259
750
for start, end, c, lines in delta:
260
751
out.append('%d,%d,%d\n' % (start, end, c))
261
out.extend([text for origin, text in lines])
265
def make_empty_knit(transport, relpath):
266
"""Construct a empty knit at the specified location."""
267
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
271
class KnitVersionedFile(VersionedFile):
272
"""Weave-like structure with faster random access.
274
A knit stores a number of texts and a summary of the relationships
275
between them. Texts are identified by a string version-id. Texts
276
are normally stored and retrieved as a series of lines, but can
277
also be passed as single strings.
279
Lines are stored with the trailing newline (if any) included, to
280
avoid special cases for files with no final newline. Lines are
281
composed of 8-bit characters, not unicode. The combination of
282
these approaches should mean any 'binary' file can be safely
283
stored and retrieved.
286
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
287
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
288
create=False, create_parent_dir=False, delay_create=False,
290
"""Construct a knit at location specified by relpath.
292
:param create: If not True, only open an existing knit.
293
:param create_parent_dir: If True, create the parent directory if
294
creating the file fails. (This is used for stores with
295
hash-prefixes that may not exist yet)
296
:param delay_create: The calling code is aware that the knit won't
297
actually be created until the first data is stored.
755
def annotate(self, knit, key):
756
annotator = _KnitAnnotator(knit)
757
return annotator.annotate_flat(key)
761
def make_file_factory(annotated, mapper):
762
"""Create a factory for creating a file based KnitVersionedFiles.
764
This is only functional enough to run interface tests, it doesn't try to
765
provide a full pack environment.
767
:param annotated: knit annotations are wanted.
768
:param mapper: The mapper from keys to paths.
770
def factory(transport):
771
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
772
access = _KnitKeyAccess(transport, mapper)
773
return KnitVersionedFiles(index, access, annotated=annotated)
777
def make_pack_factory(graph, delta, keylength):
778
"""Create a factory for creating a pack based VersionedFiles.
780
This is only functional enough to run interface tests, it doesn't try to
781
provide a full pack environment.
783
:param graph: Store a graph.
784
:param delta: Delta compress contents.
785
:param keylength: How long should keys be.
787
def factory(transport):
788
parents = graph or delta
794
max_delta_chain = 200
797
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
798
key_elements=keylength)
799
stream = transport.open_write_stream('newpack')
800
writer = pack.ContainerWriter(stream.write)
802
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
803
deltas=delta, add_callback=graph_index.add_nodes)
804
access = pack_repo._DirectPackAccess({})
805
access.set_writer(writer, graph_index, (transport, 'newpack'))
806
result = KnitVersionedFiles(index, access,
807
max_delta_chain=max_delta_chain)
808
result.stream = stream
809
result.writer = writer
814
def cleanup_pack_knit(versioned_files):
815
versioned_files.stream.close()
816
versioned_files.writer.end()
819
def _get_total_build_size(self, keys, positions):
820
"""Determine the total bytes to build these keys.
822
(helper function because _KnitGraphIndex and _KndxIndex work the same, but
823
don't inherit from a common base.)
825
:param keys: Keys that we want to build
826
:param positions: dict of {key, (info, index_memo, comp_parent)} (such
827
as returned by _get_components_positions)
828
:return: Number of bytes to build those keys
830
all_build_index_memos = {}
834
for key in build_keys:
835
# This is mostly for the 'stacked' case
836
# Where we will be getting the data from a fallback
837
if key not in positions:
839
_, index_memo, compression_parent = positions[key]
840
all_build_index_memos[key] = index_memo
841
if compression_parent not in all_build_index_memos:
842
next_keys.add(compression_parent)
843
build_keys = next_keys
844
return sum([index_memo[2] for index_memo
845
in all_build_index_memos.itervalues()])
848
class KnitVersionedFiles(VersionedFiles):
849
"""Storage for many versioned files using knit compression.
851
Backend storage is managed by indices and data objects.
853
:ivar _index: A _KnitGraphIndex or similar that can describe the
854
parents, graph, compression and data location of entries in this
855
KnitVersionedFiles. Note that this is only the index for
856
*this* vfs; if there are fallbacks they must be queried separately.
859
def __init__(self, index, data_access, max_delta_chain=200,
860
annotated=False, reload_func=None):
861
"""Create a KnitVersionedFiles with index and data_access.
863
:param index: The index for the knit data.
864
:param data_access: The access object to store and retrieve knit
866
:param max_delta_chain: The maximum number of deltas to permit during
867
insertion. Set to 0 to prohibit the use of deltas.
868
:param annotated: Set to True to cause annotations to be calculated and
869
stored during insertion.
870
:param reload_func: An function that can be called if we think we need
871
to reload the pack listing and try again. See
872
'bzrlib.repofmt.pack_repo.AggregateIndex' for the signature.
299
if deprecated_passed(basis_knit):
300
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
301
" deprecated as of bzr 0.9.",
302
DeprecationWarning, stacklevel=2)
303
if access_mode is None:
305
super(KnitVersionedFile, self).__init__(access_mode)
306
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
307
self.transport = transport
308
self.filename = relpath
309
self.factory = factory or KnitAnnotateFactory()
310
self.writable = (access_mode == 'w')
313
self._max_delta_chain = 200
315
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
316
access_mode, create=create, file_mode=file_mode,
317
create_parent_dir=create_parent_dir, delay_create=delay_create,
319
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
320
access_mode, create=create and not len(self), file_mode=file_mode,
321
create_parent_dir=create_parent_dir, delay_create=delay_create,
875
self._access = data_access
876
self._max_delta_chain = max_delta_chain
878
self._factory = KnitAnnotateFactory()
880
self._factory = KnitPlainFactory()
881
self._immediate_fallback_vfs = []
882
self._reload_func = reload_func
324
884
def __repr__(self):
325
return '%s(%s)' % (self.__class__.__name__,
326
self.transport.abspath(self.filename))
328
def _check_should_delta(self, first_parents):
885
return "%s(%r, %r)" % (
886
self.__class__.__name__,
890
def add_fallback_versioned_files(self, a_versioned_files):
891
"""Add a source of texts for texts not present in this knit.
893
:param a_versioned_files: A VersionedFiles object.
895
self._immediate_fallback_vfs.append(a_versioned_files)
897
def add_lines(self, key, parents, lines, parent_texts=None,
898
left_matching_blocks=None, nostore_sha=None, random_id=False,
900
"""See VersionedFiles.add_lines()."""
901
self._index._check_write_ok()
902
self._check_add(key, lines, random_id, check_content)
904
# The caller might pass None if there is no graph data, but kndx
905
# indexes can't directly store that, so we give them
906
# an empty tuple instead.
908
line_bytes = ''.join(lines)
909
return self._add(key, lines, parents,
910
parent_texts, left_matching_blocks, nostore_sha, random_id,
911
line_bytes=line_bytes)
913
def _add_text(self, key, parents, text, nostore_sha=None, random_id=False):
914
"""See VersionedFiles._add_text()."""
915
self._index._check_write_ok()
916
self._check_add(key, None, random_id, check_content=False)
917
if text.__class__ is not str:
918
raise errors.BzrBadParameterUnicode("text")
920
# The caller might pass None if there is no graph data, but kndx
921
# indexes can't directly store that, so we give them
922
# an empty tuple instead.
924
return self._add(key, None, parents,
925
None, None, nostore_sha, random_id,
928
def _add(self, key, lines, parents, parent_texts,
929
left_matching_blocks, nostore_sha, random_id,
931
"""Add a set of lines on top of version specified by parents.
933
Any versions not present will be converted into ghosts.
935
:param lines: A list of strings where each one is a single line (has a
936
single newline at the end of the string) This is now optional
937
(callers can pass None). It is left in its location for backwards
938
compatibility. It should ''.join(lines) must == line_bytes
939
:param line_bytes: A single string containing the content
941
We pass both lines and line_bytes because different routes bring the
942
values to this function. And for memory efficiency, we don't want to
943
have to split/join on-demand.
945
# first thing, if the content is something we don't need to store, find
947
digest = sha_string(line_bytes)
948
if nostore_sha == digest:
949
raise errors.ExistingContent
952
if parent_texts is None:
954
# Do a single query to ascertain parent presence; we only compress
955
# against parents in the same kvf.
956
present_parent_map = self._index.get_parent_map(parents)
957
for parent in parents:
958
if parent in present_parent_map:
959
present_parents.append(parent)
961
# Currently we can only compress against the left most present parent.
962
if (len(present_parents) == 0 or
963
present_parents[0] != parents[0]):
966
# To speed the extract of texts the delta chain is limited
967
# to a fixed number of deltas. This should minimize both
968
# I/O and the time spend applying deltas.
969
delta = self._check_should_delta(present_parents[0])
971
text_length = len(line_bytes)
974
# Note: line_bytes is not modified to add a newline, that is tracked
975
# via the no_eol flag. 'lines' *is* modified, because that is the
976
# general values needed by the Content code.
977
if line_bytes and line_bytes[-1] != '\n':
978
options.append('no-eol')
980
# Copy the existing list, or create a new one
982
lines = osutils.split_lines(line_bytes)
985
# Replace the last line with one that ends in a final newline
986
lines[-1] = lines[-1] + '\n'
988
lines = osutils.split_lines(line_bytes)
990
for element in key[:-1]:
991
if type(element) is not str:
992
raise TypeError("key contains non-strings: %r" % (key,))
994
key = key[:-1] + ('sha1:' + digest,)
995
elif type(key[-1]) is not str:
996
raise TypeError("key contains non-strings: %r" % (key,))
997
# Knit hunks are still last-element only
999
content = self._factory.make(lines, version_id)
1001
# Hint to the content object that its text() call should strip the
1003
content._should_strip_eol = True
1004
if delta or (self._factory.annotated and len(present_parents) > 0):
1005
# Merge annotations from parent texts if needed.
1006
delta_hunks = self._merge_annotations(content, present_parents,
1007
parent_texts, delta, self._factory.annotated,
1008
left_matching_blocks)
1011
options.append('line-delta')
1012
store_lines = self._factory.lower_line_delta(delta_hunks)
1013
size, bytes = self._record_to_data(key, digest,
1016
options.append('fulltext')
1017
# isinstance is slower and we have no hierarchy.
1018
if self._factory.__class__ is KnitPlainFactory:
1019
# Use the already joined bytes saving iteration time in
1021
dense_lines = [line_bytes]
1023
dense_lines.append('\n')
1024
size, bytes = self._record_to_data(key, digest,
1027
# get mixed annotation + content and feed it into the
1029
store_lines = self._factory.lower_fulltext(content)
1030
size, bytes = self._record_to_data(key, digest,
1033
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
1034
self._index.add_records(
1035
((key, options, access_memo, parents),),
1036
random_id=random_id)
1037
return digest, text_length, content
1039
def annotate(self, key):
1040
"""See VersionedFiles.annotate."""
1041
return self._factory.annotate(self, key)
1043
def get_annotator(self):
1044
return _KnitAnnotator(self)
1046
def check(self, progress_bar=None, keys=None):
1047
"""See VersionedFiles.check()."""
1049
return self._logical_check()
1051
# At the moment, check does not extra work over get_record_stream
1052
return self.get_record_stream(keys, 'unordered', True)
1054
def _logical_check(self):
1055
# This doesn't actually test extraction of everything, but that will
1056
# impact 'bzr check' substantially, and needs to be integrated with
1057
# care. However, it does check for the obvious problem of a delta with
1059
keys = self._index.keys()
1060
parent_map = self.get_parent_map(keys)
1062
if self._index.get_method(key) != 'fulltext':
1063
compression_parent = parent_map[key][0]
1064
if compression_parent not in parent_map:
1065
raise errors.KnitCorrupt(self,
1066
"Missing basis parent %s for %s" % (
1067
compression_parent, key))
1068
for fallback_vfs in self._immediate_fallback_vfs:
1069
fallback_vfs.check()
1071
def _check_add(self, key, lines, random_id, check_content):
1072
"""check that version_id and lines are safe to add."""
1073
version_id = key[-1]
1074
if version_id is not None:
1075
if contains_whitespace(version_id):
1076
raise InvalidRevisionId(version_id, self)
1077
self.check_not_reserved_id(version_id)
1078
# TODO: If random_id==False and the key is already present, we should
1079
# probably check that the existing content is identical to what is
1080
# being inserted, and otherwise raise an exception. This would make
1081
# the bundle code simpler.
1083
self._check_lines_not_unicode(lines)
1084
self._check_lines_are_lines(lines)
1086
def _check_header(self, key, line):
1087
rec = self._split_header(line)
1088
self._check_header_version(rec, key[-1])
1091
def _check_header_version(self, rec, version_id):
1092
"""Checks the header version on original format knit records.
1094
These have the last component of the key embedded in the record.
1096
if rec[1] != version_id:
1097
raise KnitCorrupt(self,
1098
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1100
def _check_should_delta(self, parent):
329
1101
"""Iterate back through the parent listing, looking for a fulltext.
331
1103
This is used when we want to decide whether to add a delta or a new
340
1112
fulltext_size = None
341
delta_parents = first_parents
342
1113
for count in xrange(self._max_delta_chain):
343
parent = delta_parents[0]
344
method = self._index.get_method(parent)
345
pos, size = self._index.get_position(parent)
346
if method == 'fulltext':
1115
# Note that this only looks in the index of this particular
1116
# KnitVersionedFiles, not in the fallbacks. This ensures that
1117
# we won't store a delta spanning physical repository
1119
build_details = self._index.get_build_details([parent])
1120
parent_details = build_details[parent]
1121
except (RevisionNotPresent, KeyError), e:
1122
# Some basis is not locally present: always fulltext
1124
index_memo, compression_parent, _, _ = parent_details
1125
_, _, size = index_memo
1126
if compression_parent is None:
347
1127
fulltext_size = size
349
1129
delta_size += size
350
delta_parents = self._index.get_parents(parent)
1130
# We don't explicitly check for presence because this is in an
1131
# inner loop, and if it's missing it'll fail anyhow.
1132
parent = compression_parent
352
1134
# We couldn't find a fulltext, so we must create a new one
1136
# Simple heuristic - if the total I/O wold be greater as a delta than
1137
# the originally installed fulltext, we create a new fulltext.
355
1138
return fulltext_size > delta_size
357
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
358
"""See VersionedFile._add_delta()."""
359
self._check_add(version_id, []) # should we check the lines ?
360
self._check_versions_present(parents)
364
for parent in parents:
365
if not self.has_version(parent):
366
ghosts.append(parent)
368
present_parents.append(parent)
370
if delta_parent is None:
371
# reconstitute as full text.
372
assert len(delta) == 1 or len(delta) == 0
374
assert delta[0][0] == 0
375
assert delta[0][1] == 0, delta[0][1]
376
return super(KnitVersionedFile, self)._add_delta(version_id,
387
options.append('no-eol')
389
if delta_parent is not None:
390
# determine the current delta chain length.
391
# To speed the extract of texts the delta chain is limited
392
# to a fixed number of deltas. This should minimize both
393
# I/O and the time spend applying deltas.
394
# The window was changed to a maximum of 200 deltas, but also added
395
# was a check that the total compressed size of the deltas is
396
# smaller than the compressed size of the fulltext.
397
if not self._check_should_delta([delta_parent]):
398
# We don't want a delta here, just do a normal insertion.
399
return super(KnitVersionedFile, self)._add_delta(version_id,
406
options.append('line-delta')
407
store_lines = self.factory.lower_line_delta(delta)
409
where, size = self._data.add_record(version_id, digest, store_lines)
410
self._index.add_version(version_id, options, where, size, parents)
412
def _add_raw_records(self, records, data):
413
"""Add all the records 'records' with data pre-joined in 'data'.
415
:param records: A list of tuples(version_id, options, parents, size).
416
:param data: The data for the records. When it is written, the records
417
are adjusted to have pos pointing into data by the sum of
418
the preceding records sizes.
421
pos = self._data.add_raw_record(data)
424
for (version_id, options, parents, size) in records:
425
index_entries.append((version_id, options, pos+offset,
427
if self._data._do_cache:
428
self._data._cache[version_id] = data[offset:offset+size]
430
self._index.add_versions(index_entries)
432
def enable_cache(self):
433
"""Start caching data for this knit"""
434
self._data.enable_cache()
436
def clear_cache(self):
437
"""Clear the data cache only."""
438
self._data.clear_cache()
440
def copy_to(self, name, transport):
441
"""See VersionedFile.copy_to()."""
442
# copy the current index to a temp index to avoid racing with local
444
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
445
self.transport.get(self._index._filename))
447
f = self._data._open_file()
449
transport.put_file(name + DATA_SUFFIX, f)
452
# move the copied index into place
453
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
455
def create_empty(self, name, transport, mode=None):
456
return KnitVersionedFile(name, transport, factory=self.factory,
457
delta=self.delta, create=True)
459
def _fix_parents(self, version, new_parents):
460
"""Fix the parents list for version.
