1
# Copyright (C) 2005, 2006, 2007, 2008 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
"""Knit versionedfile implementation.
19
A knit is a versioned file implementation that supports efficient append only
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
end-marker; simply "end VERSION"
28
delta can be line or full contents.a
29
... the 8's there are the index number of the annotation.
30
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
8 e.set('executable', 'yes')
36
8 if elt.get('executable') == 'yes':
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
43
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
44
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
45
09:33 < lifeless> right
46
09:33 < jrydberg> lifeless: the position and size is the range in the data file
49
so the index sequence is the dictionary compressed sequence number used
50
in the deltas to provide line annotation
55
from cStringIO import StringIO
56
from itertools import izip
61
from bzrlib.lazy_import import lazy_import
62
lazy_import(globals(), """
82
from bzrlib.errors import (
90
RevisionAlreadyPresent,
93
from bzrlib.osutils import (
100
from bzrlib.versionedfile import (
101
AbsentContentFactory,
105
ChunkedContentFactory,
112
# TODO: Split out code specific to this format into an associated object.
114
# TODO: Can we put in some kind of value to check that the index and data
115
# files belong together?
117
# TODO: accommodate binaries, perhaps by storing a byte count
119
# TODO: function to check whole file
121
# TODO: atomically append data, then measure backwards from the cursor
122
# position after writing to work out where it was located. we may need to
123
# bypass python file buffering.
125
DATA_SUFFIX = '.knit'
126
INDEX_SUFFIX = '.kndx'
127
_STREAM_MIN_BUFFER_SIZE = 5*1024*1024
130
class KnitAdapter(object):
131
"""Base class for knit record adaption."""
133
def __init__(self, basis_vf):
134
"""Create an adapter which accesses full texts from basis_vf.
136
:param basis_vf: A versioned file to access basis texts of deltas from.
137
May be None for adapters that do not need to access basis texts.
139
self._data = KnitVersionedFiles(None, None)
140
self._annotate_factory = KnitAnnotateFactory()
141
self._plain_factory = KnitPlainFactory()
142
self._basis_vf = basis_vf
145
class FTAnnotatedToUnannotated(KnitAdapter):
146
"""An adapter from FT annotated knits to unannotated ones."""
148
def get_bytes(self, factory):
149
annotated_compressed_bytes = factory._raw_record
151
self._data._parse_record_unchecked(annotated_compressed_bytes)
152
content = self._annotate_factory.parse_fulltext(contents, rec[1])
153
size, bytes = self._data._record_to_data((rec[1],), rec[3], content.text())
157
class DeltaAnnotatedToUnannotated(KnitAdapter):
158
"""An adapter for deltas from annotated to unannotated."""
160
def get_bytes(self, factory):
161
annotated_compressed_bytes = factory._raw_record
163
self._data._parse_record_unchecked(annotated_compressed_bytes)
164
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
166
contents = self._plain_factory.lower_line_delta(delta)
167
size, bytes = self._data._record_to_data((rec[1],), rec[3], contents)
171
class FTAnnotatedToFullText(KnitAdapter):
172
"""An adapter from FT annotated knits to unannotated ones."""
174
def get_bytes(self, factory):
175
annotated_compressed_bytes = factory._raw_record
177
self._data._parse_record_unchecked(annotated_compressed_bytes)
178
content, delta = self._annotate_factory.parse_record(factory.key[-1],
179
contents, factory._build_details, None)
180
return ''.join(content.text())
183
class DeltaAnnotatedToFullText(KnitAdapter):
184
"""An adapter for deltas from annotated to unannotated."""
186
def get_bytes(self, factory):
187
annotated_compressed_bytes = factory._raw_record
189
self._data._parse_record_unchecked(annotated_compressed_bytes)
190
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
192
compression_parent = factory.parents[0]
193
basis_entry = self._basis_vf.get_record_stream(
194
[compression_parent], 'unordered', True).next()
195
if basis_entry.storage_kind == 'absent':
196
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
197
basis_chunks = basis_entry.get_bytes_as('chunked')
198
basis_lines = osutils.chunks_to_lines(basis_chunks)
199
# Manually apply the delta because we have one annotated content and
201
basis_content = PlainKnitContent(basis_lines, compression_parent)
202
basis_content.apply_delta(delta, rec[1])
203
basis_content._should_strip_eol = factory._build_details[1]
204
return ''.join(basis_content.text())
207
class FTPlainToFullText(KnitAdapter):
208
"""An adapter from FT plain knits to unannotated ones."""
210
def get_bytes(self, factory):
211
compressed_bytes = factory._raw_record
213
self._data._parse_record_unchecked(compressed_bytes)
214
content, delta = self._plain_factory.parse_record(factory.key[-1],
215
contents, factory._build_details, None)
216
return ''.join(content.text())
219
class DeltaPlainToFullText(KnitAdapter):
220
"""An adapter for deltas from annotated to unannotated."""
222
def get_bytes(self, factory):
223
compressed_bytes = factory._raw_record
225
self._data._parse_record_unchecked(compressed_bytes)
226
delta = self._plain_factory.parse_line_delta(contents, rec[1])
227
compression_parent = factory.parents[0]
228
# XXX: string splitting overhead.
229
basis_entry = self._basis_vf.get_record_stream(
230
[compression_parent], 'unordered', True).next()
231
if basis_entry.storage_kind == 'absent':
232
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
233
basis_chunks = basis_entry.get_bytes_as('chunked')
234
basis_lines = osutils.chunks_to_lines(basis_chunks)
235
basis_content = PlainKnitContent(basis_lines, compression_parent)
236
# Manually apply the delta because we have one annotated content and
238
content, _ = self._plain_factory.parse_record(rec[1], contents,
239
factory._build_details, basis_content)
240
return ''.join(content.text())
243
class KnitContentFactory(ContentFactory):
244
"""Content factory for streaming from knits.
246
:seealso ContentFactory:
249
def __init__(self, key, parents, build_details, sha1, raw_record,
250
annotated, knit=None, network_bytes=None):
251
"""Create a KnitContentFactory for key.
254
:param parents: The parents.
255
:param build_details: The build details as returned from
257
:param sha1: The sha1 expected from the full text of this object.
258
:param raw_record: The bytes of the knit data from disk.
259
:param annotated: True if the raw data is annotated.
260
:param network_bytes: None to calculate the network bytes on demand,
261
not-none if they are already known.
263
ContentFactory.__init__(self)
266
self.parents = parents
267
if build_details[0] == 'line-delta':
272
annotated_kind = 'annotated-'
275
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
276
self._raw_record = raw_record
277
self._network_bytes = network_bytes
278
self._build_details = build_details
281
def _create_network_bytes(self):
282
"""Create a fully serialised network version for transmission."""
283
# storage_kind, key, parents, Noeol, raw_record
284
key_bytes = '\x00'.join(self.key)
285
if self.parents is None:
286
parent_bytes = 'None:'
288
parent_bytes = '\t'.join('\x00'.join(key) for key in self.parents)
289
if self._build_details[1]:
293
network_bytes = "%s\n%s\n%s\n%s%s" % (self.storage_kind, key_bytes,
294
parent_bytes, noeol, self._raw_record)
295
self._network_bytes = network_bytes
297
def get_bytes_as(self, storage_kind):
298
if storage_kind == self.storage_kind:
299
if self._network_bytes is None:
300
self._create_network_bytes()
301
return self._network_bytes
302
if ('-ft-' in self.storage_kind and
303
storage_kind in ('chunked', 'fulltext')):
304
adapter_key = (self.storage_kind, 'fulltext')
305
adapter_factory = adapter_registry.get(adapter_key)
306
adapter = adapter_factory(None)
307
bytes = adapter.get_bytes(self)
308
if storage_kind == 'chunked':
312
if self._knit is not None:
313
# Not redundant with direct conversion above - that only handles
315
if storage_kind == 'chunked':
316
return self._knit.get_lines(self.key[0])
317
elif storage_kind == 'fulltext':
318
return self._knit.get_text(self.key[0])
319
raise errors.UnavailableRepresentation(self.key, storage_kind,
323
class LazyKnitContentFactory(ContentFactory):
324
"""A ContentFactory which can either generate full text or a wire form.
326
:seealso ContentFactory:
329
def __init__(self, key, parents, generator, first):
330
"""Create a LazyKnitContentFactory.
332
:param key: The key of the record.
333
:param parents: The parents of the record.
334
:param generator: A _ContentMapGenerator containing the record for this
336
:param first: Is this the first content object returned from generator?
337
if it is, its storage kind is knit-delta-closure, otherwise it is
338
knit-delta-closure-ref
341
self.parents = parents
343
self._generator = generator
344
self.storage_kind = "knit-delta-closure"
346
self.storage_kind = self.storage_kind + "-ref"
349
def get_bytes_as(self, storage_kind):
350
if storage_kind == self.storage_kind:
352
return self._generator._wire_bytes()
354
# all the keys etc are contained in the bytes returned in the
357
if storage_kind in ('chunked', 'fulltext'):
358
chunks = self._generator._get_one_work(self.key).text()
359
if storage_kind == 'chunked':
362
return ''.join(chunks)
363
raise errors.UnavailableRepresentation(self.key, storage_kind,
367
def knit_delta_closure_to_records(storage_kind, bytes, line_end):
368
"""Convert a network record to a iterator over stream records.
370
:param storage_kind: The storage kind of the record.
371
Must be 'knit-delta-closure'.
372
:param bytes: The bytes of the record on the network.
374
generator = _NetworkContentMapGenerator(bytes, line_end)
375
return generator.get_record_stream()
378
def knit_network_to_record(storage_kind, bytes, line_end):
379
"""Convert a network record to a record object.
381
:param storage_kind: The storage kind of the record.
382
:param bytes: The bytes of the record on the network.
385
line_end = bytes.find('\n', start)
386
key = tuple(bytes[start:line_end].split('\x00'))
388
line_end = bytes.find('\n', start)
389
parent_line = bytes[start:line_end]
390
if parent_line == 'None:':
394
[tuple(segment.split('\x00')) for segment in parent_line.split('\t')
397
noeol = bytes[start] == 'N'
398
if 'ft' in storage_kind:
401
method = 'line-delta'
402
build_details = (method, noeol)
404
raw_record = bytes[start:]
405
annotated = 'annotated' in storage_kind
406
return [KnitContentFactory(key, parents, build_details, None, raw_record,
407
annotated, network_bytes=bytes)]
410
class KnitContent(object):
411
"""Content of a knit version to which deltas can be applied.
413
This is always stored in memory as a list of lines with \n at the end,
414
plus a flag saying if the final ending is really there or not, because that
415
corresponds to the on-disk knit representation.
419
self._should_strip_eol = False
421
def apply_delta(self, delta, new_version_id):
422
"""Apply delta to this object to become new_version_id."""
423
raise NotImplementedError(self.apply_delta)
425
def line_delta_iter(self, new_lines):
426
"""Generate line-based delta from this content to new_lines."""
427
new_texts = new_lines.text()
428
old_texts = self.text()
429
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
430
for tag, i1, i2, j1, j2 in s.get_opcodes():
433
# ofrom, oto, length, data
434
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
436
def line_delta(self, new_lines):
437
return list(self.line_delta_iter(new_lines))
440
def get_line_delta_blocks(knit_delta, source, target):
441
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
442
target_len = len(target)
445
for s_begin, s_end, t_len, new_text in knit_delta:
446
true_n = s_begin - s_pos
449
# knit deltas do not provide reliable info about whether the
450
# last line of a file matches, due to eol handling.
451
if source[s_pos + n -1] != target[t_pos + n -1]:
454
yield s_pos, t_pos, n
455
t_pos += t_len + true_n
457
n = target_len - t_pos
459
if source[s_pos + n -1] != target[t_pos + n -1]:
462
yield s_pos, t_pos, n
463
yield s_pos + (target_len - t_pos), target_len, 0
466
class AnnotatedKnitContent(KnitContent):
467
"""Annotated content."""
469
def __init__(self, lines):
470
KnitContent.__init__(self)
474
"""Return a list of (origin, text) for each content line."""
475
lines = self._lines[:]
476
if self._should_strip_eol:
477
origin, last_line = lines[-1]
478
lines[-1] = (origin, last_line.rstrip('\n'))
481
def apply_delta(self, delta, new_version_id):
482
"""Apply delta to this object to become new_version_id."""
485
for start, end, count, delta_lines in delta:
486
lines[offset+start:offset+end] = delta_lines
487
offset = offset + (start - end) + count
491
lines = [text for origin, text in self._lines]
492
except ValueError, e:
493
# most commonly (only?) caused by the internal form of the knit
494
# missing annotation information because of a bug - see thread
496
raise KnitCorrupt(self,
497
"line in annotated knit missing annotation information: %s"
499
if self._should_strip_eol:
500
lines[-1] = lines[-1].rstrip('\n')
504
return AnnotatedKnitContent(self._lines[:])
507
class PlainKnitContent(KnitContent):
508
"""Unannotated content.
510
When annotate[_iter] is called on this content, the same version is reported
511
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
515
def __init__(self, lines, version_id):
516
KnitContent.__init__(self)
518
self._version_id = version_id
521
"""Return a list of (origin, text) for each content line."""
522
return [(self._version_id, line) for line in self._lines]
524
def apply_delta(self, delta, new_version_id):
525
"""Apply delta to this object to become new_version_id."""
528
for start, end, count, delta_lines in delta:
529
lines[offset+start:offset+end] = delta_lines
530
offset = offset + (start - end) + count
531
self._version_id = new_version_id
534
return PlainKnitContent(self._lines[:], self._version_id)
538
if self._should_strip_eol:
540
lines[-1] = lines[-1].rstrip('\n')
544
class _KnitFactory(object):
545
"""Base class for common Factory functions."""
