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(), """
82
from bzrlib.errors import (
90
RevisionAlreadyPresent,
93
from bzrlib.osutils import (
100
from bzrlib.versionedfile import (
101
AbsentContentFactory,
105
ChunkedContentFactory,
112
# TODO: Split out code specific to this format into an associated object.
114
# TODO: Can we put in some kind of value to check that the index and data
115
# files belong together?
117
# TODO: accommodate binaries, perhaps by storing a byte count
119
# TODO: function to check whole file
121
# TODO: atomically append data, then measure backwards from the cursor
122
# position after writing to work out where it was located. we may need to
123
# bypass python file buffering.
125
DATA_SUFFIX = '.knit'
126
INDEX_SUFFIX = '.kndx'
127
_STREAM_MIN_BUFFER_SIZE = 5*1024*1024
130
class KnitAdapter(object):
131
"""Base class for knit record adaption."""
133
def __init__(self, basis_vf):
134
"""Create an adapter which accesses full texts from basis_vf.
136
:param basis_vf: A versioned file to access basis texts of deltas from.
137
May be None for adapters that do not need to access basis texts.
139
self._data = KnitVersionedFiles(None, None)
140
self._annotate_factory = KnitAnnotateFactory()
141
self._plain_factory = KnitPlainFactory()
142
self._basis_vf = basis_vf
145
class FTAnnotatedToUnannotated(KnitAdapter):
146
"""An adapter from FT annotated knits to unannotated ones."""
148
def get_bytes(self, factory):
149
annotated_compressed_bytes = factory._raw_record
151
self._data._parse_record_unchecked(annotated_compressed_bytes)
152
content = self._annotate_factory.parse_fulltext(contents, rec[1])
153
size, bytes = self._data._record_to_data((rec[1],), rec[3], content.text())
157
class DeltaAnnotatedToUnannotated(KnitAdapter):
158
"""An adapter for deltas from annotated to unannotated."""
160
def get_bytes(self, factory):
161
annotated_compressed_bytes = factory._raw_record
163
self._data._parse_record_unchecked(annotated_compressed_bytes)
164
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
166
contents = self._plain_factory.lower_line_delta(delta)
167
size, bytes = self._data._record_to_data((rec[1],), rec[3], contents)
171
class FTAnnotatedToFullText(KnitAdapter):
172
"""An adapter from FT annotated knits to unannotated ones."""
174
def get_bytes(self, factory):
175
annotated_compressed_bytes = factory._raw_record
177
self._data._parse_record_unchecked(annotated_compressed_bytes)
178
content, delta = self._annotate_factory.parse_record(factory.key[-1],
179
contents, factory._build_details, None)
180
return ''.join(content.text())
183
class DeltaAnnotatedToFullText(KnitAdapter):
184
"""An adapter for deltas from annotated to unannotated."""
186
def get_bytes(self, factory):
187
annotated_compressed_bytes = factory._raw_record
189
self._data._parse_record_unchecked(annotated_compressed_bytes)
190
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
192
compression_parent = factory.parents[0]
193
basis_entry = self._basis_vf.get_record_stream(
194
[compression_parent], 'unordered', True).next()
195
if basis_entry.storage_kind == 'absent':
196
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
197
basis_chunks = basis_entry.get_bytes_as('chunked')
198
basis_lines = osutils.chunks_to_lines(basis_chunks)
199
# Manually apply the delta because we have one annotated content and
201
basis_content = PlainKnitContent(basis_lines, compression_parent)
202
basis_content.apply_delta(delta, rec[1])
203
basis_content._should_strip_eol = factory._build_details[1]
204
return ''.join(basis_content.text())
207
class FTPlainToFullText(KnitAdapter):
208
"""An adapter from FT plain knits to unannotated ones."""
210
def get_bytes(self, factory):
211
compressed_bytes = factory._raw_record
213
self._data._parse_record_unchecked(compressed_bytes)
214
content, delta = self._plain_factory.parse_record(factory.key[-1],
215
contents, factory._build_details, None)
216
return ''.join(content.text())
219
class DeltaPlainToFullText(KnitAdapter):
220
"""An adapter for deltas from annotated to unannotated."""
222
def get_bytes(self, factory):
223
compressed_bytes = factory._raw_record
225
self._data._parse_record_unchecked(compressed_bytes)
226
delta = self._plain_factory.parse_line_delta(contents, rec[1])
227
compression_parent = factory.parents[0]
228
# XXX: string splitting overhead.
229
basis_entry = self._basis_vf.get_record_stream(
230
[compression_parent], 'unordered', True).next()
231
if basis_entry.storage_kind == 'absent':
232
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
233
basis_chunks = basis_entry.get_bytes_as('chunked')
234
basis_lines = osutils.chunks_to_lines(basis_chunks)
235
basis_content = PlainKnitContent(basis_lines, compression_parent)
236
# Manually apply the delta because we have one annotated content and
238
content, _ = self._plain_factory.parse_record(rec[1], contents,
239
factory._build_details, basis_content)
240
return ''.join(content.text())
243
class KnitContentFactory(ContentFactory):
244
"""Content factory for streaming from knits.
246
:seealso ContentFactory:
249
def __init__(self, key, parents, build_details, sha1, raw_record,
250
annotated, knit=None, network_bytes=None):
251
"""Create a KnitContentFactory for key.
254
:param parents: The parents.
255
:param build_details: The build details as returned from
257
:param sha1: The sha1 expected from the full text of this object.
258
:param raw_record: The bytes of the knit data from disk.
259
:param annotated: True if the raw data is annotated.
260
:param network_bytes: None to calculate the network bytes on demand,
261
not-none if they are already known.
263
ContentFactory.__init__(self)
266
self.parents = parents
267
if build_details[0] == 'line-delta':
272
annotated_kind = 'annotated-'
275
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
276
self._raw_record = raw_record
277
self._network_bytes = network_bytes
278
self._build_details = build_details
281
def _create_network_bytes(self):
282
"""Create a fully serialised network version for transmission."""
283
# storage_kind, key, parents, Noeol, raw_record
284
key_bytes = '\x00'.join(self.key)
285
if self.parents is None:
286
parent_bytes = 'None:'
288
parent_bytes = '\t'.join('\x00'.join(key) for key in self.parents)
289
if self._build_details[1]:
293
network_bytes = "%s\n%s\n%s\n%s%s" % (self.storage_kind, key_bytes,
294
parent_bytes, noeol, self._raw_record)
295
self._network_bytes = network_bytes
297
def get_bytes_as(self, storage_kind):
298
if storage_kind == self.storage_kind:
299
if self._network_bytes is None:
300
self._create_network_bytes()
301
return self._network_bytes
302
if ('-ft-' in self.storage_kind and
303
storage_kind in ('chunked', 'fulltext')):
304
adapter_key = (self.storage_kind, 'fulltext')
305
adapter_factory = adapter_registry.get(adapter_key)
306
adapter = adapter_factory(None)
307
bytes = adapter.get_bytes(self)
308
if storage_kind == 'chunked':
312
if self._knit is not None:
313
# Not redundant with direct conversion above - that only handles
315
if storage_kind == 'chunked':
316
return self._knit.get_lines(self.key[0])
317
elif storage_kind == 'fulltext':
318
return self._knit.get_text(self.key[0])
319
raise errors.UnavailableRepresentation(self.key, storage_kind,
323
class LazyKnitContentFactory(ContentFactory):
324
"""A ContentFactory which can either generate full text or a wire form.
326
:seealso ContentFactory:
329
def __init__(self, key, parents, generator, first):
330
"""Create a LazyKnitContentFactory.
332
:param key: The key of the record.
333
:param parents: The parents of the record.
334
:param generator: A _ContentMapGenerator containing the record for this
336
:param first: Is this the first content object returned from generator?
337
if it is, its storage kind is knit-delta-closure, otherwise it is
338
knit-delta-closure-ref
341
self.parents = parents
343
self._generator = generator
344
self.storage_kind = "knit-delta-closure"
346
self.storage_kind = self.storage_kind + "-ref"
349
def get_bytes_as(self, storage_kind):
350
if storage_kind == self.storage_kind:
352
return self._generator._wire_bytes()
354
# all the keys etc are contained in the bytes returned in the
357
if storage_kind in ('chunked', 'fulltext'):
358
chunks = self._generator._get_one_work(self.key).text()
359
if storage_kind == 'chunked':
362
return ''.join(chunks)
363
raise errors.UnavailableRepresentation(self.key, storage_kind,
367
def knit_delta_closure_to_records(storage_kind, bytes, line_end):
368
"""Convert a network record to a iterator over stream records.
370
:param storage_kind: The storage kind of the record.
371
Must be 'knit-delta-closure'.
372
:param bytes: The bytes of the record on the network.
374
generator = _NetworkContentMapGenerator(bytes, line_end)
375
return generator.get_record_stream()
378
def knit_network_to_record(storage_kind, bytes, line_end):
379
"""Convert a network record to a record object.
381
:param storage_kind: The storage kind of the record.
382
:param bytes: The bytes of the record on the network.
385
line_end = bytes.find('\n', start)
386
key = tuple(bytes[start:line_end].split('\x00'))
388
line_end = bytes.find('\n', start)
389
parent_line = bytes[start:line_end]
390
if parent_line == 'None:':
394
[tuple(segment.split('\x00')) for segment in parent_line.split('\t')
397
noeol = bytes[start] == 'N'
398
if 'ft' in storage_kind:
401
method = 'line-delta'
402
build_details = (method, noeol)
404
raw_record = bytes[start:]
405
annotated = 'annotated' in storage_kind
406
return [KnitContentFactory(key, parents, build_details, None, raw_record,
407
annotated, network_bytes=bytes)]
410
class KnitContent(object):
411
"""Content of a knit version to which deltas can be applied.
413
This is always stored in memory as a list of lines with \n at the end,
414
plus a flag saying if the final ending is really there or not, because that
415
corresponds to the on-disk knit representation.
419
self._should_strip_eol = False
421
def apply_delta(self, delta, new_version_id):
422
"""Apply delta to this object to become new_version_id."""
423
raise NotImplementedError(self.apply_delta)
425
def line_delta_iter(self, new_lines):
426
"""Generate line-based delta from this content to new_lines."""
427
new_texts = new_lines.text()
428
old_texts = self.text()
429
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
430
for tag, i1, i2, j1, j2 in s.get_opcodes():
433
# ofrom, oto, length, data
434
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
436
def line_delta(self, new_lines):
437
return list(self.line_delta_iter(new_lines))
440
def get_line_delta_blocks(knit_delta, source, target):
441
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
442
target_len = len(target)
445
for s_begin, s_end, t_len, new_text in knit_delta:
446
true_n = s_begin - s_pos
449
# knit deltas do not provide reliable info about whether the
450
# last line of a file matches, due to eol handling.
451
if source[s_pos + n -1] != target[t_pos + n -1]:
454
yield s_pos, t_pos, n
455
t_pos += t_len + true_n
457
n = target_len - t_pos
459
if source[s_pos + n -1] != target[t_pos + n -1]:
462
yield s_pos, t_pos, n
463
yield s_pos + (target_len - t_pos), target_len, 0
466
class AnnotatedKnitContent(KnitContent):
467
"""Annotated content."""
469
def __init__(self, lines):
470
KnitContent.__init__(self)
474
"""Return a list of (origin, text) for each content line."""
475
lines = self._lines[:]
476
if self._should_strip_eol:
477
origin, last_line = lines[-1]
478
lines[-1] = (origin, last_line.rstrip('\n'))
481
def apply_delta(self, delta, new_version_id):
482
"""Apply delta to this object to become new_version_id."""
485
for start, end, count, delta_lines in delta:
486
lines[offset+start:offset+end] = delta_lines
487
offset = offset + (start - end) + count
491
lines = [text for origin, text in self._lines]
492
except ValueError, e:
493
# most commonly (only?) caused by the internal form of the knit
494
# missing annotation information because of a bug - see thread
496
raise KnitCorrupt(self,
497
"line in annotated knit missing annotation information: %s"
499
if self._should_strip_eol:
500
lines[-1] = lines[-1].rstrip('\n')
504
return AnnotatedKnitContent(self._lines[:])
507
class PlainKnitContent(KnitContent):
508
"""Unannotated content.
510
When annotate[_iter] is called on this content, the same version is reported
511
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
515
def __init__(self, lines, version_id):
516
KnitContent.__init__(self)
518
self._version_id = version_id
521
"""Return a list of (origin, text) for each content line."""
522
return [(self._version_id, line) for line in self._lines]
524
def apply_delta(self, delta, new_version_id):
525
"""Apply delta to this object to become new_version_id."""
528
for start, end, count, delta_lines in delta:
529
lines[offset+start:offset+end] = delta_lines
530
offset = offset + (start - end) + count
531
self._version_id = new_version_id
534
return PlainKnitContent(self._lines[:], self._version_id)
538
if self._should_strip_eol:
540
lines[-1] = lines[-1].rstrip('\n')
544
class _KnitFactory(object):
545
"""Base class for common Factory functions."""
547
def parse_record(self, version_id, record, record_details,
548
base_content, copy_base_content=True):
549
"""Parse a record into a full content object.
