1
# Copyright (C) 2006-2011 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
60
from bzrlib.lazy_import import lazy_import
61
lazy_import(globals(), """
78
from bzrlib.repofmt import pack_repo
85
from bzrlib.errors import (
93
from bzrlib.osutils import (
99
from bzrlib.versionedfile import (
100
AbsentContentFactory,
109
# TODO: Split out code specific to this format into an associated object.
111
# TODO: Can we put in some kind of value to check that the index and data
112
# files belong together?
114
# TODO: accommodate binaries, perhaps by storing a byte count
116
# TODO: function to check whole file
118
# TODO: atomically append data, then measure backwards from the cursor
119
# position after writing to work out where it was located. we may need to
120
# bypass python file buffering.
122
DATA_SUFFIX = '.knit'
123
INDEX_SUFFIX = '.kndx'
124
_STREAM_MIN_BUFFER_SIZE = 5*1024*1024
127
class KnitAdapter(object):
128
"""Base class for knit record adaption."""
130
def __init__(self, basis_vf):
131
"""Create an adapter which accesses full texts from basis_vf.
133
:param basis_vf: A versioned file to access basis texts of deltas from.
134
May be None for adapters that do not need to access basis texts.
136
self._data = KnitVersionedFiles(None, None)
137
self._annotate_factory = KnitAnnotateFactory()
138
self._plain_factory = KnitPlainFactory()
139
self._basis_vf = basis_vf
142
class FTAnnotatedToUnannotated(KnitAdapter):
143
"""An adapter from FT annotated knits to unannotated ones."""
145
def get_bytes(self, factory):
146
annotated_compressed_bytes = factory._raw_record
148
self._data._parse_record_unchecked(annotated_compressed_bytes)
149
content = self._annotate_factory.parse_fulltext(contents, rec[1])
150
size, bytes = self._data._record_to_data((rec[1],), rec[3], content.text())
154
class DeltaAnnotatedToUnannotated(KnitAdapter):
155
"""An adapter for deltas from annotated to unannotated."""
157
def get_bytes(self, factory):
158
annotated_compressed_bytes = factory._raw_record
160
self._data._parse_record_unchecked(annotated_compressed_bytes)
161
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
163
contents = self._plain_factory.lower_line_delta(delta)
164
size, bytes = self._data._record_to_data((rec[1],), rec[3], contents)
168
class FTAnnotatedToFullText(KnitAdapter):
169
"""An adapter from FT annotated knits to unannotated ones."""
171
def get_bytes(self, factory):
172
annotated_compressed_bytes = factory._raw_record
174
self._data._parse_record_unchecked(annotated_compressed_bytes)
175
content, delta = self._annotate_factory.parse_record(factory.key[-1],
176
contents, factory._build_details, None)
177
return ''.join(content.text())
180
class DeltaAnnotatedToFullText(KnitAdapter):
181
"""An adapter for deltas from annotated to unannotated."""
183
def get_bytes(self, factory):
184
annotated_compressed_bytes = factory._raw_record
186
self._data._parse_record_unchecked(annotated_compressed_bytes)
187
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
189
compression_parent = factory.parents[0]
190
basis_entry = self._basis_vf.get_record_stream(
191
[compression_parent], 'unordered', True).next()
192
if basis_entry.storage_kind == 'absent':
193
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
194
basis_chunks = basis_entry.get_bytes_as('chunked')
195
basis_lines = osutils.chunks_to_lines(basis_chunks)
196
# Manually apply the delta because we have one annotated content and
198
basis_content = PlainKnitContent(basis_lines, compression_parent)
199
basis_content.apply_delta(delta, rec[1])
200
basis_content._should_strip_eol = factory._build_details[1]
201
return ''.join(basis_content.text())
204
class FTPlainToFullText(KnitAdapter):
205
"""An adapter from FT plain knits to unannotated ones."""
207
def get_bytes(self, factory):
208
compressed_bytes = factory._raw_record
210
self._data._parse_record_unchecked(compressed_bytes)
211
content, delta = self._plain_factory.parse_record(factory.key[-1],
212
contents, factory._build_details, None)
213
return ''.join(content.text())
216
class DeltaPlainToFullText(KnitAdapter):
217
"""An adapter for deltas from annotated to unannotated."""
219
def get_bytes(self, factory):
220
compressed_bytes = factory._raw_record
222
self._data._parse_record_unchecked(compressed_bytes)
223
delta = self._plain_factory.parse_line_delta(contents, rec[1])
224
compression_parent = factory.parents[0]
225
# XXX: string splitting overhead.
226
basis_entry = self._basis_vf.get_record_stream(
227
[compression_parent], 'unordered', True).next()
228
if basis_entry.storage_kind == 'absent':
229
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
230
basis_chunks = basis_entry.get_bytes_as('chunked')
231
basis_lines = osutils.chunks_to_lines(basis_chunks)
232
basis_content = PlainKnitContent(basis_lines, compression_parent)
233
# Manually apply the delta because we have one annotated content and
235
content, _ = self._plain_factory.parse_record(rec[1], contents,
236
factory._build_details, basis_content)
237
return ''.join(content.text())
240
class KnitContentFactory(ContentFactory):
241
"""Content factory for streaming from knits.
243
:seealso ContentFactory:
246
def __init__(self, key, parents, build_details, sha1, raw_record,
247
annotated, knit=None, network_bytes=None):
248
"""Create a KnitContentFactory for key.
251
:param parents: The parents.
252
:param build_details: The build details as returned from
254
:param sha1: The sha1 expected from the full text of this object.
255
:param raw_record: The bytes of the knit data from disk.
256
:param annotated: True if the raw data is annotated.
257
:param network_bytes: None to calculate the network bytes on demand,
258
not-none if they are already known.
260
ContentFactory.__init__(self)
263
self.parents = parents
264
if build_details[0] == 'line-delta':
269
annotated_kind = 'annotated-'
272
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
273
self._raw_record = raw_record
274
self._network_bytes = network_bytes
275
self._build_details = build_details
278
def _create_network_bytes(self):
279
"""Create a fully serialised network version for transmission."""
280
# storage_kind, key, parents, Noeol, raw_record
281
key_bytes = '\x00'.join(self.key)
282
if self.parents is None:
283
parent_bytes = 'None:'
285
parent_bytes = '\t'.join('\x00'.join(key) for key in self.parents)
286
if self._build_details[1]:
290
network_bytes = "%s\n%s\n%s\n%s%s" % (self.storage_kind, key_bytes,
291
parent_bytes, noeol, self._raw_record)
292
self._network_bytes = network_bytes
294
def get_bytes_as(self, storage_kind):
295
if storage_kind == self.storage_kind:
296
if self._network_bytes is None:
297
self._create_network_bytes()
298
return self._network_bytes
299
if ('-ft-' in self.storage_kind and
300
storage_kind in ('chunked', 'fulltext')):
301
adapter_key = (self.storage_kind, 'fulltext')
302
adapter_factory = adapter_registry.get(adapter_key)
303
adapter = adapter_factory(None)
304
bytes = adapter.get_bytes(self)
305
if storage_kind == 'chunked':
309
if self._knit is not None:
310
# Not redundant with direct conversion above - that only handles
312
if storage_kind == 'chunked':
313
return self._knit.get_lines(self.key[0])
314
elif storage_kind == 'fulltext':
315
return self._knit.get_text(self.key[0])
316
raise errors.UnavailableRepresentation(self.key, storage_kind,
320
class LazyKnitContentFactory(ContentFactory):
321
"""A ContentFactory which can either generate full text or a wire form.
323
:seealso ContentFactory:
326
def __init__(self, key, parents, generator, first):
327
"""Create a LazyKnitContentFactory.
329
:param key: The key of the record.
330
:param parents: The parents of the record.
331
:param generator: A _ContentMapGenerator containing the record for this
333
:param first: Is this the first content object returned from generator?
334
if it is, its storage kind is knit-delta-closure, otherwise it is
335
knit-delta-closure-ref
338
self.parents = parents
340
self._generator = generator
341
self.storage_kind = "knit-delta-closure"
343
self.storage_kind = self.storage_kind + "-ref"
346
def get_bytes_as(self, storage_kind):
347
if storage_kind == self.storage_kind:
349
return self._generator._wire_bytes()
351
# all the keys etc are contained in the bytes returned in the
354
if storage_kind in ('chunked', 'fulltext'):
355
chunks = self._generator._get_one_work(self.key).text()
356
if storage_kind == 'chunked':
359
return ''.join(chunks)
360
raise errors.UnavailableRepresentation(self.key, storage_kind,
364
def knit_delta_closure_to_records(storage_kind, bytes, line_end):
365
"""Convert a network record to a iterator over stream records.
367
:param storage_kind: The storage kind of the record.
368
Must be 'knit-delta-closure'.
369
:param bytes: The bytes of the record on the network.
371
generator = _NetworkContentMapGenerator(bytes, line_end)
372
return generator.get_record_stream()
375
def knit_network_to_record(storage_kind, bytes, line_end):
376
"""Convert a network record to a record object.
378
:param storage_kind: The storage kind of the record.
379
:param bytes: The bytes of the record on the network.
382
line_end = bytes.find('\n', start)
383
key = tuple(bytes[start:line_end].split('\x00'))
385
line_end = bytes.find('\n', start)
386
parent_line = bytes[start:line_end]
387
if parent_line == 'None:':
391
[tuple(segment.split('\x00')) for segment in parent_line.split('\t')
394
noeol = bytes[start] == 'N'
395
if 'ft' in storage_kind:
398
method = 'line-delta'
399
build_details = (method, noeol)
401
raw_record = bytes[start:]
402
annotated = 'annotated' in storage_kind
403
return [KnitContentFactory(key, parents, build_details, None, raw_record,
404
annotated, network_bytes=bytes)]
407
class KnitContent(object):
408
"""Content of a knit version to which deltas can be applied.
410
This is always stored in memory as a list of lines with \n at the end,
411
plus a flag saying if the final ending is really there or not, because that
412
corresponds to the on-disk knit representation.
416
self._should_strip_eol = False
418
def apply_delta(self, delta, new_version_id):
419
"""Apply delta to this object to become new_version_id."""
420
raise NotImplementedError(self.apply_delta)
422
def line_delta_iter(self, new_lines):
423
"""Generate line-based delta from this content to new_lines."""
424
new_texts = new_lines.text()
425
old_texts = self.text()
426
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
427
for tag, i1, i2, j1, j2 in s.get_opcodes():
430
# ofrom, oto, length, data
431
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
433
def line_delta(self, new_lines):
434
return list(self.line_delta_iter(new_lines))
437
def get_line_delta_blocks(knit_delta, source, target):
438
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
439
target_len = len(target)
442
for s_begin, s_end, t_len, new_text in knit_delta:
443
true_n = s_begin - s_pos
446
# knit deltas do not provide reliable info about whether the
447
# last line of a file matches, due to eol handling.
448
if source[s_pos + n -1] != target[t_pos + n -1]:
451
yield s_pos, t_pos, n
452
t_pos += t_len + true_n
454
n = target_len - t_pos
456
if source[s_pos + n -1] != target[t_pos + n -1]:
459
yield s_pos, t_pos, n
460
yield s_pos + (target_len - t_pos), target_len, 0
463
class AnnotatedKnitContent(KnitContent):
464
"""Annotated content."""
466
def __init__(self, lines):
467
KnitContent.__init__(self)
471
"""Return a list of (origin, text) for each content line."""
472
lines = self._lines[:]
473
if self._should_strip_eol:
474
origin, last_line = lines[-1]
475
lines[-1] = (origin, last_line.rstrip('\n'))
478
def apply_delta(self, delta, new_version_id):
479
"""Apply delta to this object to become new_version_id."""
482
for start, end, count, delta_lines in delta:
483
lines[offset+start:offset+end] = delta_lines
484
offset = offset + (start - end) + count
488
lines = [text for origin, text in self._lines]
489
except ValueError, e:
490
# most commonly (only?) caused by the internal form of the knit
491
# missing annotation information because of a bug - see thread
493
raise KnitCorrupt(self,
494
"line in annotated knit missing annotation information: %s"
496
if self._should_strip_eol:
497
lines[-1] = lines[-1].rstrip('\n')
501
return AnnotatedKnitContent(self._lines[:])
504
class PlainKnitContent(KnitContent):
505
"""Unannotated content.
507
When annotate[_iter] is called on this content, the same version is reported
508
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
512
def __init__(self, lines, version_id):
513
KnitContent.__init__(self)
515
self._version_id = version_id
518
"""Return a list of (origin, text) for each content line."""
519
return [(self._version_id, line) for line in self._lines]
521
def apply_delta(self, delta, new_version_id):
522
"""Apply delta to this object to become new_version_id."""
525
for start, end, count, delta_lines in delta:
526
lines[offset+start:offset+end] = delta_lines
527
offset = offset + (start - end) + count
528
self._version_id = new_version_id
531
return PlainKnitContent(self._lines[:], self._version_id)
535
if self._should_strip_eol:
537
lines[-1] = lines[-1].rstrip('\n')
541
class _KnitFactory(object):
542
"""Base class for common Factory functions."""
544
def parse_record(self, version_id, record, record_details,
545
base_content, copy_base_content=True):
546
"""Parse a record into a full content object.
548
:param version_id: The official version id for this content
549
:param record: The data returned by read_records_iter()
550
:param record_details: Details about the record returned by
552
:param base_content: If get_build_details returns a compression_parent,
553
you must return a base_content here, else use None
554
:param copy_base_content: When building from the base_content, decide
555
you can either copy it and return a new object, or modify it in
557
:return: (content, delta) A Content object and possibly a line-delta,
560
method, noeol = record_details
561
if method == 'line-delta':
562
if copy_base_content:
563
content = base_content.copy()
565
content = base_content
566
delta = self.parse_line_delta(record, version_id)
567
content.apply_delta(delta, version_id)
569
content = self.parse_fulltext(record, version_id)
571
content._should_strip_eol = noeol
572
return (content, delta)
575
class KnitAnnotateFactory(_KnitFactory):
576
"""Factory for creating annotated Content objects."""
