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
for fallback in self._fallback_vfs:
1197
if not missing_keys:
1199
(f_parent_map, f_missing_keys) = fallback._index.find_ancestry(
1201
parent_map.update(f_parent_map)
1202
missing_keys = f_missing_keys
1203
kg = _mod_graph.KnownGraph(parent_map)
1206
def get_parent_map(self, keys):
1207
"""Get a map of the graph parents of keys.
1209
:param keys: The keys to look up parents for.
1210
:return: A mapping from keys to parents. Absent keys are absent from
1213
return self._get_parent_map_with_sources(keys)[0]
1215
def _get_parent_map_with_sources(self, keys):
1216
"""Get a map of the parents of keys.
1218
:param keys: The keys to look up parents for.
1219
:return: A tuple. The first element is a mapping from keys to parents.
1220
Absent keys are absent from the mapping. The second element is a
1221
list with the locations each key was found in. The first element
1222
is the in-this-knit parents, the second the first fallback source,
1226
sources = [self._index] + self._fallback_vfs
1229
for source in sources:
1232
new_result = source.get_parent_map(missing)
1233
source_results.append(new_result)
1234
result.update(new_result)
1235
missing.difference_update(set(new_result))
1236
return result, source_results
1238
def _get_record_map(self, keys, allow_missing=False):
1239
"""Produce a dictionary of knit records.
1241
:return: {key:(record, record_details, digest, next)}
1243
data returned from read_records (a KnitContentobject)
1245
opaque information to pass to parse_record
1247
SHA1 digest of the full text after all steps are done
1249
build-parent of the version, i.e. the leftmost ancestor.
1250
Will be None if the record is not a delta.
1251
:param keys: The keys to build a map for
1252
:param allow_missing: If some records are missing, rather than
1253
error, just return the data that could be generated.
1255
raw_map = self._get_record_map_unparsed(keys,
1256
allow_missing=allow_missing)
1257
return self._raw_map_to_record_map(raw_map)
1259
def _raw_map_to_record_map(self, raw_map):
1260
"""Parse the contents of _get_record_map_unparsed.
1262
:return: see _get_record_map.
1266
data, record_details, next = raw_map[key]
1267
content, digest = self._parse_record(key[-1], data)
1268
result[key] = content, record_details, digest, next
1271
def _get_record_map_unparsed(self, keys, allow_missing=False):
1272
"""Get the raw data for reconstructing keys without parsing it.
1274
:return: A dict suitable for parsing via _raw_map_to_record_map.
1275
key-> raw_bytes, (method, noeol), compression_parent
1277
# This retries the whole request if anything fails. Potentially we
1278
# could be a bit more selective. We could track the keys whose records
1279
# we have successfully found, and then only request the new records
1280
# from there. However, _get_components_positions grabs the whole build
1281
# chain, which means we'll likely try to grab the same records again
1282
# anyway. Also, can the build chains change as part of a pack
1283
# operation? We wouldn't want to end up with a broken chain.
1286
position_map = self._get_components_positions(keys,
1287
allow_missing=allow_missing)
1288
# key = component_id, r = record_details, i_m = index_memo,
1290
records = [(key, i_m) for key, (r, i_m, n)
1291
in position_map.iteritems()]
1292
# Sort by the index memo, so that we request records from the
1293
# same pack file together, and in forward-sorted order
1294
records.sort(key=operator.itemgetter(1))
1296
for key, data in self._read_records_iter_unchecked(records):
1297
(record_details, index_memo, next) = position_map[key]
1298
raw_record_map[key] = data, record_details, next
1299
return raw_record_map
1300
except errors.RetryWithNewPacks, e:
1301
self._access.reload_or_raise(e)
1304
def _split_by_prefix(cls, keys):
1305
"""For the given keys, split them up based on their prefix.
1307
To keep memory pressure somewhat under control, split the
1308
requests back into per-file-id requests, otherwise "bzr co"
1309
extracts the full tree into memory before writing it to disk.
1310
This should be revisited if _get_content_maps() can ever cross
1313
The keys for a given file_id are kept in the same relative order.
1314
Ordering between file_ids is not, though prefix_order will return the
1315
order that the key was first seen.
1317
:param keys: An iterable of key tuples
1318
:return: (split_map, prefix_order)
1319
split_map A dictionary mapping prefix => keys
1320
prefix_order The order that we saw the various prefixes
1322
split_by_prefix = {}
1330
if prefix in split_by_prefix:
1331
split_by_prefix[prefix].append(key)
1333
split_by_prefix[prefix] = [key]
1334
prefix_order.append(prefix)
1335
return split_by_prefix, prefix_order
1337
def _group_keys_for_io(self, keys, non_local_keys, positions,
1338
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1339
"""For the given keys, group them into 'best-sized' requests.
1341
The idea is to avoid making 1 request per file, but to never try to
1342
unpack an entire 1.5GB source tree in a single pass. Also when
1343
possible, we should try to group requests to the same pack file
1346
:return: list of (keys, non_local) tuples that indicate what keys
1347
should be fetched next.
1349
# TODO: Ideally we would group on 2 factors. We want to extract texts
1350
# from the same pack file together, and we want to extract all
1351
# the texts for a given build-chain together. Ultimately it
1352
# probably needs a better global view.
1353
total_keys = len(keys)
1354
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1355
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1357
cur_non_local = set()
1361
for prefix in prefix_order:
1362
keys = prefix_split_keys[prefix]
1363
non_local = prefix_split_non_local_keys.get(prefix, [])
1365
this_size = self._index._get_total_build_size(keys, positions)
1366
cur_size += this_size
1367
cur_keys.extend(keys)
1368
cur_non_local.update(non_local)
1369
if cur_size > _min_buffer_size:
1370
result.append((cur_keys, cur_non_local))
1371
sizes.append(cur_size)
1373
cur_non_local = set()
1376
result.append((cur_keys, cur_non_local))
1377
sizes.append(cur_size)
1380
def get_record_stream(self, keys, ordering, include_delta_closure):
1381
"""Get a stream of records for keys.
1383
:param keys: The keys to include.
1384
:param ordering: Either 'unordered' or 'topological'. A topologically
1385
sorted stream has compression parents strictly before their
1387
:param include_delta_closure: If True then the closure across any
1388
compression parents will be included (in the opaque data).
1389
:return: An iterator of ContentFactory objects, each of which is only
1390
valid until the iterator is advanced.
1392
# keys might be a generator
1396
if not self._index.has_graph:
1397
# Cannot sort when no graph has been stored.
1398
ordering = 'unordered'
1400
remaining_keys = keys
1403
keys = set(remaining_keys)
1404
for content_factory in self._get_remaining_record_stream(keys,
1405
ordering, include_delta_closure):
1406
remaining_keys.discard(content_factory.key)
1407
yield content_factory
1409
except errors.RetryWithNewPacks, e:
1410
self._access.reload_or_raise(e)
1412
def _get_remaining_record_stream(self, keys, ordering,
1413
include_delta_closure):
1414
"""This function is the 'retry' portion for get_record_stream."""
1415
if include_delta_closure:
1416
positions = self._get_components_positions(keys, allow_missing=True)
1418
build_details = self._index.get_build_details(keys)
1420
# (record_details, access_memo, compression_parent_key)
1421
positions = dict((key, self._build_details_to_components(details))
1422
for key, details in build_details.iteritems())
1423
absent_keys = keys.difference(set(positions))
1424
# There may be more absent keys : if we're missing the basis component
1425
# and are trying to include the delta closure.
1426
# XXX: We should not ever need to examine remote sources because we do
1427
# not permit deltas across versioned files boundaries.
1428
if include_delta_closure:
1429
needed_from_fallback = set()
1430
# Build up reconstructable_keys dict. key:True in this dict means
1431
# the key can be reconstructed.
1432
reconstructable_keys = {}
1436
chain = [key, positions[key][2]]
1438
needed_from_fallback.add(key)
1441
while chain[-1] is not None:
1442
if chain[-1] in reconstructable_keys:
1443
result = reconstructable_keys[chain[-1]]
1447
chain.append(positions[chain[-1]][2])
1449
# missing basis component
1450
needed_from_fallback.add(chain[-1])
1453
for chain_key in chain[:-1]:
1454
reconstructable_keys[chain_key] = result
1456
needed_from_fallback.add(key)
1457
# Double index lookups here : need a unified api ?
1458
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1459
if ordering in ('topological', 'groupcompress'):
1460
if ordering == 'topological':
1461
# Global topological sort
1462
present_keys = tsort.topo_sort(global_map)
1464
present_keys = sort_groupcompress(global_map)
1465
# Now group by source:
1467
current_source = None
1468
for key in present_keys:
1469
for parent_map in parent_maps:
1470
if key in parent_map:
1471
key_source = parent_map
1473
if current_source is not key_source:
1474
source_keys.append((key_source, []))
1475
current_source = key_source
1476
source_keys[-1][1].append(key)
1478
if ordering != 'unordered':
1479
raise AssertionError('valid values for ordering are:'
1480
' "unordered", "groupcompress" or "topological" not: %r'
1482
# Just group by source; remote sources first.
1485
for parent_map in reversed(parent_maps):
1486
source_keys.append((parent_map, []))
1487
for key in parent_map:
1488
present_keys.append(key)
1489
source_keys[-1][1].append(key)
1490
# We have been requested to return these records in an order that
1491
# suits us. So we ask the index to give us an optimally sorted
1493
for source, sub_keys in source_keys:
1494
if source is parent_maps[0]:
1495
# Only sort the keys for this VF
1496
self._index._sort_keys_by_io(sub_keys, positions)
1497
absent_keys = keys - set(global_map)
1498
for key in absent_keys:
1499
yield AbsentContentFactory(key)
1500
# restrict our view to the keys we can answer.
1501
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1502
# XXX: At that point we need to consider the impact of double reads by
1503
# utilising components multiple times.
1504
if include_delta_closure:
1505
# XXX: get_content_maps performs its own index queries; allow state
1507
non_local_keys = needed_from_fallback - absent_keys
1508
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1511
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1514
for record in generator.get_record_stream():
1517
for source, keys in source_keys:
1518
if source is parent_maps[0]:
1519
# this KnitVersionedFiles
1520
records = [(key, positions[key][1]) for key in keys]
1521
for key, raw_data, sha1 in self._read_records_iter_raw(records):
1522
(record_details, index_memo, _) = positions[key]
1523
yield KnitContentFactory(key, global_map[key],
1524
record_details, sha1, raw_data, self._factory.annotated, None)
1526
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1527
for record in vf.get_record_stream(keys, ordering,
1528
include_delta_closure):
1531
def get_sha1s(self, keys):
1532
"""See VersionedFiles.get_sha1s()."""