551
:param version_id: The official version id for this content
552
:param record: The data returned by read_records_iter()
553
:param record_details: Details about the record returned by
555
:param base_content: If get_build_details returns a compression_parent,
556
you must return a base_content here, else use None
557
:param copy_base_content: When building from the base_content, decide
558
you can either copy it and return a new object, or modify it in
560
:return: (content, delta) A Content object and possibly a line-delta,
563
method, noeol = record_details
564
if method == 'line-delta':
565
if copy_base_content:
566
content = base_content.copy()
568
content = base_content
569
delta = self.parse_line_delta(record, version_id)
570
content.apply_delta(delta, version_id)
572
content = self.parse_fulltext(record, version_id)
574
content._should_strip_eol = noeol
575
return (content, delta)
578
class KnitAnnotateFactory(_KnitFactory):
579
"""Factory for creating annotated Content objects."""
583
def make(self, lines, version_id):
584
num_lines = len(lines)
585
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
587
def parse_fulltext(self, content, version_id):
588
"""Convert fulltext to internal representation
590
fulltext content is of the format
591
revid(utf8) plaintext\n
592
internal representation is of the format:
595
# TODO: jam 20070209 The tests expect this to be returned as tuples,
596
# but the code itself doesn't really depend on that.
597
# Figure out a way to not require the overhead of turning the
598
# list back into tuples.
599
lines = [tuple(line.split(' ', 1)) for line in content]
600
return AnnotatedKnitContent(lines)
602
def parse_line_delta_iter(self, lines):
603
return iter(self.parse_line_delta(lines))
605
def parse_line_delta(self, lines, version_id, plain=False):
606
"""Convert a line based delta into internal representation.
608
line delta is in the form of:
609
intstart intend intcount
611
revid(utf8) newline\n
612
internal representation is
613
(start, end, count, [1..count tuples (revid, newline)])
615
:param plain: If True, the lines are returned as a plain
616
list without annotations, not as a list of (origin, content) tuples, i.e.
617
(start, end, count, [1..count newline])
624
def cache_and_return(line):
625
origin, text = line.split(' ', 1)
626
return cache.setdefault(origin, origin), text
628
# walk through the lines parsing.
629
# Note that the plain test is explicitly pulled out of the
630
# loop to minimise any performance impact
633
start, end, count = [int(n) for n in header.split(',')]
634
contents = [next().split(' ', 1)[1] for i in xrange(count)]
635
result.append((start, end, count, contents))
638
start, end, count = [int(n) for n in header.split(',')]
639
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
640
result.append((start, end, count, contents))
643
def get_fulltext_content(self, lines):
644
"""Extract just the content lines from a fulltext."""
645
return (line.split(' ', 1)[1] for line in lines)
647
def get_linedelta_content(self, lines):
648
"""Extract just the content from a line delta.
650
This doesn't return all of the extra information stored in a delta.
651
Only the actual content lines.
656
header = header.split(',')
657
count = int(header[2])
658
for i in xrange(count):
659
origin, text = next().split(' ', 1)
662
def lower_fulltext(self, content):
663
"""convert a fulltext content record into a serializable form.
665
see parse_fulltext which this inverts.
667
return ['%s %s' % (o, t) for o, t in content._lines]
669
def lower_line_delta(self, delta):
670
"""convert a delta into a serializable form.
672
See parse_line_delta which this inverts.
674
# TODO: jam 20070209 We only do the caching thing to make sure that
675
# the origin is a valid utf-8 line, eventually we could remove it
677
for start, end, c, lines in delta:
678
out.append('%d,%d,%d\n' % (start, end, c))
679
out.extend(origin + ' ' + text
680
for origin, text in lines)
683
def annotate(self, knit, key):
684
content = knit._get_content(key)
685
# adjust for the fact that serialised annotations are only key suffixes
687
if type(key) is tuple:
689
origins = content.annotate()
691
for origin, line in origins:
692
result.append((prefix + (origin,), line))
695
# XXX: This smells a bit. Why would key ever be a non-tuple here?
696
# Aren't keys defined to be tuples? -- spiv 20080618
697
return content.annotate()
700
class KnitPlainFactory(_KnitFactory):
701
"""Factory for creating plain Content objects."""
705
def make(self, lines, version_id):
706
return PlainKnitContent(lines, version_id)
708
def parse_fulltext(self, content, version_id):
709
"""This parses an unannotated fulltext.
711
Note that this is not a noop - the internal representation
712
has (versionid, line) - its just a constant versionid.
714
return self.make(content, version_id)
716
def parse_line_delta_iter(self, lines, version_id):
718
num_lines = len(lines)
719
while cur < num_lines:
722
start, end, c = [int(n) for n in header.split(',')]
723
yield start, end, c, lines[cur:cur+c]
726
def parse_line_delta(self, lines, version_id):
727
return list(self.parse_line_delta_iter(lines, version_id))
729
def get_fulltext_content(self, lines):
730
"""Extract just the content lines from a fulltext."""
733
def get_linedelta_content(self, lines):
734
"""Extract just the content from a line delta.
736
This doesn't return all of the extra information stored in a delta.
737
Only the actual content lines.
742
header = header.split(',')
743
count = int(header[2])
744
for i in xrange(count):
747
def lower_fulltext(self, content):
748
return content.text()
750
def lower_line_delta(self, delta):
752
for start, end, c, lines in delta:
753
out.append('%d,%d,%d\n' % (start, end, c))
757
def annotate(self, knit, key):
758
annotator = _KnitAnnotator(knit)
759
return annotator.annotate_flat(key)
763
def make_file_factory(annotated, mapper):
764
"""Create a factory for creating a file based KnitVersionedFiles.
766
This is only functional enough to run interface tests, it doesn't try to
767
provide a full pack environment.
769
:param annotated: knit annotations are wanted.
770
:param mapper: The mapper from keys to paths.
772
def factory(transport):
773
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
774
access = _KnitKeyAccess(transport, mapper)
775
return KnitVersionedFiles(index, access, annotated=annotated)
779
def make_pack_factory(graph, delta, keylength):
780
"""Create a factory for creating a pack based VersionedFiles.
782
This is only functional enough to run interface tests, it doesn't try to
783
provide a full pack environment.
785
:param graph: Store a graph.
786
:param delta: Delta compress contents.
787
:param keylength: How long should keys be.
789
def factory(transport):
790
parents = graph or delta
796
max_delta_chain = 200
799
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
800
key_elements=keylength)
801
stream = transport.open_write_stream('newpack')
802
writer = pack.ContainerWriter(stream.write)
804
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
805
deltas=delta, add_callback=graph_index.add_nodes)
806
access = _DirectPackAccess({})
807
access.set_writer(writer, graph_index, (transport, 'newpack'))
808
result = KnitVersionedFiles(index, access,
809
max_delta_chain=max_delta_chain)
810
result.stream = stream
811
result.writer = writer
816
def cleanup_pack_knit(versioned_files):
817
versioned_files.stream.close()
818
versioned_files.writer.end()
821
def _get_total_build_size(self, keys, positions):
822
"""Determine the total bytes to build these keys.
824
(helper function because _KnitGraphIndex and _KndxIndex work the same, but
825
don't inherit from a common base.)
827
:param keys: Keys that we want to build
828
:param positions: dict of {key, (info, index_memo, comp_parent)} (such
829
as returned by _get_components_positions)
830
:return: Number of bytes to build those keys
832
all_build_index_memos = {}
836
for key in build_keys:
837
# This is mostly for the 'stacked' case
838
# Where we will be getting the data from a fallback
839
if key not in positions:
841
_, index_memo, compression_parent = positions[key]
842
all_build_index_memos[key] = index_memo
843
if compression_parent not in all_build_index_memos:
844
next_keys.add(compression_parent)
845
build_keys = next_keys
846
return sum([index_memo[2] for index_memo
847
in all_build_index_memos.itervalues()])
850
class KnitVersionedFiles(VersionedFiles):
851
"""Storage for many versioned files using knit compression.
853
Backend storage is managed by indices and data objects.
855
:ivar _index: A _KnitGraphIndex or similar that can describe the
856
parents, graph, compression and data location of entries in this
857
KnitVersionedFiles. Note that this is only the index for
858
*this* vfs; if there are fallbacks they must be queried separately.
861
def __init__(self, index, data_access, max_delta_chain=200,
862
annotated=False, reload_func=None):
863
"""Create a KnitVersionedFiles with index and data_access.
865
:param index: The index for the knit data.
866
:param data_access: The access object to store and retrieve knit
868
:param max_delta_chain: The maximum number of deltas to permit during
869
insertion. Set to 0 to prohibit the use of deltas.
870
:param annotated: Set to True to cause annotations to be calculated and
871
stored during insertion.
872
:param reload_func: An function that can be called if we think we need
873
to reload the pack listing and try again. See
874
'bzrlib.repofmt.pack_repo.AggregateIndex' for the signature.
877
self._access = data_access
878
self._max_delta_chain = max_delta_chain
880
self._factory = KnitAnnotateFactory()
882
self._factory = KnitPlainFactory()
883
self._fallback_vfs = []
884
self._reload_func = reload_func
887
return "%s(%r, %r)" % (
888
self.__class__.__name__,
892
def add_fallback_versioned_files(self, a_versioned_files):
893
"""Add a source of texts for texts not present in this knit.
895
:param a_versioned_files: A VersionedFiles object.
897
self._fallback_vfs.append(a_versioned_files)
899
def add_lines(self, key, parents, lines, parent_texts=None,
900
left_matching_blocks=None, nostore_sha=None, random_id=False,
902
"""See VersionedFiles.add_lines()."""
903
self._index._check_write_ok()
904
self._check_add(key, lines, random_id, check_content)
906
# The caller might pass None if there is no graph data, but kndx
907
# indexes can't directly store that, so we give them
908
# an empty tuple instead.
910
line_bytes = ''.join(lines)
911
return self._add(key, lines, parents,
912
parent_texts, left_matching_blocks, nostore_sha, random_id,
913
line_bytes=line_bytes)
915
def _add_text(self, key, parents, text, nostore_sha=None, random_id=False):
916
"""See VersionedFiles._add_text()."""
917
self._index._check_write_ok()
918
self._check_add(key, None, random_id, check_content=False)
919
if text.__class__ is not str:
920
raise errors.BzrBadParameterUnicode("text")
922
# The caller might pass None if there is no graph data, but kndx
923
# indexes can't directly store that, so we give them
924
# an empty tuple instead.
926
return self._add(key, None, parents,
927
None, None, nostore_sha, random_id,
930
def _add(self, key, lines, parents, parent_texts,
931
left_matching_blocks, nostore_sha, random_id,
933
"""Add a set of lines on top of version specified by parents.
935
Any versions not present will be converted into ghosts.
937
:param lines: A list of strings where each one is a single line (has a
938
single newline at the end of the string) This is now optional
939
(callers can pass None). It is left in its location for backwards
940
compatibility. It should ''.join(lines) must == line_bytes
941
:param line_bytes: A single string containing the content
943
We pass both lines and line_bytes because different routes bring the
944
values to this function. And for memory efficiency, we don't want to
945
have to split/join on-demand.
947
# first thing, if the content is something we don't need to store, find
949
digest = sha_string(line_bytes)
950
if nostore_sha == digest:
951
raise errors.ExistingContent
954
if parent_texts is None:
956
# Do a single query to ascertain parent presence; we only compress
957
# against parents in the same kvf.
958
present_parent_map = self._index.get_parent_map(parents)
959
for parent in parents:
960
if parent in present_parent_map:
961
present_parents.append(parent)
963
# Currently we can only compress against the left most present parent.
964
if (len(present_parents) == 0 or
965
present_parents[0] != parents[0]):
968
# To speed the extract of texts the delta chain is limited
969
# to a fixed number of deltas. This should minimize both
970
# I/O and the time spend applying deltas.
971
delta = self._check_should_delta(present_parents[0])
973
text_length = len(line_bytes)
976
# Note: line_bytes is not modified to add a newline, that is tracked
977
# via the no_eol flag. 'lines' *is* modified, because that is the
978
# general values needed by the Content code.
979
if line_bytes and line_bytes[-1] != '\n':
980
options.append('no-eol')
982
# Copy the existing list, or create a new one
984
lines = osutils.split_lines(line_bytes)
987
# Replace the last line with one that ends in a final newline
988
lines[-1] = lines[-1] + '\n'
990
lines = osutils.split_lines(line_bytes)
992
for element in key[:-1]:
993
if type(element) is not str:
994
raise TypeError("key contains non-strings: %r" % (key,))
996
key = key[:-1] + ('sha1:' + digest,)
997
elif type(key[-1]) is not str:
998
raise TypeError("key contains non-strings: %r" % (key,))
999
# Knit hunks are still last-element only
1000
version_id = key[-1]
1001
content = self._factory.make(lines, version_id)
1003
# Hint to the content object that its text() call should strip the
1005
content._should_strip_eol = True
1006
if delta or (self._factory.annotated and len(present_parents) > 0):
1007
# Merge annotations from parent texts if needed.
1008
delta_hunks = self._merge_annotations(content, present_parents,
1009
parent_texts, delta, self._factory.annotated,
1010
left_matching_blocks)
1013
options.append('line-delta')
1014
store_lines = self._factory.lower_line_delta(delta_hunks)
1015
size, bytes = self._record_to_data(key, digest,
1018
options.append('fulltext')
1019
# isinstance is slower and we have no hierarchy.
1020
if self._factory.__class__ is KnitPlainFactory:
1021
# Use the already joined bytes saving iteration time in
1023
dense_lines = [line_bytes]
1025
dense_lines.append('\n')
1026
size, bytes = self._record_to_data(key, digest,
1029
# get mixed annotation + content and feed it into the
1031
store_lines = self._factory.lower_fulltext(content)
1032
size, bytes = self._record_to_data(key, digest,
1035
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
1036
self._index.add_records(
1037
((key, options, access_memo, parents),),
1038
random_id=random_id)
1039
return digest, text_length, content
1041
def annotate(self, key):
1042
"""See VersionedFiles.annotate."""
