1
# Copyright (C) 2006-2010 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(), """
83
from bzrlib.errors import (
91
RevisionAlreadyPresent,
94
from bzrlib.osutils import (
101
from bzrlib.versionedfile import (
102
AbsentContentFactory,
106
ChunkedContentFactory,
113
# TODO: Split out code specific to this format into an associated object.
115
# TODO: Can we put in some kind of value to check that the index and data
116
# files belong together?
118
# TODO: accommodate binaries, perhaps by storing a byte count
120
# TODO: function to check whole file
122
# TODO: atomically append data, then measure backwards from the cursor
123
# position after writing to work out where it was located. we may need to
124
# bypass python file buffering.
126
DATA_SUFFIX = '.knit'
127
INDEX_SUFFIX = '.kndx'
128
_STREAM_MIN_BUFFER_SIZE = 5*1024*1024
131
class KnitAdapter(object):
132
"""Base class for knit record adaption."""
134
def __init__(self, basis_vf):
135
"""Create an adapter which accesses full texts from basis_vf.
137
:param basis_vf: A versioned file to access basis texts of deltas from.
138
May be None for adapters that do not need to access basis texts.
140
self._data = KnitVersionedFiles(None, None)
141
self._annotate_factory = KnitAnnotateFactory()
142
self._plain_factory = KnitPlainFactory()
143
self._basis_vf = basis_vf
146
class FTAnnotatedToUnannotated(KnitAdapter):
147
"""An adapter from FT annotated knits to unannotated ones."""
149
def get_bytes(self, factory):
150
annotated_compressed_bytes = factory._raw_record
152
self._data._parse_record_unchecked(annotated_compressed_bytes)
153
content = self._annotate_factory.parse_fulltext(contents, rec[1])
154
size, bytes = self._data._record_to_data((rec[1],), rec[3], content.text())
158
class DeltaAnnotatedToUnannotated(KnitAdapter):
159
"""An adapter for deltas from annotated to unannotated."""
161
def get_bytes(self, factory):
162
annotated_compressed_bytes = factory._raw_record
164
self._data._parse_record_unchecked(annotated_compressed_bytes)
165
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
167
contents = self._plain_factory.lower_line_delta(delta)
168
size, bytes = self._data._record_to_data((rec[1],), rec[3], contents)
172
class FTAnnotatedToFullText(KnitAdapter):
173
"""An adapter from FT annotated knits to unannotated ones."""
175
def get_bytes(self, factory):
176
annotated_compressed_bytes = factory._raw_record
178
self._data._parse_record_unchecked(annotated_compressed_bytes)
179
content, delta = self._annotate_factory.parse_record(factory.key[-1],
180
contents, factory._build_details, None)
181
return ''.join(content.text())
184
class DeltaAnnotatedToFullText(KnitAdapter):
185
"""An adapter for deltas from annotated to unannotated."""
187
def get_bytes(self, factory):
188
annotated_compressed_bytes = factory._raw_record
190
self._data._parse_record_unchecked(annotated_compressed_bytes)
191
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
193
compression_parent = factory.parents[0]
194
basis_entry = self._basis_vf.get_record_stream(
195
[compression_parent], 'unordered', True).next()
196
if basis_entry.storage_kind == 'absent':
197
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
198
basis_chunks = basis_entry.get_bytes_as('chunked')
199
basis_lines = osutils.chunks_to_lines(basis_chunks)
200
# Manually apply the delta because we have one annotated content and
202
basis_content = PlainKnitContent(basis_lines, compression_parent)
203
basis_content.apply_delta(delta, rec[1])
204
basis_content._should_strip_eol = factory._build_details[1]
205
return ''.join(basis_content.text())
208
class FTPlainToFullText(KnitAdapter):
209
"""An adapter from FT plain knits to unannotated ones."""
211
def get_bytes(self, factory):
212
compressed_bytes = factory._raw_record
214
self._data._parse_record_unchecked(compressed_bytes)
215
content, delta = self._plain_factory.parse_record(factory.key[-1],
216
contents, factory._build_details, None)
217
return ''.join(content.text())
220
class DeltaPlainToFullText(KnitAdapter):
221
"""An adapter for deltas from annotated to unannotated."""
223
def get_bytes(self, factory):
224
compressed_bytes = factory._raw_record
226
self._data._parse_record_unchecked(compressed_bytes)
227
delta = self._plain_factory.parse_line_delta(contents, rec[1])
228
compression_parent = factory.parents[0]
229
# XXX: string splitting overhead.
230
basis_entry = self._basis_vf.get_record_stream(
231
[compression_parent], 'unordered', True).next()
232
if basis_entry.storage_kind == 'absent':
233
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
234
basis_chunks = basis_entry.get_bytes_as('chunked')
235
basis_lines = osutils.chunks_to_lines(basis_chunks)
236
basis_content = PlainKnitContent(basis_lines, compression_parent)
237
# Manually apply the delta because we have one annotated content and
239
content, _ = self._plain_factory.parse_record(rec[1], contents,
240
factory._build_details, basis_content)
241
return ''.join(content.text())
244
class KnitContentFactory(ContentFactory):
245
"""Content factory for streaming from knits.
247
:seealso ContentFactory:
250
def __init__(self, key, parents, build_details, sha1, raw_record,
251
annotated, knit=None, network_bytes=None):
252
"""Create a KnitContentFactory for key.
255
:param parents: The parents.
256
:param build_details: The build details as returned from
258
:param sha1: The sha1 expected from the full text of this object.
259
:param raw_record: The bytes of the knit data from disk.
260
:param annotated: True if the raw data is annotated.
261
:param network_bytes: None to calculate the network bytes on demand,
262
not-none if they are already known.
264
ContentFactory.__init__(self)
267
self.parents = parents
268
if build_details[0] == 'line-delta':
273
annotated_kind = 'annotated-'
276
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
277
self._raw_record = raw_record
278
self._network_bytes = network_bytes
279
self._build_details = build_details
282
def _create_network_bytes(self):
283
"""Create a fully serialised network version for transmission."""
284
# storage_kind, key, parents, Noeol, raw_record
285
key_bytes = '\x00'.join(self.key)
286
if self.parents is None:
287
parent_bytes = 'None:'
289
parent_bytes = '\t'.join('\x00'.join(key) for key in self.parents)
290
if self._build_details[1]:
294
network_bytes = "%s\n%s\n%s\n%s%s" % (self.storage_kind, key_bytes,
295
parent_bytes, noeol, self._raw_record)
296
self._network_bytes = network_bytes
298
def get_bytes_as(self, storage_kind):
299
if storage_kind == self.storage_kind:
300
if self._network_bytes is None:
301
self._create_network_bytes()
302
return self._network_bytes
303
if ('-ft-' in self.storage_kind and
304
storage_kind in ('chunked', 'fulltext')):
305
adapter_key = (self.storage_kind, 'fulltext')
306
adapter_factory = adapter_registry.get(adapter_key)
307
adapter = adapter_factory(None)
308
bytes = adapter.get_bytes(self)
309
if storage_kind == 'chunked':
313
if self._knit is not None:
314
# Not redundant with direct conversion above - that only handles
316
if storage_kind == 'chunked':
317
return self._knit.get_lines(self.key[0])
318
elif storage_kind == 'fulltext':
319
return self._knit.get_text(self.key[0])
320
raise errors.UnavailableRepresentation(self.key, storage_kind,
324
class LazyKnitContentFactory(ContentFactory):
325
"""A ContentFactory which can either generate full text or a wire form.
327
:seealso ContentFactory:
330
def __init__(self, key, parents, generator, first):
331
"""Create a LazyKnitContentFactory.
333
:param key: The key of the record.
334
:param parents: The parents of the record.
335
:param generator: A _ContentMapGenerator containing the record for this
337
:param first: Is this the first content object returned from generator?
338
if it is, its storage kind is knit-delta-closure, otherwise it is
339
knit-delta-closure-ref
342
self.parents = parents
344
self._generator = generator
345
self.storage_kind = "knit-delta-closure"
347
self.storage_kind = self.storage_kind + "-ref"
350
def get_bytes_as(self, storage_kind):
351
if storage_kind == self.storage_kind:
353
return self._generator._wire_bytes()
355
# all the keys etc are contained in the bytes returned in the
358
if storage_kind in ('chunked', 'fulltext'):
359
chunks = self._generator._get_one_work(self.key).text()
360
if storage_kind == 'chunked':
363
return ''.join(chunks)
364
raise errors.UnavailableRepresentation(self.key, storage_kind,
368
def knit_delta_closure_to_records(storage_kind, bytes, line_end):
369
"""Convert a network record to a iterator over stream records.
371
:param storage_kind: The storage kind of the record.
372
Must be 'knit-delta-closure'.
373
:param bytes: The bytes of the record on the network.
375
generator = _NetworkContentMapGenerator(bytes, line_end)
376
return generator.get_record_stream()
379
def knit_network_to_record(storage_kind, bytes, line_end):
380
"""Convert a network record to a record object.
382
:param storage_kind: The storage kind of the record.
383
:param bytes: The bytes of the record on the network.
386
line_end = bytes.find('\n', start)
387
key = tuple(bytes[start:line_end].split('\x00'))
389
line_end = bytes.find('\n', start)
390
parent_line = bytes[start:line_end]
391
if parent_line == 'None:':
395
[tuple(segment.split('\x00')) for segment in parent_line.split('\t')
398
noeol = bytes[start] == 'N'
399
if 'ft' in storage_kind:
402
method = 'line-delta'
403
build_details = (method, noeol)
405
raw_record = bytes[start:]
406
annotated = 'annotated' in storage_kind
407
return [KnitContentFactory(key, parents, build_details, None, raw_record,
408
annotated, network_bytes=bytes)]
411
class KnitContent(object):
412
"""Content of a knit version to which deltas can be applied.
414
This is always stored in memory as a list of lines with \n at the end,
415
plus a flag saying if the final ending is really there or not, because that
416
corresponds to the on-disk knit representation.
420
self._should_strip_eol = False
422
def apply_delta(self, delta, new_version_id):
423
"""Apply delta to this object to become new_version_id."""
424
raise NotImplementedError(self.apply_delta)
426
def line_delta_iter(self, new_lines):
427
"""Generate line-based delta from this content to new_lines."""
428
new_texts = new_lines.text()
429
old_texts = self.text()
430
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
431
for tag, i1, i2, j1, j2 in s.get_opcodes():
434
# ofrom, oto, length, data
435
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
437
def line_delta(self, new_lines):
438
return list(self.line_delta_iter(new_lines))
441
def get_line_delta_blocks(knit_delta, source, target):
442
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
443
target_len = len(target)
446
for s_begin, s_end, t_len, new_text in knit_delta:
447
true_n = s_begin - s_pos
450
# knit deltas do not provide reliable info about whether the
451
# last line of a file matches, due to eol handling.
452
if source[s_pos + n -1] != target[t_pos + n -1]:
455
yield s_pos, t_pos, n
456
t_pos += t_len + true_n
458
n = target_len - t_pos
460
if source[s_pos + n -1] != target[t_pos + n -1]:
463
yield s_pos, t_pos, n
464
yield s_pos + (target_len - t_pos), target_len, 0
467
class AnnotatedKnitContent(KnitContent):
468
"""Annotated content."""
470
def __init__(self, lines):
471
KnitContent.__init__(self)
475
"""Return a list of (origin, text) for each content line."""
476
lines = self._lines[:]
477
if self._should_strip_eol:
478
origin, last_line = lines[-1]
479
lines[-1] = (origin, last_line.rstrip('\n'))
482
def apply_delta(self, delta, new_version_id):
483
"""Apply delta to this object to become new_version_id."""
486
for start, end, count, delta_lines in delta:
487
lines[offset+start:offset+end] = delta_lines
488
offset = offset + (start - end) + count
492
lines = [text for origin, text in self._lines]
493
except ValueError, e:
494
# most commonly (only?) caused by the internal form of the knit
495
# missing annotation information because of a bug - see thread
497
raise KnitCorrupt(self,
498
"line in annotated knit missing annotation information: %s"
500
if self._should_strip_eol:
501
lines[-1] = lines[-1].rstrip('\n')
505
return AnnotatedKnitContent(self._lines[:])
508
class PlainKnitContent(KnitContent):
509
"""Unannotated content.
511
When annotate[_iter] is called on this content, the same version is reported
512
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
516
def __init__(self, lines, version_id):
517
KnitContent.__init__(self)
519
self._version_id = version_id
522
"""Return a list of (origin, text) for each content line."""
523
return [(self._version_id, line) for line in self._lines]
525
def apply_delta(self, delta, new_version_id):
526
"""Apply delta to this object to become new_version_id."""
529
for start, end, count, delta_lines in delta:
530
lines[offset+start:offset+end] = delta_lines
531
offset = offset + (start - end) + count
532
self._version_id = new_version_id
535
return PlainKnitContent(self._lines[:], self._version_id)
539
if self._should_strip_eol:
541
lines[-1] = lines[-1].rstrip('\n')
545
class _KnitFactory(object):
546
"""Base class for common Factory functions."""
548
def parse_record(self, version_id, record, record_details,
549
base_content, copy_base_content=True):
550
"""Parse a record into a full content object.
552
:param version_id: The official version id for this content
553
:param record: The data returned by read_records_iter()
554
:param record_details: Details about the record returned by
556
:param base_content: If get_build_details returns a compression_parent,
557
you must return a base_content here, else use None
558
:param copy_base_content: When building from the base_content, decide
559
you can either copy it and return a new object, or modify it in
561
:return: (content, delta) A Content object and possibly a line-delta,
564
method, noeol = record_details
565
if method == 'line-delta':
566
if copy_base_content:
567
content = base_content.copy()
569
content = base_content
570
delta = self.parse_line_delta(record, version_id)
571
content.apply_delta(delta, version_id)
573
content = self.parse_fulltext(record, version_id)
575
content._should_strip_eol = noeol
576
return (content, delta)
579
class KnitAnnotateFactory(_KnitFactory):
580
"""Factory for creating annotated Content objects."""