580
def make(self, lines, version_id):
581
num_lines = len(lines)
582
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
584
def parse_fulltext(self, content, version_id):
585
"""Convert fulltext to internal representation
587
fulltext content is of the format
588
revid(utf8) plaintext\n
589
internal representation is of the format:
592
# TODO: jam 20070209 The tests expect this to be returned as tuples,
593
# but the code itself doesn't really depend on that.
594
# Figure out a way to not require the overhead of turning the
595
# list back into tuples.
596
lines = [tuple(line.split(' ', 1)) for line in content]
597
return AnnotatedKnitContent(lines)
599
def parse_line_delta_iter(self, lines):
600
return iter(self.parse_line_delta(lines))
602
def parse_line_delta(self, lines, version_id, plain=False):
603
"""Convert a line based delta into internal representation.
605
line delta is in the form of:
606
intstart intend intcount
608
revid(utf8) newline\n
609
internal representation is
610
(start, end, count, [1..count tuples (revid, newline)])
612
:param plain: If True, the lines are returned as a plain
613
list without annotations, not as a list of (origin, content) tuples, i.e.
614
(start, end, count, [1..count newline])
621
def cache_and_return(line):
622
origin, text = line.split(' ', 1)
623
return cache.setdefault(origin, origin), text
625
# walk through the lines parsing.
626
# Note that the plain test is explicitly pulled out of the
627
# loop to minimise any performance impact
630
start, end, count = [int(n) for n in header.split(',')]
631
contents = [next().split(' ', 1)[1] for i in xrange(count)]
632
result.append((start, end, count, contents))
635
start, end, count = [int(n) for n in header.split(',')]
636
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
637
result.append((start, end, count, contents))
640
def get_fulltext_content(self, lines):
641
"""Extract just the content lines from a fulltext."""
642
return (line.split(' ', 1)[1] for line in lines)
644
def get_linedelta_content(self, lines):
645
"""Extract just the content from a line delta.
647
This doesn't return all of the extra information stored in a delta.
648
Only the actual content lines.
653
header = header.split(',')
654
count = int(header[2])
655
for i in xrange(count):
656
origin, text = next().split(' ', 1)
659
def lower_fulltext(self, content):
660
"""convert a fulltext content record into a serializable form.
662
see parse_fulltext which this inverts.
664
return ['%s %s' % (o, t) for o, t in content._lines]
666
def lower_line_delta(self, delta):
667
"""convert a delta into a serializable form.
669
See parse_line_delta which this inverts.
671
# TODO: jam 20070209 We only do the caching thing to make sure that
672
# the origin is a valid utf-8 line, eventually we could remove it
674
for start, end, c, lines in delta:
675
out.append('%d,%d,%d\n' % (start, end, c))
676
out.extend(origin + ' ' + text
677
for origin, text in lines)
680
def annotate(self, knit, key):
681
content = knit._get_content(key)
682
# adjust for the fact that serialised annotations are only key suffixes
684
if type(key) is tuple:
686
origins = content.annotate()
688
for origin, line in origins:
689
result.append((prefix + (origin,), line))
692
# XXX: This smells a bit. Why would key ever be a non-tuple here?
693
# Aren't keys defined to be tuples? -- spiv 20080618
694
return content.annotate()
697
class KnitPlainFactory(_KnitFactory):
698
"""Factory for creating plain Content objects."""
702
def make(self, lines, version_id):
703
return PlainKnitContent(lines, version_id)
705
def parse_fulltext(self, content, version_id):
706
"""This parses an unannotated fulltext.
708
Note that this is not a noop - the internal representation
709
has (versionid, line) - its just a constant versionid.
711
return self.make(content, version_id)
713
def parse_line_delta_iter(self, lines, version_id):
715
num_lines = len(lines)
716
while cur < num_lines:
719
start, end, c = [int(n) for n in header.split(',')]
720
yield start, end, c, lines[cur:cur+c]
723
def parse_line_delta(self, lines, version_id):
724
return list(self.parse_line_delta_iter(lines, version_id))
726
def get_fulltext_content(self, lines):
727
"""Extract just the content lines from a fulltext."""
730
def get_linedelta_content(self, lines):
731
"""Extract just the content from a line delta.
733
This doesn't return all of the extra information stored in a delta.
734
Only the actual content lines.
739
header = header.split(',')
740
count = int(header[2])
741
for i in xrange(count):
744
def lower_fulltext(self, content):
745
return content.text()
747
def lower_line_delta(self, delta):
749
for start, end, c, lines in delta:
750
out.append('%d,%d,%d\n' % (start, end, c))
754
def annotate(self, knit, key):
755
annotator = _KnitAnnotator(knit)
756
return annotator.annotate_flat(key)
760
def make_file_factory(annotated, mapper):
761
"""Create a factory for creating a file based KnitVersionedFiles.
763
This is only functional enough to run interface tests, it doesn't try to
764
provide a full pack environment.
766
:param annotated: knit annotations are wanted.
767
:param mapper: The mapper from keys to paths.
769
def factory(transport):
770
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
771
access = _KnitKeyAccess(transport, mapper)
772
return KnitVersionedFiles(index, access, annotated=annotated)
776
def make_pack_factory(graph, delta, keylength):
777
"""Create a factory for creating a pack based VersionedFiles.
779
This is only functional enough to run interface tests, it doesn't try to
780
provide a full pack environment.
782
:param graph: Store a graph.
783
:param delta: Delta compress contents.
784
:param keylength: How long should keys be.
786
def factory(transport):
787
parents = graph or delta
793
max_delta_chain = 200
796
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
797
key_elements=keylength)
798
stream = transport.open_write_stream('newpack')
799
writer = pack.ContainerWriter(stream.write)
801
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
802
deltas=delta, add_callback=graph_index.add_nodes)
803
access = pack_repo._DirectPackAccess({})
804
access.set_writer(writer, graph_index, (transport, 'newpack'))
805
result = KnitVersionedFiles(index, access,
806
max_delta_chain=max_delta_chain)
807
result.stream = stream
808
result.writer = writer
813
def cleanup_pack_knit(versioned_files):
814
versioned_files.stream.close()
815
versioned_files.writer.end()
818
def _get_total_build_size(self, keys, positions):
819
"""Determine the total bytes to build these keys.
821
(helper function because _KnitGraphIndex and _KndxIndex work the same, but
822
don't inherit from a common base.)
824
:param keys: Keys that we want to build
825
:param positions: dict of {key, (info, index_memo, comp_parent)} (such
826
as returned by _get_components_positions)
827
:return: Number of bytes to build those keys
829
all_build_index_memos = {}
833
for key in build_keys:
834
# This is mostly for the 'stacked' case
835
# Where we will be getting the data from a fallback
836
if key not in positions:
838
_, index_memo, compression_parent = positions[key]
839
all_build_index_memos[key] = index_memo
840
if compression_parent not in all_build_index_memos:
841
next_keys.add(compression_parent)
842
build_keys = next_keys
843
return sum([index_memo[2] for index_memo
844
in all_build_index_memos.itervalues()])
847
class KnitVersionedFiles(VersionedFiles):
848
"""Storage for many versioned files using knit compression.
850
Backend storage is managed by indices and data objects.
852
:ivar _index: A _KnitGraphIndex or similar that can describe the
853
parents, graph, compression and data location of entries in this
854
KnitVersionedFiles. Note that this is only the index for
855
*this* vfs; if there are fallbacks they must be queried separately.
858
def __init__(self, index, data_access, max_delta_chain=200,
859
annotated=False, reload_func=None):
860
"""Create a KnitVersionedFiles with index and data_access.
862
:param index: The index for the knit data.
863
:param data_access: The access object to store and retrieve knit
865
:param max_delta_chain: The maximum number of deltas to permit during
866
insertion. Set to 0 to prohibit the use of deltas.
867
:param annotated: Set to True to cause annotations to be calculated and
868
stored during insertion.
869
:param reload_func: An function that can be called if we think we need
870
to reload the pack listing and try again. See
871
'bzrlib.repofmt.pack_repo.AggregateIndex' for the signature.
874
self._access = data_access
875
self._max_delta_chain = max_delta_chain
877
self._factory = KnitAnnotateFactory()
879
self._factory = KnitPlainFactory()
880
self._immediate_fallback_vfs = []
881
self._reload_func = reload_func
884
return "%s(%r, %r)" % (
885
self.__class__.__name__,
889
def add_fallback_versioned_files(self, a_versioned_files):
890
"""Add a source of texts for texts not present in this knit.
892
:param a_versioned_files: A VersionedFiles object.
894
self._immediate_fallback_vfs.append(a_versioned_files)
896
def add_lines(self, key, parents, lines, parent_texts=None,
897
left_matching_blocks=None, nostore_sha=None, random_id=False,
899
"""See VersionedFiles.add_lines()."""
900
self._index._check_write_ok()
901
self._check_add(key, lines, random_id, check_content)
903
# The caller might pass None if there is no graph data, but kndx
904
# indexes can't directly store that, so we give them
905
# an empty tuple instead.
907
line_bytes = ''.join(lines)
908
return self._add(key, lines, parents,
909
parent_texts, left_matching_blocks, nostore_sha, random_id,
910
line_bytes=line_bytes)
912
def _add_text(self, key, parents, text, nostore_sha=None, random_id=False):
913
"""See VersionedFiles._add_text()."""
914
self._index._check_write_ok()
915
self._check_add(key, None, random_id, check_content=False)
916
if text.__class__ is not str:
917
raise errors.BzrBadParameterUnicode("text")
919
# The caller might pass None if there is no graph data, but kndx
920
# indexes can't directly store that, so we give them
921
# an empty tuple instead.
923
return self._add(key, None, parents,
924
None, None, nostore_sha, random_id,
927
def _add(self, key, lines, parents, parent_texts,
928
left_matching_blocks, nostore_sha, random_id,
930
"""Add a set of lines on top of version specified by parents.
932
Any versions not present will be converted into ghosts.
934
:param lines: A list of strings where each one is a single line (has a
935
single newline at the end of the string) This is now optional
936
(callers can pass None). It is left in its location for backwards
937
compatibility. It should ''.join(lines) must == line_bytes
938
:param line_bytes: A single string containing the content
940
We pass both lines and line_bytes because different routes bring the
941
values to this function. And for memory efficiency, we don't want to
942
have to split/join on-demand.
944
# first thing, if the content is something we don't need to store, find
946
digest = sha_string(line_bytes)
947
if nostore_sha == digest:
948
raise errors.ExistingContent
951
if parent_texts is None:
953
# Do a single query to ascertain parent presence; we only compress
954
# against parents in the same kvf.
955
present_parent_map = self._index.get_parent_map(parents)
956
for parent in parents:
957
if parent in present_parent_map:
958
present_parents.append(parent)
960
# Currently we can only compress against the left most present parent.
961
if (len(present_parents) == 0 or
962
present_parents[0] != parents[0]):
965
# To speed the extract of texts the delta chain is limited
966
# to a fixed number of deltas. This should minimize both
967
# I/O and the time spend applying deltas.
968
delta = self._check_should_delta(present_parents[0])
970
text_length = len(line_bytes)
973
# Note: line_bytes is not modified to add a newline, that is tracked
974
# via the no_eol flag. 'lines' *is* modified, because that is the
975
# general values needed by the Content code.
976
if line_bytes and line_bytes[-1] != '\n':
977
options.append('no-eol')
979
# Copy the existing list, or create a new one
981
lines = osutils.split_lines(line_bytes)
984
# Replace the last line with one that ends in a final newline
985
lines[-1] = lines[-1] + '\n'
987
lines = osutils.split_lines(line_bytes)
989
for element in key[:-1]:
990
if type(element) is not str:
991
raise TypeError("key contains non-strings: %r" % (key,))
993
key = key[:-1] + ('sha1:' + digest,)
994
elif type(key[-1]) is not str:
995
raise TypeError("key contains non-strings: %r" % (key,))
996
# Knit hunks are still last-element only
998
content = self._factory.make(lines, version_id)
1000
# Hint to the content object that its text() call should strip the
1002
content._should_strip_eol = True
1003
if delta or (self._factory.annotated and len(present_parents) > 0):
1004
# Merge annotations from parent texts if needed.
1005
delta_hunks = self._merge_annotations(content, present_parents,
1006
parent_texts, delta, self._factory.annotated,
1007
left_matching_blocks)
1010
options.append('line-delta')
1011
store_lines = self._factory.lower_line_delta(delta_hunks)
1012
size, bytes = self._record_to_data(key, digest,
1015
options.append('fulltext')
1016
# isinstance is slower and we have no hierarchy.
1017
if self._factory.__class__ is KnitPlainFactory:
1018
# Use the already joined bytes saving iteration time in
1020
dense_lines = [line_bytes]
1022
dense_lines.append('\n')
1023
size, bytes = self._record_to_data(key, digest,
1026
# get mixed annotation + content and feed it into the
1028
store_lines = self._factory.lower_fulltext(content)
1029
size, bytes = self._record_to_data(key, digest,
1032
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
1033
self._index.add_records(
1034
((key, options, access_memo, parents),),
1035
random_id=random_id)
1036
return digest, text_length, content
1038
def annotate(self, key):
1039
"""See VersionedFiles.annotate."""
1040
return self._factory.annotate(self, key)
1042
def get_annotator(self):
1043
return _KnitAnnotator(self)
1045
def check(self, progress_bar=None, keys=None):
1046
"""See VersionedFiles.check()."""