1534
record_map = self._get_record_map(missing, allow_missing=True)
1536
for key, details in record_map.iteritems():
1537
if key not in missing:
1539
# record entry 2 is the 'digest'.
1540
result[key] = details[2]
1541
missing.difference_update(set(result))
1542
for source in self._fallback_vfs:
1545
new_result = source.get_sha1s(missing)
1546
result.update(new_result)
1547
missing.difference_update(set(new_result))
1550
def insert_record_stream(self, stream):
1551
"""Insert a record stream into this container.
1553
:param stream: A stream of records to insert.
1555
:seealso VersionedFiles.get_record_stream:
1557
def get_adapter(adapter_key):
1559
return adapters[adapter_key]
1561
adapter_factory = adapter_registry.get(adapter_key)
1562
adapter = adapter_factory(self)
1563
adapters[adapter_key] = adapter
1566
if self._factory.annotated:
1567
# self is annotated, we need annotated knits to use directly.
1568
annotated = "annotated-"
1571
# self is not annotated, but we can strip annotations cheaply.
1573
convertibles = set(["knit-annotated-ft-gz"])
1574
if self._max_delta_chain:
1575
delta_types.add("knit-annotated-delta-gz")
1576
convertibles.add("knit-annotated-delta-gz")
1577
# The set of types we can cheaply adapt without needing basis texts.
1578
native_types = set()
1579
if self._max_delta_chain:
1580
native_types.add("knit-%sdelta-gz" % annotated)
1581
delta_types.add("knit-%sdelta-gz" % annotated)
1582
native_types.add("knit-%sft-gz" % annotated)
1583
knit_types = native_types.union(convertibles)
1585
# Buffer all index entries that we can't add immediately because their
1586
# basis parent is missing. We don't buffer all because generating
1587
# annotations may require access to some of the new records. However we
1588
# can't generate annotations from new deltas until their basis parent
1589
# is present anyway, so we get away with not needing an index that
1590
# includes the new keys.
1592
# See <http://launchpad.net/bugs/300177> about ordering of compression
1593
# parents in the records - to be conservative, we insist that all
1594
# parents must be present to avoid expanding to a fulltext.
1596
# key = basis_parent, value = index entry to add
1597
buffered_index_entries = {}
1598
for record in stream:
1600
parents = record.parents
1601
if record.storage_kind in delta_types:
1602
# TODO: eventually the record itself should track
1603
# compression_parent
1604
compression_parent = parents[0]
1606
compression_parent = None
1607
# Raise an error when a record is missing.
1608
if record.storage_kind == 'absent':
1609
raise RevisionNotPresent([record.key], self)
1610
elif ((record.storage_kind in knit_types)
1611
and (compression_parent is None
1612
or not self._fallback_vfs
1613
or self._index.has_key(compression_parent)
1614
or not self.has_key(compression_parent))):
1615
# we can insert the knit record literally if either it has no
1616
# compression parent OR we already have its basis in this kvf
1617
# OR the basis is not present even in the fallbacks. In the
1618
# last case it will either turn up later in the stream and all
1619
# will be well, or it won't turn up at all and we'll raise an
1622
# TODO: self.has_key is somewhat redundant with
1623
# self._index.has_key; we really want something that directly
1624
# asks if it's only present in the fallbacks. -- mbp 20081119
1625
if record.storage_kind not in native_types:
1627
adapter_key = (record.storage_kind, "knit-delta-gz")
1628
adapter = get_adapter(adapter_key)
1630
adapter_key = (record.storage_kind, "knit-ft-gz")
1631
adapter = get_adapter(adapter_key)
1632
bytes = adapter.get_bytes(record)
1634
# It's a knit record, it has a _raw_record field (even if
1635
# it was reconstituted from a network stream).
1636
bytes = record._raw_record
1637
options = [record._build_details[0]]
1638
if record._build_details[1]:
1639
options.append('no-eol')
1640
# Just blat it across.
1641
# Note: This does end up adding data on duplicate keys. As
1642
# modern repositories use atomic insertions this should not
1643
# lead to excessive growth in the event of interrupted fetches.
1644
# 'knit' repositories may suffer excessive growth, but as a
1645
# deprecated format this is tolerable. It can be fixed if
1646
# needed by in the kndx index support raising on a duplicate
1647
# add with identical parents and options.
1648
access_memo = self._access.add_raw_records(
1649
[(record.key, len(bytes))], bytes)[0]
1650
index_entry = (record.key, options, access_memo, parents)
1651
if 'fulltext' not in options:
1652
# Not a fulltext, so we need to make sure the compression
1653
# parent will also be present.
1654
# Note that pack backed knits don't need to buffer here
1655
# because they buffer all writes to the transaction level,
1656
# but we don't expose that difference at the index level. If
1657
# the query here has sufficient cost to show up in
1658
# profiling we should do that.
1660
# They're required to be physically in this
1661
# KnitVersionedFiles, not in a fallback.
1662
if not self._index.has_key(compression_parent):
1663
pending = buffered_index_entries.setdefault(
1664
compression_parent, [])
1665
pending.append(index_entry)
1668
self._index.add_records([index_entry])
1669
elif record.storage_kind == 'chunked':
1670
self.add_lines(record.key, parents,
1671
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1673
# Not suitable for direct insertion as a
1674
# delta, either because it's not the right format, or this
1675
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1676
# 0) or because it depends on a base only present in the
1678
self._access.flush()
1680
# Try getting a fulltext directly from the record.
1681
bytes = record.get_bytes_as('fulltext')
1682
except errors.UnavailableRepresentation:
1683
adapter_key = record.storage_kind, 'fulltext'
1684
adapter = get_adapter(adapter_key)
1685
bytes = adapter.get_bytes(record)
1686
lines = split_lines(bytes)
1688
self.add_lines(record.key, parents, lines)
1689
except errors.RevisionAlreadyPresent:
1691
# Add any records whose basis parent is now available.
1693
added_keys = [record.key]
1695
key = added_keys.pop(0)
1696
if key in buffered_index_entries:
1697
index_entries = buffered_index_entries[key]
1698
self._index.add_records(index_entries)
1700
[index_entry[0] for index_entry in index_entries])
1701
del buffered_index_entries[key]
1702
if buffered_index_entries:
1703
# There were index entries buffered at the end of the stream,
1704
# So these need to be added (if the index supports holding such
1705
# entries for later insertion)
1706
for key in buffered_index_entries:
1707
index_entries = buffered_index_entries[key]
1708
self._index.add_records(index_entries,
1709
missing_compression_parents=True)
1711
def get_missing_compression_parent_keys(self):
1712
"""Return an iterable of keys of missing compression parents.
1714
Check this after calling insert_record_stream to find out if there are
1715
any missing compression parents. If there are, the records that
1716
depend on them are not able to be inserted safely. For atomic
1717
KnitVersionedFiles built on packs, the transaction should be aborted or
1718
suspended - commit will fail at this point. Nonatomic knits will error
1719
earlier because they have no staging area to put pending entries into.
1721
return self._index.get_missing_compression_parents()
1723
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1724
"""Iterate over the lines in the versioned files from keys.
1726
This may return lines from other keys. Each item the returned
1727
iterator yields is a tuple of a line and a text version that that line
1728
is present in (not introduced in).
1730
Ordering of results is in whatever order is most suitable for the
1731
underlying storage format.
1733
If a progress bar is supplied, it may be used to indicate progress.
1734
The caller is responsible for cleaning up progress bars (because this
1738
* Lines are normalised by the underlying store: they will all have \\n
1740
* Lines are returned in arbitrary order.
1741
* If a requested key did not change any lines (or didn't have any
1742
lines), it may not be mentioned at all in the result.
1744
:param pb: Progress bar supplied by caller.
1745
:return: An iterator over (line, key).
1748
pb = progress.DummyProgress()
1754
# we don't care about inclusions, the caller cares.
1755
# but we need to setup a list of records to visit.
1756
# we need key, position, length
1758
build_details = self._index.get_build_details(keys)
1759
for key, details in build_details.iteritems():
1761
key_records.append((key, details[0]))
1762
records_iter = enumerate(self._read_records_iter(key_records))
1763
for (key_idx, (key, data, sha_value)) in records_iter:
1764
pb.update('Walking content', key_idx, total)
1765
compression_parent = build_details[key][1]
1766
if compression_parent is None:
1768
line_iterator = self._factory.get_fulltext_content(data)
1771
line_iterator = self._factory.get_linedelta_content(data)
1772
# Now that we are yielding the data for this key, remove it
1775
# XXX: It might be more efficient to yield (key,
1776
# line_iterator) in the future. However for now, this is a
1777
# simpler change to integrate into the rest of the
1778
# codebase. RBC 20071110
1779
for line in line_iterator:
1782
except errors.RetryWithNewPacks, e:
1783
self._access.reload_or_raise(e)
1784
# If there are still keys we've not yet found, we look in the fallback
1785
# vfs, and hope to find them there. Note that if the keys are found
1786
# but had no changes or no content, the fallback may not return
1788
if keys and not self._fallback_vfs:
1789
# XXX: strictly the second parameter is meant to be the file id
1790
# but it's not easily accessible here.
1791
raise RevisionNotPresent(keys, repr(self))
1792
for source in self._fallback_vfs:
1796
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1797
source_keys.add(key)
1799
keys.difference_update(source_keys)
1800
pb.update('Walking content', total, total)
1802
def _make_line_delta(self, delta_seq, new_content):
1803
"""Generate a line delta from delta_seq and new_content."""
