1
# Copyright (C) 2006-2011 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
17
"""Knit versionedfile implementation.
19
A knit is a versioned file implementation that supports efficient append only
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
end-marker; simply "end VERSION"
28
delta can be line or full contents.a
29
... the 8's there are the index number of the annotation.
30
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
8 e.set('executable', 'yes')
36
8 if elt.get('executable') == 'yes':
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
43
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
44
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
45
09:33 < lifeless> right
46
09:33 < jrydberg> lifeless: the position and size is the range in the data file
49
so the index sequence is the dictionary compressed sequence number used
50
in the deltas to provide line annotation
55
from cStringIO import StringIO
56
from itertools import izip
60
from bzrlib.lazy_import import lazy_import
61
lazy_import(globals(), """
78
from bzrlib.repofmt import pack_repo
79
from bzrlib.i18n import gettext
86
from bzrlib.errors import (
94
from bzrlib.osutils import (
100
from bzrlib.versionedfile import (
102
AbsentContentFactory,
107
VersionedFilesWithFallbacks,
111
# TODO: Split out code specific to this format into an associated object.
113
# TODO: Can we put in some kind of value to check that the index and data
114
# files belong together?
116
# TODO: accommodate binaries, perhaps by storing a byte count
118
# TODO: function to check whole file
120
# TODO: atomically append data, then measure backwards from the cursor
121
# position after writing to work out where it was located. we may need to
122
# bypass python file buffering.
124
DATA_SUFFIX = '.knit'
125
INDEX_SUFFIX = '.kndx'
126
_STREAM_MIN_BUFFER_SIZE = 5*1024*1024
129
class KnitAdapter(object):
130
"""Base class for knit record adaption."""
132
def __init__(self, basis_vf):
133
"""Create an adapter which accesses full texts from basis_vf.
135
:param basis_vf: A versioned file to access basis texts of deltas from.
136
May be None for adapters that do not need to access basis texts.
138
self._data = KnitVersionedFiles(None, None)
139
self._annotate_factory = KnitAnnotateFactory()
140
self._plain_factory = KnitPlainFactory()
141
self._basis_vf = basis_vf
144
class FTAnnotatedToUnannotated(KnitAdapter):
145
"""An adapter from FT annotated knits to unannotated ones."""
147
def get_bytes(self, factory):
148
annotated_compressed_bytes = factory._raw_record
150
self._data._parse_record_unchecked(annotated_compressed_bytes)
151
content = self._annotate_factory.parse_fulltext(contents, rec[1])
152
size, bytes = self._data._record_to_data((rec[1],), rec[3], content.text())
156
class DeltaAnnotatedToUnannotated(KnitAdapter):
157
"""An adapter for deltas from annotated to unannotated."""
159
def get_bytes(self, factory):
160
annotated_compressed_bytes = factory._raw_record
162
self._data._parse_record_unchecked(annotated_compressed_bytes)
163
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
165
contents = self._plain_factory.lower_line_delta(delta)
166
size, bytes = self._data._record_to_data((rec[1],), rec[3], contents)
170
class FTAnnotatedToFullText(KnitAdapter):
171
"""An adapter from FT annotated knits to unannotated ones."""
173
def get_bytes(self, factory):
174
annotated_compressed_bytes = factory._raw_record
176
self._data._parse_record_unchecked(annotated_compressed_bytes)
177
content, delta = self._annotate_factory.parse_record(factory.key[-1],
178
contents, factory._build_details, None)
179
return ''.join(content.text())
182
class DeltaAnnotatedToFullText(KnitAdapter):
183
"""An adapter for deltas from annotated to unannotated."""
185
def get_bytes(self, factory):
186
annotated_compressed_bytes = factory._raw_record
188
self._data._parse_record_unchecked(annotated_compressed_bytes)
189
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
191
compression_parent = factory.parents[0]
192
basis_entry = self._basis_vf.get_record_stream(
193
[compression_parent], 'unordered', True).next()
194
if basis_entry.storage_kind == 'absent':
195
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
196
basis_chunks = basis_entry.get_bytes_as('chunked')
197
basis_lines = osutils.chunks_to_lines(basis_chunks)
198
# Manually apply the delta because we have one annotated content and
200
basis_content = PlainKnitContent(basis_lines, compression_parent)
201
basis_content.apply_delta(delta, rec[1])
202
basis_content._should_strip_eol = factory._build_details[1]
203
return ''.join(basis_content.text())
206
class FTPlainToFullText(KnitAdapter):
207
"""An adapter from FT plain knits to unannotated ones."""
209
def get_bytes(self, factory):
210
compressed_bytes = factory._raw_record
212
self._data._parse_record_unchecked(compressed_bytes)
213
content, delta = self._plain_factory.parse_record(factory.key[-1],
214
contents, factory._build_details, None)
215
return ''.join(content.text())
218
class DeltaPlainToFullText(KnitAdapter):
219
"""An adapter for deltas from annotated to unannotated."""
221
def get_bytes(self, factory):
222
compressed_bytes = factory._raw_record
224
self._data._parse_record_unchecked(compressed_bytes)
225
delta = self._plain_factory.parse_line_delta(contents, rec[1])
226
compression_parent = factory.parents[0]
227
# XXX: string splitting overhead.
228
basis_entry = self._basis_vf.get_record_stream(
229
[compression_parent], 'unordered', True).next()
230
if basis_entry.storage_kind == 'absent':
231
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
232
basis_chunks = basis_entry.get_bytes_as('chunked')
233
basis_lines = osutils.chunks_to_lines(basis_chunks)
234
basis_content = PlainKnitContent(basis_lines, compression_parent)
235
# Manually apply the delta because we have one annotated content and
237
content, _ = self._plain_factory.parse_record(rec[1], contents,
238
factory._build_details, basis_content)
239
return ''.join(content.text())
242
class KnitContentFactory(ContentFactory):
243
"""Content factory for streaming from knits.
245
:seealso ContentFactory:
248
def __init__(self, key, parents, build_details, sha1, raw_record,
249
annotated, knit=None, network_bytes=None):
250
"""Create a KnitContentFactory for key.
253
:param parents: The parents.
254
:param build_details: The build details as returned from
256
:param sha1: The sha1 expected from the full text of this object.
257
:param raw_record: The bytes of the knit data from disk.
258
:param annotated: True if the raw data is annotated.
259
:param network_bytes: None to calculate the network bytes on demand,
260
not-none if they are already known.
262
ContentFactory.__init__(self)
265
self.parents = parents
266
if build_details[0] == 'line-delta':
271
annotated_kind = 'annotated-'
274
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
275
self._raw_record = raw_record
276
self._network_bytes = network_bytes
277
self._build_details = build_details
280
def _create_network_bytes(self):
281
"""Create a fully serialised network version for transmission."""
282
# storage_kind, key, parents, Noeol, raw_record
283
key_bytes = '\x00'.join(self.key)
284
if self.parents is None:
285
parent_bytes = 'None:'
287
parent_bytes = '\t'.join('\x00'.join(key) for key in self.parents)
288
if self._build_details[1]:
292
network_bytes = "%s\n%s\n%s\n%s%s" % (self.storage_kind, key_bytes,
293
parent_bytes, noeol, self._raw_record)
294
self._network_bytes = network_bytes
296
def get_bytes_as(self, storage_kind):
297
if storage_kind == self.storage_kind:
298
if self._network_bytes is None:
299
self._create_network_bytes()
300
return self._network_bytes
301
if ('-ft-' in self.storage_kind and
302
storage_kind in ('chunked', 'fulltext')):
303
adapter_key = (self.storage_kind, 'fulltext')
304
adapter_factory = adapter_registry.get(adapter_key)
305
adapter = adapter_factory(None)
306
bytes = adapter.get_bytes(self)
307
if storage_kind == 'chunked':
311
if self._knit is not None:
312
# Not redundant with direct conversion above - that only handles
314
if storage_kind == 'chunked':
315
return self._knit.get_lines(self.key[0])
316
elif storage_kind == 'fulltext':
317
return self._knit.get_text(self.key[0])
318
raise errors.UnavailableRepresentation(self.key, storage_kind,
322
class LazyKnitContentFactory(ContentFactory):
323
"""A ContentFactory which can either generate full text or a wire form.
325
:seealso ContentFactory:
328
def __init__(self, key, parents, generator, first):
329
"""Create a LazyKnitContentFactory.
331
:param key: The key of the record.
332
:param parents: The parents of the record.
333
:param generator: A _ContentMapGenerator containing the record for this
335
:param first: Is this the first content object returned from generator?
336
if it is, its storage kind is knit-delta-closure, otherwise it is
337
knit-delta-closure-ref
340
self.parents = parents
342
self._generator = generator
343
self.storage_kind = "knit-delta-closure"
345
self.storage_kind = self.storage_kind + "-ref"
348
def get_bytes_as(self, storage_kind):
349
if storage_kind == self.storage_kind:
351
return self._generator._wire_bytes()
353
# all the keys etc are contained in the bytes returned in the
356
if storage_kind in ('chunked', 'fulltext'):
357
chunks = self._generator._get_one_work(self.key).text()
358
if storage_kind == 'chunked':
361
return ''.join(chunks)
362
raise errors.UnavailableRepresentation(self.key, storage_kind,
366
def knit_delta_closure_to_records(storage_kind, bytes, line_end):
367
"""Convert a network record to a iterator over stream records.
369
:param storage_kind: The storage kind of the record.
370
Must be 'knit-delta-closure'.
371
:param bytes: The bytes of the record on the network.
373
generator = _NetworkContentMapGenerator(bytes, line_end)
374
return generator.get_record_stream()
377
def knit_network_to_record(storage_kind, bytes, line_end):
378
"""Convert a network record to a record object.
380
:param storage_kind: The storage kind of the record.
381
:param bytes: The bytes of the record on the network.
384
line_end = bytes.find('\n', start)
385
key = tuple(bytes[start:line_end].split('\x00'))
387
line_end = bytes.find('\n', start)
388
parent_line = bytes[start:line_end]
389
if parent_line == 'None:':
393
[tuple(segment.split('\x00')) for segment in parent_line.split('\t')
396
noeol = bytes[start] == 'N'
397
if 'ft' in storage_kind:
400
method = 'line-delta'
401
build_details = (method, noeol)
403
raw_record = bytes[start:]
404
annotated = 'annotated' in storage_kind
405
return [KnitContentFactory(key, parents, build_details, None, raw_record,
406
annotated, network_bytes=bytes)]
409
class KnitContent(object):
410
"""Content of a knit version to which deltas can be applied.
412
This is always stored in memory as a list of lines with \\n at the end,
413
plus a flag saying if the final ending is really there or not, because that
414
corresponds to the on-disk knit representation.
418
self._should_strip_eol = False
420
def apply_delta(self, delta, new_version_id):
421
"""Apply delta to this object to become new_version_id."""
422
raise NotImplementedError(self.apply_delta)
424
def line_delta_iter(self, new_lines):
425
"""Generate line-based delta from this content to new_lines."""
426
new_texts = new_lines.text()
427
old_texts = self.text()
428
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
429
for tag, i1, i2, j1, j2 in s.get_opcodes():
432
# ofrom, oto, length, data
433
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
435
def line_delta(self, new_lines):
436
return list(self.line_delta_iter(new_lines))
439
def get_line_delta_blocks(knit_delta, source, target):
440
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
441
target_len = len(target)
444
for s_begin, s_end, t_len, new_text in knit_delta:
445
true_n = s_begin - s_pos
448
# knit deltas do not provide reliable info about whether the
449
# last line of a file matches, due to eol handling.
450
if source[s_pos + n -1] != target[t_pos + n -1]:
453
yield s_pos, t_pos, n
454
t_pos += t_len + true_n
456
n = target_len - t_pos
458
if source[s_pos + n -1] != target[t_pos + n -1]:
461
yield s_pos, t_pos, n
462
yield s_pos + (target_len - t_pos), target_len, 0
465
class AnnotatedKnitContent(KnitContent):
466
"""Annotated content."""
468
def __init__(self, lines):
469
KnitContent.__init__(self)
473
"""Return a list of (origin, text) for each content line."""
474
lines = self._lines[:]
475
if self._should_strip_eol:
476
origin, last_line = lines[-1]
477
lines[-1] = (origin, last_line.rstrip('\n'))
480
def apply_delta(self, delta, new_version_id):
481
"""Apply delta to this object to become new_version_id."""
484
for start, end, count, delta_lines in delta:
485
lines[offset+start:offset+end] = delta_lines
486
offset = offset + (start - end) + count
490
lines = [text for origin, text in self._lines]
491
except ValueError, e:
492
# most commonly (only?) caused by the internal form of the knit
493
# missing annotation information because of a bug - see thread
495
raise KnitCorrupt(self,
496
"line in annotated knit missing annotation information: %s"
498
if self._should_strip_eol:
499
lines[-1] = lines[-1].rstrip('\n')
503
return AnnotatedKnitContent(self._lines[:])
506
class PlainKnitContent(KnitContent):
507
"""Unannotated content.
509
When annotate[_iter] is called on this content, the same version is reported
510
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
514
def __init__(self, lines, version_id):
515
KnitContent.__init__(self)
517
self._version_id = version_id
520
"""Return a list of (origin, text) for each content line."""