547
def parse_record(self, version_id, record, record_details,
548
base_content, copy_base_content=True):
549
"""Parse a record into a full content object.
551
:param version_id: The official version id for this content
552
:param record: The data returned by read_records_iter()
553
:param record_details: Details about the record returned by
555
:param base_content: If get_build_details returns a compression_parent,
556
you must return a base_content here, else use None
557
:param copy_base_content: When building from the base_content, decide
558
you can either copy it and return a new object, or modify it in
560
:return: (content, delta) A Content object and possibly a line-delta,
563
method, noeol = record_details
564
if method == 'line-delta':
565
if copy_base_content:
566
content = base_content.copy()
568
content = base_content
569
delta = self.parse_line_delta(record, version_id)
570
content.apply_delta(delta, version_id)
572
content = self.parse_fulltext(record, version_id)
574
content._should_strip_eol = noeol
575
return (content, delta)
578
class KnitAnnotateFactory(_KnitFactory):
579
"""Factory for creating annotated Content objects."""
583
def make(self, lines, version_id):
584
num_lines = len(lines)
585
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
587
def parse_fulltext(self, content, version_id):
588
"""Convert fulltext to internal representation
590
fulltext content is of the format
591
revid(utf8) plaintext\n
592
internal representation is of the format:
595
# TODO: jam 20070209 The tests expect this to be returned as tuples,
596
# but the code itself doesn't really depend on that.
597
# Figure out a way to not require the overhead of turning the
598
# list back into tuples.
599
lines = [tuple(line.split(' ', 1)) for line in content]
600
return AnnotatedKnitContent(lines)
602
def parse_line_delta_iter(self, lines):
603
return iter(self.parse_line_delta(lines))
605
def parse_line_delta(self, lines, version_id, plain=False):
606
"""Convert a line based delta into internal representation.
608
line delta is in the form of:
609
intstart intend intcount
611
revid(utf8) newline\n
612
internal representation is
613
(start, end, count, [1..count tuples (revid, newline)])
615
:param plain: If True, the lines are returned as a plain
616
list without annotations, not as a list of (origin, content) tuples, i.e.
617
(start, end, count, [1..count newline])
624
def cache_and_return(line):
625
origin, text = line.split(' ', 1)
626
return cache.setdefault(origin, origin), text
628
# walk through the lines parsing.
629
# Note that the plain test is explicitly pulled out of the
630
# loop to minimise any performance impact
633
start, end, count = [int(n) for n in header.split(',')]
634
contents = [next().split(' ', 1)[1] for i in xrange(count)]
635
result.append((start, end, count, contents))
638
start, end, count = [int(n) for n in header.split(',')]
639
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
640
result.append((start, end, count, contents))
643
def get_fulltext_content(self, lines):
644
"""Extract just the content lines from a fulltext."""
645
return (line.split(' ', 1)[1] for line in lines)
647
def get_linedelta_content(self, lines):
648
"""Extract just the content from a line delta.
650
This doesn't return all of the extra information stored in a delta.
651
Only the actual content lines.
656
header = header.split(',')
657
count = int(header[2])
658
for i in xrange(count):
659
origin, text = next().split(' ', 1)
662
def lower_fulltext(self, content):
663
"""convert a fulltext content record into a serializable form.
665
see parse_fulltext which this inverts.
667
# TODO: jam 20070209 We only do the caching thing to make sure that
668
# the origin is a valid utf-8 line, eventually we could remove it
669
return ['%s %s' % (o, t) for o, t in content._lines]
671
def lower_line_delta(self, delta):
672
"""convert a delta into a serializable form.
674
See parse_line_delta which this inverts.
676
# TODO: jam 20070209 We only do the caching thing to make sure that
677
# the origin is a valid utf-8 line, eventually we could remove it
679
for start, end, c, lines in delta:
680
out.append('%d,%d,%d\n' % (start, end, c))
681
out.extend(origin + ' ' + text
682
for origin, text in lines)
685
def annotate(self, knit, key):
686
content = knit._get_content(key)
687
# adjust for the fact that serialised annotations are only key suffixes
689
if type(key) is tuple:
691
origins = content.annotate()
693
for origin, line in origins:
694
result.append((prefix + (origin,), line))
697
# XXX: This smells a bit. Why would key ever be a non-tuple here?
698
# Aren't keys defined to be tuples? -- spiv 20080618
699
return content.annotate()
702
class KnitPlainFactory(_KnitFactory):
703
"""Factory for creating plain Content objects."""
707
def make(self, lines, version_id):
708
return PlainKnitContent(lines, version_id)
710
def parse_fulltext(self, content, version_id):
711
"""This parses an unannotated fulltext.
713
Note that this is not a noop - the internal representation
714
has (versionid, line) - its just a constant versionid.
716
return self.make(content, version_id)
718
def parse_line_delta_iter(self, lines, version_id):
720
num_lines = len(lines)
721
while cur < num_lines:
724
start, end, c = [int(n) for n in header.split(',')]
725
yield start, end, c, lines[cur:cur+c]
728
def parse_line_delta(self, lines, version_id):
729
return list(self.parse_line_delta_iter(lines, version_id))
731
def get_fulltext_content(self, lines):
732
"""Extract just the content lines from a fulltext."""
735
def get_linedelta_content(self, lines):
736
"""Extract just the content from a line delta.
738
This doesn't return all of the extra information stored in a delta.
739
Only the actual content lines.
744
header = header.split(',')
745
count = int(header[2])
746
for i in xrange(count):
749
def lower_fulltext(self, content):
750
return content.text()
752
def lower_line_delta(self, delta):
754
for start, end, c, lines in delta:
755
out.append('%d,%d,%d\n' % (start, end, c))
759
def annotate(self, knit, key):
760
annotator = _KnitAnnotator(knit)
761
return annotator.annotate(key)
765
def make_file_factory(annotated, mapper):
766
"""Create a factory for creating a file based KnitVersionedFiles.
768
This is only functional enough to run interface tests, it doesn't try to
769
provide a full pack environment.
771
:param annotated: knit annotations are wanted.
772
:param mapper: The mapper from keys to paths.
774
def factory(transport):
775
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
776
access = _KnitKeyAccess(transport, mapper)
777
return KnitVersionedFiles(index, access, annotated=annotated)
781
def make_pack_factory(graph, delta, keylength):
782
"""Create a factory for creating a pack based VersionedFiles.
784
This is only functional enough to run interface tests, it doesn't try to
785
provide a full pack environment.
787
:param graph: Store a graph.
788
:param delta: Delta compress contents.
789
:param keylength: How long should keys be.
791
def factory(transport):
792
parents = graph or delta
798
max_delta_chain = 200
801
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
802
key_elements=keylength)
803
stream = transport.open_write_stream('newpack')
804
writer = pack.ContainerWriter(stream.write)
806
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
807
deltas=delta, add_callback=graph_index.add_nodes)
808
access = _DirectPackAccess({})
809
access.set_writer(writer, graph_index, (transport, 'newpack'))
810
result = KnitVersionedFiles(index, access,
811
max_delta_chain=max_delta_chain)
812
result.stream = stream
813
result.writer = writer
818
def cleanup_pack_knit(versioned_files):
819
versioned_files.stream.close()
820
versioned_files.writer.end()
823
def _get_total_build_size(self, keys, positions):
824
"""Determine the total bytes to build these keys.
826
(helper function because _KnitGraphIndex and _KndxIndex work the same, but
827
don't inherit from a common base.)
829
:param keys: Keys that we want to build
830
:param positions: dict of {key, (info, index_memo, comp_parent)} (such
831
as returned by _get_components_positions)
832
:return: Number of bytes to build those keys
834
all_build_index_memos = {}
838
for key in build_keys:
839
# This is mostly for the 'stacked' case
840
# Where we will be getting the data from a fallback
841
if key not in positions:
843
_, index_memo, compression_parent = positions[key]
844
all_build_index_memos[key] = index_memo
845
if compression_parent not in all_build_index_memos:
846
next_keys.add(compression_parent)
847
build_keys = next_keys
848
return sum([index_memo[2] for index_memo
849
in all_build_index_memos.itervalues()])
852
class KnitVersionedFiles(VersionedFiles):
853
"""Storage for many versioned files using knit compression.
855
Backend storage is managed by indices and data objects.
857
:ivar _index: A _KnitGraphIndex or similar that can describe the
858
parents, graph, compression and data location of entries in this
859
KnitVersionedFiles. Note that this is only the index for
860
*this* vfs; if there are fallbacks they must be queried separately.
863
def __init__(self, index, data_access, max_delta_chain=200,
864
annotated=False, reload_func=None):
865
"""Create a KnitVersionedFiles with index and data_access.
867
:param index: The index for the knit data.
868
:param data_access: The access object to store and retrieve knit
870
:param max_delta_chain: The maximum number of deltas to permit during
871
insertion. Set to 0 to prohibit the use of deltas.
872
:param annotated: Set to True to cause annotations to be calculated and
873
stored during insertion.
874
:param reload_func: An function that can be called if we think we need
875
to reload the pack listing and try again. See
876
'bzrlib.repofmt.pack_repo.AggregateIndex' for the signature.
879
self._access = data_access
880
self._max_delta_chain = max_delta_chain
882
self._factory = KnitAnnotateFactory()
884
self._factory = KnitPlainFactory()
885
self._fallback_vfs = []
886
self._reload_func = reload_func
889
return "%s(%r, %r)" % (
890
self.__class__.__name__,
894
def add_fallback_versioned_files(self, a_versioned_files):
895
"""Add a source of texts for texts not present in this knit.
897
:param a_versioned_files: A VersionedFiles object.
899
self._fallback_vfs.append(a_versioned_files)
901
def add_lines(self, key, parents, lines, parent_texts=None,
902
left_matching_blocks=None, nostore_sha=None, random_id=False,
904
"""See VersionedFiles.add_lines()."""
905
self._index._check_write_ok()
906
self._check_add(key, lines, random_id, check_content)
908
# The caller might pass None if there is no graph data, but kndx
909
# indexes can't directly store that, so we give them
910
# an empty tuple instead.
912
line_bytes = ''.join(lines)
913
return self._add(key, lines, parents,
914
parent_texts, left_matching_blocks, nostore_sha, random_id,
915
line_bytes=line_bytes)
917
def _add_text(self, key, parents, text, nostore_sha=None, random_id=False):
918
"""See VersionedFiles._add_text()."""
919
self._index._check_write_ok()
920
self._check_add(key, None, random_id, check_content=False)
921
if text.__class__ is not str:
922
raise errors.BzrBadParameterUnicode("text")
924
# The caller might pass None if there is no graph data, but kndx
925
# indexes can't directly store that, so we give them
926
# an empty tuple instead.
928
return self._add(key, None, parents,
929
None, None, nostore_sha, random_id,
932
def _add(self, key, lines, parents, parent_texts,
933
left_matching_blocks, nostore_sha, random_id,
935
"""Add a set of lines on top of version specified by parents.
937
Any versions not present will be converted into ghosts.
939
:param lines: A list of strings where each one is a single line (has a
940
single newline at the end of the string) This is now optional
941
(callers can pass None). It is left in its location for backwards
942
compatibility. It should ''.join(lines) must == line_bytes
943
:param line_bytes: A single string containing the content
945
We pass both lines and line_bytes because different routes bring the
946
values to this function. And for memory efficiency, we don't want to
947
have to split/join on-demand.
949
# first thing, if the content is something we don't need to store, find
951
digest = sha_string(line_bytes)
952
if nostore_sha == digest:
953
raise errors.ExistingContent
956
if parent_texts is None:
958
# Do a single query to ascertain parent presence; we only compress
959
# against parents in the same kvf.
960
present_parent_map = self._index.get_parent_map(parents)
961
for parent in parents:
962
if parent in present_parent_map:
963
present_parents.append(parent)
965
# Currently we can only compress against the left most present parent.
966
if (len(present_parents) == 0 or
967
present_parents[0] != parents[0]):
970
# To speed the extract of texts the delta chain is limited
971
# to a fixed number of deltas. This should minimize both
972
# I/O and the time spend applying deltas.
973
delta = self._check_should_delta(present_parents[0])
975
text_length = len(line_bytes)
978
# Note: line_bytes is not modified to add a newline, that is tracked
979
# via the no_eol flag. 'lines' *is* modified, because that is the
980
# general values needed by the Content code.
981
if line_bytes and line_bytes[-1] != '\n':
982
options.append('no-eol')
984
# Copy the existing list, or create a new one
986
lines = osutils.split_lines(line_bytes)
989
# Replace the last line with one that ends in a final newline
990
lines[-1] = lines[-1] + '\n'
992
lines = osutils.split_lines(line_bytes)
994
for element in key[:-1]:
995
if type(element) is not str:
996
raise TypeError("key contains non-strings: %r" % (key,))
998
key = key[:-1] + ('sha1:' + digest,)
999
elif type(key[-1]) is not str:
1000
raise TypeError("key contains non-strings: %r" % (key,))
1001
# Knit hunks are still last-element only
1002
version_id = key[-1]
1003
content = self._factory.make(lines, version_id)
1005
# Hint to the content object that its text() call should strip the
1007
content._should_strip_eol = True
1008
if delta or (self._factory.annotated and len(present_parents) > 0):
1009
# Merge annotations from parent texts if needed.
1010
delta_hunks = self._merge_annotations(content, present_parents,
1011
parent_texts, delta, self._factory.annotated,
1012
left_matching_blocks)
1015
options.append('line-delta')
1016
store_lines = self._factory.lower_line_delta(delta_hunks)
1017
size, bytes = self._record_to_data(key, digest,
1020
options.append('fulltext')
1021
# isinstance is slower and we have no hierarchy.
1022
if self._factory.__class__ is KnitPlainFactory:
1023
# Use the already joined bytes saving iteration time in
1025
dense_lines = [line_bytes]
1027
dense_lines.append('\n')
1028
size, bytes = self._record_to_data(key, digest,
1031
# get mixed annotation + content and feed it into the
1033
store_lines = self._factory.lower_fulltext(content)
1034
size, bytes = self._record_to_data(key, digest,
1037
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
1038
self._index.add_records(
1039
((key, options, access_memo, parents),),
1040
random_id=random_id)
1041
return digest, text_length, content
1043
def annotate(self, key):
1044
"""See VersionedFiles.annotate."""
