1
# Copyright (C) 2005, 2006, 2007, 2008 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
"""Knit versionedfile implementation.
19
A knit is a versioned file implementation that supports efficient append only
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
end-marker; simply "end VERSION"
28
delta can be line or full contents.a
29
... the 8's there are the index number of the annotation.
30
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
8 e.set('executable', 'yes')
36
8 if elt.get('executable') == 'yes':
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
43
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
44
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
45
09:33 < lifeless> right
46
09:33 < jrydberg> lifeless: the position and size is the range in the data file
49
so the index sequence is the dictionary compressed sequence number used
50
in the deltas to provide line annotation
55
from cStringIO import StringIO
56
from itertools import izip, chain
61
from bzrlib.lazy_import import lazy_import
62
lazy_import(globals(), """
82
from bzrlib.errors import (
90
RevisionAlreadyPresent,
93
from bzrlib.osutils import (
100
from bzrlib.versionedfile import (
101
AbsentContentFactory,
105
ChunkedContentFactory,
112
# TODO: Split out code specific to this format into an associated object.
114
# TODO: Can we put in some kind of value to check that the index and data
115
# files belong together?
117
# TODO: accommodate binaries, perhaps by storing a byte count
119
# TODO: function to check whole file
121
# TODO: atomically append data, then measure backwards from the cursor
122
# position after writing to work out where it was located. we may need to
123
# bypass python file buffering.
125
DATA_SUFFIX = '.knit'
126
INDEX_SUFFIX = '.kndx'
127
_STREAM_MIN_BUFFER_SIZE = 5*1024*1024
130
class KnitAdapter(object):
131
"""Base class for knit record adaption."""
133
def __init__(self, basis_vf):
134
"""Create an adapter which accesses full texts from basis_vf.
136
:param basis_vf: A versioned file to access basis texts of deltas from.
137
May be None for adapters that do not need to access basis texts.
139
self._data = KnitVersionedFiles(None, None)
140
self._annotate_factory = KnitAnnotateFactory()
141
self._plain_factory = KnitPlainFactory()
142
self._basis_vf = basis_vf
145
class FTAnnotatedToUnannotated(KnitAdapter):
146
"""An adapter from FT annotated knits to unannotated ones."""
148
def get_bytes(self, factory):
149
annotated_compressed_bytes = factory._raw_record
151
self._data._parse_record_unchecked(annotated_compressed_bytes)
152
content = self._annotate_factory.parse_fulltext(contents, rec[1])
153
size, bytes = self._data._record_to_data((rec[1],), rec[3], content.text())
157
class DeltaAnnotatedToUnannotated(KnitAdapter):
158
"""An adapter for deltas from annotated to unannotated."""
160
def get_bytes(self, factory):
161
annotated_compressed_bytes = factory._raw_record
163
self._data._parse_record_unchecked(annotated_compressed_bytes)
164
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
166
contents = self._plain_factory.lower_line_delta(delta)
167
size, bytes = self._data._record_to_data((rec[1],), rec[3], contents)
171
class FTAnnotatedToFullText(KnitAdapter):
172
"""An adapter from FT annotated knits to unannotated ones."""
174
def get_bytes(self, factory):
175
annotated_compressed_bytes = factory._raw_record
177
self._data._parse_record_unchecked(annotated_compressed_bytes)
178
content, delta = self._annotate_factory.parse_record(factory.key[-1],
179
contents, factory._build_details, None)
180
return ''.join(content.text())
183
class DeltaAnnotatedToFullText(KnitAdapter):
184
"""An adapter for deltas from annotated to unannotated."""
186
def get_bytes(self, factory):
187
annotated_compressed_bytes = factory._raw_record
189
self._data._parse_record_unchecked(annotated_compressed_bytes)
190
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
192
compression_parent = factory.parents[0]
193
basis_entry = self._basis_vf.get_record_stream(
194
[compression_parent], 'unordered', True).next()
195
if basis_entry.storage_kind == 'absent':
196
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
197
basis_chunks = basis_entry.get_bytes_as('chunked')
198
basis_lines = osutils.chunks_to_lines(basis_chunks)
199
# Manually apply the delta because we have one annotated content and
201
basis_content = PlainKnitContent(basis_lines, compression_parent)
202
basis_content.apply_delta(delta, rec[1])
203
basis_content._should_strip_eol = factory._build_details[1]
204
return ''.join(basis_content.text())
207
class FTPlainToFullText(KnitAdapter):
208
"""An adapter from FT plain knits to unannotated ones."""
210
def get_bytes(self, factory):
211
compressed_bytes = factory._raw_record
213
self._data._parse_record_unchecked(compressed_bytes)
214
content, delta = self._plain_factory.parse_record(factory.key[-1],
215
contents, factory._build_details, None)
216
return ''.join(content.text())
219
class DeltaPlainToFullText(KnitAdapter):
220
"""An adapter for deltas from annotated to unannotated."""
222
def get_bytes(self, factory):
223
compressed_bytes = factory._raw_record
225
self._data._parse_record_unchecked(compressed_bytes)
226
delta = self._plain_factory.parse_line_delta(contents, rec[1])
227
compression_parent = factory.parents[0]
228
# XXX: string splitting overhead.
229
basis_entry = self._basis_vf.get_record_stream(
230
[compression_parent], 'unordered', True).next()
231
if basis_entry.storage_kind == 'absent':
232
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
233
basis_chunks = basis_entry.get_bytes_as('chunked')
234
basis_lines = osutils.chunks_to_lines(basis_chunks)
235
basis_content = PlainKnitContent(basis_lines, compression_parent)
236
# Manually apply the delta because we have one annotated content and
238
content, _ = self._plain_factory.parse_record(rec[1], contents,
239
factory._build_details, basis_content)
240
return ''.join(content.text())
243
class KnitContentFactory(ContentFactory):
244
"""Content factory for streaming from knits.
246
:seealso ContentFactory:
249
def __init__(self, key, parents, build_details, sha1, raw_record,
250
annotated, knit=None, network_bytes=None):
251
"""Create a KnitContentFactory for key.
254
:param parents: The parents.
255
:param build_details: The build details as returned from
257
:param sha1: The sha1 expected from the full text of this object.
258
:param raw_record: The bytes of the knit data from disk.
259
:param annotated: True if the raw data is annotated.
260
:param network_bytes: None to calculate the network bytes on demand,
261
not-none if they are already known.
263
ContentFactory.__init__(self)
266
self.parents = parents
267
if build_details[0] == 'line-delta':
272
annotated_kind = 'annotated-'
275
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
276
self._raw_record = raw_record
277
self._network_bytes = network_bytes
278
self._build_details = build_details
281
def _create_network_bytes(self):
282
"""Create a fully serialised network version for transmission."""
283
# storage_kind, key, parents, Noeol, raw_record
284
key_bytes = '\x00'.join(self.key)
285
if self.parents is None:
286
parent_bytes = 'None:'
288
parent_bytes = '\t'.join('\x00'.join(key) for key in self.parents)
289
if self._build_details[1]:
293
network_bytes = "%s\n%s\n%s\n%s%s" % (self.storage_kind, key_bytes,
294
parent_bytes, noeol, self._raw_record)
295
self._network_bytes = network_bytes
297
def get_bytes_as(self, storage_kind):
298
if storage_kind == self.storage_kind:
299
if self._network_bytes is None:
300
self._create_network_bytes()
301
return self._network_bytes
302
if ('-ft-' in self.storage_kind and
303
storage_kind in ('chunked', 'fulltext')):
304
adapter_key = (self.storage_kind, 'fulltext')
305
adapter_factory = adapter_registry.get(adapter_key)
306
adapter = adapter_factory(None)
307
bytes = adapter.get_bytes(self)
308
if storage_kind == 'chunked':
312
if self._knit is not None:
313
# Not redundant with direct conversion above - that only handles
315
if storage_kind == 'chunked':
316
return self._knit.get_lines(self.key[0])
317
elif storage_kind == 'fulltext':
318
return self._knit.get_text(self.key[0])
319
raise errors.UnavailableRepresentation(self.key, storage_kind,
323
class LazyKnitContentFactory(ContentFactory):
324
"""A ContentFactory which can either generate full text or a wire form.
326
:seealso ContentFactory:
329
def __init__(self, key, parents, generator, first):
330
"""Create a LazyKnitContentFactory.
332
:param key: The key of the record.
333
:param parents: The parents of the record.
334
:param generator: A _ContentMapGenerator containing the record for this
336
:param first: Is this the first content object returned from generator?
337
if it is, its storage kind is knit-delta-closure, otherwise it is
338
knit-delta-closure-ref
341
self.parents = parents
343
self._generator = generator
344
self.storage_kind = "knit-delta-closure"
346
self.storage_kind = self.storage_kind + "-ref"
349
def get_bytes_as(self, storage_kind):
350
if storage_kind == self.storage_kind:
352
return self._generator._wire_bytes()
354
# all the keys etc are contained in the bytes returned in the
357
if storage_kind in ('chunked', 'fulltext'):
358
chunks = self._generator._get_one_work(self.key).text()
359
if storage_kind == 'chunked':
362
return ''.join(chunks)
363
raise errors.UnavailableRepresentation(self.key, storage_kind,
367
def knit_delta_closure_to_records(storage_kind, bytes, line_end):
368
"""Convert a network record to a iterator over stream records.
370
:param storage_kind: The storage kind of the record.
371
Must be 'knit-delta-closure'.
372
:param bytes: The bytes of the record on the network.
374
generator = _NetworkContentMapGenerator(bytes, line_end)
375
return generator.get_record_stream()
378
def knit_network_to_record(storage_kind, bytes, line_end):
379
"""Convert a network record to a record object.
381
:param storage_kind: The storage kind of the record.
382
:param bytes: The bytes of the record on the network.
385
line_end = bytes.find('\n', start)
386
key = tuple(bytes[start:line_end].split('\x00'))
388
line_end = bytes.find('\n', start)
389
parent_line = bytes[start:line_end]
390
if parent_line == 'None:':
394
[tuple(segment.split('\x00')) for segment in parent_line.split('\t')
397
noeol = bytes[start] == 'N'
398
if 'ft' in storage_kind:
401
method = 'line-delta'
402
build_details = (method, noeol)
404
raw_record = bytes[start:]
405
annotated = 'annotated' in storage_kind
406
return [KnitContentFactory(key, parents, build_details, None, raw_record,
407
annotated, network_bytes=bytes)]
410
class KnitContent(object):
411
"""Content of a knit version to which deltas can be applied.
413
This is always stored in memory as a list of lines with \n at the end,
414
plus a flag saying if the final ending is really there or not, because that
415
corresponds to the on-disk knit representation.
419
self._should_strip_eol = False
421
def apply_delta(self, delta, new_version_id):
422
"""Apply delta to this object to become new_version_id."""
423
raise NotImplementedError(self.apply_delta)
425
def line_delta_iter(self, new_lines):
426
"""Generate line-based delta from this content to new_lines."""
427
new_texts = new_lines.text()
428
old_texts = self.text()
429
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
430
for tag, i1, i2, j1, j2 in s.get_opcodes():
433
# ofrom, oto, length, data
434
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
436
def line_delta(self, new_lines):
437
return list(self.line_delta_iter(new_lines))
440
def get_line_delta_blocks(knit_delta, source, target):
441
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
442
target_len = len(target)
445
for s_begin, s_end, t_len, new_text in knit_delta:
446
true_n = s_begin - s_pos
449
# knit deltas do not provide reliable info about whether the
450
# last line of a file matches, due to eol handling.
451
if source[s_pos + n -1] != target[t_pos + n -1]:
454
yield s_pos, t_pos, n
455
t_pos += t_len + true_n
457
n = target_len - t_pos
459
if source[s_pos + n -1] != target[t_pos + n -1]:
462
yield s_pos, t_pos, n
463
yield s_pos + (target_len - t_pos), target_len, 0
466
class AnnotatedKnitContent(KnitContent):
467
"""Annotated content."""
469
def __init__(self, lines):
470
KnitContent.__init__(self)
474
"""Return a list of (origin, text) for each content line."""
475
lines = self._lines[:]
476
if self._should_strip_eol:
477
origin, last_line = lines[-1]
478
lines[-1] = (origin, last_line.rstrip('\n'))
481
def apply_delta(self, delta, new_version_id):
482
"""Apply delta to this object to become new_version_id."""
485
for start, end, count, delta_lines in delta:
486
lines[offset+start:offset+end] = delta_lines
487
offset = offset + (start - end) + count
491
lines = [text for origin, text in self._lines]
492
except ValueError, e:
493
# most commonly (only?) caused by the internal form of the knit
494
# missing annotation information because of a bug - see thread
496
raise KnitCorrupt(self,
497
"line in annotated knit missing annotation information: %s"
499
if self._should_strip_eol:
500
lines[-1] = lines[-1].rstrip('\n')
504
return AnnotatedKnitContent(self._lines[:])
507
class PlainKnitContent(KnitContent):
508
"""Unannotated content.
510
When annotate[_iter] is called on this content, the same version is reported
511
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
515
def __init__(self, lines, version_id):
516
KnitContent.__init__(self)
518
self._version_id = version_id
521
"""Return a list of (origin, text) for each content line."""
