Content that this record is built upon, may be None
2988
Logical parents of this node
2990
extra information about the content which needs to be passed to
2991
Factory.parse_record
2995
entries = self._get_entries(keys, False)
2996
for entry in entries:
2998
if not self._parents:
3001
parents = entry[3][0]
3002
if not self._deltas:
3003
compression_parent_key = None
3005
compression_parent_key = self._compression_parent(entry)
3006
noeol = (entry[2][0] == 'N')
3007
if compression_parent_key:
3008
method = 'line-delta'
3011
result[key] = (self._node_to_position(entry),
3012
compression_parent_key, parents,
3016
def _get_entries(self, keys, check_present=False):
3017
"""Get the entries for keys.
3019
:param keys: An iterable of index key tuples.
3024
for node in self._graph_index.iter_entries(keys):
3026
found_keys.add(node[1])
3028
# adapt parentless index to the rest of the code.
3029
for node in self._graph_index.iter_entries(keys):
3030
yield node[0], node[1], node[2], ()
3031
found_keys.add(node[1])
3033
missing_keys = keys.difference(found_keys)
3035
raise RevisionNotPresent(missing_keys.pop(), self)
3037
def get_method(self, key):
3038
"""Return compression method of specified key."""
3039
return self._get_method(self._get_node(key))
3041
def _get_method(self, node):
3042
if not self._deltas:
3044
if self._compression_parent(node):
3049
def _get_node(self, key):
3051
return list(self._get_entries([key]))[0]
3053
raise RevisionNotPresent(key, self)
3055
def get_options(self, key):
3056
"""Return a list representing options.
3060
node = self._get_node(key)
3061
options = [self._get_method(node)]
3062
if node[2][0] == 'N':
3063
options.append('no-eol')
3066
def find_ancestry(self, keys):
3067
"""See CombinedGraphIndex.find_ancestry()"""
3068
return self._graph_index.find_ancestry(keys, 0)
3070
def get_parent_map(self, keys):
3071
"""Get a map of the parents of keys.
3073
:param keys: The keys to look up parents for.
3074
:return: A mapping from keys to parents. Absent keys are absent from
3078
nodes = self._get_entries(keys)
3082
result[node[1]] = node[3][0]
3085
result[node[1]] = None
3088
def get_position(self, key):
3089
"""Return details needed to access the version.
3091
:return: a tuple (index, data position, size) to hand to the access
3092
logic to get the record.
3094
node = self._get_node(key)
3095
return self._node_to_position(node)
3097
has_key = _mod_index._has_key_from_parent_map
3100
"""Get all the keys in the collection.
3102
The keys are not ordered.
3105
return [node[1] for node in self._graph_index.iter_all_entries()]
3107
missing_keys = _mod_index._missing_keys_from_parent_map
3109
def _node_to_position(self, node):
3110
"""Convert an index value to position details."""
3111
bits = node[2][1:].split(' ')
3112
return node[0], int(bits[0]), int(bits[1])
3114
def _sort_keys_by_io(self, keys, positions):
3115
"""Figure out an optimal order to read the records for the given keys.
3117
Sort keys, grouped by index and sorted by position.
3119
:param keys: A list of keys whose records we want to read. This will be
3121
:param positions: A dict, such as the one returned by
3122
_get_components_positions()
3125
def get_index_memo(key):
3126
# index_memo is at offset [1]. It is made up of (GraphIndex,
3127
# position, size). GI is an object, which will be unique for each
3128
# pack file. This causes us to group by pack file, then sort by
3129
# position. Size doesn't matter, but it isn't worth breaking up the
3131
return positions[key][1]
3132
return keys.sort(key=get_index_memo)
3134
_get_total_build_size = _get_total_build_size
3137
class _KnitKeyAccess(object):
3138
"""Access to records in .knit files."""
3140
def __init__(self, transport, mapper):
3141
"""Create a _KnitKeyAccess with transport and mapper.
3143
:param transport: The transport the access object is rooted at.
3144
:param mapper: The mapper used to map keys to .knit files.
3146
self._transport = transport
3147
self._mapper = mapper
3149
def add_raw_records(self, key_sizes, raw_data):
3150
"""Add raw knit bytes to a storage area.