584
def make(self, lines, version_id):
585
num_lines = len(lines)
586
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
588
def parse_fulltext(self, content, version_id):
589
"""Convert fulltext to internal representation
591
fulltext content is of the format
592
revid(utf8) plaintext\n
593
internal representation is of the format:
596
# TODO: jam 20070209 The tests expect this to be returned as tuples,
597
# but the code itself doesn't really depend on that.
598
# Figure out a way to not require the overhead of turning the
599
# list back into tuples.
600
lines = [tuple(line.split(' ', 1)) for line in content]
601
return AnnotatedKnitContent(lines)
603
def parse_line_delta_iter(self, lines):
604
return iter(self.parse_line_delta(lines))
606
def parse_line_delta(self, lines, version_id, plain=False):
607
"""Convert a line based delta into internal representation.
609
line delta is in the form of:
610
intstart intend intcount
612
revid(utf8) newline\n
613
internal representation is
614
(start, end, count, [1..count tuples (revid, newline)])
616
:param plain: If True, the lines are returned as a plain
617
list without annotations, not as a list of (origin, content) tuples, i.e.
618
(start, end, count, [1..count newline])
625
def cache_and_return(line):
626
origin, text = line.split(' ', 1)
627
return cache.setdefault(origin, origin), text
629
# walk through the lines parsing.
630
# Note that the plain test is explicitly pulled out of the
631
# loop to minimise any performance impact
634
start, end, count = [int(n) for n in header.split(',')]
635
contents = [next().split(' ', 1)[1] for i in xrange(count)]
636
result.append((start, end, count, contents))
639
start, end, count = [int(n) for n in header.split(',')]
640
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
641
result.append((start, end, count, contents))
644
def get_fulltext_content(self, lines):
645
"""Extract just the content lines from a fulltext."""
646
return (line.split(' ', 1)[1] for line in lines)
648
def get_linedelta_content(self, lines):
649
"""Extract just the content from a line delta.
651
This doesn't return all of the extra information stored in a delta.
652
Only the actual content lines.
657
header = header.split(',')
658
count = int(header[2])
659
for i in xrange(count):
660
origin, text = next().split(' ', 1)
663
def lower_fulltext(self, content):
664
"""convert a fulltext content record into a serializable form.
666
see parse_fulltext which this inverts.
668
return ['%s %s' % (o, t) for o, t in content._lines]
670
def lower_line_delta(self, delta):
671
"""convert a delta into a serializable form.
673
See parse_line_delta which this inverts.
675
# TODO: jam 20070209 We only do the caching thing to make sure that
676
# the origin is a valid utf-8 line, eventually we could remove it
678
for start, end, c, lines in delta:
679
out.append('%d,%d,%d\n' % (start, end, c))
680
out.extend(origin + ' ' + text
681
for origin, text in lines)
684
def annotate(self, knit, key):
685
content = knit._get_content(key)
686
# adjust for the fact that serialised annotations are only key suffixes
688
if type(key) is tuple:
690
origins = content.annotate()
692
for origin, line in origins:
693
result.append((prefix + (origin,), line))
696
# XXX: This smells a bit. Why would key ever be a non-tuple here?
697
# Aren't keys defined to be tuples? -- spiv 20080618
698
return content.annotate()
701
class KnitPlainFactory(_KnitFactory):
702
"""Factory for creating plain Content objects."""
706
def make(self, lines, version_id):
707
return PlainKnitContent(lines, version_id)
709
def parse_fulltext(self, content, version_id):
710
"""This parses an unannotated fulltext.
712
Note that this is not a noop - the internal representation
713
has (versionid, line) - its just a constant versionid.
715
return self.make(content, version_id)
717
def parse_line_delta_iter(self, lines, version_id):
719
num_lines = len(lines)
720
while cur < num_lines:
723
start, end, c = [int(n) for n in header.split(',')]
724
yield start, end, c, lines[cur:cur+c]
727
def parse_line_delta(self, lines, version_id):
728
return list(self.parse_line_delta_iter(lines, version_id))
730
def get_fulltext_content(self, lines):
731
"""Extract just the content lines from a fulltext."""
734
def get_linedelta_content(self, lines):
735
"""Extract just the content from a line delta.
737
This doesn't return all of the extra information stored in a delta.
738
Only the actual content lines.
743
header = header.split(',')
744
count = int(header[2])
745
for i in xrange(count):
748
def lower_fulltext(self, content):
749
return content.text()
751
def lower_line_delta(self, delta):
753
for start, end, c, lines in delta:
754
out.append('%d,%d,%d\n' % (start, end, c))
758
def annotate(self, knit, key):
759
annotator = _KnitAnnotator(knit)
760
return annotator.annotate_flat(key)
764
def make_file_factory(annotated, mapper):
765
"""Create a factory for creating a file based KnitVersionedFiles.
767
This is only functional enough to run interface tests, it doesn't try to
768
provide a full pack environment.
770
:param annotated: knit annotations are wanted.
771
:param mapper: The mapper from keys to paths.
773
def factory(transport):
774
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
775
access = _KnitKeyAccess(transport, mapper)
776
return KnitVersionedFiles(index, access, annotated=annotated)
780
def make_pack_factory(graph, delta, keylength):
781
"""Create a factory for creating a pack based VersionedFiles.
783
This is only functional enough to run interface tests, it doesn't try to
784
provide a full pack environment.
786
:param graph: Store a graph.
787
:param delta: Delta compress contents.
788
:param keylength: How long should keys be.
790
def factory(transport):
791
parents = graph or delta
797
max_delta_chain = 200
800
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
801
key_elements=keylength)
802
stream = transport.open_write_stream('newpack')
803
writer = pack.ContainerWriter(stream.write)
805
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
806
deltas=delta, add_callback=graph_index.add_nodes)
807
access = _DirectPackAccess({})
808
access.set_writer(writer, graph_index, (transport, 'newpack'))
809
result = KnitVersionedFiles(index, access,
810
max_delta_chain=max_delta_chain)
811
result.stream = stream
812
result.writer = writer
817
def cleanup_pack_knit(versioned_files):
818
versioned_files.stream.close()
819
versioned_files.writer.end()
822
def _get_total_build_size(self, keys, positions):
823
"""Determine the total bytes to build these keys.
825
(helper function because _KnitGraphIndex and _KndxIndex work the same, but
826
don't inherit from a common base.)
828
:param keys: Keys that we want to build
829
:param positions: dict of {key, (info, index_memo, comp_parent)} (such
830
as returned by _get_components_positions)
831
:return: Number of bytes to build those keys
833
all_build_index_memos = {}
837
for key in build_keys:
838
# This is mostly for the 'stacked' case
839
# Where we will be getting the data from a fallback
840
if key not in positions:
842
_, index_memo, compression_parent = positions[key]
843
all_build_index_memos[key] = index_memo
844
if compression_parent not in all_build_index_memos:
845
next_keys.add(compression_parent)
846
build_keys = next_keys
847
return sum([index_memo[2] for index_memo
848
in all_build_index_memos.itervalues()])
851
class KnitVersionedFiles(VersionedFiles):
852
"""Storage for many versioned files using knit compression.
854
Backend storage is managed by indices and data objects.
856
:ivar _index: A _KnitGraphIndex or similar that can describe the
857
parents, graph, compression and data location of entries in this
858
KnitVersionedFiles. Note that this is only the index for
859
*this* vfs; if there are fallbacks they must be queried separately.
862
def __init__(self, index, data_access, max_delta_chain=200,
863
annotated=False, reload_func=None):
864
"""Create a KnitVersionedFiles with index and data_access.
866
:param index: The index for the knit data.
867
:param data_access: The access object to store and retrieve knit
869
:param max_delta_chain: The maximum number of deltas to permit during
870
insertion. Set to 0 to prohibit the use of deltas.
871
:param annotated: Set to True to cause annotations to be calculated and
872
stored during insertion.
873
:param reload_func: An function that can be called if we think we need
874
to reload the pack listing and try again. See
875
'bzrlib.repofmt.pack_repo.AggregateIndex' for the signature.
878
self._access = data_access
879
self._max_delta_chain = max_delta_chain
881
self._factory = KnitAnnotateFactory()
883
self._factory = KnitPlainFactory()
884
self._fallback_vfs = []
885
self._reload_func = reload_func
888
return "%s(%r, %r)" % (
889
self.__class__.__name__,
893
def add_fallback_versioned_files(self, a_versioned_files):
894
"""Add a source of texts for texts not present in this knit.
896
:param a_versioned_files: A VersionedFiles object.
898
self._fallback_vfs.append(a_versioned_files)
900
def add_lines(self, key, parents, lines, parent_texts=None,
901
left_matching_blocks=None, nostore_sha=None, random_id=False,
903
"""See VersionedFiles.add_lines()."""
904
self._index._check_write_ok()
905
self._check_add(key, lines, random_id, check_content)
907
# The caller might pass None if there is no graph data, but kndx
908
# indexes can't directly store that, so we give them
909
# an empty tuple instead.
911
line_bytes = ''.join(lines)
912
return self._add(key, lines, parents,
913
parent_texts, left_matching_blocks, nostore_sha, random_id,
914
line_bytes=line_bytes)
916
def _add_text(self, key, parents, text, nostore_sha=None, random_id=False):
917
"""See VersionedFiles._add_text()."""
918
self._index._check_write_ok()
919
self._check_add(key, None, random_id, check_content=False)
920
if text.__class__ is not str:
921
raise errors.BzrBadParameterUnicode("text")
923
# The caller might pass None if there is no graph data, but kndx
924
# indexes can't directly store that, so we give them
925
# an empty tuple instead.
927
return self._add(key, None, parents,
928
None, None, nostore_sha, random_id,
931
def _add(self, key, lines, parents, parent_texts,
932
left_matching_blocks, nostore_sha, random_id,
934
"""Add a set of lines on top of version specified by parents.
936
Any versions not present will be converted into ghosts.
938
:param lines: A list of strings where each one is a single line (has a
939
single newline at the end of the string) This is now optional
940
(callers can pass None). It is left in its location for backwards
941
compatibility. It should ''.join(lines) must == line_bytes
942
:param line_bytes: A single string containing the content
944
We pass both lines and line_bytes because different routes bring the
945
values to this function. And for memory efficiency, we don't want to
946
have to split/join on-demand.
948
# first thing, if the content is something we don't need to store, find
950
digest = sha_string(line_bytes)
951
if nostore_sha == digest:
952
raise errors.ExistingContent
955
if parent_texts is None:
957
# Do a single query to ascertain parent presence; we only compress
958
# against parents in the same kvf.
959
present_parent_map = self._index.get_parent_map(parents)
960
for parent in parents:
961
if parent in present_parent_map:
962
present_parents.append(parent)
964
# Currently we can only compress against the left most present parent.
965
if (len(present_parents) == 0 or
966
present_parents[0] != parents[0]):
969
# To speed the extract of texts the delta chain is limited
970
# to a fixed number of deltas. This should minimize both
971
# I/O and the time spend applying deltas.
972
delta = self._check_should_delta(present_parents[0])
974
text_length = len(line_bytes)
977
# Note: line_bytes is not modified to add a newline, that is tracked
978
# via the no_eol flag. 'lines' *is* modified, because that is the
979
# general values needed by the Content code.
980
if line_bytes and line_bytes[-1] != '\n':
981
options.append('no-eol')
983
# Copy the existing list, or create a new one
985
lines = osutils.split_lines(line_bytes)
988
# Replace the last line with one that ends in a final newline
989
lines[-1] = lines[-1] + '\n'
991
lines = osutils.split_lines(line_bytes)
993
for element in key[:-1]:
994
if type(element) is not str:
995
raise TypeError("key contains non-strings: %r" % (key,))
997
key = key[:-1] + ('sha1:' + digest,)
998
elif type(key[-1]) is not str:
999
raise TypeError("key contains non-strings: %r" % (key,))
1000
# Knit hunks are still last-element only
1001
version_id = key[-1]
1002
content = self._factory.make(lines, version_id)
1004
# Hint to the content object that its text() call should strip the
1006
content._should_strip_eol = True
1007
if delta or (self._factory.annotated and len(present_parents) > 0):
1008
# Merge annotations from parent texts if needed.
1009
delta_hunks = self._merge_annotations(content, present_parents,
1010
parent_texts, delta, self._factory.annotated,
1011
left_matching_blocks)
1014
options.append('line-delta')
1015
store_lines = self._factory.lower_line_delta(delta_hunks)
1016
size, bytes = self._record_to_data(key, digest,
1019
options.append('fulltext')
1020
# isinstance is slower and we have no hierarchy.
1021
if self._factory.__class__ is KnitPlainFactory:
1022
# Use the already joined bytes saving iteration time in
1024
dense_lines = [line_bytes]
1026
dense_lines.append('\n')
1027
size, bytes = self._record_to_data(key, digest,
1030
# get mixed annotation + content and feed it into the
1032
store_lines = self._factory.lower_fulltext(content)
1033
size, bytes = self._record_to_data(key, digest,
1036
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
1037
self._index.add_records(
1038
((key, options, access_memo, parents),),
1039
random_id=random_id)
1040
return digest, text_length, content
1042
def annotate(self, key):
1043
"""See VersionedFiles.annotate."""
1044
return self._factory.annotate(self, key)
1046
def get_annotator(self):
1047
return _KnitAnnotator(self)
1049
def check(self, progress_bar=None, keys=None):
1050
"""See VersionedFiles.check()."""