1045
return self._factory.annotate(self, key)
1047
def check(self, progress_bar=None):
1048
"""See VersionedFiles.check()."""
1049
# This doesn't actually test extraction of everything, but that will
1050
# impact 'bzr check' substantially, and needs to be integrated with
1051
# care. However, it does check for the obvious problem of a delta with
1053
keys = self._index.keys()
1054
parent_map = self.get_parent_map(keys)
1056
if self._index.get_method(key) != 'fulltext':
1057
compression_parent = parent_map[key][0]
1058
if compression_parent not in parent_map:
1059
raise errors.KnitCorrupt(self,
1060
"Missing basis parent %s for %s" % (
1061
compression_parent, key))
1062
for fallback_vfs in self._fallback_vfs:
1063
fallback_vfs.check()
1065
def _check_add(self, key, lines, random_id, check_content):
1066
"""check that version_id and lines are safe to add."""
1067
version_id = key[-1]
1068
if version_id is not None:
1069
if contains_whitespace(version_id):
1070
raise InvalidRevisionId(version_id, self)
1071
self.check_not_reserved_id(version_id)
1072
# TODO: If random_id==False and the key is already present, we should
1073
# probably check that the existing content is identical to what is
1074
# being inserted, and otherwise raise an exception. This would make
1075
# the bundle code simpler.
1077
self._check_lines_not_unicode(lines)
1078
self._check_lines_are_lines(lines)
1080
def _check_header(self, key, line):
1081
rec = self._split_header(line)
1082
self._check_header_version(rec, key[-1])
1085
def _check_header_version(self, rec, version_id):
1086
"""Checks the header version on original format knit records.
1088
These have the last component of the key embedded in the record.
1090
if rec[1] != version_id:
1091
raise KnitCorrupt(self,
1092
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1094
def _check_should_delta(self, parent):
1095
"""Iterate back through the parent listing, looking for a fulltext.
1097
This is used when we want to decide whether to add a delta or a new
1098
fulltext. It searches for _max_delta_chain parents. When it finds a
1099
fulltext parent, it sees if the total size of the deltas leading up to
1100
it is large enough to indicate that we want a new full text anyway.
1102
Return True if we should create a new delta, False if we should use a
1106
fulltext_size = None
1107
for count in xrange(self._max_delta_chain):
1109
# Note that this only looks in the index of this particular
1110
# KnitVersionedFiles, not in the fallbacks. This ensures that
1111
# we won't store a delta spanning physical repository
1113
build_details = self._index.get_build_details([parent])
1114
parent_details = build_details[parent]
1115
except (RevisionNotPresent, KeyError), e:
1116
# Some basis is not locally present: always fulltext
1118
index_memo, compression_parent, _, _ = parent_details
1119
_, _, size = index_memo
1120
if compression_parent is None:
1121
fulltext_size = size
1124
# We don't explicitly check for presence because this is in an
1125
# inner loop, and if it's missing it'll fail anyhow.
1126
parent = compression_parent
1128
# We couldn't find a fulltext, so we must create a new one
1130
# Simple heuristic - if the total I/O wold be greater as a delta than
1131
# the originally installed fulltext, we create a new fulltext.
1132
return fulltext_size > delta_size
1134
def _build_details_to_components(self, build_details):
1135
"""Convert a build_details tuple to a position tuple."""
1136
# record_details, access_memo, compression_parent
1137
return build_details[3], build_details[0], build_details[1]
1139
def _get_components_positions(self, keys, allow_missing=False):
1140
"""Produce a map of position data for the components of keys.
1142
This data is intended to be used for retrieving the knit records.
1144
A dict of key to (record_details, index_memo, next, parents) is
1146
method is the way referenced data should be applied.
1147
index_memo is the handle to pass to the data access to actually get the
1149
next is the build-parent of the version, or None for fulltexts.
1150
parents is the version_ids of the parents of this version
1152
:param allow_missing: If True do not raise an error on a missing component,
1156
pending_components = keys
1157
while pending_components:
1158
build_details = self._index.get_build_details(pending_components)
1159
current_components = set(pending_components)
1160
pending_components = set()
1161
for key, details in build_details.iteritems():
1162
(index_memo, compression_parent, parents,
1163
record_details) = details
1164
method = record_details[0]
1165
if compression_parent is not None:
1166
pending_components.add(compression_parent)
1167
component_data[key] = self._build_details_to_components(details)
1168
missing = current_components.difference(build_details)
1169
if missing and not allow_missing:
1170
raise errors.RevisionNotPresent(missing.pop(), self)
1171
return component_data
1173
def _get_content(self, key, parent_texts={}):
1174
"""Returns a content object that makes up the specified
1176
cached_version = parent_texts.get(key, None)
1177
if cached_version is not None:
1178
# Ensure the cache dict is valid.
1179
if not self.get_parent_map([key]):
1180
raise RevisionNotPresent(key, self)
1181
return cached_version
1182
generator = _VFContentMapGenerator(self, [key])
1183
return generator._get_content(key)
1185
def get_parent_map(self, keys):
1186
"""Get a map of the graph parents of keys.
1188
:param keys: The keys to look up parents for.
1189
:return: A mapping from keys to parents. Absent keys are absent from
1192
return self._get_parent_map_with_sources(keys)[0]
1194
def _get_parent_map_with_sources(self, keys):
1195
"""Get a map of the parents of keys.
1197
:param keys: The keys to look up parents for.
1198
:return: A tuple. The first element is a mapping from keys to parents.
1199
Absent keys are absent from the mapping. The second element is a
1200
list with the locations each key was found in. The first element
1201
is the in-this-knit parents, the second the first fallback source,
1205
sources = [self._index] + self._fallback_vfs
1208
for source in sources:
1211
new_result = source.get_parent_map(missing)
1212
source_results.append(new_result)
1213
result.update(new_result)
1214
missing.difference_update(set(new_result))
1215
return result, source_results
1217
def _get_record_map(self, keys, allow_missing=False):
1218
"""Produce a dictionary of knit records.
1220
:return: {key:(record, record_details, digest, next)}
1222
data returned from read_records (a KnitContentobject)
1224
opaque information to pass to parse_record
1226
SHA1 digest of the full text after all steps are done
1228
build-parent of the version, i.e. the leftmost ancestor.
1229
Will be None if the record is not a delta.
1230
:param keys: The keys to build a map for
1231
:param allow_missing: If some records are missing, rather than
1232
error, just return the data that could be generated.
1234
raw_map = self._get_record_map_unparsed(keys,
1235
allow_missing=allow_missing)
1236
return self._raw_map_to_record_map(raw_map)
1238
def _raw_map_to_record_map(self, raw_map):
1239
"""Parse the contents of _get_record_map_unparsed.
1241
:return: see _get_record_map.
1245
data, record_details, next = raw_map[key]
1246
content, digest = self._parse_record(key[-1], data)
1247
result[key] = content, record_details, digest, next
1250
def _get_record_map_unparsed(self, keys, allow_missing=False):
1251
"""Get the raw data for reconstructing keys without parsing it.
1253
:return: A dict suitable for parsing via _raw_map_to_record_map.
1254
key-> raw_bytes, (method, noeol), compression_parent
1256
# This retries the whole request if anything fails. Potentially we
1257
# could be a bit more selective. We could track the keys whose records
1258
# we have successfully found, and then only request the new records
1259
# from there. However, _get_components_positions grabs the whole build
1260
# chain, which means we'll likely try to grab the same records again
1261
# anyway. Also, can the build chains change as part of a pack
1262
# operation? We wouldn't want to end up with a broken chain.
1265
position_map = self._get_components_positions(keys,
1266
allow_missing=allow_missing)
1267
# key = component_id, r = record_details, i_m = index_memo,
1269
records = [(key, i_m) for key, (r, i_m, n)
1270
in position_map.iteritems()]
1271
# Sort by the index memo, so that we request records from the
1272
# same pack file together, and in forward-sorted order
1273
records.sort(key=operator.itemgetter(1))
1275
for key, data in self._read_records_iter_unchecked(records):
1276
(record_details, index_memo, next) = position_map[key]
1277
raw_record_map[key] = data, record_details, next
1278
return raw_record_map
1279
except errors.RetryWithNewPacks, e:
1280
self._access.reload_or_raise(e)
1283
def _split_by_prefix(cls, keys):
1284
"""For the given keys, split them up based on their prefix.
1286
To keep memory pressure somewhat under control, split the
1287
requests back into per-file-id requests, otherwise "bzr co"
1288
extracts the full tree into memory before writing it to disk.
1289
This should be revisited if _get_content_maps() can ever cross
1292
The keys for a given file_id are kept in the same relative order.
1293
Ordering between file_ids is not, though prefix_order will return the
1294
order that the key was first seen.
1296
:param keys: An iterable of key tuples
1297
:return: (split_map, prefix_order)
1298
split_map A dictionary mapping prefix => keys
1299
prefix_order The order that we saw the various prefixes
1301
split_by_prefix = {}
1309
if prefix in split_by_prefix:
1310
split_by_prefix[prefix].append(key)
1312
split_by_prefix[prefix] = [key]
1313
prefix_order.append(prefix)
1314
return split_by_prefix, prefix_order
1316
def _group_keys_for_io(self, keys, non_local_keys, positions,
1317
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1318
"""For the given keys, group them into 'best-sized' requests.
1320
The idea is to avoid making 1 request per file, but to never try to
1321
unpack an entire 1.5GB source tree in a single pass. Also when
1322
possible, we should try to group requests to the same pack file
1325
:return: list of (keys, non_local) tuples that indicate what keys
1326
should be fetched next.
1328
# TODO: Ideally we would group on 2 factors. We want to extract texts
1329
# from the same pack file together, and we want to extract all
1330
# the texts for a given build-chain together. Ultimately it
1331
# probably needs a better global view.
1332
total_keys = len(keys)
1333
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1334
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1336
cur_non_local = set()
1340
for prefix in prefix_order:
1341
keys = prefix_split_keys[prefix]
1342
non_local = prefix_split_non_local_keys.get(prefix, [])
1344
this_size = self._index._get_total_build_size(keys, positions)
1345
cur_size += this_size
1346
cur_keys.extend(keys)
1347
cur_non_local.update(non_local)
1348
if cur_size > _min_buffer_size:
1349
result.append((cur_keys, cur_non_local))
1350
sizes.append(cur_size)
1352
cur_non_local = set()
1355
result.append((cur_keys, cur_non_local))
1356
sizes.append(cur_size)
1359
def get_record_stream(self, keys, ordering, include_delta_closure):
1360
"""Get a stream of records for keys.
1362
:param keys: The keys to include.
1363
:param ordering: Either 'unordered' or 'topological'. A topologically
1364
sorted stream has compression parents strictly before their
1366
:param include_delta_closure: If True then the closure across any
1367
compression parents will be included (in the opaque data).
1368
:return: An iterator of ContentFactory objects, each of which is only
1369
valid until the iterator is advanced.
1371
# keys might be a generator
1375
if not self._index.has_graph:
1376
# Cannot sort when no graph has been stored.
1377
ordering = 'unordered'
1379
remaining_keys = keys
1382
keys = set(remaining_keys)
1383
for content_factory in self._get_remaining_record_stream(keys,
1384
ordering, include_delta_closure):
1385
remaining_keys.discard(content_factory.key)
1386
yield content_factory
1388
except errors.RetryWithNewPacks, e:
1389
self._access.reload_or_raise(e)
1391
def _get_remaining_record_stream(self, keys, ordering,
1392
include_delta_closure):
1393
"""This function is the 'retry' portion for get_record_stream."""