522
return [(self._version_id, line) for line in self._lines]
524
def apply_delta(self, delta, new_version_id):
525
"""Apply delta to this object to become new_version_id."""
528
for start, end, count, delta_lines in delta:
529
lines[offset+start:offset+end] = delta_lines
530
offset = offset + (start - end) + count
531
self._version_id = new_version_id
534
return PlainKnitContent(self._lines[:], self._version_id)
538
if self._should_strip_eol:
540
lines[-1] = lines[-1].rstrip('\n')
544
class _KnitFactory(object):
545
"""Base class for common Factory functions."""
547
def parse_record(self, version_id, record, record_details,
548
base_content, copy_base_content=True):
549
"""Parse a record into a full content object.
551
:param version_id: The official version id for this content
552
:param record: The data returned by read_records_iter()
553
:param record_details: Details about the record returned by
555
:param base_content: If get_build_details returns a compression_parent,
556
you must return a base_content here, else use None
557
:param copy_base_content: When building from the base_content, decide
558
you can either copy it and return a new object, or modify it in
560
:return: (content, delta) A Content object and possibly a line-delta,
563
method, noeol = record_details
564
if method == 'line-delta':
565
if copy_base_content:
566
content = base_content.copy()
568
content = base_content
569
delta = self.parse_line_delta(record, version_id)
570
content.apply_delta(delta, version_id)
572
content = self.parse_fulltext(record, version_id)
574
content._should_strip_eol = noeol
575
return (content, delta)
578
class KnitAnnotateFactory(_KnitFactory):
579
"""Factory for creating annotated Content objects."""
583
def make(self, lines, version_id):
584
num_lines = len(lines)
585
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
587
def parse_fulltext(self, content, version_id):
588
"""Convert fulltext to internal representation
590
fulltext content is of the format
591
revid(utf8) plaintext\n
592
internal representation is of the format:
595
# TODO: jam 20070209 The tests expect this to be returned as tuples,
596
# but the code itself doesn't really depend on that.
597
# Figure out a way to not require the overhead of turning the
598
# list back into tuples.
599
lines = [tuple(line.split(' ', 1)) for line in content]
600
return AnnotatedKnitContent(lines)
602
def parse_line_delta_iter(self, lines):
603
return iter(self.parse_line_delta(lines))
605
def parse_line_delta(self, lines, version_id, plain=False):
606
"""Convert a line based delta into internal representation.
608
line delta is in the form of:
609
intstart intend intcount
611
revid(utf8) newline\n
612
internal representation is
613
(start, end, count, [1..count tuples (revid, newline)])
615
:param plain: If True, the lines are returned as a plain
616
list without annotations, not as a list of (origin, content) tuples, i.e.
617
(start, end, count, [1..count newline])
624
def cache_and_return(line):
625
origin, text = line.split(' ', 1)
626
return cache.setdefault(origin, origin), text
628
# walk through the lines parsing.
629
# Note that the plain test is explicitly pulled out of the
630
# loop to minimise any performance impact
633
start, end, count = [int(n) for n in header.split(',')]
634
contents = [next().split(' ', 1)[1] for i in xrange(count)]
635
result.append((start, end, count, contents))
638
start, end, count = [int(n) for n in header.split(',')]
639
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
640
result.append((start, end, count, contents))
643
def get_fulltext_content(self, lines):
644
"""Extract just the content lines from a fulltext."""
645
return (line.split(' ', 1)[1] for line in lines)
647
def get_linedelta_content(self, lines):
648
"""Extract just the content from a line delta.
650
This doesn't return all of the extra information stored in a delta.
651
Only the actual content lines.
656
header = header.split(',')
657
count = int(header[2])
658
for i in xrange(count):
659
origin, text = next().split(' ', 1)
662
def lower_fulltext(self, content):
663
"""convert a fulltext content record into a serializable form.
665
see parse_fulltext which this inverts.
667
# TODO: jam 20070209 We only do the caching thing to make sure that
668
# the origin is a valid utf-8 line, eventually we could remove it
669
return ['%s %s' % (o, t) for o, t in content._lines]
671
def lower_line_delta(self, delta):
672
"""convert a delta into a serializable form.
674
See parse_line_delta which this inverts.
676
# TODO: jam 20070209 We only do the caching thing to make sure that
677
# the origin is a valid utf-8 line, eventually we could remove it
679
for start, end, c, lines in delta:
680
out.append('%d,%d,%d\n' % (start, end, c))
681
out.extend(origin + ' ' + text
682
for origin, text in lines)
685
def annotate(self, knit, key):
686
content = knit._get_content(key)
687
# adjust for the fact that serialised annotations are only key suffixes
689
if type(key) == tuple:
691
origins = content.annotate()
693
for origin, line in origins:
694
result.append((prefix + (origin,), line))
697
# XXX: This smells a bit. Why would key ever be a non-tuple here?
698
# Aren't keys defined to be tuples? -- spiv 20080618
699
return content.annotate()
702
class KnitPlainFactory(_KnitFactory):
703
"""Factory for creating plain Content objects."""
707
def make(self, lines, version_id):
708
return PlainKnitContent(lines, version_id)
710
def parse_fulltext(self, content, version_id):
711
"""This parses an unannotated fulltext.
713
Note that this is not a noop - the internal representation
714
has (versionid, line) - its just a constant versionid.
716
return self.make(content, version_id)
718
def parse_line_delta_iter(self, lines, version_id):
720
num_lines = len(lines)
721
while cur < num_lines:
724
start, end, c = [int(n) for n in header.split(',')]
725
yield start, end, c, lines[cur:cur+c]
728
def parse_line_delta(self, lines, version_id):
729
return list(self.parse_line_delta_iter(lines, version_id))
731
def get_fulltext_content(self, lines):
732
"""Extract just the content lines from a fulltext."""
735
def get_linedelta_content(self, lines):
736
"""Extract just the content from a line delta.
738
This doesn't return all of the extra information stored in a delta.
739
Only the actual content lines.
744
header = header.split(',')
745
count = int(header[2])
746
for i in xrange(count):
749
def lower_fulltext(self, content):
750
return content.text()
752
def lower_line_delta(self, delta):
754
for start, end, c, lines in delta:
755
out.append('%d,%d,%d\n' % (start, end, c))
759
def annotate(self, knit, key):
760
annotator = _KnitAnnotator(knit)
761
return annotator.annotate(key)
765
def make_file_factory(annotated, mapper):
766
"""Create a factory for creating a file based KnitVersionedFiles.
768
This is only functional enough to run interface tests, it doesn't try to
769
provide a full pack environment.
771
:param annotated: knit annotations are wanted.
772
:param mapper: The mapper from keys to paths.
774
def factory(transport):
775
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
776
access = _KnitKeyAccess(transport, mapper)
777
return KnitVersionedFiles(index, access, annotated=annotated)
781
def make_pack_factory(graph, delta, keylength):
782
"""Create a factory for creating a pack based VersionedFiles.
784
This is only functional enough to run interface tests, it doesn't try to
785
provide a full pack environment.
787
:param graph: Store a graph.
788
:param delta: Delta compress contents.
789
:param keylength: How long should keys be.
791
def factory(transport):
792
parents = graph or delta
798
max_delta_chain = 200
801
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
802
key_elements=keylength)
803
stream = transport.open_write_stream('newpack')
804
writer = pack.ContainerWriter(stream.write)
806
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
807
deltas=delta, add_callback=graph_index.add_nodes)
808
access = _DirectPackAccess({})
809
access.set_writer(writer, graph_index, (transport, 'newpack'))
810
result = KnitVersionedFiles(index, access,
811
max_delta_chain=max_delta_chain)
812
result.stream = stream
813
result.writer = writer
818
def cleanup_pack_knit(versioned_files):
819
versioned_files.stream.close()
820
versioned_files.writer.end()
823
def _get_total_build_size(self, keys, positions):
824
"""Determine the total bytes to build these keys.
826
(helper function because _KnitGraphIndex and _KndxIndex work the same, but
827
don't inherit from a common base.)
829
:param keys: Keys that we want to build
830
:param positions: dict of {key, (info, index_memo, comp_parent)} (such
831
as returned by _get_components_positions)
832
:return: Number of bytes to build those keys
834
all_build_index_memos = {}
838
for key in build_keys:
839
# This is mostly for the 'stacked' case
840
# Where we will be getting the data from a fallback
841
if key not in positions:
843
_, index_memo, compression_parent = positions[key]
844
all_build_index_memos[key] = index_memo
845
if compression_parent not in all_build_index_memos:
846
next_keys.add(compression_parent)
847
build_keys = next_keys
848
return sum([index_memo[2] for index_memo
849
in all_build_index_memos.itervalues()])
852
class KnitVersionedFiles(VersionedFiles):
853
"""Storage for many versioned files using knit compression.
855
Backend storage is managed by indices and data objects.
857
:ivar _index: A _KnitGraphIndex or similar that can describe the
858
parents, graph, compression and data location of entries in this
859
KnitVersionedFiles. Note that this is only the index for
860
*this* vfs; if there are fallbacks they must be queried separately.
863
def __init__(self, index, data_access, max_delta_chain=200,
864
annotated=False, reload_func=None):
865
"""Create a KnitVersionedFiles with index and data_access.
867
:param index: The index for the knit data.
868
:param data_access: The access object to store and retrieve knit
870
:param max_delta_chain: The maximum number of deltas to permit during
871
insertion. Set to 0 to prohibit the use of deltas.
872
:param annotated: Set to True to cause annotations to be calculated and
873
stored during insertion.
874
:param reload_func: An function that can be called if we think we need
875
to reload the pack listing and try again. See
876
'bzrlib.repofmt.pack_repo.AggregateIndex' for the signature.
879
self._access = data_access
880
self._max_delta_chain = max_delta_chain
882
self._factory = KnitAnnotateFactory()
884
self._factory = KnitPlainFactory()
885
self._fallback_vfs = []
886
self._reload_func = reload_func
889
return "%s(%r, %r)" % (
890
self.__class__.__name__,
894
def add_fallback_versioned_files(self, a_versioned_files):
895
"""Add a source of texts for texts not present in this knit.
897
:param a_versioned_files: A VersionedFiles object.
899
self._fallback_vfs.append(a_versioned_files)
901
def add_lines(self, key, parents, lines, parent_texts=None,
902
left_matching_blocks=None, nostore_sha=None, random_id=False,
904
"""See VersionedFiles.add_lines()."""
905
self._index._check_write_ok()
906
self._check_add(key, lines, random_id, check_content)
908
# The caller might pass None if there is no graph data, but kndx
909
# indexes can't directly store that, so we give them
910
# an empty tuple instead.
912
return self._add(key, lines, parents,
913
parent_texts, left_matching_blocks, nostore_sha, random_id)
915
def _add(self, key, lines, parents, parent_texts,
916
left_matching_blocks, nostore_sha, random_id):
917
"""Add a set of lines on top of version specified by parents.
919
Any versions not present will be converted into ghosts.
921
# first thing, if the content is something we don't need to store, find
923
line_bytes = ''.join(lines)
924
digest = sha_string(line_bytes)
925
if nostore_sha == digest:
926
raise errors.ExistingContent
929
if parent_texts is None:
931
# Do a single query to ascertain parent presence; we only compress
932
# against parents in the same kvf.
933
present_parent_map = self._index.get_parent_map(parents)
934
for parent in parents:
935
if parent in present_parent_map:
936
present_parents.append(parent)
938
# Currently we can only compress against the left most present parent.
939
if (len(present_parents) == 0 or
940
present_parents[0] != parents[0]):
943
# To speed the extract of texts the delta chain is limited
944
# to a fixed number of deltas. This should minimize both
945
# I/O and the time spend applying deltas.
946
delta = self._check_should_delta(present_parents[0])
948
text_length = len(line_bytes)
951
if lines[-1][-1] != '\n':
952
# copy the contents of lines.
954
options.append('no-eol')
955
lines[-1] = lines[-1] + '\n'
959
if type(element) != str:
960
raise TypeError("key contains non-strings: %r" % (key,))
961
# Knit hunks are still last-element only
963
content = self._factory.make(lines, version_id)
964
if 'no-eol' in options:
965
# Hint to the content object that its text() call should strip the
967
content._should_strip_eol = True
968
if delta or (self._factory.annotated and len(present_parents) > 0):
969
# Merge annotations from parent texts if needed.
970
delta_hunks = self._merge_annotations(content, present_parents,
971
parent_texts, delta, self._factory.annotated,
972
left_matching_blocks)
975
options.append('line-delta')
976
store_lines = self._factory.lower_line_delta(delta_hunks)
977
size, bytes = self._record_to_data(key, digest,
980
options.append('fulltext')
981
# isinstance is slower and we have no hierarchy.
982
if self._factory.__class__ is KnitPlainFactory:
983
# Use the already joined bytes saving iteration time in
985
size, bytes = self._record_to_data(key, digest,
988
# get mixed annotation + content and feed it into the
990
store_lines = self._factory.lower_fulltext(content)
991
size, bytes = self._record_to_data(key, digest,
994
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
995
self._index.add_records(
996
((key, options, access_memo, parents),),
998
return digest, text_length, content
1000
def annotate(self, key):
1001
"""See VersionedFiles.annotate."""