1043
return self._factory.annotate(self, key)
1045
def check(self, progress_bar=None):
1046
"""See VersionedFiles.check()."""
1047
# This doesn't actually test extraction of everything, but that will
1048
# impact 'bzr check' substantially, and needs to be integrated with
1049
# care. However, it does check for the obvious problem of a delta with
1051
keys = self._index.keys()
1052
parent_map = self.get_parent_map(keys)
1054
if self._index.get_method(key) != 'fulltext':
1055
compression_parent = parent_map[key][0]
1056
if compression_parent not in parent_map:
1057
raise errors.KnitCorrupt(self,
1058
"Missing basis parent %s for %s" % (
1059
compression_parent, key))
1060
for fallback_vfs in self._fallback_vfs:
1061
fallback_vfs.check()
1063
def _check_add(self, key, lines, random_id, check_content):
1064
"""check that version_id and lines are safe to add."""
1065
version_id = key[-1]
1066
if version_id is not None:
1067
if contains_whitespace(version_id):
1068
raise InvalidRevisionId(version_id, self)
1069
self.check_not_reserved_id(version_id)
1070
# TODO: If random_id==False and the key is already present, we should
1071
# probably check that the existing content is identical to what is
1072
# being inserted, and otherwise raise an exception. This would make
1073
# the bundle code simpler.
1075
self._check_lines_not_unicode(lines)
1076
self._check_lines_are_lines(lines)
1078
def _check_header(self, key, line):
1079
rec = self._split_header(line)
1080
self._check_header_version(rec, key[-1])
1083
def _check_header_version(self, rec, version_id):
1084
"""Checks the header version on original format knit records.
1086
These have the last component of the key embedded in the record.
1088
if rec[1] != version_id:
1089
raise KnitCorrupt(self,
1090
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1092
def _check_should_delta(self, parent):
1093
"""Iterate back through the parent listing, looking for a fulltext.
1095
This is used when we want to decide whether to add a delta or a new
1096
fulltext. It searches for _max_delta_chain parents. When it finds a
1097
fulltext parent, it sees if the total size of the deltas leading up to
1098
it is large enough to indicate that we want a new full text anyway.
1100
Return True if we should create a new delta, False if we should use a
1104
fulltext_size = None
1105
for count in xrange(self._max_delta_chain):
1107
# Note that this only looks in the index of this particular
1108
# KnitVersionedFiles, not in the fallbacks. This ensures that
1109
# we won't store a delta spanning physical repository
1111
build_details = self._index.get_build_details([parent])
1112
parent_details = build_details[parent]
1113
except (RevisionNotPresent, KeyError), e:
1114
# Some basis is not locally present: always fulltext
1116
index_memo, compression_parent, _, _ = parent_details
1117
_, _, size = index_memo
1118
if compression_parent is None:
1119
fulltext_size = size
1122
# We don't explicitly check for presence because this is in an
1123
# inner loop, and if it's missing it'll fail anyhow.
1124
parent = compression_parent
1126
# We couldn't find a fulltext, so we must create a new one
1128
# Simple heuristic - if the total I/O wold be greater as a delta than
1129
# the originally installed fulltext, we create a new fulltext.
1130
return fulltext_size > delta_size
1132
def _build_details_to_components(self, build_details):
1133
"""Convert a build_details tuple to a position tuple."""
1134
# record_details, access_memo, compression_parent
1135
return build_details[3], build_details[0], build_details[1]
1137
def _get_components_positions(self, keys, allow_missing=False):
1138
"""Produce a map of position data for the components of keys.
1140
This data is intended to be used for retrieving the knit records.
1142
A dict of key to (record_details, index_memo, next, parents) is
1144
method is the way referenced data should be applied.
1145
index_memo is the handle to pass to the data access to actually get the
1147
next is the build-parent of the version, or None for fulltexts.
1148
parents is the version_ids of the parents of this version
1150
:param allow_missing: If True do not raise an error on a missing component,
1154
pending_components = keys
1155
while pending_components:
1156
build_details = self._index.get_build_details(pending_components)
1157
current_components = set(pending_components)
1158
pending_components = set()
1159
for key, details in build_details.iteritems():
1160
(index_memo, compression_parent, parents,
1161
record_details) = details
1162
method = record_details[0]
1163
if compression_parent is not None:
1164
pending_components.add(compression_parent)
1165
component_data[key] = self._build_details_to_components(details)
1166
missing = current_components.difference(build_details)
1167
if missing and not allow_missing:
1168
raise errors.RevisionNotPresent(missing.pop(), self)
1169
return component_data
1171
def _get_content(self, key, parent_texts={}):
1172
"""Returns a content object that makes up the specified
1174
cached_version = parent_texts.get(key, None)
1175
if cached_version is not None:
1176
# Ensure the cache dict is valid.
1177
if not self.get_parent_map([key]):
1178
raise RevisionNotPresent(key, self)
1179
return cached_version
1180
generator = _VFContentMapGenerator(self, [key])
1181
return generator._get_content(key)
1183
def get_parent_map(self, keys):
1184
"""Get a map of the graph parents of keys.
1186
:param keys: The keys to look up parents for.
1187
:return: A mapping from keys to parents. Absent keys are absent from
1190
return self._get_parent_map_with_sources(keys)[0]
1192
def _get_parent_map_with_sources(self, keys):
1193
"""Get a map of the parents of keys.
1195
:param keys: The keys to look up parents for.
1196
:return: A tuple. The first element is a mapping from keys to parents.
1197
Absent keys are absent from the mapping. The second element is a
1198
list with the locations each key was found in. The first element
1199
is the in-this-knit parents, the second the first fallback source,
1203
sources = [self._index] + self._fallback_vfs
1206
for source in sources:
1209
new_result = source.get_parent_map(missing)
1210
source_results.append(new_result)
1211
result.update(new_result)
1212
missing.difference_update(set(new_result))
1213
return result, source_results
1215
def _get_record_map(self, keys, allow_missing=False):
1216
"""Produce a dictionary of knit records.
1218
:return: {key:(record, record_details, digest, next)}
1220
data returned from read_records (a KnitContentobject)
1222
opaque information to pass to parse_record
1224
SHA1 digest of the full text after all steps are done
1226
build-parent of the version, i.e. the leftmost ancestor.
1227
Will be None if the record is not a delta.
1228
:param keys: The keys to build a map for
1229
:param allow_missing: If some records are missing, rather than
1230
error, just return the data that could be generated.
1232
raw_map = self._get_record_map_unparsed(keys,
1233
allow_missing=allow_missing)
1234
return self._raw_map_to_record_map(raw_map)
1236
def _raw_map_to_record_map(self, raw_map):
1237
"""Parse the contents of _get_record_map_unparsed.
1239
:return: see _get_record_map.
1243
data, record_details, next = raw_map[key]
1244
content, digest = self._parse_record(key[-1], data)
1245
result[key] = content, record_details, digest, next
1248
def _get_record_map_unparsed(self, keys, allow_missing=False):
1249
"""Get the raw data for reconstructing keys without parsing it.
1251
:return: A dict suitable for parsing via _raw_map_to_record_map.
1252
key-> raw_bytes, (method, noeol), compression_parent
1254
# This retries the whole request if anything fails. Potentially we
1255
# could be a bit more selective. We could track the keys whose records
1256
# we have successfully found, and then only request the new records
1257
# from there. However, _get_components_positions grabs the whole build
1258
# chain, which means we'll likely try to grab the same records again
1259
# anyway. Also, can the build chains change as part of a pack
1260
# operation? We wouldn't want to end up with a broken chain.
1263
position_map = self._get_components_positions(keys,
1264
allow_missing=allow_missing)
1265
# key = component_id, r = record_details, i_m = index_memo,
1267
records = [(key, i_m) for key, (r, i_m, n)
1268
in position_map.iteritems()]
1269
# Sort by the index memo, so that we request records from the
1270
# same pack file together, and in forward-sorted order
1271
records.sort(key=operator.itemgetter(1))
1273
for key, data in self._read_records_iter_unchecked(records):
1274
(record_details, index_memo, next) = position_map[key]
1275
raw_record_map[key] = data, record_details, next
1276
return raw_record_map
1277
except errors.RetryWithNewPacks, e:
1278
self._access.reload_or_raise(e)
1281
def _split_by_prefix(cls, keys):
1282
"""For the given keys, split them up based on their prefix.
1284
To keep memory pressure somewhat under control, split the
1285
requests back into per-file-id requests, otherwise "bzr co"
1286
extracts the full tree into memory before writing it to disk.
1287
This should be revisited if _get_content_maps() can ever cross
1290
The keys for a given file_id are kept in the same relative order.
1291
Ordering between file_ids is not, though prefix_order will return the
1292
order that the key was first seen.
1294
:param keys: An iterable of key tuples
1295
:return: (split_map, prefix_order)
1296
split_map A dictionary mapping prefix => keys
1297
prefix_order The order that we saw the various prefixes
1299
split_by_prefix = {}
1307
if prefix in split_by_prefix:
1308
split_by_prefix[prefix].append(key)
1310
split_by_prefix[prefix] = [key]
1311
prefix_order.append(prefix)
1312
return split_by_prefix, prefix_order
1314
def _group_keys_for_io(self, keys, non_local_keys, positions,
1315
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1316
"""For the given keys, group them into 'best-sized' requests.
1318
The idea is to avoid making 1 request per file, but to never try to
1319
unpack an entire 1.5GB source tree in a single pass. Also when
1320
possible, we should try to group requests to the same pack file
1323
:return: list of (keys, non_local) tuples that indicate what keys
1324
should be fetched next.
1326
# TODO: Ideally we would group on 2 factors. We want to extract texts
1327
# from the same pack file together, and we want to extract all
1328
# the texts for a given build-chain together. Ultimately it
1329
# probably needs a better global view.
1330
total_keys = len(keys)
1331
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1332
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1334
cur_non_local = set()
1338
for prefix in prefix_order:
1339
keys = prefix_split_keys[prefix]
1340
non_local = prefix_split_non_local_keys.get(prefix, [])
1342
this_size = self._index._get_total_build_size(keys, positions)
1343
cur_size += this_size
1344
cur_keys.extend(keys)
1345
cur_non_local.update(non_local)
1346
if cur_size > _min_buffer_size:
1347
result.append((cur_keys, cur_non_local))
1348
sizes.append(cur_size)
1350
cur_non_local = set()
1353
result.append((cur_keys, cur_non_local))
1354
sizes.append(cur_size)
1357
def get_record_stream(self, keys, ordering, include_delta_closure):
1358
"""Get a stream of records for keys.
1360
:param keys: The keys to include.
1361
:param ordering: Either 'unordered' or 'topological'. A topologically
1362
sorted stream has compression parents strictly before their
1364
:param include_delta_closure: If True then the closure across any
1365
compression parents will be included (in the opaque data).
1366
:return: An iterator of ContentFactory objects, each of which is only
1367
valid until the iterator is advanced.
1369
# keys might be a generator
1373
if not self._index.has_graph:
1374
# Cannot sort when no graph has been stored.
1375
ordering = 'unordered'
1377
remaining_keys = keys
1380
keys = set(remaining_keys)
1381
for content_factory in self._get_remaining_record_stream(keys,
1382
ordering, include_delta_closure):
1383
remaining_keys.discard(content_factory.key)
1384
yield content_factory
1386
except errors.RetryWithNewPacks, e:
1387
self._access.reload_or_raise(e)
1389
def _get_remaining_record_stream(self, keys, ordering,
1390
include_delta_closure):
1391
"""This function is the 'retry' portion for get_record_stream."""
