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
# We have been requested to return these records in an order that
1266
# suits us. So we ask the index to give us an optimally sorted
1268
for source, sub_keys in source_keys:
1269
if source is parent_maps[0]:
1270
# Only sort the keys for this VF
1271
self._index._sort_keys_by_io(sub_keys, positions)
1272
absent_keys = keys - set(global_map)
1273
for key in absent_keys:
1274
yield AbsentContentFactory(key)
1275
# restrict our view to the keys we can answer.
1276
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1277
# XXX: At that point we need to consider the impact of double reads by
1278
# utilising components multiple times.
1279
if include_delta_closure:
1280
# XXX: get_content_maps performs its own index queries; allow state
1282
non_local_keys = needed_from_fallback - absent_keys
1283
prefix_split_keys = self._split_by_prefix(present_keys)
1284
prefix_split_non_local_keys = self._split_by_prefix(non_local_keys)
1285
for prefix, keys in prefix_split_keys.iteritems():
1286
non_local = prefix_split_non_local_keys.get(prefix, [])
1287
non_local = set(non_local)
1288
text_map, _ = self._get_content_maps(keys, non_local)
1290
lines = text_map.pop(key)
1291
text = ''.join(lines)
1292
yield FulltextContentFactory(key, global_map[key], None,
1295
for source, keys in source_keys:
1296
if source is parent_maps[0]:
1297
# this KnitVersionedFiles
1298
records = [(key, positions[key][1]) for key in keys]
1299
for key, raw_data, sha1 in self._read_records_iter_raw(records):
1300
(record_details, index_memo, _) = positions[key]
1301
yield KnitContentFactory(key, global_map[key],
1302
record_details, sha1, raw_data, self._factory.annotated, None)
1304
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1305
for record in vf.get_record_stream(keys, ordering,
1306
include_delta_closure):
1309
def get_sha1s(self, keys):
1310
"""See VersionedFiles.get_sha1s()."""
1312
record_map = self._get_record_map(missing, allow_missing=True)
1314
for key, details in record_map.iteritems():
1315
if key not in missing:
1317
# record entry 2 is the 'digest'.
1318
result[key] = details[2]
1319
missing.difference_update(set(result))
1320
for source in self._fallback_vfs:
1323
new_result = source.get_sha1s(missing)
1324
result.update(new_result)
1325
missing.difference_update(set(new_result))
1328
def insert_record_stream(self, stream):
1329
"""Insert a record stream into this container.
1331
:param stream: A stream of records to insert.
1333
:seealso VersionedFiles.get_record_stream:
1335
def get_adapter(adapter_key):
1337
return adapters[adapter_key]
1339
adapter_factory = adapter_registry.get(adapter_key)
1340
adapter = adapter_factory(self)
1341
adapters[adapter_key] = adapter
1343
if self._factory.annotated:
1344
# self is annotated, we need annotated knits to use directly.
1345
annotated = "annotated-"
1348
# self is not annotated, but we can strip annotations cheaply.
1350
convertibles = set(["knit-annotated-ft-gz"])
1351
if self._max_delta_chain:
1352
convertibles.add("knit-annotated-delta-gz")
1353
# The set of types we can cheaply adapt without needing basis texts.
1354
native_types = set()
1355
if self._max_delta_chain:
1356
native_types.add("knit-%sdelta-gz" % annotated)
1357
native_types.add("knit-%sft-gz" % annotated)
1358
knit_types = native_types.union(convertibles)
1360
# Buffer all index entries that we can't add immediately because their
1361
# basis parent is missing. We don't buffer all because generating
1362
# annotations may require access to some of the new records. However we
1363
# can't generate annotations from new deltas until their basis parent
1364
# is present anyway, so we get away with not needing an index that
1365
# includes the new keys.
1367
# See <http://launchpad.net/bugs/300177> about ordering of compression
1368
# parents in the records - to be conservative, we insist that all
1369
# parents must be present to avoid expanding to a fulltext.
1371
# key = basis_parent, value = index entry to add
1372
buffered_index_entries = {}
1373
for record in stream:
1374
parents = record.parents
1375
# Raise an error when a record is missing.
1376
if record.storage_kind == 'absent':
1377
raise RevisionNotPresent([record.key], self)
1378
elif ((record.storage_kind in knit_types)
1380
or not self._fallback_vfs
1381
or not self._index.missing_keys(parents)
1382
or self.missing_keys(parents))):
1383
# we can insert the knit record literally if either it has no
1384
# compression parent OR we already have its basis in this kvf
1385
# OR the basis is not present even in the fallbacks. In the
1386
# last case it will either turn up later in the stream and all
1387
# will be well, or it won't turn up at all and we'll raise an
1390
# TODO: self.has_key is somewhat redundant with
1391
# self._index.has_key; we really want something that directly
1392
# asks if it's only present in the fallbacks. -- mbp 20081119
1393
if record.storage_kind not in native_types:
1395
adapter_key = (record.storage_kind, "knit-delta-gz")
1396
adapter = get_adapter(adapter_key)
1398
adapter_key = (record.storage_kind, "knit-ft-gz")
1399
adapter = get_adapter(adapter_key)
1400
bytes = adapter.get_bytes(
1401
record, record.get_bytes_as(record.storage_kind))
1403
bytes = record.get_bytes_as(record.storage_kind)
1404
options = [record._build_details[0]]
1405
if record._build_details[1]:
1406
options.append('no-eol')
1407
# Just blat it across.
1408
# Note: This does end up adding data on duplicate keys. As
1409
# modern repositories use atomic insertions this should not
1410
# lead to excessive growth in the event of interrupted fetches.
1411
# 'knit' repositories may suffer excessive growth, but as a
1412
# deprecated format this is tolerable. It can be fixed if
1413
# needed by in the kndx index support raising on a duplicate
1414
# add with identical parents and options.
1415
access_memo = self._access.add_raw_records(
1416
[(record.key, len(bytes))], bytes)[0]
1417
index_entry = (record.key, options, access_memo, parents)
1419
if 'fulltext' not in options:
