1
# Copyright (C) 2005, 2006, 2007, 2008 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
"""Knit versionedfile implementation.
19
A knit is a versioned file implementation that supports efficient append only
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
end-marker; simply "end VERSION"
28
delta can be line or full contents.a
29
... the 8's there are the index number of the annotation.
30
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
8 e.set('executable', 'yes')
36
8 if elt.get('executable') == 'yes':
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
43
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
44
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
45
09:33 < lifeless> right
46
09:33 < jrydberg> lifeless: the position and size is the range in the data file
49
so the index sequence is the dictionary compressed sequence number used
50
in the deltas to provide line annotation
55
from cStringIO import StringIO
56
from itertools import izip
61
from bzrlib.lazy_import import lazy_import
62
lazy_import(globals(), """
84
from bzrlib.errors import (
92
RevisionAlreadyPresent,
95
from bzrlib.osutils import (
102
from bzrlib.versionedfile import (
103
AbsentContentFactory,
107
ChunkedContentFactory,
114
# TODO: Split out code specific to this format into an associated object.
116
# TODO: Can we put in some kind of value to check that the index and data
117
# files belong together?
119
# TODO: accommodate binaries, perhaps by storing a byte count
121
# TODO: function to check whole file
123
# TODO: atomically append data, then measure backwards from the cursor
124
# position after writing to work out where it was located. we may need to
125
# bypass python file buffering.
127
DATA_SUFFIX = '.knit'
128
INDEX_SUFFIX = '.kndx'
129
_STREAM_MIN_BUFFER_SIZE = 5*1024*1024
132
class KnitAdapter(object):
133
"""Base class for knit record adaption."""
135
def __init__(self, basis_vf):
136
"""Create an adapter which accesses full texts from basis_vf.
138
:param basis_vf: A versioned file to access basis texts of deltas from.
139
May be None for adapters that do not need to access basis texts.
141
self._data = KnitVersionedFiles(None, None)
142
self._annotate_factory = KnitAnnotateFactory()
143
self._plain_factory = KnitPlainFactory()
144
self._basis_vf = basis_vf
147
class FTAnnotatedToUnannotated(KnitAdapter):
148
"""An adapter from FT annotated knits to unannotated ones."""
150
def get_bytes(self, factory):
151
annotated_compressed_bytes = factory._raw_record
153
self._data._parse_record_unchecked(annotated_compressed_bytes)
154
content = self._annotate_factory.parse_fulltext(contents, rec[1])
155
size, bytes = self._data._record_to_data((rec[1],), rec[3], content.text())
159
class DeltaAnnotatedToUnannotated(KnitAdapter):
160
"""An adapter for deltas from annotated to unannotated."""
162
def get_bytes(self, factory):
163
annotated_compressed_bytes = factory._raw_record
165
self._data._parse_record_unchecked(annotated_compressed_bytes)
166
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
168
contents = self._plain_factory.lower_line_delta(delta)
169
size, bytes = self._data._record_to_data((rec[1],), rec[3], contents)
173
class FTAnnotatedToFullText(KnitAdapter):
174
"""An adapter from FT annotated knits to unannotated ones."""
176
def get_bytes(self, factory):
177
annotated_compressed_bytes = factory._raw_record
179
self._data._parse_record_unchecked(annotated_compressed_bytes)
180
content, delta = self._annotate_factory.parse_record(factory.key[-1],
181
contents, factory._build_details, None)
182
return ''.join(content.text())
185
class DeltaAnnotatedToFullText(KnitAdapter):
186
"""An adapter for deltas from annotated to unannotated."""
188
def get_bytes(self, factory):
189
annotated_compressed_bytes = factory._raw_record
191
self._data._parse_record_unchecked(annotated_compressed_bytes)
192
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
194
compression_parent = factory.parents[0]
195
basis_entry = self._basis_vf.get_record_stream(
196
[compression_parent], 'unordered', True).next()
197
if basis_entry.storage_kind == 'absent':
198
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
199
basis_chunks = basis_entry.get_bytes_as('chunked')
200
basis_lines = osutils.chunks_to_lines(basis_chunks)
201
# Manually apply the delta because we have one annotated content and
203
basis_content = PlainKnitContent(basis_lines, compression_parent)
204
basis_content.apply_delta(delta, rec[1])
205
basis_content._should_strip_eol = factory._build_details[1]
206
return ''.join(basis_content.text())
209
class FTPlainToFullText(KnitAdapter):
210
"""An adapter from FT plain knits to unannotated ones."""
212
def get_bytes(self, factory):
213
compressed_bytes = factory._raw_record
215
self._data._parse_record_unchecked(compressed_bytes)
216
content, delta = self._plain_factory.parse_record(factory.key[-1],
217
contents, factory._build_details, None)
218
return ''.join(content.text())
221
class DeltaPlainToFullText(KnitAdapter):
222
"""An adapter for deltas from annotated to unannotated."""
224
def get_bytes(self, factory):
225
compressed_bytes = factory._raw_record
227
self._data._parse_record_unchecked(compressed_bytes)
228
delta = self._plain_factory.parse_line_delta(contents, rec[1])
229
compression_parent = factory.parents[0]
230
# XXX: string splitting overhead.
231
basis_entry = self._basis_vf.get_record_stream(
232
[compression_parent], 'unordered', True).next()
233
if basis_entry.storage_kind == 'absent':
234
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
235
basis_chunks = basis_entry.get_bytes_as('chunked')
236
basis_lines = osutils.chunks_to_lines(basis_chunks)
237
basis_content = PlainKnitContent(basis_lines, compression_parent)
238
# Manually apply the delta because we have one annotated content and
240
content, _ = self._plain_factory.parse_record(rec[1], contents,
241
factory._build_details, basis_content)
242
return ''.join(content.text())
245
class KnitContentFactory(ContentFactory):
246
"""Content factory for streaming from knits.
248
:seealso ContentFactory:
251
def __init__(self, key, parents, build_details, sha1, raw_record,
252
annotated, knit=None, network_bytes=None):
253
"""Create a KnitContentFactory for key.
256
:param parents: The parents.
257
:param build_details: The build details as returned from
259
:param sha1: The sha1 expected from the full text of this object.
260
:param raw_record: The bytes of the knit data from disk.
261
:param annotated: True if the raw data is annotated.
262
:param network_bytes: None to calculate the network bytes on demand,
263
not-none if they are already known.
265
ContentFactory.__init__(self)
268
self.parents = parents
269
if build_details[0] == 'line-delta':
274
annotated_kind = 'annotated-'
277
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
278
self._raw_record = raw_record
279
self._network_bytes = network_bytes
280
self._build_details = build_details
283
def _create_network_bytes(self):
284
"""Create a fully serialised network version for transmission."""
285
# storage_kind, key, parents, Noeol, raw_record
286
key_bytes = '\x00'.join(self.key)
287
if self.parents is None:
288
parent_bytes = 'None:'
290
parent_bytes = '\t'.join('\x00'.join(key) for key in self.parents)
291
if self._build_details[1]:
295
network_bytes = "%s\n%s\n%s\n%s%s" % (self.storage_kind, key_bytes,
296
parent_bytes, noeol, self._raw_record)
297
self._network_bytes = network_bytes
299
def get_bytes_as(self, storage_kind):
300
if storage_kind == self.storage_kind:
301
if self._network_bytes is None:
302
self._create_network_bytes()
303
return self._network_bytes
304
if ('-ft-' in self.storage_kind and
305
storage_kind in ('chunked', 'fulltext')):
306
adapter_key = (self.storage_kind, 'fulltext')
307
adapter_factory = adapter_registry.get(adapter_key)
308
adapter = adapter_factory(None)
309
bytes = adapter.get_bytes(self)
310
if storage_kind == 'chunked':
314
if self._knit is not None:
315
# Not redundant with direct conversion above - that only handles
317
if storage_kind == 'chunked':
318
return self._knit.get_lines(self.key[0])
319
elif storage_kind == 'fulltext':
320
return self._knit.get_text(self.key[0])
321
raise errors.UnavailableRepresentation(self.key, storage_kind,
325
class LazyKnitContentFactory(ContentFactory):
326
"""A ContentFactory which can either generate full text or a wire form.
328
:seealso ContentFactory:
331
def __init__(self, key, parents, generator, first):
332
"""Create a LazyKnitContentFactory.
334
:param key: The key of the record.
335
:param parents: The parents of the record.
336
:param generator: A _ContentMapGenerator containing the record for this
338
:param first: Is this the first content object returned from generator?
339
if it is, its storage kind is knit-delta-closure, otherwise it is
340
knit-delta-closure-ref
343
self.parents = parents
345
self._generator = generator
346
self.storage_kind = "knit-delta-closure"
348
self.storage_kind = self.storage_kind + "-ref"
351
def get_bytes_as(self, storage_kind):
352
if storage_kind == self.storage_kind:
354
return self._generator._wire_bytes()
356
# all the keys etc are contained in the bytes returned in the
359
if storage_kind in ('chunked', 'fulltext'):
360
chunks = self._generator._get_one_work(self.key).text()
361
if storage_kind == 'chunked':
364
return ''.join(chunks)
365
raise errors.UnavailableRepresentation(self.key, storage_kind,
369
def knit_delta_closure_to_records(storage_kind, bytes, line_end):
370
"""Convert a network record to a iterator over stream records.
372
:param storage_kind: The storage kind of the record.
373
Must be 'knit-delta-closure'.
374
:param bytes: The bytes of the record on the network.
376
generator = _NetworkContentMapGenerator(bytes, line_end)
377
return generator.get_record_stream()
380
def knit_network_to_record(storage_kind, bytes, line_end):
381
"""Convert a network record to a record object.
383
:param storage_kind: The storage kind of the record.
384
:param bytes: The bytes of the record on the network.
387
line_end = bytes.find('\n', start)
388
key = tuple(bytes[start:line_end].split('\x00'))
390
line_end = bytes.find('\n', start)
391
parent_line = bytes[start:line_end]
392
if parent_line == 'None:':
396
[tuple(segment.split('\x00')) for segment in parent_line.split('\t')
399
noeol = bytes[start] == 'N'
400
if 'ft' in storage_kind:
403
method = 'line-delta'
404
build_details = (method, noeol)
406
raw_record = bytes[start:]
407
annotated = 'annotated' in storage_kind
408
return [KnitContentFactory(key, parents, build_details, None, raw_record,
409
annotated, network_bytes=bytes)]
412
class KnitContent(object):
413
"""Content of a knit version to which deltas can be applied.
415
This is always stored in memory as a list of lines with \n at the end,
416
plus a flag saying if the final ending is really there or not, because that
417
corresponds to the on-disk knit representation.
421
self._should_strip_eol = False
423
def apply_delta(self, delta, new_version_id):
424
"""Apply delta to this object to become new_version_id."""
425
raise NotImplementedError(self.apply_delta)
427
def line_delta_iter(self, new_lines):
428
"""Generate line-based delta from this content to new_lines."""
429
new_texts = new_lines.text()
430
old_texts = self.text()
431
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
432
for tag, i1, i2, j1, j2 in s.get_opcodes():
435
# ofrom, oto, length, data
436
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
438
def line_delta(self, new_lines):
439
return list(self.line_delta_iter(new_lines))
442
def get_line_delta_blocks(knit_delta, source, target):
443
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
444
target_len = len(target)
447
for s_begin, s_end, t_len, new_text in knit_delta:
448
true_n = s_begin - s_pos
451
# knit deltas do not provide reliable info about whether the
452
# last line of a file matches, due to eol handling.
453
if source[s_pos + n -1] != target[t_pos + n -1]:
456
yield s_pos, t_pos, n
457
t_pos += t_len + true_n
459
n = target_len - t_pos
461
if source[s_pos + n -1] != target[t_pos + n -1]:
464
yield s_pos, t_pos, n
465
yield s_pos + (target_len - t_pos), target_len, 0
468
class AnnotatedKnitContent(KnitContent):
469
"""Annotated content."""
471
def __init__(self, lines):
472
KnitContent.__init__(self)
476
"""Return a list of (origin, text) for each content line."""
477
lines = self._lines[:]
478
if self._should_strip_eol:
479
origin, last_line = lines[-1]
480
lines[-1] = (origin, last_line.rstrip('\n'))
483
def apply_delta(self, delta, new_version_id):
484
"""Apply delta to this object to become new_version_id."""
487
for start, end, count, delta_lines in delta:
488
lines[offset+start:offset+end] = delta_lines
489
offset = offset + (start - end) + count
493
lines = [text for origin, text in self._lines]
494
except ValueError, e:
495
# most commonly (only?) caused by the internal form of the knit
496
# missing annotation information because of a bug - see thread
498
raise KnitCorrupt(self,
499
"line in annotated knit missing annotation information: %s"
501
if self._should_strip_eol:
502
lines[-1] = lines[-1].rstrip('\n')
506
return AnnotatedKnitContent(self._lines[:])
509
class PlainKnitContent(KnitContent):
510
"""Unannotated content.
512
When annotate[_iter] is called on this content, the same version is reported
513
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
517
def __init__(self, lines, version_id):
518
KnitContent.__init__(self)
520
self._version_id = version_id
523
"""Return a list of (origin, text) for each content line."""
524
return [(self._version_id, line) for line in self._lines]
526
def apply_delta(self, delta, new_version_id):
527
"""Apply delta to this object to become new_version_id."""