521
return [(self._version_id, line) for line in self._lines]
523
def apply_delta(self, delta, new_version_id):
524
"""Apply delta to this object to become new_version_id."""
527
for start, end, count, delta_lines in delta:
528
lines[offset+start:offset+end] = delta_lines
529
offset = offset + (start - end) + count
530
self._version_id = new_version_id
533
return PlainKnitContent(self._lines[:], self._version_id)
537
if self._should_strip_eol:
539
lines[-1] = lines[-1].rstrip('\n')
543
class _KnitFactory(object):
544
"""Base class for common Factory functions."""
546
def parse_record(self, version_id, record, record_details,
547
base_content, copy_base_content=True):
548
"""Parse a record into a full content object.
550
:param version_id: The official version id for this content
551
:param record: The data returned by read_records_iter()
552
:param record_details: Details about the record returned by
554
:param base_content: If get_build_details returns a compression_parent,
555
you must return a base_content here, else use None
556
:param copy_base_content: When building from the base_content, decide
557
you can either copy it and return a new object, or modify it in
559
:return: (content, delta) A Content object and possibly a line-delta,
562
method, noeol = record_details
563
if method == 'line-delta':
564
if copy_base_content:
565
content = base_content.copy()
567
content = base_content
568
delta = self.parse_line_delta(record, version_id)
569
content.apply_delta(delta, version_id)
571
content = self.parse_fulltext(record, version_id)
573
content._should_strip_eol = noeol
574
return (content, delta)
577
class KnitAnnotateFactory(_KnitFactory):
578
"""Factory for creating annotated Content objects."""
582
def make(self, lines, version_id):
583
num_lines = len(lines)
584
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
586
def parse_fulltext(self, content, version_id):
587
"""Convert fulltext to internal representation
589
fulltext content is of the format
590
revid(utf8) plaintext\n
591
internal representation is of the format:
594
# TODO: jam 20070209 The tests expect this to be returned as tuples,
595
# but the code itself doesn't really depend on that.
596
# Figure out a way to not require the overhead of turning the
597
# list back into tuples.
598
lines = [tuple(line.split(' ', 1)) for line in content]
599
return AnnotatedKnitContent(lines)
601
def parse_line_delta_iter(self, lines):
602
return iter(self.parse_line_delta(lines))
604
def parse_line_delta(self, lines, version_id, plain=False):
605
"""Convert a line based delta into internal representation.
607
line delta is in the form of:
608
intstart intend intcount
610
revid(utf8) newline\n
611
internal representation is
612
(start, end, count, [1..count tuples (revid, newline)])
614
:param plain: If True, the lines are returned as a plain
615
list without annotations, not as a list of (origin, content) tuples, i.e.
616
(start, end, count, [1..count newline])
623
def cache_and_return(line):
624
origin, text = line.split(' ', 1)
625
return cache.setdefault(origin, origin), text
627
# walk through the lines parsing.
628
# Note that the plain test is explicitly pulled out of the
629
# loop to minimise any performance impact
632
start, end, count = [int(n) for n in header.split(',')]
633
contents = [next().split(' ', 1)[1] for i in xrange(count)]
634
result.append((start, end, count, contents))
637
start, end, count = [int(n) for n in header.split(',')]
638
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
639
result.append((start, end, count, contents))
642
def get_fulltext_content(self, lines):
643
"""Extract just the content lines from a fulltext."""
644
return (line.split(' ', 1)[1] for line in lines)
646
def get_linedelta_content(self, lines):
647
"""Extract just the content from a line delta.
649
This doesn't return all of the extra information stored in a delta.
650
Only the actual content lines.
655
header = header.split(',')
656
count = int(header[2])
657
for i in xrange(count):
658
origin, text = next().split(' ', 1)
661
def lower_fulltext(self, content):
662
"""convert a fulltext content record into a serializable form.
664
see parse_fulltext which this inverts.
666
return ['%s %s' % (o, t) for o, t in content._lines]
668
def lower_line_delta(self, delta):
669
"""convert a delta into a serializable form.
671
See parse_line_delta which this inverts.
673
# TODO: jam 20070209 We only do the caching thing to make sure that
674
# the origin is a valid utf-8 line, eventually we could remove it
676
for start, end, c, lines in delta:
677
out.append('%d,%d,%d\n' % (start, end, c))
678
out.extend(origin + ' ' + text
679
for origin, text in lines)
682
def annotate(self, knit, key):
683
content = knit._get_content(key)
684
# adjust for the fact that serialised annotations are only key suffixes
686
if type(key) is tuple:
688
origins = content.annotate()
690
for origin, line in origins:
691
result.append((prefix + (origin,), line))
694
# XXX: This smells a bit. Why would key ever be a non-tuple here?
695
# Aren't keys defined to be tuples? -- spiv 20080618
696
return content.annotate()
699
class KnitPlainFactory(_KnitFactory):
700
"""Factory for creating plain Content objects."""
704
def make(self, lines, version_id):
705
return PlainKnitContent(lines, version_id)
707
def parse_fulltext(self, content, version_id):
708
"""This parses an unannotated fulltext.
710
Note that this is not a noop - the internal representation
711
has (versionid, line) - its just a constant versionid.
713
return self.make(content, version_id)
715
def parse_line_delta_iter(self, lines, version_id):
717
num_lines = len(lines)
718
while cur < num_lines:
721
start, end, c = [int(n) for n in header.split(',')]
722
yield start, end, c, lines[cur:cur+c]
725
def parse_line_delta(self, lines, version_id):
726
return list(self.parse_line_delta_iter(lines, version_id))
728
def get_fulltext_content(self, lines):
729
"""Extract just the content lines from a fulltext."""
732
def get_linedelta_content(self, lines):
733
"""Extract just the content from a line delta.
735
This doesn't return all of the extra information stored in a delta.
736
Only the actual content lines.
741
header = header.split(',')
742
count = int(header[2])
743
for i in xrange(count):
746
def lower_fulltext(self, content):
747
return content.text()
749
def lower_line_delta(self, delta):
751
for start, end, c, lines in delta:
752
out.append('%d,%d,%d\n' % (start, end, c))
756
def annotate(self, knit, key):
757
annotator = _KnitAnnotator(knit)
758
return annotator.annotate_flat(key)
762
def make_file_factory(annotated, mapper):
763
"""Create a factory for creating a file based KnitVersionedFiles.
765
This is only functional enough to run interface tests, it doesn't try to
766
provide a full pack environment.
768
:param annotated: knit annotations are wanted.
769
:param mapper: The mapper from keys to paths.
771
def factory(transport):
772
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
773
access = _KnitKeyAccess(transport, mapper)
774
return KnitVersionedFiles(index, access, annotated=annotated)
778
def make_pack_factory(graph, delta, keylength):
779
"""Create a factory for creating a pack based VersionedFiles.
781
This is only functional enough to run interface tests, it doesn't try to
782
provide a full pack environment.
784
:param graph: Store a graph.
785
:param delta: Delta compress contents.
786
:param keylength: How long should keys be.
788
def factory(transport):
789
parents = graph or delta
795
max_delta_chain = 200
798
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
799
key_elements=keylength)
800
stream = transport.open_write_stream('newpack')
801
writer = pack.ContainerWriter(stream.write)
803
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
804
deltas=delta, add_callback=graph_index.add_nodes)
805
access = pack_repo._DirectPackAccess({})
806
access.set_writer(writer, graph_index, (transport, 'newpack'))
807
result = KnitVersionedFiles(index, access,
808
max_delta_chain=max_delta_chain)
809
result.stream = stream
810
result.writer = writer
815
def cleanup_pack_knit(versioned_files):
816
versioned_files.stream.close()
817
versioned_files.writer.end()
820
def _get_total_build_size(self, keys, positions):
821
"""Determine the total bytes to build these keys.
823
(helper function because _KnitGraphIndex and _KndxIndex work the same, but
824
don't inherit from a common base.)
826
:param keys: Keys that we want to build
827
:param positions: dict of {key, (info, index_memo, comp_parent)} (such
828
as returned by _get_components_positions)
829
:return: Number of bytes to build those keys
831
all_build_index_memos = {}
835
for key in build_keys:
836
# This is mostly for the 'stacked' case
837
# Where we will be getting the data from a fallback
838
if key not in positions:
840
_, index_memo, compression_parent = positions[key]
841
all_build_index_memos[key] = index_memo
842
if compression_parent not in all_build_index_memos:
843
next_keys.add(compression_parent)
844
build_keys = next_keys
845
return sum([index_memo[2] for index_memo
846
in all_build_index_memos.itervalues()])
849
class KnitVersionedFiles(VersionedFilesWithFallbacks):
850
"""Storage for many versioned files using knit compression.
852
Backend storage is managed by indices and data objects.
854
:ivar _index: A _KnitGraphIndex or similar that can describe the
855
parents, graph, compression and data location of entries in this
856
KnitVersionedFiles. Note that this is only the index for
857
*this* vfs; if there are fallbacks they must be queried separately.
860
def __init__(self, index, data_access, max_delta_chain=200,
861
annotated=False, reload_func=None):
862
"""Create a KnitVersionedFiles with index and data_access.
864
:param index: The index for the knit data.
865
:param data_access: The access object to store and retrieve knit
867
:param max_delta_chain: The maximum number of deltas to permit during
868
insertion. Set to 0 to prohibit the use of deltas.
869
:param annotated: Set to True to cause annotations to be calculated and
870
stored during insertion.
871
:param reload_func: An function that can be called if we think we need
872
to reload the pack listing and try again. See
873
'bzrlib.repofmt.pack_repo.AggregateIndex' for the signature.
876
self._access = data_access
877
self._max_delta_chain = max_delta_chain
879
self._factory = KnitAnnotateFactory()
881
self._factory = KnitPlainFactory()
882
self._immediate_fallback_vfs = []
883
self._reload_func = reload_func
886
return "%s(%r, %r)" % (
887
self.__class__.__name__,
891
def without_fallbacks(self):
892
"""Return a clone of this object without any fallbacks configured."""
893
return KnitVersionedFiles(self._index, self._access,
894
self._max_delta_chain, self._factory.annotated,
897
def add_fallback_versioned_files(self, a_versioned_files):
898
"""Add a source of texts for texts not present in this knit.
900
:param a_versioned_files: A VersionedFiles object.
902
self._immediate_fallback_vfs.append(a_versioned_files)
904
def add_lines(self, key, parents, lines, parent_texts=None,
905
left_matching_blocks=None, nostore_sha=None, random_id=False,
907
"""See VersionedFiles.add_lines()."""
908
self._index._check_write_ok()
909
self._check_add(key, lines, random_id, check_content)
911
# The caller might pass None if there is no graph data, but kndx
912
# indexes can't directly store that, so we give them
913
# an empty tuple instead.
915
line_bytes = ''.join(lines)
916
return self._add(key, lines, parents,
917
parent_texts, left_matching_blocks, nostore_sha, random_id,
918
line_bytes=line_bytes)
920
def _add_text(self, key, parents, text, nostore_sha=None, random_id=False):
921
"""See VersionedFiles._add_text()."""
922
self._index._check_write_ok()
923
self._check_add(key, None, random_id, check_content=False)
924
if text.__class__ is not str:
925
raise errors.BzrBadParameterUnicode("text")
927
# The caller might pass None if there is no graph data, but kndx
928
# indexes can't directly store that, so we give them
929
# an empty tuple instead.
931
return self._add(key, None, parents,
932
None, None, nostore_sha, random_id,
935
def _add(self, key, lines, parents, parent_texts,
936
left_matching_blocks, nostore_sha, random_id,
938
"""Add a set of lines on top of version specified by parents.
940
Any versions not present will be converted into ghosts.
942
:param lines: A list of strings where each one is a single line (has a
943
single newline at the end of the string) This is now optional
944
(callers can pass None). It is left in its location for backwards
945
compatibility. It should ''.join(lines) must == line_bytes
946
:param line_bytes: A single string containing the content
948
We pass both lines and line_bytes because different routes bring the
949
values to this function. And for memory efficiency, we don't want to
950
have to split/join on-demand.
