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 get_annotator(self):
1046
return _KnitAnnotator(self)
1048
def check(self, progress_bar=None, keys=None):
1049
"""See VersionedFiles.check()."""
1051
return self._logical_check()
1053
# At the moment, check does not extra work over get_record_stream
1054
return self.get_record_stream(keys, 'unordered', True)
1056
def _logical_check(self):
1057
# This doesn't actually test extraction of everything, but that will
1058
# impact 'bzr check' substantially, and needs to be integrated with
1059
# care. However, it does check for the obvious problem of a delta with
1061
keys = self._index.keys()
1062
parent_map = self.get_parent_map(keys)
1064
if self._index.get_method(key) != 'fulltext':
1065
compression_parent = parent_map[key][0]
1066
if compression_parent not in parent_map:
1067
raise errors.KnitCorrupt(self,
1068
"Missing basis parent %s for %s" % (
1069
compression_parent, key))
1070
for fallback_vfs in self._fallback_vfs:
1071
fallback_vfs.check()
1073
def _check_add(self, key, lines, random_id, check_content):
1074
"""check that version_id and lines are safe to add."""
1075
version_id = key[-1]
1076
if version_id is not None:
1077
if contains_whitespace(version_id):
1078
raise InvalidRevisionId(version_id, self)
1079
self.check_not_reserved_id(version_id)
1080
# TODO: If random_id==False and the key is already present, we should
1081
# probably check that the existing content is identical to what is
1082
# being inserted, and otherwise raise an exception. This would make
1083
# the bundle code simpler.
1085
self._check_lines_not_unicode(lines)
1086
self._check_lines_are_lines(lines)
1088
def _check_header(self, key, line):
1089
rec = self._split_header(line)
1090
self._check_header_version(rec, key[-1])
1093
def _check_header_version(self, rec, version_id):
1094
"""Checks the header version on original format knit records.
1096
These have the last component of the key embedded in the record.
1098
if rec[1] != version_id:
1099
raise KnitCorrupt(self,
1100
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1102
def _check_should_delta(self, parent):
1103
"""Iterate back through the parent listing, looking for a fulltext.
1105
This is used when we want to decide whether to add a delta or a new
1106
fulltext. It searches for _max_delta_chain parents. When it finds a
1107
fulltext parent, it sees if the total size of the deltas leading up to
1108
it is large enough to indicate that we want a new full text anyway.
1110
Return True if we should create a new delta, False if we should use a
1114
fulltext_size = None
1115
for count in xrange(self._max_delta_chain):
1117
# Note that this only looks in the index of this particular
1118
# KnitVersionedFiles, not in the fallbacks. This ensures that
1119
# we won't store a delta spanning physical repository
1121
build_details = self._index.get_build_details([parent])
1122
parent_details = build_details[parent]
1123
except (RevisionNotPresent, KeyError), e:
1124
# Some basis is not locally present: always fulltext
1126
index_memo, compression_parent, _, _ = parent_details
1127
_, _, size = index_memo
1128
if compression_parent is None:
1129
fulltext_size = size
1132
# We don't explicitly check for presence because this is in an
1133
# inner loop, and if it's missing it'll fail anyhow.
1134
parent = compression_parent
1136
# We couldn't find a fulltext, so we must create a new one
1138
# Simple heuristic - if the total I/O wold be greater as a delta than
1139
# the originally installed fulltext, we create a new fulltext.
1140
return fulltext_size > delta_size
1142
def _build_details_to_components(self, build_details):
1143
"""Convert a build_details tuple to a position tuple."""
1144
# record_details, access_memo, compression_parent
1145
return build_details[3], build_details[0], build_details[1]
1147
def _get_components_positions(self, keys, allow_missing=False):
1148
"""Produce a map of position data for the components of keys.
1150
This data is intended to be used for retrieving the knit records.
1152
A dict of key to (record_details, index_memo, next, parents) is
1154
method is the way referenced data should be applied.
1155
index_memo is the handle to pass to the data access to actually get the
1157
next is the build-parent of the version, or None for fulltexts.
1158
parents is the version_ids of the parents of this version
1160
:param allow_missing: If True do not raise an error on a missing component,
1164
pending_components = keys
1165
while pending_components:
1166
build_details = self._index.get_build_details(pending_components)
1167
current_components = set(pending_components)
1168
pending_components = set()
1169
for key, details in build_details.iteritems():
1170
(index_memo, compression_parent, parents,
1171
record_details) = details
1172
method = record_details[0]
1173
if compression_parent is not None:
1174
pending_components.add(compression_parent)
1175
component_data[key] = self._build_details_to_components(details)
1176
missing = current_components.difference(build_details)
1177
if missing and not allow_missing:
1178
raise errors.RevisionNotPresent(missing.pop(), self)
1179
return component_data
1181
def _get_content(self, key, parent_texts={}):
1182
"""Returns a content object that makes up the specified
1184
cached_version = parent_texts.get(key, None)
1185
if cached_version is not None:
1186
# Ensure the cache dict is valid.
1187
if not self.get_parent_map([key]):
1188
raise RevisionNotPresent(key, self)
1189
return cached_version
1190
generator = _VFContentMapGenerator(self, [key])
1191
return generator._get_content(key)
1193
def get_known_graph_ancestry(self, keys):
1194
"""Get a KnownGraph instance with the ancestry of keys."""
1195
parent_map, missing_keys = self._index.find_ancestry(keys)
1196
kg = _mod_graph.KnownGraph(parent_map)
1199
def get_parent_map(self, keys):
1200
"""Get a map of the graph parents of keys.
1202
:param keys: The keys to look up parents for.
1203
:return: A mapping from keys to parents. Absent keys are absent from
1206
return self._get_parent_map_with_sources(keys)[0]
1208
def _get_parent_map_with_sources(self, keys):
1209
"""Get a map of the parents of keys.
1211
:param keys: The keys to look up parents for.
1212
:return: A tuple. The first element is a mapping from keys to parents.
1213
Absent keys are absent from the mapping. The second element is a
1214
list with the locations each key was found in. The first element
1215
is the in-this-knit parents, the second the first fallback source,
1219
sources = [self._index] + self._fallback_vfs
1222
for source in sources:
1225
new_result = source.get_parent_map(missing)
1226
source_results.append(new_result)
1227
result.update(new_result)
1228
missing.difference_update(set(new_result))
1229
return result, source_results
1231
def _get_record_map(self, keys, allow_missing=False):
1232
"""Produce a dictionary of knit records.
1234
:return: {key:(record, record_details, digest, next)}
1236
data returned from read_records (a KnitContentobject)
1238
opaque information to pass to parse_record
1240
SHA1 digest of the full text after all steps are done
1242
build-parent of the version, i.e. the leftmost ancestor.
1243
Will be None if the record is not a delta.
1244
:param keys: The keys to build a map for
1245
:param allow_missing: If some records are missing, rather than
1246
error, just return the data that could be generated.
1248
raw_map = self._get_record_map_unparsed(keys,
1249
allow_missing=allow_missing)
1250
return self._raw_map_to_record_map(raw_map)
1252
def _raw_map_to_record_map(self, raw_map):
1253
"""Parse the contents of _get_record_map_unparsed.
1255
:return: see _get_record_map.
1259
data, record_details, next = raw_map[key]
1260
content, digest = self._parse_record(key[-1], data)
1261
result[key] = content, record_details, digest, next
1264
def _get_record_map_unparsed(self, keys, allow_missing=False):
1265
"""Get the raw data for reconstructing keys without parsing it.
1267
:return: A dict suitable for parsing via _raw_map_to_record_map.
1268
key-> raw_bytes, (method, noeol), compression_parent
1270
# This retries the whole request if anything fails. Potentially we
1271
# could be a bit more selective. We could track the keys whose records
1272
# we have successfully found, and then only request the new records
1273
# from there. However, _get_components_positions grabs the whole build
1274
# chain, which means we'll likely try to grab the same records again
1275
# anyway. Also, can the build chains change as part of a pack
1276
# operation? We wouldn't want to end up with a broken chain.
1279
position_map = self._get_components_positions(keys,
1280
allow_missing=allow_missing)
1281
# key = component_id, r = record_details, i_m = index_memo,
1283
records = [(key, i_m) for key, (r, i_m, n)
1284
in position_map.iteritems()]
1285
# Sort by the index memo, so that we request records from the
1286
# same pack file together, and in forward-sorted order
1287
records.sort(key=operator.itemgetter(1))
1289
for key, data in self._read_records_iter_unchecked(records):
1290
(record_details, index_memo, next) = position_map[key]
1291
raw_record_map[key] = data, record_details, next
1292
return raw_record_map
1293
except errors.RetryWithNewPacks, e:
1294
self._access.reload_or_raise(e)
1297
def _split_by_prefix(cls, keys):
1298
"""For the given keys, split them up based on their prefix.
1300
To keep memory pressure somewhat under control, split the
1301
requests back into per-file-id requests, otherwise "bzr co"
1302
extracts the full tree into memory before writing it to disk.
1303
This should be revisited if _get_content_maps() can ever cross
1306
The keys for a given file_id are kept in the same relative order.
1307
Ordering between file_ids is not, though prefix_order will return the
1308
order that the key was first seen.
1310
:param keys: An iterable of key tuples
1311
:return: (split_map, prefix_order)
1312
split_map A dictionary mapping prefix => keys
1313
prefix_order The order that we saw the various prefixes
1315
split_by_prefix = {}
1323
if prefix in split_by_prefix:
1324
split_by_prefix[prefix].append(key)
1326
split_by_prefix[prefix] = [key]
1327
prefix_order.append(prefix)
1328
return split_by_prefix, prefix_order
1330
def _group_keys_for_io(self, keys, non_local_keys, positions,
1331
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1332
"""For the given keys, group them into 'best-sized' requests.
1334
The idea is to avoid making 1 request per file, but to never try to
1335
unpack an entire 1.5GB source tree in a single pass. Also when
1336
possible, we should try to group requests to the same pack file
1339
:return: list of (keys, non_local) tuples that indicate what keys
1340
should be fetched next.