462
This is done by appending a new version to the index
463
with identical data except for the parents list.
464
the parents list must be a superset of the current
467
current_values = self._index._cache[version]
468
assert set(current_values[4]).difference(set(new_parents)) == set()
469
self._index.add_version(version,
475
def get_delta(self, version_id):
476
"""Get a delta for constructing version from some other version."""
477
if not self.has_version(version_id):
478
raise RevisionNotPresent(version_id, self.filename)
480
parents = self.get_parents(version_id)
485
data_pos, data_size = self._index.get_position(version_id)
486
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
487
version_idx = self._index.lookup(version_id)
488
noeol = 'no-eol' in self._index.get_options(version_id)
489
if 'fulltext' == self._index.get_method(version_id):
490
new_content = self.factory.parse_fulltext(data, version_idx)
491
if parent is not None:
492
reference_content = self._get_content(parent)
493
old_texts = reference_content.text()
496
new_texts = new_content.text()
497
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
498
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
500
delta = self.factory.parse_line_delta(data, version_idx)
501
return parent, sha1, noeol, delta
503
def get_graph_with_ghosts(self):
504
"""See VersionedFile.get_graph_with_ghosts()."""
505
graph_items = self._index.get_graph()
506
return dict(graph_items)
508
def get_sha1(self, version_id):
509
"""See VersionedFile.get_sha1()."""
510
record_map = self._get_record_map([version_id])
511
method, content, digest, next = record_map[version_id]
516
"""See VersionedFile.get_suffixes()."""
517
return [DATA_SUFFIX, INDEX_SUFFIX]
519
def has_ghost(self, version_id):
520
"""True if there is a ghost reference in the file to version_id."""
522
if self.has_version(version_id):
524
# optimisable if needed by memoising the _ghosts set.
525
items = self._index.get_graph()
526
for node, parents in items:
527
for parent in parents:
528
if parent not in self._index._cache:
529
if parent == version_id:
534
"""See VersionedFile.versions."""
535
return self._index.get_versions()
537
def has_version(self, version_id):
538
"""See VersionedFile.has_version."""
539
return self._index.has_version(version_id)
541
__contains__ = has_version
543
def _merge_annotations(self, content, parents, parent_texts={},
544
delta=None, annotated=None):
545
"""Merge annotations for content. This is done by comparing
546
the annotations based on changed to the text.
550
for parent_id in parents:
551
merge_content = self._get_content(parent_id, parent_texts)
552
seq = patiencediff.PatienceSequenceMatcher(
553
None, merge_content.text(), content.text())
554
if delta_seq is None:
555
# setup a delta seq to reuse.
557
for i, j, n in seq.get_matching_blocks():
560
# this appears to copy (origin, text) pairs across to the new
561
# content for any line that matches the last-checked parent.
562
# FIXME: save the sequence control data for delta compression
563
# against the most relevant parent rather than rediffing.
564
content._lines[j:j+n] = merge_content._lines[i:i+n]
567
reference_content = self._get_content(parents[0], parent_texts)
568
new_texts = content.text()
569
old_texts = reference_content.text()
570
delta_seq = patiencediff.PatienceSequenceMatcher(
571
None, old_texts, new_texts)
572
return self._make_line_delta(delta_seq, content)
574
def _make_line_delta(self, delta_seq, new_content):
575
"""Generate a line delta from delta_seq and new_content."""
577
for op in delta_seq.get_opcodes():
580
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
583
def _get_components_positions(self, version_ids):
584
"""Produce a map of position data for the components of versions.
1140
def _build_details_to_components(self, build_details):
1141
"""Convert a build_details tuple to a position tuple."""
1142
# record_details, access_memo, compression_parent
1143
return build_details[3], build_details[0], build_details[1]
1145
def _get_components_positions(self, keys, allow_missing=False):
1146
"""Produce a map of position data for the components of keys.
586
1148
This data is intended to be used for retrieving the knit records.
588
A dict of version_id to (method, data_pos, data_size, next) is
1150
A dict of key to (record_details, index_memo, next, parents) is
590
1152
method is the way referenced data should be applied.
591
data_pos is the position of the data in the knit.
592
data_size is the size of the data in the knit.
1153
index_memo is the handle to pass to the data access to actually get the
593
1155
next is the build-parent of the version, or None for fulltexts.
1156
parents is the version_ids of the parents of this version
1158
:param allow_missing: If True do not raise an error on a missing component,
595
1161
component_data = {}
596
for version_id in version_ids:
599
while cursor is not None and cursor not in component_data:
600
method = self._index.get_method(cursor)
601
if method == 'fulltext':
604
next = self.get_parents(cursor)[0]
605
data_pos, data_size = self._index.get_position(cursor)
606
component_data[cursor] = (method, data_pos, data_size, next)
1162
pending_components = keys
1163
while pending_components:
1164
build_details = self._index.get_build_details(pending_components)
1165
current_components = set(pending_components)
1166
pending_components = set()
1167
for key, details in build_details.iteritems():
1168
(index_memo, compression_parent, parents,
1169
record_details) = details
1170
method = record_details[0]
1171
if compression_parent is not None:
1172
pending_components.add(compression_parent)
1173
component_data[key] = self._build_details_to_components(details)
1174
missing = current_components.difference(build_details)
1175
if missing and not allow_missing:
1176
raise errors.RevisionNotPresent(missing.pop(), self)
608
1177
return component_data
610
def _get_content(self, version_id, parent_texts={}):
1179
def _get_content(self, key, parent_texts={}):
611
1180
"""Returns a content object that makes up the specified
613
if not self.has_version(version_id):
614
raise RevisionNotPresent(version_id, self.filename)
616
cached_version = parent_texts.get(version_id, None)
1182
cached_version = parent_texts.get(key, None)
617
1183
if cached_version is not None:
1184
# Ensure the cache dict is valid.
1185
if not self.get_parent_map([key]):
1186
raise RevisionNotPresent(key, self)
618
1187
return cached_version
620
text_map, contents_map = self._get_content_maps([version_id])
621
return contents_map[version_id]
623
def _check_versions_present(self, version_ids):
624
"""Check that all specified versions are present."""
625
version_ids = set(version_ids)
626
for r in list(version_ids):
627
if self._index.has_version(r):
628
version_ids.remove(r)
630
raise RevisionNotPresent(list(version_ids)[0], self.filename)
632
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
633
"""See VersionedFile.add_lines_with_ghosts()."""
634
self._check_add(version_id, lines)
635
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
637
def _add_lines(self, version_id, parents, lines, parent_texts):
638
"""See VersionedFile.add_lines."""
639
self._check_add(version_id, lines)
640
self._check_versions_present(parents)
641
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
643
def _check_add(self, version_id, lines):
644
"""check that version_id and lines are safe to add."""
645
assert self.writable, "knit is not opened for write"
646
### FIXME escape. RBC 20060228
647
if contains_whitespace(version_id):
648
raise InvalidRevisionId(version_id, self.filename)
649
if self.has_version(version_id):
650
raise RevisionAlreadyPresent(version_id, self.filename)
651
self._check_lines_not_unicode(lines)
652
self._check_lines_are_lines(lines)
654
def _add(self, version_id, lines, parents, delta, parent_texts):
655
"""Add a set of lines on top of version specified by parents.
657
If delta is true, compress the text as a line-delta against
660
Any versions not present will be converted into ghosts.
662
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
663
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
664
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
665
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
666
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
667
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
668
# +1383 0 8.0370 8.0370 +<len>
669
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
670
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
671
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
672
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
676
if parent_texts is None:
678
for parent in parents:
679
if not self.has_version(parent):
680
ghosts.append(parent)
682
present_parents.append(parent)
684
if delta and not len(present_parents):
687
digest = sha_strings(lines)
690
if lines[-1][-1] != '\n':
691
options.append('no-eol')
692
lines[-1] = lines[-1] + '\n'
694
if len(present_parents) and delta:
695
# To speed the extract of texts the delta chain is limited
696
# to a fixed number of deltas. This should minimize both
697
# I/O and the time spend applying deltas.
698
delta = self._check_should_delta(present_parents)
700
lines = self.factory.make(lines, version_id)
701
if delta or (self.factory.annotated and len(present_parents) > 0):
702
# Merge annotations from parent texts if so is needed.
703
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
704
delta, self.factory.annotated)
707
options.append('line-delta')
708
store_lines = self.factory.lower_line_delta(delta_hunks)
710
options.append('fulltext')
711
store_lines = self.factory.lower_fulltext(lines)
713
where, size = self._data.add_record(version_id, digest, store_lines)
714
self._index.add_version(version_id, options, where, size, parents)
717
def check(self, progress_bar=None):
718
"""See VersionedFile.check()."""
720
def _clone_text(self, new_version_id, old_version_id, parents):
721
"""See VersionedFile.clone_text()."""
722
# FIXME RBC 20060228 make fast by only inserting an index with null
724
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
726
def get_lines(self, version_id):
727
"""See VersionedFile.get_lines()."""
728
return self.get_line_list([version_id])[0]
730
def _get_record_map(self, version_ids):
1188
generator = _VFContentMapGenerator(self, [key])
1189
return generator._get_content(key)
1191
def get_known_graph_ancestry(self, keys):
1192
"""Get a KnownGraph instance with the ancestry of keys."""
1193
parent_map, missing_keys = self._index.find_ancestry(keys)
1194
for fallback in self._transitive_fallbacks():
1195
if not missing_keys:
1197
(f_parent_map, f_missing_keys) = fallback._index.find_ancestry(
1199
parent_map.update(f_parent_map)
1200
missing_keys = f_missing_keys
1201
kg = _mod_graph.KnownGraph(parent_map)
1204
def get_parent_map(self, keys):
1205
"""Get a map of the graph parents of keys.
1207
:param keys: The keys to look up parents for.
1208
:return: A mapping from keys to parents. Absent keys are absent from
1211
return self._get_parent_map_with_sources(keys)[0]
1213
def _get_parent_map_with_sources(self, keys):
1214
"""Get a map of the parents of keys.
1216
:param keys: The keys to look up parents for.
1217
:return: A tuple. The first element is a mapping from keys to parents.
1218
Absent keys are absent from the mapping. The second element is a
1219
list with the locations each key was found in. The first element
1220
is the in-this-knit parents, the second the first fallback source,
1224
sources = [self._index] + self._immediate_fallback_vfs
1227
for source in sources:
1230
new_result = source.get_parent_map(missing)
1231
source_results.append(new_result)
1232
result.update(new_result)
1233
missing.difference_update(set(new_result))
1234
return result, source_results
1236
def _get_record_map(self, keys, allow_missing=False):
731
1237
"""Produce a dictionary of knit records.
733
The keys are version_ids, the values are tuples of (method, content,
735
method is the way the content should be applied.
736
content is a KnitContent object.
737
digest is the SHA1 digest of this version id after all steps are done
738
next is the build-parent of the version, i.e. the leftmost ancestor.
739
If the method is fulltext, next will be None.
741
position_map = self._get_components_positions(version_ids)
742
# c = component_id, m = method, p = position, s = size, n = next
743
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
745
for component_id, content, digest in \
746
self._data.read_records_iter(records):
747
method, position, size, next = position_map[component_id]
748
record_map[component_id] = method, content, digest, next
752
def get_text(self, version_id):
753
"""See VersionedFile.get_text"""
754
return self.get_texts([version_id])[0]
756
def get_texts(self, version_ids):
757
return [''.join(l) for l in self.get_line_list(version_ids)]
759
def get_line_list(self, version_ids):
760
"""Return the texts of listed versions as a list of strings."""
761
text_map, content_map = self._get_content_maps(version_ids)
762
return [text_map[v] for v in version_ids]
764
def _get_content_maps(self, version_ids):
765
"""Produce maps of text and KnitContents
1239
:return: {key:(record, record_details, digest, next)}
1241
data returned from read_records (a KnitContentobject)
1243
opaque information to pass to parse_record
1245
SHA1 digest of the full text after all steps are done
1247
build-parent of the version, i.e. the leftmost ancestor.
1248
Will be None if the record is not a delta.
1249
:param keys: The keys to build a map for
1250
:param allow_missing: If some records are missing, rather than
1251
error, just return the data that could be generated.
1253
raw_map = self._get_record_map_unparsed(keys,
1254
allow_missing=allow_missing)
1255
return self._raw_map_to_record_map(raw_map)
1257
def _raw_map_to_record_map(self, raw_map):
1258
"""Parse the contents of _get_record_map_unparsed.
1260
:return: see _get_record_map.
1264
data, record_details, next = raw_map[key]
1265
content, digest = self._parse_record(key[-1], data)
1266
result[key] = content, record_details, digest, next
1269
def _get_record_map_unparsed(self, keys, allow_missing=False):
1270
"""Get the raw data for reconstructing keys without parsing it.
1272
:return: A dict suitable for parsing via _raw_map_to_record_map.
1273
key-> raw_bytes, (method, noeol), compression_parent
1275
# This retries the whole request if anything fails. Potentially we
1276
# could be a bit more selective. We could track the keys whose records
1277
# we have successfully found, and then only request the new records
1278
# from there. However, _get_components_positions grabs the whole build
1279
# chain, which means we'll likely try to grab the same records again
1280
# anyway. Also, can the build chains change as part of a pack
1281
# operation? We wouldn't want to end up with a broken chain.
1284
position_map = self._get_components_positions(keys,
1285
allow_missing=allow_missing)
1286
# key = component_id, r = record_details, i_m = index_memo,
1288
records = [(key, i_m) for key, (r, i_m, n)
1289
in position_map.iteritems()]
1290
# Sort by the index memo, so that we request records from the
1291
# same pack file together, and in forward-sorted order
1292
records.sort(key=operator.itemgetter(1))
1294
for key, data in self._read_records_iter_unchecked(records):
1295
(record_details, index_memo, next) = position_map[key]
1296
raw_record_map[key] = data, record_details, next
1297
return raw_record_map
1298
except errors.RetryWithNewPacks, e:
1299
self._access.reload_or_raise(e)
1302
def _split_by_prefix(cls, keys):
1303
"""For the given keys, split them up based on their prefix.
1305
To keep memory pressure somewhat under control, split the
1306
requests back into per-file-id requests, otherwise "bzr co"
1307
extracts the full tree into memory before writing it to disk.
1308
This should be revisited if _get_content_maps() can ever cross
1311
The keys for a given file_id are kept in the same relative order.
1312
Ordering between file_ids is not, though prefix_order will return the
1313
order that the key was first seen.
1315
:param keys: An iterable of key tuples
1316
:return: (split_map, prefix_order)
1317
split_map A dictionary mapping prefix => keys
1318
prefix_order The order that we saw the various prefixes
1320
split_by_prefix = {}
1328
if prefix in split_by_prefix:
1329
split_by_prefix[prefix].append(key)
1331
split_by_prefix[prefix] = [key]
1332
prefix_order.append(prefix)
1333
return split_by_prefix, prefix_order
1335
def _group_keys_for_io(self, keys, non_local_keys, positions,
1336
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1337
"""For the given keys, group them into 'best-sized' requests.
1339
The idea is to avoid making 1 request per file, but to never try to
1340
unpack an entire 1.5GB source tree in a single pass. Also when
1341
possible, we should try to group requests to the same pack file
1344
:return: list of (keys, non_local) tuples that indicate what keys
1345
should be fetched next.
1347
# TODO: Ideally we would group on 2 factors. We want to extract texts
1348
# from the same pack file together, and we want to extract all
1349
# the texts for a given build-chain together. Ultimately it
1350
# probably needs a better global view.
1351
total_keys = len(keys)
1352
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1353
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1355
cur_non_local = set()
1359
for prefix in prefix_order:
1360
keys = prefix_split_keys[prefix]
1361
non_local = prefix_split_non_local_keys.get(prefix, [])
1363
this_size = self._index._get_total_build_size(keys, positions)
1364
cur_size += this_size
1365
cur_keys.extend(keys)
1366
cur_non_local.update(non_local)
1367
if cur_size > _min_buffer_size:
1368
result.append((cur_keys, cur_non_local))
1369
sizes.append(cur_size)
1371
cur_non_local = set()
1374
result.append((cur_keys, cur_non_local))
1375
sizes.append(cur_size)
1378
def get_record_stream(self, keys, ordering, include_delta_closure):
1379
"""Get a stream of records for keys.
1381
:param keys: The keys to include.
1382
:param ordering: Either 'unordered' or 'topological'. A topologically
1383
sorted stream has compression parents strictly before their
1385
:param include_delta_closure: If True then the closure across any
1386
compression parents will be included (in the opaque data).
1387
:return: An iterator of ContentFactory objects, each of which is only
1388
valid until the iterator is advanced.
