1
# Copyright (C) 2005, 2006, 2007, 2008 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 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
# 10:16 < lifeless> make partial index writes safe
56
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
57
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
59
# move sha1 out of the content so that join is faster at verifying parents
60
# record content length ?
63
from cStringIO import StringIO
64
from itertools import izip, chain
69
from bzrlib.lazy_import import lazy_import
70
lazy_import(globals(), """
90
from bzrlib.errors import (
98
RevisionAlreadyPresent,
101
from bzrlib.osutils import (
108
from bzrlib.versionedfile import (
109
AbsentContentFactory,
113
ChunkedContentFactory,
119
# TODO: Split out code specific to this format into an associated object.
121
# TODO: Can we put in some kind of value to check that the index and data
122
# files belong together?
124
# TODO: accommodate binaries, perhaps by storing a byte count
126
# TODO: function to check whole file
128
# TODO: atomically append data, then measure backwards from the cursor
129
# position after writing to work out where it was located. we may need to
130
# bypass python file buffering.
132
DATA_SUFFIX = '.knit'
133
INDEX_SUFFIX = '.kndx'
136
class KnitAdapter(object):
137
"""Base class for knit record adaption."""
139
def __init__(self, basis_vf):
140
"""Create an adapter which accesses full texts from basis_vf.
142
:param basis_vf: A versioned file to access basis texts of deltas from.
143
May be None for adapters that do not need to access basis texts.
145
self._data = KnitVersionedFiles(None, None)
146
self._annotate_factory = KnitAnnotateFactory()
147
self._plain_factory = KnitPlainFactory()
148
self._basis_vf = basis_vf
151
class FTAnnotatedToUnannotated(KnitAdapter):
152
"""An adapter from FT annotated knits to unannotated ones."""
154
def get_bytes(self, factory, annotated_compressed_bytes):
156
self._data._parse_record_unchecked(annotated_compressed_bytes)
157
content = self._annotate_factory.parse_fulltext(contents, rec[1])
158
size, bytes = self._data._record_to_data((rec[1],), rec[3], content.text())
162
class DeltaAnnotatedToUnannotated(KnitAdapter):
163
"""An adapter for deltas from annotated to unannotated."""
165
def get_bytes(self, factory, annotated_compressed_bytes):
167
self._data._parse_record_unchecked(annotated_compressed_bytes)
168
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
170
contents = self._plain_factory.lower_line_delta(delta)
171
size, bytes = self._data._record_to_data((rec[1],), rec[3], contents)
175
class FTAnnotatedToFullText(KnitAdapter):
176
"""An adapter from FT annotated knits to unannotated ones."""
178
def get_bytes(self, factory, annotated_compressed_bytes):
180
self._data._parse_record_unchecked(annotated_compressed_bytes)
181
content, delta = self._annotate_factory.parse_record(factory.key[-1],
182
contents, factory._build_details, None)
183
return ''.join(content.text())
186
class DeltaAnnotatedToFullText(KnitAdapter):
187
"""An adapter for deltas from annotated to unannotated."""
189
def get_bytes(self, factory, annotated_compressed_bytes):
191
self._data._parse_record_unchecked(annotated_compressed_bytes)
192
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
194
compression_parent = factory.parents[0]
195
basis_entry = self._basis_vf.get_record_stream(
196
[compression_parent], 'unordered', True).next()
197
if basis_entry.storage_kind == 'absent':
198
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
199
basis_chunks = basis_entry.get_bytes_as('chunked')
200
basis_lines = osutils.chunks_to_lines(basis_chunks)
201
# Manually apply the delta because we have one annotated content and
203
basis_content = PlainKnitContent(basis_lines, compression_parent)
204
basis_content.apply_delta(delta, rec[1])
205
basis_content._should_strip_eol = factory._build_details[1]
206
return ''.join(basis_content.text())
209
class FTPlainToFullText(KnitAdapter):
210
"""An adapter from FT plain knits to unannotated ones."""
212
def get_bytes(self, factory, compressed_bytes):
214
self._data._parse_record_unchecked(compressed_bytes)
215
content, delta = self._plain_factory.parse_record(factory.key[-1],
216
contents, factory._build_details, None)
217
return ''.join(content.text())
220
class DeltaPlainToFullText(KnitAdapter):
221
"""An adapter for deltas from annotated to unannotated."""
223
def get_bytes(self, factory, compressed_bytes):
225
self._data._parse_record_unchecked(compressed_bytes)
226
delta = self._plain_factory.parse_line_delta(contents, rec[1])
227
compression_parent = factory.parents[0]
228
# XXX: string splitting overhead.
229
basis_entry = self._basis_vf.get_record_stream(
230
[compression_parent], 'unordered', True).next()
231
if basis_entry.storage_kind == 'absent':
232
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
233
basis_chunks = basis_entry.get_bytes_as('chunked')
234
basis_lines = osutils.chunks_to_lines(basis_chunks)
235
basis_content = PlainKnitContent(basis_lines, compression_parent)
236
# Manually apply the delta because we have one annotated content and
238
content, _ = self._plain_factory.parse_record(rec[1], contents,
239
factory._build_details, basis_content)
240
return ''.join(content.text())
243
class KnitContentFactory(ContentFactory):
244
"""Content factory for streaming from knits.
246
:seealso ContentFactory:
249
def __init__(self, key, parents, build_details, sha1, raw_record,
250
annotated, knit=None):
251
"""Create a KnitContentFactory for key.
254
:param parents: The parents.
255
:param build_details: The build details as returned from
257
:param sha1: The sha1 expected from the full text of this object.
258
:param raw_record: The bytes of the knit data from disk.
259
:param annotated: True if the raw data is annotated.
261
ContentFactory.__init__(self)
264
self.parents = parents
265
if build_details[0] == 'line-delta':
270
annotated_kind = 'annotated-'
273
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
274
self._raw_record = raw_record
275
self._build_details = build_details
278
def get_bytes_as(self, storage_kind):
279
if storage_kind == self.storage_kind:
280
return self._raw_record
281
if self._knit is not None:
282
if storage_kind == 'chunked':
283
return self._knit.get_lines(self.key[0])
284
elif storage_kind == 'fulltext':
285
return self._knit.get_text(self.key[0])
286
raise errors.UnavailableRepresentation(self.key, storage_kind,
290
class KnitContent(object):
291
"""Content of a knit version to which deltas can be applied.
293
This is always stored in memory as a list of lines with \n at the end,
294
plus a flag saying if the final ending is really there or not, because that
295
corresponds to the on-disk knit representation.
299
self._should_strip_eol = False
301
def apply_delta(self, delta, new_version_id):
302
"""Apply delta to this object to become new_version_id."""
303
raise NotImplementedError(self.apply_delta)
305
def line_delta_iter(self, new_lines):
306
"""Generate line-based delta from this content to new_lines."""
307
new_texts = new_lines.text()
308
old_texts = self.text()
309
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
310
for tag, i1, i2, j1, j2 in s.get_opcodes():
313
# ofrom, oto, length, data
314
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
316
def line_delta(self, new_lines):
317
return list(self.line_delta_iter(new_lines))
320
def get_line_delta_blocks(knit_delta, source, target):
321
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
322
target_len = len(target)
325
for s_begin, s_end, t_len, new_text in knit_delta:
326
true_n = s_begin - s_pos
329
# knit deltas do not provide reliable info about whether the
330
# last line of a file matches, due to eol handling.
331
if source[s_pos + n -1] != target[t_pos + n -1]:
334
yield s_pos, t_pos, n
335
t_pos += t_len + true_n
337
n = target_len - t_pos
339
if source[s_pos + n -1] != target[t_pos + n -1]:
342
yield s_pos, t_pos, n
343
yield s_pos + (target_len - t_pos), target_len, 0
346
class AnnotatedKnitContent(KnitContent):
347
"""Annotated content."""
349
def __init__(self, lines):
350
KnitContent.__init__(self)
354
"""Return a list of (origin, text) for each content line."""
355
lines = self._lines[:]
356
if self._should_strip_eol:
357
origin, last_line = lines[-1]
358
lines[-1] = (origin, last_line.rstrip('\n'))
361
def apply_delta(self, delta, new_version_id):
362
"""Apply delta to this object to become new_version_id."""
365
for start, end, count, delta_lines in delta:
366
lines[offset+start:offset+end] = delta_lines
367
offset = offset + (start - end) + count
371
lines = [text for origin, text in self._lines]
372
except ValueError, e:
373
# most commonly (only?) caused by the internal form of the knit
374
# missing annotation information because of a bug - see thread
376
raise KnitCorrupt(self,
377
"line in annotated knit missing annotation information: %s"
379
if self._should_strip_eol:
380
lines[-1] = lines[-1].rstrip('\n')
384
return AnnotatedKnitContent(self._lines[:])
387
class PlainKnitContent(KnitContent):
388
"""Unannotated content.
390
When annotate[_iter] is called on this content, the same version is reported
391
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
395
def __init__(self, lines, version_id):
396
KnitContent.__init__(self)
398
self._version_id = version_id
401
"""Return a list of (origin, text) for each content line."""
402
return [(self._version_id, line) for line in self._lines]
404
def apply_delta(self, delta, new_version_id):
405
"""Apply delta to this object to become new_version_id."""
408
for start, end, count, delta_lines in delta:
409
lines[offset+start:offset+end] = delta_lines
410
offset = offset + (start - end) + count
411
self._version_id = new_version_id
414
return PlainKnitContent(self._lines[:], self._version_id)
418
if self._should_strip_eol:
420
lines[-1] = lines[-1].rstrip('\n')
424
class _KnitFactory(object):
425
"""Base class for common Factory functions."""
427
def parse_record(self, version_id, record, record_details,
428
base_content, copy_base_content=True):
429
"""Parse a record into a full content object.
431
:param version_id: The official version id for this content
432
:param record: The data returned by read_records_iter()
433
:param record_details: Details about the record returned by
435
:param base_content: If get_build_details returns a compression_parent,
436
you must return a base_content here, else use None
437
:param copy_base_content: When building from the base_content, decide
438
you can either copy it and return a new object, or modify it in
440
:return: (content, delta) A Content object and possibly a line-delta,
443
method, noeol = record_details
444
if method == 'line-delta':
445
if copy_base_content:
446
content = base_content.copy()
448
content = base_content
449
delta = self.parse_line_delta(record, version_id)
450
content.apply_delta(delta, version_id)
452
content = self.parse_fulltext(record, version_id)
454
content._should_strip_eol = noeol
455
return (content, delta)
458
class KnitAnnotateFactory(_KnitFactory):
459
"""Factory for creating annotated Content objects."""
463
def make(self, lines, version_id):
464
num_lines = len(lines)
465
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
467
def parse_fulltext(self, content, version_id):
468
"""Convert fulltext to internal representation
470
fulltext content is of the format
471
revid(utf8) plaintext\n
472
internal representation is of the format:
475
# TODO: jam 20070209 The tests expect this to be returned as tuples,
476
# but the code itself doesn't really depend on that.
477
# Figure out a way to not require the overhead of turning the
478
# list back into tuples.
479
lines = [tuple(line.split(' ', 1)) for line in content]
480
return AnnotatedKnitContent(lines)
482
def parse_line_delta_iter(self, lines):
483
return iter(self.parse_line_delta(lines))
485
def parse_line_delta(self, lines, version_id, plain=False):
486
"""Convert a line based delta into internal representation.