1420
# Not a fulltext, so we need to make sure the compression
1421
# parent will also be present.
1422
# Note that pack backed knits don't need to buffer here
1423
# because they buffer all writes to the transaction level,
1424
# but we don't expose that difference at the index level. If
1425
# the query here has sufficient cost to show up in
1426
# profiling we should do that.
1428
# They're required to be physically in this
1429
# KnitVersionedFiles, not in a fallback.
1430
compression_parent = parents[0]
1431
if self.missing_keys([compression_parent]):
1432
pending = buffered_index_entries.setdefault(
1433
compression_parent, [])
1434
pending.append(index_entry)
1437
self._index.add_records([index_entry])
1438
elif record.storage_kind == 'fulltext':
1439
self.add_lines(record.key, parents,
1440
split_lines(record.get_bytes_as('fulltext')))
1442
# Not a fulltext, and not suitable for direct insertion as a
1443
# delta, either because it's not the right format, or this
1444
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1445
# 0) or because it depends on a base only present in the
1447
adapter_key = record.storage_kind, 'fulltext'
1448
adapter = get_adapter(adapter_key)
1449
lines = split_lines(adapter.get_bytes(
1450
record, record.get_bytes_as(record.storage_kind)))
1452
self.add_lines(record.key, parents, lines)
1453
except errors.RevisionAlreadyPresent:
1455
# Add any records whose basis parent is now available.
1456
added_keys = [record.key]
1458
key = added_keys.pop(0)
1459
if key in buffered_index_entries:
1460
index_entries = buffered_index_entries[key]
1461
self._index.add_records(index_entries)
1463
[index_entry[0] for index_entry in index_entries])
1464
del buffered_index_entries[key]
1465
# If there were any deltas which had a missing basis parent, error.
1466
if buffered_index_entries:
1467
from pprint import pformat
1468
raise errors.BzrCheckError(
1469
"record_stream refers to compression parents not in %r:\n%s"
1470
% (self, pformat(sorted(buffered_index_entries.keys()))))
1472
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1473
"""Iterate over the lines in the versioned files from keys.
1475
This may return lines from other keys. Each item the returned
1476
iterator yields is a tuple of a line and a text version that that line
1477
is present in (not introduced in).
1479
Ordering of results is in whatever order is most suitable for the
1480
underlying storage format.
1482
If a progress bar is supplied, it may be used to indicate progress.
1483
The caller is responsible for cleaning up progress bars (because this
1487
* Lines are normalised by the underlying store: they will all have \\n
1489
* Lines are returned in arbitrary order.
1490
* If a requested key did not change any lines (or didn't have any
1491
lines), it may not be mentioned at all in the result.
1493
:return: An iterator over (line, key).
1496
pb = progress.DummyProgress()
1502
# we don't care about inclusions, the caller cares.
1503
# but we need to setup a list of records to visit.
1504
# we need key, position, length
1506
build_details = self._index.get_build_details(keys)
1507
for key, details in build_details.iteritems():
1509
key_records.append((key, details[0]))
1510
records_iter = enumerate(self._read_records_iter(key_records))
1511
for (key_idx, (key, data, sha_value)) in records_iter:
1512
pb.update('Walking content.', key_idx, total)
1513
compression_parent = build_details[key][1]
1514
if compression_parent is None:
1516
line_iterator = self._factory.get_fulltext_content(data)
1519
line_iterator = self._factory.get_linedelta_content(data)
1520
# Now that we are yielding the data for this key, remove it
1523
# XXX: It might be more efficient to yield (key,
1524
# line_iterator) in the future. However for now, this is a
1525
# simpler change to integrate into the rest of the
1526
# codebase. RBC 20071110
1527
for line in line_iterator:
1530
except errors.RetryWithNewPacks, e:
1531
self._access.reload_or_raise(e)
1532
# If there are still keys we've not yet found, we look in the fallback
1533
# vfs, and hope to find them there. Note that if the keys are found
1534
# but had no changes or no content, the fallback may not return
1536
if keys and not self._fallback_vfs:
1537
# XXX: strictly the second parameter is meant to be the file id
1538
# but it's not easily accessible here.
1539
raise RevisionNotPresent(keys, repr(self))
1540
for source in self._fallback_vfs:
1544
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1545
source_keys.add(key)
1547
keys.difference_update(source_keys)
1548
pb.update('Walking content.', total, total)
1550
def _make_line_delta(self, delta_seq, new_content):
1551
"""Generate a line delta from delta_seq and new_content."""
1553
for op in delta_seq.get_opcodes():
1554
if op[0] == 'equal':
1556
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1559
def _merge_annotations(self, content, parents, parent_texts={},
1560
delta=None, annotated=None,
1561
left_matching_blocks=None):
1562
"""Merge annotations for content and generate deltas.
1564
This is done by comparing the annotations based on changes to the text
1565
and generating a delta on the resulting full texts. If annotations are
1566
not being created then a simple delta is created.
1568
if left_matching_blocks is not None:
1569
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1573
for parent_key in parents:
1574
merge_content = self._get_content(parent_key, parent_texts)
1575
if (parent_key == parents[0] and delta_seq is not None):
1578
seq = patiencediff.PatienceSequenceMatcher(
1579
None, merge_content.text(), content.text())
1580
for i, j, n in seq.get_matching_blocks():
1583
# this copies (origin, text) pairs across to the new
1584
# content for any line that matches the last-checked
1586
content._lines[j:j+n] = merge_content._lines[i:i+n]
1587
# XXX: Robert says the following block is a workaround for a
1588
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1589
if content._lines and content._lines[-1][1][-1] != '\n':
1590
# The copied annotation was from a line without a trailing EOL,
1591
# reinstate one for the content object, to ensure correct
1593
line = content._lines[-1][1] + '\n'
1594
content._lines[-1] = (content._lines[-1][0], line)
1596
if delta_seq is None:
1597
reference_content = self._get_content(parents[0], parent_texts)
1598
new_texts = content.text()
1599
old_texts = reference_content.text()
1600
delta_seq = patiencediff.PatienceSequenceMatcher(
1601
None, old_texts, new_texts)
1602
return self._make_line_delta(delta_seq, content)
1604
def _parse_record(self, version_id, data):
1605
"""Parse an original format knit record.
1607
These have the last element of the key only present in the stored data.
1609
rec, record_contents = self._parse_record_unchecked(data)
1610
self._check_header_version(rec, version_id)
1611
return record_contents, rec[3]
1613
def _parse_record_header(self, key, raw_data):
1614
"""Parse a record header for consistency.
1616
:return: the header and the decompressor stream.
1617
as (stream, header_record)
1619
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1622
rec = self._check_header(key, df.readline())
1623
except Exception, e:
1624
raise KnitCorrupt(self,
1625
"While reading {%s} got %s(%s)"
1626
% (key, e.__class__.__name__, str(e)))
1629
def _parse_record_unchecked(self, data):
1631
# 4168 calls in 2880 217 internal
1632
# 4168 calls to _parse_record_header in 2121
1633
# 4168 calls to readlines in 330
1634
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1636
record_contents = df.readlines()
1637
except Exception, e:
1638
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1639
(data, e.__class__.__name__, str(e)))
1640
header = record_contents.pop(0)
1641
rec = self._split_header(header)
1642
last_line = record_contents.pop()
1643
if len(record_contents) != int(rec[2]):
1644
raise KnitCorrupt(self,
1645
'incorrect number of lines %s != %s'
1646
' for version {%s} %s'
1647
% (len(record_contents), int(rec[2]),
1648
rec[1], record_contents))
1649
if last_line != 'end %s\n' % rec[1]:
1650
raise KnitCorrupt(self,
1651
'unexpected version end line %r, wanted %r'
1652
% (last_line, rec[1]))
1654
return rec, record_contents
1656
def _read_records_iter(self, records):
1657
"""Read text records from data file and yield result.