530
for start, end, count, delta_lines in delta:
531
lines[offset+start:offset+end] = delta_lines
532
offset = offset + (start - end) + count
533
self._version_id = new_version_id
536
return PlainKnitContent(self._lines[:], self._version_id)
540
if self._should_strip_eol:
542
lines[-1] = lines[-1].rstrip('\n')
546
class _KnitFactory(object):
547
"""Base class for common Factory functions."""
549
def parse_record(self, version_id, record, record_details,
550
base_content, copy_base_content=True):
551
"""Parse a record into a full content object.
553
:param version_id: The official version id for this content
554
:param record: The data returned by read_records_iter()
555
:param record_details: Details about the record returned by
557
:param base_content: If get_build_details returns a compression_parent,
558
you must return a base_content here, else use None
559
:param copy_base_content: When building from the base_content, decide
560
you can either copy it and return a new object, or modify it in
562
:return: (content, delta) A Content object and possibly a line-delta,
565
method, noeol = record_details
566
if method == 'line-delta':
567
if copy_base_content:
568
content = base_content.copy()
570
content = base_content
571
delta = self.parse_line_delta(record, version_id)
572
content.apply_delta(delta, version_id)
574
content = self.parse_fulltext(record, version_id)
576
content._should_strip_eol = noeol
577
return (content, delta)
580
class KnitAnnotateFactory(_KnitFactory):
581
"""Factory for creating annotated Content objects."""
585
def make(self, lines, version_id):
586
num_lines = len(lines)
587
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
589
def parse_fulltext(self, content, version_id):
590
"""Convert fulltext to internal representation
592
fulltext content is of the format
593
revid(utf8) plaintext\n
594
internal representation is of the format:
597
# TODO: jam 20070209 The tests expect this to be returned as tuples,
598
# but the code itself doesn't really depend on that.
599
# Figure out a way to not require the overhead of turning the
600
# list back into tuples.
601
lines = [tuple(line.split(' ', 1)) for line in content]
602
return AnnotatedKnitContent(lines)
604
def parse_line_delta_iter(self, lines):
605
return iter(self.parse_line_delta(lines))
607
def parse_line_delta(self, lines, version_id, plain=False):
608
"""Convert a line based delta into internal representation.
610
line delta is in the form of:
611
intstart intend intcount
613
revid(utf8) newline\n
614
internal representation is
615
(start, end, count, [1..count tuples (revid, newline)])
617
:param plain: If True, the lines are returned as a plain
618
list without annotations, not as a list of (origin, content) tuples, i.e.
619
(start, end, count, [1..count newline])
626
def cache_and_return(line):
627
origin, text = line.split(' ', 1)
628
return cache.setdefault(origin, origin), text
630
# walk through the lines parsing.
631
# Note that the plain test is explicitly pulled out of the
632
# loop to minimise any performance impact
635
start, end, count = [int(n) for n in header.split(',')]
636
contents = [next().split(' ', 1)[1] for i in xrange(count)]
637
result.append((start, end, count, contents))
640
start, end, count = [int(n) for n in header.split(',')]
641
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
642
result.append((start, end, count, contents))
645
def get_fulltext_content(self, lines):
646
"""Extract just the content lines from a fulltext."""
647
return (line.split(' ', 1)[1] for line in lines)
649
def get_linedelta_content(self, lines):
650
"""Extract just the content from a line delta.
652
This doesn't return all of the extra information stored in a delta.
653
Only the actual content lines.
658
header = header.split(',')
659
count = int(header[2])
660
for i in xrange(count):
661
origin, text = next().split(' ', 1)
664
def lower_fulltext(self, content):
665
"""convert a fulltext content record into a serializable form.
667
see parse_fulltext which this inverts.
669
return ['%s %s' % (o, t) for o, t in content._lines]
671
def lower_line_delta(self, delta):
672
"""convert a delta into a serializable form.
674
See parse_line_delta which this inverts.
676
# TODO: jam 20070209 We only do the caching thing to make sure that
677
# the origin is a valid utf-8 line, eventually we could remove it
679
for start, end, c, lines in delta:
680
out.append('%d,%d,%d\n' % (start, end, c))
681
out.extend(origin + ' ' + text
682
for origin, text in lines)
685
def annotate(self, knit, key):
686
content = knit._get_content(key)
687
# adjust for the fact that serialised annotations are only key suffixes
689
if type(key) is tuple:
691
origins = content.annotate()
693
for origin, line in origins:
694
result.append((prefix + (origin,), line))
697
# XXX: This smells a bit. Why would key ever be a non-tuple here?
698
# Aren't keys defined to be tuples? -- spiv 20080618
699
return content.annotate()
702
class KnitPlainFactory(_KnitFactory):
703
"""Factory for creating plain Content objects."""
707
def make(self, lines, version_id):
708
return PlainKnitContent(lines, version_id)
710
def parse_fulltext(self, content, version_id):
711
"""This parses an unannotated fulltext.
713
Note that this is not a noop - the internal representation
714
has (versionid, line) - its just a constant versionid.
716
return self.make(content, version_id)
718
def parse_line_delta_iter(self, lines, version_id):
720
num_lines = len(lines)
721
while cur < num_lines:
724
start, end, c = [int(n) for n in header.split(',')]
725
yield start, end, c, lines[cur:cur+c]
728
def parse_line_delta(self, lines, version_id):
729
return list(self.parse_line_delta_iter(lines, version_id))
731
def get_fulltext_content(self, lines):
732
"""Extract just the content lines from a fulltext."""
735
def get_linedelta_content(self, lines):
736
"""Extract just the content from a line delta.
738
This doesn't return all of the extra information stored in a delta.
739
Only the actual content lines.
744
header = header.split(',')
745
count = int(header[2])
746
for i in xrange(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 = _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] for index_memo
849
in all_build_index_memos.itervalues()])
852
class KnitVersionedFiles(VersionedFiles):
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
'bzrlib.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._fallback_vfs = []
886
self._reload_func = reload_func
889
return "%s(%r, %r)" % (
890
self.__class__.__name__,
894
def add_fallback_versioned_files(self, a_versioned_files):
895
"""Add a source of texts for texts not present in this knit.
897
:param a_versioned_files: A VersionedFiles object.
899
self._fallback_vfs.append(a_versioned_files)
901
def add_lines(self, key, parents, lines, parent_texts=None,
902
left_matching_blocks=None, nostore_sha=None, random_id=False,
904
"""See VersionedFiles.add_lines()."""
905
self._index._check_write_ok()
906
self._check_add(key, lines, random_id, check_content)
908
# The caller might pass None if there is no graph data, but kndx
909
# indexes can't directly store that, so we give them
910
# an empty tuple instead.
912
line_bytes = ''.join(lines)
913
return self._add(key, lines, parents,
914
parent_texts, left_matching_blocks, nostore_sha, random_id,
915
line_bytes=line_bytes)
917
def _add_text(self, key, parents, text, nostore_sha=None, random_id=False):
918
"""See VersionedFiles._add_text()."""
919
self._index._check_write_ok()
920
self._check_add(key, None, random_id, check_content=False)
921
if text.__class__ is not str:
922
raise errors.BzrBadParameterUnicode("text")
924
# The caller might pass None if there is no graph data, but kndx
925
# indexes can't directly store that, so we give them
926
# an empty tuple instead.
928
return self._add(key, None, parents,
929
None, None, nostore_sha, random_id,
932
def _add(self, key, lines, parents, parent_texts,
933
left_matching_blocks, nostore_sha, random_id,
935
"""Add a set of lines on top of version specified by parents.
937
Any versions not present will be converted into ghosts.
939
:param lines: A list of strings where each one is a single line (has a
940
single newline at the end of the string) This is now optional
941
(callers can pass None). It is left in its location for backwards
942
compatibility. It should ''.join(lines) must == line_bytes
943
:param line_bytes: A single string containing the content
945
We pass both lines and line_bytes because different routes bring the
946
values to this function. And for memory efficiency, we don't want to
947
have to split/join on-demand.
949
# first thing, if the content is something we don't need to store, find
951
digest = sha_string(line_bytes)
952
if nostore_sha == digest:
953
raise errors.ExistingContent
956
if parent_texts is None:
958
# Do a single query to ascertain parent presence; we only compress
959
# against parents in the same kvf.
960
present_parent_map = self._index.get_parent_map(parents)
961
for parent in parents:
962
if parent in present_parent_map:
963
present_parents.append(parent)
965
# Currently we can only compress against the left most present parent.
966
if (len(present_parents) == 0 or
967
present_parents[0] != parents[0]):
970
# To speed the extract of texts the delta chain is limited
971
# to a fixed number of deltas. This should minimize both
972
# I/O and the time spend applying deltas.
973
delta = self._check_should_delta(present_parents[0])
975
text_length = len(line_bytes)
978
# Note: line_bytes is not modified to add a newline, that is tracked
979
# via the no_eol flag. 'lines' *is* modified, because that is the
980
# general values needed by the Content code.
981
if line_bytes and line_bytes[-1] != '\n':
982
options.append('no-eol')
984
# Copy the existing list, or create a new one
986
lines = osutils.split_lines(line_bytes)
989
# Replace the last line with one that ends in a final newline
990
lines[-1] = lines[-1] + '\n'
992
lines = osutils.split_lines(line_bytes)
994
for element in key[:-1]:
995
if type(element) is not str:
996
raise TypeError("key contains non-strings: %r" % (key,))
998
key = key[:-1] + ('sha1:' + digest,)
999
elif type(key[-1]) is not str:
1000
raise TypeError("key contains non-strings: %r" % (key,))
1001
# Knit hunks are still last-element only
1002
version_id = key[-1]
1003
content = self._factory.make(lines, version_id)
1005
# Hint to the content object that its text() call should strip the
1007
content._should_strip_eol = True
1008
if delta or (self._factory.annotated and len(present_parents) > 0):
1009
# Merge annotations from parent texts if needed.
1010
delta_hunks = self._merge_annotations(content, present_parents,
1011
parent_texts, delta, self._factory.annotated,
1012
left_matching_blocks)
1015
options.append('line-delta')
1016
store_lines = self._factory.lower_line_delta(delta_hunks)
1017
size, bytes = self._record_to_data(key, digest,
1020
options.append('fulltext')
1021
# isinstance is slower and we have no hierarchy.
1022
if self._factory.__class__ is KnitPlainFactory:
1023
# Use the already joined bytes saving iteration time in
1025
dense_lines = [line_bytes]
1027
dense_lines.append('\n')
1028
size, bytes = self._record_to_data(key, digest,
1031
# get mixed annotation + content and feed it into the
1033
store_lines = self._factory.lower_fulltext(content)
1034
size, bytes = self._record_to_data(key, digest,
1037
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
1038
self._index.add_records(
1039
((key, options, access_memo, parents),),
1040
random_id=random_id)
1041
return digest, text_length, content
1043
def annotate(self, key):
1044
"""See VersionedFiles.annotate."""
1045
return self._factory.annotate(self, key)
1047
def get_annotator(self):
1048
return _KnitAnnotator(self)
1050
def check(self, progress_bar=None, keys=None):
1051
"""See VersionedFiles.check()."""
1053
return self._logical_check()
1055
# At the moment, check does not extra work over get_record_stream
1056
return self.get_record_stream(keys, 'unordered', True)
1058
def _logical_check(self):
1059
# This doesn't actually test extraction of everything, but that will
1060
# impact 'bzr check' substantially, and needs to be integrated with
1061
# care. However, it does check for the obvious problem of a delta with
1063
keys = self._index.keys()
1064
parent_map = self.get_parent_map(keys)
1066
if self._index.get_method(key) != 'fulltext':
1067
compression_parent = parent_map[key][0]
1068
if compression_parent not in parent_map:
1069
raise errors.KnitCorrupt(self,
1070
"Missing basis parent %s for %s" % (
1071
compression_parent, key))
1072
for fallback_vfs in self._fallback_vfs:
1073
fallback_vfs.check()
1075
def _check_add(self, key, lines, random_id, check_content):
1076
"""check that version_id and lines are safe to add."""
1077
version_id = key[-1]
1078
if version_id is not None:
1079
if contains_whitespace(version_id):
1080
raise InvalidRevisionId(version_id, self)
1081
self.check_not_reserved_id(version_id)
1082
# TODO: If random_id==False and the key is already present, we should
1083
# probably check that the existing content is identical to what is
1084
# being inserted, and otherwise raise an exception. This would make
1085
# the bundle code simpler.
1087
self._check_lines_not_unicode(lines)
1088
self._check_lines_are_lines(lines)
1090
def _check_header(self, key, line):
1091
rec = self._split_header(line)
1092
self._check_header_version(rec, key[-1])
1095
def _check_header_version(self, rec, version_id):
1096
"""Checks the header version on original format knit records.
1098
These have the last component of the key embedded in the record.
1100
if rec[1] != version_id:
1101
raise KnitCorrupt(self,
1102
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1104
def _check_should_delta(self, parent):
1105
"""Iterate back through the parent listing, looking for a fulltext.
1107
This is used when we want to decide whether to add a delta or a new
1108
fulltext. It searches for _max_delta_chain parents. When it finds a
1109
fulltext parent, it sees if the total size of the deltas leading up to
1110
it is large enough to indicate that we want a new full text anyway.