488
line delta is in the form of:
489
intstart intend intcount
491
revid(utf8) newline\n
492
internal representation is
493
(start, end, count, [1..count tuples (revid, newline)])
495
:param plain: If True, the lines are returned as a plain
496
list without annotations, not as a list of (origin, content) tuples, i.e.
497
(start, end, count, [1..count newline])
504
def cache_and_return(line):
505
origin, text = line.split(' ', 1)
506
return cache.setdefault(origin, origin), text
508
# walk through the lines parsing.
509
# Note that the plain test is explicitly pulled out of the
510
# loop to minimise any performance impact
513
start, end, count = [int(n) for n in header.split(',')]
514
contents = [next().split(' ', 1)[1] for i in xrange(count)]
515
result.append((start, end, count, contents))
518
start, end, count = [int(n) for n in header.split(',')]
519
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
520
result.append((start, end, count, contents))
523
def get_fulltext_content(self, lines):
524
"""Extract just the content lines from a fulltext."""
525
return (line.split(' ', 1)[1] for line in lines)
527
def get_linedelta_content(self, lines):
528
"""Extract just the content from a line delta.
530
This doesn't return all of the extra information stored in a delta.
531
Only the actual content lines.
536
header = header.split(',')
537
count = int(header[2])
538
for i in xrange(count):
539
origin, text = next().split(' ', 1)
542
def lower_fulltext(self, content):
543
"""convert a fulltext content record into a serializable form.
545
see parse_fulltext which this inverts.
547
# TODO: jam 20070209 We only do the caching thing to make sure that
548
# the origin is a valid utf-8 line, eventually we could remove it
549
return ['%s %s' % (o, t) for o, t in content._lines]
551
def lower_line_delta(self, delta):
552
"""convert a delta into a serializable form.
554
See parse_line_delta which this inverts.
556
# TODO: jam 20070209 We only do the caching thing to make sure that
557
# the origin is a valid utf-8 line, eventually we could remove it
559
for start, end, c, lines in delta:
560
out.append('%d,%d,%d\n' % (start, end, c))
561
out.extend(origin + ' ' + text
562
for origin, text in lines)
565
def annotate(self, knit, key):
566
content = knit._get_content(key)
567
# adjust for the fact that serialised annotations are only key suffixes
569
if type(key) == tuple:
571
origins = content.annotate()
573
for origin, line in origins:
574
result.append((prefix + (origin,), line))
577
# XXX: This smells a bit. Why would key ever be a non-tuple here?
578
# Aren't keys defined to be tuples? -- spiv 20080618
579
return content.annotate()
582
class KnitPlainFactory(_KnitFactory):
583
"""Factory for creating plain Content objects."""
587
def make(self, lines, version_id):
588
return PlainKnitContent(lines, version_id)
590
def parse_fulltext(self, content, version_id):
591
"""This parses an unannotated fulltext.
593
Note that this is not a noop - the internal representation
594
has (versionid, line) - its just a constant versionid.
596
return self.make(content, version_id)
598
def parse_line_delta_iter(self, lines, version_id):
600
num_lines = len(lines)
601
while cur < num_lines:
604
start, end, c = [int(n) for n in header.split(',')]
605
yield start, end, c, lines[cur:cur+c]
608
def parse_line_delta(self, lines, version_id):
609
return list(self.parse_line_delta_iter(lines, version_id))
611
def get_fulltext_content(self, lines):
612
"""Extract just the content lines from a fulltext."""
615
def get_linedelta_content(self, lines):
616
"""Extract just the content from a line delta.
618
This doesn't return all of the extra information stored in a delta.
619
Only the actual content lines.
624
header = header.split(',')
625
count = int(header[2])
626
for i in xrange(count):
629
def lower_fulltext(self, content):
630
return content.text()
632
def lower_line_delta(self, delta):
634
for start, end, c, lines in delta:
635
out.append('%d,%d,%d\n' % (start, end, c))
639
def annotate(self, knit, key):
640
annotator = _KnitAnnotator(knit)
641
return annotator.annotate(key)
645
def make_file_factory(annotated, mapper):
646
"""Create a factory for creating a file based KnitVersionedFiles.
648
This is only functional enough to run interface tests, it doesn't try to
649
provide a full pack environment.
651
:param annotated: knit annotations are wanted.
652
:param mapper: The mapper from keys to paths.
654
def factory(transport):
655
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
656
access = _KnitKeyAccess(transport, mapper)
657
return KnitVersionedFiles(index, access, annotated=annotated)
661
def make_pack_factory(graph, delta, keylength):
662
"""Create a factory for creating a pack based VersionedFiles.
664
This is only functional enough to run interface tests, it doesn't try to
665
provide a full pack environment.
667
:param graph: Store a graph.
668
:param delta: Delta compress contents.
669
:param keylength: How long should keys be.
671
def factory(transport):
672
parents = graph or delta
678
max_delta_chain = 200
681
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
682
key_elements=keylength)
683
stream = transport.open_write_stream('newpack')
684
writer = pack.ContainerWriter(stream.write)
686
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
687
deltas=delta, add_callback=graph_index.add_nodes)
688
access = _DirectPackAccess({})
689
access.set_writer(writer, graph_index, (transport, 'newpack'))
690
result = KnitVersionedFiles(index, access,
691
max_delta_chain=max_delta_chain)
692
result.stream = stream
693
result.writer = writer
698
def cleanup_pack_knit(versioned_files):
699
versioned_files.stream.close()
700
versioned_files.writer.end()
703
class KnitVersionedFiles(VersionedFiles):
704
"""Storage for many versioned files using knit compression.
706
Backend storage is managed by indices and data objects.
708
:ivar _index: A _KnitGraphIndex or similar that can describe the
709
parents, graph, compression and data location of entries in this
710
KnitVersionedFiles. Note that this is only the index for
711
*this* vfs; if there are fallbacks they must be queried separately.
714
def __init__(self, index, data_access, max_delta_chain=200,
715
annotated=False, reload_func=None):
716
"""Create a KnitVersionedFiles with index and data_access.
718
:param index: The index for the knit data.
719
:param data_access: The access object to store and retrieve knit
721
:param max_delta_chain: The maximum number of deltas to permit during
722
insertion. Set to 0 to prohibit the use of deltas.
723
:param annotated: Set to True to cause annotations to be calculated and
724
stored during insertion.
725
:param reload_func: An function that can be called if we think we need
726
to reload the pack listing and try again. See
727
'bzrlib.repofmt.pack_repo.AggregateIndex' for the signature.
730
self._access = data_access
731
self._max_delta_chain = max_delta_chain
733
self._factory = KnitAnnotateFactory()
735
self._factory = KnitPlainFactory()
736
self._fallback_vfs = []
737
self._reload_func = reload_func
740
return "%s(%r, %r)" % (
741
self.__class__.__name__,
745
def add_fallback_versioned_files(self, a_versioned_files):
746
"""Add a source of texts for texts not present in this knit.
748
:param a_versioned_files: A VersionedFiles object.
750
self._fallback_vfs.append(a_versioned_files)
752
def add_lines(self, key, parents, lines, parent_texts=None,
753
left_matching_blocks=None, nostore_sha=None, random_id=False,
755
"""See VersionedFiles.add_lines()."""
756
self._index._check_write_ok()
757
self._check_add(key, lines, random_id, check_content)
759
# The caller might pass None if there is no graph data, but kndx
760
# indexes can't directly store that, so we give them
761
# an empty tuple instead.
763
return self._add(key, lines, parents,
764
parent_texts, left_matching_blocks, nostore_sha, random_id)
766
def _add(self, key, lines, parents, parent_texts,
767
left_matching_blocks, nostore_sha, random_id):
768
"""Add a set of lines on top of version specified by parents.
770
Any versions not present will be converted into ghosts.
772
# first thing, if the content is something we don't need to store, find
774
line_bytes = ''.join(lines)
775
digest = sha_string(line_bytes)
776
if nostore_sha == digest:
777
raise errors.ExistingContent
780
if parent_texts is None:
782
# Do a single query to ascertain parent presence; we only compress
783
# against parents in the same kvf.
784
present_parent_map = self._index.get_parent_map(parents)
785
for parent in parents:
786
if parent in present_parent_map:
787
present_parents.append(parent)
789
# Currently we can only compress against the left most present parent.
790
if (len(present_parents) == 0 or
791
present_parents[0] != parents[0]):
794
# To speed the extract of texts the delta chain is limited
795
# to a fixed number of deltas. This should minimize both
796
# I/O and the time spend applying deltas.
797
delta = self._check_should_delta(present_parents[0])
799
text_length = len(line_bytes)
802
if lines[-1][-1] != '\n':
803
# copy the contents of lines.
805
options.append('no-eol')
806
lines[-1] = lines[-1] + '\n'
810
if type(element) != str:
811
raise TypeError("key contains non-strings: %r" % (key,))
812
# Knit hunks are still last-element only
814
content = self._factory.make(lines, version_id)
815
if 'no-eol' in options:
816
# Hint to the content object that its text() call should strip the
818
content._should_strip_eol = True
819
if delta or (self._factory.annotated and len(present_parents) > 0):
820
# Merge annotations from parent texts if needed.
821
delta_hunks = self._merge_annotations(content, present_parents,
822
parent_texts, delta, self._factory.annotated,
823
left_matching_blocks)
826
options.append('line-delta')
827
store_lines = self._factory.lower_line_delta(delta_hunks)
828
size, bytes = self._record_to_data(key, digest,
831
options.append('fulltext')
832
# isinstance is slower and we have no hierarchy.
833
if self._factory.__class__ == KnitPlainFactory:
834
# Use the already joined bytes saving iteration time in
836
size, bytes = self._record_to_data(key, digest,
839
# get mixed annotation + content and feed it into the
841
store_lines = self._factory.lower_fulltext(content)
842
size, bytes = self._record_to_data(key, digest,
845
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
846
self._index.add_records(
847
((key, options, access_memo, parents),),
849
return digest, text_length, content
851
def annotate(self, key):
852
"""See VersionedFiles.annotate."""
853
return self._factory.annotate(self, key)
855
def check(self, progress_bar=None):
856
"""See VersionedFiles.check()."""
857
# This doesn't actually test extraction of everything, but that will
858
# impact 'bzr check' substantially, and needs to be integrated with
859
# care. However, it does check for the obvious problem of a delta with
861
keys = self._index.keys()
862
parent_map = self.get_parent_map(keys)
864
if self._index.get_method(key) != 'fulltext':
865
compression_parent = parent_map[key][0]
866
if compression_parent not in parent_map:
867
raise errors.KnitCorrupt(self,
868
"Missing basis parent %s for %s" % (
869
compression_parent, key))
870
for fallback_vfs in self._fallback_vfs:
873
def _check_add(self, key, lines, random_id, check_content):
874
"""check that version_id and lines are safe to add."""