1394
if include_delta_closure:
1395
positions = self._get_components_positions(keys, allow_missing=True)
1397
build_details = self._index.get_build_details(keys)
1399
# (record_details, access_memo, compression_parent_key)
1400
positions = dict((key, self._build_details_to_components(details))
1401
for key, details in build_details.iteritems())
1402
absent_keys = keys.difference(set(positions))
1403
# There may be more absent keys : if we're missing the basis component
1404
# and are trying to include the delta closure.
1405
# XXX: We should not ever need to examine remote sources because we do
1406
# not permit deltas across versioned files boundaries.
1407
if include_delta_closure:
1408
needed_from_fallback = set()
1409
# Build up reconstructable_keys dict. key:True in this dict means
1410
# the key can be reconstructed.
1411
reconstructable_keys = {}
1415
chain = [key, positions[key][2]]
1417
needed_from_fallback.add(key)
1420
while chain[-1] is not None:
1421
if chain[-1] in reconstructable_keys:
1422
result = reconstructable_keys[chain[-1]]
1426
chain.append(positions[chain[-1]][2])
1428
# missing basis component
1429
needed_from_fallback.add(chain[-1])
1432
for chain_key in chain[:-1]:
1433
reconstructable_keys[chain_key] = result
1435
needed_from_fallback.add(key)
1436
# Double index lookups here : need a unified api ?
1437
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1438
if ordering in ('topological', 'groupcompress'):
1439
if ordering == 'topological':
1440
# Global topological sort
1441
present_keys = tsort.topo_sort(global_map)
1443
present_keys = sort_groupcompress(global_map)
1444
# Now group by source:
1446
current_source = None
1447
for key in present_keys:
1448
for parent_map in parent_maps:
1449
if key in parent_map:
1450
key_source = parent_map
1452
if current_source is not key_source:
1453
source_keys.append((key_source, []))
1454
current_source = key_source
1455
source_keys[-1][1].append(key)
1457
if ordering != 'unordered':
1458
raise AssertionError('valid values for ordering are:'
1459
' "unordered", "groupcompress" or "topological" not: %r'
1461
# Just group by source; remote sources first.
1464
for parent_map in reversed(parent_maps):
1465
source_keys.append((parent_map, []))
1466
for key in parent_map:
1467
present_keys.append(key)
1468
source_keys[-1][1].append(key)
1469
# We have been requested to return these records in an order that
1470
# suits us. So we ask the index to give us an optimally sorted
1472
for source, sub_keys in source_keys:
1473
if source is parent_maps[0]:
1474
# Only sort the keys for this VF
1475
self._index._sort_keys_by_io(sub_keys, positions)
1476
absent_keys = keys - set(global_map)
1477
for key in absent_keys:
1478
yield AbsentContentFactory(key)
1479
# restrict our view to the keys we can answer.
1480
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1481
# XXX: At that point we need to consider the impact of double reads by
1482
# utilising components multiple times.
1483
if include_delta_closure:
1484
# XXX: get_content_maps performs its own index queries; allow state
1486
non_local_keys = needed_from_fallback - absent_keys
1487
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1490
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1492
for record in generator.get_record_stream():
1495
for source, keys in source_keys:
1496
if source is parent_maps[0]:
1497
# this KnitVersionedFiles
1498
records = [(key, positions[key][1]) for key in keys]
1499
for key, raw_data, sha1 in self._read_records_iter_raw(records):
1500
(record_details, index_memo, _) = positions[key]
1501
yield KnitContentFactory(key, global_map[key],
1502
record_details, sha1, raw_data, self._factory.annotated, None)
1504
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1505
for record in vf.get_record_stream(keys, ordering,
1506
include_delta_closure):
1509
def get_sha1s(self, keys):
1510
"""See VersionedFiles.get_sha1s()."""
1512
record_map = self._get_record_map(missing, allow_missing=True)
1514
for key, details in record_map.iteritems():
1515
if key not in missing:
1517
# record entry 2 is the 'digest'.
1518
result[key] = details[2]
1519
missing.difference_update(set(result))
1520
for source in self._fallback_vfs:
1523
new_result = source.get_sha1s(missing)
1524
result.update(new_result)
1525
missing.difference_update(set(new_result))
1528
def insert_record_stream(self, stream):
1529
"""Insert a record stream into this container.
1531
:param stream: A stream of records to insert.
1533
:seealso VersionedFiles.get_record_stream:
1535
def get_adapter(adapter_key):
1537
return adapters[adapter_key]
1539
adapter_factory = adapter_registry.get(adapter_key)
1540
adapter = adapter_factory(self)
1541
adapters[adapter_key] = adapter
1544
if self._factory.annotated:
1545
# self is annotated, we need annotated knits to use directly.
1546
annotated = "annotated-"
1549
# self is not annotated, but we can strip annotations cheaply.
1551
convertibles = set(["knit-annotated-ft-gz"])
1552
if self._max_delta_chain:
1553
delta_types.add("knit-annotated-delta-gz")
1554
convertibles.add("knit-annotated-delta-gz")
1555
# The set of types we can cheaply adapt without needing basis texts.
1556
native_types = set()
1557
if self._max_delta_chain:
1558
native_types.add("knit-%sdelta-gz" % annotated)
1559
delta_types.add("knit-%sdelta-gz" % annotated)
1560
native_types.add("knit-%sft-gz" % annotated)
1561
knit_types = native_types.union(convertibles)
1563
# Buffer all index entries that we can't add immediately because their
1564
# basis parent is missing. We don't buffer all because generating
1565
# annotations may require access to some of the new records. However we
1566
# can't generate annotations from new deltas until their basis parent
1567
# is present anyway, so we get away with not needing an index that
1568
# includes the new keys.
1570
# See <http://launchpad.net/bugs/300177> about ordering of compression
1571
# parents in the records - to be conservative, we insist that all
1572
# parents must be present to avoid expanding to a fulltext.
1574
# key = basis_parent, value = index entry to add
1575
buffered_index_entries = {}
1576
for record in stream:
1578
parents = record.parents
1579
if record.storage_kind in delta_types:
1580
# TODO: eventually the record itself should track
1581
# compression_parent
1582
compression_parent = parents[0]
1584
compression_parent = None
1585
# Raise an error when a record is missing.
1586
if record.storage_kind == 'absent':
1587
raise RevisionNotPresent([record.key], self)
1588
elif ((record.storage_kind in knit_types)
1589
and (compression_parent is None
1590
or not self._fallback_vfs
1591
or self._index.has_key(compression_parent)
1592
or not self.has_key(compression_parent))):
1593
# we can insert the knit record literally if either it has no
1594
# compression parent OR we already have its basis in this kvf
1595
# OR the basis is not present even in the fallbacks. In the
1596
# last case it will either turn up later in the stream and all
1597
# will be well, or it won't turn up at all and we'll raise an
1600
# TODO: self.has_key is somewhat redundant with
1601
# self._index.has_key; we really want something that directly
1602
# asks if it's only present in the fallbacks. -- mbp 20081119
1603
if record.storage_kind not in native_types:
1605
adapter_key = (record.storage_kind, "knit-delta-gz")
1606
adapter = get_adapter(adapter_key)
1608
adapter_key = (record.storage_kind, "knit-ft-gz")
1609
adapter = get_adapter(adapter_key)
1610
bytes = adapter.get_bytes(record)
1612
# It's a knit record, it has a _raw_record field (even if
1613
# it was reconstituted from a network stream).
1614
bytes = record._raw_record
1615
options = [record._build_details[0]]
1616
if record._build_details[1]:
1617
options.append('no-eol')
1618
# Just blat it across.
1619
# Note: This does end up adding data on duplicate keys. As
1620
# modern repositories use atomic insertions this should not
1621
# lead to excessive growth in the event of interrupted fetches.
1622
# 'knit' repositories may suffer excessive growth, but as a
1623
# deprecated format this is tolerable. It can be fixed if
1624
# needed by in the kndx index support raising on a duplicate
1625
# add with identical parents and options.
1626
access_memo = self._access.add_raw_records(
1627
[(record.key, len(bytes))], bytes)[0]
1628
index_entry = (record.key, options, access_memo, parents)
1629
if 'fulltext' not in options:
1630
# Not a fulltext, so we need to make sure the compression
1631
# parent will also be present.
1632
# Note that pack backed knits don't need to buffer here
1633
# because they buffer all writes to the transaction level,
1634
# but we don't expose that difference at the index level. If
1635
# the query here has sufficient cost to show up in
1636
# profiling we should do that.
1638
# They're required to be physically in this
1639
# KnitVersionedFiles, not in a fallback.
1640
if not self._index.has_key(compression_parent):
1641
pending = buffered_index_entries.setdefault(
1642
compression_parent, [])
1643
pending.append(index_entry)
1646
self._index.add_records([index_entry])
1647
elif record.storage_kind == 'chunked':
1648
self.add_lines(record.key, parents,
1649
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1651
# Not suitable for direct insertion as a
1652
# delta, either because it's not the right format, or this
1653
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1654
# 0) or because it depends on a base only present in the
1656
self._access.flush()
1658
# Try getting a fulltext directly from the record.
1659
bytes = record.get_bytes_as('fulltext')
1660
except errors.UnavailableRepresentation:
1661
adapter_key = record.storage_kind, 'fulltext'
1662
adapter = get_adapter(adapter_key)
1663
bytes = adapter.get_bytes(record)
1664
lines = split_lines(bytes)
1666
self.add_lines(record.key, parents, lines)
1667
except errors.RevisionAlreadyPresent:
1669
# Add any records whose basis parent is now available.
1671
added_keys = [record.key]
1673
key = added_keys.pop(0)
1674
if key in buffered_index_entries:
1675
index_entries = buffered_index_entries[key]
1676
self._index.add_records(index_entries)
1678
[index_entry[0] for index_entry in index_entries])
1679
del buffered_index_entries[key]
1680
if buffered_index_entries:
1681
# There were index entries buffered at the end of the stream,
1682
# So these need to be added (if the index supports holding such
1683
# entries for later insertion)
1684
for key in buffered_index_entries:
1685
index_entries = buffered_index_entries[key]
1686
self._index.add_records(index_entries,
1687
missing_compression_parents=True)
1689
def get_missing_compression_parent_keys(self):
1690
"""Return an iterable of keys of missing compression parents.
1692
Check this after calling insert_record_stream to find out if there are
1693
any missing compression parents. If there are, the records that
1694
depend on them are not able to be inserted safely. For atomic
1695
KnitVersionedFiles built on packs, the transaction should be aborted or
1696
suspended - commit will fail at this point. Nonatomic knits will error
1697
earlier because they have no staging area to put pending entries into.
1699
return self._index.get_missing_compression_parents()
1701
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1702
"""Iterate over the lines in the versioned files from keys.
1704
This may return lines from other keys. Each item the returned
1705
iterator yields is a tuple of a line and a text version that that line
1706
is present in (not introduced in).
1708
Ordering of results is in whatever order is most suitable for the
1709
underlying storage format.
1711
If a progress bar is supplied, it may be used to indicate progress.
1712
The caller is responsible for cleaning up progress bars (because this
1716
* Lines are normalised by the underlying store: they will all have \\n
1718
* Lines are returned in arbitrary order.
1719
* If a requested key did not change any lines (or didn't have any
1720
lines), it may not be mentioned at all in the result.
1722
:param pb: Progress bar supplied by caller.
1723
:return: An iterator over (line, key).
1726
pb = progress.DummyProgress()
1732
# we don't care about inclusions, the caller cares.
1733
# but we need to setup a list of records to visit.
1734
# we need key, position, length
1736
build_details = self._index.get_build_details(keys)
1737
for key, details in build_details.iteritems():
1739
key_records.append((key, details[0]))
1740
records_iter = enumerate(self._read_records_iter(key_records))
1741
for (key_idx, (key, data, sha_value)) in records_iter:
1742
pb.update('Walking content', key_idx, total)
1743
compression_parent = build_details[key][1]
1744
if compression_parent is None:
1746
line_iterator = self._factory.get_fulltext_content(data)
1749
line_iterator = self._factory.get_linedelta_content(data)
1750
# Now that we are yielding the data for this key, remove it
1753
# XXX: It might be more efficient to yield (key,
1754
# line_iterator) in the future. However for now, this is a
1755
# simpler change to integrate into the rest of the
1756
# codebase. RBC 20071110
1757
for line in line_iterator:
1760
except errors.RetryWithNewPacks, e:
1761
self._access.reload_or_raise(e)
1762
# If there are still keys we've not yet found, we look in the fallback
1763
# vfs, and hope to find them there. Note that if the keys are found
1764
# but had no changes or no content, the fallback may not return
1766
if keys and not self._fallback_vfs:
1767
# XXX: strictly the second parameter is meant to be the file id
1768
# but it's not easily accessible here.
1769
raise RevisionNotPresent(keys, repr(self))
1770
for source in self._fallback_vfs:
1774
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1775
source_keys.add(key)
1777
keys.difference_update(source_keys)
1778
pb.update('Walking content', total, total)
1780
def _make_line_delta(self, delta_seq, new_content):
1781
"""Generate a line delta from delta_seq and new_content."""
1783
for op in delta_seq.get_opcodes():
1784
if op[0] == 'equal':
1786
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1789
def _merge_annotations(self, content, parents, parent_texts={},
1790
delta=None, annotated=None,
1791
left_matching_blocks=None):
1792
"""Merge annotations for content and generate deltas.
1794
This is done by comparing the annotations based on changes to the text
1795
and generating a delta on the resulting full texts. If annotations are
1796
not being created then a simple delta is created.
1798
if left_matching_blocks is not None:
1799
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1803
for parent_key in parents:
1804
merge_content = self._get_content(parent_key, parent_texts)
1805
if (parent_key == parents[0] and delta_seq is not None):
1808
seq = patiencediff.PatienceSequenceMatcher(
1809
None, merge_content.text(), content.text())
1810
for i, j, n in seq.get_matching_blocks():
1813
# this copies (origin, text) pairs across to the new
1814
# content for any line that matches the last-checked
1816
content._lines[j:j+n] = merge_content._lines[i:i+n]
1817
# XXX: Robert says the following block is a workaround for a
1818
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1819
if content._lines and content._lines[-1][1][-1] != '\n':
1820
# The copied annotation was from a line without a trailing EOL,
1821
# reinstate one for the content object, to ensure correct
1823
line = content._lines[-1][1] + '\n'
1824
content._lines[-1] = (content._lines[-1][0], line)
1826
if delta_seq is None:
1827
reference_content = self._get_content(parents[0], parent_texts)
1828
new_texts = content.text()
1829
old_texts = reference_content.text()
1830
delta_seq = patiencediff.PatienceSequenceMatcher(
1831
None, old_texts, new_texts)
1832
return self._make_line_delta(delta_seq, content)
1834
def _parse_record(self, version_id, data):
1835
"""Parse an original format knit record.
