750
248
for start, end, c, lines in delta:
751
249
out.append('%d,%d,%d\n' % (start, end, c))
250
out.extend([text for origin, text in lines])
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(VersionedFilesWithFallbacks):
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.
254
def make_empty_knit(transport, relpath):
255
"""Construct a empty knit at the specified location."""
256
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
260
class KnitVersionedFile(VersionedFile):
261
"""Weave-like structure with faster random access.
263
A knit stores a number of texts and a summary of the relationships
264
between them. Texts are identified by a string version-id. Texts
265
are normally stored and retrieved as a series of lines, but can
266
also be passed as single strings.
268
Lines are stored with the trailing newline (if any) included, to
269
avoid special cases for files with no final newline. Lines are
270
composed of 8-bit characters, not unicode. The combination of
271
these approaches should mean any 'binary' file can be safely
272
stored and retrieved.
275
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
276
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
278
"""Construct a knit at location specified by relpath.
280
:param create: If not True, only open an existing knit.
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
282
if deprecated_passed(basis_knit):
283
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
284
" deprecated as of bzr 0.9.",
285
DeprecationWarning, stacklevel=2)
286
if access_mode is None:
288
super(KnitVersionedFile, self).__init__(access_mode)
289
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
290
self.transport = transport
291
self.filename = relpath
292
self.factory = factory or KnitAnnotateFactory()
293
self.writable = (access_mode == 'w')
296
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
297
access_mode, create=create, file_mode=file_mode)
298
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
299
access_mode, create=create and not len(self), file_mode=file_mode)
884
301
def __repr__(self):
885
return "%s(%r, %r)" % (
886
self.__class__.__name__,
890
def without_fallbacks(self):
891
"""Return a clone of this object without any fallbacks configured."""
892
return KnitVersionedFiles(self._index, self._access,
893
self._max_delta_chain, self._factory.annotated,
896
def add_fallback_versioned_files(self, a_versioned_files):
897
"""Add a source of texts for texts not present in this knit.
899
:param a_versioned_files: A VersionedFiles object.
901
self._immediate_fallback_vfs.append(a_versioned_files)
903
def add_lines(self, key, parents, lines, parent_texts=None,
904
left_matching_blocks=None, nostore_sha=None, random_id=False,
906
"""See VersionedFiles.add_lines()."""
907
self._index._check_write_ok()
908
self._check_add(key, lines, random_id, check_content)
910
# The caller might pass None if there is no graph data, but kndx
911
# indexes can't directly store that, so we give them
912
# an empty tuple instead.
914
line_bytes = ''.join(lines)
915
return self._add(key, lines, parents,
916
parent_texts, left_matching_blocks, nostore_sha, random_id,
917
line_bytes=line_bytes)
919
def _add_text(self, key, parents, text, nostore_sha=None, random_id=False):
920
"""See VersionedFiles._add_text()."""
921
self._index._check_write_ok()
922
self._check_add(key, None, random_id, check_content=False)
923
if text.__class__ is not str:
924
raise errors.BzrBadParameterUnicode("text")
926
# The caller might pass None if there is no graph data, but kndx
927
# indexes can't directly store that, so we give them
928
# an empty tuple instead.
930
return self._add(key, None, parents,
931
None, None, nostore_sha, random_id,
934
def _add(self, key, lines, parents, parent_texts,
935
left_matching_blocks, nostore_sha, random_id,
937
"""Add a set of lines on top of version specified by parents.
939
Any versions not present will be converted into ghosts.
941
:param lines: A list of strings where each one is a single line (has a
942
single newline at the end of the string) This is now optional
943
(callers can pass None). It is left in its location for backwards
944
compatibility. It should ''.join(lines) must == line_bytes
945
:param line_bytes: A single string containing the content
947
We pass both lines and line_bytes because different routes bring the
948
values to this function. And for memory efficiency, we don't want to
949
have to split/join on-demand.
951
# first thing, if the content is something we don't need to store, find
953
digest = sha_string(line_bytes)
954
if nostore_sha == digest:
955
raise errors.ExistingContent
302
return '%s(%s)' % (self.__class__.__name__,
303
self.transport.abspath(self.filename))
305
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
306
"""See VersionedFile._add_delta()."""