1112
Return True if we should create a new delta, False if we should use a
1116
fulltext_size = None
1117
for count in xrange(self._max_delta_chain):
1119
# Note that this only looks in the index of this particular
1120
# KnitVersionedFiles, not in the fallbacks. This ensures that
1121
# we won't store a delta spanning physical repository
1123
build_details = self._index.get_build_details([parent])
1124
parent_details = build_details[parent]
1125
except (RevisionNotPresent, KeyError), e:
1126
# Some basis is not locally present: always fulltext
1128
index_memo, compression_parent, _, _ = parent_details
1129
_, _, size = index_memo
1130
if compression_parent is None:
1131
fulltext_size = size
1134
# We don't explicitly check for presence because this is in an
1135
# inner loop, and if it's missing it'll fail anyhow.
1136
parent = compression_parent
1138
# We couldn't find a fulltext, so we must create a new one
1140
# Simple heuristic - if the total I/O wold be greater as a delta than
1141
# the originally installed fulltext, we create a new fulltext.
1142
return fulltext_size > delta_size
1144
def _build_details_to_components(self, build_details):
1145
"""Convert a build_details tuple to a position tuple."""
1146
# record_details, access_memo, compression_parent
1147
return build_details[3], build_details[0], build_details[1]
1149
def _get_components_positions(self, keys, allow_missing=False):
1150
"""Produce a map of position data for the components of keys.
1152
This data is intended to be used for retrieving the knit records.
1154
A dict of key to (record_details, index_memo, next, parents) is
1156
method is the way referenced data should be applied.
1157
index_memo is the handle to pass to the data access to actually get the
1159
next is the build-parent of the version, or None for fulltexts.
1160
parents is the version_ids of the parents of this version
1162
:param allow_missing: If True do not raise an error on a missing component,
1166
pending_components = keys
1167
while pending_components:
1168
build_details = self._index.get_build_details(pending_components)
1169
current_components = set(pending_components)
1170
pending_components = set()
1171
for key, details in build_details.iteritems():
1172
(index_memo, compression_parent, parents,
1173
record_details) = details
1174
method = record_details[0]
1175
if compression_parent is not None:
1176
pending_components.add(compression_parent)
1177
component_data[key] = self._build_details_to_components(details)
1178
missing = current_components.difference(build_details)
1179
if missing and not allow_missing:
1180
raise errors.RevisionNotPresent(missing.pop(), self)
1181
return component_data
1183
def _get_content(self, key, parent_texts={}):
1184
"""Returns a content object that makes up the specified
1186
cached_version = parent_texts.get(key, None)
1187
if cached_version is not None:
1188
# Ensure the cache dict is valid.
1189
if not self.get_parent_map([key]):
1190
raise RevisionNotPresent(key, self)
1191
return cached_version
1192
generator = _VFContentMapGenerator(self, [key])
1193
return generator._get_content(key)
1195
def get_known_graph_ancestry(self, keys):
1196
"""Get a KnownGraph instance with the ancestry of keys."""
1197
parent_map, missing_keys = self._index.find_ancestry(keys)
1198
for fallback in self._fallback_vfs:
1199
if not missing_keys:
1201
(f_parent_map, f_missing_keys) = fallback._index.find_ancestry(
1203
parent_map.update(f_parent_map)
1204
missing_keys = f_missing_keys
1205
kg = _mod_graph.KnownGraph(parent_map)
1208
def get_parent_map(self, keys):
1209
"""Get a map of the graph parents of keys.
1211
:param keys: The keys to look up parents for.
1212
:return: A mapping from keys to parents. Absent keys are absent from
1215
return self._get_parent_map_with_sources(keys)[0]
1217
def _get_parent_map_with_sources(self, keys):
1218
"""Get a map of the parents of keys.
1220
:param keys: The keys to look up parents for.
1221
:return: A tuple. The first element is a mapping from keys to parents.
1222
Absent keys are absent from the mapping. The second element is a
1223
list with the locations each key was found in. The first element
1224
is the in-this-knit parents, the second the first fallback source,
1228
sources = [self._index] + self._fallback_vfs
1231
for source in sources:
1234
new_result = source.get_parent_map(missing)
1235
source_results.append(new_result)
1236
result.update(new_result)
1237
missing.difference_update(set(new_result))
1238
return result, source_results
1240
def _get_record_map(self, keys, allow_missing=False):
1241
"""Produce a dictionary of knit records.
1243
:return: {key:(record, record_details, digest, next)}
1245
data returned from read_records (a KnitContentobject)
1247
opaque information to pass to parse_record
1249
SHA1 digest of the full text after all steps are done
1251
build-parent of the version, i.e. the leftmost ancestor.
1252
Will be None if the record is not a delta.
1253
:param keys: The keys to build a map for
1254
:param allow_missing: If some records are missing, rather than
1255
error, just return the data that could be generated.
1257
raw_map = self._get_record_map_unparsed(keys,
1258
allow_missing=allow_missing)
1259
return self._raw_map_to_record_map(raw_map)
1261
def _raw_map_to_record_map(self, raw_map):
1262
"""Parse the contents of _get_record_map_unparsed.
1264
:return: see _get_record_map.
1268
data, record_details, next = raw_map[key]
1269
content, digest = self._parse_record(key[-1], data)
1270
result[key] = content, record_details, digest, next
1273
def _get_record_map_unparsed(self, keys, allow_missing=False):
1274
"""Get the raw data for reconstructing keys without parsing it.
1276
:return: A dict suitable for parsing via _raw_map_to_record_map.
1277
key-> raw_bytes, (method, noeol), compression_parent
1279
# This retries the whole request if anything fails. Potentially we
1280
# could be a bit more selective. We could track the keys whose records
1281
# we have successfully found, and then only request the new records
1282
# from there. However, _get_components_positions grabs the whole build
1283
# chain, which means we'll likely try to grab the same records again
1284
# anyway. Also, can the build chains change as part of a pack
1285
# operation? We wouldn't want to end up with a broken chain.
1288
position_map = self._get_components_positions(keys,
1289
allow_missing=allow_missing)
1290
# key = component_id, r = record_details, i_m = index_memo,
1292
records = [(key, i_m) for key, (r, i_m, n)
1293
in position_map.iteritems()]
1294
# Sort by the index memo, so that we request records from the
1295
# same pack file together, and in forward-sorted order
1296
records.sort(key=operator.itemgetter(1))
1298
for key, data in self._read_records_iter_unchecked(records):
1299
(record_details, index_memo, next) = position_map[key]
1300
raw_record_map[key] = data, record_details, next
1301
return raw_record_map
1302
except errors.RetryWithNewPacks, e:
1303
self._access.reload_or_raise(e)
1306
def _split_by_prefix(cls, keys):
1307
"""For the given keys, split them up based on their prefix.
1309
To keep memory pressure somewhat under control, split the
1310
requests back into per-file-id requests, otherwise "bzr co"
1311
extracts the full tree into memory before writing it to disk.
1312
This should be revisited if _get_content_maps() can ever cross
1315
The keys for a given file_id are kept in the same relative order.
1316
Ordering between file_ids is not, though prefix_order will return the
1317
order that the key was first seen.
1319
:param keys: An iterable of key tuples
1320
:return: (split_map, prefix_order)
1321
split_map A dictionary mapping prefix => keys
1322
prefix_order The order that we saw the various prefixes
1324
split_by_prefix = {}
1332
if prefix in split_by_prefix:
1333
split_by_prefix[prefix].append(key)
1335
split_by_prefix[prefix] = [key]
1336
prefix_order.append(prefix)
1337
return split_by_prefix, prefix_order
1339
def _group_keys_for_io(self, keys, non_local_keys, positions,
1340
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1341
"""For the given keys, group them into 'best-sized' requests.
1343
The idea is to avoid making 1 request per file, but to never try to
1344
unpack an entire 1.5GB source tree in a single pass. Also when
1345
possible, we should try to group requests to the same pack file
1348
:return: list of (keys, non_local) tuples that indicate what keys
1349
should be fetched next.
1351
# TODO: Ideally we would group on 2 factors. We want to extract texts
1352
# from the same pack file together, and we want to extract all
1353
# the texts for a given build-chain together. Ultimately it
1354
# probably needs a better global view.
1355
total_keys = len(keys)
1356
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1357
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1359
cur_non_local = set()
1363
for prefix in prefix_order:
1364
keys = prefix_split_keys[prefix]
1365
non_local = prefix_split_non_local_keys.get(prefix, [])
1367
this_size = self._index._get_total_build_size(keys, positions)
1368
cur_size += this_size
1369
cur_keys.extend(keys)
1370
cur_non_local.update(non_local)
1371
if cur_size > _min_buffer_size:
1372
result.append((cur_keys, cur_non_local))
1373
sizes.append(cur_size)
1375
cur_non_local = set()
1378
result.append((cur_keys, cur_non_local))
1379
sizes.append(cur_size)
1382
def get_record_stream(self, keys, ordering, include_delta_closure):
1383
"""Get a stream of records for keys.
1385
:param keys: The keys to include.
1386
:param ordering: Either 'unordered' or 'topological'. A topologically
1387
sorted stream has compression parents strictly before their
1389
:param include_delta_closure: If True then the closure across any
1390
compression parents will be included (in the opaque data).
1391
:return: An iterator of ContentFactory objects, each of which is only
1392
valid until the iterator is advanced.
1394
# keys might be a generator
1398
if not self._index.has_graph:
1399
# Cannot sort when no graph has been stored.
1400
ordering = 'unordered'
1402
remaining_keys = keys
1405
keys = set(remaining_keys)
1406
for content_factory in self._get_remaining_record_stream(keys,
1407
ordering, include_delta_closure):
1408
remaining_keys.discard(content_factory.key)
1409
yield content_factory
1411
except errors.RetryWithNewPacks, e:
1412
self._access.reload_or_raise(e)
1414
def _get_remaining_record_stream(self, keys, ordering,
1415
include_delta_closure):
1416
"""This function is the 'retry' portion for get_record_stream."""
1417
if include_delta_closure:
1418
positions = self._get_components_positions(keys, allow_missing=True)
1420
build_details = self._index.get_build_details(keys)
1422
# (record_details, access_memo, compression_parent_key)
1423
positions = dict((key, self._build_details_to_components(details))
1424
for key, details in build_details.iteritems())
1425
absent_keys = keys.difference(set(positions))
1426
# There may be more absent keys : if we're missing the basis component
1427
# and are trying to include the delta closure.
1428
# XXX: We should not ever need to examine remote sources because we do
1429
# not permit deltas across versioned files boundaries.
1430
if include_delta_closure:
1431
needed_from_fallback = set()
1432
# Build up reconstructable_keys dict. key:True in this dict means
1433
# the key can be reconstructed.
1434
reconstructable_keys = {}
1438
chain = [key, positions[key][2]]
1440
needed_from_fallback.add(key)
1443
while chain[-1] is not None:
1444
if chain[-1] in reconstructable_keys:
1445
result = reconstructable_keys[chain[-1]]
1449
chain.append(positions[chain[-1]][2])
1451
# missing basis component
1452
needed_from_fallback.add(chain[-1])
1455
for chain_key in chain[:-1]:
1456
reconstructable_keys[chain_key] = result
1458
needed_from_fallback.add(key)
1459
# Double index lookups here : need a unified api ?
1460
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1461
if ordering in ('topological', 'groupcompress'):
1462
if ordering == 'topological':
1463
# Global topological sort
1464
present_keys = tsort.topo_sort(global_map)
1466
present_keys = sort_groupcompress(global_map)
1467
# Now group by source:
1469
current_source = None
1470
for key in present_keys:
1471
for parent_map in parent_maps:
1472
if key in parent_map:
1473
key_source = parent_map
1475
if current_source is not key_source:
1476
source_keys.append((key_source, []))
1477
current_source = key_source
1478
source_keys[-1][1].append(key)
1480
if ordering != 'unordered':
1481
raise AssertionError('valid values for ordering are:'
1482
' "unordered", "groupcompress" or "topological" not: %r'
1484
# Just group by source; remote sources first.
1487
for parent_map in reversed(parent_maps):
1488
source_keys.append((parent_map, []))
1489
for key in parent_map:
1490
present_keys.append(key)
1491
source_keys[-1][1].append(key)
1492
# We have been requested to return these records in an order that
1493
# suits us. So we ask the index to give us an optimally sorted
1495
for source, sub_keys in source_keys:
1496
if source is parent_maps[0]:
1497
# Only sort the keys for this VF
1498
self._index._sort_keys_by_io(sub_keys, positions)
1499
absent_keys = keys - set(global_map)
1500
for key in absent_keys:
1501
yield AbsentContentFactory(key)
1502
# restrict our view to the keys we can answer.
1503
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1504
# XXX: At that point we need to consider the impact of double reads by
1505
# utilising components multiple times.
1506
if include_delta_closure:
1507
# XXX: get_content_maps performs its own index queries; allow state
1509
non_local_keys = needed_from_fallback - absent_keys
1510
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1513
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1516
for record in generator.get_record_stream():
1519
for source, keys in source_keys:
1520
if source is parent_maps[0]:
1521
# this KnitVersionedFiles
1522
records = [(key, positions[key][1]) for key in keys]
1523
for key, raw_data in self._read_records_iter_unchecked(records):
1524
(record_details, index_memo, _) = positions[key]
1525
yield KnitContentFactory(key, global_map[key],
1526
record_details, None, raw_data, self._factory.annotated, None)
1528
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1529
for record in vf.get_record_stream(keys, ordering,
1530
include_delta_closure):
1533
def get_sha1s(self, keys):
1534
"""See VersionedFiles.get_sha1s()."""
1536
record_map = self._get_record_map(missing, allow_missing=True)
1538
for key, details in record_map.iteritems():
1539
if key not in missing:
1541
# record entry 2 is the 'digest'.
1542
result[key] = details[2]
1543
missing.difference_update(set(result))
1544
for source in self._fallback_vfs:
1547
new_result = source.get_sha1s(missing)
1548
result.update(new_result)
1549
missing.difference_update(set(new_result))
1552
def insert_record_stream(self, stream):
1553
"""Insert a record stream into this container.
