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
FulltextContentFactory,
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_lines = split_lines(basis_entry.get_bytes_as('fulltext'))
200
# Manually apply the delta because we have one annotated content and
202
basis_content = PlainKnitContent(basis_lines, compression_parent)
203
basis_content.apply_delta(delta, rec[1])
204
basis_content._should_strip_eol = factory._build_details[1]
205
return ''.join(basis_content.text())
208
class FTPlainToFullText(KnitAdapter):
209
"""An adapter from FT plain knits to unannotated ones."""
211
def get_bytes(self, factory, compressed_bytes):
213
self._data._parse_record_unchecked(compressed_bytes)
214
content, delta = self._plain_factory.parse_record(factory.key[-1],
215
contents, factory._build_details, None)
216
return ''.join(content.text())
219
class DeltaPlainToFullText(KnitAdapter):
220
"""An adapter for deltas from annotated to unannotated."""
222
def get_bytes(self, factory, compressed_bytes):
224
self._data._parse_record_unchecked(compressed_bytes)
225
delta = self._plain_factory.parse_line_delta(contents, rec[1])
226
compression_parent = factory.parents[0]
227
# XXX: string splitting overhead.
228
basis_entry = self._basis_vf.get_record_stream(
229
[compression_parent], 'unordered', True).next()
230
if basis_entry.storage_kind == 'absent':
231
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
232
basis_lines = split_lines(basis_entry.get_bytes_as('fulltext'))
233
basis_content = PlainKnitContent(basis_lines, compression_parent)
234
# Manually apply the delta because we have one annotated content and
236
content, _ = self._plain_factory.parse_record(rec[1], contents,
237
factory._build_details, basis_content)
238
return ''.join(content.text())
241
class KnitContentFactory(ContentFactory):
242
"""Content factory for streaming from knits.
244
:seealso ContentFactory:
247
def __init__(self, key, parents, build_details, sha1, raw_record,
248
annotated, knit=None):
249
"""Create a KnitContentFactory for key.
252
:param parents: The parents.
253
:param build_details: The build details as returned from
255
:param sha1: The sha1 expected from the full text of this object.
256
:param raw_record: The bytes of the knit data from disk.
257
:param annotated: True if the raw data is annotated.
259
ContentFactory.__init__(self)
262
self.parents = parents
263
if build_details[0] == 'line-delta':
268
annotated_kind = 'annotated-'
271
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
272
self._raw_record = raw_record
273
self._build_details = build_details
276
def get_bytes_as(self, storage_kind):
277
if storage_kind == self.storage_kind:
278
return self._raw_record
279
if storage_kind == 'fulltext' and self._knit is not None:
280
return self._knit.get_text(self.key[0])
282
raise errors.UnavailableRepresentation(self.key, storage_kind,
286
class KnitContent(object):
287
"""Content of a knit version to which deltas can be applied.
289
This is always stored in memory as a list of lines with \n at the end,
290
plus a flag saying if the final ending is really there or not, because that
291
corresponds to the on-disk knit representation.
295
self._should_strip_eol = False
297
def apply_delta(self, delta, new_version_id):
298
"""Apply delta to this object to become new_version_id."""
299
raise NotImplementedError(self.apply_delta)
301
def line_delta_iter(self, new_lines):
302
"""Generate line-based delta from this content to new_lines."""
303
new_texts = new_lines.text()
304
old_texts = self.text()
305
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
306
for tag, i1, i2, j1, j2 in s.get_opcodes():
309
# ofrom, oto, length, data
310
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
312
def line_delta(self, new_lines):
313
return list(self.line_delta_iter(new_lines))
316
def get_line_delta_blocks(knit_delta, source, target):
317
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
318
target_len = len(target)
321
for s_begin, s_end, t_len, new_text in knit_delta:
322
true_n = s_begin - s_pos
325
# knit deltas do not provide reliable info about whether the
326
# last line of a file matches, due to eol handling.
327
if source[s_pos + n -1] != target[t_pos + n -1]:
330
yield s_pos, t_pos, n
331
t_pos += t_len + true_n
333
n = target_len - t_pos
335
if source[s_pos + n -1] != target[t_pos + n -1]:
338
yield s_pos, t_pos, n
339
yield s_pos + (target_len - t_pos), target_len, 0
342
class AnnotatedKnitContent(KnitContent):
343
"""Annotated content."""
345
def __init__(self, lines):
346
KnitContent.__init__(self)
350
"""Return a list of (origin, text) for each content line."""
351
lines = self._lines[:]
352
if self._should_strip_eol:
353
origin, last_line = lines[-1]
354
lines[-1] = (origin, last_line.rstrip('\n'))
357
def apply_delta(self, delta, new_version_id):
358
"""Apply delta to this object to become new_version_id."""
361
for start, end, count, delta_lines in delta:
362
lines[offset+start:offset+end] = delta_lines
363
offset = offset + (start - end) + count
367
lines = [text for origin, text in self._lines]
368
except ValueError, e:
369
# most commonly (only?) caused by the internal form of the knit
370
# missing annotation information because of a bug - see thread
372
raise KnitCorrupt(self,
373
"line in annotated knit missing annotation information: %s"
375
if self._should_strip_eol:
376
lines[-1] = lines[-1].rstrip('\n')
380
return AnnotatedKnitContent(self._lines[:])
383
class PlainKnitContent(KnitContent):
384
"""Unannotated content.
386
When annotate[_iter] is called on this content, the same version is reported
387
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
391
def __init__(self, lines, version_id):
392
KnitContent.__init__(self)
394
self._version_id = version_id
397
"""Return a list of (origin, text) for each content line."""
398
return [(self._version_id, line) for line in self._lines]
400
def apply_delta(self, delta, new_version_id):
401
"""Apply delta to this object to become new_version_id."""
404
for start, end, count, delta_lines in delta:
405
lines[offset+start:offset+end] = delta_lines
406
offset = offset + (start - end) + count
407
self._version_id = new_version_id
410
return PlainKnitContent(self._lines[:], self._version_id)
414
if self._should_strip_eol:
416
lines[-1] = lines[-1].rstrip('\n')
420
class _KnitFactory(object):
421
"""Base class for common Factory functions."""
423
def parse_record(self, version_id, record, record_details,
424
base_content, copy_base_content=True):
425
"""Parse a record into a full content object.
427
:param version_id: The official version id for this content
428
:param record: The data returned by read_records_iter()
429
:param record_details: Details about the record returned by
431
:param base_content: If get_build_details returns a compression_parent,
432
you must return a base_content here, else use None
433
:param copy_base_content: When building from the base_content, decide
434
you can either copy it and return a new object, or modify it in
436
:return: (content, delta) A Content object and possibly a line-delta,
439
method, noeol = record_details
440
if method == 'line-delta':
441
if copy_base_content:
442
content = base_content.copy()
444
content = base_content
445
delta = self.parse_line_delta(record, version_id)
446
content.apply_delta(delta, version_id)
448
content = self.parse_fulltext(record, version_id)
450
content._should_strip_eol = noeol
451
return (content, delta)
454
class KnitAnnotateFactory(_KnitFactory):
455
"""Factory for creating annotated Content objects."""
459
def make(self, lines, version_id):
460
num_lines = len(lines)
461
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
463
def parse_fulltext(self, content, version_id):
464
"""Convert fulltext to internal representation
466
fulltext content is of the format
467
revid(utf8) plaintext\n
468
internal representation is of the format:
471
# TODO: jam 20070209 The tests expect this to be returned as tuples,
472
# but the code itself doesn't really depend on that.
473
# Figure out a way to not require the overhead of turning the
474
# list back into tuples.
475
lines = [tuple(line.split(' ', 1)) for line in content]
476
return AnnotatedKnitContent(lines)
478
def parse_line_delta_iter(self, lines):
479
return iter(self.parse_line_delta(lines))
481
def parse_line_delta(self, lines, version_id, plain=False):
482
"""Convert a line based delta into internal representation.
484
line delta is in the form of:
485
intstart intend intcount
487
revid(utf8) newline\n
488
internal representation is
489
(start, end, count, [1..count tuples (revid, newline)])
491
:param plain: If True, the lines are returned as a plain
492
list without annotations, not as a list of (origin, content) tuples, i.e.
493
(start, end, count, [1..count newline])
500
def cache_and_return(line):
501
origin, text = line.split(' ', 1)
502
return cache.setdefault(origin, origin), text
504
# walk through the lines parsing.
505
# Note that the plain test is explicitly pulled out of the
506
# loop to minimise any performance impact
509
start, end, count = [int(n) for n in header.split(',')]
510
contents = [next().split(' ', 1)[1] for i in xrange(count)]
511
result.append((start, end, count, contents))
514
start, end, count = [int(n) for n in header.split(',')]
515
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
516
result.append((start, end, count, contents))
519
def get_fulltext_content(self, lines):
520
"""Extract just the content lines from a fulltext."""
521
return (line.split(' ', 1)[1] for line in lines)
523
def get_linedelta_content(self, lines):
524
"""Extract just the content from a line delta.
526
This doesn't return all of the extra information stored in a delta.
527
Only the actual content lines.
532
header = header.split(',')
533
count = int(header[2])
534
for i in xrange(count):
535
origin, text = next().split(' ', 1)
538
def lower_fulltext(self, content):
539
"""convert a fulltext content record into a serializable form.
541
see parse_fulltext which this inverts.
543
# TODO: jam 20070209 We only do the caching thing to make sure that
544
# the origin is a valid utf-8 line, eventually we could remove it
545
return ['%s %s' % (o, t) for o, t in content._lines]
547
def lower_line_delta(self, delta):
548
"""convert a delta into a serializable form.
550
See parse_line_delta which this inverts.
552
# TODO: jam 20070209 We only do the caching thing to make sure that
553
# the origin is a valid utf-8 line, eventually we could remove it
555
for start, end, c, lines in delta:
556
out.append('%d,%d,%d\n' % (start, end, c))
557
out.extend(origin + ' ' + text
558
for origin, text in lines)
561
def annotate(self, knit, key):
562
content = knit._get_content(key)
563
# adjust for the fact that serialised annotations are only key suffixes
565
if type(key) == tuple:
567
origins = content.annotate()
569
for origin, line in origins:
570
result.append((prefix + (origin,), line))
573
# XXX: This smells a bit. Why would key ever be a non-tuple here?
574
# Aren't keys defined to be tuples? -- spiv 20080618
575
return content.annotate()
578
class KnitPlainFactory(_KnitFactory):
579
"""Factory for creating plain Content objects."""