1392
if include_delta_closure:
1393
positions = self._get_components_positions(keys, allow_missing=True)
1395
build_details = self._index.get_build_details(keys)
1397
# (record_details, access_memo, compression_parent_key)
1398
positions = dict((key, self._build_details_to_components(details))
1399
for key, details in build_details.iteritems())
1400
absent_keys = keys.difference(set(positions))
1401
# There may be more absent keys : if we're missing the basis component
1402
# and are trying to include the delta closure.
1403
# XXX: We should not ever need to examine remote sources because we do
1404
# not permit deltas across versioned files boundaries.
1405
if include_delta_closure:
1406
needed_from_fallback = set()
1407
# Build up reconstructable_keys dict. key:True in this dict means
1408
# the key can be reconstructed.
1409
reconstructable_keys = {}
1413
chain = [key, positions[key][2]]
1415
needed_from_fallback.add(key)
1418
while chain[-1] is not None:
1419
if chain[-1] in reconstructable_keys:
1420
result = reconstructable_keys[chain[-1]]
1424
chain.append(positions[chain[-1]][2])
1426
# missing basis component
1427
needed_from_fallback.add(chain[-1])
1430
for chain_key in chain[:-1]:
1431
reconstructable_keys[chain_key] = result
1433
needed_from_fallback.add(key)
1434
# Double index lookups here : need a unified api ?
1435
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1436
if ordering in ('topological', 'groupcompress'):
1437
if ordering == 'topological':
1438
# Global topological sort
1439
present_keys = tsort.topo_sort(global_map)
1441
present_keys = sort_groupcompress(global_map)
1442
# Now group by source:
1444
current_source = None
1445
for key in present_keys:
1446
for parent_map in parent_maps:
1447
if key in parent_map:
1448
key_source = parent_map
1450
if current_source is not key_source:
1451
source_keys.append((key_source, []))
1452
current_source = key_source
1453
source_keys[-1][1].append(key)
1455
if ordering != 'unordered':
1456
raise AssertionError('valid values for ordering are:'
1457
' "unordered", "groupcompress" or "topological" not: %r'
1459
# Just group by source; remote sources first.
1462
for parent_map in reversed(parent_maps):
1463
source_keys.append((parent_map, []))
1464
for key in parent_map:
1465
present_keys.append(key)
1466
source_keys[-1][1].append(key)
1467
# We have been requested to return these records in an order that
1468
# suits us. So we ask the index to give us an optimally sorted
1470
for source, sub_keys in source_keys:
1471
if source is parent_maps[0]:
1472
# Only sort the keys for this VF
1473
self._index._sort_keys_by_io(sub_keys, positions)
1474
absent_keys = keys - set(global_map)
1475
for key in absent_keys:
1476
yield AbsentContentFactory(key)
1477
# restrict our view to the keys we can answer.
1478
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1479
# XXX: At that point we need to consider the impact of double reads by
1480
# utilising components multiple times.
1481
if include_delta_closure:
1482
# XXX: get_content_maps performs its own index queries; allow state
1484
non_local_keys = needed_from_fallback - absent_keys
1485
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1488
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1490
for record in generator.get_record_stream():
1493
for source, keys in source_keys:
1494
if source is parent_maps[0]:
1495
# this KnitVersionedFiles
1496
records = [(key, positions[key][1]) for key in keys]
1497
for key, raw_data, sha1 in self._read_records_iter_raw(records):
1498
(record_details, index_memo, _) = positions[key]
1499
yield KnitContentFactory(key, global_map[key],
1500
record_details, sha1, raw_data, self._factory.annotated, None)
1502
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1503
for record in vf.get_record_stream(keys, ordering,
1504
include_delta_closure):
1507
def get_sha1s(self, keys):
1508
"""See VersionedFiles.get_sha1s()."""
1510
record_map = self._get_record_map(missing, allow_missing=True)
1512
for key, details in record_map.iteritems():
1513
if key not in missing:
1515
# record entry 2 is the 'digest'.
1516
result[key] = details[2]
1517
missing.difference_update(set(result))
1518
for source in self._fallback_vfs:
1521
new_result = source.get_sha1s(missing)
1522
result.update(new_result)
1523
missing.difference_update(set(new_result))
1526
def insert_record_stream(self, stream):
1527
"""Insert a record stream into this container.
1529
:param stream: A stream of records to insert.
1531
:seealso VersionedFiles.get_record_stream:
1533
def get_adapter(adapter_key):
1535
return adapters[adapter_key]
1537
adapter_factory = adapter_registry.get(adapter_key)
1538
adapter = adapter_factory(self)
1539
adapters[adapter_key] = adapter
1542
if self._factory.annotated:
1543
# self is annotated, we need annotated knits to use directly.
1544
annotated = "annotated-"
1547
# self is not annotated, but we can strip annotations cheaply.
1549
convertibles = set(["knit-annotated-ft-gz"])
1550
if self._max_delta_chain:
1551
delta_types.add("knit-annotated-delta-gz")
1552
convertibles.add("knit-annotated-delta-gz")
1553
# The set of types we can cheaply adapt without needing basis texts.
1554
native_types = set()
1555
if self._max_delta_chain:
1556
native_types.add("knit-%sdelta-gz" % annotated)
1557
delta_types.add("knit-%sdelta-gz" % annotated)
1558
native_types.add("knit-%sft-gz" % annotated)
1559
knit_types = native_types.union(convertibles)
1561
# Buffer all index entries that we can't add immediately because their
1562
# basis parent is missing. We don't buffer all because generating
1563
# annotations may require access to some of the new records. However we
1564
# can't generate annotations from new deltas until their basis parent
1565
# is present anyway, so we get away with not needing an index that
1566
# includes the new keys.
1568
# See <http://launchpad.net/bugs/300177> about ordering of compression
1569
# parents in the records - to be conservative, we insist that all
1570
# parents must be present to avoid expanding to a fulltext.
1572
# key = basis_parent, value = index entry to add
1573
buffered_index_entries = {}
1574
for record in stream:
1576
parents = record.parents
1577
if record.storage_kind in delta_types:
1578
# TODO: eventually the record itself should track
1579
# compression_parent
1580
compression_parent = parents[0]
1582
compression_parent = None
1583
# Raise an error when a record is missing.
1584
if record.storage_kind == 'absent':
1585
raise RevisionNotPresent([record.key], self)
1586
elif ((record.storage_kind in knit_types)
1587
and (compression_parent is None
1588
or not self._fallback_vfs
1589
or self._index.has_key(compression_parent)
1590
or not self.has_key(compression_parent))):
1591
# we can insert the knit record literally if either it has no
1592
# compression parent OR we already have its basis in this kvf
1593
# OR the basis is not present even in the fallbacks. In the
1594
# last case it will either turn up later in the stream and all
1595
# will be well, or it won't turn up at all and we'll raise an
1598
# TODO: self.has_key is somewhat redundant with
1599
# self._index.has_key; we really want something that directly
1600
# asks if it's only present in the fallbacks. -- mbp 20081119
1601
if record.storage_kind not in native_types:
1603
adapter_key = (record.storage_kind, "knit-delta-gz")
1604
adapter = get_adapter(adapter_key)
1606
adapter_key = (record.storage_kind, "knit-ft-gz")
1607
adapter = get_adapter(adapter_key)
1608
bytes = adapter.get_bytes(record)
1610
# It's a knit record, it has a _raw_record field (even if
1611
# it was reconstituted from a network stream).
1612
bytes = record._raw_record
1613
options = [record._build_details[0]]
1614
if record._build_details[1]:
1615
options.append('no-eol')
1616
# Just blat it across.
1617
# Note: This does end up adding data on duplicate keys. As
1618
# modern repositories use atomic insertions this should not
1619
# lead to excessive growth in the event of interrupted fetches.
1620
# 'knit' repositories may suffer excessive growth, but as a
1621
# deprecated format this is tolerable. It can be fixed if
1622
# needed by in the kndx index support raising on a duplicate
1623
# add with identical parents and options.
1624
access_memo = self._access.add_raw_records(
1625
[(record.key, len(bytes))], bytes)[0]
1626
index_entry = (record.key, options, access_memo, parents)
1627
if 'fulltext' not in options:
1628
# Not a fulltext, so we need to make sure the compression
1629
# parent will also be present.
1630
# Note that pack backed knits don't need to buffer here
1631
# because they buffer all writes to the transaction level,
1632
# but we don't expose that difference at the index level. If
1633
# the query here has sufficient cost to show up in
1634
# profiling we should do that.
1636
# They're required to be physically in this
1637
# KnitVersionedFiles, not in a fallback.
1638
if not self._index.has_key(compression_parent):
1639
pending = buffered_index_entries.setdefault(
1640
compression_parent, [])
1641
pending.append(index_entry)
1644
self._index.add_records([index_entry])
1645
elif record.storage_kind == 'chunked':
1646
self.add_lines(record.key, parents,
1647
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1649
# Not suitable for direct insertion as a
1650
# delta, either because it's not the right format, or this
1651
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1652
# 0) or because it depends on a base only present in the
1654
self._access.flush()
1656
# Try getting a fulltext directly from the record.
1657
bytes = record.get_bytes_as('fulltext')
1658
except errors.UnavailableRepresentation:
1659
adapter_key = record.storage_kind, 'fulltext'
1660
adapter = get_adapter(adapter_key)
1661
bytes = adapter.get_bytes(record)
1662
lines = split_lines(bytes)
1664
self.add_lines(record.key, parents, lines)
1665
except errors.RevisionAlreadyPresent:
1667
# Add any records whose basis parent is now available.
1669
added_keys = [record.key]
1671
key = added_keys.pop(0)
1672
if key in buffered_index_entries:
1673
index_entries = buffered_index_entries[key]
1674
self._index.add_records(index_entries)
1676
[index_entry[0] for index_entry in index_entries])
1677
del buffered_index_entries[key]
1678
if buffered_index_entries:
1679
# There were index entries buffered at the end of the stream,
1680
# So these need to be added (if the index supports holding such
1681
# entries for later insertion)
1682
for key in buffered_index_entries:
1683
index_entries = buffered_index_entries[key]
1684
self._index.add_records(index_entries,
1685
missing_compression_parents=True)
1687
def get_missing_compression_parent_keys(self):
1688
"""Return an iterable of keys of missing compression parents.
1690
Check this after calling insert_record_stream to find out if there are
1691
any missing compression parents. If there are, the records that
1692
depend on them are not able to be inserted safely. For atomic
1693
KnitVersionedFiles built on packs, the transaction should be aborted or
1694
suspended - commit will fail at this point. Nonatomic knits will error
1695
earlier because they have no staging area to put pending entries into.
1697
return self._index.get_missing_compression_parents()
1699
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1700
"""Iterate over the lines in the versioned files from keys.
1702
This may return lines from other keys. Each item the returned
1703
iterator yields is a tuple of a line and a text version that that line
1704
is present in (not introduced in).
1706
Ordering of results is in whatever order is most suitable for the
1707
underlying storage format.
1709
If a progress bar is supplied, it may be used to indicate progress.
1710
The caller is responsible for cleaning up progress bars (because this
1714
* Lines are normalised by the underlying store: they will all have \\n
1716
* Lines are returned in arbitrary order.
1717
* If a requested key did not change any lines (or didn't have any
1718
lines), it may not be mentioned at all in the result.
1720
:param pb: Progress bar supplied by caller.
1721
:return: An iterator over (line, key).
1724
pb = progress.DummyProgress()
1730
# we don't care about inclusions, the caller cares.
1731
# but we need to setup a list of records to visit.
1732
# we need key, position, length
1734
build_details = self._index.get_build_details(keys)
1735
for key, details in build_details.iteritems():
1737
key_records.append((key, details[0]))
1738
records_iter = enumerate(self._read_records_iter(key_records))
1739
for (key_idx, (key, data, sha_value)) in records_iter:
1740
pb.update('Walking content', key_idx, total)
1741
compression_parent = build_details[key][1]
1742
if compression_parent is None:
1744
line_iterator = self._factory.get_fulltext_content(data)
1747
line_iterator = self._factory.get_linedelta_content(data)
1748
# Now that we are yielding the data for this key, remove it
1751
# XXX: It might be more efficient to yield (key,
1752
# line_iterator) in the future. However for now, this is a
1753
# simpler change to integrate into the rest of the
1754
# codebase. RBC 20071110
1755
for line in line_iterator:
1758
except errors.RetryWithNewPacks, e:
1759
self._access.reload_or_raise(e)
1760
# If there are still keys we've not yet found, we look in the fallback
1761
# vfs, and hope to find them there. Note that if the keys are found
1762
# but had no changes or no content, the fallback may not return
1764
if keys and not self._fallback_vfs:
1765
# XXX: strictly the second parameter is meant to be the file id
1766
# but it's not easily accessible here.
1767
raise RevisionNotPresent(keys, repr(self))
1768
for source in self._fallback_vfs:
1772
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1773
source_keys.add(key)
1775
keys.difference_update(source_keys)
1776
pb.update('Walking content', total, total)
1778
def _make_line_delta(self, delta_seq, new_content):
1779
"""Generate a line delta from delta_seq and new_content."""
1781
for op in delta_seq.get_opcodes():
1782
if op[0] == 'equal':
1784
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1787
def _merge_annotations(self, content, parents, parent_texts={},
1788
delta=None, annotated=None,
1789
left_matching_blocks=None):
1790
"""Merge annotations for content and generate deltas.
1792
This is done by comparing the annotations based on changes to the text
1793
and generating a delta on the resulting full texts. If annotations are
1794
not being created then a simple delta is created.
1796
if left_matching_blocks is not None:
1797
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1801
for parent_key in parents:
1802
merge_content = self._get_content(parent_key, parent_texts)
1803
if (parent_key == parents[0] and delta_seq is not None):
1806
seq = patiencediff.PatienceSequenceMatcher(
1807
None, merge_content.text(), content.text())
1808
for i, j, n in seq.get_matching_blocks():
1811
# this copies (origin, text) pairs across to the new
1812
# content for any line that matches the last-checked
1814
content._lines[j:j+n] = merge_content._lines[i:i+n]
1815
# XXX: Robert says the following block is a workaround for a
1816
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1817
if content._lines and content._lines[-1][1][-1] != '\n':
1818
# The copied annotation was from a line without a trailing EOL,
1819
# reinstate one for the content object, to ensure correct
1821
line = content._lines[-1][1] + '\n'
1822
content._lines[-1] = (content._lines[-1][0], line)
1824
if delta_seq is None:
1825
reference_content = self._get_content(parents[0], parent_texts)
1826
new_texts = content.text()
1827
old_texts = reference_content.text()
1828
delta_seq = patiencediff.PatienceSequenceMatcher(
1829
None, old_texts, new_texts)
1830
return self._make_line_delta(delta_seq, content)
1832
def _parse_record(self, version_id, data):
1833
"""Parse an original format knit record.