1052
return self._logical_check()
1054
# At the moment, check does not extra work over get_record_stream
1055
return self.get_record_stream(keys, 'unordered', True)
1057
def _logical_check(self):
1058
# This doesn't actually test extraction of everything, but that will
1059
# impact 'bzr check' substantially, and needs to be integrated with
1060
# care. However, it does check for the obvious problem of a delta with
1062
keys = self._index.keys()
1063
parent_map = self.get_parent_map(keys)
1065
if self._index.get_method(key) != 'fulltext':
1066
compression_parent = parent_map[key][0]
1067
if compression_parent not in parent_map:
1068
raise errors.KnitCorrupt(self,
1069
"Missing basis parent %s for %s" % (
1070
compression_parent, key))
1071
for fallback_vfs in self._fallback_vfs:
1072
fallback_vfs.check()
1074
def _check_add(self, key, lines, random_id, check_content):
1075
"""check that version_id and lines are safe to add."""
1076
version_id = key[-1]
1077
if version_id is not None:
1078
if contains_whitespace(version_id):
1079
raise InvalidRevisionId(version_id, self)
1080
self.check_not_reserved_id(version_id)
1081
# TODO: If random_id==False and the key is already present, we should
1082
# probably check that the existing content is identical to what is
1083
# being inserted, and otherwise raise an exception. This would make
1084
# the bundle code simpler.
1086
self._check_lines_not_unicode(lines)
1087
self._check_lines_are_lines(lines)
1089
def _check_header(self, key, line):
1090
rec = self._split_header(line)
1091
self._check_header_version(rec, key[-1])
1094
def _check_header_version(self, rec, version_id):
1095
"""Checks the header version on original format knit records.
1097
These have the last component of the key embedded in the record.
1099
if rec[1] != version_id:
1100
raise KnitCorrupt(self,
1101
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1103
def _check_should_delta(self, parent):
1104
"""Iterate back through the parent listing, looking for a fulltext.
1106
This is used when we want to decide whether to add a delta or a new
1107
fulltext. It searches for _max_delta_chain parents. When it finds a
1108
fulltext parent, it sees if the total size of the deltas leading up to
1109
it is large enough to indicate that we want a new full text anyway.
1111
Return True if we should create a new delta, False if we should use a
1115
fulltext_size = None
1116
for count in xrange(self._max_delta_chain):
1118
# Note that this only looks in the index of this particular
1119
# KnitVersionedFiles, not in the fallbacks. This ensures that
1120
# we won't store a delta spanning physical repository
1122
build_details = self._index.get_build_details([parent])
1123
parent_details = build_details[parent]
1124
except (RevisionNotPresent, KeyError), e:
1125
# Some basis is not locally present: always fulltext
1127
index_memo, compression_parent, _, _ = parent_details
1128
_, _, size = index_memo
1129
if compression_parent is None:
1130
fulltext_size = size
1133
# We don't explicitly check for presence because this is in an
1134
# inner loop, and if it's missing it'll fail anyhow.
1135
parent = compression_parent
1137
# We couldn't find a fulltext, so we must create a new one
1139
# Simple heuristic - if the total I/O wold be greater as a delta than
1140
# the originally installed fulltext, we create a new fulltext.
1141
return fulltext_size > delta_size
1143
def _build_details_to_components(self, build_details):
1144
"""Convert a build_details tuple to a position tuple."""
1145
# record_details, access_memo, compression_parent
1146
return build_details[3], build_details[0], build_details[1]
1148
def _get_components_positions(self, keys, allow_missing=False):
1149
"""Produce a map of position data for the components of keys.
1151
This data is intended to be used for retrieving the knit records.
1153
A dict of key to (record_details, index_memo, next, parents) is
1155
method is the way referenced data should be applied.
1156
index_memo is the handle to pass to the data access to actually get the
1158
next is the build-parent of the version, or None for fulltexts.
1159
parents is the version_ids of the parents of this version
1161
:param allow_missing: If True do not raise an error on a missing component,
1165
pending_components = keys
1166
while pending_components:
1167
build_details = self._index.get_build_details(pending_components)
1168
current_components = set(pending_components)
1169
pending_components = set()
1170
for key, details in build_details.iteritems():
1171
(index_memo, compression_parent, parents,
1172
record_details) = details
1173
method = record_details[0]
1174
if compression_parent is not None:
1175
pending_components.add(compression_parent)
1176
component_data[key] = self._build_details_to_components(details)
1177
missing = current_components.difference(build_details)
1178
if missing and not allow_missing:
1179
raise errors.RevisionNotPresent(missing.pop(), self)
1180
return component_data
1182
def _get_content(self, key, parent_texts={}):
1183
"""Returns a content object that makes up the specified
1185
cached_version = parent_texts.get(key, None)
1186
if cached_version is not None:
1187
# Ensure the cache dict is valid.
1188
if not self.get_parent_map([key]):
1189
raise RevisionNotPresent(key, self)
1190
return cached_version
1191
generator = _VFContentMapGenerator(self, [key])
1192
return generator._get_content(key)
1194
def get_known_graph_ancestry(self, keys):
1195
"""Get a KnownGraph instance with the ancestry of keys."""
1196
parent_map, missing_keys = self._index.find_ancestry(keys)
1197
for fallback in self._fallback_vfs:
1198
if not missing_keys:
1200
(f_parent_map, f_missing_keys) = fallback._index.find_ancestry(
1202
parent_map.update(f_parent_map)
1203
missing_keys = f_missing_keys
1204
kg = _mod_graph.KnownGraph(parent_map)
1207
def get_parent_map(self, keys):
1208
"""Get a map of the graph parents of keys.
1210
:param keys: The keys to look up parents for.
1211
:return: A mapping from keys to parents. Absent keys are absent from
1214
return self._get_parent_map_with_sources(keys)[0]
1216
def _get_parent_map_with_sources(self, keys):
1217
"""Get a map of the parents of keys.
1219
:param keys: The keys to look up parents for.
1220
:return: A tuple. The first element is a mapping from keys to parents.
1221
Absent keys are absent from the mapping. The second element is a
1222
list with the locations each key was found in. The first element
1223
is the in-this-knit parents, the second the first fallback source,
1227
sources = [self._index] + self._fallback_vfs
1230
for source in sources:
1233
new_result = source.get_parent_map(missing)
1234
source_results.append(new_result)
1235
result.update(new_result)
1236
missing.difference_update(set(new_result))
1237
return result, source_results
1239
def _get_record_map(self, keys, allow_missing=False):
1240
"""Produce a dictionary of knit records.
1242
:return: {key:(record, record_details, digest, next)}
1244
data returned from read_records (a KnitContentobject)
1246
opaque information to pass to parse_record
1248
SHA1 digest of the full text after all steps are done
1250
build-parent of the version, i.e. the leftmost ancestor.
1251
Will be None if the record is not a delta.
1252
:param keys: The keys to build a map for
1253
:param allow_missing: If some records are missing, rather than
1254
error, just return the data that could be generated.
1256
raw_map = self._get_record_map_unparsed(keys,
1257
allow_missing=allow_missing)
1258
return self._raw_map_to_record_map(raw_map)
1260
def _raw_map_to_record_map(self, raw_map):
1261
"""Parse the contents of _get_record_map_unparsed.
1263
:return: see _get_record_map.
1267
data, record_details, next = raw_map[key]
1268
content, digest = self._parse_record(key[-1], data)
1269
result[key] = content, record_details, digest, next
1272
def _get_record_map_unparsed(self, keys, allow_missing=False):
1273
"""Get the raw data for reconstructing keys without parsing it.
1275
:return: A dict suitable for parsing via _raw_map_to_record_map.
1276
key-> raw_bytes, (method, noeol), compression_parent
1278
# This retries the whole request if anything fails. Potentially we
1279
# could be a bit more selective. We could track the keys whose records
1280
# we have successfully found, and then only request the new records
1281
# from there. However, _get_components_positions grabs the whole build
1282
# chain, which means we'll likely try to grab the same records again
1283
# anyway. Also, can the build chains change as part of a pack
1284
# operation? We wouldn't want to end up with a broken chain.
1287
position_map = self._get_components_positions(keys,
1288
allow_missing=allow_missing)
1289
# key = component_id, r = record_details, i_m = index_memo,
1291
records = [(key, i_m) for key, (r, i_m, n)
1292
in position_map.iteritems()]
1293
# Sort by the index memo, so that we request records from the
1294
# same pack file together, and in forward-sorted order
1295
records.sort(key=operator.itemgetter(1))
1297
for key, data in self._read_records_iter_unchecked(records):
1298
(record_details, index_memo, next) = position_map[key]
1299
raw_record_map[key] = data, record_details, next
1300
return raw_record_map
1301
except errors.RetryWithNewPacks, e:
1302
self._access.reload_or_raise(e)
1305
def _split_by_prefix(cls, keys):
1306
"""For the given keys, split them up based on their prefix.
1308
To keep memory pressure somewhat under control, split the
1309
requests back into per-file-id requests, otherwise "bzr co"
1310
extracts the full tree into memory before writing it to disk.
1311
This should be revisited if _get_content_maps() can ever cross
1314
The keys for a given file_id are kept in the same relative order.
1315
Ordering between file_ids is not, though prefix_order will return the
1316
order that the key was first seen.
1318
:param keys: An iterable of key tuples
1319
:return: (split_map, prefix_order)
1320
split_map A dictionary mapping prefix => keys
1321
prefix_order The order that we saw the various prefixes
1323
split_by_prefix = {}
1331
if prefix in split_by_prefix:
1332
split_by_prefix[prefix].append(key)
1334
split_by_prefix[prefix] = [key]
1335
prefix_order.append(prefix)
1336
return split_by_prefix, prefix_order
1338
def _group_keys_for_io(self, keys, non_local_keys, positions,
1339
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1340
"""For the given keys, group them into 'best-sized' requests.
1342
The idea is to avoid making 1 request per file, but to never try to
1343
unpack an entire 1.5GB source tree in a single pass. Also when
1344
possible, we should try to group requests to the same pack file
1347
:return: list of (keys, non_local) tuples that indicate what keys
1348
should be fetched next.
1350
# TODO: Ideally we would group on 2 factors. We want to extract texts
1351
# from the same pack file together, and we want to extract all
1352
# the texts for a given build-chain together. Ultimately it
1353
# probably needs a better global view.
1354
total_keys = len(keys)
1355
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1356
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1358
cur_non_local = set()
1362
for prefix in prefix_order:
1363
keys = prefix_split_keys[prefix]
1364
non_local = prefix_split_non_local_keys.get(prefix, [])
1366
this_size = self._index._get_total_build_size(keys, positions)
1367
cur_size += this_size
1368
cur_keys.extend(keys)
1369
cur_non_local.update(non_local)
1370
if cur_size > _min_buffer_size:
1371
result.append((cur_keys, cur_non_local))
1372
sizes.append(cur_size)
1374
cur_non_local = set()
1377
result.append((cur_keys, cur_non_local))
1378
sizes.append(cur_size)
1381
def get_record_stream(self, keys, ordering, include_delta_closure):
1382
"""Get a stream of records for keys.
1384
:param keys: The keys to include.
1385
:param ordering: Either 'unordered' or 'topological'. A topologically
1386
sorted stream has compression parents strictly before their
1388
:param include_delta_closure: If True then the closure across any
1389
compression parents will be included (in the opaque data).
1390
:return: An iterator of ContentFactory objects, each of which is only
1391
valid until the iterator is advanced.
1393
# keys might be a generator
1397
if not self._index.has_graph:
1398
# Cannot sort when no graph has been stored.
1399
ordering = 'unordered'
1401
remaining_keys = keys
1404
keys = set(remaining_keys)
1405
for content_factory in self._get_remaining_record_stream(keys,
1406
ordering, include_delta_closure):
1407
remaining_keys.discard(content_factory.key)
1408
yield content_factory
1410
except errors.RetryWithNewPacks, e:
1411
self._access.reload_or_raise(e)
1413
def _get_remaining_record_stream(self, keys, ordering,
1414
include_delta_closure):
1415
"""This function is the 'retry' portion for get_record_stream."""
1416
if include_delta_closure:
1417
positions = self._get_components_positions(keys, allow_missing=True)
1419
build_details = self._index.get_build_details(keys)
1421
# (record_details, access_memo, compression_parent_key)
1422
positions = dict((key, self._build_details_to_components(details))
1423
for key, details in build_details.iteritems())
1424
absent_keys = keys.difference(set(positions))
1425
# There may be more absent keys : if we're missing the basis component
1426
# and are trying to include the delta closure.
1427
# XXX: We should not ever need to examine remote sources because we do
1428
# not permit deltas across versioned files boundaries.
1429
if include_delta_closure:
1430
needed_from_fallback = set()
1431
# Build up reconstructable_keys dict. key:True in this dict means
1432
# the key can be reconstructed.
1433
reconstructable_keys = {}
1437
chain = [key, positions[key][2]]
1439
needed_from_fallback.add(key)
1442
while chain[-1] is not None:
1443
if chain[-1] in reconstructable_keys:
1444
result = reconstructable_keys[chain[-1]]
1448
chain.append(positions[chain[-1]][2])
1450
# missing basis component
1451
needed_from_fallback.add(chain[-1])
1454
for chain_key in chain[:-1]:
1455
reconstructable_keys[chain_key] = result
1457
needed_from_fallback.add(key)
1458
# Double index lookups here : need a unified api ?
1459
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1460
if ordering in ('topological', 'groupcompress'):
1461
if ordering == 'topological':
1462
# Global topological sort
1463
present_keys = tsort.topo_sort(global_map)
1465
present_keys = sort_groupcompress(global_map)
1466
# Now group by source:
1468
current_source = None
1469
for key in present_keys:
1470
for parent_map in parent_maps:
1471
if key in parent_map:
1472
key_source = parent_map
1474
if current_source is not key_source:
1475
source_keys.append((key_source, []))
1476
current_source = key_source
1477
source_keys[-1][1].append(key)
1479
if ordering != 'unordered':
1480
raise AssertionError('valid values for ordering are:'
1481
' "unordered", "groupcompress" or "topological" not: %r'
1483
# Just group by source; remote sources first.
1486
for parent_map in reversed(parent_maps):
1487
source_keys.append((parent_map, []))
1488
for key in parent_map:
1489
present_keys.append(key)
1490
source_keys[-1][1].append(key)
1491
# We have been requested to return these records in an order that
1492
# suits us. So we ask the index to give us an optimally sorted
1494
for source, sub_keys in source_keys:
1495
if source is parent_maps[0]:
1496
# Only sort the keys for this VF
1497
self._index._sort_keys_by_io(sub_keys, positions)
1498
absent_keys = keys - set(global_map)
1499
for key in absent_keys:
1500
yield AbsentContentFactory(key)
1501
# restrict our view to the keys we can answer.
1502
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1503
# XXX: At that point we need to consider the impact of double reads by
1504
# utilising components multiple times.
1505
if include_delta_closure:
1506
# XXX: get_content_maps performs its own index queries; allow state
1508
non_local_keys = needed_from_fallback - absent_keys
1509
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1512
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1515
for record in generator.get_record_stream():
1518
for source, keys in source_keys:
1519
if source is parent_maps[0]:
1520
# this KnitVersionedFiles
1521
records = [(key, positions[key][1]) for key in keys]
1522
for key, raw_data in self._read_records_iter_unchecked(records):
1523
(record_details, index_memo, _) = positions[key]
1524
yield KnitContentFactory(key, global_map[key],
1525
record_details, None, raw_data, self._factory.annotated, None)
1527
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1528
for record in vf.get_record_stream(keys, ordering,
1529
include_delta_closure):
1532
def get_sha1s(self, keys):
1533
"""See VersionedFiles.get_sha1s()."""