1390
# keys might be a generator
1394
if not self._index.has_graph:
1395
# Cannot sort when no graph has been stored.
1396
ordering = 'unordered'
1398
remaining_keys = keys
1401
keys = set(remaining_keys)
1402
for content_factory in self._get_remaining_record_stream(keys,
1403
ordering, include_delta_closure):
1404
remaining_keys.discard(content_factory.key)
1405
yield content_factory
1407
except errors.RetryWithNewPacks, e:
1408
self._access.reload_or_raise(e)
1410
def _get_remaining_record_stream(self, keys, ordering,
1411
include_delta_closure):
1412
"""This function is the 'retry' portion for get_record_stream."""
1413
if include_delta_closure:
1414
positions = self._get_components_positions(keys, allow_missing=True)
1416
build_details = self._index.get_build_details(keys)
1418
# (record_details, access_memo, compression_parent_key)
1419
positions = dict((key, self._build_details_to_components(details))
1420
for key, details in build_details.iteritems())
1421
absent_keys = keys.difference(set(positions))
1422
# There may be more absent keys : if we're missing the basis component
1423
# and are trying to include the delta closure.
1424
# XXX: We should not ever need to examine remote sources because we do
1425
# not permit deltas across versioned files boundaries.
1426
if include_delta_closure:
1427
needed_from_fallback = set()
1428
# Build up reconstructable_keys dict. key:True in this dict means
1429
# the key can be reconstructed.
1430
reconstructable_keys = {}
1434
chain = [key, positions[key][2]]
1436
needed_from_fallback.add(key)
1439
while chain[-1] is not None:
1440
if chain[-1] in reconstructable_keys:
1441
result = reconstructable_keys[chain[-1]]
1445
chain.append(positions[chain[-1]][2])
1447
# missing basis component
1448
needed_from_fallback.add(chain[-1])
1451
for chain_key in chain[:-1]:
1452
reconstructable_keys[chain_key] = result
1454
needed_from_fallback.add(key)
1455
# Double index lookups here : need a unified api ?
1456
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1457
if ordering in ('topological', 'groupcompress'):
1458
if ordering == 'topological':
1459
# Global topological sort
1460
present_keys = tsort.topo_sort(global_map)
1462
present_keys = sort_groupcompress(global_map)
1463
# Now group by source:
1465
current_source = None
1466
for key in present_keys:
1467
for parent_map in parent_maps:
1468
if key in parent_map:
1469
key_source = parent_map
1471
if current_source is not key_source:
1472
source_keys.append((key_source, []))
1473
current_source = key_source
1474
source_keys[-1][1].append(key)
1476
if ordering != 'unordered':
1477
raise AssertionError('valid values for ordering are:'
1478
' "unordered", "groupcompress" or "topological" not: %r'
1480
# Just group by source; remote sources first.
1483
for parent_map in reversed(parent_maps):
1484
source_keys.append((parent_map, []))
1485
for key in parent_map:
1486
present_keys.append(key)
1487
source_keys[-1][1].append(key)
1488
# We have been requested to return these records in an order that
1489
# suits us. So we ask the index to give us an optimally sorted
1491
for source, sub_keys in source_keys:
1492
if source is parent_maps[0]:
1493
# Only sort the keys for this VF
1494
self._index._sort_keys_by_io(sub_keys, positions)
1495
absent_keys = keys - set(global_map)
1496
for key in absent_keys:
1497
yield AbsentContentFactory(key)
1498
# restrict our view to the keys we can answer.
1499
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1500
# XXX: At that point we need to consider the impact of double reads by
1501
# utilising components multiple times.
1502
if include_delta_closure:
1503
# XXX: get_content_maps performs its own index queries; allow state
1505
non_local_keys = needed_from_fallback - absent_keys
1506
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1509
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1512
for record in generator.get_record_stream():
1515
for source, keys in source_keys:
1516
if source is parent_maps[0]:
1517
# this KnitVersionedFiles
1518
records = [(key, positions[key][1]) for key in keys]
1519
for key, raw_data in self._read_records_iter_unchecked(records):
1520
(record_details, index_memo, _) = positions[key]
1521
yield KnitContentFactory(key, global_map[key],
1522
record_details, None, raw_data, self._factory.annotated, None)
1524
vf = self._immediate_fallback_vfs[parent_maps.index(source) - 1]
1525
for record in vf.get_record_stream(keys, ordering,
1526
include_delta_closure):
1529
def get_sha1s(self, keys):
1530
"""See VersionedFiles.get_sha1s()."""
1532
record_map = self._get_record_map(missing, allow_missing=True)
1534
for key, details in record_map.iteritems():
1535
if key not in missing:
1537
# record entry 2 is the 'digest'.
1538
result[key] = details[2]
1539
missing.difference_update(set(result))
1540
for source in self._immediate_fallback_vfs:
1543
new_result = source.get_sha1s(missing)
1544
result.update(new_result)
1545
missing.difference_update(set(new_result))
1548
def insert_record_stream(self, stream):
1549
"""Insert a record stream into this container.
1551
:param stream: A stream of records to insert.
1553
:seealso VersionedFiles.get_record_stream:
1555
def get_adapter(adapter_key):
1557
return adapters[adapter_key]
1559
adapter_factory = adapter_registry.get(adapter_key)
1560
adapter = adapter_factory(self)
1561
adapters[adapter_key] = adapter
1564
if self._factory.annotated:
1565
# self is annotated, we need annotated knits to use directly.
1566
annotated = "annotated-"
1569
# self is not annotated, but we can strip annotations cheaply.
1571
convertibles = set(["knit-annotated-ft-gz"])
1572
if self._max_delta_chain:
1573
delta_types.add("knit-annotated-delta-gz")
1574
convertibles.add("knit-annotated-delta-gz")
1575
# The set of types we can cheaply adapt without needing basis texts.
1576
native_types = set()
1577
if self._max_delta_chain:
1578
native_types.add("knit-%sdelta-gz" % annotated)
1579
delta_types.add("knit-%sdelta-gz" % annotated)
1580
native_types.add("knit-%sft-gz" % annotated)
1581
knit_types = native_types.union(convertibles)
1583
# Buffer all index entries that we can't add immediately because their
1584
# basis parent is missing. We don't buffer all because generating
1585
# annotations may require access to some of the new records. However we
1586
# can't generate annotations from new deltas until their basis parent
1587
# is present anyway, so we get away with not needing an index that
1588
# includes the new keys.
1590
# See <http://launchpad.net/bugs/300177> about ordering of compression
1591
# parents in the records - to be conservative, we insist that all
1592
# parents must be present to avoid expanding to a fulltext.
1594
# key = basis_parent, value = index entry to add
1595
buffered_index_entries = {}
1596
for record in stream:
1597
kind = record.storage_kind
1598
if kind.startswith('knit-') and kind.endswith('-gz'):
1599
# Check that the ID in the header of the raw knit bytes matches
1600
# the record metadata.
1601
raw_data = record._raw_record
1602
df, rec = self._parse_record_header(record.key, raw_data)
1605
parents = record.parents
1606
if record.storage_kind in delta_types:
1607
# TODO: eventually the record itself should track
1608
# compression_parent
1609
compression_parent = parents[0]
1611
compression_parent = None
1612
# Raise an error when a record is missing.
1613
if record.storage_kind == 'absent':
1614
raise RevisionNotPresent([record.key], self)
1615
elif ((record.storage_kind in knit_types)
1616
and (compression_parent is None
1617
or not self._immediate_fallback_vfs
1618
or self._index.has_key(compression_parent)
1619
or not self.has_key(compression_parent))):
1620
# we can insert the knit record literally if either it has no
1621
# compression parent OR we already have its basis in this kvf
1622
# OR the basis is not present even in the fallbacks. In the
1623
# last case it will either turn up later in the stream and all
1624
# will be well, or it won't turn up at all and we'll raise an
1627
# TODO: self.has_key is somewhat redundant with
1628
# self._index.has_key; we really want something that directly
1629
# asks if it's only present in the fallbacks. -- mbp 20081119
1630
if record.storage_kind not in native_types:
1632
adapter_key = (record.storage_kind, "knit-delta-gz")
1633
adapter = get_adapter(adapter_key)
1635
adapter_key = (record.storage_kind, "knit-ft-gz")
1636
adapter = get_adapter(adapter_key)
1637
bytes = adapter.get_bytes(record)
1639
# It's a knit record, it has a _raw_record field (even if
1640
# it was reconstituted from a network stream).
1641
bytes = record._raw_record
1642
options = [record._build_details[0]]
1643
if record._build_details[1]:
1644
options.append('no-eol')
1645
# Just blat it across.
1646
# Note: This does end up adding data on duplicate keys. As
1647
# modern repositories use atomic insertions this should not
1648
# lead to excessive growth in the event of interrupted fetches.
1649
# 'knit' repositories may suffer excessive growth, but as a
1650
# deprecated format this is tolerable. It can be fixed if
1651
# needed by in the kndx index support raising on a duplicate
1652
# add with identical parents and options.
1653
access_memo = self._access.add_raw_records(
1654
[(record.key, len(bytes))], bytes)[0]
1655
index_entry = (record.key, options, access_memo, parents)
1656
if 'fulltext' not in options:
1657
# Not a fulltext, so we need to make sure the compression
1658
# parent will also be present.
1659
# Note that pack backed knits don't need to buffer here
1660
# because they buffer all writes to the transaction level,
1661
# but we don't expose that difference at the index level. If
1662
# the query here has sufficient cost to show up in
1663
# profiling we should do that.
1665
# They're required to be physically in this
1666
# KnitVersionedFiles, not in a fallback.
1667
if not self._index.has_key(compression_parent):
1668
pending = buffered_index_entries.setdefault(
1669
compression_parent, [])
1670
pending.append(index_entry)
1673
self._index.add_records([index_entry])
1674
elif record.storage_kind == 'chunked':
1675
self.add_lines(record.key, parents,
1676
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1678
# Not suitable for direct insertion as a
1679
# delta, either because it's not the right format, or this
1680
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1681
# 0) or because it depends on a base only present in the
1683
self._access.flush()
1685
# Try getting a fulltext directly from the record.
1686
bytes = record.get_bytes_as('fulltext')
1687
except errors.UnavailableRepresentation:
1688
adapter_key = record.storage_kind, 'fulltext'
1689
adapter = get_adapter(adapter_key)
1690
bytes = adapter.get_bytes(record)
1691
lines = split_lines(bytes)
1693
self.add_lines(record.key, parents, lines)
1694
except errors.RevisionAlreadyPresent:
1696
# Add any records whose basis parent is now available.
1698
added_keys = [record.key]
1700
key = added_keys.pop(0)
1701
if key in buffered_index_entries:
1702
index_entries = buffered_index_entries[key]
1703
self._index.add_records(index_entries)
1705
[index_entry[0] for index_entry in index_entries])
1706
del buffered_index_entries[key]
1707
if buffered_index_entries:
1708
# There were index entries buffered at the end of the stream,
1709
# So these need to be added (if the index supports holding such
1710
# entries for later insertion)
1712
for key in buffered_index_entries:
1713
index_entries = buffered_index_entries[key]
1714
all_entries.extend(index_entries)
1715
self._index.add_records(
1716
all_entries, missing_compression_parents=True)
1718
def get_missing_compression_parent_keys(self):
1719
"""Return an iterable of keys of missing compression parents.
1721
Check this after calling insert_record_stream to find out if there are
1722
any missing compression parents. If there are, the records that
1723
depend on them are not able to be inserted safely. For atomic
1724
KnitVersionedFiles built on packs, the transaction should be aborted or
1725
suspended - commit will fail at this point. Nonatomic knits will error
1726
earlier because they have no staging area to put pending entries into.
1728
return self._index.get_missing_compression_parents()
1730
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1731
"""Iterate over the lines in the versioned files from keys.
1733
This may return lines from other keys. Each item the returned
1734
iterator yields is a tuple of a line and a text version that that line
1735
is present in (not introduced in).
1737
Ordering of results is in whatever order is most suitable for the
1738
underlying storage format.
1740
If a progress bar is supplied, it may be used to indicate progress.
1741
The caller is responsible for cleaning up progress bars (because this
1745
* Lines are normalised by the underlying store: they will all have \\n
1747
* Lines are returned in arbitrary order.
1748
* If a requested key did not change any lines (or didn't have any
1749
lines), it may not be mentioned at all in the result.
1751
:param pb: Progress bar supplied by caller.
1752
:return: An iterator over (line, key).
1755
pb = ui.ui_factory.nested_progress_bar()
1761
# we don't care about inclusions, the caller cares.
1762
# but we need to setup a list of records to visit.
1763
# we need key, position, length
1765
build_details = self._index.get_build_details(keys)
1766
for key, details in build_details.iteritems():
1768
key_records.append((key, details[0]))
1769
records_iter = enumerate(self._read_records_iter(key_records))
1770
for (key_idx, (key, data, sha_value)) in records_iter:
1771
pb.update('Walking content', key_idx, total)
1772
compression_parent = build_details[key][1]
1773
if compression_parent is None:
1775
line_iterator = self._factory.get_fulltext_content(data)
1778
line_iterator = self._factory.get_linedelta_content(data)
1779
# Now that we are yielding the data for this key, remove it
1782
# XXX: It might be more efficient to yield (key,
1783
# line_iterator) in the future. However for now, this is a
1784
# simpler change to integrate into the rest of the
1785
# codebase. RBC 20071110
1786
for line in line_iterator:
1789
except errors.RetryWithNewPacks, e:
1790
self._access.reload_or_raise(e)
1791
# If there are still keys we've not yet found, we look in the fallback
1792
# vfs, and hope to find them there. Note that if the keys are found
1793
# but had no changes or no content, the fallback may not return
1795
if keys and not self._immediate_fallback_vfs:
1796
# XXX: strictly the second parameter is meant to be the file id
1797
# but it's not easily accessible here.
1798
raise RevisionNotPresent(keys, repr(self))
1799
for source in self._immediate_fallback_vfs:
1803
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1804
source_keys.add(key)
1806
keys.difference_update(source_keys)
1807
pb.update('Walking content', total, total)
1809
def _make_line_delta(self, delta_seq, new_content):
1810
"""Generate a line delta from delta_seq and new_content."""
1812
for op in delta_seq.get_opcodes():
1813
if op[0] == 'equal':
1815
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1818
def _merge_annotations(self, content, parents, parent_texts={},
1819
delta=None, annotated=None,
1820
left_matching_blocks=None):
1821
"""Merge annotations for content and generate deltas.
1823
This is done by comparing the annotations based on changes to the text
1824
and generating a delta on the resulting full texts. If annotations are
1825
not being created then a simple delta is created.
1827
if left_matching_blocks is not None:
1828
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1832
for parent_key in parents:
1833
merge_content = self._get_content(parent_key, parent_texts)
1834
if (parent_key == parents[0] and delta_seq is not None):
1837
seq = patiencediff.PatienceSequenceMatcher(
1838
None, merge_content.text(), content.text())
1839
for i, j, n in seq.get_matching_blocks():
1842
# this copies (origin, text) pairs across to the new
1843
# content for any line that matches the last-checked
1845
content._lines[j:j+n] = merge_content._lines[i:i+n]
1846
# XXX: Robert says the following block is a workaround for a
1847
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1848
if content._lines and content._lines[-1][1][-1] != '\n':
1849
# The copied annotation was from a line without a trailing EOL,
1850
# reinstate one for the content object, to ensure correct
1852
line = content._lines[-1][1] + '\n'
1853
content._lines[-1] = (content._lines[-1][0], line)
1855
if delta_seq is None:
1856
reference_content = self._get_content(parents[0], parent_texts)
1857
new_texts = content.text()
1858
old_texts = reference_content.text()
1859
delta_seq = patiencediff.PatienceSequenceMatcher(
1860
None, old_texts, new_texts)
1861
return self._make_line_delta(delta_seq, content)
1863
def _parse_record(self, version_id, data):
1864
"""Parse an original format knit record.
1866
These have the last element of the key only present in the stored data.
1868
rec, record_contents = self._parse_record_unchecked(data)
1869
self._check_header_version(rec, version_id)
1870
return record_contents, rec[3]
1872
def _parse_record_header(self, key, raw_data):
1873
"""Parse a record header for consistency.
1875
:return: the header and the decompressor stream.
1876
as (stream, header_record)
1878
df = gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1881
rec = self._check_header(key, df.readline())
1882
except Exception, e:
1883
raise KnitCorrupt(self,
1884
"While reading {%s} got %s(%s)"
1885
% (key, e.__class__.__name__, str(e)))
1888
def _parse_record_unchecked(self, data):
1890
# 4168 calls in 2880 217 internal
1891
# 4168 calls to _parse_record_header in 2121
1892
# 4168 calls to readlines in 330
1893
df = gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1895
record_contents = df.readlines()
1896
except Exception, e:
1897
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1898
(data, e.__class__.__name__, str(e)))
1899
header = record_contents.pop(0)
1900
rec = self._split_header(header)
1901
last_line = record_contents.pop()
1902
if len(record_contents) != int(rec[2]):
1903
raise KnitCorrupt(self,
1904
'incorrect number of lines %s != %s'
1905
' for version {%s} %s'
1906
% (len(record_contents), int(rec[2]),
1907
rec[1], record_contents))
1908
if last_line != 'end %s\n' % rec[1]:
1909
raise KnitCorrupt(self,
1910
'unexpected version end line %r, wanted %r'
1911
% (last_line, rec[1]))
1913
return rec, record_contents
1915
def _read_records_iter(self, records):
1916
"""Read text records from data file and yield result.