1555
:param stream: A stream of records to insert.
1557
:seealso VersionedFiles.get_record_stream:
1559
def get_adapter(adapter_key):
1561
return adapters[adapter_key]
1563
adapter_factory = adapter_registry.get(adapter_key)
1564
adapter = adapter_factory(self)
1565
adapters[adapter_key] = adapter
1568
if self._factory.annotated:
1569
# self is annotated, we need annotated knits to use directly.
1570
annotated = "annotated-"
1573
# self is not annotated, but we can strip annotations cheaply.
1575
convertibles = set(["knit-annotated-ft-gz"])
1576
if self._max_delta_chain:
1577
delta_types.add("knit-annotated-delta-gz")
1578
convertibles.add("knit-annotated-delta-gz")
1579
# The set of types we can cheaply adapt without needing basis texts.
1580
native_types = set()
1581
if self._max_delta_chain:
1582
native_types.add("knit-%sdelta-gz" % annotated)
1583
delta_types.add("knit-%sdelta-gz" % annotated)
1584
native_types.add("knit-%sft-gz" % annotated)
1585
knit_types = native_types.union(convertibles)
1587
# Buffer all index entries that we can't add immediately because their
1588
# basis parent is missing. We don't buffer all because generating
1589
# annotations may require access to some of the new records. However we
1590
# can't generate annotations from new deltas until their basis parent
1591
# is present anyway, so we get away with not needing an index that
1592
# includes the new keys.
1594
# See <http://launchpad.net/bugs/300177> about ordering of compression
1595
# parents in the records - to be conservative, we insist that all
1596
# parents must be present to avoid expanding to a fulltext.
1598
# key = basis_parent, value = index entry to add
1599
buffered_index_entries = {}
1600
for record in stream:
1601
kind = record.storage_kind
1602
if kind.startswith('knit-') and kind.endswith('-gz'):
1603
# Check that the ID in the header of the raw knit bytes matches
1604
# the record metadata.
1605
raw_data = record._raw_record
1606
df, rec = self._parse_record_header(record.key, raw_data)
1609
parents = record.parents
1610
if record.storage_kind in delta_types:
1611
# TODO: eventually the record itself should track
1612
# compression_parent
1613
compression_parent = parents[0]
1615
compression_parent = None
1616
# Raise an error when a record is missing.
1617
if record.storage_kind == 'absent':
1618
raise RevisionNotPresent([record.key], self)
1619
elif ((record.storage_kind in knit_types)
1620
and (compression_parent is None
1621
or not self._fallback_vfs
1622
or self._index.has_key(compression_parent)
1623
or not self.has_key(compression_parent))):
1624
# we can insert the knit record literally if either it has no
1625
# compression parent OR we already have its basis in this kvf
1626
# OR the basis is not present even in the fallbacks. In the
1627
# last case it will either turn up later in the stream and all
1628
# will be well, or it won't turn up at all and we'll raise an
1631
# TODO: self.has_key is somewhat redundant with
1632
# self._index.has_key; we really want something that directly
1633
# asks if it's only present in the fallbacks. -- mbp 20081119
1634
if record.storage_kind not in native_types:
1636
adapter_key = (record.storage_kind, "knit-delta-gz")
1637
adapter = get_adapter(adapter_key)
1639
adapter_key = (record.storage_kind, "knit-ft-gz")
1640
adapter = get_adapter(adapter_key)
1641
bytes = adapter.get_bytes(record)
1643
# It's a knit record, it has a _raw_record field (even if
1644
# it was reconstituted from a network stream).
1645
bytes = record._raw_record
1646
options = [record._build_details[0]]
1647
if record._build_details[1]:
1648
options.append('no-eol')
1649
# Just blat it across.
1650
# Note: This does end up adding data on duplicate keys. As
1651
# modern repositories use atomic insertions this should not
1652
# lead to excessive growth in the event of interrupted fetches.
1653
# 'knit' repositories may suffer excessive growth, but as a
1654
# deprecated format this is tolerable. It can be fixed if
1655
# needed by in the kndx index support raising on a duplicate
1656
# add with identical parents and options.
1657
access_memo = self._access.add_raw_records(
1658
[(record.key, len(bytes))], bytes)[0]
1659
index_entry = (record.key, options, access_memo, parents)
1660
if 'fulltext' not in options:
1661
# Not a fulltext, so we need to make sure the compression
1662
# parent will also be present.
1663
# Note that pack backed knits don't need to buffer here
1664
# because they buffer all writes to the transaction level,
1665
# but we don't expose that difference at the index level. If
1666
# the query here has sufficient cost to show up in
1667
# profiling we should do that.
1669
# They're required to be physically in this
1670
# KnitVersionedFiles, not in a fallback.
1671
if not self._index.has_key(compression_parent):
1672
pending = buffered_index_entries.setdefault(
1673
compression_parent, [])
1674
pending.append(index_entry)
1677
self._index.add_records([index_entry])
1678
elif record.storage_kind == 'chunked':
1679
self.add_lines(record.key, parents,
1680
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1682
# Not suitable for direct insertion as a
1683
# delta, either because it's not the right format, or this
1684
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1685
# 0) or because it depends on a base only present in the
1687
self._access.flush()
1689
# Try getting a fulltext directly from the record.
1690
bytes = record.get_bytes_as('fulltext')
1691
except errors.UnavailableRepresentation:
1692
adapter_key = record.storage_kind, 'fulltext'
1693
adapter = get_adapter(adapter_key)
1694
bytes = adapter.get_bytes(record)
1695
lines = split_lines(bytes)
1697
self.add_lines(record.key, parents, lines)
1698
except errors.RevisionAlreadyPresent:
1700
# Add any records whose basis parent is now available.
1702
added_keys = [record.key]
1704
key = added_keys.pop(0)
1705
if key in buffered_index_entries:
1706
index_entries = buffered_index_entries[key]
1707
self._index.add_records(index_entries)
1709
[index_entry[0] for index_entry in index_entries])
1710
del buffered_index_entries[key]
1711
if buffered_index_entries:
1712
# There were index entries buffered at the end of the stream,
1713
# So these need to be added (if the index supports holding such
1714
# entries for later insertion)
1716
for key in buffered_index_entries:
1717
index_entries = buffered_index_entries[key]
1718
all_entries.extend(index_entries)
1719
self._index.add_records(
1720
all_entries, missing_compression_parents=True)
1722
def get_missing_compression_parent_keys(self):
1723
"""Return an iterable of keys of missing compression parents.
1725
Check this after calling insert_record_stream to find out if there are
1726
any missing compression parents. If there are, the records that
1727
depend on them are not able to be inserted safely. For atomic
1728
KnitVersionedFiles built on packs, the transaction should be aborted or
1729
suspended - commit will fail at this point. Nonatomic knits will error
1730
earlier because they have no staging area to put pending entries into.
1732
return self._index.get_missing_compression_parents()
1734
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1735
"""Iterate over the lines in the versioned files from keys.
1737
This may return lines from other keys. Each item the returned
1738
iterator yields is a tuple of a line and a text version that that line
1739
is present in (not introduced in).
1741
Ordering of results is in whatever order is most suitable for the
1742
underlying storage format.
1744
If a progress bar is supplied, it may be used to indicate progress.
1745
The caller is responsible for cleaning up progress bars (because this
1749
* Lines are normalised by the underlying store: they will all have \\n
1751
* Lines are returned in arbitrary order.
1752
* If a requested key did not change any lines (or didn't have any
1753
lines), it may not be mentioned at all in the result.
1755
:param pb: Progress bar supplied by caller.
1756
:return: An iterator over (line, key).
1759
pb = ui.ui_factory.nested_progress_bar()
1765
# we don't care about inclusions, the caller cares.
1766
# but we need to setup a list of records to visit.
1767
# we need key, position, length
1769
build_details = self._index.get_build_details(keys)
1770
for key, details in build_details.iteritems():
1772
key_records.append((key, details[0]))
1773
records_iter = enumerate(self._read_records_iter(key_records))
1774
for (key_idx, (key, data, sha_value)) in records_iter:
1775
pb.update('Walking content', key_idx, total)
1776
compression_parent = build_details[key][1]
1777
if compression_parent is None:
1779
line_iterator = self._factory.get_fulltext_content(data)
1782
line_iterator = self._factory.get_linedelta_content(data)
1783
# Now that we are yielding the data for this key, remove it
1786
# XXX: It might be more efficient to yield (key,
1787
# line_iterator) in the future. However for now, this is a
1788
# simpler change to integrate into the rest of the
1789
# codebase. RBC 20071110
1790
for line in line_iterator:
1793
except errors.RetryWithNewPacks, e:
1794
self._access.reload_or_raise(e)
1795
# If there are still keys we've not yet found, we look in the fallback
1796
# vfs, and hope to find them there. Note that if the keys are found
1797
# but had no changes or no content, the fallback may not return
1799
if keys and not self._fallback_vfs:
1800
# XXX: strictly the second parameter is meant to be the file id
1801
# but it's not easily accessible here.
1802
raise RevisionNotPresent(keys, repr(self))
1803
for source in self._fallback_vfs:
1807
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1808
source_keys.add(key)
1810
keys.difference_update(source_keys)
1811
pb.update('Walking content', total, total)
1813
def _make_line_delta(self, delta_seq, new_content):
1814
"""Generate a line delta from delta_seq and new_content."""
1816
for op in delta_seq.get_opcodes():
1817
if op[0] == 'equal':
1819
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1822
def _merge_annotations(self, content, parents, parent_texts={},
1823
delta=None, annotated=None,
1824
left_matching_blocks=None):
1825
"""Merge annotations for content and generate deltas.
1827
This is done by comparing the annotations based on changes to the text
1828
and generating a delta on the resulting full texts. If annotations are
1829
not being created then a simple delta is created.
1831
if left_matching_blocks is not None:
1832
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1836
for parent_key in parents:
1837
merge_content = self._get_content(parent_key, parent_texts)
1838
if (parent_key == parents[0] and delta_seq is not None):
1841
seq = patiencediff.PatienceSequenceMatcher(
1842
None, merge_content.text(), content.text())
1843
for i, j, n in seq.get_matching_blocks():
1846
# this copies (origin, text) pairs across to the new
1847
# content for any line that matches the last-checked
1849
content._lines[j:j+n] = merge_content._lines[i:i+n]
1850
# XXX: Robert says the following block is a workaround for a
1851
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1852
if content._lines and content._lines[-1][1][-1] != '\n':
1853
# The copied annotation was from a line without a trailing EOL,
1854
# reinstate one for the content object, to ensure correct
1856
line = content._lines[-1][1] + '\n'
1857
content._lines[-1] = (content._lines[-1][0], line)
1859
if delta_seq is None:
1860
reference_content = self._get_content(parents[0], parent_texts)
1861
new_texts = content.text()
1862
old_texts = reference_content.text()
1863
delta_seq = patiencediff.PatienceSequenceMatcher(
1864
None, old_texts, new_texts)
1865
return self._make_line_delta(delta_seq, content)
1867
def _parse_record(self, version_id, data):
1868
"""Parse an original format knit record.
1870
These have the last element of the key only present in the stored data.
1872
rec, record_contents = self._parse_record_unchecked(data)
1873
self._check_header_version(rec, version_id)
1874
return record_contents, rec[3]
1876
def _parse_record_header(self, key, raw_data):
1877
"""Parse a record header for consistency.
1879
:return: the header and the decompressor stream.
1880
as (stream, header_record)
1882
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1885
rec = self._check_header(key, df.readline())
1886
except Exception, e:
1887
raise KnitCorrupt(self,
1888
"While reading {%s} got %s(%s)"
1889
% (key, e.__class__.__name__, str(e)))
1892
def _parse_record_unchecked(self, data):
1894
# 4168 calls in 2880 217 internal
1895
# 4168 calls to _parse_record_header in 2121
1896
# 4168 calls to readlines in 330
1897
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1899
record_contents = df.readlines()
1900
except Exception, e:
1901
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1902
(data, e.__class__.__name__, str(e)))
1903
header = record_contents.pop(0)
1904
rec = self._split_header(header)
1905
last_line = record_contents.pop()
1906
if len(record_contents) != int(rec[2]):
1907
raise KnitCorrupt(self,
1908
'incorrect number of lines %s != %s'
1909
' for version {%s} %s'
1910
% (len(record_contents), int(rec[2]),
1911
rec[1], record_contents))
1912
if last_line != 'end %s\n' % rec[1]:
1913
raise KnitCorrupt(self,
1914
'unexpected version end line %r, wanted %r'
1915
% (last_line, rec[1]))
1917
return rec, record_contents
1919
def _read_records_iter(self, records):
1920
"""Read text records from data file and yield result.
1922
The result will be returned in whatever is the fastest to read.
1923
Not by the order requested. Also, multiple requests for the same
1924
record will only yield 1 response.
1925
:param records: A list of (key, access_memo) entries
1926
:return: Yields (key, contents, digest) in the order
1927
read, not the order requested
1932
# XXX: This smells wrong, IO may not be getting ordered right.
1933
needed_records = sorted(set(records), key=operator.itemgetter(1))
1934
if not needed_records:
1937
# The transport optimizes the fetching as well
1938
# (ie, reads continuous ranges.)
1939
raw_data = self._access.get_raw_records(
1940
[index_memo for key, index_memo in needed_records])
1942
for (key, index_memo), data in \
1943
izip(iter(needed_records), raw_data):
1944
content, digest = self._parse_record(key[-1], data)
1945
yield key, content, digest
1947
def _read_records_iter_raw(self, records):
1948
"""Read text records from data file and yield raw data.