583
def make(self, lines, version_id):
584
return PlainKnitContent(lines, version_id)
586
def parse_fulltext(self, content, version_id):
587
"""This parses an unannotated fulltext.
589
Note that this is not a noop - the internal representation
590
has (versionid, line) - its just a constant versionid.
592
return self.make(content, version_id)
594
def parse_line_delta_iter(self, lines, version_id):
596
num_lines = len(lines)
597
while cur < num_lines:
600
start, end, c = [int(n) for n in header.split(',')]
601
yield start, end, c, lines[cur:cur+c]
604
def parse_line_delta(self, lines, version_id):
605
return list(self.parse_line_delta_iter(lines, version_id))
607
def get_fulltext_content(self, lines):
608
"""Extract just the content lines from a fulltext."""
611
def get_linedelta_content(self, lines):
612
"""Extract just the content from a line delta.
614
This doesn't return all of the extra information stored in a delta.
615
Only the actual content lines.
620
header = header.split(',')
621
count = int(header[2])
622
for i in xrange(count):
625
def lower_fulltext(self, content):
626
return content.text()
628
def lower_line_delta(self, delta):
630
for start, end, c, lines in delta:
631
out.append('%d,%d,%d\n' % (start, end, c))
635
def annotate(self, knit, key):
636
annotator = _KnitAnnotator(knit)
637
return annotator.annotate(key)
641
def make_file_factory(annotated, mapper):
642
"""Create a factory for creating a file based KnitVersionedFiles.
644
This is only functional enough to run interface tests, it doesn't try to
645
provide a full pack environment.
647
:param annotated: knit annotations are wanted.
648
:param mapper: The mapper from keys to paths.
650
def factory(transport):
651
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
652
access = _KnitKeyAccess(transport, mapper)
653
return KnitVersionedFiles(index, access, annotated=annotated)
657
def make_pack_factory(graph, delta, keylength):
658
"""Create a factory for creating a pack based VersionedFiles.
660
This is only functional enough to run interface tests, it doesn't try to
661
provide a full pack environment.
663
:param graph: Store a graph.
664
:param delta: Delta compress contents.
665
:param keylength: How long should keys be.
667
def factory(transport):
668
parents = graph or delta
674
max_delta_chain = 200
677
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
678
key_elements=keylength)
679
stream = transport.open_write_stream('newpack')
680
writer = pack.ContainerWriter(stream.write)
682
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
683
deltas=delta, add_callback=graph_index.add_nodes)
684
access = _DirectPackAccess({})
685
access.set_writer(writer, graph_index, (transport, 'newpack'))
686
result = KnitVersionedFiles(index, access,
687
max_delta_chain=max_delta_chain)
688
result.stream = stream
689
result.writer = writer
694
def cleanup_pack_knit(versioned_files):
695
versioned_files.stream.close()
696
versioned_files.writer.end()
699
class KnitVersionedFiles(VersionedFiles):
700
"""Storage for many versioned files using knit compression.
702
Backend storage is managed by indices and data objects.
704
:ivar _index: A _KnitGraphIndex or similar that can describe the
705
parents, graph, compression and data location of entries in this
706
KnitVersionedFiles. Note that this is only the index for
707
*this* vfs; if there are fallbacks they must be queried separately.
710
def __init__(self, index, data_access, max_delta_chain=200,
711
annotated=False, reload_func=None):
712
"""Create a KnitVersionedFiles with index and data_access.
714
:param index: The index for the knit data.
715
:param data_access: The access object to store and retrieve knit
717
:param max_delta_chain: The maximum number of deltas to permit during
718
insertion. Set to 0 to prohibit the use of deltas.
719
:param annotated: Set to True to cause annotations to be calculated and
720
stored during insertion.
721
:param reload_func: An function that can be called if we think we need
722
to reload the pack listing and try again. See
723
'bzrlib.repofmt.pack_repo.AggregateIndex' for the signature.
726
self._access = data_access
727
self._max_delta_chain = max_delta_chain
729
self._factory = KnitAnnotateFactory()
731
self._factory = KnitPlainFactory()
732
self._fallback_vfs = []
733
self._reload_func = reload_func
736
return "%s(%r, %r)" % (
737
self.__class__.__name__,
741
def add_fallback_versioned_files(self, a_versioned_files):
742
"""Add a source of texts for texts not present in this knit.
744
:param a_versioned_files: A VersionedFiles object.
746
self._fallback_vfs.append(a_versioned_files)
748
def add_lines(self, key, parents, lines, parent_texts=None,
749
left_matching_blocks=None, nostore_sha=None, random_id=False,
751
"""See VersionedFiles.add_lines()."""
752
self._index._check_write_ok()
753
self._check_add(key, lines, random_id, check_content)
755
# The caller might pass None if there is no graph data, but kndx
756
# indexes can't directly store that, so we give them
757
# an empty tuple instead.
759
return self._add(key, lines, parents,
760
parent_texts, left_matching_blocks, nostore_sha, random_id)
762
def _add(self, key, lines, parents, parent_texts,
763
left_matching_blocks, nostore_sha, random_id):
764
"""Add a set of lines on top of version specified by parents.
766
Any versions not present will be converted into ghosts.
768
# first thing, if the content is something we don't need to store, find
770
line_bytes = ''.join(lines)
771
digest = sha_string(line_bytes)
772
if nostore_sha == digest:
773
raise errors.ExistingContent
776
if parent_texts is None:
778
# Do a single query to ascertain parent presence; we only compress
779
# against parents in the same kvf.
780
present_parent_map = self._index.get_parent_map(parents)
781
for parent in parents:
782
if parent in present_parent_map:
783
present_parents.append(parent)
785
# Currently we can only compress against the left most present parent.
786
if (len(present_parents) == 0 or
787
present_parents[0] != parents[0]):
790
# To speed the extract of texts the delta chain is limited
791
# to a fixed number of deltas. This should minimize both
792
# I/O and the time spend applying deltas.
793
delta = self._check_should_delta(present_parents[0])
795
text_length = len(line_bytes)
798
if lines[-1][-1] != '\n':
799
# copy the contents of lines.
801
options.append('no-eol')
802
lines[-1] = lines[-1] + '\n'
806
if type(element) != str:
807
raise TypeError("key contains non-strings: %r" % (key,))
808
# Knit hunks are still last-element only
810
content = self._factory.make(lines, version_id)
811
if 'no-eol' in options:
812
# Hint to the content object that its text() call should strip the
814
content._should_strip_eol = True
815
if delta or (self._factory.annotated and len(present_parents) > 0):
816
# Merge annotations from parent texts if needed.
817
delta_hunks = self._merge_annotations(content, present_parents,
818
parent_texts, delta, self._factory.annotated,
819
left_matching_blocks)
822
options.append('line-delta')
823
store_lines = self._factory.lower_line_delta(delta_hunks)
824
size, bytes = self._record_to_data(key, digest,
827
options.append('fulltext')
828
# isinstance is slower and we have no hierarchy.
829
if self._factory.__class__ == KnitPlainFactory:
830
# Use the already joined bytes saving iteration time in
832
size, bytes = self._record_to_data(key, digest,
835
# get mixed annotation + content and feed it into the
837
store_lines = self._factory.lower_fulltext(content)
838
size, bytes = self._record_to_data(key, digest,
841
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
842
self._index.add_records(
843
((key, options, access_memo, parents),),
845
return digest, text_length, content
847
def annotate(self, key):
848
"""See VersionedFiles.annotate."""
849
return self._factory.annotate(self, key)
851
def check(self, progress_bar=None):
852
"""See VersionedFiles.check()."""
853
# This doesn't actually test extraction of everything, but that will
854
# impact 'bzr check' substantially, and needs to be integrated with
855
# care. However, it does check for the obvious problem of a delta with
857
keys = self._index.keys()
858
parent_map = self.get_parent_map(keys)
860
if self._index.get_method(key) != 'fulltext':
861
compression_parent = parent_map[key][0]
862
if compression_parent not in parent_map:
863
raise errors.KnitCorrupt(self,
864
"Missing basis parent %s for %s" % (
865
compression_parent, key))
866
for fallback_vfs in self._fallback_vfs:
869
def _check_add(self, key, lines, random_id, check_content):
870
"""check that version_id and lines are safe to add."""
872
if contains_whitespace(version_id):
873
raise InvalidRevisionId(version_id, self)
874
self.check_not_reserved_id(version_id)
875
# TODO: If random_id==False and the key is already present, we should
876
# probably check that the existing content is identical to what is
877
# being inserted, and otherwise raise an exception. This would make
878
# the bundle code simpler.
880
self._check_lines_not_unicode(lines)
881
self._check_lines_are_lines(lines)
883
def _check_header(self, key, line):
884
rec = self._split_header(line)
885
self._check_header_version(rec, key[-1])
888
def _check_header_version(self, rec, version_id):
889
"""Checks the header version on original format knit records.
891
These have the last component of the key embedded in the record.
893
if rec[1] != version_id:
894
raise KnitCorrupt(self,
895
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
897
def _check_should_delta(self, parent):
898
"""Iterate back through the parent listing, looking for a fulltext.
900
This is used when we want to decide whether to add a delta or a new
901
fulltext. It searches for _max_delta_chain parents. When it finds a
902
fulltext parent, it sees if the total size of the deltas leading up to
903
it is large enough to indicate that we want a new full text anyway.
905
Return True if we should create a new delta, False if we should use a
910
for count in xrange(self._max_delta_chain):
911
# XXX: Collapse these two queries:
913
# Note that this only looks in the index of this particular
914
# KnitVersionedFiles, not in the fallbacks. This ensures that
915
# we won't store a delta spanning physical repository
917
method = self._index.get_method(parent)
918
except RevisionNotPresent:
919
# Some basis is not locally present: always delta
921
index, pos, size = self._index.get_position(parent)
922
if method == 'fulltext':
926
# We don't explicitly check for presence because this is in an
927
# inner loop, and if it's missing it'll fail anyhow.
928
# TODO: This should be asking for compression parent, not graph
930
parent = self._index.get_parent_map([parent])[parent][0]
932
# We couldn't find a fulltext, so we must create a new one
934
# Simple heuristic - if the total I/O wold be greater as a delta than
935
# the originally installed fulltext, we create a new fulltext.
936
return fulltext_size > delta_size
938
def _build_details_to_components(self, build_details):
939
"""Convert a build_details tuple to a position tuple."""