1805
for op in delta_seq.get_opcodes():
1806
if op[0] == 'equal':
1808
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1811
def _merge_annotations(self, content, parents, parent_texts={},
1812
delta=None, annotated=None,
1813
left_matching_blocks=None):
1814
"""Merge annotations for content and generate deltas.
1816
This is done by comparing the annotations based on changes to the text
1817
and generating a delta on the resulting full texts. If annotations are
1818
not being created then a simple delta is created.
1820
if left_matching_blocks is not None:
1821
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1825
for parent_key in parents:
1826
merge_content = self._get_content(parent_key, parent_texts)
1827
if (parent_key == parents[0] and delta_seq is not None):
1830
seq = patiencediff.PatienceSequenceMatcher(
1831
None, merge_content.text(), content.text())
1832
for i, j, n in seq.get_matching_blocks():
1835
# this copies (origin, text) pairs across to the new
1836
# content for any line that matches the last-checked
1838
content._lines[j:j+n] = merge_content._lines[i:i+n]
1839
# XXX: Robert says the following block is a workaround for a
1840
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1841
if content._lines and content._lines[-1][1][-1] != '\n':
1842
# The copied annotation was from a line without a trailing EOL,
1843
# reinstate one for the content object, to ensure correct
1845
line = content._lines[-1][1] + '\n'
1846
content._lines[-1] = (content._lines[-1][0], line)
1848
if delta_seq is None:
1849
reference_content = self._get_content(parents[0], parent_texts)
1850
new_texts = content.text()
1851
old_texts = reference_content.text()
1852
delta_seq = patiencediff.PatienceSequenceMatcher(
1853
None, old_texts, new_texts)
1854
return self._make_line_delta(delta_seq, content)
1856
def _parse_record(self, version_id, data):
1857
"""Parse an original format knit record.
1859
These have the last element of the key only present in the stored data.
1861
rec, record_contents = self._parse_record_unchecked(data)
1862
self._check_header_version(rec, version_id)
1863
return record_contents, rec[3]
1865
def _parse_record_header(self, key, raw_data):
1866
"""Parse a record header for consistency.
1868
:return: the header and the decompressor stream.
1869
as (stream, header_record)
1871
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1874
rec = self._check_header(key, df.readline())
1875
except Exception, e:
1876
raise KnitCorrupt(self,
1877
"While reading {%s} got %s(%s)"
1878
% (key, e.__class__.__name__, str(e)))
1881
def _parse_record_unchecked(self, data):
1883
# 4168 calls in 2880 217 internal
1884
# 4168 calls to _parse_record_header in 2121
1885
# 4168 calls to readlines in 330
1886
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1888
record_contents = df.readlines()
1889
except Exception, e:
1890
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1891
(data, e.__class__.__name__, str(e)))
1892
header = record_contents.pop(0)
1893
rec = self._split_header(header)
1894
last_line = record_contents.pop()
1895
if len(record_contents) != int(rec[2]):
1896
raise KnitCorrupt(self,
1897
'incorrect number of lines %s != %s'
1898
' for version {%s} %s'
1899
% (len(record_contents), int(rec[2]),
1900
rec[1], record_contents))
1901
if last_line != 'end %s\n' % rec[1]:
1902
raise KnitCorrupt(self,
1903
'unexpected version end line %r, wanted %r'
1904
% (last_line, rec[1]))
1906
return rec, record_contents
1908
def _read_records_iter(self, records):
1909
"""Read text records from data file and yield result.
1911
The result will be returned in whatever is the fastest to read.
1912
Not by the order requested. Also, multiple requests for the same
1913
record will only yield 1 response.
1914
:param records: A list of (key, access_memo) entries
1915
:return: Yields (key, contents, digest) in the order
1916
read, not the order requested
1921
# XXX: This smells wrong, IO may not be getting ordered right.
1922
needed_records = sorted(set(records), key=operator.itemgetter(1))
1923
if not needed_records:
1926
# The transport optimizes the fetching as well
1927
# (ie, reads continuous ranges.)
1928
raw_data = self._access.get_raw_records(
1929
[index_memo for key, index_memo in needed_records])
1931
for (key, index_memo), data in \
1932
izip(iter(needed_records), raw_data):
1933
content, digest = self._parse_record(key[-1], data)
1934
yield key, content, digest
1936
def _read_records_iter_raw(self, records):
1937
"""Read text records from data file and yield raw data.
1939
This unpacks enough of the text record to validate the id is
1940
as expected but thats all.
1942
Each item the iterator yields is (key, bytes,
1943
expected_sha1_of_full_text).
1945
for key, data in self._read_records_iter_unchecked(records):
1946
# validate the header (note that we can only use the suffix in
1947
# current knit records).
1948
df, rec = self._parse_record_header(key, data)
1950
yield key, data, rec[3]
1952
def _read_records_iter_unchecked(self, records):
1953
"""Read text records from data file and yield raw data.
1955
No validation is done.
1957
Yields tuples of (key, data).
1959
# setup an iterator of the external records:
1960
# uses readv so nice and fast we hope.
1962
# grab the disk data needed.
1963
needed_offsets = [index_memo for key, index_memo
1965
raw_records = self._access.get_raw_records(needed_offsets)
1967
for key, index_memo in records:
1968
data = raw_records.next()
1971
def _record_to_data(self, key, digest, lines, dense_lines=None):
1972
"""Convert key, digest, lines into a raw data block.
1974
:param key: The key of the record. Currently keys are always serialised
1975
using just the trailing component.
1976
:param dense_lines: The bytes of lines but in a denser form. For
1977
instance, if lines is a list of 1000 bytestrings each ending in \n,
1978
dense_lines may be a list with one line in it, containing all the
1979
1000's lines and their \n's. Using dense_lines if it is already
1980
known is a win because the string join to create bytes in this
1981
function spends less time resizing the final string.
1982
:return: (len, a StringIO instance with the raw data ready to read.)
1984
chunks = ["version %s %d %s\n" % (key[-1], len(lines), digest)]
1985
chunks.extend(dense_lines or lines)
1986
chunks.append("end %s\n" % key[-1])
1987
for chunk in chunks:
1988
if type(chunk) is not str:
1989
raise AssertionError(
1990
'data must be plain bytes was %s' % type(chunk))
1991
if lines and lines[-1][-1] != '\n':
1992
raise ValueError('corrupt lines value %r' % lines)
1993
compressed_bytes = tuned_gzip.chunks_to_gzip(chunks)
1994
return len(compressed_bytes), compressed_bytes
1996
def _split_header(self, line):
1999
raise KnitCorrupt(self,
2000
'unexpected number of elements in record header')
2004
"""See VersionedFiles.keys."""
2005
if 'evil' in debug.debug_flags:
2006
trace.mutter_callsite(2, "keys scales with size of history")
2007
sources = [self._index] + self._fallback_vfs
2009
for source in sources:
2010
result.update(source.keys())
2014
class _ContentMapGenerator(object):
2015
"""Generate texts or expose raw deltas for a set of texts."""
2017
def __init__(self, ordering='unordered'):
2018
self._ordering = ordering
2020
def _get_content(self, key):
2021
"""Get the content object for key."""
2022
# Note that _get_content is only called when the _ContentMapGenerator
2023
# has been constructed with just one key requested for reconstruction.
2024
if key in self.nonlocal_keys:
2025
record = self.get_record_stream().next()
2026
# Create a content object on the fly
2027
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
2028
return PlainKnitContent(lines, record.key)
2030
# local keys we can ask for directly
2031
return self._get_one_work(key)
2033
def get_record_stream(self):
2034
"""Get a record stream for the keys requested during __init__."""
2035
for record in self._work():
2039
"""Produce maps of text and KnitContents as dicts.
2041
:return: (text_map, content_map) where text_map contains the texts for
2042
the requested versions and content_map contains the KnitContents.