1918
The result will be returned in whatever is the fastest to read.
1919
Not by the order requested. Also, multiple requests for the same
1920
record will only yield 1 response.
1921
:param records: A list of (key, access_memo) entries
1922
:return: Yields (key, contents, digest) in the order
1923
read, not the order requested
1928
# XXX: This smells wrong, IO may not be getting ordered right.
1929
needed_records = sorted(set(records), key=operator.itemgetter(1))
1930
if not needed_records:
1933
# The transport optimizes the fetching as well
1934
# (ie, reads continuous ranges.)
1935
raw_data = self._access.get_raw_records(
1936
[index_memo for key, index_memo in needed_records])
1938
for (key, index_memo), data in \
1939
izip(iter(needed_records), raw_data):
1940
content, digest = self._parse_record(key[-1], data)
1941
yield key, content, digest
1943
def _read_records_iter_raw(self, records):
1944
"""Read text records from data file and yield raw data.
1946
This unpacks enough of the text record to validate the id is
1947
as expected but thats all.
1949
Each item the iterator yields is (key, bytes,
1950
expected_sha1_of_full_text).
1952
for key, data in self._read_records_iter_unchecked(records):
1953
# validate the header (note that we can only use the suffix in
1954
# current knit records).
1955
df, rec = self._parse_record_header(key, data)
1957
yield key, data, rec[3]
1959
def _read_records_iter_unchecked(self, records):
1960
"""Read text records from data file and yield raw data.
1962
No validation is done.
1964
Yields tuples of (key, data).
1966
# setup an iterator of the external records:
1967
# uses readv so nice and fast we hope.
1969
# grab the disk data needed.
1970
needed_offsets = [index_memo for key, index_memo
1972
raw_records = self._access.get_raw_records(needed_offsets)
1974
for key, index_memo in records:
1975
data = raw_records.next()
1978
def _record_to_data(self, key, digest, lines, dense_lines=None):
1979
"""Convert key, digest, lines into a raw data block.
1981
:param key: The key of the record. Currently keys are always serialised
1982
using just the trailing component.
1983
:param dense_lines: The bytes of lines but in a denser form. For
1984
instance, if lines is a list of 1000 bytestrings each ending in \n,
1985
dense_lines may be a list with one line in it, containing all the
1986
1000's lines and their \n's. Using dense_lines if it is already
1987
known is a win because the string join to create bytes in this
1988
function spends less time resizing the final string.
1989
:return: (len, a StringIO instance with the raw data ready to read.)
1991
chunks = ["version %s %d %s\n" % (key[-1], len(lines), digest)]
1992
chunks.extend(dense_lines or lines)
1993
chunks.append("end %s\n" % key[-1])
1994
for chunk in chunks:
1995
if type(chunk) is not str:
1996
raise AssertionError(
1997
'data must be plain bytes was %s' % type(chunk))
1998
if lines and lines[-1][-1] != '\n':
1999
raise ValueError('corrupt lines value %r' % lines)
2000
compressed_bytes = tuned_gzip.chunks_to_gzip(chunks)
2001
return len(compressed_bytes), compressed_bytes
2003
def _split_header(self, line):
2006
raise KnitCorrupt(self,
2007
'unexpected number of elements in record header')
2011
"""See VersionedFiles.keys."""
2012
if 'evil' in debug.debug_flags:
2013
trace.mutter_callsite(2, "keys scales with size of history")
2014
sources = [self._index] + self._immediate_fallback_vfs
2016
for source in sources:
2017
result.update(source.keys())
2021
class _ContentMapGenerator(object):
2022
"""Generate texts or expose raw deltas for a set of texts."""
2024
def __init__(self, ordering='unordered'):
2025
self._ordering = ordering
2027
def _get_content(self, key):
2028
"""Get the content object for key."""
2029
# Note that _get_content is only called when the _ContentMapGenerator
2030
# has been constructed with just one key requested for reconstruction.
2031
if key in self.nonlocal_keys:
2032
record = self.get_record_stream().next()
2033
# Create a content object on the fly
2034
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
2035
return PlainKnitContent(lines, record.key)
2037
# local keys we can ask for directly
2038
return self._get_one_work(key)
2040
def get_record_stream(self):
2041
"""Get a record stream for the keys requested during __init__."""
2042
for record in self._work():
2046
"""Produce maps of text and KnitContents as dicts.
767
2048
:return: (text_map, content_map) where text_map contains the texts for
768
the requested versions and content_map contains the KnitContents.
769
Both dicts take version_ids as their keys.
2049
the requested versions and content_map contains the KnitContents.
771
for version_id in version_ids:
772
if not self.has_version(version_id):
773
raise RevisionNotPresent(version_id, self.filename)
774
record_map = self._get_record_map(version_ids)
779
for version_id in version_ids:
2051
# NB: By definition we never need to read remote sources unless texts
2052
# are requested from them: we don't delta across stores - and we
2053
# explicitly do not want to to prevent data loss situations.
2054
if self.global_map is None:
2055
self.global_map = self.vf.get_parent_map(self.keys)
2056
nonlocal_keys = self.nonlocal_keys
2058
missing_keys = set(nonlocal_keys)
2059
# Read from remote versioned file instances and provide to our caller.
2060
for source in self.vf._immediate_fallback_vfs:
2061
if not missing_keys:
2063
# Loop over fallback repositories asking them for texts - ignore
2064
# any missing from a particular fallback.
2065
for record in source.get_record_stream(missing_keys,
2066
self._ordering, True):
2067
if record.storage_kind == 'absent':
2068
# Not in thie particular stream, may be in one of the
2069
# other fallback vfs objects.
2071
missing_keys.remove(record.key)
2074
if self._raw_record_map is None:
2075
raise AssertionError('_raw_record_map should have been filled')
2077
for key in self.keys:
2078
if key in self.nonlocal_keys:
2080
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2083
def _get_one_work(self, requested_key):
2084
# Now, if we have calculated everything already, just return the
2086
if requested_key in self._contents_map:
2087
return self._contents_map[requested_key]
2088
# To simplify things, parse everything at once - code that wants one text
2089
# probably wants them all.
2090
# FUTURE: This function could be improved for the 'extract many' case
2091
# by tracking each component and only doing the copy when the number of
2092
# children than need to apply delta's to it is > 1 or it is part of the
2094
multiple_versions = len(self.keys) != 1
2095
if self._record_map is None:
2096
self._record_map = self.vf._raw_map_to_record_map(
2097
self._raw_record_map)
2098
record_map = self._record_map
2099
# raw_record_map is key:
2100
# Have read and parsed records at this point.
2101
for key in self.keys:
2102
if key in self.nonlocal_keys:
782
2107
while cursor is not None:
783
method, data, digest, next = record_map[cursor]
784
components.append((cursor, method, data, digest))
785
if cursor in content_map:
2109
record, record_details, digest, next = record_map[cursor]
2111
raise RevisionNotPresent(cursor, self)
2112
components.append((cursor, record, record_details, digest))
2114
if cursor in self._contents_map:
2115
# no need to plan further back
2116
components.append((cursor, None, None, None))
790
for component_id, method, data, digest in reversed(components):
791
if component_id in content_map:
792
content = content_map[component_id]
2120
for (component_id, record, record_details,
2121
digest) in reversed(components):
2122
if component_id in self._contents_map:
2123
content = self._contents_map[component_id]
794
version_idx = self._index.lookup(component_id)
795
if method == 'fulltext':
796
assert content is None
797
content = self.factory.parse_fulltext(data, version_idx)
798
elif method == 'line-delta':
799
delta = self.factory.parse_line_delta(data, version_idx)
800
content = content.copy()
801
content._lines = self._apply_delta(content._lines,
803
content_map[component_id] = content
805
if 'no-eol' in self._index.get_options(version_id):
806
content = content.copy()
807
line = content._lines[-1][1].rstrip('\n')
808
content._lines[-1] = (content._lines[-1][0], line)
809
final_content[version_id] = content
2125
content, delta = self._factory.parse_record(key[-1],
2126
record, record_details, content,
2127
copy_base_content=multiple_versions)
2128
if multiple_versions:
2129
self._contents_map[component_id] = content
811
2131
# digest here is the digest from the last applied component.
812
2132
text = content.text()
813
if sha_strings(text) != digest:
814
raise KnitCorrupt(self.filename,
815
'sha-1 does not match %s' % version_id)
817
text_map[version_id] = text
818
return text_map, final_content
820
def iter_lines_added_or_present_in_versions(self, version_ids=None,
822
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
823
if version_ids is None:
824
version_ids = self.versions()
826
pb = progress.DummyProgress()
827
# we don't care about inclusions, the caller cares.
828
# but we need to setup a list of records to visit.
829
# we need version_id, position, length
830
version_id_records = []
831
requested_versions = set(version_ids)
832
# filter for available versions
833
for version_id in requested_versions:
834
if not self.has_version(version_id):
835
raise RevisionNotPresent(version_id, self.filename)
836
# get a in-component-order queue:
837
for version_id in self.versions():
838
if version_id in requested_versions:
839
data_pos, length = self._index.get_position(version_id)
840
version_id_records.append((version_id, data_pos, length))
842
total = len(version_id_records)
843
for version_idx, (version_id, data, sha_value) in \
844
enumerate(self._data.read_records_iter(version_id_records)):
845
pb.update('Walking content.', version_idx, total)
846
method = self._index.get_method(version_id)
847
version_idx = self._index.lookup(version_id)
848
assert method in ('fulltext', 'line-delta')
849
if method == 'fulltext':
850
content = self.factory.parse_fulltext(data, version_idx)
851
for line in content.text():
854
delta = self.factory.parse_line_delta(data, version_idx)
855
for start, end, count, lines in delta:
856
for origin, line in lines:
858
pb.update('Walking content.', total, total)
860
def num_versions(self):
861
"""See VersionedFile.num_versions()."""
862
return self._index.num_versions()
864
__len__ = num_versions
866
def annotate_iter(self, version_id):
867
"""See VersionedFile.annotate_iter."""
868
content = self._get_content(version_id)
869
for origin, text in content.annotate_iter():
872
def get_parents(self, version_id):
873
"""See VersionedFile.get_parents."""
876
# 52554 calls in 1264 872 internal down from 3674
878
return self._index.get_parents(version_id)
880
raise RevisionNotPresent(version_id, self.filename)
882
def get_parents_with_ghosts(self, version_id):
883
"""See VersionedFile.get_parents."""
885
return self._index.get_parents_with_ghosts(version_id)
887
raise RevisionNotPresent(version_id, self.filename)
889
def get_ancestry(self, versions):
890
"""See VersionedFile.get_ancestry."""
891
if isinstance(versions, basestring):
892
versions = [versions]
895
self._check_versions_present(versions)
896
return self._index.get_ancestry(versions)
898
def get_ancestry_with_ghosts(self, versions):
899
"""See VersionedFile.get_ancestry_with_ghosts."""
900
if isinstance(versions, basestring):
901
versions = [versions]
904
self._check_versions_present(versions)
905
return self._index.get_ancestry_with_ghosts(versions)
907
#@deprecated_method(zero_eight)
908
def walk(self, version_ids):
909
"""See VersionedFile.walk."""
910
# We take the short path here, and extract all relevant texts
911
# and put them in a weave and let that do all the work. Far
912
# from optimal, but is much simpler.
913
# FIXME RB 20060228 this really is inefficient!
914
from bzrlib.weave import Weave
916
w = Weave(self.filename)
917
ancestry = self.get_ancestry(version_ids)
918
sorted_graph = topo_sort(self._index.get_graph())
919
version_list = [vid for vid in sorted_graph if vid in ancestry]
921
for version_id in version_list:
922
lines = self.get_lines(version_id)
923
w.add_lines(version_id, self.get_parents(version_id), lines)
925
for lineno, insert_id, dset, line in w.walk(version_ids):
926
yield lineno, insert_id, dset, line
928
def plan_merge(self, ver_a, ver_b):
929
"""See VersionedFile.plan_merge."""
930
ancestors_b = set(self.get_ancestry(ver_b))
931
def status_a(revision, text):
932
if revision in ancestors_b:
933
return 'killed-b', text
937
ancestors_a = set(self.get_ancestry(ver_a))
938
def status_b(revision, text):
939
if revision in ancestors_a:
940
return 'killed-a', text
944
annotated_a = self.annotate(ver_a)
945
annotated_b = self.annotate(ver_b)
946
plain_a = [t for (a, t) in annotated_a]
947
plain_b = [t for (a, t) in annotated_b]
948
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
951
for ai, bi, l in blocks:
952
# process all mismatched sections
953
# (last mismatched section is handled because blocks always
954
# includes a 0-length last block)
955
for revision, text in annotated_a[a_cur:ai]:
956
yield status_a(revision, text)
957
for revision, text in annotated_b[b_cur:bi]:
958
yield status_b(revision, text)
960
# and now the matched section
963
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
964
assert text_a == text_b
965
yield "unchanged", text_a
968
class _KnitComponentFile(object):
969
"""One of the files used to implement a knit database"""
971
def __init__(self, transport, filename, mode, file_mode=None,
972
create_parent_dir=False, dir_mode=None):
973
self._transport = transport
974
self._filename = filename
976
self._file_mode = file_mode
977
self._dir_mode = dir_mode
978
self._create_parent_dir = create_parent_dir
979
self._need_to_create = False
981
def check_header(self, fp):
983
if line != self.HEADER:
984
raise KnitHeaderError(badline=line)
987
"""Commit is a nop."""
990
return '%s(%s)' % (self.__class__.__name__, self._filename)
993
class _KnitIndex(_KnitComponentFile):
994
"""Manages knit index file.
996
The index is already kept in memory and read on startup, to enable
2133
actual_sha = sha_strings(text)
2134
if actual_sha != digest:
2135
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2136
if multiple_versions:
2137
return self._contents_map[requested_key]
2141
def _wire_bytes(self):
2142
"""Get the bytes to put on the wire for 'key'.
2144
The first collection of bytes asked for returns the serialised
2145
raw_record_map and the additional details (key, parent) for key.
2146
Subsequent calls return just the additional details (key, parent).
2147
The wire storage_kind given for the first key is 'knit-delta-closure',
2148
For subsequent keys it is 'knit-delta-closure-ref'.
2150
:param key: A key from the content generator.
2151
:return: Bytes to put on the wire.
2154
# kind marker for dispatch on the far side,
2155
lines.append('knit-delta-closure')
2157
if self.vf._factory.annotated:
2158
lines.append('annotated')
2161
# then the list of keys
2162
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2163
if key not in self.nonlocal_keys]))
2164
# then the _raw_record_map in serialised form:
2166
# for each item in the map:
2168
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2169
# one line with method
2170
# one line with noeol
2171
# one line with next ('' for None)
2172
# one line with byte count of the record bytes
2174
for key, (record_bytes, (method, noeol), next) in \
2175
self._raw_record_map.iteritems():
2176
key_bytes = '\x00'.join(key)
2177
parents = self.global_map.get(key, None)
2179
parent_bytes = 'None:'
2181
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2182
method_bytes = method
2188
next_bytes = '\x00'.join(next)
2191
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2192
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2193
len(record_bytes), record_bytes))
2194
map_bytes = ''.join(map_byte_list)
2195
lines.append(map_bytes)
2196
bytes = '\n'.join(lines)
2200
class _VFContentMapGenerator(_ContentMapGenerator):
2201
"""Content map generator reading from a VersionedFiles object."""
2203
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2204
global_map=None, raw_record_map=None, ordering='unordered'):
2205
"""Create a _ContentMapGenerator.
2207
:param versioned_files: The versioned files that the texts are being
2209
:param keys: The keys to produce content maps for.
2210
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2211
which are known to not be in this knit, but rather in one of the
2213
:param global_map: The result of get_parent_map(keys) (or a supermap).
2214
This is required if get_record_stream() is to be used.
2215
:param raw_record_map: A unparsed raw record map to use for answering
2218
_ContentMapGenerator.__init__(self, ordering=ordering)
2219
# The vf to source data from
2220
self.vf = versioned_files
2222
self.keys = list(keys)
2223
# Keys known to be in fallback vfs objects
2224
if nonlocal_keys is None:
2225
self.nonlocal_keys = set()
2227
self.nonlocal_keys = frozenset(nonlocal_keys)
2228
# Parents data for keys to be returned in get_record_stream
2229
self.global_map = global_map
2230
# The chunked lists for self.keys in text form
2232
# A cache of KnitContent objects used in extracting texts.
2233
self._contents_map = {}
2234
# All the knit records needed to assemble the requested keys as full
2236
self._record_map = None
2237
if raw_record_map is None:
2238
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2241
self._raw_record_map = raw_record_map
2242
# the factory for parsing records
2243
self._factory = self.vf._factory
2246
class _NetworkContentMapGenerator(_ContentMapGenerator):
2247
"""Content map generator sourced from a network stream."""
