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'
958
for element in key[:-1]:
959
if type(element) != str:
960
raise TypeError("key contains non-strings: %r" % (key,))
962
key = key[:-1] + ('sha1:' + digest,)
963
elif type(key[-1]) != str:
964
raise TypeError("key contains non-strings: %r" % (key,))
965
# Knit hunks are still last-element only
967
content = self._factory.make(lines, version_id)
968
if 'no-eol' in options:
969
# Hint to the content object that its text() call should strip the
971
content._should_strip_eol = True
972
if delta or (self._factory.annotated and len(present_parents) > 0):
973
# Merge annotations from parent texts if needed.
974
delta_hunks = self._merge_annotations(content, present_parents,
975
parent_texts, delta, self._factory.annotated,
976
left_matching_blocks)
979
options.append('line-delta')
980
store_lines = self._factory.lower_line_delta(delta_hunks)
981
size, bytes = self._record_to_data(key, digest,
984
options.append('fulltext')
985
# isinstance is slower and we have no hierarchy.
986
if self._factory.__class__ is KnitPlainFactory:
987
# Use the already joined bytes saving iteration time in
989
size, bytes = self._record_to_data(key, digest,
992
# get mixed annotation + content and feed it into the
994
store_lines = self._factory.lower_fulltext(content)
995
size, bytes = self._record_to_data(key, digest,
998
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
999
self._index.add_records(
1000
((key, options, access_memo, parents),),
1001
random_id=random_id)
1002
return digest, text_length, content
1004
def annotate(self, key):
1005
"""See VersionedFiles.annotate."""
1006
return self._factory.annotate(self, key)
1008
def check(self, progress_bar=None):
1009
"""See VersionedFiles.check()."""
1010
# This doesn't actually test extraction of everything, but that will
1011
# impact 'bzr check' substantially, and needs to be integrated with
1012
# care. However, it does check for the obvious problem of a delta with
1014
keys = self._index.keys()
1015
parent_map = self.get_parent_map(keys)
1017
if self._index.get_method(key) != 'fulltext':
1018
compression_parent = parent_map[key][0]
1019
if compression_parent not in parent_map:
1020
raise errors.KnitCorrupt(self,
1021
"Missing basis parent %s for %s" % (
1022
compression_parent, key))
1023
for fallback_vfs in self._fallback_vfs:
1024
fallback_vfs.check()
1026
def _check_add(self, key, lines, random_id, check_content):
1027
"""check that version_id and lines are safe to add."""
1028
version_id = key[-1]
1029
if version_id is not None:
1030
if contains_whitespace(version_id):
1031
raise InvalidRevisionId(version_id, self)
1032
self.check_not_reserved_id(version_id)
1033
# TODO: If random_id==False and the key is already present, we should
1034
# probably check that the existing content is identical to what is
1035
# being inserted, and otherwise raise an exception. This would make
1036
# the bundle code simpler.
1038
self._check_lines_not_unicode(lines)
1039
self._check_lines_are_lines(lines)
1041
def _check_header(self, key, line):
1042
rec = self._split_header(line)
1043
self._check_header_version(rec, key[-1])
1046
def _check_header_version(self, rec, version_id):
1047
"""Checks the header version on original format knit records.
1049
These have the last component of the key embedded in the record.
1051
if rec[1] != version_id:
1052
raise KnitCorrupt(self,
1053
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1055
def _check_should_delta(self, parent):
1056
"""Iterate back through the parent listing, looking for a fulltext.
1058
This is used when we want to decide whether to add a delta or a new
1059
fulltext. It searches for _max_delta_chain parents. When it finds a
1060
fulltext parent, it sees if the total size of the deltas leading up to
1061
it is large enough to indicate that we want a new full text anyway.
1063
Return True if we should create a new delta, False if we should use a
1067
fulltext_size = None
1068
for count in xrange(self._max_delta_chain):
1070
# Note that this only looks in the index of this particular
1071
# KnitVersionedFiles, not in the fallbacks. This ensures that
1072
# we won't store a delta spanning physical repository
1074
build_details = self._index.get_build_details([parent])
1075
parent_details = build_details[parent]
1076
except (RevisionNotPresent, KeyError), e:
1077
# Some basis is not locally present: always fulltext
1079
index_memo, compression_parent, _, _ = parent_details
1080
_, _, size = index_memo
1081
if compression_parent is None:
1082
fulltext_size = size
1085
# We don't explicitly check for presence because this is in an
1086
# inner loop, and if it's missing it'll fail anyhow.
1087
parent = compression_parent
1089
# We couldn't find a fulltext, so we must create a new one
1091
# Simple heuristic - if the total I/O wold be greater as a delta than
1092
# the originally installed fulltext, we create a new fulltext.
1093
return fulltext_size > delta_size
1095
def _build_details_to_components(self, build_details):
1096
"""Convert a build_details tuple to a position tuple."""
1097
# record_details, access_memo, compression_parent
1098
return build_details[3], build_details[0], build_details[1]
1100
def _get_components_positions(self, keys, allow_missing=False):
1101
"""Produce a map of position data for the components of keys.
1103
This data is intended to be used for retrieving the knit records.
1105
A dict of key to (record_details, index_memo, next, parents) is
1107
method is the way referenced data should be applied.
1108
index_memo is the handle to pass to the data access to actually get the
1110
next is the build-parent of the version, or None for fulltexts.
1111
parents is the version_ids of the parents of this version
1113
:param allow_missing: If True do not raise an error on a missing component,
1117
pending_components = keys
1118
while pending_components:
1119
build_details = self._index.get_build_details(pending_components)
1120
current_components = set(pending_components)
1121
pending_components = set()
1122
for key, details in build_details.iteritems():
1123
(index_memo, compression_parent, parents,
1124
record_details) = details
1125
method = record_details[0]
1126
if compression_parent is not None:
1127
pending_components.add(compression_parent)
1128
component_data[key] = self._build_details_to_components(details)
1129
missing = current_components.difference(build_details)
1130
if missing and not allow_missing:
1131
raise errors.RevisionNotPresent(missing.pop(), self)
1132
return component_data
1134
def _get_content(self, key, parent_texts={}):
1135
"""Returns a content object that makes up the specified
1137
cached_version = parent_texts.get(key, None)
1138
if cached_version is not None:
1139
# Ensure the cache dict is valid.
1140
if not self.get_parent_map([key]):
1141
raise RevisionNotPresent(key, self)
1142
return cached_version
1143
generator = _VFContentMapGenerator(self, [key])
1144
return generator._get_content(key)
1146
def get_parent_map(self, keys):
1147
"""Get a map of the graph parents of keys.
1149
:param keys: The keys to look up parents for.
1150
:return: A mapping from keys to parents. Absent keys are absent from
1153
return self._get_parent_map_with_sources(keys)[0]
1155
def _get_parent_map_with_sources(self, keys):
1156
"""Get a map of the parents of keys.
1158
:param keys: The keys to look up parents for.
1159
:return: A tuple. The first element is a mapping from keys to parents.
1160
Absent keys are absent from the mapping. The second element is a
1161
list with the locations each key was found in. The first element
1162
is the in-this-knit parents, the second the first fallback source,
1166
sources = [self._index] + self._fallback_vfs
1169
for source in sources:
1172
new_result = source.get_parent_map(missing)
1173
source_results.append(new_result)
1174
result.update(new_result)
1175
missing.difference_update(set(new_result))
1176
return result, source_results
1178
def _get_record_map(self, keys, allow_missing=False):
1179
"""Produce a dictionary of knit records.
1181
:return: {key:(record, record_details, digest, next)}
1183
data returned from read_records (a KnitContentobject)
1185
opaque information to pass to parse_record
1187
SHA1 digest of the full text after all steps are done
1189
build-parent of the version, i.e. the leftmost ancestor.
1190
Will be None if the record is not a delta.
1191
:param keys: The keys to build a map for
1192
:param allow_missing: If some records are missing, rather than
1193
error, just return the data that could be generated.
1195
raw_map = self._get_record_map_unparsed(keys,
1196
allow_missing=allow_missing)
1197
return self._raw_map_to_record_map(raw_map)
1199
def _raw_map_to_record_map(self, raw_map):
1200
"""Parse the contents of _get_record_map_unparsed.
1202
:return: see _get_record_map.
1206
data, record_details, next = raw_map[key]
1207
content, digest = self._parse_record(key[-1], data)
1208
result[key] = content, record_details, digest, next
1211
def _get_record_map_unparsed(self, keys, allow_missing=False):
1212
"""Get the raw data for reconstructing keys without parsing it.
1214
:return: A dict suitable for parsing via _raw_map_to_record_map.
1215
key-> raw_bytes, (method, noeol), compression_parent
1217
# This retries the whole request if anything fails. Potentially we
1218
# could be a bit more selective. We could track the keys whose records
1219
# we have successfully found, and then only request the new records
1220
# from there. However, _get_components_positions grabs the whole build
1221
# chain, which means we'll likely try to grab the same records again
1222
# anyway. Also, can the build chains change as part of a pack
1223
# operation? We wouldn't want to end up with a broken chain.
1226
position_map = self._get_components_positions(keys,
1227
allow_missing=allow_missing)
1228
# key = component_id, r = record_details, i_m = index_memo,
1230
records = [(key, i_m) for key, (r, i_m, n)
1231
in position_map.iteritems()]
1232
# Sort by the index memo, so that we request records from the
1233
# same pack file together, and in forward-sorted order
1234
records.sort(key=operator.itemgetter(1))
1236
for key, data in self._read_records_iter_unchecked(records):
1237
(record_details, index_memo, next) = position_map[key]
1238
raw_record_map[key] = data, record_details, next
1239
return raw_record_map
1240
except errors.RetryWithNewPacks, e:
1241
self._access.reload_or_raise(e)
1244
def _split_by_prefix(cls, keys):
1245
"""For the given keys, split them up based on their prefix.
1247
To keep memory pressure somewhat under control, split the
1248
requests back into per-file-id requests, otherwise "bzr co"
1249
extracts the full tree into memory before writing it to disk.
1250
This should be revisited if _get_content_maps() can ever cross
1253
The keys for a given file_id are kept in the same relative order.
1254
Ordering between file_ids is not, though prefix_order will return the
1255
order that the key was first seen.