3152
The data is spooled to the container writer in one bytes-record per
3155
:param sizes: An iterable of tuples containing the key and size of each
3157
:param raw_data: A bytestring containing the data.
3158
:return: A list of memos to retrieve the record later. Each memo is an
3159
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3160
length), where the key is the record key.
3162
if type(raw_data) is not str:
3163
raise AssertionError(
3164
'data must be plain bytes was %s' % type(raw_data))
3167
# TODO: This can be tuned for writing to sftp and other servers where
3168
# append() is relatively expensive by grouping the writes to each key
3170
for key, size in key_sizes:
3171
path = self._mapper.map(key)
3173
base = self._transport.append_bytes(path + '.knit',
3174
raw_data[offset:offset+size])
3175
except errors.NoSuchFile:
3176
self._transport.mkdir(osutils.dirname(path))
3177
base = self._transport.append_bytes(path + '.knit',
3178
raw_data[offset:offset+size])
3182
result.append((key, base, size))
3186
"""Flush pending writes on this access object.
3188
For .knit files this is a no-op.
3192
def get_raw_records(self, memos_for_retrieval):
3193
"""Get the raw bytes for a records.
3195
:param memos_for_retrieval: An iterable containing the access memo for
3196
retrieving the bytes.
3197
:return: An iterator over the bytes of the records.
3199
# first pass, group into same-index request to minimise readv's issued.
3201
current_prefix = None
3202
for (key, offset, length) in memos_for_retrieval:
3203
if current_prefix == key[:-1]:
3204
current_list.append((offset, length))
3206
if current_prefix is not None:
3207
request_lists.append((current_prefix, current_list))
3208
current_prefix = key[:-1]
3209
current_list = [(offset, length)]
3210
# handle the last entry
3211
if current_prefix is not None:
3212
request_lists.append((current_prefix, current_list))
3213
for prefix, read_vector in request_lists:
3214
path = self._mapper.map(prefix) + '.knit'
3215
for pos, data in self._transport.readv(path, read_vector):
3219
class _DirectPackAccess(object):
3220
"""Access to data in one or more packs with less translation."""
3222
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
3223
"""Create a _DirectPackAccess object.
3225
:param index_to_packs: A dict mapping index objects to the transport
3226
and file names for obtaining data.
3227
:param reload_func: A function to call if we determine that the pack
3228
files have moved and we need to reload our caches. See
3229
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
3231
self._container_writer = None
3232
self._write_index = None
3233
self._indices = index_to_packs
3234
self._reload_func = reload_func
3235
self._flush_func = flush_func
3237
def add_raw_records(self, key_sizes, raw_data):
3238
"""Add raw knit bytes to a storage area.
3240
The data is spooled to the container writer in one bytes-record per
3243
:param sizes: An iterable of tuples containing the key and size of each
3245
:param raw_data: A bytestring containing the data.
3246
:return: A list of memos to retrieve the record later. Each memo is an
3247
opaque index memo. For _DirectPackAccess the memo is (index, pos,
3248
length), where the index field is the write_index object supplied
3249
to the PackAccess object.
3251
if type(raw_data) is not str:
3252
raise AssertionError(
3253
'data must be plain bytes was %s' % type(raw_data))
3256
for key, size in key_sizes:
3257
p_offset, p_length = self._container_writer.add_bytes_record(
3258
raw_data[offset:offset+size], [])
3260
result.append((self._write_index, p_offset, p_length))
3264
"""Flush pending writes on this access object.
3266
This will flush any buffered writes to a NewPack.
3268
if self._flush_func is not None:
3271
def get_raw_records(self, memos_for_retrieval):
3272
"""Get the raw bytes for a records.
3274
:param memos_for_retrieval: An iterable containing the (index, pos,
3275
length) memo for retrieving the bytes. The Pack access method
3276
looks up the pack to use for a given record in its index_to_pack
3278
:return: An iterator over the bytes of the records.