1048
return self._logical_check()
1050
# At the moment, check does not extra work over get_record_stream
1051
return self.get_record_stream(keys, 'unordered', True)
1053
def _logical_check(self):
1054
# This doesn't actually test extraction of everything, but that will
1055
# impact 'bzr check' substantially, and needs to be integrated with
1056
# care. However, it does check for the obvious problem of a delta with
1058
keys = self._index.keys()
1059
parent_map = self.get_parent_map(keys)
1061
if self._index.get_method(key) != 'fulltext':
1062
compression_parent = parent_map[key][0]
1063
if compression_parent not in parent_map:
1064
raise errors.KnitCorrupt(self,
1065
"Missing basis parent %s for %s" % (
1066
compression_parent, key))
1067
for fallback_vfs in self._immediate_fallback_vfs:
1068
fallback_vfs.check()
1070
def _check_add(self, key, lines, random_id, check_content):
1071
"""check that version_id and lines are safe to add."""
1072
version_id = key[-1]
1073
if version_id is not None:
1074
if contains_whitespace(version_id):
1075
raise InvalidRevisionId(version_id, self)
1076
self.check_not_reserved_id(version_id)
1077
# TODO: If random_id==False and the key is already present, we should
1078
# probably check that the existing content is identical to what is
1079
# being inserted, and otherwise raise an exception. This would make
1080
# the bundle code simpler.
1082
self._check_lines_not_unicode(lines)
1083
self._check_lines_are_lines(lines)
1085
def _check_header(self, key, line):
1086
rec = self._split_header(line)
1087
self._check_header_version(rec, key[-1])
1090
def _check_header_version(self, rec, version_id):
1091
"""Checks the header version on original format knit records.
1093
These have the last component of the key embedded in the record.
1095
if rec[1] != version_id:
1096
raise KnitCorrupt(self,
1097
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1099
def _check_should_delta(self, parent):
1100
"""Iterate back through the parent listing, looking for a fulltext.
1102
This is used when we want to decide whether to add a delta or a new
1103
fulltext. It searches for _max_delta_chain parents. When it finds a
1104
fulltext parent, it sees if the total size of the deltas leading up to
1105
it is large enough to indicate that we want a new full text anyway.
1107
Return True if we should create a new delta, False if we should use a
1111
fulltext_size = None
1112
for count in xrange(self._max_delta_chain):
1114
# Note that this only looks in the index of this particular
1115
# KnitVersionedFiles, not in the fallbacks. This ensures that
1116
# we won't store a delta spanning physical repository
1118
build_details = self._index.get_build_details([parent])
1119
parent_details = build_details[parent]
1120
except (RevisionNotPresent, KeyError), e:
1121
# Some basis is not locally present: always fulltext
1123
index_memo, compression_parent, _, _ = parent_details
1124
_, _, size = index_memo
1125
if compression_parent is None:
1126
fulltext_size = size
1129
# We don't explicitly check for presence because this is in an
1130
# inner loop, and if it's missing it'll fail anyhow.
1131
parent = compression_parent
1133
# We couldn't find a fulltext, so we must create a new one
1135
# Simple heuristic - if the total I/O wold be greater as a delta than
1136
# the originally installed fulltext, we create a new fulltext.
1137
return fulltext_size > delta_size
1139
def _build_details_to_components(self, build_details):
1140
"""Convert a build_details tuple to a position tuple."""
1141
# record_details, access_memo, compression_parent
1142
return build_details[3], build_details[0], build_details[1]
1144
def _get_components_positions(self, keys, allow_missing=False):
1145
"""Produce a map of position data for the components of keys.
1147
This data is intended to be used for retrieving the knit records.
1149
A dict of key to (record_details, index_memo, next, parents) is
1151
method is the way referenced data should be applied.
1152
index_memo is the handle to pass to the data access to actually get the
1154
next is the build-parent of the version, or None for fulltexts.
1155
parents is the version_ids of the parents of this version
1157
:param allow_missing: If True do not raise an error on a missing component,
1161
pending_components = keys
1162
while pending_components:
1163
build_details = self._index.get_build_details(pending_components)
1164
current_components = set(pending_components)
1165
pending_components = set()
1166
for key, details in build_details.iteritems():
1167
(index_memo, compression_parent, parents,
1168
record_details) = details
1169
method = record_details[0]
1170
if compression_parent is not None:
1171
pending_components.add(compression_parent)
1172
component_data[key] = self._build_details_to_components(details)
1173
missing = current_components.difference(build_details)
1174
if missing and not allow_missing:
1175
raise errors.RevisionNotPresent(missing.pop(), self)
1176
return component_data
1178
def _get_content(self, key, parent_texts={}):
1179
"""Returns a content object that makes up the specified
1181
cached_version = parent_texts.get(key, None)
1182
if cached_version is not None:
1183
# Ensure the cache dict is valid.
1184
if not self.get_parent_map([key]):
1185
raise RevisionNotPresent(key, self)
1186
return cached_version
1187
generator = _VFContentMapGenerator(self, [key])
1188
return generator._get_content(key)
1190
def get_known_graph_ancestry(self, keys):
1191
"""Get a KnownGraph instance with the ancestry of keys."""
1192
parent_map, missing_keys = self._index.find_ancestry(keys)
1193
for fallback in self._transitive_fallbacks():
1194
if not missing_keys:
1196
(f_parent_map, f_missing_keys) = fallback._index.find_ancestry(
1198
parent_map.update(f_parent_map)
1199
missing_keys = f_missing_keys
1200
kg = _mod_graph.KnownGraph(parent_map)
1203
def get_parent_map(self, keys):
1204
"""Get a map of the graph parents of keys.
1206
:param keys: The keys to look up parents for.
1207
:return: A mapping from keys to parents. Absent keys are absent from
1210
return self._get_parent_map_with_sources(keys)[0]
1212
def _get_parent_map_with_sources(self, keys):
1213
"""Get a map of the parents of keys.
1215
:param keys: The keys to look up parents for.
1216
:return: A tuple. The first element is a mapping from keys to parents.
1217
Absent keys are absent from the mapping. The second element is a
1218
list with the locations each key was found in. The first element
1219
is the in-this-knit parents, the second the first fallback source,
1223
sources = [self._index] + self._immediate_fallback_vfs
1226
for source in sources:
1229
new_result = source.get_parent_map(missing)
1230
source_results.append(new_result)
1231
result.update(new_result)
1232
missing.difference_update(set(new_result))
1233
return result, source_results
1235
def _get_record_map(self, keys, allow_missing=False):
1236
"""Produce a dictionary of knit records.
1238
:return: {key:(record, record_details, digest, next)}
1240
data returned from read_records (a KnitContentobject)
1242
opaque information to pass to parse_record
1244
SHA1 digest of the full text after all steps are done
1246
build-parent of the version, i.e. the leftmost ancestor.
1247
Will be None if the record is not a delta.
1248
:param keys: The keys to build a map for
1249
:param allow_missing: If some records are missing, rather than
1250
error, just return the data that could be generated.
1252
raw_map = self._get_record_map_unparsed(keys,
1253
allow_missing=allow_missing)
1254
return self._raw_map_to_record_map(raw_map)
1256
def _raw_map_to_record_map(self, raw_map):
1257
"""Parse the contents of _get_record_map_unparsed.
1259
:return: see _get_record_map.
1263
data, record_details, next = raw_map[key]
1264
content, digest = self._parse_record(key[-1], data)
1265
result[key] = content, record_details, digest, next
1268
def _get_record_map_unparsed(self, keys, allow_missing=False):
1269
"""Get the raw data for reconstructing keys without parsing it.
1271
:return: A dict suitable for parsing via _raw_map_to_record_map.
1272
key-> raw_bytes, (method, noeol), compression_parent
1274
# This retries the whole request if anything fails. Potentially we
1275
# could be a bit more selective. We could track the keys whose records
1276
# we have successfully found, and then only request the new records
1277
# from there. However, _get_components_positions grabs the whole build
1278
# chain, which means we'll likely try to grab the same records again
1279
# anyway. Also, can the build chains change as part of a pack
1280
# operation? We wouldn't want to end up with a broken chain.
1283
position_map = self._get_components_positions(keys,
1284
allow_missing=allow_missing)
1285
# key = component_id, r = record_details, i_m = index_memo,
1287
records = [(key, i_m) for key, (r, i_m, n)
1288
in position_map.iteritems()]
1289
# Sort by the index memo, so that we request records from the
1290
# same pack file together, and in forward-sorted order
1291
records.sort(key=operator.itemgetter(1))
1293
for key, data in self._read_records_iter_unchecked(records):
1294
(record_details, index_memo, next) = position_map[key]
1295
raw_record_map[key] = data, record_details, next
1296
return raw_record_map
1297
except errors.RetryWithNewPacks, e:
1298
self._access.reload_or_raise(e)
1301
def _split_by_prefix(cls, keys):
1302
"""For the given keys, split them up based on their prefix.
1304
To keep memory pressure somewhat under control, split the
1305
requests back into per-file-id requests, otherwise "bzr co"
1306
extracts the full tree into memory before writing it to disk.
1307
This should be revisited if _get_content_maps() can ever cross
1310
The keys for a given file_id are kept in the same relative order.
1311
Ordering between file_ids is not, though prefix_order will return the
1312
order that the key was first seen.
1314
:param keys: An iterable of key tuples
1315
:return: (split_map, prefix_order)
1316
split_map A dictionary mapping prefix => keys
1317
prefix_order The order that we saw the various prefixes
1319
split_by_prefix = {}
1327
if prefix in split_by_prefix:
1328
split_by_prefix[prefix].append(key)
1330
split_by_prefix[prefix] = [key]
1331
prefix_order.append(prefix)
1332
return split_by_prefix, prefix_order
1334
def _group_keys_for_io(self, keys, non_local_keys, positions,
1335
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1336
"""For the given keys, group them into 'best-sized' requests.
1338
The idea is to avoid making 1 request per file, but to never try to
1339
unpack an entire 1.5GB source tree in a single pass. Also when
1340
possible, we should try to group requests to the same pack file
1343
:return: list of (keys, non_local) tuples that indicate what keys
1344
should be fetched next.
1346
# TODO: Ideally we would group on 2 factors. We want to extract texts
1347
# from the same pack file together, and we want to extract all
1348
# the texts for a given build-chain together. Ultimately it
1349
# probably needs a better global view.
1350
total_keys = len(keys)
1351
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1352
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1354
cur_non_local = set()
1358
for prefix in prefix_order:
1359
keys = prefix_split_keys[prefix]
1360
non_local = prefix_split_non_local_keys.get(prefix, [])
1362
this_size = self._index._get_total_build_size(keys, positions)
1363
cur_size += this_size
1364
cur_keys.extend(keys)
1365
cur_non_local.update(non_local)
1366
if cur_size > _min_buffer_size:
1367
result.append((cur_keys, cur_non_local))
1368
sizes.append(cur_size)
1370
cur_non_local = set()
1373
result.append((cur_keys, cur_non_local))
1374
sizes.append(cur_size)
1377
def get_record_stream(self, keys, ordering, include_delta_closure):
1378
"""Get a stream of records for keys.
1380
:param keys: The keys to include.
1381
:param ordering: Either 'unordered' or 'topological'. A topologically
1382
sorted stream has compression parents strictly before their
1384
:param include_delta_closure: If True then the closure across any
1385
compression parents will be included (in the opaque data).
1386
:return: An iterator of ContentFactory objects, each of which is only
1387
valid until the iterator is advanced.
1389
# keys might be a generator
1393
if not self._index.has_graph:
1394
# Cannot sort when no graph has been stored.
1395
ordering = 'unordered'
1397
remaining_keys = keys
1400
keys = set(remaining_keys)
1401
for content_factory in self._get_remaining_record_stream(keys,
1402
ordering, include_delta_closure):
1403
remaining_keys.discard(content_factory.key)
1404
yield content_factory
1406
except errors.RetryWithNewPacks, e:
1407
self._access.reload_or_raise(e)
1409
def _get_remaining_record_stream(self, keys, ordering,
1410
include_delta_closure):
1411
"""This function is the 'retry' portion for get_record_stream."""
1412
if include_delta_closure:
1413
positions = self._get_components_positions(keys, allow_missing=True)
1415
build_details = self._index.get_build_details(keys)
1417
# (record_details, access_memo, compression_parent_key)
1418
positions = dict((key, self._build_details_to_components(details))
1419
for key, details in build_details.iteritems())
1420
absent_keys = keys.difference(set(positions))
1421
# There may be more absent keys : if we're missing the basis component
1422
# and are trying to include the delta closure.
1423
# XXX: We should not ever need to examine remote sources because we do
1424
# not permit deltas across versioned files boundaries.
1425
if include_delta_closure:
1426
needed_from_fallback = set()
1427
# Build up reconstructable_keys dict. key:True in this dict means
1428
# the key can be reconstructed.
1429
reconstructable_keys = {}
1433
chain = [key, positions[key][2]]
1435
needed_from_fallback.add(key)
1438
while chain[-1] is not None:
1439
if chain[-1] in reconstructable_keys:
1440
result = reconstructable_keys[chain[-1]]
1444
chain.append(positions[chain[-1]][2])
1446
# missing basis component
1447
needed_from_fallback.add(chain[-1])
1450
for chain_key in chain[:-1]:
1451
reconstructable_keys[chain_key] = result
1453
needed_from_fallback.add(key)
1454
# Double index lookups here : need a unified api ?
1455
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1456
if ordering in ('topological', 'groupcompress'):
1457
if ordering == 'topological':
1458
# Global topological sort
1459
present_keys = tsort.topo_sort(global_map)
1461
present_keys = sort_groupcompress(global_map)
1462
# Now group by source:
1464
current_source = None
1465
for key in present_keys:
1466
for parent_map in parent_maps:
1467
if key in parent_map:
1468
key_source = parent_map
1470
if current_source is not key_source:
1471
source_keys.append((key_source, []))
1472
current_source = key_source
1473
source_keys[-1][1].append(key)
1475
if ordering != 'unordered':
1476
raise AssertionError('valid values for ordering are:'
1477
' "unordered", "groupcompress" or "topological" not: %r'
1479
# Just group by source; remote sources first.
1482
for parent_map in reversed(parent_maps):
1483
source_keys.append((parent_map, []))
1484
for key in parent_map:
1485
present_keys.append(key)
1486
source_keys[-1][1].append(key)
1487
# We have been requested to return these records in an order that
1488
# suits us. So we ask the index to give us an optimally sorted
1490
for source, sub_keys in source_keys:
1491
if source is parent_maps[0]:
1492
# Only sort the keys for this VF
1493
self._index._sort_keys_by_io(sub_keys, positions)
1494
absent_keys = keys - set(global_map)
1495
for key in absent_keys:
1496
yield AbsentContentFactory(key)
1497
# restrict our view to the keys we can answer.
1498
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1499
# XXX: At that point we need to consider the impact of double reads by
1500
# utilising components multiple times.
1501
if include_delta_closure:
1502
# XXX: get_content_maps performs its own index queries; allow state
1504
non_local_keys = needed_from_fallback - absent_keys
1505
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1508
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1511
for record in generator.get_record_stream():
1514
for source, keys in source_keys:
1515
if source is parent_maps[0]:
1516
# this KnitVersionedFiles
1517
records = [(key, positions[key][1]) for key in keys]
1518
for key, raw_data in self._read_records_iter_unchecked(records):
1519
(record_details, index_memo, _) = positions[key]
1520
yield KnitContentFactory(key, global_map[key],
1521
record_details, None, raw_data, self._factory.annotated, None)
1523
vf = self._immediate_fallback_vfs[parent_maps.index(source) - 1]
1524
for record in vf.get_record_stream(keys, ordering,
1525
include_delta_closure):
1528
def get_sha1s(self, keys):
1529
"""See VersionedFiles.get_sha1s()."""