1342
# TODO: Ideally we would group on 2 factors. We want to extract texts
1343
# from the same pack file together, and we want to extract all
1344
# the texts for a given build-chain together. Ultimately it
1345
# probably needs a better global view.
1346
total_keys = len(keys)
1347
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1348
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1350
cur_non_local = set()
1354
for prefix in prefix_order:
1355
keys = prefix_split_keys[prefix]
1356
non_local = prefix_split_non_local_keys.get(prefix, [])
1358
this_size = self._index._get_total_build_size(keys, positions)
1359
cur_size += this_size
1360
cur_keys.extend(keys)
1361
cur_non_local.update(non_local)
1362
if cur_size > _min_buffer_size:
1363
result.append((cur_keys, cur_non_local))
1364
sizes.append(cur_size)
1366
cur_non_local = set()
1369
result.append((cur_keys, cur_non_local))
1370
sizes.append(cur_size)
1373
def get_record_stream(self, keys, ordering, include_delta_closure):
1374
"""Get a stream of records for keys.
1376
:param keys: The keys to include.
1377
:param ordering: Either 'unordered' or 'topological'. A topologically
1378
sorted stream has compression parents strictly before their
1380
:param include_delta_closure: If True then the closure across any
1381
compression parents will be included (in the opaque data).
1382
:return: An iterator of ContentFactory objects, each of which is only
1383
valid until the iterator is advanced.
1385
# keys might be a generator
1389
if not self._index.has_graph:
1390
# Cannot sort when no graph has been stored.
1391
ordering = 'unordered'
1393
remaining_keys = keys
1396
keys = set(remaining_keys)
1397
for content_factory in self._get_remaining_record_stream(keys,
1398
ordering, include_delta_closure):
1399
remaining_keys.discard(content_factory.key)
1400
yield content_factory
1402
except errors.RetryWithNewPacks, e:
1403
self._access.reload_or_raise(e)
1405
def _get_remaining_record_stream(self, keys, ordering,
1406
include_delta_closure):
1407
"""This function is the 'retry' portion for get_record_stream."""
1408
if include_delta_closure:
1409
positions = self._get_components_positions(keys, allow_missing=True)
1411
build_details = self._index.get_build_details(keys)
1413
# (record_details, access_memo, compression_parent_key)
1414
positions = dict((key, self._build_details_to_components(details))
1415
for key, details in build_details.iteritems())
1416
absent_keys = keys.difference(set(positions))
1417
# There may be more absent keys : if we're missing the basis component
1418
# and are trying to include the delta closure.
1419
# XXX: We should not ever need to examine remote sources because we do
1420
# not permit deltas across versioned files boundaries.
1421
if include_delta_closure:
1422
needed_from_fallback = set()
1423
# Build up reconstructable_keys dict. key:True in this dict means
1424
# the key can be reconstructed.
1425
reconstructable_keys = {}
1429
chain = [key, positions[key][2]]
1431
needed_from_fallback.add(key)
1434
while chain[-1] is not None:
1435
if chain[-1] in reconstructable_keys:
1436
result = reconstructable_keys[chain[-1]]
1440
chain.append(positions[chain[-1]][2])
1442
# missing basis component
1443
needed_from_fallback.add(chain[-1])
1446
for chain_key in chain[:-1]:
1447
reconstructable_keys[chain_key] = result
1449
needed_from_fallback.add(key)
1450
# Double index lookups here : need a unified api ?
1451
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1452
if ordering in ('topological', 'groupcompress'):
1453
if ordering == 'topological':
1454
# Global topological sort
1455
present_keys = tsort.topo_sort(global_map)
1457
present_keys = sort_groupcompress(global_map)
1458
# Now group by source:
1460
current_source = None
1461
for key in present_keys:
1462
for parent_map in parent_maps:
1463
if key in parent_map:
1464
key_source = parent_map
1466
if current_source is not key_source:
1467
source_keys.append((key_source, []))
1468
current_source = key_source
1469
source_keys[-1][1].append(key)
1471
if ordering != 'unordered':
1472
raise AssertionError('valid values for ordering are:'
1473
' "unordered", "groupcompress" or "topological" not: %r'
1475
# Just group by source; remote sources first.
1478
for parent_map in reversed(parent_maps):
1479
source_keys.append((parent_map, []))
1480
for key in parent_map:
1481
present_keys.append(key)
1482
source_keys[-1][1].append(key)
1483
# We have been requested to return these records in an order that
1484
# suits us. So we ask the index to give us an optimally sorted
1486
for source, sub_keys in source_keys:
1487
if source is parent_maps[0]:
1488
# Only sort the keys for this VF
1489
self._index._sort_keys_by_io(sub_keys, positions)
1490
absent_keys = keys - set(global_map)
1491
for key in absent_keys:
1492
yield AbsentContentFactory(key)
1493
# restrict our view to the keys we can answer.
1494
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1495
# XXX: At that point we need to consider the impact of double reads by
1496
# utilising components multiple times.
1497
if include_delta_closure:
1498
# XXX: get_content_maps performs its own index queries; allow state
1500
non_local_keys = needed_from_fallback - absent_keys
1501
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1504
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1507
for record in generator.get_record_stream():
1510
for source, keys in source_keys:
1511
if source is parent_maps[0]:
1512
# this KnitVersionedFiles
1513
records = [(key, positions[key][1]) for key in keys]
1514
for key, raw_data, sha1 in self._read_records_iter_raw(records):
1515
(record_details, index_memo, _) = positions[key]
1516
yield KnitContentFactory(key, global_map[key],
1517
record_details, sha1, raw_data, self._factory.annotated, None)
1519
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1520
for record in vf.get_record_stream(keys, ordering,
1521
include_delta_closure):
1524
def get_sha1s(self, keys):
1525
"""See VersionedFiles.get_sha1s()."""
1527
record_map = self._get_record_map(missing, allow_missing=True)
1529
for key, details in record_map.iteritems():
1530
if key not in missing:
1532
# record entry 2 is the 'digest'.
1533
result[key] = details[2]
1534
missing.difference_update(set(result))
1535
for source in self._fallback_vfs:
1538
new_result = source.get_sha1s(missing)
1539
result.update(new_result)
1540
missing.difference_update(set(new_result))
1543
def insert_record_stream(self, stream):
1544
"""Insert a record stream into this container.
1546
:param stream: A stream of records to insert.
1548
:seealso VersionedFiles.get_record_stream:
1550
def get_adapter(adapter_key):
1552
return adapters[adapter_key]
1554
adapter_factory = adapter_registry.get(adapter_key)
1555
adapter = adapter_factory(self)
1556
adapters[adapter_key] = adapter
1559
if self._factory.annotated:
1560
# self is annotated, we need annotated knits to use directly.
1561
annotated = "annotated-"
1564
# self is not annotated, but we can strip annotations cheaply.
1566
convertibles = set(["knit-annotated-ft-gz"])
1567
if self._max_delta_chain:
1568
delta_types.add("knit-annotated-delta-gz")
1569
convertibles.add("knit-annotated-delta-gz")
1570
# The set of types we can cheaply adapt without needing basis texts.
1571
native_types = set()
1572
if self._max_delta_chain:
1573
native_types.add("knit-%sdelta-gz" % annotated)
1574
delta_types.add("knit-%sdelta-gz" % annotated)
1575
native_types.add("knit-%sft-gz" % annotated)
1576
knit_types = native_types.union(convertibles)
1578
# Buffer all index entries that we can't add immediately because their
1579
# basis parent is missing. We don't buffer all because generating
1580
# annotations may require access to some of the new records. However we
1581
# can't generate annotations from new deltas until their basis parent
1582
# is present anyway, so we get away with not needing an index that
1583
# includes the new keys.
1585
# See <http://launchpad.net/bugs/300177> about ordering of compression
1586
# parents in the records - to be conservative, we insist that all
1587
# parents must be present to avoid expanding to a fulltext.
1589
# key = basis_parent, value = index entry to add
1590
buffered_index_entries = {}
1591
for record in stream:
1593
parents = record.parents
1594
if record.storage_kind in delta_types:
1595
# TODO: eventually the record itself should track
1596
# compression_parent
1597
compression_parent = parents[0]
1599
compression_parent = None
1600
# Raise an error when a record is missing.
1601
if record.storage_kind == 'absent':
1602
raise RevisionNotPresent([record.key], self)
1603
elif ((record.storage_kind in knit_types)
1604
and (compression_parent is None
1605
or not self._fallback_vfs
1606
or self._index.has_key(compression_parent)
1607
or not self.has_key(compression_parent))):
1608
# we can insert the knit record literally if either it has no
1609
# compression parent OR we already have its basis in this kvf
1610
# OR the basis is not present even in the fallbacks. In the
1611
# last case it will either turn up later in the stream and all
1612
# will be well, or it won't turn up at all and we'll raise an
1615
# TODO: self.has_key is somewhat redundant with
1616
# self._index.has_key; we really want something that directly
1617
# asks if it's only present in the fallbacks. -- mbp 20081119
1618
if record.storage_kind not in native_types:
1620
adapter_key = (record.storage_kind, "knit-delta-gz")
1621
adapter = get_adapter(adapter_key)
1623
adapter_key = (record.storage_kind, "knit-ft-gz")
1624
adapter = get_adapter(adapter_key)
1625
bytes = adapter.get_bytes(record)
1627
# It's a knit record, it has a _raw_record field (even if
1628
# it was reconstituted from a network stream).
1629
bytes = record._raw_record
1630
options = [record._build_details[0]]
1631
if record._build_details[1]:
1632
options.append('no-eol')
1633
# Just blat it across.
1634
# Note: This does end up adding data on duplicate keys. As
1635
# modern repositories use atomic insertions this should not
1636
# lead to excessive growth in the event of interrupted fetches.
1637
# 'knit' repositories may suffer excessive growth, but as a
1638
# deprecated format this is tolerable. It can be fixed if
1639
# needed by in the kndx index support raising on a duplicate
1640
# add with identical parents and options.
1641
access_memo = self._access.add_raw_records(
1642
[(record.key, len(bytes))], bytes)[0]
1643
index_entry = (record.key, options, access_memo, parents)
1644
if 'fulltext' not in options:
1645
# Not a fulltext, so we need to make sure the compression
1646
# parent will also be present.
1647
# Note that pack backed knits don't need to buffer here
1648
# because they buffer all writes to the transaction level,
1649
# but we don't expose that difference at the index level. If
1650
# the query here has sufficient cost to show up in
1651
# profiling we should do that.
1653
# They're required to be physically in this
1654
# KnitVersionedFiles, not in a fallback.
1655
if not self._index.has_key(compression_parent):
1656
pending = buffered_index_entries.setdefault(
1657
compression_parent, [])
1658
pending.append(index_entry)
1661
self._index.add_records([index_entry])
1662
elif record.storage_kind == 'chunked':
1663
self.add_lines(record.key, parents,
1664
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1666
# Not suitable for direct insertion as a
1667
# delta, either because it's not the right format, or this
1668
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1669
# 0) or because it depends on a base only present in the
1671
self._access.flush()
1673
# Try getting a fulltext directly from the record.
1674
bytes = record.get_bytes_as('fulltext')
1675
except errors.UnavailableRepresentation:
1676
adapter_key = record.storage_kind, 'fulltext'
1677
adapter = get_adapter(adapter_key)
1678
bytes = adapter.get_bytes(record)
1679
lines = split_lines(bytes)
1681
self.add_lines(record.key, parents, lines)
1682
except errors.RevisionAlreadyPresent:
1684
# Add any records whose basis parent is now available.
1686
added_keys = [record.key]
1688
key = added_keys.pop(0)
1689
if key in buffered_index_entries:
1690
index_entries = buffered_index_entries[key]
1691
self._index.add_records(index_entries)
1693
[index_entry[0] for index_entry in index_entries])
1694
del buffered_index_entries[key]
1695
if buffered_index_entries:
1696
# There were index entries buffered at the end of the stream,
1697
# So these need to be added (if the index supports holding such
1698
# entries for later insertion)
1699
for key in buffered_index_entries:
1700
index_entries = buffered_index_entries[key]
1701
self._index.add_records(index_entries,
1702
missing_compression_parents=True)
1704
def get_missing_compression_parent_keys(self):
1705
"""Return an iterable of keys of missing compression parents.
1707
Check this after calling insert_record_stream to find out if there are
1708
any missing compression parents. If there are, the records that
1709
depend on them are not able to be inserted safely. For atomic
1710
KnitVersionedFiles built on packs, the transaction should be aborted or
1711
suspended - commit will fail at this point. Nonatomic knits will error
1712
earlier because they have no staging area to put pending entries into.
1714
return self._index.get_missing_compression_parents()
1716
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1717
"""Iterate over the lines in the versioned files from keys.
1719
This may return lines from other keys. Each item the returned
1720
iterator yields is a tuple of a line and a text version that that line
1721
is present in (not introduced in).
1723
Ordering of results is in whatever order is most suitable for the
1724
underlying storage format.
1726
If a progress bar is supplied, it may be used to indicate progress.
1727
The caller is responsible for cleaning up progress bars (because this
1731
* Lines are normalised by the underlying store: they will all have \\n
1733
* Lines are returned in arbitrary order.
1734
* If a requested key did not change any lines (or didn't have any
1735
lines), it may not be mentioned at all in the result.
1737
:param pb: Progress bar supplied by caller.
1738
:return: An iterator over (line, key).
1741
pb = progress.DummyProgress()
1747
# we don't care about inclusions, the caller cares.
1748
# but we need to setup a list of records to visit.
1749
# we need key, position, length
1751
build_details = self._index.get_build_details(keys)
1752
for key, details in build_details.iteritems():
1754
key_records.append((key, details[0]))
1755
records_iter = enumerate(self._read_records_iter(key_records))
1756
for (key_idx, (key, data, sha_value)) in records_iter:
1757
pb.update('Walking content', key_idx, total)
1758
compression_parent = build_details[key][1]
1759
if compression_parent is None:
1761
line_iterator = self._factory.get_fulltext_content(data)
1764
line_iterator = self._factory.get_linedelta_content(data)
1765
# Now that we are yielding the data for this key, remove it
1768
# XXX: It might be more efficient to yield (key,
1769
# line_iterator) in the future. However for now, this is a
1770
# simpler change to integrate into the rest of the
1771
# codebase. RBC 20071110
1772
for line in line_iterator:
1775
except errors.RetryWithNewPacks, e:
1776
self._access.reload_or_raise(e)
1777
# If there are still keys we've not yet found, we look in the fallback
1778
# vfs, and hope to find them there. Note that if the keys are found
1779
# but had no changes or no content, the fallback may not return
1781
if keys and not self._fallback_vfs:
1782
# XXX: strictly the second parameter is meant to be the file id
1783
# but it's not easily accessible here.
1784
raise RevisionNotPresent(keys, repr(self))
1785
for source in self._fallback_vfs:
1789
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1790
source_keys.add(key)
1792
keys.difference_update(source_keys)
1793
pb.update('Walking content', total, total)
1795
def _make_line_delta(self, delta_seq, new_content):
1796
"""Generate a line delta from delta_seq and new_content."""
1798
for op in delta_seq.get_opcodes():
1799
if op[0] == 'equal':
1801
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1804
def _merge_annotations(self, content, parents, parent_texts={},
1805
delta=None, annotated=None,
1806
left_matching_blocks=None):
1807
"""Merge annotations for content and generate deltas.
1809
This is done by comparing the annotations based on changes to the text
1810
and generating a delta on the resulting full texts. If annotations are
1811
not being created then a simple delta is created.
1813
if left_matching_blocks is not None:
1814
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1818
for parent_key in parents:
1819
merge_content = self._get_content(parent_key, parent_texts)
1820
if (parent_key == parents[0] and delta_seq is not None):
1823
seq = patiencediff.PatienceSequenceMatcher(
1824
None, merge_content.text(), content.text())
1825
for i, j, n in seq.get_matching_blocks():
1828
# this copies (origin, text) pairs across to the new
1829
# content for any line that matches the last-checked
1831
content._lines[j:j+n] = merge_content._lines[i:i+n]
1832
# XXX: Robert says the following block is a workaround for a
1833
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1834
if content._lines and content._lines[-1][1][-1] != '\n':
1835
# The copied annotation was from a line without a trailing EOL,
1836
# reinstate one for the content object, to ensure correct
1838
line = content._lines[-1][1] + '\n'
1839
content._lines[-1] = (content._lines[-1][0], line)
1841
if delta_seq is None:
1842
reference_content = self._get_content(parents[0], parent_texts)
1843
new_texts = content.text()
1844
old_texts = reference_content.text()
1845
delta_seq = patiencediff.PatienceSequenceMatcher(
1846
None, old_texts, new_texts)
1847
return self._make_line_delta(delta_seq, content)
1849
def _parse_record(self, version_id, data):
1850
"""Parse an original format knit record.
1852
These have the last element of the key only present in the stored data.
1854
rec, record_contents = self._parse_record_unchecked(data)
1855
self._check_header_version(rec, version_id)
1856
return record_contents, rec[3]
1858
def _parse_record_header(self, key, raw_data):
1859
"""Parse a record header for consistency.
1861
:return: the header and the decompressor stream.
1862
as (stream, header_record)
1864
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1867
rec = self._check_header(key, df.readline())
1868
except Exception, e:
1869
raise KnitCorrupt(self,
1870
"While reading {%s} got %s(%s)"
1871
% (key, e.__class__.__name__, str(e)))
1874
def _parse_record_unchecked(self, data):
1876
# 4168 calls in 2880 217 internal
1877
# 4168 calls to _parse_record_header in 2121
1878
# 4168 calls to readlines in 330
1879
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1881
record_contents = df.readlines()
1882
except Exception, e:
1883
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1884
(data, e.__class__.__name__, str(e)))
1885
header = record_contents.pop(0)
1886
rec = self._split_header(header)
1887
last_line = record_contents.pop()
1888
if len(record_contents) != int(rec[2]):
1889
raise KnitCorrupt(self,
1890
'incorrect number of lines %s != %s'
1891
' for version {%s} %s'
1892
% (len(record_contents), int(rec[2]),
1893
rec[1], record_contents))
1894
if last_line != 'end %s\n' % rec[1]:
1895
raise KnitCorrupt(self,
1896
'unexpected version end line %r, wanted %r'
1897
% (last_line, rec[1]))
1899
return rec, record_contents
1901
def _read_records_iter(self, records):
1902
"""Read text records from data file and yield result.
1904
The result will be returned in whatever is the fastest to read.
1905
Not by the order requested. Also, multiple requests for the same
1906
record will only yield 1 response.
1907
:param records: A list of (key, access_memo) entries
1908
:return: Yields (key, contents, digest) in the order
1909
read, not the order requested
1914
# XXX: This smells wrong, IO may not be getting ordered right.
1915
needed_records = sorted(set(records), key=operator.itemgetter(1))
1916
if not needed_records:
1919
# The transport optimizes the fetching as well
1920
# (ie, reads continuous ranges.)
1921
raw_data = self._access.get_raw_records(
1922
[index_memo for key, index_memo in needed_records])
1924
for (key, index_memo), data in \
1925
izip(iter(needed_records), raw_data):
1926
content, digest = self._parse_record(key[-1], data)
1927
yield key, content, digest
1929
def _read_records_iter_raw(self, records):
1930
"""Read text records from data file and yield raw data.
1932
This unpacks enough of the text record to validate the id is
1933
as expected but thats all.
1935
Each item the iterator yields is (key, bytes,
1936
expected_sha1_of_full_text).
1938
for key, data in self._read_records_iter_unchecked(records):
1939
# validate the header (note that we can only use the suffix in
1940
# current knit records).
1941
df, rec = self._parse_record_header(key, data)
1943
yield key, data, rec[3]
1945
def _read_records_iter_unchecked(self, records):
1946
"""Read text records from data file and yield raw data.
1948
No validation is done.
1950
Yields tuples of (key, data).
1952
# setup an iterator of the external records:
1953
# uses readv so nice and fast we hope.
1955
# grab the disk data needed.
1956
needed_offsets = [index_memo for key, index_memo
1958
raw_records = self._access.get_raw_records(needed_offsets)
1960
for key, index_memo in records:
1961
data = raw_records.next()
1964
def _record_to_data(self, key, digest, lines, dense_lines=None):
1965
"""Convert key, digest, lines into a raw data block.
1967
:param key: The key of the record. Currently keys are always serialised
1968
using just the trailing component.
1969
:param dense_lines: The bytes of lines but in a denser form. For
1970
instance, if lines is a list of 1000 bytestrings each ending in \n,
1971
dense_lines may be a list with one line in it, containing all the
1972
1000's lines and their \n's. Using dense_lines if it is already
1973
known is a win because the string join to create bytes in this
1974
function spends less time resizing the final string.