940
# record_details, access_memo, compression_parent
941
return build_details[3], build_details[0], build_details[1]
943
def _get_components_positions(self, keys, allow_missing=False):
944
"""Produce a map of position data for the components of keys.
946
This data is intended to be used for retrieving the knit records.
948
A dict of key to (record_details, index_memo, next, parents) is
950
method is the way referenced data should be applied.
951
index_memo is the handle to pass to the data access to actually get the
953
next is the build-parent of the version, or None for fulltexts.
954
parents is the version_ids of the parents of this version
956
:param allow_missing: If True do not raise an error on a missing component,
960
pending_components = keys
961
while pending_components:
962
build_details = self._index.get_build_details(pending_components)
963
current_components = set(pending_components)
964
pending_components = set()
965
for key, details in build_details.iteritems():
966
(index_memo, compression_parent, parents,
967
record_details) = details
968
method = record_details[0]
969
if compression_parent is not None:
970
pending_components.add(compression_parent)
971
component_data[key] = self._build_details_to_components(details)
972
missing = current_components.difference(build_details)
973
if missing and not allow_missing:
974
raise errors.RevisionNotPresent(missing.pop(), self)
975
return component_data
977
def _get_content(self, key, parent_texts={}):
978
"""Returns a content object that makes up the specified
980
cached_version = parent_texts.get(key, None)
981
if cached_version is not None:
982
# Ensure the cache dict is valid.
983
if not self.get_parent_map([key]):
984
raise RevisionNotPresent(key, self)
985
return cached_version
986
text_map, contents_map = self._get_content_maps([key])
987
return contents_map[key]
989
def _get_content_maps(self, keys, nonlocal_keys=None):
990
"""Produce maps of text and KnitContents
992
:param keys: The keys to produce content maps for.
993
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
994
which are known to not be in this knit, but rather in one of the
996
:return: (text_map, content_map) where text_map contains the texts for
997
the requested versions and content_map contains the KnitContents.
999
# FUTURE: This function could be improved for the 'extract many' case
1000
# by tracking each component and only doing the copy when the number of
1001
# children than need to apply delta's to it is > 1 or it is part of the
1004
multiple_versions = len(keys) != 1
1005
record_map = self._get_record_map(keys, allow_missing=True)
1010
if nonlocal_keys is None:
1011
nonlocal_keys = set()
1013
nonlocal_keys = frozenset(nonlocal_keys)
1014
missing_keys = set(nonlocal_keys)
1015
for source in self._fallback_vfs:
1016
if not missing_keys:
1018
for record in source.get_record_stream(missing_keys,
1020
if record.storage_kind == 'absent':
1022
missing_keys.remove(record.key)
1023
lines = split_lines(record.get_bytes_as('fulltext'))
1024
text_map[record.key] = lines
1025
content_map[record.key] = PlainKnitContent(lines, record.key)
1026
if record.key in keys:
1027
final_content[record.key] = content_map[record.key]
1029
if key in nonlocal_keys:
1034
while cursor is not None:
1036
record, record_details, digest, next = record_map[cursor]
1038
raise RevisionNotPresent(cursor, self)
1039
components.append((cursor, record, record_details, digest))
1041
if cursor in content_map:
1042
# no need to plan further back
1043
components.append((cursor, None, None, None))
1047
for (component_id, record, record_details,
1048
digest) in reversed(components):
1049
if component_id in content_map:
1050
content = content_map[component_id]
1052
content, delta = self._factory.parse_record(key[-1],
1053
record, record_details, content,
1054
copy_base_content=multiple_versions)
1055
if multiple_versions:
1056
content_map[component_id] = content
1058
final_content[key] = content
1060
# digest here is the digest from the last applied component.
1061
text = content.text()
1062
actual_sha = sha_strings(text)
1063
if actual_sha != digest:
1064
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
1065
text_map[key] = text
1066
return text_map, final_content
1068
def get_parent_map(self, keys):
1069
"""Get a map of the graph parents of keys.
1071
:param keys: The keys to look up parents for.
1072
:return: A mapping from keys to parents. Absent keys are absent from
1075
return self._get_parent_map_with_sources(keys)[0]
1077
def _get_parent_map_with_sources(self, keys):
1078
"""Get a map of the parents of keys.
1080
:param keys: The keys to look up parents for.
1081
:return: A tuple. The first element is a mapping from keys to parents.
1082
Absent keys are absent from the mapping. The second element is a
1083
list with the locations each key was found in. The first element
1084
is the in-this-knit parents, the second the first fallback source,
1088
sources = [self._index] + self._fallback_vfs
1091
for source in sources:
1094
new_result = source.get_parent_map(missing)
1095
source_results.append(new_result)
1096
result.update(new_result)
1097
missing.difference_update(set(new_result))
1098
return result, source_results
1100
def _get_record_map(self, keys, allow_missing=False):
1101
"""Produce a dictionary of knit records.
1103
:return: {key:(record, record_details, digest, next)}
1105
data returned from read_records
1107
opaque information to pass to parse_record
1109
SHA1 digest of the full text after all steps are done
1111
build-parent of the version, i.e. the leftmost ancestor.
1112
Will be None if the record is not a delta.
1113
:param keys: The keys to build a map for
1114
:param allow_missing: If some records are missing, rather than
1115
error, just return the data that could be generated.
1117
# This retries the whole request if anything fails. Potentially we
1118
# could be a bit more selective. We could track the keys whose records
1119
# we have successfully found, and then only request the new records
1120
# from there. However, _get_components_positions grabs the whole build
1121
# chain, which means we'll likely try to grab the same records again
1122
# anyway. Also, can the build chains change as part of a pack
1123
# operation? We wouldn't want to end up with a broken chain.
1126
position_map = self._get_components_positions(keys,
1127
allow_missing=allow_missing)
1128
# key = component_id, r = record_details, i_m = index_memo,
1130
records = [(key, i_m) for key, (r, i_m, n)
1131
in position_map.iteritems()]
1133
for key, record, digest in self._read_records_iter(records):
1134
(record_details, index_memo, next) = position_map[key]
1135
record_map[key] = record, record_details, digest, next
1137
except errors.RetryWithNewPacks, e:
1138
self._access.reload_or_raise(e)
1140
def _split_by_prefix(self, keys):
1141
"""For the given keys, split them up based on their prefix.
1143
To keep memory pressure somewhat under control, split the
1144
requests back into per-file-id requests, otherwise "bzr co"
1145
extracts the full tree into memory before writing it to disk.
1146
This should be revisited if _get_content_maps() can ever cross
1149
:param keys: An iterable of key tuples
1150
:return: A dict of {prefix: [key_list]}
1152
split_by_prefix = {}
1155
split_by_prefix.setdefault('', []).append(key)
1157
split_by_prefix.setdefault(key[0], []).append(key)
1158
return split_by_prefix
1160
def get_record_stream(self, keys, ordering, include_delta_closure):
1161
"""Get a stream of records for keys.
1163
:param keys: The keys to include.
1164
:param ordering: Either 'unordered' or 'topological'. A topologically
1165
sorted stream has compression parents strictly before their
1167
:param include_delta_closure: If True then the closure across any
1168
compression parents will be included (in the opaque data).
1169
:return: An iterator of ContentFactory objects, each of which is only
1170
valid until the iterator is advanced.
1172
# keys might be a generator
1176
if not self._index.has_graph:
1177
# Cannot topological order when no graph has been stored.
1178
ordering = 'unordered'
1180
remaining_keys = keys
1183
keys = set(remaining_keys)
1184
for content_factory in self._get_remaining_record_stream(keys,
1185
ordering, include_delta_closure):
1186
remaining_keys.discard(content_factory.key)
1187
yield content_factory
1189
except errors.RetryWithNewPacks, e:
1190
self._access.reload_or_raise(e)
1192
def _get_remaining_record_stream(self, keys, ordering,
1193
include_delta_closure):
1194
"""This function is the 'retry' portion for get_record_stream."""
1195
if include_delta_closure:
1196
positions = self._get_components_positions(keys, allow_missing=True)
1198
build_details = self._index.get_build_details(keys)
1200
# (record_details, access_memo, compression_parent_key)
1201
positions = dict((key, self._build_details_to_components(details))
1202
for key, details in build_details.iteritems())
1203
absent_keys = keys.difference(set(positions))
1204
# There may be more absent keys : if we're missing the basis component
1205
# and are trying to include the delta closure.
1206
if include_delta_closure:
1207
needed_from_fallback = set()
1208
# Build up reconstructable_keys dict. key:True in this dict means
1209
# the key can be reconstructed.
1210
reconstructable_keys = {}
1214
chain = [key, positions[key][2]]
1216
needed_from_fallback.add(key)
1219
while chain[-1] is not None:
1220
if chain[-1] in reconstructable_keys:
1221
result = reconstructable_keys[chain[-1]]
1225
chain.append(positions[chain[-1]][2])
1227
# missing basis component
1228
needed_from_fallback.add(chain[-1])
1231
for chain_key in chain[:-1]:
1232
reconstructable_keys[chain_key] = result
1234
needed_from_fallback.add(key)
1235
# Double index lookups here : need a unified api ?
1236
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1237
if ordering == 'topological':
1238
# Global topological sort
1239
present_keys = tsort.topo_sort(global_map)
1240
# Now group by source:
1242
current_source = None
1243
for key in present_keys:
1244
for parent_map in parent_maps:
1245
if key in parent_map:
1246
key_source = parent_map
1248
if current_source is not key_source:
1249
source_keys.append((key_source, []))
1250
current_source = key_source
1251
source_keys[-1][1].append(key)
1253
if ordering != 'unordered':
1254
raise AssertionError('valid values for ordering are:'
1255
' "unordered" or "topological" not: %r'
1257
# Just group by source; remote sources first.
1260
for parent_map in reversed(parent_maps):
1261
source_keys.append((parent_map, []))
1262
for key in parent_map:
1263
present_keys.append(key)
1264
source_keys[-1][1].append(key)
1265
absent_keys = keys - set(global_map)
1266
for key in absent_keys:
1267
yield AbsentContentFactory(key)
1268
# restrict our view to the keys we can answer.
1269
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1270
# XXX: At that point we need to consider the impact of double reads by
1271
# utilising components multiple times.
1272
if include_delta_closure:
1273
# XXX: get_content_maps performs its own index queries; allow state
1275
non_local_keys = needed_from_fallback - absent_keys
1276
prefix_split_keys = self._split_by_prefix(present_keys)
1277
prefix_split_non_local_keys = self._split_by_prefix(non_local_keys)
1278
for prefix, keys in prefix_split_keys.iteritems():
1279
non_local = prefix_split_non_local_keys.get(prefix, [])
1280
non_local = set(non_local)
1281
text_map, _ = self._get_content_maps(keys, non_local)
1283
lines = text_map.pop(key)
1284
text = ''.join(lines)
1285
yield FulltextContentFactory(key, global_map[key], None,
1288
for source, keys in source_keys:
1289
if source is parent_maps[0]:
1290
# this KnitVersionedFiles
1291
records = [(key, positions[key][1]) for key in keys]
1292
for key, raw_data, sha1 in self._read_records_iter_raw(records):
1293
(record_details, index_memo, _) = positions[key]
1294
yield KnitContentFactory(key, global_map[key],
1295
record_details, sha1, raw_data, self._factory.annotated, None)
1297
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1298
for record in vf.get_record_stream(keys, ordering,
1299
include_delta_closure):
1302
def get_sha1s(self, keys):
1303
"""See VersionedFiles.get_sha1s()."""