1837
These have the last element of the key only present in the stored data.
1839
rec, record_contents = self._parse_record_unchecked(data)
1840
self._check_header_version(rec, version_id)
1841
return record_contents, rec[3]
1843
def _parse_record_header(self, key, raw_data):
1844
"""Parse a record header for consistency.
1846
:return: the header and the decompressor stream.
1847
as (stream, header_record)
1849
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1852
rec = self._check_header(key, df.readline())
1853
except Exception, e:
1854
raise KnitCorrupt(self,
1855
"While reading {%s} got %s(%s)"
1856
% (key, e.__class__.__name__, str(e)))
1859
def _parse_record_unchecked(self, data):
1861
# 4168 calls in 2880 217 internal
1862
# 4168 calls to _parse_record_header in 2121
1863
# 4168 calls to readlines in 330
1864
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1866
record_contents = df.readlines()
1867
except Exception, e:
1868
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1869
(data, e.__class__.__name__, str(e)))
1870
header = record_contents.pop(0)
1871
rec = self._split_header(header)
1872
last_line = record_contents.pop()
1873
if len(record_contents) != int(rec[2]):
1874
raise KnitCorrupt(self,
1875
'incorrect number of lines %s != %s'
1876
' for version {%s} %s'
1877
% (len(record_contents), int(rec[2]),
1878
rec[1], record_contents))
1879
if last_line != 'end %s\n' % rec[1]:
1880
raise KnitCorrupt(self,
1881
'unexpected version end line %r, wanted %r'
1882
% (last_line, rec[1]))
1884
return rec, record_contents
1886
def _read_records_iter(self, records):
1887
"""Read text records from data file and yield result.
1889
The result will be returned in whatever is the fastest to read.
1890
Not by the order requested. Also, multiple requests for the same
1891
record will only yield 1 response.
1892
:param records: A list of (key, access_memo) entries
1893
:return: Yields (key, contents, digest) in the order
1894
read, not the order requested
1899
# XXX: This smells wrong, IO may not be getting ordered right.
1900
needed_records = sorted(set(records), key=operator.itemgetter(1))
1901
if not needed_records:
1904
# The transport optimizes the fetching as well
1905
# (ie, reads continuous ranges.)
1906
raw_data = self._access.get_raw_records(
1907
[index_memo for key, index_memo in needed_records])
1909
for (key, index_memo), data in \
1910
izip(iter(needed_records), raw_data):
1911
content, digest = self._parse_record(key[-1], data)
1912
yield key, content, digest
1914
def _read_records_iter_raw(self, records):
1915
"""Read text records from data file and yield raw data.
1917
This unpacks enough of the text record to validate the id is
1918
as expected but thats all.
1920
Each item the iterator yields is (key, bytes,
1921
expected_sha1_of_full_text).
1923
for key, data in self._read_records_iter_unchecked(records):
1924
# validate the header (note that we can only use the suffix in
1925
# current knit records).
1926
df, rec = self._parse_record_header(key, data)
1928
yield key, data, rec[3]
1930
def _read_records_iter_unchecked(self, records):
1931
"""Read text records from data file and yield raw data.
1933
No validation is done.
1935
Yields tuples of (key, data).
1937
# setup an iterator of the external records:
1938
# uses readv so nice and fast we hope.
1940
# grab the disk data needed.
1941
needed_offsets = [index_memo for key, index_memo
1943
raw_records = self._access.get_raw_records(needed_offsets)
1945
for key, index_memo in records:
1946
data = raw_records.next()
1949
def _record_to_data(self, key, digest, lines, dense_lines=None):
1950
"""Convert key, digest, lines into a raw data block.
1952
:param key: The key of the record. Currently keys are always serialised
1953
using just the trailing component.
1954
:param dense_lines: The bytes of lines but in a denser form. For
1955
instance, if lines is a list of 1000 bytestrings each ending in \n,
1956
dense_lines may be a list with one line in it, containing all the
1957
1000's lines and their \n's. Using dense_lines if it is already
1958
known is a win because the string join to create bytes in this
1959
function spends less time resizing the final string.
1960
:return: (len, a StringIO instance with the raw data ready to read.)
1962
chunks = ["version %s %d %s\n" % (key[-1], len(lines), digest)]
1963
chunks.extend(dense_lines or lines)
1964
chunks.append("end %s\n" % key[-1])
1965
for chunk in chunks:
1966
if type(chunk) is not str:
1967
raise AssertionError(
1968
'data must be plain bytes was %s' % type(chunk))
1969
if lines and lines[-1][-1] != '\n':
1970
raise ValueError('corrupt lines value %r' % lines)
1971
compressed_bytes = tuned_gzip.chunks_to_gzip(chunks)
1972
return len(compressed_bytes), compressed_bytes
1974
def _split_header(self, line):
1977
raise KnitCorrupt(self,
1978
'unexpected number of elements in record header')
1982
"""See VersionedFiles.keys."""
1983
if 'evil' in debug.debug_flags:
1984
trace.mutter_callsite(2, "keys scales with size of history")
1985
sources = [self._index] + self._fallback_vfs
1987
for source in sources:
1988
result.update(source.keys())
1992
class _ContentMapGenerator(object):
1993
"""Generate texts or expose raw deltas for a set of texts."""
1995
def _get_content(self, key):
1996
"""Get the content object for key."""
1997
# Note that _get_content is only called when the _ContentMapGenerator
1998
# has been constructed with just one key requested for reconstruction.
1999
if key in self.nonlocal_keys:
2000
record = self.get_record_stream().next()
2001
# Create a content object on the fly
2002
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
2003
return PlainKnitContent(lines, record.key)
2005
# local keys we can ask for directly
2006
return self._get_one_work(key)
2008
def get_record_stream(self):
2009
"""Get a record stream for the keys requested during __init__."""
2010
for record in self._work():
2014
"""Produce maps of text and KnitContents as dicts.
2016
:return: (text_map, content_map) where text_map contains the texts for
2017
the requested versions and content_map contains the KnitContents.
2019
# NB: By definition we never need to read remote sources unless texts
2020
# are requested from them: we don't delta across stores - and we
2021
# explicitly do not want to to prevent data loss situations.
2022
if self.global_map is None:
2023
self.global_map = self.vf.get_parent_map(self.keys)
2024
nonlocal_keys = self.nonlocal_keys
2026
missing_keys = set(nonlocal_keys)
2027
# Read from remote versioned file instances and provide to our caller.
2028
for source in self.vf._fallback_vfs:
2029
if not missing_keys:
2031
# Loop over fallback repositories asking them for texts - ignore
2032
# any missing from a particular fallback.
2033
for record in source.get_record_stream(missing_keys,
2035
if record.storage_kind == 'absent':
2036
# Not in thie particular stream, may be in one of the
2037
# other fallback vfs objects.
2039
missing_keys.remove(record.key)
2042
if self._raw_record_map is None:
2043
raise AssertionError('_raw_record_map should have been filled')
2045
for key in self.keys:
2046
if key in self.nonlocal_keys:
2048
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2051
def _get_one_work(self, requested_key):
2052
# Now, if we have calculated everything already, just return the
2054
if requested_key in self._contents_map:
2055
return self._contents_map[requested_key]
2056
# To simplify things, parse everything at once - code that wants one text
2057
# probably wants them all.
2058
# FUTURE: This function could be improved for the 'extract many' case
2059
# by tracking each component and only doing the copy when the number of
2060
# children than need to apply delta's to it is > 1 or it is part of the
2062
multiple_versions = len(self.keys) != 1
2063
if self._record_map is None:
2064
self._record_map = self.vf._raw_map_to_record_map(
2065
self._raw_record_map)
2066
record_map = self._record_map
2067
# raw_record_map is key:
2068
# Have read and parsed records at this point.
2069
for key in self.keys:
2070
if key in self.nonlocal_keys:
2075
while cursor is not None:
2077
record, record_details, digest, next = record_map[cursor]
2079
raise RevisionNotPresent(cursor, self)
2080
components.append((cursor, record, record_details, digest))
2082
if cursor in self._contents_map:
2083
# no need to plan further back
2084
components.append((cursor, None, None, None))
2088
for (component_id, record, record_details,
2089
digest) in reversed(components):
2090
if component_id in self._contents_map:
2091
content = self._contents_map[component_id]
2093
content, delta = self._factory.parse_record(key[-1],
2094
record, record_details, content,
2095
copy_base_content=multiple_versions)
2096
if multiple_versions:
2097
self._contents_map[component_id] = content
2099
# digest here is the digest from the last applied component.
2100
text = content.text()
2101
actual_sha = sha_strings(text)
2102
if actual_sha != digest:
2103
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2104
if multiple_versions:
2105
return self._contents_map[requested_key]
2109
def _wire_bytes(self):
2110
"""Get the bytes to put on the wire for 'key'.
2112
The first collection of bytes asked for returns the serialised
2113
raw_record_map and the additional details (key, parent) for key.
2114
Subsequent calls return just the additional details (key, parent).
2115
The wire storage_kind given for the first key is 'knit-delta-closure',
2116
For subsequent keys it is 'knit-delta-closure-ref'.
2118
:param key: A key from the content generator.
2119
:return: Bytes to put on the wire.
2122
# kind marker for dispatch on the far side,
2123
lines.append('knit-delta-closure')
2125
if self.vf._factory.annotated:
2126
lines.append('annotated')
2129
# then the list of keys
2130
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2131
if key not in self.nonlocal_keys]))
2132
# then the _raw_record_map in serialised form:
2134
# for each item in the map:
2136
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2137
# one line with method
2138
# one line with noeol
2139
# one line with next ('' for None)
2140
# one line with byte count of the record bytes
2142
for key, (record_bytes, (method, noeol), next) in \
2143
self._raw_record_map.iteritems():
2144
key_bytes = '\x00'.join(key)
2145
parents = self.global_map.get(key, None)
2147
parent_bytes = 'None:'
2149
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2150
method_bytes = method
2156
next_bytes = '\x00'.join(next)
2159
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2160
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2161
len(record_bytes), record_bytes))
2162
map_bytes = ''.join(map_byte_list)
2163
lines.append(map_bytes)
2164
bytes = '\n'.join(lines)
2168
class _VFContentMapGenerator(_ContentMapGenerator):
2169
"""Content map generator reading from a VersionedFiles object."""
2171
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2172
global_map=None, raw_record_map=None):
2173
"""Create a _ContentMapGenerator.
2175
:param versioned_files: The versioned files that the texts are being
2177
:param keys: The keys to produce content maps for.
2178
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2179
which are known to not be in this knit, but rather in one of the
2181
:param global_map: The result of get_parent_map(keys) (or a supermap).
2182
This is required if get_record_stream() is to be used.
2183
:param raw_record_map: A unparsed raw record map to use for answering
2186
# The vf to source data from
2187
self.vf = versioned_files
2189
self.keys = list(keys)
2190
# Keys known to be in fallback vfs objects
2191
if nonlocal_keys is None:
2192
self.nonlocal_keys = set()
2194
self.nonlocal_keys = frozenset(nonlocal_keys)
2195
# Parents data for keys to be returned in get_record_stream
2196
self.global_map = global_map
2197
# The chunked lists for self.keys in text form
2199
# A cache of KnitContent objects used in extracting texts.
2200
self._contents_map = {}
2201
# All the knit records needed to assemble the requested keys as full
2203
self._record_map = None
2204
if raw_record_map is None:
2205
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2208
self._raw_record_map = raw_record_map
2209
# the factory for parsing records
2210
self._factory = self.vf._factory
2213
class _NetworkContentMapGenerator(_ContentMapGenerator):
2214
"""Content map generator sourced from a network stream."""
2216
def __init__(self, bytes, line_end):
2217
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2219
self.global_map = {}
2220
self._raw_record_map = {}
2221
self._contents_map = {}
2222
self._record_map = None
2223
self.nonlocal_keys = []
2224
# Get access to record parsing facilities
2225
self.vf = KnitVersionedFiles(None, None)
2228
line_end = bytes.find('\n', start)
2229
line = bytes[start:line_end]
2230
start = line_end + 1
2231
if line == 'annotated':
2232
self._factory = KnitAnnotateFactory()
2234
self._factory = KnitPlainFactory()
2235
# list of keys to emit in get_record_stream
2236
line_end = bytes.find('\n', start)
2237
line = bytes[start:line_end]
2238
start = line_end + 1
2240
tuple(segment.split('\x00')) for segment in line.split('\t')
2242
# now a loop until the end. XXX: It would be nice if this was just a
2243
# bunch of the same records as get_record_stream(..., False) gives, but
2244
# there is a decent sized gap stopping that at the moment.
2248
line_end = bytes.find('\n', start)
2249
key = tuple(bytes[start:line_end].split('\x00'))
2250
start = line_end + 1
2251
# 1 line with parents (None: for None, '' for ())
2252
line_end = bytes.find('\n', start)
2253
line = bytes[start:line_end]
2258
[tuple(segment.split('\x00')) for segment in line.split('\t')
2260
self.global_map[key] = parents
2261
start = line_end + 1
2262
# one line with method
2263
line_end = bytes.find('\n', start)
2264
line = bytes[start:line_end]
2266
start = line_end + 1
2267
# one line with noeol
2268
line_end = bytes.find('\n', start)
2269
line = bytes[start:line_end]
2271
start = line_end + 1
2272
# one line with next ('' for None)
2273
line_end = bytes.find('\n', start)
2274
line = bytes[start:line_end]
2278
next = tuple(bytes[start:line_end].split('\x00'))
2279
start = line_end + 1
2280
# one line with byte count of the record bytes
2281
line_end = bytes.find('\n', start)
2282
line = bytes[start:line_end]
2284
start = line_end + 1
2286
record_bytes = bytes[start:start+count]
2287
start = start + count
2289
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2291
def get_record_stream(self):
2292
"""Get a record stream for for keys requested by the bytestream."""