1535
record_map = self._get_record_map(missing, allow_missing=True)
1537
for key, details in record_map.iteritems():
1538
if key not in missing:
1540
# record entry 2 is the 'digest'.
1541
result[key] = details[2]
1542
missing.difference_update(set(result))
1543
for source in self._fallback_vfs:
1546
new_result = source.get_sha1s(missing)
1547
result.update(new_result)
1548
missing.difference_update(set(new_result))
1551
def insert_record_stream(self, stream):
1552
"""Insert a record stream into this container.
1554
:param stream: A stream of records to insert.
1556
:seealso VersionedFiles.get_record_stream:
1558
def get_adapter(adapter_key):
1560
return adapters[adapter_key]
1562
adapter_factory = adapter_registry.get(adapter_key)
1563
adapter = adapter_factory(self)
1564
adapters[adapter_key] = adapter
1567
if self._factory.annotated:
1568
# self is annotated, we need annotated knits to use directly.
1569
annotated = "annotated-"
1572
# self is not annotated, but we can strip annotations cheaply.
1574
convertibles = set(["knit-annotated-ft-gz"])
1575
if self._max_delta_chain:
1576
delta_types.add("knit-annotated-delta-gz")
1577
convertibles.add("knit-annotated-delta-gz")
1578
# The set of types we can cheaply adapt without needing basis texts.
1579
native_types = set()
1580
if self._max_delta_chain:
1581
native_types.add("knit-%sdelta-gz" % annotated)
1582
delta_types.add("knit-%sdelta-gz" % annotated)
1583
native_types.add("knit-%sft-gz" % annotated)
1584
knit_types = native_types.union(convertibles)
1586
# Buffer all index entries that we can't add immediately because their
1587
# basis parent is missing. We don't buffer all because generating
1588
# annotations may require access to some of the new records. However we
1589
# can't generate annotations from new deltas until their basis parent
1590
# is present anyway, so we get away with not needing an index that
1591
# includes the new keys.
1593
# See <http://launchpad.net/bugs/300177> about ordering of compression
1594
# parents in the records - to be conservative, we insist that all
1595
# parents must be present to avoid expanding to a fulltext.
1597
# key = basis_parent, value = index entry to add
1598
buffered_index_entries = {}
1599
for record in stream:
1600
kind = record.storage_kind
1601
if kind.startswith('knit-') and kind.endswith('-gz'):
1602
# Check that the ID in the header of the raw knit bytes matches
1603
# the record metadata.
1604
raw_data = record._raw_record
1605
df, rec = self._parse_record_header(record.key, raw_data)
1608
parents = record.parents
1609
if record.storage_kind in delta_types:
1610
# TODO: eventually the record itself should track
1611
# compression_parent
1612
compression_parent = parents[0]
1614
compression_parent = None
1615
# Raise an error when a record is missing.
1616
if record.storage_kind == 'absent':
1617
raise RevisionNotPresent([record.key], self)
1618
elif ((record.storage_kind in knit_types)
1619
and (compression_parent is None
1620
or not self._fallback_vfs
1621
or self._index.has_key(compression_parent)
1622
or not self.has_key(compression_parent))):
1623
# we can insert the knit record literally if either it has no
1624
# compression parent OR we already have its basis in this kvf
1625
# OR the basis is not present even in the fallbacks. In the
1626
# last case it will either turn up later in the stream and all
1627
# will be well, or it won't turn up at all and we'll raise an
1630
# TODO: self.has_key is somewhat redundant with
1631
# self._index.has_key; we really want something that directly
1632
# asks if it's only present in the fallbacks. -- mbp 20081119
1633
if record.storage_kind not in native_types:
1635
adapter_key = (record.storage_kind, "knit-delta-gz")
1636
adapter = get_adapter(adapter_key)
1638
adapter_key = (record.storage_kind, "knit-ft-gz")
1639
adapter = get_adapter(adapter_key)
1640
bytes = adapter.get_bytes(record)
1642
# It's a knit record, it has a _raw_record field (even if
1643
# it was reconstituted from a network stream).
1644
bytes = record._raw_record
1645
options = [record._build_details[0]]
1646
if record._build_details[1]:
1647
options.append('no-eol')
1648
# Just blat it across.
1649
# Note: This does end up adding data on duplicate keys. As
1650
# modern repositories use atomic insertions this should not
1651
# lead to excessive growth in the event of interrupted fetches.
1652
# 'knit' repositories may suffer excessive growth, but as a
1653
# deprecated format this is tolerable. It can be fixed if
1654
# needed by in the kndx index support raising on a duplicate
1655
# add with identical parents and options.
1656
access_memo = self._access.add_raw_records(
1657
[(record.key, len(bytes))], bytes)[0]
1658
index_entry = (record.key, options, access_memo, parents)
1659
if 'fulltext' not in options:
1660
# Not a fulltext, so we need to make sure the compression
1661
# parent will also be present.
1662
# Note that pack backed knits don't need to buffer here
1663
# because they buffer all writes to the transaction level,
1664
# but we don't expose that difference at the index level. If
1665
# the query here has sufficient cost to show up in
1666
# profiling we should do that.
1668
# They're required to be physically in this
1669
# KnitVersionedFiles, not in a fallback.
1670
if not self._index.has_key(compression_parent):
1671
pending = buffered_index_entries.setdefault(
1672
compression_parent, [])
1673
pending.append(index_entry)
1676
self._index.add_records([index_entry])
1677
elif record.storage_kind == 'chunked':
1678
self.add_lines(record.key, parents,
1679
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1681
# Not suitable for direct insertion as a
1682
# delta, either because it's not the right format, or this
1683
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1684
# 0) or because it depends on a base only present in the
1686
self._access.flush()
1688
# Try getting a fulltext directly from the record.
1689
bytes = record.get_bytes_as('fulltext')
1690
except errors.UnavailableRepresentation:
1691
adapter_key = record.storage_kind, 'fulltext'
1692
adapter = get_adapter(adapter_key)
1693
bytes = adapter.get_bytes(record)
1694
lines = split_lines(bytes)
1696
self.add_lines(record.key, parents, lines)
1697
except errors.RevisionAlreadyPresent:
1699
# Add any records whose basis parent is now available.
1701
added_keys = [record.key]
1703
key = added_keys.pop(0)
1704
if key in buffered_index_entries:
1705
index_entries = buffered_index_entries[key]
1706
self._index.add_records(index_entries)
1708
[index_entry[0] for index_entry in index_entries])
1709
del buffered_index_entries[key]
1710
if buffered_index_entries:
1711
# There were index entries buffered at the end of the stream,
1712
# So these need to be added (if the index supports holding such
1713
# entries for later insertion)
1715
for key in buffered_index_entries:
1716
index_entries = buffered_index_entries[key]
1717
all_entries.extend(index_entries)
1718
self._index.add_records(
1719
all_entries, missing_compression_parents=True)
1721
def get_missing_compression_parent_keys(self):
1722
"""Return an iterable of keys of missing compression parents.
1724
Check this after calling insert_record_stream to find out if there are
1725
any missing compression parents. If there are, the records that
1726
depend on them are not able to be inserted safely. For atomic
1727
KnitVersionedFiles built on packs, the transaction should be aborted or
1728
suspended - commit will fail at this point. Nonatomic knits will error
1729
earlier because they have no staging area to put pending entries into.
1731
return self._index.get_missing_compression_parents()
1733
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1734
"""Iterate over the lines in the versioned files from keys.
1736
This may return lines from other keys. Each item the returned
1737
iterator yields is a tuple of a line and a text version that that line
1738
is present in (not introduced in).
1740
Ordering of results is in whatever order is most suitable for the
1741
underlying storage format.
1743
If a progress bar is supplied, it may be used to indicate progress.
1744
The caller is responsible for cleaning up progress bars (because this
1748
* Lines are normalised by the underlying store: they will all have \\n
1750
* Lines are returned in arbitrary order.
1751
* If a requested key did not change any lines (or didn't have any
1752
lines), it may not be mentioned at all in the result.
1754
:param pb: Progress bar supplied by caller.
1755
:return: An iterator over (line, key).
1758
pb = progress.DummyProgress()
1764
# we don't care about inclusions, the caller cares.
1765
# but we need to setup a list of records to visit.
1766
# we need key, position, length
1768
build_details = self._index.get_build_details(keys)
1769
for key, details in build_details.iteritems():
1771
key_records.append((key, details[0]))
1772
records_iter = enumerate(self._read_records_iter(key_records))
1773
for (key_idx, (key, data, sha_value)) in records_iter:
1774
pb.update('Walking content', key_idx, total)
1775
compression_parent = build_details[key][1]
1776
if compression_parent is None:
1778
line_iterator = self._factory.get_fulltext_content(data)
1781
line_iterator = self._factory.get_linedelta_content(data)
1782
# Now that we are yielding the data for this key, remove it
1785
# XXX: It might be more efficient to yield (key,
1786
# line_iterator) in the future. However for now, this is a
1787
# simpler change to integrate into the rest of the
1788
# codebase. RBC 20071110
1789
for line in line_iterator:
1792
except errors.RetryWithNewPacks, e:
1793
self._access.reload_or_raise(e)
1794
# If there are still keys we've not yet found, we look in the fallback
1795
# vfs, and hope to find them there. Note that if the keys are found
1796
# but had no changes or no content, the fallback may not return
1798
if keys and not self._fallback_vfs:
1799
# XXX: strictly the second parameter is meant to be the file id
1800
# but it's not easily accessible here.
1801
raise RevisionNotPresent(keys, repr(self))
1802
for source in self._fallback_vfs:
1806
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1807
source_keys.add(key)
1809
keys.difference_update(source_keys)
1810
pb.update('Walking content', total, total)
1812
def _make_line_delta(self, delta_seq, new_content):
1813
"""Generate a line delta from delta_seq and new_content."""
1815
for op in delta_seq.get_opcodes():
1816
if op[0] == 'equal':
1818
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1821
def _merge_annotations(self, content, parents, parent_texts={},
1822
delta=None, annotated=None,
1823
left_matching_blocks=None):
1824
"""Merge annotations for content and generate deltas.
1826
This is done by comparing the annotations based on changes to the text
1827
and generating a delta on the resulting full texts. If annotations are
1828
not being created then a simple delta is created.
1830
if left_matching_blocks is not None:
1831
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1835
for parent_key in parents:
1836
merge_content = self._get_content(parent_key, parent_texts)
1837
if (parent_key == parents[0] and delta_seq is not None):
1840
seq = patiencediff.PatienceSequenceMatcher(
1841
None, merge_content.text(), content.text())
1842
for i, j, n in seq.get_matching_blocks():
1845
# this copies (origin, text) pairs across to the new
1846
# content for any line that matches the last-checked
1848
content._lines[j:j+n] = merge_content._lines[i:i+n]
1849
# XXX: Robert says the following block is a workaround for a
1850
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1851
if content._lines and content._lines[-1][1][-1] != '\n':
1852
# The copied annotation was from a line without a trailing EOL,
1853
# reinstate one for the content object, to ensure correct
1855
line = content._lines[-1][1] + '\n'
1856
content._lines[-1] = (content._lines[-1][0], line)
1858
if delta_seq is None:
1859
reference_content = self._get_content(parents[0], parent_texts)
1860
new_texts = content.text()
1861
old_texts = reference_content.text()
1862
delta_seq = patiencediff.PatienceSequenceMatcher(
1863
None, old_texts, new_texts)
1864
return self._make_line_delta(delta_seq, content)
1866
def _parse_record(self, version_id, data):
1867
"""Parse an original format knit record.
1869
These have the last element of the key only present in the stored data.
1871
rec, record_contents = self._parse_record_unchecked(data)
1872
self._check_header_version(rec, version_id)
1873
return record_contents, rec[3]
1875
def _parse_record_header(self, key, raw_data):
1876
"""Parse a record header for consistency.
1878
:return: the header and the decompressor stream.