1257
:param keys: An iterable of key tuples
1258
:return: (split_map, prefix_order)
1259
split_map A dictionary mapping prefix => keys
1260
prefix_order The order that we saw the various prefixes
1262
split_by_prefix = {}
1270
if prefix in split_by_prefix:
1271
split_by_prefix[prefix].append(key)
1273
split_by_prefix[prefix] = [key]
1274
prefix_order.append(prefix)
1275
return split_by_prefix, prefix_order
1277
def _group_keys_for_io(self, keys, non_local_keys, positions,
1278
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1279
"""For the given keys, group them into 'best-sized' requests.
1281
The idea is to avoid making 1 request per file, but to never try to
1282
unpack an entire 1.5GB source tree in a single pass. Also when
1283
possible, we should try to group requests to the same pack file
1286
:return: list of (keys, non_local) tuples that indicate what keys
1287
should be fetched next.
1289
# TODO: Ideally we would group on 2 factors. We want to extract texts
1290
# from the same pack file together, and we want to extract all
1291
# the texts for a given build-chain together. Ultimately it
1292
# probably needs a better global view.
1293
total_keys = len(keys)
1294
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1295
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1297
cur_non_local = set()
1301
for prefix in prefix_order:
1302
keys = prefix_split_keys[prefix]
1303
non_local = prefix_split_non_local_keys.get(prefix, [])
1305
this_size = self._index._get_total_build_size(keys, positions)
1306
cur_size += this_size
1307
cur_keys.extend(keys)
1308
cur_non_local.update(non_local)
1309
if cur_size > _min_buffer_size:
1310
result.append((cur_keys, cur_non_local))
1311
sizes.append(cur_size)
1313
cur_non_local = set()
1316
result.append((cur_keys, cur_non_local))
1317
sizes.append(cur_size)
1320
def get_record_stream(self, keys, ordering, include_delta_closure):
1321
"""Get a stream of records for keys.
1323
:param keys: The keys to include.
1324
:param ordering: Either 'unordered' or 'topological'. A topologically
1325
sorted stream has compression parents strictly before their
1327
:param include_delta_closure: If True then the closure across any
1328
compression parents will be included (in the opaque data).
1329
:return: An iterator of ContentFactory objects, each of which is only
1330
valid until the iterator is advanced.
1332
# keys might be a generator
1336
if not self._index.has_graph:
1337
# Cannot sort when no graph has been stored.
1338
ordering = 'unordered'
1340
remaining_keys = keys
1343
keys = set(remaining_keys)
1344
for content_factory in self._get_remaining_record_stream(keys,
1345
ordering, include_delta_closure):
1346
remaining_keys.discard(content_factory.key)
1347
yield content_factory
1349
except errors.RetryWithNewPacks, e:
1350
self._access.reload_or_raise(e)
1352
def _get_remaining_record_stream(self, keys, ordering,
1353
include_delta_closure):
1354
"""This function is the 'retry' portion for get_record_stream."""
1355
if include_delta_closure:
1356
positions = self._get_components_positions(keys, allow_missing=True)
1358
build_details = self._index.get_build_details(keys)
1360
# (record_details, access_memo, compression_parent_key)
1361
positions = dict((key, self._build_details_to_components(details))
1362
for key, details in build_details.iteritems())
1363
absent_keys = keys.difference(set(positions))
1364
# There may be more absent keys : if we're missing the basis component
1365
# and are trying to include the delta closure.
1366
# XXX: We should not ever need to examine remote sources because we do
1367
# not permit deltas across versioned files boundaries.
1368
if include_delta_closure:
1369
needed_from_fallback = set()
1370
# Build up reconstructable_keys dict. key:True in this dict means
1371
# the key can be reconstructed.
1372
reconstructable_keys = {}
1376
chain = [key, positions[key][2]]
1378
needed_from_fallback.add(key)
1381
while chain[-1] is not None:
1382
if chain[-1] in reconstructable_keys:
1383
result = reconstructable_keys[chain[-1]]
1387
chain.append(positions[chain[-1]][2])
1389
# missing basis component
1390
needed_from_fallback.add(chain[-1])
1393
for chain_key in chain[:-1]:
1394
reconstructable_keys[chain_key] = result
1396
needed_from_fallback.add(key)
1397
# Double index lookups here : need a unified api ?
1398
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1399
if ordering in ('topological', 'groupcompress'):
1400
if ordering == 'topological':
1401
# Global topological sort
1402
present_keys = tsort.topo_sort(global_map)
1404
present_keys = sort_groupcompress(global_map)
1405
# Now group by source:
1407
current_source = None
1408
for key in present_keys:
1409
for parent_map in parent_maps:
1410
if key in parent_map:
1411
key_source = parent_map
1413
if current_source is not key_source:
1414
source_keys.append((key_source, []))
1415
current_source = key_source
1416
source_keys[-1][1].append(key)
1418
if ordering != 'unordered':
1419
raise AssertionError('valid values for ordering are:'
1420
' "unordered", "groupcompress" or "topological" not: %r'
1422
# Just group by source; remote sources first.
1425
for parent_map in reversed(parent_maps):
1426
source_keys.append((parent_map, []))
1427
for key in parent_map:
1428
present_keys.append(key)
1429
source_keys[-1][1].append(key)
1430
# We have been requested to return these records in an order that
1431
# suits us. So we ask the index to give us an optimally sorted
1433
for source, sub_keys in source_keys:
1434
if source is parent_maps[0]:
1435
# Only sort the keys for this VF
1436
self._index._sort_keys_by_io(sub_keys, positions)
1437
absent_keys = keys - set(global_map)
1438
for key in absent_keys:
1439
yield AbsentContentFactory(key)
1440
# restrict our view to the keys we can answer.
1441
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1442
# XXX: At that point we need to consider the impact of double reads by
1443
# utilising components multiple times.
1444
if include_delta_closure:
1445
# XXX: get_content_maps performs its own index queries; allow state
1447
non_local_keys = needed_from_fallback - absent_keys
1448
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1451
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1453
for record in generator.get_record_stream():
1456
for source, keys in source_keys:
1457
if source is parent_maps[0]:
1458
# this KnitVersionedFiles
1459
records = [(key, positions[key][1]) for key in keys]
1460
for key, raw_data, sha1 in self._read_records_iter_raw(records):
1461
(record_details, index_memo, _) = positions[key]
1462
yield KnitContentFactory(key, global_map[key],
1463
record_details, sha1, raw_data, self._factory.annotated, None)
1465
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1466
for record in vf.get_record_stream(keys, ordering,
1467
include_delta_closure):
1470
def get_sha1s(self, keys):
1471
"""See VersionedFiles.get_sha1s()."""
1473
record_map = self._get_record_map(missing, allow_missing=True)
1475
for key, details in record_map.iteritems():
1476
if key not in missing:
1478
# record entry 2 is the 'digest'.
1479
result[key] = details[2]
1480
missing.difference_update(set(result))
1481
for source in self._fallback_vfs:
1484
new_result = source.get_sha1s(missing)
1485
result.update(new_result)
1486
missing.difference_update(set(new_result))
1489
def insert_record_stream(self, stream):
1490
"""Insert a record stream into this container.
1492
:param stream: A stream of records to insert.
1494
:seealso VersionedFiles.get_record_stream:
1496
def get_adapter(adapter_key):
1498
return adapters[adapter_key]
1500
adapter_factory = adapter_registry.get(adapter_key)
1501
adapter = adapter_factory(self)
1502
adapters[adapter_key] = adapter
1505
if self._factory.annotated:
1506
# self is annotated, we need annotated knits to use directly.
1507
annotated = "annotated-"
1510
# self is not annotated, but we can strip annotations cheaply.
1512
convertibles = set(["knit-annotated-ft-gz"])
1513
if self._max_delta_chain:
1514
delta_types.add("knit-annotated-delta-gz")
1515
convertibles.add("knit-annotated-delta-gz")
1516
# The set of types we can cheaply adapt without needing basis texts.
1517
native_types = set()
1518
if self._max_delta_chain:
1519
native_types.add("knit-%sdelta-gz" % annotated)
1520
delta_types.add("knit-%sdelta-gz" % annotated)
1521
native_types.add("knit-%sft-gz" % annotated)
1522
knit_types = native_types.union(convertibles)
1524
# Buffer all index entries that we can't add immediately because their
1525
# basis parent is missing. We don't buffer all because generating
1526
# annotations may require access to some of the new records. However we
1527
# can't generate annotations from new deltas until their basis parent
1528
# is present anyway, so we get away with not needing an index that
1529
# includes the new keys.
1531
# See <http://launchpad.net/bugs/300177> about ordering of compression
1532
# parents in the records - to be conservative, we insist that all
1533
# parents must be present to avoid expanding to a fulltext.
1535
# key = basis_parent, value = index entry to add
1536
buffered_index_entries = {}
1537
for record in stream:
1539
parents = record.parents
1540
if record.storage_kind in delta_types:
1541
# TODO: eventually the record itself should track
1542
# compression_parent
1543
compression_parent = parents[0]
1545
compression_parent = None
1546
# Raise an error when a record is missing.
1547
if record.storage_kind == 'absent':
1548
raise RevisionNotPresent([record.key], self)
1549
elif ((record.storage_kind in knit_types)
1550
and (compression_parent is None
1551
or not self._fallback_vfs
1552
or self._index.has_key(compression_parent)
1553
or not self.has_key(compression_parent))):
1554
# we can insert the knit record literally if either it has no
1555
# compression parent OR we already have its basis in this kvf
1556
# OR the basis is not present even in the fallbacks. In the
1557
# last case it will either turn up later in the stream and all
1558
# will be well, or it won't turn up at all and we'll raise an
1561
# TODO: self.has_key is somewhat redundant with
1562
# self._index.has_key; we really want something that directly
1563
# asks if it's only present in the fallbacks. -- mbp 20081119
1564
if record.storage_kind not in native_types:
1566
adapter_key = (record.storage_kind, "knit-delta-gz")
1567
adapter = get_adapter(adapter_key)
1569
adapter_key = (record.storage_kind, "knit-ft-gz")
1570
adapter = get_adapter(adapter_key)
1571
bytes = adapter.get_bytes(record)
1573
# It's a knit record, it has a _raw_record field (even if
1574
# it was reconstituted from a network stream).
1575
bytes = record._raw_record
1576
options = [record._build_details[0]]
1577
if record._build_details[1]:
1578
options.append('no-eol')