952
# first thing, if the content is something we don't need to store, find
954
digest = sha_string(line_bytes)
955
if nostore_sha == digest:
956
raise errors.ExistingContent
959
if parent_texts is None:
961
# Do a single query to ascertain parent presence; we only compress
962
# against parents in the same kvf.
963
present_parent_map = self._index.get_parent_map(parents)
964
for parent in parents:
965
if parent in present_parent_map:
966
present_parents.append(parent)
968
# Currently we can only compress against the left most present parent.
969
if (len(present_parents) == 0 or
970
present_parents[0] != parents[0]):
973
# To speed the extract of texts the delta chain is limited
974
# to a fixed number of deltas. This should minimize both
975
# I/O and the time spend applying deltas.
976
delta = self._check_should_delta(present_parents[0])
978
text_length = len(line_bytes)
981
# Note: line_bytes is not modified to add a newline, that is tracked
982
# via the no_eol flag. 'lines' *is* modified, because that is the
983
# general values needed by the Content code.
984
if line_bytes and line_bytes[-1] != '\n':
985
options.append('no-eol')
987
# Copy the existing list, or create a new one
989
lines = osutils.split_lines(line_bytes)
992
# Replace the last line with one that ends in a final newline
993
lines[-1] = lines[-1] + '\n'
995
lines = osutils.split_lines(line_bytes)
997
for element in key[:-1]:
998
if type(element) is not str:
999
raise TypeError("key contains non-strings: %r" % (key,))
1001
key = key[:-1] + ('sha1:' + digest,)
1002
elif type(key[-1]) is not str:
1003
raise TypeError("key contains non-strings: %r" % (key,))
1004
# Knit hunks are still last-element only
1005
version_id = key[-1]
1006
content = self._factory.make(lines, version_id)
1008
# Hint to the content object that its text() call should strip the
1010
content._should_strip_eol = True
1011
if delta or (self._factory.annotated and len(present_parents) > 0):
1012
# Merge annotations from parent texts if needed.
1013
delta_hunks = self._merge_annotations(content, present_parents,
1014
parent_texts, delta, self._factory.annotated,
1015
left_matching_blocks)
1018
options.append('line-delta')
1019
store_lines = self._factory.lower_line_delta(delta_hunks)
1020
size, bytes = self._record_to_data(key, digest,
1023
options.append('fulltext')
1024
# isinstance is slower and we have no hierarchy.
1025
if self._factory.__class__ is KnitPlainFactory:
1026
# Use the already joined bytes saving iteration time in
1028
dense_lines = [line_bytes]
1030
dense_lines.append('\n')
1031
size, bytes = self._record_to_data(key, digest,
1034
# get mixed annotation + content and feed it into the
1036
store_lines = self._factory.lower_fulltext(content)
1037
size, bytes = self._record_to_data(key, digest,
1040
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
1041
self._index.add_records(
1042
((key, options, access_memo, parents),),
1043
random_id=random_id)
1044
return digest, text_length, content
1046
def annotate(self, key):
1047
"""See VersionedFiles.annotate."""
1048
return self._factory.annotate(self, key)
1050
def get_annotator(self):
1051
return _KnitAnnotator(self)
1053
def check(self, progress_bar=None, keys=None):
1054
"""See VersionedFiles.check()."""
1056
return self._logical_check()
1058
# At the moment, check does not extra work over get_record_stream
1059
return self.get_record_stream(keys, 'unordered', True)
1061
def _logical_check(self):
1062
# This doesn't actually test extraction of everything, but that will
1063
# impact 'bzr check' substantially, and needs to be integrated with
1064
# care. However, it does check for the obvious problem of a delta with
1066
keys = self._index.keys()
1067
parent_map = self.get_parent_map(keys)
1069
if self._index.get_method(key) != 'fulltext':
1070
compression_parent = parent_map[key][0]
1071
if compression_parent not in parent_map:
1072
raise errors.KnitCorrupt(self,
1073
"Missing basis parent %s for %s" % (
1074
compression_parent, key))
1075
for fallback_vfs in self._immediate_fallback_vfs:
1076
fallback_vfs.check()
1078
def _check_add(self, key, lines, random_id, check_content):
1079
"""check that version_id and lines are safe to add."""
1080
version_id = key[-1]
1081
if version_id is not None:
1082
if contains_whitespace(version_id):
1083
raise InvalidRevisionId(version_id, self)
1084
self.check_not_reserved_id(version_id)
1085
# TODO: If random_id==False and the key is already present, we should
1086
# probably check that the existing content is identical to what is
1087
# being inserted, and otherwise raise an exception. This would make
1088
# the bundle code simpler.
1090
self._check_lines_not_unicode(lines)
1091
self._check_lines_are_lines(lines)
1093
def _check_header(self, key, line):
1094
rec = self._split_header(line)
1095
self._check_header_version(rec, key[-1])
1098
def _check_header_version(self, rec, version_id):
1099
"""Checks the header version on original format knit records.
1101
These have the last component of the key embedded in the record.
1103
if rec[1] != version_id:
1104
raise KnitCorrupt(self,
1105
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1107
def _check_should_delta(self, parent):
1108
"""Iterate back through the parent listing, looking for a fulltext.
1110
This is used when we want to decide whether to add a delta or a new
1111
fulltext. It searches for _max_delta_chain parents. When it finds a
1112
fulltext parent, it sees if the total size of the deltas leading up to
1113
it is large enough to indicate that we want a new full text anyway.
1115
Return True if we should create a new delta, False if we should use a
1119
fulltext_size = None
1120
for count in xrange(self._max_delta_chain):
1122
# Note that this only looks in the index of this particular
1123
# KnitVersionedFiles, not in the fallbacks. This ensures that
1124
# we won't store a delta spanning physical repository
1126
build_details = self._index.get_build_details([parent])
1127
parent_details = build_details[parent]
1128
except (RevisionNotPresent, KeyError), e:
1129
# Some basis is not locally present: always fulltext
1131
index_memo, compression_parent, _, _ = parent_details
1132
_, _, size = index_memo
1133
if compression_parent is None:
1134
fulltext_size = size
1137
# We don't explicitly check for presence because this is in an
1138
# inner loop, and if it's missing it'll fail anyhow.
1139
parent = compression_parent
1141
# We couldn't find a fulltext, so we must create a new one
1143
# Simple heuristic - if the total I/O wold be greater as a delta than
1144
# the originally installed fulltext, we create a new fulltext.
1145
return fulltext_size > delta_size
1147
def _build_details_to_components(self, build_details):
1148
"""Convert a build_details tuple to a position tuple."""
1149
# record_details, access_memo, compression_parent
1150
return build_details[3], build_details[0], build_details[1]
1152
def _get_components_positions(self, keys, allow_missing=False):
1153
"""Produce a map of position data for the components of keys.
1155
This data is intended to be used for retrieving the knit records.
1157
A dict of key to (record_details, index_memo, next, parents) is
1160
* method is the way referenced data should be applied.
1161
* index_memo is the handle to pass to the data access to actually get
1163
* next is the build-parent of the version, or None for fulltexts.
1164
* parents is the version_ids of the parents of this version
1166
:param allow_missing: If True do not raise an error on a missing
1167
component, just ignore it.
1170
pending_components = keys
1171
while pending_components:
1172
build_details = self._index.get_build_details(pending_components)
1173
current_components = set(pending_components)
1174
pending_components = set()
1175
for key, details in build_details.iteritems():
1176
(index_memo, compression_parent, parents,
1177
record_details) = details
1178
method = record_details[0]
1179
if compression_parent is not None:
1180
pending_components.add(compression_parent)
1181
component_data[key] = self._build_details_to_components(details)
1182
missing = current_components.difference(build_details)
1183
if missing and not allow_missing:
1184
raise errors.RevisionNotPresent(missing.pop(), self)
1185
return component_data
1187
def _get_content(self, key, parent_texts={}):
1188
"""Returns a content object that makes up the specified
1190
cached_version = parent_texts.get(key, None)
1191
if cached_version is not None:
1192
# Ensure the cache dict is valid.
1193
if not self.get_parent_map([key]):
1194
raise RevisionNotPresent(key, self)
1195
return cached_version
1196
generator = _VFContentMapGenerator(self, [key])
1197
return generator._get_content(key)
1199
def get_parent_map(self, keys):
1200
"""Get a map of the graph parents of keys.
1202
:param keys: The keys to look up parents for.
1203
:return: A mapping from keys to parents. Absent keys are absent from
1206
return self._get_parent_map_with_sources(keys)[0]
1208
def _get_parent_map_with_sources(self, keys):
1209
"""Get a map of the parents of keys.
1211
:param keys: The keys to look up parents for.
1212
:return: A tuple. The first element is a mapping from keys to parents.
1213
Absent keys are absent from the mapping. The second element is a
1214
list with the locations each key was found in. The first element
1215
is the in-this-knit parents, the second the first fallback source,
1219
sources = [self._index] + self._immediate_fallback_vfs
1222
for source in sources:
1225
new_result = source.get_parent_map(missing)
1226
source_results.append(new_result)
1227
result.update(new_result)
1228
missing.difference_update(set(new_result))
1229
return result, source_results
1231
def _get_record_map(self, keys, allow_missing=False):
1232
"""Produce a dictionary of knit records.
1234
:return: {key:(record, record_details, digest, next)}
1236
* record: data returned from read_records (a KnitContentobject)
1237
* record_details: opaque information to pass to parse_record
1238
* digest: SHA1 digest of the full text after all steps are done
1239
* next: build-parent of the version, i.e. the leftmost ancestor.
1240
Will be None if the record is not a delta.
1242
:param keys: The keys to build a map for
1243
:param allow_missing: If some records are missing, rather than
1244
error, just return the data that could be generated.
1246
raw_map = self._get_record_map_unparsed(keys,
1247
allow_missing=allow_missing)
1248
return self._raw_map_to_record_map(raw_map)
1250
def _raw_map_to_record_map(self, raw_map):
1251
"""Parse the contents of _get_record_map_unparsed.
1253
:return: see _get_record_map.
1257
data, record_details, next = raw_map[key]
1258
content, digest = self._parse_record(key[-1], data)
1259
result[key] = content, record_details, digest, next
1262
def _get_record_map_unparsed(self, keys, allow_missing=False):
1263
"""Get the raw data for reconstructing keys without parsing it.
1265
:return: A dict suitable for parsing via _raw_map_to_record_map.
1266
key-> raw_bytes, (method, noeol), compression_parent
1268
# This retries the whole request if anything fails. Potentially we
1269
# could be a bit more selective. We could track the keys whose records
1270
# we have successfully found, and then only request the new records
1271
# from there. However, _get_components_positions grabs the whole build
1272
# chain, which means we'll likely try to grab the same records again
1273
# anyway. Also, can the build chains change as part of a pack
1274
# operation? We wouldn't want to end up with a broken chain.
1277
position_map = self._get_components_positions(keys,
1278
allow_missing=allow_missing)
1279
# key = component_id, r = record_details, i_m = index_memo,
1281
records = [(key, i_m) for key, (r, i_m, n)
1282
in position_map.iteritems()]
1283
# Sort by the index memo, so that we request records from the
1284
# same pack file together, and in forward-sorted order
1285
records.sort(key=operator.itemgetter(1))
1287
for key, data in self._read_records_iter_unchecked(records):
1288
(record_details, index_memo, next) = position_map[key]
1289
raw_record_map[key] = data, record_details, next
1290
return raw_record_map
1291
except errors.RetryWithNewPacks, e:
1292
self._access.reload_or_raise(e)
1295
def _split_by_prefix(cls, keys):
1296
"""For the given keys, split them up based on their prefix.
1298
To keep memory pressure somewhat under control, split the
1299
requests back into per-file-id requests, otherwise "bzr co"
1300
extracts the full tree into memory before writing it to disk.
1301
This should be revisited if _get_content_maps() can ever cross
1304
The keys for a given file_id are kept in the same relative order.
1305
Ordering between file_ids is not, though prefix_order will return the
1306
order that the key was first seen.
1308
:param keys: An iterable of key tuples
1309
:return: (split_map, prefix_order)
1310
split_map A dictionary mapping prefix => keys
1311
prefix_order The order that we saw the various prefixes
1313
split_by_prefix = {}
1321
if prefix in split_by_prefix:
1322
split_by_prefix[prefix].append(key)
1324
split_by_prefix[prefix] = [key]
1325
prefix_order.append(prefix)
1326
return split_by_prefix, prefix_order
1328
def _group_keys_for_io(self, keys, non_local_keys, positions,
1329
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1330
"""For the given keys, group them into 'best-sized' requests.
1332
The idea is to avoid making 1 request per file, but to never try to
1333
unpack an entire 1.5GB source tree in a single pass. Also when
1334
possible, we should try to group requests to the same pack file
1337
:return: list of (keys, non_local) tuples that indicate what keys
1338
should be fetched next.
1340
# TODO: Ideally we would group on 2 factors. We want to extract texts
1341
# from the same pack file together, and we want to extract all
1342
# the texts for a given build-chain together. Ultimately it
1343
# probably needs a better global view.
1344
total_keys = len(keys)
1345
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1346
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1348
cur_non_local = set()
1352
for prefix in prefix_order:
1353
keys = prefix_split_keys[prefix]
1354
non_local = prefix_split_non_local_keys.get(prefix, [])
1356
this_size = self._index._get_total_build_size(keys, positions)
1357
cur_size += this_size
1358
cur_keys.extend(keys)
1359
cur_non_local.update(non_local)
1360
if cur_size > _min_buffer_size:
1361
result.append((cur_keys, cur_non_local))
1362
sizes.append(cur_size)
1364
cur_non_local = set()
1367
result.append((cur_keys, cur_non_local))
1368
sizes.append(cur_size)
1371
def get_record_stream(self, keys, ordering, include_delta_closure):
1372
"""Get a stream of records for keys.
1374
:param keys: The keys to include.
1375
:param ordering: Either 'unordered' or 'topological'. A topologically
1376
sorted stream has compression parents strictly before their
1378
:param include_delta_closure: If True then the closure across any
1379
compression parents will be included (in the opaque data).
1380
:return: An iterator of ContentFactory objects, each of which is only
1381
valid until the iterator is advanced.
1383
# keys might be a generator
1387
if not self._index.has_graph:
1388
# Cannot sort when no graph has been stored.
1389
ordering = 'unordered'
1391
remaining_keys = keys
1394
keys = set(remaining_keys)
1395
for content_factory in self._get_remaining_record_stream(keys,
1396
ordering, include_delta_closure):
1397
remaining_keys.discard(content_factory.key)
1398
yield content_factory
1400
except errors.RetryWithNewPacks, e:
1401
self._access.reload_or_raise(e)
1403
def _get_remaining_record_stream(self, keys, ordering,
1404
include_delta_closure):
1405
"""This function is the 'retry' portion for get_record_stream."""