1002
return self._factory.annotate(self, key)
1004
def check(self, progress_bar=None):
1005
"""See VersionedFiles.check()."""
1006
# This doesn't actually test extraction of everything, but that will
1007
# impact 'bzr check' substantially, and needs to be integrated with
1008
# care. However, it does check for the obvious problem of a delta with
1010
keys = self._index.keys()
1011
parent_map = self.get_parent_map(keys)
1013
if self._index.get_method(key) != 'fulltext':
1014
compression_parent = parent_map[key][0]
1015
if compression_parent not in parent_map:
1016
raise errors.KnitCorrupt(self,
1017
"Missing basis parent %s for %s" % (
1018
compression_parent, key))
1019
for fallback_vfs in self._fallback_vfs:
1020
fallback_vfs.check()
1022
def _check_add(self, key, lines, random_id, check_content):
1023
"""check that version_id and lines are safe to add."""
1024
version_id = key[-1]
1025
if contains_whitespace(version_id):
1026
raise InvalidRevisionId(version_id, self)
1027
self.check_not_reserved_id(version_id)
1028
# TODO: If random_id==False and the key is already present, we should
1029
# probably check that the existing content is identical to what is
1030
# being inserted, and otherwise raise an exception. This would make
1031
# the bundle code simpler.
1033
self._check_lines_not_unicode(lines)
1034
self._check_lines_are_lines(lines)
1036
def _check_header(self, key, line):
1037
rec = self._split_header(line)
1038
self._check_header_version(rec, key[-1])
1041
def _check_header_version(self, rec, version_id):
1042
"""Checks the header version on original format knit records.
1044
These have the last component of the key embedded in the record.
1046
if rec[1] != version_id:
1047
raise KnitCorrupt(self,
1048
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1050
def _check_should_delta(self, parent):
1051
"""Iterate back through the parent listing, looking for a fulltext.
1053
This is used when we want to decide whether to add a delta or a new
1054
fulltext. It searches for _max_delta_chain parents. When it finds a
1055
fulltext parent, it sees if the total size of the deltas leading up to
1056
it is large enough to indicate that we want a new full text anyway.
1058
Return True if we should create a new delta, False if we should use a
1062
fulltext_size = None
1063
for count in xrange(self._max_delta_chain):
1065
# Note that this only looks in the index of this particular
1066
# KnitVersionedFiles, not in the fallbacks. This ensures that
1067
# we won't store a delta spanning physical repository
1069
build_details = self._index.get_build_details([parent])
1070
parent_details = build_details[parent]
1071
except (RevisionNotPresent, KeyError), e:
1072
# Some basis is not locally present: always fulltext
1074
index_memo, compression_parent, _, _ = parent_details
1075
_, _, size = index_memo
1076
if compression_parent is None:
1077
fulltext_size = size
1080
# We don't explicitly check for presence because this is in an
1081
# inner loop, and if it's missing it'll fail anyhow.
1082
parent = compression_parent
1084
# We couldn't find a fulltext, so we must create a new one
1086
# Simple heuristic - if the total I/O wold be greater as a delta than
1087
# the originally installed fulltext, we create a new fulltext.
1088
return fulltext_size > delta_size
1090
def _build_details_to_components(self, build_details):
1091
"""Convert a build_details tuple to a position tuple."""
1092
# record_details, access_memo, compression_parent
1093
return build_details[3], build_details[0], build_details[1]
1095
def _get_components_positions(self, keys, allow_missing=False):
1096
"""Produce a map of position data for the components of keys.
1098
This data is intended to be used for retrieving the knit records.
1100
A dict of key to (record_details, index_memo, next, parents) is
1102
method is the way referenced data should be applied.
1103
index_memo is the handle to pass to the data access to actually get the
1105
next is the build-parent of the version, or None for fulltexts.
1106
parents is the version_ids of the parents of this version
1108
:param allow_missing: If True do not raise an error on a missing component,
1112
pending_components = keys
1113
while pending_components:
1114
build_details = self._index.get_build_details(pending_components)
1115
current_components = set(pending_components)
1116
pending_components = set()
1117
for key, details in build_details.iteritems():
1118
(index_memo, compression_parent, parents,
1119
record_details) = details
1120
method = record_details[0]
1121
if compression_parent is not None:
1122
pending_components.add(compression_parent)
1123
component_data[key] = self._build_details_to_components(details)
1124
missing = current_components.difference(build_details)
1125
if missing and not allow_missing:
1126
raise errors.RevisionNotPresent(missing.pop(), self)
1127
return component_data
1129
def _get_content(self, key, parent_texts={}):
1130
"""Returns a content object that makes up the specified
1132
cached_version = parent_texts.get(key, None)
1133
if cached_version is not None:
1134
# Ensure the cache dict is valid.
1135
if not self.get_parent_map([key]):
1136
raise RevisionNotPresent(key, self)
1137
return cached_version
1138
generator = _VFContentMapGenerator(self, [key])
1139
return generator._get_content(key)
1141
def get_parent_map(self, keys):
1142
"""Get a map of the graph parents of keys.
1144
:param keys: The keys to look up parents for.
1145
:return: A mapping from keys to parents. Absent keys are absent from
1148
return self._get_parent_map_with_sources(keys)[0]
1150
def _get_parent_map_with_sources(self, keys):
1151
"""Get a map of the parents of keys.
1153
:param keys: The keys to look up parents for.
1154
:return: A tuple. The first element is a mapping from keys to parents.
1155
Absent keys are absent from the mapping. The second element is a
1156
list with the locations each key was found in. The first element
1157
is the in-this-knit parents, the second the first fallback source,
1161
sources = [self._index] + self._fallback_vfs
1164
for source in sources:
1167
new_result = source.get_parent_map(missing)
1168
source_results.append(new_result)
1169
result.update(new_result)
1170
missing.difference_update(set(new_result))
1171
return result, source_results
1173
def _get_record_map(self, keys, allow_missing=False):
1174
"""Produce a dictionary of knit records.
1176
:return: {key:(record, record_details, digest, next)}
1178
data returned from read_records (a KnitContentobject)
1180
opaque information to pass to parse_record
1182
SHA1 digest of the full text after all steps are done
1184
build-parent of the version, i.e. the leftmost ancestor.
1185
Will be None if the record is not a delta.
1186
:param keys: The keys to build a map for
1187
:param allow_missing: If some records are missing, rather than
1188
error, just return the data that could be generated.
1190
raw_map = self._get_record_map_unparsed(keys,
1191
allow_missing=allow_missing)
1192
return self._raw_map_to_record_map(raw_map)
1194
def _raw_map_to_record_map(self, raw_map):
1195
"""Parse the contents of _get_record_map_unparsed.
1197
:return: see _get_record_map.
1201
data, record_details, next = raw_map[key]
1202
content, digest = self._parse_record(key[-1], data)
1203
result[key] = content, record_details, digest, next
1206
def _get_record_map_unparsed(self, keys, allow_missing=False):
1207
"""Get the raw data for reconstructing keys without parsing it.
1209
:return: A dict suitable for parsing via _raw_map_to_record_map.
1210
key-> raw_bytes, (method, noeol), compression_parent
1212
# This retries the whole request if anything fails. Potentially we
1213
# could be a bit more selective. We could track the keys whose records
1214
# we have successfully found, and then only request the new records
1215
# from there. However, _get_components_positions grabs the whole build
1216
# chain, which means we'll likely try to grab the same records again
1217
# anyway. Also, can the build chains change as part of a pack
1218
# operation? We wouldn't want to end up with a broken chain.
1221
position_map = self._get_components_positions(keys,
1222
allow_missing=allow_missing)
1223
# key = component_id, r = record_details, i_m = index_memo,
1225
records = [(key, i_m) for key, (r, i_m, n)
1226
in position_map.iteritems()]
1227
# Sort by the index memo, so that we request records from the
1228
# same pack file together, and in forward-sorted order
1229
records.sort(key=operator.itemgetter(1))
1231
for key, data in self._read_records_iter_unchecked(records):
1232
(record_details, index_memo, next) = position_map[key]
1233
raw_record_map[key] = data, record_details, next
1234
return raw_record_map
1235
except errors.RetryWithNewPacks, e:
1236
self._access.reload_or_raise(e)
1239
def _split_by_prefix(cls, keys):
1240
"""For the given keys, split them up based on their prefix.
1242
To keep memory pressure somewhat under control, split the
1243
requests back into per-file-id requests, otherwise "bzr co"
1244
extracts the full tree into memory before writing it to disk.
1245
This should be revisited if _get_content_maps() can ever cross
1248
The keys for a given file_id are kept in the same relative order.
1249
Ordering between file_ids is not, though prefix_order will return the
1250
order that the key was first seen.
1252
:param keys: An iterable of key tuples
1253
:return: (split_map, prefix_order)
1254
split_map A dictionary mapping prefix => keys
1255
prefix_order The order that we saw the various prefixes
1257
split_by_prefix = {}
1265
if prefix in split_by_prefix:
1266
split_by_prefix[prefix].append(key)
1268
split_by_prefix[prefix] = [key]
1269
prefix_order.append(prefix)
1270
return split_by_prefix, prefix_order
1272
def _group_keys_for_io(self, keys, non_local_keys, positions,
1273
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1274
"""For the given keys, group them into 'best-sized' requests.
1276
The idea is to avoid making 1 request per file, but to never try to
1277
unpack an entire 1.5GB source tree in a single pass. Also when
1278
possible, we should try to group requests to the same pack file
1281
:return: list of (keys, non_local) tuples that indicate what keys
1282
should be fetched next.
1284
# TODO: Ideally we would group on 2 factors. We want to extract texts
1285
# from the same pack file together, and we want to extract all
1286
# the texts for a given build-chain together. Ultimately it
1287
# probably needs a better global view.
1288
total_keys = len(keys)
1289
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1290
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1292
cur_non_local = set()
1296
for prefix in prefix_order:
1297
keys = prefix_split_keys[prefix]
1298
non_local = prefix_split_non_local_keys.get(prefix, [])
1300
this_size = self._index._get_total_build_size(keys, positions)
1301
cur_size += this_size
1302
cur_keys.extend(keys)
1303
cur_non_local.update(non_local)
1304
if cur_size > _min_buffer_size:
1305
result.append((cur_keys, cur_non_local))
1306
sizes.append(cur_size)
1308
cur_non_local = set()
1311
result.append((cur_keys, cur_non_local))
1312
sizes.append(cur_size)
1315
def get_record_stream(self, keys, ordering, include_delta_closure):
1316
"""Get a stream of records for keys.
1318
:param keys: The keys to include.
1319
:param ordering: Either 'unordered' or 'topological'. A topologically
1320
sorted stream has compression parents strictly before their
1322
:param include_delta_closure: If True then the closure across any
1323
compression parents will be included (in the opaque data).
1324
:return: An iterator of ContentFactory objects, each of which is only
1325
valid until the iterator is advanced.
1327
# keys might be a generator
1331
if not self._index.has_graph:
1332
# Cannot sort when no graph has been stored.
1333
ordering = 'unordered'
1335
remaining_keys = keys
1338
keys = set(remaining_keys)
1339
for content_factory in self._get_remaining_record_stream(keys,
1340
ordering, include_delta_closure):
1341
remaining_keys.discard(content_factory.key)
1342
yield content_factory
1344
except errors.RetryWithNewPacks, e:
1345
self._access.reload_or_raise(e)
1347
def _get_remaining_record_stream(self, keys, ordering,
1348
include_delta_closure):
1349
"""This function is the 'retry' portion for get_record_stream."""