876
if contains_whitespace(version_id):
877
raise InvalidRevisionId(version_id, self)
878
self.check_not_reserved_id(version_id)
879
# TODO: If random_id==False and the key is already present, we should
880
# probably check that the existing content is identical to what is
881
# being inserted, and otherwise raise an exception. This would make
882
# the bundle code simpler.
884
self._check_lines_not_unicode(lines)
885
self._check_lines_are_lines(lines)
887
def _check_header(self, key, line):
888
rec = self._split_header(line)
889
self._check_header_version(rec, key[-1])
892
def _check_header_version(self, rec, version_id):
893
"""Checks the header version on original format knit records.
895
These have the last component of the key embedded in the record.
897
if rec[1] != version_id:
898
raise KnitCorrupt(self,
899
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
901
def _check_should_delta(self, parent):
902
"""Iterate back through the parent listing, looking for a fulltext.
904
This is used when we want to decide whether to add a delta or a new
905
fulltext. It searches for _max_delta_chain parents. When it finds a
906
fulltext parent, it sees if the total size of the deltas leading up to
907
it is large enough to indicate that we want a new full text anyway.
909
Return True if we should create a new delta, False if we should use a
914
for count in xrange(self._max_delta_chain):
915
# XXX: Collapse these two queries:
917
# Note that this only looks in the index of this particular
918
# KnitVersionedFiles, not in the fallbacks. This ensures that
919
# we won't store a delta spanning physical repository
921
method = self._index.get_method(parent)
922
except RevisionNotPresent:
923
# Some basis is not locally present: always delta
925
index, pos, size = self._index.get_position(parent)
926
if method == 'fulltext':
930
# We don't explicitly check for presence because this is in an
931
# inner loop, and if it's missing it'll fail anyhow.
932
# TODO: This should be asking for compression parent, not graph
934
parent = self._index.get_parent_map([parent])[parent][0]
936
# We couldn't find a fulltext, so we must create a new one
938
# Simple heuristic - if the total I/O wold be greater as a delta than
939
# the originally installed fulltext, we create a new fulltext.
940
return fulltext_size > delta_size
942
def _build_details_to_components(self, build_details):
943
"""Convert a build_details tuple to a position tuple."""
944
# record_details, access_memo, compression_parent
945
return build_details[3], build_details[0], build_details[1]
947
def _get_components_positions(self, keys, allow_missing=False):
948
"""Produce a map of position data for the components of keys.
950
This data is intended to be used for retrieving the knit records.
952
A dict of key to (record_details, index_memo, next, parents) is
954
method is the way referenced data should be applied.
955
index_memo is the handle to pass to the data access to actually get the
957
next is the build-parent of the version, or None for fulltexts.
958
parents is the version_ids of the parents of this version
960
:param allow_missing: If True do not raise an error on a missing component,
964
pending_components = keys
965
while pending_components:
966
build_details = self._index.get_build_details(pending_components)
967
current_components = set(pending_components)
968
pending_components = set()
969
for key, details in build_details.iteritems():
970
(index_memo, compression_parent, parents,
971
record_details) = details
972
method = record_details[0]
973
if compression_parent is not None:
974
pending_components.add(compression_parent)
975
component_data[key] = self._build_details_to_components(details)
976
missing = current_components.difference(build_details)
977
if missing and not allow_missing:
978
raise errors.RevisionNotPresent(missing.pop(), self)
979
return component_data
981
def _get_content(self, key, parent_texts={}):
982
"""Returns a content object that makes up the specified
984
cached_version = parent_texts.get(key, None)
985
if cached_version is not None:
986
# Ensure the cache dict is valid.
987
if not self.get_parent_map([key]):
988
raise RevisionNotPresent(key, self)
989
return cached_version
990
text_map, contents_map = self._get_content_maps([key])
991
return contents_map[key]
993
def _get_content_maps(self, keys, nonlocal_keys=None):
994
"""Produce maps of text and KnitContents
996
:param keys: The keys to produce content maps for.
997
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
998
which are known to not be in this knit, but rather in one of the
1000
:return: (text_map, content_map) where text_map contains the texts for
1001
the requested versions and content_map contains the KnitContents.
1003
# FUTURE: This function could be improved for the 'extract many' case
1004
# by tracking each component and only doing the copy when the number of
1005
# children than need to apply delta's to it is > 1 or it is part of the
1008
multiple_versions = len(keys) != 1
1009
record_map = self._get_record_map(keys, allow_missing=True)
1014
if nonlocal_keys is None:
1015
nonlocal_keys = set()
1017
nonlocal_keys = frozenset(nonlocal_keys)
1018
missing_keys = set(nonlocal_keys)
1019
for source in self._fallback_vfs:
1020
if not missing_keys:
1022
for record in source.get_record_stream(missing_keys,
1024
if record.storage_kind == 'absent':
1026
missing_keys.remove(record.key)
1027
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
1028
text_map[record.key] = lines
1029
content_map[record.key] = PlainKnitContent(lines, record.key)
1030
if record.key in keys:
1031
final_content[record.key] = content_map[record.key]
1033
if key in nonlocal_keys:
1038
while cursor is not None:
1040
record, record_details, digest, next = record_map[cursor]
1042
raise RevisionNotPresent(cursor, self)
1043
components.append((cursor, record, record_details, digest))
1045
if cursor in content_map:
1046
# no need to plan further back
1047
components.append((cursor, None, None, None))
1051
for (component_id, record, record_details,
1052
digest) in reversed(components):
1053
if component_id in content_map:
1054
content = content_map[component_id]
1056
content, delta = self._factory.parse_record(key[-1],
1057
record, record_details, content,
1058
copy_base_content=multiple_versions)
1059
if multiple_versions:
1060
content_map[component_id] = content
1062
final_content[key] = content
1064
# digest here is the digest from the last applied component.
1065
text = content.text()
1066
actual_sha = sha_strings(text)
1067
if actual_sha != digest:
1068
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
1069
text_map[key] = text
1070
return text_map, final_content
1072
def get_parent_map(self, keys):
1073
"""Get a map of the graph parents of keys.
1075
:param keys: The keys to look up parents for.
1076
:return: A mapping from keys to parents. Absent keys are absent from
1079
return self._get_parent_map_with_sources(keys)[0]
1081
def _get_parent_map_with_sources(self, keys):
1082
"""Get a map of the parents of keys.
1084
:param keys: The keys to look up parents for.
1085
:return: A tuple. The first element is a mapping from keys to parents.
1086
Absent keys are absent from the mapping. The second element is a
1087
list with the locations each key was found in. The first element
1088
is the in-this-knit parents, the second the first fallback source,
1092
sources = [self._index] + self._fallback_vfs
1095
for source in sources:
1098
new_result = source.get_parent_map(missing)
1099
source_results.append(new_result)
1100
result.update(new_result)
1101
missing.difference_update(set(new_result))
1102
return result, source_results
1104
def _get_record_map(self, keys, allow_missing=False):
1105
"""Produce a dictionary of knit records.
1107
:return: {key:(record, record_details, digest, next)}
1109
data returned from read_records
1111
opaque information to pass to parse_record
1113
SHA1 digest of the full text after all steps are done
1115
build-parent of the version, i.e. the leftmost ancestor.
1116
Will be None if the record is not a delta.
1117
:param keys: The keys to build a map for
1118
:param allow_missing: If some records are missing, rather than
1119
error, just return the data that could be generated.
1121
# This retries the whole request if anything fails. Potentially we
1122
# could be a bit more selective. We could track the keys whose records
1123
# we have successfully found, and then only request the new records
1124
# from there. However, _get_components_positions grabs the whole build
1125
# chain, which means we'll likely try to grab the same records again
1126
# anyway. Also, can the build chains change as part of a pack
1127
# operation? We wouldn't want to end up with a broken chain.
1130
position_map = self._get_components_positions(keys,
1131
allow_missing=allow_missing)
1132
# key = component_id, r = record_details, i_m = index_memo,
1134
records = [(key, i_m) for key, (r, i_m, n)
1135
in position_map.iteritems()]
1137
for key, record, digest in self._read_records_iter(records):
1138
(record_details, index_memo, next) = position_map[key]
1139
record_map[key] = record, record_details, digest, next
1141
except errors.RetryWithNewPacks, e:
1142
self._access.reload_or_raise(e)
1144
def _split_by_prefix(self, keys):
1145
"""For the given keys, split them up based on their prefix.
1147
To keep memory pressure somewhat under control, split the
1148
requests back into per-file-id requests, otherwise "bzr co"
1149
extracts the full tree into memory before writing it to disk.
1150
This should be revisited if _get_content_maps() can ever cross
1153
:param keys: An iterable of key tuples
1154
:return: A dict of {prefix: [key_list]}
1156
split_by_prefix = {}
1159
split_by_prefix.setdefault('', []).append(key)
1161
split_by_prefix.setdefault(key[0], []).append(key)
1162
return split_by_prefix
1164
def get_record_stream(self, keys, ordering, include_delta_closure):
1165
"""Get a stream of records for keys.
1167
:param keys: The keys to include.
1168
:param ordering: Either 'unordered' or 'topological'. A topologically
1169
sorted stream has compression parents strictly before their
1171
:param include_delta_closure: If True then the closure across any
1172
compression parents will be included (in the opaque data).
1173
:return: An iterator of ContentFactory objects, each of which is only
1174
valid until the iterator is advanced.
1176
# keys might be a generator
1180
if not self._index.has_graph:
1181
# Cannot topological order when no graph has been stored.
1182
ordering = 'unordered'
1184
remaining_keys = keys
1187
keys = set(remaining_keys)
1188
for content_factory in self._get_remaining_record_stream(keys,
1189
ordering, include_delta_closure):
1190
remaining_keys.discard(content_factory.key)
1191
yield content_factory
1193
except errors.RetryWithNewPacks, e:
1194
self._access.reload_or_raise(e)
1196
def _get_remaining_record_stream(self, keys, ordering,
1197
include_delta_closure):
1198
"""This function is the 'retry' portion for get_record_stream."""