1406
if include_delta_closure:
1407
positions = self._get_components_positions(keys, allow_missing=True)
1409
build_details = self._index.get_build_details(keys)
1411
# (record_details, access_memo, compression_parent_key)
1412
positions = dict((key, self._build_details_to_components(details))
1413
for key, details in build_details.iteritems())
1414
absent_keys = keys.difference(set(positions))
1415
# There may be more absent keys : if we're missing the basis component
1416
# and are trying to include the delta closure.
1417
# XXX: We should not ever need to examine remote sources because we do
1418
# not permit deltas across versioned files boundaries.
1419
if include_delta_closure:
1420
needed_from_fallback = set()
1421
# Build up reconstructable_keys dict. key:True in this dict means
1422
# the key can be reconstructed.
1423
reconstructable_keys = {}
1427
chain = [key, positions[key][2]]
1429
needed_from_fallback.add(key)
1432
while chain[-1] is not None:
1433
if chain[-1] in reconstructable_keys:
1434
result = reconstructable_keys[chain[-1]]
1438
chain.append(positions[chain[-1]][2])
1440
# missing basis component
1441
needed_from_fallback.add(chain[-1])
1444
for chain_key in chain[:-1]:
1445
reconstructable_keys[chain_key] = result
1447
needed_from_fallback.add(key)
1448
# Double index lookups here : need a unified api ?
1449
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1450
if ordering in ('topological', 'groupcompress'):
1451
if ordering == 'topological':
1452
# Global topological sort
1453
present_keys = tsort.topo_sort(global_map)
1455
present_keys = sort_groupcompress(global_map)
1456
# Now group by source:
1458
current_source = None
1459
for key in present_keys:
1460
for parent_map in parent_maps:
1461
if key in parent_map:
1462
key_source = parent_map
1464
if current_source is not key_source:
1465
source_keys.append((key_source, []))
1466
current_source = key_source
1467
source_keys[-1][1].append(key)
1469
if ordering != 'unordered':
1470
raise AssertionError('valid values for ordering are:'
1471
' "unordered", "groupcompress" or "topological" not: %r'
1473
# Just group by source; remote sources first.
1476
for parent_map in reversed(parent_maps):
1477
source_keys.append((parent_map, []))
1478
for key in parent_map:
1479
present_keys.append(key)
1480
source_keys[-1][1].append(key)
1481
# We have been requested to return these records in an order that
1482
# suits us. So we ask the index to give us an optimally sorted
1484
for source, sub_keys in source_keys:
1485
if source is parent_maps[0]:
1486
# Only sort the keys for this VF
1487
self._index._sort_keys_by_io(sub_keys, positions)
1488
absent_keys = keys - set(global_map)
1489
for key in absent_keys:
1490
yield AbsentContentFactory(key)
1491
# restrict our view to the keys we can answer.
1492
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1493
# XXX: At that point we need to consider the impact of double reads by
1494
# utilising components multiple times.
1495
if include_delta_closure:
1496
# XXX: get_content_maps performs its own index queries; allow state
1498
non_local_keys = needed_from_fallback - absent_keys
1499
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1502
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1505
for record in generator.get_record_stream():
1508
for source, keys in source_keys:
1509
if source is parent_maps[0]:
1510
# this KnitVersionedFiles
1511
records = [(key, positions[key][1]) for key in keys]
1512
for key, raw_data in self._read_records_iter_unchecked(records):
1513
(record_details, index_memo, _) = positions[key]
1514
yield KnitContentFactory(key, global_map[key],
1515
record_details, None, raw_data, self._factory.annotated, None)
1517
vf = self._immediate_fallback_vfs[parent_maps.index(source) - 1]
1518
for record in vf.get_record_stream(keys, ordering,
1519
include_delta_closure):
1522
def get_sha1s(self, keys):
1523
"""See VersionedFiles.get_sha1s()."""
1525
record_map = self._get_record_map(missing, allow_missing=True)
1527
for key, details in record_map.iteritems():
1528
if key not in missing:
1530
# record entry 2 is the 'digest'.
1531
result[key] = details[2]
1532
missing.difference_update(set(result))
1533
for source in self._immediate_fallback_vfs:
1536
new_result = source.get_sha1s(missing)
1537
result.update(new_result)
1538
missing.difference_update(set(new_result))
1541
def insert_record_stream(self, stream):
1542
"""Insert a record stream into this container.
1544
:param stream: A stream of records to insert.
1546
:seealso VersionedFiles.get_record_stream:
1548
def get_adapter(adapter_key):
1550
return adapters[adapter_key]
1552
adapter_factory = adapter_registry.get(adapter_key)
1553
adapter = adapter_factory(self)
1554
adapters[adapter_key] = adapter
1557
if self._factory.annotated:
1558
# self is annotated, we need annotated knits to use directly.
1559
annotated = "annotated-"
1562
# self is not annotated, but we can strip annotations cheaply.
1564
convertibles = set(["knit-annotated-ft-gz"])
1565
if self._max_delta_chain:
1566
delta_types.add("knit-annotated-delta-gz")
1567
convertibles.add("knit-annotated-delta-gz")
1568
# The set of types we can cheaply adapt without needing basis texts.
1569
native_types = set()
1570
if self._max_delta_chain:
1571
native_types.add("knit-%sdelta-gz" % annotated)
1572
delta_types.add("knit-%sdelta-gz" % annotated)
1573
native_types.add("knit-%sft-gz" % annotated)
1574
knit_types = native_types.union(convertibles)
1576
# Buffer all index entries that we can't add immediately because their
1577
# basis parent is missing. We don't buffer all because generating
1578
# annotations may require access to some of the new records. However we
1579
# can't generate annotations from new deltas until their basis parent
1580
# is present anyway, so we get away with not needing an index that
1581
# includes the new keys.
1583
# See <http://launchpad.net/bugs/300177> about ordering of compression
1584
# parents in the records - to be conservative, we insist that all
1585
# parents must be present to avoid expanding to a fulltext.
1587
# key = basis_parent, value = index entry to add
1588
buffered_index_entries = {}
1589
for record in stream:
1590
kind = record.storage_kind
1591
if kind.startswith('knit-') and kind.endswith('-gz'):
1592
# Check that the ID in the header of the raw knit bytes matches
1593
# the record metadata.
1594
raw_data = record._raw_record
1595
df, rec = self._parse_record_header(record.key, raw_data)
1598
parents = record.parents
1599
if record.storage_kind in delta_types:
1600
# TODO: eventually the record itself should track
1601
# compression_parent
1602
compression_parent = parents[0]
1604
compression_parent = None
1605
# Raise an error when a record is missing.
1606
if record.storage_kind == 'absent':
1607
raise RevisionNotPresent([record.key], self)
1608
elif ((record.storage_kind in knit_types)
1609
and (compression_parent is None
1610
or not self._immediate_fallback_vfs
1611
or self._index.has_key(compression_parent)
1612
or not self.has_key(compression_parent))):
1613
# we can insert the knit record literally if either it has no
1614
# compression parent OR we already have its basis in this kvf
1615
# OR the basis is not present even in the fallbacks. In the
1616
# last case it will either turn up later in the stream and all
1617
# will be well, or it won't turn up at all and we'll raise an
1620
# TODO: self.has_key is somewhat redundant with
1621
# self._index.has_key; we really want something that directly
1622
# asks if it's only present in the fallbacks. -- mbp 20081119
1623
if record.storage_kind not in native_types:
1625
adapter_key = (record.storage_kind, "knit-delta-gz")
1626
adapter = get_adapter(adapter_key)
1628
adapter_key = (record.storage_kind, "knit-ft-gz")
1629
adapter = get_adapter(adapter_key)
1630
bytes = adapter.get_bytes(record)
1632
# It's a knit record, it has a _raw_record field (even if
1633
# it was reconstituted from a network stream).
1634
bytes = record._raw_record
1635
options = [record._build_details[0]]
1636
if record._build_details[1]:
1637
options.append('no-eol')
1638
# Just blat it across.
1639
# Note: This does end up adding data on duplicate keys. As
1640
# modern repositories use atomic insertions this should not
1641
# lead to excessive growth in the event of interrupted fetches.
1642
# 'knit' repositories may suffer excessive growth, but as a
1643
# deprecated format this is tolerable. It can be fixed if
1644
# needed by in the kndx index support raising on a duplicate
1645
# add with identical parents and options.
1646
access_memo = self._access.add_raw_records(
1647
[(record.key, len(bytes))], bytes)[0]
1648
index_entry = (record.key, options, access_memo, parents)
1649
if 'fulltext' not in options:
1650
# Not a fulltext, so we need to make sure the compression
1651
# parent will also be present.
1652
# Note that pack backed knits don't need to buffer here
1653
# because they buffer all writes to the transaction level,
1654
# but we don't expose that difference at the index level. If
1655
# the query here has sufficient cost to show up in
1656
# profiling we should do that.
1658
# They're required to be physically in this
1659
# KnitVersionedFiles, not in a fallback.
1660
if not self._index.has_key(compression_parent):
1661
pending = buffered_index_entries.setdefault(
1662
compression_parent, [])
1663
pending.append(index_entry)
1666
self._index.add_records([index_entry])
1667
elif record.storage_kind == 'chunked':
1668
self.add_lines(record.key, parents,
1669
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1671
# Not suitable for direct insertion as a
1672
# delta, either because it's not the right format, or this
1673
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1674
# 0) or because it depends on a base only present in the
1676
self._access.flush()
1678
# Try getting a fulltext directly from the record.
1679
bytes = record.get_bytes_as('fulltext')
1680
except errors.UnavailableRepresentation:
1681
adapter_key = record.storage_kind, 'fulltext'
1682
adapter = get_adapter(adapter_key)
1683
bytes = adapter.get_bytes(record)
1684
lines = split_lines(bytes)
1686
self.add_lines(record.key, parents, lines)
1687
except errors.RevisionAlreadyPresent:
1689
# Add any records whose basis parent is now available.
1691
added_keys = [record.key]
1693
key = added_keys.pop(0)
1694
if key in buffered_index_entries:
1695
index_entries = buffered_index_entries[key]
1696
self._index.add_records(index_entries)
1698
[index_entry[0] for index_entry in index_entries])
1699
del buffered_index_entries[key]
1700
if buffered_index_entries:
1701
# There were index entries buffered at the end of the stream,
1702
# So these need to be added (if the index supports holding such
1703
# entries for later insertion)
1705
for key in buffered_index_entries:
1706
index_entries = buffered_index_entries[key]
1707
all_entries.extend(index_entries)
1708
self._index.add_records(
1709
all_entries, missing_compression_parents=True)
1711
def get_missing_compression_parent_keys(self):
1712
"""Return an iterable of keys of missing compression parents.
1714
Check this after calling insert_record_stream to find out if there are
1715
any missing compression parents. If there are, the records that
1716
depend on them are not able to be inserted safely. For atomic
1717
KnitVersionedFiles built on packs, the transaction should be aborted or
1718
suspended - commit will fail at this point. Nonatomic knits will error
1719
earlier because they have no staging area to put pending entries into.
1721
return self._index.get_missing_compression_parents()
1723
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1724
"""Iterate over the lines in the versioned files from keys.
1726
This may return lines from other keys. Each item the returned
1727
iterator yields is a tuple of a line and a text version that that line
1728
is present in (not introduced in).
1730
Ordering of results is in whatever order is most suitable for the
1731
underlying storage format.
1733
If a progress bar is supplied, it may be used to indicate progress.
1734
The caller is responsible for cleaning up progress bars (because this
1738
* Lines are normalised by the underlying store: they will all have \\n
1740
* Lines are returned in arbitrary order.
1741
* If a requested key did not change any lines (or didn't have any
1742
lines), it may not be mentioned at all in the result.
1744
:param pb: Progress bar supplied by caller.
1745
:return: An iterator over (line, key).
1748
pb = ui.ui_factory.nested_progress_bar()
1754
# we don't care about inclusions, the caller cares.
1755
# but we need to setup a list of records to visit.
1756
# we need key, position, length
1758
build_details = self._index.get_build_details(keys)
1759
for key, details in build_details.iteritems():
1761
key_records.append((key, details[0]))
1762
records_iter = enumerate(self._read_records_iter(key_records))
1763
for (key_idx, (key, data, sha_value)) in records_iter:
1764
pb.update(gettext('Walking content'), key_idx, total)
1765
compression_parent = build_details[key][1]
1766
if compression_parent is None:
1768
line_iterator = self._factory.get_fulltext_content(data)
1771
line_iterator = self._factory.get_linedelta_content(data)
1772
# Now that we are yielding the data for this key, remove it
1775
# XXX: It might be more efficient to yield (key,
1776
# line_iterator) in the future. However for now, this is a
1777
# simpler change to integrate into the rest of the
1778
# codebase. RBC 20071110
1779
for line in line_iterator:
1782
except errors.RetryWithNewPacks, e:
1783
self._access.reload_or_raise(e)
1784
# If there are still keys we've not yet found, we look in the fallback
1785
# vfs, and hope to find them there. Note that if the keys are found
1786
# but had no changes or no content, the fallback may not return
1788
if keys and not self._immediate_fallback_vfs:
1789
# XXX: strictly the second parameter is meant to be the file id
1790
# but it's not easily accessible here.
1791
raise RevisionNotPresent(keys, repr(self))
1792
for source in self._immediate_fallback_vfs:
1796
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1797
source_keys.add(key)
1799
keys.difference_update(source_keys)
1800
pb.update(gettext('Walking content'), total, total)
1802
def _make_line_delta(self, delta_seq, new_content):
1803
"""Generate a line delta from delta_seq and new_content."""