3280
# first pass, group into same-index requests
3282
current_index = None
3283
for (index, offset, length) in memos_for_retrieval:
3284
if current_index == index:
3285
current_list.append((offset, length))
3287
if current_index is not None:
3288
request_lists.append((current_index, current_list))
3289
current_index = index
3290
current_list = [(offset, length)]
3291
# handle the last entry
3292
if current_index is not None:
3293
request_lists.append((current_index, current_list))
3294
for index, offsets in request_lists:
3296
transport, path = self._indices[index]
3298
# A KeyError here indicates that someone has triggered an index
3299
# reload, and this index has gone missing, we need to start
3301
if self._reload_func is None:
3302
# If we don't have a _reload_func there is nothing that can
3305
raise errors.RetryWithNewPacks(index,
3306
reload_occurred=True,
3307
exc_info=sys.exc_info())
3309
reader = pack.make_readv_reader(transport, path, offsets)
3310
for names, read_func in reader.iter_records():
3311
yield read_func(None)
3312
except errors.NoSuchFile:
3313
# A NoSuchFile error indicates that a pack file has gone
3314
# missing on disk, we need to trigger a reload, and start over.
3315
if self._reload_func is None:
3317
raise errors.RetryWithNewPacks(transport.abspath(path),
3318
reload_occurred=False,
3319
exc_info=sys.exc_info())
3321
def set_writer(self, writer, index, transport_packname):
3322
"""Set a writer to use for adding data."""
3323
if index is not None:
3324
self._indices[index] = transport_packname
3325
self._container_writer = writer
3326
self._write_index = index
3328
def reload_or_raise(self, retry_exc):
3329
"""Try calling the reload function, or re-raise the original exception.
3331
This should be called after _DirectPackAccess raises a
3332
RetryWithNewPacks exception. This function will handle the common logic
3333
of determining when the error is fatal versus being temporary.
3334
It will also make sure that the original exception is raised, rather
3335
than the RetryWithNewPacks exception.
3337
If this function returns, then the calling function should retry
3338
whatever operation was being performed. Otherwise an exception will
3341
:param retry_exc: A RetryWithNewPacks exception.
3344
if self._reload_func is None:
3346
elif not self._reload_func():
3347
# The reload claimed that nothing changed
3348
if not retry_exc.reload_occurred:
3349
# If there wasn't an earlier reload, then we really were
3350
# expecting to find changes. We didn't find them, so this is a
3354
exc_class, exc_value, exc_traceback = retry_exc.exc_info
3355
raise exc_class, exc_value, exc_traceback
3358
def annotate_knit(knit, revision_id):
3359
"""Annotate a knit with no cached annotations.
3361
This implementation is for knits with no cached annotations.
3362
It will work for knits with cached annotations, but this is not
3365
annotator = _KnitAnnotator(knit)
3366
return iter(annotator.annotate_flat(revision_id))
3369
class _KnitAnnotator(annotate.Annotator):
3370
"""Build up the annotations for a text."""
3372
def __init__(self, vf):
3373
annotate.Annotator.__init__(self, vf)
3375
# TODO: handle Nodes which cannot be extracted
3376
# self._ghosts = set()
3378
# Map from (key, parent_key) => matching_blocks, should be 'use once'
3379
self._matching_blocks = {}
3381
# KnitContent objects
3382
self._content_objects = {}
3383
# The number of children that depend on this fulltext content object
3384
self._num_compression_children = {}
3385
# Delta records that need their compression parent before they can be
3387
self._pending_deltas = {}
3388
# Fulltext records that are waiting for their parents fulltexts before
3389
# they can be yielded for annotation
3390
self._pending_annotation = {}
3392
self._all_build_details = {}
3394
def _get_build_graph(self, key):
3395
"""Get the graphs for building texts and annotations.
3397
The data you need for creating a full text may be different than the
3398
data you need to annotate that text. (At a minimum, you need both
3399
parents to create an annotation, but only need 1 parent to generate the
3402
:return: A list of (key, index_memo) records, suitable for
3403
passing to read_records_iter to start reading in the raw data from
3406
pending = set([key])
3409
self._num_needed_children[key] = 1
3411
# get all pending nodes
3412
this_iteration = pending
3413
build_details = self._vf._index.get_build_details(this_iteration)
3414
self._all_build_details.update(build_details)
3415
# new_nodes = self._vf._index._get_entries(this_iteration)
3417
for key, details in build_details.iteritems():
3418
(index_memo, compression_parent, parent_keys,
3419
record_details) = details
3420
self._parent_map[key] = parent_keys
3421
self._heads_provider = None
3422
records.append((key, index_memo))