2044
# NB: By definition we never need to read remote sources unless texts
2045
# are requested from them: we don't delta across stores - and we
2046
# explicitly do not want to to prevent data loss situations.
2047
if self.global_map is None:
2048
self.global_map = self.vf.get_parent_map(self.keys)
2049
nonlocal_keys = self.nonlocal_keys
2051
missing_keys = set(nonlocal_keys)
2052
# Read from remote versioned file instances and provide to our caller.
2053
for source in self.vf._fallback_vfs:
2054
if not missing_keys:
2056
# Loop over fallback repositories asking them for texts - ignore
2057
# any missing from a particular fallback.
2058
for record in source.get_record_stream(missing_keys,
2059
self._ordering, True):
2060
if record.storage_kind == 'absent':
2061
# Not in thie particular stream, may be in one of the
2062
# other fallback vfs objects.
2064
missing_keys.remove(record.key)
2067
if self._raw_record_map is None:
2068
raise AssertionError('_raw_record_map should have been filled')
2070
for key in self.keys:
2071
if key in self.nonlocal_keys:
2073
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2076
def _get_one_work(self, requested_key):
2077
# Now, if we have calculated everything already, just return the
2079
if requested_key in self._contents_map:
2080
return self._contents_map[requested_key]
2081
# To simplify things, parse everything at once - code that wants one text
2082
# probably wants them all.
2083
# FUTURE: This function could be improved for the 'extract many' case
2084
# by tracking each component and only doing the copy when the number of
2085
# children than need to apply delta's to it is > 1 or it is part of the
2087
multiple_versions = len(self.keys) != 1
2088
if self._record_map is None:
2089
self._record_map = self.vf._raw_map_to_record_map(
2090
self._raw_record_map)
2091
record_map = self._record_map
2092
# raw_record_map is key:
2093
# Have read and parsed records at this point.
2094
for key in self.keys:
2095
if key in self.nonlocal_keys:
2100
while cursor is not None:
2102
record, record_details, digest, next = record_map[cursor]
2104
raise RevisionNotPresent(cursor, self)
2105
components.append((cursor, record, record_details, digest))
2107
if cursor in self._contents_map:
2108
# no need to plan further back
2109
components.append((cursor, None, None, None))
2113
for (component_id, record, record_details,
2114
digest) in reversed(components):
2115
if component_id in self._contents_map:
2116
content = self._contents_map[component_id]
2118
content, delta = self._factory.parse_record(key[-1],
2119
record, record_details, content,
2120
copy_base_content=multiple_versions)
2121
if multiple_versions:
2122
self._contents_map[component_id] = content
2124
# digest here is the digest from the last applied component.
2125
text = content.text()
2126
actual_sha = sha_strings(text)
2127
if actual_sha != digest:
2128
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2129
if multiple_versions:
2130
return self._contents_map[requested_key]
2134
def _wire_bytes(self):
2135
"""Get the bytes to put on the wire for 'key'.
2137
The first collection of bytes asked for returns the serialised
2138
raw_record_map and the additional details (key, parent) for key.
2139
Subsequent calls return just the additional details (key, parent).
2140
The wire storage_kind given for the first key is 'knit-delta-closure',
2141
For subsequent keys it is 'knit-delta-closure-ref'.
2143
:param key: A key from the content generator.
2144
:return: Bytes to put on the wire.
2147
# kind marker for dispatch on the far side,
2148
lines.append('knit-delta-closure')
2150
if self.vf._factory.annotated:
2151
lines.append('annotated')
2154
# then the list of keys
2155
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2156
if key not in self.nonlocal_keys]))
2157
# then the _raw_record_map in serialised form:
2159
# for each item in the map:
2161
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2162
# one line with method
2163
# one line with noeol
2164
# one line with next ('' for None)
2165
# one line with byte count of the record bytes
2167
for key, (record_bytes, (method, noeol), next) in \
2168
self._raw_record_map.iteritems():
2169
key_bytes = '\x00'.join(key)
2170
parents = self.global_map.get(key, None)
2172
parent_bytes = 'None:'
2174
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2175
method_bytes = method
2181
next_bytes = '\x00'.join(next)
2184
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2185
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2186
len(record_bytes), record_bytes))
2187
map_bytes = ''.join(map_byte_list)
2188
lines.append(map_bytes)
2189
bytes = '\n'.join(lines)
2193
class _VFContentMapGenerator(_ContentMapGenerator):
2194
"""Content map generator reading from a VersionedFiles object."""
2196
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2197
global_map=None, raw_record_map=None, ordering='unordered'):
2198
"""Create a _ContentMapGenerator.
2200
:param versioned_files: The versioned files that the texts are being
2202
:param keys: The keys to produce content maps for.
2203
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2204
which are known to not be in this knit, but rather in one of the
2206
:param global_map: The result of get_parent_map(keys) (or a supermap).
2207
This is required if get_record_stream() is to be used.
2208
:param raw_record_map: A unparsed raw record map to use for answering
2211
_ContentMapGenerator.__init__(self, ordering=ordering)
2212
# The vf to source data from
2213
self.vf = versioned_files
2215
self.keys = list(keys)
2216
# Keys known to be in fallback vfs objects
2217
if nonlocal_keys is None:
2218
self.nonlocal_keys = set()
2220
self.nonlocal_keys = frozenset(nonlocal_keys)
2221
# Parents data for keys to be returned in get_record_stream
2222
self.global_map = global_map
2223
# The chunked lists for self.keys in text form
2225
# A cache of KnitContent objects used in extracting texts.
2226
self._contents_map = {}
2227
# All the knit records needed to assemble the requested keys as full
2229
self._record_map = None
2230
if raw_record_map is None:
2231
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2234
self._raw_record_map = raw_record_map
2235
# the factory for parsing records
2236
self._factory = self.vf._factory
2239
class _NetworkContentMapGenerator(_ContentMapGenerator):
2240
"""Content map generator sourced from a network stream."""
2242
def __init__(self, bytes, line_end):
2243
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2245
self.global_map = {}
2246
self._raw_record_map = {}
2247
self._contents_map = {}
2248
self._record_map = None
2249
self.nonlocal_keys = []
2250
# Get access to record parsing facilities
2251
self.vf = KnitVersionedFiles(None, None)
2254
line_end = bytes.find('\n', start)
2255
line = bytes[start:line_end]
2256
start = line_end + 1
2257
if line == 'annotated':
2258
self._factory = KnitAnnotateFactory()
2260
self._factory = KnitPlainFactory()
2261
# list of keys to emit in get_record_stream
2262
line_end = bytes.find('\n', start)
2263
line = bytes[start:line_end]
2264
start = line_end + 1
2266
tuple(segment.split('\x00')) for segment in line.split('\t')
2268
# now a loop until the end. XXX: It would be nice if this was just a
2269
# bunch of the same records as get_record_stream(..., False) gives, but
2270
# there is a decent sized gap stopping that at the moment.
2274
line_end = bytes.find('\n', start)
2275
key = tuple(bytes[start:line_end].split('\x00'))
2276
start = line_end + 1
2277
# 1 line with parents (None: for None, '' for ())
2278
line_end = bytes.find('\n', start)
2279
line = bytes[start:line_end]
2284
[tuple(segment.split('\x00')) for segment in line.split('\t')
2286
self.global_map[key] = parents
2287
start = line_end + 1
2288
# one line with method
2289
line_end = bytes.find('\n', start)
2290
line = bytes[start:line_end]
2292
start = line_end + 1
2293
# one line with noeol
2294
line_end = bytes.find('\n', start)
2295
line = bytes[start:line_end]
2297
start = line_end + 1
2298
# one line with next ('' for None)
2299
line_end = bytes.find('\n', start)
2300
line = bytes[start:line_end]
2304
next = tuple(bytes[start:line_end].split('\x00'))
2305
start = line_end + 1
2306
# one line with byte count of the record bytes
2307
line_end = bytes.find('\n', start)
2308
line = bytes[start:line_end]
2310
start = line_end + 1
2312
record_bytes = bytes[start:start+count]
2313
start = start + count
2315
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2317
def get_record_stream(self):
2318
"""Get a record stream for for keys requested by the bytestream."""
2320
for key in self.keys:
2321
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2324
def _wire_bytes(self):
2328
class _KndxIndex(object):
2329
"""Manages knit index files
2331
The index is kept in memory and read on startup, to enable
2332
fast lookups of revision information. The cursor of the index
2333
file is always pointing to the end, making it easy to append
2336
_cache is a cache for fast mapping from version id to a Index
2339
_history is a cache for fast mapping from indexes to version ids.
2341
The index data format is dictionary compressed when it comes to
2342
parent references; a index entry may only have parents that with a
2343
lover index number. As a result, the index is topological sorted.
2345
Duplicate entries may be written to the index for a single version id
2346
if this is done then the latter one completely replaces the former:
2347
this allows updates to correct version and parent information.
2348
Note that the two entries may share the delta, and that successive
2349
annotations and references MUST point to the first entry.
2351
The index file on disc contains a header, followed by one line per knit
2352
record. The same revision can be present in an index file more than once.
2353
The first occurrence gets assigned a sequence number starting from 0.
2355
The format of a single line is
2356
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
2357
REVISION_ID is a utf8-encoded revision id
2358
FLAGS is a comma separated list of flags about the record. Values include
2359
no-eol, line-delta, fulltext.
2360
BYTE_OFFSET is the ascii representation of the byte offset in the data file
2361
that the the compressed data starts at.
2362
LENGTH is the ascii representation of the length of the data file.
2363
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
2365
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
2366
revision id already in the knit that is a parent of REVISION_ID.
2367
The ' :' marker is the end of record marker.
2370
when a write is interrupted to the index file, it will result in a line
2371
that does not end in ' :'. If the ' :' is not present at the end of a line,
2372
or at the end of the file, then the record that is missing it will be
2373
ignored by the parser.
2375
When writing new records to the index file, the data is preceded by '\n'
2376
to ensure that records always start on new lines even if the last write was
2377
interrupted. As a result its normal for the last line in the index to be
2378
missing a trailing newline. One can be added with no harmful effects.
2380
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2381
where prefix is e.g. the (fileid,) for .texts instances or () for
2382
constant-mapped things like .revisions, and the old state is
2383
tuple(history_vector, cache_dict). This is used to prevent having an
2384
ABI change with the C extension that reads .kndx files.
2387
HEADER = "# bzr knit index 8\n"
2389
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2390
"""Create a _KndxIndex on transport using mapper."""
2391
self._transport = transport
2392
self._mapper = mapper
2393
self._get_scope = get_scope
2394
self._allow_writes = allow_writes
2395
self._is_locked = is_locked
2397
self.has_graph = True
2399
def add_records(self, records, random_id=False, missing_compression_parents=False):
2400
"""Add multiple records to the index.
2402
:param records: a list of tuples:
2403
(key, options, access_memo, parents).
2404
:param random_id: If True the ids being added were randomly generated
2405
and no check for existence will be performed.
2406
:param missing_compression_parents: If True the records being added are
2407
only compressed against texts already in the index (or inside
2408
records). If False the records all refer to unavailable texts (or
2409
texts inside records) as compression parents.