1531
record_map = self._get_record_map(missing, allow_missing=True)
1533
for key, details in record_map.iteritems():
1534
if key not in missing:
1536
# record entry 2 is the 'digest'.
1537
result[key] = details[2]
1538
missing.difference_update(set(result))
1539
for source in self._immediate_fallback_vfs:
1542
new_result = source.get_sha1s(missing)
1543
result.update(new_result)
1544
missing.difference_update(set(new_result))
1547
def insert_record_stream(self, stream):
1548
"""Insert a record stream into this container.
1550
:param stream: A stream of records to insert.
1552
:seealso VersionedFiles.get_record_stream:
1554
def get_adapter(adapter_key):
1556
return adapters[adapter_key]
1558
adapter_factory = adapter_registry.get(adapter_key)
1559
adapter = adapter_factory(self)
1560
adapters[adapter_key] = adapter
1563
if self._factory.annotated:
1564
# self is annotated, we need annotated knits to use directly.
1565
annotated = "annotated-"
1568
# self is not annotated, but we can strip annotations cheaply.
1570
convertibles = set(["knit-annotated-ft-gz"])
1571
if self._max_delta_chain:
1572
delta_types.add("knit-annotated-delta-gz")
1573
convertibles.add("knit-annotated-delta-gz")
1574
# The set of types we can cheaply adapt without needing basis texts.
1575
native_types = set()
1576
if self._max_delta_chain:
1577
native_types.add("knit-%sdelta-gz" % annotated)
1578
delta_types.add("knit-%sdelta-gz" % annotated)
1579
native_types.add("knit-%sft-gz" % annotated)
1580
knit_types = native_types.union(convertibles)
1582
# Buffer all index entries that we can't add immediately because their
1583
# basis parent is missing. We don't buffer all because generating
1584
# annotations may require access to some of the new records. However we
1585
# can't generate annotations from new deltas until their basis parent
1586
# is present anyway, so we get away with not needing an index that
1587
# includes the new keys.
1589
# See <http://launchpad.net/bugs/300177> about ordering of compression
1590
# parents in the records - to be conservative, we insist that all
1591
# parents must be present to avoid expanding to a fulltext.
1593
# key = basis_parent, value = index entry to add
1594
buffered_index_entries = {}
1595
for record in stream:
1596
kind = record.storage_kind
1597
if kind.startswith('knit-') and kind.endswith('-gz'):
1598
# Check that the ID in the header of the raw knit bytes matches
1599
# the record metadata.
1600
raw_data = record._raw_record
1601
df, rec = self._parse_record_header(record.key, raw_data)
1604
parents = record.parents
1605
if record.storage_kind in delta_types:
1606
# TODO: eventually the record itself should track
1607
# compression_parent
1608
compression_parent = parents[0]
1610
compression_parent = None
1611
# Raise an error when a record is missing.
1612
if record.storage_kind == 'absent':
1613
raise RevisionNotPresent([record.key], self)
1614
elif ((record.storage_kind in knit_types)
1615
and (compression_parent is None
1616
or not self._immediate_fallback_vfs
1617
or self._index.has_key(compression_parent)
1618
or not self.has_key(compression_parent))):
1619
# we can insert the knit record literally if either it has no
1620
# compression parent OR we already have its basis in this kvf
1621
# OR the basis is not present even in the fallbacks. In the
1622
# last case it will either turn up later in the stream and all
1623
# will be well, or it won't turn up at all and we'll raise an
1626
# TODO: self.has_key is somewhat redundant with
1627
# self._index.has_key; we really want something that directly
1628
# asks if it's only present in the fallbacks. -- mbp 20081119
1629
if record.storage_kind not in native_types:
1631
adapter_key = (record.storage_kind, "knit-delta-gz")
1632
adapter = get_adapter(adapter_key)
1634
adapter_key = (record.storage_kind, "knit-ft-gz")
1635
adapter = get_adapter(adapter_key)
1636
bytes = adapter.get_bytes(record)
1638
# It's a knit record, it has a _raw_record field (even if
1639
# it was reconstituted from a network stream).
1640
bytes = record._raw_record
1641
options = [record._build_details[0]]
1642
if record._build_details[1]:
1643
options.append('no-eol')
1644
# Just blat it across.
1645
# Note: This does end up adding data on duplicate keys. As
1646
# modern repositories use atomic insertions this should not
1647
# lead to excessive growth in the event of interrupted fetches.
1648
# 'knit' repositories may suffer excessive growth, but as a
1649
# deprecated format this is tolerable. It can be fixed if
1650
# needed by in the kndx index support raising on a duplicate
1651
# add with identical parents and options.
1652
access_memo = self._access.add_raw_records(
1653
[(record.key, len(bytes))], bytes)[0]
1654
index_entry = (record.key, options, access_memo, parents)
1655
if 'fulltext' not in options:
1656
# Not a fulltext, so we need to make sure the compression
1657
# parent will also be present.
1658
# Note that pack backed knits don't need to buffer here
1659
# because they buffer all writes to the transaction level,
1660
# but we don't expose that difference at the index level. If
1661
# the query here has sufficient cost to show up in
1662
# profiling we should do that.
1664
# They're required to be physically in this
1665
# KnitVersionedFiles, not in a fallback.
1666
if not self._index.has_key(compression_parent):
1667
pending = buffered_index_entries.setdefault(
1668
compression_parent, [])
1669
pending.append(index_entry)
1672
self._index.add_records([index_entry])
1673
elif record.storage_kind == 'chunked':
1674
self.add_lines(record.key, parents,
1675
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1677
# Not suitable for direct insertion as a
1678
# delta, either because it's not the right format, or this
1679
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1680
# 0) or because it depends on a base only present in the
1682
self._access.flush()
1684
# Try getting a fulltext directly from the record.
1685
bytes = record.get_bytes_as('fulltext')
1686
except errors.UnavailableRepresentation:
1687
adapter_key = record.storage_kind, 'fulltext'
1688
adapter = get_adapter(adapter_key)
1689
bytes = adapter.get_bytes(record)
1690
lines = split_lines(bytes)
1692
self.add_lines(record.key, parents, lines)
1693
except errors.RevisionAlreadyPresent:
1695
# Add any records whose basis parent is now available.
1697
added_keys = [record.key]
1699
key = added_keys.pop(0)
1700
if key in buffered_index_entries:
1701
index_entries = buffered_index_entries[key]
1702
self._index.add_records(index_entries)
1704
[index_entry[0] for index_entry in index_entries])
1705
del buffered_index_entries[key]
1706
if buffered_index_entries:
1707
# There were index entries buffered at the end of the stream,
1708
# So these need to be added (if the index supports holding such
1709
# entries for later insertion)
1711
for key in buffered_index_entries:
1712
index_entries = buffered_index_entries[key]
1713
all_entries.extend(index_entries)
1714
self._index.add_records(
1715
all_entries, missing_compression_parents=True)
1717
def get_missing_compression_parent_keys(self):
1718
"""Return an iterable of keys of missing compression parents.
1720
Check this after calling insert_record_stream to find out if there are
1721
any missing compression parents. If there are, the records that
1722
depend on them are not able to be inserted safely. For atomic
1723
KnitVersionedFiles built on packs, the transaction should be aborted or
1724
suspended - commit will fail at this point. Nonatomic knits will error
1725
earlier because they have no staging area to put pending entries into.
1727
return self._index.get_missing_compression_parents()
1729
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1730
"""Iterate over the lines in the versioned files from keys.
1732
This may return lines from other keys. Each item the returned
1733
iterator yields is a tuple of a line and a text version that that line
1734
is present in (not introduced in).
1736
Ordering of results is in whatever order is most suitable for the
1737
underlying storage format.
1739
If a progress bar is supplied, it may be used to indicate progress.
1740
The caller is responsible for cleaning up progress bars (because this
1744
* Lines are normalised by the underlying store: they will all have \\n
1746
* Lines are returned in arbitrary order.
1747
* If a requested key did not change any lines (or didn't have any
1748
lines), it may not be mentioned at all in the result.
1750
:param pb: Progress bar supplied by caller.
1751
:return: An iterator over (line, key).
1754
pb = ui.ui_factory.nested_progress_bar()
1760
# we don't care about inclusions, the caller cares.
1761
# but we need to setup a list of records to visit.
1762
# we need key, position, length
1764
build_details = self._index.get_build_details(keys)
1765
for key, details in build_details.iteritems():
1767
key_records.append((key, details[0]))
1768
records_iter = enumerate(self._read_records_iter(key_records))
1769
for (key_idx, (key, data, sha_value)) in records_iter:
1770
pb.update('Walking content', key_idx, total)
1771
compression_parent = build_details[key][1]
1772
if compression_parent is None:
1774
line_iterator = self._factory.get_fulltext_content(data)
1777
line_iterator = self._factory.get_linedelta_content(data)
1778
# Now that we are yielding the data for this key, remove it
1781
# XXX: It might be more efficient to yield (key,
1782
# line_iterator) in the future. However for now, this is a
1783
# simpler change to integrate into the rest of the
1784
# codebase. RBC 20071110
1785
for line in line_iterator:
1788
except errors.RetryWithNewPacks, e:
1789
self._access.reload_or_raise(e)
1790
# If there are still keys we've not yet found, we look in the fallback
1791
# vfs, and hope to find them there. Note that if the keys are found
1792
# but had no changes or no content, the fallback may not return
1794
if keys and not self._immediate_fallback_vfs:
1795
# XXX: strictly the second parameter is meant to be the file id
1796
# but it's not easily accessible here.
1797
raise RevisionNotPresent(keys, repr(self))
1798
for source in self._immediate_fallback_vfs:
1802
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1803
source_keys.add(key)
1805
keys.difference_update(source_keys)
1806
pb.update('Walking content', total, total)
1808
def _make_line_delta(self, delta_seq, new_content):
1809
"""Generate a line delta from delta_seq and new_content."""
1811
for op in delta_seq.get_opcodes():
1812
if op[0] == 'equal':
1814
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1817
def _merge_annotations(self, content, parents, parent_texts={},
1818
delta=None, annotated=None,
1819
left_matching_blocks=None):
1820
"""Merge annotations for content and generate deltas.
1822
This is done by comparing the annotations based on changes to the text
1823
and generating a delta on the resulting full texts. If annotations are
1824
not being created then a simple delta is created.
1826
if left_matching_blocks is not None:
1827
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1831
for parent_key in parents:
1832
merge_content = self._get_content(parent_key, parent_texts)
1833
if (parent_key == parents[0] and delta_seq is not None):
1836
seq = patiencediff.PatienceSequenceMatcher(
1837
None, merge_content.text(), content.text())
1838
for i, j, n in seq.get_matching_blocks():
1841
# this copies (origin, text) pairs across to the new
1842
# content for any line that matches the last-checked
1844
content._lines[j:j+n] = merge_content._lines[i:i+n]
1845
# XXX: Robert says the following block is a workaround for a
1846
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1847
if content._lines and content._lines[-1][1][-1] != '\n':
1848
# The copied annotation was from a line without a trailing EOL,
1849
# reinstate one for the content object, to ensure correct
1851
line = content._lines[-1][1] + '\n'
1852
content._lines[-1] = (content._lines[-1][0], line)
1854
if delta_seq is None:
1855
reference_content = self._get_content(parents[0], parent_texts)
1856
new_texts = content.text()
1857
old_texts = reference_content.text()
1858
delta_seq = patiencediff.PatienceSequenceMatcher(
1859
None, old_texts, new_texts)
1860
return self._make_line_delta(delta_seq, content)
1862
def _parse_record(self, version_id, data):
1863
"""Parse an original format knit record.
1865
These have the last element of the key only present in the stored data.
1867
rec, record_contents = self._parse_record_unchecked(data)
1868
self._check_header_version(rec, version_id)
1869
return record_contents, rec[3]
1871
def _parse_record_header(self, key, raw_data):
1872
"""Parse a record header for consistency.
1874
:return: the header and the decompressor stream.