1579
# Just blat it across.
1580
# Note: This does end up adding data on duplicate keys. As
1581
# modern repositories use atomic insertions this should not
1582
# lead to excessive growth in the event of interrupted fetches.
1583
# 'knit' repositories may suffer excessive growth, but as a
1584
# deprecated format this is tolerable. It can be fixed if
1585
# needed by in the kndx index support raising on a duplicate
1586
# add with identical parents and options.
1587
access_memo = self._access.add_raw_records(
1588
[(record.key, len(bytes))], bytes)[0]
1589
index_entry = (record.key, options, access_memo, parents)
1590
if 'fulltext' not in options:
1591
# Not a fulltext, so we need to make sure the compression
1592
# parent will also be present.
1593
# Note that pack backed knits don't need to buffer here
1594
# because they buffer all writes to the transaction level,
1595
# but we don't expose that difference at the index level. If
1596
# the query here has sufficient cost to show up in
1597
# profiling we should do that.
1599
# They're required to be physically in this
1600
# KnitVersionedFiles, not in a fallback.
1601
if not self._index.has_key(compression_parent):
1602
pending = buffered_index_entries.setdefault(
1603
compression_parent, [])
1604
pending.append(index_entry)
1607
self._index.add_records([index_entry])
1608
elif record.storage_kind == 'chunked':
1609
self.add_lines(record.key, parents,
1610
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1612
# Not suitable for direct insertion as a
1613
# delta, either because it's not the right format, or this
1614
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1615
# 0) or because it depends on a base only present in the
1617
self._access.flush()
1619
# Try getting a fulltext directly from the record.
1620
bytes = record.get_bytes_as('fulltext')
1621
except errors.UnavailableRepresentation:
1622
adapter_key = record.storage_kind, 'fulltext'
1623
adapter = get_adapter(adapter_key)
1624
bytes = adapter.get_bytes(record)
1625
lines = split_lines(bytes)
1627
self.add_lines(record.key, parents, lines)
1628
except errors.RevisionAlreadyPresent:
1630
# Add any records whose basis parent is now available.
1632
added_keys = [record.key]
1634
key = added_keys.pop(0)
1635
if key in buffered_index_entries:
1636
index_entries = buffered_index_entries[key]
1637
self._index.add_records(index_entries)
1639
[index_entry[0] for index_entry in index_entries])
1640
del buffered_index_entries[key]
1641
if buffered_index_entries:
1642
# There were index entries buffered at the end of the stream,
1643
# So these need to be added (if the index supports holding such
1644
# entries for later insertion)
1645
for key in buffered_index_entries:
1646
index_entries = buffered_index_entries[key]
1647
self._index.add_records(index_entries,
1648
missing_compression_parents=True)
1650
def get_missing_compression_parent_keys(self):
1651
"""Return an iterable of keys of missing compression parents.
1653
Check this after calling insert_record_stream to find out if there are
1654
any missing compression parents. If there are, the records that
1655
depend on them are not able to be inserted safely. For atomic
1656
KnitVersionedFiles built on packs, the transaction should be aborted or
1657
suspended - commit will fail at this point. Nonatomic knits will error
1658
earlier because they have no staging area to put pending entries into.
1660
return self._index.get_missing_compression_parents()
1662
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1663
"""Iterate over the lines in the versioned files from keys.
1665
This may return lines from other keys. Each item the returned
1666
iterator yields is a tuple of a line and a text version that that line
1667
is present in (not introduced in).
1669
Ordering of results is in whatever order is most suitable for the
1670
underlying storage format.
1672
If a progress bar is supplied, it may be used to indicate progress.
1673
The caller is responsible for cleaning up progress bars (because this
1677
* Lines are normalised by the underlying store: they will all have \\n
1679
* Lines are returned in arbitrary order.
1680
* If a requested key did not change any lines (or didn't have any
1681
lines), it may not be mentioned at all in the result.
1683
:param pb: Progress bar supplied by caller.
1684
:return: An iterator over (line, key).
1687
pb = progress.DummyProgress()
1693
# we don't care about inclusions, the caller cares.
1694
# but we need to setup a list of records to visit.
1695
# we need key, position, length
1697
build_details = self._index.get_build_details(keys)
1698
for key, details in build_details.iteritems():
1700
key_records.append((key, details[0]))
1701
records_iter = enumerate(self._read_records_iter(key_records))
1702
for (key_idx, (key, data, sha_value)) in records_iter:
1703
pb.update('Walking content', key_idx, total)
1704
compression_parent = build_details[key][1]
1705
if compression_parent is None:
1707
line_iterator = self._factory.get_fulltext_content(data)
1710
line_iterator = self._factory.get_linedelta_content(data)
1711
# Now that we are yielding the data for this key, remove it
1714
# XXX: It might be more efficient to yield (key,
1715
# line_iterator) in the future. However for now, this is a
1716
# simpler change to integrate into the rest of the
1717
# codebase. RBC 20071110
1718
for line in line_iterator:
1721
except errors.RetryWithNewPacks, e:
1722
self._access.reload_or_raise(e)
1723
# If there are still keys we've not yet found, we look in the fallback
1724
# vfs, and hope to find them there. Note that if the keys are found
1725
# but had no changes or no content, the fallback may not return
1727
if keys and not self._fallback_vfs:
1728
# XXX: strictly the second parameter is meant to be the file id
1729
# but it's not easily accessible here.
1730
raise RevisionNotPresent(keys, repr(self))
1731
for source in self._fallback_vfs:
1735
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1736
source_keys.add(key)
1738
keys.difference_update(source_keys)
1739
pb.update('Walking content', total, total)
1741
def _make_line_delta(self, delta_seq, new_content):
1742
"""Generate a line delta from delta_seq and new_content."""
1744
for op in delta_seq.get_opcodes():
1745
if op[0] == 'equal':
1747
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1750
def _merge_annotations(self, content, parents, parent_texts={},
1751
delta=None, annotated=None,
1752
left_matching_blocks=None):
1753
"""Merge annotations for content and generate deltas.
1755
This is done by comparing the annotations based on changes to the text
1756
and generating a delta on the resulting full texts. If annotations are
1757
not being created then a simple delta is created.
1759
if left_matching_blocks is not None:
1760
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1764
for parent_key in parents:
1765
merge_content = self._get_content(parent_key, parent_texts)
1766
if (parent_key == parents[0] and delta_seq is not None):
1769
seq = patiencediff.PatienceSequenceMatcher(
1770
None, merge_content.text(), content.text())
1771
for i, j, n in seq.get_matching_blocks():
1774
# this copies (origin, text) pairs across to the new
1775
# content for any line that matches the last-checked
1777
content._lines[j:j+n] = merge_content._lines[i:i+n]
1778
# XXX: Robert says the following block is a workaround for a
1779
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1780
if content._lines and content._lines[-1][1][-1] != '\n':
1781
# The copied annotation was from a line without a trailing EOL,
1782
# reinstate one for the content object, to ensure correct
1784
line = content._lines[-1][1] + '\n'
1785
content._lines[-1] = (content._lines[-1][0], line)
1787
if delta_seq is None:
1788
reference_content = self._get_content(parents[0], parent_texts)
1789
new_texts = content.text()
1790
old_texts = reference_content.text()
1791
delta_seq = patiencediff.PatienceSequenceMatcher(
1792
None, old_texts, new_texts)
1793
return self._make_line_delta(delta_seq, content)
1795
def _parse_record(self, version_id, data):
1796
"""Parse an original format knit record.
1798
These have the last element of the key only present in the stored data.
1800
rec, record_contents = self._parse_record_unchecked(data)
1801
self._check_header_version(rec, version_id)
1802
return record_contents, rec[3]
1804
def _parse_record_header(self, key, raw_data):
1805
"""Parse a record header for consistency.
1807
:return: the header and the decompressor stream.
1808
as (stream, header_record)
1810
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1813
rec = self._check_header(key, df.readline())
1814
except Exception, e:
1815
raise KnitCorrupt(self,
1816
"While reading {%s} got %s(%s)"
1817
% (key, e.__class__.__name__, str(e)))
1820
def _parse_record_unchecked(self, data):
1822
# 4168 calls in 2880 217 internal
1823
# 4168 calls to _parse_record_header in 2121
1824
# 4168 calls to readlines in 330
1825
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1827
record_contents = df.readlines()
1828
except Exception, e:
1829
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1830
(data, e.__class__.__name__, str(e)))
1831
header = record_contents.pop(0)
1832
rec = self._split_header(header)
1833
last_line = record_contents.pop()
1834
if len(record_contents) != int(rec[2]):
1835
raise KnitCorrupt(self,
1836
'incorrect number of lines %s != %s'
1837
' for version {%s} %s'
1838
% (len(record_contents), int(rec[2]),
1839
rec[1], record_contents))
1840
if last_line != 'end %s\n' % rec[1]:
1841
raise KnitCorrupt(self,
1842
'unexpected version end line %r, wanted %r'
1843
% (last_line, rec[1]))
1845
return rec, record_contents
1847
def _read_records_iter(self, records):
1848
"""Read text records from data file and yield result.
1850
The result will be returned in whatever is the fastest to read.
1851
Not by the order requested. Also, multiple requests for the same
1852
record will only yield 1 response.
1853
:param records: A list of (key, access_memo) entries
1854
:return: Yields (key, contents, digest) in the order
1855
read, not the order requested
1860
# XXX: This smells wrong, IO may not be getting ordered right.
1861
needed_records = sorted(set(records), key=operator.itemgetter(1))
1862
if not needed_records:
1865
# The transport optimizes the fetching as well
1866
# (ie, reads continuous ranges.)
1867
raw_data = self._access.get_raw_records(
1868
[index_memo for key, index_memo in needed_records])
1870
for (key, index_memo), data in \
1871
izip(iter(needed_records), raw_data):
1872
content, digest = self._parse_record(key[-1], data)
1873
yield key, content, digest
1875
def _read_records_iter_raw(self, records):
1876
"""Read text records from data file and yield raw data.
1878
This unpacks enough of the text record to validate the id is
1879
as expected but thats all.
1881
Each item the iterator yields is (key, bytes,
1882
expected_sha1_of_full_text).
1884
for key, data in self._read_records_iter_unchecked(records):
1885
# validate the header (note that we can only use the suffix in
1886
# current knit records).
1887
df, rec = self._parse_record_header(key, data)
1889
yield key, data, rec[3]
1891
def _read_records_iter_unchecked(self, records):
1892
"""Read text records from data file and yield raw data.