1835
These have the last element of the key only present in the stored data.
1837
rec, record_contents = self._parse_record_unchecked(data)
1838
self._check_header_version(rec, version_id)
1839
return record_contents, rec[3]
1841
def _parse_record_header(self, key, raw_data):
1842
"""Parse a record header for consistency.
1844
:return: the header and the decompressor stream.
1845
as (stream, header_record)
1847
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1850
rec = self._check_header(key, df.readline())
1851
except Exception, e:
1852
raise KnitCorrupt(self,
1853
"While reading {%s} got %s(%s)"
1854
% (key, e.__class__.__name__, str(e)))
1857
def _parse_record_unchecked(self, data):
1859
# 4168 calls in 2880 217 internal
1860
# 4168 calls to _parse_record_header in 2121
1861
# 4168 calls to readlines in 330
1862
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1864
record_contents = df.readlines()
1865
except Exception, e:
1866
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1867
(data, e.__class__.__name__, str(e)))
1868
header = record_contents.pop(0)
1869
rec = self._split_header(header)
1870
last_line = record_contents.pop()
1871
if len(record_contents) != int(rec[2]):
1872
raise KnitCorrupt(self,
1873
'incorrect number of lines %s != %s'
1874
' for version {%s} %s'
1875
% (len(record_contents), int(rec[2]),
1876
rec[1], record_contents))
1877
if last_line != 'end %s\n' % rec[1]:
1878
raise KnitCorrupt(self,
1879
'unexpected version end line %r, wanted %r'
1880
% (last_line, rec[1]))
1882
return rec, record_contents
1884
def _read_records_iter(self, records):
1885
"""Read text records from data file and yield result.
1887
The result will be returned in whatever is the fastest to read.
1888
Not by the order requested. Also, multiple requests for the same
1889
record will only yield 1 response.
1890
:param records: A list of (key, access_memo) entries
1891
:return: Yields (key, contents, digest) in the order
1892
read, not the order requested
1897
# XXX: This smells wrong, IO may not be getting ordered right.
1898
needed_records = sorted(set(records), key=operator.itemgetter(1))
1899
if not needed_records:
1902
# The transport optimizes the fetching as well
1903
# (ie, reads continuous ranges.)
1904
raw_data = self._access.get_raw_records(
1905
[index_memo for key, index_memo in needed_records])
1907
for (key, index_memo), data in \
1908
izip(iter(needed_records), raw_data):
1909
content, digest = self._parse_record(key[-1], data)
1910
yield key, content, digest
1912
def _read_records_iter_raw(self, records):
1913
"""Read text records from data file and yield raw data.
1915
This unpacks enough of the text record to validate the id is
1916
as expected but thats all.
1918
Each item the iterator yields is (key, bytes,
1919
expected_sha1_of_full_text).
1921
for key, data in self._read_records_iter_unchecked(records):
1922
# validate the header (note that we can only use the suffix in
1923
# current knit records).
1924
df, rec = self._parse_record_header(key, data)
1926
yield key, data, rec[3]
1928
def _read_records_iter_unchecked(self, records):
1929
"""Read text records from data file and yield raw data.
1931
No validation is done.
1933
Yields tuples of (key, data).
1935
# setup an iterator of the external records:
1936
# uses readv so nice and fast we hope.
1938
# grab the disk data needed.
1939
needed_offsets = [index_memo for key, index_memo
1941
raw_records = self._access.get_raw_records(needed_offsets)
1943
for key, index_memo in records:
1944
data = raw_records.next()
1947
def _record_to_data(self, key, digest, lines, dense_lines=None):
1948
"""Convert key, digest, lines into a raw data block.
1950
:param key: The key of the record. Currently keys are always serialised
1951
using just the trailing component.
1952
:param dense_lines: The bytes of lines but in a denser form. For
1953
instance, if lines is a list of 1000 bytestrings each ending in \n,
1954
dense_lines may be a list with one line in it, containing all the
1955
1000's lines and their \n's. Using dense_lines if it is already
1956
known is a win because the string join to create bytes in this
1957
function spends less time resizing the final string.
1958
:return: (len, a StringIO instance with the raw data ready to read.)
1960
chunks = ["version %s %d %s\n" % (key[-1], len(lines), digest)]
1961
chunks.extend(dense_lines or lines)
1962
chunks.append("end %s\n" % key[-1])
1963
for chunk in chunks:
1964
if type(chunk) is not str:
1965
raise AssertionError(
1966
'data must be plain bytes was %s' % type(chunk))
1967
if lines and lines[-1][-1] != '\n':
1968
raise ValueError('corrupt lines value %r' % lines)
1969
compressed_bytes = tuned_gzip.chunks_to_gzip(chunks)
1970
return len(compressed_bytes), compressed_bytes
1972
def _split_header(self, line):
1975
raise KnitCorrupt(self,
1976
'unexpected number of elements in record header')
1980
"""See VersionedFiles.keys."""
1981
if 'evil' in debug.debug_flags:
1982
trace.mutter_callsite(2, "keys scales with size of history")
1983
sources = [self._index] + self._fallback_vfs
1985
for source in sources:
1986
result.update(source.keys())
1990
class _ContentMapGenerator(object):
1991
"""Generate texts or expose raw deltas for a set of texts."""
1993
def _get_content(self, key):
1994
"""Get the content object for key."""
1995
# Note that _get_content is only called when the _ContentMapGenerator
1996
# has been constructed with just one key requested for reconstruction.
1997
if key in self.nonlocal_keys:
1998
record = self.get_record_stream().next()
1999
# Create a content object on the fly
2000
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
2001
return PlainKnitContent(lines, record.key)
2003
# local keys we can ask for directly
2004
return self._get_one_work(key)
2006
def get_record_stream(self):
2007
"""Get a record stream for the keys requested during __init__."""
2008
for record in self._work():
2012
"""Produce maps of text and KnitContents as dicts.
2014
:return: (text_map, content_map) where text_map contains the texts for
2015
the requested versions and content_map contains the KnitContents.
2017
# NB: By definition we never need to read remote sources unless texts
2018
# are requested from them: we don't delta across stores - and we
2019
# explicitly do not want to to prevent data loss situations.
2020
if self.global_map is None:
2021
self.global_map = self.vf.get_parent_map(self.keys)
2022
nonlocal_keys = self.nonlocal_keys
2024
missing_keys = set(nonlocal_keys)
2025
# Read from remote versioned file instances and provide to our caller.
2026
for source in self.vf._fallback_vfs:
2027
if not missing_keys:
2029
# Loop over fallback repositories asking them for texts - ignore
2030
# any missing from a particular fallback.
2031
for record in source.get_record_stream(missing_keys,
2033
if record.storage_kind == 'absent':
2034
# Not in thie particular stream, may be in one of the
2035
# other fallback vfs objects.
2037
missing_keys.remove(record.key)
2040
if self._raw_record_map is None:
2041
raise AssertionError('_raw_record_map should have been filled')
2043
for key in self.keys:
2044
if key in self.nonlocal_keys:
2046
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2049
def _get_one_work(self, requested_key):
2050
# Now, if we have calculated everything already, just return the
2052
if requested_key in self._contents_map:
2053
return self._contents_map[requested_key]
2054
# To simplify things, parse everything at once - code that wants one text
2055
# probably wants them all.
2056
# FUTURE: This function could be improved for the 'extract many' case
2057
# by tracking each component and only doing the copy when the number of
2058
# children than need to apply delta's to it is > 1 or it is part of the
2060
multiple_versions = len(self.keys) != 1
2061
if self._record_map is None:
2062
self._record_map = self.vf._raw_map_to_record_map(
2063
self._raw_record_map)
2064
record_map = self._record_map
2065
# raw_record_map is key:
2066
# Have read and parsed records at this point.
2067
for key in self.keys:
2068
if key in self.nonlocal_keys:
2073
while cursor is not None:
2075
record, record_details, digest, next = record_map[cursor]
2077
raise RevisionNotPresent(cursor, self)
2078
components.append((cursor, record, record_details, digest))
2080
if cursor in self._contents_map:
2081
# no need to plan further back
2082
components.append((cursor, None, None, None))
2086
for (component_id, record, record_details,
2087
digest) in reversed(components):
2088
if component_id in self._contents_map:
2089
content = self._contents_map[component_id]
2091
content, delta = self._factory.parse_record(key[-1],
2092
record, record_details, content,
2093
copy_base_content=multiple_versions)
2094
if multiple_versions:
2095
self._contents_map[component_id] = content
2097
# digest here is the digest from the last applied component.
2098
text = content.text()
2099
actual_sha = sha_strings(text)
2100
if actual_sha != digest:
2101
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2102
if multiple_versions:
2103
return self._contents_map[requested_key]
2107
def _wire_bytes(self):
2108
"""Get the bytes to put on the wire for 'key'.
2110
The first collection of bytes asked for returns the serialised
2111
raw_record_map and the additional details (key, parent) for key.
2112
Subsequent calls return just the additional details (key, parent).
2113
The wire storage_kind given for the first key is 'knit-delta-closure',
2114
For subsequent keys it is 'knit-delta-closure-ref'.
2116
:param key: A key from the content generator.
2117
:return: Bytes to put on the wire.
2120
# kind marker for dispatch on the far side,
2121
lines.append('knit-delta-closure')
2123
if self.vf._factory.annotated:
2124
lines.append('annotated')
2127
# then the list of keys
2128
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2129
if key not in self.nonlocal_keys]))
2130
# then the _raw_record_map in serialised form:
2132
# for each item in the map:
2134
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2135
# one line with method
2136
# one line with noeol
2137
# one line with next ('' for None)
2138
# one line with byte count of the record bytes
2140
for key, (record_bytes, (method, noeol), next) in \
2141
self._raw_record_map.iteritems():
2142
key_bytes = '\x00'.join(key)
2143
parents = self.global_map.get(key, None)
2145
parent_bytes = 'None:'
2147
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2148
method_bytes = method
2154
next_bytes = '\x00'.join(next)
2157
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2158
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2159
len(record_bytes), record_bytes))
2160
map_bytes = ''.join(map_byte_list)
2161
lines.append(map_bytes)
2162
bytes = '\n'.join(lines)
2166
class _VFContentMapGenerator(_ContentMapGenerator):
2167
"""Content map generator reading from a VersionedFiles object."""
2169
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2170
global_map=None, raw_record_map=None):
2171
"""Create a _ContentMapGenerator.
2173
:param versioned_files: The versioned files that the texts are being
2175
:param keys: The keys to produce content maps for.
2176
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2177
which are known to not be in this knit, but rather in one of the
2179
:param global_map: The result of get_parent_map(keys) (or a supermap).
2180
This is required if get_record_stream() is to be used.
2181
:param raw_record_map: A unparsed raw record map to use for answering
2184
# The vf to source data from
2185
self.vf = versioned_files
2187
self.keys = list(keys)
2188
# Keys known to be in fallback vfs objects
2189
if nonlocal_keys is None:
2190
self.nonlocal_keys = set()
2192
self.nonlocal_keys = frozenset(nonlocal_keys)
2193
# Parents data for keys to be returned in get_record_stream
2194
self.global_map = global_map
2195
# The chunked lists for self.keys in text form
2197
# A cache of KnitContent objects used in extracting texts.
2198
self._contents_map = {}
2199
# All the knit records needed to assemble the requested keys as full
2201
self._record_map = None
2202
if raw_record_map is None:
2203
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2206
self._raw_record_map = raw_record_map
2207
# the factory for parsing records
2208
self._factory = self.vf._factory
2211
class _NetworkContentMapGenerator(_ContentMapGenerator):
2212
"""Content map generator sourced from a network stream."""
2214
def __init__(self, bytes, line_end):
2215
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2217
self.global_map = {}
2218
self._raw_record_map = {}
2219
self._contents_map = {}
2220
self._record_map = None
2221
self.nonlocal_keys = []
2222
# Get access to record parsing facilities
2223
self.vf = KnitVersionedFiles(None, None)
2226
line_end = bytes.find('\n', start)
2227
line = bytes[start:line_end]
2228
start = line_end + 1
2229
if line == 'annotated':
2230
self._factory = KnitAnnotateFactory()
2232
self._factory = KnitPlainFactory()
2233
# list of keys to emit in get_record_stream
2234
line_end = bytes.find('\n', start)
2235
line = bytes[start:line_end]
2236
start = line_end + 1
2238
tuple(segment.split('\x00')) for segment in line.split('\t')
2240
# now a loop until the end. XXX: It would be nice if this was just a
2241
# bunch of the same records as get_record_stream(..., False) gives, but
2242
# there is a decent sized gap stopping that at the moment.
2246
line_end = bytes.find('\n', start)
2247
key = tuple(bytes[start:line_end].split('\x00'))
2248
start = line_end + 1
2249
# 1 line with parents (None: for None, '' for ())
2250
line_end = bytes.find('\n', start)
2251
line = bytes[start:line_end]
2256
[tuple(segment.split('\x00')) for segment in line.split('\t')
2258
self.global_map[key] = parents
2259
start = line_end + 1
2260
# one line with method
2261
line_end = bytes.find('\n', start)
2262
line = bytes[start:line_end]
2264
start = line_end + 1
2265
# one line with noeol
2266
line_end = bytes.find('\n', start)
2267
line = bytes[start:line_end]
2269
start = line_end + 1
2270
# one line with next ('' for None)
2271
line_end = bytes.find('\n', start)
2272
line = bytes[start:line_end]
2276
next = tuple(bytes[start:line_end].split('\x00'))
2277
start = line_end + 1
2278
# one line with byte count of the record bytes
2279
line_end = bytes.find('\n', start)
2280
line = bytes[start:line_end]
2282
start = line_end + 1
2284
record_bytes = bytes[start:start+count]
2285
start = start + count
2287
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2289
def get_record_stream(self):
2290
"""Get a record stream for for keys requested by the bytestream."""
