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
54
from __future__ import absolute_import
59
from .lazy_import import lazy_import
60
lazy_import(globals(), """
77
from breezy.repofmt import pack_repo
78
from breezy.i18n import gettext
93
from .osutils import (
105
from .versionedfile import (
107
AbsentContentFactory,
112
VersionedFilesWithFallbacks,
116
# TODO: Split out code specific to this format into an associated object.
118
# TODO: Can we put in some kind of value to check that the index and data
119
# files belong together?
121
# TODO: accommodate binaries, perhaps by storing a byte count
123
# TODO: function to check whole file
125
# TODO: atomically append data, then measure backwards from the cursor
126
# position after writing to work out where it was located. we may need to
127
# bypass python file buffering.
129
DATA_SUFFIX = '.knit'
130
INDEX_SUFFIX = '.kndx'
131
_STREAM_MIN_BUFFER_SIZE = 5*1024*1024
134
class KnitAdapter(object):
135
"""Base class for knit record adaption."""
137
def __init__(self, basis_vf):
138
"""Create an adapter which accesses full texts from basis_vf.
140
:param basis_vf: A versioned file to access basis texts of deltas from.
141
May be None for adapters that do not need to access basis texts.
143
self._data = KnitVersionedFiles(None, None)
144
self._annotate_factory = KnitAnnotateFactory()
145
self._plain_factory = KnitPlainFactory()
146
self._basis_vf = basis_vf
149
class FTAnnotatedToUnannotated(KnitAdapter):
150
"""An adapter from FT annotated knits to unannotated ones."""
152
def get_bytes(self, factory):
153
annotated_compressed_bytes = factory._raw_record
155
self._data._parse_record_unchecked(annotated_compressed_bytes)
156
content = self._annotate_factory.parse_fulltext(contents, rec[1])
157
size, bytes = self._data._record_to_data((rec[1],), rec[3], content.text())
161
class DeltaAnnotatedToUnannotated(KnitAdapter):
162
"""An adapter for deltas from annotated to unannotated."""
164
def get_bytes(self, factory):
165
annotated_compressed_bytes = factory._raw_record
167
self._data._parse_record_unchecked(annotated_compressed_bytes)
168
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
170
contents = self._plain_factory.lower_line_delta(delta)
171
size, bytes = self._data._record_to_data((rec[1],), rec[3], contents)
175
class FTAnnotatedToFullText(KnitAdapter):
176
"""An adapter from FT annotated knits to unannotated ones."""
178
def get_bytes(self, factory):
179
annotated_compressed_bytes = factory._raw_record
181
self._data._parse_record_unchecked(annotated_compressed_bytes)
182
content, delta = self._annotate_factory.parse_record(factory.key[-1],
183
contents, factory._build_details, None)
184
return ''.join(content.text())
187
class DeltaAnnotatedToFullText(KnitAdapter):
188
"""An adapter for deltas from annotated to unannotated."""
190
def get_bytes(self, factory):
191
annotated_compressed_bytes = factory._raw_record
193
self._data._parse_record_unchecked(annotated_compressed_bytes)
194
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
196
compression_parent = factory.parents[0]
197
basis_entry = next(self._basis_vf.get_record_stream(
198
[compression_parent], 'unordered', True))
199
if basis_entry.storage_kind == 'absent':
200
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
201
basis_chunks = basis_entry.get_bytes_as('chunked')
202
basis_lines = osutils.chunks_to_lines(basis_chunks)
203
# Manually apply the delta because we have one annotated content and
205
basis_content = PlainKnitContent(basis_lines, compression_parent)
206
basis_content.apply_delta(delta, rec[1])
207
basis_content._should_strip_eol = factory._build_details[1]
208
return ''.join(basis_content.text())
211
class FTPlainToFullText(KnitAdapter):
212
"""An adapter from FT plain knits to unannotated ones."""
214
def get_bytes(self, factory):
215
compressed_bytes = factory._raw_record
217
self._data._parse_record_unchecked(compressed_bytes)
218
content, delta = self._plain_factory.parse_record(factory.key[-1],
219
contents, factory._build_details, None)
220
return ''.join(content.text())
223
class DeltaPlainToFullText(KnitAdapter):
224
"""An adapter for deltas from annotated to unannotated."""
226
def get_bytes(self, factory):
227
compressed_bytes = factory._raw_record
229
self._data._parse_record_unchecked(compressed_bytes)
230
delta = self._plain_factory.parse_line_delta(contents, rec[1])
231
compression_parent = factory.parents[0]
232
# XXX: string splitting overhead.
233
basis_entry = next(self._basis_vf.get_record_stream(
234
[compression_parent], 'unordered', True))
235
if basis_entry.storage_kind == 'absent':
236
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
237
basis_chunks = basis_entry.get_bytes_as('chunked')
238
basis_lines = osutils.chunks_to_lines(basis_chunks)
239
basis_content = PlainKnitContent(basis_lines, compression_parent)
240
# Manually apply the delta because we have one annotated content and
242
content, _ = self._plain_factory.parse_record(rec[1], contents,
243
factory._build_details, basis_content)
244
return ''.join(content.text())
247
class KnitContentFactory(ContentFactory):
248
"""Content factory for streaming from knits.
250
:seealso ContentFactory:
253
def __init__(self, key, parents, build_details, sha1, raw_record,
254
annotated, knit=None, network_bytes=None):
255
"""Create a KnitContentFactory for key.
258
:param parents: The parents.
259
:param build_details: The build details as returned from
261
:param sha1: The sha1 expected from the full text of this object.
262
:param raw_record: The bytes of the knit data from disk.
263
:param annotated: True if the raw data is annotated.
264
:param network_bytes: None to calculate the network bytes on demand,
265
not-none if they are already known.
267
ContentFactory.__init__(self)
270
self.parents = parents
271
if build_details[0] == 'line-delta':
276
annotated_kind = 'annotated-'
279
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
280
self._raw_record = raw_record
281
self._network_bytes = network_bytes
282
self._build_details = build_details
285
def _create_network_bytes(self):
286
"""Create a fully serialised network version for transmission."""
287
# storage_kind, key, parents, Noeol, raw_record
288
key_bytes = '\x00'.join(self.key)
289
if self.parents is None:
290
parent_bytes = 'None:'
292
parent_bytes = '\t'.join('\x00'.join(key) for key in self.parents)
293
if self._build_details[1]:
297
network_bytes = "%s\n%s\n%s\n%s%s" % (self.storage_kind, key_bytes,
298
parent_bytes, noeol, self._raw_record)
299
self._network_bytes = network_bytes
301
def get_bytes_as(self, storage_kind):
302
if storage_kind == self.storage_kind:
303
if self._network_bytes is None:
304
self._create_network_bytes()
305
return self._network_bytes
306
if ('-ft-' in self.storage_kind and
307
storage_kind in ('chunked', 'fulltext')):
308
adapter_key = (self.storage_kind, 'fulltext')
309
adapter_factory = adapter_registry.get(adapter_key)
310
adapter = adapter_factory(None)
311
bytes = adapter.get_bytes(self)
312
if storage_kind == 'chunked':
316
if self._knit is not None:
317
# Not redundant with direct conversion above - that only handles
319
if storage_kind == 'chunked':
320
return self._knit.get_lines(self.key[0])
321
elif storage_kind == 'fulltext':
322
return self._knit.get_text(self.key[0])
323
raise errors.UnavailableRepresentation(self.key, storage_kind,
327
class LazyKnitContentFactory(ContentFactory):
328
"""A ContentFactory which can either generate full text or a wire form.
330
:seealso ContentFactory:
333
def __init__(self, key, parents, generator, first):
334
"""Create a LazyKnitContentFactory.
336
:param key: The key of the record.
337
:param parents: The parents of the record.
338
:param generator: A _ContentMapGenerator containing the record for this
340
:param first: Is this the first content object returned from generator?
341
if it is, its storage kind is knit-delta-closure, otherwise it is
342
knit-delta-closure-ref
345
self.parents = parents
347
self._generator = generator
348
self.storage_kind = "knit-delta-closure"
350
self.storage_kind = self.storage_kind + "-ref"
353
def get_bytes_as(self, storage_kind):
354
if storage_kind == self.storage_kind:
356
return self._generator._wire_bytes()
358
# all the keys etc are contained in the bytes returned in the
361
if storage_kind in ('chunked', 'fulltext'):
362
chunks = self._generator._get_one_work(self.key).text()
363
if storage_kind == 'chunked':
366
return ''.join(chunks)
367
raise errors.UnavailableRepresentation(self.key, storage_kind,
371
def knit_delta_closure_to_records(storage_kind, bytes, line_end):
372
"""Convert a network record to a iterator over stream records.
374
:param storage_kind: The storage kind of the record.
375
Must be 'knit-delta-closure'.
376
:param bytes: The bytes of the record on the network.
378
generator = _NetworkContentMapGenerator(bytes, line_end)
379
return generator.get_record_stream()
382
def knit_network_to_record(storage_kind, bytes, line_end):
383
"""Convert a network record to a record object.
385
:param storage_kind: The storage kind of the record.
386
:param bytes: The bytes of the record on the network.
389
line_end = bytes.find('\n', start)
390
key = tuple(bytes[start:line_end].split('\x00'))
392
line_end = bytes.find('\n', start)
393
parent_line = bytes[start:line_end]
394
if parent_line == 'None:':
398
[tuple(segment.split('\x00')) for segment in parent_line.split('\t')
401
noeol = bytes[start] == 'N'
402
if 'ft' in storage_kind:
405
method = 'line-delta'
406
build_details = (method, noeol)
408
raw_record = bytes[start:]
409
annotated = 'annotated' in storage_kind
410
return [KnitContentFactory(key, parents, build_details, None, raw_record,
411
annotated, network_bytes=bytes)]
414
class KnitContent(object):
415
"""Content of a knit version to which deltas can be applied.
417
This is always stored in memory as a list of lines with \\n at the end,
418
plus a flag saying if the final ending is really there or not, because that
419
corresponds to the on-disk knit representation.
423
self._should_strip_eol = False
425
def apply_delta(self, delta, new_version_id):
426
"""Apply delta to this object to become new_version_id."""
427
raise NotImplementedError(self.apply_delta)
429
def line_delta_iter(self, new_lines):
430
"""Generate line-based delta from this content to new_lines."""
431
new_texts = new_lines.text()
432
old_texts = self.text()
433
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
434
for tag, i1, i2, j1, j2 in s.get_opcodes():
437
# ofrom, oto, length, data
438
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
440
def line_delta(self, new_lines):
441
return list(self.line_delta_iter(new_lines))
444
def get_line_delta_blocks(knit_delta, source, target):
445
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
446
target_len = len(target)
449
for s_begin, s_end, t_len, new_text in knit_delta:
450
true_n = s_begin - s_pos
453
# knit deltas do not provide reliable info about whether the
454
# last line of a file matches, due to eol handling.
455
if source[s_pos + n -1] != target[t_pos + n -1]:
458
yield s_pos, t_pos, n
459
t_pos += t_len + true_n
461
n = target_len - t_pos
463
if source[s_pos + n -1] != target[t_pos + n -1]:
466
yield s_pos, t_pos, n
467
yield s_pos + (target_len - t_pos), target_len, 0
470
class AnnotatedKnitContent(KnitContent):
471
"""Annotated content."""
473
def __init__(self, lines):
474
KnitContent.__init__(self)
475
self._lines = list(lines)
478
"""Return a list of (origin, text) for each content line."""
479
lines = self._lines[:]
480
if self._should_strip_eol:
481
origin, last_line = lines[-1]
482
lines[-1] = (origin, last_line.rstrip('\n'))
485
def apply_delta(self, delta, new_version_id):
486
"""Apply delta to this object to become new_version_id."""
489
for start, end, count, delta_lines in delta:
490
lines[offset+start:offset+end] = delta_lines
491
offset = offset + (start - end) + count
495
lines = [text for origin, text in self._lines]
496
except ValueError as e:
497
# most commonly (only?) caused by the internal form of the knit
498
# missing annotation information because of a bug - see thread
500
raise KnitCorrupt(self,
501
"line in annotated knit missing annotation information: %s"
503
if self._should_strip_eol:
504
lines[-1] = lines[-1].rstrip('\n')
508
return AnnotatedKnitContent(self._lines)
511
class PlainKnitContent(KnitContent):
512
"""Unannotated content.
514
When annotate[_iter] is called on this content, the same version is reported
515
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
519
def __init__(self, lines, version_id):
520
KnitContent.__init__(self)
522
self._version_id = version_id
525
"""Return a list of (origin, text) for each content line."""
526
return [(self._version_id, line) for line in self._lines]
528
def apply_delta(self, delta, new_version_id):
529
"""Apply delta to this object to become new_version_id."""
532
for start, end, count, delta_lines in delta:
533
lines[offset+start:offset+end] = delta_lines
534
offset = offset + (start - end) + count
535
self._version_id = new_version_id
538
return PlainKnitContent(self._lines[:], self._version_id)
542
if self._should_strip_eol:
544
lines[-1] = lines[-1].rstrip('\n')
548
class _KnitFactory(object):
549
"""Base class for common Factory functions."""
551
def parse_record(self, version_id, record, record_details,
552
base_content, copy_base_content=True):
553
"""Parse a record into a full content object.