1950
This unpacks enough of the text record to validate the id is
1951
as expected but thats all.
1953
Each item the iterator yields is (key, bytes,
1954
expected_sha1_of_full_text).
1956
for key, data in self._read_records_iter_unchecked(records):
1957
# validate the header (note that we can only use the suffix in
1958
# current knit records).
1959
df, rec = self._parse_record_header(key, data)
1961
yield key, data, rec[3]
1963
def _read_records_iter_unchecked(self, records):
1964
"""Read text records from data file and yield raw data.
1966
No validation is done.
1968
Yields tuples of (key, data).
1970
# setup an iterator of the external records:
1971
# uses readv so nice and fast we hope.
1973
# grab the disk data needed.
1974
needed_offsets = [index_memo for key, index_memo
1976
raw_records = self._access.get_raw_records(needed_offsets)
1978
for key, index_memo in records:
1979
data = raw_records.next()
1982
def _record_to_data(self, key, digest, lines, dense_lines=None):
1983
"""Convert key, digest, lines into a raw data block.
1985
:param key: The key of the record. Currently keys are always serialised
1986
using just the trailing component.
1987
:param dense_lines: The bytes of lines but in a denser form. For
1988
instance, if lines is a list of 1000 bytestrings each ending in \n,
1989
dense_lines may be a list with one line in it, containing all the
1990
1000's lines and their \n's. Using dense_lines if it is already
1991
known is a win because the string join to create bytes in this
1992
function spends less time resizing the final string.
1993
:return: (len, a StringIO instance with the raw data ready to read.)
1995
chunks = ["version %s %d %s\n" % (key[-1], len(lines), digest)]
1996
chunks.extend(dense_lines or lines)
1997
chunks.append("end %s\n" % key[-1])
1998
for chunk in chunks:
1999
if type(chunk) is not str:
2000
raise AssertionError(
2001
'data must be plain bytes was %s' % type(chunk))
2002
if lines and lines[-1][-1] != '\n':
2003
raise ValueError('corrupt lines value %r' % lines)
2004
compressed_bytes = tuned_gzip.chunks_to_gzip(chunks)
2005
return len(compressed_bytes), compressed_bytes
2007
def _split_header(self, line):
2010
raise KnitCorrupt(self,
2011
'unexpected number of elements in record header')
2015
"""See VersionedFiles.keys."""
2016
if 'evil' in debug.debug_flags:
2017
trace.mutter_callsite(2, "keys scales with size of history")
2018
sources = [self._index] + self._fallback_vfs
2020
for source in sources:
2021
result.update(source.keys())
2025
class _ContentMapGenerator(object):
2026
"""Generate texts or expose raw deltas for a set of texts."""
2028
def __init__(self, ordering='unordered'):
2029
self._ordering = ordering
2031
def _get_content(self, key):
2032
"""Get the content object for key."""
2033
# Note that _get_content is only called when the _ContentMapGenerator
2034
# has been constructed with just one key requested for reconstruction.
2035
if key in self.nonlocal_keys:
2036
record = self.get_record_stream().next()
2037
# Create a content object on the fly
2038
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
2039
return PlainKnitContent(lines, record.key)
2041
# local keys we can ask for directly
2042
return self._get_one_work(key)
2044
def get_record_stream(self):
2045
"""Get a record stream for the keys requested during __init__."""
2046
for record in self._work():
2050
"""Produce maps of text and KnitContents as dicts.
2052
:return: (text_map, content_map) where text_map contains the texts for
2053
the requested versions and content_map contains the KnitContents.
2055
# NB: By definition we never need to read remote sources unless texts
2056
# are requested from them: we don't delta across stores - and we
2057
# explicitly do not want to to prevent data loss situations.
2058
if self.global_map is None:
2059
self.global_map = self.vf.get_parent_map(self.keys)
2060
nonlocal_keys = self.nonlocal_keys
2062
missing_keys = set(nonlocal_keys)
2063
# Read from remote versioned file instances and provide to our caller.
2064
for source in self.vf._fallback_vfs:
2065
if not missing_keys:
2067
# Loop over fallback repositories asking them for texts - ignore
2068
# any missing from a particular fallback.
2069
for record in source.get_record_stream(missing_keys,
2070
self._ordering, True):
2071
if record.storage_kind == 'absent':
2072
# Not in thie particular stream, may be in one of the
2073
# other fallback vfs objects.
2075
missing_keys.remove(record.key)
2078
if self._raw_record_map is None:
2079
raise AssertionError('_raw_record_map should have been filled')
2081
for key in self.keys:
2082
if key in self.nonlocal_keys:
2084
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2087
def _get_one_work(self, requested_key):
2088
# Now, if we have calculated everything already, just return the
2090
if requested_key in self._contents_map:
2091
return self._contents_map[requested_key]
2092
# To simplify things, parse everything at once - code that wants one text
2093
# probably wants them all.
2094
# FUTURE: This function could be improved for the 'extract many' case
2095
# by tracking each component and only doing the copy when the number of
2096
# children than need to apply delta's to it is > 1 or it is part of the
2098
multiple_versions = len(self.keys) != 1
2099
if self._record_map is None:
2100
self._record_map = self.vf._raw_map_to_record_map(
2101
self._raw_record_map)
2102
record_map = self._record_map
2103
# raw_record_map is key:
2104
# Have read and parsed records at this point.
2105
for key in self.keys:
2106
if key in self.nonlocal_keys:
2111
while cursor is not None:
2113
record, record_details, digest, next = record_map[cursor]
2115
raise RevisionNotPresent(cursor, self)
2116
components.append((cursor, record, record_details, digest))
2118
if cursor in self._contents_map:
2119
# no need to plan further back
2120
components.append((cursor, None, None, None))
2124
for (component_id, record, record_details,
2125
digest) in reversed(components):
2126
if component_id in self._contents_map:
2127
content = self._contents_map[component_id]
2129
content, delta = self._factory.parse_record(key[-1],
2130
record, record_details, content,
2131
copy_base_content=multiple_versions)
2132
if multiple_versions:
2133
self._contents_map[component_id] = content
2135
# digest here is the digest from the last applied component.
2136
text = content.text()
2137
actual_sha = sha_strings(text)
2138
if actual_sha != digest:
2139
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2140
if multiple_versions:
2141
return self._contents_map[requested_key]
2145
def _wire_bytes(self):
2146
"""Get the bytes to put on the wire for 'key'.
2148
The first collection of bytes asked for returns the serialised
2149
raw_record_map and the additional details (key, parent) for key.
2150
Subsequent calls return just the additional details (key, parent).
2151
The wire storage_kind given for the first key is 'knit-delta-closure',
2152
For subsequent keys it is 'knit-delta-closure-ref'.
2154
:param key: A key from the content generator.
2155
:return: Bytes to put on the wire.
2158
# kind marker for dispatch on the far side,
2159
lines.append('knit-delta-closure')
2161
if self.vf._factory.annotated:
2162
lines.append('annotated')
2165
# then the list of keys
2166
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2167
if key not in self.nonlocal_keys]))
2168
# then the _raw_record_map in serialised form:
2170
# for each item in the map:
2172
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2173
# one line with method
2174
# one line with noeol
2175
# one line with next ('' for None)
2176
# one line with byte count of the record bytes
2178
for key, (record_bytes, (method, noeol), next) in \
2179
self._raw_record_map.iteritems():
2180
key_bytes = '\x00'.join(key)
2181
parents = self.global_map.get(key, None)
2183
parent_bytes = 'None:'
2185
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2186
method_bytes = method
2192
next_bytes = '\x00'.join(next)
2195
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2196
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2197
len(record_bytes), record_bytes))
2198
map_bytes = ''.join(map_byte_list)
2199
lines.append(map_bytes)
2200
bytes = '\n'.join(lines)
2204
class _VFContentMapGenerator(_ContentMapGenerator):
2205
"""Content map generator reading from a VersionedFiles object."""
2207
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2208
global_map=None, raw_record_map=None, ordering='unordered'):
2209
"""Create a _ContentMapGenerator.
2211
:param versioned_files: The versioned files that the texts are being
2213
:param keys: The keys to produce content maps for.
2214
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2215
which are known to not be in this knit, but rather in one of the
2217
:param global_map: The result of get_parent_map(keys) (or a supermap).
2218
This is required if get_record_stream() is to be used.
2219
:param raw_record_map: A unparsed raw record map to use for answering
2222
_ContentMapGenerator.__init__(self, ordering=ordering)
2223
# The vf to source data from
2224
self.vf = versioned_files
2226
self.keys = list(keys)
2227
# Keys known to be in fallback vfs objects
2228
if nonlocal_keys is None:
2229
self.nonlocal_keys = set()
2231
self.nonlocal_keys = frozenset(nonlocal_keys)
2232
# Parents data for keys to be returned in get_record_stream
2233
self.global_map = global_map
2234
# The chunked lists for self.keys in text form
2236
# A cache of KnitContent objects used in extracting texts.
2237
self._contents_map = {}
2238
# All the knit records needed to assemble the requested keys as full
2240
self._record_map = None
2241
if raw_record_map is None:
2242
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2245
self._raw_record_map = raw_record_map
2246
# the factory for parsing records
2247
self._factory = self.vf._factory
2250
class _NetworkContentMapGenerator(_ContentMapGenerator):
2251
"""Content map generator sourced from a network stream."""
2253
def __init__(self, bytes, line_end):
2254
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2256
self.global_map = {}
2257
self._raw_record_map = {}
2258
self._contents_map = {}
2259
self._record_map = None
2260
self.nonlocal_keys = []
2261
# Get access to record parsing facilities
2262
self.vf = KnitVersionedFiles(None, None)
2265
line_end = bytes.find('\n', start)
2266
line = bytes[start:line_end]
2267
start = line_end + 1
2268
if line == 'annotated':
2269
self._factory = KnitAnnotateFactory()
2271
self._factory = KnitPlainFactory()
2272
# list of keys to emit in get_record_stream
2273
line_end = bytes.find('\n', start)
2274
line = bytes[start:line_end]
2275
start = line_end + 1
2277
tuple(segment.split('\x00')) for segment in line.split('\t')
2279
# now a loop until the end. XXX: It would be nice if this was just a
2280
# bunch of the same records as get_record_stream(..., False) gives, but
2281
# there is a decent sized gap stopping that at the moment.
2285
line_end = bytes.find('\n', start)
2286
key = tuple(bytes[start:line_end].split('\x00'))
2287
start = line_end + 1
2288
# 1 line with parents (None: for None, '' for ())
2289
line_end = bytes.find('\n', start)
2290
line = bytes[start:line_end]
2295
[tuple(segment.split('\x00')) for segment in line.split('\t')
2297
self.global_map[key] = parents
2298
start = line_end + 1
2299
# one line with method
2300
line_end = bytes.find('\n', start)
2301
line = bytes[start:line_end]
2303
start = line_end + 1
2304
# one line with noeol
2305
line_end = bytes.find('\n', start)
2306
line = bytes[start:line_end]
2308
start = line_end + 1
2309
# one line with next ('' for None)
2310
line_end = bytes.find('\n', start)
2311
line = bytes[start:line_end]
2315
next = tuple(bytes[start:line_end].split('\x00'))
2316
start = line_end + 1
2317
# one line with byte count of the record bytes
2318
line_end = bytes.find('\n', start)
2319
line = bytes[start:line_end]
2321
start = line_end + 1
2323
record_bytes = bytes[start:start+count]
2324
start = start + count
2326
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2328
def get_record_stream(self):
2329
"""Get a record stream for for keys requested by the bytestream."""
2331
for key in self.keys:
2332
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2335
def _wire_bytes(self):
2339
class _KndxIndex(object):
2340
"""Manages knit index files
2342
The index is kept in memory and read on startup, to enable
2343
fast lookups of revision information. The cursor of the index
2344
file is always pointing to the end, making it easy to append
2347
_cache is a cache for fast mapping from version id to a Index
2350
_history is a cache for fast mapping from indexes to version ids.
2352
The index data format is dictionary compressed when it comes to
2353
parent references; a index entry may only have parents that with a
2354
lover index number. As a result, the index is topological sorted.
2356
Duplicate entries may be written to the index for a single version id
2357
if this is done then the latter one completely replaces the former:
2358
this allows updates to correct version and parent information.
2359
Note that the two entries may share the delta, and that successive
2360
annotations and references MUST point to the first entry.
2362
The index file on disc contains a header, followed by one line per knit
2363
record. The same revision can be present in an index file more than once.
2364
The first occurrence gets assigned a sequence number starting from 0.
2366
The format of a single line is
2367
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
2368
REVISION_ID is a utf8-encoded revision id
2369
FLAGS is a comma separated list of flags about the record. Values include
2370
no-eol, line-delta, fulltext.
2371
BYTE_OFFSET is the ascii representation of the byte offset in the data file
2372
that the compressed data starts at.
2373
LENGTH is the ascii representation of the length of the data file.
2374
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
2376
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
2377
revision id already in the knit that is a parent of REVISION_ID.
2378
The ' :' marker is the end of record marker.
2381
when a write is interrupted to the index file, it will result in a line
2382
that does not end in ' :'. If the ' :' is not present at the end of a line,
2383
or at the end of the file, then the record that is missing it will be
2384
ignored by the parser.