1894
No validation is done.
1896
Yields tuples of (key, data).
1898
# setup an iterator of the external records:
1899
# uses readv so nice and fast we hope.
1901
# grab the disk data needed.
1902
needed_offsets = [index_memo for key, index_memo
1904
raw_records = self._access.get_raw_records(needed_offsets)
1906
for key, index_memo in records:
1907
data = raw_records.next()
1910
def _record_to_data(self, key, digest, lines, dense_lines=None):
1911
"""Convert key, digest, lines into a raw data block.
1913
:param key: The key of the record. Currently keys are always serialised
1914
using just the trailing component.
1915
:param dense_lines: The bytes of lines but in a denser form. For
1916
instance, if lines is a list of 1000 bytestrings each ending in \n,
1917
dense_lines may be a list with one line in it, containing all the
1918
1000's lines and their \n's. Using dense_lines if it is already
1919
known is a win because the string join to create bytes in this
1920
function spends less time resizing the final string.
1921
:return: (len, a StringIO instance with the raw data ready to read.)
1923
# Note: using a string copy here increases memory pressure with e.g.
1924
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
1925
# when doing the initial commit of a mozilla tree. RBC 20070921
1926
bytes = ''.join(chain(
1927
["version %s %d %s\n" % (key[-1],
1930
dense_lines or lines,
1931
["end %s\n" % key[-1]]))
1932
if type(bytes) != str:
1933
raise AssertionError(
1934
'data must be plain bytes was %s' % type(bytes))
1935
if lines and lines[-1][-1] != '\n':
1936
raise ValueError('corrupt lines value %r' % lines)
1937
compressed_bytes = tuned_gzip.bytes_to_gzip(bytes)
1938
return len(compressed_bytes), compressed_bytes
1940
def _split_header(self, line):
1943
raise KnitCorrupt(self,
1944
'unexpected number of elements in record header')
1948
"""See VersionedFiles.keys."""
1949
if 'evil' in debug.debug_flags:
1950
trace.mutter_callsite(2, "keys scales with size of history")
1951
sources = [self._index] + self._fallback_vfs
1953
for source in sources:
1954
result.update(source.keys())
1958
class _ContentMapGenerator(object):
1959
"""Generate texts or expose raw deltas for a set of texts."""
1961
def _get_content(self, key):
1962
"""Get the content object for key."""
1963
# Note that _get_content is only called when the _ContentMapGenerator
1964
# has been constructed with just one key requested for reconstruction.
1965
if key in self.nonlocal_keys:
1966
record = self.get_record_stream().next()
1967
# Create a content object on the fly
1968
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
1969
return PlainKnitContent(lines, record.key)
1971
# local keys we can ask for directly
1972
return self._get_one_work(key)
1974
def get_record_stream(self):
1975
"""Get a record stream for the keys requested during __init__."""
1976
for record in self._work():
1980
"""Produce maps of text and KnitContents as dicts.
1982
:return: (text_map, content_map) where text_map contains the texts for
1983
the requested versions and content_map contains the KnitContents.
1985
# NB: By definition we never need to read remote sources unless texts
1986
# are requested from them: we don't delta across stores - and we
1987
# explicitly do not want to to prevent data loss situations.
1988
if self.global_map is None:
1989
self.global_map = self.vf.get_parent_map(self.keys)
1990
nonlocal_keys = self.nonlocal_keys
1992
missing_keys = set(nonlocal_keys)
1993
# Read from remote versioned file instances and provide to our caller.
1994
for source in self.vf._fallback_vfs:
1995
if not missing_keys:
1997
# Loop over fallback repositories asking them for texts - ignore
1998
# any missing from a particular fallback.
1999
for record in source.get_record_stream(missing_keys,
2001
if record.storage_kind == 'absent':
2002
# Not in thie particular stream, may be in one of the
2003
# other fallback vfs objects.
2005
missing_keys.remove(record.key)
2008
self._raw_record_map = self.vf._get_record_map_unparsed(self.keys,
2011
for key in self.keys:
2012
if key in self.nonlocal_keys:
2014
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2017
def _get_one_work(self, requested_key):
2018
# Now, if we have calculated everything already, just return the
2020
if requested_key in self._contents_map:
2021
return self._contents_map[requested_key]
2022
# To simplify things, parse everything at once - code that wants one text
2023
# probably wants them all.
2024
# FUTURE: This function could be improved for the 'extract many' case
2025
# by tracking each component and only doing the copy when the number of
2026
# children than need to apply delta's to it is > 1 or it is part of the
2028
multiple_versions = len(self.keys) != 1
2029
if self._record_map is None:
2030
self._record_map = self.vf._raw_map_to_record_map(
2031
self._raw_record_map)
2032
record_map = self._record_map
2033
# raw_record_map is key:
2034
# Have read and parsed records at this point.
2035
for key in self.keys:
2036
if key in self.nonlocal_keys:
2041
while cursor is not None:
2043
record, record_details, digest, next = record_map[cursor]
2045
raise RevisionNotPresent(cursor, self)
2046
components.append((cursor, record, record_details, digest))
2048
if cursor in self._contents_map:
2049
# no need to plan further back
2050
components.append((cursor, None, None, None))
2054
for (component_id, record, record_details,
2055
digest) in reversed(components):
2056
if component_id in self._contents_map:
2057
content = self._contents_map[component_id]
2059
content, delta = self._factory.parse_record(key[-1],
2060
record, record_details, content,
2061
copy_base_content=multiple_versions)
2062
if multiple_versions:
2063
self._contents_map[component_id] = content
2065
# digest here is the digest from the last applied component.
2066
text = content.text()
2067
actual_sha = sha_strings(text)
2068
if actual_sha != digest:
2069
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2070
if multiple_versions:
2071
return self._contents_map[requested_key]
2075
def _wire_bytes(self):
2076
"""Get the bytes to put on the wire for 'key'.
2078
The first collection of bytes asked for returns the serialised
2079
raw_record_map and the additional details (key, parent) for key.
2080
Subsequent calls return just the additional details (key, parent).
2081
The wire storage_kind given for the first key is 'knit-delta-closure',
2082
For subsequent keys it is 'knit-delta-closure-ref'.
2084
:param key: A key from the content generator.
2085
:return: Bytes to put on the wire.
2088
# kind marker for dispatch on the far side,
2089
lines.append('knit-delta-closure')
2091
if self.vf._factory.annotated:
2092
lines.append('annotated')
2095
# then the list of keys
2096
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2097
if key not in self.nonlocal_keys]))
2098
# then the _raw_record_map in serialised form:
2100
# for each item in the map:
2102
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2103
# one line with method
2104
# one line with noeol
2105
# one line with next ('' for None)
2106
# one line with byte count of the record bytes
2108
for key, (record_bytes, (method, noeol), next) in \
2109
self._raw_record_map.iteritems():
2110
key_bytes = '\x00'.join(key)
2111
parents = self.global_map.get(key, None)
2113
parent_bytes = 'None:'
2115
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2116
method_bytes = method
2122
next_bytes = '\x00'.join(next)
2125
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2126
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2127
len(record_bytes), record_bytes))
2128
map_bytes = ''.join(map_byte_list)
2129
lines.append(map_bytes)
2130
bytes = '\n'.join(lines)
2134
class _VFContentMapGenerator(_ContentMapGenerator):
2135
"""Content map generator reading from a VersionedFiles object."""
2137
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2138
global_map=None, raw_record_map=None):
2139
"""Create a _ContentMapGenerator.
2141
:param versioned_files: The versioned files that the texts are being
2143
:param keys: The keys to produce content maps for.
2144
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2145
which are known to not be in this knit, but rather in one of the
2147
:param global_map: The result of get_parent_map(keys) (or a supermap).
2148
This is required if get_record_stream() is to be used.
2149
:param raw_record_map: A unparsed raw record map to use for answering
2152
# The vf to source data from
2153
self.vf = versioned_files
2155
self.keys = list(keys)
2156
# Keys known to be in fallback vfs objects
2157
if nonlocal_keys is None:
2158
self.nonlocal_keys = set()
2160
self.nonlocal_keys = frozenset(nonlocal_keys)
2161
# Parents data for keys to be returned in get_record_stream
2162
self.global_map = global_map
2163
# The chunked lists for self.keys in text form
2165
# A cache of KnitContent objects used in extracting texts.
2166
self._contents_map = {}
2167
# All the knit records needed to assemble the requested keys as full
2169
self._record_map = None
2170
if raw_record_map is None:
2171
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2174
self._raw_record_map = raw_record_map
2175
# the factory for parsing records
2176
self._factory = self.vf._factory
2179
class _NetworkContentMapGenerator(_ContentMapGenerator):
2180
"""Content map generator sourced from a network stream."""
2182
def __init__(self, bytes, line_end):
2183
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2185
self.global_map = {}
2186
self._raw_record_map = {}
2187
self._contents_map = {}
2188
self._record_map = None
2189
self.nonlocal_keys = []
2190
# Get access to record parsing facilities
2191
self.vf = KnitVersionedFiles(None, None)
2194
line_end = bytes.find('\n', start)
2195
line = bytes[start:line_end]
2196
start = line_end + 1
2197
if line == 'annotated':
2198
self._factory = KnitAnnotateFactory()
2200
self._factory = KnitPlainFactory()
2201
# list of keys to emit in get_record_stream
2202
line_end = bytes.find('\n', start)
2203
line = bytes[start:line_end]
2204
start = line_end + 1
2206
tuple(segment.split('\x00')) for segment in line.split('\t')
2208
# now a loop until the end. XXX: It would be nice if this was just a
2209
# bunch of the same records as get_record_stream(..., False) gives, but
2210
# there is a decent sized gap stopping that at the moment.
2214
line_end = bytes.find('\n', start)
2215
key = tuple(bytes[start:line_end].split('\x00'))
2216
start = line_end + 1
2217
# 1 line with parents (None: for None, '' for ())
2218
line_end = bytes.find('\n', start)
2219
line = bytes[start:line_end]
2224
[tuple(segment.split('\x00')) for segment in line.split('\t')
2226
self.global_map[key] = parents
2227
start = line_end + 1
2228
# one line with method
2229
line_end = bytes.find('\n', start)
2230
line = bytes[start:line_end]
2232
start = line_end + 1
2233
# one line with noeol
2234
line_end = bytes.find('\n', start)
2235
line = bytes[start:line_end]
2237
start = line_end + 1
2238
# one line with next ('' for None)
2239
line_end = bytes.find('\n', start)
2240
line = bytes[start:line_end]
2244
next = tuple(bytes[start:line_end].split('\x00'))
2245
start = line_end + 1
2246
# one line with byte count of the record bytes
2247
line_end = bytes.find('\n', start)
2248
line = bytes[start:line_end]
2250
start = line_end + 1
2252
record_bytes = bytes[start:start+count]
2253
start = start + count
2255
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2257
def get_record_stream(self):
2258
"""Get a record stream for for keys requested by the bytestream."""