1975
:return: (len, a StringIO instance with the raw data ready to read.)
1977
chunks = ["version %s %d %s\n" % (key[-1], len(lines), digest)]
1978
chunks.extend(dense_lines or lines)
1979
chunks.append("end %s\n" % key[-1])
1980
for chunk in chunks:
1981
if type(chunk) is not str:
1982
raise AssertionError(
1983
'data must be plain bytes was %s' % type(chunk))
1984
if lines and lines[-1][-1] != '\n':
1985
raise ValueError('corrupt lines value %r' % lines)
1986
compressed_bytes = tuned_gzip.chunks_to_gzip(chunks)
1987
return len(compressed_bytes), compressed_bytes
1989
def _split_header(self, line):
1992
raise KnitCorrupt(self,
1993
'unexpected number of elements in record header')
1997
"""See VersionedFiles.keys."""
1998
if 'evil' in debug.debug_flags:
1999
trace.mutter_callsite(2, "keys scales with size of history")
2000
sources = [self._index] + self._fallback_vfs
2002
for source in sources:
2003
result.update(source.keys())
2007
class _ContentMapGenerator(object):
2008
"""Generate texts or expose raw deltas for a set of texts."""
2010
def __init__(self, ordering='unordered'):
2011
self._ordering = ordering
2013
def _get_content(self, key):
2014
"""Get the content object for key."""
2015
# Note that _get_content is only called when the _ContentMapGenerator
2016
# has been constructed with just one key requested for reconstruction.
2017
if key in self.nonlocal_keys:
2018
record = self.get_record_stream().next()
2019
# Create a content object on the fly
2020
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
2021
return PlainKnitContent(lines, record.key)
2023
# local keys we can ask for directly
2024
return self._get_one_work(key)
2026
def get_record_stream(self):
2027
"""Get a record stream for the keys requested during __init__."""
2028
for record in self._work():
2032
"""Produce maps of text and KnitContents as dicts.
2034
:return: (text_map, content_map) where text_map contains the texts for
2035
the requested versions and content_map contains the KnitContents.
2037
# NB: By definition we never need to read remote sources unless texts
2038
# are requested from them: we don't delta across stores - and we
2039
# explicitly do not want to to prevent data loss situations.
2040
if self.global_map is None:
2041
self.global_map = self.vf.get_parent_map(self.keys)
2042
nonlocal_keys = self.nonlocal_keys
2044
missing_keys = set(nonlocal_keys)
2045
# Read from remote versioned file instances and provide to our caller.
2046
for source in self.vf._fallback_vfs:
2047
if not missing_keys:
2049
# Loop over fallback repositories asking them for texts - ignore
2050
# any missing from a particular fallback.
2051
for record in source.get_record_stream(missing_keys,
2052
self._ordering, True):
2053
if record.storage_kind == 'absent':
2054
# Not in thie particular stream, may be in one of the
2055
# other fallback vfs objects.
2057
missing_keys.remove(record.key)
2060
if self._raw_record_map is None:
2061
raise AssertionError('_raw_record_map should have been filled')
2063
for key in self.keys:
2064
if key in self.nonlocal_keys:
2066
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2069
def _get_one_work(self, requested_key):
2070
# Now, if we have calculated everything already, just return the
2072
if requested_key in self._contents_map:
2073
return self._contents_map[requested_key]
2074
# To simplify things, parse everything at once - code that wants one text
2075
# probably wants them all.
2076
# FUTURE: This function could be improved for the 'extract many' case
2077
# by tracking each component and only doing the copy when the number of
2078
# children than need to apply delta's to it is > 1 or it is part of the
2080
multiple_versions = len(self.keys) != 1
2081
if self._record_map is None:
2082
self._record_map = self.vf._raw_map_to_record_map(
2083
self._raw_record_map)
2084
record_map = self._record_map
2085
# raw_record_map is key:
2086
# Have read and parsed records at this point.
2087
for key in self.keys:
2088
if key in self.nonlocal_keys:
2093
while cursor is not None:
2095
record, record_details, digest, next = record_map[cursor]
2097
raise RevisionNotPresent(cursor, self)
2098
components.append((cursor, record, record_details, digest))
2100
if cursor in self._contents_map:
2101
# no need to plan further back
2102
components.append((cursor, None, None, None))
2106
for (component_id, record, record_details,
2107
digest) in reversed(components):
2108
if component_id in self._contents_map:
2109
content = self._contents_map[component_id]
2111
content, delta = self._factory.parse_record(key[-1],
2112
record, record_details, content,
2113
copy_base_content=multiple_versions)
2114
if multiple_versions:
2115
self._contents_map[component_id] = content
2117
# digest here is the digest from the last applied component.
2118
text = content.text()
2119
actual_sha = sha_strings(text)
2120
if actual_sha != digest:
2121
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2122
if multiple_versions:
2123
return self._contents_map[requested_key]
2127
def _wire_bytes(self):
2128
"""Get the bytes to put on the wire for 'key'.
2130
The first collection of bytes asked for returns the serialised
2131
raw_record_map and the additional details (key, parent) for key.
2132
Subsequent calls return just the additional details (key, parent).
2133
The wire storage_kind given for the first key is 'knit-delta-closure',
2134
For subsequent keys it is 'knit-delta-closure-ref'.
2136
:param key: A key from the content generator.
2137
:return: Bytes to put on the wire.
2140
# kind marker for dispatch on the far side,
2141
lines.append('knit-delta-closure')
2143
if self.vf._factory.annotated:
2144
lines.append('annotated')
2147
# then the list of keys
2148
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2149
if key not in self.nonlocal_keys]))
2150
# then the _raw_record_map in serialised form:
2152
# for each item in the map:
2154
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2155
# one line with method
2156
# one line with noeol
2157
# one line with next ('' for None)
2158
# one line with byte count of the record bytes
2160
for key, (record_bytes, (method, noeol), next) in \
2161
self._raw_record_map.iteritems():
2162
key_bytes = '\x00'.join(key)
2163
parents = self.global_map.get(key, None)
2165
parent_bytes = 'None:'
2167
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2168
method_bytes = method
2174
next_bytes = '\x00'.join(next)
2177
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2178
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2179
len(record_bytes), record_bytes))
2180
map_bytes = ''.join(map_byte_list)
2181
lines.append(map_bytes)
2182
bytes = '\n'.join(lines)
2186
class _VFContentMapGenerator(_ContentMapGenerator):
2187
"""Content map generator reading from a VersionedFiles object."""
2189
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2190
global_map=None, raw_record_map=None, ordering='unordered'):
2191
"""Create a _ContentMapGenerator.
2193
:param versioned_files: The versioned files that the texts are being
2195
:param keys: The keys to produce content maps for.
2196
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2197
which are known to not be in this knit, but rather in one of the
2199
:param global_map: The result of get_parent_map(keys) (or a supermap).
2200
This is required if get_record_stream() is to be used.
2201
:param raw_record_map: A unparsed raw record map to use for answering
2204
_ContentMapGenerator.__init__(self, ordering=ordering)
2205
# The vf to source data from
2206
self.vf = versioned_files
2208
self.keys = list(keys)
2209
# Keys known to be in fallback vfs objects
2210
if nonlocal_keys is None:
2211
self.nonlocal_keys = set()
2213
self.nonlocal_keys = frozenset(nonlocal_keys)
2214
# Parents data for keys to be returned in get_record_stream
2215
self.global_map = global_map
2216
# The chunked lists for self.keys in text form
2218
# A cache of KnitContent objects used in extracting texts.
2219
self._contents_map = {}
2220
# All the knit records needed to assemble the requested keys as full
2222
self._record_map = None
2223
if raw_record_map is None:
2224
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2227
self._raw_record_map = raw_record_map
2228
# the factory for parsing records
2229
self._factory = self.vf._factory
2232
class _NetworkContentMapGenerator(_ContentMapGenerator):
2233
"""Content map generator sourced from a network stream."""
2235
def __init__(self, bytes, line_end):
2236
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2238
self.global_map = {}
2239
self._raw_record_map = {}
2240
self._contents_map = {}
2241
self._record_map = None
2242
self.nonlocal_keys = []
2243
# Get access to record parsing facilities
2244
self.vf = KnitVersionedFiles(None, None)
2247
line_end = bytes.find('\n', start)
2248
line = bytes[start:line_end]
2249
start = line_end + 1
2250
if line == 'annotated':
2251
self._factory = KnitAnnotateFactory()
2253
self._factory = KnitPlainFactory()
2254
# list of keys to emit in get_record_stream
2255
line_end = bytes.find('\n', start)
2256
line = bytes[start:line_end]
2257
start = line_end + 1
2259
tuple(segment.split('\x00')) for segment in line.split('\t')
2261
# now a loop until the end. XXX: It would be nice if this was just a
2262
# bunch of the same records as get_record_stream(..., False) gives, but
2263
# there is a decent sized gap stopping that at the moment.
2267
line_end = bytes.find('\n', start)
2268
key = tuple(bytes[start:line_end].split('\x00'))
2269
start = line_end + 1
2270
# 1 line with parents (None: for None, '' for ())
2271
line_end = bytes.find('\n', start)
2272
line = bytes[start:line_end]
2277
[tuple(segment.split('\x00')) for segment in line.split('\t')
2279
self.global_map[key] = parents
2280
start = line_end + 1
2281
# one line with method
2282
line_end = bytes.find('\n', start)
2283
line = bytes[start:line_end]
2285
start = line_end + 1
2286
# one line with noeol
2287
line_end = bytes.find('\n', start)
2288
line = bytes[start:line_end]
2290
start = line_end + 1
2291
# one line with next ('' for None)
2292
line_end = bytes.find('\n', start)
2293
line = bytes[start:line_end]
2297
next = tuple(bytes[start:line_end].split('\x00'))
2298
start = line_end + 1
2299
# one line with byte count of the record bytes
2300
line_end = bytes.find('\n', start)
2301
line = bytes[start:line_end]
2303
start = line_end + 1
2305
record_bytes = bytes[start:start+count]
2306
start = start + count
2308
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2310
def get_record_stream(self):
2311
"""Get a record stream for for keys requested by the bytestream."""
2313
for key in self.keys:
2314
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2317
def _wire_bytes(self):
2321
class _KndxIndex(object):
2322
"""Manages knit index files
2324
The index is kept in memory and read on startup, to enable
2325
fast lookups of revision information. The cursor of the index
2326
file is always pointing to the end, making it easy to append
2329
_cache is a cache for fast mapping from version id to a Index
2332
_history is a cache for fast mapping from indexes to version ids.