1879
as (stream, header_record)
1881
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1884
rec = self._check_header(key, df.readline())
1885
except Exception, e:
1886
raise KnitCorrupt(self,
1887
"While reading {%s} got %s(%s)"
1888
% (key, e.__class__.__name__, str(e)))
1891
def _parse_record_unchecked(self, data):
1893
# 4168 calls in 2880 217 internal
1894
# 4168 calls to _parse_record_header in 2121
1895
# 4168 calls to readlines in 330
1896
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1898
record_contents = df.readlines()
1899
except Exception, e:
1900
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1901
(data, e.__class__.__name__, str(e)))
1902
header = record_contents.pop(0)
1903
rec = self._split_header(header)
1904
last_line = record_contents.pop()
1905
if len(record_contents) != int(rec[2]):
1906
raise KnitCorrupt(self,
1907
'incorrect number of lines %s != %s'
1908
' for version {%s} %s'
1909
% (len(record_contents), int(rec[2]),
1910
rec[1], record_contents))
1911
if last_line != 'end %s\n' % rec[1]:
1912
raise KnitCorrupt(self,
1913
'unexpected version end line %r, wanted %r'
1914
% (last_line, rec[1]))
1916
return rec, record_contents
1918
def _read_records_iter(self, records):
1919
"""Read text records from data file and yield result.
1921
The result will be returned in whatever is the fastest to read.
1922
Not by the order requested. Also, multiple requests for the same
1923
record will only yield 1 response.
1924
:param records: A list of (key, access_memo) entries
1925
:return: Yields (key, contents, digest) in the order
1926
read, not the order requested
1931
# XXX: This smells wrong, IO may not be getting ordered right.
1932
needed_records = sorted(set(records), key=operator.itemgetter(1))
1933
if not needed_records:
1936
# The transport optimizes the fetching as well
1937
# (ie, reads continuous ranges.)
1938
raw_data = self._access.get_raw_records(
1939
[index_memo for key, index_memo in needed_records])
1941
for (key, index_memo), data in \
1942
izip(iter(needed_records), raw_data):
1943
content, digest = self._parse_record(key[-1], data)
1944
yield key, content, digest
1946
def _read_records_iter_raw(self, records):
1947
"""Read text records from data file and yield raw data.
1949
This unpacks enough of the text record to validate the id is
1950
as expected but thats all.
1952
Each item the iterator yields is (key, bytes,
1953
expected_sha1_of_full_text).
1955
for key, data in self._read_records_iter_unchecked(records):
1956
# validate the header (note that we can only use the suffix in
1957
# current knit records).
1958
df, rec = self._parse_record_header(key, data)
1960
yield key, data, rec[3]
1962
def _read_records_iter_unchecked(self, records):
1963
"""Read text records from data file and yield raw data.
1965
No validation is done.
1967
Yields tuples of (key, data).
1969
# setup an iterator of the external records:
1970
# uses readv so nice and fast we hope.
1972
# grab the disk data needed.
1973
needed_offsets = [index_memo for key, index_memo
1975
raw_records = self._access.get_raw_records(needed_offsets)
1977
for key, index_memo in records:
1978
data = raw_records.next()
1981
def _record_to_data(self, key, digest, lines, dense_lines=None):
1982
"""Convert key, digest, lines into a raw data block.
1984
:param key: The key of the record. Currently keys are always serialised
1985
using just the trailing component.
1986
:param dense_lines: The bytes of lines but in a denser form. For
1987
instance, if lines is a list of 1000 bytestrings each ending in \n,
1988
dense_lines may be a list with one line in it, containing all the
1989
1000's lines and their \n's. Using dense_lines if it is already
1990
known is a win because the string join to create bytes in this
1991
function spends less time resizing the final string.
1992
:return: (len, a StringIO instance with the raw data ready to read.)
1994
chunks = ["version %s %d %s\n" % (key[-1], len(lines), digest)]
1995
chunks.extend(dense_lines or lines)
1996
chunks.append("end %s\n" % key[-1])
1997
for chunk in chunks:
1998
if type(chunk) is not str:
1999
raise AssertionError(
2000
'data must be plain bytes was %s' % type(chunk))
2001
if lines and lines[-1][-1] != '\n':
2002
raise ValueError('corrupt lines value %r' % lines)
2003
compressed_bytes = tuned_gzip.chunks_to_gzip(chunks)
2004
return len(compressed_bytes), compressed_bytes
2006
def _split_header(self, line):
2009
raise KnitCorrupt(self,
2010
'unexpected number of elements in record header')
2014
"""See VersionedFiles.keys."""
2015
if 'evil' in debug.debug_flags:
2016
trace.mutter_callsite(2, "keys scales with size of history")
2017
sources = [self._index] + self._fallback_vfs
2019
for source in sources:
2020
result.update(source.keys())
2024
class _ContentMapGenerator(object):
2025
"""Generate texts or expose raw deltas for a set of texts."""
2027
def __init__(self, ordering='unordered'):
2028
self._ordering = ordering
2030
def _get_content(self, key):
2031
"""Get the content object for key."""
2032
# Note that _get_content is only called when the _ContentMapGenerator
2033
# has been constructed with just one key requested for reconstruction.
2034
if key in self.nonlocal_keys:
2035
record = self.get_record_stream().next()
2036
# Create a content object on the fly
2037
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
2038
return PlainKnitContent(lines, record.key)
2040
# local keys we can ask for directly
2041
return self._get_one_work(key)
2043
def get_record_stream(self):
2044
"""Get a record stream for the keys requested during __init__."""
2045
for record in self._work():
2049
"""Produce maps of text and KnitContents as dicts.
2051
:return: (text_map, content_map) where text_map contains the texts for
2052
the requested versions and content_map contains the KnitContents.
2054
# NB: By definition we never need to read remote sources unless texts
2055
# are requested from them: we don't delta across stores - and we
2056
# explicitly do not want to to prevent data loss situations.
2057
if self.global_map is None:
2058
self.global_map = self.vf.get_parent_map(self.keys)
2059
nonlocal_keys = self.nonlocal_keys
2061
missing_keys = set(nonlocal_keys)
2062
# Read from remote versioned file instances and provide to our caller.
2063
for source in self.vf._fallback_vfs:
2064
if not missing_keys:
2066
# Loop over fallback repositories asking them for texts - ignore
2067
# any missing from a particular fallback.
2068
for record in source.get_record_stream(missing_keys,
2069
self._ordering, True):
2070
if record.storage_kind == 'absent':
2071
# Not in thie particular stream, may be in one of the
2072
# other fallback vfs objects.
2074
missing_keys.remove(record.key)
2077
if self._raw_record_map is None:
2078
raise AssertionError('_raw_record_map should have been filled')
2080
for key in self.keys:
2081
if key in self.nonlocal_keys:
2083
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2086
def _get_one_work(self, requested_key):
2087
# Now, if we have calculated everything already, just return the
2089
if requested_key in self._contents_map:
2090
return self._contents_map[requested_key]
2091
# To simplify things, parse everything at once - code that wants one text
2092
# probably wants them all.
2093
# FUTURE: This function could be improved for the 'extract many' case
2094
# by tracking each component and only doing the copy when the number of
2095
# children than need to apply delta's to it is > 1 or it is part of the
2097
multiple_versions = len(self.keys) != 1
2098
if self._record_map is None:
2099
self._record_map = self.vf._raw_map_to_record_map(
2100
self._raw_record_map)
2101
record_map = self._record_map
2102
# raw_record_map is key:
2103
# Have read and parsed records at this point.
2104
for key in self.keys:
2105
if key in self.nonlocal_keys:
2110
while cursor is not None:
2112
record, record_details, digest, next = record_map[cursor]
2114
raise RevisionNotPresent(cursor, self)
2115
components.append((cursor, record, record_details, digest))
2117
if cursor in self._contents_map:
2118
# no need to plan further back
2119
components.append((cursor, None, None, None))
2123
for (component_id, record, record_details,
2124
digest) in reversed(components):
2125
if component_id in self._contents_map:
2126
content = self._contents_map[component_id]
2128
content, delta = self._factory.parse_record(key[-1],
2129
record, record_details, content,
2130
copy_base_content=multiple_versions)
2131
if multiple_versions:
2132
self._contents_map[component_id] = content
2134
# digest here is the digest from the last applied component.
2135
text = content.text()
2136
actual_sha = sha_strings(text)
2137
if actual_sha != digest:
2138
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2139
if multiple_versions:
2140
return self._contents_map[requested_key]
2144
def _wire_bytes(self):
2145
"""Get the bytes to put on the wire for 'key'.
2147
The first collection of bytes asked for returns the serialised
2148
raw_record_map and the additional details (key, parent) for key.
2149
Subsequent calls return just the additional details (key, parent).
2150
The wire storage_kind given for the first key is 'knit-delta-closure',
2151
For subsequent keys it is 'knit-delta-closure-ref'.
2153
:param key: A key from the content generator.
2154
:return: Bytes to put on the wire.
2157
# kind marker for dispatch on the far side,
2158
lines.append('knit-delta-closure')
2160
if self.vf._factory.annotated:
2161
lines.append('annotated')
2164
# then the list of keys
2165
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2166
if key not in self.nonlocal_keys]))
2167
# then the _raw_record_map in serialised form:
2169
# for each item in the map:
2171
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2172
# one line with method
2173
# one line with noeol
2174
# one line with next ('' for None)
2175
# one line with byte count of the record bytes
2177
for key, (record_bytes, (method, noeol), next) in \
2178
self._raw_record_map.iteritems():
2179
key_bytes = '\x00'.join(key)
2180
parents = self.global_map.get(key, None)
2182
parent_bytes = 'None:'
2184
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2185
method_bytes = method
2191
next_bytes = '\x00'.join(next)
2194
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2195
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2196
len(record_bytes), record_bytes))
2197
map_bytes = ''.join(map_byte_list)
2198
lines.append(map_bytes)
2199
bytes = '\n'.join(lines)
2203
class _VFContentMapGenerator(_ContentMapGenerator):
2204
"""Content map generator reading from a VersionedFiles object."""
2206
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2207
global_map=None, raw_record_map=None, ordering='unordered'):
2208
"""Create a _ContentMapGenerator.
2210
:param versioned_files: The versioned files that the texts are being
2212
:param keys: The keys to produce content maps for.
2213
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2214
which are known to not be in this knit, but rather in one of the
2216
:param global_map: The result of get_parent_map(keys) (or a supermap).
2217
This is required if get_record_stream() is to be used.
2218
:param raw_record_map: A unparsed raw record map to use for answering
2221
_ContentMapGenerator.__init__(self, ordering=ordering)
2222
# The vf to source data from
2223
self.vf = versioned_files
2225
self.keys = list(keys)
2226
# Keys known to be in fallback vfs objects
2227
if nonlocal_keys is None:
2228
self.nonlocal_keys = set()
2230
self.nonlocal_keys = frozenset(nonlocal_keys)
2231
# Parents data for keys to be returned in get_record_stream
2232
self.global_map = global_map
2233
# The chunked lists for self.keys in text form
2235
# A cache of KnitContent objects used in extracting texts.
2236
self._contents_map = {}
2237
# All the knit records needed to assemble the requested keys as full
2239
self._record_map = None
2240
if raw_record_map is None:
2241
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2244
self._raw_record_map = raw_record_map
2245
# the factory for parsing records
2246
self._factory = self.vf._factory
2249
class _NetworkContentMapGenerator(_ContentMapGenerator):
2250
"""Content map generator sourced from a network stream."""
2252
def __init__(self, bytes, line_end):
2253
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2255
self.global_map = {}
2256
self._raw_record_map = {}
2257
self._contents_map = {}
2258
self._record_map = None
2259
self.nonlocal_keys = []
2260
# Get access to record parsing facilities
2261
self.vf = KnitVersionedFiles(None, None)
2264
line_end = bytes.find('\n', start)
2265
line = bytes[start:line_end]
2266
start = line_end + 1
2267
if line == 'annotated':
2268
self._factory = KnitAnnotateFactory()
2270
self._factory = KnitPlainFactory()
2271
# list of keys to emit in get_record_stream
2272
line_end = bytes.find('\n', start)
2273
line = bytes[start:line_end]
2274
start = line_end + 1
2276
tuple(segment.split('\x00')) for segment in line.split('\t')
2278
# now a loop until the end. XXX: It would be nice if this was just a
2279
# bunch of the same records as get_record_stream(..., False) gives, but
2280
# there is a decent sized gap stopping that at the moment.
2284
line_end = bytes.find('\n', start)
2285
key = tuple(bytes[start:line_end].split('\x00'))
2286
start = line_end + 1
2287
# 1 line with parents (None: for None, '' for ())
2288
line_end = bytes.find('\n', start)
2289
line = bytes[start:line_end]
2294
[tuple(segment.split('\x00')) for segment in line.split('\t')
2296
self.global_map[key] = parents
2297
start = line_end + 1
2298
# one line with method
2299
line_end = bytes.find('\n', start)
2300
line = bytes[start:line_end]
2302
start = line_end + 1
2303
# one line with noeol
2304
line_end = bytes.find('\n', start)
2305
line = bytes[start:line_end]
2307
start = line_end + 1
2308
# one line with next ('' for None)
2309
line_end = bytes.find('\n', start)
2310
line = bytes[start:line_end]
2314
next = tuple(bytes[start:line_end].split('\x00'))
2315
start = line_end + 1
2316
# one line with byte count of the record bytes
2317
line_end = bytes.find('\n', start)
2318
line = bytes[start:line_end]
2320
start = line_end + 1
2322
record_bytes = bytes[start:start+count]
2323
start = start + count
2325
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2327
def get_record_stream(self):
2328
"""Get a record stream for for keys requested by the bytestream."""
2330
for key in self.keys:
2331
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2334
def _wire_bytes(self):
2338
class _KndxIndex(object):
2339
"""Manages knit index files
2341
The index is kept in memory and read on startup, to enable
2342
fast lookups of revision information. The cursor of the index
2343
file is always pointing to the end, making it easy to append
2346
_cache is a cache for fast mapping from version id to a Index
2349
_history is a cache for fast mapping from indexes to version ids.
2351
The index data format is dictionary compressed when it comes to
2352
parent references; a index entry may only have parents that with a
2353
lover index number. As a result, the index is topological sorted.