2260
for key in self.keys:
2261
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2264
def _wire_bytes(self):
2268
class _KndxIndex(object):
2269
"""Manages knit index files
2271
The index is kept in memory and read on startup, to enable
2272
fast lookups of revision information. The cursor of the index
2273
file is always pointing to the end, making it easy to append
2276
_cache is a cache for fast mapping from version id to a Index
2279
_history is a cache for fast mapping from indexes to version ids.
2281
The index data format is dictionary compressed when it comes to
2282
parent references; a index entry may only have parents that with a
2283
lover index number. As a result, the index is topological sorted.
2285
Duplicate entries may be written to the index for a single version id
2286
if this is done then the latter one completely replaces the former:
2287
this allows updates to correct version and parent information.
2288
Note that the two entries may share the delta, and that successive
2289
annotations and references MUST point to the first entry.
2291
The index file on disc contains a header, followed by one line per knit
2292
record. The same revision can be present in an index file more than once.
2293
The first occurrence gets assigned a sequence number starting from 0.
2295
The format of a single line is
2296
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
2297
REVISION_ID is a utf8-encoded revision id
2298
FLAGS is a comma separated list of flags about the record. Values include
2299
no-eol, line-delta, fulltext.
2300
BYTE_OFFSET is the ascii representation of the byte offset in the data file
2301
that the the compressed data starts at.
2302
LENGTH is the ascii representation of the length of the data file.
2303
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
2305
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
2306
revision id already in the knit that is a parent of REVISION_ID.
2307
The ' :' marker is the end of record marker.
2310
when a write is interrupted to the index file, it will result in a line
2311
that does not end in ' :'. If the ' :' is not present at the end of a line,
2312
or at the end of the file, then the record that is missing it will be
2313
ignored by the parser.
2315
When writing new records to the index file, the data is preceded by '\n'
2316
to ensure that records always start on new lines even if the last write was
2317
interrupted. As a result its normal for the last line in the index to be
2318
missing a trailing newline. One can be added with no harmful effects.
2320
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2321
where prefix is e.g. the (fileid,) for .texts instances or () for
2322
constant-mapped things like .revisions, and the old state is
2323
tuple(history_vector, cache_dict). This is used to prevent having an
2324
ABI change with the C extension that reads .kndx files.
2327
HEADER = "# bzr knit index 8\n"
2329
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2330
"""Create a _KndxIndex on transport using mapper."""
2331
self._transport = transport
2332
self._mapper = mapper
2333
self._get_scope = get_scope
2334
self._allow_writes = allow_writes
2335
self._is_locked = is_locked
2337
self.has_graph = True
2339
def add_records(self, records, random_id=False, missing_compression_parents=False):
2340
"""Add multiple records to the index.
2342
:param records: a list of tuples:
2343
(key, options, access_memo, parents).
2344
:param random_id: If True the ids being added were randomly generated
2345
and no check for existence will be performed.
2346
:param missing_compression_parents: If True the records being added are
2347
only compressed against texts already in the index (or inside
2348
records). If False the records all refer to unavailable texts (or
2349
texts inside records) as compression parents.
2351
if missing_compression_parents:
2352
# It might be nice to get the edge of the records. But keys isn't
2354
keys = sorted(record[0] for record in records)
2355
raise errors.RevisionNotPresent(keys, self)
2357
for record in records:
2360
path = self._mapper.map(key) + '.kndx'
2361
path_keys = paths.setdefault(path, (prefix, []))
2362
path_keys[1].append(record)
2363
for path in sorted(paths):
2364
prefix, path_keys = paths[path]
2365
self._load_prefixes([prefix])
2367
orig_history = self._kndx_cache[prefix][1][:]
2368
orig_cache = self._kndx_cache[prefix][0].copy()
2371
for key, options, (_, pos, size), parents in path_keys:
2373
# kndx indices cannot be parentless.
2375
line = "\n%s %s %s %s %s :" % (
2376
key[-1], ','.join(options), pos, size,
2377
self._dictionary_compress(parents))
2378
if type(line) != str:
2379
raise AssertionError(
2380
'data must be utf8 was %s' % type(line))
2382
self._cache_key(key, options, pos, size, parents)
2383
if len(orig_history):
2384
self._transport.append_bytes(path, ''.join(lines))
2386
self._init_index(path, lines)
2388
# If any problems happen, restore the original values and re-raise
2389
self._kndx_cache[prefix] = (orig_cache, orig_history)
2392
def scan_unvalidated_index(self, graph_index):
2393
"""See _KnitGraphIndex.scan_unvalidated_index."""
2394
# Because kndx files do not support atomic insertion via separate index
2395
# files, they do not support this method.
2396
raise NotImplementedError(self.scan_unvalidated_index)
2398
def get_missing_compression_parents(self):
2399
"""See _KnitGraphIndex.get_missing_compression_parents."""
2400
# Because kndx files do not support atomic insertion via separate index
2401
# files, they do not support this method.
2402
raise NotImplementedError(self.get_missing_compression_parents)
2404
def _cache_key(self, key, options, pos, size, parent_keys):
2405
"""Cache a version record in the history array and index cache.
2407
This is inlined into _load_data for performance. KEEP IN SYNC.
2408
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2412
version_id = key[-1]
2413
# last-element only for compatibilty with the C load_data.
2414
parents = tuple(parent[-1] for parent in parent_keys)
2415
for parent in parent_keys:
2416
if parent[:-1] != prefix:
2417
raise ValueError("mismatched prefixes for %r, %r" % (
2419
cache, history = self._kndx_cache[prefix]
2420
# only want the _history index to reference the 1st index entry
2422
if version_id not in cache:
2423
index = len(history)
2424
history.append(version_id)
2426
index = cache[version_id][5]
2427
cache[version_id] = (version_id,
2434
def check_header(self, fp):
2435
line = fp.readline()
2437
# An empty file can actually be treated as though the file doesn't
2439
raise errors.NoSuchFile(self)
2440
if line != self.HEADER:
2441
raise KnitHeaderError(badline=line, filename=self)
2443
def _check_read(self):
2444
if not self._is_locked():
2445
raise errors.ObjectNotLocked(self)
2446
if self._get_scope() != self._scope:
2449
def _check_write_ok(self):
2450
"""Assert if not writes are permitted."""
2451
if not self._is_locked():
2452
raise errors.ObjectNotLocked(self)
2453
if self._get_scope() != self._scope:
2455
if self._mode != 'w':
2456
raise errors.ReadOnlyObjectDirtiedError(self)
2458
def get_build_details(self, keys):
2459
"""Get the method, index_memo and compression parent for keys.
2461
Ghosts are omitted from the result.
2463
:param keys: An iterable of keys.
2464
:return: A dict of key:(index_memo, compression_parent, parents,
2467
opaque structure to pass to read_records to extract the raw
2470
Content that this record is built upon, may be None
2472
Logical parents of this node
2474
extra information about the content which needs to be passed to
2475
Factory.parse_record
2477
parent_map = self.get_parent_map(keys)
2480
if key not in parent_map:
2482
method = self.get_method(key)
2483
parents = parent_map[key]
2484
if method == 'fulltext':
2485
compression_parent = None
2487
compression_parent = parents[0]
2488
noeol = 'no-eol' in self.get_options(key)
2489
index_memo = self.get_position(key)
2490
result[key] = (index_memo, compression_parent,
2491
parents, (method, noeol))
2494
def get_method(self, key):
2495
"""Return compression method of specified key."""
2496
options = self.get_options(key)
2497
if 'fulltext' in options:
2499
elif 'line-delta' in options:
2502
raise errors.KnitIndexUnknownMethod(self, options)
2504
def get_options(self, key):
2505
"""Return a list representing options.
2509
prefix, suffix = self._split_key(key)
2510
self._load_prefixes([prefix])
2512
return self._kndx_cache[prefix][0][suffix][1]
2514
raise RevisionNotPresent(key, self)
2516
def get_parent_map(self, keys):
2517
"""Get a map of the parents of keys.
2519
:param keys: The keys to look up parents for.
2520
:return: A mapping from keys to parents. Absent keys are absent from
2523
# Parse what we need to up front, this potentially trades off I/O
2524
# locality (.kndx and .knit in the same block group for the same file
2525
# id) for less checking in inner loops.
2526
prefixes = set(key[:-1] for key in keys)
2527
self._load_prefixes(prefixes)
2532
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2536
result[key] = tuple(prefix + (suffix,) for
2537
suffix in suffix_parents)
2540
def get_position(self, key):
2541
"""Return details needed to access the version.
2543
:return: a tuple (key, data position, size) to hand to the access
2544
logic to get the record.
2546
prefix, suffix = self._split_key(key)
2547
self._load_prefixes([prefix])
2548
entry = self._kndx_cache[prefix][0][suffix]
2549
return key, entry[2], entry[3]
2551
has_key = _mod_index._has_key_from_parent_map
2553
def _init_index(self, path, extra_lines=[]):
2554
"""Initialize an index."""
2556
sio.write(self.HEADER)
2557
sio.writelines(extra_lines)
2559
self._transport.put_file_non_atomic(path, sio,
2560
create_parent_dir=True)
2561
# self._create_parent_dir)
2562
# mode=self._file_mode,
2563
# dir_mode=self._dir_mode)
2566
"""Get all the keys in the collection.
2568
The keys are not ordered.
2571
# Identify all key prefixes.
2572
# XXX: A bit hacky, needs polish.
2573
if type(self._mapper) == ConstantMapper:
2577
for quoted_relpath in self._transport.iter_files_recursive():
2578
path, ext = os.path.splitext(quoted_relpath)
2580
prefixes = [self._mapper.unmap(path) for path in relpaths]
2581
self._load_prefixes(prefixes)
2582
for prefix in prefixes:
2583
for suffix in self._kndx_cache[prefix][1]:
2584
result.add(prefix + (suffix,))
2587
def _load_prefixes(self, prefixes):
2588
"""Load the indices for prefixes."""