555
:param version_id: The official version id for this content
556
:param record: The data returned by read_records_iter()
557
:param record_details: Details about the record returned by
559
:param base_content: If get_build_details returns a compression_parent,
560
you must return a base_content here, else use None
561
:param copy_base_content: When building from the base_content, decide
562
you can either copy it and return a new object, or modify it in
564
:return: (content, delta) A Content object and possibly a line-delta,
567
method, noeol = record_details
568
if method == 'line-delta':
569
if copy_base_content:
570
content = base_content.copy()
572
content = base_content
573
delta = self.parse_line_delta(record, version_id)
574
content.apply_delta(delta, version_id)
576
content = self.parse_fulltext(record, version_id)
578
content._should_strip_eol = noeol
579
return (content, delta)
582
class KnitAnnotateFactory(_KnitFactory):
583
"""Factory for creating annotated Content objects."""
587
def make(self, lines, version_id):
588
num_lines = len(lines)
589
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
591
def parse_fulltext(self, content, version_id):
592
"""Convert fulltext to internal representation
594
fulltext content is of the format
595
revid(utf8) plaintext\n
596
internal representation is of the format:
599
# TODO: jam 20070209 The tests expect this to be returned as tuples,
600
# but the code itself doesn't really depend on that.
601
# Figure out a way to not require the overhead of turning the
602
# list back into tuples.
603
lines = (tuple(line.split(' ', 1)) for line in content)
604
return AnnotatedKnitContent(lines)
606
def parse_line_delta_iter(self, lines):
607
return iter(self.parse_line_delta(lines))
609
def parse_line_delta(self, lines, version_id, plain=False):
610
"""Convert a line based delta into internal representation.
612
line delta is in the form of:
613
intstart intend intcount
615
revid(utf8) newline\n
616
internal representation is
617
(start, end, count, [1..count tuples (revid, newline)])
619
:param plain: If True, the lines are returned as a plain
620
list without annotations, not as a list of (origin, content) tuples, i.e.
621
(start, end, count, [1..count newline])
627
def cache_and_return(line):
628
origin, text = line.split(' ', 1)
629
return cache.setdefault(origin, origin), text
631
# walk through the lines parsing.
632
# Note that the plain test is explicitly pulled out of the
633
# loop to minimise any performance impact
636
start, end, count = [int(n) for n in header.split(',')]
637
contents = [next(lines).split(' ', 1)[1] for _ in range(count)]
638
result.append((start, end, count, contents))
641
start, end, count = [int(n) for n in header.split(',')]
642
contents = [tuple(next(lines).split(' ', 1))
643
for _ in range(count)]
644
result.append((start, end, count, contents))
647
def get_fulltext_content(self, lines):
648
"""Extract just the content lines from a fulltext."""
649
return (line.split(' ', 1)[1] for line in lines)
651
def get_linedelta_content(self, lines):
652
"""Extract just the content from a line delta.
654
This doesn't return all of the extra information stored in a delta.
655
Only the actual content lines.
659
header = header.split(',')
660
count = int(header[2])
661
for _ in range(count):
662
origin, text = next(lines).split(' ', 1)
665
def lower_fulltext(self, content):
666
"""convert a fulltext content record into a serializable form.
668
see parse_fulltext which this inverts.
670
return ['%s %s' % (o, t) for o, t in content._lines]
672
def lower_line_delta(self, delta):
673
"""convert a delta into a serializable form.
675
See parse_line_delta which this inverts.
677
# TODO: jam 20070209 We only do the caching thing to make sure that
678
# the origin is a valid utf-8 line, eventually we could remove it
680
for start, end, c, lines in delta:
681
out.append('%d,%d,%d\n' % (start, end, c))
682
out.extend(origin + ' ' + text
683
for origin, text in lines)
686
def annotate(self, knit, key):
687
content = knit._get_content(key)
688
# adjust for the fact that serialised annotations are only key suffixes
690
if isinstance(key, tuple):
692
origins = content.annotate()
694
for origin, line in origins:
695
result.append((prefix + (origin,), line))
698
# XXX: This smells a bit. Why would key ever be a non-tuple here?
699
# Aren't keys defined to be tuples? -- spiv 20080618
700
return content.annotate()
703
class KnitPlainFactory(_KnitFactory):
704
"""Factory for creating plain Content objects."""
708
def make(self, lines, version_id):
709
return PlainKnitContent(lines, version_id)
711
def parse_fulltext(self, content, version_id):
712
"""This parses an unannotated fulltext.
714
Note that this is not a noop - the internal representation
715
has (versionid, line) - its just a constant versionid.
717
return self.make(content, version_id)
719
def parse_line_delta_iter(self, lines, version_id):
721
num_lines = len(lines)
722
while cur < num_lines:
725
start, end, c = [int(n) for n in header.split(',')]
726
yield start, end, c, lines[cur:cur+c]
729
def parse_line_delta(self, lines, version_id):
730
return list(self.parse_line_delta_iter(lines, version_id))
732
def get_fulltext_content(self, lines):
733
"""Extract just the content lines from a fulltext."""
736
def get_linedelta_content(self, lines):
737
"""Extract just the content from a line delta.
739
This doesn't return all of the extra information stored in a delta.
740
Only the actual content lines.
744
header = header.split(',')
745
count = int(header[2])
746
for _ in range(count):
749
def lower_fulltext(self, content):
750
return content.text()
752
def lower_line_delta(self, delta):
754
for start, end, c, lines in delta:
755
out.append('%d,%d,%d\n' % (start, end, c))
759
def annotate(self, knit, key):
760
annotator = _KnitAnnotator(knit)
761
return annotator.annotate_flat(key)
765
def make_file_factory(annotated, mapper):
766
"""Create a factory for creating a file based KnitVersionedFiles.
768
This is only functional enough to run interface tests, it doesn't try to
769
provide a full pack environment.
771
:param annotated: knit annotations are wanted.
772
:param mapper: The mapper from keys to paths.
774
def factory(transport):
775
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
776
access = _KnitKeyAccess(transport, mapper)
777
return KnitVersionedFiles(index, access, annotated=annotated)
781
def make_pack_factory(graph, delta, keylength):
782
"""Create a factory for creating a pack based VersionedFiles.
784
This is only functional enough to run interface tests, it doesn't try to
785
provide a full pack environment.
787
:param graph: Store a graph.
788
:param delta: Delta compress contents.
789
:param keylength: How long should keys be.
791
def factory(transport):
792
parents = graph or delta
798
max_delta_chain = 200
801
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
802
key_elements=keylength)
803
stream = transport.open_write_stream('newpack')
804
writer = pack.ContainerWriter(stream.write)
806
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
807
deltas=delta, add_callback=graph_index.add_nodes)
808
access = pack_repo._DirectPackAccess({})
809
access.set_writer(writer, graph_index, (transport, 'newpack'))
810
result = KnitVersionedFiles(index, access,
811
max_delta_chain=max_delta_chain)
812
result.stream = stream
813
result.writer = writer
818
def cleanup_pack_knit(versioned_files):
819
versioned_files.stream.close()
820
versioned_files.writer.end()
823
def _get_total_build_size(self, keys, positions):
824
"""Determine the total bytes to build these keys.
826
(helper function because _KnitGraphIndex and _KndxIndex work the same, but
827
don't inherit from a common base.)
829
:param keys: Keys that we want to build
830
:param positions: dict of {key, (info, index_memo, comp_parent)} (such
831
as returned by _get_components_positions)
832
:return: Number of bytes to build those keys
834
all_build_index_memos = {}
838
for key in build_keys:
839
# This is mostly for the 'stacked' case
840
# Where we will be getting the data from a fallback
841
if key not in positions:
843
_, index_memo, compression_parent = positions[key]
844
all_build_index_memos[key] = index_memo
845
if compression_parent not in all_build_index_memos:
846
next_keys.add(compression_parent)
847
build_keys = next_keys
848
return sum(index_memo[2]
849
for index_memo in viewvalues(all_build_index_memos))
852
class KnitVersionedFiles(VersionedFilesWithFallbacks):
853
"""Storage for many versioned files using knit compression.
855
Backend storage is managed by indices and data objects.
857
:ivar _index: A _KnitGraphIndex or similar that can describe the
858
parents, graph, compression and data location of entries in this
859
KnitVersionedFiles. Note that this is only the index for
860
*this* vfs; if there are fallbacks they must be queried separately.
863
def __init__(self, index, data_access, max_delta_chain=200,
864
annotated=False, reload_func=None):
865
"""Create a KnitVersionedFiles with index and data_access.
867
:param index: The index for the knit data.
868
:param data_access: The access object to store and retrieve knit
870
:param max_delta_chain: The maximum number of deltas to permit during
871
insertion. Set to 0 to prohibit the use of deltas.
872
:param annotated: Set to True to cause annotations to be calculated and
873
stored during insertion.
874
:param reload_func: An function that can be called if we think we need
875
to reload the pack listing and try again. See
876
'breezy.repofmt.pack_repo.AggregateIndex' for the signature.
879
self._access = data_access
880
self._max_delta_chain = max_delta_chain
882
self._factory = KnitAnnotateFactory()
884
self._factory = KnitPlainFactory()
885
self._immediate_fallback_vfs = []
886
self._reload_func = reload_func
889
return "%s(%r, %r)" % (
890
self.__class__.__name__,
894
def without_fallbacks(self):
895
"""Return a clone of this object without any fallbacks configured."""
896
return KnitVersionedFiles(self._index, self._access,
897
self._max_delta_chain, self._factory.annotated,
900
def add_fallback_versioned_files(self, a_versioned_files):
901
"""Add a source of texts for texts not present in this knit.
903
:param a_versioned_files: A VersionedFiles object.
905
self._immediate_fallback_vfs.append(a_versioned_files)
907
def add_lines(self, key, parents, lines, parent_texts=None,
908
left_matching_blocks=None, nostore_sha=None, random_id=False,
910
"""See VersionedFiles.add_lines()."""
911
self._index._check_write_ok()
912
self._check_add(key, lines, random_id, check_content)
914
# The caller might pass None if there is no graph data, but kndx
915
# indexes can't directly store that, so we give them
916
# an empty tuple instead.
918
line_bytes = ''.join(lines)
919
return self._add(key, lines, parents,
920
parent_texts, left_matching_blocks, nostore_sha, random_id,
921
line_bytes=line_bytes)
923
def _add_text(self, key, parents, text, nostore_sha=None, random_id=False):
924
"""See VersionedFiles._add_text()."""
925
self._index._check_write_ok()
926
self._check_add(key, None, random_id, check_content=False)
927
if text.__class__ is not str:
928
raise errors.BzrBadParameterUnicode("text")
930
# The caller might pass None if there is no graph data, but kndx
931
# indexes can't directly store that, so we give them
932
# an empty tuple instead.
934
return self._add(key, None, parents,
935
None, None, nostore_sha, random_id,
938
def _add(self, key, lines, parents, parent_texts,
939
left_matching_blocks, nostore_sha, random_id,
941
"""Add a set of lines on top of version specified by parents.
943
Any versions not present will be converted into ghosts.
945
:param lines: A list of strings where each one is a single line (has a
946
single newline at the end of the string) This is now optional
947
(callers can pass None). It is left in its location for backwards
948
compatibility. It should ''.join(lines) must == line_bytes
949
:param line_bytes: A single string containing the content
951
We pass both lines and line_bytes because different routes bring the
952
values to this function. And for memory efficiency, we don't want to
953
have to split/join on-demand.
955
# first thing, if the content is something we don't need to store, find
957
digest = sha_string(line_bytes)
958
if nostore_sha == digest:
959
raise errors.ExistingContent
962
if parent_texts is None:
964
# Do a single query to ascertain parent presence; we only compress
965
# against parents in the same kvf.
966
present_parent_map = self._index.get_parent_map(parents)
967
for parent in parents:
968
if parent in present_parent_map:
969
present_parents.append(parent)
971
# Currently we can only compress against the left most present parent.
972
if (len(present_parents) == 0 or
973
present_parents[0] != parents[0]):
976
# To speed the extract of texts the delta chain is limited
977
# to a fixed number of deltas. This should minimize both
978
# I/O and the time spend applying deltas.
979
delta = self._check_should_delta(present_parents[0])
981
text_length = len(line_bytes)
984
# Note: line_bytes is not modified to add a newline, that is tracked
985
# via the no_eol flag. 'lines' *is* modified, because that is the
986
# general values needed by the Content code.
987
if line_bytes and line_bytes[-1] != '\n':
988
options.append('no-eol')
990
# Copy the existing list, or create a new one
992
lines = osutils.split_lines(line_bytes)
995
# Replace the last line with one that ends in a final newline
996
lines[-1] = lines[-1] + '\n'
998
lines = osutils.split_lines(line_bytes)
1000
for element in key[:-1]:
1001
if not isinstance(element, str):
1002
raise TypeError("key contains non-strings: %r" % (key,))
1004
key = key[:-1] + ('sha1:' + digest,)
1005
elif not isinstance(key[-1], str):
1006
raise TypeError("key contains non-strings: %r" % (key,))
1007
# Knit hunks are still last-element only
1008
version_id = key[-1]
1009
content = self._factory.make(lines, version_id)
1011
# Hint to the content object that its text() call should strip the
1013
content._should_strip_eol = True
1014
if delta or (self._factory.annotated and len(present_parents) > 0):
1015
# Merge annotations from parent texts if needed.
1016
delta_hunks = self._merge_annotations(content, present_parents,
1017
parent_texts, delta, self._factory.annotated,
1018
left_matching_blocks)
1021
options.append('line-delta')
1022
store_lines = self._factory.lower_line_delta(delta_hunks)
1023
size, bytes = self._record_to_data(key, digest,
1026
options.append('fulltext')
1027
# isinstance is slower and we have no hierarchy.
1028
if self._factory.__class__ is KnitPlainFactory:
1029
# Use the already joined bytes saving iteration time in
1031
dense_lines = [line_bytes]
1033
dense_lines.append('\n')
1034
size, bytes = self._record_to_data(key, digest,
1037
# get mixed annotation + content and feed it into the
1039
store_lines = self._factory.lower_fulltext(content)
1040
size, bytes = self._record_to_data(key, digest,
1043
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
1044
self._index.add_records(
1045
((key, options, access_memo, parents),),
1046
random_id=random_id)
1047
return digest, text_length, content
1049
def annotate(self, key):
1050
"""See VersionedFiles.annotate."""