1875
as (stream, header_record)
1877
df = gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1880
rec = self._check_header(key, df.readline())
1881
except Exception, e:
1882
raise KnitCorrupt(self,
1883
"While reading {%s} got %s(%s)"
1884
% (key, e.__class__.__name__, str(e)))
1887
def _parse_record_unchecked(self, data):
1889
# 4168 calls in 2880 217 internal
1890
# 4168 calls to _parse_record_header in 2121
1891
# 4168 calls to readlines in 330
1892
df = gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1894
record_contents = df.readlines()
1895
except Exception, e:
1896
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1897
(data, e.__class__.__name__, str(e)))
1898
header = record_contents.pop(0)
1899
rec = self._split_header(header)
1900
last_line = record_contents.pop()
1901
if len(record_contents) != int(rec[2]):
1902
raise KnitCorrupt(self,
1903
'incorrect number of lines %s != %s'
1904
' for version {%s} %s'
1905
% (len(record_contents), int(rec[2]),
1906
rec[1], record_contents))
1907
if last_line != 'end %s\n' % rec[1]:
1908
raise KnitCorrupt(self,
1909
'unexpected version end line %r, wanted %r'
1910
% (last_line, rec[1]))
1912
return rec, record_contents
1914
def _read_records_iter(self, records):
1915
"""Read text records from data file and yield result.
1917
The result will be returned in whatever is the fastest to read.
1918
Not by the order requested. Also, multiple requests for the same
1919
record will only yield 1 response.
1920
:param records: A list of (key, access_memo) entries
1921
:return: Yields (key, contents, digest) in the order
1922
read, not the order requested
1927
# XXX: This smells wrong, IO may not be getting ordered right.
1928
needed_records = sorted(set(records), key=operator.itemgetter(1))
1929
if not needed_records:
1932
# The transport optimizes the fetching as well
1933
# (ie, reads continuous ranges.)
1934
raw_data = self._access.get_raw_records(
1935
[index_memo for key, index_memo in needed_records])
1937
for (key, index_memo), data in \
1938
izip(iter(needed_records), raw_data):
1939
content, digest = self._parse_record(key[-1], data)
1940
yield key, content, digest
1942
def _read_records_iter_raw(self, records):
1943
"""Read text records from data file and yield raw data.
1945
This unpacks enough of the text record to validate the id is
1946
as expected but thats all.
1948
Each item the iterator yields is (key, bytes,
1949
expected_sha1_of_full_text).
1951
for key, data in self._read_records_iter_unchecked(records):
1952
# validate the header (note that we can only use the suffix in
1953
# current knit records).
1954
df, rec = self._parse_record_header(key, data)
1956
yield key, data, rec[3]
1958
def _read_records_iter_unchecked(self, records):
1959
"""Read text records from data file and yield raw data.
1961
No validation is done.
1963
Yields tuples of (key, data).
1965
# setup an iterator of the external records:
1966
# uses readv so nice and fast we hope.
1968
# grab the disk data needed.
1969
needed_offsets = [index_memo for key, index_memo
1971
raw_records = self._access.get_raw_records(needed_offsets)
1973
for key, index_memo in records:
1974
data = raw_records.next()
1977
def _record_to_data(self, key, digest, lines, dense_lines=None):
1978
"""Convert key, digest, lines into a raw data block.
1980
:param key: The key of the record. Currently keys are always serialised
1981
using just the trailing component.
1982
:param dense_lines: The bytes of lines but in a denser form. For
1983
instance, if lines is a list of 1000 bytestrings each ending in \n,
1984
dense_lines may be a list with one line in it, containing all the
1985
1000's lines and their \n's. Using dense_lines if it is already
1986
known is a win because the string join to create bytes in this
1987
function spends less time resizing the final string.
1988
:return: (len, a StringIO instance with the raw data ready to read.)
1990
chunks = ["version %s %d %s\n" % (key[-1], len(lines), digest)]
1991
chunks.extend(dense_lines or lines)
1992
chunks.append("end %s\n" % key[-1])
1993
for chunk in chunks:
1994
if type(chunk) is not str:
1995
raise AssertionError(
1996
'data must be plain bytes was %s' % type(chunk))
1997
if lines and lines[-1][-1] != '\n':
1998
raise ValueError('corrupt lines value %r' % lines)
1999
compressed_bytes = tuned_gzip.chunks_to_gzip(chunks)
2000
return len(compressed_bytes), compressed_bytes
2002
def _split_header(self, line):
2005
raise KnitCorrupt(self,
2006
'unexpected number of elements in record header')
2010
"""See VersionedFiles.keys."""
2011
if 'evil' in debug.debug_flags:
2012
trace.mutter_callsite(2, "keys scales with size of history")
2013
sources = [self._index] + self._immediate_fallback_vfs
2015
for source in sources:
2016
result.update(source.keys())
2020
class _ContentMapGenerator(object):
2021
"""Generate texts or expose raw deltas for a set of texts."""
2023
def __init__(self, ordering='unordered'):
2024
self._ordering = ordering
2026
def _get_content(self, key):
2027
"""Get the content object for key."""
2028
# Note that _get_content is only called when the _ContentMapGenerator
2029
# has been constructed with just one key requested for reconstruction.
2030
if key in self.nonlocal_keys:
2031
record = self.get_record_stream().next()
2032
# Create a content object on the fly
2033
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
2034
return PlainKnitContent(lines, record.key)
2036
# local keys we can ask for directly
2037
return self._get_one_work(key)
2039
def get_record_stream(self):
2040
"""Get a record stream for the keys requested during __init__."""
2041
for record in self._work():
2045
"""Produce maps of text and KnitContents as dicts.
2047
:return: (text_map, content_map) where text_map contains the texts for
2048
the requested versions and content_map contains the KnitContents.
2050
# NB: By definition we never need to read remote sources unless texts
2051
# are requested from them: we don't delta across stores - and we
2052
# explicitly do not want to to prevent data loss situations.
2053
if self.global_map is None:
2054
self.global_map = self.vf.get_parent_map(self.keys)
2055
nonlocal_keys = self.nonlocal_keys
2057
missing_keys = set(nonlocal_keys)
2058
# Read from remote versioned file instances and provide to our caller.
2059
for source in self.vf._immediate_fallback_vfs:
2060
if not missing_keys:
2062
# Loop over fallback repositories asking them for texts - ignore
2063
# any missing from a particular fallback.
2064
for record in source.get_record_stream(missing_keys,
2065
self._ordering, True):
2066
if record.storage_kind == 'absent':
2067
# Not in thie particular stream, may be in one of the
2068
# other fallback vfs objects.
2070
missing_keys.remove(record.key)
2073
if self._raw_record_map is None:
2074
raise AssertionError('_raw_record_map should have been filled')
2076
for key in self.keys:
2077
if key in self.nonlocal_keys:
2079
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2082
def _get_one_work(self, requested_key):
2083
# Now, if we have calculated everything already, just return the
2085
if requested_key in self._contents_map:
2086
return self._contents_map[requested_key]
2087
# To simplify things, parse everything at once - code that wants one text
2088
# probably wants them all.
2089
# FUTURE: This function could be improved for the 'extract many' case
2090
# by tracking each component and only doing the copy when the number of
2091
# children than need to apply delta's to it is > 1 or it is part of the
2093
multiple_versions = len(self.keys) != 1
2094
if self._record_map is None:
2095
self._record_map = self.vf._raw_map_to_record_map(
2096
self._raw_record_map)
2097
record_map = self._record_map
2098
# raw_record_map is key:
2099
# Have read and parsed records at this point.
2100
for key in self.keys:
2101
if key in self.nonlocal_keys:
2106
while cursor is not None:
2108
record, record_details, digest, next = record_map[cursor]
2110
raise RevisionNotPresent(cursor, self)
2111
components.append((cursor, record, record_details, digest))
2113
if cursor in self._contents_map:
2114
# no need to plan further back
2115
components.append((cursor, None, None, None))
2119
for (component_id, record, record_details,
2120
digest) in reversed(components):
2121
if component_id in self._contents_map:
2122
content = self._contents_map[component_id]
2124
content, delta = self._factory.parse_record(key[-1],
2125
record, record_details, content,
2126
copy_base_content=multiple_versions)
2127
if multiple_versions:
2128
self._contents_map[component_id] = content
2130
# digest here is the digest from the last applied component.
2131
text = content.text()
2132
actual_sha = sha_strings(text)
2133
if actual_sha != digest:
2134
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2135
if multiple_versions:
2136
return self._contents_map[requested_key]
2140
def _wire_bytes(self):
2141
"""Get the bytes to put on the wire for 'key'.
2143
The first collection of bytes asked for returns the serialised
2144
raw_record_map and the additional details (key, parent) for key.
2145
Subsequent calls return just the additional details (key, parent).
2146
The wire storage_kind given for the first key is 'knit-delta-closure',
2147
For subsequent keys it is 'knit-delta-closure-ref'.
2149
:param key: A key from the content generator.
2150
:return: Bytes to put on the wire.
2153
# kind marker for dispatch on the far side,
2154
lines.append('knit-delta-closure')
2156
if self.vf._factory.annotated:
2157
lines.append('annotated')
2160
# then the list of keys
2161
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2162
if key not in self.nonlocal_keys]))
2163
# then the _raw_record_map in serialised form:
2165
# for each item in the map:
2167
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2168
# one line with method
2169
# one line with noeol
2170
# one line with next ('' for None)
2171
# one line with byte count of the record bytes
2173
for key, (record_bytes, (method, noeol), next) in \
2174
self._raw_record_map.iteritems():
2175
key_bytes = '\x00'.join(key)
2176
parents = self.global_map.get(key, None)
2178
parent_bytes = 'None:'
2180
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2181
method_bytes = method
2187
next_bytes = '\x00'.join(next)
2190
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2191
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2192
len(record_bytes), record_bytes))
2193
map_bytes = ''.join(map_byte_list)
2194
lines.append(map_bytes)
2195
bytes = '\n'.join(lines)
2199
class _VFContentMapGenerator(_ContentMapGenerator):
2200
"""Content map generator reading from a VersionedFiles object."""
2202
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2203
global_map=None, raw_record_map=None, ordering='unordered'):
2204
"""Create a _ContentMapGenerator.
2206
:param versioned_files: The versioned files that the texts are being
2208
:param keys: The keys to produce content maps for.
2209
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2210
which are known to not be in this knit, but rather in one of the
2212
:param global_map: The result of get_parent_map(keys) (or a supermap).
2213
This is required if get_record_stream() is to be used.
2214
:param raw_record_map: A unparsed raw record map to use for answering
2217
_ContentMapGenerator.__init__(self, ordering=ordering)
2218
# The vf to source data from
2219
self.vf = versioned_files
2221
self.keys = list(keys)
2222
# Keys known to be in fallback vfs objects
2223
if nonlocal_keys is None:
2224
self.nonlocal_keys = set()
2226
self.nonlocal_keys = frozenset(nonlocal_keys)
2227
# Parents data for keys to be returned in get_record_stream
2228
self.global_map = global_map
2229
# The chunked lists for self.keys in text form
2231
# A cache of KnitContent objects used in extracting texts.
2232
self._contents_map = {}
2233
# All the knit records needed to assemble the requested keys as full
2235
self._record_map = None
2236
if raw_record_map is None:
2237
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2240
self._raw_record_map = raw_record_map
2241
# the factory for parsing records
2242
self._factory = self.vf._factory
2245
class _NetworkContentMapGenerator(_ContentMapGenerator):
2246
"""Content map generator sourced from a network stream."""
2248
def __init__(self, bytes, line_end):
2249
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2251
self.global_map = {}
2252
self._raw_record_map = {}
2253
self._contents_map = {}
2254
self._record_map = None
2255
self.nonlocal_keys = []
2256
# Get access to record parsing facilities
2257
self.vf = KnitVersionedFiles(None, None)
2260
line_end = bytes.find('\n', start)
2261
line = bytes[start:line_end]
2262
start = line_end + 1
2263
if line == 'annotated':
2264
self._factory = KnitAnnotateFactory()
2266
self._factory = KnitPlainFactory()
2267
# list of keys to emit in get_record_stream
2268
line_end = bytes.find('\n', start)
2269
line = bytes[start:line_end]
2270
start = line_end + 1
2272
tuple(segment.split('\x00')) for segment in line.split('\t')
2274
# now a loop until the end. XXX: It would be nice if this was just a
2275
# bunch of the same records as get_record_stream(..., False) gives, but
2276
# there is a decent sized gap stopping that at the moment.
2280
line_end = bytes.find('\n', start)
2281
key = tuple(bytes[start:line_end].split('\x00'))
2282
start = line_end + 1
2283
# 1 line with parents (None: for None, '' for ())
2284
line_end = bytes.find('\n', start)
2285
line = bytes[start:line_end]
2290
[tuple(segment.split('\x00')) for segment in line.split('\t')
2292
self.global_map[key] = parents
2293
start = line_end + 1
2294
# one line with method
2295
line_end = bytes.find('\n', start)
2296
line = bytes[start:line_end]
2298
start = line_end + 1
2299
# one line with noeol
2300
line_end = bytes.find('\n', start)
2301
line = bytes[start:line_end]
2303
start = line_end + 1
2304
# one line with next ('' for None)
2305
line_end = bytes.find('\n', start)
2306
line = bytes[start:line_end]
2310
next = tuple(bytes[start:line_end].split('\x00'))
2311
start = line_end + 1
2312
# one line with byte count of the record bytes
2313
line_end = bytes.find('\n', start)
2314
line = bytes[start:line_end]
2316
start = line_end + 1
2318
record_bytes = bytes[start:start+count]
2319
start = start + count
2321
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2323
def get_record_stream(self):
2324
"""Get a record stream for for keys requested by the bytestream."""
2326
for key in self.keys:
2327
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2330
def _wire_bytes(self):
2334
class _KndxIndex(object):
2335
"""Manages knit index files
2337
The index is kept in memory and read on startup, to enable
2338
fast lookups of revision information. The cursor of the index
2339
file is always pointing to the end, making it easy to append
2342
_cache is a cache for fast mapping from version id to a Index
2345
_history is a cache for fast mapping from indexes to version ids.