2590
for prefix in prefixes:
2591
if prefix not in self._kndx_cache:
2592
# the load_data interface writes to these variables.
2595
self._filename = prefix
2597
path = self._mapper.map(prefix) + '.kndx'
2598
fp = self._transport.get(path)
2600
# _load_data may raise NoSuchFile if the target knit is
2602
_load_data(self, fp)
2605
self._kndx_cache[prefix] = (self._cache, self._history)
2610
self._kndx_cache[prefix] = ({}, [])
2611
if type(self._mapper) == ConstantMapper:
2612
# preserve behaviour for revisions.kndx etc.
2613
self._init_index(path)
2618
missing_keys = _mod_index._missing_keys_from_parent_map
2620
def _partition_keys(self, keys):
2621
"""Turn keys into a dict of prefix:suffix_list."""
2624
prefix_keys = result.setdefault(key[:-1], [])
2625
prefix_keys.append(key[-1])
2628
def _dictionary_compress(self, keys):
2629
"""Dictionary compress keys.
2631
:param keys: The keys to generate references to.
2632
:return: A string representation of keys. keys which are present are
2633
dictionary compressed, and others are emitted as fulltext with a
2639
prefix = keys[0][:-1]
2640
cache = self._kndx_cache[prefix][0]
2642
if key[:-1] != prefix:
2643
# kndx indices cannot refer across partitioned storage.
2644
raise ValueError("mismatched prefixes for %r" % keys)
2645
if key[-1] in cache:
2646
# -- inlined lookup() --
2647
result_list.append(str(cache[key[-1]][5]))
2648
# -- end lookup () --
2650
result_list.append('.' + key[-1])
2651
return ' '.join(result_list)
2653
def _reset_cache(self):
2654
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2655
# (cache_dict, history_vector) for parsed kndx files.
2656
self._kndx_cache = {}
2657
self._scope = self._get_scope()
2658
allow_writes = self._allow_writes()
2664
def _sort_keys_by_io(self, keys, positions):
2665
"""Figure out an optimal order to read the records for the given keys.
2667
Sort keys, grouped by index and sorted by position.
2669
:param keys: A list of keys whose records we want to read. This will be
2671
:param positions: A dict, such as the one returned by
2672
_get_components_positions()
2675
def get_sort_key(key):
2676
index_memo = positions[key][1]
2677
# Group by prefix and position. index_memo[0] is the key, so it is
2678
# (file_id, revision_id) and we don't want to sort on revision_id,
2679
# index_memo[1] is the position, and index_memo[2] is the size,
2680
# which doesn't matter for the sort
2681
return index_memo[0][:-1], index_memo[1]
2682
return keys.sort(key=get_sort_key)
2684
_get_total_build_size = _get_total_build_size
2686
def _split_key(self, key):
2687
"""Split key into a prefix and suffix."""
2688
return key[:-1], key[-1]
2691
class _KeyRefs(object):
2694
# dict mapping 'key' to 'set of keys referring to that key'
2697
def add_references(self, key, refs):
2698
# Record the new references
2699
for referenced in refs:
2701
needed_by = self.refs[referenced]
2703
needed_by = self.refs[referenced] = set()
2705
# Discard references satisfied by the new key
2708
def get_unsatisfied_refs(self):
2709
return self.refs.iterkeys()
2711
def add_key(self, key):
2715
# No keys depended on this key. That's ok.
2718
def add_keys(self, keys):
2722
def get_referrers(self):
2724
for referrers in self.refs.itervalues():
2725
result.update(referrers)
2729
class _KnitGraphIndex(object):
2730
"""A KnitVersionedFiles index layered on GraphIndex."""
2732
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2733
add_callback=None, track_external_parent_refs=False):
2734
"""Construct a KnitGraphIndex on a graph_index.
2736
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2737
:param is_locked: A callback to check whether the object should answer
2739
:param deltas: Allow delta-compressed records.
2740
:param parents: If True, record knits parents, if not do not record
2742
:param add_callback: If not None, allow additions to the index and call
2743
this callback with a list of added GraphIndex nodes:
2744
[(node, value, node_refs), ...]
2745
:param is_locked: A callback, returns True if the index is locked and
2747
:param track_external_parent_refs: If True, record all external parent
2748
references parents from added records. These can be retrieved
2749
later by calling get_missing_parents().
2751
self._add_callback = add_callback
2752
self._graph_index = graph_index
2753
self._deltas = deltas
2754
self._parents = parents
2755
if deltas and not parents:
2756
# XXX: TODO: Delta tree and parent graph should be conceptually
2758
raise KnitCorrupt(self, "Cannot do delta compression without "
2760
self.has_graph = parents
2761
self._is_locked = is_locked
2762
self._missing_compression_parents = set()
2763
if track_external_parent_refs:
2764
self._key_dependencies = _KeyRefs()
2766
self._key_dependencies = None
2769
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2771
def add_records(self, records, random_id=False,
2772
missing_compression_parents=False):
2773
"""Add multiple records to the index.
2775
This function does not insert data into the Immutable GraphIndex
2776
backing the KnitGraphIndex, instead it prepares data for insertion by
2777
the caller and checks that it is safe to insert then calls
2778
self._add_callback with the prepared GraphIndex nodes.
2780
:param records: a list of tuples:
2781
(key, options, access_memo, parents).
2782
:param random_id: If True the ids being added were randomly generated
2783
and no check for existence will be performed.
2784
:param missing_compression_parents: If True the records being added are
2785
only compressed against texts already in the index (or inside
2786
records). If False the records all refer to unavailable texts (or
2787
texts inside records) as compression parents.
2789
if not self._add_callback:
2790
raise errors.ReadOnlyError(self)
2791
# we hope there are no repositories with inconsistent parentage
2795
compression_parents = set()
2796
key_dependencies = self._key_dependencies
2797
for (key, options, access_memo, parents) in records:
2799
parents = tuple(parents)
2800
if key_dependencies is not None:
2801
key_dependencies.add_references(key, parents)
2802
index, pos, size = access_memo
2803
if 'no-eol' in options:
2807
value += "%d %d" % (pos, size)
2808
if not self._deltas:
2809
if 'line-delta' in options:
2810
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2813
if 'line-delta' in options:
2814
node_refs = (parents, (parents[0],))
2815
if missing_compression_parents:
2816
compression_parents.add(parents[0])
2818
node_refs = (parents, ())
2820
node_refs = (parents, )
2823
raise KnitCorrupt(self, "attempt to add node with parents "
2824
"in parentless index.")
2826
keys[key] = (value, node_refs)
2829
present_nodes = self._get_entries(keys)
2830
for (index, key, value, node_refs) in present_nodes:
2831
if (value[0] != keys[key][0][0] or
2832
node_refs[:1] != keys[key][1][:1]):
2833
raise KnitCorrupt(self, "inconsistent details in add_records"
2834
": %s %s" % ((value, node_refs), keys[key]))
2838
for key, (value, node_refs) in keys.iteritems():
2839
result.append((key, value, node_refs))
2841
for key, (value, node_refs) in keys.iteritems():
2842
result.append((key, value))
2843
self._add_callback(result)
2844
if missing_compression_parents:
2845
# This may appear to be incorrect (it does not check for
2846
# compression parents that are in the existing graph index),
2847
# but such records won't have been buffered, so this is
2848
# actually correct: every entry when
2849
# missing_compression_parents==True either has a missing parent, or
2850
# a parent that is one of the keys in records.
2851
compression_parents.difference_update(keys)
2852
self._missing_compression_parents.update(compression_parents)
2853
# Adding records may have satisfied missing compression parents.
2854
self._missing_compression_parents.difference_update(keys)
2856
def scan_unvalidated_index(self, graph_index):
2857
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2859
This allows this _KnitGraphIndex to keep track of any missing
2860
compression parents we may want to have filled in to make those
2863
:param graph_index: A GraphIndex
2866
new_missing = graph_index.external_references(ref_list_num=1)
2867
new_missing.difference_update(self.get_parent_map(new_missing))
2868
self._missing_compression_parents.update(new_missing)
2869
if self._key_dependencies is not None:
2870
# Add parent refs from graph_index (and discard parent refs that
2871
# the graph_index has).
2872
for node in graph_index.iter_all_entries():
2873
self._key_dependencies.add_references(node[1], node[3][0])
2875
def get_missing_compression_parents(self):
2876
"""Return the keys of missing compression parents.
2878
Missing compression parents occur when a record stream was missing
2879
basis texts, or a index was scanned that had missing basis texts.
2881
return frozenset(self._missing_compression_parents)
2883
def get_missing_parents(self):
2884
"""Return the keys of missing parents."""
2885
# We may have false positives, so filter those out.
2886
self._key_dependencies.add_keys(
2887
self.get_parent_map(self._key_dependencies.get_unsatisfied_refs()))
2888
return frozenset(self._key_dependencies.get_unsatisfied_refs())
2890
def _check_read(self):
2891
"""raise if reads are not permitted."""
2892
if not self._is_locked():
2893
raise errors.ObjectNotLocked(self)
2895
def _check_write_ok(self):
2896
"""Assert if writes are not permitted."""
2897
if not self._is_locked():
2898
raise errors.ObjectNotLocked(self)
2900
def _compression_parent(self, an_entry):
2901
# return the key that an_entry is compressed against, or None
2902
# Grab the second parent list (as deltas implies parents currently)
2903
compression_parents = an_entry[3][1]
2904
if not compression_parents:
2906
if len(compression_parents) != 1:
2907
raise AssertionError(
2908
"Too many compression parents: %r" % compression_parents)
2909
return compression_parents[0]
2911
def get_build_details(self, keys):
2912
"""Get the method, index_memo and compression parent for version_ids.
2914
Ghosts are omitted from the result.
2916
:param keys: An iterable of keys.
2917
:return: A dict of key:
2918
(index_memo, compression_parent, parents, record_details).
2920
opaque structure to pass to read_records to extract the raw
2923
Content that this record is built upon, may be None
2925
Logical parents of this node
2927
extra information about the content which needs to be passed to
2928
Factory.parse_record
2932
entries = self._get_entries(keys, False)
2933
for entry in entries:
2935
if not self._parents:
2938
parents = entry[3][0]
2939
if not self._deltas:
2940
compression_parent_key = None
2942
compression_parent_key = self._compression_parent(entry)
2943
noeol = (entry[2][0] == 'N')
2944
if compression_parent_key:
2945
method = 'line-delta'
2948
result[key] = (self._node_to_position(entry),
2949
compression_parent_key, parents,
2953
def _get_entries(self, keys, check_present=False):
2954
"""Get the entries for keys.