1051
return self._factory.annotate(self, key)
1053
def get_annotator(self):
1054
return _KnitAnnotator(self)
1056
def check(self, progress_bar=None, keys=None):
1057
"""See VersionedFiles.check()."""
1059
return self._logical_check()
1061
# At the moment, check does not extra work over get_record_stream
1062
return self.get_record_stream(keys, 'unordered', True)
1064
def _logical_check(self):
1065
# This doesn't actually test extraction of everything, but that will
1066
# impact 'bzr check' substantially, and needs to be integrated with
1067
# care. However, it does check for the obvious problem of a delta with
1069
keys = self._index.keys()
1070
parent_map = self.get_parent_map(keys)
1072
if self._index.get_method(key) != 'fulltext':
1073
compression_parent = parent_map[key][0]
1074
if compression_parent not in parent_map:
1075
raise errors.KnitCorrupt(self,
1076
"Missing basis parent %s for %s" % (
1077
compression_parent, key))
1078
for fallback_vfs in self._immediate_fallback_vfs:
1079
fallback_vfs.check()
1081
def _check_add(self, key, lines, random_id, check_content):
1082
"""check that version_id and lines are safe to add."""
1083
version_id = key[-1]
1084
if version_id is not None:
1085
if contains_whitespace(version_id):
1086
raise InvalidRevisionId(version_id, self)
1087
self.check_not_reserved_id(version_id)
1088
# TODO: If random_id==False and the key is already present, we should
1089
# probably check that the existing content is identical to what is
1090
# being inserted, and otherwise raise an exception. This would make
1091
# the bundle code simpler.
1093
self._check_lines_not_unicode(lines)
1094
self._check_lines_are_lines(lines)
1096
def _check_header(self, key, line):
1097
rec = self._split_header(line)
1098
self._check_header_version(rec, key[-1])
1101
def _check_header_version(self, rec, version_id):
1102
"""Checks the header version on original format knit records.
1104
These have the last component of the key embedded in the record.
1106
if rec[1] != version_id:
1107
raise KnitCorrupt(self,
1108
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1110
def _check_should_delta(self, parent):
1111
"""Iterate back through the parent listing, looking for a fulltext.
1113
This is used when we want to decide whether to add a delta or a new
1114
fulltext. It searches for _max_delta_chain parents. When it finds a
1115
fulltext parent, it sees if the total size of the deltas leading up to
1116
it is large enough to indicate that we want a new full text anyway.
1118
Return True if we should create a new delta, False if we should use a
1122
fulltext_size = None
1123
for count in range(self._max_delta_chain):
1125
# Note that this only looks in the index of this particular
1126
# KnitVersionedFiles, not in the fallbacks. This ensures that
1127
# we won't store a delta spanning physical repository
1129
build_details = self._index.get_build_details([parent])
1130
parent_details = build_details[parent]
1131
except (RevisionNotPresent, KeyError) as e:
1132
# Some basis is not locally present: always fulltext
1134
index_memo, compression_parent, _, _ = parent_details
1135
_, _, size = index_memo
1136
if compression_parent is None:
1137
fulltext_size = size
1140
# We don't explicitly check for presence because this is in an
1141
# inner loop, and if it's missing it'll fail anyhow.
1142
parent = compression_parent
1144
# We couldn't find a fulltext, so we must create a new one
1146
# Simple heuristic - if the total I/O wold be greater as a delta than
1147
# the originally installed fulltext, we create a new fulltext.
1148
return fulltext_size > delta_size
1150
def _build_details_to_components(self, build_details):
1151
"""Convert a build_details tuple to a position tuple."""
1152
# record_details, access_memo, compression_parent
1153
return build_details[3], build_details[0], build_details[1]
1155
def _get_components_positions(self, keys, allow_missing=False):
1156
"""Produce a map of position data for the components of keys.
1158
This data is intended to be used for retrieving the knit records.
1160
A dict of key to (record_details, index_memo, next, parents) is
1163
* method is the way referenced data should be applied.
1164
* index_memo is the handle to pass to the data access to actually get
1166
* next is the build-parent of the version, or None for fulltexts.
1167
* parents is the version_ids of the parents of this version
1169
:param allow_missing: If True do not raise an error on a missing
1170
component, just ignore it.
1173
pending_components = keys
1174
while pending_components:
1175
build_details = self._index.get_build_details(pending_components)
1176
current_components = set(pending_components)
1177
pending_components = set()
1178
for key, details in viewitems(build_details):
1179
(index_memo, compression_parent, parents,
1180
record_details) = details
1181
method = record_details[0]
1182
if compression_parent is not None:
1183
pending_components.add(compression_parent)
1184
component_data[key] = self._build_details_to_components(details)
1185
missing = current_components.difference(build_details)
1186
if missing and not allow_missing:
1187
raise errors.RevisionNotPresent(missing.pop(), self)
1188
return component_data
1190
def _get_content(self, key, parent_texts={}):
1191
"""Returns a content object that makes up the specified
1193
cached_version = parent_texts.get(key, None)
1194
if cached_version is not None:
1195
# Ensure the cache dict is valid.
1196
if not self.get_parent_map([key]):
1197
raise RevisionNotPresent(key, self)
1198
return cached_version
1199
generator = _VFContentMapGenerator(self, [key])
1200
return generator._get_content(key)
1202
def get_parent_map(self, keys):
1203
"""Get a map of the graph parents of keys.
1205
:param keys: The keys to look up parents for.
1206
:return: A mapping from keys to parents. Absent keys are absent from
1209
return self._get_parent_map_with_sources(keys)[0]
1211
def _get_parent_map_with_sources(self, keys):
1212
"""Get a map of the parents of keys.
1214
:param keys: The keys to look up parents for.
1215
:return: A tuple. The first element is a mapping from keys to parents.
1216
Absent keys are absent from the mapping. The second element is a
1217
list with the locations each key was found in. The first element
1218
is the in-this-knit parents, the second the first fallback source,
1222
sources = [self._index] + self._immediate_fallback_vfs
1225
for source in sources:
1228
new_result = source.get_parent_map(missing)
1229
source_results.append(new_result)
1230
result.update(new_result)
1231
missing.difference_update(set(new_result))
1232
return result, source_results
1234
def _get_record_map(self, keys, allow_missing=False):
1235
"""Produce a dictionary of knit records.
1237
:return: {key:(record, record_details, digest, next)}
1239
* record: data returned from read_records (a KnitContentobject)
1240
* record_details: opaque information to pass to parse_record
1241
* digest: SHA1 digest of the full text after all steps are done
1242
* next: build-parent of the version, i.e. the leftmost ancestor.
1243
Will be None if the record is not a delta.
1245
:param keys: The keys to build a map for
1246
:param allow_missing: If some records are missing, rather than
1247
error, just return the data that could be generated.
1249
raw_map = self._get_record_map_unparsed(keys,
1250
allow_missing=allow_missing)
1251
return self._raw_map_to_record_map(raw_map)
1253
def _raw_map_to_record_map(self, raw_map):
1254
"""Parse the contents of _get_record_map_unparsed.
1256
:return: see _get_record_map.
1260
data, record_details, next = raw_map[key]
1261
content, digest = self._parse_record(key[-1], data)
1262
result[key] = content, record_details, digest, next
1265
def _get_record_map_unparsed(self, keys, allow_missing=False):
1266
"""Get the raw data for reconstructing keys without parsing it.
1268
:return: A dict suitable for parsing via _raw_map_to_record_map.
1269
key-> raw_bytes, (method, noeol), compression_parent
1271
# This retries the whole request if anything fails. Potentially we
1272
# could be a bit more selective. We could track the keys whose records
1273
# we have successfully found, and then only request the new records
1274
# from there. However, _get_components_positions grabs the whole build
1275
# chain, which means we'll likely try to grab the same records again
1276
# anyway. Also, can the build chains change as part of a pack
1277
# operation? We wouldn't want to end up with a broken chain.
1280
position_map = self._get_components_positions(keys,
1281
allow_missing=allow_missing)
1282
# key = component_id, r = record_details, i_m = index_memo,
1284
records = [(key, i_m) for key, (r, i_m, n)
1285
in viewitems(position_map)]
1286
# Sort by the index memo, so that we request records from the
1287
# same pack file together, and in forward-sorted order
1288
records.sort(key=operator.itemgetter(1))
1290
for key, data in self._read_records_iter_unchecked(records):
1291
(record_details, index_memo, next) = position_map[key]
1292
raw_record_map[key] = data, record_details, next
1293
return raw_record_map
1294
except errors.RetryWithNewPacks as e:
1295
self._access.reload_or_raise(e)
1298
def _split_by_prefix(cls, keys):
1299
"""For the given keys, split them up based on their prefix.
1301
To keep memory pressure somewhat under control, split the
1302
requests back into per-file-id requests, otherwise "bzr co"
1303
extracts the full tree into memory before writing it to disk.
1304
This should be revisited if _get_content_maps() can ever cross
1307
The keys for a given file_id are kept in the same relative order.
1308
Ordering between file_ids is not, though prefix_order will return the
1309
order that the key was first seen.
1311
:param keys: An iterable of key tuples
1312
:return: (split_map, prefix_order)
1313
split_map A dictionary mapping prefix => keys
1314
prefix_order The order that we saw the various prefixes
1316
split_by_prefix = {}
1324
if prefix in split_by_prefix:
1325
split_by_prefix[prefix].append(key)
1327
split_by_prefix[prefix] = [key]
1328
prefix_order.append(prefix)
1329
return split_by_prefix, prefix_order
1331
def _group_keys_for_io(self, keys, non_local_keys, positions,
1332
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1333
"""For the given keys, group them into 'best-sized' requests.
1335
The idea is to avoid making 1 request per file, but to never try to
1336
unpack an entire 1.5GB source tree in a single pass. Also when
1337
possible, we should try to group requests to the same pack file
1340
:return: list of (keys, non_local) tuples that indicate what keys
1341
should be fetched next.
1343
# TODO: Ideally we would group on 2 factors. We want to extract texts
1344
# from the same pack file together, and we want to extract all
1345
# the texts for a given build-chain together. Ultimately it
1346
# probably needs a better global view.
1347
total_keys = len(keys)
1348
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1349
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1351
cur_non_local = set()
1355
for prefix in prefix_order:
1356
keys = prefix_split_keys[prefix]
1357
non_local = prefix_split_non_local_keys.get(prefix, [])
1359
this_size = self._index._get_total_build_size(keys, positions)
1360
cur_size += this_size
1361
cur_keys.extend(keys)
1362
cur_non_local.update(non_local)
1363
if cur_size > _min_buffer_size:
1364
result.append((cur_keys, cur_non_local))
1365
sizes.append(cur_size)
1367
cur_non_local = set()
1370
result.append((cur_keys, cur_non_local))
1371
sizes.append(cur_size)
1374
def get_record_stream(self, keys, ordering, include_delta_closure):
1375
"""Get a stream of records for keys.
1377
:param keys: The keys to include.
1378
:param ordering: Either 'unordered' or 'topological'. A topologically
1379
sorted stream has compression parents strictly before their
1381
:param include_delta_closure: If True then the closure across any
1382
compression parents will be included (in the opaque data).
1383
:return: An iterator of ContentFactory objects, each of which is only
1384
valid until the iterator is advanced.
1386
# keys might be a generator
1390
if not self._index.has_graph:
1391
# Cannot sort when no graph has been stored.
1392
ordering = 'unordered'
1394
remaining_keys = keys
1397
keys = set(remaining_keys)
1398
for content_factory in self._get_remaining_record_stream(keys,
1399
ordering, include_delta_closure):
1400
remaining_keys.discard(content_factory.key)
1401
yield content_factory
1403
except errors.RetryWithNewPacks as e:
1404
self._access.reload_or_raise(e)
1406
def _get_remaining_record_stream(self, keys, ordering,
1407
include_delta_closure):
1408
"""This function is the 'retry' portion for get_record_stream."""