1350
if include_delta_closure:
1351
positions = self._get_components_positions(keys, allow_missing=True)
1353
build_details = self._index.get_build_details(keys)
1355
# (record_details, access_memo, compression_parent_key)
1356
positions = dict((key, self._build_details_to_components(details))
1357
for key, details in build_details.iteritems())
1358
absent_keys = keys.difference(set(positions))
1359
# There may be more absent keys : if we're missing the basis component
1360
# and are trying to include the delta closure.
1361
# XXX: We should not ever need to examine remote sources because we do
1362
# not permit deltas across versioned files boundaries.
1363
if include_delta_closure:
1364
needed_from_fallback = set()
1365
# Build up reconstructable_keys dict. key:True in this dict means
1366
# the key can be reconstructed.
1367
reconstructable_keys = {}
1371
chain = [key, positions[key][2]]
1373
needed_from_fallback.add(key)
1376
while chain[-1] is not None:
1377
if chain[-1] in reconstructable_keys:
1378
result = reconstructable_keys[chain[-1]]
1382
chain.append(positions[chain[-1]][2])
1384
# missing basis component
1385
needed_from_fallback.add(chain[-1])
1388
for chain_key in chain[:-1]:
1389
reconstructable_keys[chain_key] = result
1391
needed_from_fallback.add(key)
1392
# Double index lookups here : need a unified api ?
1393
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1394
if ordering in ('topological', 'groupcompress'):
1395
if ordering == 'topological':
1396
# Global topological sort
1397
present_keys = tsort.topo_sort(global_map)
1399
present_keys = sort_groupcompress(global_map)
1400
# Now group by source:
1402
current_source = None
1403
for key in present_keys:
1404
for parent_map in parent_maps:
1405
if key in parent_map:
1406
key_source = parent_map
1408
if current_source is not key_source:
1409
source_keys.append((key_source, []))
1410
current_source = key_source
1411
source_keys[-1][1].append(key)
1413
if ordering != 'unordered':
1414
raise AssertionError('valid values for ordering are:'
1415
' "unordered", "groupcompress" or "topological" not: %r'
1417
# Just group by source; remote sources first.
1420
for parent_map in reversed(parent_maps):
1421
source_keys.append((parent_map, []))
1422
for key in parent_map:
1423
present_keys.append(key)
1424
source_keys[-1][1].append(key)
1425
# We have been requested to return these records in an order that
1426
# suits us. So we ask the index to give us an optimally sorted
1428
for source, sub_keys in source_keys:
1429
if source is parent_maps[0]:
1430
# Only sort the keys for this VF
1431
self._index._sort_keys_by_io(sub_keys, positions)
1432
absent_keys = keys - set(global_map)
1433
for key in absent_keys:
1434
yield AbsentContentFactory(key)
1435
# restrict our view to the keys we can answer.
1436
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1437
# XXX: At that point we need to consider the impact of double reads by
1438
# utilising components multiple times.
1439
if include_delta_closure:
1440
# XXX: get_content_maps performs its own index queries; allow state
1442
non_local_keys = needed_from_fallback - absent_keys
1443
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1446
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1448
for record in generator.get_record_stream():
1451
for source, keys in source_keys:
1452
if source is parent_maps[0]:
1453
# this KnitVersionedFiles
1454
records = [(key, positions[key][1]) for key in keys]
1455
for key, raw_data, sha1 in self._read_records_iter_raw(records):
1456
(record_details, index_memo, _) = positions[key]
1457
yield KnitContentFactory(key, global_map[key],
1458
record_details, sha1, raw_data, self._factory.annotated, None)
1460
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1461
for record in vf.get_record_stream(keys, ordering,
1462
include_delta_closure):
1465
def get_sha1s(self, keys):
1466
"""See VersionedFiles.get_sha1s()."""
1468
record_map = self._get_record_map(missing, allow_missing=True)
1470
for key, details in record_map.iteritems():
1471
if key not in missing:
1473
# record entry 2 is the 'digest'.
1474
result[key] = details[2]
1475
missing.difference_update(set(result))
1476
for source in self._fallback_vfs:
1479
new_result = source.get_sha1s(missing)
1480
result.update(new_result)
1481
missing.difference_update(set(new_result))
1484
def insert_record_stream(self, stream):
1485
"""Insert a record stream into this container.
1487
:param stream: A stream of records to insert.
1489
:seealso VersionedFiles.get_record_stream:
1491
def get_adapter(adapter_key):
1493
return adapters[adapter_key]
1495
adapter_factory = adapter_registry.get(adapter_key)
1496
adapter = adapter_factory(self)
1497
adapters[adapter_key] = adapter
1500
if self._factory.annotated:
1501
# self is annotated, we need annotated knits to use directly.
1502
annotated = "annotated-"
1505
# self is not annotated, but we can strip annotations cheaply.
1507
convertibles = set(["knit-annotated-ft-gz"])
1508
if self._max_delta_chain:
1509
delta_types.add("knit-annotated-delta-gz")
1510
convertibles.add("knit-annotated-delta-gz")
1511
# The set of types we can cheaply adapt without needing basis texts.
1512
native_types = set()
1513
if self._max_delta_chain:
1514
native_types.add("knit-%sdelta-gz" % annotated)
1515
delta_types.add("knit-%sdelta-gz" % annotated)
1516
native_types.add("knit-%sft-gz" % annotated)
1517
knit_types = native_types.union(convertibles)
1519
# Buffer all index entries that we can't add immediately because their
1520
# basis parent is missing. We don't buffer all because generating
1521
# annotations may require access to some of the new records. However we
1522
# can't generate annotations from new deltas until their basis parent
1523
# is present anyway, so we get away with not needing an index that
1524
# includes the new keys.
1526
# See <http://launchpad.net/bugs/300177> about ordering of compression
1527
# parents in the records - to be conservative, we insist that all
1528
# parents must be present to avoid expanding to a fulltext.
1530
# key = basis_parent, value = index entry to add
1531
buffered_index_entries = {}
1532
for record in stream:
1534
parents = record.parents
1535
if record.storage_kind in delta_types:
1536
# TODO: eventually the record itself should track
1537
# compression_parent
1538
compression_parent = parents[0]
1540
compression_parent = None
1541
# Raise an error when a record is missing.
1542
if record.storage_kind == 'absent':
1543
raise RevisionNotPresent([record.key], self)
1544
elif ((record.storage_kind in knit_types)
1545
and (compression_parent is None
1546
or not self._fallback_vfs
1547
or self._index.has_key(compression_parent)
1548
or not self.has_key(compression_parent))):
1549
# we can insert the knit record literally if either it has no
1550
# compression parent OR we already have its basis in this kvf
1551
# OR the basis is not present even in the fallbacks. In the
1552
# last case it will either turn up later in the stream and all
1553
# will be well, or it won't turn up at all and we'll raise an
1556
# TODO: self.has_key is somewhat redundant with
1557
# self._index.has_key; we really want something that directly
1558
# asks if it's only present in the fallbacks. -- mbp 20081119
1559
if record.storage_kind not in native_types:
1561
adapter_key = (record.storage_kind, "knit-delta-gz")
1562
adapter = get_adapter(adapter_key)
1564
adapter_key = (record.storage_kind, "knit-ft-gz")
1565
adapter = get_adapter(adapter_key)
1566
bytes = adapter.get_bytes(record)
1568
# It's a knit record, it has a _raw_record field (even if
1569
# it was reconstituted from a network stream).
1570
bytes = record._raw_record
1571
options = [record._build_details[0]]
1572
if record._build_details[1]:
1573
options.append('no-eol')
1574
# Just blat it across.
1575
# Note: This does end up adding data on duplicate keys. As
1576
# modern repositories use atomic insertions this should not
1577
# lead to excessive growth in the event of interrupted fetches.
1578
# 'knit' repositories may suffer excessive growth, but as a
1579
# deprecated format this is tolerable. It can be fixed if
1580
# needed by in the kndx index support raising on a duplicate
1581
# add with identical parents and options.
1582
access_memo = self._access.add_raw_records(
1583
[(record.key, len(bytes))], bytes)[0]
1584
index_entry = (record.key, options, access_memo, parents)
1585
if 'fulltext' not in options:
1586
# Not a fulltext, so we need to make sure the compression
1587
# parent will also be present.
1588
# Note that pack backed knits don't need to buffer here
1589
# because they buffer all writes to the transaction level,
1590
# but we don't expose that difference at the index level. If
1591
# the query here has sufficient cost to show up in
1592
# profiling we should do that.
1594
# They're required to be physically in this
1595
# KnitVersionedFiles, not in a fallback.
1596
if not self._index.has_key(compression_parent):
1597
pending = buffered_index_entries.setdefault(
1598
compression_parent, [])
1599
pending.append(index_entry)
1602
self._index.add_records([index_entry])
1603
elif record.storage_kind == 'chunked':
1604
self.add_lines(record.key, parents,
1605
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1607
# Not suitable for direct insertion as a
1608
# delta, either because it's not the right format, or this
1609
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1610
# 0) or because it depends on a base only present in the
1613
# Try getting a fulltext directly from the record.
1614
bytes = record.get_bytes_as('fulltext')
1615
except errors.UnavailableRepresentation:
1616
adapter_key = record.storage_kind, 'fulltext'
1617
adapter = get_adapter(adapter_key)
1618
bytes = adapter.get_bytes(record)
1619
lines = split_lines(bytes)
1621
self.add_lines(record.key, parents, lines)
1622
except errors.RevisionAlreadyPresent:
1624
# Add any records whose basis parent is now available.
1626
added_keys = [record.key]
1628
key = added_keys.pop(0)
1629
if key in buffered_index_entries:
1630
index_entries = buffered_index_entries[key]
1631
self._index.add_records(index_entries)
1633
[index_entry[0] for index_entry in index_entries])
1634
del buffered_index_entries[key]
1635
if buffered_index_entries:
1636
# There were index entries buffered at the end of the stream,
1637
# So these need to be added (if the index supports holding such
1638
# entries for later insertion)
1639
for key in buffered_index_entries:
1640
index_entries = buffered_index_entries[key]
1641
self._index.add_records(index_entries,
1642
missing_compression_parents=True)
1644
def get_missing_compression_parent_keys(self):
1645
"""Return an iterable of keys of missing compression parents.
1647
Check this after calling insert_record_stream to find out if there are
1648
any missing compression parents. If there are, the records that
1649
depend on them are not able to be inserted safely. For atomic
1650
KnitVersionedFiles built on packs, the transaction should be aborted or
1651
suspended - commit will fail at this point. Nonatomic knits will error
1652
earlier because they have no staging area to put pending entries into.
1654
return self._index.get_missing_compression_parents()
1656
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1657
"""Iterate over the lines in the versioned files from keys.
1659
This may return lines from other keys. Each item the returned
1660
iterator yields is a tuple of a line and a text version that that line
1661
is present in (not introduced in).
1663
Ordering of results is in whatever order is most suitable for the
1664
underlying storage format.
1666
If a progress bar is supplied, it may be used to indicate progress.
1667
The caller is responsible for cleaning up progress bars (because this
1671
* Lines are normalised by the underlying store: they will all have \\n
1673
* Lines are returned in arbitrary order.
1674
* If a requested key did not change any lines (or didn't have any
1675
lines), it may not be mentioned at all in the result.
1677
:param pb: Progress bar supplied by caller.
1678
:return: An iterator over (line, key).
1681
pb = progress.DummyProgress()
1687
# we don't care about inclusions, the caller cares.
1688
# but we need to setup a list of records to visit.
1689
# we need key, position, length
1691
build_details = self._index.get_build_details(keys)
1692
for key, details in build_details.iteritems():
1694
key_records.append((key, details[0]))
1695
records_iter = enumerate(self._read_records_iter(key_records))
1696
for (key_idx, (key, data, sha_value)) in records_iter:
1697
pb.update('Walking content', key_idx, total)
1698
compression_parent = build_details[key][1]
1699
if compression_parent is None:
1701
line_iterator = self._factory.get_fulltext_content(data)
1704
line_iterator = self._factory.get_linedelta_content(data)
1705
# Now that we are yielding the data for this key, remove it
1708
# XXX: It might be more efficient to yield (key,
1709
# line_iterator) in the future. However for now, this is a
1710
# simpler change to integrate into the rest of the
1711
# codebase. RBC 20071110
1712
for line in line_iterator:
1715
except errors.RetryWithNewPacks, e:
1716
self._access.reload_or_raise(e)
1717
# If there are still keys we've not yet found, we look in the fallback
1718
# vfs, and hope to find them there. Note that if the keys are found
1719
# but had no changes or no content, the fallback may not return
1721
if keys and not self._fallback_vfs:
1722
# XXX: strictly the second parameter is meant to be the file id
1723
# but it's not easily accessible here.
1724
raise RevisionNotPresent(keys, repr(self))
1725
for source in self._fallback_vfs:
1729
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1730
source_keys.add(key)
1732
keys.difference_update(source_keys)
1733
pb.update('Walking content', total, total)
1735
def _make_line_delta(self, delta_seq, new_content):
1736
"""Generate a line delta from delta_seq and new_content."""
1738
for op in delta_seq.get_opcodes():
1739
if op[0] == 'equal':
1741
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1744
def _merge_annotations(self, content, parents, parent_texts={},
1745
delta=None, annotated=None,
1746
left_matching_blocks=None):
1747
"""Merge annotations for content and generate deltas.
1749
This is done by comparing the annotations based on changes to the text
1750
and generating a delta on the resulting full texts. If annotations are
1751
not being created then a simple delta is created.
1753
if left_matching_blocks is not None:
1754
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1758
for parent_key in parents:
1759
merge_content = self._get_content(parent_key, parent_texts)
1760
if (parent_key == parents[0] and delta_seq is not None):
1763
seq = patiencediff.PatienceSequenceMatcher(
1764
None, merge_content.text(), content.text())
1765
for i, j, n in seq.get_matching_blocks():
1768
# this copies (origin, text) pairs across to the new
1769
# content for any line that matches the last-checked
1771
content._lines[j:j+n] = merge_content._lines[i:i+n]
1772
# XXX: Robert says the following block is a workaround for a
1773
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1774
if content._lines and content._lines[-1][1][-1] != '\n':
1775
# The copied annotation was from a line without a trailing EOL,
1776
# reinstate one for the content object, to ensure correct
1778
line = content._lines[-1][1] + '\n'
1779
content._lines[-1] = (content._lines[-1][0], line)
1781
if delta_seq is None:
1782
reference_content = self._get_content(parents[0], parent_texts)
1783
new_texts = content.text()
1784
old_texts = reference_content.text()
1785
delta_seq = patiencediff.PatienceSequenceMatcher(
1786
None, old_texts, new_texts)
1787
return self._make_line_delta(delta_seq, content)
1789
def _parse_record(self, version_id, data):
1790
"""Parse an original format knit record.