1305
record_map = self._get_record_map(missing, allow_missing=True)
1307
for key, details in record_map.iteritems():
1308
if key not in missing:
1310
# record entry 2 is the 'digest'.
1311
result[key] = details[2]
1312
missing.difference_update(set(result))
1313
for source in self._fallback_vfs:
1316
new_result = source.get_sha1s(missing)
1317
result.update(new_result)
1318
missing.difference_update(set(new_result))
1321
def insert_record_stream(self, stream):
1322
"""Insert a record stream into this container.
1324
:param stream: A stream of records to insert.
1326
:seealso VersionedFiles.get_record_stream:
1328
def get_adapter(adapter_key):
1330
return adapters[adapter_key]
1332
adapter_factory = adapter_registry.get(adapter_key)
1333
adapter = adapter_factory(self)
1334
adapters[adapter_key] = adapter
1336
if self._factory.annotated:
1337
# self is annotated, we need annotated knits to use directly.
1338
annotated = "annotated-"
1341
# self is not annotated, but we can strip annotations cheaply.
1343
convertibles = set(["knit-annotated-ft-gz"])
1344
if self._max_delta_chain:
1345
convertibles.add("knit-annotated-delta-gz")
1346
# The set of types we can cheaply adapt without needing basis texts.
1347
native_types = set()
1348
if self._max_delta_chain:
1349
native_types.add("knit-%sdelta-gz" % annotated)
1350
native_types.add("knit-%sft-gz" % annotated)
1351
knit_types = native_types.union(convertibles)
1353
# Buffer all index entries that we can't add immediately because their
1354
# basis parent is missing. We don't buffer all because generating
1355
# annotations may require access to some of the new records. However we
1356
# can't generate annotations from new deltas until their basis parent
1357
# is present anyway, so we get away with not needing an index that
1358
# includes the new keys.
1360
# See <http://launchpad.net/bugs/300177> about ordering of compression
1361
# parents in the records - to be conservative, we insist that all
1362
# parents must be present to avoid expanding to a fulltext.
1364
# key = basis_parent, value = index entry to add
1365
buffered_index_entries = {}
1366
for record in stream:
1367
parents = record.parents
1368
# Raise an error when a record is missing.
1369
if record.storage_kind == 'absent':
1370
raise RevisionNotPresent([record.key], self)
1371
elif ((record.storage_kind in knit_types)
1373
or not self._fallback_vfs
1374
or not self._index.missing_keys(parents)
1375
or self.missing_keys(parents))):
1376
# we can insert the knit record literally if either it has no
1377
# compression parent OR we already have its basis in this kvf
1378
# OR the basis is not present even in the fallbacks. In the
1379
# last case it will either turn up later in the stream and all
1380
# will be well, or it won't turn up at all and we'll raise an
1383
# TODO: self.has_key is somewhat redundant with
1384
# self._index.has_key; we really want something that directly
1385
# asks if it's only present in the fallbacks. -- mbp 20081119
1386
if record.storage_kind not in native_types:
1388
adapter_key = (record.storage_kind, "knit-delta-gz")
1389
adapter = get_adapter(adapter_key)
1391
adapter_key = (record.storage_kind, "knit-ft-gz")
1392
adapter = get_adapter(adapter_key)
1393
bytes = adapter.get_bytes(
1394
record, record.get_bytes_as(record.storage_kind))
1396
bytes = record.get_bytes_as(record.storage_kind)
1397
options = [record._build_details[0]]
1398
if record._build_details[1]:
1399
options.append('no-eol')
1400
# Just blat it across.
1401
# Note: This does end up adding data on duplicate keys. As
1402
# modern repositories use atomic insertions this should not
1403
# lead to excessive growth in the event of interrupted fetches.
1404
# 'knit' repositories may suffer excessive growth, but as a
1405
# deprecated format this is tolerable. It can be fixed if
1406
# needed by in the kndx index support raising on a duplicate
1407
# add with identical parents and options.
1408
access_memo = self._access.add_raw_records(
1409
[(record.key, len(bytes))], bytes)[0]
1410
index_entry = (record.key, options, access_memo, parents)
1412
if 'fulltext' not in options:
1413
# Not a fulltext, so we need to make sure the compression
1414
# parent will also be present.
1415
# Note that pack backed knits don't need to buffer here
1416
# because they buffer all writes to the transaction level,
1417
# but we don't expose that difference at the index level. If
1418
# the query here has sufficient cost to show up in
1419
# profiling we should do that.
1421
# They're required to be physically in this
1422
# KnitVersionedFiles, not in a fallback.
1423
compression_parent = parents[0]
1424
if self.missing_keys([compression_parent]):
1425
pending = buffered_index_entries.setdefault(
1426
compression_parent, [])
1427
pending.append(index_entry)
1430
self._index.add_records([index_entry])
1431
elif record.storage_kind == 'fulltext':
1432
self.add_lines(record.key, parents,
1433
split_lines(record.get_bytes_as('fulltext')))
1435
# Not a fulltext, and not suitable for direct insertion as a
1436
# delta, either because it's not the right format, or this
1437
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1438
# 0) or because it depends on a base only present in the
1440
adapter_key = record.storage_kind, 'fulltext'
1441
adapter = get_adapter(adapter_key)
1442
lines = split_lines(adapter.get_bytes(
1443
record, record.get_bytes_as(record.storage_kind)))
1445
self.add_lines(record.key, parents, lines)
1446
except errors.RevisionAlreadyPresent:
1448
# Add any records whose basis parent is now available.
1449
added_keys = [record.key]
1451
key = added_keys.pop(0)
1452
if key in buffered_index_entries:
1453
index_entries = buffered_index_entries[key]
1454
self._index.add_records(index_entries)
1456
[index_entry[0] for index_entry in index_entries])
1457
del buffered_index_entries[key]
1458
# If there were any deltas which had a missing basis parent, error.
1459
if buffered_index_entries:
1460
from pprint import pformat
1461
raise errors.BzrCheckError(
1462
"record_stream refers to compression parents not in %r:\n%s"
1463
% (self, pformat(sorted(buffered_index_entries.keys()))))
1465
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1466
"""Iterate over the lines in the versioned files from keys.
1468
This may return lines from other keys. Each item the returned
1469
iterator yields is a tuple of a line and a text version that that line
1470
is present in (not introduced in).
1472
Ordering of results is in whatever order is most suitable for the
1473
underlying storage format.
1475
If a progress bar is supplied, it may be used to indicate progress.
1476
The caller is responsible for cleaning up progress bars (because this
1480
* Lines are normalised by the underlying store: they will all have \\n
1482
* Lines are returned in arbitrary order.
1483
* If a requested key did not change any lines (or didn't have any
1484
lines), it may not be mentioned at all in the result.
1486
:return: An iterator over (line, key).
1489
pb = progress.DummyProgress()
1495
# we don't care about inclusions, the caller cares.
1496
# but we need to setup a list of records to visit.
1497
# we need key, position, length
1499
build_details = self._index.get_build_details(keys)
1500
for key, details in build_details.iteritems():
1502
key_records.append((key, details[0]))
1503
records_iter = enumerate(self._read_records_iter(key_records))
1504
for (key_idx, (key, data, sha_value)) in records_iter:
1505
pb.update('Walking content.', key_idx, total)
1506
compression_parent = build_details[key][1]
1507
if compression_parent is None:
1509
line_iterator = self._factory.get_fulltext_content(data)
1512
line_iterator = self._factory.get_linedelta_content(data)
1513
# Now that we are yielding the data for this key, remove it
1516
# XXX: It might be more efficient to yield (key,
1517
# line_iterator) in the future. However for now, this is a
1518
# simpler change to integrate into the rest of the
1519
# codebase. RBC 20071110
1520
for line in line_iterator:
1523
except errors.RetryWithNewPacks, e:
1524
self._access.reload_or_raise(e)
1525
# If there are still keys we've not yet found, we look in the fallback
1526
# vfs, and hope to find them there. Note that if the keys are found
1527
# but had no changes or no content, the fallback may not return
1529
if keys and not self._fallback_vfs:
1530
# XXX: strictly the second parameter is meant to be the file id
1531
# but it's not easily accessible here.
1532
raise RevisionNotPresent(keys, repr(self))
1533
for source in self._fallback_vfs:
1537
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1538
source_keys.add(key)
1540
keys.difference_update(source_keys)
1541
pb.update('Walking content.', total, total)
1543
def _make_line_delta(self, delta_seq, new_content):
1544
"""Generate a line delta from delta_seq and new_content."""
1546
for op in delta_seq.get_opcodes():
1547
if op[0] == 'equal':
1549
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1552
def _merge_annotations(self, content, parents, parent_texts={},
1553
delta=None, annotated=None,
1554
left_matching_blocks=None):
1555
"""Merge annotations for content and generate deltas.