2249
def __init__(self, bytes, line_end):
2250
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2252
self.global_map = {}
2253
self._raw_record_map = {}
2254
self._contents_map = {}
2255
self._record_map = None
2256
self.nonlocal_keys = []
2257
# Get access to record parsing facilities
2258
self.vf = KnitVersionedFiles(None, None)
2261
line_end = bytes.find('\n', start)
2262
line = bytes[start:line_end]
2263
start = line_end + 1
2264
if line == 'annotated':
2265
self._factory = KnitAnnotateFactory()
2267
self._factory = KnitPlainFactory()
2268
# list of keys to emit in get_record_stream
2269
line_end = bytes.find('\n', start)
2270
line = bytes[start:line_end]
2271
start = line_end + 1
2273
tuple(segment.split('\x00')) for segment in line.split('\t')
2275
# now a loop until the end. XXX: It would be nice if this was just a
2276
# bunch of the same records as get_record_stream(..., False) gives, but
2277
# there is a decent sized gap stopping that at the moment.
2281
line_end = bytes.find('\n', start)
2282
key = tuple(bytes[start:line_end].split('\x00'))
2283
start = line_end + 1
2284
# 1 line with parents (None: for None, '' for ())
2285
line_end = bytes.find('\n', start)
2286
line = bytes[start:line_end]
2291
[tuple(segment.split('\x00')) for segment in line.split('\t')
2293
self.global_map[key] = parents
2294
start = line_end + 1
2295
# one line with method
2296
line_end = bytes.find('\n', start)
2297
line = bytes[start:line_end]
2299
start = line_end + 1
2300
# one line with noeol
2301
line_end = bytes.find('\n', start)
2302
line = bytes[start:line_end]
2304
start = line_end + 1
2305
# one line with next ('' for None)
2306
line_end = bytes.find('\n', start)
2307
line = bytes[start:line_end]
2311
next = tuple(bytes[start:line_end].split('\x00'))
2312
start = line_end + 1
2313
# one line with byte count of the record bytes
2314
line_end = bytes.find('\n', start)
2315
line = bytes[start:line_end]
2317
start = line_end + 1
2319
record_bytes = bytes[start:start+count]
2320
start = start + count
2322
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2324
def get_record_stream(self):
2325
"""Get a record stream for for keys requested by the bytestream."""
2327
for key in self.keys:
2328
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2331
def _wire_bytes(self):
2335
class _KndxIndex(object):
2336
"""Manages knit index files
2338
The index is kept in memory and read on startup, to enable
997
2339
fast lookups of revision information. The cursor of the index
998
2340
file is always pointing to the end, making it easy to append
1010
2352
Duplicate entries may be written to the index for a single version id
1011
2353
if this is done then the latter one completely replaces the former:
1012
this allows updates to correct version and parent information.
2354
this allows updates to correct version and parent information.
1013
2355
Note that the two entries may share the delta, and that successive
1014
2356
annotations and references MUST point to the first entry.
1016
2358
The index file on disc contains a header, followed by one line per knit
1017
2359
record. The same revision can be present in an index file more than once.
1018
The first occurrence gets assigned a sequence number starting from 0.
2360
The first occurrence gets assigned a sequence number starting from 0.
1020
2362
The format of a single line is
1021
2363
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1022
2364
REVISION_ID is a utf8-encoded revision id
1023
FLAGS is a comma separated list of flags about the record. Values include
2365
FLAGS is a comma separated list of flags about the record. Values include
1024
2366
no-eol, line-delta, fulltext.
1025
2367
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1026
that the the compressed data starts at.
2368
that the compressed data starts at.
1027
2369
LENGTH is the ascii representation of the length of the data file.
1028
2370
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1030
2372
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1031
2373
revision id already in the knit that is a parent of REVISION_ID.
1032
2374
The ' :' marker is the end of record marker.
1034
2376
partial writes:
1035
when a write is interrupted to the index file, it will result in a line that
1036
does not end in ' :'. If the ' :' is not present at the end of a line, or at
1037
the end of the file, then the record that is missing it will be ignored by
2377
when a write is interrupted to the index file, it will result in a line
2378
that does not end in ' :'. If the ' :' is not present at the end of a line,
2379
or at the end of the file, then the record that is missing it will be
2380
ignored by the parser.
1040
2382
When writing new records to the index file, the data is preceded by '\n'
1041
2383
to ensure that records always start on new lines even if the last write was
1042
2384
interrupted. As a result its normal for the last line in the index to be
1043
2385
missing a trailing newline. One can be added with no harmful effects.
2387
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2388
where prefix is e.g. the (fileid,) for .texts instances or () for
2389
constant-mapped things like .revisions, and the old state is
2390
tuple(history_vector, cache_dict). This is used to prevent having an
2391
ABI change with the C extension that reads .kndx files.
1046
2394
HEADER = "# bzr knit index 8\n"
1048
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1049
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1051
def _cache_version(self, version_id, options, pos, size, parents):
2396
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2397
"""Create a _KndxIndex on transport using mapper."""
2398
self._transport = transport
2399
self._mapper = mapper
2400
self._get_scope = get_scope
2401
self._allow_writes = allow_writes
2402
self._is_locked = is_locked
2404
self.has_graph = True
2406
def add_records(self, records, random_id=False, missing_compression_parents=False):
2407
"""Add multiple records to the index.
2409
:param records: a list of tuples:
2410
(key, options, access_memo, parents).
2411
:param random_id: If True the ids being added were randomly generated
2412
and no check for existence will be performed.
2413
:param missing_compression_parents: If True the records being added are
2414
only compressed against texts already in the index (or inside
2415
records). If False the records all refer to unavailable texts (or
2416
texts inside records) as compression parents.
2418
if missing_compression_parents:
2419
# It might be nice to get the edge of the records. But keys isn't
2421
keys = sorted(record[0] for record in records)
2422
raise errors.RevisionNotPresent(keys, self)
2424
for record in records:
2427
path = self._mapper.map(key) + '.kndx'
2428
path_keys = paths.setdefault(path, (prefix, []))
2429
path_keys[1].append(record)
2430
for path in sorted(paths):
2431
prefix, path_keys = paths[path]
2432
self._load_prefixes([prefix])
2434
orig_history = self._kndx_cache[prefix][1][:]
2435
orig_cache = self._kndx_cache[prefix][0].copy()
2438
for key, options, (_, pos, size), parents in path_keys:
2440
# kndx indices cannot be parentless.
2442
line = "\n%s %s %s %s %s :" % (
2443
key[-1], ','.join(options), pos, size,
2444
self._dictionary_compress(parents))
2445
if type(line) is not str:
2446
raise AssertionError(
2447
'data must be utf8 was %s' % type(line))
2449
self._cache_key(key, options, pos, size, parents)
2450
if len(orig_history):
2451
self._transport.append_bytes(path, ''.join(lines))
2453
self._init_index(path, lines)
2455
# If any problems happen, restore the original values and re-raise
2456
self._kndx_cache[prefix] = (orig_cache, orig_history)
2459
def scan_unvalidated_index(self, graph_index):
2460
"""See _KnitGraphIndex.scan_unvalidated_index."""
2461
# Because kndx files do not support atomic insertion via separate index
2462
# files, they do not support this method.
2463
raise NotImplementedError(self.scan_unvalidated_index)
2465
def get_missing_compression_parents(self):
2466
"""See _KnitGraphIndex.get_missing_compression_parents."""
2467
# Because kndx files do not support atomic insertion via separate index
2468
# files, they do not support this method.
2469
raise NotImplementedError(self.get_missing_compression_parents)
2471
def _cache_key(self, key, options, pos, size, parent_keys):
1052
2472
"""Cache a version record in the history array and index cache.
1054
This is inlined into __init__ for performance. KEEP IN SYNC.
2474
This is inlined into _load_data for performance. KEEP IN SYNC.
1055
2475
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2479
version_id = key[-1]
2480
# last-element only for compatibilty with the C load_data.
2481
parents = tuple(parent[-1] for parent in parent_keys)
2482
for parent in parent_keys:
2483
if parent[:-1] != prefix:
2484
raise ValueError("mismatched prefixes for %r, %r" % (
2486
cache, history = self._kndx_cache[prefix]
1058
2487
# only want the _history index to reference the 1st index entry
1059
2488
# for version_id
1060
if version_id not in self._cache:
1061
index = len(self._history)
1062
self._history.append(version_id)
2489
if version_id not in cache:
2490
index = len(history)
2491
history.append(version_id)
1064
index = self._cache[version_id][5]
1065
self._cache[version_id] = (version_id,
2493
index = cache[version_id][5]
2494
cache[version_id] = (version_id,
1072
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1073
create_parent_dir=False, delay_create=False, dir_mode=None):
1074
_KnitComponentFile.__init__(self, transport, filename, mode,
1075
file_mode=file_mode,
1076
create_parent_dir=create_parent_dir,
1079
# position in _history is the 'official' index for a revision
1080
# but the values may have come from a newer entry.
1081
# so - wc -l of a knit index is != the number of unique names
1084
decode_utf8 = cache_utf8.decode
1085
pb = bzrlib.ui.ui_factory.nested_progress_bar()
2501
def check_header(self, fp):
2502
line = fp.readline()
2504
# An empty file can actually be treated as though the file doesn't
2506
raise errors.NoSuchFile(self)
2507
if line != self.HEADER:
2508
raise KnitHeaderError(badline=line, filename=self)
2510
def _check_read(self):
2511
if not self._is_locked():
2512
raise errors.ObjectNotLocked(self)
2513
if self._get_scope() != self._scope:
2516
def _check_write_ok(self):
2517
"""Assert if not writes are permitted."""
2518
if not self._is_locked():
2519
raise errors.ObjectNotLocked(self)
2520
if self._get_scope() != self._scope:
2522
if self._mode != 'w':
2523
raise errors.ReadOnlyObjectDirtiedError(self)
2525
def get_build_details(self, keys):
2526
"""Get the method, index_memo and compression parent for keys.
2528
Ghosts are omitted from the result.
2530
:param keys: An iterable of keys.
2531
:return: A dict of key:(index_memo, compression_parent, parents,
2534
opaque structure to pass to read_records to extract the raw
2537
Content that this record is built upon, may be None
2539
Logical parents of this node
2541
extra information about the content which needs to be passed to
2542
Factory.parse_record
2544
parent_map = self.get_parent_map(keys)
2547
if key not in parent_map:
2549
method = self.get_method(key)
2550
parents = parent_map[key]
2551
if method == 'fulltext':
2552
compression_parent = None
2554
compression_parent = parents[0]
2555
noeol = 'no-eol' in self.get_options(key)
2556
index_memo = self.get_position(key)
2557
result[key] = (index_memo, compression_parent,
2558
parents, (method, noeol))
2561
def get_method(self, key):
2562
"""Return compression method of specified key."""
2563
options = self.get_options(key)
2564
if 'fulltext' in options:
2566
elif 'line-delta' in options:
2569
raise errors.KnitIndexUnknownMethod(self, options)
2571
def get_options(self, key):
2572
"""Return a list representing options.
2576
prefix, suffix = self._split_key(key)
2577
self._load_prefixes([prefix])
1090
pb.update('read knit index', count, total)
1091
fp = self._transport.get(self._filename)
2579
return self._kndx_cache[prefix][0][suffix][1]
2581
raise RevisionNotPresent(key, self)
2583
def find_ancestry(self, keys):
2584
"""See CombinedGraphIndex.find_ancestry()"""
2585
prefixes = set(key[:-1] for key in keys)
2586
self._load_prefixes(prefixes)
2589
missing_keys = set()
2590
pending_keys = list(keys)
2591
# This assumes that keys will not reference parents in a different
2592
# prefix, which is accurate so far.
2594
key = pending_keys.pop()
2595
if key in parent_map:
2599
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2601
missing_keys.add(key)
2603
parent_keys = tuple([prefix + (suffix,)
2604
for suffix in suffix_parents])
2605
parent_map[key] = parent_keys
2606
pending_keys.extend([p for p in parent_keys
2607
if p not in parent_map])
2608
return parent_map, missing_keys
2610
def get_parent_map(self, keys):
2611
"""Get a map of the parents of keys.
2613
:param keys: The keys to look up parents for.
2614
:return: A mapping from keys to parents. Absent keys are absent from
2617
# Parse what we need to up front, this potentially trades off I/O
2618
# locality (.kndx and .knit in the same block group for the same file
2619
# id) for less checking in inner loops.
2620
prefixes = set(key[:-1] for key in keys)
2621
self._load_prefixes(prefixes)
2626
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2630
result[key] = tuple(prefix + (suffix,) for
2631
suffix in suffix_parents)
2634
def get_position(self, key):
2635
"""Return details needed to access the version.
2637
:return: a tuple (key, data position, size) to hand to the access
2638
logic to get the record.
2640
prefix, suffix = self._split_key(key)
2641
self._load_prefixes([prefix])
2642
entry = self._kndx_cache[prefix][0][suffix]
2643
return key, entry[2], entry[3]
2645
has_key = _mod_index._has_key_from_parent_map
2647
def _init_index(self, path, extra_lines=[]):
2648
"""Initialize an index."""
2650
sio.write(self.HEADER)
2651
sio.writelines(extra_lines)
2653
self._transport.put_file_non_atomic(path, sio,
2654
create_parent_dir=True)
2655
# self._create_parent_dir)
2656
# mode=self._file_mode,
2657
# dir_mode=self._dir_mode)
2660
"""Get all the keys in the collection.
2662
The keys are not ordered.
2665
# Identify all key prefixes.
2666
# XXX: A bit hacky, needs polish.
2667
if type(self._mapper) is ConstantMapper:
2671
for quoted_relpath in self._transport.iter_files_recursive():
2672
path, ext = os.path.splitext(quoted_relpath)
2674
prefixes = [self._mapper.unmap(path) for path in relpaths]
2675
self._load_prefixes(prefixes)
2676
for prefix in prefixes:
2677
for suffix in self._kndx_cache[prefix][1]:
2678
result.add(prefix + (suffix,))
2681
def _load_prefixes(self, prefixes):
2682
"""Load the indices for prefixes."""
2684
for prefix in prefixes:
2685
if prefix not in self._kndx_cache:
2686
# the load_data interface writes to these variables.
2689
self._filename = prefix
1093
self.check_header(fp)
1094
# readlines reads the whole file at once:
1095
# bad for transports like http, good for local disk
1096
# we save 60 ms doing this one change (
1097
# from calling readline each time to calling
1099
# probably what we want for nice behaviour on
1100
# http is a incremental readlines that yields, or
1101
# a check for local vs non local indexes,
1102
for l in fp.readlines():
1104
if len(rec) < 5 or rec[-1] != ':':
1106
# FIXME: in the future we should determine if its a
1107
# short write - and ignore it
1108
# or a different failure, and raise. RBC 20060407
1112
#pb.update('read knit index', count, total)
1113
# See self._parse_parents
1115
for value in rec[4:-1]:
1117
# uncompressed reference
1118
parents.append(decode_utf8(value[1:]))
1120
# this is 15/4000ms faster than isinstance,
1122
# this function is called thousands of times a
1123
# second so small variations add up.
1124
assert value.__class__ is str
1125
parents.append(self._history[int(value)])
1126
# end self._parse_parents
1127
# self._cache_version(decode_utf8(rec[0]),
1128
# rec[1].split(','),
1132
# --- self._cache_version
1133
# only want the _history index to reference the 1st
1134
# index entry for version_id
1135
version_id = decode_utf8(rec[0])
1136
if version_id not in self._cache:
1137
index = len(self._history)
1138
self._history.append(version_id)
1140
index = self._cache[version_id][5]
1141
self._cache[version_id] = (version_id,
1147
# --- self._cache_version
1150
except NoSuchFile, e:
1151
if mode != 'w' or not create:
1154
self._need_to_create = True
1156
self._transport.put_bytes_non_atomic(self._filename,
1157
self.HEADER, mode=self._file_mode)
1160
pb.update('read knit index', total, total)
1163
def _parse_parents(self, compressed_parents):
1164
"""convert a list of string parent values into version ids.
1166
ints are looked up in the index.
1167
.FOO values are ghosts and converted in to FOO.
1169
NOTE: the function is retained here for clarity, and for possible
1170
use in partial index reads. However bulk processing now has
1171
it inlined in __init__ for inner-loop optimisation.