2292
for key in self.keys:
2293
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2296
def _wire_bytes(self):
2300
class _KndxIndex(object):
2301
"""Manages knit index files
2303
The index is kept in memory and read on startup, to enable
2304
fast lookups of revision information. The cursor of the index
2305
file is always pointing to the end, making it easy to append
2308
_cache is a cache for fast mapping from version id to a Index
2311
_history is a cache for fast mapping from indexes to version ids.
2313
The index data format is dictionary compressed when it comes to
2314
parent references; a index entry may only have parents that with a
2315
lover index number. As a result, the index is topological sorted.
2317
Duplicate entries may be written to the index for a single version id
2318
if this is done then the latter one completely replaces the former:
2319
this allows updates to correct version and parent information.
2320
Note that the two entries may share the delta, and that successive
2321
annotations and references MUST point to the first entry.
2323
The index file on disc contains a header, followed by one line per knit
2324
record. The same revision can be present in an index file more than once.
2325
The first occurrence gets assigned a sequence number starting from 0.
2327
The format of a single line is
2328
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
2329
REVISION_ID is a utf8-encoded revision id
2330
FLAGS is a comma separated list of flags about the record. Values include
2331
no-eol, line-delta, fulltext.
2332
BYTE_OFFSET is the ascii representation of the byte offset in the data file
2333
that the the compressed data starts at.
2334
LENGTH is the ascii representation of the length of the data file.
2335
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
2337
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
2338
revision id already in the knit that is a parent of REVISION_ID.
2339
The ' :' marker is the end of record marker.
2342
when a write is interrupted to the index file, it will result in a line
2343
that does not end in ' :'. If the ' :' is not present at the end of a line,
2344
or at the end of the file, then the record that is missing it will be
2345
ignored by the parser.
2347
When writing new records to the index file, the data is preceded by '\n'
2348
to ensure that records always start on new lines even if the last write was
2349
interrupted. As a result its normal for the last line in the index to be
2350
missing a trailing newline. One can be added with no harmful effects.
2352
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2353
where prefix is e.g. the (fileid,) for .texts instances or () for
2354
constant-mapped things like .revisions, and the old state is
2355
tuple(history_vector, cache_dict). This is used to prevent having an
2356
ABI change with the C extension that reads .kndx files.
2359
HEADER = "# bzr knit index 8\n"
2361
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2362
"""Create a _KndxIndex on transport using mapper."""
2363
self._transport = transport
2364
self._mapper = mapper
2365
self._get_scope = get_scope
2366
self._allow_writes = allow_writes
2367
self._is_locked = is_locked
2369
self.has_graph = True
2371
def add_records(self, records, random_id=False, missing_compression_parents=False):
2372
"""Add multiple records to the index.
2374
:param records: a list of tuples:
2375
(key, options, access_memo, parents).
2376
:param random_id: If True the ids being added were randomly generated
2377
and no check for existence will be performed.
2378
:param missing_compression_parents: If True the records being added are
2379
only compressed against texts already in the index (or inside
2380
records). If False the records all refer to unavailable texts (or
2381
texts inside records) as compression parents.
2383
if missing_compression_parents:
2384
# It might be nice to get the edge of the records. But keys isn't
2386
keys = sorted(record[0] for record in records)
2387
raise errors.RevisionNotPresent(keys, self)
2389
for record in records:
2392
path = self._mapper.map(key) + '.kndx'
2393
path_keys = paths.setdefault(path, (prefix, []))
2394
path_keys[1].append(record)
2395
for path in sorted(paths):
2396
prefix, path_keys = paths[path]
2397
self._load_prefixes([prefix])
2399
orig_history = self._kndx_cache[prefix][1][:]
2400
orig_cache = self._kndx_cache[prefix][0].copy()
2403
for key, options, (_, pos, size), parents in path_keys:
2405
# kndx indices cannot be parentless.
2407
line = "\n%s %s %s %s %s :" % (
2408
key[-1], ','.join(options), pos, size,
2409
self._dictionary_compress(parents))
2410
if type(line) is not str:
2411
raise AssertionError(
2412
'data must be utf8 was %s' % type(line))
2414
self._cache_key(key, options, pos, size, parents)
2415
if len(orig_history):
2416
self._transport.append_bytes(path, ''.join(lines))
2418
self._init_index(path, lines)
2420
# If any problems happen, restore the original values and re-raise
2421
self._kndx_cache[prefix] = (orig_cache, orig_history)
2424
def scan_unvalidated_index(self, graph_index):
2425
"""See _KnitGraphIndex.scan_unvalidated_index."""
2426
# Because kndx files do not support atomic insertion via separate index
2427
# files, they do not support this method.
2428
raise NotImplementedError(self.scan_unvalidated_index)
2430
def get_missing_compression_parents(self):
2431
"""See _KnitGraphIndex.get_missing_compression_parents."""
2432
# Because kndx files do not support atomic insertion via separate index
2433
# files, they do not support this method.
2434
raise NotImplementedError(self.get_missing_compression_parents)
2436
def _cache_key(self, key, options, pos, size, parent_keys):
2437
"""Cache a version record in the history array and index cache.
2439
This is inlined into _load_data for performance. KEEP IN SYNC.
2440
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2444
version_id = key[-1]
2445
# last-element only for compatibilty with the C load_data.
2446
parents = tuple(parent[-1] for parent in parent_keys)
2447
for parent in parent_keys:
2448
if parent[:-1] != prefix:
2449
raise ValueError("mismatched prefixes for %r, %r" % (
2451
cache, history = self._kndx_cache[prefix]
2452
# only want the _history index to reference the 1st index entry
2454
if version_id not in cache:
2455
index = len(history)
2456
history.append(version_id)
2458
index = cache[version_id][5]
2459
cache[version_id] = (version_id,
2466
def check_header(self, fp):
2467
line = fp.readline()
2469
# An empty file can actually be treated as though the file doesn't
2471
raise errors.NoSuchFile(self)
2472
if line != self.HEADER:
2473
raise KnitHeaderError(badline=line, filename=self)
2475
def _check_read(self):
2476
if not self._is_locked():
2477
raise errors.ObjectNotLocked(self)
2478
if self._get_scope() != self._scope:
2481
def _check_write_ok(self):
2482
"""Assert if not writes are permitted."""
2483
if not self._is_locked():
2484
raise errors.ObjectNotLocked(self)
2485
if self._get_scope() != self._scope:
2487
if self._mode != 'w':
2488
raise errors.ReadOnlyObjectDirtiedError(self)
2490
def get_build_details(self, keys):
2491
"""Get the method, index_memo and compression parent for keys.
2493
Ghosts are omitted from the result.
2495
:param keys: An iterable of keys.
2496
:return: A dict of key:(index_memo, compression_parent, parents,
2499
opaque structure to pass to read_records to extract the raw
2502
Content that this record is built upon, may be None
2504
Logical parents of this node
2506
extra information about the content which needs to be passed to
2507
Factory.parse_record
2509
parent_map = self.get_parent_map(keys)
2512
if key not in parent_map:
2514
method = self.get_method(key)
2515
parents = parent_map[key]
2516
if method == 'fulltext':
2517
compression_parent = None
2519
compression_parent = parents[0]
2520
noeol = 'no-eol' in self.get_options(key)
2521
index_memo = self.get_position(key)
2522
result[key] = (index_memo, compression_parent,
2523
parents, (method, noeol))
2526
def get_method(self, key):
2527
"""Return compression method of specified key."""
2528
options = self.get_options(key)
2529
if 'fulltext' in options:
2531
elif 'line-delta' in options:
2534
raise errors.KnitIndexUnknownMethod(self, options)
2536
def get_options(self, key):
2537
"""Return a list representing options.
2541
prefix, suffix = self._split_key(key)
2542
self._load_prefixes([prefix])
2544
return self._kndx_cache[prefix][0][suffix][1]
2546
raise RevisionNotPresent(key, self)
2548
def get_parent_map(self, keys):
2549
"""Get a map of the parents of keys.
2551
:param keys: The keys to look up parents for.
2552
:return: A mapping from keys to parents. Absent keys are absent from
2555
# Parse what we need to up front, this potentially trades off I/O
2556
# locality (.kndx and .knit in the same block group for the same file
2557
# id) for less checking in inner loops.
2558
prefixes = set(key[:-1] for key in keys)
2559
self._load_prefixes(prefixes)
2564
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2568
result[key] = tuple(prefix + (suffix,) for
2569
suffix in suffix_parents)
2572
def get_position(self, key):
2573
"""Return details needed to access the version.
2575
:return: a tuple (key, data position, size) to hand to the access
2576
logic to get the record.
2578
prefix, suffix = self._split_key(key)
2579
self._load_prefixes([prefix])
2580
entry = self._kndx_cache[prefix][0][suffix]
2581
return key, entry[2], entry[3]
2583
has_key = _mod_index._has_key_from_parent_map
2585
def _init_index(self, path, extra_lines=[]):
2586
"""Initialize an index."""
2588
sio.write(self.HEADER)
2589
sio.writelines(extra_lines)
2591
self._transport.put_file_non_atomic(path, sio,
2592
create_parent_dir=True)
2593
# self._create_parent_dir)
2594
# mode=self._file_mode,
2595
# dir_mode=self._dir_mode)
2598
"""Get all the keys in the collection.
2600
The keys are not ordered.
2603
# Identify all key prefixes.
2604
# XXX: A bit hacky, needs polish.
2605
if type(self._mapper) is ConstantMapper:
2609
for quoted_relpath in self._transport.iter_files_recursive():
2610
path, ext = os.path.splitext(quoted_relpath)
2612
prefixes = [self._mapper.unmap(path) for path in relpaths]
2613
self._load_prefixes(prefixes)
2614
for prefix in prefixes:
2615
for suffix in self._kndx_cache[prefix][1]:
2616
result.add(prefix + (suffix,))
2619
def _load_prefixes(self, prefixes):
2620
"""Load the indices for prefixes."""
2622
for prefix in prefixes:
2623
if prefix not in self._kndx_cache:
2624
# the load_data interface writes to these variables.
2627
self._filename = prefix
2629
path = self._mapper.map(prefix) + '.kndx'
2630
fp = self._transport.get(path)
2632
# _load_data may raise NoSuchFile if the target knit is
2634
_load_data(self, fp)
2637
self._kndx_cache[prefix] = (self._cache, self._history)
2642
self._kndx_cache[prefix] = ({}, [])
2643
if type(self._mapper) is ConstantMapper:
2644
# preserve behaviour for revisions.kndx etc.
2645
self._init_index(path)
2650
missing_keys = _mod_index._missing_keys_from_parent_map
2652
def _partition_keys(self, keys):
2653
"""Turn keys into a dict of prefix:suffix_list."""
2656
prefix_keys = result.setdefault(key[:-1], [])
2657
prefix_keys.append(key[-1])
2660
def _dictionary_compress(self, keys):
2661
"""Dictionary compress keys.
2663
:param keys: The keys to generate references to.
2664
:return: A string representation of keys. keys which are present are
2665
dictionary compressed, and others are emitted as fulltext with a
2671
prefix = keys[0][:-1]
2672
cache = self._kndx_cache[prefix][0]
2674
if key[:-1] != prefix:
2675
# kndx indices cannot refer across partitioned storage.
2676
raise ValueError("mismatched prefixes for %r" % keys)
2677
if key[-1] in cache:
2678
# -- inlined lookup() --
2679
result_list.append(str(cache[key[-1]][5]))
2680
# -- end lookup () --
2682
result_list.append('.' + key[-1])
2683
return ' '.join(result_list)
2685
def _reset_cache(self):
2686
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2687
# (cache_dict, history_vector) for parsed kndx files.
2688
self._kndx_cache = {}
2689
self._scope = self._get_scope()
2690
allow_writes = self._allow_writes()
2696
def _sort_keys_by_io(self, keys, positions):
2697
"""Figure out an optimal order to read the records for the given keys.
2699
Sort keys, grouped by index and sorted by position.
2701
:param keys: A list of keys whose records we want to read. This will be
2703
:param positions: A dict, such as the one returned by
2704
_get_components_positions()
2707
def get_sort_key(key):
2708
index_memo = positions[key][1]
2709
# Group by prefix and position. index_memo[0] is the key, so it is
2710
# (file_id, revision_id) and we don't want to sort on revision_id,
2711
# index_memo[1] is the position, and index_memo[2] is the size,
2712
# which doesn't matter for the sort
2713
return index_memo[0][:-1], index_memo[1]
2714
return keys.sort(key=get_sort_key)
2716
_get_total_build_size = _get_total_build_size
2718
def _split_key(self, key):
2719
"""Split key into a prefix and suffix."""