1557
This is done by comparing the annotations based on changes to the text
1558
and generating a delta on the resulting full texts. If annotations are
1559
not being created then a simple delta is created.
1561
if left_matching_blocks is not None:
1562
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1566
for parent_key in parents:
1567
merge_content = self._get_content(parent_key, parent_texts)
1568
if (parent_key == parents[0] and delta_seq is not None):
1571
seq = patiencediff.PatienceSequenceMatcher(
1572
None, merge_content.text(), content.text())
1573
for i, j, n in seq.get_matching_blocks():
1576
# this copies (origin, text) pairs across to the new
1577
# content for any line that matches the last-checked
1579
content._lines[j:j+n] = merge_content._lines[i:i+n]
1580
# XXX: Robert says the following block is a workaround for a
1581
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1582
if content._lines and content._lines[-1][1][-1] != '\n':
1583
# The copied annotation was from a line without a trailing EOL,
1584
# reinstate one for the content object, to ensure correct
1586
line = content._lines[-1][1] + '\n'
1587
content._lines[-1] = (content._lines[-1][0], line)
1589
if delta_seq is None:
1590
reference_content = self._get_content(parents[0], parent_texts)
1591
new_texts = content.text()
1592
old_texts = reference_content.text()
1593
delta_seq = patiencediff.PatienceSequenceMatcher(
1594
None, old_texts, new_texts)
1595
return self._make_line_delta(delta_seq, content)
1597
def _parse_record(self, version_id, data):
1598
"""Parse an original format knit record.
1600
These have the last element of the key only present in the stored data.
1602
rec, record_contents = self._parse_record_unchecked(data)
1603
self._check_header_version(rec, version_id)
1604
return record_contents, rec[3]
1606
def _parse_record_header(self, key, raw_data):
1607
"""Parse a record header for consistency.
1609
:return: the header and the decompressor stream.
1610
as (stream, header_record)
1612
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1615
rec = self._check_header(key, df.readline())
1616
except Exception, e:
1617
raise KnitCorrupt(self,
1618
"While reading {%s} got %s(%s)"
1619
% (key, e.__class__.__name__, str(e)))
1622
def _parse_record_unchecked(self, data):
1624
# 4168 calls in 2880 217 internal
1625
# 4168 calls to _parse_record_header in 2121
1626
# 4168 calls to readlines in 330
1627
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1629
record_contents = df.readlines()
1630
except Exception, e:
1631
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1632
(data, e.__class__.__name__, str(e)))
1633
header = record_contents.pop(0)
1634
rec = self._split_header(header)
1635
last_line = record_contents.pop()
1636
if len(record_contents) != int(rec[2]):
1637
raise KnitCorrupt(self,
1638
'incorrect number of lines %s != %s'
1639
' for version {%s} %s'
1640
% (len(record_contents), int(rec[2]),
1641
rec[1], record_contents))
1642
if last_line != 'end %s\n' % rec[1]:
1643
raise KnitCorrupt(self,
1644
'unexpected version end line %r, wanted %r'
1645
% (last_line, rec[1]))
1647
return rec, record_contents
1649
def _read_records_iter(self, records):
1650
"""Read text records from data file and yield result.
1652
The result will be returned in whatever is the fastest to read.
1653
Not by the order requested. Also, multiple requests for the same
1654
record will only yield 1 response.
1655
:param records: A list of (key, access_memo) entries
1656
:return: Yields (key, contents, digest) in the order
1657
read, not the order requested
1662
# XXX: This smells wrong, IO may not be getting ordered right.
1663
needed_records = sorted(set(records), key=operator.itemgetter(1))
1664
if not needed_records:
1667
# The transport optimizes the fetching as well
1668
# (ie, reads continuous ranges.)
1669
raw_data = self._access.get_raw_records(
1670
[index_memo for key, index_memo in needed_records])
1672
for (key, index_memo), data in \
1673
izip(iter(needed_records), raw_data):
1674
content, digest = self._parse_record(key[-1], data)
1675
yield key, content, digest
1677
def _read_records_iter_raw(self, records):
1678
"""Read text records from data file and yield raw data.
1680
This unpacks enough of the text record to validate the id is
1681
as expected but thats all.
1683
Each item the iterator yields is (key, bytes, sha1_of_full_text).
1685
# setup an iterator of the external records:
1686
# uses readv so nice and fast we hope.
1688
# grab the disk data needed.
1689
needed_offsets = [index_memo for key, index_memo
1691
raw_records = self._access.get_raw_records(needed_offsets)
1693
for key, index_memo in records:
1694
data = raw_records.next()
1695
# validate the header (note that we can only use the suffix in
1696
# current knit records).
1697
df, rec = self._parse_record_header(key, data)
1699
yield key, data, rec[3]
1701
def _record_to_data(self, key, digest, lines, dense_lines=None):
1702
"""Convert key, digest, lines into a raw data block.
1704
:param key: The key of the record. Currently keys are always serialised
1705
using just the trailing component.
1706
:param dense_lines: The bytes of lines but in a denser form. For
1707
instance, if lines is a list of 1000 bytestrings each ending in \n,
1708
dense_lines may be a list with one line in it, containing all the
1709
1000's lines and their \n's. Using dense_lines if it is already
1710
known is a win because the string join to create bytes in this
1711
function spends less time resizing the final string.
1712
:return: (len, a StringIO instance with the raw data ready to read.)
1714
# Note: using a string copy here increases memory pressure with e.g.
1715
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
1716
# when doing the initial commit of a mozilla tree. RBC 20070921
1717
bytes = ''.join(chain(
1718
["version %s %d %s\n" % (key[-1],
1721
dense_lines or lines,
1722
["end %s\n" % key[-1]]))
1723
if type(bytes) != str:
1724
raise AssertionError(
1725
'data must be plain bytes was %s' % type(bytes))
1726
if lines and lines[-1][-1] != '\n':
1727
raise ValueError('corrupt lines value %r' % lines)
1728
compressed_bytes = tuned_gzip.bytes_to_gzip(bytes)
1729
return len(compressed_bytes), compressed_bytes
1731
def _split_header(self, line):
1734
raise KnitCorrupt(self,
1735
'unexpected number of elements in record header')
1739
"""See VersionedFiles.keys."""
1740
if 'evil' in debug.debug_flags:
1741
trace.mutter_callsite(2, "keys scales with size of history")
1742
sources = [self._index] + self._fallback_vfs
1744
for source in sources:
1745
result.update(source.keys())
1749
class _KndxIndex(object):
1750
"""Manages knit index files
1752
The index is kept in memory and read on startup, to enable
1753
fast lookups of revision information. The cursor of the index
1754
file is always pointing to the end, making it easy to append
1757
_cache is a cache for fast mapping from version id to a Index
1760
_history is a cache for fast mapping from indexes to version ids.
1762
The index data format is dictionary compressed when it comes to
1763
parent references; a index entry may only have parents that with a
1764
lover index number. As a result, the index is topological sorted.
1766
Duplicate entries may be written to the index for a single version id
1767
if this is done then the latter one completely replaces the former:
1768
this allows updates to correct version and parent information.
1769
Note that the two entries may share the delta, and that successive
1770
annotations and references MUST point to the first entry.
1772
The index file on disc contains a header, followed by one line per knit
1773
record. The same revision can be present in an index file more than once.
1774
The first occurrence gets assigned a sequence number starting from 0.
1776
The format of a single line is
1777
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1778
REVISION_ID is a utf8-encoded revision id
1779
FLAGS is a comma separated list of flags about the record. Values include
1780
no-eol, line-delta, fulltext.
1781
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1782
that the the compressed data starts at.
1783
LENGTH is the ascii representation of the length of the data file.
1784
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1786
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1787
revision id already in the knit that is a parent of REVISION_ID.
1788
The ' :' marker is the end of record marker.
1791
when a write is interrupted to the index file, it will result in a line
1792
that does not end in ' :'. If the ' :' is not present at the end of a line,
1793
or at the end of the file, then the record that is missing it will be
1794
ignored by the parser.
1796
When writing new records to the index file, the data is preceded by '\n'
1797
to ensure that records always start on new lines even if the last write was
1798
interrupted. As a result its normal for the last line in the index to be
1799
missing a trailing newline. One can be added with no harmful effects.
1801
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
1802
where prefix is e.g. the (fileid,) for .texts instances or () for
1803
constant-mapped things like .revisions, and the old state is
1804
tuple(history_vector, cache_dict). This is used to prevent having an
1805
ABI change with the C extension that reads .kndx files.
1808
HEADER = "# bzr knit index 8\n"
1810
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
1811
"""Create a _KndxIndex on transport using mapper."""
1812
self._transport = transport
1813
self._mapper = mapper
1814
self._get_scope = get_scope
1815
self._allow_writes = allow_writes
1816
self._is_locked = is_locked
1818
self.has_graph = True
1820
def add_records(self, records, random_id=False):
1821
"""Add multiple records to the index.
1823
:param records: a list of tuples:
1824
(key, options, access_memo, parents).
1825
:param random_id: If True the ids being added were randomly generated
1826
and no check for existence will be performed.
1829
for record in records:
1832
path = self._mapper.map(key) + '.kndx'
1833
path_keys = paths.setdefault(path, (prefix, []))
1834
path_keys[1].append(record)
1835
for path in sorted(paths):
1836
prefix, path_keys = paths[path]
1837
self._load_prefixes([prefix])
1839
orig_history = self._kndx_cache[prefix][1][:]
1840
orig_cache = self._kndx_cache[prefix][0].copy()
1843
for key, options, (_, pos, size), parents in path_keys:
1845
# kndx indices cannot be parentless.
1847
line = "\n%s %s %s %s %s :" % (
1848
key[-1], ','.join(options), pos, size,
1849
self._dictionary_compress(parents))
1850
if type(line) != str:
1851
raise AssertionError(
1852
'data must be utf8 was %s' % type(line))
1854
self._cache_key(key, options, pos, size, parents)
1855
if len(orig_history):
1856
self._transport.append_bytes(path, ''.join(lines))
1858
self._init_index(path, lines)
1860
# If any problems happen, restore the original values and re-raise
1861
self._kndx_cache[prefix] = (orig_cache, orig_history)
1864
def _cache_key(self, key, options, pos, size, parent_keys):
1865
"""Cache a version record in the history array and index cache.