2956
:param keys: An iterable of index key tuples.
2961
for node in self._graph_index.iter_entries(keys):
2963
found_keys.add(node[1])
2965
# adapt parentless index to the rest of the code.
2966
for node in self._graph_index.iter_entries(keys):
2967
yield node[0], node[1], node[2], ()
2968
found_keys.add(node[1])
2970
missing_keys = keys.difference(found_keys)
2972
raise RevisionNotPresent(missing_keys.pop(), self)
2974
def get_method(self, key):
2975
"""Return compression method of specified key."""
2976
return self._get_method(self._get_node(key))
2978
def _get_method(self, node):
2979
if not self._deltas:
2981
if self._compression_parent(node):
2986
def _get_node(self, key):
2988
return list(self._get_entries([key]))[0]
2990
raise RevisionNotPresent(key, self)
2992
def get_options(self, key):
2993
"""Return a list representing options.
2997
node = self._get_node(key)
2998
options = [self._get_method(node)]
2999
if node[2][0] == 'N':
3000
options.append('no-eol')
3003
def get_parent_map(self, keys):
3004
"""Get a map of the parents of keys.
3006
:param keys: The keys to look up parents for.
3007
:return: A mapping from keys to parents. Absent keys are absent from
3011
nodes = self._get_entries(keys)
3015
result[node[1]] = node[3][0]
3018
result[node[1]] = None
3021
def get_position(self, key):
3022
"""Return details needed to access the version.
3024
:return: a tuple (index, data position, size) to hand to the access
3025
logic to get the record.
3027
node = self._get_node(key)
3028
return self._node_to_position(node)
3030
has_key = _mod_index._has_key_from_parent_map
3033
"""Get all the keys in the collection.
3035
The keys are not ordered.
3038
return [node[1] for node in self._graph_index.iter_all_entries()]
3040
missing_keys = _mod_index._missing_keys_from_parent_map
3042
def _node_to_position(self, node):
3043
"""Convert an index value to position details."""
3044
bits = node[2][1:].split(' ')
3045
return node[0], int(bits[0]), int(bits[1])
3047
def _sort_keys_by_io(self, keys, positions):
3048
"""Figure out an optimal order to read the records for the given keys.
3050
Sort keys, grouped by index and sorted by position.
3052
:param keys: A list of keys whose records we want to read. This will be
3054
:param positions: A dict, such as the one returned by
3055
_get_components_positions()
3058
def get_index_memo(key):
3059
# index_memo is at offset [1]. It is made up of (GraphIndex,
3060
# position, size). GI is an object, which will be unique for each
3061
# pack file. This causes us to group by pack file, then sort by
3062
# position. Size doesn't matter, but it isn't worth breaking up the
3064
return positions[key][1]
3065
return keys.sort(key=get_index_memo)
3067
_get_total_build_size = _get_total_build_size
3070
class _KnitKeyAccess(object):
3071
"""Access to records in .knit files."""
3073
def __init__(self, transport, mapper):
3074
"""Create a _KnitKeyAccess with transport and mapper.
3076
:param transport: The transport the access object is rooted at.
3077
:param mapper: The mapper used to map keys to .knit files.
3079
self._transport = transport
3080
self._mapper = mapper
3082
def add_raw_records(self, key_sizes, raw_data):
3083
"""Add raw knit bytes to a storage area.
3085
The data is spooled to the container writer in one bytes-record per
3088
:param sizes: An iterable of tuples containing the key and size of each
3090
:param raw_data: A bytestring containing the data.
3091
:return: A list of memos to retrieve the record later. Each memo is an
3092
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3093
length), where the key is the record key.
3095
if type(raw_data) != str:
3096
raise AssertionError(
3097
'data must be plain bytes was %s' % type(raw_data))
3100
# TODO: This can be tuned for writing to sftp and other servers where
3101
# append() is relatively expensive by grouping the writes to each key
3103
for key, size in key_sizes:
3104
path = self._mapper.map(key)
3106
base = self._transport.append_bytes(path + '.knit',
3107
raw_data[offset:offset+size])
3108
except errors.NoSuchFile:
3109
self._transport.mkdir(osutils.dirname(path))
3110
base = self._transport.append_bytes(path + '.knit',
3111
raw_data[offset:offset+size])
3115
result.append((key, base, size))
3119
"""Flush pending writes on this access object.
3121
For .knit files this is a no-op.
3125
def get_raw_records(self, memos_for_retrieval):
3126
"""Get the raw bytes for a records.
3128
:param memos_for_retrieval: An iterable containing the access memo for
3129
retrieving the bytes.
3130
:return: An iterator over the bytes of the records.
3132
# first pass, group into same-index request to minimise readv's issued.
3134
current_prefix = None
3135
for (key, offset, length) in memos_for_retrieval:
3136
if current_prefix == key[:-1]:
3137
current_list.append((offset, length))
3139
if current_prefix is not None:
3140
request_lists.append((current_prefix, current_list))
3141
current_prefix = key[:-1]
3142
current_list = [(offset, length)]
3143
# handle the last entry
3144
if current_prefix is not None:
3145
request_lists.append((current_prefix, current_list))
3146
for prefix, read_vector in request_lists:
3147
path = self._mapper.map(prefix) + '.knit'
3148
for pos, data in self._transport.readv(path, read_vector):
3152
class _DirectPackAccess(object):
3153
"""Access to data in one or more packs with less translation."""
3155
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
3156
"""Create a _DirectPackAccess object.
3158
:param index_to_packs: A dict mapping index objects to the transport
3159
and file names for obtaining data.
3160
:param reload_func: A function to call if we determine that the pack
3161
files have moved and we need to reload our caches. See
3162
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
3164
self._container_writer = None
3165
self._write_index = None
3166
self._indices = index_to_packs
3167
self._reload_func = reload_func
3168
self._flush_func = flush_func
3170
def add_raw_records(self, key_sizes, raw_data):
3171
"""Add raw knit bytes to a storage area.
3173
The data is spooled to the container writer in one bytes-record per
3176
:param sizes: An iterable of tuples containing the key and size of each
3178
:param raw_data: A bytestring containing the data.
3179
:return: A list of memos to retrieve the record later. Each memo is an
3180
opaque index memo. For _DirectPackAccess the memo is (index, pos,
3181
length), where the index field is the write_index object supplied
3182
to the PackAccess object.
3184
if type(raw_data) != str:
3185
raise AssertionError(
3186
'data must be plain bytes was %s' % type(raw_data))
3189
for key, size in key_sizes:
3190
p_offset, p_length = self._container_writer.add_bytes_record(
3191
raw_data[offset:offset+size], [])
3193
result.append((self._write_index, p_offset, p_length))
3197
"""Flush pending writes on this access object.
3199
This will flush any buffered writes to a NewPack.
3201
if self._flush_func is not None:
3204
def get_raw_records(self, memos_for_retrieval):
3205
"""Get the raw bytes for a records.
3207
:param memos_for_retrieval: An iterable containing the (index, pos,
3208
length) memo for retrieving the bytes. The Pack access method
3209
looks up the pack to use for a given record in its index_to_pack
3211
:return: An iterator over the bytes of the records.
3213
# first pass, group into same-index requests
3215
current_index = None
3216
for (index, offset, length) in memos_for_retrieval:
3217
if current_index == index:
3218
current_list.append((offset, length))
3220
if current_index is not None:
3221
request_lists.append((current_index, current_list))
3222
current_index = index
3223
current_list = [(offset, length)]
3224
# handle the last entry
3225
if current_index is not None:
3226
request_lists.append((current_index, current_list))
3227
for index, offsets in request_lists:
3229
transport, path = self._indices[index]
3231
# A KeyError here indicates that someone has triggered an index
3232
# reload, and this index has gone missing, we need to start
3234
if self._reload_func is None:
3235
# If we don't have a _reload_func there is nothing that can
3238
raise errors.RetryWithNewPacks(index,
3239
reload_occurred=True,
3240
exc_info=sys.exc_info())
3242
reader = pack.make_readv_reader(transport, path, offsets)
3243
for names, read_func in reader.iter_records():
3244
yield read_func(None)
3245
except errors.NoSuchFile:
3246
# A NoSuchFile error indicates that a pack file has gone
3247
# missing on disk, we need to trigger a reload, and start over.
3248
if self._reload_func is None:
3250
raise errors.RetryWithNewPacks(transport.abspath(path),
3251
reload_occurred=False,
3252
exc_info=sys.exc_info())
3254
def set_writer(self, writer, index, transport_packname):
3255
"""Set a writer to use for adding data."""
3256
if index is not None:
3257
self._indices[index] = transport_packname
3258
self._container_writer = writer
3259
self._write_index = index
3261
def reload_or_raise(self, retry_exc):
3262
"""Try calling the reload function, or re-raise the original exception.
3264
This should be called after _DirectPackAccess raises a
3265
RetryWithNewPacks exception. This function will handle the common logic
3266
of determining when the error is fatal versus being temporary.
3267
It will also make sure that the original exception is raised, rather
3268
than the RetryWithNewPacks exception.
3270
If this function returns, then the calling function should retry
3271
whatever operation was being performed. Otherwise an exception will
3274
:param retry_exc: A RetryWithNewPacks exception.
3277
if self._reload_func is None:
3279
elif not self._reload_func():
3280
# The reload claimed that nothing changed
3281
if not retry_exc.reload_occurred:
3282
# If there wasn't an earlier reload, then we really were
3283
# expecting to find changes. We didn't find them, so this is a
3287
exc_class, exc_value, exc_traceback = retry_exc.exc_info
3288
raise exc_class, exc_value, exc_traceback
3291
# Deprecated, use PatienceSequenceMatcher instead
3292
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
3295
def annotate_knit(knit, revision_id):
3296
"""Annotate a knit with no cached annotations.
3298
This implementation is for knits with no cached annotations.
3299
It will work for knits with cached annotations, but this is not
3302
annotator = _KnitAnnotator(knit)
3303
return iter(annotator.annotate(revision_id))
3306
class _KnitAnnotator(object):
3307
"""Build up the annotations for a text."""