1805
for op in delta_seq.get_opcodes():
1806
if op[0] == 'equal':
1808
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1811
def _merge_annotations(self, content, parents, parent_texts={},
1812
delta=None, annotated=None,
1813
left_matching_blocks=None):
1814
"""Merge annotations for content and generate deltas.
1816
This is done by comparing the annotations based on changes to the text
1817
and generating a delta on the resulting full texts. If annotations are
1818
not being created then a simple delta is created.
1820
if left_matching_blocks is not None:
1821
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1825
for parent_key in parents:
1826
merge_content = self._get_content(parent_key, parent_texts)
1827
if (parent_key == parents[0] and delta_seq is not None):
1830
seq = patiencediff.PatienceSequenceMatcher(
1831
None, merge_content.text(), content.text())
1832
for i, j, n in seq.get_matching_blocks():
1835
# this copies (origin, text) pairs across to the new
1836
# content for any line that matches the last-checked
1838
content._lines[j:j+n] = merge_content._lines[i:i+n]
1839
# XXX: Robert says the following block is a workaround for a
1840
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1841
if content._lines and content._lines[-1][1][-1] != '\n':
1842
# The copied annotation was from a line without a trailing EOL,
1843
# reinstate one for the content object, to ensure correct
1845
line = content._lines[-1][1] + '\n'
1846
content._lines[-1] = (content._lines[-1][0], line)
1848
if delta_seq is None:
1849
reference_content = self._get_content(parents[0], parent_texts)
1850
new_texts = content.text()
1851
old_texts = reference_content.text()
1852
delta_seq = patiencediff.PatienceSequenceMatcher(
1853
None, old_texts, new_texts)
1854
return self._make_line_delta(delta_seq, content)
1856
def _parse_record(self, version_id, data):
1857
"""Parse an original format knit record.
1859
These have the last element of the key only present in the stored data.
1861
rec, record_contents = self._parse_record_unchecked(data)
1862
self._check_header_version(rec, version_id)
1863
return record_contents, rec[3]
1865
def _parse_record_header(self, key, raw_data):
1866
"""Parse a record header for consistency.
1868
:return: the header and the decompressor stream.
1869
as (stream, header_record)
1871
df = gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1874
rec = self._check_header(key, df.readline())
1875
except Exception, e:
1876
raise KnitCorrupt(self,
1877
"While reading {%s} got %s(%s)"
1878
% (key, e.__class__.__name__, str(e)))
1881
def _parse_record_unchecked(self, data):
1883
# 4168 calls in 2880 217 internal
1884
# 4168 calls to _parse_record_header in 2121
1885
# 4168 calls to readlines in 330
1886
df = gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1888
record_contents = df.readlines()
1889
except Exception, e:
1890
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1891
(data, e.__class__.__name__, str(e)))
1892
header = record_contents.pop(0)
1893
rec = self._split_header(header)
1894
last_line = record_contents.pop()
1895
if len(record_contents) != int(rec[2]):
1896
raise KnitCorrupt(self,
1897
'incorrect number of lines %s != %s'
1898
' for version {%s} %s'
1899
% (len(record_contents), int(rec[2]),
1900
rec[1], record_contents))
1901
if last_line != 'end %s\n' % rec[1]:
1902
raise KnitCorrupt(self,
1903
'unexpected version end line %r, wanted %r'
1904
% (last_line, rec[1]))
1906
return rec, record_contents
1908
def _read_records_iter(self, records):
1909
"""Read text records from data file and yield result.
1911
The result will be returned in whatever is the fastest to read.
1912
Not by the order requested. Also, multiple requests for the same
1913
record will only yield 1 response.
1915
:param records: A list of (key, access_memo) entries
1916
:return: Yields (key, contents, digest) in the order
1917
read, not the order requested
1922
# XXX: This smells wrong, IO may not be getting ordered right.
1923
needed_records = sorted(set(records), key=operator.itemgetter(1))
1924
if not needed_records:
1927
# The transport optimizes the fetching as well
1928
# (ie, reads continuous ranges.)
1929
raw_data = self._access.get_raw_records(
1930
[index_memo for key, index_memo in needed_records])
1932
for (key, index_memo), data in \
1933
izip(iter(needed_records), raw_data):
1934
content, digest = self._parse_record(key[-1], data)
1935
yield key, content, digest
1937
def _read_records_iter_raw(self, records):
1938
"""Read text records from data file and yield raw data.
1940
This unpacks enough of the text record to validate the id is
1941
as expected but thats all.
1943
Each item the iterator yields is (key, bytes,
1944
expected_sha1_of_full_text).
1946
for key, data in self._read_records_iter_unchecked(records):
1947
# validate the header (note that we can only use the suffix in
1948
# current knit records).
1949
df, rec = self._parse_record_header(key, data)
1951
yield key, data, rec[3]
1953
def _read_records_iter_unchecked(self, records):
1954
"""Read text records from data file and yield raw data.
1956
No validation is done.
1958
Yields tuples of (key, data).
1960
# setup an iterator of the external records:
1961
# uses readv so nice and fast we hope.
1963
# grab the disk data needed.
1964
needed_offsets = [index_memo for key, index_memo
1966
raw_records = self._access.get_raw_records(needed_offsets)
1968
for key, index_memo in records:
1969
data = raw_records.next()
1972
def _record_to_data(self, key, digest, lines, dense_lines=None):
1973
"""Convert key, digest, lines into a raw data block.
1975
:param key: The key of the record. Currently keys are always serialised
1976
using just the trailing component.
1977
:param dense_lines: The bytes of lines but in a denser form. For
1978
instance, if lines is a list of 1000 bytestrings each ending in
1979
\\n, dense_lines may be a list with one line in it, containing all
1980
the 1000's lines and their \\n's. Using dense_lines if it is
1981
already known is a win because the string join to create bytes in
1982
this function spends less time resizing the final string.
1983
:return: (len, a StringIO instance with the raw data ready to read.)
1985
chunks = ["version %s %d %s\n" % (key[-1], len(lines), digest)]
1986
chunks.extend(dense_lines or lines)
1987
chunks.append("end %s\n" % key[-1])
1988
for chunk in chunks:
1989
if type(chunk) is not str:
1990
raise AssertionError(
1991
'data must be plain bytes was %s' % type(chunk))
1992
if lines and lines[-1][-1] != '\n':
1993
raise ValueError('corrupt lines value %r' % lines)
1994
compressed_bytes = tuned_gzip.chunks_to_gzip(chunks)
1995
return len(compressed_bytes), compressed_bytes
1997
def _split_header(self, line):
2000
raise KnitCorrupt(self,
2001
'unexpected number of elements in record header')
2005
"""See VersionedFiles.keys."""
2006
if 'evil' in debug.debug_flags:
2007
trace.mutter_callsite(2, "keys scales with size of history")
2008
sources = [self._index] + self._immediate_fallback_vfs
2010
for source in sources:
2011
result.update(source.keys())
2015
class _ContentMapGenerator(object):
2016
"""Generate texts or expose raw deltas for a set of texts."""
2018
def __init__(self, ordering='unordered'):
2019
self._ordering = ordering
2021
def _get_content(self, key):
2022
"""Get the content object for key."""
2023
# Note that _get_content is only called when the _ContentMapGenerator
2024
# has been constructed with just one key requested for reconstruction.
2025
if key in self.nonlocal_keys:
2026
record = self.get_record_stream().next()
2027
# Create a content object on the fly
2028
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
2029
return PlainKnitContent(lines, record.key)
2031
# local keys we can ask for directly
2032
return self._get_one_work(key)
2034
def get_record_stream(self):
2035
"""Get a record stream for the keys requested during __init__."""
2036
for record in self._work():
2040
"""Produce maps of text and KnitContents as dicts.
2042
:return: (text_map, content_map) where text_map contains the texts for
2043
the requested versions and content_map contains the KnitContents.
2045
# NB: By definition we never need to read remote sources unless texts
2046
# are requested from them: we don't delta across stores - and we
2047
# explicitly do not want to to prevent data loss situations.
2048
if self.global_map is None:
2049
self.global_map = self.vf.get_parent_map(self.keys)
2050
nonlocal_keys = self.nonlocal_keys
2052
missing_keys = set(nonlocal_keys)
2053
# Read from remote versioned file instances and provide to our caller.
2054
for source in self.vf._immediate_fallback_vfs:
2055
if not missing_keys:
2057
# Loop over fallback repositories asking them for texts - ignore
2058
# any missing from a particular fallback.
2059
for record in source.get_record_stream(missing_keys,
2060
self._ordering, True):
2061
if record.storage_kind == 'absent':
2062
# Not in thie particular stream, may be in one of the
2063
# other fallback vfs objects.
2065
missing_keys.remove(record.key)
2068
if self._raw_record_map is None:
2069
raise AssertionError('_raw_record_map should have been filled')
2071
for key in self.keys:
2072
if key in self.nonlocal_keys:
2074
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2077
def _get_one_work(self, requested_key):
2078
# Now, if we have calculated everything already, just return the
2080
if requested_key in self._contents_map:
2081
return self._contents_map[requested_key]
2082
# To simplify things, parse everything at once - code that wants one text
2083
# probably wants them all.
2084
# FUTURE: This function could be improved for the 'extract many' case
2085
# by tracking each component and only doing the copy when the number of
2086
# children than need to apply delta's to it is > 1 or it is part of the
2088
multiple_versions = len(self.keys) != 1
2089
if self._record_map is None:
2090
self._record_map = self.vf._raw_map_to_record_map(
2091
self._raw_record_map)
2092
record_map = self._record_map
2093
# raw_record_map is key:
2094
# Have read and parsed records at this point.
2095
for key in self.keys:
2096
if key in self.nonlocal_keys:
2101
while cursor is not None:
2103
record, record_details, digest, next = record_map[cursor]
2105
raise RevisionNotPresent(cursor, self)
2106
components.append((cursor, record, record_details, digest))
2108
if cursor in self._contents_map:
2109
# no need to plan further back
2110
components.append((cursor, None, None, None))
2114
for (component_id, record, record_details,
2115
digest) in reversed(components):
2116
if component_id in self._contents_map:
2117
content = self._contents_map[component_id]
2119
content, delta = self._factory.parse_record(key[-1],
2120
record, record_details, content,
2121
copy_base_content=multiple_versions)
2122
if multiple_versions:
2123
self._contents_map[component_id] = content
2125
# digest here is the digest from the last applied component.
2126
text = content.text()
2127
actual_sha = sha_strings(text)
2128
if actual_sha != digest:
2129
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2130
if multiple_versions:
2131
return self._contents_map[requested_key]
2135
def _wire_bytes(self):
2136
"""Get the bytes to put on the wire for 'key'.
2138
The first collection of bytes asked for returns the serialised
2139
raw_record_map and the additional details (key, parent) for key.
2140
Subsequent calls return just the additional details (key, parent).
2141
The wire storage_kind given for the first key is 'knit-delta-closure',
2142
For subsequent keys it is 'knit-delta-closure-ref'.
2144
:param key: A key from the content generator.
2145
:return: Bytes to put on the wire.
2148
# kind marker for dispatch on the far side,
2149
lines.append('knit-delta-closure')
2151
if self.vf._factory.annotated:
2152
lines.append('annotated')
2155
# then the list of keys
2156
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2157
if key not in self.nonlocal_keys]))
2158
# then the _raw_record_map in serialised form:
2160
# for each item in the map:
2162
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2163
# one line with method
2164
# one line with noeol
2165
# one line with next ('' for None)
2166
# one line with byte count of the record bytes
2168
for key, (record_bytes, (method, noeol), next) in \
2169
self._raw_record_map.iteritems():
2170
key_bytes = '\x00'.join(key)
2171
parents = self.global_map.get(key, None)
2173
parent_bytes = 'None:'
2175
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2176
method_bytes = method
2182
next_bytes = '\x00'.join(next)
2185
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2186
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2187
len(record_bytes), record_bytes))
2188
map_bytes = ''.join(map_byte_list)
2189
lines.append(map_bytes)
2190
bytes = '\n'.join(lines)
2194
class _VFContentMapGenerator(_ContentMapGenerator):
2195
"""Content map generator reading from a VersionedFiles object."""
2197
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2198
global_map=None, raw_record_map=None, ordering='unordered'):
2199
"""Create a _ContentMapGenerator.
2201
:param versioned_files: The versioned files that the texts are being
2203
:param keys: The keys to produce content maps for.
2204
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2205
which are known to not be in this knit, but rather in one of the
2207
:param global_map: The result of get_parent_map(keys) (or a supermap).
2208
This is required if get_record_stream() is to be used.