1199
if include_delta_closure:
1200
positions = self._get_components_positions(keys, allow_missing=True)
1202
build_details = self._index.get_build_details(keys)
1204
# (record_details, access_memo, compression_parent_key)
1205
positions = dict((key, self._build_details_to_components(details))
1206
for key, details in build_details.iteritems())
1207
absent_keys = keys.difference(set(positions))
1208
# There may be more absent keys : if we're missing the basis component
1209
# and are trying to include the delta closure.
1210
if include_delta_closure:
1211
needed_from_fallback = set()
1212
# Build up reconstructable_keys dict. key:True in this dict means
1213
# the key can be reconstructed.
1214
reconstructable_keys = {}
1218
chain = [key, positions[key][2]]
1220
needed_from_fallback.add(key)
1223
while chain[-1] is not None:
1224
if chain[-1] in reconstructable_keys:
1225
result = reconstructable_keys[chain[-1]]
1229
chain.append(positions[chain[-1]][2])
1231
# missing basis component
1232
needed_from_fallback.add(chain[-1])
1235
for chain_key in chain[:-1]:
1236
reconstructable_keys[chain_key] = result
1238
needed_from_fallback.add(key)
1239
# Double index lookups here : need a unified api ?
1240
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1241
if ordering == 'topological':
1242
# Global topological sort
1243
present_keys = tsort.topo_sort(global_map)
1244
# Now group by source:
1246
current_source = None
1247
for key in present_keys:
1248
for parent_map in parent_maps:
1249
if key in parent_map:
1250
key_source = parent_map
1252
if current_source is not key_source:
1253
source_keys.append((key_source, []))
1254
current_source = key_source
1255
source_keys[-1][1].append(key)
1257
if ordering != 'unordered':
1258
raise AssertionError('valid values for ordering are:'
1259
' "unordered" or "topological" not: %r'
1261
# Just group by source; remote sources first.
1264
for parent_map in reversed(parent_maps):
1265
source_keys.append((parent_map, []))
1266
for key in parent_map:
1267
present_keys.append(key)
1268
source_keys[-1][1].append(key)
1269
# We have been requested to return these records in an order that
1270
# suits us. So we ask the index to give us an optimally sorted
1272
for source, sub_keys in source_keys:
1273
if source is parent_maps[0]:
1274
# Only sort the keys for this VF
1275
self._index._sort_keys_by_io(sub_keys, positions)
1276
absent_keys = keys - set(global_map)
1277
for key in absent_keys:
1278
yield AbsentContentFactory(key)
1279
# restrict our view to the keys we can answer.
1280
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1281
# XXX: At that point we need to consider the impact of double reads by
1282
# utilising components multiple times.
1283
if include_delta_closure:
1284
# XXX: get_content_maps performs its own index queries; allow state
1286
non_local_keys = needed_from_fallback - absent_keys
1287
prefix_split_keys = self._split_by_prefix(present_keys)
1288
prefix_split_non_local_keys = self._split_by_prefix(non_local_keys)
1289
for prefix, keys in prefix_split_keys.iteritems():
1290
non_local = prefix_split_non_local_keys.get(prefix, [])
1291
non_local = set(non_local)
1292
text_map, _ = self._get_content_maps(keys, non_local)
1294
lines = text_map.pop(key)
1295
yield ChunkedContentFactory(key, global_map[key], None,
1298
for source, keys in source_keys:
1299
if source is parent_maps[0]:
1300
# this KnitVersionedFiles
1301
records = [(key, positions[key][1]) for key in keys]
1302
for key, raw_data, sha1 in self._read_records_iter_raw(records):
1303
(record_details, index_memo, _) = positions[key]
1304
yield KnitContentFactory(key, global_map[key],
1305
record_details, sha1, raw_data, self._factory.annotated, None)
1307
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1308
for record in vf.get_record_stream(keys, ordering,
1309
include_delta_closure):
1312
def get_sha1s(self, keys):
1313
"""See VersionedFiles.get_sha1s()."""
1315
record_map = self._get_record_map(missing, allow_missing=True)
1317
for key, details in record_map.iteritems():
1318
if key not in missing:
1320
# record entry 2 is the 'digest'.
1321
result[key] = details[2]
1322
missing.difference_update(set(result))
1323
for source in self._fallback_vfs:
1326
new_result = source.get_sha1s(missing)
1327
result.update(new_result)
1328
missing.difference_update(set(new_result))
1331
def insert_record_stream(self, stream):
1332
"""Insert a record stream into this container.
1334
:param stream: A stream of records to insert.
1336
:seealso VersionedFiles.get_record_stream:
1338
def get_adapter(adapter_key):
1340
return adapters[adapter_key]
1342
adapter_factory = adapter_registry.get(adapter_key)
1343
adapter = adapter_factory(self)
1344
adapters[adapter_key] = adapter
1347
if self._factory.annotated:
1348
# self is annotated, we need annotated knits to use directly.
1349
annotated = "annotated-"
1352
# self is not annotated, but we can strip annotations cheaply.
1354
convertibles = set(["knit-annotated-ft-gz"])
1355
if self._max_delta_chain:
1356
delta_types.add("knit-annotated-delta-gz")
1357
convertibles.add("knit-annotated-delta-gz")
1358
# The set of types we can cheaply adapt without needing basis texts.
1359
native_types = set()
1360
if self._max_delta_chain:
1361
native_types.add("knit-%sdelta-gz" % annotated)
1362
delta_types.add("knit-%sdelta-gz" % annotated)
1363
native_types.add("knit-%sft-gz" % annotated)
1364
knit_types = native_types.union(convertibles)
1366
# Buffer all index entries that we can't add immediately because their
1367
# basis parent is missing. We don't buffer all because generating
1368
# annotations may require access to some of the new records. However we
1369
# can't generate annotations from new deltas until their basis parent
1370
# is present anyway, so we get away with not needing an index that
1371
# includes the new keys.
1373
# See <http://launchpad.net/bugs/300177> about ordering of compression
1374
# parents in the records - to be conservative, we insist that all
1375
# parents must be present to avoid expanding to a fulltext.
1377
# key = basis_parent, value = index entry to add
1378
buffered_index_entries = {}
1379
for record in stream:
1380
parents = record.parents
1381
# trace.mutter('inserting record %s (kind: %s, parents: %r)',
1382
# record.key, record.storage_kind, parents)
1383
if record.storage_kind in delta_types:
1384
# TODO: eventually the record itself should track
1385
# compression_parent
1386
compression_parent = parents[0]
1388
compression_parent = None
1389
# Raise an error when a record is missing.
1390
if record.storage_kind == 'absent':
1391
raise RevisionNotPresent([record.key], self)
1392
elif ((record.storage_kind in knit_types)
1393
and (compression_parent is None
1394
or not self._fallback_vfs
1395
or self._index.has_key(compression_parent)
1396
or not self.has_key(compression_parent))):
1397
# we can insert the knit record literally if either it has no
1398
# compression parent OR we already have its basis in this kvf
1399
# OR the basis is not present even in the fallbacks. In the
1400
# last case it will either turn up later in the stream and all
1401
# will be well, or it won't turn up at all and we'll raise an
1404
# TODO: self.has_key is somewhat redundant with
1405
# self._index.has_key; we really want something that directly
1406
# asks if it's only present in the fallbacks. -- mbp 20081119
1407
if record.storage_kind not in native_types:
1409
adapter_key = (record.storage_kind, "knit-delta-gz")
1410
adapter = get_adapter(adapter_key)
1412
adapter_key = (record.storage_kind, "knit-ft-gz")
1413
adapter = get_adapter(adapter_key)
1414
bytes = adapter.get_bytes(
1415
record, record.get_bytes_as(record.storage_kind))
1417
bytes = record.get_bytes_as(record.storage_kind)
1418
options = [record._build_details[0]]
1419
if record._build_details[1]:
1420
options.append('no-eol')
1421
# Just blat it across.
1422
# Note: This does end up adding data on duplicate keys. As
1423
# modern repositories use atomic insertions this should not
1424
# lead to excessive growth in the event of interrupted fetches.
1425
# 'knit' repositories may suffer excessive growth, but as a
1426
# deprecated format this is tolerable. It can be fixed if
1427
# needed by in the kndx index support raising on a duplicate
1428
# add with identical parents and options.
1429
access_memo = self._access.add_raw_records(
1430
[(record.key, len(bytes))], bytes)[0]
1431
index_entry = (record.key, options, access_memo, parents)
1433
if 'fulltext' not in options:
1434
# Not a fulltext, so we need to make sure the compression
1435
# parent will also be present.
1436
# Note that pack backed knits don't need to buffer here
1437
# because they buffer all writes to the transaction level,
1438
# but we don't expose that difference at the index level. If
1439
# the query here has sufficient cost to show up in
1440
# profiling we should do that.
1442
# They're required to be physically in this
1443
# KnitVersionedFiles, not in a fallback.
1444
if not self._index.has_key(compression_parent):
1445
pending = buffered_index_entries.setdefault(
1446
compression_parent, [])
1447
pending.append(index_entry)
1450
self._index.add_records([index_entry])
1451
elif record.storage_kind == 'chunked':
1452
self.add_lines(record.key, parents,
1453
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1454
elif record.storage_kind == 'fulltext':
1455
self.add_lines(record.key, parents,
1456
split_lines(record.get_bytes_as('fulltext')))
1458
# Not a fulltext, and not suitable for direct insertion as a
1459
# delta, either because it's not the right format, or this
1460
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1461
# 0) or because it depends on a base only present in the
1463
adapter_key = record.storage_kind, 'fulltext'
1464
adapter = get_adapter(adapter_key)
1465
lines = split_lines(adapter.get_bytes(
1466
record, record.get_bytes_as(record.storage_kind)))
1468
self.add_lines(record.key, parents, lines)
1469
except errors.RevisionAlreadyPresent:
1471
# Add any records whose basis parent is now available.
1472
added_keys = [record.key]
1474
key = added_keys.pop(0)
1475
if key in buffered_index_entries:
1476
index_entries = buffered_index_entries[key]
1477
self._index.add_records(index_entries)
1479
[index_entry[0] for index_entry in index_entries])
1480
del buffered_index_entries[key]
1481
# If there were any deltas which had a missing basis parent, error.
1482
if buffered_index_entries:
1483
from pprint import pformat
1484
raise errors.BzrCheckError(
1485
"record_stream refers to compression parents not in %r:\n%s"
1486
% (self, pformat(sorted(buffered_index_entries.keys()))))
1488
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1489
"""Iterate over the lines in the versioned files from keys.
1491
This may return lines from other keys. Each item the returned
1492
iterator yields is a tuple of a line and a text version that that line
1493
is present in (not introduced in).
1495
Ordering of results is in whatever order is most suitable for the
1496
underlying storage format.
1498
If a progress bar is supplied, it may be used to indicate progress.
1499
The caller is responsible for cleaning up progress bars (because this
1503
* Lines are normalised by the underlying store: they will all have \\n
1505
* Lines are returned in arbitrary order.
1506
* If a requested key did not change any lines (or didn't have any
1507
lines), it may not be mentioned at all in the result.
1509
:return: An iterator over (line, key).