3423
# Do we actually need to check _annotated_lines?
3424
pending.update([p for p in parent_keys
3425
if p not in self._all_build_details])
3427
for parent_key in parent_keys:
3428
if parent_key in self._num_needed_children:
3429
self._num_needed_children[parent_key] += 1
3431
self._num_needed_children[parent_key] = 1
3432
if compression_parent:
3433
if compression_parent in self._num_compression_children:
3434
self._num_compression_children[compression_parent] += 1
3436
self._num_compression_children[compression_parent] = 1
3438
missing_versions = this_iteration.difference(build_details.keys())
3439
if missing_versions:
3440
for key in missing_versions:
3441
if key in self._parent_map and key in self._text_cache:
3442
# We already have this text ready, we just need to
3443
# yield it later so we get it annotated
3445
parent_keys = self._parent_map[key]
3446
for parent_key in parent_keys:
3447
if parent_key in self._num_needed_children:
3448
self._num_needed_children[parent_key] += 1
3450
self._num_needed_children[parent_key] = 1
3451
pending.update([p for p in parent_keys
3452
if p not in self._all_build_details])
3454
raise errors.RevisionNotPresent(key, self._vf)
3455
# Generally we will want to read the records in reverse order, because
3456
# we find the parent nodes after the children
3458
return records, ann_keys
3460
def _get_needed_texts(self, key, pb=None):
3461
# if True or len(self._vf._immediate_fallback_vfs) > 0:
3462
if len(self._vf._immediate_fallback_vfs) > 0:
3463
# If we have fallbacks, go to the generic path
3464
for v in annotate.Annotator._get_needed_texts(self, key, pb=pb):
3469
records, ann_keys = self._get_build_graph(key)
3470
for idx, (sub_key, text, num_lines) in enumerate(
3471
self._extract_texts(records)):
3473
pb.update('annotating', idx, len(records))
3474
yield sub_key, text, num_lines
3475
for sub_key in ann_keys:
3476
text = self._text_cache[sub_key]
3477
num_lines = len(text) # bad assumption
3478
yield sub_key, text, num_lines
3480
except errors.RetryWithNewPacks, e:
3481
self._vf._access.reload_or_raise(e)
3482
# The cached build_details are no longer valid
3483
self._all_build_details.clear()
3485
def _cache_delta_blocks(self, key, compression_parent, delta, lines):
3486
parent_lines = self._text_cache[compression_parent]
3487
blocks = list(KnitContent.get_line_delta_blocks(delta, parent_lines, lines))
3488
self._matching_blocks[(key, compression_parent)] = blocks
3490
def _expand_record(self, key, parent_keys, compression_parent, record,
3493
if compression_parent:
3494
if compression_parent not in self._content_objects:
3495
# Waiting for the parent
3496
self._pending_deltas.setdefault(compression_parent, []).append(
3497
(key, parent_keys, record, record_details))
3499
# We have the basis parent, so expand the delta
3500
num = self._num_compression_children[compression_parent]
3503
base_content = self._content_objects.pop(compression_parent)
3504
self._num_compression_children.pop(compression_parent)
3506
self._num_compression_children[compression_parent] = num
3507
base_content = self._content_objects[compression_parent]
3508
# It is tempting to want to copy_base_content=False for the last
3509
# child object. However, whenever noeol=False,
3510
# self._text_cache[parent_key] is content._lines. So mutating it
3511
# gives very bad results.
3512
# The alternative is to copy the lines into text cache, but then we
3513
# are copying anyway, so just do it here.
3514
content, delta = self._vf._factory.parse_record(
3515
key, record, record_details, base_content,
3516
copy_base_content=True)
3519
content, _ = self._vf._factory.parse_record(
3520
key, record, record_details, None)
3521
if self._num_compression_children.get(key, 0) > 0:
3522
self._content_objects[key] = content
3523
lines = content.text()
3524
self._text_cache[key] = lines
3525
if delta is not None:
3526
self._cache_delta_blocks(key, compression_parent, delta, lines)
3529
def _get_parent_annotations_and_matches(self, key, text, parent_key):
3530
"""Get the list of annotations for the parent, and the matching lines.