2355
Duplicate entries may be written to the index for a single version id
2356
if this is done then the latter one completely replaces the former:
2357
this allows updates to correct version and parent information.
2358
Note that the two entries may share the delta, and that successive
2359
annotations and references MUST point to the first entry.
2361
The index file on disc contains a header, followed by one line per knit
2362
record. The same revision can be present in an index file more than once.
2363
The first occurrence gets assigned a sequence number starting from 0.
2365
The format of a single line is
2366
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
2367
REVISION_ID is a utf8-encoded revision id
2368
FLAGS is a comma separated list of flags about the record. Values include
2369
no-eol, line-delta, fulltext.
2370
BYTE_OFFSET is the ascii representation of the byte offset in the data file
2371
that the compressed data starts at.
2372
LENGTH is the ascii representation of the length of the data file.
2373
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
2375
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
2376
revision id already in the knit that is a parent of REVISION_ID.
2377
The ' :' marker is the end of record marker.
2380
when a write is interrupted to the index file, it will result in a line
2381
that does not end in ' :'. If the ' :' is not present at the end of a line,
2382
or at the end of the file, then the record that is missing it will be
2383
ignored by the parser.
2385
When writing new records to the index file, the data is preceded by '\n'
2386
to ensure that records always start on new lines even if the last write was
2387
interrupted. As a result its normal for the last line in the index to be
2388
missing a trailing newline. One can be added with no harmful effects.
2390
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2391
where prefix is e.g. the (fileid,) for .texts instances or () for
2392
constant-mapped things like .revisions, and the old state is
2393
tuple(history_vector, cache_dict). This is used to prevent having an
2394
ABI change with the C extension that reads .kndx files.
2397
HEADER = "# bzr knit index 8\n"
2399
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2400
"""Create a _KndxIndex on transport using mapper."""
2401
self._transport = transport
2402
self._mapper = mapper
2403
self._get_scope = get_scope
2404
self._allow_writes = allow_writes
2405
self._is_locked = is_locked
2407
self.has_graph = True
2409
def add_records(self, records, random_id=False, missing_compression_parents=False):
2410
"""Add multiple records to the index.
2412
:param records: a list of tuples:
2413
(key, options, access_memo, parents).
2414
:param random_id: If True the ids being added were randomly generated
2415
and no check for existence will be performed.
2416
:param missing_compression_parents: If True the records being added are
2417
only compressed against texts already in the index (or inside
2418
records). If False the records all refer to unavailable texts (or
2419
texts inside records) as compression parents.
2421
if missing_compression_parents:
2422
# It might be nice to get the edge of the records. But keys isn't
2424
keys = sorted(record[0] for record in records)
2425
raise errors.RevisionNotPresent(keys, self)
2427
for record in records:
2430
path = self._mapper.map(key) + '.kndx'
2431
path_keys = paths.setdefault(path, (prefix, []))
2432
path_keys[1].append(record)
2433
for path in sorted(paths):
2434
prefix, path_keys = paths[path]
2435
self._load_prefixes([prefix])
2437
orig_history = self._kndx_cache[prefix][1][:]
2438
orig_cache = self._kndx_cache[prefix][0].copy()
2441
for key, options, (_, pos, size), parents in path_keys:
2443
# kndx indices cannot be parentless.
2445
line = "\n%s %s %s %s %s :" % (
2446
key[-1], ','.join(options), pos, size,
2447
self._dictionary_compress(parents))
2448
if type(line) is not str:
2449
raise AssertionError(
2450
'data must be utf8 was %s' % type(line))
2452
self._cache_key(key, options, pos, size, parents)
2453
if len(orig_history):
2454
self._transport.append_bytes(path, ''.join(lines))
2456
self._init_index(path, lines)
2458
# If any problems happen, restore the original values and re-raise
2459
self._kndx_cache[prefix] = (orig_cache, orig_history)
2462
def scan_unvalidated_index(self, graph_index):
2463
"""See _KnitGraphIndex.scan_unvalidated_index."""
2464
# Because kndx files do not support atomic insertion via separate index
2465
# files, they do not support this method.
2466
raise NotImplementedError(self.scan_unvalidated_index)
2468
def get_missing_compression_parents(self):
2469
"""See _KnitGraphIndex.get_missing_compression_parents."""
2470
# Because kndx files do not support atomic insertion via separate index
2471
# files, they do not support this method.
2472
raise NotImplementedError(self.get_missing_compression_parents)
2474
def _cache_key(self, key, options, pos, size, parent_keys):
2475
"""Cache a version record in the history array and index cache.
2477
This is inlined into _load_data for performance. KEEP IN SYNC.
2478
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2482
version_id = key[-1]
2483
# last-element only for compatibilty with the C load_data.
2484
parents = tuple(parent[-1] for parent in parent_keys)
2485
for parent in parent_keys:
2486
if parent[:-1] != prefix:
2487
raise ValueError("mismatched prefixes for %r, %r" % (
2489
cache, history = self._kndx_cache[prefix]
2490
# only want the _history index to reference the 1st index entry
2492
if version_id not in cache:
2493
index = len(history)
2494
history.append(version_id)
2496
index = cache[version_id][5]
2497
cache[version_id] = (version_id,
2504
def check_header(self, fp):
2505
line = fp.readline()
2507
# An empty file can actually be treated as though the file doesn't
2509
raise errors.NoSuchFile(self)
2510
if line != self.HEADER:
2511
raise KnitHeaderError(badline=line, filename=self)
2513
def _check_read(self):
2514
if not self._is_locked():
2515
raise errors.ObjectNotLocked(self)
2516
if self._get_scope() != self._scope:
2519
def _check_write_ok(self):
2520
"""Assert if not writes are permitted."""
2521
if not self._is_locked():
2522
raise errors.ObjectNotLocked(self)
2523
if self._get_scope() != self._scope:
2525
if self._mode != 'w':
2526
raise errors.ReadOnlyObjectDirtiedError(self)
2528
def get_build_details(self, keys):
2529
"""Get the method, index_memo and compression parent for keys.
2531
Ghosts are omitted from the result.
2533
:param keys: An iterable of keys.
2534
:return: A dict of key:(index_memo, compression_parent, parents,
2537
opaque structure to pass to read_records to extract the raw
2540
Content that this record is built upon, may be None
2542
Logical parents of this node
2544
extra information about the content which needs to be passed to
2545
Factory.parse_record
2547
parent_map = self.get_parent_map(keys)
2550
if key not in parent_map:
2552
method = self.get_method(key)
2553
parents = parent_map[key]
2554
if method == 'fulltext':
2555
compression_parent = None
2557
compression_parent = parents[0]
2558
noeol = 'no-eol' in self.get_options(key)
2559
index_memo = self.get_position(key)
2560
result[key] = (index_memo, compression_parent,
2561
parents, (method, noeol))
2564
def get_method(self, key):
2565
"""Return compression method of specified key."""
2566
options = self.get_options(key)
2567
if 'fulltext' in options:
2569
elif 'line-delta' in options:
2572
raise errors.KnitIndexUnknownMethod(self, options)
2574
def get_options(self, key):
2575
"""Return a list representing options.
2579
prefix, suffix = self._split_key(key)
2580
self._load_prefixes([prefix])
2582
return self._kndx_cache[prefix][0][suffix][1]
2584
raise RevisionNotPresent(key, self)
2586
def find_ancestry(self, keys):
2587
"""See CombinedGraphIndex.find_ancestry()"""
2588
prefixes = set(key[:-1] for key in keys)
2589
self._load_prefixes(prefixes)
2592
missing_keys = set()
2593
pending_keys = list(keys)
2594
# This assumes that keys will not reference parents in a different
2595
# prefix, which is accurate so far.
2597
key = pending_keys.pop()
2598
if key in parent_map:
2602
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2604
missing_keys.add(key)
2606
parent_keys = tuple([prefix + (suffix,)
2607
for suffix in suffix_parents])
2608
parent_map[key] = parent_keys
2609
pending_keys.extend([p for p in parent_keys
2610
if p not in parent_map])
2611
return parent_map, missing_keys
2613
def get_parent_map(self, keys):
2614
"""Get a map of the parents of keys.
2616
:param keys: The keys to look up parents for.
2617
:return: A mapping from keys to parents. Absent keys are absent from
2620
# Parse what we need to up front, this potentially trades off I/O
2621
# locality (.kndx and .knit in the same block group for the same file
2622
# id) for less checking in inner loops.
2623
prefixes = set(key[:-1] for key in keys)
2624
self._load_prefixes(prefixes)
2629
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2633
result[key] = tuple(prefix + (suffix,) for
2634
suffix in suffix_parents)
2637
def get_position(self, key):
2638
"""Return details needed to access the version.
2640
:return: a tuple (key, data position, size) to hand to the access
2641
logic to get the record.
2643
prefix, suffix = self._split_key(key)
2644
self._load_prefixes([prefix])
2645
entry = self._kndx_cache[prefix][0][suffix]
2646
return key, entry[2], entry[3]
2648
has_key = _mod_index._has_key_from_parent_map
2650
def _init_index(self, path, extra_lines=[]):
2651
"""Initialize an index."""
2653
sio.write(self.HEADER)
2654
sio.writelines(extra_lines)
2656
self._transport.put_file_non_atomic(path, sio,
2657
create_parent_dir=True)
2658
# self._create_parent_dir)
2659
# mode=self._file_mode,
2660
# dir_mode=self._dir_mode)
2663
"""Get all the keys in the collection.
2665
The keys are not ordered.
2668
# Identify all key prefixes.
2669
# XXX: A bit hacky, needs polish.
2670
if type(self._mapper) is ConstantMapper:
2674
for quoted_relpath in self._transport.iter_files_recursive():
2675
path, ext = os.path.splitext(quoted_relpath)
2677
prefixes = [self._mapper.unmap(path) for path in relpaths]
2678
self._load_prefixes(prefixes)
2679
for prefix in prefixes:
2680
for suffix in self._kndx_cache[prefix][1]:
2681
result.add(prefix + (suffix,))
2684
def _load_prefixes(self, prefixes):
2685
"""Load the indices for prefixes."""
2687
for prefix in prefixes:
2688
if prefix not in self._kndx_cache:
2689
# the load_data interface writes to these variables.
2692
self._filename = prefix
2694
path = self._mapper.map(prefix) + '.kndx'
2695
fp = self._transport.get(path)
2697
# _load_data may raise NoSuchFile if the target knit is
2699
_load_data(self, fp)
2702
self._kndx_cache[prefix] = (self._cache, self._history)
2707
self._kndx_cache[prefix] = ({}, [])
2708
if type(self._mapper) is ConstantMapper:
2709
# preserve behaviour for revisions.kndx etc.
2710
self._init_index(path)
2715
missing_keys = _mod_index._missing_keys_from_parent_map
2717
def _partition_keys(self, keys):
2718
"""Turn keys into a dict of prefix:suffix_list."""
2721
prefix_keys = result.setdefault(key[:-1], [])
2722
prefix_keys.append(key[-1])
2725
def _dictionary_compress(self, keys):
2726
"""Dictionary compress keys.
2728
:param keys: The keys to generate references to.
2729
:return: A string representation of keys. keys which are present are
2730
dictionary compressed, and others are emitted as fulltext with a
2736
prefix = keys[0][:-1]
2737
cache = self._kndx_cache[prefix][0]
2739
if key[:-1] != prefix:
2740
# kndx indices cannot refer across partitioned storage.
2741
raise ValueError("mismatched prefixes for %r" % keys)
2742
if key[-1] in cache:
2743
# -- inlined lookup() --
2744
result_list.append(str(cache[key[-1]][5]))
2745
# -- end lookup () --
2747
result_list.append('.' + key[-1])
2748
return ' '.join(result_list)
2750
def _reset_cache(self):
2751
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2752
# (cache_dict, history_vector) for parsed kndx files.
2753
self._kndx_cache = {}
2754
self._scope = self._get_scope()
2755
allow_writes = self._allow_writes()
2761
def _sort_keys_by_io(self, keys, positions):
2762
"""Figure out an optimal order to read the records for the given keys.
2764
Sort keys, grouped by index and sorted by position.
2766
:param keys: A list of keys whose records we want to read. This will be
2768
:param positions: A dict, such as the one returned by
2769
_get_components_positions()
2772
def get_sort_key(key):
2773
index_memo = positions[key][1]
2774
# Group by prefix and position. index_memo[0] is the key, so it is
2775
# (file_id, revision_id) and we don't want to sort on revision_id,
2776
# index_memo[1] is the position, and index_memo[2] is the size,
2777
# which doesn't matter for the sort
2778
return index_memo[0][:-1], index_memo[1]
2779
return keys.sort(key=get_sort_key)
2781
_get_total_build_size = _get_total_build_size
2783
def _split_key(self, key):
2784
"""Split key into a prefix and suffix."""
2785
return key[:-1], key[-1]
2788
class _KeyRefs(object):
2790
def __init__(self, track_new_keys=False):
2791
# dict mapping 'key' to 'set of keys referring to that key'
2794
# set remembering all new keys
2795
self.new_keys = set()
2797
self.new_keys = None
2803
self.new_keys.clear()
2805
def add_references(self, key, refs):
2806
# Record the new references
2807
for referenced in refs:
2809
needed_by = self.refs[referenced]
2811
needed_by = self.refs[referenced] = set()
2813
# Discard references satisfied by the new key
2816
def get_new_keys(self):
2817
return self.new_keys
2819
def get_unsatisfied_refs(self):
2820
return self.refs.iterkeys()
2822
def _satisfy_refs_for_key(self, key):
2826
# No keys depended on this key. That's ok.
2829
def add_key(self, key):
2830
# satisfy refs for key, and remember that we've seen this key.
2831
self._satisfy_refs_for_key(key)
2832
if self.new_keys is not None:
2833
self.new_keys.add(key)
2835
def satisfy_refs_for_keys(self, keys):
2837
self._satisfy_refs_for_key(key)
2839
def get_referrers(self):
2841
for referrers in self.refs.itervalues():
2842
result.update(referrers)
2846
class _KnitGraphIndex(object):
2847
"""A KnitVersionedFiles index layered on GraphIndex."""