2411
if missing_compression_parents:
2412
# It might be nice to get the edge of the records. But keys isn't
2414
keys = sorted(record[0] for record in records)
2415
raise errors.RevisionNotPresent(keys, self)
2417
for record in records:
2420
path = self._mapper.map(key) + '.kndx'
2421
path_keys = paths.setdefault(path, (prefix, []))
2422
path_keys[1].append(record)
2423
for path in sorted(paths):
2424
prefix, path_keys = paths[path]
2425
self._load_prefixes([prefix])
2427
orig_history = self._kndx_cache[prefix][1][:]
2428
orig_cache = self._kndx_cache[prefix][0].copy()
2431
for key, options, (_, pos, size), parents in path_keys:
2433
# kndx indices cannot be parentless.
2435
line = "\n%s %s %s %s %s :" % (
2436
key[-1], ','.join(options), pos, size,
2437
self._dictionary_compress(parents))
2438
if type(line) is not str:
2439
raise AssertionError(
2440
'data must be utf8 was %s' % type(line))
2442
self._cache_key(key, options, pos, size, parents)
2443
if len(orig_history):
2444
self._transport.append_bytes(path, ''.join(lines))
2446
self._init_index(path, lines)
2448
# If any problems happen, restore the original values and re-raise
2449
self._kndx_cache[prefix] = (orig_cache, orig_history)
2452
def scan_unvalidated_index(self, graph_index):
2453
"""See _KnitGraphIndex.scan_unvalidated_index."""
2454
# Because kndx files do not support atomic insertion via separate index
2455
# files, they do not support this method.
2456
raise NotImplementedError(self.scan_unvalidated_index)
2458
def get_missing_compression_parents(self):
2459
"""See _KnitGraphIndex.get_missing_compression_parents."""
2460
# Because kndx files do not support atomic insertion via separate index
2461
# files, they do not support this method.
2462
raise NotImplementedError(self.get_missing_compression_parents)
2464
def _cache_key(self, key, options, pos, size, parent_keys):
2465
"""Cache a version record in the history array and index cache.
2467
This is inlined into _load_data for performance. KEEP IN SYNC.
2468
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2472
version_id = key[-1]
2473
# last-element only for compatibilty with the C load_data.
2474
parents = tuple(parent[-1] for parent in parent_keys)
2475
for parent in parent_keys:
2476
if parent[:-1] != prefix:
2477
raise ValueError("mismatched prefixes for %r, %r" % (
2479
cache, history = self._kndx_cache[prefix]
2480
# only want the _history index to reference the 1st index entry
2482
if version_id not in cache:
2483
index = len(history)
2484
history.append(version_id)
2486
index = cache[version_id][5]
2487
cache[version_id] = (version_id,
2494
def check_header(self, fp):
2495
line = fp.readline()
2497
# An empty file can actually be treated as though the file doesn't
2499
raise errors.NoSuchFile(self)
2500
if line != self.HEADER:
2501
raise KnitHeaderError(badline=line, filename=self)
2503
def _check_read(self):
2504
if not self._is_locked():
2505
raise errors.ObjectNotLocked(self)
2506
if self._get_scope() != self._scope:
2509
def _check_write_ok(self):
2510
"""Assert if not writes are permitted."""
2511
if not self._is_locked():
2512
raise errors.ObjectNotLocked(self)
2513
if self._get_scope() != self._scope:
2515
if self._mode != 'w':
2516
raise errors.ReadOnlyObjectDirtiedError(self)
2518
def get_build_details(self, keys):
2519
"""Get the method, index_memo and compression parent for keys.
2521
Ghosts are omitted from the result.
2523
:param keys: An iterable of keys.
2524
:return: A dict of key:(index_memo, compression_parent, parents,
2527
opaque structure to pass to read_records to extract the raw
2530
Content that this record is built upon, may be None
2532
Logical parents of this node
2534
extra information about the content which needs to be passed to
2535
Factory.parse_record
2537
parent_map = self.get_parent_map(keys)
2540
if key not in parent_map:
2542
method = self.get_method(key)
2543
parents = parent_map[key]
2544
if method == 'fulltext':
2545
compression_parent = None
2547
compression_parent = parents[0]
2548
noeol = 'no-eol' in self.get_options(key)
2549
index_memo = self.get_position(key)
2550
result[key] = (index_memo, compression_parent,
2551
parents, (method, noeol))
2554
def get_method(self, key):
2555
"""Return compression method of specified key."""
2556
options = self.get_options(key)
2557
if 'fulltext' in options:
2559
elif 'line-delta' in options:
2562
raise errors.KnitIndexUnknownMethod(self, options)
2564
def get_options(self, key):
2565
"""Return a list representing options.
2569
prefix, suffix = self._split_key(key)
2570
self._load_prefixes([prefix])
2572
return self._kndx_cache[prefix][0][suffix][1]
2574
raise RevisionNotPresent(key, self)
2576
def find_ancestry(self, keys):
2577
"""See CombinedGraphIndex.find_ancestry()"""
2578
prefixes = set(key[:-1] for key in keys)
2579
self._load_prefixes(prefixes)
2582
missing_keys = set()
2583
pending_keys = list(keys)
2584
# This assumes that keys will not reference parents in a different
2585
# prefix, which is accurate so far.
2587
key = pending_keys.pop()
2588
if key in parent_map:
2592
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2594
missing_keys.add(key)
2596
parent_keys = tuple([prefix + (suffix,)
2597
for suffix in suffix_parents])
2598
parent_map[key] = parent_keys
2599
pending_keys.extend([p for p in parent_keys
2600
if p not in parent_map])
2601
return parent_map, missing_keys
2603
def get_parent_map(self, keys):
2604
"""Get a map of the parents of keys.
2606
:param keys: The keys to look up parents for.
2607
:return: A mapping from keys to parents. Absent keys are absent from
2610
# Parse what we need to up front, this potentially trades off I/O
2611
# locality (.kndx and .knit in the same block group for the same file
2612
# id) for less checking in inner loops.
2613
prefixes = set(key[:-1] for key in keys)
2614
self._load_prefixes(prefixes)
2619
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2623
result[key] = tuple(prefix + (suffix,) for
2624
suffix in suffix_parents)
2627
def get_position(self, key):
2628
"""Return details needed to access the version.
2630
:return: a tuple (key, data position, size) to hand to the access
2631
logic to get the record.
2633
prefix, suffix = self._split_key(key)
2634
self._load_prefixes([prefix])
2635
entry = self._kndx_cache[prefix][0][suffix]
2636
return key, entry[2], entry[3]
2638
has_key = _mod_index._has_key_from_parent_map
2640
def _init_index(self, path, extra_lines=[]):
2641
"""Initialize an index."""
2643
sio.write(self.HEADER)
2644
sio.writelines(extra_lines)
2646
self._transport.put_file_non_atomic(path, sio,
2647
create_parent_dir=True)
2648
# self._create_parent_dir)
2649
# mode=self._file_mode,
2650
# dir_mode=self._dir_mode)
2653
"""Get all the keys in the collection.
2655
The keys are not ordered.
2658
# Identify all key prefixes.
2659
# XXX: A bit hacky, needs polish.
2660
if type(self._mapper) is ConstantMapper:
2664
for quoted_relpath in self._transport.iter_files_recursive():
2665
path, ext = os.path.splitext(quoted_relpath)
2667
prefixes = [self._mapper.unmap(path) for path in relpaths]
2668
self._load_prefixes(prefixes)
2669
for prefix in prefixes:
2670
for suffix in self._kndx_cache[prefix][1]:
2671
result.add(prefix + (suffix,))
2674
def _load_prefixes(self, prefixes):
2675
"""Load the indices for prefixes."""
2677
for prefix in prefixes:
2678
if prefix not in self._kndx_cache:
2679
# the load_data interface writes to these variables.
2682
self._filename = prefix
2684
path = self._mapper.map(prefix) + '.kndx'
2685
fp = self._transport.get(path)
2687
# _load_data may raise NoSuchFile if the target knit is
2689
_load_data(self, fp)
2692
self._kndx_cache[prefix] = (self._cache, self._history)
2697
self._kndx_cache[prefix] = ({}, [])
2698
if type(self._mapper) is ConstantMapper:
2699
# preserve behaviour for revisions.kndx etc.
2700
self._init_index(path)
2705
missing_keys = _mod_index._missing_keys_from_parent_map
2707
def _partition_keys(self, keys):
2708
"""Turn keys into a dict of prefix:suffix_list."""
2711
prefix_keys = result.setdefault(key[:-1], [])
2712
prefix_keys.append(key[-1])
2715
def _dictionary_compress(self, keys):
2716
"""Dictionary compress keys.
2718
:param keys: The keys to generate references to.
2719
:return: A string representation of keys. keys which are present are
2720
dictionary compressed, and others are emitted as fulltext with a
2726
prefix = keys[0][:-1]
2727
cache = self._kndx_cache[prefix][0]
2729
if key[:-1] != prefix:
2730
# kndx indices cannot refer across partitioned storage.
2731
raise ValueError("mismatched prefixes for %r" % keys)
2732
if key[-1] in cache:
2733
# -- inlined lookup() --
2734
result_list.append(str(cache[key[-1]][5]))
2735
# -- end lookup () --
2737
result_list.append('.' + key[-1])
2738
return ' '.join(result_list)
2740
def _reset_cache(self):
2741
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2742
# (cache_dict, history_vector) for parsed kndx files.
2743
self._kndx_cache = {}
2744
self._scope = self._get_scope()
2745
allow_writes = self._allow_writes()
2751
def _sort_keys_by_io(self, keys, positions):
2752
"""Figure out an optimal order to read the records for the given keys.
2754
Sort keys, grouped by index and sorted by position.