3532
:param text: The opaque value given by _get_needed_texts
3533
:param parent_key: The key for the parent text
3534
:return: (parent_annotations, matching_blocks)
3535
parent_annotations is a list as long as the number of lines in
3537
matching_blocks is a list of (parent_idx, text_idx, len) tuples
3538
indicating which lines match between the two texts
3540
block_key = (key, parent_key)
3541
if block_key in self._matching_blocks:
3542
blocks = self._matching_blocks.pop(block_key)
3543
parent_annotations = self._annotations_cache[parent_key]
3544
return parent_annotations, blocks
3545
return annotate.Annotator._get_parent_annotations_and_matches(self,
3546
key, text, parent_key)
3548
def _process_pending(self, key):
3549
"""The content for 'key' was just processed.
3551
Determine if there is any more pending work to be processed.
3554
if key in self._pending_deltas:
3555
compression_parent = key
3556
children = self._pending_deltas.pop(key)
3557
for child_key, parent_keys, record, record_details in children:
3558
lines = self._expand_record(child_key, parent_keys,
3560
record, record_details)
3561
if self._check_ready_for_annotations(child_key, parent_keys):
3562
to_return.append(child_key)
3563
# Also check any children that are waiting for this parent to be
3565
if key in self._pending_annotation:
3566
children = self._pending_annotation.pop(key)
3567
to_return.extend([c for c, p_keys in children
3568
if self._check_ready_for_annotations(c, p_keys)])
3571
def _check_ready_for_annotations(self, key, parent_keys):
3572
"""return true if this text is ready to be yielded.
3574
Otherwise, this will return False, and queue the text into
3575
self._pending_annotation
3577
for parent_key in parent_keys:
3578
if parent_key not in self._annotations_cache:
3579
# still waiting on at least one parent text, so queue it up
3580
# Note that if there are multiple parents, we need to wait
3582
self._pending_annotation.setdefault(parent_key,
3583
[]).append((key, parent_keys))
3587
def _extract_texts(self, records):
3588
"""Extract the various texts needed based on records"""
3589
# We iterate in the order read, rather than a strict order requested
3590
# However, process what we can, and put off to the side things that
3591
# still need parents, cleaning them up when those parents are
3594
# 1) As 'records' are read, see if we can expand these records into
3595
# Content objects (and thus lines)
3596
# 2) If a given line-delta is waiting on its compression parent, it
3597
# gets queued up into self._pending_deltas, otherwise we expand
3598
# it, and put it into self._text_cache and self._content_objects
3599
# 3) If we expanded the text, we will then check to see if all
3600
# parents have also been processed. If so, this text gets yielded,
3601
# else this record gets set aside into pending_annotation
3602
# 4) Further, if we expanded the text in (2), we will then check to
3603
# see if there are any children in self._pending_deltas waiting to
3604
# also be processed. If so, we go back to (2) for those
3605
# 5) Further again, if we yielded the text, we can then check if that
3606
# 'unlocks' any of the texts in pending_annotations, which should
3607
# then get yielded as well
3608
# Note that both steps 4 and 5 are 'recursive' in that unlocking one
3609
# compression child could unlock yet another, and yielding a fulltext
3610
# will also 'unlock' the children that are waiting on that annotation.
3611
# (Though also, unlocking 1 parent's fulltext, does not unlock a child
3612
# if other parents are also waiting.)
3613
# We want to yield content before expanding child content objects, so
3614
# that we know when we can re-use the content lines, and the annotation
3615
# code can know when it can stop caching fulltexts, as well.
3617
# Children that are missing their compression parent
3619
for (key, record, digest) in self._vf._read_records_iter(records):
3621
details = self._all_build_details[key]
3622
(_, compression_parent, parent_keys, record_details) = details
3623
lines = self._expand_record(key, parent_keys, compression_parent,
3624
record, record_details)
3626
# Pending delta should be queued up
3628
# At this point, we may be able to yield this content, if all
3629
# parents are also finished
3630
yield_this_text = self._check_ready_for_annotations(key,
3633
# All parents present
3634
yield key, lines, len(lines)
3635
to_process = self._process_pending(key)
3637
this_process = to_process
3639
for key in this_process:
3640
lines = self._text_cache[key]
3641
yield key, lines, len(lines)
3642
to_process.extend(self._process_pending(key))
3645
from bzrlib._knit_load_data_pyx import _load_data_c as _load_data
3646
except ImportError, e:
3647
osutils.failed_to_load_extension(e)
3648
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