2347
The index data format is dictionary compressed when it comes to
2348
parent references; a index entry may only have parents that with a
2349
lover index number. As a result, the index is topological sorted.
2351
Duplicate entries may be written to the index for a single version id
2352
if this is done then the latter one completely replaces the former:
2353
this allows updates to correct version and parent information.
2354
Note that the two entries may share the delta, and that successive
2355
annotations and references MUST point to the first entry.
2357
The index file on disc contains a header, followed by one line per knit
2358
record. The same revision can be present in an index file more than once.
2359
The first occurrence gets assigned a sequence number starting from 0.
2361
The format of a single line is
2362
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
2363
REVISION_ID is a utf8-encoded revision id
2364
FLAGS is a comma separated list of flags about the record. Values include
2365
no-eol, line-delta, fulltext.
2366
BYTE_OFFSET is the ascii representation of the byte offset in the data file
2367
that the compressed data starts at.
2368
LENGTH is the ascii representation of the length of the data file.
2369
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
2371
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
2372
revision id already in the knit that is a parent of REVISION_ID.
2373
The ' :' marker is the end of record marker.
2376
when a write is interrupted to the index file, it will result in a line
2377
that does not end in ' :'. If the ' :' is not present at the end of a line,
2378
or at the end of the file, then the record that is missing it will be
2379
ignored by the parser.
2381
When writing new records to the index file, the data is preceded by '\n'
2382
to ensure that records always start on new lines even if the last write was
2383
interrupted. As a result its normal for the last line in the index to be
2384
missing a trailing newline. One can be added with no harmful effects.
2386
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2387
where prefix is e.g. the (fileid,) for .texts instances or () for
2388
constant-mapped things like .revisions, and the old state is
2389
tuple(history_vector, cache_dict). This is used to prevent having an
2390
ABI change with the C extension that reads .kndx files.
2393
HEADER = "# bzr knit index 8\n"
2395
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2396
"""Create a _KndxIndex on transport using mapper."""
2397
self._transport = transport
2398
self._mapper = mapper
2399
self._get_scope = get_scope
2400
self._allow_writes = allow_writes
2401
self._is_locked = is_locked
2403
self.has_graph = True
2405
def add_records(self, records, random_id=False, missing_compression_parents=False):
2406
"""Add multiple records to the index.
2408
:param records: a list of tuples:
2409
(key, options, access_memo, parents).
2410
:param random_id: If True the ids being added were randomly generated
2411
and no check for existence will be performed.
2412
:param missing_compression_parents: If True the records being added are
2413
only compressed against texts already in the index (or inside
2414
records). If False the records all refer to unavailable texts (or
2415
texts inside records) as compression parents.
2417
if missing_compression_parents:
2418
# It might be nice to get the edge of the records. But keys isn't
2420
keys = sorted(record[0] for record in records)
2421
raise errors.RevisionNotPresent(keys, self)
2423
for record in records:
2426
path = self._mapper.map(key) + '.kndx'
2427
path_keys = paths.setdefault(path, (prefix, []))
2428
path_keys[1].append(record)
2429
for path in sorted(paths):
2430
prefix, path_keys = paths[path]
2431
self._load_prefixes([prefix])
2433
orig_history = self._kndx_cache[prefix][1][:]
2434
orig_cache = self._kndx_cache[prefix][0].copy()
2437
for key, options, (_, pos, size), parents in path_keys:
2439
# kndx indices cannot be parentless.
2441
line = "\n%s %s %s %s %s :" % (
2442
key[-1], ','.join(options), pos, size,
2443
self._dictionary_compress(parents))
2444
if type(line) is not str:
2445
raise AssertionError(
2446
'data must be utf8 was %s' % type(line))
2448
self._cache_key(key, options, pos, size, parents)
2449
if len(orig_history):
2450
self._transport.append_bytes(path, ''.join(lines))
2452
self._init_index(path, lines)
2454
# If any problems happen, restore the original values and re-raise
2455
self._kndx_cache[prefix] = (orig_cache, orig_history)
2458
def scan_unvalidated_index(self, graph_index):
2459
"""See _KnitGraphIndex.scan_unvalidated_index."""
2460
# Because kndx files do not support atomic insertion via separate index
2461
# files, they do not support this method.
2462
raise NotImplementedError(self.scan_unvalidated_index)
2464
def get_missing_compression_parents(self):
2465
"""See _KnitGraphIndex.get_missing_compression_parents."""
2466
# Because kndx files do not support atomic insertion via separate index
2467
# files, they do not support this method.
2468
raise NotImplementedError(self.get_missing_compression_parents)
2470
def _cache_key(self, key, options, pos, size, parent_keys):
2471
"""Cache a version record in the history array and index cache.
2473
This is inlined into _load_data for performance. KEEP IN SYNC.
2474
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2478
version_id = key[-1]
2479
# last-element only for compatibilty with the C load_data.
2480
parents = tuple(parent[-1] for parent in parent_keys)
2481
for parent in parent_keys:
2482
if parent[:-1] != prefix:
2483
raise ValueError("mismatched prefixes for %r, %r" % (
2485
cache, history = self._kndx_cache[prefix]
2486
# only want the _history index to reference the 1st index entry
2488
if version_id not in cache:
2489
index = len(history)
2490
history.append(version_id)
2492
index = cache[version_id][5]
2493
cache[version_id] = (version_id,
2500
def check_header(self, fp):
2501
line = fp.readline()
2503
# An empty file can actually be treated as though the file doesn't
2505
raise errors.NoSuchFile(self)
2506
if line != self.HEADER:
2507
raise KnitHeaderError(badline=line, filename=self)
2509
def _check_read(self):
2510
if not self._is_locked():
2511
raise errors.ObjectNotLocked(self)
2512
if self._get_scope() != self._scope:
2515
def _check_write_ok(self):
2516
"""Assert if not writes are permitted."""
2517
if not self._is_locked():
2518
raise errors.ObjectNotLocked(self)
2519
if self._get_scope() != self._scope:
2521
if self._mode != 'w':
2522
raise errors.ReadOnlyObjectDirtiedError(self)
2524
def get_build_details(self, keys):
2525
"""Get the method, index_memo and compression parent for keys.
2527
Ghosts are omitted from the result.
2529
:param keys: An iterable of keys.
2530
:return: A dict of key:(index_memo, compression_parent, parents,
2533
opaque structure to pass to read_records to extract the raw
2536
Content that this record is built upon, may be None
2538
Logical parents of this node
2540
extra information about the content which needs to be passed to
2541
Factory.parse_record
2543
parent_map = self.get_parent_map(keys)
2546
if key not in parent_map:
2548
method = self.get_method(key)
2549
parents = parent_map[key]
2550
if method == 'fulltext':
2551
compression_parent = None
2553
compression_parent = parents[0]
2554
noeol = 'no-eol' in self.get_options(key)
2555
index_memo = self.get_position(key)
2556
result[key] = (index_memo, compression_parent,
2557
parents, (method, noeol))
2560
def get_method(self, key):
2561
"""Return compression method of specified key."""
2562
options = self.get_options(key)
2563
if 'fulltext' in options:
2565
elif 'line-delta' in options:
2568
raise errors.KnitIndexUnknownMethod(self, options)
2570
def get_options(self, key):
2571
"""Return a list representing options.
2575
prefix, suffix = self._split_key(key)
2576
self._load_prefixes([prefix])
2578
return self._kndx_cache[prefix][0][suffix][1]
2580
raise RevisionNotPresent(key, self)
2582
def find_ancestry(self, keys):
2583
"""See CombinedGraphIndex.find_ancestry()"""
2584
prefixes = set(key[:-1] for key in keys)
2585
self._load_prefixes(prefixes)
2588
missing_keys = set()
2589
pending_keys = list(keys)
2590
# This assumes that keys will not reference parents in a different
2591
# prefix, which is accurate so far.
2593
key = pending_keys.pop()
2594
if key in parent_map:
2598
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2600
missing_keys.add(key)
2602
parent_keys = tuple([prefix + (suffix,)
2603
for suffix in suffix_parents])
2604
parent_map[key] = parent_keys
2605
pending_keys.extend([p for p in parent_keys
2606
if p not in parent_map])
2607
return parent_map, missing_keys
2609
def get_parent_map(self, keys):
2610
"""Get a map of the parents of keys.
2612
:param keys: The keys to look up parents for.
2613
:return: A mapping from keys to parents. Absent keys are absent from
2616
# Parse what we need to up front, this potentially trades off I/O
2617
# locality (.kndx and .knit in the same block group for the same file
2618
# id) for less checking in inner loops.
2619
prefixes = set(key[:-1] for key in keys)
2620
self._load_prefixes(prefixes)
2625
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2629
result[key] = tuple(prefix + (suffix,) for
2630
suffix in suffix_parents)
2633
def get_position(self, key):
2634
"""Return details needed to access the version.
2636
:return: a tuple (key, data position, size) to hand to the access
2637
logic to get the record.
2639
prefix, suffix = self._split_key(key)
2640
self._load_prefixes([prefix])
2641
entry = self._kndx_cache[prefix][0][suffix]
2642
return key, entry[2], entry[3]
2644
has_key = _mod_index._has_key_from_parent_map
2646
def _init_index(self, path, extra_lines=[]):
2647
"""Initialize an index."""
2649
sio.write(self.HEADER)
2650
sio.writelines(extra_lines)
2652
self._transport.put_file_non_atomic(path, sio,
2653
create_parent_dir=True)
2654
# self._create_parent_dir)
2655
# mode=self._file_mode,
2656
# dir_mode=self._dir_mode)
2659
"""Get all the keys in the collection.
2661
The keys are not ordered.
2664
# Identify all key prefixes.
2665
# XXX: A bit hacky, needs polish.
2666
if type(self._mapper) is ConstantMapper:
2670
for quoted_relpath in self._transport.iter_files_recursive():
2671
path, ext = os.path.splitext(quoted_relpath)
2673
prefixes = [self._mapper.unmap(path) for path in relpaths]
2674
self._load_prefixes(prefixes)
2675
for prefix in prefixes:
2676
for suffix in self._kndx_cache[prefix][1]:
2677
result.add(prefix + (suffix,))
2680
def _load_prefixes(self, prefixes):
2681
"""Load the indices for prefixes."""
2683
for prefix in prefixes:
2684
if prefix not in self._kndx_cache:
2685
# the load_data interface writes to these variables.
2688
self._filename = prefix
2690
path = self._mapper.map(prefix) + '.kndx'
2691
fp = self._transport.get(path)
2693
# _load_data may raise NoSuchFile if the target knit is
2695
_load_data(self, fp)
2698
self._kndx_cache[prefix] = (self._cache, self._history)
2703
self._kndx_cache[prefix] = ({}, [])
2704
if type(self._mapper) is ConstantMapper:
2705
# preserve behaviour for revisions.kndx etc.
2706
self._init_index(path)
2711
missing_keys = _mod_index._missing_keys_from_parent_map
2713
def _partition_keys(self, keys):
2714
"""Turn keys into a dict of prefix:suffix_list."""
2717
prefix_keys = result.setdefault(key[:-1], [])
2718
prefix_keys.append(key[-1])
2721
def _dictionary_compress(self, keys):
2722
"""Dictionary compress keys.
2724
:param keys: The keys to generate references to.
2725
:return: A string representation of keys. keys which are present are
2726
dictionary compressed, and others are emitted as fulltext with a
2732
prefix = keys[0][:-1]
2733
cache = self._kndx_cache[prefix][0]
2735
if key[:-1] != prefix:
2736
# kndx indices cannot refer across partitioned storage.
2737
raise ValueError("mismatched prefixes for %r" % keys)
2738
if key[-1] in cache:
2739
# -- inlined lookup() --
2740
result_list.append(str(cache[key[-1]][5]))
2741
# -- end lookup () --
2743
result_list.append('.' + key[-1])
2744
return ' '.join(result_list)
2746
def _reset_cache(self):
2747
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2748
# (cache_dict, history_vector) for parsed kndx files.
2749
self._kndx_cache = {}
2750
self._scope = self._get_scope()
2751
allow_writes = self._allow_writes()
2757
def _sort_keys_by_io(self, keys, positions):
2758
"""Figure out an optimal order to read the records for the given keys.
2760
Sort keys, grouped by index and sorted by position.
2762
:param keys: A list of keys whose records we want to read. This will be
2764
:param positions: A dict, such as the one returned by
2765
_get_components_positions()
2768
def get_sort_key(key):
2769
index_memo = positions[key][1]
2770
# Group by prefix and position. index_memo[0] is the key, so it is
2771
# (file_id, revision_id) and we don't want to sort on revision_id,
2772
# index_memo[1] is the position, and index_memo[2] is the size,
2773
# which doesn't matter for the sort
2774
return index_memo[0][:-1], index_memo[1]
2775
return keys.sort(key=get_sort_key)
2777
_get_total_build_size = _get_total_build_size
2779
def _split_key(self, key):
2780
"""Split key into a prefix and suffix."""
2781
return key[:-1], key[-1]
2784
class _KeyRefs(object):
2786
def __init__(self, track_new_keys=False):
2787
# dict mapping 'key' to 'set of keys referring to that key'
2790
# set remembering all new keys
2791
self.new_keys = set()
2793
self.new_keys = None
2799
self.new_keys.clear()
2801
def add_references(self, key, refs):
2802
# Record the new references
2803
for referenced in refs:
2805
needed_by = self.refs[referenced]
2807
needed_by = self.refs[referenced] = set()
2809
# Discard references satisfied by the new key
2812
def get_new_keys(self):
2813
return self.new_keys
2815
def get_unsatisfied_refs(self):
2816
return self.refs.iterkeys()
2818
def _satisfy_refs_for_key(self, key):
2822
# No keys depended on this key. That's ok.
2825
def add_key(self, key):
2826
# satisfy refs for key, and remember that we've seen this key.
2827
self._satisfy_refs_for_key(key)
2828
if self.new_keys is not None:
2829
self.new_keys.add(key)
2831
def satisfy_refs_for_keys(self, keys):
2833
self._satisfy_refs_for_key(key)
2835
def get_referrers(self):
2837
for referrers in self.refs.itervalues():
2838
result.update(referrers)
2842
class _KnitGraphIndex(object):
2843
"""A KnitVersionedFiles index layered on GraphIndex."""