2756
:param keys: A list of keys whose records we want to read. This will be
2758
:param positions: A dict, such as the one returned by
2759
_get_components_positions()
2762
def get_sort_key(key):
2763
index_memo = positions[key][1]
2764
# Group by prefix and position. index_memo[0] is the key, so it is
2765
# (file_id, revision_id) and we don't want to sort on revision_id,
2766
# index_memo[1] is the position, and index_memo[2] is the size,
2767
# which doesn't matter for the sort
2768
return index_memo[0][:-1], index_memo[1]
2769
return keys.sort(key=get_sort_key)
2771
_get_total_build_size = _get_total_build_size
2773
def _split_key(self, key):
2774
"""Split key into a prefix and suffix."""
2775
return key[:-1], key[-1]
2778
class _KeyRefs(object):
2780
def __init__(self, track_new_keys=False):
2781
# dict mapping 'key' to 'set of keys referring to that key'
2784
# set remembering all new keys
2785
self.new_keys = set()
2787
self.new_keys = None
2793
self.new_keys.clear()
2795
def add_references(self, key, refs):
2796
# Record the new references
2797
for referenced in refs:
2799
needed_by = self.refs[referenced]
2801
needed_by = self.refs[referenced] = set()
2803
# Discard references satisfied by the new key
2806
def get_new_keys(self):
2807
return self.new_keys
2809
def get_unsatisfied_refs(self):
2810
return self.refs.iterkeys()
2812
def _satisfy_refs_for_key(self, key):
2816
# No keys depended on this key. That's ok.
2819
def add_key(self, key):
2820
# satisfy refs for key, and remember that we've seen this key.
2821
self._satisfy_refs_for_key(key)
2822
if self.new_keys is not None:
2823
self.new_keys.add(key)
2825
def satisfy_refs_for_keys(self, keys):
2827
self._satisfy_refs_for_key(key)
2829
def get_referrers(self):
2831
for referrers in self.refs.itervalues():
2832
result.update(referrers)
2836
class _KnitGraphIndex(object):
2837
"""A KnitVersionedFiles index layered on GraphIndex."""
2839
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2840
add_callback=None, track_external_parent_refs=False):
2841
"""Construct a KnitGraphIndex on a graph_index.
2843
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2844
:param is_locked: A callback to check whether the object should answer
2846
:param deltas: Allow delta-compressed records.
2847
:param parents: If True, record knits parents, if not do not record
2849
:param add_callback: If not None, allow additions to the index and call
2850
this callback with a list of added GraphIndex nodes:
2851
[(node, value, node_refs), ...]
2852
:param is_locked: A callback, returns True if the index is locked and
2854
:param track_external_parent_refs: If True, record all external parent
2855
references parents from added records. These can be retrieved
2856
later by calling get_missing_parents().
2858
self._add_callback = add_callback
2859
self._graph_index = graph_index
2860
self._deltas = deltas
2861
self._parents = parents
2862
if deltas and not parents:
2863
# XXX: TODO: Delta tree and parent graph should be conceptually
2865
raise KnitCorrupt(self, "Cannot do delta compression without "
2867
self.has_graph = parents
2868
self._is_locked = is_locked
2869
self._missing_compression_parents = set()
2870
if track_external_parent_refs:
2871
self._key_dependencies = _KeyRefs()
2873
self._key_dependencies = None
2876
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2878
def add_records(self, records, random_id=False,
2879
missing_compression_parents=False):
2880
"""Add multiple records to the index.
2882
This function does not insert data into the Immutable GraphIndex
2883
backing the KnitGraphIndex, instead it prepares data for insertion by
2884
the caller and checks that it is safe to insert then calls
2885
self._add_callback with the prepared GraphIndex nodes.
2887
:param records: a list of tuples:
2888
(key, options, access_memo, parents).
2889
:param random_id: If True the ids being added were randomly generated
2890
and no check for existence will be performed.
2891
:param missing_compression_parents: If True the records being added are
2892
only compressed against texts already in the index (or inside
2893
records). If False the records all refer to unavailable texts (or
2894
texts inside records) as compression parents.
2896
if not self._add_callback:
2897
raise errors.ReadOnlyError(self)
2898
# we hope there are no repositories with inconsistent parentage
2902
compression_parents = set()
2903
key_dependencies = self._key_dependencies
2904
for (key, options, access_memo, parents) in records:
2906
parents = tuple(parents)
2907
if key_dependencies is not None:
2908
key_dependencies.add_references(key, parents)
2909
index, pos, size = access_memo
2910
if 'no-eol' in options:
2914
value += "%d %d" % (pos, size)
2915
if not self._deltas:
2916
if 'line-delta' in options:
2917
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2920
if 'line-delta' in options:
2921
node_refs = (parents, (parents[0],))
2922
if missing_compression_parents:
2923
compression_parents.add(parents[0])
2925
node_refs = (parents, ())
2927
node_refs = (parents, )
2930
raise KnitCorrupt(self, "attempt to add node with parents "
2931
"in parentless index.")
2933
keys[key] = (value, node_refs)
2936
present_nodes = self._get_entries(keys)
2937
for (index, key, value, node_refs) in present_nodes:
2938
if (value[0] != keys[key][0][0] or
2939
node_refs[:1] != keys[key][1][:1]):
2940
raise KnitCorrupt(self, "inconsistent details in add_records"
2941
": %s %s" % ((value, node_refs), keys[key]))
2945
for key, (value, node_refs) in keys.iteritems():
2946
result.append((key, value, node_refs))
2948
for key, (value, node_refs) in keys.iteritems():
2949
result.append((key, value))
2950
self._add_callback(result)
2951
if missing_compression_parents:
2952
# This may appear to be incorrect (it does not check for
2953
# compression parents that are in the existing graph index),
2954
# but such records won't have been buffered, so this is
2955
# actually correct: every entry when
2956
# missing_compression_parents==True either has a missing parent, or
2957
# a parent that is one of the keys in records.
2958
compression_parents.difference_update(keys)
2959
self._missing_compression_parents.update(compression_parents)
2960
# Adding records may have satisfied missing compression parents.
2961
self._missing_compression_parents.difference_update(keys)
2963
def scan_unvalidated_index(self, graph_index):
2964
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2966
This allows this _KnitGraphIndex to keep track of any missing
2967
compression parents we may want to have filled in to make those
2970
:param graph_index: A GraphIndex
2973
new_missing = graph_index.external_references(ref_list_num=1)
2974
new_missing.difference_update(self.get_parent_map(new_missing))
2975
self._missing_compression_parents.update(new_missing)
2976
if self._key_dependencies is not None:
2977
# Add parent refs from graph_index (and discard parent refs that
2978
# the graph_index has).
2979
for node in graph_index.iter_all_entries():
2980
self._key_dependencies.add_references(node[1], node[3][0])
2982
def get_missing_compression_parents(self):
2983
"""Return the keys of missing compression parents.
2985
Missing compression parents occur when a record stream was missing
2986
basis texts, or a index was scanned that had missing basis texts.
2988
return frozenset(self._missing_compression_parents)
2990
def get_missing_parents(self):
2991
"""Return the keys of missing parents."""
2992
# If updating this, you should also update
2993
# groupcompress._GCGraphIndex.get_missing_parents
2994
# We may have false positives, so filter those out.
2995
self._key_dependencies.satisfy_refs_for_keys(
2996
self.get_parent_map(self._key_dependencies.get_unsatisfied_refs()))
2997
return frozenset(self._key_dependencies.get_unsatisfied_refs())
2999
def _check_read(self):
3000
"""raise if reads are not permitted."""
3001
if not self._is_locked():
3002
raise errors.ObjectNotLocked(self)
3004
def _check_write_ok(self):
3005
"""Assert if writes are not permitted."""
3006
if not self._is_locked():
3007
raise errors.ObjectNotLocked(self)
3009
def _compression_parent(self, an_entry):
3010
# return the key that an_entry is compressed against, or None
3011
# Grab the second parent list (as deltas implies parents currently)
3012
compression_parents = an_entry[3][1]
3013
if not compression_parents:
3015
if len(compression_parents) != 1:
3016
raise AssertionError(
3017
"Too many compression parents: %r" % compression_parents)
3018
return compression_parents[0]
3020
def get_build_details(self, keys):
3021
"""Get the method, index_memo and compression parent for version_ids.
3023
Ghosts are omitted from the result.
3025
:param keys: An iterable of keys.
3026
:return: A dict of key:
3027
(index_memo, compression_parent, parents, record_details).
3029
opaque structure to pass to read_records to extract the raw
3032
Content that this record is built upon, may be None
3034
Logical parents of this node
3036
extra information about the content which needs to be passed to
3037
Factory.parse_record
3041
entries = self._get_entries(keys, False)
3042
for entry in entries:
3044
if not self._parents:
3047
parents = entry[3][0]
3048
if not self._deltas:
3049
compression_parent_key = None
3051
compression_parent_key = self._compression_parent(entry)
3052
noeol = (entry[2][0] == 'N')
3053
if compression_parent_key:
3054
method = 'line-delta'
3057
result[key] = (self._node_to_position(entry),
3058
compression_parent_key, parents,
3062
def _get_entries(self, keys, check_present=False):
3063
"""Get the entries for keys.
3065
:param keys: An iterable of index key tuples.
3070
for node in self._graph_index.iter_entries(keys):
3072
found_keys.add(node[1])
3074
# adapt parentless index to the rest of the code.
3075
for node in self._graph_index.iter_entries(keys):
3076
yield node[0], node[1], node[2], ()
3077
found_keys.add(node[1])
3079
missing_keys = keys.difference(found_keys)
3081
raise RevisionNotPresent(missing_keys.pop(), self)
3083
def get_method(self, key):
3084
"""Return compression method of specified key."""