1512
pb = progress.DummyProgress()
1518
# we don't care about inclusions, the caller cares.
1519
# but we need to setup a list of records to visit.
1520
# we need key, position, length
1522
build_details = self._index.get_build_details(keys)
1523
for key, details in build_details.iteritems():
1525
key_records.append((key, details[0]))
1526
records_iter = enumerate(self._read_records_iter(key_records))
1527
for (key_idx, (key, data, sha_value)) in records_iter:
1528
pb.update('Walking content.', key_idx, total)
1529
compression_parent = build_details[key][1]
1530
if compression_parent is None:
1532
line_iterator = self._factory.get_fulltext_content(data)
1535
line_iterator = self._factory.get_linedelta_content(data)
1536
# Now that we are yielding the data for this key, remove it
1539
# XXX: It might be more efficient to yield (key,
1540
# line_iterator) in the future. However for now, this is a
1541
# simpler change to integrate into the rest of the
1542
# codebase. RBC 20071110
1543
for line in line_iterator:
1546
except errors.RetryWithNewPacks, e:
1547
self._access.reload_or_raise(e)
1548
# If there are still keys we've not yet found, we look in the fallback
1549
# vfs, and hope to find them there. Note that if the keys are found
1550
# but had no changes or no content, the fallback may not return
1552
if keys and not self._fallback_vfs:
1553
# XXX: strictly the second parameter is meant to be the file id
1554
# but it's not easily accessible here.
1555
raise RevisionNotPresent(keys, repr(self))
1556
for source in self._fallback_vfs:
1560
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1561
source_keys.add(key)
1563
keys.difference_update(source_keys)
1564
pb.update('Walking content.', total, total)
1566
def _make_line_delta(self, delta_seq, new_content):
1567
"""Generate a line delta from delta_seq and new_content."""
1569
for op in delta_seq.get_opcodes():
1570
if op[0] == 'equal':
1572
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1575
def _merge_annotations(self, content, parents, parent_texts={},
1576
delta=None, annotated=None,
1577
left_matching_blocks=None):
1578
"""Merge annotations for content and generate deltas.
1580
This is done by comparing the annotations based on changes to the text
1581
and generating a delta on the resulting full texts. If annotations are
1582
not being created then a simple delta is created.
1584
if left_matching_blocks is not None:
1585
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1589
for parent_key in parents:
1590
merge_content = self._get_content(parent_key, parent_texts)
1591
if (parent_key == parents[0] and delta_seq is not None):
1594
seq = patiencediff.PatienceSequenceMatcher(
1595
None, merge_content.text(), content.text())
1596
for i, j, n in seq.get_matching_blocks():
1599
# this copies (origin, text) pairs across to the new
1600
# content for any line that matches the last-checked
1602
content._lines[j:j+n] = merge_content._lines[i:i+n]
1603
# XXX: Robert says the following block is a workaround for a
1604
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1605
if content._lines and content._lines[-1][1][-1] != '\n':
1606
# The copied annotation was from a line without a trailing EOL,
1607
# reinstate one for the content object, to ensure correct
1609
line = content._lines[-1][1] + '\n'
1610
content._lines[-1] = (content._lines[-1][0], line)
1612
if delta_seq is None:
1613
reference_content = self._get_content(parents[0], parent_texts)
1614
new_texts = content.text()
1615
old_texts = reference_content.text()
1616
delta_seq = patiencediff.PatienceSequenceMatcher(
1617
None, old_texts, new_texts)
1618
return self._make_line_delta(delta_seq, content)
1620
def _parse_record(self, version_id, data):
1621
"""Parse an original format knit record.
1623
These have the last element of the key only present in the stored data.
1625
rec, record_contents = self._parse_record_unchecked(data)
1626
self._check_header_version(rec, version_id)
1627
return record_contents, rec[3]
1629
def _parse_record_header(self, key, raw_data):
1630
"""Parse a record header for consistency.
1632
:return: the header and the decompressor stream.
1633
as (stream, header_record)
1635
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1638
rec = self._check_header(key, df.readline())
1639
except Exception, e:
1640
raise KnitCorrupt(self,
1641
"While reading {%s} got %s(%s)"
1642
% (key, e.__class__.__name__, str(e)))
1645
def _parse_record_unchecked(self, data):
1647
# 4168 calls in 2880 217 internal
1648
# 4168 calls to _parse_record_header in 2121
1649
# 4168 calls to readlines in 330
1650
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1652
record_contents = df.readlines()
1653
except Exception, e:
1654
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1655
(data, e.__class__.__name__, str(e)))
1656
header = record_contents.pop(0)
1657
rec = self._split_header(header)
1658
last_line = record_contents.pop()
1659
if len(record_contents) != int(rec[2]):
1660
raise KnitCorrupt(self,
1661
'incorrect number of lines %s != %s'
1662
' for version {%s} %s'
1663
% (len(record_contents), int(rec[2]),
1664
rec[1], record_contents))
1665
if last_line != 'end %s\n' % rec[1]:
1666
raise KnitCorrupt(self,
1667
'unexpected version end line %r, wanted %r'
1668
% (last_line, rec[1]))
1670
return rec, record_contents
1672
def _read_records_iter(self, records):
1673
"""Read text records from data file and yield result.
1675
The result will be returned in whatever is the fastest to read.
1676
Not by the order requested. Also, multiple requests for the same
1677
record will only yield 1 response.
1678
:param records: A list of (key, access_memo) entries
1679
:return: Yields (key, contents, digest) in the order
1680
read, not the order requested
1685
# XXX: This smells wrong, IO may not be getting ordered right.
1686
needed_records = sorted(set(records), key=operator.itemgetter(1))
1687
if not needed_records:
1690
# The transport optimizes the fetching as well
1691
# (ie, reads continuous ranges.)
1692
raw_data = self._access.get_raw_records(
1693
[index_memo for key, index_memo in needed_records])
1695
for (key, index_memo), data in \
1696
izip(iter(needed_records), raw_data):
1697
content, digest = self._parse_record(key[-1], data)
1698
yield key, content, digest
1700
def _read_records_iter_raw(self, records):
1701
"""Read text records from data file and yield raw data.
1703
This unpacks enough of the text record to validate the id is
1704
as expected but thats all.
1706
Each item the iterator yields is (key, bytes, sha1_of_full_text).
1708
# setup an iterator of the external records:
1709
# uses readv so nice and fast we hope.
1711
# grab the disk data needed.
1712
needed_offsets = [index_memo for key, index_memo
1714
raw_records = self._access.get_raw_records(needed_offsets)
1716
for key, index_memo in records:
1717
data = raw_records.next()
1718
# validate the header (note that we can only use the suffix in
1719
# current knit records).
1720
df, rec = self._parse_record_header(key, data)
1722
yield key, data, rec[3]
1724
def _record_to_data(self, key, digest, lines, dense_lines=None):
1725
"""Convert key, digest, lines into a raw data block.
1727
:param key: The key of the record. Currently keys are always serialised
1728
using just the trailing component.
1729
:param dense_lines: The bytes of lines but in a denser form. For
1730
instance, if lines is a list of 1000 bytestrings each ending in \n,
1731
dense_lines may be a list with one line in it, containing all the
1732
1000's lines and their \n's. Using dense_lines if it is already
1733
known is a win because the string join to create bytes in this
1734
function spends less time resizing the final string.
1735
:return: (len, a StringIO instance with the raw data ready to read.)
1737
# Note: using a string copy here increases memory pressure with e.g.
1738
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
1739
# when doing the initial commit of a mozilla tree. RBC 20070921
1740
bytes = ''.join(chain(
1741
["version %s %d %s\n" % (key[-1],
1744
dense_lines or lines,
1745
["end %s\n" % key[-1]]))
1746
if type(bytes) != str:
1747
raise AssertionError(
1748
'data must be plain bytes was %s' % type(bytes))
1749
if lines and lines[-1][-1] != '\n':
1750
raise ValueError('corrupt lines value %r' % lines)
1751
compressed_bytes = tuned_gzip.bytes_to_gzip(bytes)
1752
return len(compressed_bytes), compressed_bytes
1754
def _split_header(self, line):
1757
raise KnitCorrupt(self,
1758
'unexpected number of elements in record header')
1762
"""See VersionedFiles.keys."""
1763
if 'evil' in debug.debug_flags:
1764
trace.mutter_callsite(2, "keys scales with size of history")
1765
sources = [self._index] + self._fallback_vfs
1767
for source in sources:
1768
result.update(source.keys())
1772
class _KndxIndex(object):
1773
"""Manages knit index files
1775
The index is kept in memory and read on startup, to enable
1776
fast lookups of revision information. The cursor of the index
1777
file is always pointing to the end, making it easy to append
1780
_cache is a cache for fast mapping from version id to a Index
1783
_history is a cache for fast mapping from indexes to version ids.
1785
The index data format is dictionary compressed when it comes to
1786
parent references; a index entry may only have parents that with a
1787
lover index number. As a result, the index is topological sorted.
1789
Duplicate entries may be written to the index for a single version id
1790
if this is done then the latter one completely replaces the former:
1791
this allows updates to correct version and parent information.
1792
Note that the two entries may share the delta, and that successive
1793
annotations and references MUST point to the first entry.
1795
The index file on disc contains a header, followed by one line per knit
1796
record. The same revision can be present in an index file more than once.
1797
The first occurrence gets assigned a sequence number starting from 0.
1799
The format of a single line is
1800
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1801
REVISION_ID is a utf8-encoded revision id
1802
FLAGS is a comma separated list of flags about the record. Values include
1803
no-eol, line-delta, fulltext.
1804
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1805
that the the compressed data starts at.
1806
LENGTH is the ascii representation of the length of the data file.
1807
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1809
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1810
revision id already in the knit that is a parent of REVISION_ID.
1811
The ' :' marker is the end of record marker.
1814
when a write is interrupted to the index file, it will result in a line
1815
that does not end in ' :'. If the ' :' is not present at the end of a line,
1816
or at the end of the file, then the record that is missing it will be
1817
ignored by the parser.
1819
When writing new records to the index file, the data is preceded by '\n'
1820
to ensure that records always start on new lines even if the last write was
1821
interrupted. As a result its normal for the last line in the index to be
1822
missing a trailing newline. One can be added with no harmful effects.
1824
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
1825
where prefix is e.g. the (fileid,) for .texts instances or () for
1826
constant-mapped things like .revisions, and the old state is
1827
tuple(history_vector, cache_dict). This is used to prevent having an
1828
ABI change with the C extension that reads .kndx files.
1831
HEADER = "# bzr knit index 8\n"
1833
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
1834
"""Create a _KndxIndex on transport using mapper."""
1835
self._transport = transport
1836
self._mapper = mapper
1837
self._get_scope = get_scope
1838
self._allow_writes = allow_writes
1839
self._is_locked = is_locked
1841
self.has_graph = True
1843
def add_records(self, records, random_id=False):
1844
"""Add multiple records to the index.
1846
:param records: a list of tuples:
1847
(key, options, access_memo, parents).
1848
:param random_id: If True the ids being added were randomly generated
1849
and no check for existence will be performed.