1659
The result will be returned in whatever is the fastest to read.
1660
Not by the order requested. Also, multiple requests for the same
1661
record will only yield 1 response.
1662
:param records: A list of (key, access_memo) entries
1663
:return: Yields (key, contents, digest) in the order
1664
read, not the order requested
1669
# XXX: This smells wrong, IO may not be getting ordered right.
1670
needed_records = sorted(set(records), key=operator.itemgetter(1))
1671
if not needed_records:
1674
# The transport optimizes the fetching as well
1675
# (ie, reads continuous ranges.)
1676
raw_data = self._access.get_raw_records(
1677
[index_memo for key, index_memo in needed_records])
1679
for (key, index_memo), data in \
1680
izip(iter(needed_records), raw_data):
1681
content, digest = self._parse_record(key[-1], data)
1682
yield key, content, digest
1684
def _read_records_iter_raw(self, records):
1685
"""Read text records from data file and yield raw data.
1687
This unpacks enough of the text record to validate the id is
1688
as expected but thats all.
1690
Each item the iterator yields is (key, bytes, sha1_of_full_text).
1692
# setup an iterator of the external records:
1693
# uses readv so nice and fast we hope.
1695
# grab the disk data needed.
1696
needed_offsets = [index_memo for key, index_memo
1698
raw_records = self._access.get_raw_records(needed_offsets)
1700
for key, index_memo in records:
1701
data = raw_records.next()
1702
# validate the header (note that we can only use the suffix in
1703
# current knit records).
1704
df, rec = self._parse_record_header(key, data)
1706
yield key, data, rec[3]
1708
def _record_to_data(self, key, digest, lines, dense_lines=None):
1709
"""Convert key, digest, lines into a raw data block.
1711
:param key: The key of the record. Currently keys are always serialised
1712
using just the trailing component.
1713
:param dense_lines: The bytes of lines but in a denser form. For
1714
instance, if lines is a list of 1000 bytestrings each ending in \n,
1715
dense_lines may be a list with one line in it, containing all the
1716
1000's lines and their \n's. Using dense_lines if it is already
1717
known is a win because the string join to create bytes in this
1718
function spends less time resizing the final string.
1719
:return: (len, a StringIO instance with the raw data ready to read.)
1721
# Note: using a string copy here increases memory pressure with e.g.
1722
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
1723
# when doing the initial commit of a mozilla tree. RBC 20070921
1724
bytes = ''.join(chain(
1725
["version %s %d %s\n" % (key[-1],
1728
dense_lines or lines,
1729
["end %s\n" % key[-1]]))
1730
if type(bytes) != str:
1731
raise AssertionError(
1732
'data must be plain bytes was %s' % type(bytes))
1733
if lines and lines[-1][-1] != '\n':
1734
raise ValueError('corrupt lines value %r' % lines)
1735
compressed_bytes = tuned_gzip.bytes_to_gzip(bytes)
1736
return len(compressed_bytes), compressed_bytes
1738
def _split_header(self, line):
1741
raise KnitCorrupt(self,
1742
'unexpected number of elements in record header')
1746
"""See VersionedFiles.keys."""
1747
if 'evil' in debug.debug_flags:
1748
trace.mutter_callsite(2, "keys scales with size of history")
1749
sources = [self._index] + self._fallback_vfs
1751
for source in sources:
1752
result.update(source.keys())
1756
class _KndxIndex(object):
1757
"""Manages knit index files
1759
The index is kept in memory and read on startup, to enable
1760
fast lookups of revision information. The cursor of the index
1761
file is always pointing to the end, making it easy to append
1764
_cache is a cache for fast mapping from version id to a Index
1767
_history is a cache for fast mapping from indexes to version ids.
1769
The index data format is dictionary compressed when it comes to
1770
parent references; a index entry may only have parents that with a
1771
lover index number. As a result, the index is topological sorted.
1773
Duplicate entries may be written to the index for a single version id
1774
if this is done then the latter one completely replaces the former:
1775
this allows updates to correct version and parent information.
1776
Note that the two entries may share the delta, and that successive
1777
annotations and references MUST point to the first entry.
1779
The index file on disc contains a header, followed by one line per knit
1780
record. The same revision can be present in an index file more than once.
1781
The first occurrence gets assigned a sequence number starting from 0.
1783
The format of a single line is
1784
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1785
REVISION_ID is a utf8-encoded revision id
1786
FLAGS is a comma separated list of flags about the record. Values include
1787
no-eol, line-delta, fulltext.
1788
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1789
that the the compressed data starts at.
1790
LENGTH is the ascii representation of the length of the data file.
1791
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1793
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1794
revision id already in the knit that is a parent of REVISION_ID.
1795
The ' :' marker is the end of record marker.
1798
when a write is interrupted to the index file, it will result in a line
1799
that does not end in ' :'. If the ' :' is not present at the end of a line,
1800
or at the end of the file, then the record that is missing it will be
1801
ignored by the parser.
1803
When writing new records to the index file, the data is preceded by '\n'
1804
to ensure that records always start on new lines even if the last write was
1805
interrupted. As a result its normal for the last line in the index to be
1806
missing a trailing newline. One can be added with no harmful effects.
1808
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
1809
where prefix is e.g. the (fileid,) for .texts instances or () for
1810
constant-mapped things like .revisions, and the old state is
1811
tuple(history_vector, cache_dict). This is used to prevent having an
1812
ABI change with the C extension that reads .kndx files.
1815
HEADER = "# bzr knit index 8\n"
1817
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
1818
"""Create a _KndxIndex on transport using mapper."""
1819
self._transport = transport
1820
self._mapper = mapper
1821
self._get_scope = get_scope
1822
self._allow_writes = allow_writes
1823
self._is_locked = is_locked
1825
self.has_graph = True
1827
def add_records(self, records, random_id=False):
1828
"""Add multiple records to the index.
1830
:param records: a list of tuples:
1831
(key, options, access_memo, parents).
1832
:param random_id: If True the ids being added were randomly generated
1833
and no check for existence will be performed.