1867
This is inlined into _load_data for performance. KEEP IN SYNC.
1868
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1872
version_id = key[-1]
1873
# last-element only for compatibilty with the C load_data.
1874
parents = tuple(parent[-1] for parent in parent_keys)
1875
for parent in parent_keys:
1876
if parent[:-1] != prefix:
1877
raise ValueError("mismatched prefixes for %r, %r" % (
1879
cache, history = self._kndx_cache[prefix]
1880
# only want the _history index to reference the 1st index entry
1882
if version_id not in cache:
1883
index = len(history)
1884
history.append(version_id)
1886
index = cache[version_id][5]
1887
cache[version_id] = (version_id,
1894
def check_header(self, fp):
1895
line = fp.readline()
1897
# An empty file can actually be treated as though the file doesn't
1899
raise errors.NoSuchFile(self)
1900
if line != self.HEADER:
1901
raise KnitHeaderError(badline=line, filename=self)
1903
def _check_read(self):
1904
if not self._is_locked():
1905
raise errors.ObjectNotLocked(self)
1906
if self._get_scope() != self._scope:
1909
def _check_write_ok(self):
1910
"""Assert if not writes are permitted."""
1911
if not self._is_locked():
1912
raise errors.ObjectNotLocked(self)
1913
if self._get_scope() != self._scope:
1915
if self._mode != 'w':
1916
raise errors.ReadOnlyObjectDirtiedError(self)
1918
def get_build_details(self, keys):
1919
"""Get the method, index_memo and compression parent for keys.
1921
Ghosts are omitted from the result.
1923
:param keys: An iterable of keys.
1924
:return: A dict of key:(index_memo, compression_parent, parents,
1927
opaque structure to pass to read_records to extract the raw
1930
Content that this record is built upon, may be None
1932
Logical parents of this node
1934
extra information about the content which needs to be passed to
1935
Factory.parse_record
1937
parent_map = self.get_parent_map(keys)
1940
if key not in parent_map:
1942
method = self.get_method(key)
1943
parents = parent_map[key]
1944
if method == 'fulltext':
1945
compression_parent = None
1947
compression_parent = parents[0]
1948
noeol = 'no-eol' in self.get_options(key)
1949
index_memo = self.get_position(key)
1950
result[key] = (index_memo, compression_parent,
1951
parents, (method, noeol))
1954
def get_method(self, key):
1955
"""Return compression method of specified key."""
1956
options = self.get_options(key)
1957
if 'fulltext' in options:
1959
elif 'line-delta' in options:
1962
raise errors.KnitIndexUnknownMethod(self, options)
1964
def get_options(self, key):
1965
"""Return a list representing options.
1969
prefix, suffix = self._split_key(key)
1970
self._load_prefixes([prefix])
1972
return self._kndx_cache[prefix][0][suffix][1]
1974
raise RevisionNotPresent(key, self)
1976
def get_parent_map(self, keys):
1977
"""Get a map of the parents of keys.
1979
:param keys: The keys to look up parents for.
1980
:return: A mapping from keys to parents. Absent keys are absent from
1983
# Parse what we need to up front, this potentially trades off I/O
1984
# locality (.kndx and .knit in the same block group for the same file
1985
# id) for less checking in inner loops.
1986
prefixes = set(key[:-1] for key in keys)
1987
self._load_prefixes(prefixes)
1992
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
1996
result[key] = tuple(prefix + (suffix,) for
1997
suffix in suffix_parents)
2000
def get_position(self, key):
2001
"""Return details needed to access the version.
2003
:return: a tuple (key, data position, size) to hand to the access
2004
logic to get the record.
2006
prefix, suffix = self._split_key(key)
2007
self._load_prefixes([prefix])
2008
entry = self._kndx_cache[prefix][0][suffix]
2009
return key, entry[2], entry[3]
2011
has_key = _mod_index._has_key_from_parent_map
2013
def _init_index(self, path, extra_lines=[]):
2014
"""Initialize an index."""
2016
sio.write(self.HEADER)
2017
sio.writelines(extra_lines)
2019
self._transport.put_file_non_atomic(path, sio,
2020
create_parent_dir=True)
2021
# self._create_parent_dir)
2022
# mode=self._file_mode,
2023
# dir_mode=self._dir_mode)
2026
"""Get all the keys in the collection.
2028
The keys are not ordered.
2031
# Identify all key prefixes.
2032
# XXX: A bit hacky, needs polish.
2033
if type(self._mapper) == ConstantMapper:
2037
for quoted_relpath in self._transport.iter_files_recursive():
2038
path, ext = os.path.splitext(quoted_relpath)
2040
prefixes = [self._mapper.unmap(path) for path in relpaths]
2041
self._load_prefixes(prefixes)
2042
for prefix in prefixes:
2043
for suffix in self._kndx_cache[prefix][1]:
2044
result.add(prefix + (suffix,))
2047
def _load_prefixes(self, prefixes):
2048
"""Load the indices for prefixes."""
2050
for prefix in prefixes:
2051
if prefix not in self._kndx_cache:
2052
# the load_data interface writes to these variables.
2055
self._filename = prefix
2057
path = self._mapper.map(prefix) + '.kndx'
2058
fp = self._transport.get(path)
2060
# _load_data may raise NoSuchFile if the target knit is
2062
_load_data(self, fp)
2065
self._kndx_cache[prefix] = (self._cache, self._history)
2070
self._kndx_cache[prefix] = ({}, [])
2071
if type(self._mapper) == ConstantMapper:
2072
# preserve behaviour for revisions.kndx etc.
2073
self._init_index(path)
2078
missing_keys = _mod_index._missing_keys_from_parent_map
2080
def _partition_keys(self, keys):
2081
"""Turn keys into a dict of prefix:suffix_list."""
2084
prefix_keys = result.setdefault(key[:-1], [])
2085
prefix_keys.append(key[-1])
2088
def _dictionary_compress(self, keys):
2089
"""Dictionary compress keys.
2091
:param keys: The keys to generate references to.
2092
:return: A string representation of keys. keys which are present are
2093
dictionary compressed, and others are emitted as fulltext with a
2099
prefix = keys[0][:-1]
2100
cache = self._kndx_cache[prefix][0]
2102
if key[:-1] != prefix:
2103
# kndx indices cannot refer across partitioned storage.
2104
raise ValueError("mismatched prefixes for %r" % keys)
2105
if key[-1] in cache:
2106
# -- inlined lookup() --
2107
result_list.append(str(cache[key[-1]][5]))
2108
# -- end lookup () --
2110
result_list.append('.' + key[-1])
2111
return ' '.join(result_list)
2113
def _reset_cache(self):
2114
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2115
# (cache_dict, history_vector) for parsed kndx files.
2116
self._kndx_cache = {}
2117
self._scope = self._get_scope()
2118
allow_writes = self._allow_writes()
2124
def _split_key(self, key):
2125
"""Split key into a prefix and suffix."""
2126
return key[:-1], key[-1]
2129
class _KnitGraphIndex(object):
2130
"""A KnitVersionedFiles index layered on GraphIndex."""
2132
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2134
"""Construct a KnitGraphIndex on a graph_index.
2136
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2137
:param is_locked: A callback to check whether the object should answer
2139
:param deltas: Allow delta-compressed records.
2140
:param parents: If True, record knits parents, if not do not record
2142
:param add_callback: If not None, allow additions to the index and call
2143
this callback with a list of added GraphIndex nodes:
2144
[(node, value, node_refs), ...]
2145
:param is_locked: A callback, returns True if the index is locked and
2148
self._add_callback = add_callback
2149
self._graph_index = graph_index
2150
self._deltas = deltas
2151
self._parents = parents
2152
if deltas and not parents:
2153
# XXX: TODO: Delta tree and parent graph should be conceptually
2155
raise KnitCorrupt(self, "Cannot do delta compression without "
2157
self.has_graph = parents
2158
self._is_locked = is_locked
2161
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2163
def add_records(self, records, random_id=False):
2164
"""Add multiple records to the index.
2166
This function does not insert data into the Immutable GraphIndex
2167
backing the KnitGraphIndex, instead it prepares data for insertion by
2168
the caller and checks that it is safe to insert then calls
2169
self._add_callback with the prepared GraphIndex nodes.
2171
:param records: a list of tuples:
2172
(key, options, access_memo, parents).
2173
:param random_id: If True the ids being added were randomly generated
2174
and no check for existence will be performed.
2176
if not self._add_callback:
2177
raise errors.ReadOnlyError(self)
2178
# we hope there are no repositories with inconsistent parentage
2182
for (key, options, access_memo, parents) in records:
2184
parents = tuple(parents)
2185
index, pos, size = access_memo
2186
if 'no-eol' in options:
2190
value += "%d %d" % (pos, size)
2191
if not self._deltas:
2192
if 'line-delta' in options:
2193
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2196
if 'line-delta' in options:
2197
node_refs = (parents, (parents[0],))
2199
node_refs = (parents, ())
2201
node_refs = (parents, )
2204
raise KnitCorrupt(self, "attempt to add node with parents "
2205
"in parentless index.")
2207
keys[key] = (value, node_refs)
2210
present_nodes = self._get_entries(keys)
2211
for (index, key, value, node_refs) in present_nodes:
2212
if (value[0] != keys[key][0][0] or
2213
node_refs != keys[key][1]):
2214
raise KnitCorrupt(self, "inconsistent details in add_records"
2215
": %s %s" % ((value, node_refs), keys[key]))
2219
for key, (value, node_refs) in keys.iteritems():
2220
result.append((key, value, node_refs))
2222
for key, (value, node_refs) in keys.iteritems():
2223
result.append((key, value))
2224
self._add_callback(result)
2226
def _check_read(self):
2227
"""raise if reads are not permitted."""
2228
if not self._is_locked():
2229
raise errors.ObjectNotLocked(self)
2231
def _check_write_ok(self):
2232
"""Assert if writes are not permitted."""
2233
if not self._is_locked():
2234
raise errors.ObjectNotLocked(self)
2236
def _compression_parent(self, an_entry):
2237
# return the key that an_entry is compressed against, or None
2238
# Grab the second parent list (as deltas implies parents currently)
2239
compression_parents = an_entry[3][1]
2240
if not compression_parents:
2242
if len(compression_parents) != 1:
2243
raise AssertionError(
2244
"Too many compression parents: %r" % compression_parents)
2245
return compression_parents[0]
2247
def get_build_details(self, keys):
2248
"""Get the method, index_memo and compression parent for version_ids.