1852
for record in records:
1855
path = self._mapper.map(key) + '.kndx'
1856
path_keys = paths.setdefault(path, (prefix, []))
1857
path_keys[1].append(record)
1858
for path in sorted(paths):
1859
prefix, path_keys = paths[path]
1860
self._load_prefixes([prefix])
1862
orig_history = self._kndx_cache[prefix][1][:]
1863
orig_cache = self._kndx_cache[prefix][0].copy()
1866
for key, options, (_, pos, size), parents in path_keys:
1868
# kndx indices cannot be parentless.
1870
line = "\n%s %s %s %s %s :" % (
1871
key[-1], ','.join(options), pos, size,
1872
self._dictionary_compress(parents))
1873
if type(line) != str:
1874
raise AssertionError(
1875
'data must be utf8 was %s' % type(line))
1877
self._cache_key(key, options, pos, size, parents)
1878
if len(orig_history):
1879
self._transport.append_bytes(path, ''.join(lines))
1881
self._init_index(path, lines)
1883
# If any problems happen, restore the original values and re-raise
1884
self._kndx_cache[prefix] = (orig_cache, orig_history)
1887
def _cache_key(self, key, options, pos, size, parent_keys):
1888
"""Cache a version record in the history array and index cache.
1890
This is inlined into _load_data for performance. KEEP IN SYNC.
1891
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1895
version_id = key[-1]
1896
# last-element only for compatibilty with the C load_data.
1897
parents = tuple(parent[-1] for parent in parent_keys)
1898
for parent in parent_keys:
1899
if parent[:-1] != prefix:
1900
raise ValueError("mismatched prefixes for %r, %r" % (
1902
cache, history = self._kndx_cache[prefix]
1903
# only want the _history index to reference the 1st index entry
1905
if version_id not in cache:
1906
index = len(history)
1907
history.append(version_id)
1909
index = cache[version_id][5]
1910
cache[version_id] = (version_id,
1917
def check_header(self, fp):
1918
line = fp.readline()
1920
# An empty file can actually be treated as though the file doesn't
1922
raise errors.NoSuchFile(self)
1923
if line != self.HEADER:
1924
raise KnitHeaderError(badline=line, filename=self)
1926
def _check_read(self):
1927
if not self._is_locked():
1928
raise errors.ObjectNotLocked(self)
1929
if self._get_scope() != self._scope:
1932
def _check_write_ok(self):
1933
"""Assert if not writes are permitted."""
1934
if not self._is_locked():
1935
raise errors.ObjectNotLocked(self)
1936
if self._get_scope() != self._scope:
1938
if self._mode != 'w':
1939
raise errors.ReadOnlyObjectDirtiedError(self)
1941
def get_build_details(self, keys):
1942
"""Get the method, index_memo and compression parent for keys.
1944
Ghosts are omitted from the result.
1946
:param keys: An iterable of keys.
1947
:return: A dict of key:(index_memo, compression_parent, parents,
1950
opaque structure to pass to read_records to extract the raw
1953
Content that this record is built upon, may be None
1955
Logical parents of this node
1957
extra information about the content which needs to be passed to
1958
Factory.parse_record
1960
parent_map = self.get_parent_map(keys)
1963
if key not in parent_map:
1965
method = self.get_method(key)
1966
parents = parent_map[key]
1967
if method == 'fulltext':
1968
compression_parent = None
1970
compression_parent = parents[0]
1971
noeol = 'no-eol' in self.get_options(key)
1972
index_memo = self.get_position(key)
1973
result[key] = (index_memo, compression_parent,
1974
parents, (method, noeol))
1977
def get_method(self, key):
1978
"""Return compression method of specified key."""
1979
options = self.get_options(key)
1980
if 'fulltext' in options:
1982
elif 'line-delta' in options:
1985
raise errors.KnitIndexUnknownMethod(self, options)
1987
def get_options(self, key):
1988
"""Return a list representing options.
1992
prefix, suffix = self._split_key(key)
1993
self._load_prefixes([prefix])
1995
return self._kndx_cache[prefix][0][suffix][1]
1997
raise RevisionNotPresent(key, self)
1999
def get_parent_map(self, keys):
2000
"""Get a map of the parents of keys.
2002
:param keys: The keys to look up parents for.
2003
:return: A mapping from keys to parents. Absent keys are absent from
2006
# Parse what we need to up front, this potentially trades off I/O
2007
# locality (.kndx and .knit in the same block group for the same file
2008
# id) for less checking in inner loops.
2009
prefixes = set(key[:-1] for key in keys)
2010
self._load_prefixes(prefixes)
2015
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2019
result[key] = tuple(prefix + (suffix,) for
2020
suffix in suffix_parents)
2023
def get_position(self, key):
2024
"""Return details needed to access the version.
2026
:return: a tuple (key, data position, size) to hand to the access
2027
logic to get the record.
2029
prefix, suffix = self._split_key(key)
2030
self._load_prefixes([prefix])
2031
entry = self._kndx_cache[prefix][0][suffix]
2032
return key, entry[2], entry[3]
2034
has_key = _mod_index._has_key_from_parent_map
2036
def _init_index(self, path, extra_lines=[]):
2037
"""Initialize an index."""
2039
sio.write(self.HEADER)
2040
sio.writelines(extra_lines)
2042
self._transport.put_file_non_atomic(path, sio,
2043
create_parent_dir=True)
2044
# self._create_parent_dir)
2045
# mode=self._file_mode,
2046
# dir_mode=self._dir_mode)
2049
"""Get all the keys in the collection.
2051
The keys are not ordered.
2054
# Identify all key prefixes.
2055
# XXX: A bit hacky, needs polish.
2056
if type(self._mapper) == ConstantMapper:
2060
for quoted_relpath in self._transport.iter_files_recursive():
2061
path, ext = os.path.splitext(quoted_relpath)
2063
prefixes = [self._mapper.unmap(path) for path in relpaths]
2064
self._load_prefixes(prefixes)
2065
for prefix in prefixes:
2066
for suffix in self._kndx_cache[prefix][1]:
2067
result.add(prefix + (suffix,))
2070
def _load_prefixes(self, prefixes):
2071
"""Load the indices for prefixes."""
2073
for prefix in prefixes:
2074
if prefix not in self._kndx_cache:
2075
# the load_data interface writes to these variables.
2078
self._filename = prefix
2080
path = self._mapper.map(prefix) + '.kndx'
2081
fp = self._transport.get(path)
2083
# _load_data may raise NoSuchFile if the target knit is
2085
_load_data(self, fp)
2088
self._kndx_cache[prefix] = (self._cache, self._history)
2093
self._kndx_cache[prefix] = ({}, [])
2094
if type(self._mapper) == ConstantMapper:
2095
# preserve behaviour for revisions.kndx etc.
2096
self._init_index(path)
2101
missing_keys = _mod_index._missing_keys_from_parent_map
2103
def _partition_keys(self, keys):
2104
"""Turn keys into a dict of prefix:suffix_list."""
2107
prefix_keys = result.setdefault(key[:-1], [])
2108
prefix_keys.append(key[-1])
2111
def _dictionary_compress(self, keys):
2112
"""Dictionary compress keys.
2114
:param keys: The keys to generate references to.
2115
:return: A string representation of keys. keys which are present are
2116
dictionary compressed, and others are emitted as fulltext with a
2122
prefix = keys[0][:-1]
2123
cache = self._kndx_cache[prefix][0]
2125
if key[:-1] != prefix:
2126
# kndx indices cannot refer across partitioned storage.
2127
raise ValueError("mismatched prefixes for %r" % keys)
2128
if key[-1] in cache:
2129
# -- inlined lookup() --
2130
result_list.append(str(cache[key[-1]][5]))
2131
# -- end lookup () --
2133
result_list.append('.' + key[-1])
2134
return ' '.join(result_list)
2136
def _reset_cache(self):
2137
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2138
# (cache_dict, history_vector) for parsed kndx files.
2139
self._kndx_cache = {}
2140
self._scope = self._get_scope()
2141
allow_writes = self._allow_writes()
2147
def _sort_keys_by_io(self, keys, positions):
2148
"""Figure out an optimal order to read the records for the given keys.
2150
Sort keys, grouped by index and sorted by position.
2152
:param keys: A list of keys whose records we want to read. This will be
2154
:param positions: A dict, such as the one returned by
2155
_get_components_positions()
2158
def get_sort_key(key):
2159
index_memo = positions[key][1]
2160
# Group by prefix and position. index_memo[0] is the key, so it is
2161
# (file_id, revision_id) and we don't want to sort on revision_id,
2162
# index_memo[1] is the position, and index_memo[2] is the size,
2163
# which doesn't matter for the sort
2164
return index_memo[0][:-1], index_memo[1]
2165
return keys.sort(key=get_sort_key)
2167
def _split_key(self, key):
2168
"""Split key into a prefix and suffix."""
2169
return key[:-1], key[-1]
2172
class _KnitGraphIndex(object):
2173
"""A KnitVersionedFiles index layered on GraphIndex."""
2175
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2177
"""Construct a KnitGraphIndex on a graph_index.
2179
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2180
:param is_locked: A callback to check whether the object should answer
2182
:param deltas: Allow delta-compressed records.
2183
:param parents: If True, record knits parents, if not do not record
2185
:param add_callback: If not None, allow additions to the index and call
2186
this callback with a list of added GraphIndex nodes:
2187
[(node, value, node_refs), ...]
2188
:param is_locked: A callback, returns True if the index is locked and
2191
self._add_callback = add_callback
2192
self._graph_index = graph_index
2193
self._deltas = deltas
2194
self._parents = parents
2195
if deltas and not parents:
2196
# XXX: TODO: Delta tree and parent graph should be conceptually
2198
raise KnitCorrupt(self, "Cannot do delta compression without "
2200
self.has_graph = parents
2201
self._is_locked = is_locked
2204
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2206
def add_records(self, records, random_id=False):
2207
"""Add multiple records to the index.
2209
This function does not insert data into the Immutable GraphIndex
2210
backing the KnitGraphIndex, instead it prepares data for insertion by
2211
the caller and checks that it is safe to insert then calls
2212
self._add_callback with the prepared GraphIndex nodes.
2214
:param records: a list of tuples:
2215
(key, options, access_memo, parents).
2216
:param random_id: If True the ids being added were randomly generated
2217
and no check for existence will be performed.
2219
if not self._add_callback:
2220
raise errors.ReadOnlyError(self)
2221
# we hope there are no repositories with inconsistent parentage
2225
for (key, options, access_memo, parents) in records:
2227
parents = tuple(parents)
2228
index, pos, size = access_memo
2229
if 'no-eol' in options:
2233
value += "%d %d" % (pos, size)
2234
if not self._deltas:
2235
if 'line-delta' in options:
2236
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2239
if 'line-delta' in options:
2240
node_refs = (parents, (parents[0],))
2242
node_refs = (parents, ())
2244
node_refs = (parents, )
2247
raise KnitCorrupt(self, "attempt to add node with parents "
2248
"in parentless index.")
2250
keys[key] = (value, node_refs)
2253
present_nodes = self._get_entries(keys)
2254
for (index, key, value, node_refs) in present_nodes:
2255
if (value[0] != keys[key][0][0] or
2256
node_refs != keys[key][1]):
2257
raise KnitCorrupt(self, "inconsistent details in add_records"
2258
": %s %s" % ((value, node_refs), keys[key]))
2262
for key, (value, node_refs) in keys.iteritems():
2263
result.append((key, value, node_refs))
2265
for key, (value, node_refs) in keys.iteritems():
2266
result.append((key, value))
2267
self._add_callback(result)
2269
def _check_read(self):
2270
"""raise if reads are not permitted."""