3085
return self._get_method(self._get_node(key))
3087
def _get_method(self, node):
3088
if not self._deltas:
3090
if self._compression_parent(node):
3095
def _get_node(self, key):
3097
return list(self._get_entries([key]))[0]
3099
raise RevisionNotPresent(key, self)
3101
def get_options(self, key):
3102
"""Return a list representing options.
3106
node = self._get_node(key)
3107
options = [self._get_method(node)]
3108
if node[2][0] == 'N':
3109
options.append('no-eol')
3112
def find_ancestry(self, keys):
3113
"""See CombinedGraphIndex.find_ancestry()"""
3114
return self._graph_index.find_ancestry(keys, 0)
3116
def get_parent_map(self, keys):
3117
"""Get a map of the parents of keys.
3119
:param keys: The keys to look up parents for.
3120
:return: A mapping from keys to parents. Absent keys are absent from
3124
nodes = self._get_entries(keys)
3128
result[node[1]] = node[3][0]
3131
result[node[1]] = None
3134
def get_position(self, key):
3135
"""Return details needed to access the version.
3137
:return: a tuple (index, data position, size) to hand to the access
3138
logic to get the record.
3140
node = self._get_node(key)
3141
return self._node_to_position(node)
3143
has_key = _mod_index._has_key_from_parent_map
3146
"""Get all the keys in the collection.
3148
The keys are not ordered.
3151
return [node[1] for node in self._graph_index.iter_all_entries()]
3153
missing_keys = _mod_index._missing_keys_from_parent_map
3155
def _node_to_position(self, node):
3156
"""Convert an index value to position details."""
3157
bits = node[2][1:].split(' ')
3158
return node[0], int(bits[0]), int(bits[1])
3160
def _sort_keys_by_io(self, keys, positions):
3161
"""Figure out an optimal order to read the records for the given keys.
3163
Sort keys, grouped by index and sorted by position.
3165
:param keys: A list of keys whose records we want to read. This will be
3167
:param positions: A dict, such as the one returned by
3168
_get_components_positions()
3171
def get_index_memo(key):
3172
# index_memo is at offset [1]. It is made up of (GraphIndex,
3173
# position, size). GI is an object, which will be unique for each
3174
# pack file. This causes us to group by pack file, then sort by
3175
# position. Size doesn't matter, but it isn't worth breaking up the
3177
return positions[key][1]
3178
return keys.sort(key=get_index_memo)
3180
_get_total_build_size = _get_total_build_size
3183
class _KnitKeyAccess(object):
3184
"""Access to records in .knit files."""
3186
def __init__(self, transport, mapper):
3187
"""Create a _KnitKeyAccess with transport and mapper.
3189
:param transport: The transport the access object is rooted at.
3190
:param mapper: The mapper used to map keys to .knit files.
3192
self._transport = transport
3193
self._mapper = mapper
3195
def add_raw_records(self, key_sizes, raw_data):
3196
"""Add raw knit bytes to a storage area.
3198
The data is spooled to the container writer in one bytes-record per
3201
:param sizes: An iterable of tuples containing the key and size of each
3203
:param raw_data: A bytestring containing the data.
3204
:return: A list of memos to retrieve the record later. Each memo is an
3205
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3206
length), where the key is the record key.
3208
if type(raw_data) is not str:
3209
raise AssertionError(
3210
'data must be plain bytes was %s' % type(raw_data))
3213
# TODO: This can be tuned for writing to sftp and other servers where
3214
# append() is relatively expensive by grouping the writes to each key
3216
for key, size in key_sizes:
3217
path = self._mapper.map(key)
3219
base = self._transport.append_bytes(path + '.knit',
3220
raw_data[offset:offset+size])
3221
except errors.NoSuchFile:
3222
self._transport.mkdir(osutils.dirname(path))
3223
base = self._transport.append_bytes(path + '.knit',
3224
raw_data[offset:offset+size])
3228
result.append((key, base, size))
3232
"""Flush pending writes on this access object.
3234
For .knit files this is a no-op.
3238
def get_raw_records(self, memos_for_retrieval):
3239
"""Get the raw bytes for a records.
3241
:param memos_for_retrieval: An iterable containing the access memo for
3242
retrieving the bytes.
3243
:return: An iterator over the bytes of the records.
3245
# first pass, group into same-index request to minimise readv's issued.
3247
current_prefix = None
3248
for (key, offset, length) in memos_for_retrieval:
3249
if current_prefix == key[:-1]:
3250
current_list.append((offset, length))
3252
if current_prefix is not None:
3253
request_lists.append((current_prefix, current_list))
3254
current_prefix = key[:-1]
3255
current_list = [(offset, length)]
3256
# handle the last entry
3257
if current_prefix is not None:
3258
request_lists.append((current_prefix, current_list))
3259
for prefix, read_vector in request_lists:
3260
path = self._mapper.map(prefix) + '.knit'
3261
for pos, data in self._transport.readv(path, read_vector):
3265
class _DirectPackAccess(object):
3266
"""Access to data in one or more packs with less translation."""
3268
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
3269
"""Create a _DirectPackAccess object.
3271
:param index_to_packs: A dict mapping index objects to the transport
3272
and file names for obtaining data.
3273
:param reload_func: A function to call if we determine that the pack
3274
files have moved and we need to reload our caches. See
3275
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
3277
self._container_writer = None
3278
self._write_index = None
3279
self._indices = index_to_packs
3280
self._reload_func = reload_func
3281
self._flush_func = flush_func
3283
def add_raw_records(self, key_sizes, raw_data):
3284
"""Add raw knit bytes to a storage area.
3286
The data is spooled to the container writer in one bytes-record per
3289
:param sizes: An iterable of tuples containing the key and size of each
3291
:param raw_data: A bytestring containing the data.
3292
:return: A list of memos to retrieve the record later. Each memo is an
3293
opaque index memo. For _DirectPackAccess the memo is (index, pos,
3294
length), where the index field is the write_index object supplied
3295
to the PackAccess object.
3297
if type(raw_data) is not str:
3298
raise AssertionError(
3299
'data must be plain bytes was %s' % type(raw_data))
3302
for key, size in key_sizes:
3303
p_offset, p_length = self._container_writer.add_bytes_record(
3304
raw_data[offset:offset+size], [])
3306
result.append((self._write_index, p_offset, p_length))
3310
"""Flush pending writes on this access object.
3312
This will flush any buffered writes to a NewPack.
3314
if self._flush_func is not None:
3317
def get_raw_records(self, memos_for_retrieval):
3318
"""Get the raw bytes for a records.
3320
:param memos_for_retrieval: An iterable containing the (index, pos,
3321
length) memo for retrieving the bytes. The Pack access method
3322
looks up the pack to use for a given record in its index_to_pack
3324
:return: An iterator over the bytes of the records.
3326
# first pass, group into same-index requests
3328
current_index = None
3329
for (index, offset, length) in memos_for_retrieval:
3330
if current_index == index:
3331
current_list.append((offset, length))
3333
if current_index is not None:
3334
request_lists.append((current_index, current_list))
3335
current_index = index
3336
current_list = [(offset, length)]
3337
# handle the last entry
3338
if current_index is not None:
3339
request_lists.append((current_index, current_list))
3340
for index, offsets in request_lists:
3342
transport, path = self._indices[index]
3344
# A KeyError here indicates that someone has triggered an index
3345
# reload, and this index has gone missing, we need to start
3347
if self._reload_func is None:
3348
# If we don't have a _reload_func there is nothing that can
3351
raise errors.RetryWithNewPacks(index,
3352
reload_occurred=True,
3353
exc_info=sys.exc_info())
3355
reader = pack.make_readv_reader(transport, path, offsets)
3356
for names, read_func in reader.iter_records():
3357
yield read_func(None)
3358
except errors.NoSuchFile:
3359
# A NoSuchFile error indicates that a pack file has gone
3360
# missing on disk, we need to trigger a reload, and start over.
3361
if self._reload_func is None:
3363
raise errors.RetryWithNewPacks(transport.abspath(path),
3364
reload_occurred=False,
3365
exc_info=sys.exc_info())
3367
def set_writer(self, writer, index, transport_packname):
3368
"""Set a writer to use for adding data."""
3369
if index is not None:
3370
self._indices[index] = transport_packname
3371
self._container_writer = writer
3372
self._write_index = index
3374
def reload_or_raise(self, retry_exc):
3375
"""Try calling the reload function, or re-raise the original exception.
3377
This should be called after _DirectPackAccess raises a
3378
RetryWithNewPacks exception. This function will handle the common logic
3379
of determining when the error is fatal versus being temporary.
3380
It will also make sure that the original exception is raised, rather
3381
than the RetryWithNewPacks exception.
3383
If this function returns, then the calling function should retry
3384
whatever operation was being performed. Otherwise an exception will
3387
:param retry_exc: A RetryWithNewPacks exception.
3390
if self._reload_func is None:
3392
elif not self._reload_func():
3393
# The reload claimed that nothing changed
3394
if not retry_exc.reload_occurred:
3395
# If there wasn't an earlier reload, then we really were
3396
# expecting to find changes. We didn't find them, so this is a
3400
exc_class, exc_value, exc_traceback = retry_exc.exc_info
3401
raise exc_class, exc_value, exc_traceback
3404
# Deprecated, use PatienceSequenceMatcher instead
3405
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
3408
def annotate_knit(knit, revision_id):
3409
"""Annotate a knit with no cached annotations.
3411
This implementation is for knits with no cached annotations.
3412
It will work for knits with cached annotations, but this is not
3415
annotator = _KnitAnnotator(knit)
3416
return iter(annotator.annotate_flat(revision_id))
3419
class _KnitAnnotator(annotate.Annotator):
3420
"""Build up the annotations for a text."""