2250
Ghosts are omitted from the result.
2252
:param keys: An iterable of keys.
2253
:return: A dict of key:
2254
(index_memo, compression_parent, parents, record_details).
2256
opaque structure to pass to read_records to extract the raw
2259
Content that this record is built upon, may be None
2261
Logical parents of this node
2263
extra information about the content which needs to be passed to
2264
Factory.parse_record
2268
entries = self._get_entries(keys, False)
2269
for entry in entries:
2271
if not self._parents:
2274
parents = entry[3][0]
2275
if not self._deltas:
2276
compression_parent_key = None
2278
compression_parent_key = self._compression_parent(entry)
2279
noeol = (entry[2][0] == 'N')
2280
if compression_parent_key:
2281
method = 'line-delta'
2284
result[key] = (self._node_to_position(entry),
2285
compression_parent_key, parents,
2289
def _get_entries(self, keys, check_present=False):
2290
"""Get the entries for keys.
2292
:param keys: An iterable of index key tuples.
2297
for node in self._graph_index.iter_entries(keys):
2299
found_keys.add(node[1])
2301
# adapt parentless index to the rest of the code.
2302
for node in self._graph_index.iter_entries(keys):
2303
yield node[0], node[1], node[2], ()
2304
found_keys.add(node[1])
2306
missing_keys = keys.difference(found_keys)
2308
raise RevisionNotPresent(missing_keys.pop(), self)
2310
def get_method(self, key):
2311
"""Return compression method of specified key."""
2312
return self._get_method(self._get_node(key))
2314
def _get_method(self, node):
2315
if not self._deltas:
2317
if self._compression_parent(node):
2322
def _get_node(self, key):
2324
return list(self._get_entries([key]))[0]
2326
raise RevisionNotPresent(key, self)
2328
def get_options(self, key):
2329
"""Return a list representing options.
2333
node = self._get_node(key)
2334
options = [self._get_method(node)]
2335
if node[2][0] == 'N':
2336
options.append('no-eol')
2339
def get_parent_map(self, keys):
2340
"""Get a map of the parents of keys.
2342
:param keys: The keys to look up parents for.
2343
:return: A mapping from keys to parents. Absent keys are absent from
2347
nodes = self._get_entries(keys)
2351
result[node[1]] = node[3][0]
2354
result[node[1]] = None
2357
def get_position(self, key):
2358
"""Return details needed to access the version.
2360
:return: a tuple (index, data position, size) to hand to the access
2361
logic to get the record.
2363
node = self._get_node(key)
2364
return self._node_to_position(node)
2366
has_key = _mod_index._has_key_from_parent_map
2369
"""Get all the keys in the collection.
2371
The keys are not ordered.
2374
return [node[1] for node in self._graph_index.iter_all_entries()]
2376
missing_keys = _mod_index._missing_keys_from_parent_map
2378
def _node_to_position(self, node):
2379
"""Convert an index value to position details."""
2380
bits = node[2][1:].split(' ')
2381
return node[0], int(bits[0]), int(bits[1])
2384
class _KnitKeyAccess(object):
2385
"""Access to records in .knit files."""
2387
def __init__(self, transport, mapper):
2388
"""Create a _KnitKeyAccess with transport and mapper.
2390
:param transport: The transport the access object is rooted at.
2391
:param mapper: The mapper used to map keys to .knit files.
2393
self._transport = transport
2394
self._mapper = mapper
2396
def add_raw_records(self, key_sizes, raw_data):
2397
"""Add raw knit bytes to a storage area.
2399
The data is spooled to the container writer in one bytes-record per
2402
:param sizes: An iterable of tuples containing the key and size of each
2404
:param raw_data: A bytestring containing the data.
2405
:return: A list of memos to retrieve the record later. Each memo is an
2406
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
2407
length), where the key is the record key.
2409
if type(raw_data) != str:
2410
raise AssertionError(
2411
'data must be plain bytes was %s' % type(raw_data))
2414
# TODO: This can be tuned for writing to sftp and other servers where
2415
# append() is relatively expensive by grouping the writes to each key
2417
for key, size in key_sizes:
2418
path = self._mapper.map(key)
2420
base = self._transport.append_bytes(path + '.knit',
2421
raw_data[offset:offset+size])
2422
except errors.NoSuchFile:
2423
self._transport.mkdir(osutils.dirname(path))
2424
base = self._transport.append_bytes(path + '.knit',
2425
raw_data[offset:offset+size])
2429
result.append((key, base, size))
2432
def get_raw_records(self, memos_for_retrieval):
2433
"""Get the raw bytes for a records.
2435
:param memos_for_retrieval: An iterable containing the access memo for
2436
retrieving the bytes.
2437
:return: An iterator over the bytes of the records.
2439
# first pass, group into same-index request to minimise readv's issued.
2441
current_prefix = None
2442
for (key, offset, length) in memos_for_retrieval:
2443
if current_prefix == key[:-1]:
2444
current_list.append((offset, length))
2446
if current_prefix is not None:
2447
request_lists.append((current_prefix, current_list))
2448
current_prefix = key[:-1]
2449
current_list = [(offset, length)]
2450
# handle the last entry
2451
if current_prefix is not None:
2452
request_lists.append((current_prefix, current_list))
2453
for prefix, read_vector in request_lists:
2454
path = self._mapper.map(prefix) + '.knit'
2455
for pos, data in self._transport.readv(path, read_vector):
2459
class _DirectPackAccess(object):
2460
"""Access to data in one or more packs with less translation."""
2462
def __init__(self, index_to_packs, reload_func=None):
2463
"""Create a _DirectPackAccess object.
2465
:param index_to_packs: A dict mapping index objects to the transport
2466
and file names for obtaining data.
2467
:param reload_func: A function to call if we determine that the pack
2468
files have moved and we need to reload our caches. See
2469
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
2471
self._container_writer = None
2472
self._write_index = None
2473
self._indices = index_to_packs
2474
self._reload_func = reload_func
2476
def add_raw_records(self, key_sizes, raw_data):
2477
"""Add raw knit bytes to a storage area.
2479
The data is spooled to the container writer in one bytes-record per
2482
:param sizes: An iterable of tuples containing the key and size of each
2484
:param raw_data: A bytestring containing the data.
2485
:return: A list of memos to retrieve the record later. Each memo is an
2486
opaque index memo. For _DirectPackAccess the memo is (index, pos,
2487
length), where the index field is the write_index object supplied
2488
to the PackAccess object.
2490
if type(raw_data) != str:
2491
raise AssertionError(
2492
'data must be plain bytes was %s' % type(raw_data))
2495
for key, size in key_sizes:
2496
p_offset, p_length = self._container_writer.add_bytes_record(
2497
raw_data[offset:offset+size], [])
2499
result.append((self._write_index, p_offset, p_length))
2502
def get_raw_records(self, memos_for_retrieval):
2503
"""Get the raw bytes for a records.
2505
:param memos_for_retrieval: An iterable containing the (index, pos,
2506
length) memo for retrieving the bytes. The Pack access method
2507
looks up the pack to use for a given record in its index_to_pack
2509
:return: An iterator over the bytes of the records.
2511
# first pass, group into same-index requests
2513
current_index = None
2514
for (index, offset, length) in memos_for_retrieval:
2515
if current_index == index:
2516
current_list.append((offset, length))
2518
if current_index is not None:
2519
request_lists.append((current_index, current_list))
2520
current_index = index
2521
current_list = [(offset, length)]
2522
# handle the last entry
2523
if current_index is not None:
2524
request_lists.append((current_index, current_list))
2525
for index, offsets in request_lists:
2527
transport, path = self._indices[index]
2529
# A KeyError here indicates that someone has triggered an index
2530
# reload, and this index has gone missing, we need to start
2532
if self._reload_func is None:
2533
# If we don't have a _reload_func there is nothing that can
2536
raise errors.RetryWithNewPacks(reload_occurred=True,
2537
exc_info=sys.exc_info())
2539
reader = pack.make_readv_reader(transport, path, offsets)
2540
for names, read_func in reader.iter_records():
2541
yield read_func(None)
2542
except errors.NoSuchFile:
2543
# A NoSuchFile error indicates that a pack file has gone
2544
# missing on disk, we need to trigger a reload, and start over.
2545
if self._reload_func is None:
2547
raise errors.RetryWithNewPacks(reload_occurred=False,
2548
exc_info=sys.exc_info())
2550
def set_writer(self, writer, index, transport_packname):
2551
"""Set a writer to use for adding data."""
2552
if index is not None:
2553
self._indices[index] = transport_packname
2554
self._container_writer = writer
2555
self._write_index = index
2557
def reload_or_raise(self, retry_exc):
2558
"""Try calling the reload function, or re-raise the original exception.
2560
This should be called after _DirectPackAccess raises a
2561
RetryWithNewPacks exception. This function will handle the common logic
2562
of determining when the error is fatal versus being temporary.
2563
It will also make sure that the original exception is raised, rather
2564
than the RetryWithNewPacks exception.
2566
If this function returns, then the calling function should retry
2567
whatever operation was being performed. Otherwise an exception will
2570
:param retry_exc: A RetryWithNewPacks exception.
2573
if self._reload_func is None:
2575
elif not self._reload_func():
2576
# The reload claimed that nothing changed
2577
if not retry_exc.reload_occurred:
2578
# If there wasn't an earlier reload, then we really were
2579
# expecting to find changes. We didn't find them, so this is a
2583
exc_class, exc_value, exc_traceback = retry_exc.exc_info
2584
raise exc_class, exc_value, exc_traceback
2587
# Deprecated, use PatienceSequenceMatcher instead
2588
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2591
def annotate_knit(knit, revision_id):
2592
"""Annotate a knit with no cached annotations.
2594
This implementation is for knits with no cached annotations.
2595
It will work for knits with cached annotations, but this is not
2598
annotator = _KnitAnnotator(knit)
2599
return iter(annotator.annotate(revision_id))
2602
class _KnitAnnotator(object):
2603
"""Build up the annotations for a text."""