1409
if include_delta_closure:
1410
positions = self._get_components_positions(keys, allow_missing=True)
1412
build_details = self._index.get_build_details(keys)
1414
# (record_details, access_memo, compression_parent_key)
1415
positions = dict((key, self._build_details_to_components(details))
1416
for key, details in viewitems(build_details))
1417
absent_keys = keys.difference(set(positions))
1418
# There may be more absent keys : if we're missing the basis component
1419
# and are trying to include the delta closure.
1420
# XXX: We should not ever need to examine remote sources because we do
1421
# not permit deltas across versioned files boundaries.
1422
if include_delta_closure:
1423
needed_from_fallback = set()
1424
# Build up reconstructable_keys dict. key:True in this dict means
1425
# the key can be reconstructed.
1426
reconstructable_keys = {}
1430
chain = [key, positions[key][2]]
1432
needed_from_fallback.add(key)
1435
while chain[-1] is not None:
1436
if chain[-1] in reconstructable_keys:
1437
result = reconstructable_keys[chain[-1]]
1441
chain.append(positions[chain[-1]][2])
1443
# missing basis component
1444
needed_from_fallback.add(chain[-1])
1447
for chain_key in chain[:-1]:
1448
reconstructable_keys[chain_key] = result
1450
needed_from_fallback.add(key)
1451
# Double index lookups here : need a unified api ?
1452
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1453
if ordering in ('topological', 'groupcompress'):
1454
if ordering == 'topological':
1455
# Global topological sort
1456
present_keys = tsort.topo_sort(global_map)
1458
present_keys = sort_groupcompress(global_map)
1459
# Now group by source:
1461
current_source = None
1462
for key in present_keys:
1463
for parent_map in parent_maps:
1464
if key in parent_map:
1465
key_source = parent_map
1467
if current_source is not key_source:
1468
source_keys.append((key_source, []))
1469
current_source = key_source
1470
source_keys[-1][1].append(key)
1472
if ordering != 'unordered':
1473
raise AssertionError('valid values for ordering are:'
1474
' "unordered", "groupcompress" or "topological" not: %r'
1476
# Just group by source; remote sources first.
1479
for parent_map in reversed(parent_maps):
1480
source_keys.append((parent_map, []))
1481
for key in parent_map:
1482
present_keys.append(key)
1483
source_keys[-1][1].append(key)
1484
# We have been requested to return these records in an order that
1485
# suits us. So we ask the index to give us an optimally sorted
1487
for source, sub_keys in source_keys:
1488
if source is parent_maps[0]:
1489
# Only sort the keys for this VF
1490
self._index._sort_keys_by_io(sub_keys, positions)
1491
absent_keys = keys - set(global_map)
1492
for key in absent_keys:
1493
yield AbsentContentFactory(key)
1494
# restrict our view to the keys we can answer.
1495
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1496
# XXX: At that point we need to consider the impact of double reads by
1497
# utilising components multiple times.
1498
if include_delta_closure:
1499
# XXX: get_content_maps performs its own index queries; allow state
1501
non_local_keys = needed_from_fallback - absent_keys
1502
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1505
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1508
for record in generator.get_record_stream():
1511
for source, keys in source_keys:
1512
if source is parent_maps[0]:
1513
# this KnitVersionedFiles
1514
records = [(key, positions[key][1]) for key in keys]
1515
for key, raw_data in self._read_records_iter_unchecked(records):
1516
(record_details, index_memo, _) = positions[key]
1517
yield KnitContentFactory(key, global_map[key],
1518
record_details, None, raw_data, self._factory.annotated, None)
1520
vf = self._immediate_fallback_vfs[parent_maps.index(source) - 1]
1521
for record in vf.get_record_stream(keys, ordering,
1522
include_delta_closure):
1525
def get_sha1s(self, keys):
1526
"""See VersionedFiles.get_sha1s()."""
1528
record_map = self._get_record_map(missing, allow_missing=True)
1530
for key, details in viewitems(record_map):
1531
if key not in missing:
1533
# record entry 2 is the 'digest'.
1534
result[key] = details[2]
1535
missing.difference_update(set(result))
1536
for source in self._immediate_fallback_vfs:
1539
new_result = source.get_sha1s(missing)
1540
result.update(new_result)
1541
missing.difference_update(set(new_result))
1544
def insert_record_stream(self, stream):
1545
"""Insert a record stream into this container.
1547
:param stream: A stream of records to insert.
1549
:seealso VersionedFiles.get_record_stream:
1551
def get_adapter(adapter_key):
1553
return adapters[adapter_key]
1555
adapter_factory = adapter_registry.get(adapter_key)
1556
adapter = adapter_factory(self)
1557
adapters[adapter_key] = adapter
1560
if self._factory.annotated:
1561
# self is annotated, we need annotated knits to use directly.
1562
annotated = "annotated-"
1565
# self is not annotated, but we can strip annotations cheaply.
1567
convertibles = {"knit-annotated-ft-gz"}
1568
if self._max_delta_chain:
1569
delta_types.add("knit-annotated-delta-gz")
1570
convertibles.add("knit-annotated-delta-gz")
1571
# The set of types we can cheaply adapt without needing basis texts.
1572
native_types = set()
1573
if self._max_delta_chain:
1574
native_types.add("knit-%sdelta-gz" % annotated)
1575
delta_types.add("knit-%sdelta-gz" % annotated)
1576
native_types.add("knit-%sft-gz" % annotated)
1577
knit_types = native_types.union(convertibles)
1579
# Buffer all index entries that we can't add immediately because their
1580
# basis parent is missing. We don't buffer all because generating
1581
# annotations may require access to some of the new records. However we
1582
# can't generate annotations from new deltas until their basis parent
1583
# is present anyway, so we get away with not needing an index that
1584
# includes the new keys.
1586
# See <http://launchpad.net/bugs/300177> about ordering of compression
1587
# parents in the records - to be conservative, we insist that all
1588
# parents must be present to avoid expanding to a fulltext.
1590
# key = basis_parent, value = index entry to add
1591
buffered_index_entries = {}
1592
for record in stream:
1593
kind = record.storage_kind
1594
if kind.startswith('knit-') and kind.endswith('-gz'):
1595
# Check that the ID in the header of the raw knit bytes matches
1596
# the record metadata.
1597
raw_data = record._raw_record
1598
df, rec = self._parse_record_header(record.key, raw_data)
1601
parents = record.parents
1602
if record.storage_kind in delta_types:
1603
# TODO: eventually the record itself should track
1604
# compression_parent
1605
compression_parent = parents[0]
1607
compression_parent = None
1608
# Raise an error when a record is missing.
1609
if record.storage_kind == 'absent':
1610
raise RevisionNotPresent([record.key], self)
1611
elif ((record.storage_kind in knit_types)
1612
and (compression_parent is None
1613
or not self._immediate_fallback_vfs
1614
or compression_parent in self._index
1615
or compression_parent not in self)):
1616
# we can insert the knit record literally if either it has no
1617
# compression parent OR we already have its basis in this kvf
1618
# OR the basis is not present even in the fallbacks. In the
1619
# last case it will either turn up later in the stream and all
1620
# will be well, or it won't turn up at all and we'll raise an
1623
# TODO: self.__contains__ is somewhat redundant with
1624
# self._index.__contains__; we really want something that directly
1625
# asks if it's only present in the fallbacks. -- mbp 20081119
1626
if record.storage_kind not in native_types:
1628
adapter_key = (record.storage_kind, "knit-delta-gz")
1629
adapter = get_adapter(adapter_key)
1631
adapter_key = (record.storage_kind, "knit-ft-gz")
1632
adapter = get_adapter(adapter_key)
1633
bytes = adapter.get_bytes(record)
1635
# It's a knit record, it has a _raw_record field (even if
1636
# it was reconstituted from a network stream).
1637
bytes = record._raw_record
1638
options = [record._build_details[0]]
1639
if record._build_details[1]:
1640
options.append('no-eol')
1641
# Just blat it across.
1642
# Note: This does end up adding data on duplicate keys. As
1643
# modern repositories use atomic insertions this should not
1644
# lead to excessive growth in the event of interrupted fetches.
1645
# 'knit' repositories may suffer excessive growth, but as a
1646
# deprecated format this is tolerable. It can be fixed if
1647
# needed by in the kndx index support raising on a duplicate
1648
# add with identical parents and options.
1649
access_memo = self._access.add_raw_records(
1650
[(record.key, len(bytes))], bytes)[0]
1651
index_entry = (record.key, options, access_memo, parents)
1652
if 'fulltext' not in options:
1653
# Not a fulltext, so we need to make sure the compression
1654
# parent will also be present.
1655
# Note that pack backed knits don't need to buffer here
1656
# because they buffer all writes to the transaction level,
1657
# but we don't expose that difference at the index level. If
1658
# the query here has sufficient cost to show up in
1659
# profiling we should do that.
1661
# They're required to be physically in this
1662
# KnitVersionedFiles, not in a fallback.
1663
if compression_parent not in self._index:
1664
pending = buffered_index_entries.setdefault(
1665
compression_parent, [])
1666
pending.append(index_entry)
1669
self._index.add_records([index_entry])
1670
elif record.storage_kind == 'chunked':
1671
self.add_lines(record.key, parents,
1672
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1674
# Not suitable for direct insertion as a
1675
# delta, either because it's not the right format, or this
1676
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1677
# 0) or because it depends on a base only present in the
1679
self._access.flush()
1681
# Try getting a fulltext directly from the record.
1682
bytes = record.get_bytes_as('fulltext')
1683
except errors.UnavailableRepresentation:
1684
adapter_key = record.storage_kind, 'fulltext'
1685
adapter = get_adapter(adapter_key)
1686
bytes = adapter.get_bytes(record)
1687
lines = split_lines(bytes)
1689
self.add_lines(record.key, parents, lines)
1690
except errors.RevisionAlreadyPresent:
1692
# Add any records whose basis parent is now available.
1694
added_keys = [record.key]
1696
key = added_keys.pop(0)
1697
if key in buffered_index_entries:
1698
index_entries = buffered_index_entries[key]
1699
self._index.add_records(index_entries)
1701
[index_entry[0] for index_entry in index_entries])
1702
del buffered_index_entries[key]
1703
if buffered_index_entries:
1704
# There were index entries buffered at the end of the stream,
1705
# So these need to be added (if the index supports holding such
1706
# entries for later insertion)
1708
for key in buffered_index_entries:
1709
index_entries = buffered_index_entries[key]
1710
all_entries.extend(index_entries)
1711
self._index.add_records(
1712
all_entries, missing_compression_parents=True)
1714
def get_missing_compression_parent_keys(self):
1715
"""Return an iterable of keys of missing compression parents.
1717
Check this after calling insert_record_stream to find out if there are
1718
any missing compression parents. If there are, the records that
1719
depend on them are not able to be inserted safely. For atomic
1720
KnitVersionedFiles built on packs, the transaction should be aborted or
1721
suspended - commit will fail at this point. Nonatomic knits will error
1722
earlier because they have no staging area to put pending entries into.
1724
return self._index.get_missing_compression_parents()
1726
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1727
"""Iterate over the lines in the versioned files from keys.
1729
This may return lines from other keys. Each item the returned
1730
iterator yields is a tuple of a line and a text version that that line
1731
is present in (not introduced in).
1733
Ordering of results is in whatever order is most suitable for the
1734
underlying storage format.
1736
If a progress bar is supplied, it may be used to indicate progress.
1737
The caller is responsible for cleaning up progress bars (because this
1741
* Lines are normalised by the underlying store: they will all have \\n
1743
* Lines are returned in arbitrary order.
1744
* If a requested key did not change any lines (or didn't have any
1745
lines), it may not be mentioned at all in the result.
1747
:param pb: Progress bar supplied by caller.
1748
:return: An iterator over (line, key).
1751
pb = ui.ui_factory.nested_progress_bar()
1757
# we don't care about inclusions, the caller cares.
1758
# but we need to setup a list of records to visit.
1759
# we need key, position, length
1761
build_details = self._index.get_build_details(keys)
1762
for key, details in viewitems(build_details):
1764
key_records.append((key, details[0]))
1765
records_iter = enumerate(self._read_records_iter(key_records))
1766
for (key_idx, (key, data, sha_value)) in records_iter:
1767
pb.update(gettext('Walking content'), key_idx, total)
1768
compression_parent = build_details[key][1]
1769
if compression_parent is None:
1771
line_iterator = self._factory.get_fulltext_content(data)
1774
line_iterator = self._factory.get_linedelta_content(data)
1775
# Now that we are yielding the data for this key, remove it
1778
# XXX: It might be more efficient to yield (key,
1779
# line_iterator) in the future. However for now, this is a
1780
# simpler change to integrate into the rest of the
1781
# codebase. RBC 20071110
1782
for line in line_iterator:
1785
except errors.RetryWithNewPacks as e:
1786
self._access.reload_or_raise(e)
1787
# If there are still keys we've not yet found, we look in the fallback
1788
# vfs, and hope to find them there. Note that if the keys are found
1789
# but had no changes or no content, the fallback may not return
1791
if keys and not self._immediate_fallback_vfs:
1792
# XXX: strictly the second parameter is meant to be the file id
1793
# but it's not easily accessible here.
1794
raise RevisionNotPresent(keys, repr(self))
1795
for source in self._immediate_fallback_vfs:
1799
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1800
source_keys.add(key)
1802
keys.difference_update(source_keys)
1803
pb.update(gettext('Walking content'), total, total)
1805
def _make_line_delta(self, delta_seq, new_content):
1806
"""Generate a line delta from delta_seq and new_content."""