2294
for key in self.keys:
2295
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2298
def _wire_bytes(self):
2302
class _KndxIndex(object):
2303
"""Manages knit index files
2305
The index is kept in memory and read on startup, to enable
2306
fast lookups of revision information. The cursor of the index
2307
file is always pointing to the end, making it easy to append
2310
_cache is a cache for fast mapping from version id to a Index
2313
_history is a cache for fast mapping from indexes to version ids.
2315
The index data format is dictionary compressed when it comes to
2316
parent references; a index entry may only have parents that with a
2317
lover index number. As a result, the index is topological sorted.
2319
Duplicate entries may be written to the index for a single version id
2320
if this is done then the latter one completely replaces the former:
2321
this allows updates to correct version and parent information.
2322
Note that the two entries may share the delta, and that successive
2323
annotations and references MUST point to the first entry.
2325
The index file on disc contains a header, followed by one line per knit
2326
record. The same revision can be present in an index file more than once.
2327
The first occurrence gets assigned a sequence number starting from 0.
2329
The format of a single line is
2330
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
2331
REVISION_ID is a utf8-encoded revision id
2332
FLAGS is a comma separated list of flags about the record. Values include
2333
no-eol, line-delta, fulltext.
2334
BYTE_OFFSET is the ascii representation of the byte offset in the data file
2335
that the the compressed data starts at.
2336
LENGTH is the ascii representation of the length of the data file.
2337
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
2339
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
2340
revision id already in the knit that is a parent of REVISION_ID.
2341
The ' :' marker is the end of record marker.
2344
when a write is interrupted to the index file, it will result in a line
2345
that does not end in ' :'. If the ' :' is not present at the end of a line,
2346
or at the end of the file, then the record that is missing it will be
2347
ignored by the parser.
2349
When writing new records to the index file, the data is preceded by '\n'
2350
to ensure that records always start on new lines even if the last write was
2351
interrupted. As a result its normal for the last line in the index to be
2352
missing a trailing newline. One can be added with no harmful effects.
2354
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2355
where prefix is e.g. the (fileid,) for .texts instances or () for
2356
constant-mapped things like .revisions, and the old state is
2357
tuple(history_vector, cache_dict). This is used to prevent having an
2358
ABI change with the C extension that reads .kndx files.
2361
HEADER = "# bzr knit index 8\n"
2363
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2364
"""Create a _KndxIndex on transport using mapper."""
2365
self._transport = transport
2366
self._mapper = mapper
2367
self._get_scope = get_scope
2368
self._allow_writes = allow_writes
2369
self._is_locked = is_locked
2371
self.has_graph = True
2373
def add_records(self, records, random_id=False, missing_compression_parents=False):
2374
"""Add multiple records to the index.
2376
:param records: a list of tuples:
2377
(key, options, access_memo, parents).
2378
:param random_id: If True the ids being added were randomly generated
2379
and no check for existence will be performed.
2380
:param missing_compression_parents: If True the records being added are
2381
only compressed against texts already in the index (or inside
2382
records). If False the records all refer to unavailable texts (or
2383
texts inside records) as compression parents.
2385
if missing_compression_parents:
2386
# It might be nice to get the edge of the records. But keys isn't
2388
keys = sorted(record[0] for record in records)
2389
raise errors.RevisionNotPresent(keys, self)
2391
for record in records:
2394
path = self._mapper.map(key) + '.kndx'
2395
path_keys = paths.setdefault(path, (prefix, []))
2396
path_keys[1].append(record)
2397
for path in sorted(paths):
2398
prefix, path_keys = paths[path]
2399
self._load_prefixes([prefix])
2401
orig_history = self._kndx_cache[prefix][1][:]
2402
orig_cache = self._kndx_cache[prefix][0].copy()
2405
for key, options, (_, pos, size), parents in path_keys:
2407
# kndx indices cannot be parentless.
2409
line = "\n%s %s %s %s %s :" % (
2410
key[-1], ','.join(options), pos, size,
2411
self._dictionary_compress(parents))
2412
if type(line) is not str:
2413
raise AssertionError(
2414
'data must be utf8 was %s' % type(line))
2416
self._cache_key(key, options, pos, size, parents)
2417
if len(orig_history):
2418
self._transport.append_bytes(path, ''.join(lines))
2420
self._init_index(path, lines)
2422
# If any problems happen, restore the original values and re-raise
2423
self._kndx_cache[prefix] = (orig_cache, orig_history)
2426
def scan_unvalidated_index(self, graph_index):
2427
"""See _KnitGraphIndex.scan_unvalidated_index."""
2428
# Because kndx files do not support atomic insertion via separate index
2429
# files, they do not support this method.
2430
raise NotImplementedError(self.scan_unvalidated_index)
2432
def get_missing_compression_parents(self):
2433
"""See _KnitGraphIndex.get_missing_compression_parents."""
2434
# Because kndx files do not support atomic insertion via separate index
2435
# files, they do not support this method.
2436
raise NotImplementedError(self.get_missing_compression_parents)
2438
def _cache_key(self, key, options, pos, size, parent_keys):
2439
"""Cache a version record in the history array and index cache.
2441
This is inlined into _load_data for performance. KEEP IN SYNC.
2442
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2446
version_id = key[-1]
2447
# last-element only for compatibilty with the C load_data.
2448
parents = tuple(parent[-1] for parent in parent_keys)
2449
for parent in parent_keys:
2450
if parent[:-1] != prefix:
2451
raise ValueError("mismatched prefixes for %r, %r" % (
2453
cache, history = self._kndx_cache[prefix]
2454
# only want the _history index to reference the 1st index entry
2456
if version_id not in cache:
2457
index = len(history)
2458
history.append(version_id)
2460
index = cache[version_id][5]
2461
cache[version_id] = (version_id,
2468
def check_header(self, fp):
2469
line = fp.readline()
2471
# An empty file can actually be treated as though the file doesn't
2473
raise errors.NoSuchFile(self)
2474
if line != self.HEADER:
2475
raise KnitHeaderError(badline=line, filename=self)
2477
def _check_read(self):
2478
if not self._is_locked():
2479
raise errors.ObjectNotLocked(self)
2480
if self._get_scope() != self._scope:
2483
def _check_write_ok(self):
2484
"""Assert if not writes are permitted."""
2485
if not self._is_locked():
2486
raise errors.ObjectNotLocked(self)
2487
if self._get_scope() != self._scope:
2489
if self._mode != 'w':
2490
raise errors.ReadOnlyObjectDirtiedError(self)
2492
def get_build_details(self, keys):
2493
"""Get the method, index_memo and compression parent for keys.
2495
Ghosts are omitted from the result.
2497
:param keys: An iterable of keys.
2498
:return: A dict of key:(index_memo, compression_parent, parents,
2501
opaque structure to pass to read_records to extract the raw
2504
Content that this record is built upon, may be None
2506
Logical parents of this node
2508
extra information about the content which needs to be passed to
2509
Factory.parse_record
2511
parent_map = self.get_parent_map(keys)
2514
if key not in parent_map:
2516
method = self.get_method(key)
2517
parents = parent_map[key]
2518
if method == 'fulltext':
2519
compression_parent = None
2521
compression_parent = parents[0]
2522
noeol = 'no-eol' in self.get_options(key)
2523
index_memo = self.get_position(key)
2524
result[key] = (index_memo, compression_parent,
2525
parents, (method, noeol))
2528
def get_method(self, key):
2529
"""Return compression method of specified key."""
2530
options = self.get_options(key)
2531
if 'fulltext' in options:
2533
elif 'line-delta' in options:
2536
raise errors.KnitIndexUnknownMethod(self, options)
2538
def get_options(self, key):
2539
"""Return a list representing options.
2543
prefix, suffix = self._split_key(key)
2544
self._load_prefixes([prefix])
2546
return self._kndx_cache[prefix][0][suffix][1]
2548
raise RevisionNotPresent(key, self)
2550
def get_parent_map(self, keys):
2551
"""Get a map of the parents of keys.
2553
:param keys: The keys to look up parents for.
2554
:return: A mapping from keys to parents. Absent keys are absent from
2557
# Parse what we need to up front, this potentially trades off I/O
2558
# locality (.kndx and .knit in the same block group for the same file
2559
# id) for less checking in inner loops.
2560
prefixes = set(key[:-1] for key in keys)
2561
self._load_prefixes(prefixes)
2566
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2570
result[key] = tuple(prefix + (suffix,) for
2571
suffix in suffix_parents)
2574
def get_position(self, key):
2575
"""Return details needed to access the version.
2577
:return: a tuple (key, data position, size) to hand to the access
2578
logic to get the record.
2580
prefix, suffix = self._split_key(key)
2581
self._load_prefixes([prefix])
2582
entry = self._kndx_cache[prefix][0][suffix]
2583
return key, entry[2], entry[3]
2585
has_key = _mod_index._has_key_from_parent_map
2587
def _init_index(self, path, extra_lines=[]):
2588
"""Initialize an index."""
2590
sio.write(self.HEADER)
2591
sio.writelines(extra_lines)
2593
self._transport.put_file_non_atomic(path, sio,
2594
create_parent_dir=True)
2595
# self._create_parent_dir)
2596
# mode=self._file_mode,
2597
# dir_mode=self._dir_mode)
2600
"""Get all the keys in the collection.
2602
The keys are not ordered.
2605
# Identify all key prefixes.
2606
# XXX: A bit hacky, needs polish.
2607
if type(self._mapper) is ConstantMapper:
2611
for quoted_relpath in self._transport.iter_files_recursive():
2612
path, ext = os.path.splitext(quoted_relpath)
2614
prefixes = [self._mapper.unmap(path) for path in relpaths]
2615
self._load_prefixes(prefixes)
2616
for prefix in prefixes:
2617
for suffix in self._kndx_cache[prefix][1]:
2618
result.add(prefix + (suffix,))
2621
def _load_prefixes(self, prefixes):
2622
"""Load the indices for prefixes."""
2624
for prefix in prefixes:
2625
if prefix not in self._kndx_cache:
2626
# the load_data interface writes to these variables.
2629
self._filename = prefix
2631
path = self._mapper.map(prefix) + '.kndx'
2632
fp = self._transport.get(path)
2634
# _load_data may raise NoSuchFile if the target knit is
2636
_load_data(self, fp)
2639
self._kndx_cache[prefix] = (self._cache, self._history)
2644
self._kndx_cache[prefix] = ({}, [])
2645
if type(self._mapper) is ConstantMapper:
2646
# preserve behaviour for revisions.kndx etc.
2647
self._init_index(path)
2652
missing_keys = _mod_index._missing_keys_from_parent_map
2654
def _partition_keys(self, keys):
2655
"""Turn keys into a dict of prefix:suffix_list."""
2658
prefix_keys = result.setdefault(key[:-1], [])
2659
prefix_keys.append(key[-1])
2662
def _dictionary_compress(self, keys):
2663
"""Dictionary compress keys.
2665
:param keys: The keys to generate references to.
2666
:return: A string representation of keys. keys which are present are
2667
dictionary compressed, and others are emitted as fulltext with a
2673
prefix = keys[0][:-1]
2674
cache = self._kndx_cache[prefix][0]
2676
if key[:-1] != prefix:
2677
# kndx indices cannot refer across partitioned storage.
2678
raise ValueError("mismatched prefixes for %r" % keys)
2679
if key[-1] in cache:
2680
# -- inlined lookup() --
2681
result_list.append(str(cache[key[-1]][5]))
2682
# -- end lookup () --
2684
result_list.append('.' + key[-1])
2685
return ' '.join(result_list)
2687
def _reset_cache(self):
2688
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2689
# (cache_dict, history_vector) for parsed kndx files.
2690
self._kndx_cache = {}
2691
self._scope = self._get_scope()
2692
allow_writes = self._allow_writes()
2698
def _sort_keys_by_io(self, keys, positions):
2699
"""Figure out an optimal order to read the records for the given keys.
2701
Sort keys, grouped by index and sorted by position.
2703
:param keys: A list of keys whose records we want to read. This will be
2705
:param positions: A dict, such as the one returned by
2706
_get_components_positions()
2709
def get_sort_key(key):
2710
index_memo = positions[key][1]
2711
# Group by prefix and position. index_memo[0] is the key, so it is
2712
# (file_id, revision_id) and we don't want to sort on revision_id,
2713
# index_memo[1] is the position, and index_memo[2] is the size,
2714
# which doesn't matter for the sort
2715
return index_memo[0][:-1], index_memo[1]
2716
return keys.sort(key=get_sort_key)
2718
_get_total_build_size = _get_total_build_size
2720
def _split_key(self, key):
2721
"""Split key into a prefix and suffix."""
2722
return key[:-1], key[-1]
2725
class _KeyRefs(object):
2728
# dict mapping 'key' to 'set of keys referring to that key'
2731
def add_references(self, key, refs):
2732
# Record the new references
2733
for referenced in refs:
2735
needed_by = self.refs[referenced]
2737
needed_by = self.refs[referenced] = set()
2739
# Discard references satisfied by the new key
2742
def get_unsatisfied_refs(self):
2743
return self.refs.iterkeys()
2745
def add_key(self, key):
2749
# No keys depended on this key. That's ok.
2752
def add_keys(self, keys):
2756
def get_referrers(self):
2758
for referrers in self.refs.itervalues():
2759
result.update(referrers)
2763
class _KnitGraphIndex(object):
2764
"""A KnitVersionedFiles index layered on GraphIndex."""