1792
These have the last element of the key only present in the stored data.
1794
rec, record_contents = self._parse_record_unchecked(data)
1795
self._check_header_version(rec, version_id)
1796
return record_contents, rec[3]
1798
def _parse_record_header(self, key, raw_data):
1799
"""Parse a record header for consistency.
1801
:return: the header and the decompressor stream.
1802
as (stream, header_record)
1804
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1807
rec = self._check_header(key, df.readline())
1808
except Exception, e:
1809
raise KnitCorrupt(self,
1810
"While reading {%s} got %s(%s)"
1811
% (key, e.__class__.__name__, str(e)))
1814
def _parse_record_unchecked(self, data):
1816
# 4168 calls in 2880 217 internal
1817
# 4168 calls to _parse_record_header in 2121
1818
# 4168 calls to readlines in 330
1819
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1821
record_contents = df.readlines()
1822
except Exception, e:
1823
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1824
(data, e.__class__.__name__, str(e)))
1825
header = record_contents.pop(0)
1826
rec = self._split_header(header)
1827
last_line = record_contents.pop()
1828
if len(record_contents) != int(rec[2]):
1829
raise KnitCorrupt(self,
1830
'incorrect number of lines %s != %s'
1831
' for version {%s} %s'
1832
% (len(record_contents), int(rec[2]),
1833
rec[1], record_contents))
1834
if last_line != 'end %s\n' % rec[1]:
1835
raise KnitCorrupt(self,
1836
'unexpected version end line %r, wanted %r'
1837
% (last_line, rec[1]))
1839
return rec, record_contents
1841
def _read_records_iter(self, records):
1842
"""Read text records from data file and yield result.
1844
The result will be returned in whatever is the fastest to read.
1845
Not by the order requested. Also, multiple requests for the same
1846
record will only yield 1 response.
1847
:param records: A list of (key, access_memo) entries
1848
:return: Yields (key, contents, digest) in the order
1849
read, not the order requested
1854
# XXX: This smells wrong, IO may not be getting ordered right.
1855
needed_records = sorted(set(records), key=operator.itemgetter(1))
1856
if not needed_records:
1859
# The transport optimizes the fetching as well
1860
# (ie, reads continuous ranges.)
1861
raw_data = self._access.get_raw_records(
1862
[index_memo for key, index_memo in needed_records])
1864
for (key, index_memo), data in \
1865
izip(iter(needed_records), raw_data):
1866
content, digest = self._parse_record(key[-1], data)
1867
yield key, content, digest
1869
def _read_records_iter_raw(self, records):
1870
"""Read text records from data file and yield raw data.
1872
This unpacks enough of the text record to validate the id is
1873
as expected but thats all.
1875
Each item the iterator yields is (key, bytes,
1876
expected_sha1_of_full_text).
1878
for key, data in self._read_records_iter_unchecked(records):
1879
# validate the header (note that we can only use the suffix in
1880
# current knit records).
1881
df, rec = self._parse_record_header(key, data)
1883
yield key, data, rec[3]
1885
def _read_records_iter_unchecked(self, records):
1886
"""Read text records from data file and yield raw data.
1888
No validation is done.
1890
Yields tuples of (key, data).
1892
# setup an iterator of the external records:
1893
# uses readv so nice and fast we hope.
1895
# grab the disk data needed.
1896
needed_offsets = [index_memo for key, index_memo
1898
raw_records = self._access.get_raw_records(needed_offsets)
1900
for key, index_memo in records:
1901
data = raw_records.next()
1904
def _record_to_data(self, key, digest, lines, dense_lines=None):
1905
"""Convert key, digest, lines into a raw data block.
1907
:param key: The key of the record. Currently keys are always serialised
1908
using just the trailing component.
1909
:param dense_lines: The bytes of lines but in a denser form. For
1910
instance, if lines is a list of 1000 bytestrings each ending in \n,
1911
dense_lines may be a list with one line in it, containing all the
1912
1000's lines and their \n's. Using dense_lines if it is already
1913
known is a win because the string join to create bytes in this
1914
function spends less time resizing the final string.
1915
:return: (len, a StringIO instance with the raw data ready to read.)
1917
# Note: using a string copy here increases memory pressure with e.g.
1918
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
1919
# when doing the initial commit of a mozilla tree. RBC 20070921
1920
bytes = ''.join(chain(
1921
["version %s %d %s\n" % (key[-1],
1924
dense_lines or lines,
1925
["end %s\n" % key[-1]]))
1926
if type(bytes) != str:
1927
raise AssertionError(
1928
'data must be plain bytes was %s' % type(bytes))
1929
if lines and lines[-1][-1] != '\n':
1930
raise ValueError('corrupt lines value %r' % lines)
1931
compressed_bytes = tuned_gzip.bytes_to_gzip(bytes)
1932
return len(compressed_bytes), compressed_bytes
1934
def _split_header(self, line):
1937
raise KnitCorrupt(self,
1938
'unexpected number of elements in record header')
1942
"""See VersionedFiles.keys."""
1943
if 'evil' in debug.debug_flags:
1944
trace.mutter_callsite(2, "keys scales with size of history")
1945
sources = [self._index] + self._fallback_vfs
1947
for source in sources:
1948
result.update(source.keys())
1952
class _ContentMapGenerator(object):
1953
"""Generate texts or expose raw deltas for a set of texts."""
1955
def _get_content(self, key):
1956
"""Get the content object for key."""
1957
# Note that _get_content is only called when the _ContentMapGenerator
1958
# has been constructed with just one key requested for reconstruction.
1959
if key in self.nonlocal_keys:
1960
record = self.get_record_stream().next()
1961
# Create a content object on the fly
1962
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
1963
return PlainKnitContent(lines, record.key)
1965
# local keys we can ask for directly
1966
return self._get_one_work(key)
1968
def get_record_stream(self):
1969
"""Get a record stream for the keys requested during __init__."""
1970
for record in self._work():
1974
"""Produce maps of text and KnitContents as dicts.
1976
:return: (text_map, content_map) where text_map contains the texts for
1977
the requested versions and content_map contains the KnitContents.
1979
# NB: By definition we never need to read remote sources unless texts
1980
# are requested from them: we don't delta across stores - and we
1981
# explicitly do not want to to prevent data loss situations.
1982
if self.global_map is None:
1983
self.global_map = self.vf.get_parent_map(self.keys)
1984
nonlocal_keys = self.nonlocal_keys
1986
missing_keys = set(nonlocal_keys)
1987
# Read from remote versioned file instances and provide to our caller.
1988
for source in self.vf._fallback_vfs:
1989
if not missing_keys:
1991
# Loop over fallback repositories asking them for texts - ignore
1992
# any missing from a particular fallback.
1993
for record in source.get_record_stream(missing_keys,
1995
if record.storage_kind == 'absent':
1996
# Not in thie particular stream, may be in one of the
1997
# other fallback vfs objects.
1999
missing_keys.remove(record.key)
2002
self._raw_record_map = self.vf._get_record_map_unparsed(self.keys,
2005
for key in self.keys:
2006
if key in self.nonlocal_keys:
2008
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2011
def _get_one_work(self, requested_key):
2012
# Now, if we have calculated everything already, just return the
2014
if requested_key in self._contents_map:
2015
return self._contents_map[requested_key]
2016
# To simplify things, parse everything at once - code that wants one text
2017
# probably wants them all.
2018
# FUTURE: This function could be improved for the 'extract many' case
2019
# by tracking each component and only doing the copy when the number of
2020
# children than need to apply delta's to it is > 1 or it is part of the
2022
multiple_versions = len(self.keys) != 1
2023
if self._record_map is None:
2024
self._record_map = self.vf._raw_map_to_record_map(
2025
self._raw_record_map)
2026
record_map = self._record_map
2027
# raw_record_map is key:
2028
# Have read and parsed records at this point.
2029
for key in self.keys:
2030
if key in self.nonlocal_keys:
2035
while cursor is not None:
2037
record, record_details, digest, next = record_map[cursor]
2039
raise RevisionNotPresent(cursor, self)
2040
components.append((cursor, record, record_details, digest))
2042
if cursor in self._contents_map:
2043
# no need to plan further back
2044
components.append((cursor, None, None, None))
2048
for (component_id, record, record_details,
2049
digest) in reversed(components):
2050
if component_id in self._contents_map:
2051
content = self._contents_map[component_id]
2053
content, delta = self._factory.parse_record(key[-1],
2054
record, record_details, content,
2055
copy_base_content=multiple_versions)
2056
if multiple_versions:
2057
self._contents_map[component_id] = content
2059
# digest here is the digest from the last applied component.
2060
text = content.text()
2061
actual_sha = sha_strings(text)
2062
if actual_sha != digest:
2063
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2064
if multiple_versions:
2065
return self._contents_map[requested_key]
2069
def _wire_bytes(self):
2070
"""Get the bytes to put on the wire for 'key'.
2072
The first collection of bytes asked for returns the serialised
2073
raw_record_map and the additional details (key, parent) for key.
2074
Subsequent calls return just the additional details (key, parent).
2075
The wire storage_kind given for the first key is 'knit-delta-closure',
2076
For subsequent keys it is 'knit-delta-closure-ref'.
2078
:param key: A key from the content generator.
2079
:return: Bytes to put on the wire.
2082
# kind marker for dispatch on the far side,
2083
lines.append('knit-delta-closure')
2085
if self.vf._factory.annotated:
2086
lines.append('annotated')
2089
# then the list of keys
2090
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2091
if key not in self.nonlocal_keys]))
2092
# then the _raw_record_map in serialised form:
2094
# for each item in the map:
2096
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2097
# one line with method
2098
# one line with noeol
2099
# one line with next ('' for None)
2100
# one line with byte count of the record bytes
2102
for key, (record_bytes, (method, noeol), next) in \
2103
self._raw_record_map.iteritems():
2104
key_bytes = '\x00'.join(key)
2105
parents = self.global_map.get(key, None)
2107
parent_bytes = 'None:'
2109
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2110
method_bytes = method
2116
next_bytes = '\x00'.join(next)
2119
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2120
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2121
len(record_bytes), record_bytes))
2122
map_bytes = ''.join(map_byte_list)
2123
lines.append(map_bytes)
2124
bytes = '\n'.join(lines)
2128
class _VFContentMapGenerator(_ContentMapGenerator):
2129
"""Content map generator reading from a VersionedFiles object."""
2131
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2132
global_map=None, raw_record_map=None):
2133
"""Create a _ContentMapGenerator.
2135
:param versioned_files: The versioned files that the texts are being
2137
:param keys: The keys to produce content maps for.
2138
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2139
which are known to not be in this knit, but rather in one of the
2141
:param global_map: The result of get_parent_map(keys) (or a supermap).
2142
This is required if get_record_stream() is to be used.
2143
:param raw_record_map: A unparsed raw record map to use for answering
2146
# The vf to source data from
2147
self.vf = versioned_files
2149
self.keys = list(keys)
2150
# Keys known to be in fallback vfs objects
2151
if nonlocal_keys is None:
2152
self.nonlocal_keys = set()
2154
self.nonlocal_keys = frozenset(nonlocal_keys)
2155
# Parents data for keys to be returned in get_record_stream
2156
self.global_map = global_map
2157
# The chunked lists for self.keys in text form
2159
# A cache of KnitContent objects used in extracting texts.
2160
self._contents_map = {}
2161
# All the knit records needed to assemble the requested keys as full
2163
self._record_map = None
2164
if raw_record_map is None:
2165
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2168
self._raw_record_map = raw_record_map
2169
# the factory for parsing records
2170
self._factory = self.vf._factory
2173
class _NetworkContentMapGenerator(_ContentMapGenerator):
2174
"""Content map generator sourced from a network stream."""
2176
def __init__(self, bytes, line_end):
2177
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2179
self.global_map = {}
2180
self._raw_record_map = {}
2181
self._contents_map = {}
2182
self._record_map = None
2183
self.nonlocal_keys = []
2184
# Get access to record parsing facilities
2185
self.vf = KnitVersionedFiles(None, None)
2188
line_end = bytes.find('\n', start)
2189
line = bytes[start:line_end]
2190
start = line_end + 1
2191
if line == 'annotated':
2192
self._factory = KnitAnnotateFactory()
2194
self._factory = KnitPlainFactory()
2195
# list of keys to emit in get_record_stream
2196
line_end = bytes.find('\n', start)
2197
line = bytes[start:line_end]
2198
start = line_end + 1
2200
tuple(segment.split('\x00')) for segment in line.split('\t')
2202
# now a loop until the end. XXX: It would be nice if this was just a
2203
# bunch of the same records as get_record_stream(..., False) gives, but
2204
# there is a decent sized gap stopping that at the moment.
2208
line_end = bytes.find('\n', start)
2209
key = tuple(bytes[start:line_end].split('\x00'))
2210
start = line_end + 1
2211
# 1 line with parents (None: for None, '' for ())
2212
line_end = bytes.find('\n', start)
2213
line = bytes[start:line_end]
2218
[tuple(segment.split('\x00')) for segment in line.split('\t')
2220
self.global_map[key] = parents
2221
start = line_end + 1
2222
# one line with method
2223
line_end = bytes.find('\n', start)
2224
line = bytes[start:line_end]
2226
start = line_end + 1
2227
# one line with noeol
2228
line_end = bytes.find('\n', start)
2229
line = bytes[start:line_end]
2231
start = line_end + 1
2232
# one line with next ('' for None)
2233
line_end = bytes.find('\n', start)
2234
line = bytes[start:line_end]
2238
next = tuple(bytes[start:line_end].split('\x00'))
2239
start = line_end + 1
2240
# one line with byte count of the record bytes
2241
line_end = bytes.find('\n', start)
2242
line = bytes[start:line_end]
2244
start = line_end + 1
2246
record_bytes = bytes[start:start+count]
2247
start = start + count
2249
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2251
def get_record_stream(self):
2252
"""Get a record stream for for keys requested by the bytestream."""