2691
path = self._mapper.map(prefix) + '.kndx'
2692
fp = self._transport.get(path)
2694
# _load_data may raise NoSuchFile if the target knit is
2696
_load_data(self, fp)
2699
self._kndx_cache[prefix] = (self._cache, self._history)
2704
self._kndx_cache[prefix] = ({}, [])
2705
if type(self._mapper) is ConstantMapper:
2706
# preserve behaviour for revisions.kndx etc.
2707
self._init_index(path)
2712
missing_keys = _mod_index._missing_keys_from_parent_map
2714
def _partition_keys(self, keys):
2715
"""Turn keys into a dict of prefix:suffix_list."""
2718
prefix_keys = result.setdefault(key[:-1], [])
2719
prefix_keys.append(key[-1])
2722
def _dictionary_compress(self, keys):
2723
"""Dictionary compress keys.
2725
:param keys: The keys to generate references to.
2726
:return: A string representation of keys. keys which are present are
2727
dictionary compressed, and others are emitted as fulltext with a
1174
for value in compressed_parents:
1175
if value[-1] == '.':
1176
# uncompressed reference
1177
result.append(cache_utf8.decode_utf8(value[1:]))
1179
# this is 15/4000ms faster than isinstance,
1180
# this function is called thousands of times a
1181
# second so small variations add up.
1182
assert value.__class__ is str
1183
result.append(self._history[int(value)])
1186
def get_graph(self):
1188
for version_id, index in self._cache.iteritems():
1189
graph.append((version_id, index[4]))
1192
def get_ancestry(self, versions):
1193
"""See VersionedFile.get_ancestry."""
1194
# get a graph of all the mentioned versions:
1196
pending = set(versions)
1198
version = pending.pop()
1199
parents = self._cache[version][4]
1200
# got the parents ok
1202
parents = [parent for parent in parents if parent in self._cache]
1203
for parent in parents:
1204
# if not completed and not a ghost
1205
if parent not in graph:
1207
graph[version] = parents
1208
return topo_sort(graph.items())
1210
def get_ancestry_with_ghosts(self, versions):
1211
"""See VersionedFile.get_ancestry_with_ghosts."""
1212
# get a graph of all the mentioned versions:
1214
pending = set(versions)
1216
version = pending.pop()
1218
parents = self._cache[version][4]
1224
# got the parents ok
1225
for parent in parents:
1226
if parent not in graph:
1228
graph[version] = parents
1229
return topo_sort(graph.items())
1231
def num_versions(self):
1232
return len(self._history)
1234
__len__ = num_versions
1236
def get_versions(self):
1237
return self._history
1239
def idx_to_name(self, idx):
1240
return self._history[idx]
1242
def lookup(self, version_id):
1243
assert version_id in self._cache
1244
return self._cache[version_id][5]
1246
def _version_list_to_index(self, versions):
1247
encode_utf8 = cache_utf8.encode
1248
2732
result_list = []
1249
for version in versions:
1250
if version in self._cache:
2733
prefix = keys[0][:-1]
2734
cache = self._kndx_cache[prefix][0]
2736
if key[:-1] != prefix:
2737
# kndx indices cannot refer across partitioned storage.
2738
raise ValueError("mismatched prefixes for %r" % keys)
2739
if key[-1] in cache:
1251
2740
# -- inlined lookup() --
1252
result_list.append(str(self._cache[version][5]))
2741
result_list.append(str(cache[key[-1]][5]))
1253
2742
# -- end lookup () --
1255
result_list.append('.' + encode_utf8(version))
2744
result_list.append('.' + key[-1])
1256
2745
return ' '.join(result_list)
1258
def add_version(self, version_id, options, pos, size, parents):
1259
"""Add a version record to the index."""
1260
self.add_versions(((version_id, options, pos, size, parents),))
1262
def add_versions(self, versions):
1263
"""Add multiple versions to the index.
1265
:param versions: a list of tuples:
1266
(version_id, options, pos, size, parents).
1269
encode_utf8 = cache_utf8.encode
1270
orig_history = self._history[:]
1271
orig_cache = self._cache.copy()
1274
for version_id, options, pos, size, parents in versions:
1275
line = "\n%s %s %s %s %s :" % (encode_utf8(version_id),
1279
self._version_list_to_index(parents))
1280
assert isinstance(line, str), \
1281
'content must be utf-8 encoded: %r' % (line,)
1283
self._cache_version(version_id, options, pos, size, parents)
1284
if not self._need_to_create:
1285
self._transport.append_bytes(self._filename, ''.join(lines))
1288
sio.write(self.HEADER)
1289
sio.writelines(lines)
1291
self._transport.put_file_non_atomic(self._filename, sio,
1292
create_parent_dir=self._create_parent_dir,
1293
mode=self._file_mode,
1294
dir_mode=self._dir_mode)
1295
self._need_to_create = False
1297
# If any problems happen, restore the original values and re-raise
1298
self._history = orig_history
1299
self._cache = orig_cache
1302
def has_version(self, version_id):
1303
"""True if the version is in the index."""
1304
return (version_id in self._cache)
1306
def get_position(self, version_id):
1307
"""Return data position and size of specified version."""
1308
return (self._cache[version_id][2], \
1309
self._cache[version_id][3])
1311
def get_method(self, version_id):
1312
"""Return compression method of specified version."""
1313
options = self._cache[version_id][1]
1314
if 'fulltext' in options:
2747
def _reset_cache(self):
2748
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2749
# (cache_dict, history_vector) for parsed kndx files.
2750
self._kndx_cache = {}
2751
self._scope = self._get_scope()
2752
allow_writes = self._allow_writes()
2758
def _sort_keys_by_io(self, keys, positions):
2759
"""Figure out an optimal order to read the records for the given keys.
2761
Sort keys, grouped by index and sorted by position.
2763
:param keys: A list of keys whose records we want to read. This will be
2765
:param positions: A dict, such as the one returned by
2766
_get_components_positions()
2769
def get_sort_key(key):
2770
index_memo = positions[key][1]
2771
# Group by prefix and position. index_memo[0] is the key, so it is
2772
# (file_id, revision_id) and we don't want to sort on revision_id,
2773
# index_memo[1] is the position, and index_memo[2] is the size,
2774
# which doesn't matter for the sort
2775
return index_memo[0][:-1], index_memo[1]
2776
return keys.sort(key=get_sort_key)
2778
_get_total_build_size = _get_total_build_size
2780
def _split_key(self, key):
2781
"""Split key into a prefix and suffix."""
2782
return key[:-1], key[-1]
2785
class _KnitGraphIndex(object):
2786
"""A KnitVersionedFiles index layered on GraphIndex."""
2788
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2789
add_callback=None, track_external_parent_refs=False):
2790
"""Construct a KnitGraphIndex on a graph_index.
2792
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2793
:param is_locked: A callback to check whether the object should answer
2795
:param deltas: Allow delta-compressed records.
2796
:param parents: If True, record knits parents, if not do not record
2798
:param add_callback: If not None, allow additions to the index and call
2799
this callback with a list of added GraphIndex nodes:
2800
[(node, value, node_refs), ...]
2801
:param is_locked: A callback, returns True if the index is locked and
2803
:param track_external_parent_refs: If True, record all external parent
2804
references parents from added records. These can be retrieved
2805
later by calling get_missing_parents().
2807
self._add_callback = add_callback
2808
self._graph_index = graph_index
2809
self._deltas = deltas
2810
self._parents = parents
2811
if deltas and not parents:
2812
# XXX: TODO: Delta tree and parent graph should be conceptually
2814
raise KnitCorrupt(self, "Cannot do delta compression without "
2816
self.has_graph = parents
2817
self._is_locked = is_locked
2818
self._missing_compression_parents = set()
2819
if track_external_parent_refs:
2820
self._key_dependencies = _KeyRefs()
2822
self._key_dependencies = None
2825
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2827
def add_records(self, records, random_id=False,
2828
missing_compression_parents=False):
2829
"""Add multiple records to the index.
2831
This function does not insert data into the Immutable GraphIndex
2832
backing the KnitGraphIndex, instead it prepares data for insertion by
2833
the caller and checks that it is safe to insert then calls
2834
self._add_callback with the prepared GraphIndex nodes.
2836
:param records: a list of tuples:
2837
(key, options, access_memo, parents).
2838
:param random_id: If True the ids being added were randomly generated
2839
and no check for existence will be performed.
2840
:param missing_compression_parents: If True the records being added are
2841
only compressed against texts already in the index (or inside
2842
records). If False the records all refer to unavailable texts (or
2843
texts inside records) as compression parents.
2845
if not self._add_callback:
2846
raise errors.ReadOnlyError(self)
2847
# we hope there are no repositories with inconsistent parentage
2851
compression_parents = set()
2852
key_dependencies = self._key_dependencies
2853
for (key, options, access_memo, parents) in records:
2855
parents = tuple(parents)
2856
if key_dependencies is not None:
2857
key_dependencies.add_references(key, parents)
2858
index, pos, size = access_memo
2859
if 'no-eol' in options:
2863
value += "%d %d" % (pos, size)
2864
if not self._deltas:
2865
if 'line-delta' in options:
2866
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2869
if 'line-delta' in options:
2870
node_refs = (parents, (parents[0],))
2871
if missing_compression_parents:
2872
compression_parents.add(parents[0])
2874
node_refs = (parents, ())
2876
node_refs = (parents, )
2879
raise KnitCorrupt(self, "attempt to add node with parents "
2880
"in parentless index.")
2882
keys[key] = (value, node_refs)
2885
present_nodes = self._get_entries(keys)
2886
for (index, key, value, node_refs) in present_nodes:
2887
parents = node_refs[:1]
2888
# Sometimes these are passed as a list rather than a tuple
2889
passed = static_tuple.as_tuples(keys[key])
2890
passed_parents = passed[1][:1]
2891
if (value[0] != keys[key][0][0] or
2892
parents != passed_parents):
2893
node_refs = static_tuple.as_tuples(node_refs)
2894
raise KnitCorrupt(self, "inconsistent details in add_records"
2895
": %s %s" % ((value, node_refs), passed))
2899
for key, (value, node_refs) in keys.iteritems():
2900
result.append((key, value, node_refs))
2902
for key, (value, node_refs) in keys.iteritems():
2903
result.append((key, value))
2904
self._add_callback(result)
2905
if missing_compression_parents:
2906
# This may appear to be incorrect (it does not check for
2907
# compression parents that are in the existing graph index),
2908
# but such records won't have been buffered, so this is
2909
# actually correct: every entry when
2910
# missing_compression_parents==True either has a missing parent, or
2911
# a parent that is one of the keys in records.
2912
compression_parents.difference_update(keys)
2913
self._missing_compression_parents.update(compression_parents)
2914
# Adding records may have satisfied missing compression parents.
2915
self._missing_compression_parents.difference_update(keys)
2917
def scan_unvalidated_index(self, graph_index):
2918
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2920
This allows this _KnitGraphIndex to keep track of any missing
2921
compression parents we may want to have filled in to make those
2924
:param graph_index: A GraphIndex
2927
new_missing = graph_index.external_references(ref_list_num=1)
2928
new_missing.difference_update(self.get_parent_map(new_missing))
2929
self._missing_compression_parents.update(new_missing)
2930
if self._key_dependencies is not None:
2931
# Add parent refs from graph_index (and discard parent refs that
2932
# the graph_index has).
2933
for node in graph_index.iter_all_entries():
2934
self._key_dependencies.add_references(node[1], node[3][0])
2936
def get_missing_compression_parents(self):
2937
"""Return the keys of missing compression parents.
2939
Missing compression parents occur when a record stream was missing
2940
basis texts, or a index was scanned that had missing basis texts.
2942
return frozenset(self._missing_compression_parents)
2944
def get_missing_parents(self):
2945
"""Return the keys of missing parents."""
2946
# If updating this, you should also update
2947
# groupcompress._GCGraphIndex.get_missing_parents
2948
# We may have false positives, so filter those out.
2949
self._key_dependencies.satisfy_refs_for_keys(
2950
self.get_parent_map(self._key_dependencies.get_unsatisfied_refs()))
2951
return frozenset(self._key_dependencies.get_unsatisfied_refs())
2953
def _check_read(self):
2954
"""raise if reads are not permitted."""
2955
if not self._is_locked():
2956
raise errors.ObjectNotLocked(self)
2958
def _check_write_ok(self):
2959
"""Assert if writes are not permitted."""
2960
if not self._is_locked():
2961
raise errors.ObjectNotLocked(self)
2963
def _compression_parent(self, an_entry):
2964
# return the key that an_entry is compressed against, or None
2965
# Grab the second parent list (as deltas implies parents currently)
2966
compression_parents = an_entry[3][1]
2967
if not compression_parents:
2969
if len(compression_parents) != 1:
2970
raise AssertionError(
2971
"Too many compression parents: %r" % compression_parents)
2972
return compression_parents[0]
2974
def get_build_details(self, keys):
2975
"""Get the method, index_memo and compression parent for version_ids.
2977
Ghosts are omitted from the result.
2979
:param keys: An iterable of keys.
2980
:return: A dict of key:
2981
(index_memo, compression_parent, parents, record_details).
2983
opaque structure to pass to read_records to extract the raw
2986
Content that this record is built upon, may be None
2988
Logical parents of this node
2990
extra information about the content which needs to be passed to
2991
Factory.parse_record
2995
entries = self._get_entries(keys, False)
2996
for entry in entries:
2998
if not self._parents:
3001
parents = entry[3][0]
3002
if not self._deltas:
3003
compression_parent_key = None
3005
compression_parent_key = self._compression_parent(entry)
3006
noeol = (entry[2][0] == 'N')
3007
if compression_parent_key:
3008
method = 'line-delta'
3011
result[key] = (self._node_to_position(entry),
3012
compression_parent_key, parents,
3016
def _get_entries(self, keys, check_present=False):
3017
"""Get the entries for keys.
3019
:param keys: An iterable of index key tuples.
3024
for node in self._graph_index.iter_entries(keys):
3026
found_keys.add(node[1])
3028
# adapt parentless index to the rest of the code.
3029
for node in self._graph_index.iter_entries(keys):
3030
yield node[0], node[1], node[2], ()
3031
found_keys.add(node[1])
3033
missing_keys = keys.difference(found_keys)
3035
raise RevisionNotPresent(missing_keys.pop(), self)
3037
def get_method(self, key):
3038
"""Return compression method of specified key."""
3039
return self._get_method(self._get_node(key))
3041
def _get_method(self, node):
3042
if not self._deltas:
1315
3043
return 'fulltext'
1317
assert 'line-delta' in options
3044
if self._compression_parent(node):
1318
3045
return 'line-delta'
1320
def get_options(self, version_id):
1321
return self._cache[version_id][1]
1323
def get_parents(self, version_id):
1324
"""Return parents of specified version ignoring ghosts."""
1325
return [parent for parent in self._cache[version_id][4]
1326
if parent in self._cache]
1328
def get_parents_with_ghosts(self, version_id):
1329
"""Return parents of specified version with ghosts."""
1330
return self._cache[version_id][4]
1332
def check_versions_present(self, version_ids):
1333
"""Check that all specified versions are present."""
1334
version_ids = set(version_ids)
1335
for version_id in list(version_ids):
1336
if version_id in self._cache:
1337
version_ids.remove(version_id)
1339
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1342
class _KnitData(_KnitComponentFile):
1343
"""Contents of the knit data file"""
1345
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1346
create_parent_dir=False, delay_create=False,
1348
_KnitComponentFile.__init__(self, transport, filename, mode,
1349
file_mode=file_mode,
1350
create_parent_dir=create_parent_dir,
1352
self._checked = False
1353
# TODO: jam 20060713 conceptually, this could spill to disk
1354
# if the cached size gets larger than a certain amount
1355
# but it complicates the model a bit, so for now just use
1356
# a simple dictionary
1358
self._do_cache = False
1361
self._need_to_create = create
1363
self._transport.put_bytes_non_atomic(self._filename, '',
1364
mode=self._file_mode)
1366
def enable_cache(self):
1367
"""Enable caching of reads."""
1368
self._do_cache = True
1370
def clear_cache(self):
1371
"""Clear the record cache."""
1372
self._do_cache = False
1375
def _open_file(self):
1377
return self._transport.get(self._filename)
1382
def _record_to_data(self, version_id, digest, lines):
1383
"""Convert version_id, digest, lines into a raw data block.
1385
:return: (len, a StringIO instance with the raw data ready to read.)
1388
data_file = GzipFile(None, mode='wb', fileobj=sio)
1390
version_id_utf8 = cache_utf8.encode(version_id)
1391
data_file.writelines(chain(
1392
["version %s %d %s\n" % (version_id_utf8,
1396
["end %s\n" % version_id_utf8]))
1403
def add_raw_record(self, raw_data):
1404
"""Append a prepared record to the data file.
1406
:return: the offset in the data file raw_data was written.