2334
The index data format is dictionary compressed when it comes to
2335
parent references; a index entry may only have parents that with a
2336
lover index number. As a result, the index is topological sorted.
2338
Duplicate entries may be written to the index for a single version id
2339
if this is done then the latter one completely replaces the former:
2340
this allows updates to correct version and parent information.
2341
Note that the two entries may share the delta, and that successive
2342
annotations and references MUST point to the first entry.
2344
The index file on disc contains a header, followed by one line per knit
2345
record. The same revision can be present in an index file more than once.
2346
The first occurrence gets assigned a sequence number starting from 0.
2348
The format of a single line is
2349
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
2350
REVISION_ID is a utf8-encoded revision id
2351
FLAGS is a comma separated list of flags about the record. Values include
2352
no-eol, line-delta, fulltext.
2353
BYTE_OFFSET is the ascii representation of the byte offset in the data file
2354
that the the compressed data starts at.
2355
LENGTH is the ascii representation of the length of the data file.
2356
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
2358
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
2359
revision id already in the knit that is a parent of REVISION_ID.
2360
The ' :' marker is the end of record marker.
2363
when a write is interrupted to the index file, it will result in a line
2364
that does not end in ' :'. If the ' :' is not present at the end of a line,
2365
or at the end of the file, then the record that is missing it will be
2366
ignored by the parser.
2368
When writing new records to the index file, the data is preceded by '\n'
2369
to ensure that records always start on new lines even if the last write was
2370
interrupted. As a result its normal for the last line in the index to be
2371
missing a trailing newline. One can be added with no harmful effects.
2373
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2374
where prefix is e.g. the (fileid,) for .texts instances or () for
2375
constant-mapped things like .revisions, and the old state is
2376
tuple(history_vector, cache_dict). This is used to prevent having an
2377
ABI change with the C extension that reads .kndx files.
2380
HEADER = "# bzr knit index 8\n"
2382
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2383
"""Create a _KndxIndex on transport using mapper."""
2384
self._transport = transport
2385
self._mapper = mapper
2386
self._get_scope = get_scope
2387
self._allow_writes = allow_writes
2388
self._is_locked = is_locked
2390
self.has_graph = True
2392
def add_records(self, records, random_id=False, missing_compression_parents=False):
2393
"""Add multiple records to the index.
2395
:param records: a list of tuples:
2396
(key, options, access_memo, parents).
2397
:param random_id: If True the ids being added were randomly generated
2398
and no check for existence will be performed.
2399
:param missing_compression_parents: If True the records being added are
2400
only compressed against texts already in the index (or inside
2401
records). If False the records all refer to unavailable texts (or
2402
texts inside records) as compression parents.
2404
if missing_compression_parents:
2405
# It might be nice to get the edge of the records. But keys isn't
2407
keys = sorted(record[0] for record in records)
2408
raise errors.RevisionNotPresent(keys, self)
2410
for record in records:
2413
path = self._mapper.map(key) + '.kndx'
2414
path_keys = paths.setdefault(path, (prefix, []))
2415
path_keys[1].append(record)
2416
for path in sorted(paths):
2417
prefix, path_keys = paths[path]
2418
self._load_prefixes([prefix])
2420
orig_history = self._kndx_cache[prefix][1][:]
2421
orig_cache = self._kndx_cache[prefix][0].copy()
2424
for key, options, (_, pos, size), parents in path_keys:
2426
# kndx indices cannot be parentless.
2428
line = "\n%s %s %s %s %s :" % (
2429
key[-1], ','.join(options), pos, size,
2430
self._dictionary_compress(parents))
2431
if type(line) is not str:
2432
raise AssertionError(
2433
'data must be utf8 was %s' % type(line))
2435
self._cache_key(key, options, pos, size, parents)
2436
if len(orig_history):
2437
self._transport.append_bytes(path, ''.join(lines))
2439
self._init_index(path, lines)
2441
# If any problems happen, restore the original values and re-raise
2442
self._kndx_cache[prefix] = (orig_cache, orig_history)
2445
def scan_unvalidated_index(self, graph_index):
2446
"""See _KnitGraphIndex.scan_unvalidated_index."""
2447
# Because kndx files do not support atomic insertion via separate index
2448
# files, they do not support this method.
2449
raise NotImplementedError(self.scan_unvalidated_index)
2451
def get_missing_compression_parents(self):
2452
"""See _KnitGraphIndex.get_missing_compression_parents."""
2453
# Because kndx files do not support atomic insertion via separate index
2454
# files, they do not support this method.
2455
raise NotImplementedError(self.get_missing_compression_parents)
2457
def _cache_key(self, key, options, pos, size, parent_keys):
2458
"""Cache a version record in the history array and index cache.
2460
This is inlined into _load_data for performance. KEEP IN SYNC.
2461
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2465
version_id = key[-1]
2466
# last-element only for compatibilty with the C load_data.
2467
parents = tuple(parent[-1] for parent in parent_keys)
2468
for parent in parent_keys:
2469
if parent[:-1] != prefix:
2470
raise ValueError("mismatched prefixes for %r, %r" % (
2472
cache, history = self._kndx_cache[prefix]
2473
# only want the _history index to reference the 1st index entry
2475
if version_id not in cache:
2476
index = len(history)
2477
history.append(version_id)
2479
index = cache[version_id][5]
2480
cache[version_id] = (version_id,
2487
def check_header(self, fp):
2488
line = fp.readline()
2490
# An empty file can actually be treated as though the file doesn't
2492
raise errors.NoSuchFile(self)
2493
if line != self.HEADER:
2494
raise KnitHeaderError(badline=line, filename=self)
2496
def _check_read(self):
2497
if not self._is_locked():
2498
raise errors.ObjectNotLocked(self)
2499
if self._get_scope() != self._scope:
2502
def _check_write_ok(self):
2503
"""Assert if not writes are permitted."""
2504
if not self._is_locked():
2505
raise errors.ObjectNotLocked(self)
2506
if self._get_scope() != self._scope:
2508
if self._mode != 'w':
2509
raise errors.ReadOnlyObjectDirtiedError(self)
2511
def get_build_details(self, keys):
2512
"""Get the method, index_memo and compression parent for keys.
2514
Ghosts are omitted from the result.
2516
:param keys: An iterable of keys.
2517
:return: A dict of key:(index_memo, compression_parent, parents,
2520
opaque structure to pass to read_records to extract the raw
2523
Content that this record is built upon, may be None
2525
Logical parents of this node
2527
extra information about the content which needs to be passed to
2528
Factory.parse_record
2530
parent_map = self.get_parent_map(keys)
2533
if key not in parent_map:
2535
method = self.get_method(key)
2536
parents = parent_map[key]
2537
if method == 'fulltext':
2538
compression_parent = None
2540
compression_parent = parents[0]
2541
noeol = 'no-eol' in self.get_options(key)
2542
index_memo = self.get_position(key)
2543
result[key] = (index_memo, compression_parent,
2544
parents, (method, noeol))
2547
def get_method(self, key):
2548
"""Return compression method of specified key."""
2549
options = self.get_options(key)
2550
if 'fulltext' in options:
2552
elif 'line-delta' in options:
2555
raise errors.KnitIndexUnknownMethod(self, options)
2557
def get_options(self, key):
2558
"""Return a list representing options.
2562
prefix, suffix = self._split_key(key)
2563
self._load_prefixes([prefix])
2565
return self._kndx_cache[prefix][0][suffix][1]
2567
raise RevisionNotPresent(key, self)
2569
def find_ancestry(self, keys):
2570
"""See CombinedGraphIndex.find_ancestry()"""
2571
prefixes = set(key[:-1] for key in keys)
2572
self._load_prefixes(prefixes)
2575
missing_keys = set()
2576
pending_keys = list(keys)
2577
# This assumes that keys will not reference parents in a different
2578
# prefix, which is accurate so far.
2580
key = pending_keys.pop()
2581
if key in parent_map:
2585
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2587
missing_keys.add(key)
2589
parent_keys = tuple([prefix + (suffix,)
2590
for suffix in suffix_parents])
2591
parent_map[key] = parent_keys
2592
pending_keys.extend([p for p in parent_keys
2593
if p not in parent_map])
2594
return parent_map, missing_keys
2596
def get_parent_map(self, keys):
2597
"""Get a map of the parents of keys.
2599
:param keys: The keys to look up parents for.
2600
:return: A mapping from keys to parents. Absent keys are absent from
2603
# Parse what we need to up front, this potentially trades off I/O
2604
# locality (.kndx and .knit in the same block group for the same file
2605
# id) for less checking in inner loops.
2606
prefixes = set(key[:-1] for key in keys)
2607
self._load_prefixes(prefixes)
2612
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2616
result[key] = tuple(prefix + (suffix,) for
2617
suffix in suffix_parents)
2620
def get_position(self, key):
2621
"""Return details needed to access the version.
2623
:return: a tuple (key, data position, size) to hand to the access
2624
logic to get the record.
2626
prefix, suffix = self._split_key(key)
2627
self._load_prefixes([prefix])
2628
entry = self._kndx_cache[prefix][0][suffix]
2629
return key, entry[2], entry[3]
2631
has_key = _mod_index._has_key_from_parent_map
2633
def _init_index(self, path, extra_lines=[]):
2634
"""Initialize an index."""
2636
sio.write(self.HEADER)
2637
sio.writelines(extra_lines)
2639
self._transport.put_file_non_atomic(path, sio,
2640
create_parent_dir=True)
2641
# self._create_parent_dir)
2642
# mode=self._file_mode,
2643
# dir_mode=self._dir_mode)
2646
"""Get all the keys in the collection.
2648
The keys are not ordered.
2651
# Identify all key prefixes.
2652
# XXX: A bit hacky, needs polish.
2653
if type(self._mapper) is ConstantMapper:
2657
for quoted_relpath in self._transport.iter_files_recursive():
2658
path, ext = os.path.splitext(quoted_relpath)
2660
prefixes = [self._mapper.unmap(path) for path in relpaths]
2661
self._load_prefixes(prefixes)
2662
for prefix in prefixes:
2663
for suffix in self._kndx_cache[prefix][1]:
2664
result.add(prefix + (suffix,))
2667
def _load_prefixes(self, prefixes):
2668
"""Load the indices for prefixes."""