1836
for record in records:
1839
path = self._mapper.map(key) + '.kndx'
1840
path_keys = paths.setdefault(path, (prefix, []))
1841
path_keys[1].append(record)
1842
for path in sorted(paths):
1843
prefix, path_keys = paths[path]
1844
self._load_prefixes([prefix])
1846
orig_history = self._kndx_cache[prefix][1][:]
1847
orig_cache = self._kndx_cache[prefix][0].copy()
1850
for key, options, (_, pos, size), parents in path_keys:
1852
# kndx indices cannot be parentless.
1854
line = "\n%s %s %s %s %s :" % (
1855
key[-1], ','.join(options), pos, size,
1856
self._dictionary_compress(parents))
1857
if type(line) != str:
1858
raise AssertionError(
1859
'data must be utf8 was %s' % type(line))
1861
self._cache_key(key, options, pos, size, parents)
1862
if len(orig_history):
1863
self._transport.append_bytes(path, ''.join(lines))
1865
self._init_index(path, lines)
1867
# If any problems happen, restore the original values and re-raise
1868
self._kndx_cache[prefix] = (orig_cache, orig_history)
1871
def _cache_key(self, key, options, pos, size, parent_keys):
1872
"""Cache a version record in the history array and index cache.
1874
This is inlined into _load_data for performance. KEEP IN SYNC.
1875
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1879
version_id = key[-1]
1880
# last-element only for compatibilty with the C load_data.
1881
parents = tuple(parent[-1] for parent in parent_keys)
1882
for parent in parent_keys:
1883
if parent[:-1] != prefix:
1884
raise ValueError("mismatched prefixes for %r, %r" % (
1886
cache, history = self._kndx_cache[prefix]
1887
# only want the _history index to reference the 1st index entry
1889
if version_id not in cache:
1890
index = len(history)
1891
history.append(version_id)
1893
index = cache[version_id][5]
1894
cache[version_id] = (version_id,
1901
def check_header(self, fp):
1902
line = fp.readline()
1904
# An empty file can actually be treated as though the file doesn't
1906
raise errors.NoSuchFile(self)
1907
if line != self.HEADER:
1908
raise KnitHeaderError(badline=line, filename=self)
1910
def _check_read(self):
1911
if not self._is_locked():
1912
raise errors.ObjectNotLocked(self)
1913
if self._get_scope() != self._scope:
1916
def _check_write_ok(self):
1917
"""Assert if not writes are permitted."""
1918
if not self._is_locked():
1919
raise errors.ObjectNotLocked(self)
1920
if self._get_scope() != self._scope:
1922
if self._mode != 'w':
1923
raise errors.ReadOnlyObjectDirtiedError(self)
1925
def get_build_details(self, keys):
1926
"""Get the method, index_memo and compression parent for keys.
1928
Ghosts are omitted from the result.
1930
:param keys: An iterable of keys.
1931
:return: A dict of key:(index_memo, compression_parent, parents,
1934
opaque structure to pass to read_records to extract the raw
1937
Content that this record is built upon, may be None
1939
Logical parents of this node
1941
extra information about the content which needs to be passed to
1942
Factory.parse_record
1944
parent_map = self.get_parent_map(keys)
1947
if key not in parent_map:
1949
method = self.get_method(key)
1950
parents = parent_map[key]
1951
if method == 'fulltext':
1952
compression_parent = None
1954
compression_parent = parents[0]
1955
noeol = 'no-eol' in self.get_options(key)
1956
index_memo = self.get_position(key)
1957
result[key] = (index_memo, compression_parent,
1958
parents, (method, noeol))
1961
def get_method(self, key):
1962
"""Return compression method of specified key."""
1963
options = self.get_options(key)
1964
if 'fulltext' in options:
1966
elif 'line-delta' in options:
1969
raise errors.KnitIndexUnknownMethod(self, options)
1971
def get_options(self, key):
1972
"""Return a list representing options.
1976
prefix, suffix = self._split_key(key)
1977
self._load_prefixes([prefix])
1979
return self._kndx_cache[prefix][0][suffix][1]
1981
raise RevisionNotPresent(key, self)
1983
def get_parent_map(self, keys):
1984
"""Get a map of the parents of keys.
1986
:param keys: The keys to look up parents for.
1987
:return: A mapping from keys to parents. Absent keys are absent from
1990
# Parse what we need to up front, this potentially trades off I/O
1991
# locality (.kndx and .knit in the same block group for the same file
1992
# id) for less checking in inner loops.
1993
prefixes = set(key[:-1] for key in keys)
1994
self._load_prefixes(prefixes)
1999
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2003
result[key] = tuple(prefix + (suffix,) for
2004
suffix in suffix_parents)
2007
def get_position(self, key):
2008
"""Return details needed to access the version.
2010
:return: a tuple (key, data position, size) to hand to the access
2011
logic to get the record.
2013
prefix, suffix = self._split_key(key)
2014
self._load_prefixes([prefix])
2015
entry = self._kndx_cache[prefix][0][suffix]
2016
return key, entry[2], entry[3]
2018
has_key = _mod_index._has_key_from_parent_map
2020
def _init_index(self, path, extra_lines=[]):
2021
"""Initialize an index."""
2023
sio.write(self.HEADER)
2024
sio.writelines(extra_lines)
2026
self._transport.put_file_non_atomic(path, sio,
2027
create_parent_dir=True)
2028
# self._create_parent_dir)
2029
# mode=self._file_mode,
2030
# dir_mode=self._dir_mode)
2033
"""Get all the keys in the collection.
2035
The keys are not ordered.
2038
# Identify all key prefixes.
2039
# XXX: A bit hacky, needs polish.
2040
if type(self._mapper) == ConstantMapper:
2044
for quoted_relpath in self._transport.iter_files_recursive():
2045
path, ext = os.path.splitext(quoted_relpath)
2047
prefixes = [self._mapper.unmap(path) for path in relpaths]
2048
self._load_prefixes(prefixes)
2049
for prefix in prefixes:
2050
for suffix in self._kndx_cache[prefix][1]:
2051
result.add(prefix + (suffix,))
2054
def _load_prefixes(self, prefixes):
2055
"""Load the indices for prefixes."""
2057
for prefix in prefixes:
2058
if prefix not in self._kndx_cache:
2059
# the load_data interface writes to these variables.
2062
self._filename = prefix
2064
path = self._mapper.map(prefix) + '.kndx'
2065
fp = self._transport.get(path)
2067
# _load_data may raise NoSuchFile if the target knit is
2069
_load_data(self, fp)
2072
self._kndx_cache[prefix] = (self._cache, self._history)
2077
self._kndx_cache[prefix] = ({}, [])
2078
if type(self._mapper) == ConstantMapper:
2079
# preserve behaviour for revisions.kndx etc.
2080
self._init_index(path)
2085
missing_keys = _mod_index._missing_keys_from_parent_map
2087
def _partition_keys(self, keys):
2088
"""Turn keys into a dict of prefix:suffix_list."""