2271
if not self._is_locked():
2272
raise errors.ObjectNotLocked(self)
2274
def _check_write_ok(self):
2275
"""Assert if writes are not permitted."""
2276
if not self._is_locked():
2277
raise errors.ObjectNotLocked(self)
2279
def _compression_parent(self, an_entry):
2280
# return the key that an_entry is compressed against, or None
2281
# Grab the second parent list (as deltas implies parents currently)
2282
compression_parents = an_entry[3][1]
2283
if not compression_parents:
2285
if len(compression_parents) != 1:
2286
raise AssertionError(
2287
"Too many compression parents: %r" % compression_parents)
2288
return compression_parents[0]
2290
def get_build_details(self, keys):
2291
"""Get the method, index_memo and compression parent for version_ids.
2293
Ghosts are omitted from the result.
2295
:param keys: An iterable of keys.
2296
:return: A dict of key:
2297
(index_memo, compression_parent, parents, record_details).
2299
opaque structure to pass to read_records to extract the raw
2302
Content that this record is built upon, may be None
2304
Logical parents of this node
2306
extra information about the content which needs to be passed to
2307
Factory.parse_record
2311
entries = self._get_entries(keys, False)
2312
for entry in entries:
2314
if not self._parents:
2317
parents = entry[3][0]
2318
if not self._deltas:
2319
compression_parent_key = None
2321
compression_parent_key = self._compression_parent(entry)
2322
noeol = (entry[2][0] == 'N')
2323
if compression_parent_key:
2324
method = 'line-delta'
2327
result[key] = (self._node_to_position(entry),
2328
compression_parent_key, parents,
2332
def _get_entries(self, keys, check_present=False):
2333
"""Get the entries for keys.
2335
:param keys: An iterable of index key tuples.
2340
for node in self._graph_index.iter_entries(keys):
2342
found_keys.add(node[1])
2344
# adapt parentless index to the rest of the code.
2345
for node in self._graph_index.iter_entries(keys):
2346
yield node[0], node[1], node[2], ()
2347
found_keys.add(node[1])
2349
missing_keys = keys.difference(found_keys)
2351
raise RevisionNotPresent(missing_keys.pop(), self)
2353
def get_method(self, key):
2354
"""Return compression method of specified key."""
2355
return self._get_method(self._get_node(key))
2357
def _get_method(self, node):
2358
if not self._deltas:
2360
if self._compression_parent(node):
2365
def _get_node(self, key):
2367
return list(self._get_entries([key]))[0]
2369
raise RevisionNotPresent(key, self)
2371
def get_options(self, key):
2372
"""Return a list representing options.
2376
node = self._get_node(key)
2377
options = [self._get_method(node)]
2378
if node[2][0] == 'N':
2379
options.append('no-eol')
2382
def get_parent_map(self, keys):
2383
"""Get a map of the parents of keys.
2385
:param keys: The keys to look up parents for.
2386
:return: A mapping from keys to parents. Absent keys are absent from
2390
nodes = self._get_entries(keys)
2394
result[node[1]] = node[3][0]
2397
result[node[1]] = None
2400
def get_position(self, key):
2401
"""Return details needed to access the version.
2403
:return: a tuple (index, data position, size) to hand to the access
2404
logic to get the record.
2406
node = self._get_node(key)
2407
return self._node_to_position(node)
2409
has_key = _mod_index._has_key_from_parent_map
2412
"""Get all the keys in the collection.
2414
The keys are not ordered.
2417
return [node[1] for node in self._graph_index.iter_all_entries()]
2419
missing_keys = _mod_index._missing_keys_from_parent_map
2421
def _node_to_position(self, node):
2422
"""Convert an index value to position details."""
2423
bits = node[2][1:].split(' ')
2424
return node[0], int(bits[0]), int(bits[1])
2426
def _sort_keys_by_io(self, keys, positions):
2427
"""Figure out an optimal order to read the records for the given keys.
2429
Sort keys, grouped by index and sorted by position.
2431
:param keys: A list of keys whose records we want to read. This will be
2433
:param positions: A dict, such as the one returned by
2434
_get_components_positions()
2437
def get_index_memo(key):
2438
# index_memo is at offset [1]. It is made up of (GraphIndex,
2439
# position, size). GI is an object, which will be unique for each
2440
# pack file. This causes us to group by pack file, then sort by
2441
# position. Size doesn't matter, but it isn't worth breaking up the
2443
return positions[key][1]
2444
return keys.sort(key=get_index_memo)
2447
class _KnitKeyAccess(object):
2448
"""Access to records in .knit files."""
2450
def __init__(self, transport, mapper):
2451
"""Create a _KnitKeyAccess with transport and mapper.
2453
:param transport: The transport the access object is rooted at.
2454
:param mapper: The mapper used to map keys to .knit files.
2456
self._transport = transport
2457
self._mapper = mapper
2459
def add_raw_records(self, key_sizes, raw_data):
2460
"""Add raw knit bytes to a storage area.
2462
The data is spooled to the container writer in one bytes-record per
2465
:param sizes: An iterable of tuples containing the key and size of each
2467
:param raw_data: A bytestring containing the data.
2468
:return: A list of memos to retrieve the record later. Each memo is an
2469
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
2470
length), where the key is the record key.
2472
if type(raw_data) != str:
2473
raise AssertionError(
2474
'data must be plain bytes was %s' % type(raw_data))
2477
# TODO: This can be tuned for writing to sftp and other servers where
2478
# append() is relatively expensive by grouping the writes to each key
2480
for key, size in key_sizes:
2481
path = self._mapper.map(key)
2483
base = self._transport.append_bytes(path + '.knit',
2484
raw_data[offset:offset+size])
2485
except errors.NoSuchFile:
2486
self._transport.mkdir(osutils.dirname(path))
2487
base = self._transport.append_bytes(path + '.knit',
2488
raw_data[offset:offset+size])
2492
result.append((key, base, size))
2495
def get_raw_records(self, memos_for_retrieval):
2496
"""Get the raw bytes for a records.
2498
:param memos_for_retrieval: An iterable containing the access memo for
2499
retrieving the bytes.
2500
:return: An iterator over the bytes of the records.
2502
# first pass, group into same-index request to minimise readv's issued.
2504
current_prefix = None
2505
for (key, offset, length) in memos_for_retrieval:
2506
if current_prefix == key[:-1]:
2507
current_list.append((offset, length))
2509
if current_prefix is not None:
2510
request_lists.append((current_prefix, current_list))
2511
current_prefix = key[:-1]
2512
current_list = [(offset, length)]
2513
# handle the last entry
2514
if current_prefix is not None:
2515
request_lists.append((current_prefix, current_list))
2516
for prefix, read_vector in request_lists:
2517
path = self._mapper.map(prefix) + '.knit'
2518
for pos, data in self._transport.readv(path, read_vector):
2522
class _DirectPackAccess(object):
2523
"""Access to data in one or more packs with less translation."""
2525
def __init__(self, index_to_packs, reload_func=None):
2526
"""Create a _DirectPackAccess object.
2528
:param index_to_packs: A dict mapping index objects to the transport
2529
and file names for obtaining data.
2530
:param reload_func: A function to call if we determine that the pack
2531
files have moved and we need to reload our caches. See
2532
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
2534
self._container_writer = None
2535
self._write_index = None
2536
self._indices = index_to_packs
2537
self._reload_func = reload_func
2539
def add_raw_records(self, key_sizes, raw_data):
2540
"""Add raw knit bytes to a storage area.
2542
The data is spooled to the container writer in one bytes-record per
2545
:param sizes: An iterable of tuples containing the key and size of each
2547
:param raw_data: A bytestring containing the data.
2548
:return: A list of memos to retrieve the record later. Each memo is an
2549
opaque index memo. For _DirectPackAccess the memo is (index, pos,
2550
length), where the index field is the write_index object supplied
2551
to the PackAccess object.
2553
if type(raw_data) != str:
2554
raise AssertionError(
2555
'data must be plain bytes was %s' % type(raw_data))
2558
for key, size in key_sizes:
2559
p_offset, p_length = self._container_writer.add_bytes_record(
2560
raw_data[offset:offset+size], [])
2562
result.append((self._write_index, p_offset, p_length))
2565
def get_raw_records(self, memos_for_retrieval):
2566
"""Get the raw bytes for a records.
2568
:param memos_for_retrieval: An iterable containing the (index, pos,
2569
length) memo for retrieving the bytes. The Pack access method
2570
looks up the pack to use for a given record in its index_to_pack
2572
:return: An iterator over the bytes of the records.
2574
# first pass, group into same-index requests
2576
current_index = None
2577
for (index, offset, length) in memos_for_retrieval:
2578
if current_index == index:
2579
current_list.append((offset, length))
2581
if current_index is not None:
2582
request_lists.append((current_index, current_list))
2583
current_index = index
2584
current_list = [(offset, length)]
2585
# handle the last entry
2586
if current_index is not None:
2587
request_lists.append((current_index, current_list))
2588
for index, offsets in request_lists:
2590
transport, path = self._indices[index]
2592
# A KeyError here indicates that someone has triggered an index
2593
# reload, and this index has gone missing, we need to start
2595
if self._reload_func is None:
2596
# If we don't have a _reload_func there is nothing that can
2599
raise errors.RetryWithNewPacks(index,
2600
reload_occurred=True,
2601
exc_info=sys.exc_info())
2603
reader = pack.make_readv_reader(transport, path, offsets)
2604
for names, read_func in reader.iter_records():
2605
yield read_func(None)
2606
except errors.NoSuchFile:
2607
# A NoSuchFile error indicates that a pack file has gone
2608
# missing on disk, we need to trigger a reload, and start over.
2609
if self._reload_func is None:
2611
raise errors.RetryWithNewPacks(transport.abspath(path),
2612
reload_occurred=False,
2613
exc_info=sys.exc_info())
2615
def set_writer(self, writer, index, transport_packname):
2616
"""Set a writer to use for adding data."""
2617
if index is not None:
2618
self._indices[index] = transport_packname
2619
self._container_writer = writer
2620
self._write_index = index
2622
def reload_or_raise(self, retry_exc):
2623
"""Try calling the reload function, or re-raise the original exception.
2625
This should be called after _DirectPackAccess raises a
2626
RetryWithNewPacks exception. This function will handle the common logic
2627
of determining when the error is fatal versus being temporary.
2628
It will also make sure that the original exception is raised, rather
2629
than the RetryWithNewPacks exception.
2631
If this function returns, then the calling function should retry
2632
whatever operation was being performed. Otherwise an exception will
2635
:param retry_exc: A RetryWithNewPacks exception.
2638
if self._reload_func is None:
2640
elif not self._reload_func():
2641
# The reload claimed that nothing changed
2642
if not retry_exc.reload_occurred:
2643
# If there wasn't an earlier reload, then we really were
2644
# expecting to find changes. We didn't find them, so this is a
2648
exc_class, exc_value, exc_traceback = retry_exc.exc_info
2649
raise exc_class, exc_value, exc_traceback
2652
# Deprecated, use PatienceSequenceMatcher instead
2653
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2656
def annotate_knit(knit, revision_id):
2657
"""Annotate a knit with no cached annotations.