2845
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2846
add_callback=None, track_external_parent_refs=False):
2847
"""Construct a KnitGraphIndex on a graph_index.
2849
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2850
:param is_locked: A callback to check whether the object should answer
2852
:param deltas: Allow delta-compressed records.
2853
:param parents: If True, record knits parents, if not do not record
2855
:param add_callback: If not None, allow additions to the index and call
2856
this callback with a list of added GraphIndex nodes:
2857
[(node, value, node_refs), ...]
2858
:param is_locked: A callback, returns True if the index is locked and
2860
:param track_external_parent_refs: If True, record all external parent
2861
references parents from added records. These can be retrieved
2862
later by calling get_missing_parents().
2864
self._add_callback = add_callback
2865
self._graph_index = graph_index
2866
self._deltas = deltas
2867
self._parents = parents
2868
if deltas and not parents:
2869
# XXX: TODO: Delta tree and parent graph should be conceptually
2871
raise KnitCorrupt(self, "Cannot do delta compression without "
2873
self.has_graph = parents
2874
self._is_locked = is_locked
2875
self._missing_compression_parents = set()
2876
if track_external_parent_refs:
2877
self._key_dependencies = _KeyRefs()
2879
self._key_dependencies = None
2882
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2884
def add_records(self, records, random_id=False,
2885
missing_compression_parents=False):
2886
"""Add multiple records to the index.
2888
This function does not insert data into the Immutable GraphIndex
2889
backing the KnitGraphIndex, instead it prepares data for insertion by
2890
the caller and checks that it is safe to insert then calls
2891
self._add_callback with the prepared GraphIndex nodes.
2893
:param records: a list of tuples:
2894
(key, options, access_memo, parents).
2895
:param random_id: If True the ids being added were randomly generated
2896
and no check for existence will be performed.
2897
:param missing_compression_parents: If True the records being added are
2898
only compressed against texts already in the index (or inside
2899
records). If False the records all refer to unavailable texts (or
2900
texts inside records) as compression parents.
2902
if not self._add_callback:
2903
raise errors.ReadOnlyError(self)
2904
# we hope there are no repositories with inconsistent parentage
2908
compression_parents = set()
2909
key_dependencies = self._key_dependencies
2910
for (key, options, access_memo, parents) in records:
2912
parents = tuple(parents)
2913
if key_dependencies is not None:
2914
key_dependencies.add_references(key, parents)
2915
index, pos, size = access_memo
2916
if 'no-eol' in options:
2920
value += "%d %d" % (pos, size)
2921
if not self._deltas:
2922
if 'line-delta' in options:
2923
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2926
if 'line-delta' in options:
2927
node_refs = (parents, (parents[0],))
2928
if missing_compression_parents:
2929
compression_parents.add(parents[0])
2931
node_refs = (parents, ())
2933
node_refs = (parents, )
2936
raise KnitCorrupt(self, "attempt to add node with parents "
2937
"in parentless index.")
2939
keys[key] = (value, node_refs)
2942
present_nodes = self._get_entries(keys)
2943
for (index, key, value, node_refs) in present_nodes:
2944
parents = node_refs[:1]
2945
# Sometimes these are passed as a list rather than a tuple
2946
passed = static_tuple.as_tuples(keys[key])
2947
passed_parents = passed[1][:1]
2948
if (value[0] != keys[key][0][0] or
2949
parents != passed_parents):
2950
node_refs = static_tuple.as_tuples(node_refs)
2951
raise KnitCorrupt(self, "inconsistent details in add_records"
2952
": %s %s" % ((value, node_refs), passed))
2956
for key, (value, node_refs) in keys.iteritems():
2957
result.append((key, value, node_refs))
2959
for key, (value, node_refs) in keys.iteritems():
2960
result.append((key, value))
2961
self._add_callback(result)
2962
if missing_compression_parents:
2963
# This may appear to be incorrect (it does not check for
2964
# compression parents that are in the existing graph index),
2965
# but such records won't have been buffered, so this is
2966
# actually correct: every entry when
2967
# missing_compression_parents==True either has a missing parent, or
2968
# a parent that is one of the keys in records.
2969
compression_parents.difference_update(keys)
2970
self._missing_compression_parents.update(compression_parents)
2971
# Adding records may have satisfied missing compression parents.
2972
self._missing_compression_parents.difference_update(keys)
2974
def scan_unvalidated_index(self, graph_index):
2975
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2977
This allows this _KnitGraphIndex to keep track of any missing
2978
compression parents we may want to have filled in to make those
2981
:param graph_index: A GraphIndex
2984
new_missing = graph_index.external_references(ref_list_num=1)
2985
new_missing.difference_update(self.get_parent_map(new_missing))
2986
self._missing_compression_parents.update(new_missing)
2987
if self._key_dependencies is not None:
2988
# Add parent refs from graph_index (and discard parent refs that
2989
# the graph_index has).
2990
for node in graph_index.iter_all_entries():
2991
self._key_dependencies.add_references(node[1], node[3][0])
2993
def get_missing_compression_parents(self):
2994
"""Return the keys of missing compression parents.
2996
Missing compression parents occur when a record stream was missing
2997
basis texts, or a index was scanned that had missing basis texts.
2999
return frozenset(self._missing_compression_parents)
3001
def get_missing_parents(self):
3002
"""Return the keys of missing parents."""
3003
# If updating this, you should also update
3004
# groupcompress._GCGraphIndex.get_missing_parents
3005
# We may have false positives, so filter those out.
3006
self._key_dependencies.satisfy_refs_for_keys(
3007
self.get_parent_map(self._key_dependencies.get_unsatisfied_refs()))
3008
return frozenset(self._key_dependencies.get_unsatisfied_refs())
3010
def _check_read(self):
3011
"""raise if reads are not permitted."""
3012
if not self._is_locked():
3013
raise errors.ObjectNotLocked(self)
3015
def _check_write_ok(self):
3016
"""Assert if writes are not permitted."""
3017
if not self._is_locked():
3018
raise errors.ObjectNotLocked(self)
3020
def _compression_parent(self, an_entry):
3021
# return the key that an_entry is compressed against, or None
3022
# Grab the second parent list (as deltas implies parents currently)
3023
compression_parents = an_entry[3][1]
3024
if not compression_parents:
3026
if len(compression_parents) != 1:
3027
raise AssertionError(
3028
"Too many compression parents: %r" % compression_parents)
3029
return compression_parents[0]
3031
def get_build_details(self, keys):
3032
"""Get the method, index_memo and compression parent for version_ids.
3034
Ghosts are omitted from the result.
3036
:param keys: An iterable of keys.
3037
:return: A dict of key:
3038
(index_memo, compression_parent, parents, record_details).
3040
opaque structure to pass to read_records to extract the raw
3043
Content that this record is built upon, may be None
3045
Logical parents of this node
3047
extra information about the content which needs to be passed to
3048
Factory.parse_record
3052
entries = self._get_entries(keys, False)
3053
for entry in entries:
3055
if not self._parents:
3058
parents = entry[3][0]
3059
if not self._deltas:
3060
compression_parent_key = None
3062
compression_parent_key = self._compression_parent(entry)
3063
noeol = (entry[2][0] == 'N')
3064
if compression_parent_key:
3065
method = 'line-delta'
3068
result[key] = (self._node_to_position(entry),
3069
compression_parent_key, parents,
3073
def _get_entries(self, keys, check_present=False):
3074
"""Get the entries for keys.
3076
:param keys: An iterable of index key tuples.
3081
for node in self._graph_index.iter_entries(keys):
3083
found_keys.add(node[1])
3085
# adapt parentless index to the rest of the code.
3086
for node in self._graph_index.iter_entries(keys):
3087
yield node[0], node[1], node[2], ()
3088
found_keys.add(node[1])
3090
missing_keys = keys.difference(found_keys)
3092
raise RevisionNotPresent(missing_keys.pop(), self)
3094
def get_method(self, key):
3095
"""Return compression method of specified key."""
3096
return self._get_method(self._get_node(key))
3098
def _get_method(self, node):
3099
if not self._deltas:
3101
if self._compression_parent(node):
3106
def _get_node(self, key):
3108
return list(self._get_entries([key]))[0]
3110
raise RevisionNotPresent(key, self)
3112
def get_options(self, key):
3113
"""Return a list representing options.
3117
node = self._get_node(key)
3118
options = [self._get_method(node)]
3119
if node[2][0] == 'N':
3120
options.append('no-eol')
3123
def find_ancestry(self, keys):
3124
"""See CombinedGraphIndex.find_ancestry()"""
3125
return self._graph_index.find_ancestry(keys, 0)
3127
def get_parent_map(self, keys):
3128
"""Get a map of the parents of keys.
3130
:param keys: The keys to look up parents for.
3131
:return: A mapping from keys to parents. Absent keys are absent from
3135
nodes = self._get_entries(keys)
3139
result[node[1]] = node[3][0]
3142
result[node[1]] = None
3145
def get_position(self, key):
3146
"""Return details needed to access the version.
3148
:return: a tuple (index, data position, size) to hand to the access
3149
logic to get the record.
3151
node = self._get_node(key)
3152
return self._node_to_position(node)
3154
has_key = _mod_index._has_key_from_parent_map
3157
"""Get all the keys in the collection.
3159
The keys are not ordered.
3162
return [node[1] for node in self._graph_index.iter_all_entries()]
3164
missing_keys = _mod_index._missing_keys_from_parent_map
3166
def _node_to_position(self, node):
3167
"""Convert an index value to position details."""
3168
bits = node[2][1:].split(' ')
3169
return node[0], int(bits[0]), int(bits[1])
3171
def _sort_keys_by_io(self, keys, positions):
3172
"""Figure out an optimal order to read the records for the given keys.
3174
Sort keys, grouped by index and sorted by position.
3176
:param keys: A list of keys whose records we want to read. This will be
3178
:param positions: A dict, such as the one returned by
3179
_get_components_positions()
3182
def get_index_memo(key):
3183
# index_memo is at offset [1]. It is made up of (GraphIndex,
3184
# position, size). GI is an object, which will be unique for each
3185
# pack file. This causes us to group by pack file, then sort by
3186
# position. Size doesn't matter, but it isn't worth breaking up the
3188
return positions[key][1]
3189
return keys.sort(key=get_index_memo)
3191
_get_total_build_size = _get_total_build_size
3194
class _KnitKeyAccess(object):
3195
"""Access to records in .knit files."""
3197
def __init__(self, transport, mapper):
3198
"""Create a _KnitKeyAccess with transport and mapper.
3200
:param transport: The transport the access object is rooted at.
3201
:param mapper: The mapper used to map keys to .knit files.
3203
self._transport = transport
3204
self._mapper = mapper
3206
def add_raw_records(self, key_sizes, raw_data):
3207
"""Add raw knit bytes to a storage area.
3209
The data is spooled to the container writer in one bytes-record per
3212
:param sizes: An iterable of tuples containing the key and size of each
3214
:param raw_data: A bytestring containing the data.
3215
:return: A list of memos to retrieve the record later. Each memo is an
3216
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3217
length), where the key is the record key.
3219
if type(raw_data) is not str:
3220
raise AssertionError(
3221
'data must be plain bytes was %s' % type(raw_data))
3224
# TODO: This can be tuned for writing to sftp and other servers where
3225
# append() is relatively expensive by grouping the writes to each key
3227
for key, size in key_sizes:
3228
path = self._mapper.map(key)
3230
base = self._transport.append_bytes(path + '.knit',
3231
raw_data[offset:offset+size])
3232
except errors.NoSuchFile:
3233
self._transport.mkdir(osutils.dirname(path))
3234
base = self._transport.append_bytes(path + '.knit',
3235
raw_data[offset:offset+size])
3239
result.append((key, base, size))
3243
"""Flush pending writes on this access object.
3245
For .knit files this is a no-op.
3249
def get_raw_records(self, memos_for_retrieval):
3250
"""Get the raw bytes for a records.
3252
:param memos_for_retrieval: An iterable containing the access memo for
3253
retrieving the bytes.
3254
:return: An iterator over the bytes of the records.
3256
# first pass, group into same-index request to minimise readv's issued.
3258
current_prefix = None
3259
for (key, offset, length) in memos_for_retrieval:
3260
if current_prefix == key[:-1]:
3261
current_list.append((offset, length))
3263
if current_prefix is not None:
3264
request_lists.append((current_prefix, current_list))
3265
current_prefix = key[:-1]
3266
current_list = [(offset, length)]
3267
# handle the last entry
3268
if current_prefix is not None:
3269
request_lists.append((current_prefix, current_list))
3270
for prefix, read_vector in request_lists:
3271
path = self._mapper.map(prefix) + '.knit'
3272
for pos, data in self._transport.readv(path, read_vector):
3276
class _DirectPackAccess(object):
3277
"""Access to data in one or more packs with less translation."""
3279
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
3280
"""Create a _DirectPackAccess object.
3282
:param index_to_packs: A dict mapping index objects to the transport
3283
and file names for obtaining data.
3284
:param reload_func: A function to call if we determine that the pack
3285
files have moved and we need to reload our caches. See
3286
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
3288
self._container_writer = None
3289
self._write_index = None
3290
self._indices = index_to_packs
3291
self._reload_func = reload_func
3292
self._flush_func = flush_func
3294
def add_raw_records(self, key_sizes, raw_data):
3295
"""Add raw knit bytes to a storage area.
3297
The data is spooled to the container writer in one bytes-record per
3300
:param sizes: An iterable of tuples containing the key and size of each
3302
:param raw_data: A bytestring containing the data.