1808
for op in delta_seq.get_opcodes():
1809
if op[0] == 'equal':
1811
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1814
def _merge_annotations(self, content, parents, parent_texts={},
1815
delta=None, annotated=None,
1816
left_matching_blocks=None):
1817
"""Merge annotations for content and generate deltas.
1819
This is done by comparing the annotations based on changes to the text
1820
and generating a delta on the resulting full texts. If annotations are
1821
not being created then a simple delta is created.
1823
if left_matching_blocks is not None:
1824
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1828
for parent_key in parents:
1829
merge_content = self._get_content(parent_key, parent_texts)
1830
if (parent_key == parents[0] and delta_seq is not None):
1833
seq = patiencediff.PatienceSequenceMatcher(
1834
None, merge_content.text(), content.text())
1835
for i, j, n in seq.get_matching_blocks():
1838
# this copies (origin, text) pairs across to the new
1839
# content for any line that matches the last-checked
1841
content._lines[j:j+n] = merge_content._lines[i:i+n]
1842
# XXX: Robert says the following block is a workaround for a
1843
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1844
if content._lines and content._lines[-1][1][-1] != '\n':
1845
# The copied annotation was from a line without a trailing EOL,
1846
# reinstate one for the content object, to ensure correct
1848
line = content._lines[-1][1] + '\n'
1849
content._lines[-1] = (content._lines[-1][0], line)
1851
if delta_seq is None:
1852
reference_content = self._get_content(parents[0], parent_texts)
1853
new_texts = content.text()
1854
old_texts = reference_content.text()
1855
delta_seq = patiencediff.PatienceSequenceMatcher(
1856
None, old_texts, new_texts)
1857
return self._make_line_delta(delta_seq, content)
1859
def _parse_record(self, version_id, data):
1860
"""Parse an original format knit record.
1862
These have the last element of the key only present in the stored data.
1864
rec, record_contents = self._parse_record_unchecked(data)
1865
self._check_header_version(rec, version_id)
1866
return record_contents, rec[3]
1868
def _parse_record_header(self, key, raw_data):
1869
"""Parse a record header for consistency.
1871
:return: the header and the decompressor stream.
1872
as (stream, header_record)
1874
df = gzip.GzipFile(mode='rb', fileobj=BytesIO(raw_data))
1877
rec = self._check_header(key, df.readline())
1878
except Exception as e:
1879
raise KnitCorrupt(self,
1880
"While reading {%s} got %s(%s)"
1881
% (key, e.__class__.__name__, str(e)))
1884
def _parse_record_unchecked(self, data):
1886
# 4168 calls in 2880 217 internal
1887
# 4168 calls to _parse_record_header in 2121
1888
# 4168 calls to readlines in 330
1889
df = gzip.GzipFile(mode='rb', fileobj=BytesIO(data))
1891
record_contents = df.readlines()
1892
except Exception as e:
1893
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1894
(data, e.__class__.__name__, str(e)))
1895
header = record_contents.pop(0)
1896
rec = self._split_header(header)
1897
last_line = record_contents.pop()
1898
if len(record_contents) != int(rec[2]):
1899
raise KnitCorrupt(self,
1900
'incorrect number of lines %s != %s'
1901
' for version {%s} %s'
1902
% (len(record_contents), int(rec[2]),
1903
rec[1], record_contents))
1904
if last_line != 'end %s\n' % rec[1]:
1905
raise KnitCorrupt(self,
1906
'unexpected version end line %r, wanted %r'
1907
% (last_line, rec[1]))
1909
return rec, record_contents
1911
def _read_records_iter(self, records):
1912
"""Read text records from data file and yield result.
1914
The result will be returned in whatever is the fastest to read.
1915
Not by the order requested. Also, multiple requests for the same
1916
record will only yield 1 response.
1918
:param records: A list of (key, access_memo) entries
1919
:return: Yields (key, contents, digest) in the order
1920
read, not the order requested
1925
# XXX: This smells wrong, IO may not be getting ordered right.
1926
needed_records = sorted(set(records), key=operator.itemgetter(1))
1927
if not needed_records:
1930
# The transport optimizes the fetching as well
1931
# (ie, reads continuous ranges.)
1932
raw_data = self._access.get_raw_records(
1933
[index_memo for key, index_memo in needed_records])
1935
for (key, index_memo), data in zip(needed_records, raw_data):
1936
content, digest = self._parse_record(key[-1], data)
1937
yield key, content, digest
1939
def _read_records_iter_raw(self, records):
1940
"""Read text records from data file and yield raw data.
1942
This unpacks enough of the text record to validate the id is
1943
as expected but thats all.
1945
Each item the iterator yields is (key, bytes,
1946
expected_sha1_of_full_text).
1948
for key, data in self._read_records_iter_unchecked(records):
1949
# validate the header (note that we can only use the suffix in
1950
# current knit records).
1951
df, rec = self._parse_record_header(key, data)
1953
yield key, data, rec[3]
1955
def _read_records_iter_unchecked(self, records):
1956
"""Read text records from data file and yield raw data.
1958
No validation is done.
1960
Yields tuples of (key, data).
1962
# setup an iterator of the external records:
1963
# uses readv so nice and fast we hope.
1965
# grab the disk data needed.
1966
needed_offsets = [index_memo for key, index_memo
1968
raw_records = self._access.get_raw_records(needed_offsets)
1970
for key, index_memo in records:
1971
data = next(raw_records)
1974
def _record_to_data(self, key, digest, lines, dense_lines=None):
1975
"""Convert key, digest, lines into a raw data block.
1977
:param key: The key of the record. Currently keys are always serialised
1978
using just the trailing component.
1979
:param dense_lines: The bytes of lines but in a denser form. For
1980
instance, if lines is a list of 1000 bytestrings each ending in
1981
\\n, dense_lines may be a list with one line in it, containing all
1982
the 1000's lines and their \\n's. Using dense_lines if it is
1983
already known is a win because the string join to create bytes in
1984
this function spends less time resizing the final string.
1985
:return: (len, a BytesIO instance with the raw data ready to read.)
1987
chunks = ["version %s %d %s\n" % (key[-1], len(lines), digest)]
1988
chunks.extend(dense_lines or lines)
1989
chunks.append("end %s\n" % key[-1])
1990
for chunk in chunks:
1991
if not isinstance(chunk, str):
1992
raise AssertionError(
1993
'data must be plain bytes was %s' % type(chunk))
1994
if lines and lines[-1][-1] != '\n':
1995
raise ValueError('corrupt lines value %r' % lines)
1996
compressed_bytes = tuned_gzip.chunks_to_gzip(chunks)
1997
return len(compressed_bytes), compressed_bytes
1999
def _split_header(self, line):
2002
raise KnitCorrupt(self,
2003
'unexpected number of elements in record header')
2007
"""See VersionedFiles.keys."""
2008
if 'evil' in debug.debug_flags:
2009
trace.mutter_callsite(2, "keys scales with size of history")
2010
sources = [self._index] + self._immediate_fallback_vfs
2012
for source in sources:
2013
result.update(source.keys())
2017
class _ContentMapGenerator(object):
2018
"""Generate texts or expose raw deltas for a set of texts."""
2020
def __init__(self, ordering='unordered'):
2021
self._ordering = ordering
2023
def _get_content(self, key):
2024
"""Get the content object for key."""
2025
# Note that _get_content is only called when the _ContentMapGenerator
2026
# has been constructed with just one key requested for reconstruction.
2027
if key in self.nonlocal_keys:
2028
record = next(self.get_record_stream())
2029
# Create a content object on the fly
2030
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
2031
return PlainKnitContent(lines, record.key)
2033
# local keys we can ask for directly
2034
return self._get_one_work(key)
2036
def get_record_stream(self):
2037
"""Get a record stream for the keys requested during __init__."""
2038
for record in self._work():
2042
"""Produce maps of text and KnitContents as dicts.
2044
:return: (text_map, content_map) where text_map contains the texts for
2045
the requested versions and content_map contains the KnitContents.
2047
# NB: By definition we never need to read remote sources unless texts
2048
# are requested from them: we don't delta across stores - and we
2049
# explicitly do not want to to prevent data loss situations.
2050
if self.global_map is None:
2051
self.global_map = self.vf.get_parent_map(self.keys)
2052
nonlocal_keys = self.nonlocal_keys
2054
missing_keys = set(nonlocal_keys)
2055
# Read from remote versioned file instances and provide to our caller.
2056
for source in self.vf._immediate_fallback_vfs:
2057
if not missing_keys:
2059
# Loop over fallback repositories asking them for texts - ignore
2060
# any missing from a particular fallback.
2061
for record in source.get_record_stream(missing_keys,
2062
self._ordering, True):
2063
if record.storage_kind == 'absent':
2064
# Not in thie particular stream, may be in one of the
2065
# other fallback vfs objects.
2067
missing_keys.remove(record.key)
2070
if self._raw_record_map is None:
2071
raise AssertionError('_raw_record_map should have been filled')
2073
for key in self.keys:
2074
if key in self.nonlocal_keys:
2076
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2079
def _get_one_work(self, requested_key):
2080
# Now, if we have calculated everything already, just return the
2082
if requested_key in self._contents_map:
2083
return self._contents_map[requested_key]
2084
# To simplify things, parse everything at once - code that wants one text
2085
# probably wants them all.
2086
# FUTURE: This function could be improved for the 'extract many' case
2087
# by tracking each component and only doing the copy when the number of
2088
# children than need to apply delta's to it is > 1 or it is part of the
2090
multiple_versions = len(self.keys) != 1
2091
if self._record_map is None:
2092
self._record_map = self.vf._raw_map_to_record_map(
2093
self._raw_record_map)
2094
record_map = self._record_map
2095
# raw_record_map is key:
2096
# Have read and parsed records at this point.
2097
for key in self.keys:
2098
if key in self.nonlocal_keys:
2103
while cursor is not None:
2105
record, record_details, digest, next = record_map[cursor]
2107
raise RevisionNotPresent(cursor, self)
2108
components.append((cursor, record, record_details, digest))
2110
if cursor in self._contents_map:
2111
# no need to plan further back
2112
components.append((cursor, None, None, None))
2116
for (component_id, record, record_details,
2117
digest) in reversed(components):
2118
if component_id in self._contents_map:
2119
content = self._contents_map[component_id]
2121
content, delta = self._factory.parse_record(key[-1],
2122
record, record_details, content,
2123
copy_base_content=multiple_versions)
2124
if multiple_versions:
2125
self._contents_map[component_id] = content
2127
# digest here is the digest from the last applied component.
2128
text = content.text()
2129
actual_sha = sha_strings(text)
2130
if actual_sha != digest:
2131
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2132
if multiple_versions:
2133
return self._contents_map[requested_key]
2137
def _wire_bytes(self):
2138
"""Get the bytes to put on the wire for 'key'.
2140
The first collection of bytes asked for returns the serialised
2141
raw_record_map and the additional details (key, parent) for key.
2142
Subsequent calls return just the additional details (key, parent).
2143
The wire storage_kind given for the first key is 'knit-delta-closure',
2144
For subsequent keys it is 'knit-delta-closure-ref'.
2146
:param key: A key from the content generator.
2147
:return: Bytes to put on the wire.
2150
# kind marker for dispatch on the far side,
2151
lines.append('knit-delta-closure')
2153
if self.vf._factory.annotated:
2154
lines.append('annotated')
2157
# then the list of keys
2158
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2159
if key not in self.nonlocal_keys]))
2160
# then the _raw_record_map in serialised form:
2162
# for each item in the map:
2164
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2165
# one line with method
2166
# one line with noeol
2167
# one line with next ('' for None)
2168
# one line with byte count of the record bytes
2170
for key, (record_bytes, (method, noeol), next) in viewitems(
2171
self._raw_record_map):
2172
key_bytes = '\x00'.join(key)
2173
parents = self.global_map.get(key, None)
2175
parent_bytes = 'None:'
2177
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2178
method_bytes = method
2184
next_bytes = '\x00'.join(next)
2187
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2188
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2189
len(record_bytes), record_bytes))
2190
map_bytes = ''.join(map_byte_list)
2191
lines.append(map_bytes)
2192
bytes = '\n'.join(lines)
2196
class _VFContentMapGenerator(_ContentMapGenerator):
2197
"""Content map generator reading from a VersionedFiles object."""
2199
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2200
global_map=None, raw_record_map=None, ordering='unordered'):
2201
"""Create a _ContentMapGenerator.
2203
:param versioned_files: The versioned files that the texts are being
2205
:param keys: The keys to produce content maps for.
2206
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2207
which are known to not be in this knit, but rather in one of the
2209
:param global_map: The result of get_parent_map(keys) (or a supermap).
2210
This is required if get_record_stream() is to be used.