2091
prefix_keys = result.setdefault(key[:-1], [])
2092
prefix_keys.append(key[-1])
2095
def _dictionary_compress(self, keys):
2096
"""Dictionary compress keys.
2098
:param keys: The keys to generate references to.
2099
:return: A string representation of keys. keys which are present are
2100
dictionary compressed, and others are emitted as fulltext with a
2106
prefix = keys[0][:-1]
2107
cache = self._kndx_cache[prefix][0]
2109
if key[:-1] != prefix:
2110
# kndx indices cannot refer across partitioned storage.
2111
raise ValueError("mismatched prefixes for %r" % keys)
2112
if key[-1] in cache:
2113
# -- inlined lookup() --
2114
result_list.append(str(cache[key[-1]][5]))
2115
# -- end lookup () --
2117
result_list.append('.' + key[-1])
2118
return ' '.join(result_list)
2120
def _reset_cache(self):
2121
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2122
# (cache_dict, history_vector) for parsed kndx files.
2123
self._kndx_cache = {}
2124
self._scope = self._get_scope()
2125
allow_writes = self._allow_writes()
2131
def _sort_keys_by_io(self, keys, positions):
2132
"""Figure out an optimal order to read the records for the given keys.
2134
Sort keys, grouped by index and sorted by position.
2136
:param keys: A list of keys whose records we want to read. This will be
2138
:param positions: A dict, such as the one returned by
2139
_get_components_positions()
2142
def get_sort_key(key):
2143
index_memo = positions[key][1]
2144
# Group by prefix and position. index_memo[0] is the key, so it is
2145
# (file_id, revision_id) and we don't want to sort on revision_id,
2146
# index_memo[1] is the position, and index_memo[2] is the size,
2147
# which doesn't matter for the sort
2148
return index_memo[0][:-1], index_memo[1]
2149
return keys.sort(key=get_sort_key)
2151
def _split_key(self, key):
2152
"""Split key into a prefix and suffix."""
2153
return key[:-1], key[-1]
2156
class _KnitGraphIndex(object):
2157
"""A KnitVersionedFiles index layered on GraphIndex."""
2159
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2161
"""Construct a KnitGraphIndex on a graph_index.
2163
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2164
:param is_locked: A callback to check whether the object should answer
2166
:param deltas: Allow delta-compressed records.
2167
:param parents: If True, record knits parents, if not do not record
2169
:param add_callback: If not None, allow additions to the index and call
2170
this callback with a list of added GraphIndex nodes:
2171
[(node, value, node_refs), ...]
2172
:param is_locked: A callback, returns True if the index is locked and
2175
self._add_callback = add_callback
2176
self._graph_index = graph_index
2177
self._deltas = deltas
2178
self._parents = parents
2179
if deltas and not parents:
2180
# XXX: TODO: Delta tree and parent graph should be conceptually
2182
raise KnitCorrupt(self, "Cannot do delta compression without "
2184
self.has_graph = parents
2185
self._is_locked = is_locked
2188
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2190
def add_records(self, records, random_id=False):
2191
"""Add multiple records to the index.
2193
This function does not insert data into the Immutable GraphIndex
2194
backing the KnitGraphIndex, instead it prepares data for insertion by
2195
the caller and checks that it is safe to insert then calls
2196
self._add_callback with the prepared GraphIndex nodes.
2198
:param records: a list of tuples:
2199
(key, options, access_memo, parents).
2200
:param random_id: If True the ids being added were randomly generated
2201
and no check for existence will be performed.
2203
if not self._add_callback:
2204
raise errors.ReadOnlyError(self)
2205
# we hope there are no repositories with inconsistent parentage
2209
for (key, options, access_memo, parents) in records:
2211
parents = tuple(parents)
2212
index, pos, size = access_memo
2213
if 'no-eol' in options:
2217
value += "%d %d" % (pos, size)
2218
if not self._deltas:
2219
if 'line-delta' in options:
2220
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2223
if 'line-delta' in options:
2224
node_refs = (parents, (parents[0],))
2226
node_refs = (parents, ())
2228
node_refs = (parents, )
2231
raise KnitCorrupt(self, "attempt to add node with parents "
2232
"in parentless index.")
2234
keys[key] = (value, node_refs)
2237
present_nodes = self._get_entries(keys)
2238
for (index, key, value, node_refs) in present_nodes:
2239
if (value[0] != keys[key][0][0] or
2240
node_refs != keys[key][1]):
2241
raise KnitCorrupt(self, "inconsistent details in add_records"
2242
": %s %s" % ((value, node_refs), keys[key]))
2246
for key, (value, node_refs) in keys.iteritems():
2247
result.append((key, value, node_refs))
2249
for key, (value, node_refs) in keys.iteritems():
2250
result.append((key, value))
2251
self._add_callback(result)
2253
def _check_read(self):
2254
"""raise if reads are not permitted."""
2255
if not self._is_locked():
2256
raise errors.ObjectNotLocked(self)
2258
def _check_write_ok(self):
2259
"""Assert if writes are not permitted."""
2260
if not self._is_locked():
2261
raise errors.ObjectNotLocked(self)
2263
def _compression_parent(self, an_entry):
2264
# return the key that an_entry is compressed against, or None
2265
# Grab the second parent list (as deltas implies parents currently)
2266
compression_parents = an_entry[3][1]
2267
if not compression_parents:
2269
if len(compression_parents) != 1:
2270
raise AssertionError(
2271
"Too many compression parents: %r" % compression_parents)
2272
return compression_parents[0]
2274
def get_build_details(self, keys):
2275
"""Get the method, index_memo and compression parent for version_ids.
2277
Ghosts are omitted from the result.
2279
:param keys: An iterable of keys.
2280
:return: A dict of key:
2281
(index_memo, compression_parent, parents, record_details).
2283
opaque structure to pass to read_records to extract the raw
2286
Content that this record is built upon, may be None
2288
Logical parents of this node
2290
extra information about the content which needs to be passed to
2291
Factory.parse_record
2295
entries = self._get_entries(keys, False)
2296
for entry in entries:
2298
if not self._parents:
2301
parents = entry[3][0]
2302
if not self._deltas:
2303
compression_parent_key = None
2305
compression_parent_key = self._compression_parent(entry)
2306
noeol = (entry[2][0] == 'N')
2307
if compression_parent_key:
2308
method = 'line-delta'
2311
result[key] = (self._node_to_position(entry),
2312
compression_parent_key, parents,
2316
def _get_entries(self, keys, check_present=False):
2317
"""Get the entries for keys.
2319
:param keys: An iterable of index key tuples.