2254
for key in self.keys:
2255
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2258
def _wire_bytes(self):
2262
class _KndxIndex(object):
2263
"""Manages knit index files
2265
The index is kept in memory and read on startup, to enable
2266
fast lookups of revision information. The cursor of the index
2267
file is always pointing to the end, making it easy to append
2270
_cache is a cache for fast mapping from version id to a Index
2273
_history is a cache for fast mapping from indexes to version ids.
2275
The index data format is dictionary compressed when it comes to
2276
parent references; a index entry may only have parents that with a
2277
lover index number. As a result, the index is topological sorted.
2279
Duplicate entries may be written to the index for a single version id
2280
if this is done then the latter one completely replaces the former:
2281
this allows updates to correct version and parent information.
2282
Note that the two entries may share the delta, and that successive
2283
annotations and references MUST point to the first entry.
2285
The index file on disc contains a header, followed by one line per knit
2286
record. The same revision can be present in an index file more than once.
2287
The first occurrence gets assigned a sequence number starting from 0.
2289
The format of a single line is
2290
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
2291
REVISION_ID is a utf8-encoded revision id
2292
FLAGS is a comma separated list of flags about the record. Values include
2293
no-eol, line-delta, fulltext.
2294
BYTE_OFFSET is the ascii representation of the byte offset in the data file
2295
that the the compressed data starts at.
2296
LENGTH is the ascii representation of the length of the data file.
2297
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
2299
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
2300
revision id already in the knit that is a parent of REVISION_ID.
2301
The ' :' marker is the end of record marker.
2304
when a write is interrupted to the index file, it will result in a line
2305
that does not end in ' :'. If the ' :' is not present at the end of a line,
2306
or at the end of the file, then the record that is missing it will be
2307
ignored by the parser.
2309
When writing new records to the index file, the data is preceded by '\n'
2310
to ensure that records always start on new lines even if the last write was
2311
interrupted. As a result its normal for the last line in the index to be
2312
missing a trailing newline. One can be added with no harmful effects.
2314
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2315
where prefix is e.g. the (fileid,) for .texts instances or () for
2316
constant-mapped things like .revisions, and the old state is
2317
tuple(history_vector, cache_dict). This is used to prevent having an
2318
ABI change with the C extension that reads .kndx files.
2321
HEADER = "# bzr knit index 8\n"
2323
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2324
"""Create a _KndxIndex on transport using mapper."""
2325
self._transport = transport
2326
self._mapper = mapper
2327
self._get_scope = get_scope
2328
self._allow_writes = allow_writes
2329
self._is_locked = is_locked
2331
self.has_graph = True
2333
def add_records(self, records, random_id=False, missing_compression_parents=False):
2334
"""Add multiple records to the index.
2336
:param records: a list of tuples:
2337
(key, options, access_memo, parents).
2338
:param random_id: If True the ids being added were randomly generated
2339
and no check for existence will be performed.
2340
:param missing_compression_parents: If True the records being added are
2341
only compressed against texts already in the index (or inside
2342
records). If False the records all refer to unavailable texts (or
2343
texts inside records) as compression parents.
2345
if missing_compression_parents:
2346
# It might be nice to get the edge of the records. But keys isn't
2348
keys = sorted(record[0] for record in records)
2349
raise errors.RevisionNotPresent(keys, self)
2351
for record in records:
2354
path = self._mapper.map(key) + '.kndx'
2355
path_keys = paths.setdefault(path, (prefix, []))
2356
path_keys[1].append(record)
2357
for path in sorted(paths):
2358
prefix, path_keys = paths[path]
2359
self._load_prefixes([prefix])
2361
orig_history = self._kndx_cache[prefix][1][:]
2362
orig_cache = self._kndx_cache[prefix][0].copy()
2365
for key, options, (_, pos, size), parents in path_keys:
2367
# kndx indices cannot be parentless.
2369
line = "\n%s %s %s %s %s :" % (
2370
key[-1], ','.join(options), pos, size,
2371
self._dictionary_compress(parents))
2372
if type(line) != str:
2373
raise AssertionError(
2374
'data must be utf8 was %s' % type(line))
2376
self._cache_key(key, options, pos, size, parents)
2377
if len(orig_history):
2378
self._transport.append_bytes(path, ''.join(lines))
2380
self._init_index(path, lines)
2382
# If any problems happen, restore the original values and re-raise
2383
self._kndx_cache[prefix] = (orig_cache, orig_history)
2386
def scan_unvalidated_index(self, graph_index):
2387
"""See _KnitGraphIndex.scan_unvalidated_index."""
2388
# Because kndx files do not support atomic insertion via separate index
2389
# files, they do not support this method.
2390
raise NotImplementedError(self.scan_unvalidated_index)
2392
def get_missing_compression_parents(self):
2393
"""See _KnitGraphIndex.get_missing_compression_parents."""
2394
# Because kndx files do not support atomic insertion via separate index
2395
# files, they do not support this method.
2396
raise NotImplementedError(self.get_missing_compression_parents)
2398
def _cache_key(self, key, options, pos, size, parent_keys):
2399
"""Cache a version record in the history array and index cache.
2401
This is inlined into _load_data for performance. KEEP IN SYNC.
2402
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2406
version_id = key[-1]
2407
# last-element only for compatibilty with the C load_data.
2408
parents = tuple(parent[-1] for parent in parent_keys)
2409
for parent in parent_keys:
2410
if parent[:-1] != prefix:
2411
raise ValueError("mismatched prefixes for %r, %r" % (
2413
cache, history = self._kndx_cache[prefix]
2414
# only want the _history index to reference the 1st index entry
2416
if version_id not in cache:
2417
index = len(history)
2418
history.append(version_id)
2420
index = cache[version_id][5]
2421
cache[version_id] = (version_id,
2428
def check_header(self, fp):
2429
line = fp.readline()
2431
# An empty file can actually be treated as though the file doesn't
2433
raise errors.NoSuchFile(self)
2434
if line != self.HEADER:
2435
raise KnitHeaderError(badline=line, filename=self)
2437
def _check_read(self):
2438
if not self._is_locked():
2439
raise errors.ObjectNotLocked(self)
2440
if self._get_scope() != self._scope:
2443
def _check_write_ok(self):
2444
"""Assert if not writes are permitted."""
2445
if not self._is_locked():
2446
raise errors.ObjectNotLocked(self)
2447
if self._get_scope() != self._scope:
2449
if self._mode != 'w':
2450
raise errors.ReadOnlyObjectDirtiedError(self)
2452
def get_build_details(self, keys):
2453
"""Get the method, index_memo and compression parent for keys.
2455
Ghosts are omitted from the result.
2457
:param keys: An iterable of keys.
2458
:return: A dict of key:(index_memo, compression_parent, parents,
2461
opaque structure to pass to read_records to extract the raw
2464
Content that this record is built upon, may be None
2466
Logical parents of this node
2468
extra information about the content which needs to be passed to
2469
Factory.parse_record
2471
parent_map = self.get_parent_map(keys)
2474
if key not in parent_map:
2476
method = self.get_method(key)
2477
parents = parent_map[key]
2478
if method == 'fulltext':
2479
compression_parent = None
2481
compression_parent = parents[0]
2482
noeol = 'no-eol' in self.get_options(key)
2483
index_memo = self.get_position(key)
2484
result[key] = (index_memo, compression_parent,
2485
parents, (method, noeol))
2488
def get_method(self, key):
2489
"""Return compression method of specified key."""
2490
options = self.get_options(key)
2491
if 'fulltext' in options:
2493
elif 'line-delta' in options:
2496
raise errors.KnitIndexUnknownMethod(self, options)
2498
def get_options(self, key):
2499
"""Return a list representing options.
2503
prefix, suffix = self._split_key(key)
2504
self._load_prefixes([prefix])
2506
return self._kndx_cache[prefix][0][suffix][1]
2508
raise RevisionNotPresent(key, self)
2510
def get_parent_map(self, keys):
2511
"""Get a map of the parents of keys.
2513
:param keys: The keys to look up parents for.
2514
:return: A mapping from keys to parents. Absent keys are absent from
2517
# Parse what we need to up front, this potentially trades off I/O
2518
# locality (.kndx and .knit in the same block group for the same file
2519
# id) for less checking in inner loops.
2520
prefixes = set(key[:-1] for key in keys)
2521
self._load_prefixes(prefixes)
2526
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2530
result[key] = tuple(prefix + (suffix,) for
2531
suffix in suffix_parents)
2534
def get_position(self, key):
2535
"""Return details needed to access the version.
2537
:return: a tuple (key, data position, size) to hand to the access
2538
logic to get the record.
2540
prefix, suffix = self._split_key(key)
2541
self._load_prefixes([prefix])
2542
entry = self._kndx_cache[prefix][0][suffix]
2543
return key, entry[2], entry[3]
2545
has_key = _mod_index._has_key_from_parent_map
2547
def _init_index(self, path, extra_lines=[]):
2548
"""Initialize an index."""
2550
sio.write(self.HEADER)
2551
sio.writelines(extra_lines)
2553
self._transport.put_file_non_atomic(path, sio,
2554
create_parent_dir=True)
2555
# self._create_parent_dir)
2556
# mode=self._file_mode,
2557
# dir_mode=self._dir_mode)
2560
"""Get all the keys in the collection.
2562
The keys are not ordered.
2565
# Identify all key prefixes.
2566
# XXX: A bit hacky, needs polish.
2567
if type(self._mapper) == ConstantMapper:
2571
for quoted_relpath in self._transport.iter_files_recursive():
2572
path, ext = os.path.splitext(quoted_relpath)
2574
prefixes = [self._mapper.unmap(path) for path in relpaths]
2575
self._load_prefixes(prefixes)
2576
for prefix in prefixes:
2577
for suffix in self._kndx_cache[prefix][1]:
2578
result.add(prefix + (suffix,))
2581
def _load_prefixes(self, prefixes):
2582
"""Load the indices for prefixes."""
2584
for prefix in prefixes:
2585
if prefix not in self._kndx_cache:
2586
# the load_data interface writes to these variables.
2589
self._filename = prefix
2591
path = self._mapper.map(prefix) + '.kndx'
2592
fp = self._transport.get(path)
2594
# _load_data may raise NoSuchFile if the target knit is
2596
_load_data(self, fp)
2599
self._kndx_cache[prefix] = (self._cache, self._history)
2604
self._kndx_cache[prefix] = ({}, [])
2605
if type(self._mapper) == ConstantMapper:
2606
# preserve behaviour for revisions.kndx etc.
2607
self._init_index(path)
2612
missing_keys = _mod_index._missing_keys_from_parent_map
2614
def _partition_keys(self, keys):
2615
"""Turn keys into a dict of prefix:suffix_list."""
2618
prefix_keys = result.setdefault(key[:-1], [])
2619
prefix_keys.append(key[-1])
2622
def _dictionary_compress(self, keys):
2623
"""Dictionary compress keys.
2625
:param keys: The keys to generate references to.
2626
:return: A string representation of keys. keys which are present are
2627
dictionary compressed, and others are emitted as fulltext with a
2633
prefix = keys[0][:-1]
2634
cache = self._kndx_cache[prefix][0]
2636
if key[:-1] != prefix:
2637
# kndx indices cannot refer across partitioned storage.
2638
raise ValueError("mismatched prefixes for %r" % keys)
2639
if key[-1] in cache:
2640
# -- inlined lookup() --
2641
result_list.append(str(cache[key[-1]][5]))
2642
# -- end lookup () --
2644
result_list.append('.' + key[-1])
2645
return ' '.join(result_list)
2647
def _reset_cache(self):
2648
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2649
# (cache_dict, history_vector) for parsed kndx files.
2650
self._kndx_cache = {}
2651
self._scope = self._get_scope()
2652
allow_writes = self._allow_writes()
2658
def _sort_keys_by_io(self, keys, positions):
2659
"""Figure out an optimal order to read the records for the given keys.
2661
Sort keys, grouped by index and sorted by position.
2663
:param keys: A list of keys whose records we want to read. This will be
2665
:param positions: A dict, such as the one returned by
2666
_get_components_positions()
2669
def get_sort_key(key):
2670
index_memo = positions[key][1]
2671
# Group by prefix and position. index_memo[0] is the key, so it is
2672
# (file_id, revision_id) and we don't want to sort on revision_id,
2673
# index_memo[1] is the position, and index_memo[2] is the size,
2674
# which doesn't matter for the sort
2675
return index_memo[0][:-1], index_memo[1]
2676
return keys.sort(key=get_sort_key)
2678
_get_total_build_size = _get_total_build_size
2680
def _split_key(self, key):
2681
"""Split key into a prefix and suffix."""