2766
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2767
add_callback=None, track_external_parent_refs=False):
2768
"""Construct a KnitGraphIndex on a graph_index.
2770
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2771
:param is_locked: A callback to check whether the object should answer
2773
:param deltas: Allow delta-compressed records.
2774
:param parents: If True, record knits parents, if not do not record
2776
:param add_callback: If not None, allow additions to the index and call
2777
this callback with a list of added GraphIndex nodes:
2778
[(node, value, node_refs), ...]
2779
:param is_locked: A callback, returns True if the index is locked and
2781
:param track_external_parent_refs: If True, record all external parent
2782
references parents from added records. These can be retrieved
2783
later by calling get_missing_parents().
2785
self._add_callback = add_callback
2786
self._graph_index = graph_index
2787
self._deltas = deltas
2788
self._parents = parents
2789
if deltas and not parents:
2790
# XXX: TODO: Delta tree and parent graph should be conceptually
2792
raise KnitCorrupt(self, "Cannot do delta compression without "
2794
self.has_graph = parents
2795
self._is_locked = is_locked
2796
self._missing_compression_parents = set()
2797
if track_external_parent_refs:
2798
self._key_dependencies = _KeyRefs()
2800
self._key_dependencies = None
2803
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2805
def add_records(self, records, random_id=False,
2806
missing_compression_parents=False):
2807
"""Add multiple records to the index.
2809
This function does not insert data into the Immutable GraphIndex
2810
backing the KnitGraphIndex, instead it prepares data for insertion by
2811
the caller and checks that it is safe to insert then calls
2812
self._add_callback with the prepared GraphIndex nodes.
2814
:param records: a list of tuples:
2815
(key, options, access_memo, parents).
2816
:param random_id: If True the ids being added were randomly generated
2817
and no check for existence will be performed.
2818
:param missing_compression_parents: If True the records being added are
2819
only compressed against texts already in the index (or inside
2820
records). If False the records all refer to unavailable texts (or
2821
texts inside records) as compression parents.
2823
if not self._add_callback:
2824
raise errors.ReadOnlyError(self)
2825
# we hope there are no repositories with inconsistent parentage
2829
compression_parents = set()
2830
key_dependencies = self._key_dependencies
2831
for (key, options, access_memo, parents) in records:
2833
parents = tuple(parents)
2834
if key_dependencies is not None:
2835
key_dependencies.add_references(key, parents)
2836
index, pos, size = access_memo
2837
if 'no-eol' in options:
2841
value += "%d %d" % (pos, size)
2842
if not self._deltas:
2843
if 'line-delta' in options:
2844
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2847
if 'line-delta' in options:
2848
node_refs = (parents, (parents[0],))
2849
if missing_compression_parents:
2850
compression_parents.add(parents[0])
2852
node_refs = (parents, ())
2854
node_refs = (parents, )
2857
raise KnitCorrupt(self, "attempt to add node with parents "
2858
"in parentless index.")
2860
keys[key] = (value, node_refs)
2863
present_nodes = self._get_entries(keys)
2864
for (index, key, value, node_refs) in present_nodes:
2865
if (value[0] != keys[key][0][0] or
2866
node_refs[:1] != keys[key][1][:1]):
2867
raise KnitCorrupt(self, "inconsistent details in add_records"
2868
": %s %s" % ((value, node_refs), keys[key]))
2872
for key, (value, node_refs) in keys.iteritems():
2873
result.append((key, value, node_refs))
2875
for key, (value, node_refs) in keys.iteritems():
2876
result.append((key, value))
2877
self._add_callback(result)
2878
if missing_compression_parents:
2879
# This may appear to be incorrect (it does not check for
2880
# compression parents that are in the existing graph index),
2881
# but such records won't have been buffered, so this is
2882
# actually correct: every entry when
2883
# missing_compression_parents==True either has a missing parent, or
2884
# a parent that is one of the keys in records.
2885
compression_parents.difference_update(keys)
2886
self._missing_compression_parents.update(compression_parents)
2887
# Adding records may have satisfied missing compression parents.
2888
self._missing_compression_parents.difference_update(keys)
2890
def scan_unvalidated_index(self, graph_index):
2891
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2893
This allows this _KnitGraphIndex to keep track of any missing
2894
compression parents we may want to have filled in to make those
2897
:param graph_index: A GraphIndex
2900
new_missing = graph_index.external_references(ref_list_num=1)
2901
new_missing.difference_update(self.get_parent_map(new_missing))
2902
self._missing_compression_parents.update(new_missing)
2903
if self._key_dependencies is not None:
2904
# Add parent refs from graph_index (and discard parent refs that
2905
# the graph_index has).
2906
for node in graph_index.iter_all_entries():
2907
self._key_dependencies.add_references(node[1], node[3][0])
2909
def get_missing_compression_parents(self):
2910
"""Return the keys of missing compression parents.
2912
Missing compression parents occur when a record stream was missing
2913
basis texts, or a index was scanned that had missing basis texts.
2915
return frozenset(self._missing_compression_parents)
2917
def get_missing_parents(self):
2918
"""Return the keys of missing parents."""
2919
# If updating this, you should also update
2920
# groupcompress._GCGraphIndex.get_missing_parents
2921
# We may have false positives, so filter those out.
2922
self._key_dependencies.add_keys(
2923
self.get_parent_map(self._key_dependencies.get_unsatisfied_refs()))
2924
return frozenset(self._key_dependencies.get_unsatisfied_refs())
2926
def _check_read(self):
2927
"""raise if reads are not permitted."""
2928
if not self._is_locked():
2929
raise errors.ObjectNotLocked(self)
2931
def _check_write_ok(self):
2932
"""Assert if writes are not permitted."""
2933
if not self._is_locked():
2934
raise errors.ObjectNotLocked(self)
2936
def _compression_parent(self, an_entry):
2937
# return the key that an_entry is compressed against, or None
2938
# Grab the second parent list (as deltas implies parents currently)
2939
compression_parents = an_entry[3][1]
2940
if not compression_parents:
2942
if len(compression_parents) != 1:
2943
raise AssertionError(
2944
"Too many compression parents: %r" % compression_parents)
2945
return compression_parents[0]
2947
def get_build_details(self, keys):
2948
"""Get the method, index_memo and compression parent for version_ids.
2950
Ghosts are omitted from the result.
2952
:param keys: An iterable of keys.
2953
:return: A dict of key:
2954
(index_memo, compression_parent, parents, record_details).
2956
opaque structure to pass to read_records to extract the raw
2959
Content that this record is built upon, may be None
2961
Logical parents of this node
2963
extra information about the content which needs to be passed to
2964
Factory.parse_record
2968
entries = self._get_entries(keys, False)
2969
for entry in entries:
2971
if not self._parents:
2974
parents = entry[3][0]
2975
if not self._deltas:
2976
compression_parent_key = None
2978
compression_parent_key = self._compression_parent(entry)
2979
noeol = (entry[2][0] == 'N')
2980
if compression_parent_key:
2981
method = 'line-delta'
2984
result[key] = (self._node_to_position(entry),
2985
compression_parent_key, parents,
2989
def _get_entries(self, keys, check_present=False):
2990
"""Get the entries for keys.
2992
:param keys: An iterable of index key tuples.
2997
for node in self._graph_index.iter_entries(keys):
2999
found_keys.add(node[1])
3001
# adapt parentless index to the rest of the code.
3002
for node in self._graph_index.iter_entries(keys):
3003
yield node[0], node[1], node[2], ()
3004
found_keys.add(node[1])
3006
missing_keys = keys.difference(found_keys)
3008
raise RevisionNotPresent(missing_keys.pop(), self)
3010
def get_method(self, key):
3011
"""Return compression method of specified key."""
3012
return self._get_method(self._get_node(key))
3014
def _get_method(self, node):
3015
if not self._deltas:
3017
if self._compression_parent(node):
3022
def _get_node(self, key):
3024
return list(self._get_entries([key]))[0]
3026
raise RevisionNotPresent(key, self)
3028
def get_options(self, key):
3029
"""Return a list representing options.
3033
node = self._get_node(key)
3034
options = [self._get_method(node)]
3035
if node[2][0] == 'N':
3036
options.append('no-eol')
3039
def get_parent_map(self, keys):
3040
"""Get a map of the parents of keys.
3042
:param keys: The keys to look up parents for.
3043
:return: A mapping from keys to parents. Absent keys are absent from
3047
nodes = self._get_entries(keys)
3051
result[node[1]] = node[3][0]
3054
result[node[1]] = None
3057
def get_position(self, key):
3058
"""Return details needed to access the version.
3060
:return: a tuple (index, data position, size) to hand to the access
3061
logic to get the record.
3063
node = self._get_node(key)
3064
return self._node_to_position(node)
3066
has_key = _mod_index._has_key_from_parent_map
3069
"""Get all the keys in the collection.
3071
The keys are not ordered.
3074
return [node[1] for node in self._graph_index.iter_all_entries()]
3076
missing_keys = _mod_index._missing_keys_from_parent_map
3078
def _node_to_position(self, node):
3079
"""Convert an index value to position details."""
3080
bits = node[2][1:].split(' ')
3081
return node[0], int(bits[0]), int(bits[1])
3083
def _sort_keys_by_io(self, keys, positions):
3084
"""Figure out an optimal order to read the records for the given keys.
3086
Sort keys, grouped by index and sorted by position.
3088
:param keys: A list of keys whose records we want to read. This will be
3090
:param positions: A dict, such as the one returned by
3091
_get_components_positions()
3094
def get_index_memo(key):
3095
# index_memo is at offset [1]. It is made up of (GraphIndex,
3096
# position, size). GI is an object, which will be unique for each
3097
# pack file. This causes us to group by pack file, then sort by
3098
# position. Size doesn't matter, but it isn't worth breaking up the
3100
return positions[key][1]
3101
return keys.sort(key=get_index_memo)
3103
_get_total_build_size = _get_total_build_size
3106
class _KnitKeyAccess(object):
3107
"""Access to records in .knit files."""
3109
def __init__(self, transport, mapper):
3110
"""Create a _KnitKeyAccess with transport and mapper.
3112
:param transport: The transport the access object is rooted at.
3113
:param mapper: The mapper used to map keys to .knit files.
3115
self._transport = transport
3116
self._mapper = mapper
3118
def add_raw_records(self, key_sizes, raw_data):
3119
"""Add raw knit bytes to a storage area.
3121
The data is spooled to the container writer in one bytes-record per
3124
:param sizes: An iterable of tuples containing the key and size of each
3126
:param raw_data: A bytestring containing the data.
3127
:return: A list of memos to retrieve the record later. Each memo is an
3128
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3129
length), where the key is the record key.
3131
if type(raw_data) is not str:
3132
raise AssertionError(
3133
'data must be plain bytes was %s' % type(raw_data))
3136
# TODO: This can be tuned for writing to sftp and other servers where
3137
# append() is relatively expensive by grouping the writes to each key
3139
for key, size in key_sizes:
3140
path = self._mapper.map(key)
3142
base = self._transport.append_bytes(path + '.knit',
3143
raw_data[offset:offset+size])
3144
except errors.NoSuchFile:
3145
self._transport.mkdir(osutils.dirname(path))
3146
base = self._transport.append_bytes(path + '.knit',
3147
raw_data[offset:offset+size])
3151
result.append((key, base, size))
3155
"""Flush pending writes on this access object.
3157
For .knit files this is a no-op.
3161
def get_raw_records(self, memos_for_retrieval):
3162
"""Get the raw bytes for a records.
3164
:param memos_for_retrieval: An iterable containing the access memo for
3165
retrieving the bytes.
3166
:return: An iterator over the bytes of the records.
3168
# first pass, group into same-index request to minimise readv's issued.
3170
current_prefix = None
3171
for (key, offset, length) in memos_for_retrieval:
3172
if current_prefix == key[:-1]:
3173
current_list.append((offset, length))
3175
if current_prefix is not None:
3176
request_lists.append((current_prefix, current_list))
3177
current_prefix = key[:-1]
3178
current_list = [(offset, length)]
3179
# handle the last entry
3180
if current_prefix is not None:
3181
request_lists.append((current_prefix, current_list))
3182
for prefix, read_vector in request_lists:
3183
path = self._mapper.map(prefix) + '.knit'
3184
for pos, data in self._transport.readv(path, read_vector):
3188
class _DirectPackAccess(object):
3189
"""Access to data in one or more packs with less translation."""
3191
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
3192
"""Create a _DirectPackAccess object.
3194
:param index_to_packs: A dict mapping index objects to the transport
3195
and file names for obtaining data.
3196
:param reload_func: A function to call if we determine that the pack
3197
files have moved and we need to reload our caches. See
3198
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
3200
self._container_writer = None
3201
self._write_index = None
3202
self._indices = index_to_packs
3203
self._reload_func = reload_func
3204
self._flush_func = flush_func
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 _DirectPackAccess the memo is (index, pos,
3217
length), where the index field is the write_index object supplied
3218
to the PackAccess object.
3220
if type(raw_data) is not str:
3221
raise AssertionError(
3222
'data must be plain bytes was %s' % type(raw_data))
3225
for key, size in key_sizes:
3226
p_offset, p_length = self._container_writer.add_bytes_record(
3227
raw_data[offset:offset+size], [])
3229
result.append((self._write_index, p_offset, p_length))
3233
"""Flush pending writes on this access object.