2324
for node in self._graph_index.iter_entries(keys):
2326
found_keys.add(node[1])
2328
# adapt parentless index to the rest of the code.
2329
for node in self._graph_index.iter_entries(keys):
2330
yield node[0], node[1], node[2], ()
2331
found_keys.add(node[1])
2333
missing_keys = keys.difference(found_keys)
2335
raise RevisionNotPresent(missing_keys.pop(), self)
2337
def get_method(self, key):
2338
"""Return compression method of specified key."""
2339
return self._get_method(self._get_node(key))
2341
def _get_method(self, node):
2342
if not self._deltas:
2344
if self._compression_parent(node):
2349
def _get_node(self, key):
2351
return list(self._get_entries([key]))[0]
2353
raise RevisionNotPresent(key, self)
2355
def get_options(self, key):
2356
"""Return a list representing options.
2360
node = self._get_node(key)
2361
options = [self._get_method(node)]
2362
if node[2][0] == 'N':
2363
options.append('no-eol')
2366
def get_parent_map(self, keys):
2367
"""Get a map of the parents of keys.
2369
:param keys: The keys to look up parents for.
2370
:return: A mapping from keys to parents. Absent keys are absent from
2374
nodes = self._get_entries(keys)
2378
result[node[1]] = node[3][0]
2381
result[node[1]] = None
2384
def get_position(self, key):
2385
"""Return details needed to access the version.
2387
:return: a tuple (index, data position, size) to hand to the access
2388
logic to get the record.
2390
node = self._get_node(key)
2391
return self._node_to_position(node)
2393
has_key = _mod_index._has_key_from_parent_map
2396
"""Get all the keys in the collection.
2398
The keys are not ordered.
2401
return [node[1] for node in self._graph_index.iter_all_entries()]
2403
missing_keys = _mod_index._missing_keys_from_parent_map
2405
def _node_to_position(self, node):
2406
"""Convert an index value to position details."""
2407
bits = node[2][1:].split(' ')
2408
return node[0], int(bits[0]), int(bits[1])
2410
def _sort_keys_by_io(self, keys, positions):
2411
"""Figure out an optimal order to read the records for the given keys.
2413
Sort keys, grouped by index and sorted by position.
2415
:param keys: A list of keys whose records we want to read. This will be
2417
:param positions: A dict, such as the one returned by
2418
_get_components_positions()
2421
def get_index_memo(key):
2422
# index_memo is at offset [1]. It is made up of (GraphIndex,
2423
# position, size). GI is an object, which will be unique for each
2424
# pack file. This causes us to group by pack file, then sort by
2425
# position. Size doesn't matter, but it isn't worth breaking up the
2427
return positions[key][1]
2428
return keys.sort(key=get_index_memo)
2431
class _KnitKeyAccess(object):
2432
"""Access to records in .knit files."""
2434
def __init__(self, transport, mapper):
2435
"""Create a _KnitKeyAccess with transport and mapper.
2437
:param transport: The transport the access object is rooted at.
2438
:param mapper: The mapper used to map keys to .knit files.
2440
self._transport = transport
2441
self._mapper = mapper
2443
def add_raw_records(self, key_sizes, raw_data):
2444
"""Add raw knit bytes to a storage area.
2446
The data is spooled to the container writer in one bytes-record per
2449
:param sizes: An iterable of tuples containing the key and size of each
2451
:param raw_data: A bytestring containing the data.
2452
:return: A list of memos to retrieve the record later. Each memo is an
2453
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
2454
length), where the key is the record key.
2456
if type(raw_data) != str:
2457
raise AssertionError(
2458
'data must be plain bytes was %s' % type(raw_data))
2461
# TODO: This can be tuned for writing to sftp and other servers where
2462
# append() is relatively expensive by grouping the writes to each key
2464
for key, size in key_sizes:
2465
path = self._mapper.map(key)
2467
base = self._transport.append_bytes(path + '.knit',
2468
raw_data[offset:offset+size])
2469
except errors.NoSuchFile:
2470
self._transport.mkdir(osutils.dirname(path))
2471
base = self._transport.append_bytes(path + '.knit',
2472
raw_data[offset:offset+size])
2476
result.append((key, base, size))
2479
def get_raw_records(self, memos_for_retrieval):
2480
"""Get the raw bytes for a records.
2482
:param memos_for_retrieval: An iterable containing the access memo for
2483
retrieving the bytes.
2484
:return: An iterator over the bytes of the records.
2486
# first pass, group into same-index request to minimise readv's issued.
2488
current_prefix = None
2489
for (key, offset, length) in memos_for_retrieval:
2490
if current_prefix == key[:-1]:
2491
current_list.append((offset, length))
2493
if current_prefix is not None:
2494
request_lists.append((current_prefix, current_list))
2495
current_prefix = key[:-1]
2496
current_list = [(offset, length)]
2497
# handle the last entry
2498
if current_prefix is not None:
2499
request_lists.append((current_prefix, current_list))
2500
for prefix, read_vector in request_lists:
2501
path = self._mapper.map(prefix) + '.knit'
2502
for pos, data in self._transport.readv(path, read_vector):
2506
class _DirectPackAccess(object):
2507
"""Access to data in one or more packs with less translation."""
2509
def __init__(self, index_to_packs, reload_func=None):
2510
"""Create a _DirectPackAccess object.
2512
:param index_to_packs: A dict mapping index objects to the transport
2513
and file names for obtaining data.
2514
:param reload_func: A function to call if we determine that the pack
2515
files have moved and we need to reload our caches. See
2516
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
2518
self._container_writer = None
2519
self._write_index = None
2520
self._indices = index_to_packs
2521
self._reload_func = reload_func
2523
def add_raw_records(self, key_sizes, raw_data):
2524
"""Add raw knit bytes to a storage area.
2526
The data is spooled to the container writer in one bytes-record per
2529
:param sizes: An iterable of tuples containing the key and size of each
2531
:param raw_data: A bytestring containing the data.
2532
:return: A list of memos to retrieve the record later. Each memo is an
2533
opaque index memo. For _DirectPackAccess the memo is (index, pos,
2534
length), where the index field is the write_index object supplied
2535
to the PackAccess object.
2537
if type(raw_data) != str:
2538
raise AssertionError(
2539
'data must be plain bytes was %s' % type(raw_data))
2542
for key, size in key_sizes:
2543
p_offset, p_length = self._container_writer.add_bytes_record(
2544
raw_data[offset:offset+size], [])
2546
result.append((self._write_index, p_offset, p_length))
2549
def get_raw_records(self, memos_for_retrieval):
2550
"""Get the raw bytes for a records.
2552
:param memos_for_retrieval: An iterable containing the (index, pos,
2553
length) memo for retrieving the bytes. The Pack access method
2554
looks up the pack to use for a given record in its index_to_pack
2556
:return: An iterator over the bytes of the records.