2682
return key[:-1], key[-1]
2685
class _KnitGraphIndex(object):
2686
"""A KnitVersionedFiles index layered on GraphIndex."""
2688
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2690
"""Construct a KnitGraphIndex on a graph_index.
2692
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2693
:param is_locked: A callback to check whether the object should answer
2695
:param deltas: Allow delta-compressed records.
2696
:param parents: If True, record knits parents, if not do not record
2698
:param add_callback: If not None, allow additions to the index and call
2699
this callback with a list of added GraphIndex nodes:
2700
[(node, value, node_refs), ...]
2701
:param is_locked: A callback, returns True if the index is locked and
2704
self._add_callback = add_callback
2705
self._graph_index = graph_index
2706
self._deltas = deltas
2707
self._parents = parents
2708
if deltas and not parents:
2709
# XXX: TODO: Delta tree and parent graph should be conceptually
2711
raise KnitCorrupt(self, "Cannot do delta compression without "
2713
self.has_graph = parents
2714
self._is_locked = is_locked
2715
self._missing_compression_parents = set()
2718
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2720
def add_records(self, records, random_id=False,
2721
missing_compression_parents=False):
2722
"""Add multiple records to the index.
2724
This function does not insert data into the Immutable GraphIndex
2725
backing the KnitGraphIndex, instead it prepares data for insertion by
2726
the caller and checks that it is safe to insert then calls
2727
self._add_callback with the prepared GraphIndex nodes.
2729
:param records: a list of tuples:
2730
(key, options, access_memo, parents).
2731
:param random_id: If True the ids being added were randomly generated
2732
and no check for existence will be performed.
2733
:param missing_compression_parents: If True the records being added are
2734
only compressed against texts already in the index (or inside
2735
records). If False the records all refer to unavailable texts (or
2736
texts inside records) as compression parents.
2738
if not self._add_callback:
2739
raise errors.ReadOnlyError(self)
2740
# we hope there are no repositories with inconsistent parentage
2744
compression_parents = set()
2745
for (key, options, access_memo, parents) in records:
2747
parents = tuple(parents)
2748
index, pos, size = access_memo
2749
if 'no-eol' in options:
2753
value += "%d %d" % (pos, size)
2754
if not self._deltas:
2755
if 'line-delta' in options:
2756
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2759
if 'line-delta' in options:
2760
node_refs = (parents, (parents[0],))
2761
if missing_compression_parents:
2762
compression_parents.add(parents[0])
2764
node_refs = (parents, ())
2766
node_refs = (parents, )
2769
raise KnitCorrupt(self, "attempt to add node with parents "
2770
"in parentless index.")
2772
keys[key] = (value, node_refs)
2775
present_nodes = self._get_entries(keys)
2776
for (index, key, value, node_refs) in present_nodes:
2777
if (value[0] != keys[key][0][0] or
2778
node_refs[:1] != keys[key][1][:1]):
2779
raise KnitCorrupt(self, "inconsistent details in add_records"
2780
": %s %s" % ((value, node_refs), keys[key]))
2784
for key, (value, node_refs) in keys.iteritems():
2785
result.append((key, value, node_refs))
2787
for key, (value, node_refs) in keys.iteritems():
2788
result.append((key, value))
2789
self._add_callback(result)
2790
if missing_compression_parents:
2791
# This may appear to be incorrect (it does not check for
2792
# compression parents that are in the existing graph index),
2793
# but such records won't have been buffered, so this is
2794
# actually correct: every entry when
2795
# missing_compression_parents==True either has a missing parent, or
2796
# a parent that is one of the keys in records.
2797
compression_parents.difference_update(keys)
2798
self._missing_compression_parents.update(compression_parents)
2799
# Adding records may have satisfied missing compression parents.
2800
self._missing_compression_parents.difference_update(keys)
2802
def scan_unvalidated_index(self, graph_index):
2803
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2805
This allows this _KnitGraphIndex to keep track of any missing
2806
compression parents we may want to have filled in to make those
2809
:param graph_index: A GraphIndex
2812
new_missing = graph_index.external_references(ref_list_num=1)
2813
new_missing.difference_update(self.get_parent_map(new_missing))
2814
self._missing_compression_parents.update(new_missing)
2816
def get_missing_compression_parents(self):
2817
"""Return the keys of missing compression parents.
2819
Missing compression parents occur when a record stream was missing
2820
basis texts, or a index was scanned that had missing basis texts.
2822
return frozenset(self._missing_compression_parents)
2824
def _check_read(self):
2825
"""raise if reads are not permitted."""
2826
if not self._is_locked():
2827
raise errors.ObjectNotLocked(self)
2829
def _check_write_ok(self):
2830
"""Assert if writes are not permitted."""
2831
if not self._is_locked():
2832
raise errors.ObjectNotLocked(self)
2834
def _compression_parent(self, an_entry):
2835
# return the key that an_entry is compressed against, or None
2836
# Grab the second parent list (as deltas implies parents currently)
2837
compression_parents = an_entry[3][1]
2838
if not compression_parents:
2840
if len(compression_parents) != 1:
2841
raise AssertionError(
2842
"Too many compression parents: %r" % compression_parents)
2843
return compression_parents[0]
2845
def get_build_details(self, keys):
2846
"""Get the method, index_memo and compression parent for version_ids.
2848
Ghosts are omitted from the result.
2850
:param keys: An iterable of keys.
2851
:return: A dict of key:
2852
(index_memo, compression_parent, parents, record_details).
2854
opaque structure to pass to read_records to extract the raw
2857
Content that this record is built upon, may be None
2859
Logical parents of this node
2861
extra information about the content which needs to be passed to
2862
Factory.parse_record
2866
entries = self._get_entries(keys, False)
2867
for entry in entries:
2869
if not self._parents:
2872
parents = entry[3][0]
2873
if not self._deltas:
2874
compression_parent_key = None
2876
compression_parent_key = self._compression_parent(entry)
2877
noeol = (entry[2][0] == 'N')
2878
if compression_parent_key:
2879
method = 'line-delta'
2882
result[key] = (self._node_to_position(entry),
2883
compression_parent_key, parents,
2887
def _get_entries(self, keys, check_present=False):
2888
"""Get the entries for keys.
2890
:param keys: An iterable of index key tuples.
2895
for node in self._graph_index.iter_entries(keys):
2897
found_keys.add(node[1])
2899
# adapt parentless index to the rest of the code.
2900
for node in self._graph_index.iter_entries(keys):
2901
yield node[0], node[1], node[2], ()
2902
found_keys.add(node[1])
2904
missing_keys = keys.difference(found_keys)
2906
raise RevisionNotPresent(missing_keys.pop(), self)
2908
def get_method(self, key):
2909
"""Return compression method of specified key."""
2910
return self._get_method(self._get_node(key))
2912
def _get_method(self, node):
2913
if not self._deltas:
2915
if self._compression_parent(node):
2920
def _get_node(self, key):
2922
return list(self._get_entries([key]))[0]
2924
raise RevisionNotPresent(key, self)
2926
def get_options(self, key):
2927
"""Return a list representing options.
2931
node = self._get_node(key)
2932
options = [self._get_method(node)]
2933
if node[2][0] == 'N':
2934
options.append('no-eol')
2937
def get_parent_map(self, keys):
2938
"""Get a map of the parents of keys.
2940
:param keys: The keys to look up parents for.
2941
:return: A mapping from keys to parents. Absent keys are absent from
2945
nodes = self._get_entries(keys)
2949
result[node[1]] = node[3][0]
2952
result[node[1]] = None
2955
def get_position(self, key):
2956
"""Return details needed to access the version.
2958
:return: a tuple (index, data position, size) to hand to the access
2959
logic to get the record.
2961
node = self._get_node(key)
2962
return self._node_to_position(node)
2964
has_key = _mod_index._has_key_from_parent_map
2967
"""Get all the keys in the collection.
2969
The keys are not ordered.
2972
return [node[1] for node in self._graph_index.iter_all_entries()]
2974
missing_keys = _mod_index._missing_keys_from_parent_map
2976
def _node_to_position(self, node):
2977
"""Convert an index value to position details."""
2978
bits = node[2][1:].split(' ')
2979
return node[0], int(bits[0]), int(bits[1])
2981
def _sort_keys_by_io(self, keys, positions):
2982
"""Figure out an optimal order to read the records for the given keys.
2984
Sort keys, grouped by index and sorted by position.
2986
:param keys: A list of keys whose records we want to read. This will be
2988
:param positions: A dict, such as the one returned by
2989
_get_components_positions()
2992
def get_index_memo(key):
2993
# index_memo is at offset [1]. It is made up of (GraphIndex,
2994
# position, size). GI is an object, which will be unique for each
2995
# pack file. This causes us to group by pack file, then sort by
2996
# position. Size doesn't matter, but it isn't worth breaking up the
2998
return positions[key][1]
2999
return keys.sort(key=get_index_memo)
3001
_get_total_build_size = _get_total_build_size
3004
class _KnitKeyAccess(object):
3005
"""Access to records in .knit files."""
3007
def __init__(self, transport, mapper):
3008
"""Create a _KnitKeyAccess with transport and mapper.
3010
:param transport: The transport the access object is rooted at.
3011
:param mapper: The mapper used to map keys to .knit files.
3013
self._transport = transport
3014
self._mapper = mapper
3016
def add_raw_records(self, key_sizes, raw_data):
3017
"""Add raw knit bytes to a storage area.
3019
The data is spooled to the container writer in one bytes-record per
3022
:param sizes: An iterable of tuples containing the key and size of each
3024
:param raw_data: A bytestring containing the data.
3025
:return: A list of memos to retrieve the record later. Each memo is an
3026
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3027
length), where the key is the record key.
3029
if type(raw_data) != str:
3030
raise AssertionError(
3031
'data must be plain bytes was %s' % type(raw_data))
3034
# TODO: This can be tuned for writing to sftp and other servers where
3035
# append() is relatively expensive by grouping the writes to each key
3037
for key, size in key_sizes:
3038
path = self._mapper.map(key)
3040
base = self._transport.append_bytes(path + '.knit',
3041
raw_data[offset:offset+size])
3042
except errors.NoSuchFile:
3043
self._transport.mkdir(osutils.dirname(path))
3044
base = self._transport.append_bytes(path + '.knit',
3045
raw_data[offset:offset+size])
3049
result.append((key, base, size))
3052
def get_raw_records(self, memos_for_retrieval):
3053
"""Get the raw bytes for a records.
3055
:param memos_for_retrieval: An iterable containing the access memo for
3056
retrieving the bytes.
3057
:return: An iterator over the bytes of the records.
3059
# first pass, group into same-index request to minimise readv's issued.
3061
current_prefix = None
3062
for (key, offset, length) in memos_for_retrieval:
3063
if current_prefix == key[:-1]:
3064
current_list.append((offset, length))
3066
if current_prefix is not None:
3067
request_lists.append((current_prefix, current_list))
3068
current_prefix = key[:-1]
3069
current_list = [(offset, length)]
3070
# handle the last entry
3071
if current_prefix is not None:
3072
request_lists.append((current_prefix, current_list))
3073
for prefix, read_vector in request_lists:
3074
path = self._mapper.map(prefix) + '.knit'
3075
for pos, data in self._transport.readv(path, read_vector):
3079
class _DirectPackAccess(object):
3080
"""Access to data in one or more packs with less translation."""
3082
def __init__(self, index_to_packs, reload_func=None):
3083
"""Create a _DirectPackAccess object.
3085
:param index_to_packs: A dict mapping index objects to the transport
3086
and file names for obtaining data.
3087
:param reload_func: A function to call if we determine that the pack
3088
files have moved and we need to reload our caches. See
3089
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
3091
self._container_writer = None
3092
self._write_index = None
3093
self._indices = index_to_packs
3094
self._reload_func = reload_func
3096
def add_raw_records(self, key_sizes, raw_data):
3097
"""Add raw knit bytes to a storage area.
3099
The data is spooled to the container writer in one bytes-record per
3102
:param sizes: An iterable of tuples containing the key and size of each
3104
:param raw_data: A bytestring containing the data.
3105
:return: A list of memos to retrieve the record later. Each memo is an
3106
opaque index memo. For _DirectPackAccess the memo is (index, pos,
3107
length), where the index field is the write_index object supplied
3108
to the PackAccess object.
3110
if type(raw_data) != str:
3111
raise AssertionError(
3112
'data must be plain bytes was %s' % type(raw_data))
3115
for key, size in key_sizes:
3116
p_offset, p_length = self._container_writer.add_bytes_record(
3117
raw_data[offset:offset+size], [])
3119
result.append((self._write_index, p_offset, p_length))
3122
def get_raw_records(self, memos_for_retrieval):
3123
"""Get the raw bytes for a records.
3125
:param memos_for_retrieval: An iterable containing the (index, pos,
3126
length) memo for retrieving the bytes. The Pack access method
3127
looks up the pack to use for a given record in its index_to_pack
3129
:return: An iterator over the bytes of the records.