3309
def __init__(self, knit):
3312
# Content objects, differs from fulltexts because of how final newlines
3313
# are treated by knits. the content objects here will always have a
3315
self._fulltext_contents = {}
3317
# Annotated lines of specific revisions
3318
self._annotated_lines = {}
3320
# Track the raw data for nodes that we could not process yet.
3321
# This maps the revision_id of the base to a list of children that will
3322
# annotated from it.
3323
self._pending_children = {}
3325
# Nodes which cannot be extracted
3326
self._ghosts = set()
3328
# Track how many children this node has, so we know if we need to keep
3330
self._annotate_children = {}
3331
self._compression_children = {}
3333
self._all_build_details = {}
3334
# The children => parent revision_id graph
3335
self._revision_id_graph = {}
3337
self._heads_provider = None
3339
self._nodes_to_keep_annotations = set()
3340
self._generations_until_keep = 100
3342
def set_generations_until_keep(self, value):
3343
"""Set the number of generations before caching a node.
3345
Setting this to -1 will cache every merge node, setting this higher
3346
will cache fewer nodes.
3348
self._generations_until_keep = value
3350
def _add_fulltext_content(self, revision_id, content_obj):
3351
self._fulltext_contents[revision_id] = content_obj
3352
# TODO: jam 20080305 It might be good to check the sha1digest here
3353
return content_obj.text()
3355
def _check_parents(self, child, nodes_to_annotate):
3356
"""Check if all parents have been processed.
3358
:param child: A tuple of (rev_id, parents, raw_content)
3359
:param nodes_to_annotate: If child is ready, add it to
3360
nodes_to_annotate, otherwise put it back in self._pending_children
3362
for parent_id in child[1]:
3363
if (parent_id not in self._annotated_lines):
3364
# This parent is present, but another parent is missing
3365
self._pending_children.setdefault(parent_id,
3369
# This one is ready to be processed
3370
nodes_to_annotate.append(child)
3372
def _add_annotation(self, revision_id, fulltext, parent_ids,
3373
left_matching_blocks=None):
3374
"""Add an annotation entry.
3376
All parents should already have been annotated.
3377
:return: A list of children that now have their parents satisfied.
3379
a = self._annotated_lines
3380
annotated_parent_lines = [a[p] for p in parent_ids]
3381
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
3382
fulltext, revision_id, left_matching_blocks,
3383
heads_provider=self._get_heads_provider()))
3384
self._annotated_lines[revision_id] = annotated_lines
3385
for p in parent_ids:
3386
ann_children = self._annotate_children[p]
3387
ann_children.remove(revision_id)
3388
if (not ann_children
3389
and p not in self._nodes_to_keep_annotations):
3390
del self._annotated_lines[p]
3391
del self._all_build_details[p]
3392
if p in self._fulltext_contents:
3393
del self._fulltext_contents[p]
3394
# Now that we've added this one, see if there are any pending
3395
# deltas to be done, certainly this parent is finished
3396
nodes_to_annotate = []
3397
for child in self._pending_children.pop(revision_id, []):
3398
self._check_parents(child, nodes_to_annotate)
3399
return nodes_to_annotate
3401
def _get_build_graph(self, key):
3402
"""Get the graphs for building texts and annotations.
3404
The data you need for creating a full text may be different than the
3405
data you need to annotate that text. (At a minimum, you need both
3406
parents to create an annotation, but only need 1 parent to generate the
3409
:return: A list of (key, index_memo) records, suitable for
3410
passing to read_records_iter to start reading in the raw data fro/
3413
if key in self._annotated_lines:
3416
pending = set([key])
3421
# get all pending nodes
3423
this_iteration = pending
3424
build_details = self._knit._index.get_build_details(this_iteration)
3425
self._all_build_details.update(build_details)
3426
# new_nodes = self._knit._index._get_entries(this_iteration)
3428
for key, details in build_details.iteritems():
3429
(index_memo, compression_parent, parents,
3430
record_details) = details
3431
self._revision_id_graph[key] = parents
3432
records.append((key, index_memo))
3433
# Do we actually need to check _annotated_lines?
3434
pending.update(p for p in parents
3435
if p not in self._all_build_details)
3436
if compression_parent:
3437
self._compression_children.setdefault(compression_parent,
3440
for parent in parents:
3441
self._annotate_children.setdefault(parent,
3443
num_gens = generation - kept_generation
3444
if ((num_gens >= self._generations_until_keep)
3445
and len(parents) > 1):
3446
kept_generation = generation
3447
self._nodes_to_keep_annotations.add(key)
3449
missing_versions = this_iteration.difference(build_details.keys())
3450
self._ghosts.update(missing_versions)
3451
for missing_version in missing_versions:
3452
# add a key, no parents
3453
self._revision_id_graph[missing_version] = ()
3454
pending.discard(missing_version) # don't look for it
3455
if self._ghosts.intersection(self._compression_children):
3457
"We cannot have nodes which have a ghost compression parent:\n"
3459
"compression children: %r"
3460
% (self._ghosts, self._compression_children))
3461
# Cleanout anything that depends on a ghost so that we don't wait for
3462
# the ghost to show up
3463
for node in self._ghosts:
3464
if node in self._annotate_children:
3465
# We won't be building this node
3466
del self._annotate_children[node]
3467
# Generally we will want to read the records in reverse order, because
3468
# we find the parent nodes after the children
3472
def _annotate_records(self, records):
3473
"""Build the annotations for the listed records."""
3474
# We iterate in the order read, rather than a strict order requested
3475
# However, process what we can, and put off to the side things that
3476
# still need parents, cleaning them up when those parents are
3478
for (rev_id, record,
3479
digest) in self._knit._read_records_iter(records):
3480
if rev_id in self._annotated_lines:
3482
parent_ids = self._revision_id_graph[rev_id]
3483
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3484
details = self._all_build_details[rev_id]
3485
(index_memo, compression_parent, parents,
3486
record_details) = details
3487
nodes_to_annotate = []
3488
# TODO: Remove the punning between compression parents, and
3489
# parent_ids, we should be able to do this without assuming
3491
if len(parent_ids) == 0:
3492
# There are no parents for this node, so just add it
3493
# TODO: This probably needs to be decoupled
3494
fulltext_content, delta = self._knit._factory.parse_record(
3495
rev_id, record, record_details, None)
3496
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3497
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3498
parent_ids, left_matching_blocks=None))
3500
child = (rev_id, parent_ids, record)
3501
# Check if all the parents are present
3502
self._check_parents(child, nodes_to_annotate)
3503
while nodes_to_annotate:
3504
# Should we use a queue here instead of a stack?
3505
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3506
(index_memo, compression_parent, parents,
3507
record_details) = self._all_build_details[rev_id]
3509
if compression_parent is not None:
3510
comp_children = self._compression_children[compression_parent]
3511
if rev_id not in comp_children:
3512
raise AssertionError("%r not in compression children %r"
3513
% (rev_id, comp_children))
3514
# If there is only 1 child, it is safe to reuse this
3516
reuse_content = (len(comp_children) == 1
3517
and compression_parent not in
3518
self._nodes_to_keep_annotations)
3520
# Remove it from the cache since it will be changing
3521
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3522
# Make sure to copy the fulltext since it might be
3524
parent_fulltext = list(parent_fulltext_content.text())
3526
parent_fulltext_content = self._fulltext_contents[compression_parent]
3527
parent_fulltext = parent_fulltext_content.text()
3528
comp_children.remove(rev_id)
3529
fulltext_content, delta = self._knit._factory.parse_record(
3530
rev_id, record, record_details,
3531
parent_fulltext_content,
3532
copy_base_content=(not reuse_content))
3533
fulltext = self._add_fulltext_content(rev_id,
3535
if compression_parent == parent_ids[0]:
3536
# the compression_parent is the left parent, so we can
3538
blocks = KnitContent.get_line_delta_blocks(delta,
3539
parent_fulltext, fulltext)
3541
fulltext_content = self._knit._factory.parse_fulltext(
3543
fulltext = self._add_fulltext_content(rev_id,
3545
nodes_to_annotate.extend(
3546
self._add_annotation(rev_id, fulltext, parent_ids,
3547
left_matching_blocks=blocks))
3549
def _get_heads_provider(self):
3550
"""Create a heads provider for resolving ancestry issues."""
3551
if self._heads_provider is not None:
3552
return self._heads_provider
3553
parent_provider = _mod_graph.DictParentsProvider(
3554
self._revision_id_graph)
3555
graph_obj = _mod_graph.Graph(parent_provider)
3556
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3557
self._heads_provider = head_cache
3560
def annotate(self, key):
3561
"""Return the annotated fulltext at the given key.
3563
:param key: The key to annotate.
3565
if len(self._knit._fallback_vfs) > 0:
3566
# stacked knits can't use the fast path at present.
3567
return self._simple_annotate(key)
3570
records = self._get_build_graph(key)
3571
if key in self._ghosts:
3572
raise errors.RevisionNotPresent(key, self._knit)
3573
self._annotate_records(records)
3574
return self._annotated_lines[key]
3575
except errors.RetryWithNewPacks, e:
3576
self._knit._access.reload_or_raise(e)
3577
# The cached build_details are no longer valid
3578
self._all_build_details.clear()
3580
def _simple_annotate(self, key):
3581
"""Return annotated fulltext, rediffing from the full texts.
3583
This is slow but makes no assumptions about the repository
3584
being able to produce line deltas.
3586
# TODO: this code generates a parent maps of present ancestors; it
3587
# could be split out into a separate method, and probably should use
3588
# iter_ancestry instead. -- mbp and robertc 20080704
3589
graph = _mod_graph.Graph(self._knit)
3590
head_cache = _mod_graph.FrozenHeadsCache(graph)
3591
search = graph._make_breadth_first_searcher([key])
3595
present, ghosts = search.next_with_ghosts()
3596
except StopIteration:
3598
keys.update(present)
3599
parent_map = self._knit.get_parent_map(keys)
3601
reannotate = annotate.reannotate
3602
for record in self._knit.get_record_stream(keys, 'topological', True):
3604
fulltext = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
3605
parents = parent_map[key]
3606
if parents is not None:
3607
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
3610
parent_cache[key] = list(
3611
reannotate(parent_lines, fulltext, key, None, head_cache))
3613
return parent_cache[key]
3615
raise errors.RevisionNotPresent(key, self._knit)
3619
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3621
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