2211
:param raw_record_map: A unparsed raw record map to use for answering
2214
_ContentMapGenerator.__init__(self, ordering=ordering)
2215
# The vf to source data from
2216
self.vf = versioned_files
2218
self.keys = list(keys)
2219
# Keys known to be in fallback vfs objects
2220
if nonlocal_keys is None:
2221
self.nonlocal_keys = set()
2223
self.nonlocal_keys = frozenset(nonlocal_keys)
2224
# Parents data for keys to be returned in get_record_stream
2225
self.global_map = global_map
2226
# The chunked lists for self.keys in text form
2228
# A cache of KnitContent objects used in extracting texts.
2229
self._contents_map = {}
2230
# All the knit records needed to assemble the requested keys as full
2232
self._record_map = None
2233
if raw_record_map is None:
2234
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2237
self._raw_record_map = raw_record_map
2238
# the factory for parsing records
2239
self._factory = self.vf._factory
2242
class _NetworkContentMapGenerator(_ContentMapGenerator):
2243
"""Content map generator sourced from a network stream."""
2245
def __init__(self, bytes, line_end):
2246
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2248
self.global_map = {}
2249
self._raw_record_map = {}
2250
self._contents_map = {}
2251
self._record_map = None
2252
self.nonlocal_keys = []
2253
# Get access to record parsing facilities
2254
self.vf = KnitVersionedFiles(None, None)
2257
line_end = bytes.find('\n', start)
2258
line = bytes[start:line_end]
2259
start = line_end + 1
2260
if line == 'annotated':
2261
self._factory = KnitAnnotateFactory()
2263
self._factory = KnitPlainFactory()
2264
# list of keys to emit in get_record_stream
2265
line_end = bytes.find('\n', start)
2266
line = bytes[start:line_end]
2267
start = line_end + 1
2269
tuple(segment.split('\x00')) for segment in line.split('\t')
2271
# now a loop until the end. XXX: It would be nice if this was just a
2272
# bunch of the same records as get_record_stream(..., False) gives, but
2273
# there is a decent sized gap stopping that at the moment.
2277
line_end = bytes.find('\n', start)
2278
key = tuple(bytes[start:line_end].split('\x00'))
2279
start = line_end + 1
2280
# 1 line with parents (None: for None, '' for ())
2281
line_end = bytes.find('\n', start)
2282
line = bytes[start:line_end]
2287
[tuple(segment.split('\x00')) for segment in line.split('\t')
2289
self.global_map[key] = parents
2290
start = line_end + 1
2291
# one line with method
2292
line_end = bytes.find('\n', start)
2293
line = bytes[start:line_end]
2295
start = line_end + 1
2296
# one line with noeol
2297
line_end = bytes.find('\n', start)
2298
line = bytes[start:line_end]
2300
start = line_end + 1
2301
# one line with next ('' for None)
2302
line_end = bytes.find('\n', start)
2303
line = bytes[start:line_end]
2307
next = tuple(bytes[start:line_end].split('\x00'))
2308
start = line_end + 1
2309
# one line with byte count of the record bytes
2310
line_end = bytes.find('\n', start)
2311
line = bytes[start:line_end]
2313
start = line_end + 1
2315
record_bytes = bytes[start:start+count]
2316
start = start + count
2318
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2320
def get_record_stream(self):
2321
"""Get a record stream for for keys requested by the bytestream."""
2323
for key in self.keys:
2324
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2327
def _wire_bytes(self):
2331
class _KndxIndex(object):
2332
"""Manages knit index files
2334
The index is kept in memory and read on startup, to enable
2335
fast lookups of revision information. The cursor of the index
2336
file is always pointing to the end, making it easy to append
2339
_cache is a cache for fast mapping from version id to a Index
2342
_history is a cache for fast mapping from indexes to version ids.
2344
The index data format is dictionary compressed when it comes to
2345
parent references; a index entry may only have parents that with a
2346
lover index number. As a result, the index is topological sorted.
2348
Duplicate entries may be written to the index for a single version id
2349
if this is done then the latter one completely replaces the former:
2350
this allows updates to correct version and parent information.
2351
Note that the two entries may share the delta, and that successive
2352
annotations and references MUST point to the first entry.
2354
The index file on disc contains a header, followed by one line per knit
2355
record. The same revision can be present in an index file more than once.
2356
The first occurrence gets assigned a sequence number starting from 0.
2358
The format of a single line is
2359
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
2360
REVISION_ID is a utf8-encoded revision id
2361
FLAGS is a comma separated list of flags about the record. Values include
2362
no-eol, line-delta, fulltext.
2363
BYTE_OFFSET is the ascii representation of the byte offset in the data file
2364
that the compressed data starts at.
2365
LENGTH is the ascii representation of the length of the data file.
2366
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
2368
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
2369
revision id already in the knit that is a parent of REVISION_ID.
2370
The ' :' marker is the end of record marker.
2373
when a write is interrupted to the index file, it will result in a line
2374
that does not end in ' :'. If the ' :' is not present at the end of a line,
2375
or at the end of the file, then the record that is missing it will be
2376
ignored by the parser.
2378
When writing new records to the index file, the data is preceded by '\n'
2379
to ensure that records always start on new lines even if the last write was
2380
interrupted. As a result its normal for the last line in the index to be
2381
missing a trailing newline. One can be added with no harmful effects.
2383
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2384
where prefix is e.g. the (fileid,) for .texts instances or () for
2385
constant-mapped things like .revisions, and the old state is
2386
tuple(history_vector, cache_dict). This is used to prevent having an
2387
ABI change with the C extension that reads .kndx files.
2390
HEADER = "# bzr knit index 8\n"
2392
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2393
"""Create a _KndxIndex on transport using mapper."""
2394
self._transport = transport
2395
self._mapper = mapper
2396
self._get_scope = get_scope
2397
self._allow_writes = allow_writes
2398
self._is_locked = is_locked
2400
self.has_graph = True
2402
def add_records(self, records, random_id=False, missing_compression_parents=False):
2403
"""Add multiple records to the index.
2405
:param records: a list of tuples:
2406
(key, options, access_memo, parents).
2407
:param random_id: If True the ids being added were randomly generated
2408
and no check for existence will be performed.
2409
:param missing_compression_parents: If True the records being added are
2410
only compressed against texts already in the index (or inside
2411
records). If False the records all refer to unavailable texts (or
2412
texts inside records) as compression parents.
2414
if missing_compression_parents:
2415
# It might be nice to get the edge of the records. But keys isn't
2417
keys = sorted(record[0] for record in records)
2418
raise errors.RevisionNotPresent(keys, self)
2420
for record in records:
2423
path = self._mapper.map(key) + '.kndx'
2424
path_keys = paths.setdefault(path, (prefix, []))
2425
path_keys[1].append(record)
2426
for path in sorted(paths):
2427
prefix, path_keys = paths[path]
2428
self._load_prefixes([prefix])
2430
orig_history = self._kndx_cache[prefix][1][:]
2431
orig_cache = self._kndx_cache[prefix][0].copy()
2434
for key, options, (_, pos, size), parents in path_keys:
2436
# kndx indices cannot be parentless.
2438
line = "\n%s %s %s %s %s :" % (
2439
key[-1], ','.join(options), pos, size,
2440
self._dictionary_compress(parents))
2441
if not isinstance(line, str):
2442
raise AssertionError(
2443
'data must be utf8 was %s' % type(line))
2445
self._cache_key(key, options, pos, size, parents)
2446
if len(orig_history):
2447
self._transport.append_bytes(path, ''.join(lines))
2449
self._init_index(path, lines)
2451
# If any problems happen, restore the original values and re-raise
2452
self._kndx_cache[prefix] = (orig_cache, orig_history)
2455
def scan_unvalidated_index(self, graph_index):
2456
"""See _KnitGraphIndex.scan_unvalidated_index."""
2457
# Because kndx files do not support atomic insertion via separate index
2458
# files, they do not support this method.
2459
raise NotImplementedError(self.scan_unvalidated_index)
2461
def get_missing_compression_parents(self):
2462
"""See _KnitGraphIndex.get_missing_compression_parents."""
2463
# Because kndx files do not support atomic insertion via separate index
2464
# files, they do not support this method.
2465
raise NotImplementedError(self.get_missing_compression_parents)
2467
def _cache_key(self, key, options, pos, size, parent_keys):
2468
"""Cache a version record in the history array and index cache.
2470
This is inlined into _load_data for performance. KEEP IN SYNC.
2471
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2475
version_id = key[-1]
2476
# last-element only for compatibilty with the C load_data.
2477
parents = tuple(parent[-1] for parent in parent_keys)
2478
for parent in parent_keys:
2479
if parent[:-1] != prefix:
2480
raise ValueError("mismatched prefixes for %r, %r" % (
2482
cache, history = self._kndx_cache[prefix]
2483
# only want the _history index to reference the 1st index entry
2485
if version_id not in cache:
2486
index = len(history)
2487
history.append(version_id)
2489
index = cache[version_id][5]
2490
cache[version_id] = (version_id,
2497
def check_header(self, fp):
2498
line = fp.readline()
2500
# An empty file can actually be treated as though the file doesn't
2502
raise errors.NoSuchFile(self)
2503
if line != self.HEADER:
2504
raise KnitHeaderError(badline=line, filename=self)
2506
def _check_read(self):
2507
if not self._is_locked():
2508
raise errors.ObjectNotLocked(self)
2509
if self._get_scope() != self._scope:
2512
def _check_write_ok(self):
2513
"""Assert if not writes are permitted."""
2514
if not self._is_locked():
2515
raise errors.ObjectNotLocked(self)
2516
if self._get_scope() != self._scope:
2518
if self._mode != 'w':
2519
raise errors.ReadOnlyObjectDirtiedError(self)
2521
def get_build_details(self, keys):
2522
"""Get the method, index_memo and compression parent for keys.
2524
Ghosts are omitted from the result.
2526
:param keys: An iterable of keys.
2527
:return: A dict of key:(index_memo, compression_parent, parents,
2530
opaque structure to pass to read_records to extract the raw
2533
Content that this record is built upon, may be None
2535
Logical parents of this node
2537
extra information about the content which needs to be passed to
2538
Factory.parse_record
2540
parent_map = self.get_parent_map(keys)
2543
if key not in parent_map:
2545
method = self.get_method(key)
2546
parents = parent_map[key]
2547
if method == 'fulltext':
2548
compression_parent = None
2550
compression_parent = parents[0]
2551
noeol = 'no-eol' in self.get_options(key)
2552
index_memo = self.get_position(key)
2553
result[key] = (index_memo, compression_parent,
2554
parents, (method, noeol))
2557
def get_method(self, key):
2558
"""Return compression method of specified key."""
2559
options = self.get_options(key)
2560
if 'fulltext' in options:
2562
elif 'line-delta' in options:
2565
raise errors.KnitIndexUnknownMethod(self, options)
2567
def get_options(self, key):
2568
"""Return a list representing options.
2572
prefix, suffix = self._split_key(key)
2573
self._load_prefixes([prefix])
2575
return self._kndx_cache[prefix][0][suffix][1]
2577
raise RevisionNotPresent(key, self)
2579
def find_ancestry(self, keys):
2580
"""See CombinedGraphIndex.find_ancestry()"""
2581
prefixes = set(key[:-1] for key in keys)
2582
self._load_prefixes(prefixes)
2585
missing_keys = set()
2586
pending_keys = list(keys)
2587
# This assumes that keys will not reference parents in a different
2588
# prefix, which is accurate so far.
2590
key = pending_keys.pop()
2591
if key in parent_map:
2595
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2597
missing_keys.add(key)
2599
parent_keys = tuple([prefix + (suffix,)
2600
for suffix in suffix_parents])
2601
parent_map[key] = parent_keys
2602
pending_keys.extend([p for p in parent_keys
2603
if p not in parent_map])
2604
return parent_map, missing_keys
2606
def get_parent_map(self, keys):
2607
"""Get a map of the parents of keys.
2609
:param keys: The keys to look up parents for.
2610
:return: A mapping from keys to parents. Absent keys are absent from
2613
# Parse what we need to up front, this potentially trades off I/O
2614
# locality (.kndx and .knit in the same block group for the same file
2615
# id) for less checking in inner loops.
2616
prefixes = set(key[:-1] for key in keys)
2617
self._load_prefixes(prefixes)
2622
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2626
result[key] = tuple(prefix + (suffix,) for
2627
suffix in suffix_parents)
2630
def get_position(self, key):
2631
"""Return details needed to access the version.
2633
:return: a tuple (key, data position, size) to hand to the access
2634
logic to get the record.
2636
prefix, suffix = self._split_key(key)
2637
self._load_prefixes([prefix])
2638
entry = self._kndx_cache[prefix][0][suffix]
2639
return key, entry[2], entry[3]
2641
__contains__ = _mod_index._has_key_from_parent_map
2643
def _init_index(self, path, extra_lines=[]):
2644
"""Initialize an index."""
2646
sio.write(self.HEADER)
2647
sio.writelines(extra_lines)
2649
self._transport.put_file_non_atomic(path, sio,
2650
create_parent_dir=True)
2651
# self._create_parent_dir)
2652
# mode=self._file_mode,
2653
# dir_mode=self._dir_mode)
2656
"""Get all the keys in the collection.
2658
The keys are not ordered.
2661
# Identify all key prefixes.
2662
# XXX: A bit hacky, needs polish.
2663
if isinstance(self._mapper, ConstantMapper):
2667
for quoted_relpath in self._transport.iter_files_recursive():
2668
path, ext = os.path.splitext(quoted_relpath)
2670
prefixes = [self._mapper.unmap(path) for path in relpaths]
2671
self._load_prefixes(prefixes)
2672
for prefix in prefixes:
2673
for suffix in self._kndx_cache[prefix][1]:
2674
result.add(prefix + (suffix,))
2677
def _load_prefixes(self, prefixes):
2678
"""Load the indices for prefixes."""