2849
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2850
add_callback=None, track_external_parent_refs=False):
2851
"""Construct a KnitGraphIndex on a graph_index.
2853
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2854
:param is_locked: A callback to check whether the object should answer
2856
:param deltas: Allow delta-compressed records.
2857
:param parents: If True, record knits parents, if not do not record
2859
:param add_callback: If not None, allow additions to the index and call
2860
this callback with a list of added GraphIndex nodes:
2861
[(node, value, node_refs), ...]
2862
:param is_locked: A callback, returns True if the index is locked and
2864
:param track_external_parent_refs: If True, record all external parent
2865
references parents from added records. These can be retrieved
2866
later by calling get_missing_parents().
2868
self._add_callback = add_callback
2869
self._graph_index = graph_index
2870
self._deltas = deltas
2871
self._parents = parents
2872
if deltas and not parents:
2873
# XXX: TODO: Delta tree and parent graph should be conceptually
2875
raise KnitCorrupt(self, "Cannot do delta compression without "
2877
self.has_graph = parents
2878
self._is_locked = is_locked
2879
self._missing_compression_parents = set()
2880
if track_external_parent_refs:
2881
self._key_dependencies = _KeyRefs()
2883
self._key_dependencies = None
2886
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2888
def add_records(self, records, random_id=False,
2889
missing_compression_parents=False):
2890
"""Add multiple records to the index.
2892
This function does not insert data into the Immutable GraphIndex
2893
backing the KnitGraphIndex, instead it prepares data for insertion by
2894
the caller and checks that it is safe to insert then calls
2895
self._add_callback with the prepared GraphIndex nodes.
2897
:param records: a list of tuples:
2898
(key, options, access_memo, parents).
2899
:param random_id: If True the ids being added were randomly generated
2900
and no check for existence will be performed.
2901
:param missing_compression_parents: If True the records being added are
2902
only compressed against texts already in the index (or inside
2903
records). If False the records all refer to unavailable texts (or
2904
texts inside records) as compression parents.
2906
if not self._add_callback:
2907
raise errors.ReadOnlyError(self)
2908
# we hope there are no repositories with inconsistent parentage
2912
compression_parents = set()
2913
key_dependencies = self._key_dependencies
2914
for (key, options, access_memo, parents) in records:
2916
parents = tuple(parents)
2917
if key_dependencies is not None:
2918
key_dependencies.add_references(key, parents)
2919
index, pos, size = access_memo
2920
if 'no-eol' in options:
2924
value += "%d %d" % (pos, size)
2925
if not self._deltas:
2926
if 'line-delta' in options:
2927
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2930
if 'line-delta' in options:
2931
node_refs = (parents, (parents[0],))
2932
if missing_compression_parents:
2933
compression_parents.add(parents[0])
2935
node_refs = (parents, ())
2937
node_refs = (parents, )
2940
raise KnitCorrupt(self, "attempt to add node with parents "
2941
"in parentless index.")
2943
keys[key] = (value, node_refs)
2946
present_nodes = self._get_entries(keys)
2947
for (index, key, value, node_refs) in present_nodes:
2948
parents = node_refs[:1]
2949
# Sometimes these are passed as a list rather than a tuple
2950
passed = static_tuple.as_tuples(keys[key])
2951
passed_parents = passed[1][:1]
2952
if (value[0] != keys[key][0][0] or
2953
parents != passed_parents):
2954
node_refs = static_tuple.as_tuples(node_refs)
2955
raise KnitCorrupt(self, "inconsistent details in add_records"
2956
": %s %s" % ((value, node_refs), passed))
2960
for key, (value, node_refs) in keys.iteritems():
2961
result.append((key, value, node_refs))
2963
for key, (value, node_refs) in keys.iteritems():
2964
result.append((key, value))
2965
self._add_callback(result)
2966
if missing_compression_parents:
2967
# This may appear to be incorrect (it does not check for
2968
# compression parents that are in the existing graph index),
2969
# but such records won't have been buffered, so this is
2970
# actually correct: every entry when
2971
# missing_compression_parents==True either has a missing parent, or
2972
# a parent that is one of the keys in records.
2973
compression_parents.difference_update(keys)
2974
self._missing_compression_parents.update(compression_parents)
2975
# Adding records may have satisfied missing compression parents.
2976
self._missing_compression_parents.difference_update(keys)
2978
def scan_unvalidated_index(self, graph_index):
2979
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2981
This allows this _KnitGraphIndex to keep track of any missing
2982
compression parents we may want to have filled in to make those
2985
:param graph_index: A GraphIndex
2988
new_missing = graph_index.external_references(ref_list_num=1)
2989
new_missing.difference_update(self.get_parent_map(new_missing))
2990
self._missing_compression_parents.update(new_missing)
2991
if self._key_dependencies is not None:
2992
# Add parent refs from graph_index (and discard parent refs that
2993
# the graph_index has).
2994
for node in graph_index.iter_all_entries():
2995
self._key_dependencies.add_references(node[1], node[3][0])
2997
def get_missing_compression_parents(self):
2998
"""Return the keys of missing compression parents.
3000
Missing compression parents occur when a record stream was missing
3001
basis texts, or a index was scanned that had missing basis texts.
3003
return frozenset(self._missing_compression_parents)
3005
def get_missing_parents(self):
3006
"""Return the keys of missing parents."""
3007
# If updating this, you should also update
3008
# groupcompress._GCGraphIndex.get_missing_parents
3009
# We may have false positives, so filter those out.
3010
self._key_dependencies.satisfy_refs_for_keys(
3011
self.get_parent_map(self._key_dependencies.get_unsatisfied_refs()))
3012
return frozenset(self._key_dependencies.get_unsatisfied_refs())
3014
def _check_read(self):
3015
"""raise if reads are not permitted."""
3016
if not self._is_locked():
3017
raise errors.ObjectNotLocked(self)
3019
def _check_write_ok(self):
3020
"""Assert if writes are not permitted."""
3021
if not self._is_locked():
3022
raise errors.ObjectNotLocked(self)
3024
def _compression_parent(self, an_entry):
3025
# return the key that an_entry is compressed against, or None
3026
# Grab the second parent list (as deltas implies parents currently)
3027
compression_parents = an_entry[3][1]
3028
if not compression_parents:
3030
if len(compression_parents) != 1:
3031
raise AssertionError(
3032
"Too many compression parents: %r" % compression_parents)
3033
return compression_parents[0]
3035
def get_build_details(self, keys):
3036
"""Get the method, index_memo and compression parent for version_ids.
3038
Ghosts are omitted from the result.
3040
:param keys: An iterable of keys.
3041
:return: A dict of key:
3042
(index_memo, compression_parent, parents, record_details).
3044
opaque structure to pass to read_records to extract the raw
3047
Content that this record is built upon, may be None
3049
Logical parents of this node
3051
extra information about the content which needs to be passed to
3052
Factory.parse_record
3056
entries = self._get_entries(keys, False)
3057
for entry in entries:
3059
if not self._parents:
3062
parents = entry[3][0]
3063
if not self._deltas:
3064
compression_parent_key = None
3066
compression_parent_key = self._compression_parent(entry)
3067
noeol = (entry[2][0] == 'N')
3068
if compression_parent_key:
3069
method = 'line-delta'
3072
result[key] = (self._node_to_position(entry),
3073
compression_parent_key, parents,
3077
def _get_entries(self, keys, check_present=False):
3078
"""Get the entries for keys.
3080
:param keys: An iterable of index key tuples.
3085
for node in self._graph_index.iter_entries(keys):
3087
found_keys.add(node[1])
3089
# adapt parentless index to the rest of the code.
3090
for node in self._graph_index.iter_entries(keys):
3091
yield node[0], node[1], node[2], ()
3092
found_keys.add(node[1])
3094
missing_keys = keys.difference(found_keys)
3096
raise RevisionNotPresent(missing_keys.pop(), self)
3098
def get_method(self, key):
3099
"""Return compression method of specified key."""
3100
return self._get_method(self._get_node(key))
3102
def _get_method(self, node):
3103
if not self._deltas:
3105
if self._compression_parent(node):
3110
def _get_node(self, key):
3112
return list(self._get_entries([key]))[0]
3114
raise RevisionNotPresent(key, self)
3116
def get_options(self, key):
3117
"""Return a list representing options.
3121
node = self._get_node(key)
3122
options = [self._get_method(node)]
3123
if node[2][0] == 'N':
3124
options.append('no-eol')
3127
def find_ancestry(self, keys):
3128
"""See CombinedGraphIndex.find_ancestry()"""
3129
return self._graph_index.find_ancestry(keys, 0)
3131
def get_parent_map(self, keys):
3132
"""Get a map of the parents of keys.
3134
:param keys: The keys to look up parents for.
3135
:return: A mapping from keys to parents. Absent keys are absent from
3139
nodes = self._get_entries(keys)
3143
result[node[1]] = node[3][0]
3146
result[node[1]] = None
3149
def get_position(self, key):
3150
"""Return details needed to access the version.
3152
:return: a tuple (index, data position, size) to hand to the access
3153
logic to get the record.
3155
node = self._get_node(key)
3156
return self._node_to_position(node)
3158
has_key = _mod_index._has_key_from_parent_map
3161
"""Get all the keys in the collection.
3163
The keys are not ordered.
3166
return [node[1] for node in self._graph_index.iter_all_entries()]
3168
missing_keys = _mod_index._missing_keys_from_parent_map
3170
def _node_to_position(self, node):
3171
"""Convert an index value to position details."""
3172
bits = node[2][1:].split(' ')
3173
return node[0], int(bits[0]), int(bits[1])
3175
def _sort_keys_by_io(self, keys, positions):
3176
"""Figure out an optimal order to read the records for the given keys.
3178
Sort keys, grouped by index and sorted by position.
3180
:param keys: A list of keys whose records we want to read. This will be
3182
:param positions: A dict, such as the one returned by
3183
_get_components_positions()
3186
def get_index_memo(key):
3187
# index_memo is at offset [1]. It is made up of (GraphIndex,
3188
# position, size). GI is an object, which will be unique for each
3189
# pack file. This causes us to group by pack file, then sort by
3190
# position. Size doesn't matter, but it isn't worth breaking up the
3192
return positions[key][1]
3193
return keys.sort(key=get_index_memo)
3195
_get_total_build_size = _get_total_build_size
3198
class _KnitKeyAccess(object):
3199
"""Access to records in .knit files."""
3201
def __init__(self, transport, mapper):
3202
"""Create a _KnitKeyAccess with transport and mapper.
3204
:param transport: The transport the access object is rooted at.
3205
:param mapper: The mapper used to map keys to .knit files.
3207
self._transport = transport
3208
self._mapper = mapper
3210
def add_raw_records(self, key_sizes, raw_data):
3211
"""Add raw knit bytes to a storage area.
3213
The data is spooled to the container writer in one bytes-record per
3216
:param sizes: An iterable of tuples containing the key and size of each
3218
:param raw_data: A bytestring containing the data.
3219
:return: A list of memos to retrieve the record later. Each memo is an
3220
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3221
length), where the key is the record key.
3223
if type(raw_data) is not str:
3224
raise AssertionError(
3225
'data must be plain bytes was %s' % type(raw_data))
3228
# TODO: This can be tuned for writing to sftp and other servers where
3229
# append() is relatively expensive by grouping the writes to each key
3231
for key, size in key_sizes:
3232
path = self._mapper.map(key)
3234
base = self._transport.append_bytes(path + '.knit',
3235
raw_data[offset:offset+size])
3236
except errors.NoSuchFile:
3237
self._transport.mkdir(osutils.dirname(path))
3238
base = self._transport.append_bytes(path + '.knit',
3239
raw_data[offset:offset+size])
3243
result.append((key, base, size))
3247
"""Flush pending writes on this access object.
3249
For .knit files this is a no-op.
3253
def get_raw_records(self, memos_for_retrieval):
3254
"""Get the raw bytes for a records.
3256
:param memos_for_retrieval: An iterable containing the access memo for
3257
retrieving the bytes.
3258
:return: An iterator over the bytes of the records.
3260
# first pass, group into same-index request to minimise readv's issued.
3262
current_prefix = None
3263
for (key, offset, length) in memos_for_retrieval:
3264
if current_prefix == key[:-1]:
3265
current_list.append((offset, length))
3267
if current_prefix is not None:
3268
request_lists.append((current_prefix, current_list))
3269
current_prefix = key[:-1]
3270
current_list = [(offset, length)]
3271
# handle the last entry
3272
if current_prefix is not None:
3273
request_lists.append((current_prefix, current_list))
3274
for prefix, read_vector in request_lists:
3275
path = self._mapper.map(prefix) + '.knit'
3276
for pos, data in self._transport.readv(path, read_vector):
3280
class _DirectPackAccess(object):
3281
"""Access to data in one or more packs with less translation."""
3283
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
3284
"""Create a _DirectPackAccess object.
3286
:param index_to_packs: A dict mapping index objects to the transport
3287
and file names for obtaining data.
3288
:param reload_func: A function to call if we determine that the pack
3289
files have moved and we need to reload our caches. See
3290
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
3292
self._container_writer = None
3293
self._write_index = None
3294
self._indices = index_to_packs
3295
self._reload_func = reload_func
3296
self._flush_func = flush_func
3298
def add_raw_records(self, key_sizes, raw_data):
3299
"""Add raw knit bytes to a storage area.
3301
The data is spooled to the container writer in one bytes-record per
3304
:param sizes: An iterable of tuples containing the key and size of each
3306
:param raw_data: A bytestring containing the data.
3307
:return: A list of memos to retrieve the record later. Each memo is an
3308
opaque index memo. For _DirectPackAccess the memo is (index, pos,
3309
length), where the index field is the write_index object supplied
3310
to the PackAccess object.