307
self._check_add(version_id, []) # should we check the lines ?
308
self._check_versions_present(parents)
957
309
present_parents = []
958
if parent_texts is None:
960
# Do a single query to ascertain parent presence; we only compress
961
# against parents in the same kvf.
962
present_parent_map = self._index.get_parent_map(parents)
963
312
for parent in parents:
964
if parent in present_parent_map:
313
if not self.has_version(parent):
314
ghosts.append(parent)
965
316
present_parents.append(parent)
967
# Currently we can only compress against the left most present parent.
968
if (len(present_parents) == 0 or
969
present_parents[0] != parents[0]):
318
if delta_parent is None:
319
# reconstitute as full text.
320
assert len(delta) == 1 or len(delta) == 0
322
assert delta[0][0] == 0
323
assert delta[0][1] == 0, delta[0][1]
324
return super(KnitVersionedFile, self)._add_delta(version_id,
335
options.append('no-eol')
337
if delta_parent is not None:
338
# determine the current delta chain length.
972
339
# To speed the extract of texts the delta chain is limited
973
340
# to a fixed number of deltas. This should minimize both
974
341
# I/O and the time spend applying deltas.
975
delta = self._check_should_delta(present_parents[0])
977
text_length = len(line_bytes)
980
# Note: line_bytes is not modified to add a newline, that is tracked
981
# via the no_eol flag. 'lines' *is* modified, because that is the
982
# general values needed by the Content code.
983
if line_bytes and line_bytes[-1] != '\n':
984
options.append('no-eol')
986
# Copy the existing list, or create a new one
988
lines = osutils.split_lines(line_bytes)
991
# Replace the last line with one that ends in a final newline
992
lines[-1] = lines[-1] + '\n'
994
lines = osutils.split_lines(line_bytes)
996
for element in key[:-1]:
997
if type(element) is not str:
998
raise TypeError("key contains non-strings: %r" % (key,))
1000
key = key[:-1] + ('sha1:' + digest,)
1001
elif type(key[-1]) is not str:
1002
raise TypeError("key contains non-strings: %r" % (key,))
1003
# Knit hunks are still last-element only
1004
version_id = key[-1]
1005
content = self._factory.make(lines, version_id)
1007
# Hint to the content object that its text() call should strip the
1009
content._should_strip_eol = True
1010
if delta or (self._factory.annotated and len(present_parents) > 0):
1011
# Merge annotations from parent texts if needed.
1012
delta_hunks = self._merge_annotations(content, present_parents,
1013
parent_texts, delta, self._factory.annotated,
1014
left_matching_blocks)
1017
options.append('line-delta')
1018
store_lines = self._factory.lower_line_delta(delta_hunks)
1019
size, bytes = self._record_to_data(key, digest,
1022
options.append('fulltext')
1023
# isinstance is slower and we have no hierarchy.
1024
if self._factory.__class__ is KnitPlainFactory:
1025
# Use the already joined bytes saving iteration time in
1027
dense_lines = [line_bytes]
1029
dense_lines.append('\n')
1030
size, bytes = self._record_to_data(key, digest,
1033
# get mixed annotation + content and feed it into the
1035
store_lines = self._factory.lower_fulltext(content)
1036
size, bytes = self._record_to_data(key, digest,
1039
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
1040
self._index.add_records(
1041
((key, options, access_memo, parents),),
1042
random_id=random_id)
1043
return digest, text_length, content
1045
def annotate(self, key):
1046
"""See VersionedFiles.annotate."""
1047
return self._factory.annotate(self, key)
1049
def get_annotator(self):
1050
return _KnitAnnotator(self)
1052
def check(self, progress_bar=None, keys=None):
1053
"""See VersionedFiles.check()."""
1055
return self._logical_check()
1057
# At the moment, check does not extra work over get_record_stream
1058
return self.get_record_stream(keys, 'unordered', True)
1060
def _logical_check(self):
1061
# This doesn't actually test extraction of everything, but that will
1062
# impact 'bzr check' substantially, and needs to be integrated with
1063
# care. However, it does check for the obvious problem of a delta with
1065
keys = self._index.keys()
1066
parent_map = self.get_parent_map(keys)
1068
if self._index.get_method(key) != 'fulltext':
1069
compression_parent = parent_map[key][0]
1070
if compression_parent not in parent_map:
1071
raise errors.KnitCorrupt(self,
1072
"Missing basis parent %s for %s" % (
1073
compression_parent, key))
1074
for fallback_vfs in self._immediate_fallback_vfs:
1075
fallback_vfs.check()
1077
def _check_add(self, key, lines, random_id, check_content):
1078
"""check that version_id and lines are safe to add."""