3235
This will flush any buffered writes to a NewPack.
3237
if self._flush_func is not None:
3240
def get_raw_records(self, memos_for_retrieval):
3241
"""Get the raw bytes for a records.
3243
:param memos_for_retrieval: An iterable containing the (index, pos,
3244
length) memo for retrieving the bytes. The Pack access method
3245
looks up the pack to use for a given record in its index_to_pack
3247
:return: An iterator over the bytes of the records.
3249
# first pass, group into same-index requests
3251
current_index = None
3252
for (index, offset, length) in memos_for_retrieval:
3253
if current_index == index:
3254
current_list.append((offset, length))
3256
if current_index is not None:
3257
request_lists.append((current_index, current_list))
3258
current_index = index
3259
current_list = [(offset, length)]
3260
# handle the last entry
3261
if current_index is not None:
3262
request_lists.append((current_index, current_list))
3263
for index, offsets in request_lists:
3265
transport, path = self._indices[index]
3267
# A KeyError here indicates that someone has triggered an index
3268
# reload, and this index has gone missing, we need to start
3270
if self._reload_func is None:
3271
# If we don't have a _reload_func there is nothing that can
3274
raise errors.RetryWithNewPacks(index,
3275
reload_occurred=True,
3276
exc_info=sys.exc_info())
3278
reader = pack.make_readv_reader(transport, path, offsets)
3279
for names, read_func in reader.iter_records():
3280
yield read_func(None)
3281
except errors.NoSuchFile:
3282
# A NoSuchFile error indicates that a pack file has gone
3283
# missing on disk, we need to trigger a reload, and start over.
3284
if self._reload_func is None:
3286
raise errors.RetryWithNewPacks(transport.abspath(path),
3287
reload_occurred=False,
3288
exc_info=sys.exc_info())
3290
def set_writer(self, writer, index, transport_packname):
3291
"""Set a writer to use for adding data."""
3292
if index is not None:
3293
self._indices[index] = transport_packname
3294
self._container_writer = writer
3295
self._write_index = index
3297
def reload_or_raise(self, retry_exc):
3298
"""Try calling the reload function, or re-raise the original exception.
3300
This should be called after _DirectPackAccess raises a
3301
RetryWithNewPacks exception. This function will handle the common logic
3302
of determining when the error is fatal versus being temporary.
3303
It will also make sure that the original exception is raised, rather
3304
than the RetryWithNewPacks exception.
3306
If this function returns, then the calling function should retry
3307
whatever operation was being performed. Otherwise an exception will
3310
:param retry_exc: A RetryWithNewPacks exception.
3313
if self._reload_func is None:
3315
elif not self._reload_func():
3316
# The reload claimed that nothing changed
3317
if not retry_exc.reload_occurred:
3318
# If there wasn't an earlier reload, then we really were
3319
# expecting to find changes. We didn't find them, so this is a
3323
exc_class, exc_value, exc_traceback = retry_exc.exc_info
3324
raise exc_class, exc_value, exc_traceback
3327
# Deprecated, use PatienceSequenceMatcher instead
3328
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
3331
def annotate_knit(knit, revision_id):
3332
"""Annotate a knit with no cached annotations.
3334
This implementation is for knits with no cached annotations.
3335
It will work for knits with cached annotations, but this is not
3338
annotator = _KnitAnnotator(knit)
3339
return iter(annotator.annotate(revision_id))
3342
class _KnitAnnotator(object):
3343
"""Build up the annotations for a text."""
3345
def __init__(self, knit):
3348
# Content objects, differs from fulltexts because of how final newlines
3349
# are treated by knits. the content objects here will always have a
3351
self._fulltext_contents = {}
3353
# Annotated lines of specific revisions
3354
self._annotated_lines = {}
3356
# Track the raw data for nodes that we could not process yet.
3357
# This maps the revision_id of the base to a list of children that will
3358
# annotated from it.
3359
self._pending_children = {}
3361
# Nodes which cannot be extracted
3362
self._ghosts = set()
3364
# Track how many children this node has, so we know if we need to keep
3366
self._annotate_children = {}
3367
self._compression_children = {}
3369
self._all_build_details = {}
3370
# The children => parent revision_id graph
3371
self._revision_id_graph = {}
3373
self._heads_provider = None
3375
self._nodes_to_keep_annotations = set()
3376
self._generations_until_keep = 100
3378
def set_generations_until_keep(self, value):
3379
"""Set the number of generations before caching a node.
3381
Setting this to -1 will cache every merge node, setting this higher
3382
will cache fewer nodes.
3384
self._generations_until_keep = value
3386
def _add_fulltext_content(self, revision_id, content_obj):
3387
self._fulltext_contents[revision_id] = content_obj
3388
# TODO: jam 20080305 It might be good to check the sha1digest here
3389
return content_obj.text()
3391
def _check_parents(self, child, nodes_to_annotate):
3392
"""Check if all parents have been processed.
3394
:param child: A tuple of (rev_id, parents, raw_content)
3395
:param nodes_to_annotate: If child is ready, add it to
3396
nodes_to_annotate, otherwise put it back in self._pending_children
3398
for parent_id in child[1]:
3399
if (parent_id not in self._annotated_lines):
3400
# This parent is present, but another parent is missing
3401
self._pending_children.setdefault(parent_id,
3405
# This one is ready to be processed
3406
nodes_to_annotate.append(child)
3408
def _add_annotation(self, revision_id, fulltext, parent_ids,
3409
left_matching_blocks=None):
3410
"""Add an annotation entry.
3412
All parents should already have been annotated.
3413
:return: A list of children that now have their parents satisfied.
3415
a = self._annotated_lines
3416
annotated_parent_lines = [a[p] for p in parent_ids]
3417
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
3418
fulltext, revision_id, left_matching_blocks,
3419
heads_provider=self._get_heads_provider()))
3420
self._annotated_lines[revision_id] = annotated_lines
3421
for p in parent_ids:
3422
ann_children = self._annotate_children[p]
3423
ann_children.remove(revision_id)
3424
if (not ann_children
3425
and p not in self._nodes_to_keep_annotations):
3426
del self._annotated_lines[p]
3427
del self._all_build_details[p]
3428
if p in self._fulltext_contents:
3429
del self._fulltext_contents[p]
3430
# Now that we've added this one, see if there are any pending
3431
# deltas to be done, certainly this parent is finished
3432
nodes_to_annotate = []
3433
for child in self._pending_children.pop(revision_id, []):
3434
self._check_parents(child, nodes_to_annotate)
3435
return nodes_to_annotate
3437
def _get_build_graph(self, key):
3438
"""Get the graphs for building texts and annotations.
3440
The data you need for creating a full text may be different than the
3441
data you need to annotate that text. (At a minimum, you need both
3442
parents to create an annotation, but only need 1 parent to generate the
3445
:return: A list of (key, index_memo) records, suitable for
3446
passing to read_records_iter to start reading in the raw data from
3449
if key in self._annotated_lines:
3452
pending = set([key])
3457
# get all pending nodes
3459
this_iteration = pending
3460
build_details = self._knit._index.get_build_details(this_iteration)
3461
self._all_build_details.update(build_details)
3462
# new_nodes = self._knit._index._get_entries(this_iteration)
3464
for key, details in build_details.iteritems():
3465
(index_memo, compression_parent, parents,
3466
record_details) = details
3467
self._revision_id_graph[key] = parents
3468
records.append((key, index_memo))
3469
# Do we actually need to check _annotated_lines?
3470
pending.update(p for p in parents
3471
if p not in self._all_build_details)
3472
if compression_parent:
3473
self._compression_children.setdefault(compression_parent,
3476
for parent in parents:
3477
self._annotate_children.setdefault(parent,
3479
num_gens = generation - kept_generation
3480
if ((num_gens >= self._generations_until_keep)
3481
and len(parents) > 1):
3482
kept_generation = generation
3483
self._nodes_to_keep_annotations.add(key)
3485
missing_versions = this_iteration.difference(build_details.keys())
3486
self._ghosts.update(missing_versions)
3487
for missing_version in missing_versions:
3488
# add a key, no parents
3489
self._revision_id_graph[missing_version] = ()
3490
pending.discard(missing_version) # don't look for it
3491
if self._ghosts.intersection(self._compression_children):
3493
"We cannot have nodes which have a ghost compression parent:\n"
3495
"compression children: %r"
3496
% (self._ghosts, self._compression_children))
3497
# Cleanout anything that depends on a ghost so that we don't wait for
3498
# the ghost to show up
3499
for node in self._ghosts:
3500
if node in self._annotate_children:
3501
# We won't be building this node
3502
del self._annotate_children[node]
3503
# Generally we will want to read the records in reverse order, because
3504
# we find the parent nodes after the children
3508
def _annotate_records(self, records):
3509
"""Build the annotations for the listed records."""
3510
# We iterate in the order read, rather than a strict order requested
3511
# However, process what we can, and put off to the side things that
3512
# still need parents, cleaning them up when those parents are
3514
for (rev_id, record,
3515
digest) in self._knit._read_records_iter(records):
3516
if rev_id in self._annotated_lines:
3518
parent_ids = self._revision_id_graph[rev_id]
3519
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3520
details = self._all_build_details[rev_id]
3521
(index_memo, compression_parent, parents,
3522
record_details) = details
3523
nodes_to_annotate = []
3524
# TODO: Remove the punning between compression parents, and
3525
# parent_ids, we should be able to do this without assuming
3527
if len(parent_ids) == 0:
3528
# There are no parents for this node, so just add it
3529
# TODO: This probably needs to be decoupled
3530
fulltext_content, delta = self._knit._factory.parse_record(
3531
rev_id, record, record_details, None)
3532
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3533
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3534
parent_ids, left_matching_blocks=None))
3536
child = (rev_id, parent_ids, record)
3537
# Check if all the parents are present
3538
self._check_parents(child, nodes_to_annotate)
3539
while nodes_to_annotate:
3540
# Should we use a queue here instead of a stack?
3541
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3542
(index_memo, compression_parent, parents,
3543
record_details) = self._all_build_details[rev_id]
3545
if compression_parent is not None:
3546
comp_children = self._compression_children[compression_parent]
3547
if rev_id not in comp_children:
3548
raise AssertionError("%r not in compression children %r"
3549
% (rev_id, comp_children))
3550
# If there is only 1 child, it is safe to reuse this
3552
reuse_content = (len(comp_children) == 1
3553
and compression_parent not in
3554
self._nodes_to_keep_annotations)
3556
# Remove it from the cache since it will be changing
3557
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3558
# Make sure to copy the fulltext since it might be
3560
parent_fulltext = list(parent_fulltext_content.text())
3562
parent_fulltext_content = self._fulltext_contents[compression_parent]
3563
parent_fulltext = parent_fulltext_content.text()
3564
comp_children.remove(rev_id)
3565
fulltext_content, delta = self._knit._factory.parse_record(
3566
rev_id, record, record_details,
3567
parent_fulltext_content,
3568
copy_base_content=(not reuse_content))
3569
fulltext = self._add_fulltext_content(rev_id,
3571
if compression_parent == parent_ids[0]:
3572
# the compression_parent is the left parent, so we can
3574
blocks = KnitContent.get_line_delta_blocks(delta,
3575
parent_fulltext, fulltext)
3577
fulltext_content = self._knit._factory.parse_fulltext(
3579
fulltext = self._add_fulltext_content(rev_id,
3581
nodes_to_annotate.extend(
3582
self._add_annotation(rev_id, fulltext, parent_ids,
3583
left_matching_blocks=blocks))
3585
def _get_heads_provider(self):
3586
"""Create a heads provider for resolving ancestry issues."""
3587
if self._heads_provider is not None:
3588
return self._heads_provider
3589
self._heads_provider = _mod_graph.KnownGraph(self._revision_id_graph)
3590
return self._heads_provider
3592
def annotate(self, key):
3593
"""Return the annotated fulltext at the given key.
3595
:param key: The key to annotate.
3597
if len(self._knit._fallback_vfs) > 0:
3598
# stacked knits can't use the fast path at present.
3599
return self._simple_annotate(key)
3602
records = self._get_build_graph(key)
3603
if key in self._ghosts:
3604
raise errors.RevisionNotPresent(key, self._knit)
3605
self._annotate_records(records)
3606
return self._annotated_lines[key]
3607
except errors.RetryWithNewPacks, e:
3608
self._knit._access.reload_or_raise(e)
3609
# The cached build_details are no longer valid
3610
self._all_build_details.clear()
3612
def _simple_annotate(self, key):
3613
"""Return annotated fulltext, rediffing from the full texts.
3615
This is slow but makes no assumptions about the repository
3616
being able to produce line deltas.
3618
# TODO: this code generates a parent maps of present ancestors; it
3619
# could be split out into a separate method
3620
# -- mbp and robertc 20080704
3621
graph = _mod_graph.Graph(self._knit)
3622
parent_map = dict((k, v) for k, v in graph.iter_ancestry([key])
3625
raise errors.RevisionNotPresent(key, self)
3626
keys = parent_map.keys()
3627
heads_provider = _mod_graph.KnownGraph(parent_map)
3629
reannotate = annotate.reannotate
3630
for record in self._knit.get_record_stream(keys, 'topological', True):
3632
fulltext = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
3633
parents = parent_map[key]
3634
if parents is not None:
3635
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
3638
parent_cache[key] = list(
3639
reannotate(parent_lines, fulltext, key, None, heads_provider))
3641
return parent_cache[key]
3643
raise errors.RevisionNotPresent(key, self._knit)
3647
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3649
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