3312
if type(raw_data) is not str:
3313
raise AssertionError(
3314
'data must be plain bytes was %s' % type(raw_data))
3317
for key, size in key_sizes:
3318
p_offset, p_length = self._container_writer.add_bytes_record(
3319
raw_data[offset:offset+size], [])
3321
result.append((self._write_index, p_offset, p_length))
3325
"""Flush pending writes on this access object.
3327
This will flush any buffered writes to a NewPack.
3329
if self._flush_func is not None:
3332
def get_raw_records(self, memos_for_retrieval):
3333
"""Get the raw bytes for a records.
3335
:param memos_for_retrieval: An iterable containing the (index, pos,
3336
length) memo for retrieving the bytes. The Pack access method
3337
looks up the pack to use for a given record in its index_to_pack
3339
:return: An iterator over the bytes of the records.
3341
# first pass, group into same-index requests
3343
current_index = None
3344
for (index, offset, length) in memos_for_retrieval:
3345
if current_index == index:
3346
current_list.append((offset, length))
3348
if current_index is not None:
3349
request_lists.append((current_index, current_list))
3350
current_index = index
3351
current_list = [(offset, length)]
3352
# handle the last entry
3353
if current_index is not None:
3354
request_lists.append((current_index, current_list))
3355
for index, offsets in request_lists:
3357
transport, path = self._indices[index]
3359
# A KeyError here indicates that someone has triggered an index
3360
# reload, and this index has gone missing, we need to start
3362
if self._reload_func is None:
3363
# If we don't have a _reload_func there is nothing that can
3366
raise errors.RetryWithNewPacks(index,
3367
reload_occurred=True,
3368
exc_info=sys.exc_info())
3370
reader = pack.make_readv_reader(transport, path, offsets)
3371
for names, read_func in reader.iter_records():
3372
yield read_func(None)
3373
except errors.NoSuchFile:
3374
# A NoSuchFile error indicates that a pack file has gone
3375
# missing on disk, we need to trigger a reload, and start over.
3376
if self._reload_func is None:
3378
raise errors.RetryWithNewPacks(transport.abspath(path),
3379
reload_occurred=False,
3380
exc_info=sys.exc_info())
3382
def set_writer(self, writer, index, transport_packname):
3383
"""Set a writer to use for adding data."""
3384
if index is not None:
3385
self._indices[index] = transport_packname
3386
self._container_writer = writer
3387
self._write_index = index
3389
def reload_or_raise(self, retry_exc):
3390
"""Try calling the reload function, or re-raise the original exception.
3392
This should be called after _DirectPackAccess raises a
3393
RetryWithNewPacks exception. This function will handle the common logic
3394
of determining when the error is fatal versus being temporary.
3395
It will also make sure that the original exception is raised, rather
3396
than the RetryWithNewPacks exception.
3398
If this function returns, then the calling function should retry
3399
whatever operation was being performed. Otherwise an exception will
3402
:param retry_exc: A RetryWithNewPacks exception.
3405
if self._reload_func is None:
3407
elif not self._reload_func():
3408
# The reload claimed that nothing changed
3409
if not retry_exc.reload_occurred:
3410
# If there wasn't an earlier reload, then we really were
3411
# expecting to find changes. We didn't find them, so this is a
3415
exc_class, exc_value, exc_traceback = retry_exc.exc_info
3416
raise exc_class, exc_value, exc_traceback
3419
# Deprecated, use PatienceSequenceMatcher instead
3420
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
3423
def annotate_knit(knit, revision_id):
3424
"""Annotate a knit with no cached annotations.
3426
This implementation is for knits with no cached annotations.
3427
It will work for knits with cached annotations, but this is not
3430
annotator = _KnitAnnotator(knit)
3431
return iter(annotator.annotate_flat(revision_id))
3434
class _KnitAnnotator(annotate.Annotator):
3435
"""Build up the annotations for a text."""
3437
def __init__(self, vf):
3438
annotate.Annotator.__init__(self, vf)
3440
# TODO: handle Nodes which cannot be extracted
3441
# self._ghosts = set()
3443
# Map from (key, parent_key) => matching_blocks, should be 'use once'
3444
self._matching_blocks = {}
3446
# KnitContent objects
3447
self._content_objects = {}
3448
# The number of children that depend on this fulltext content object
3449
self._num_compression_children = {}
3450
# Delta records that need their compression parent before they can be
3452
self._pending_deltas = {}
3453
# Fulltext records that are waiting for their parents fulltexts before
3454
# they can be yielded for annotation
3455
self._pending_annotation = {}
3457
self._all_build_details = {}
3459
def _get_build_graph(self, key):
3460
"""Get the graphs for building texts and annotations.
3462
The data you need for creating a full text may be different than the
3463
data you need to annotate that text. (At a minimum, you need both
3464
parents to create an annotation, but only need 1 parent to generate the
3467
:return: A list of (key, index_memo) records, suitable for
3468
passing to read_records_iter to start reading in the raw data from
3471
pending = set([key])
3474
self._num_needed_children[key] = 1
3476
# get all pending nodes
3477
this_iteration = pending
3478
build_details = self._vf._index.get_build_details(this_iteration)
3479
self._all_build_details.update(build_details)
3480
# new_nodes = self._vf._index._get_entries(this_iteration)
3482
for key, details in build_details.iteritems():
3483
(index_memo, compression_parent, parent_keys,
3484
record_details) = details
3485
self._parent_map[key] = parent_keys
3486
self._heads_provider = None
3487
records.append((key, index_memo))
3488
# Do we actually need to check _annotated_lines?
3489
pending.update([p for p in parent_keys
3490
if p not in self._all_build_details])
3492
for parent_key in parent_keys:
3493
if parent_key in self._num_needed_children:
3494
self._num_needed_children[parent_key] += 1
3496
self._num_needed_children[parent_key] = 1
3497
if compression_parent:
3498
if compression_parent in self._num_compression_children:
3499
self._num_compression_children[compression_parent] += 1
3501
self._num_compression_children[compression_parent] = 1
3503
missing_versions = this_iteration.difference(build_details.keys())
3504
if missing_versions:
3505
for key in missing_versions:
3506
if key in self._parent_map and key in self._text_cache:
3507
# We already have this text ready, we just need to
3508
# yield it later so we get it annotated
3510
parent_keys = self._parent_map[key]
3511
for parent_key in parent_keys:
3512
if parent_key in self._num_needed_children:
3513
self._num_needed_children[parent_key] += 1
3515
self._num_needed_children[parent_key] = 1
3516
pending.update([p for p in parent_keys
3517
if p not in self._all_build_details])
3519
raise errors.RevisionNotPresent(key, self._vf)
3520
# Generally we will want to read the records in reverse order, because
3521
# we find the parent nodes after the children
3523
return records, ann_keys
3525
def _get_needed_texts(self, key, pb=None):
3526
# if True or len(self._vf._fallback_vfs) > 0:
3527
if len(self._vf._fallback_vfs) > 0:
3528
# If we have fallbacks, go to the generic path
3529
for v in annotate.Annotator._get_needed_texts(self, key, pb=pb):
3534
records, ann_keys = self._get_build_graph(key)
3535
for idx, (sub_key, text, num_lines) in enumerate(
3536
self._extract_texts(records)):
3538
pb.update('annotating', idx, len(records))
3539
yield sub_key, text, num_lines
3540
for sub_key in ann_keys:
3541
text = self._text_cache[sub_key]
3542
num_lines = len(text) # bad assumption
3543
yield sub_key, text, num_lines
3545
except errors.RetryWithNewPacks, e:
3546
self._vf._access.reload_or_raise(e)
3547
# The cached build_details are no longer valid
3548
self._all_build_details.clear()
3550
def _cache_delta_blocks(self, key, compression_parent, delta, lines):
3551
parent_lines = self._text_cache[compression_parent]
3552
blocks = list(KnitContent.get_line_delta_blocks(delta, parent_lines, lines))
3553
self._matching_blocks[(key, compression_parent)] = blocks
3555
def _expand_record(self, key, parent_keys, compression_parent, record,
3558
if compression_parent:
3559
if compression_parent not in self._content_objects:
3560
# Waiting for the parent
3561
self._pending_deltas.setdefault(compression_parent, []).append(
3562
(key, parent_keys, record, record_details))
3564
# We have the basis parent, so expand the delta
3565
num = self._num_compression_children[compression_parent]
3568
base_content = self._content_objects.pop(compression_parent)
3569
self._num_compression_children.pop(compression_parent)
3571
self._num_compression_children[compression_parent] = num
3572
base_content = self._content_objects[compression_parent]
3573
# It is tempting to want to copy_base_content=False for the last
3574
# child object. However, whenever noeol=False,
3575
# self._text_cache[parent_key] is content._lines. So mutating it
3576
# gives very bad results.
3577
# The alternative is to copy the lines into text cache, but then we
3578
# are copying anyway, so just do it here.
3579
content, delta = self._vf._factory.parse_record(
3580
key, record, record_details, base_content,
3581
copy_base_content=True)
3584
content, _ = self._vf._factory.parse_record(
3585
key, record, record_details, None)
3586
if self._num_compression_children.get(key, 0) > 0:
3587
self._content_objects[key] = content
3588
lines = content.text()
3589
self._text_cache[key] = lines
3590
if delta is not None:
3591
self._cache_delta_blocks(key, compression_parent, delta, lines)
3594
def _get_parent_annotations_and_matches(self, key, text, parent_key):
3595
"""Get the list of annotations for the parent, and the matching lines.
3597
:param text: The opaque value given by _get_needed_texts
3598
:param parent_key: The key for the parent text
3599
:return: (parent_annotations, matching_blocks)
3600
parent_annotations is a list as long as the number of lines in
3602
matching_blocks is a list of (parent_idx, text_idx, len) tuples
3603
indicating which lines match between the two texts
3605
block_key = (key, parent_key)
3606
if block_key in self._matching_blocks:
3607
blocks = self._matching_blocks.pop(block_key)
3608
parent_annotations = self._annotations_cache[parent_key]
3609
return parent_annotations, blocks
3610
return annotate.Annotator._get_parent_annotations_and_matches(self,
3611
key, text, parent_key)
3613
def _process_pending(self, key):
3614
"""The content for 'key' was just processed.
3616
Determine if there is any more pending work to be processed.
3619
if key in self._pending_deltas:
3620
compression_parent = key
3621
children = self._pending_deltas.pop(key)
3622
for child_key, parent_keys, record, record_details in children:
3623
lines = self._expand_record(child_key, parent_keys,
3625
record, record_details)
3626
if self._check_ready_for_annotations(child_key, parent_keys):
3627
to_return.append(child_key)
3628
# Also check any children that are waiting for this parent to be
3630
if key in self._pending_annotation:
3631
children = self._pending_annotation.pop(key)
3632
to_return.extend([c for c, p_keys in children
3633
if self._check_ready_for_annotations(c, p_keys)])
3636
def _check_ready_for_annotations(self, key, parent_keys):
3637
"""return true if this text is ready to be yielded.
3639
Otherwise, this will return False, and queue the text into
3640
self._pending_annotation
3642
for parent_key in parent_keys:
3643
if parent_key not in self._annotations_cache:
3644
# still waiting on at least one parent text, so queue it up
3645
# Note that if there are multiple parents, we need to wait
3647
self._pending_annotation.setdefault(parent_key,
3648
[]).append((key, parent_keys))
3652
def _extract_texts(self, records):
3653
"""Extract the various texts needed based on records"""
3654
# We iterate in the order read, rather than a strict order requested
3655
# However, process what we can, and put off to the side things that
3656
# still need parents, cleaning them up when those parents are
3659
# 1) As 'records' are read, see if we can expand these records into
3660
# Content objects (and thus lines)
3661
# 2) If a given line-delta is waiting on its compression parent, it
3662
# gets queued up into self._pending_deltas, otherwise we expand
3663
# it, and put it into self._text_cache and self._content_objects
3664
# 3) If we expanded the text, we will then check to see if all
3665
# parents have also been processed. If so, this text gets yielded,
3666
# else this record gets set aside into pending_annotation
3667
# 4) Further, if we expanded the text in (2), we will then check to
3668
# see if there are any children in self._pending_deltas waiting to
3669
# also be processed. If so, we go back to (2) for those
3670
# 5) Further again, if we yielded the text, we can then check if that
3671
# 'unlocks' any of the texts in pending_annotations, which should
3672
# then get yielded as well
3673
# Note that both steps 4 and 5 are 'recursive' in that unlocking one
3674
# compression child could unlock yet another, and yielding a fulltext
3675
# will also 'unlock' the children that are waiting on that annotation.
3676
# (Though also, unlocking 1 parent's fulltext, does not unlock a child
3677
# if other parents are also waiting.)
3678
# We want to yield content before expanding child content objects, so
3679
# that we know when we can re-use the content lines, and the annotation
3680
# code can know when it can stop caching fulltexts, as well.
3682
# Children that are missing their compression parent
3684
for (key, record, digest) in self._vf._read_records_iter(records):
3686
details = self._all_build_details[key]
3687
(_, compression_parent, parent_keys, record_details) = details
3688
lines = self._expand_record(key, parent_keys, compression_parent,
3689
record, record_details)
3691
# Pending delta should be queued up
3693
# At this point, we may be able to yield this content, if all
3694
# parents are also finished
3695
yield_this_text = self._check_ready_for_annotations(key,
3698
# All parents present
3699
yield key, lines, len(lines)
3700
to_process = self._process_pending(key)
3702
this_process = to_process
3704
for key in this_process:
3705
lines = self._text_cache[key]
3706
yield key, lines, len(lines)
3707
to_process.extend(self._process_pending(key))
3710
from bzrlib._knit_load_data_pyx import _load_data_c as _load_data
3711
except ImportError, e:
3712
osutils.failed_to_load_extension(e)
3713
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