2680
for prefix in prefixes:
2681
if prefix not in self._kndx_cache:
2682
# the load_data interface writes to these variables.
2685
self._filename = prefix
2687
path = self._mapper.map(prefix) + '.kndx'
2688
fp = self._transport.get(path)
2690
# _load_data may raise NoSuchFile if the target knit is
2692
_load_data(self, fp)
2695
self._kndx_cache[prefix] = (self._cache, self._history)
2700
self._kndx_cache[prefix] = ({}, [])
2701
if isinstance(self._mapper, ConstantMapper):
2702
# preserve behaviour for revisions.kndx etc.
2703
self._init_index(path)
2708
missing_keys = _mod_index._missing_keys_from_parent_map
2710
def _partition_keys(self, keys):
2711
"""Turn keys into a dict of prefix:suffix_list."""
2714
prefix_keys = result.setdefault(key[:-1], [])
2715
prefix_keys.append(key[-1])
2718
def _dictionary_compress(self, keys):
2719
"""Dictionary compress keys.
2721
:param keys: The keys to generate references to.
2722
:return: A string representation of keys. keys which are present are
2723
dictionary compressed, and others are emitted as fulltext with a
2729
prefix = keys[0][:-1]
2730
cache = self._kndx_cache[prefix][0]
2732
if key[:-1] != prefix:
2733
# kndx indices cannot refer across partitioned storage.
2734
raise ValueError("mismatched prefixes for %r" % keys)
2735
if key[-1] in cache:
2736
# -- inlined lookup() --
2737
result_list.append(str(cache[key[-1]][5]))
2738
# -- end lookup () --
2740
result_list.append('.' + key[-1])
2741
return ' '.join(result_list)
2743
def _reset_cache(self):
2744
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2745
# (cache_dict, history_vector) for parsed kndx files.
2746
self._kndx_cache = {}
2747
self._scope = self._get_scope()
2748
allow_writes = self._allow_writes()
2754
def _sort_keys_by_io(self, keys, positions):
2755
"""Figure out an optimal order to read the records for the given keys.
2757
Sort keys, grouped by index and sorted by position.
2759
:param keys: A list of keys whose records we want to read. This will be
2761
:param positions: A dict, such as the one returned by
2762
_get_components_positions()
2765
def get_sort_key(key):
2766
index_memo = positions[key][1]
2767
# Group by prefix and position. index_memo[0] is the key, so it is
2768
# (file_id, revision_id) and we don't want to sort on revision_id,
2769
# index_memo[1] is the position, and index_memo[2] is the size,
2770
# which doesn't matter for the sort
2771
return index_memo[0][:-1], index_memo[1]
2772
return keys.sort(key=get_sort_key)
2774
_get_total_build_size = _get_total_build_size
2776
def _split_key(self, key):
2777
"""Split key into a prefix and suffix."""
2778
return key[:-1], key[-1]
2781
class _KnitGraphIndex(object):
2782
"""A KnitVersionedFiles index layered on GraphIndex."""
2784
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2785
add_callback=None, track_external_parent_refs=False):
2786
"""Construct a KnitGraphIndex on a graph_index.
2788
:param graph_index: An implementation of breezy.index.GraphIndex.
2789
:param is_locked: A callback to check whether the object should answer
2791
:param deltas: Allow delta-compressed records.
2792
:param parents: If True, record knits parents, if not do not record
2794
:param add_callback: If not None, allow additions to the index and call
2795
this callback with a list of added GraphIndex nodes:
2796
[(node, value, node_refs), ...]
2797
:param is_locked: A callback, returns True if the index is locked and
2799
:param track_external_parent_refs: If True, record all external parent
2800
references parents from added records. These can be retrieved
2801
later by calling get_missing_parents().
2803
self._add_callback = add_callback
2804
self._graph_index = graph_index
2805
self._deltas = deltas
2806
self._parents = parents
2807
if deltas and not parents:
2808
# XXX: TODO: Delta tree and parent graph should be conceptually
2810
raise KnitCorrupt(self, "Cannot do delta compression without "
2812
self.has_graph = parents
2813
self._is_locked = is_locked
2814
self._missing_compression_parents = set()
2815
if track_external_parent_refs:
2816
self._key_dependencies = _KeyRefs()
2818
self._key_dependencies = None
2821
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2823
def add_records(self, records, random_id=False,
2824
missing_compression_parents=False):
2825
"""Add multiple records to the index.
2827
This function does not insert data into the Immutable GraphIndex
2828
backing the KnitGraphIndex, instead it prepares data for insertion by
2829
the caller and checks that it is safe to insert then calls
2830
self._add_callback with the prepared GraphIndex nodes.
2832
:param records: a list of tuples:
2833
(key, options, access_memo, parents).
2834
:param random_id: If True the ids being added were randomly generated
2835
and no check for existence will be performed.
2836
:param missing_compression_parents: If True the records being added are
2837
only compressed against texts already in the index (or inside
2838
records). If False the records all refer to unavailable texts (or
2839
texts inside records) as compression parents.
2841
if not self._add_callback:
2842
raise errors.ReadOnlyError(self)
2843
# we hope there are no repositories with inconsistent parentage
2847
compression_parents = set()
2848
key_dependencies = self._key_dependencies
2849
for (key, options, access_memo, parents) in records:
2851
parents = tuple(parents)
2852
if key_dependencies is not None:
2853
key_dependencies.add_references(key, parents)
2854
index, pos, size = access_memo
2855
if 'no-eol' in options:
2859
value += "%d %d" % (pos, size)
2860
if not self._deltas:
2861
if 'line-delta' in options:
2862
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2865
if 'line-delta' in options:
2866
node_refs = (parents, (parents[0],))
2867
if missing_compression_parents:
2868
compression_parents.add(parents[0])
2870
node_refs = (parents, ())
2872
node_refs = (parents, )
2875
raise KnitCorrupt(self, "attempt to add node with parents "
2876
"in parentless index.")
2878
keys[key] = (value, node_refs)
2881
present_nodes = self._get_entries(keys)
2882
for (index, key, value, node_refs) in present_nodes:
2883
parents = node_refs[:1]
2884
# Sometimes these are passed as a list rather than a tuple
2885
passed = static_tuple.as_tuples(keys[key])
2886
passed_parents = passed[1][:1]
2887
if (value[0] != keys[key][0][0] or
2888
parents != passed_parents):
2889
node_refs = static_tuple.as_tuples(node_refs)
2890
raise KnitCorrupt(self, "inconsistent details in add_records"
2891
": %s %s" % ((value, node_refs), passed))
2895
for key, (value, node_refs) in viewitems(keys):
2896
result.append((key, value, node_refs))
2898
for key, (value, node_refs) in viewitems(keys):
2899
result.append((key, value))
2900
self._add_callback(result)
2901
if missing_compression_parents:
2902
# This may appear to be incorrect (it does not check for
2903
# compression parents that are in the existing graph index),
2904
# but such records won't have been buffered, so this is
2905
# actually correct: every entry when
2906
# missing_compression_parents==True either has a missing parent, or
2907
# a parent that is one of the keys in records.
2908
compression_parents.difference_update(keys)
2909
self._missing_compression_parents.update(compression_parents)
2910
# Adding records may have satisfied missing compression parents.
2911
self._missing_compression_parents.difference_update(keys)
2913
def scan_unvalidated_index(self, graph_index):
2914
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2916
This allows this _KnitGraphIndex to keep track of any missing
2917
compression parents we may want to have filled in to make those
2920
:param graph_index: A GraphIndex
2923
new_missing = graph_index.external_references(ref_list_num=1)
2924
new_missing.difference_update(self.get_parent_map(new_missing))
2925
self._missing_compression_parents.update(new_missing)
2926
if self._key_dependencies is not None:
2927
# Add parent refs from graph_index (and discard parent refs that
2928
# the graph_index has).
2929
for node in graph_index.iter_all_entries():
2930
self._key_dependencies.add_references(node[1], node[3][0])
2932
def get_missing_compression_parents(self):
2933
"""Return the keys of missing compression parents.
2935
Missing compression parents occur when a record stream was missing
2936
basis texts, or a index was scanned that had missing basis texts.
2938
return frozenset(self._missing_compression_parents)
2940
def get_missing_parents(self):
2941
"""Return the keys of missing parents."""
2942
# If updating this, you should also update
2943
# groupcompress._GCGraphIndex.get_missing_parents
2944
# We may have false positives, so filter those out.
2945
self._key_dependencies.satisfy_refs_for_keys(
2946
self.get_parent_map(self._key_dependencies.get_unsatisfied_refs()))
2947
return frozenset(self._key_dependencies.get_unsatisfied_refs())
2949
def _check_read(self):
2950
"""raise if reads are not permitted."""
2951
if not self._is_locked():
2952
raise errors.ObjectNotLocked(self)
2954
def _check_write_ok(self):
2955
"""Assert if writes are not permitted."""
2956
if not self._is_locked():
2957
raise errors.ObjectNotLocked(self)
2959
def _compression_parent(self, an_entry):
2960
# return the key that an_entry is compressed against, or None
2961
# Grab the second parent list (as deltas implies parents currently)
2962
compression_parents = an_entry[3][1]
2963
if not compression_parents:
2965
if len(compression_parents) != 1:
2966
raise AssertionError(
2967
"Too many compression parents: %r" % compression_parents)
2968
return compression_parents[0]
2970
def get_build_details(self, keys):
2971
"""Get the method, index_memo and compression parent for version_ids.
2973
Ghosts are omitted from the result.
2975
:param keys: An iterable of keys.
2976
:return: A dict of key:
2977
(index_memo, compression_parent, parents, record_details).
2979
opaque structure to pass to read_records to extract the raw
2982
Content that this record is built upon, may be None
2984
Logical parents of this node
2986
extra information about the content which needs to be passed to
2987
Factory.parse_record
2991
entries = self._get_entries(keys, False)
2992
for entry in entries:
2994
if not self._parents:
2997
parents = entry[3][0]
2998
if not self._deltas:
2999
compression_parent_key = None
3001
compression_parent_key = self._compression_parent(entry)
3002
noeol = (entry[2][0] == 'N')
3003
if compression_parent_key:
3004
method = 'line-delta'
3007
result[key] = (self._node_to_position(entry),
3008
compression_parent_key, parents,
3012
def _get_entries(self, keys, check_present=False):
3013
"""Get the entries for keys.
3015
:param keys: An iterable of index key tuples.
3020
for node in self._graph_index.iter_entries(keys):
3022
found_keys.add(node[1])
3024
# adapt parentless index to the rest of the code.
3025
for node in self._graph_index.iter_entries(keys):
3026
yield node[0], node[1], node[2], ()
3027
found_keys.add(node[1])
3029
missing_keys = keys.difference(found_keys)
3031
raise RevisionNotPresent(missing_keys.pop(), self)
3033
def get_method(self, key):
3034
"""Return compression method of specified key."""
3035
return self._get_method(self._get_node(key))
3037
def _get_method(self, node):
3038
if not self._deltas:
3040
if self._compression_parent(node):
3045
def _get_node(self, key):
3047
return list(self._get_entries([key]))[0]
3049
raise RevisionNotPresent(key, self)
3051
def get_options(self, key):
3052
"""Return a list representing options.
3056
node = self._get_node(key)
3057
options = [self._get_method(node)]
3058
if node[2][0] == 'N':
3059
options.append('no-eol')
3062
def find_ancestry(self, keys):
3063
"""See CombinedGraphIndex.find_ancestry()"""
3064
return self._graph_index.find_ancestry(keys, 0)
3066
def get_parent_map(self, keys):
3067
"""Get a map of the parents of keys.
3069
:param keys: The keys to look up parents for.
3070
:return: A mapping from keys to parents. Absent keys are absent from
3074
nodes = self._get_entries(keys)
3078
result[node[1]] = node[3][0]
3081
result[node[1]] = None
3084
def get_position(self, key):
3085
"""Return details needed to access the version.
3087
:return: a tuple (index, data position, size) to hand to the access
3088
logic to get the record.
3090
node = self._get_node(key)
3091
return self._node_to_position(node)
3093
__contains__ = _mod_index._has_key_from_parent_map
3096
"""Get all the keys in the collection.
3098
The keys are not ordered.
3101
return [node[1] for node in self._graph_index.iter_all_entries()]
3103
missing_keys = _mod_index._missing_keys_from_parent_map
3105
def _node_to_position(self, node):
3106
"""Convert an index value to position details."""
3107
bits = node[2][1:].split(' ')
3108
return node[0], int(bits[0]), int(bits[1])
3110
def _sort_keys_by_io(self, keys, positions):
3111
"""Figure out an optimal order to read the records for the given keys.
3113
Sort keys, grouped by index and sorted by position.