2558
# first pass, group into same-index requests
2560
current_index = None
2561
for (index, offset, length) in memos_for_retrieval:
2562
if current_index == index:
2563
current_list.append((offset, length))
2565
if current_index is not None:
2566
request_lists.append((current_index, current_list))
2567
current_index = index
2568
current_list = [(offset, length)]
2569
# handle the last entry
2570
if current_index is not None:
2571
request_lists.append((current_index, current_list))
2572
for index, offsets in request_lists:
2574
transport, path = self._indices[index]
2576
# A KeyError here indicates that someone has triggered an index
2577
# reload, and this index has gone missing, we need to start
2579
if self._reload_func is None:
2580
# If we don't have a _reload_func there is nothing that can
2583
raise errors.RetryWithNewPacks(reload_occurred=True,
2584
exc_info=sys.exc_info())
2586
reader = pack.make_readv_reader(transport, path, offsets)
2587
for names, read_func in reader.iter_records():
2588
yield read_func(None)
2589
except errors.NoSuchFile:
2590
# A NoSuchFile error indicates that a pack file has gone
2591
# missing on disk, we need to trigger a reload, and start over.
2592
if self._reload_func is None:
2594
raise errors.RetryWithNewPacks(reload_occurred=False,
2595
exc_info=sys.exc_info())
2597
def set_writer(self, writer, index, transport_packname):
2598
"""Set a writer to use for adding data."""
2599
if index is not None:
2600
self._indices[index] = transport_packname
2601
self._container_writer = writer
2602
self._write_index = index
2604
def reload_or_raise(self, retry_exc):
2605
"""Try calling the reload function, or re-raise the original exception.
2607
This should be called after _DirectPackAccess raises a
2608
RetryWithNewPacks exception. This function will handle the common logic
2609
of determining when the error is fatal versus being temporary.
2610
It will also make sure that the original exception is raised, rather
2611
than the RetryWithNewPacks exception.
2613
If this function returns, then the calling function should retry
2614
whatever operation was being performed. Otherwise an exception will
2617
:param retry_exc: A RetryWithNewPacks exception.
2620
if self._reload_func is None:
2622
elif not self._reload_func():
2623
# The reload claimed that nothing changed
2624
if not retry_exc.reload_occurred:
2625
# If there wasn't an earlier reload, then we really were
2626
# expecting to find changes. We didn't find them, so this is a
2630
exc_class, exc_value, exc_traceback = retry_exc.exc_info
2631
raise exc_class, exc_value, exc_traceback
2634
# Deprecated, use PatienceSequenceMatcher instead
2635
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2638
def annotate_knit(knit, revision_id):
2639
"""Annotate a knit with no cached annotations.
2641
This implementation is for knits with no cached annotations.
2642
It will work for knits with cached annotations, but this is not
2645
annotator = _KnitAnnotator(knit)
2646
return iter(annotator.annotate(revision_id))
2649
class _KnitAnnotator(object):
2650
"""Build up the annotations for a text."""
2652
def __init__(self, knit):
2655
# Content objects, differs from fulltexts because of how final newlines
2656
# are treated by knits. the content objects here will always have a
2658
self._fulltext_contents = {}
2660
# Annotated lines of specific revisions
2661
self._annotated_lines = {}
2663
# Track the raw data for nodes that we could not process yet.
2664
# This maps the revision_id of the base to a list of children that will
2665
# annotated from it.
2666
self._pending_children = {}
2668
# Nodes which cannot be extracted
2669
self._ghosts = set()
2671
# Track how many children this node has, so we know if we need to keep
2673
self._annotate_children = {}
2674
self._compression_children = {}
2676
self._all_build_details = {}
2677
# The children => parent revision_id graph
2678
self._revision_id_graph = {}
2680
self._heads_provider = None
2682
self._nodes_to_keep_annotations = set()
2683
self._generations_until_keep = 100
2685
def set_generations_until_keep(self, value):
2686
"""Set the number of generations before caching a node.
2688
Setting this to -1 will cache every merge node, setting this higher
2689
will cache fewer nodes.
2691
self._generations_until_keep = value
2693
def _add_fulltext_content(self, revision_id, content_obj):
2694
self._fulltext_contents[revision_id] = content_obj
2695
# TODO: jam 20080305 It might be good to check the sha1digest here
2696
return content_obj.text()
2698
def _check_parents(self, child, nodes_to_annotate):
2699
"""Check if all parents have been processed.
2701
:param child: A tuple of (rev_id, parents, raw_content)
2702
:param nodes_to_annotate: If child is ready, add it to
2703
nodes_to_annotate, otherwise put it back in self._pending_children
2705
for parent_id in child[1]:
2706
if (parent_id not in self._annotated_lines):
2707
# This parent is present, but another parent is missing
2708
self._pending_children.setdefault(parent_id,
2712
# This one is ready to be processed
2713
nodes_to_annotate.append(child)
2715
def _add_annotation(self, revision_id, fulltext, parent_ids,
2716
left_matching_blocks=None):
2717
"""Add an annotation entry.
2719
All parents should already have been annotated.
2720
:return: A list of children that now have their parents satisfied.
2722
a = self._annotated_lines
2723
annotated_parent_lines = [a[p] for p in parent_ids]
2724
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2725
fulltext, revision_id, left_matching_blocks,
2726
heads_provider=self._get_heads_provider()))
2727
self._annotated_lines[revision_id] = annotated_lines
2728
for p in parent_ids:
2729
ann_children = self._annotate_children[p]
2730
ann_children.remove(revision_id)
2731
if (not ann_children
2732
and p not in self._nodes_to_keep_annotations):
2733
del self._annotated_lines[p]
2734
del self._all_build_details[p]
2735
if p in self._fulltext_contents:
2736
del self._fulltext_contents[p]
2737
# Now that we've added this one, see if there are any pending
2738
# deltas to be done, certainly this parent is finished
2739
nodes_to_annotate = []
2740
for child in self._pending_children.pop(revision_id, []):
2741
self._check_parents(child, nodes_to_annotate)
2742
return nodes_to_annotate
2744
def _get_build_graph(self, key):
2745
"""Get the graphs for building texts and annotations.