3422
def __init__(self, vf):
3423
annotate.Annotator.__init__(self, vf)
3425
# TODO: handle Nodes which cannot be extracted
3426
# self._ghosts = set()
3428
# Map from (key, parent_key) => matching_blocks, should be 'use once'
3429
self._matching_blocks = {}
3431
# KnitContent objects
3432
self._content_objects = {}
3433
# The number of children that depend on this fulltext content object
3434
self._num_compression_children = {}
3435
# Delta records that need their compression parent before they can be
3437
self._pending_deltas = {}
3438
# Fulltext records that are waiting for their parents fulltexts before
3439
# they can be yielded for annotation
3440
self._pending_annotation = {}
3442
self._all_build_details = {}
3444
def _get_build_graph(self, key):
3445
"""Get the graphs for building texts and annotations.
3447
The data you need for creating a full text may be different than the
3448
data you need to annotate that text. (At a minimum, you need both
3449
parents to create an annotation, but only need 1 parent to generate the
3452
:return: A list of (key, index_memo) records, suitable for
3453
passing to read_records_iter to start reading in the raw data from
3456
pending = set([key])
3459
self._num_needed_children[key] = 1
3461
# get all pending nodes
3462
this_iteration = pending
3463
build_details = self._vf._index.get_build_details(this_iteration)
3464
self._all_build_details.update(build_details)
3465
# new_nodes = self._vf._index._get_entries(this_iteration)
3467
for key, details in build_details.iteritems():
3468
(index_memo, compression_parent, parent_keys,
3469
record_details) = details
3470
self._parent_map[key] = parent_keys
3471
self._heads_provider = None
3472
records.append((key, index_memo))
3473
# Do we actually need to check _annotated_lines?
3474
pending.update([p for p in parent_keys
3475
if p not in self._all_build_details])
3477
for parent_key in parent_keys:
3478
if parent_key in self._num_needed_children:
3479
self._num_needed_children[parent_key] += 1
3481
self._num_needed_children[parent_key] = 1
3482
if compression_parent:
3483
if compression_parent in self._num_compression_children:
3484
self._num_compression_children[compression_parent] += 1
3486
self._num_compression_children[compression_parent] = 1
3488
missing_versions = this_iteration.difference(build_details.keys())
3489
if missing_versions:
3490
for key in missing_versions:
3491
if key in self._parent_map and key in self._text_cache:
3492
# We already have this text ready, we just need to
3493
# yield it later so we get it annotated
3495
parent_keys = self._parent_map[key]
3496
for parent_key in parent_keys:
3497
if parent_key in self._num_needed_children:
3498
self._num_needed_children[parent_key] += 1
3500
self._num_needed_children[parent_key] = 1
3501
pending.update([p for p in parent_keys
3502
if p not in self._all_build_details])
3504
raise errors.RevisionNotPresent(key, self._vf)
3505
# Generally we will want to read the records in reverse order, because
3506
# we find the parent nodes after the children
3508
return records, ann_keys
3510
def _get_needed_texts(self, key, pb=None):
3511
# if True or len(self._vf._fallback_vfs) > 0:
3512
if len(self._vf._fallback_vfs) > 0:
3513
# If we have fallbacks, go to the generic path
3514
for v in annotate.Annotator._get_needed_texts(self, key, pb=pb):
3519
records, ann_keys = self._get_build_graph(key)
3520
for idx, (sub_key, text, num_lines) in enumerate(
3521
self._extract_texts(records)):
3523
pb.update('annotating', idx, len(records))
3524
yield sub_key, text, num_lines
3525
for sub_key in ann_keys:
3526
text = self._text_cache[sub_key]
3527
num_lines = len(text) # bad assumption
3528
yield sub_key, text, num_lines
3530
except errors.RetryWithNewPacks, e:
3531
self._vf._access.reload_or_raise(e)
3532
# The cached build_details are no longer valid
3533
self._all_build_details.clear()
3535
def _cache_delta_blocks(self, key, compression_parent, delta, lines):
3536
parent_lines = self._text_cache[compression_parent]
3537
blocks = list(KnitContent.get_line_delta_blocks(delta, parent_lines, lines))
3538
self._matching_blocks[(key, compression_parent)] = blocks
3540
def _expand_record(self, key, parent_keys, compression_parent, record,
3543
if compression_parent:
3544
if compression_parent not in self._content_objects:
3545
# Waiting for the parent
3546
self._pending_deltas.setdefault(compression_parent, []).append(
3547
(key, parent_keys, record, record_details))
3549
# We have the basis parent, so expand the delta
3550
num = self._num_compression_children[compression_parent]
3553
base_content = self._content_objects.pop(compression_parent)
3554
self._num_compression_children.pop(compression_parent)
3556
self._num_compression_children[compression_parent] = num
3557
base_content = self._content_objects[compression_parent]
3558
# It is tempting to want to copy_base_content=False for the last
3559
# child object. However, whenever noeol=False,
3560
# self._text_cache[parent_key] is content._lines. So mutating it
3561
# gives very bad results.
3562
# The alternative is to copy the lines into text cache, but then we
3563
# are copying anyway, so just do it here.
3564
content, delta = self._vf._factory.parse_record(
3565
key, record, record_details, base_content,
3566
copy_base_content=True)
3569
content, _ = self._vf._factory.parse_record(
3570
key, record, record_details, None)
3571
if self._num_compression_children.get(key, 0) > 0:
3572
self._content_objects[key] = content
3573
lines = content.text()
3574
self._text_cache[key] = lines
3575
if delta is not None:
3576
self._cache_delta_blocks(key, compression_parent, delta, lines)
3579
def _get_parent_annotations_and_matches(self, key, text, parent_key):
3580
"""Get the list of annotations for the parent, and the matching lines.
3582
:param text: The opaque value given by _get_needed_texts
3583
:param parent_key: The key for the parent text
3584
:return: (parent_annotations, matching_blocks)
3585
parent_annotations is a list as long as the number of lines in
3587
matching_blocks is a list of (parent_idx, text_idx, len) tuples
3588
indicating which lines match between the two texts
3590
block_key = (key, parent_key)
3591
if block_key in self._matching_blocks:
3592
blocks = self._matching_blocks.pop(block_key)
3593
parent_annotations = self._annotations_cache[parent_key]
3594
return parent_annotations, blocks
3595
return annotate.Annotator._get_parent_annotations_and_matches(self,
3596
key, text, parent_key)
3598
def _process_pending(self, key):
3599
"""The content for 'key' was just processed.
3601
Determine if there is any more pending work to be processed.
3604
if key in self._pending_deltas:
3605
compression_parent = key
3606
children = self._pending_deltas.pop(key)
3607
for child_key, parent_keys, record, record_details in children:
3608
lines = self._expand_record(child_key, parent_keys,
3610
record, record_details)
3611
if self._check_ready_for_annotations(child_key, parent_keys):
3612
to_return.append(child_key)
3613
# Also check any children that are waiting for this parent to be
3615
if key in self._pending_annotation:
3616
children = self._pending_annotation.pop(key)
3617
to_return.extend([c for c, p_keys in children
3618
if self._check_ready_for_annotations(c, p_keys)])
3621
def _check_ready_for_annotations(self, key, parent_keys):
3622
"""return true if this text is ready to be yielded.
3624
Otherwise, this will return False, and queue the text into
3625
self._pending_annotation
3627
for parent_key in parent_keys:
3628
if parent_key not in self._annotations_cache:
3629
# still waiting on at least one parent text, so queue it up
3630
# Note that if there are multiple parents, we need to wait
3632
self._pending_annotation.setdefault(parent_key,
3633
[]).append((key, parent_keys))
3637
def _extract_texts(self, records):
3638
"""Extract the various texts needed based on records"""
3639
# We iterate in the order read, rather than a strict order requested
3640
# However, process what we can, and put off to the side things that
3641
# still need parents, cleaning them up when those parents are
3644
# 1) As 'records' are read, see if we can expand these records into
3645
# Content objects (and thus lines)
3646
# 2) If a given line-delta is waiting on its compression parent, it
3647
# gets queued up into self._pending_deltas, otherwise we expand
3648
# it, and put it into self._text_cache and self._content_objects
3649
# 3) If we expanded the text, we will then check to see if all
3650
# parents have also been processed. If so, this text gets yielded,
3651
# else this record gets set aside into pending_annotation
3652
# 4) Further, if we expanded the text in (2), we will then check to
3653
# see if there are any children in self._pending_deltas waiting to
3654
# also be processed. If so, we go back to (2) for those
3655
# 5) Further again, if we yielded the text, we can then check if that
3656
# 'unlocks' any of the texts in pending_annotations, which should
3657
# then get yielded as well
3658
# Note that both steps 4 and 5 are 'recursive' in that unlocking one
3659
# compression child could unlock yet another, and yielding a fulltext
3660
# will also 'unlock' the children that are waiting on that annotation.
3661
# (Though also, unlocking 1 parent's fulltext, does not unlock a child
3662
# if other parents are also waiting.)
3663
# We want to yield content before expanding child content objects, so
3664
# that we know when we can re-use the content lines, and the annotation
3665
# code can know when it can stop caching fulltexts, as well.
3667
# Children that are missing their compression parent
3669
for (key, record, digest) in self._vf._read_records_iter(records):
3671
details = self._all_build_details[key]
3672
(_, compression_parent, parent_keys, record_details) = details
3673
lines = self._expand_record(key, parent_keys, compression_parent,
3674
record, record_details)
3676
# Pending delta should be queued up
3678
# At this point, we may be able to yield this content, if all
3679
# parents are also finished
3680
yield_this_text = self._check_ready_for_annotations(key,
3683
# All parents present
3684
yield key, lines, len(lines)
3685
to_process = self._process_pending(key)
3687
this_process = to_process
3689
for key in this_process:
3690
lines = self._text_cache[key]
3691
yield key, lines, len(lines)
3692
to_process.extend(self._process_pending(key))
3695
from bzrlib._knit_load_data_pyx import _load_data_c as _load_data
3697
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