2386
When writing new records to the index file, the data is preceded by '\n'
2387
to ensure that records always start on new lines even if the last write was
2388
interrupted. As a result its normal for the last line in the index to be
2389
missing a trailing newline. One can be added with no harmful effects.
2391
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2392
where prefix is e.g. the (fileid,) for .texts instances or () for
2393
constant-mapped things like .revisions, and the old state is
2394
tuple(history_vector, cache_dict). This is used to prevent having an
2395
ABI change with the C extension that reads .kndx files.
2398
HEADER = "# bzr knit index 8\n"
2400
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2401
"""Create a _KndxIndex on transport using mapper."""
2402
self._transport = transport
2403
self._mapper = mapper
2404
self._get_scope = get_scope
2405
self._allow_writes = allow_writes
2406
self._is_locked = is_locked
2408
self.has_graph = True
2410
def add_records(self, records, random_id=False, missing_compression_parents=False):
2411
"""Add multiple records to the index.
2413
:param records: a list of tuples:
2414
(key, options, access_memo, parents).
2415
:param random_id: If True the ids being added were randomly generated
2416
and no check for existence will be performed.
2417
:param missing_compression_parents: If True the records being added are
2418
only compressed against texts already in the index (or inside
2419
records). If False the records all refer to unavailable texts (or
2420
texts inside records) as compression parents.
2422
if missing_compression_parents:
2423
# It might be nice to get the edge of the records. But keys isn't
2425
keys = sorted(record[0] for record in records)
2426
raise errors.RevisionNotPresent(keys, self)
2428
for record in records:
2431
path = self._mapper.map(key) + '.kndx'
2432
path_keys = paths.setdefault(path, (prefix, []))
2433
path_keys[1].append(record)
2434
for path in sorted(paths):
2435
prefix, path_keys = paths[path]
2436
self._load_prefixes([prefix])
2438
orig_history = self._kndx_cache[prefix][1][:]
2439
orig_cache = self._kndx_cache[prefix][0].copy()
2442
for key, options, (_, pos, size), parents in path_keys:
2444
# kndx indices cannot be parentless.
2446
line = "\n%s %s %s %s %s :" % (
2447
key[-1], ','.join(options), pos, size,
2448
self._dictionary_compress(parents))
2449
if type(line) is not str:
2450
raise AssertionError(
2451
'data must be utf8 was %s' % type(line))
2453
self._cache_key(key, options, pos, size, parents)
2454
if len(orig_history):
2455
self._transport.append_bytes(path, ''.join(lines))
2457
self._init_index(path, lines)
2459
# If any problems happen, restore the original values and re-raise
2460
self._kndx_cache[prefix] = (orig_cache, orig_history)
2463
def scan_unvalidated_index(self, graph_index):
2464
"""See _KnitGraphIndex.scan_unvalidated_index."""
2465
# Because kndx files do not support atomic insertion via separate index
2466
# files, they do not support this method.
2467
raise NotImplementedError(self.scan_unvalidated_index)
2469
def get_missing_compression_parents(self):
2470
"""See _KnitGraphIndex.get_missing_compression_parents."""
2471
# Because kndx files do not support atomic insertion via separate index
2472
# files, they do not support this method.
2473
raise NotImplementedError(self.get_missing_compression_parents)
2475
def _cache_key(self, key, options, pos, size, parent_keys):
2476
"""Cache a version record in the history array and index cache.
2478
This is inlined into _load_data for performance. KEEP IN SYNC.
2479
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2483
version_id = key[-1]
2484
# last-element only for compatibilty with the C load_data.
2485
parents = tuple(parent[-1] for parent in parent_keys)
2486
for parent in parent_keys:
2487
if parent[:-1] != prefix:
2488
raise ValueError("mismatched prefixes for %r, %r" % (
2490
cache, history = self._kndx_cache[prefix]
2491
# only want the _history index to reference the 1st index entry
2493
if version_id not in cache:
2494
index = len(history)
2495
history.append(version_id)
2497
index = cache[version_id][5]
2498
cache[version_id] = (version_id,
2505
def check_header(self, fp):
2506
line = fp.readline()
2508
# An empty file can actually be treated as though the file doesn't
2510
raise errors.NoSuchFile(self)
2511
if line != self.HEADER:
2512
raise KnitHeaderError(badline=line, filename=self)
2514
def _check_read(self):
2515
if not self._is_locked():
2516
raise errors.ObjectNotLocked(self)
2517
if self._get_scope() != self._scope:
2520
def _check_write_ok(self):
2521
"""Assert if not writes are permitted."""
2522
if not self._is_locked():
2523
raise errors.ObjectNotLocked(self)
2524
if self._get_scope() != self._scope:
2526
if self._mode != 'w':
2527
raise errors.ReadOnlyObjectDirtiedError(self)
2529
def get_build_details(self, keys):
2530
"""Get the method, index_memo and compression parent for keys.
2532
Ghosts are omitted from the result.
2534
:param keys: An iterable of keys.
2535
:return: A dict of key:(index_memo, compression_parent, parents,
2538
opaque structure to pass to read_records to extract the raw
2541
Content that this record is built upon, may be None
2543
Logical parents of this node
2545
extra information about the content which needs to be passed to
2546
Factory.parse_record
2548
parent_map = self.get_parent_map(keys)
2551
if key not in parent_map:
2553
method = self.get_method(key)
2554
parents = parent_map[key]
2555
if method == 'fulltext':
2556
compression_parent = None
2558
compression_parent = parents[0]
2559
noeol = 'no-eol' in self.get_options(key)
2560
index_memo = self.get_position(key)
2561
result[key] = (index_memo, compression_parent,
2562
parents, (method, noeol))
2565
def get_method(self, key):
2566
"""Return compression method of specified key."""
2567
options = self.get_options(key)
2568
if 'fulltext' in options:
2570
elif 'line-delta' in options:
2573
raise errors.KnitIndexUnknownMethod(self, options)
2575
def get_options(self, key):
2576
"""Return a list representing options.
2580
prefix, suffix = self._split_key(key)
2581
self._load_prefixes([prefix])
2583
return self._kndx_cache[prefix][0][suffix][1]
2585
raise RevisionNotPresent(key, self)
2587
def find_ancestry(self, keys):
2588
"""See CombinedGraphIndex.find_ancestry()"""
2589
prefixes = set(key[:-1] for key in keys)
2590
self._load_prefixes(prefixes)
2593
missing_keys = set()
2594
pending_keys = list(keys)
2595
# This assumes that keys will not reference parents in a different
2596
# prefix, which is accurate so far.
2598
key = pending_keys.pop()
2599
if key in parent_map:
2603
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2605
missing_keys.add(key)
2607
parent_keys = tuple([prefix + (suffix,)
2608
for suffix in suffix_parents])
2609
parent_map[key] = parent_keys
2610
pending_keys.extend([p for p in parent_keys
2611
if p not in parent_map])
2612
return parent_map, missing_keys
2614
def get_parent_map(self, keys):
2615
"""Get a map of the parents of keys.
2617
:param keys: The keys to look up parents for.
2618
:return: A mapping from keys to parents. Absent keys are absent from
2621
# Parse what we need to up front, this potentially trades off I/O
2622
# locality (.kndx and .knit in the same block group for the same file
2623
# id) for less checking in inner loops.
2624
prefixes = set(key[:-1] for key in keys)
2625
self._load_prefixes(prefixes)
2630
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2634
result[key] = tuple(prefix + (suffix,) for
2635
suffix in suffix_parents)
2638
def get_position(self, key):
2639
"""Return details needed to access the version.
2641
:return: a tuple (key, data position, size) to hand to the access
2642
logic to get the record.
2644
prefix, suffix = self._split_key(key)
2645
self._load_prefixes([prefix])
2646
entry = self._kndx_cache[prefix][0][suffix]
2647
return key, entry[2], entry[3]
2649
has_key = _mod_index._has_key_from_parent_map
2651
def _init_index(self, path, extra_lines=[]):
2652
"""Initialize an index."""
2654
sio.write(self.HEADER)
2655
sio.writelines(extra_lines)
2657
self._transport.put_file_non_atomic(path, sio,
2658
create_parent_dir=True)
2659
# self._create_parent_dir)
2660
# mode=self._file_mode,
2661
# dir_mode=self._dir_mode)
2664
"""Get all the keys in the collection.
2666
The keys are not ordered.
2669
# Identify all key prefixes.
2670
# XXX: A bit hacky, needs polish.
2671
if type(self._mapper) is ConstantMapper:
2675
for quoted_relpath in self._transport.iter_files_recursive():
2676
path, ext = os.path.splitext(quoted_relpath)
2678
prefixes = [self._mapper.unmap(path) for path in relpaths]
2679
self._load_prefixes(prefixes)
2680
for prefix in prefixes:
2681
for suffix in self._kndx_cache[prefix][1]:
2682
result.add(prefix + (suffix,))
2685
def _load_prefixes(self, prefixes):
2686
"""Load the indices for prefixes."""
2688
for prefix in prefixes:
2689
if prefix not in self._kndx_cache:
2690
# the load_data interface writes to these variables.
2693
self._filename = prefix
2695
path = self._mapper.map(prefix) + '.kndx'
2696
fp = self._transport.get(path)
2698
# _load_data may raise NoSuchFile if the target knit is
2700
_load_data(self, fp)
2703
self._kndx_cache[prefix] = (self._cache, self._history)
2708
self._kndx_cache[prefix] = ({}, [])
2709
if type(self._mapper) is ConstantMapper:
2710
# preserve behaviour for revisions.kndx etc.
2711
self._init_index(path)
2716
missing_keys = _mod_index._missing_keys_from_parent_map
2718
def _partition_keys(self, keys):
2719
"""Turn keys into a dict of prefix:suffix_list."""
2722
prefix_keys = result.setdefault(key[:-1], [])
2723
prefix_keys.append(key[-1])
2726
def _dictionary_compress(self, keys):
2727
"""Dictionary compress keys.
2729
:param keys: The keys to generate references to.
2730
:return: A string representation of keys. keys which are present are
2731
dictionary compressed, and others are emitted as fulltext with a
2737
prefix = keys[0][:-1]
2738
cache = self._kndx_cache[prefix][0]
2740
if key[:-1] != prefix:
2741
# kndx indices cannot refer across partitioned storage.
2742
raise ValueError("mismatched prefixes for %r" % keys)
2743
if key[-1] in cache:
2744
# -- inlined lookup() --
2745
result_list.append(str(cache[key[-1]][5]))
2746
# -- end lookup () --
2748
result_list.append('.' + key[-1])
2749
return ' '.join(result_list)
2751
def _reset_cache(self):
2752
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2753
# (cache_dict, history_vector) for parsed kndx files.
2754
self._kndx_cache = {}
2755
self._scope = self._get_scope()
2756
allow_writes = self._allow_writes()
2762
def _sort_keys_by_io(self, keys, positions):
2763
"""Figure out an optimal order to read the records for the given keys.
2765
Sort keys, grouped by index and sorted by position.
2767
:param keys: A list of keys whose records we want to read. This will be
2769
:param positions: A dict, such as the one returned by
2770
_get_components_positions()
2773
def get_sort_key(key):
2774
index_memo = positions[key][1]
2775
# Group by prefix and position. index_memo[0] is the key, so it is
2776
# (file_id, revision_id) and we don't want to sort on revision_id,
2777
# index_memo[1] is the position, and index_memo[2] is the size,
2778
# which doesn't matter for the sort
2779
return index_memo[0][:-1], index_memo[1]
2780
return keys.sort(key=get_sort_key)
2782
_get_total_build_size = _get_total_build_size
2784
def _split_key(self, key):
2785
"""Split key into a prefix and suffix."""
2786
return key[:-1], key[-1]
2789
class _KeyRefs(object):
2791
def __init__(self, track_new_keys=False):
2792
# dict mapping 'key' to 'set of keys referring to that key'
2795
# set remembering all new keys
2796
self.new_keys = set()
2798
self.new_keys = None
2804
self.new_keys.clear()
2806
def add_references(self, key, refs):
2807
# Record the new references
2808
for referenced in refs:
2810
needed_by = self.refs[referenced]
2812
needed_by = self.refs[referenced] = set()
2814
# Discard references satisfied by the new key
2817
def get_new_keys(self):
2818
return self.new_keys
2820
def get_unsatisfied_refs(self):
2821
return self.refs.iterkeys()
2823
def _satisfy_refs_for_key(self, key):
2827
# No keys depended on this key. That's ok.
2830
def add_key(self, key):
2831
# satisfy refs for key, and remember that we've seen this key.
2832
self._satisfy_refs_for_key(key)
2833
if self.new_keys is not None:
2834
self.new_keys.add(key)
2836
def satisfy_refs_for_keys(self, keys):
2838
self._satisfy_refs_for_key(key)
2840
def get_referrers(self):
2842
for referrers in self.refs.itervalues():
2843
result.update(referrers)
2847
class _KnitGraphIndex(object):
2848
"""A KnitVersionedFiles index layered on GraphIndex."""
2850
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2851
add_callback=None, track_external_parent_refs=False):
2852
"""Construct a KnitGraphIndex on a graph_index.
2854
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2855
:param is_locked: A callback to check whether the object should answer
2857
:param deltas: Allow delta-compressed records.
2858
:param parents: If True, record knits parents, if not do not record
2860
:param add_callback: If not None, allow additions to the index and call
2861
this callback with a list of added GraphIndex nodes:
2862
[(node, value, node_refs), ...]