2747
The data you need for creating a full text may be different than the
2748
data you need to annotate that text. (At a minimum, you need both
2749
parents to create an annotation, but only need 1 parent to generate the
2752
:return: A list of (key, index_memo) records, suitable for
2753
passing to read_records_iter to start reading in the raw data fro/
2756
if key in self._annotated_lines:
2759
pending = set([key])
2764
# get all pending nodes
2766
this_iteration = pending
2767
build_details = self._knit._index.get_build_details(this_iteration)
2768
self._all_build_details.update(build_details)
2769
# new_nodes = self._knit._index._get_entries(this_iteration)
2771
for key, details in build_details.iteritems():
2772
(index_memo, compression_parent, parents,
2773
record_details) = details
2774
self._revision_id_graph[key] = parents
2775
records.append((key, index_memo))
2776
# Do we actually need to check _annotated_lines?
2777
pending.update(p for p in parents
2778
if p not in self._all_build_details)
2779
if compression_parent:
2780
self._compression_children.setdefault(compression_parent,
2783
for parent in parents:
2784
self._annotate_children.setdefault(parent,
2786
num_gens = generation - kept_generation
2787
if ((num_gens >= self._generations_until_keep)
2788
and len(parents) > 1):
2789
kept_generation = generation
2790
self._nodes_to_keep_annotations.add(key)
2792
missing_versions = this_iteration.difference(build_details.keys())
2793
self._ghosts.update(missing_versions)
2794
for missing_version in missing_versions:
2795
# add a key, no parents
2796
self._revision_id_graph[missing_version] = ()
2797
pending.discard(missing_version) # don't look for it
2798
if self._ghosts.intersection(self._compression_children):
2800
"We cannot have nodes which have a ghost compression parent:\n"
2802
"compression children: %r"
2803
% (self._ghosts, self._compression_children))
2804
# Cleanout anything that depends on a ghost so that we don't wait for
2805
# the ghost to show up
2806
for node in self._ghosts:
2807
if node in self._annotate_children:
2808
# We won't be building this node
2809
del self._annotate_children[node]
2810
# Generally we will want to read the records in reverse order, because
2811
# we find the parent nodes after the children
2815
def _annotate_records(self, records):
2816
"""Build the annotations for the listed records."""
2817
# We iterate in the order read, rather than a strict order requested
2818
# However, process what we can, and put off to the side things that
2819
# still need parents, cleaning them up when those parents are
2821
for (rev_id, record,
2822
digest) in self._knit._read_records_iter(records):
2823
if rev_id in self._annotated_lines:
2825
parent_ids = self._revision_id_graph[rev_id]
2826
parent_ids = [p for p in parent_ids if p not in self._ghosts]
2827
details = self._all_build_details[rev_id]
2828
(index_memo, compression_parent, parents,
2829
record_details) = details
2830
nodes_to_annotate = []
2831
# TODO: Remove the punning between compression parents, and
2832
# parent_ids, we should be able to do this without assuming
2834
if len(parent_ids) == 0:
2835
# There are no parents for this node, so just add it
2836
# TODO: This probably needs to be decoupled
2837
fulltext_content, delta = self._knit._factory.parse_record(
2838
rev_id, record, record_details, None)
2839
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
2840
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
2841
parent_ids, left_matching_blocks=None))
2843
child = (rev_id, parent_ids, record)
2844
# Check if all the parents are present
2845
self._check_parents(child, nodes_to_annotate)
2846
while nodes_to_annotate:
2847
# Should we use a queue here instead of a stack?
2848
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
2849
(index_memo, compression_parent, parents,
2850
record_details) = self._all_build_details[rev_id]
2852
if compression_parent is not None:
2853
comp_children = self._compression_children[compression_parent]
2854
if rev_id not in comp_children:
2855
raise AssertionError("%r not in compression children %r"
2856
% (rev_id, comp_children))
2857
# If there is only 1 child, it is safe to reuse this
2859
reuse_content = (len(comp_children) == 1
2860
and compression_parent not in
2861
self._nodes_to_keep_annotations)
2863
# Remove it from the cache since it will be changing
2864
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
2865
# Make sure to copy the fulltext since it might be
2867
parent_fulltext = list(parent_fulltext_content.text())
2869
parent_fulltext_content = self._fulltext_contents[compression_parent]
2870
parent_fulltext = parent_fulltext_content.text()
2871
comp_children.remove(rev_id)
2872
fulltext_content, delta = self._knit._factory.parse_record(
2873
rev_id, record, record_details,
2874
parent_fulltext_content,
2875
copy_base_content=(not reuse_content))
2876
fulltext = self._add_fulltext_content(rev_id,
2878
if compression_parent == parent_ids[0]:
2879
# the compression_parent is the left parent, so we can
2881
blocks = KnitContent.get_line_delta_blocks(delta,
2882
parent_fulltext, fulltext)
2884
fulltext_content = self._knit._factory.parse_fulltext(
2886
fulltext = self._add_fulltext_content(rev_id,
2888
nodes_to_annotate.extend(
2889
self._add_annotation(rev_id, fulltext, parent_ids,
2890
left_matching_blocks=blocks))
2892
def _get_heads_provider(self):
2893
"""Create a heads provider for resolving ancestry issues."""
2894
if self._heads_provider is not None:
2895
return self._heads_provider
2896
parent_provider = _mod_graph.DictParentsProvider(
2897
self._revision_id_graph)
2898
graph_obj = _mod_graph.Graph(parent_provider)
2899
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
2900
self._heads_provider = head_cache
2903
def annotate(self, key):
2904
"""Return the annotated fulltext at the given key.
2906
:param key: The key to annotate.
2908
if len(self._knit._fallback_vfs) > 0:
2909
# stacked knits can't use the fast path at present.
2910
return self._simple_annotate(key)
2913
records = self._get_build_graph(key)
2914
if key in self._ghosts:
2915
raise errors.RevisionNotPresent(key, self._knit)
2916
self._annotate_records(records)
2917
return self._annotated_lines[key]
2918
except errors.RetryWithNewPacks, e:
2919
self._knit._access.reload_or_raise(e)
2920
# The cached build_details are no longer valid
2921
self._all_build_details.clear()
2923
def _simple_annotate(self, key):
2924
"""Return annotated fulltext, rediffing from the full texts.
2926
This is slow but makes no assumptions about the repository
2927
being able to produce line deltas.
2929
# TODO: this code generates a parent maps of present ancestors; it
2930
# could be split out into a separate method, and probably should use
2931
# iter_ancestry instead. -- mbp and robertc 20080704
2932
graph = _mod_graph.Graph(self._knit)
2933
head_cache = _mod_graph.FrozenHeadsCache(graph)
2934
search = graph._make_breadth_first_searcher([key])
2938
present, ghosts = search.next_with_ghosts()
2939
except StopIteration:
2941
keys.update(present)
2942
parent_map = self._knit.get_parent_map(keys)
2944
reannotate = annotate.reannotate
2945
for record in self._knit.get_record_stream(keys, 'topological', True):
2947
fulltext = split_lines(record.get_bytes_as('fulltext'))
2948
parents = parent_map[key]
2949
if parents is not None:
2950
parent_lines = [parent_cache[parent] for parent in parent_map[key]]
2953
parent_cache[key] = list(
2954
reannotate(parent_lines, fulltext, key, None, head_cache))
2956
return parent_cache[key]
2958
raise errors.RevisionNotPresent(key, self._knit)
2962
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2964
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