2720
return key[:-1], key[-1]
2723
class _KeyRefs(object):
2726
# dict mapping 'key' to 'set of keys referring to that key'
2729
def add_references(self, key, refs):
2730
# Record the new references
2731
for referenced in refs:
2733
needed_by = self.refs[referenced]
2735
needed_by = self.refs[referenced] = set()
2737
# Discard references satisfied by the new key
2740
def get_unsatisfied_refs(self):
2741
return self.refs.iterkeys()
2743
def add_key(self, key):
2747
# No keys depended on this key. That's ok.
2750
def add_keys(self, keys):
2754
def get_referrers(self):
2756
for referrers in self.refs.itervalues():
2757
result.update(referrers)
2761
class _KnitGraphIndex(object):
2762
"""A KnitVersionedFiles index layered on GraphIndex."""
2764
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2765
add_callback=None, track_external_parent_refs=False):
2766
"""Construct a KnitGraphIndex on a graph_index.
2768
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2769
:param is_locked: A callback to check whether the object should answer
2771
:param deltas: Allow delta-compressed records.
2772
:param parents: If True, record knits parents, if not do not record
2774
:param add_callback: If not None, allow additions to the index and call
2775
this callback with a list of added GraphIndex nodes:
2776
[(node, value, node_refs), ...]
2777
:param is_locked: A callback, returns True if the index is locked and
2779
:param track_external_parent_refs: If True, record all external parent
2780
references parents from added records. These can be retrieved
2781
later by calling get_missing_parents().
2783
self._add_callback = add_callback
2784
self._graph_index = graph_index
2785
self._deltas = deltas
2786
self._parents = parents
2787
if deltas and not parents:
2788
# XXX: TODO: Delta tree and parent graph should be conceptually
2790
raise KnitCorrupt(self, "Cannot do delta compression without "
2792
self.has_graph = parents
2793
self._is_locked = is_locked
2794
self._missing_compression_parents = set()
2795
if track_external_parent_refs:
2796
self._key_dependencies = _KeyRefs()
2798
self._key_dependencies = None
2801
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2803
def add_records(self, records, random_id=False,
2804
missing_compression_parents=False):
2805
"""Add multiple records to the index.
2807
This function does not insert data into the Immutable GraphIndex
2808
backing the KnitGraphIndex, instead it prepares data for insertion by
2809
the caller and checks that it is safe to insert then calls
2810
self._add_callback with the prepared GraphIndex nodes.
2812
:param records: a list of tuples:
2813
(key, options, access_memo, parents).
2814
:param random_id: If True the ids being added were randomly generated
2815
and no check for existence will be performed.
2816
:param missing_compression_parents: If True the records being added are
2817
only compressed against texts already in the index (or inside
2818
records). If False the records all refer to unavailable texts (or
2819
texts inside records) as compression parents.
2821
if not self._add_callback:
2822
raise errors.ReadOnlyError(self)
2823
# we hope there are no repositories with inconsistent parentage
2827
compression_parents = set()
2828
key_dependencies = self._key_dependencies
2829
for (key, options, access_memo, parents) in records:
2831
parents = tuple(parents)
2832
if key_dependencies is not None:
2833
key_dependencies.add_references(key, parents)
2834
index, pos, size = access_memo
2835
if 'no-eol' in options:
2839
value += "%d %d" % (pos, size)
2840
if not self._deltas:
2841
if 'line-delta' in options:
2842
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2845
if 'line-delta' in options:
2846
node_refs = (parents, (parents[0],))
2847
if missing_compression_parents:
2848
compression_parents.add(parents[0])
2850
node_refs = (parents, ())
2852
node_refs = (parents, )
2855
raise KnitCorrupt(self, "attempt to add node with parents "
2856
"in parentless index.")
2858
keys[key] = (value, node_refs)
2861
present_nodes = self._get_entries(keys)
2862
for (index, key, value, node_refs) in present_nodes:
2863
if (value[0] != keys[key][0][0] or
2864
node_refs[:1] != keys[key][1][:1]):
2865
raise KnitCorrupt(self, "inconsistent details in add_records"
2866
": %s %s" % ((value, node_refs), keys[key]))
2870
for key, (value, node_refs) in keys.iteritems():
2871
result.append((key, value, node_refs))
2873
for key, (value, node_refs) in keys.iteritems():
2874
result.append((key, value))
2875
self._add_callback(result)
2876
if missing_compression_parents:
2877
# This may appear to be incorrect (it does not check for
2878
# compression parents that are in the existing graph index),
2879
# but such records won't have been buffered, so this is
2880
# actually correct: every entry when
2881
# missing_compression_parents==True either has a missing parent, or
2882
# a parent that is one of the keys in records.
2883
compression_parents.difference_update(keys)
2884
self._missing_compression_parents.update(compression_parents)
2885
# Adding records may have satisfied missing compression parents.
2886
self._missing_compression_parents.difference_update(keys)
2888
def scan_unvalidated_index(self, graph_index):
2889
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2891
This allows this _KnitGraphIndex to keep track of any missing
2892
compression parents we may want to have filled in to make those
2895
:param graph_index: A GraphIndex
2898
new_missing = graph_index.external_references(ref_list_num=1)
2899
new_missing.difference_update(self.get_parent_map(new_missing))
2900
self._missing_compression_parents.update(new_missing)
2901
if self._key_dependencies is not None:
2902
# Add parent refs from graph_index (and discard parent refs that
2903
# the graph_index has).
2904
for node in graph_index.iter_all_entries():
2905
self._key_dependencies.add_references(node[1], node[3][0])
2907
def get_missing_compression_parents(self):
2908
"""Return the keys of missing compression parents.
2910
Missing compression parents occur when a record stream was missing
2911
basis texts, or a index was scanned that had missing basis texts.
2913
return frozenset(self._missing_compression_parents)
2915
def get_missing_parents(self):
2916
"""Return the keys of missing parents."""
2917
# If updating this, you should also update
2918
# groupcompress._GCGraphIndex.get_missing_parents
2919
# We may have false positives, so filter those out.
2920
self._key_dependencies.add_keys(
2921
self.get_parent_map(self._key_dependencies.get_unsatisfied_refs()))
2922
return frozenset(self._key_dependencies.get_unsatisfied_refs())
2924
def _check_read(self):
2925
"""raise if reads are not permitted."""
2926
if not self._is_locked():
2927
raise errors.ObjectNotLocked(self)
2929
def _check_write_ok(self):
2930
"""Assert if writes are not permitted."""
2931
if not self._is_locked():
2932
raise errors.ObjectNotLocked(self)
2934
def _compression_parent(self, an_entry):
2935
# return the key that an_entry is compressed against, or None
2936
# Grab the second parent list (as deltas implies parents currently)
2937
compression_parents = an_entry[3][1]
2938
if not compression_parents:
2940
if len(compression_parents) != 1:
2941
raise AssertionError(
2942
"Too many compression parents: %r" % compression_parents)
2943
return compression_parents[0]
2945
def get_build_details(self, keys):
2946
"""Get the method, index_memo and compression parent for version_ids.
2948
Ghosts are omitted from the result.
2950
:param keys: An iterable of keys.
2951
:return: A dict of key:
2952
(index_memo, compression_parent, parents, record_details).
2954
opaque structure to pass to read_records to extract the raw
2957
Content that this record is built upon, may be None
2959
Logical parents of this node
2961
extra information about the content which needs to be passed to
2962
Factory.parse_record
2966
entries = self._get_entries(keys, False)
2967
for entry in entries:
2969
if not self._parents:
2972
parents = entry[3][0]
2973
if not self._deltas:
2974
compression_parent_key = None
2976
compression_parent_key = self._compression_parent(entry)
2977
noeol = (entry[2][0] == 'N')
2978
if compression_parent_key:
2979
method = 'line-delta'
2982
result[key] = (self._node_to_position(entry),
2983
compression_parent_key, parents,
2987
def _get_entries(self, keys, check_present=False):
2988
"""Get the entries for keys.
2990
:param keys: An iterable of index key tuples.
2995
for node in self._graph_index.iter_entries(keys):
2997
found_keys.add(node[1])
2999
# adapt parentless index to the rest of the code.
3000
for node in self._graph_index.iter_entries(keys):
3001
yield node[0], node[1], node[2], ()
3002
found_keys.add(node[1])
3004
missing_keys = keys.difference(found_keys)
3006
raise RevisionNotPresent(missing_keys.pop(), self)
3008
def get_method(self, key):
3009
"""Return compression method of specified key."""
3010
return self._get_method(self._get_node(key))
3012
def _get_method(self, node):
3013
if not self._deltas:
3015
if self._compression_parent(node):
3020
def _get_node(self, key):
3022
return list(self._get_entries([key]))[0]
3024
raise RevisionNotPresent(key, self)
3026
def get_options(self, key):
3027
"""Return a list representing options.
3031
node = self._get_node(key)
3032
options = [self._get_method(node)]
3033
if node[2][0] == 'N':
3034
options.append('no-eol')
3037
def get_parent_map(self, keys):
3038
"""Get a map of the parents of keys.
3040
:param keys: The keys to look up parents for.
3041
:return: A mapping from keys to parents. Absent keys are absent from
3045
nodes = self._get_entries(keys)
3049
result[node[1]] = node[3][0]
3052
result[node[1]] = None
3055
def get_position(self, key):
3056
"""Return details needed to access the version.
3058
:return: a tuple (index, data position, size) to hand to the access
3059
logic to get the record.
3061
node = self._get_node(key)
3062
return self._node_to_position(node)
3064
has_key = _mod_index._has_key_from_parent_map
3067
"""Get all the keys in the collection.
3069
The keys are not ordered.
3072
return [node[1] for node in self._graph_index.iter_all_entries()]
3074
missing_keys = _mod_index._missing_keys_from_parent_map
3076
def _node_to_position(self, node):
3077
"""Convert an index value to position details."""
3078
bits = node[2][1:].split(' ')
3079
return node[0], int(bits[0]), int(bits[1])
3081
def _sort_keys_by_io(self, keys, positions):
3082
"""Figure out an optimal order to read the records for the given keys.
3084
Sort keys, grouped by index and sorted by position.
3086
:param keys: A list of keys whose records we want to read. This will be
3088
:param positions: A dict, such as the one returned by
3089
_get_components_positions()
3092
def get_index_memo(key):
3093
# index_memo is at offset [1]. It is made up of (GraphIndex,
3094
# position, size). GI is an object, which will be unique for each
3095
# pack file. This causes us to group by pack file, then sort by
3096
# position. Size doesn't matter, but it isn't worth breaking up the
3098
return positions[key][1]
3099
return keys.sort(key=get_index_memo)
3101
_get_total_build_size = _get_total_build_size
3104
class _KnitKeyAccess(object):
3105
"""Access to records in .knit files."""
3107
def __init__(self, transport, mapper):
3108
"""Create a _KnitKeyAccess with transport and mapper.
3110
:param transport: The transport the access object is rooted at.
3111
:param mapper: The mapper used to map keys to .knit files.
3113
self._transport = transport
3114
self._mapper = mapper
3116
def add_raw_records(self, key_sizes, raw_data):
3117
"""Add raw knit bytes to a storage area.
3119
The data is spooled to the container writer in one bytes-record per
3122
:param sizes: An iterable of tuples containing the key and size of each
3124
:param raw_data: A bytestring containing the data.
3125
:return: A list of memos to retrieve the record later. Each memo is an
3126
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3127
length), where the key is the record key.
3129
if type(raw_data) is not str:
3130
raise AssertionError(
3131
'data must be plain bytes was %s' % type(raw_data))
3134
# TODO: This can be tuned for writing to sftp and other servers where
3135
# append() is relatively expensive by grouping the writes to each key
3137
for key, size in key_sizes:
3138
path = self._mapper.map(key)
3140
base = self._transport.append_bytes(path + '.knit',
3141
raw_data[offset:offset+size])
3142
except errors.NoSuchFile:
3143
self._transport.mkdir(osutils.dirname(path))
3144
base = self._transport.append_bytes(path + '.knit',
3145
raw_data[offset:offset+size])
3149
result.append((key, base, size))
3153
"""Flush pending writes on this access object.
3155
For .knit files this is a no-op.
3159
def get_raw_records(self, memos_for_retrieval):
3160
"""Get the raw bytes for a records.
3162
:param memos_for_retrieval: An iterable containing the access memo for
3163
retrieving the bytes.
3164
:return: An iterator over the bytes of the records.
3166
# first pass, group into same-index request to minimise readv's issued.
3168
current_prefix = None
3169
for (key, offset, length) in memos_for_retrieval:
3170
if current_prefix == key[:-1]:
3171
current_list.append((offset, length))
3173
if current_prefix is not None:
3174
request_lists.append((current_prefix, current_list))
3175
current_prefix = key[:-1]
3176
current_list = [(offset, length)]
3177
# handle the last entry
3178
if current_prefix is not None:
3179
request_lists.append((current_prefix, current_list))
3180
for prefix, read_vector in request_lists:
3181
path = self._mapper.map(prefix) + '.knit'
3182
for pos, data in self._transport.readv(path, read_vector):
3186
class _DirectPackAccess(object):
3187
"""Access to data in one or more packs with less translation."""
3189
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
3190
"""Create a _DirectPackAccess object.
3192
:param index_to_packs: A dict mapping index objects to the transport
3193
and file names for obtaining data.
3194
:param reload_func: A function to call if we determine that the pack
3195
files have moved and we need to reload our caches. See
3196
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
3198
self._container_writer = None
3199
self._write_index = None
3200
self._indices = index_to_packs
3201
self._reload_func = reload_func
3202
self._flush_func = flush_func
3204
def add_raw_records(self, key_sizes, raw_data):
3205
"""Add raw knit bytes to a storage area.