1408
assert isinstance(raw_data, str), 'data must be plain bytes'
1409
if not self._need_to_create:
1410
return self._transport.append_bytes(self._filename, raw_data)
1412
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1413
create_parent_dir=self._create_parent_dir,
1414
mode=self._file_mode,
1415
dir_mode=self._dir_mode)
1416
self._need_to_create = False
1419
def add_record(self, version_id, digest, lines):
1420
"""Write new text record to disk. Returns the position in the
1421
file where it was written."""
1422
size, sio = self._record_to_data(version_id, digest, lines)
1424
if not self._need_to_create:
1425
start_pos = self._transport.append_file(self._filename, sio)
1427
self._transport.put_file_non_atomic(self._filename, sio,
1428
create_parent_dir=self._create_parent_dir,
1429
mode=self._file_mode,
1430
dir_mode=self._dir_mode)
1431
self._need_to_create = False
1434
self._cache[version_id] = sio.getvalue()
1435
return start_pos, size
1437
def _parse_record_header(self, version_id, raw_data):
1438
"""Parse a record header for consistency.
1440
:return: the header and the decompressor stream.
1441
as (stream, header_record)
1443
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1444
rec = df.readline().split()
1446
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
1447
if cache_utf8.decode(rec[1]) != version_id:
1448
raise KnitCorrupt(self._filename,
1449
'unexpected version, wanted %r, got %r' % (
1450
version_id, rec[1]))
1453
def _parse_record(self, version_id, data):
1455
# 4168 calls in 2880 217 internal
1456
# 4168 calls to _parse_record_header in 2121
1457
# 4168 calls to readlines in 330
1458
df, rec = self._parse_record_header(version_id, data)
1459
record_contents = df.readlines()
1460
l = record_contents.pop()
1461
assert len(record_contents) == int(rec[2])
1462
if l != 'end %s\n' % cache_utf8.encode(version_id):
1463
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
1466
return record_contents, rec[3]
1468
def read_records_iter_raw(self, records):
1469
"""Read text records from data file and yield raw data.
1471
This unpacks enough of the text record to validate the id is
1472
as expected but thats all.
1474
# setup an iterator of the external records:
1475
# uses readv so nice and fast we hope.
1477
# grab the disk data needed.
1479
# Don't check _cache if it is empty
1480
needed_offsets = [(pos, size) for version_id, pos, size
1482
if version_id not in self._cache]
1484
needed_offsets = [(pos, size) for version_id, pos, size
1487
raw_records = self._transport.readv(self._filename, needed_offsets)
1490
for version_id, pos, size in records:
1491
if version_id in self._cache:
1492
# This data has already been validated
1493
data = self._cache[version_id]
1495
pos, data = raw_records.next()
1497
self._cache[version_id] = data
1499
# validate the header
1500
df, rec = self._parse_record_header(version_id, data)
1502
yield version_id, data
1504
def read_records_iter(self, records):
1505
"""Read text records from data file and yield result.
1507
The result will be returned in whatever is the fastest to read.
1508
Not by the order requested. Also, multiple requests for the same
1509
record will only yield 1 response.
1510
:param records: A list of (version_id, pos, len) entries
1511
:return: Yields (version_id, contents, digest) in the order
1512
read, not the order requested
1518
# Skip records we have alread seen
1519
yielded_records = set()
1520
needed_records = set()
1521
for record in records:
1522
if record[0] in self._cache:
1523
if record[0] in yielded_records:
1525
yielded_records.add(record[0])
1526
data = self._cache[record[0]]
1527
content, digest = self._parse_record(record[0], data)
1528
yield (record[0], content, digest)
1530
needed_records.add(record)
1531
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1533
needed_records = sorted(set(records), key=operator.itemgetter(1))
1535
if not needed_records:
1538
# The transport optimizes the fetching as well
1539
# (ie, reads continuous ranges.)
1540
readv_response = self._transport.readv(self._filename,
1541
[(pos, size) for version_id, pos, size in needed_records])
1543
for (version_id, pos, size), (pos, data) in \
1544
izip(iter(needed_records), readv_response):
1545
content, digest = self._parse_record(version_id, data)
1547
self._cache[version_id] = data
1548
yield version_id, content, digest
1550
def read_records(self, records):
1551
"""Read records into a dictionary."""
1553
for record_id, content, digest in \
1554
self.read_records_iter(records):
1555
components[record_id] = (content, digest)
1559
class InterKnit(InterVersionedFile):
1560
"""Optimised code paths for knit to knit operations."""
1562
_matching_file_from_factory = KnitVersionedFile
1563
_matching_file_to_factory = KnitVersionedFile
1566
def is_compatible(source, target):
1567
"""Be compatible with knits. """
1569
return (isinstance(source, KnitVersionedFile) and
1570
isinstance(target, KnitVersionedFile))
1571
except AttributeError:
1574
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1575
"""See InterVersionedFile.join."""
1576
assert isinstance(self.source, KnitVersionedFile)
1577
assert isinstance(self.target, KnitVersionedFile)
1579
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1584
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1586
version_ids = list(version_ids)
1587
if None in version_ids:
1588
version_ids.remove(None)
1590
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1591
this_versions = set(self.target._index.get_versions())
1592
needed_versions = self.source_ancestry - this_versions
1593
cross_check_versions = self.source_ancestry.intersection(this_versions)
1594
mismatched_versions = set()
1595
for version in cross_check_versions:
1596
# scan to include needed parents.
1597
n1 = set(self.target.get_parents_with_ghosts(version))
1598
n2 = set(self.source.get_parents_with_ghosts(version))
1600
# FIXME TEST this check for cycles being introduced works
1601
# the logic is we have a cycle if in our graph we are an
1602
# ancestor of any of the n2 revisions.
1608
parent_ancestors = self.source.get_ancestry(parent)
1609
if version in parent_ancestors:
1610
raise errors.GraphCycleError([parent, version])
1611
# ensure this parent will be available later.
1612
new_parents = n2.difference(n1)
1613
needed_versions.update(new_parents.difference(this_versions))
1614
mismatched_versions.add(version)
1616
if not needed_versions and not mismatched_versions:
1618
full_list = topo_sort(self.source.get_graph())
1620
version_list = [i for i in full_list if (not self.target.has_version(i)
1621
and i in needed_versions)]
1625
copy_queue_records = []
1627
for version_id in version_list:
1628
options = self.source._index.get_options(version_id)
1629
parents = self.source._index.get_parents_with_ghosts(version_id)
1630
# check that its will be a consistent copy:
1631
for parent in parents:
1632
# if source has the parent, we must :
1633
# * already have it or
1634
# * have it scheduled already
1635
# otherwise we don't care
1636
assert (self.target.has_version(parent) or
1637
parent in copy_set or
1638
not self.source.has_version(parent))
1639
data_pos, data_size = self.source._index.get_position(version_id)
1640
copy_queue_records.append((version_id, data_pos, data_size))
1641
copy_queue.append((version_id, options, parents))
1642
copy_set.add(version_id)
1644
# data suck the join:
1646
total = len(version_list)
1649
for (version_id, raw_data), \
1650
(version_id2, options, parents) in \
1651
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1653
assert version_id == version_id2, 'logic error, inconsistent results'
1655
pb.update("Joining knit", count, total)
1656
raw_records.append((version_id, options, parents, len(raw_data)))
1657
raw_datum.append(raw_data)
1658
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1660
for version in mismatched_versions:
1661
# FIXME RBC 20060309 is this needed?
1662
n1 = set(self.target.get_parents_with_ghosts(version))
1663
n2 = set(self.source.get_parents_with_ghosts(version))
1664
# write a combined record to our history preserving the current
1665
# parents as first in the list
1666
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1667
self.target.fix_parents(version, new_parents)
1673
InterVersionedFile.register_optimiser(InterKnit)
1676
class WeaveToKnit(InterVersionedFile):
1677
"""Optimised code paths for weave to knit operations."""
1679
_matching_file_from_factory = bzrlib.weave.WeaveFile
1680
_matching_file_to_factory = KnitVersionedFile
1683
def is_compatible(source, target):
1684
"""Be compatible with weaves to knits."""
1686
return (isinstance(source, bzrlib.weave.Weave) and
1687
isinstance(target, KnitVersionedFile))
1688
except AttributeError:
1691
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1692
"""See InterVersionedFile.join."""
1693
assert isinstance(self.source, bzrlib.weave.Weave)
1694
assert isinstance(self.target, KnitVersionedFile)
1696
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1701
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1703
version_ids = list(version_ids)
1705
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1706
this_versions = set(self.target._index.get_versions())
1707
needed_versions = self.source_ancestry - this_versions
1708
cross_check_versions = self.source_ancestry.intersection(this_versions)
1709
mismatched_versions = set()
1710
for version in cross_check_versions:
1711
# scan to include needed parents.
1712
n1 = set(self.target.get_parents_with_ghosts(version))
1713
n2 = set(self.source.get_parents(version))
1714
# if all of n2's parents are in n1, then its fine.
1715
if n2.difference(n1):
1716
# FIXME TEST this check for cycles being introduced works
1717
# the logic is we have a cycle if in our graph we are an
1718
# ancestor of any of the n2 revisions.
1724
parent_ancestors = self.source.get_ancestry(parent)
1725
if version in parent_ancestors:
1726
raise errors.GraphCycleError([parent, version])
1727
# ensure this parent will be available later.
1728
new_parents = n2.difference(n1)
1729
needed_versions.update(new_parents.difference(this_versions))
1730
mismatched_versions.add(version)
1732
if not needed_versions and not mismatched_versions:
1734
full_list = topo_sort(self.source.get_graph())
1736
version_list = [i for i in full_list if (not self.target.has_version(i)
1737
and i in needed_versions)]
1741
total = len(version_list)
1742
for version_id in version_list:
1743
pb.update("Converting to knit", count, total)
1744
parents = self.source.get_parents(version_id)
1745
# check that its will be a consistent copy:
1746
for parent in parents:
1747
# if source has the parent, we must already have it
1748
assert (self.target.has_version(parent))
1749
self.target.add_lines(
1750
version_id, parents, self.source.get_lines(version_id))
1753
for version in mismatched_versions:
1754
# FIXME RBC 20060309 is this needed?
1755
n1 = set(self.target.get_parents_with_ghosts(version))
1756
n2 = set(self.source.get_parents(version))
1757
# write a combined record to our history preserving the current
1758
# parents as first in the list
1759
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1760
self.target.fix_parents(version, new_parents)
1766
InterVersionedFile.register_optimiser(WeaveToKnit)
1769
class KnitSequenceMatcher(difflib.SequenceMatcher):
1770
"""Knit tuned sequence matcher.
1772
This is based on profiling of difflib which indicated some improvements
1773
for our usage pattern.
1776
def find_longest_match(self, alo, ahi, blo, bhi):
1777
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1779
If isjunk is not defined:
1781
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1782
alo <= i <= i+k <= ahi
1783
blo <= j <= j+k <= bhi
1784
and for all (i',j',k') meeting those conditions,
1787
and if i == i', j <= j'
1789
In other words, of all maximal matching blocks, return one that
1790
starts earliest in a, and of all those maximal matching blocks that
1791
start earliest in a, return the one that starts earliest in b.
1793
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1794
>>> s.find_longest_match(0, 5, 0, 9)
1797
If isjunk is defined, first the longest matching block is
1798
determined as above, but with the additional restriction that no
1799
junk element appears in the block. Then that block is extended as
1800
far as possible by matching (only) junk elements on both sides. So
1801
the resulting block never matches on junk except as identical junk
1802
happens to be adjacent to an "interesting" match.
1804
Here's the same example as before, but considering blanks to be
1805
junk. That prevents " abcd" from matching the " abcd" at the tail
1806
end of the second sequence directly. Instead only the "abcd" can
1807
match, and matches the leftmost "abcd" in the second sequence:
1809
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1810
>>> s.find_longest_match(0, 5, 0, 9)
1813
If no blocks match, return (alo, blo, 0).
1815
>>> s = SequenceMatcher(None, "ab", "c")
1816
>>> s.find_longest_match(0, 2, 0, 1)
1820
# CAUTION: stripping common prefix or suffix would be incorrect.
1824
# Longest matching block is "ab", but if common prefix is
1825
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1826
# strip, so ends up claiming that ab is changed to acab by
1827
# inserting "ca" in the middle. That's minimal but unintuitive:
1828
# "it's obvious" that someone inserted "ac" at the front.
1829
# Windiff ends up at the same place as diff, but by pairing up
1830
# the unique 'b's and then matching the first two 'a's.
1832
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1833
besti, bestj, bestsize = alo, blo, 0
1834
# find longest junk-free match
1835
# during an iteration of the loop, j2len[j] = length of longest
1836
# junk-free match ending with a[i-1] and b[j]
1840
for i in xrange(alo, ahi):
1841
# look at all instances of a[i] in b; note that because
1842
# b2j has no junk keys, the loop is skipped if a[i] is junk
1843
j2lenget = j2len.get
3049
def _get_node(self, key):
3051
return list(self._get_entries([key]))[0]
3053
raise RevisionNotPresent(key, self)
3055
def get_options(self, key):
3056
"""Return a list representing options.
3060
node = self._get_node(key)
3061
options = [self._get_method(node)]
3062
if node[2][0] == 'N':
3063
options.append('no-eol')
3066
def find_ancestry(self, keys):
3067
"""See CombinedGraphIndex.find_ancestry()"""
3068
return self._graph_index.find_ancestry(keys, 0)
3070
def get_parent_map(self, keys):
3071
"""Get a map of the parents of keys.
3073
:param keys: The keys to look up parents for.
3074
:return: A mapping from keys to parents. Absent keys are absent from
3078
nodes = self._get_entries(keys)
3082
result[node[1]] = node[3][0]
3085
result[node[1]] = None
3088
def get_position(self, key):
3089
"""Return details needed to access the version.
3091
:return: a tuple (index, data position, size) to hand to the access
3092
logic to get the record.
3094
node = self._get_node(key)
3095
return self._node_to_position(node)
3097
has_key = _mod_index._has_key_from_parent_map
3100
"""Get all the keys in the collection.
3102
The keys are not ordered.
3105
return [node[1] for node in self._graph_index.iter_all_entries()]
3107
missing_keys = _mod_index._missing_keys_from_parent_map
3109
def _node_to_position(self, node):
3110
"""Convert an index value to position details."""
3111
bits = node[2][1:].split(' ')
3112
return node[0], int(bits[0]), int(bits[1])
3114
def _sort_keys_by_io(self, keys, positions):
3115
"""Figure out an optimal order to read the records for the given keys.
3117
Sort keys, grouped by index and sorted by position.
3119
:param keys: A list of keys whose records we want to read. This will be
3121
:param positions: A dict, such as the one returned by
3122
_get_components_positions()
3125
def get_index_memo(key):
3126
# index_memo is at offset [1]. It is made up of (GraphIndex,
3127
# position, size). GI is an object, which will be unique for each
3128
# pack file. This causes us to group by pack file, then sort by
3129
# position. Size doesn't matter, but it isn't worth breaking up the
3131
return positions[key][1]
3132
return keys.sort(key=get_index_memo)
3134
_get_total_build_size = _get_total_build_size
3137
class _KnitKeyAccess(object):
3138
"""Access to records in .knit files."""
3140
def __init__(self, transport, mapper):
3141
"""Create a _KnitKeyAccess with transport and mapper.
3143
:param transport: The transport the access object is rooted at.
3144
:param mapper: The mapper used to map keys to .knit files.
3146
self._transport = transport
3147
self._mapper = mapper
3149
def add_raw_records(self, key_sizes, raw_data):
3150
"""Add raw knit bytes to a storage area.
3152
The data is spooled to the container writer in one bytes-record per
3155
:param sizes: An iterable of tuples containing the key and size of each
3157
:param raw_data: A bytestring containing the data.
3158
:return: A list of memos to retrieve the record later. Each memo is an
3159
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3160
length), where the key is the record key.
3162
if type(raw_data) is not str:
3163
raise AssertionError(
3164
'data must be plain bytes was %s' % type(raw_data))
3167
# TODO: This can be tuned for writing to sftp and other servers where
3168
# append() is relatively expensive by grouping the writes to each key
3170
for key, size in key_sizes:
3171
path = self._mapper.map(key)
3173
base = self._transport.append_bytes(path + '.knit',
3174
raw_data[offset:offset+size])
3175
except errors.NoSuchFile:
3176
self._transport.mkdir(osutils.dirname(path))
3177
base = self._transport.append_bytes(path + '.knit',
3178
raw_data[offset:offset+size])
3182
result.append((key, base, size))
3186
"""Flush pending writes on this access object.
3188
For .knit files this is a no-op.
3192
def get_raw_records(self, memos_for_retrieval):
3193
"""Get the raw bytes for a records.
3195
:param memos_for_retrieval: An iterable containing the access memo for
3196
retrieving the bytes.
3197
:return: An iterator over the bytes of the records.