2670
for prefix in prefixes:
2671
if prefix not in self._kndx_cache:
2672
# the load_data interface writes to these variables.
2675
self._filename = prefix
2677
path = self._mapper.map(prefix) + '.kndx'
2678
fp = self._transport.get(path)
2680
# _load_data may raise NoSuchFile if the target knit is
2682
_load_data(self, fp)
2685
self._kndx_cache[prefix] = (self._cache, self._history)
2690
self._kndx_cache[prefix] = ({}, [])
2691
if type(self._mapper) is ConstantMapper:
2692
# preserve behaviour for revisions.kndx etc.
2693
self._init_index(path)
2698
missing_keys = _mod_index._missing_keys_from_parent_map
2700
def _partition_keys(self, keys):
2701
"""Turn keys into a dict of prefix:suffix_list."""
2704
prefix_keys = result.setdefault(key[:-1], [])
2705
prefix_keys.append(key[-1])
2708
def _dictionary_compress(self, keys):
2709
"""Dictionary compress keys.
2711
:param keys: The keys to generate references to.
2712
:return: A string representation of keys. keys which are present are
2713
dictionary compressed, and others are emitted as fulltext with a
2719
prefix = keys[0][:-1]
2720
cache = self._kndx_cache[prefix][0]
2722
if key[:-1] != prefix:
2723
# kndx indices cannot refer across partitioned storage.
2724
raise ValueError("mismatched prefixes for %r" % keys)
2725
if key[-1] in cache:
2726
# -- inlined lookup() --
2727
result_list.append(str(cache[key[-1]][5]))
2728
# -- end lookup () --
2730
result_list.append('.' + key[-1])
2731
return ' '.join(result_list)
2733
def _reset_cache(self):
2734
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2735
# (cache_dict, history_vector) for parsed kndx files.
2736
self._kndx_cache = {}
2737
self._scope = self._get_scope()
2738
allow_writes = self._allow_writes()
2744
def _sort_keys_by_io(self, keys, positions):
2745
"""Figure out an optimal order to read the records for the given keys.
2747
Sort keys, grouped by index and sorted by position.
2749
:param keys: A list of keys whose records we want to read. This will be
2751
:param positions: A dict, such as the one returned by
2752
_get_components_positions()
2755
def get_sort_key(key):
2756
index_memo = positions[key][1]
2757
# Group by prefix and position. index_memo[0] is the key, so it is
2758
# (file_id, revision_id) and we don't want to sort on revision_id,
2759
# index_memo[1] is the position, and index_memo[2] is the size,
2760
# which doesn't matter for the sort
2761
return index_memo[0][:-1], index_memo[1]
2762
return keys.sort(key=get_sort_key)
2764
_get_total_build_size = _get_total_build_size
2766
def _split_key(self, key):
2767
"""Split key into a prefix and suffix."""
2768
return key[:-1], key[-1]
2771
class _KeyRefs(object):
2774
# dict mapping 'key' to 'set of keys referring to that key'
2777
def add_references(self, key, refs):
2778
# Record the new references
2779
for referenced in refs:
2781
needed_by = self.refs[referenced]
2783
needed_by = self.refs[referenced] = set()
2785
# Discard references satisfied by the new key
2788
def get_unsatisfied_refs(self):
2789
return self.refs.iterkeys()
2791
def add_key(self, key):
2795
# No keys depended on this key. That's ok.
2798
def add_keys(self, keys):
2802
def get_referrers(self):
2804
for referrers in self.refs.itervalues():
2805
result.update(referrers)
2809
class _KnitGraphIndex(object):
2810
"""A KnitVersionedFiles index layered on GraphIndex."""
2812
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2813
add_callback=None, track_external_parent_refs=False):
2814
"""Construct a KnitGraphIndex on a graph_index.
2816
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2817
:param is_locked: A callback to check whether the object should answer
2819
:param deltas: Allow delta-compressed records.
2820
:param parents: If True, record knits parents, if not do not record
2822
:param add_callback: If not None, allow additions to the index and call
2823
this callback with a list of added GraphIndex nodes:
2824
[(node, value, node_refs), ...]
2825
:param is_locked: A callback, returns True if the index is locked and
2827
:param track_external_parent_refs: If True, record all external parent
2828
references parents from added records. These can be retrieved
2829
later by calling get_missing_parents().
2831
self._add_callback = add_callback
2832
self._graph_index = graph_index
2833
self._deltas = deltas
2834
self._parents = parents
2835
if deltas and not parents:
2836
# XXX: TODO: Delta tree and parent graph should be conceptually
2838
raise KnitCorrupt(self, "Cannot do delta compression without "
2840
self.has_graph = parents
2841
self._is_locked = is_locked
2842
self._missing_compression_parents = set()
2843
if track_external_parent_refs:
2844
self._key_dependencies = _KeyRefs()
2846
self._key_dependencies = None
2849
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2851
def add_records(self, records, random_id=False,
2852
missing_compression_parents=False):
2853
"""Add multiple records to the index.
2855
This function does not insert data into the Immutable GraphIndex
2856
backing the KnitGraphIndex, instead it prepares data for insertion by
2857
the caller and checks that it is safe to insert then calls
2858
self._add_callback with the prepared GraphIndex nodes.
2860
:param records: a list of tuples:
2861
(key, options, access_memo, parents).
2862
:param random_id: If True the ids being added were randomly generated
2863
and no check for existence will be performed.
2864
:param missing_compression_parents: If True the records being added are
2865
only compressed against texts already in the index (or inside
2866
records). If False the records all refer to unavailable texts (or
2867
texts inside records) as compression parents.
2869
if not self._add_callback:
2870
raise errors.ReadOnlyError(self)
2871
# we hope there are no repositories with inconsistent parentage
2875
compression_parents = set()
2876
key_dependencies = self._key_dependencies
2877
for (key, options, access_memo, parents) in records:
2879
parents = tuple(parents)
2880
if key_dependencies is not None:
2881
key_dependencies.add_references(key, parents)
2882
index, pos, size = access_memo
2883
if 'no-eol' in options:
2887
value += "%d %d" % (pos, size)
2888
if not self._deltas:
2889
if 'line-delta' in options:
2890
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2893
if 'line-delta' in options:
2894
node_refs = (parents, (parents[0],))
2895
if missing_compression_parents:
2896
compression_parents.add(parents[0])
2898
node_refs = (parents, ())
2900
node_refs = (parents, )
2903
raise KnitCorrupt(self, "attempt to add node with parents "
2904
"in parentless index.")
2906
keys[key] = (value, node_refs)
2909
present_nodes = self._get_entries(keys)
2910
for (index, key, value, node_refs) in present_nodes:
2911
if (value[0] != keys[key][0][0] or
2912
node_refs[:1] != keys[key][1][:1]):
2913
raise KnitCorrupt(self, "inconsistent details in add_records"
2914
": %s %s" % ((value, node_refs), keys[key]))
2918
for key, (value, node_refs) in keys.iteritems():
2919
result.append((key, value, node_refs))
2921
for key, (value, node_refs) in keys.iteritems():
2922
result.append((key, value))
2923
self._add_callback(result)
2924
if missing_compression_parents:
2925
# This may appear to be incorrect (it does not check for
2926
# compression parents that are in the existing graph index),
2927
# but such records won't have been buffered, so this is
2928
# actually correct: every entry when
2929
# missing_compression_parents==True either has a missing parent, or
2930
# a parent that is one of the keys in records.
2931
compression_parents.difference_update(keys)
2932
self._missing_compression_parents.update(compression_parents)
2933
# Adding records may have satisfied missing compression parents.
2934
self._missing_compression_parents.difference_update(keys)
2936
def scan_unvalidated_index(self, graph_index):
2937
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2939
This allows this _KnitGraphIndex to keep track of any missing
2940
compression parents we may want to have filled in to make those
2943
:param graph_index: A GraphIndex
2946
new_missing = graph_index.external_references(ref_list_num=1)
2947
new_missing.difference_update(self.get_parent_map(new_missing))
2948
self._missing_compression_parents.update(new_missing)
2949
if self._key_dependencies is not None:
2950
# Add parent refs from graph_index (and discard parent refs that
2951
# the graph_index has).
2952
for node in graph_index.iter_all_entries():
2953
self._key_dependencies.add_references(node[1], node[3][0])
2955
def get_missing_compression_parents(self):
2956
"""Return the keys of missing compression parents.
2958
Missing compression parents occur when a record stream was missing
2959
basis texts, or a index was scanned that had missing basis texts.
2961
return frozenset(self._missing_compression_parents)
2963
def get_missing_parents(self):
2964
"""Return the keys of missing parents."""
2965
# If updating this, you should also update
2966
# groupcompress._GCGraphIndex.get_missing_parents
2967
# We may have false positives, so filter those out.
2968
self._key_dependencies.add_keys(
2969
self.get_parent_map(self._key_dependencies.get_unsatisfied_refs()))
2970
return frozenset(self._key_dependencies.get_unsatisfied_refs())
2972
def _check_read(self):
2973
"""raise if reads are not permitted."""
2974
if not self._is_locked():
2975
raise errors.ObjectNotLocked(self)
2977
def _check_write_ok(self):
2978
"""Assert if writes are not permitted."""
2979
if not self._is_locked():
2980
raise errors.ObjectNotLocked(self)
2982
def _compression_parent(self, an_entry):
2983
# return the key that an_entry is compressed against, or None
2984
# Grab the second parent list (as deltas implies parents currently)
2985
compression_parents = an_entry[3][1]
2986
if not compression_parents:
2988
if len(compression_parents) != 1:
2989
raise AssertionError(
2990
"Too many compression parents: %r" % compression_parents)
2991
return compression_parents[0]
2993
def get_build_details(self, keys):
2994
"""Get the method, index_memo and compression parent for version_ids.
2996
Ghosts are omitted from the result.
2998
:param keys: An iterable of keys.
2999
:return: A dict of key:
3000
(index_memo, compression_parent, parents, record_details).