2659
This implementation is for knits with no cached annotations.
2660
It will work for knits with cached annotations, but this is not
2663
annotator = _KnitAnnotator(knit)
2664
return iter(annotator.annotate(revision_id))
2667
class _KnitAnnotator(object):
2668
"""Build up the annotations for a text."""
2670
def __init__(self, knit):
2673
# Content objects, differs from fulltexts because of how final newlines
2674
# are treated by knits. the content objects here will always have a
2676
self._fulltext_contents = {}
2678
# Annotated lines of specific revisions
2679
self._annotated_lines = {}
2681
# Track the raw data for nodes that we could not process yet.
2682
# This maps the revision_id of the base to a list of children that will
2683
# annotated from it.
2684
self._pending_children = {}
2686
# Nodes which cannot be extracted
2687
self._ghosts = set()
2689
# Track how many children this node has, so we know if we need to keep
2691
self._annotate_children = {}
2692
self._compression_children = {}
2694
self._all_build_details = {}
2695
# The children => parent revision_id graph
2696
self._revision_id_graph = {}
2698
self._heads_provider = None
2700
self._nodes_to_keep_annotations = set()
2701
self._generations_until_keep = 100
2703
def set_generations_until_keep(self, value):
2704
"""Set the number of generations before caching a node.
2706
Setting this to -1 will cache every merge node, setting this higher
2707
will cache fewer nodes.
2709
self._generations_until_keep = value
2711
def _add_fulltext_content(self, revision_id, content_obj):
2712
self._fulltext_contents[revision_id] = content_obj
2713
# TODO: jam 20080305 It might be good to check the sha1digest here
2714
return content_obj.text()
2716
def _check_parents(self, child, nodes_to_annotate):
2717
"""Check if all parents have been processed.
2719
:param child: A tuple of (rev_id, parents, raw_content)
2720
:param nodes_to_annotate: If child is ready, add it to
2721
nodes_to_annotate, otherwise put it back in self._pending_children
2723
for parent_id in child[1]:
2724
if (parent_id not in self._annotated_lines):
2725
# This parent is present, but another parent is missing
2726
self._pending_children.setdefault(parent_id,
2730
# This one is ready to be processed
2731
nodes_to_annotate.append(child)
2733
def _add_annotation(self, revision_id, fulltext, parent_ids,
2734
left_matching_blocks=None):
2735
"""Add an annotation entry.
2737
All parents should already have been annotated.
2738
:return: A list of children that now have their parents satisfied.
2740
a = self._annotated_lines
2741
annotated_parent_lines = [a[p] for p in parent_ids]
2742
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2743
fulltext, revision_id, left_matching_blocks,
2744
heads_provider=self._get_heads_provider()))
2745
self._annotated_lines[revision_id] = annotated_lines
2746
for p in parent_ids:
2747
ann_children = self._annotate_children[p]
2748
ann_children.remove(revision_id)
2749
if (not ann_children
2750
and p not in self._nodes_to_keep_annotations):
2751
del self._annotated_lines[p]
2752
del self._all_build_details[p]
2753
if p in self._fulltext_contents:
2754
del self._fulltext_contents[p]
2755
# Now that we've added this one, see if there are any pending
2756
# deltas to be done, certainly this parent is finished
2757
nodes_to_annotate = []
2758
for child in self._pending_children.pop(revision_id, []):
2759
self._check_parents(child, nodes_to_annotate)
2760
return nodes_to_annotate
2762
def _get_build_graph(self, key):
2763
"""Get the graphs for building texts and annotations.
2765
The data you need for creating a full text may be different than the
2766
data you need to annotate that text. (At a minimum, you need both
2767
parents to create an annotation, but only need 1 parent to generate the
2770
:return: A list of (key, index_memo) records, suitable for
2771
passing to read_records_iter to start reading in the raw data fro/
2774
if key in self._annotated_lines:
2777
pending = set([key])
2782
# get all pending nodes
2784
this_iteration = pending
2785
build_details = self._knit._index.get_build_details(this_iteration)
2786
self._all_build_details.update(build_details)
2787
# new_nodes = self._knit._index._get_entries(this_iteration)
2789
for key, details in build_details.iteritems():
2790
(index_memo, compression_parent, parents,
2791
record_details) = details
2792
self._revision_id_graph[key] = parents
2793
records.append((key, index_memo))
2794
# Do we actually need to check _annotated_lines?
2795
pending.update(p for p in parents
2796
if p not in self._all_build_details)
2797
if compression_parent:
2798
self._compression_children.setdefault(compression_parent,
2801
for parent in parents:
2802
self._annotate_children.setdefault(parent,
2804
num_gens = generation - kept_generation
2805
if ((num_gens >= self._generations_until_keep)
2806
and len(parents) > 1):
2807
kept_generation = generation
2808
self._nodes_to_keep_annotations.add(key)
2810
missing_versions = this_iteration.difference(build_details.keys())
2811
self._ghosts.update(missing_versions)
2812
for missing_version in missing_versions:
2813
# add a key, no parents
2814
self._revision_id_graph[missing_version] = ()
2815
pending.discard(missing_version) # don't look for it
2816
if self._ghosts.intersection(self._compression_children):
2818
"We cannot have nodes which have a ghost compression parent:\n"
2820
"compression children: %r"
2821
% (self._ghosts, self._compression_children))
2822
# Cleanout anything that depends on a ghost so that we don't wait for
2823
# the ghost to show up
2824
for node in self._ghosts:
2825
if node in self._annotate_children:
2826
# We won't be building this node
2827
del self._annotate_children[node]
2828
# Generally we will want to read the records in reverse order, because
2829
# we find the parent nodes after the children
2833
def _annotate_records(self, records):
2834
"""Build the annotations for the listed records."""
2835
# We iterate in the order read, rather than a strict order requested
2836
# However, process what we can, and put off to the side things that
2837
# still need parents, cleaning them up when those parents are
2839
for (rev_id, record,
2840
digest) in self._knit._read_records_iter(records):
2841
if rev_id in self._annotated_lines:
2843
parent_ids = self._revision_id_graph[rev_id]
2844
parent_ids = [p for p in parent_ids if p not in self._ghosts]
2845
details = self._all_build_details[rev_id]
2846
(index_memo, compression_parent, parents,
2847
record_details) = details
2848
nodes_to_annotate = []
2849
# TODO: Remove the punning between compression parents, and
2850
# parent_ids, we should be able to do this without assuming
2852
if len(parent_ids) == 0:
2853
# There are no parents for this node, so just add it
2854
# TODO: This probably needs to be decoupled
2855
fulltext_content, delta = self._knit._factory.parse_record(
2856
rev_id, record, record_details, None)
2857
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
2858
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
2859
parent_ids, left_matching_blocks=None))
2861
child = (rev_id, parent_ids, record)
2862
# Check if all the parents are present
2863
self._check_parents(child, nodes_to_annotate)
2864
while nodes_to_annotate:
2865
# Should we use a queue here instead of a stack?
2866
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
2867
(index_memo, compression_parent, parents,
2868
record_details) = self._all_build_details[rev_id]
2870
if compression_parent is not None:
2871
comp_children = self._compression_children[compression_parent]
2872
if rev_id not in comp_children:
2873
raise AssertionError("%r not in compression children %r"
2874
% (rev_id, comp_children))
2875
# If there is only 1 child, it is safe to reuse this
2877
reuse_content = (len(comp_children) == 1
2878
and compression_parent not in
2879
self._nodes_to_keep_annotations)
2881
# Remove it from the cache since it will be changing
2882
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
2883
# Make sure to copy the fulltext since it might be
2885
parent_fulltext = list(parent_fulltext_content.text())
2887
parent_fulltext_content = self._fulltext_contents[compression_parent]
2888
parent_fulltext = parent_fulltext_content.text()
2889
comp_children.remove(rev_id)
2890
fulltext_content, delta = self._knit._factory.parse_record(
2891
rev_id, record, record_details,
2892
parent_fulltext_content,
2893
copy_base_content=(not reuse_content))
2894
fulltext = self._add_fulltext_content(rev_id,
2896
if compression_parent == parent_ids[0]:
2897
# the compression_parent is the left parent, so we can
2899
blocks = KnitContent.get_line_delta_blocks(delta,
2900
parent_fulltext, fulltext)
2902
fulltext_content = self._knit._factory.parse_fulltext(
2904
fulltext = self._add_fulltext_content(rev_id,
2906
nodes_to_annotate.extend(
2907
self._add_annotation(rev_id, fulltext, parent_ids,
2908
left_matching_blocks=blocks))
2910
def _get_heads_provider(self):
2911
"""Create a heads provider for resolving ancestry issues."""
2912
if self._heads_provider is not None:
2913
return self._heads_provider
2914
parent_provider = _mod_graph.DictParentsProvider(
2915
self._revision_id_graph)
2916
graph_obj = _mod_graph.Graph(parent_provider)
2917
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
2918
self._heads_provider = head_cache
2921
def annotate(self, key):
2922
"""Return the annotated fulltext at the given key.
2924
:param key: The key to annotate.
2926
if len(self._knit._fallback_vfs) > 0:
2927
# stacked knits can't use the fast path at present.
2928
return self._simple_annotate(key)
2931
records = self._get_build_graph(key)
2932
if key in self._ghosts:
2933
raise errors.RevisionNotPresent(key, self._knit)
2934
self._annotate_records(records)
2935
return self._annotated_lines[key]
2936
except errors.RetryWithNewPacks, e:
2937
self._knit._access.reload_or_raise(e)
2938
# The cached build_details are no longer valid
2939
self._all_build_details.clear()
2941
def _simple_annotate(self, key):
2942
"""Return annotated fulltext, rediffing from the full texts.
2944
This is slow but makes no assumptions about the repository
2945
being able to produce line deltas.
2947
# TODO: this code generates a parent maps of present ancestors; it
2948
# could be split out into a separate method, and probably should use
2949
# iter_ancestry instead. -- mbp and robertc 20080704
2950
graph = _mod_graph.Graph(self._knit)
2951
head_cache = _mod_graph.FrozenHeadsCache(graph)
2952
search = graph._make_breadth_first_searcher([key])
2956
present, ghosts = search.next_with_ghosts()
2957
except StopIteration:
2959
keys.update(present)
2960
parent_map = self._knit.get_parent_map(keys)
2962
reannotate = annotate.reannotate
2963
for record in self._knit.get_record_stream(keys, 'topological', True):
2965
fulltext = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
2966
parents = parent_map[key]
2967
if parents is not None:
2968
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
2971
parent_cache[key] = list(
2972
reannotate(parent_lines, fulltext, key, None, head_cache))
2974
return parent_cache[key]
2976
raise errors.RevisionNotPresent(key, self._knit)
2980
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2982
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