3199
# first pass, group into same-index request to minimise readv's issued.
3201
current_prefix = None
3202
for (key, offset, length) in memos_for_retrieval:
3203
if current_prefix == key[:-1]:
3204
current_list.append((offset, length))
3206
if current_prefix is not None:
3207
request_lists.append((current_prefix, current_list))
3208
current_prefix = key[:-1]
3209
current_list = [(offset, length)]
3210
# handle the last entry
3211
if current_prefix is not None:
3212
request_lists.append((current_prefix, current_list))
3213
for prefix, read_vector in request_lists:
3214
path = self._mapper.map(prefix) + '.knit'
3215
for pos, data in self._transport.readv(path, read_vector):
3219
class _DirectPackAccess(object):
3220
"""Access to data in one or more packs with less translation."""
3222
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
3223
"""Create a _DirectPackAccess object.
3225
:param index_to_packs: A dict mapping index objects to the transport
3226
and file names for obtaining data.
3227
:param reload_func: A function to call if we determine that the pack
3228
files have moved and we need to reload our caches. See
3229
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
3231
self._container_writer = None
3232
self._write_index = None
3233
self._indices = index_to_packs
3234
self._reload_func = reload_func
3235
self._flush_func = flush_func
3237
def add_raw_records(self, key_sizes, raw_data):
3238
"""Add raw knit bytes to a storage area.
3240
The data is spooled to the container writer in one bytes-record per
3243
:param sizes: An iterable of tuples containing the key and size of each
3245
:param raw_data: A bytestring containing the data.
3246
:return: A list of memos to retrieve the record later. Each memo is an
3247
opaque index memo. For _DirectPackAccess the memo is (index, pos,
3248
length), where the index field is the write_index object supplied
3249
to the PackAccess object.
3251
if type(raw_data) is not str:
3252
raise AssertionError(
3253
'data must be plain bytes was %s' % type(raw_data))
3256
for key, size in key_sizes:
3257
p_offset, p_length = self._container_writer.add_bytes_record(
3258
raw_data[offset:offset+size], [])
3260
result.append((self._write_index, p_offset, p_length))
3264
"""Flush pending writes on this access object.
3266
This will flush any buffered writes to a NewPack.
3268
if self._flush_func is not None:
1846
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1847
# following improvement
1848
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1849
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1850
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1852
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1853
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1854
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
3271
def get_raw_records(self, memos_for_retrieval):
3272
"""Get the raw bytes for a records.
3274
:param memos_for_retrieval: An iterable containing the (index, pos,
3275
length) memo for retrieving the bytes. The Pack access method
3276
looks up the pack to use for a given record in its index_to_pack
3278
:return: An iterator over the bytes of the records.
3280
# first pass, group into same-index requests
3282
current_index = None
3283
for (index, offset, length) in memos_for_retrieval:
3284
if current_index == index:
3285
current_list.append((offset, length))
3287
if current_index is not None:
3288
request_lists.append((current_index, current_list))
3289
current_index = index
3290
current_list = [(offset, length)]
3291
# handle the last entry
3292
if current_index is not None:
3293
request_lists.append((current_index, current_list))
3294
for index, offsets in request_lists:
3296
transport, path = self._indices[index]
1858
3297
except KeyError:
3298
# A KeyError here indicates that someone has triggered an index
3299
# reload, and this index has gone missing, we need to start
3301
if self._reload_func is None:
3302
# If we don't have a _reload_func there is nothing that can
3305
raise errors.RetryWithNewPacks(index,
3306
reload_occurred=True,
3307
exc_info=sys.exc_info())
3309
reader = pack.make_readv_reader(transport, path, offsets)
3310
for names, read_func in reader.iter_records():
3311
yield read_func(None)
3312
except errors.NoSuchFile:
3313
# A NoSuchFile error indicates that a pack file has gone
3314
# missing on disk, we need to trigger a reload, and start over.
3315
if self._reload_func is None:
3317
raise errors.RetryWithNewPacks(transport.abspath(path),
3318
reload_occurred=False,
3319
exc_info=sys.exc_info())
3321
def set_writer(self, writer, index, transport_packname):
3322
"""Set a writer to use for adding data."""
3323
if index is not None:
3324
self._indices[index] = transport_packname
3325
self._container_writer = writer
3326
self._write_index = index
3328
def reload_or_raise(self, retry_exc):
3329
"""Try calling the reload function, or re-raise the original exception.
3331
This should be called after _DirectPackAccess raises a
3332
RetryWithNewPacks exception. This function will handle the common logic
3333
of determining when the error is fatal versus being temporary.
3334
It will also make sure that the original exception is raised, rather
3335
than the RetryWithNewPacks exception.
3337
If this function returns, then the calling function should retry
3338
whatever operation was being performed. Otherwise an exception will
3341
:param retry_exc: A RetryWithNewPacks exception.
3344
if self._reload_func is None:
3346
elif not self._reload_func():
3347
# The reload claimed that nothing changed
3348
if not retry_exc.reload_occurred:
3349
# If there wasn't an earlier reload, then we really were
3350
# expecting to find changes. We didn't find them, so this is a
3354
exc_class, exc_value, exc_traceback = retry_exc.exc_info
3355
raise exc_class, exc_value, exc_traceback
3358
def annotate_knit(knit, revision_id):
3359
"""Annotate a knit with no cached annotations.
3361
This implementation is for knits with no cached annotations.
3362
It will work for knits with cached annotations, but this is not
3365
annotator = _KnitAnnotator(knit)
3366
return iter(annotator.annotate_flat(revision_id))
3369
class _KnitAnnotator(annotate.Annotator):
3370
"""Build up the annotations for a text."""
3372
def __init__(self, vf):
3373
annotate.Annotator.__init__(self, vf)
3375
# TODO: handle Nodes which cannot be extracted
3376
# self._ghosts = set()
3378
# Map from (key, parent_key) => matching_blocks, should be 'use once'
3379
self._matching_blocks = {}
3381
# KnitContent objects
3382
self._content_objects = {}
3383
# The number of children that depend on this fulltext content object
3384
self._num_compression_children = {}
3385
# Delta records that need their compression parent before they can be
3387
self._pending_deltas = {}
3388
# Fulltext records that are waiting for their parents fulltexts before
3389
# they can be yielded for annotation
3390
self._pending_annotation = {}
3392
self._all_build_details = {}
3394
def _get_build_graph(self, key):
3395
"""Get the graphs for building texts and annotations.
3397
The data you need for creating a full text may be different than the
3398
data you need to annotate that text. (At a minimum, you need both
3399
parents to create an annotation, but only need 1 parent to generate the
3402
:return: A list of (key, index_memo) records, suitable for
3403
passing to read_records_iter to start reading in the raw data from
3406
pending = set([key])
3409
self._num_needed_children[key] = 1
3411
# get all pending nodes
3412
this_iteration = pending
3413
build_details = self._vf._index.get_build_details(this_iteration)
3414
self._all_build_details.update(build_details)
3415
# new_nodes = self._vf._index._get_entries(this_iteration)
3417
for key, details in build_details.iteritems():
3418
(index_memo, compression_parent, parent_keys,
3419
record_details) = details
3420
self._parent_map[key] = parent_keys
3421
self._heads_provider = None
3422
records.append((key, index_memo))
3423
# Do we actually need to check _annotated_lines?
3424
pending.update([p for p in parent_keys
3425
if p not in self._all_build_details])
3427
for parent_key in parent_keys:
3428
if parent_key in self._num_needed_children:
3429
self._num_needed_children[parent_key] += 1
3431
self._num_needed_children[parent_key] = 1
3432
if compression_parent:
3433
if compression_parent in self._num_compression_children:
3434
self._num_compression_children[compression_parent] += 1
3436
self._num_compression_children[compression_parent] = 1
3438
missing_versions = this_iteration.difference(build_details.keys())
3439
if missing_versions:
3440
for key in missing_versions:
3441
if key in self._parent_map and key in self._text_cache:
3442
# We already have this text ready, we just need to
3443
# yield it later so we get it annotated
3445
parent_keys = self._parent_map[key]
3446
for parent_key in parent_keys:
3447
if parent_key in self._num_needed_children:
3448
self._num_needed_children[parent_key] += 1
3450
self._num_needed_children[parent_key] = 1
3451
pending.update([p for p in parent_keys
3452
if p not in self._all_build_details])
3454
raise errors.RevisionNotPresent(key, self._vf)
3455
# Generally we will want to read the records in reverse order, because
3456
# we find the parent nodes after the children
3458
return records, ann_keys
3460
def _get_needed_texts(self, key, pb=None):
3461
# if True or len(self._vf._immediate_fallback_vfs) > 0:
3462
if len(self._vf._immediate_fallback_vfs) > 0:
3463
# If we have fallbacks, go to the generic path
3464
for v in annotate.Annotator._get_needed_texts(self, key, pb=pb):
3469
records, ann_keys = self._get_build_graph(key)
3470
for idx, (sub_key, text, num_lines) in enumerate(
3471
self._extract_texts(records)):
3473
pb.update('annotating', idx, len(records))
3474
yield sub_key, text, num_lines
3475
for sub_key in ann_keys:
3476
text = self._text_cache[sub_key]
3477
num_lines = len(text) # bad assumption
3478
yield sub_key, text, num_lines
3480
except errors.RetryWithNewPacks, e:
3481
self._vf._access.reload_or_raise(e)
3482
# The cached build_details are no longer valid
3483
self._all_build_details.clear()
3485
def _cache_delta_blocks(self, key, compression_parent, delta, lines):
3486
parent_lines = self._text_cache[compression_parent]
3487
blocks = list(KnitContent.get_line_delta_blocks(delta, parent_lines, lines))
3488
self._matching_blocks[(key, compression_parent)] = blocks
3490
def _expand_record(self, key, parent_keys, compression_parent, record,
3493
if compression_parent:
3494
if compression_parent not in self._content_objects:
3495
# Waiting for the parent
3496
self._pending_deltas.setdefault(compression_parent, []).append(
3497
(key, parent_keys, record, record_details))
3499
# We have the basis parent, so expand the delta
3500
num = self._num_compression_children[compression_parent]
3503
base_content = self._content_objects.pop(compression_parent)
3504
self._num_compression_children.pop(compression_parent)
1866
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1868
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1871
# Extend the best by non-junk elements on each end. In particular,
1872
# "popular" non-junk elements aren't in b2j, which greatly speeds
1873
# the inner loop above, but also means "the best" match so far
1874
# doesn't contain any junk *or* popular non-junk elements.
1875
while besti > alo and bestj > blo and \
1876
not isbjunk(b[bestj-1]) and \
1877
a[besti-1] == b[bestj-1]:
1878
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1879
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1880
not isbjunk(b[bestj+bestsize]) and \
1881
a[besti+bestsize] == b[bestj+bestsize]:
1884
# Now that we have a wholly interesting match (albeit possibly
1885
# empty!), we may as well suck up the matching junk on each
1886
# side of it too. Can't think of a good reason not to, and it
1887
# saves post-processing the (possibly considerable) expense of
1888
# figuring out what to do with it. In the case of an empty
1889
# interesting match, this is clearly the right thing to do,
1890
# because no other kind of match is possible in the regions.
1891
while besti > alo and bestj > blo and \
1892
isbjunk(b[bestj-1]) and \
1893
a[besti-1] == b[bestj-1]:
1894
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1895
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1896
isbjunk(b[bestj+bestsize]) and \
1897
a[besti+bestsize] == b[bestj+bestsize]:
1898
bestsize = bestsize + 1
1900
return besti, bestj, bestsize
3506
self._num_compression_children[compression_parent] = num
3507
base_content = self._content_objects[compression_parent]
3508
# It is tempting to want to copy_base_content=False for the last
3509
# child object. However, whenever noeol=False,
3510
# self._text_cache[parent_key] is content._lines. So mutating it
3511
# gives very bad results.
3512
# The alternative is to copy the lines into text cache, but then we
3513
# are copying anyway, so just do it here.
3514
content, delta = self._vf._factory.parse_record(
3515
key, record, record_details, base_content,
3516
copy_base_content=True)
3519
content, _ = self._vf._factory.parse_record(
3520
key, record, record_details, None)
3521
if self._num_compression_children.get(key, 0) > 0:
3522
self._content_objects[key] = content
3523
lines = content.text()
3524
self._text_cache[key] = lines
3525
if delta is not None:
3526
self._cache_delta_blocks(key, compression_parent, delta, lines)
3529
def _get_parent_annotations_and_matches(self, key, text, parent_key):
3530
"""Get the list of annotations for the parent, and the matching lines.
3532
:param text: The opaque value given by _get_needed_texts
3533
:param parent_key: The key for the parent text
3534
:return: (parent_annotations, matching_blocks)
3535
parent_annotations is a list as long as the number of lines in
3537
matching_blocks is a list of (parent_idx, text_idx, len) tuples
3538
indicating which lines match between the two texts
3540
block_key = (key, parent_key)
3541
if block_key in self._matching_blocks:
3542
blocks = self._matching_blocks.pop(block_key)
3543
parent_annotations = self._annotations_cache[parent_key]
3544
return parent_annotations, blocks
3545
return annotate.Annotator._get_parent_annotations_and_matches(self,
3546
key, text, parent_key)
3548
def _process_pending(self, key):
3549
"""The content for 'key' was just processed.
3551
Determine if there is any more pending work to be processed.
3554
if key in self._pending_deltas:
3555
compression_parent = key
3556
children = self._pending_deltas.pop(key)
3557
for child_key, parent_keys, record, record_details in children:
3558
lines = self._expand_record(child_key, parent_keys,
3560
record, record_details)
3561
if self._check_ready_for_annotations(child_key, parent_keys):
3562
to_return.append(child_key)
3563
# Also check any children that are waiting for this parent to be
3565
if key in self._pending_annotation:
3566
children = self._pending_annotation.pop(key)
3567
to_return.extend([c for c, p_keys in children
3568
if self._check_ready_for_annotations(c, p_keys)])
3571
def _check_ready_for_annotations(self, key, parent_keys):
3572
"""return true if this text is ready to be yielded.
3574
Otherwise, this will return False, and queue the text into
3575
self._pending_annotation
3577
for parent_key in parent_keys:
3578
if parent_key not in self._annotations_cache:
3579
# still waiting on at least one parent text, so queue it up
3580
# Note that if there are multiple parents, we need to wait
3582
self._pending_annotation.setdefault(parent_key,
3583
[]).append((key, parent_keys))
3587
def _extract_texts(self, records):
3588
"""Extract the various texts needed based on records"""
3589
# We iterate in the order read, rather than a strict order requested
3590
# However, process what we can, and put off to the side things that
3591
# still need parents, cleaning them up when those parents are
3594
# 1) As 'records' are read, see if we can expand these records into
3595
# Content objects (and thus lines)
3596
# 2) If a given line-delta is waiting on its compression parent, it
3597
# gets queued up into self._pending_deltas, otherwise we expand
3598
# it, and put it into self._text_cache and self._content_objects
3599
# 3) If we expanded the text, we will then check to see if all
3600
# parents have also been processed. If so, this text gets yielded,
3601
# else this record gets set aside into pending_annotation
3602
# 4) Further, if we expanded the text in (2), we will then check to
3603
# see if there are any children in self._pending_deltas waiting to
3604
# also be processed. If so, we go back to (2) for those
3605
# 5) Further again, if we yielded the text, we can then check if that
3606
# 'unlocks' any of the texts in pending_annotations, which should
3607
# then get yielded as well
3608
# Note that both steps 4 and 5 are 'recursive' in that unlocking one
3609
# compression child could unlock yet another, and yielding a fulltext
3610
# will also 'unlock' the children that are waiting on that annotation.
3611
# (Though also, unlocking 1 parent's fulltext, does not unlock a child
3612
# if other parents are also waiting.)
3613
# We want to yield content before expanding child content objects, so
3614
# that we know when we can re-use the content lines, and the annotation
3615
# code can know when it can stop caching fulltexts, as well.
3617
# Children that are missing their compression parent
3619
for (key, record, digest) in self._vf._read_records_iter(records):
3621
details = self._all_build_details[key]
3622
(_, compression_parent, parent_keys, record_details) = details
3623
lines = self._expand_record(key, parent_keys, compression_parent,
3624
record, record_details)
3626
# Pending delta should be queued up
3628
# At this point, we may be able to yield this content, if all
3629
# parents are also finished
3630
yield_this_text = self._check_ready_for_annotations(key,
3633
# All parents present
3634
yield key, lines, len(lines)
3635
to_process = self._process_pending(key)
3637
this_process = to_process
3639
for key in this_process:
3640
lines = self._text_cache[key]
3641
yield key, lines, len(lines)
3642
to_process.extend(self._process_pending(key))
3645
from bzrlib._knit_load_data_pyx import _load_data_c as _load_data
3646
except ImportError, e:
3647
osutils.failed_to_load_extension(e)
3648
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