2209
:param raw_record_map: A unparsed raw record map to use for answering
2212
_ContentMapGenerator.__init__(self, ordering=ordering)
2213
# The vf to source data from
2214
self.vf = versioned_files
2216
self.keys = list(keys)
2217
# Keys known to be in fallback vfs objects
2218
if nonlocal_keys is None:
2219
self.nonlocal_keys = set()
2221
self.nonlocal_keys = frozenset(nonlocal_keys)
2222
# Parents data for keys to be returned in get_record_stream
2223
self.global_map = global_map
2224
# The chunked lists for self.keys in text form
2226
# A cache of KnitContent objects used in extracting texts.
2227
self._contents_map = {}
2228
# All the knit records needed to assemble the requested keys as full
2230
self._record_map = None
2231
if raw_record_map is None:
2232
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2235
self._raw_record_map = raw_record_map
2236
# the factory for parsing records
2237
self._factory = self.vf._factory
2240
class _NetworkContentMapGenerator(_ContentMapGenerator):
2241
"""Content map generator sourced from a network stream."""
2243
def __init__(self, bytes, line_end):
2244
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2246
self.global_map = {}
2247
self._raw_record_map = {}
2248
self._contents_map = {}
2249
self._record_map = None
2250
self.nonlocal_keys = []
2251
# Get access to record parsing facilities
2252
self.vf = KnitVersionedFiles(None, None)
2255
line_end = bytes.find('\n', start)
2256
line = bytes[start:line_end]
2257
start = line_end + 1
2258
if line == 'annotated':
2259
self._factory = KnitAnnotateFactory()
2261
self._factory = KnitPlainFactory()
2262
# list of keys to emit in get_record_stream
2263
line_end = bytes.find('\n', start)
2264
line = bytes[start:line_end]
2265
start = line_end + 1
2267
tuple(segment.split('\x00')) for segment in line.split('\t')
2269
# now a loop until the end. XXX: It would be nice if this was just a
2270
# bunch of the same records as get_record_stream(..., False) gives, but
2271
# there is a decent sized gap stopping that at the moment.
2275
line_end = bytes.find('\n', start)
2276
key = tuple(bytes[start:line_end].split('\x00'))
2277
start = line_end + 1
2278
# 1 line with parents (None: for None, '' for ())
2279
line_end = bytes.find('\n', start)
2280
line = bytes[start:line_end]
2285
[tuple(segment.split('\x00')) for segment in line.split('\t')
2287
self.global_map[key] = parents
2288
start = line_end + 1
2289
# one line with method
2290
line_end = bytes.find('\n', start)
2291
line = bytes[start:line_end]
2293
start = line_end + 1
2294
# one line with noeol
2295
line_end = bytes.find('\n', start)
2296
line = bytes[start:line_end]
2298
start = line_end + 1
2299
# one line with next ('' for None)
2300
line_end = bytes.find('\n', start)
2301
line = bytes[start:line_end]
2305
next = tuple(bytes[start:line_end].split('\x00'))
2306
start = line_end + 1
2307
# one line with byte count of the record bytes
2308
line_end = bytes.find('\n', start)
2309
line = bytes[start:line_end]
2311
start = line_end + 1
2313
record_bytes = bytes[start:start+count]
2314
start = start + count
2316
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2318
def get_record_stream(self):
2319
"""Get a record stream for for keys requested by the bytestream."""
2321
for key in self.keys:
2322
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2325
def _wire_bytes(self):
2329
class _KndxIndex(object):
2330
"""Manages knit index files
2332
The index is kept in memory and read on startup, to enable
2333
fast lookups of revision information. The cursor of the index
2334
file is always pointing to the end, making it easy to append
2337
_cache is a cache for fast mapping from version id to a Index
2340
_history is a cache for fast mapping from indexes to version ids.
2342
The index data format is dictionary compressed when it comes to
2343
parent references; a index entry may only have parents that with a
2344
lover index number. As a result, the index is topological sorted.
2346
Duplicate entries may be written to the index for a single version id
2347
if this is done then the latter one completely replaces the former:
2348
this allows updates to correct version and parent information.
2349
Note that the two entries may share the delta, and that successive
2350
annotations and references MUST point to the first entry.
2352
The index file on disc contains a header, followed by one line per knit
2353
record. The same revision can be present in an index file more than once.
2354
The first occurrence gets assigned a sequence number starting from 0.
2356
The format of a single line is
2357
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
2358
REVISION_ID is a utf8-encoded revision id
2359
FLAGS is a comma separated list of flags about the record. Values include
2360
no-eol, line-delta, fulltext.
2361
BYTE_OFFSET is the ascii representation of the byte offset in the data file
2362
that the compressed data starts at.
2363
LENGTH is the ascii representation of the length of the data file.
2364
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
2366
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
2367
revision id already in the knit that is a parent of REVISION_ID.
2368
The ' :' marker is the end of record marker.
2371
when a write is interrupted to the index file, it will result in a line
2372
that does not end in ' :'. If the ' :' is not present at the end of a line,
2373
or at the end of the file, then the record that is missing it will be
2374
ignored by the parser.
2376
When writing new records to the index file, the data is preceded by '\n'
2377
to ensure that records always start on new lines even if the last write was
2378
interrupted. As a result its normal for the last line in the index to be
2379
missing a trailing newline. One can be added with no harmful effects.
2381
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2382
where prefix is e.g. the (fileid,) for .texts instances or () for
2383
constant-mapped things like .revisions, and the old state is
2384
tuple(history_vector, cache_dict). This is used to prevent having an
2385
ABI change with the C extension that reads .kndx files.
2388
HEADER = "# bzr knit index 8\n"
2390
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2391
"""Create a _KndxIndex on transport using mapper."""
2392
self._transport = transport
2393
self._mapper = mapper
2394
self._get_scope = get_scope
2395
self._allow_writes = allow_writes
2396
self._is_locked = is_locked
2398
self.has_graph = True
2400
def add_records(self, records, random_id=False, missing_compression_parents=False):
2401
"""Add multiple records to the index.
2403
:param records: a list of tuples:
2404
(key, options, access_memo, parents).
2405
:param random_id: If True the ids being added were randomly generated
2406
and no check for existence will be performed.
2407
:param missing_compression_parents: If True the records being added are
2408
only compressed against texts already in the index (or inside
2409
records). If False the records all refer to unavailable texts (or
2410
texts inside records) as compression parents.
2412
if missing_compression_parents:
2413
# It might be nice to get the edge of the records. But keys isn't
2415
keys = sorted(record[0] for record in records)
2416
raise errors.RevisionNotPresent(keys, self)
2418
for record in records:
2421
path = self._mapper.map(key) + '.kndx'
2422
path_keys = paths.setdefault(path, (prefix, []))
2423
path_keys[1].append(record)
2424
for path in sorted(paths):
2425
prefix, path_keys = paths[path]
2426
self._load_prefixes([prefix])
2428
orig_history = self._kndx_cache[prefix][1][:]
2429
orig_cache = self._kndx_cache[prefix][0].copy()
2432
for key, options, (_, pos, size), parents in path_keys:
2434
# kndx indices cannot be parentless.
2436
line = "\n%s %s %s %s %s :" % (
2437
key[-1], ','.join(options), pos, size,
2438
self._dictionary_compress(parents))
2439
if type(line) is not str:
2440
raise AssertionError(
2441
'data must be utf8 was %s' % type(line))
2443
self._cache_key(key, options, pos, size, parents)
2444
if len(orig_history):
2445
self._transport.append_bytes(path, ''.join(lines))
2447
self._init_index(path, lines)
2449
# If any problems happen, restore the original values and re-raise
2450
self._kndx_cache[prefix] = (orig_cache, orig_history)
2453
def scan_unvalidated_index(self, graph_index):
2454
"""See _KnitGraphIndex.scan_unvalidated_index."""
2455
# Because kndx files do not support atomic insertion via separate index
2456
# files, they do not support this method.
2457
raise NotImplementedError(self.scan_unvalidated_index)
2459
def get_missing_compression_parents(self):
2460
"""See _KnitGraphIndex.get_missing_compression_parents."""
2461
# Because kndx files do not support atomic insertion via separate index
2462
# files, they do not support this method.
2463
raise NotImplementedError(self.get_missing_compression_parents)
2465
def _cache_key(self, key, options, pos, size, parent_keys):
2466
"""Cache a version record in the history array and index cache.
2468
This is inlined into _load_data for performance. KEEP IN SYNC.
2469
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2473
version_id = key[-1]
2474
# last-element only for compatibilty with the C load_data.
2475
parents = tuple(parent[-1] for parent in parent_keys)
2476
for parent in parent_keys:
2477
if parent[:-1] != prefix:
2478
raise ValueError("mismatched prefixes for %r, %r" % (
2480
cache, history = self._kndx_cache[prefix]
2481
# only want the _history index to reference the 1st index entry
2483
if version_id not in cache:
2484
index = len(history)
2485
history.append(version_id)
2487
index = cache[version_id][5]
2488
cache[version_id] = (version_id,
2495
def check_header(self, fp):
2496
line = fp.readline()
2498
# An empty file can actually be treated as though the file doesn't
2500
raise errors.NoSuchFile(self)
2501
if line != self.HEADER:
2502
raise KnitHeaderError(badline=line, filename=self)
2504
def _check_read(self):
2505
if not self._is_locked():
2506
raise errors.ObjectNotLocked(self)
2507
if self._get_scope() != self._scope:
2510
def _check_write_ok(self):
2511
"""Assert if not writes are permitted."""
2512
if not self._is_locked():
2513
raise errors.ObjectNotLocked(self)
2514
if self._get_scope() != self._scope:
2516
if self._mode != 'w':
2517
raise errors.ReadOnlyObjectDirtiedError(self)
2519
def get_build_details(self, keys):
2520
"""Get the method, index_memo and compression parent for keys.
2522
Ghosts are omitted from the result.
2524
:param keys: An iterable of keys.
2525
:return: A dict of key:(index_memo, compression_parent, parents,
2528
opaque structure to pass to read_records to extract the raw
2531
Content that this record is built upon, may be None
2533
Logical parents of this node
2535
extra information about the content which needs to be passed to
2536
Factory.parse_record
2538
parent_map = self.get_parent_map(keys)
2541
if key not in parent_map:
2543
method = self.get_method(key)
2544
parents = parent_map[key]
2545
if method == 'fulltext':
2546
compression_parent = None
2548
compression_parent = parents[0]
2549
noeol = 'no-eol' in self.get_options(key)
2550
index_memo = self.get_position(key)
2551
result[key] = (index_memo, compression_parent,
2552
parents, (method, noeol))
2555
def get_method(self, key):
2556
"""Return compression method of specified key."""
2557
options = self.get_options(key)
2558
if 'fulltext' in options:
2560
elif 'line-delta' in options:
2563
raise errors.KnitIndexUnknownMethod(self, options)
2565
def get_options(self, key):
2566
"""Return a list representing options.
2570
prefix, suffix = self._split_key(key)
2571
self._load_prefixes([prefix])
2573
return self._kndx_cache[prefix][0][suffix][1]
2575
raise RevisionNotPresent(key, self)
2577
def find_ancestry(self, keys):
2578
"""See CombinedGraphIndex.find_ancestry()"""
2579
prefixes = set(key[:-1] for key in keys)
2580
self._load_prefixes(prefixes)
2583
missing_keys = set()
2584
pending_keys = list(keys)
2585
# This assumes that keys will not reference parents in a different
2586
# prefix, which is accurate so far.
2588
key = pending_keys.pop()
2589
if key in parent_map:
2593
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2595
missing_keys.add(key)
2597
parent_keys = tuple([prefix + (suffix,)
2598
for suffix in suffix_parents])
2599
parent_map[key] = parent_keys
2600
pending_keys.extend([p for p in parent_keys
2601
if p not in parent_map])
2602
return parent_map, missing_keys
2604
def get_parent_map(self, keys):
2605
"""Get a map of the parents of keys.
2607
:param keys: The keys to look up parents for.
2608
:return: A mapping from keys to parents. Absent keys are absent from
2611
# Parse what we need to up front, this potentially trades off I/O
2612
# locality (.kndx and .knit in the same block group for the same file
2613
# id) for less checking in inner loops.
2614
prefixes = set(key[:-1] for key in keys)
2615
self._load_prefixes(prefixes)
2620
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2624
result[key] = tuple(prefix + (suffix,) for
2625
suffix in suffix_parents)
2628
def get_position(self, key):
2629
"""Return details needed to access the version.
2631
:return: a tuple (key, data position, size) to hand to the access
2632
logic to get the record.
2634
prefix, suffix = self._split_key(key)
2635
self._load_prefixes([prefix])
2636
entry = self._kndx_cache[prefix][0][suffix]
2637
return key, entry[2], entry[3]
2639
has_key = _mod_index._has_key_from_parent_map
2641
def _init_index(self, path, extra_lines=[]):
2642
"""Initialize an index."""
2644
sio.write(self.HEADER)
2645
sio.writelines(extra_lines)
2647
self._transport.put_file_non_atomic(path, sio,
2648
create_parent_dir=True)
2649
# self._create_parent_dir)
2650
# mode=self._file_mode,
2651
# dir_mode=self._dir_mode)
2654
"""Get all the keys in the collection.
2656
The keys are not ordered.