3002
opaque structure to pass to read_records to extract the raw
3005
Content that this record is built upon, may be None
3007
Logical parents of this node
3009
extra information about the content which needs to be passed to
3010
Factory.parse_record
3014
entries = self._get_entries(keys, False)
3015
for entry in entries:
3017
if not self._parents:
3020
parents = entry[3][0]
3021
if not self._deltas:
3022
compression_parent_key = None
3024
compression_parent_key = self._compression_parent(entry)
3025
noeol = (entry[2][0] == 'N')
3026
if compression_parent_key:
3027
method = 'line-delta'
3030
result[key] = (self._node_to_position(entry),
3031
compression_parent_key, parents,
3035
def _get_entries(self, keys, check_present=False):
3036
"""Get the entries for keys.
3038
:param keys: An iterable of index key tuples.
3043
for node in self._graph_index.iter_entries(keys):
3045
found_keys.add(node[1])
3047
# adapt parentless index to the rest of the code.
3048
for node in self._graph_index.iter_entries(keys):
3049
yield node[0], node[1], node[2], ()
3050
found_keys.add(node[1])
3052
missing_keys = keys.difference(found_keys)
3054
raise RevisionNotPresent(missing_keys.pop(), self)
3056
def get_method(self, key):
3057
"""Return compression method of specified key."""
3058
return self._get_method(self._get_node(key))
3060
def _get_method(self, node):
3061
if not self._deltas:
3063
if self._compression_parent(node):
3068
def _get_node(self, key):
3070
return list(self._get_entries([key]))[0]
3072
raise RevisionNotPresent(key, self)
3074
def get_options(self, key):
3075
"""Return a list representing options.
3079
node = self._get_node(key)
3080
options = [self._get_method(node)]
3081
if node[2][0] == 'N':
3082
options.append('no-eol')
3085
def find_ancestry(self, keys):
3086
"""See CombinedGraphIndex.find_ancestry()"""
3087
return self._graph_index.find_ancestry(keys, 0)
3089
def get_parent_map(self, keys):
3090
"""Get a map of the parents of keys.
3092
:param keys: The keys to look up parents for.
3093
:return: A mapping from keys to parents. Absent keys are absent from
3097
nodes = self._get_entries(keys)
3101
result[node[1]] = node[3][0]
3104
result[node[1]] = None
3107
def get_position(self, key):
3108
"""Return details needed to access the version.
3110
:return: a tuple (index, data position, size) to hand to the access
3111
logic to get the record.
3113
node = self._get_node(key)
3114
return self._node_to_position(node)
3116
has_key = _mod_index._has_key_from_parent_map
3119
"""Get all the keys in the collection.
3121
The keys are not ordered.
3124
return [node[1] for node in self._graph_index.iter_all_entries()]
3126
missing_keys = _mod_index._missing_keys_from_parent_map
3128
def _node_to_position(self, node):
3129
"""Convert an index value to position details."""
3130
bits = node[2][1:].split(' ')
3131
return node[0], int(bits[0]), int(bits[1])
3133
def _sort_keys_by_io(self, keys, positions):
3134
"""Figure out an optimal order to read the records for the given keys.
3136
Sort keys, grouped by index and sorted by position.
3138
:param keys: A list of keys whose records we want to read. This will be
3140
:param positions: A dict, such as the one returned by
3141
_get_components_positions()
3144
def get_index_memo(key):
3145
# index_memo is at offset [1]. It is made up of (GraphIndex,
3146
# position, size). GI is an object, which will be unique for each
3147
# pack file. This causes us to group by pack file, then sort by
3148
# position. Size doesn't matter, but it isn't worth breaking up the
3150
return positions[key][1]
3151
return keys.sort(key=get_index_memo)
3153
_get_total_build_size = _get_total_build_size
3156
class _KnitKeyAccess(object):
3157
"""Access to records in .knit files."""
3159
def __init__(self, transport, mapper):
3160
"""Create a _KnitKeyAccess with transport and mapper.
3162
:param transport: The transport the access object is rooted at.
3163
:param mapper: The mapper used to map keys to .knit files.
3165
self._transport = transport
3166
self._mapper = mapper
3168
def add_raw_records(self, key_sizes, raw_data):
3169
"""Add raw knit bytes to a storage area.
3171
The data is spooled to the container writer in one bytes-record per
3174
:param sizes: An iterable of tuples containing the key and size of each
3176
:param raw_data: A bytestring containing the data.
3177
:return: A list of memos to retrieve the record later. Each memo is an
3178
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3179
length), where the key is the record key.
3181
if type(raw_data) is not str:
3182
raise AssertionError(
3183
'data must be plain bytes was %s' % type(raw_data))
3186
# TODO: This can be tuned for writing to sftp and other servers where
3187
# append() is relatively expensive by grouping the writes to each key
3189
for key, size in key_sizes:
3190
path = self._mapper.map(key)
3192
base = self._transport.append_bytes(path + '.knit',
3193
raw_data[offset:offset+size])
3194
except errors.NoSuchFile:
3195
self._transport.mkdir(osutils.dirname(path))
3196
base = self._transport.append_bytes(path + '.knit',
3197
raw_data[offset:offset+size])
3201
result.append((key, base, size))
3205
"""Flush pending writes on this access object.
3207
For .knit files this is a no-op.
3211
def get_raw_records(self, memos_for_retrieval):
3212
"""Get the raw bytes for a records.
3214
:param memos_for_retrieval: An iterable containing the access memo for
3215
retrieving the bytes.
3216
:return: An iterator over the bytes of the records.
3218
# first pass, group into same-index request to minimise readv's issued.
3220
current_prefix = None
3221
for (key, offset, length) in memos_for_retrieval:
3222
if current_prefix == key[:-1]:
3223
current_list.append((offset, length))
3225
if current_prefix is not None:
3226
request_lists.append((current_prefix, current_list))
3227
current_prefix = key[:-1]
3228
current_list = [(offset, length)]
3229
# handle the last entry
3230
if current_prefix is not None:
3231
request_lists.append((current_prefix, current_list))
3232
for prefix, read_vector in request_lists:
3233
path = self._mapper.map(prefix) + '.knit'
3234
for pos, data in self._transport.readv(path, read_vector):
3238
class _DirectPackAccess(object):
3239
"""Access to data in one or more packs with less translation."""
3241
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
3242
"""Create a _DirectPackAccess object.
3244
:param index_to_packs: A dict mapping index objects to the transport
3245
and file names for obtaining data.
3246
:param reload_func: A function to call if we determine that the pack
3247
files have moved and we need to reload our caches. See
3248
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
3250
self._container_writer = None
3251
self._write_index = None
3252
self._indices = index_to_packs
3253
self._reload_func = reload_func
3254
self._flush_func = flush_func
3256
def add_raw_records(self, key_sizes, raw_data):
3257
"""Add raw knit bytes to a storage area.
3259
The data is spooled to the container writer in one bytes-record per
3262
:param sizes: An iterable of tuples containing the key and size of each
3264
:param raw_data: A bytestring containing the data.
3265
:return: A list of memos to retrieve the record later. Each memo is an
3266
opaque index memo. For _DirectPackAccess the memo is (index, pos,
3267
length), where the index field is the write_index object supplied
3268
to the PackAccess object.
3270
if type(raw_data) is not str:
3271
raise AssertionError(
3272
'data must be plain bytes was %s' % type(raw_data))
3275
for key, size in key_sizes:
3276
p_offset, p_length = self._container_writer.add_bytes_record(
3277
raw_data[offset:offset+size], [])
3279
result.append((self._write_index, p_offset, p_length))
3283
"""Flush pending writes on this access object.
3285
This will flush any buffered writes to a NewPack.
3287
if self._flush_func is not None:
3290
def get_raw_records(self, memos_for_retrieval):
3291
"""Get the raw bytes for a records.
3293
:param memos_for_retrieval: An iterable containing the (index, pos,
3294
length) memo for retrieving the bytes. The Pack access method
3295
looks up the pack to use for a given record in its index_to_pack
3297
:return: An iterator over the bytes of the records.
3299
# first pass, group into same-index requests
3301
current_index = None
3302
for (index, offset, length) in memos_for_retrieval:
3303
if current_index == index:
3304
current_list.append((offset, length))
3306
if current_index is not None:
3307
request_lists.append((current_index, current_list))
3308
current_index = index
3309
current_list = [(offset, length)]
3310
# handle the last entry
3311
if current_index is not None:
3312
request_lists.append((current_index, current_list))
3313
for index, offsets in request_lists:
3315
transport, path = self._indices[index]
3317
# A KeyError here indicates that someone has triggered an index
3318
# reload, and this index has gone missing, we need to start
3320
if self._reload_func is None:
3321
# If we don't have a _reload_func there is nothing that can
3324
raise errors.RetryWithNewPacks(index,
3325
reload_occurred=True,
3326
exc_info=sys.exc_info())
3328
reader = pack.make_readv_reader(transport, path, offsets)
3329
for names, read_func in reader.iter_records():
3330
yield read_func(None)
3331
except errors.NoSuchFile:
3332
# A NoSuchFile error indicates that a pack file has gone
3333
# missing on disk, we need to trigger a reload, and start over.
3334
if self._reload_func is None:
3336
raise errors.RetryWithNewPacks(transport.abspath(path),
3337
reload_occurred=False,
3338
exc_info=sys.exc_info())
3340
def set_writer(self, writer, index, transport_packname):
3341
"""Set a writer to use for adding data."""
3342
if index is not None:
3343
self._indices[index] = transport_packname
3344
self._container_writer = writer
3345
self._write_index = index
3347
def reload_or_raise(self, retry_exc):
3348
"""Try calling the reload function, or re-raise the original exception.
3350
This should be called after _DirectPackAccess raises a
3351
RetryWithNewPacks exception. This function will handle the common logic
3352
of determining when the error is fatal versus being temporary.
3353
It will also make sure that the original exception is raised, rather
3354
than the RetryWithNewPacks exception.
3356
If this function returns, then the calling function should retry
3357
whatever operation was being performed. Otherwise an exception will
3360
:param retry_exc: A RetryWithNewPacks exception.
3363
if self._reload_func is None:
3365
elif not self._reload_func():
3366
# The reload claimed that nothing changed
3367
if not retry_exc.reload_occurred:
3368
# If there wasn't an earlier reload, then we really were
3369
# expecting to find changes. We didn't find them, so this is a
3373
exc_class, exc_value, exc_traceback = retry_exc.exc_info
3374
raise exc_class, exc_value, exc_traceback
3377
# Deprecated, use PatienceSequenceMatcher instead
3378
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
3381
def annotate_knit(knit, revision_id):
3382
"""Annotate a knit with no cached annotations.