2863
:param is_locked: A callback, returns True if the index is locked and
2865
:param track_external_parent_refs: If True, record all external parent
2866
references parents from added records. These can be retrieved
2867
later by calling get_missing_parents().
2869
self._add_callback = add_callback
2870
self._graph_index = graph_index
2871
self._deltas = deltas
2872
self._parents = parents
2873
if deltas and not parents:
2874
# XXX: TODO: Delta tree and parent graph should be conceptually
2876
raise KnitCorrupt(self, "Cannot do delta compression without "
2878
self.has_graph = parents
2879
self._is_locked = is_locked
2880
self._missing_compression_parents = set()
2881
if track_external_parent_refs:
2882
self._key_dependencies = _KeyRefs()
2884
self._key_dependencies = None
2887
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2889
def add_records(self, records, random_id=False,
2890
missing_compression_parents=False):
2891
"""Add multiple records to the index.
2893
This function does not insert data into the Immutable GraphIndex
2894
backing the KnitGraphIndex, instead it prepares data for insertion by
2895
the caller and checks that it is safe to insert then calls
2896
self._add_callback with the prepared GraphIndex nodes.
2898
:param records: a list of tuples:
2899
(key, options, access_memo, parents).
2900
:param random_id: If True the ids being added were randomly generated
2901
and no check for existence will be performed.
2902
:param missing_compression_parents: If True the records being added are
2903
only compressed against texts already in the index (or inside
2904
records). If False the records all refer to unavailable texts (or
2905
texts inside records) as compression parents.
2907
if not self._add_callback:
2908
raise errors.ReadOnlyError(self)
2909
# we hope there are no repositories with inconsistent parentage
2913
compression_parents = set()
2914
key_dependencies = self._key_dependencies
2915
for (key, options, access_memo, parents) in records:
2917
parents = tuple(parents)
2918
if key_dependencies is not None:
2919
key_dependencies.add_references(key, parents)
2920
index, pos, size = access_memo
2921
if 'no-eol' in options:
2925
value += "%d %d" % (pos, size)
2926
if not self._deltas:
2927
if 'line-delta' in options:
2928
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2931
if 'line-delta' in options:
2932
node_refs = (parents, (parents[0],))
2933
if missing_compression_parents:
2934
compression_parents.add(parents[0])
2936
node_refs = (parents, ())
2938
node_refs = (parents, )
2941
raise KnitCorrupt(self, "attempt to add node with parents "
2942
"in parentless index.")
2944
keys[key] = (value, node_refs)
2947
present_nodes = self._get_entries(keys)
2948
for (index, key, value, node_refs) in present_nodes:
2949
parents = node_refs[:1]
2950
# Sometimes these are passed as a list rather than a tuple
2951
passed = static_tuple.as_tuples(keys[key])
2952
passed_parents = passed[1][:1]
2953
if (value[0] != keys[key][0][0] or
2954
parents != passed_parents):
2955
node_refs = static_tuple.as_tuples(node_refs)
2956
raise KnitCorrupt(self, "inconsistent details in add_records"
2957
": %s %s" % ((value, node_refs), passed))
2961
for key, (value, node_refs) in keys.iteritems():
2962
result.append((key, value, node_refs))
2964
for key, (value, node_refs) in keys.iteritems():
2965
result.append((key, value))
2966
self._add_callback(result)
2967
if missing_compression_parents:
2968
# This may appear to be incorrect (it does not check for
2969
# compression parents that are in the existing graph index),
2970
# but such records won't have been buffered, so this is
2971
# actually correct: every entry when
2972
# missing_compression_parents==True either has a missing parent, or
2973
# a parent that is one of the keys in records.
2974
compression_parents.difference_update(keys)
2975
self._missing_compression_parents.update(compression_parents)
2976
# Adding records may have satisfied missing compression parents.
2977
self._missing_compression_parents.difference_update(keys)
2979
def scan_unvalidated_index(self, graph_index):
2980
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2982
This allows this _KnitGraphIndex to keep track of any missing
2983
compression parents we may want to have filled in to make those
2986
:param graph_index: A GraphIndex
2989
new_missing = graph_index.external_references(ref_list_num=1)
2990
new_missing.difference_update(self.get_parent_map(new_missing))
2991
self._missing_compression_parents.update(new_missing)
2992
if self._key_dependencies is not None:
2993
# Add parent refs from graph_index (and discard parent refs that
2994
# the graph_index has).
2995
for node in graph_index.iter_all_entries():
2996
self._key_dependencies.add_references(node[1], node[3][0])
2998
def get_missing_compression_parents(self):
2999
"""Return the keys of missing compression parents.
3001
Missing compression parents occur when a record stream was missing
3002
basis texts, or a index was scanned that had missing basis texts.
3004
return frozenset(self._missing_compression_parents)
3006
def get_missing_parents(self):
3007
"""Return the keys of missing parents."""
3008
# If updating this, you should also update
3009
# groupcompress._GCGraphIndex.get_missing_parents
3010
# We may have false positives, so filter those out.
3011
self._key_dependencies.satisfy_refs_for_keys(
3012
self.get_parent_map(self._key_dependencies.get_unsatisfied_refs()))
3013
return frozenset(self._key_dependencies.get_unsatisfied_refs())
3015
def _check_read(self):
3016
"""raise if reads are not permitted."""
3017
if not self._is_locked():
3018
raise errors.ObjectNotLocked(self)
3020
def _check_write_ok(self):
3021
"""Assert if writes are not permitted."""
3022
if not self._is_locked():
3023
raise errors.ObjectNotLocked(self)
3025
def _compression_parent(self, an_entry):
3026
# return the key that an_entry is compressed against, or None
3027
# Grab the second parent list (as deltas implies parents currently)
3028
compression_parents = an_entry[3][1]
3029
if not compression_parents:
3031
if len(compression_parents) != 1:
3032
raise AssertionError(
3033
"Too many compression parents: %r" % compression_parents)
3034
return compression_parents[0]
3036
def get_build_details(self, keys):
3037
"""Get the method, index_memo and compression parent for version_ids.
3039
Ghosts are omitted from the result.
3041
:param keys: An iterable of keys.
3042
:return: A dict of key:
3043
(index_memo, compression_parent, parents, record_details).
3045
opaque structure to pass to read_records to extract the raw
3048
Content that this record is built upon, may be None
3050
Logical parents of this node
3052
extra information about the content which needs to be passed to
3053
Factory.parse_record
3057
entries = self._get_entries(keys, False)
3058
for entry in entries:
3060
if not self._parents:
3063
parents = entry[3][0]
3064
if not self._deltas:
3065
compression_parent_key = None
3067
compression_parent_key = self._compression_parent(entry)
3068
noeol = (entry[2][0] == 'N')
3069
if compression_parent_key:
3070
method = 'line-delta'
3073
result[key] = (self._node_to_position(entry),
3074
compression_parent_key, parents,
3078
def _get_entries(self, keys, check_present=False):
3079
"""Get the entries for keys.
3081
:param keys: An iterable of index key tuples.
3086
for node in self._graph_index.iter_entries(keys):
3088
found_keys.add(node[1])
3090
# adapt parentless index to the rest of the code.
3091
for node in self._graph_index.iter_entries(keys):
3092
yield node[0], node[1], node[2], ()
3093
found_keys.add(node[1])
3095
missing_keys = keys.difference(found_keys)
3097
raise RevisionNotPresent(missing_keys.pop(), self)
3099
def get_method(self, key):
3100
"""Return compression method of specified key."""
3101
return self._get_method(self._get_node(key))
3103
def _get_method(self, node):
3104
if not self._deltas:
3106
if self._compression_parent(node):
3111
def _get_node(self, key):
3113
return list(self._get_entries([key]))[0]
3115
raise RevisionNotPresent(key, self)
3117
def get_options(self, key):
3118
"""Return a list representing options.
3122
node = self._get_node(key)
3123
options = [self._get_method(node)]
3124
if node[2][0] == 'N':
3125
options.append('no-eol')
3128
def find_ancestry(self, keys):
3129
"""See CombinedGraphIndex.find_ancestry()"""
3130
return self._graph_index.find_ancestry(keys, 0)
3132
def get_parent_map(self, keys):
3133
"""Get a map of the parents of keys.
3135
:param keys: The keys to look up parents for.
3136
:return: A mapping from keys to parents. Absent keys are absent from
3140
nodes = self._get_entries(keys)
3144
result[node[1]] = node[3][0]
3147
result[node[1]] = None
3150
def get_position(self, key):
3151
"""Return details needed to access the version.
3153
:return: a tuple (index, data position, size) to hand to the access
3154
logic to get the record.
3156
node = self._get_node(key)
3157
return self._node_to_position(node)
3159
has_key = _mod_index._has_key_from_parent_map
3162
"""Get all the keys in the collection.
3164
The keys are not ordered.
3167
return [node[1] for node in self._graph_index.iter_all_entries()]
3169
missing_keys = _mod_index._missing_keys_from_parent_map
3171
def _node_to_position(self, node):
3172
"""Convert an index value to position details."""
3173
bits = node[2][1:].split(' ')
3174
return node[0], int(bits[0]), int(bits[1])
3176
def _sort_keys_by_io(self, keys, positions):
3177
"""Figure out an optimal order to read the records for the given keys.
3179
Sort keys, grouped by index and sorted by position.
3181
:param keys: A list of keys whose records we want to read. This will be
3183
:param positions: A dict, such as the one returned by
3184
_get_components_positions()
3187
def get_index_memo(key):
3188
# index_memo is at offset [1]. It is made up of (GraphIndex,
3189
# position, size). GI is an object, which will be unique for each
3190
# pack file. This causes us to group by pack file, then sort by
3191
# position. Size doesn't matter, but it isn't worth breaking up the
3193
return positions[key][1]
3194
return keys.sort(key=get_index_memo)
3196
_get_total_build_size = _get_total_build_size
3199
class _KnitKeyAccess(object):
3200
"""Access to records in .knit files."""
3202
def __init__(self, transport, mapper):
3203
"""Create a _KnitKeyAccess with transport and mapper.
3205
:param transport: The transport the access object is rooted at.
3206
:param mapper: The mapper used to map keys to .knit files.
3208
self._transport = transport
3209
self._mapper = mapper
3211
def add_raw_records(self, key_sizes, raw_data):
3212
"""Add raw knit bytes to a storage area.
3214
The data is spooled to the container writer in one bytes-record per
3217
:param sizes: An iterable of tuples containing the key and size of each
3219
:param raw_data: A bytestring containing the data.
3220
:return: A list of memos to retrieve the record later. Each memo is an
3221
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3222
length), where the key is the record key.
3224
if type(raw_data) is not str:
3225
raise AssertionError(
3226
'data must be plain bytes was %s' % type(raw_data))
3229
# TODO: This can be tuned for writing to sftp and other servers where
3230
# append() is relatively expensive by grouping the writes to each key
3232
for key, size in key_sizes:
3233
path = self._mapper.map(key)
3235
base = self._transport.append_bytes(path + '.knit',
3236
raw_data[offset:offset+size])
3237
except errors.NoSuchFile:
3238
self._transport.mkdir(osutils.dirname(path))
3239
base = self._transport.append_bytes(path + '.knit',
3240
raw_data[offset:offset+size])
3244
result.append((key, base, size))
3248
"""Flush pending writes on this access object.
3250
For .knit files this is a no-op.
3254
def get_raw_records(self, memos_for_retrieval):
3255
"""Get the raw bytes for a records.
3257
:param memos_for_retrieval: An iterable containing the access memo for
3258
retrieving the bytes.
3259
:return: An iterator over the bytes of the records.
3261
# first pass, group into same-index request to minimise readv's issued.
3263
current_prefix = None
3264
for (key, offset, length) in memos_for_retrieval:
3265
if current_prefix == key[:-1]:
3266
current_list.append((offset, length))
3268
if current_prefix is not None:
3269
request_lists.append((current_prefix, current_list))
3270
current_prefix = key[:-1]
3271
current_list = [(offset, length)]
3272
# handle the last entry
3273
if current_prefix is not None:
3274
request_lists.append((current_prefix, current_list))
3275
for prefix, read_vector in request_lists:
3276
path = self._mapper.map(prefix) + '.knit'
3277
for pos, data in self._transport.readv(path, read_vector):
3281
class _DirectPackAccess(object):
3282
"""Access to data in one or more packs with less translation."""
3284
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
3285
"""Create a _DirectPackAccess object.
3287
:param index_to_packs: A dict mapping index objects to the transport
3288
and file names for obtaining data.
3289
:param reload_func: A function to call if we determine that the pack
3290
files have moved and we need to reload our caches. See
3291
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
3293
self._container_writer = None
3294
self._write_index = None
3295
self._indices = index_to_packs
3296
self._reload_func = reload_func
3297
self._flush_func = flush_func
3299
def add_raw_records(self, key_sizes, raw_data):
3300
"""Add raw knit bytes to a storage area.
3302
The data is spooled to the container writer in one bytes-record per
3305
:param sizes: An iterable of tuples containing the key and size of each
3307
:param raw_data: A bytestring containing the data.
3308
:return: A list of memos to retrieve the record later. Each memo is an
3309
opaque index memo. For _DirectPackAccess the memo is (index, pos,
3310
length), where the index field is the write_index object supplied
3311
to the PackAccess object.
3313
if type(raw_data) is not str:
3314
raise AssertionError(
3315
'data must be plain bytes was %s' % type(raw_data))
3318
for key, size in key_sizes:
3319
p_offset, p_length = self._container_writer.add_bytes_record(
3320
raw_data[offset:offset+size], [])
3322
result.append((self._write_index, p_offset, p_length))
3326
"""Flush pending writes on this access object.