3303
:return: A list of memos to retrieve the record later. Each memo is an
3304
opaque index memo. For _DirectPackAccess the memo is (index, pos,
3305
length), where the index field is the write_index object supplied
3306
to the PackAccess object.
3308
if type(raw_data) is not str:
3309
raise AssertionError(
3310
'data must be plain bytes was %s' % type(raw_data))
3313
for key, size in key_sizes:
3314
p_offset, p_length = self._container_writer.add_bytes_record(
3315
raw_data[offset:offset+size], [])
3317
result.append((self._write_index, p_offset, p_length))
3321
"""Flush pending writes on this access object.
3323
This will flush any buffered writes to a NewPack.
3325
if self._flush_func is not None:
3328
def get_raw_records(self, memos_for_retrieval):
3329
"""Get the raw bytes for a records.
3331
:param memos_for_retrieval: An iterable containing the (index, pos,
3332
length) memo for retrieving the bytes. The Pack access method
3333
looks up the pack to use for a given record in its index_to_pack
3335
:return: An iterator over the bytes of the records.
3337
# first pass, group into same-index requests
3339
current_index = None
3340
for (index, offset, length) in memos_for_retrieval:
3341
if current_index == index:
3342
current_list.append((offset, length))
3344
if current_index is not None:
3345
request_lists.append((current_index, current_list))
3346
current_index = index
3347
current_list = [(offset, length)]
3348
# handle the last entry
3349
if current_index is not None:
3350
request_lists.append((current_index, current_list))
3351
for index, offsets in request_lists:
3353
transport, path = self._indices[index]
3355
# A KeyError here indicates that someone has triggered an index
3356
# reload, and this index has gone missing, we need to start
3358
if self._reload_func is None:
3359
# If we don't have a _reload_func there is nothing that can
3362
raise errors.RetryWithNewPacks(index,
3363
reload_occurred=True,
3364
exc_info=sys.exc_info())
3366
reader = pack.make_readv_reader(transport, path, offsets)
3367
for names, read_func in reader.iter_records():
3368
yield read_func(None)
3369
except errors.NoSuchFile:
3370
# A NoSuchFile error indicates that a pack file has gone
3371
# missing on disk, we need to trigger a reload, and start over.
3372
if self._reload_func is None:
3374
raise errors.RetryWithNewPacks(transport.abspath(path),
3375
reload_occurred=False,
3376
exc_info=sys.exc_info())
3378
def set_writer(self, writer, index, transport_packname):
3379
"""Set a writer to use for adding data."""
3380
if index is not None:
3381
self._indices[index] = transport_packname
3382
self._container_writer = writer
3383
self._write_index = index
3385
def reload_or_raise(self, retry_exc):
3386
"""Try calling the reload function, or re-raise the original exception.
3388
This should be called after _DirectPackAccess raises a
3389
RetryWithNewPacks exception. This function will handle the common logic
3390
of determining when the error is fatal versus being temporary.
3391
It will also make sure that the original exception is raised, rather
3392
than the RetryWithNewPacks exception.
3394
If this function returns, then the calling function should retry
3395
whatever operation was being performed. Otherwise an exception will
3398
:param retry_exc: A RetryWithNewPacks exception.
3401
if self._reload_func is None:
3403
elif not self._reload_func():
3404
# The reload claimed that nothing changed
3405
if not retry_exc.reload_occurred:
3406
# If there wasn't an earlier reload, then we really were
3407
# expecting to find changes. We didn't find them, so this is a
3411
exc_class, exc_value, exc_traceback = retry_exc.exc_info
3412
raise exc_class, exc_value, exc_traceback
3415
def annotate_knit(knit, revision_id):
3416
"""Annotate a knit with no cached annotations.
3418
This implementation is for knits with no cached annotations.
3419
It will work for knits with cached annotations, but this is not
3422
annotator = _KnitAnnotator(knit)
3423
return iter(annotator.annotate_flat(revision_id))
3426
class _KnitAnnotator(annotate.Annotator):
3427
"""Build up the annotations for a text."""
3429
def __init__(self, vf):
3430
annotate.Annotator.__init__(self, vf)
3432
# TODO: handle Nodes which cannot be extracted
3433
# self._ghosts = set()
3435
# Map from (key, parent_key) => matching_blocks, should be 'use once'
3436
self._matching_blocks = {}
3438
# KnitContent objects
3439
self._content_objects = {}
3440
# The number of children that depend on this fulltext content object
3441
self._num_compression_children = {}
3442
# Delta records that need their compression parent before they can be
3444
self._pending_deltas = {}
3445
# Fulltext records that are waiting for their parents fulltexts before
3446
# they can be yielded for annotation
3447
self._pending_annotation = {}
3449
self._all_build_details = {}
3451
def _get_build_graph(self, key):
3452
"""Get the graphs for building texts and annotations.
3454
The data you need for creating a full text may be different than the
3455
data you need to annotate that text. (At a minimum, you need both
3456
parents to create an annotation, but only need 1 parent to generate the
3459
:return: A list of (key, index_memo) records, suitable for
3460
passing to read_records_iter to start reading in the raw data from
3463
pending = set([key])
3466
self._num_needed_children[key] = 1
3468
# get all pending nodes
3469
this_iteration = pending
3470
build_details = self._vf._index.get_build_details(this_iteration)
3471
self._all_build_details.update(build_details)
3472
# new_nodes = self._vf._index._get_entries(this_iteration)
3474
for key, details in build_details.iteritems():
3475
(index_memo, compression_parent, parent_keys,
3476
record_details) = details
3477
self._parent_map[key] = parent_keys
3478
self._heads_provider = None
3479
records.append((key, index_memo))
3480
# Do we actually need to check _annotated_lines?
3481
pending.update([p for p in parent_keys
3482
if p not in self._all_build_details])
3484
for parent_key in parent_keys:
3485
if parent_key in self._num_needed_children:
3486
self._num_needed_children[parent_key] += 1
3488
self._num_needed_children[parent_key] = 1
3489
if compression_parent:
3490
if compression_parent in self._num_compression_children:
3491
self._num_compression_children[compression_parent] += 1
3493
self._num_compression_children[compression_parent] = 1
3495
missing_versions = this_iteration.difference(build_details.keys())
3496
if missing_versions:
3497
for key in missing_versions:
3498
if key in self._parent_map and key in self._text_cache:
3499
# We already have this text ready, we just need to
3500
# yield it later so we get it annotated
3502
parent_keys = self._parent_map[key]
3503
for parent_key in parent_keys:
3504
if parent_key in self._num_needed_children:
3505
self._num_needed_children[parent_key] += 1
3507
self._num_needed_children[parent_key] = 1
3508
pending.update([p for p in parent_keys
3509
if p not in self._all_build_details])
3511
raise errors.RevisionNotPresent(key, self._vf)
3512
# Generally we will want to read the records in reverse order, because
3513
# we find the parent nodes after the children
3515
return records, ann_keys
3517
def _get_needed_texts(self, key, pb=None):
3518
# if True or len(self._vf._immediate_fallback_vfs) > 0:
3519
if len(self._vf._immediate_fallback_vfs) > 0:
3520
# If we have fallbacks, go to the generic path
3521
for v in annotate.Annotator._get_needed_texts(self, key, pb=pb):
3526
records, ann_keys = self._get_build_graph(key)
3527
for idx, (sub_key, text, num_lines) in enumerate(
3528
self._extract_texts(records)):
3530
pb.update('annotating', idx, len(records))
3531
yield sub_key, text, num_lines
3532
for sub_key in ann_keys:
3533
text = self._text_cache[sub_key]
3534
num_lines = len(text) # bad assumption
3535
yield sub_key, text, num_lines
3537
except errors.RetryWithNewPacks, e:
3538
self._vf._access.reload_or_raise(e)
3539
# The cached build_details are no longer valid
3540
self._all_build_details.clear()
3542
def _cache_delta_blocks(self, key, compression_parent, delta, lines):
3543
parent_lines = self._text_cache[compression_parent]
3544
blocks = list(KnitContent.get_line_delta_blocks(delta, parent_lines, lines))
3545
self._matching_blocks[(key, compression_parent)] = blocks
3547
def _expand_record(self, key, parent_keys, compression_parent, record,
3550
if compression_parent:
3551
if compression_parent not in self._content_objects:
3552
# Waiting for the parent
3553
self._pending_deltas.setdefault(compression_parent, []).append(
3554
(key, parent_keys, record, record_details))
3556
# We have the basis parent, so expand the delta
3557
num = self._num_compression_children[compression_parent]
3560
base_content = self._content_objects.pop(compression_parent)
3561
self._num_compression_children.pop(compression_parent)
3563
self._num_compression_children[compression_parent] = num
3564
base_content = self._content_objects[compression_parent]
3565
# It is tempting to want to copy_base_content=False for the last
3566
# child object. However, whenever noeol=False,
3567
# self._text_cache[parent_key] is content._lines. So mutating it
3568
# gives very bad results.
3569
# The alternative is to copy the lines into text cache, but then we
3570
# are copying anyway, so just do it here.
3571
content, delta = self._vf._factory.parse_record(
3572
key, record, record_details, base_content,
3573
copy_base_content=True)
3576
content, _ = self._vf._factory.parse_record(
3577
key, record, record_details, None)
3578
if self._num_compression_children.get(key, 0) > 0:
3579
self._content_objects[key] = content
3580
lines = content.text()
3581
self._text_cache[key] = lines
3582
if delta is not None:
3583
self._cache_delta_blocks(key, compression_parent, delta, lines)
3586
def _get_parent_annotations_and_matches(self, key, text, parent_key):
3587
"""Get the list of annotations for the parent, and the matching lines.
3589
:param text: The opaque value given by _get_needed_texts
3590
:param parent_key: The key for the parent text
3591
:return: (parent_annotations, matching_blocks)
3592
parent_annotations is a list as long as the number of lines in
3594
matching_blocks is a list of (parent_idx, text_idx, len) tuples
3595
indicating which lines match between the two texts
3597
block_key = (key, parent_key)
3598
if block_key in self._matching_blocks:
3599
blocks = self._matching_blocks.pop(block_key)
3600
parent_annotations = self._annotations_cache[parent_key]
3601
return parent_annotations, blocks
3602
return annotate.Annotator._get_parent_annotations_and_matches(self,
3603
key, text, parent_key)
3605
def _process_pending(self, key):
3606
"""The content for 'key' was just processed.
3608
Determine if there is any more pending work to be processed.
3611
if key in self._pending_deltas:
3612
compression_parent = key
3613
children = self._pending_deltas.pop(key)
3614
for child_key, parent_keys, record, record_details in children:
3615
lines = self._expand_record(child_key, parent_keys,
3617
record, record_details)
3618
if self._check_ready_for_annotations(child_key, parent_keys):
3619
to_return.append(child_key)
3620
# Also check any children that are waiting for this parent to be
3622
if key in self._pending_annotation:
3623
children = self._pending_annotation.pop(key)
3624
to_return.extend([c for c, p_keys in children
3625
if self._check_ready_for_annotations(c, p_keys)])
3628
def _check_ready_for_annotations(self, key, parent_keys):
3629
"""return true if this text is ready to be yielded.
3631
Otherwise, this will return False, and queue the text into
3632
self._pending_annotation
3634
for parent_key in parent_keys:
3635
if parent_key not in self._annotations_cache:
3636
# still waiting on at least one parent text, so queue it up
3637
# Note that if there are multiple parents, we need to wait
3639
self._pending_annotation.setdefault(parent_key,
3640
[]).append((key, parent_keys))
3644
def _extract_texts(self, records):
3645
"""Extract the various texts needed based on records"""
3646
# We iterate in the order read, rather than a strict order requested
3647
# However, process what we can, and put off to the side things that
3648
# still need parents, cleaning them up when those parents are
3651
# 1) As 'records' are read, see if we can expand these records into
3652
# Content objects (and thus lines)
3653
# 2) If a given line-delta is waiting on its compression parent, it
3654
# gets queued up into self._pending_deltas, otherwise we expand
3655
# it, and put it into self._text_cache and self._content_objects
3656
# 3) If we expanded the text, we will then check to see if all
3657
# parents have also been processed. If so, this text gets yielded,
3658
# else this record gets set aside into pending_annotation
3659
# 4) Further, if we expanded the text in (2), we will then check to
3660
# see if there are any children in self._pending_deltas waiting to
3661
# also be processed. If so, we go back to (2) for those
3662
# 5) Further again, if we yielded the text, we can then check if that
3663
# 'unlocks' any of the texts in pending_annotations, which should
3664
# then get yielded as well
3665
# Note that both steps 4 and 5 are 'recursive' in that unlocking one
3666
# compression child could unlock yet another, and yielding a fulltext
3667
# will also 'unlock' the children that are waiting on that annotation.
3668
# (Though also, unlocking 1 parent's fulltext, does not unlock a child
3669
# if other parents are also waiting.)
3670
# We want to yield content before expanding child content objects, so
3671
# that we know when we can re-use the content lines, and the annotation
3672
# code can know when it can stop caching fulltexts, as well.
3674
# Children that are missing their compression parent
3676
for (key, record, digest) in self._vf._read_records_iter(records):
3678
details = self._all_build_details[key]
3679
(_, compression_parent, parent_keys, record_details) = details
3680
lines = self._expand_record(key, parent_keys, compression_parent,
3681
record, record_details)
3683
# Pending delta should be queued up
3685
# At this point, we may be able to yield this content, if all
3686
# parents are also finished
3687
yield_this_text = self._check_ready_for_annotations(key,
3690
# All parents present
3691
yield key, lines, len(lines)
3692
to_process = self._process_pending(key)
3694
this_process = to_process
3696
for key in this_process:
3697
lines = self._text_cache[key]
3698
yield key, lines, len(lines)
3699
to_process.extend(self._process_pending(key))
3702
from bzrlib._knit_load_data_pyx import _load_data_c as _load_data
3703
except ImportError, e:
3704
osutils.failed_to_load_extension(e)
3705
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