3384
This implementation is for knits with no cached annotations.
3385
It will work for knits with cached annotations, but this is not
3388
annotator = _KnitAnnotator(knit)
3389
return iter(annotator.annotate_flat(revision_id))
3392
class _KnitAnnotator(annotate.Annotator):
3393
"""Build up the annotations for a text."""
3395
def __init__(self, vf):
3396
annotate.Annotator.__init__(self, vf)
3398
# TODO: handle Nodes which cannot be extracted
3399
# self._ghosts = set()
3401
# Map from (key, parent_key) => matching_blocks, should be 'use once'
3402
self._matching_blocks = {}
3404
# KnitContent objects
3405
self._content_objects = {}
3406
# The number of children that depend on this fulltext content object
3407
self._num_compression_children = {}
3408
# Delta records that need their compression parent before they can be
3410
self._pending_deltas = {}
3411
# Fulltext records that are waiting for their parents fulltexts before
3412
# they can be yielded for annotation
3413
self._pending_annotation = {}
3415
self._all_build_details = {}
3417
def _get_build_graph(self, key):
3418
"""Get the graphs for building texts and annotations.
3420
The data you need for creating a full text may be different than the
3421
data you need to annotate that text. (At a minimum, you need both
3422
parents to create an annotation, but only need 1 parent to generate the
3425
:return: A list of (key, index_memo) records, suitable for
3426
passing to read_records_iter to start reading in the raw data from
3429
pending = set([key])
3432
self._num_needed_children[key] = 1
3434
# get all pending nodes
3435
this_iteration = pending
3436
build_details = self._vf._index.get_build_details(this_iteration)
3437
self._all_build_details.update(build_details)
3438
# new_nodes = self._vf._index._get_entries(this_iteration)
3440
for key, details in build_details.iteritems():
3441
(index_memo, compression_parent, parent_keys,
3442
record_details) = details
3443
self._parent_map[key] = parent_keys
3444
self._heads_provider = None
3445
records.append((key, index_memo))
3446
# Do we actually need to check _annotated_lines?
3447
pending.update([p for p in parent_keys
3448
if p not in self._all_build_details])
3450
for parent_key in parent_keys:
3451
if parent_key in self._num_needed_children:
3452
self._num_needed_children[parent_key] += 1
3454
self._num_needed_children[parent_key] = 1
3455
if compression_parent:
3456
if compression_parent in self._num_compression_children:
3457
self._num_compression_children[compression_parent] += 1
3459
self._num_compression_children[compression_parent] = 1
3461
missing_versions = this_iteration.difference(build_details.keys())
3462
if missing_versions:
3463
for key in missing_versions:
3464
if key in self._parent_map and key in self._text_cache:
3465
# We already have this text ready, we just need to
3466
# yield it later so we get it annotated
3468
parent_keys = self._parent_map[key]
3469
for parent_key in parent_keys:
3470
if parent_key in self._num_needed_children:
3471
self._num_needed_children[parent_key] += 1
3473
self._num_needed_children[parent_key] = 1
3474
pending.update([p for p in parent_keys
3475
if p not in self._all_build_details])
3477
raise errors.RevisionNotPresent(key, self._vf)
3478
# Generally we will want to read the records in reverse order, because
3479
# we find the parent nodes after the children
3481
return records, ann_keys
3483
def _get_needed_texts(self, key, pb=None):
3484
# if True or len(self._vf._fallback_vfs) > 0:
3485
if len(self._vf._fallback_vfs) > 0:
3486
# If we have fallbacks, go to the generic path
3487
for v in annotate.Annotator._get_needed_texts(self, key, pb=pb):
3492
records, ann_keys = self._get_build_graph(key)
3493
for idx, (sub_key, text, num_lines) in enumerate(
3494
self._extract_texts(records)):
3496
pb.update('annotating', idx, len(records))
3497
yield sub_key, text, num_lines
3498
for sub_key in ann_keys:
3499
text = self._text_cache[sub_key]
3500
num_lines = len(text) # bad assumption
3501
yield sub_key, text, num_lines
3503
except errors.RetryWithNewPacks, e:
3504
self._vf._access.reload_or_raise(e)
3505
# The cached build_details are no longer valid
3506
self._all_build_details.clear()
3508
def _cache_delta_blocks(self, key, compression_parent, delta, lines):
3509
parent_lines = self._text_cache[compression_parent]
3510
blocks = list(KnitContent.get_line_delta_blocks(delta, parent_lines, lines))
3511
self._matching_blocks[(key, compression_parent)] = blocks
3513
def _expand_record(self, key, parent_keys, compression_parent, record,
3516
if compression_parent:
3517
if compression_parent not in self._content_objects:
3518
# Waiting for the parent
3519
self._pending_deltas.setdefault(compression_parent, []).append(
3520
(key, parent_keys, record, record_details))
3522
# We have the basis parent, so expand the delta
3523
num = self._num_compression_children[compression_parent]
3526
base_content = self._content_objects.pop(compression_parent)
3527
self._num_compression_children.pop(compression_parent)
3529
self._num_compression_children[compression_parent] = num
3530
base_content = self._content_objects[compression_parent]
3531
# It is tempting to want to copy_base_content=False for the last
3532
# child object. However, whenever noeol=False,
3533
# self._text_cache[parent_key] is content._lines. So mutating it
3534
# gives very bad results.
3535
# The alternative is to copy the lines into text cache, but then we
3536
# are copying anyway, so just do it here.
3537
content, delta = self._vf._factory.parse_record(
3538
key, record, record_details, base_content,
3539
copy_base_content=True)
3542
content, _ = self._vf._factory.parse_record(
3543
key, record, record_details, None)
3544
if self._num_compression_children.get(key, 0) > 0:
3545
self._content_objects[key] = content
3546
lines = content.text()
3547
self._text_cache[key] = lines
3548
if delta is not None:
3549
self._cache_delta_blocks(key, compression_parent, delta, lines)
3552
def _get_parent_annotations_and_matches(self, key, text, parent_key):
3553
"""Get the list of annotations for the parent, and the matching lines.
3555
:param text: The opaque value given by _get_needed_texts
3556
:param parent_key: The key for the parent text
3557
:return: (parent_annotations, matching_blocks)
3558
parent_annotations is a list as long as the number of lines in
3560
matching_blocks is a list of (parent_idx, text_idx, len) tuples
3561
indicating which lines match between the two texts
3563
block_key = (key, parent_key)
3564
if block_key in self._matching_blocks:
3565
blocks = self._matching_blocks.pop(block_key)
3566
parent_annotations = self._annotations_cache[parent_key]
3567
return parent_annotations, blocks
3568
return annotate.Annotator._get_parent_annotations_and_matches(self,
3569
key, text, parent_key)
3571
def _process_pending(self, key):
3572
"""The content for 'key' was just processed.
3574
Determine if there is any more pending work to be processed.
3577
if key in self._pending_deltas:
3578
compression_parent = key
3579
children = self._pending_deltas.pop(key)
3580
for child_key, parent_keys, record, record_details in children:
3581
lines = self._expand_record(child_key, parent_keys,
3583
record, record_details)
3584
if self._check_ready_for_annotations(child_key, parent_keys):
3585
to_return.append(child_key)
3586
# Also check any children that are waiting for this parent to be
3588
if key in self._pending_annotation:
3589
children = self._pending_annotation.pop(key)
3590
to_return.extend([c for c, p_keys in children
3591
if self._check_ready_for_annotations(c, p_keys)])
3594
def _check_ready_for_annotations(self, key, parent_keys):
3595
"""return true if this text is ready to be yielded.
3597
Otherwise, this will return False, and queue the text into
3598
self._pending_annotation
3600
for parent_key in parent_keys:
3601
if parent_key not in self._annotations_cache:
3602
# still waiting on at least one parent text, so queue it up
3603
# Note that if there are multiple parents, we need to wait
3605
self._pending_annotation.setdefault(parent_key,
3606
[]).append((key, parent_keys))
3610
def _extract_texts(self, records):
3611
"""Extract the various texts needed based on records"""
3612
# We iterate in the order read, rather than a strict order requested
3613
# However, process what we can, and put off to the side things that
3614
# still need parents, cleaning them up when those parents are
3617
# 1) As 'records' are read, see if we can expand these records into
3618
# Content objects (and thus lines)
3619
# 2) If a given line-delta is waiting on its compression parent, it
3620
# gets queued up into self._pending_deltas, otherwise we expand
3621
# it, and put it into self._text_cache and self._content_objects
3622
# 3) If we expanded the text, we will then check to see if all
3623
# parents have also been processed. If so, this text gets yielded,
3624
# else this record gets set aside into pending_annotation
3625
# 4) Further, if we expanded the text in (2), we will then check to
3626
# see if there are any children in self._pending_deltas waiting to
3627
# also be processed. If so, we go back to (2) for those
3628
# 5) Further again, if we yielded the text, we can then check if that
3629
# 'unlocks' any of the texts in pending_annotations, which should
3630
# then get yielded as well
3631
# Note that both steps 4 and 5 are 'recursive' in that unlocking one
3632
# compression child could unlock yet another, and yielding a fulltext
3633
# will also 'unlock' the children that are waiting on that annotation.
3634
# (Though also, unlocking 1 parent's fulltext, does not unlock a child
3635
# if other parents are also waiting.)
3636
# We want to yield content before expanding child content objects, so
3637
# that we know when we can re-use the content lines, and the annotation
3638
# code can know when it can stop caching fulltexts, as well.
3640
# Children that are missing their compression parent
3642
for (key, record, digest) in self._vf._read_records_iter(records):
3644
details = self._all_build_details[key]
3645
(_, compression_parent, parent_keys, record_details) = details
3646
lines = self._expand_record(key, parent_keys, compression_parent,
3647
record, record_details)
3649
# Pending delta should be queued up
3651
# At this point, we may be able to yield this content, if all
3652
# parents are also finished
3653
yield_this_text = self._check_ready_for_annotations(key,
3656
# All parents present
3657
yield key, lines, len(lines)
3658
to_process = self._process_pending(key)
3660
this_process = to_process
3662
for key in this_process:
3663
lines = self._text_cache[key]
3664
yield key, lines, len(lines)
3665
to_process.extend(self._process_pending(key))
3668
from bzrlib._knit_load_data_pyx import _load_data_c as _load_data
3670
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