3328
This will flush any buffered writes to a NewPack.
3330
if self._flush_func is not None:
3333
def get_raw_records(self, memos_for_retrieval):
3334
"""Get the raw bytes for a records.
3336
:param memos_for_retrieval: An iterable containing the (index, pos,
3337
length) memo for retrieving the bytes. The Pack access method
3338
looks up the pack to use for a given record in its index_to_pack
3340
:return: An iterator over the bytes of the records.
3342
# first pass, group into same-index requests
3344
current_index = None
3345
for (index, offset, length) in memos_for_retrieval:
3346
if current_index == index:
3347
current_list.append((offset, length))
3349
if current_index is not None:
3350
request_lists.append((current_index, current_list))
3351
current_index = index
3352
current_list = [(offset, length)]
3353
# handle the last entry
3354
if current_index is not None:
3355
request_lists.append((current_index, current_list))
3356
for index, offsets in request_lists:
3358
transport, path = self._indices[index]
3360
# A KeyError here indicates that someone has triggered an index
3361
# reload, and this index has gone missing, we need to start
3363
if self._reload_func is None:
3364
# If we don't have a _reload_func there is nothing that can
3367
raise errors.RetryWithNewPacks(index,
3368
reload_occurred=True,
3369
exc_info=sys.exc_info())
3371
reader = pack.make_readv_reader(transport, path, offsets)
3372
for names, read_func in reader.iter_records():
3373
yield read_func(None)
3374
except errors.NoSuchFile:
3375
# A NoSuchFile error indicates that a pack file has gone
3376
# missing on disk, we need to trigger a reload, and start over.
3377
if self._reload_func is None:
3379
raise errors.RetryWithNewPacks(transport.abspath(path),
3380
reload_occurred=False,
3381
exc_info=sys.exc_info())
3383
def set_writer(self, writer, index, transport_packname):
3384
"""Set a writer to use for adding data."""
3385
if index is not None:
3386
self._indices[index] = transport_packname
3387
self._container_writer = writer
3388
self._write_index = index
3390
def reload_or_raise(self, retry_exc):
3391
"""Try calling the reload function, or re-raise the original exception.
3393
This should be called after _DirectPackAccess raises a
3394
RetryWithNewPacks exception. This function will handle the common logic
3395
of determining when the error is fatal versus being temporary.
3396
It will also make sure that the original exception is raised, rather
3397
than the RetryWithNewPacks exception.
3399
If this function returns, then the calling function should retry
3400
whatever operation was being performed. Otherwise an exception will
3403
:param retry_exc: A RetryWithNewPacks exception.
3406
if self._reload_func is None:
3408
elif not self._reload_func():
3409
# The reload claimed that nothing changed
3410
if not retry_exc.reload_occurred:
3411
# If there wasn't an earlier reload, then we really were
3412
# expecting to find changes. We didn't find them, so this is a
3416
exc_class, exc_value, exc_traceback = retry_exc.exc_info
3417
raise exc_class, exc_value, exc_traceback
3420
# Deprecated, use PatienceSequenceMatcher instead
3421
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
3424
def annotate_knit(knit, revision_id):
3425
"""Annotate a knit with no cached annotations.
3427
This implementation is for knits with no cached annotations.
3428
It will work for knits with cached annotations, but this is not
3431
annotator = _KnitAnnotator(knit)
3432
return iter(annotator.annotate_flat(revision_id))
3435
class _KnitAnnotator(annotate.Annotator):
3436
"""Build up the annotations for a text."""
3438
def __init__(self, vf):
3439
annotate.Annotator.__init__(self, vf)
3441
# TODO: handle Nodes which cannot be extracted
3442
# self._ghosts = set()
3444
# Map from (key, parent_key) => matching_blocks, should be 'use once'
3445
self._matching_blocks = {}
3447
# KnitContent objects
3448
self._content_objects = {}
3449
# The number of children that depend on this fulltext content object
3450
self._num_compression_children = {}
3451
# Delta records that need their compression parent before they can be
3453
self._pending_deltas = {}
3454
# Fulltext records that are waiting for their parents fulltexts before
3455
# they can be yielded for annotation
3456
self._pending_annotation = {}
3458
self._all_build_details = {}
3460
def _get_build_graph(self, key):
3461
"""Get the graphs for building texts and annotations.
3463
The data you need for creating a full text may be different than the
3464
data you need to annotate that text. (At a minimum, you need both
3465
parents to create an annotation, but only need 1 parent to generate the
3468
:return: A list of (key, index_memo) records, suitable for
3469
passing to read_records_iter to start reading in the raw data from
3472
pending = set([key])
3475
self._num_needed_children[key] = 1
3477
# get all pending nodes
3478
this_iteration = pending
3479
build_details = self._vf._index.get_build_details(this_iteration)
3480
self._all_build_details.update(build_details)
3481
# new_nodes = self._vf._index._get_entries(this_iteration)
3483
for key, details in build_details.iteritems():
3484
(index_memo, compression_parent, parent_keys,
3485
record_details) = details
3486
self._parent_map[key] = parent_keys
3487
self._heads_provider = None
3488
records.append((key, index_memo))
3489
# Do we actually need to check _annotated_lines?
3490
pending.update([p for p in parent_keys
3491
if p not in self._all_build_details])
3493
for parent_key in parent_keys:
3494
if parent_key in self._num_needed_children:
3495
self._num_needed_children[parent_key] += 1
3497
self._num_needed_children[parent_key] = 1
3498
if compression_parent:
3499
if compression_parent in self._num_compression_children:
3500
self._num_compression_children[compression_parent] += 1
3502
self._num_compression_children[compression_parent] = 1
3504
missing_versions = this_iteration.difference(build_details.keys())
3505
if missing_versions:
3506
for key in missing_versions:
3507
if key in self._parent_map and key in self._text_cache:
3508
# We already have this text ready, we just need to
3509
# yield it later so we get it annotated
3511
parent_keys = self._parent_map[key]
3512
for parent_key in parent_keys:
3513
if parent_key in self._num_needed_children:
3514
self._num_needed_children[parent_key] += 1
3516
self._num_needed_children[parent_key] = 1
3517
pending.update([p for p in parent_keys
3518
if p not in self._all_build_details])
3520
raise errors.RevisionNotPresent(key, self._vf)
3521
# Generally we will want to read the records in reverse order, because
3522
# we find the parent nodes after the children
3524
return records, ann_keys
3526
def _get_needed_texts(self, key, pb=None):
3527
# if True or len(self._vf._fallback_vfs) > 0:
3528
if len(self._vf._fallback_vfs) > 0:
3529
# If we have fallbacks, go to the generic path
3530
for v in annotate.Annotator._get_needed_texts(self, key, pb=pb):
3535
records, ann_keys = self._get_build_graph(key)
3536
for idx, (sub_key, text, num_lines) in enumerate(
3537
self._extract_texts(records)):
3539
pb.update('annotating', idx, len(records))
3540
yield sub_key, text, num_lines
3541
for sub_key in ann_keys:
3542
text = self._text_cache[sub_key]
3543
num_lines = len(text) # bad assumption
3544
yield sub_key, text, num_lines
3546
except errors.RetryWithNewPacks, e:
3547
self._vf._access.reload_or_raise(e)
3548
# The cached build_details are no longer valid
3549
self._all_build_details.clear()
3551
def _cache_delta_blocks(self, key, compression_parent, delta, lines):
3552
parent_lines = self._text_cache[compression_parent]
3553
blocks = list(KnitContent.get_line_delta_blocks(delta, parent_lines, lines))
3554
self._matching_blocks[(key, compression_parent)] = blocks
3556
def _expand_record(self, key, parent_keys, compression_parent, record,
3559
if compression_parent:
3560
if compression_parent not in self._content_objects:
3561
# Waiting for the parent
3562
self._pending_deltas.setdefault(compression_parent, []).append(
3563
(key, parent_keys, record, record_details))
3565
# We have the basis parent, so expand the delta
3566
num = self._num_compression_children[compression_parent]
3569
base_content = self._content_objects.pop(compression_parent)
3570
self._num_compression_children.pop(compression_parent)
3572
self._num_compression_children[compression_parent] = num
3573
base_content = self._content_objects[compression_parent]
3574
# It is tempting to want to copy_base_content=False for the last
3575
# child object. However, whenever noeol=False,
3576
# self._text_cache[parent_key] is content._lines. So mutating it
3577
# gives very bad results.
3578
# The alternative is to copy the lines into text cache, but then we
3579
# are copying anyway, so just do it here.
3580
content, delta = self._vf._factory.parse_record(
3581
key, record, record_details, base_content,
3582
copy_base_content=True)
3585
content, _ = self._vf._factory.parse_record(
3586
key, record, record_details, None)
3587
if self._num_compression_children.get(key, 0) > 0:
3588
self._content_objects[key] = content
3589
lines = content.text()
3590
self._text_cache[key] = lines
3591
if delta is not None:
3592
self._cache_delta_blocks(key, compression_parent, delta, lines)
3595
def _get_parent_annotations_and_matches(self, key, text, parent_key):
3596
"""Get the list of annotations for the parent, and the matching lines.
3598
:param text: The opaque value given by _get_needed_texts
3599
:param parent_key: The key for the parent text
3600
:return: (parent_annotations, matching_blocks)
3601
parent_annotations is a list as long as the number of lines in
3603
matching_blocks is a list of (parent_idx, text_idx, len) tuples
3604
indicating which lines match between the two texts
3606
block_key = (key, parent_key)
3607
if block_key in self._matching_blocks:
3608
blocks = self._matching_blocks.pop(block_key)
3609
parent_annotations = self._annotations_cache[parent_key]
3610
return parent_annotations, blocks
3611
return annotate.Annotator._get_parent_annotations_and_matches(self,
3612
key, text, parent_key)
3614
def _process_pending(self, key):
3615
"""The content for 'key' was just processed.
3617
Determine if there is any more pending work to be processed.
3620
if key in self._pending_deltas:
3621
compression_parent = key
3622
children = self._pending_deltas.pop(key)
3623
for child_key, parent_keys, record, record_details in children:
3624
lines = self._expand_record(child_key, parent_keys,
3626
record, record_details)
3627
if self._check_ready_for_annotations(child_key, parent_keys):
3628
to_return.append(child_key)
3629
# Also check any children that are waiting for this parent to be
3631
if key in self._pending_annotation:
3632
children = self._pending_annotation.pop(key)
3633
to_return.extend([c for c, p_keys in children
3634
if self._check_ready_for_annotations(c, p_keys)])
3637
def _check_ready_for_annotations(self, key, parent_keys):
3638
"""return true if this text is ready to be yielded.
3640
Otherwise, this will return False, and queue the text into
3641
self._pending_annotation
3643
for parent_key in parent_keys:
3644
if parent_key not in self._annotations_cache:
3645
# still waiting on at least one parent text, so queue it up
3646
# Note that if there are multiple parents, we need to wait
3648
self._pending_annotation.setdefault(parent_key,
3649
[]).append((key, parent_keys))
3653
def _extract_texts(self, records):
3654
"""Extract the various texts needed based on records"""
3655
# We iterate in the order read, rather than a strict order requested
3656
# However, process what we can, and put off to the side things that
3657
# still need parents, cleaning them up when those parents are
3660
# 1) As 'records' are read, see if we can expand these records into
3661
# Content objects (and thus lines)
3662
# 2) If a given line-delta is waiting on its compression parent, it
3663
# gets queued up into self._pending_deltas, otherwise we expand
3664
# it, and put it into self._text_cache and self._content_objects
3665
# 3) If we expanded the text, we will then check to see if all
3666
# parents have also been processed. If so, this text gets yielded,
3667
# else this record gets set aside into pending_annotation
3668
# 4) Further, if we expanded the text in (2), we will then check to
3669
# see if there are any children in self._pending_deltas waiting to
3670
# also be processed. If so, we go back to (2) for those
3671
# 5) Further again, if we yielded the text, we can then check if that
3672
# 'unlocks' any of the texts in pending_annotations, which should
3673
# then get yielded as well
3674
# Note that both steps 4 and 5 are 'recursive' in that unlocking one
3675
# compression child could unlock yet another, and yielding a fulltext
3676
# will also 'unlock' the children that are waiting on that annotation.
3677
# (Though also, unlocking 1 parent's fulltext, does not unlock a child
3678
# if other parents are also waiting.)
3679
# We want to yield content before expanding child content objects, so
3680
# that we know when we can re-use the content lines, and the annotation
3681
# code can know when it can stop caching fulltexts, as well.
3683
# Children that are missing their compression parent
3685
for (key, record, digest) in self._vf._read_records_iter(records):
3687
details = self._all_build_details[key]
3688
(_, compression_parent, parent_keys, record_details) = details
3689
lines = self._expand_record(key, parent_keys, compression_parent,
3690
record, record_details)
3692
# Pending delta should be queued up
3694
# At this point, we may be able to yield this content, if all
3695
# parents are also finished
3696
yield_this_text = self._check_ready_for_annotations(key,
3699
# All parents present
3700
yield key, lines, len(lines)
3701
to_process = self._process_pending(key)
3703
this_process = to_process
3705
for key in this_process:
3706
lines = self._text_cache[key]
3707
yield key, lines, len(lines)
3708
to_process.extend(self._process_pending(key))
3711
from bzrlib._knit_load_data_pyx import _load_data_c as _load_data
3712
except ImportError, e:
3713
osutils.failed_to_load_extension(e)
3714
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