3131
# first pass, group into same-index requests
3133
current_index = None
3134
for (index, offset, length) in memos_for_retrieval:
3135
if current_index == index:
3136
current_list.append((offset, length))
3138
if current_index is not None:
3139
request_lists.append((current_index, current_list))
3140
current_index = index
3141
current_list = [(offset, length)]
3142
# handle the last entry
3143
if current_index is not None:
3144
request_lists.append((current_index, current_list))
3145
for index, offsets in request_lists:
3147
transport, path = self._indices[index]
3149
# A KeyError here indicates that someone has triggered an index
3150
# reload, and this index has gone missing, we need to start
3152
if self._reload_func is None:
3153
# If we don't have a _reload_func there is nothing that can
3156
raise errors.RetryWithNewPacks(index,
3157
reload_occurred=True,
3158
exc_info=sys.exc_info())
3160
reader = pack.make_readv_reader(transport, path, offsets)
3161
for names, read_func in reader.iter_records():
3162
yield read_func(None)
3163
except errors.NoSuchFile:
3164
# A NoSuchFile error indicates that a pack file has gone
3165
# missing on disk, we need to trigger a reload, and start over.
3166
if self._reload_func is None:
3168
raise errors.RetryWithNewPacks(transport.abspath(path),
3169
reload_occurred=False,
3170
exc_info=sys.exc_info())
3172
def set_writer(self, writer, index, transport_packname):
3173
"""Set a writer to use for adding data."""
3174
if index is not None:
3175
self._indices[index] = transport_packname
3176
self._container_writer = writer
3177
self._write_index = index
3179
def reload_or_raise(self, retry_exc):
3180
"""Try calling the reload function, or re-raise the original exception.
3182
This should be called after _DirectPackAccess raises a
3183
RetryWithNewPacks exception. This function will handle the common logic
3184
of determining when the error is fatal versus being temporary.
3185
It will also make sure that the original exception is raised, rather
3186
than the RetryWithNewPacks exception.
3188
If this function returns, then the calling function should retry
3189
whatever operation was being performed. Otherwise an exception will
3192
:param retry_exc: A RetryWithNewPacks exception.
3195
if self._reload_func is None:
3197
elif not self._reload_func():
3198
# The reload claimed that nothing changed
3199
if not retry_exc.reload_occurred:
3200
# If there wasn't an earlier reload, then we really were
3201
# expecting to find changes. We didn't find them, so this is a
3205
exc_class, exc_value, exc_traceback = retry_exc.exc_info
3206
raise exc_class, exc_value, exc_traceback
3209
# Deprecated, use PatienceSequenceMatcher instead
3210
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
3213
def annotate_knit(knit, revision_id):
3214
"""Annotate a knit with no cached annotations.
3216
This implementation is for knits with no cached annotations.
3217
It will work for knits with cached annotations, but this is not
3220
annotator = _KnitAnnotator(knit)
3221
return iter(annotator.annotate(revision_id))
3224
class _KnitAnnotator(object):
3225
"""Build up the annotations for a text."""
3227
def __init__(self, knit):
3230
# Content objects, differs from fulltexts because of how final newlines
3231
# are treated by knits. the content objects here will always have a
3233
self._fulltext_contents = {}
3235
# Annotated lines of specific revisions
3236
self._annotated_lines = {}
3238
# Track the raw data for nodes that we could not process yet.
3239
# This maps the revision_id of the base to a list of children that will
3240
# annotated from it.
3241
self._pending_children = {}
3243
# Nodes which cannot be extracted
3244
self._ghosts = set()
3246
# Track how many children this node has, so we know if we need to keep
3248
self._annotate_children = {}
3249
self._compression_children = {}
3251
self._all_build_details = {}
3252
# The children => parent revision_id graph
3253
self._revision_id_graph = {}
3255
self._heads_provider = None
3257
self._nodes_to_keep_annotations = set()
3258
self._generations_until_keep = 100
3260
def set_generations_until_keep(self, value):
3261
"""Set the number of generations before caching a node.
3263
Setting this to -1 will cache every merge node, setting this higher
3264
will cache fewer nodes.
3266
self._generations_until_keep = value
3268
def _add_fulltext_content(self, revision_id, content_obj):
3269
self._fulltext_contents[revision_id] = content_obj
3270
# TODO: jam 20080305 It might be good to check the sha1digest here
3271
return content_obj.text()
3273
def _check_parents(self, child, nodes_to_annotate):
3274
"""Check if all parents have been processed.
3276
:param child: A tuple of (rev_id, parents, raw_content)
3277
:param nodes_to_annotate: If child is ready, add it to
3278
nodes_to_annotate, otherwise put it back in self._pending_children
3280
for parent_id in child[1]:
3281
if (parent_id not in self._annotated_lines):
3282
# This parent is present, but another parent is missing
3283
self._pending_children.setdefault(parent_id,
3287
# This one is ready to be processed
3288
nodes_to_annotate.append(child)
3290
def _add_annotation(self, revision_id, fulltext, parent_ids,
3291
left_matching_blocks=None):
3292
"""Add an annotation entry.
3294
All parents should already have been annotated.
3295
:return: A list of children that now have their parents satisfied.
3297
a = self._annotated_lines
3298
annotated_parent_lines = [a[p] for p in parent_ids]
3299
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
3300
fulltext, revision_id, left_matching_blocks,
3301
heads_provider=self._get_heads_provider()))
3302
self._annotated_lines[revision_id] = annotated_lines
3303
for p in parent_ids:
3304
ann_children = self._annotate_children[p]
3305
ann_children.remove(revision_id)
3306
if (not ann_children
3307
and p not in self._nodes_to_keep_annotations):
3308
del self._annotated_lines[p]
3309
del self._all_build_details[p]
3310
if p in self._fulltext_contents:
3311
del self._fulltext_contents[p]
3312
# Now that we've added this one, see if there are any pending
3313
# deltas to be done, certainly this parent is finished
3314
nodes_to_annotate = []
3315
for child in self._pending_children.pop(revision_id, []):
3316
self._check_parents(child, nodes_to_annotate)
3317
return nodes_to_annotate
3319
def _get_build_graph(self, key):
3320
"""Get the graphs for building texts and annotations.
3322
The data you need for creating a full text may be different than the
3323
data you need to annotate that text. (At a minimum, you need both
3324
parents to create an annotation, but only need 1 parent to generate the
3327
:return: A list of (key, index_memo) records, suitable for
3328
passing to read_records_iter to start reading in the raw data fro/
3331
if key in self._annotated_lines:
3334
pending = set([key])
3339
# get all pending nodes
3341
this_iteration = pending
3342
build_details = self._knit._index.get_build_details(this_iteration)
3343
self._all_build_details.update(build_details)
3344
# new_nodes = self._knit._index._get_entries(this_iteration)
3346
for key, details in build_details.iteritems():
3347
(index_memo, compression_parent, parents,
3348
record_details) = details
3349
self._revision_id_graph[key] = parents
3350
records.append((key, index_memo))
3351
# Do we actually need to check _annotated_lines?
3352
pending.update(p for p in parents
3353
if p not in self._all_build_details)
3354
if compression_parent:
3355
self._compression_children.setdefault(compression_parent,
3358
for parent in parents:
3359
self._annotate_children.setdefault(parent,
3361
num_gens = generation - kept_generation
3362
if ((num_gens >= self._generations_until_keep)
3363
and len(parents) > 1):
3364
kept_generation = generation
3365
self._nodes_to_keep_annotations.add(key)
3367
missing_versions = this_iteration.difference(build_details.keys())
3368
self._ghosts.update(missing_versions)
3369
for missing_version in missing_versions:
3370
# add a key, no parents
3371
self._revision_id_graph[missing_version] = ()
3372
pending.discard(missing_version) # don't look for it
3373
if self._ghosts.intersection(self._compression_children):
3375
"We cannot have nodes which have a ghost compression parent:\n"
3377
"compression children: %r"
3378
% (self._ghosts, self._compression_children))
3379
# Cleanout anything that depends on a ghost so that we don't wait for
3380
# the ghost to show up
3381
for node in self._ghosts:
3382
if node in self._annotate_children:
3383
# We won't be building this node
3384
del self._annotate_children[node]
3385
# Generally we will want to read the records in reverse order, because
3386
# we find the parent nodes after the children
3390
def _annotate_records(self, records):
3391
"""Build the annotations for the listed records."""
3392
# We iterate in the order read, rather than a strict order requested
3393
# However, process what we can, and put off to the side things that
3394
# still need parents, cleaning them up when those parents are
3396
for (rev_id, record,
3397
digest) in self._knit._read_records_iter(records):
3398
if rev_id in self._annotated_lines:
3400
parent_ids = self._revision_id_graph[rev_id]
3401
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3402
details = self._all_build_details[rev_id]
3403
(index_memo, compression_parent, parents,
3404
record_details) = details
3405
nodes_to_annotate = []
3406
# TODO: Remove the punning between compression parents, and
3407
# parent_ids, we should be able to do this without assuming
3409
if len(parent_ids) == 0:
3410
# There are no parents for this node, so just add it
3411
# TODO: This probably needs to be decoupled
3412
fulltext_content, delta = self._knit._factory.parse_record(
3413
rev_id, record, record_details, None)
3414
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3415
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3416
parent_ids, left_matching_blocks=None))
3418
child = (rev_id, parent_ids, record)
3419
# Check if all the parents are present
3420
self._check_parents(child, nodes_to_annotate)
3421
while nodes_to_annotate:
3422
# Should we use a queue here instead of a stack?
3423
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3424
(index_memo, compression_parent, parents,
3425
record_details) = self._all_build_details[rev_id]
3427
if compression_parent is not None:
3428
comp_children = self._compression_children[compression_parent]
3429
if rev_id not in comp_children:
3430
raise AssertionError("%r not in compression children %r"
3431
% (rev_id, comp_children))
3432
# If there is only 1 child, it is safe to reuse this
3434
reuse_content = (len(comp_children) == 1
3435
and compression_parent not in
3436
self._nodes_to_keep_annotations)
3438
# Remove it from the cache since it will be changing
3439
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3440
# Make sure to copy the fulltext since it might be
3442
parent_fulltext = list(parent_fulltext_content.text())
3444
parent_fulltext_content = self._fulltext_contents[compression_parent]
3445
parent_fulltext = parent_fulltext_content.text()
3446
comp_children.remove(rev_id)
3447
fulltext_content, delta = self._knit._factory.parse_record(
3448
rev_id, record, record_details,
3449
parent_fulltext_content,
3450
copy_base_content=(not reuse_content))
3451
fulltext = self._add_fulltext_content(rev_id,
3453
if compression_parent == parent_ids[0]:
3454
# the compression_parent is the left parent, so we can
3456
blocks = KnitContent.get_line_delta_blocks(delta,
3457
parent_fulltext, fulltext)
3459
fulltext_content = self._knit._factory.parse_fulltext(
3461
fulltext = self._add_fulltext_content(rev_id,
3463
nodes_to_annotate.extend(
3464
self._add_annotation(rev_id, fulltext, parent_ids,
3465
left_matching_blocks=blocks))
3467
def _get_heads_provider(self):
3468
"""Create a heads provider for resolving ancestry issues."""
3469
if self._heads_provider is not None:
3470
return self._heads_provider
3471
parent_provider = _mod_graph.DictParentsProvider(
3472
self._revision_id_graph)
3473
graph_obj = _mod_graph.Graph(parent_provider)
3474
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3475
self._heads_provider = head_cache
3478
def annotate(self, key):
3479
"""Return the annotated fulltext at the given key.
3481
:param key: The key to annotate.
3483
if len(self._knit._fallback_vfs) > 0:
3484
# stacked knits can't use the fast path at present.
3485
return self._simple_annotate(key)
3488
records = self._get_build_graph(key)
3489
if key in self._ghosts:
3490
raise errors.RevisionNotPresent(key, self._knit)
3491
self._annotate_records(records)
3492
return self._annotated_lines[key]
3493
except errors.RetryWithNewPacks, e:
3494
self._knit._access.reload_or_raise(e)
3495
# The cached build_details are no longer valid
3496
self._all_build_details.clear()
3498
def _simple_annotate(self, key):
3499
"""Return annotated fulltext, rediffing from the full texts.
3501
This is slow but makes no assumptions about the repository
3502
being able to produce line deltas.
3504
# TODO: this code generates a parent maps of present ancestors; it
3505
# could be split out into a separate method, and probably should use
3506
# iter_ancestry instead. -- mbp and robertc 20080704
3507
graph = _mod_graph.Graph(self._knit)
3508
head_cache = _mod_graph.FrozenHeadsCache(graph)
3509
search = graph._make_breadth_first_searcher([key])
3513
present, ghosts = search.next_with_ghosts()
3514
except StopIteration:
3516
keys.update(present)
3517
parent_map = self._knit.get_parent_map(keys)
3519
reannotate = annotate.reannotate
3520
for record in self._knit.get_record_stream(keys, 'topological', True):
3522
fulltext = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
3523
parents = parent_map[key]
3524
if parents is not None:
3525
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
3528
parent_cache[key] = list(
3529
reannotate(parent_lines, fulltext, key, None, head_cache))
3531
return parent_cache[key]
3533
raise errors.RevisionNotPresent(key, self._knit)
3537
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3539
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