2659
# Identify all key prefixes.
2660
# XXX: A bit hacky, needs polish.
2661
if type(self._mapper) is ConstantMapper:
2665
for quoted_relpath in self._transport.iter_files_recursive():
2666
path, ext = os.path.splitext(quoted_relpath)
2668
prefixes = [self._mapper.unmap(path) for path in relpaths]
2669
self._load_prefixes(prefixes)
2670
for prefix in prefixes:
2671
for suffix in self._kndx_cache[prefix][1]:
2672
result.add(prefix + (suffix,))
2675
def _load_prefixes(self, prefixes):
2676
"""Load the indices for prefixes."""
2678
for prefix in prefixes:
2679
if prefix not in self._kndx_cache:
2680
# the load_data interface writes to these variables.
2683
self._filename = prefix
2685
path = self._mapper.map(prefix) + '.kndx'
2686
fp = self._transport.get(path)
2688
# _load_data may raise NoSuchFile if the target knit is
2690
_load_data(self, fp)
2693
self._kndx_cache[prefix] = (self._cache, self._history)
2698
self._kndx_cache[prefix] = ({}, [])
2699
if type(self._mapper) is ConstantMapper:
2700
# preserve behaviour for revisions.kndx etc.
2701
self._init_index(path)
2706
missing_keys = _mod_index._missing_keys_from_parent_map
2708
def _partition_keys(self, keys):
2709
"""Turn keys into a dict of prefix:suffix_list."""
2712
prefix_keys = result.setdefault(key[:-1], [])
2713
prefix_keys.append(key[-1])
2716
def _dictionary_compress(self, keys):
2717
"""Dictionary compress keys.
2719
:param keys: The keys to generate references to.
2720
:return: A string representation of keys. keys which are present are
2721
dictionary compressed, and others are emitted as fulltext with a
2727
prefix = keys[0][:-1]
2728
cache = self._kndx_cache[prefix][0]
2730
if key[:-1] != prefix:
2731
# kndx indices cannot refer across partitioned storage.
2732
raise ValueError("mismatched prefixes for %r" % keys)
2733
if key[-1] in cache:
2734
# -- inlined lookup() --
2735
result_list.append(str(cache[key[-1]][5]))
2736
# -- end lookup () --
2738
result_list.append('.' + key[-1])
2739
return ' '.join(result_list)
2741
def _reset_cache(self):
2742
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2743
# (cache_dict, history_vector) for parsed kndx files.
2744
self._kndx_cache = {}
2745
self._scope = self._get_scope()
2746
allow_writes = self._allow_writes()
2752
def _sort_keys_by_io(self, keys, positions):
2753
"""Figure out an optimal order to read the records for the given keys.
2755
Sort keys, grouped by index and sorted by position.
2757
:param keys: A list of keys whose records we want to read. This will be
2759
:param positions: A dict, such as the one returned by
2760
_get_components_positions()
2763
def get_sort_key(key):
2764
index_memo = positions[key][1]
2765
# Group by prefix and position. index_memo[0] is the key, so it is
2766
# (file_id, revision_id) and we don't want to sort on revision_id,
2767
# index_memo[1] is the position, and index_memo[2] is the size,
2768
# which doesn't matter for the sort
2769
return index_memo[0][:-1], index_memo[1]
2770
return keys.sort(key=get_sort_key)
2772
_get_total_build_size = _get_total_build_size
2774
def _split_key(self, key):
2775
"""Split key into a prefix and suffix."""
2776
return key[:-1], key[-1]
2779
class _KnitGraphIndex(object):
2780
"""A KnitVersionedFiles index layered on GraphIndex."""
2782
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2783
add_callback=None, track_external_parent_refs=False):
2784
"""Construct a KnitGraphIndex on a graph_index.
2786
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2787
:param is_locked: A callback to check whether the object should answer
2789
:param deltas: Allow delta-compressed records.
2790
:param parents: If True, record knits parents, if not do not record
2792
:param add_callback: If not None, allow additions to the index and call
2793
this callback with a list of added GraphIndex nodes:
2794
[(node, value, node_refs), ...]
2795
:param is_locked: A callback, returns True if the index is locked and
2797
:param track_external_parent_refs: If True, record all external parent
2798
references parents from added records. These can be retrieved
2799
later by calling get_missing_parents().
2801
self._add_callback = add_callback
2802
self._graph_index = graph_index
2803
self._deltas = deltas
2804
self._parents = parents
2805
if deltas and not parents:
2806
# XXX: TODO: Delta tree and parent graph should be conceptually
2808
raise KnitCorrupt(self, "Cannot do delta compression without "
2810
self.has_graph = parents
2811
self._is_locked = is_locked
2812
self._missing_compression_parents = set()
2813
if track_external_parent_refs:
2814
self._key_dependencies = _KeyRefs()
2816
self._key_dependencies = None
2819
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2821
def add_records(self, records, random_id=False,
2822
missing_compression_parents=False):
2823
"""Add multiple records to the index.
2825
This function does not insert data into the Immutable GraphIndex
2826
backing the KnitGraphIndex, instead it prepares data for insertion by
2827
the caller and checks that it is safe to insert then calls
2828
self._add_callback with the prepared GraphIndex nodes.
2830
:param records: a list of tuples:
2831
(key, options, access_memo, parents).
2832
:param random_id: If True the ids being added were randomly generated
2833
and no check for existence will be performed.
2834
:param missing_compression_parents: If True the records being added are
2835
only compressed against texts already in the index (or inside
2836
records). If False the records all refer to unavailable texts (or
2837
texts inside records) as compression parents.
2839
if not self._add_callback:
2840
raise errors.ReadOnlyError(self)
2841
# we hope there are no repositories with inconsistent parentage
2845
compression_parents = set()
2846
key_dependencies = self._key_dependencies
2847
for (key, options, access_memo, parents) in records:
2849
parents = tuple(parents)
2850
if key_dependencies is not None:
2851
key_dependencies.add_references(key, parents)
2852
index, pos, size = access_memo
2853
if 'no-eol' in options:
2857
value += "%d %d" % (pos, size)
2858
if not self._deltas:
2859
if 'line-delta' in options:
2860
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2863
if 'line-delta' in options:
2864
node_refs = (parents, (parents[0],))
2865
if missing_compression_parents:
2866
compression_parents.add(parents[0])
2868
node_refs = (parents, ())
2870
node_refs = (parents, )
2873
raise KnitCorrupt(self, "attempt to add node with parents "
2874
"in parentless index.")
2876
keys[key] = (value, node_refs)
2879
present_nodes = self._get_entries(keys)
2880
for (index, key, value, node_refs) in present_nodes:
2881
parents = node_refs[:1]
2882
# Sometimes these are passed as a list rather than a tuple
2883
passed = static_tuple.as_tuples(keys[key])
2884
passed_parents = passed[1][:1]
2885
if (value[0] != keys[key][0][0] or
2886
parents != passed_parents):
2887
node_refs = static_tuple.as_tuples(node_refs)
2888
raise KnitCorrupt(self, "inconsistent details in add_records"
2889
": %s %s" % ((value, node_refs), passed))
2893
for key, (value, node_refs) in keys.iteritems():
2894
result.append((key, value, node_refs))
2896
for key, (value, node_refs) in keys.iteritems():
2897
result.append((key, value))
2898
self._add_callback(result)
2899
if missing_compression_parents:
2900
# This may appear to be incorrect (it does not check for
2901
# compression parents that are in the existing graph index),
2902
# but such records won't have been buffered, so this is
2903
# actually correct: every entry when
2904
# missing_compression_parents==True either has a missing parent, or
2905
# a parent that is one of the keys in records.
2906
compression_parents.difference_update(keys)
2907
self._missing_compression_parents.update(compression_parents)
2908
# Adding records may have satisfied missing compression parents.
2909
self._missing_compression_parents.difference_update(keys)
2911
def scan_unvalidated_index(self, graph_index):
2912
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2914
This allows this _KnitGraphIndex to keep track of any missing
2915
compression parents we may want to have filled in to make those
2918
:param graph_index: A GraphIndex
2921
new_missing = graph_index.external_references(ref_list_num=1)
2922
new_missing.difference_update(self.get_parent_map(new_missing))
2923
self._missing_compression_parents.update(new_missing)
2924
if self._key_dependencies is not None:
2925
# Add parent refs from graph_index (and discard parent refs that
2926
# the graph_index has).
2927
for node in graph_index.iter_all_entries():
2928
self._key_dependencies.add_references(node[1], node[3][0])
2930
def get_missing_compression_parents(self):
2931
"""Return the keys of missing compression parents.
2933
Missing compression parents occur when a record stream was missing
2934
basis texts, or a index was scanned that had missing basis texts.
2936
return frozenset(self._missing_compression_parents)
2938
def get_missing_parents(self):
2939
"""Return the keys of missing parents."""
2940
# If updating this, you should also update
2941
# groupcompress._GCGraphIndex.get_missing_parents
2942
# We may have false positives, so filter those out.
2943
self._key_dependencies.satisfy_refs_for_keys(
2944
self.get_parent_map(self._key_dependencies.get_unsatisfied_refs()))
2945
return frozenset(self._key_dependencies.get_unsatisfied_refs())
2947
def _check_read(self):
2948
"""raise if reads are not permitted."""
2949
if not self._is_locked():
2950
raise errors.ObjectNotLocked(self)
2952
def _check_write_ok(self):
2953
"""Assert if writes are not permitted."""
2954
if not self._is_locked():
2955
raise errors.ObjectNotLocked(self)
2957
def _compression_parent(self, an_entry):
2958
# return the key that an_entry is compressed against, or None
2959
# Grab the second parent list (as deltas implies parents currently)
2960
compression_parents = an_entry[3][1]
2961
if not compression_parents:
2963
if len(compression_parents) != 1:
2964
raise AssertionError(
2965
"Too many compression parents: %r" % compression_parents)
2966
return compression_parents[0]
2968
def get_build_details(self, keys):
2969
"""Get the method, index_memo and compression parent for version_ids.
2971
Ghosts are omitted from the result.
2973
:param keys: An iterable of keys.
2974
:return: A dict of key:
2975
(index_memo, compression_parent, parents, record_details).
2977
opaque structure to pass to read_records to extract the raw
2980
Content that this record is built upon, may be None
2982
Logical parents of this node
2984
extra information about the content which needs to be passed to
2985
Factory.parse_record
2989
entries = self._get_entries(keys, False)
2990
for entry in entries:
2992
if not self._parents:
2995
parents = entry[3][0]
2996
if not self._deltas:
2997
compression_parent_key = None
2999
compression_parent_key = self._compression_parent(entry)
3000
noeol = (entry[2][0] == 'N')
3001
if compression_parent_key:
3002
method = 'line-delta'
3005
result[key] = (self._node_to_position(entry),
3006
compression_parent_key, parents,
3010
def _get_entries(self, keys, check_present=False):
3011
"""Get the entries for keys.
3013
:param keys: An iterable of index key tuples.
3018
for node in self._graph_index.iter_entries(keys):
3020
found_keys.add(node[1])
3022
# adapt parentless index to the rest of the code.
3023
for node in self._graph_index.iter_entries(keys):
3024
yield node[0], node[1], node[2], ()
3025
found_keys.add(node[1])
3027
missing_keys = keys.difference(found_keys)
3029
raise RevisionNotPresent(missing_keys.pop(), self)
3031
def get_method(self, key):
3032
"""Return compression method of specified key."""
3033
return self._get_method(self._get_node(key))
3035
def _get_method(self, node):
3036
if not self._deltas:
3038
if self._compression_parent(node):
3043
def _get_node(self, key):
3045
return list(self._get_entries([key]))[0]
3047
raise RevisionNotPresent(key, self)
3049
def get_options(self, key):
3050
"""Return a list representing options.
3054
node = self._get_node(key)
3055
options = [self._get_method(node)]
3056
if node[2][0] == 'N':
3057
options.append('no-eol')
3060
def find_ancestry(self, keys):
3061
"""See CombinedGraphIndex.find_ancestry()"""
3062
return self._graph_index.find_ancestry(keys, 0)
3064
def get_parent_map(self, keys):
3065
"""Get a map of the parents of keys.
3067
:param keys: The keys to look up parents for.
3068
:return: A mapping from keys to parents. Absent keys are absent from
3072
nodes = self._get_entries(keys)
3076
result[node[1]] = node[3][0]
3079
result[node[1]] = None
3082
def get_position(self, key):
3083
"""Return details needed to access the version.
3085
:return: a tuple (index, data position, size) to hand to the access
3086
logic to get the record.
3088
node = self._get_node(key)
3089
return self._node_to_position(node)
3091
has_key = _mod_index._has_key_from_parent_map
3094
"""Get all the keys in the collection.
3096
The keys are not ordered.
3099
return [node[1] for node in self._graph_index.iter_all_entries()]
3101
missing_keys = _mod_index._missing_keys_from_parent_map
3103
def _node_to_position(self, node):
3104
"""Convert an index value to position details."""
3105
bits = node[2][1:].split(' ')
3106
return node[0], int(bits[0]), int(bits[1])
3108
def _sort_keys_by_io(self, keys, positions):
3109
"""Figure out an optimal order to read the records for the given keys.
3111
Sort keys, grouped by index and sorted by position.