3115
:param keys: A list of keys whose records we want to read. This will be
3117
:param positions: A dict, such as the one returned by
3118
_get_components_positions()
3121
def get_index_memo(key):
3122
# index_memo is at offset [1]. It is made up of (GraphIndex,
3123
# position, size). GI is an object, which will be unique for each
3124
# pack file. This causes us to group by pack file, then sort by
3125
# position. Size doesn't matter, but it isn't worth breaking up the
3127
return positions[key][1]
3128
return keys.sort(key=get_index_memo)
3130
_get_total_build_size = _get_total_build_size
3133
class _KnitKeyAccess(object):
3134
"""Access to records in .knit files."""
3136
def __init__(self, transport, mapper):
3137
"""Create a _KnitKeyAccess with transport and mapper.
3139
:param transport: The transport the access object is rooted at.
3140
:param mapper: The mapper used to map keys to .knit files.
3142
self._transport = transport
3143
self._mapper = mapper
3145
def add_raw_records(self, key_sizes, raw_data):
3146
"""Add raw knit bytes to a storage area.
3148
The data is spooled to the container writer in one bytes-record per
3151
:param sizes: An iterable of tuples containing the key and size of each
3153
:param raw_data: A bytestring containing the data.
3154
:return: A list of memos to retrieve the record later. Each memo is an
3155
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3156
length), where the key is the record key.
3158
if not isinstance(raw_data, str):
3159
raise AssertionError(
3160
'data must be plain bytes was %s' % type(raw_data))
3163
# TODO: This can be tuned for writing to sftp and other servers where
3164
# append() is relatively expensive by grouping the writes to each key
3166
for key, size in key_sizes:
3167
path = self._mapper.map(key)
3169
base = self._transport.append_bytes(path + '.knit',
3170
raw_data[offset:offset+size])
3171
except errors.NoSuchFile:
3172
self._transport.mkdir(osutils.dirname(path))
3173
base = self._transport.append_bytes(path + '.knit',
3174
raw_data[offset:offset+size])
3178
result.append((key, base, size))
3182
"""Flush pending writes on this access object.
3184
For .knit files this is a no-op.
3188
def get_raw_records(self, memos_for_retrieval):
3189
"""Get the raw bytes for a records.
3191
:param memos_for_retrieval: An iterable containing the access memo for
3192
retrieving the bytes.
3193
:return: An iterator over the bytes of the records.
3195
# first pass, group into same-index request to minimise readv's issued.
3197
current_prefix = None
3198
for (key, offset, length) in memos_for_retrieval:
3199
if current_prefix == key[:-1]:
3200
current_list.append((offset, length))
3202
if current_prefix is not None:
3203
request_lists.append((current_prefix, current_list))
3204
current_prefix = key[:-1]
3205
current_list = [(offset, length)]
3206
# handle the last entry
3207
if current_prefix is not None:
3208
request_lists.append((current_prefix, current_list))
3209
for prefix, read_vector in request_lists:
3210
path = self._mapper.map(prefix) + '.knit'
3211
for pos, data in self._transport.readv(path, read_vector):
3215
def annotate_knit(knit, revision_id):
3216
"""Annotate a knit with no cached annotations.
3218
This implementation is for knits with no cached annotations.
3219
It will work for knits with cached annotations, but this is not
3222
annotator = _KnitAnnotator(knit)
3223
return iter(annotator.annotate_flat(revision_id))
3226
class _KnitAnnotator(annotate.Annotator):
3227
"""Build up the annotations for a text."""
3229
def __init__(self, vf):
3230
annotate.Annotator.__init__(self, vf)
3232
# TODO: handle Nodes which cannot be extracted
3233
# self._ghosts = set()
3235
# Map from (key, parent_key) => matching_blocks, should be 'use once'
3236
self._matching_blocks = {}
3238
# KnitContent objects
3239
self._content_objects = {}
3240
# The number of children that depend on this fulltext content object
3241
self._num_compression_children = {}
3242
# Delta records that need their compression parent before they can be
3244
self._pending_deltas = {}
3245
# Fulltext records that are waiting for their parents fulltexts before
3246
# they can be yielded for annotation
3247
self._pending_annotation = {}
3249
self._all_build_details = {}
3251
def _get_build_graph(self, key):
3252
"""Get the graphs for building texts and annotations.
3254
The data you need for creating a full text may be different than the
3255
data you need to annotate that text. (At a minimum, you need both
3256
parents to create an annotation, but only need 1 parent to generate the
3259
:return: A list of (key, index_memo) records, suitable for
3260
passing to read_records_iter to start reading in the raw data from
3266
self._num_needed_children[key] = 1
3268
# get all pending nodes
3269
this_iteration = pending
3270
build_details = self._vf._index.get_build_details(this_iteration)
3271
self._all_build_details.update(build_details)
3272
# new_nodes = self._vf._index._get_entries(this_iteration)
3274
for key, details in viewitems(build_details):
3275
(index_memo, compression_parent, parent_keys,
3276
record_details) = details
3277
self._parent_map[key] = parent_keys
3278
self._heads_provider = None
3279
records.append((key, index_memo))
3280
# Do we actually need to check _annotated_lines?
3281
pending.update([p for p in parent_keys
3282
if p not in self._all_build_details])
3284
for parent_key in parent_keys:
3285
if parent_key in self._num_needed_children:
3286
self._num_needed_children[parent_key] += 1
3288
self._num_needed_children[parent_key] = 1
3289
if compression_parent:
3290
if compression_parent in self._num_compression_children:
3291
self._num_compression_children[compression_parent] += 1
3293
self._num_compression_children[compression_parent] = 1
3295
missing_versions = this_iteration.difference(build_details)
3296
if missing_versions:
3297
for key in missing_versions:
3298
if key in self._parent_map and key in self._text_cache:
3299
# We already have this text ready, we just need to
3300
# yield it later so we get it annotated
3302
parent_keys = self._parent_map[key]
3303
for parent_key in parent_keys:
3304
if parent_key in self._num_needed_children:
3305
self._num_needed_children[parent_key] += 1
3307
self._num_needed_children[parent_key] = 1
3308
pending.update([p for p in parent_keys
3309
if p not in self._all_build_details])
3311
raise errors.RevisionNotPresent(key, self._vf)
3312
# Generally we will want to read the records in reverse order, because
3313
# we find the parent nodes after the children
3315
return records, ann_keys
3317
def _get_needed_texts(self, key, pb=None):
3318
# if True or len(self._vf._immediate_fallback_vfs) > 0:
3319
if len(self._vf._immediate_fallback_vfs) > 0:
3320
# If we have fallbacks, go to the generic path
3321
for v in annotate.Annotator._get_needed_texts(self, key, pb=pb):
3326
records, ann_keys = self._get_build_graph(key)
3327
for idx, (sub_key, text, num_lines) in enumerate(
3328
self._extract_texts(records)):
3330
pb.update(gettext('annotating'), idx, len(records))
3331
yield sub_key, text, num_lines
3332
for sub_key in ann_keys:
3333
text = self._text_cache[sub_key]
3334
num_lines = len(text) # bad assumption
3335
yield sub_key, text, num_lines
3337
except errors.RetryWithNewPacks as e:
3338
self._vf._access.reload_or_raise(e)
3339
# The cached build_details are no longer valid
3340
self._all_build_details.clear()
3342
def _cache_delta_blocks(self, key, compression_parent, delta, lines):
3343
parent_lines = self._text_cache[compression_parent]
3344
blocks = list(KnitContent.get_line_delta_blocks(delta, parent_lines, lines))
3345
self._matching_blocks[(key, compression_parent)] = blocks
3347
def _expand_record(self, key, parent_keys, compression_parent, record,
3350
if compression_parent:
3351
if compression_parent not in self._content_objects:
3352
# Waiting for the parent
3353
self._pending_deltas.setdefault(compression_parent, []).append(
3354
(key, parent_keys, record, record_details))
3356
# We have the basis parent, so expand the delta
3357
num = self._num_compression_children[compression_parent]
3360
base_content = self._content_objects.pop(compression_parent)
3361
self._num_compression_children.pop(compression_parent)
3363
self._num_compression_children[compression_parent] = num
3364
base_content = self._content_objects[compression_parent]
3365
# It is tempting to want to copy_base_content=False for the last
3366
# child object. However, whenever noeol=False,
3367
# self._text_cache[parent_key] is content._lines. So mutating it
3368
# gives very bad results.
3369
# The alternative is to copy the lines into text cache, but then we
3370
# are copying anyway, so just do it here.
3371
content, delta = self._vf._factory.parse_record(
3372
key, record, record_details, base_content,
3373
copy_base_content=True)
3376
content, _ = self._vf._factory.parse_record(
3377
key, record, record_details, None)
3378
if self._num_compression_children.get(key, 0) > 0:
3379
self._content_objects[key] = content
3380
lines = content.text()
3381
self._text_cache[key] = lines
3382
if delta is not None:
3383
self._cache_delta_blocks(key, compression_parent, delta, lines)
3386
def _get_parent_annotations_and_matches(self, key, text, parent_key):
3387
"""Get the list of annotations for the parent, and the matching lines.
3389
:param text: The opaque value given by _get_needed_texts
3390
:param parent_key: The key for the parent text
3391
:return: (parent_annotations, matching_blocks)
3392
parent_annotations is a list as long as the number of lines in
3394
matching_blocks is a list of (parent_idx, text_idx, len) tuples
3395
indicating which lines match between the two texts
3397
block_key = (key, parent_key)
3398
if block_key in self._matching_blocks:
3399
blocks = self._matching_blocks.pop(block_key)
3400
parent_annotations = self._annotations_cache[parent_key]
3401
return parent_annotations, blocks
3402
return annotate.Annotator._get_parent_annotations_and_matches(self,
3403
key, text, parent_key)
3405
def _process_pending(self, key):
3406
"""The content for 'key' was just processed.
3408
Determine if there is any more pending work to be processed.
3411
if key in self._pending_deltas:
3412
compression_parent = key
3413
children = self._pending_deltas.pop(key)
3414
for child_key, parent_keys, record, record_details in children:
3415
lines = self._expand_record(child_key, parent_keys,
3417
record, record_details)
3418
if self._check_ready_for_annotations(child_key, parent_keys):
3419
to_return.append(child_key)
3420
# Also check any children that are waiting for this parent to be
3422
if key in self._pending_annotation:
3423
children = self._pending_annotation.pop(key)
3424
to_return.extend([c for c, p_keys in children
3425
if self._check_ready_for_annotations(c, p_keys)])
3428
def _check_ready_for_annotations(self, key, parent_keys):
3429
"""return true if this text is ready to be yielded.
3431
Otherwise, this will return False, and queue the text into
3432
self._pending_annotation
3434
for parent_key in parent_keys:
3435
if parent_key not in self._annotations_cache:
3436
# still waiting on at least one parent text, so queue it up
3437
# Note that if there are multiple parents, we need to wait
3439
self._pending_annotation.setdefault(parent_key,
3440
[]).append((key, parent_keys))
3444
def _extract_texts(self, records):
3445
"""Extract the various texts needed based on records"""
3446
# We iterate in the order read, rather than a strict order requested
3447
# However, process what we can, and put off to the side things that
3448
# still need parents, cleaning them up when those parents are
3451
# 1) As 'records' are read, see if we can expand these records into
3452
# Content objects (and thus lines)
3453
# 2) If a given line-delta is waiting on its compression parent, it
3454
# gets queued up into self._pending_deltas, otherwise we expand
3455
# it, and put it into self._text_cache and self._content_objects
3456
# 3) If we expanded the text, we will then check to see if all
3457
# parents have also been processed. If so, this text gets yielded,
3458
# else this record gets set aside into pending_annotation
3459
# 4) Further, if we expanded the text in (2), we will then check to
3460
# see if there are any children in self._pending_deltas waiting to
3461
# also be processed. If so, we go back to (2) for those
3462
# 5) Further again, if we yielded the text, we can then check if that
3463
# 'unlocks' any of the texts in pending_annotations, which should
3464
# then get yielded as well
3465
# Note that both steps 4 and 5 are 'recursive' in that unlocking one
3466
# compression child could unlock yet another, and yielding a fulltext
3467
# will also 'unlock' the children that are waiting on that annotation.
3468
# (Though also, unlocking 1 parent's fulltext, does not unlock a child
3469
# if other parents are also waiting.)
3470
# We want to yield content before expanding child content objects, so
3471
# that we know when we can re-use the content lines, and the annotation
3472
# code can know when it can stop caching fulltexts, as well.
3474
# Children that are missing their compression parent
3476
for (key, record, digest) in self._vf._read_records_iter(records):
3478
details = self._all_build_details[key]
3479
(_, compression_parent, parent_keys, record_details) = details
3480
lines = self._expand_record(key, parent_keys, compression_parent,
3481
record, record_details)
3483
# Pending delta should be queued up
3485
# At this point, we may be able to yield this content, if all
3486
# parents are also finished
3487
yield_this_text = self._check_ready_for_annotations(key,
3490
# All parents present
3491
yield key, lines, len(lines)
3492
to_process = self._process_pending(key)
3494
this_process = to_process
3496
for key in this_process:
3497
lines = self._text_cache[key]
3498
yield key, lines, len(lines)
3499
to_process.extend(self._process_pending(key))
3502
from ._knit_load_data_pyx import _load_data_c as _load_data
3503
except ImportError as e:
3504
osutils.failed_to_load_extension(e)
3505
from ._knit_load_data_py import _load_data_py as _load_data
added
removed
breezy/knit.py