3207
The data is spooled to the container writer in one bytes-record per
3210
:param sizes: An iterable of tuples containing the key and size of each
3212
:param raw_data: A bytestring containing the data.
3213
:return: A list of memos to retrieve the record later. Each memo is an
3214
opaque index memo. For _DirectPackAccess the memo is (index, pos,
3215
length), where the index field is the write_index object supplied
3216
to the PackAccess object.
3218
if type(raw_data) is not str:
3219
raise AssertionError(
3220
'data must be plain bytes was %s' % type(raw_data))
3223
for key, size in key_sizes:
3224
p_offset, p_length = self._container_writer.add_bytes_record(
3225
raw_data[offset:offset+size], [])
3227
result.append((self._write_index, p_offset, p_length))
3231
"""Flush pending writes on this access object.
3233
This will flush any buffered writes to a NewPack.
3235
if self._flush_func is not None:
3238
def get_raw_records(self, memos_for_retrieval):
3239
"""Get the raw bytes for a records.
3241
:param memos_for_retrieval: An iterable containing the (index, pos,
3242
length) memo for retrieving the bytes. The Pack access method
3243
looks up the pack to use for a given record in its index_to_pack
3245
:return: An iterator over the bytes of the records.
3247
# first pass, group into same-index requests
3249
current_index = None
3250
for (index, offset, length) in memos_for_retrieval:
3251
if current_index == index:
3252
current_list.append((offset, length))
3254
if current_index is not None:
3255
request_lists.append((current_index, current_list))
3256
current_index = index
3257
current_list = [(offset, length)]
3258
# handle the last entry
3259
if current_index is not None:
3260
request_lists.append((current_index, current_list))
3261
for index, offsets in request_lists:
3263
transport, path = self._indices[index]
3265
# A KeyError here indicates that someone has triggered an index
3266
# reload, and this index has gone missing, we need to start
3268
if self._reload_func is None:
3269
# If we don't have a _reload_func there is nothing that can
3272
raise errors.RetryWithNewPacks(index,
3273
reload_occurred=True,
3274
exc_info=sys.exc_info())
3276
reader = pack.make_readv_reader(transport, path, offsets)
3277
for names, read_func in reader.iter_records():
3278
yield read_func(None)
3279
except errors.NoSuchFile:
3280
# A NoSuchFile error indicates that a pack file has gone
3281
# missing on disk, we need to trigger a reload, and start over.
3282
if self._reload_func is None:
3284
raise errors.RetryWithNewPacks(transport.abspath(path),
3285
reload_occurred=False,
3286
exc_info=sys.exc_info())
3288
def set_writer(self, writer, index, transport_packname):
3289
"""Set a writer to use for adding data."""
3290
if index is not None:
3291
self._indices[index] = transport_packname
3292
self._container_writer = writer
3293
self._write_index = index
3295
def reload_or_raise(self, retry_exc):
3296
"""Try calling the reload function, or re-raise the original exception.
3298
This should be called after _DirectPackAccess raises a
3299
RetryWithNewPacks exception. This function will handle the common logic
3300
of determining when the error is fatal versus being temporary.
3301
It will also make sure that the original exception is raised, rather
3302
than the RetryWithNewPacks exception.
3304
If this function returns, then the calling function should retry
3305
whatever operation was being performed. Otherwise an exception will
3308
:param retry_exc: A RetryWithNewPacks exception.
3311
if self._reload_func is None:
3313
elif not self._reload_func():
3314
# The reload claimed that nothing changed
3315
if not retry_exc.reload_occurred:
3316
# If there wasn't an earlier reload, then we really were
3317
# expecting to find changes. We didn't find them, so this is a
3321
exc_class, exc_value, exc_traceback = retry_exc.exc_info
3322
raise exc_class, exc_value, exc_traceback
3325
# Deprecated, use PatienceSequenceMatcher instead
3326
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
3329
def annotate_knit(knit, revision_id):
3330
"""Annotate a knit with no cached annotations.
3332
This implementation is for knits with no cached annotations.
3333
It will work for knits with cached annotations, but this is not
3336
annotator = _KnitAnnotator(knit)
3337
return iter(annotator.annotate_flat(revision_id))
3340
class _KnitAnnotator(annotate.Annotator):
3341
"""Build up the annotations for a text."""
3343
def __init__(self, vf):
3344
annotate.Annotator.__init__(self, vf)
3346
# TODO: handle Nodes which cannot be extracted
3347
# self._ghosts = set()
3349
# Map from (key, parent_key) => matching_blocks, should be 'use once'
3350
self._matching_blocks = {}
3352
# KnitContent objects
3353
self._content_objects = {}
3354
# The number of children that depend on this fulltext content object
3355
self._num_compression_children = {}
3356
# Delta records that need their compression parent before they can be
3358
self._pending_deltas = {}
3359
# Fulltext records that are waiting for their parents fulltexts before
3360
# they can be yielded for annotation
3361
self._pending_annotation = {}
3363
self._all_build_details = {}
3365
def _get_build_graph(self, key):
3366
"""Get the graphs for building texts and annotations.
3368
The data you need for creating a full text may be different than the
3369
data you need to annotate that text. (At a minimum, you need both
3370
parents to create an annotation, but only need 1 parent to generate the
3373
:return: A list of (key, index_memo) records, suitable for
3374
passing to read_records_iter to start reading in the raw data from
3377
pending = set([key])
3381
self._num_needed_children[key] = 1
3383
# get all pending nodes
3385
this_iteration = pending
3386
build_details = self._vf._index.get_build_details(this_iteration)
3387
self._all_build_details.update(build_details)
3388
# new_nodes = self._vf._index._get_entries(this_iteration)
3390
for key, details in build_details.iteritems():
3391
(index_memo, compression_parent, parent_keys,
3392
record_details) = details
3393
self._parent_map[key] = parent_keys
3394
self._heads_provider = None
3395
records.append((key, index_memo))
3396
# Do we actually need to check _annotated_lines?
3397
pending.update(p for p in parent_keys
3398
if p not in self._all_build_details)
3400
for parent_key in parent_keys:
3401
if parent_key in self._num_needed_children:
3402
self._num_needed_children[parent_key] += 1
3404
self._num_needed_children[parent_key] = 1
3405
if compression_parent:
3406
if compression_parent in self._num_compression_children:
3407
self._num_compression_children[compression_parent] += 1
3409
self._num_compression_children[compression_parent] = 1
3411
missing_versions = this_iteration.difference(build_details.keys())
3412
if missing_versions:
3413
raise ValueError('i dont handle ghosts')
3414
# Generally we will want to read the records in reverse order, because
3415
# we find the parent nodes after the children
3419
def _get_needed_texts(self, key, pb=None):
3420
# if True or len(self._vf._fallback_vfs) > 0:
3421
if len(self._vf._fallback_vfs) > 0:
3422
# If we have fallbacks, go to the generic path
3423
for v in annotate.Annotator._get_needed_texts(self, key, pb=pb):
3428
records = self._get_build_graph(key)
3429
for idx, (key, text, num_lines) in enumerate(
3430
self._extract_texts(records)):
3432
pb.update('annotating', idx, len(records))
3433
yield key, text, num_lines
3435
except errors.RetryWithNewPacks, e:
3436
self._vf._access.reload_or_raise(e)
3437
# The cached build_details are no longer valid
3438
self._all_build_details.clear()
3440
def _cache_delta_blocks(self, key, compression_parent, delta, lines):
3441
parent_lines = self._text_cache[compression_parent]
3442
blocks = list(KnitContent.get_line_delta_blocks(delta, parent_lines, lines))
3443
self._matching_blocks[(key, compression_parent)] = blocks
3445
def _expand_record(self, key, parent_keys, compression_parent, record,
3448
if compression_parent:
3449
if compression_parent not in self._content_objects:
3450
# Waiting for the parent
3451
self._pending_deltas.setdefault(compression_parent, []).append(
3452
(key, parent_keys, record, record_details))
3454
# We have the basis parent, so expand the delta
3455
num = self._num_compression_children[compression_parent]
3458
base_content = self._content_objects.pop(compression_parent)
3459
self._num_compression_children.pop(compression_parent)
3461
self._num_compression_children[compression_parent] = num
3462
base_content = self._content_objects[compression_parent]
3463
# It is tempting to want to copy_base_content=False for the last
3464
# child object. However, whenever noeol=False,
3465
# self._text_cache[parent_key] is content._lines. So mutating it
3466
# gives very bad results.
3467
# The alternative is to copy the lines into text cache, but then we
3468
# are copying anyway, so just do it here.
3469
content, delta = self._vf._factory.parse_record(
3470
key, record, record_details, base_content,
3471
copy_base_content=True)
3474
content, _ = self._vf._factory.parse_record(
3475
key, record, record_details, None)
3476
if self._num_compression_children.get(key, 0) > 0:
3477
self._content_objects[key] = content
3478
lines = content.text()
3479
self._text_cache[key] = lines
3480
if delta is not None:
3481
self._cache_delta_blocks(key, compression_parent, delta, lines)
3484
def _get_parent_annotations_and_matches(self, key, text, parent_key):
3485
"""Get the list of annotations for the parent, and the matching lines.
3487
:param text: The opaque value given by _get_needed_texts
3488
:param parent_key: The key for the parent text
3489
:return: (parent_annotations, matching_blocks)
3490
parent_annotations is a list as long as the number of lines in
3492
matching_blocks is a list of (parent_idx, text_idx, len) tuples
3493
indicating which lines match between the two texts
3495
block_key = (key, parent_key)
3496
if block_key in self._matching_blocks:
3497
blocks = self._matching_blocks.pop(block_key)
3498
parent_annotations = self._annotations_cache[parent_key]
3499
return parent_annotations, blocks
3500
return annotate.Annotator._get_parent_annotations_and_matches(self,
3501
key, text, parent_key)
3503
def _process_pending(self, key):
3504
"""The content for 'key' was just processed.
3506
Determine if there is any more pending work to be processed.
3509
if key in self._pending_deltas:
3510
compression_parent = key
3511
children = self._pending_deltas.pop(key)
3512
for child_key, parent_keys, record, record_details in children:
3513
lines = self._expand_record(child_key, parent_keys,
3515
record, record_details)
3516
assert lines is not None
3517
if self._check_ready_for_annotations(child_key, parent_keys):
3518
to_return.append(child_key)
3519
# Also check any children that are waiting for this parent to be
3521
if key in self._pending_annotation:
3522
children = self._pending_annotation.pop(key)
3523
to_return.extend([c for c, p_keys in children
3524
if self._check_ready_for_annotations(c, p_keys)])
3527
def _check_ready_for_annotations(self, key, parent_keys):
3528
"""return true if this text is ready to be yielded.
3530
Otherwise, this will return False, and queue the text into
3531
self._pending_annotation
3533
for parent_key in parent_keys:
3534
if parent_key not in self._annotations_cache:
3535
# still waiting on at least one parent text, so queue it up
3536
# Note that if there are multiple parents, we need to wait
3538
self._pending_annotation.setdefault(parent_key,
3539
[]).append((key, parent_keys))
3543
def _extract_texts(self, records):
3544
"""Extract the various texts needed based on records"""
3545
# We iterate in the order read, rather than a strict order requested
3546
# However, process what we can, and put off to the side things that
3547
# still need parents, cleaning them up when those parents are
3550
# 1) As 'records' are read, see if we can expand these records into
3551
# Content objects (and thus lines)
3552
# 2) If a given line-delta is waiting on its compression parent, it
3553
# gets queued up into self._pending_deltas, otherwise we expand
3554
# it, and put it into self._text_cache and self._content_objects
3555
# 3) If we expanded the text, we will then check to see if all
3556
# parents have also been processed. If so, this text gets yielded,
3557
# else this record gets set aside into pending_annotation
3558
# 4) Further, if we expanded the text in (2), we will then check to
3559
# see if there are any children in self._pending_deltas waiting to
3560
# also be processed. If so, we go back to (2) for those
3561
# 5) Further again, if we yielded the text, we can then check if that
3562
# 'unlocks' any of the texts in pending_annotations, which should
3563
# then get yielded as well
3564
# Note that both steps 4 and 5 are 'recursive' in that unlocking one
3565
# compression child could unlock yet another, and yielding a fulltext
3566
# will also 'unlock' the children that are waiting on that annotation.
3567
# (Though also, unlocking 1 parent's fulltext, does not unlock a child
3568
# if other parents are also waiting.)
3569
# We want to yield content before expanding child content objects, so
3570
# that we know when we can re-use the content lines, and the annotation
3571
# code can know when it can stop caching fulltexts, as well.
3573
# Children that are missing their compression parent
3575
for (key, record, digest) in self._vf._read_records_iter(records):
3577
details = self._all_build_details[key]
3578
(_, compression_parent, parent_keys, record_details) = details
3579
assert parent_keys == self._parent_map[key]
3580
lines = self._expand_record(key, parent_keys, compression_parent,
3581
record, record_details)
3583
# Pending delta should be queued up
3585
# At this point, we may be able to yield this content, if all
3586
# parents are also finished
3587
yield_this_text = self._check_ready_for_annotations(key,
3590
# All parents present
3591
yield key, lines, len(lines)
3592
to_process = self._process_pending(key)
3594
this_process = to_process
3596
for key in this_process:
3597
lines = self._text_cache[key]
3598
yield key, lines, len(lines)
3599
to_process.extend(self._process_pending(key))
3602
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3604
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