3113
:param keys: A list of keys whose records we want to read. This will be
3115
:param positions: A dict, such as the one returned by
3116
_get_components_positions()
3119
def get_index_memo(key):
3120
# index_memo is at offset [1]. It is made up of (GraphIndex,
3121
# position, size). GI is an object, which will be unique for each
3122
# pack file. This causes us to group by pack file, then sort by
3123
# position. Size doesn't matter, but it isn't worth breaking up the
3125
return positions[key][1]
3126
return keys.sort(key=get_index_memo)
3128
_get_total_build_size = _get_total_build_size
3131
class _KnitKeyAccess(object):
3132
"""Access to records in .knit files."""
3134
def __init__(self, transport, mapper):
3135
"""Create a _KnitKeyAccess with transport and mapper.
3137
:param transport: The transport the access object is rooted at.
3138
:param mapper: The mapper used to map keys to .knit files.
3140
self._transport = transport
3141
self._mapper = mapper
3143
def add_raw_records(self, key_sizes, raw_data):
3144
"""Add raw knit bytes to a storage area.
3146
The data is spooled to the container writer in one bytes-record per
3149
:param sizes: An iterable of tuples containing the key and size of each
3151
:param raw_data: A bytestring containing the data.
3152
:return: A list of memos to retrieve the record later. Each memo is an
3153
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3154
length), where the key is the record key.
3156
if type(raw_data) is not str:
3157
raise AssertionError(
3158
'data must be plain bytes was %s' % type(raw_data))
3161
# TODO: This can be tuned for writing to sftp and other servers where
3162
# append() is relatively expensive by grouping the writes to each key
3164
for key, size in key_sizes:
3165
path = self._mapper.map(key)
3167
base = self._transport.append_bytes(path + '.knit',
3168
raw_data[offset:offset+size])
3169
except errors.NoSuchFile:
3170
self._transport.mkdir(osutils.dirname(path))
3171
base = self._transport.append_bytes(path + '.knit',
3172
raw_data[offset:offset+size])
3176
result.append((key, base, size))
3180
"""Flush pending writes on this access object.
3182
For .knit files this is a no-op.
3186
def get_raw_records(self, memos_for_retrieval):
3187
"""Get the raw bytes for a records.
3189
:param memos_for_retrieval: An iterable containing the access memo for
3190
retrieving the bytes.
3191
:return: An iterator over the bytes of the records.
3193
# first pass, group into same-index request to minimise readv's issued.
3195
current_prefix = None
3196
for (key, offset, length) in memos_for_retrieval:
3197
if current_prefix == key[:-1]:
3198
current_list.append((offset, length))
3200
if current_prefix is not None:
3201
request_lists.append((current_prefix, current_list))
3202
current_prefix = key[:-1]
3203
current_list = [(offset, length)]
3204
# handle the last entry
3205
if current_prefix is not None:
3206
request_lists.append((current_prefix, current_list))
3207
for prefix, read_vector in request_lists:
3208
path = self._mapper.map(prefix) + '.knit'
3209
for pos, data in self._transport.readv(path, read_vector):
3213
def annotate_knit(knit, revision_id):
3214
"""Annotate a knit with no cached annotations.
3216
This implementation is for knits with no cached annotations.
3217
It will work for knits with cached annotations, but this is not
3220
annotator = _KnitAnnotator(knit)
3221
return iter(annotator.annotate_flat(revision_id))
3224
class _KnitAnnotator(annotate.Annotator):
3225
"""Build up the annotations for a text."""
3227
def __init__(self, vf):
3228
annotate.Annotator.__init__(self, vf)
3230
# TODO: handle Nodes which cannot be extracted
3231
# self._ghosts = set()
3233
# Map from (key, parent_key) => matching_blocks, should be 'use once'
3234
self._matching_blocks = {}
3236
# KnitContent objects
3237
self._content_objects = {}
3238
# The number of children that depend on this fulltext content object
3239
self._num_compression_children = {}
3240
# Delta records that need their compression parent before they can be
3242
self._pending_deltas = {}
3243
# Fulltext records that are waiting for their parents fulltexts before
3244
# they can be yielded for annotation
3245
self._pending_annotation = {}
3247
self._all_build_details = {}
3249
def _get_build_graph(self, key):
3250
"""Get the graphs for building texts and annotations.
3252
The data you need for creating a full text may be different than the
3253
data you need to annotate that text. (At a minimum, you need both
3254
parents to create an annotation, but only need 1 parent to generate the
3257
:return: A list of (key, index_memo) records, suitable for
3258
passing to read_records_iter to start reading in the raw data from
3261
pending = set([key])
3264
self._num_needed_children[key] = 1
3266
# get all pending nodes
3267
this_iteration = pending
3268
build_details = self._vf._index.get_build_details(this_iteration)
3269
self._all_build_details.update(build_details)
3270
# new_nodes = self._vf._index._get_entries(this_iteration)
3272
for key, details in build_details.iteritems():
3273
(index_memo, compression_parent, parent_keys,
3274
record_details) = details
3275
self._parent_map[key] = parent_keys
3276
self._heads_provider = None
3277
records.append((key, index_memo))
3278
# Do we actually need to check _annotated_lines?
3279
pending.update([p for p in parent_keys
3280
if p not in self._all_build_details])
3282
for parent_key in parent_keys:
3283
if parent_key in self._num_needed_children:
3284
self._num_needed_children[parent_key] += 1
3286
self._num_needed_children[parent_key] = 1
3287
if compression_parent:
3288
if compression_parent in self._num_compression_children:
3289
self._num_compression_children[compression_parent] += 1
3291
self._num_compression_children[compression_parent] = 1
3293
missing_versions = this_iteration.difference(build_details.keys())
3294
if missing_versions:
3295
for key in missing_versions:
3296
if key in self._parent_map and key in self._text_cache:
3297
# We already have this text ready, we just need to
3298
# yield it later so we get it annotated
3300
parent_keys = self._parent_map[key]
3301
for parent_key in parent_keys:
3302
if parent_key in self._num_needed_children:
3303
self._num_needed_children[parent_key] += 1
3305
self._num_needed_children[parent_key] = 1
3306
pending.update([p for p in parent_keys
3307
if p not in self._all_build_details])
3309
raise errors.RevisionNotPresent(key, self._vf)
3310
# Generally we will want to read the records in reverse order, because
3311
# we find the parent nodes after the children
3313
return records, ann_keys
3315
def _get_needed_texts(self, key, pb=None):
3316
# if True or len(self._vf._immediate_fallback_vfs) > 0:
3317
if len(self._vf._immediate_fallback_vfs) > 0:
3318
# If we have fallbacks, go to the generic path
3319
for v in annotate.Annotator._get_needed_texts(self, key, pb=pb):
3324
records, ann_keys = self._get_build_graph(key)
3325
for idx, (sub_key, text, num_lines) in enumerate(
3326
self._extract_texts(records)):
3328
pb.update(gettext('annotating'), idx, len(records))
3329
yield sub_key, text, num_lines
3330
for sub_key in ann_keys:
3331
text = self._text_cache[sub_key]
3332
num_lines = len(text) # bad assumption
3333
yield sub_key, text, num_lines
3335
except errors.RetryWithNewPacks, e:
3336
self._vf._access.reload_or_raise(e)
3337
# The cached build_details are no longer valid
3338
self._all_build_details.clear()
3340
def _cache_delta_blocks(self, key, compression_parent, delta, lines):
3341
parent_lines = self._text_cache[compression_parent]
3342
blocks = list(KnitContent.get_line_delta_blocks(delta, parent_lines, lines))
3343
self._matching_blocks[(key, compression_parent)] = blocks
3345
def _expand_record(self, key, parent_keys, compression_parent, record,
3348
if compression_parent:
3349
if compression_parent not in self._content_objects:
3350
# Waiting for the parent
3351
self._pending_deltas.setdefault(compression_parent, []).append(
3352
(key, parent_keys, record, record_details))
3354
# We have the basis parent, so expand the delta
3355
num = self._num_compression_children[compression_parent]
3358
base_content = self._content_objects.pop(compression_parent)
3359
self._num_compression_children.pop(compression_parent)
3361
self._num_compression_children[compression_parent] = num
3362
base_content = self._content_objects[compression_parent]
3363
# It is tempting to want to copy_base_content=False for the last
3364
# child object. However, whenever noeol=False,
3365
# self._text_cache[parent_key] is content._lines. So mutating it
3366
# gives very bad results.
3367
# The alternative is to copy the lines into text cache, but then we
3368
# are copying anyway, so just do it here.
3369
content, delta = self._vf._factory.parse_record(
3370
key, record, record_details, base_content,
3371
copy_base_content=True)
3374
content, _ = self._vf._factory.parse_record(
3375
key, record, record_details, None)
3376
if self._num_compression_children.get(key, 0) > 0:
3377
self._content_objects[key] = content
3378
lines = content.text()
3379
self._text_cache[key] = lines
3380
if delta is not None:
3381
self._cache_delta_blocks(key, compression_parent, delta, lines)
3384
def _get_parent_annotations_and_matches(self, key, text, parent_key):
3385
"""Get the list of annotations for the parent, and the matching lines.
3387
:param text: The opaque value given by _get_needed_texts
3388
:param parent_key: The key for the parent text
3389
:return: (parent_annotations, matching_blocks)
3390
parent_annotations is a list as long as the number of lines in
3392
matching_blocks is a list of (parent_idx, text_idx, len) tuples
3393
indicating which lines match between the two texts
3395
block_key = (key, parent_key)
3396
if block_key in self._matching_blocks:
3397
blocks = self._matching_blocks.pop(block_key)
3398
parent_annotations = self._annotations_cache[parent_key]
3399
return parent_annotations, blocks
3400
return annotate.Annotator._get_parent_annotations_and_matches(self,
3401
key, text, parent_key)
3403
def _process_pending(self, key):
3404
"""The content for 'key' was just processed.
3406
Determine if there is any more pending work to be processed.
3409
if key in self._pending_deltas:
3410
compression_parent = key
3411
children = self._pending_deltas.pop(key)
3412
for child_key, parent_keys, record, record_details in children:
3413
lines = self._expand_record(child_key, parent_keys,
3415
record, record_details)
3416
if self._check_ready_for_annotations(child_key, parent_keys):
3417
to_return.append(child_key)
3418
# Also check any children that are waiting for this parent to be
3420
if key in self._pending_annotation:
3421
children = self._pending_annotation.pop(key)
3422
to_return.extend([c for c, p_keys in children
3423
if self._check_ready_for_annotations(c, p_keys)])
3426
def _check_ready_for_annotations(self, key, parent_keys):
3427
"""return true if this text is ready to be yielded.
3429
Otherwise, this will return False, and queue the text into
3430
self._pending_annotation
3432
for parent_key in parent_keys:
3433
if parent_key not in self._annotations_cache:
3434
# still waiting on at least one parent text, so queue it up
3435
# Note that if there are multiple parents, we need to wait
3437
self._pending_annotation.setdefault(parent_key,
3438
[]).append((key, parent_keys))
3442
def _extract_texts(self, records):
3443
"""Extract the various texts needed based on records"""
3444
# We iterate in the order read, rather than a strict order requested
3445
# However, process what we can, and put off to the side things that
3446
# still need parents, cleaning them up when those parents are
3449
# 1) As 'records' are read, see if we can expand these records into
3450
# Content objects (and thus lines)
3451
# 2) If a given line-delta is waiting on its compression parent, it
3452
# gets queued up into self._pending_deltas, otherwise we expand
3453
# it, and put it into self._text_cache and self._content_objects
3454
# 3) If we expanded the text, we will then check to see if all
3455
# parents have also been processed. If so, this text gets yielded,
3456
# else this record gets set aside into pending_annotation
3457
# 4) Further, if we expanded the text in (2), we will then check to
3458
# see if there are any children in self._pending_deltas waiting to
3459
# also be processed. If so, we go back to (2) for those
3460
# 5) Further again, if we yielded the text, we can then check if that
3461
# 'unlocks' any of the texts in pending_annotations, which should
3462
# then get yielded as well
3463
# Note that both steps 4 and 5 are 'recursive' in that unlocking one
3464
# compression child could unlock yet another, and yielding a fulltext
3465
# will also 'unlock' the children that are waiting on that annotation.
3466
# (Though also, unlocking 1 parent's fulltext, does not unlock a child
3467
# if other parents are also waiting.)
3468
# We want to yield content before expanding child content objects, so
3469
# that we know when we can re-use the content lines, and the annotation
3470
# code can know when it can stop caching fulltexts, as well.
3472
# Children that are missing their compression parent
3474
for (key, record, digest) in self._vf._read_records_iter(records):
3476
details = self._all_build_details[key]
3477
(_, compression_parent, parent_keys, record_details) = details
3478
lines = self._expand_record(key, parent_keys, compression_parent,
3479
record, record_details)
3481
# Pending delta should be queued up
3483
# At this point, we may be able to yield this content, if all
3484
# parents are also finished
3485
yield_this_text = self._check_ready_for_annotations(key,
3488
# All parents present
3489
yield key, lines, len(lines)
3490
to_process = self._process_pending(key)
3492
this_process = to_process
3494
for key in this_process:
3495
lines = self._text_cache[key]
3496
yield key, lines, len(lines)
3497
to_process.extend(self._process_pending(key))
3500
from bzrlib._knit_load_data_pyx import _load_data_c as _load_data
3501
except ImportError, e:
3502
osutils.failed_to_load_extension(e)
3503
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