2605
def __init__(self, knit):
2608
# Content objects, differs from fulltexts because of how final newlines
2609
# are treated by knits. the content objects here will always have a
2611
self._fulltext_contents = {}
2613
# Annotated lines of specific revisions
2614
self._annotated_lines = {}
2616
# Track the raw data for nodes that we could not process yet.
2617
# This maps the revision_id of the base to a list of children that will
2618
# annotated from it.
2619
self._pending_children = {}
2621
# Nodes which cannot be extracted
2622
self._ghosts = set()
2624
# Track how many children this node has, so we know if we need to keep
2626
self._annotate_children = {}
2627
self._compression_children = {}
2629
self._all_build_details = {}
2630
# The children => parent revision_id graph
2631
self._revision_id_graph = {}
2633
self._heads_provider = None
2635
self._nodes_to_keep_annotations = set()
2636
self._generations_until_keep = 100
2638
def set_generations_until_keep(self, value):
2639
"""Set the number of generations before caching a node.
2641
Setting this to -1 will cache every merge node, setting this higher
2642
will cache fewer nodes.
2644
self._generations_until_keep = value
2646
def _add_fulltext_content(self, revision_id, content_obj):
2647
self._fulltext_contents[revision_id] = content_obj
2648
# TODO: jam 20080305 It might be good to check the sha1digest here
2649
return content_obj.text()
2651
def _check_parents(self, child, nodes_to_annotate):
2652
"""Check if all parents have been processed.
2654
:param child: A tuple of (rev_id, parents, raw_content)
2655
:param nodes_to_annotate: If child is ready, add it to
2656
nodes_to_annotate, otherwise put it back in self._pending_children
2658
for parent_id in child[1]:
2659
if (parent_id not in self._annotated_lines):
2660
# This parent is present, but another parent is missing
2661
self._pending_children.setdefault(parent_id,
2665
# This one is ready to be processed
2666
nodes_to_annotate.append(child)
2668
def _add_annotation(self, revision_id, fulltext, parent_ids,
2669
left_matching_blocks=None):
2670
"""Add an annotation entry.
2672
All parents should already have been annotated.
2673
:return: A list of children that now have their parents satisfied.
2675
a = self._annotated_lines
2676
annotated_parent_lines = [a[p] for p in parent_ids]
2677
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2678
fulltext, revision_id, left_matching_blocks,
2679
heads_provider=self._get_heads_provider()))
2680
self._annotated_lines[revision_id] = annotated_lines
2681
for p in parent_ids:
2682
ann_children = self._annotate_children[p]
2683
ann_children.remove(revision_id)
2684
if (not ann_children
2685
and p not in self._nodes_to_keep_annotations):
2686
del self._annotated_lines[p]
2687
del self._all_build_details[p]
2688
if p in self._fulltext_contents:
2689
del self._fulltext_contents[p]
2690
# Now that we've added this one, see if there are any pending
2691
# deltas to be done, certainly this parent is finished
2692
nodes_to_annotate = []
2693
for child in self._pending_children.pop(revision_id, []):
2694
self._check_parents(child, nodes_to_annotate)
2695
return nodes_to_annotate
2697
def _get_build_graph(self, key):
2698
"""Get the graphs for building texts and annotations.
2700
The data you need for creating a full text may be different than the
2701
data you need to annotate that text. (At a minimum, you need both
2702
parents to create an annotation, but only need 1 parent to generate the
2705
:return: A list of (key, index_memo) records, suitable for
2706
passing to read_records_iter to start reading in the raw data fro/
2709
if key in self._annotated_lines:
2712
pending = set([key])
2717
# get all pending nodes
2719
this_iteration = pending
2720
build_details = self._knit._index.get_build_details(this_iteration)
2721
self._all_build_details.update(build_details)
2722
# new_nodes = self._knit._index._get_entries(this_iteration)
2724
for key, details in build_details.iteritems():
2725
(index_memo, compression_parent, parents,
2726
record_details) = details
2727
self._revision_id_graph[key] = parents
2728
records.append((key, index_memo))
2729
# Do we actually need to check _annotated_lines?
2730
pending.update(p for p in parents
2731
if p not in self._all_build_details)
2732
if compression_parent:
2733
self._compression_children.setdefault(compression_parent,
2736
for parent in parents:
2737
self._annotate_children.setdefault(parent,
2739
num_gens = generation - kept_generation
2740
if ((num_gens >= self._generations_until_keep)
2741
and len(parents) > 1):
2742
kept_generation = generation
2743
self._nodes_to_keep_annotations.add(key)
2745
missing_versions = this_iteration.difference(build_details.keys())
2746
self._ghosts.update(missing_versions)
2747
for missing_version in missing_versions:
2748
# add a key, no parents
2749
self._revision_id_graph[missing_version] = ()
2750
pending.discard(missing_version) # don't look for it
2751
if self._ghosts.intersection(self._compression_children):
2753
"We cannot have nodes which have a ghost compression parent:\n"
2755
"compression children: %r"
2756
% (self._ghosts, self._compression_children))
2757
# Cleanout anything that depends on a ghost so that we don't wait for
2758
# the ghost to show up
2759
for node in self._ghosts:
2760
if node in self._annotate_children:
2761
# We won't be building this node
2762
del self._annotate_children[node]
2763
# Generally we will want to read the records in reverse order, because
2764
# we find the parent nodes after the children
2768
def _annotate_records(self, records):
2769
"""Build the annotations for the listed records."""
2770
# We iterate in the order read, rather than a strict order requested
2771
# However, process what we can, and put off to the side things that
2772
# still need parents, cleaning them up when those parents are
2774
for (rev_id, record,
2775
digest) in self._knit._read_records_iter(records):
2776
if rev_id in self._annotated_lines:
2778
parent_ids = self._revision_id_graph[rev_id]
2779
parent_ids = [p for p in parent_ids if p not in self._ghosts]
2780
details = self._all_build_details[rev_id]
2781
(index_memo, compression_parent, parents,
2782
record_details) = details
2783
nodes_to_annotate = []
2784
# TODO: Remove the punning between compression parents, and
2785
# parent_ids, we should be able to do this without assuming
2787
if len(parent_ids) == 0:
2788
# There are no parents for this node, so just add it
2789
# TODO: This probably needs to be decoupled
2790
fulltext_content, delta = self._knit._factory.parse_record(
2791
rev_id, record, record_details, None)
2792
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
2793
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
2794
parent_ids, left_matching_blocks=None))
2796
child = (rev_id, parent_ids, record)
2797
# Check if all the parents are present
2798
self._check_parents(child, nodes_to_annotate)
2799
while nodes_to_annotate:
2800
# Should we use a queue here instead of a stack?
2801
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
2802
(index_memo, compression_parent, parents,
2803
record_details) = self._all_build_details[rev_id]
2805
if compression_parent is not None:
2806
comp_children = self._compression_children[compression_parent]
2807
if rev_id not in comp_children:
2808
raise AssertionError("%r not in compression children %r"
2809
% (rev_id, comp_children))
2810
# If there is only 1 child, it is safe to reuse this
2812
reuse_content = (len(comp_children) == 1
2813
and compression_parent not in
2814
self._nodes_to_keep_annotations)
2816
# Remove it from the cache since it will be changing
2817
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
2818
# Make sure to copy the fulltext since it might be
2820
parent_fulltext = list(parent_fulltext_content.text())
2822
parent_fulltext_content = self._fulltext_contents[compression_parent]
2823
parent_fulltext = parent_fulltext_content.text()
2824
comp_children.remove(rev_id)
2825
fulltext_content, delta = self._knit._factory.parse_record(
2826
rev_id, record, record_details,
2827
parent_fulltext_content,
2828
copy_base_content=(not reuse_content))
2829
fulltext = self._add_fulltext_content(rev_id,
2831
if compression_parent == parent_ids[0]:
2832
# the compression_parent is the left parent, so we can
2834
blocks = KnitContent.get_line_delta_blocks(delta,
2835
parent_fulltext, fulltext)
2837
fulltext_content = self._knit._factory.parse_fulltext(
2839
fulltext = self._add_fulltext_content(rev_id,
2841
nodes_to_annotate.extend(
2842
self._add_annotation(rev_id, fulltext, parent_ids,
2843
left_matching_blocks=blocks))
2845
def _get_heads_provider(self):
2846
"""Create a heads provider for resolving ancestry issues."""
2847
if self._heads_provider is not None:
2848
return self._heads_provider
2849
parent_provider = _mod_graph.DictParentsProvider(
2850
self._revision_id_graph)
2851
graph_obj = _mod_graph.Graph(parent_provider)
2852
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
2853
self._heads_provider = head_cache
2856
def annotate(self, key):
2857
"""Return the annotated fulltext at the given key.
2859
:param key: The key to annotate.
2861
if len(self._knit._fallback_vfs) > 0:
2862
# stacked knits can't use the fast path at present.
2863
return self._simple_annotate(key)
2866
records = self._get_build_graph(key)
2867
if key in self._ghosts:
2868
raise errors.RevisionNotPresent(key, self._knit)
2869
self._annotate_records(records)
2870
return self._annotated_lines[key]
2871
except errors.RetryWithNewPacks, e:
2872
self._knit._access.reload_or_raise(e)
2873
# The cached build_details are no longer valid
2874
self._all_build_details.clear()
2876
def _simple_annotate(self, key):
2877
"""Return annotated fulltext, rediffing from the full texts.
2879
This is slow but makes no assumptions about the repository
2880
being able to produce line deltas.
2882
# TODO: this code generates a parent maps of present ancestors; it
2883
# could be split out into a separate method, and probably should use
2884
# iter_ancestry instead. -- mbp and robertc 20080704
2885
graph = _mod_graph.Graph(self._knit)
2886
head_cache = _mod_graph.FrozenHeadsCache(graph)
2887
search = graph._make_breadth_first_searcher([key])
2891
present, ghosts = search.next_with_ghosts()
2892
except StopIteration:
2894
keys.update(present)
2895
parent_map = self._knit.get_parent_map(keys)
2897
reannotate = annotate.reannotate
2898
for record in self._knit.get_record_stream(keys, 'topological', True):
2900
fulltext = split_lines(record.get_bytes_as('fulltext'))
2901
parents = parent_map[key]
2902
if parents is not None:
2903
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
2906
parent_cache[key] = list(
2907
reannotate(parent_lines, fulltext, key, None, head_cache))
2909
return parent_cache[key]
2911
raise errors.RevisionNotPresent(key, self._knit)
2915
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2917
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