1079
version_id = key[-1]
1080
if version_id is not None:
1081
if contains_whitespace(version_id):
1082
raise InvalidRevisionId(version_id, self)
1083
self.check_not_reserved_id(version_id)
1084
# TODO: If random_id==False and the key is already present, we should
1085
# probably check that the existing content is identical to what is
1086
# being inserted, and otherwise raise an exception. This would make
1087
# the bundle code simpler.
1089
self._check_lines_not_unicode(lines)
1090
self._check_lines_are_lines(lines)
1092
def _check_header(self, key, line):
1093
rec = self._split_header(line)
1094
self._check_header_version(rec, key[-1])
1097
def _check_header_version(self, rec, version_id):
1098
"""Checks the header version on original format knit records.
1100
These have the last component of the key embedded in the record.
1102
if rec[1] != version_id:
1103
raise KnitCorrupt(self,
1104
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1106
def _check_should_delta(self, parent):
1107
"""Iterate back through the parent listing, looking for a fulltext.
1109
This is used when we want to decide whether to add a delta or a new
1110
fulltext. It searches for _max_delta_chain parents. When it finds a
1111
fulltext parent, it sees if the total size of the deltas leading up to
1112
it is large enough to indicate that we want a new full text anyway.
1114
Return True if we should create a new delta, False if we should use a
1118
fulltext_size = None
1119
for count in xrange(self._max_delta_chain):
1121
# Note that this only looks in the index of this particular
1122
# KnitVersionedFiles, not in the fallbacks. This ensures that
1123
# we won't store a delta spanning physical repository
1125
build_details = self._index.get_build_details([parent])
1126
parent_details = build_details[parent]
1127
except (RevisionNotPresent, KeyError), e:
1128
# Some basis is not locally present: always fulltext
1130
index_memo, compression_parent, _, _ = parent_details
1131
_, _, size = index_memo
1132
if compression_parent is None:
1133
fulltext_size = size
1136
# We don't explicitly check for presence because this is in an
1137
# inner loop, and if it's missing it'll fail anyhow.
1138
parent = compression_parent
1140
# We couldn't find a fulltext, so we must create a new one
343
delta_parents = [delta_parent]
345
parent = delta_parents[0]
346
method = self._index.get_method(parent)
347
if method == 'fulltext':
349
delta_parents = self._index.get_parents(parent)
351
if method == 'line-delta':
352
# did not find a fulltext in the delta limit.
353
# just do a normal insertion.
354
return super(KnitVersionedFile, self)._add_delta(version_id,
361
options.append('line-delta')
362
store_lines = self.factory.lower_line_delta(delta)
364
where, size = self._data.add_record(version_id, digest, store_lines)
365
self._index.add_version(version_id, options, where, size, parents)
367
def _add_raw_records(self, records, data):
368
"""Add all the records 'records' with data pre-joined in 'data'.
370
:param records: A list of tuples(version_id, options, parents, size).
371
:param data: The data for the records. When it is written, the records
372
are adjusted to have pos pointing into data by the sum of
373
the preceding records sizes.
376
pos = self._data.add_raw_record(data)
378
for (version_id, options, parents, size) in records:
379
index_entries.append((version_id, options, pos, size, parents))
381
self._index.add_versions(index_entries)
383
def clear_cache(self):
384
"""Clear the data cache only."""
385
self._data.clear_cache()
387
def copy_to(self, name, transport):
388
"""See VersionedFile.copy_to()."""
389
# copy the current index to a temp index to avoid racing with local
391
transport.put(name + INDEX_SUFFIX + '.tmp', self.transport.get(self._index._filename),)
393
f = self._data._open_file()
395
transport.put(name + DATA_SUFFIX, f)
398
# move the copied index into place
399
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
401
def create_empty(self, name, transport, mode=None):
402
return KnitVersionedFile(name, transport, factory=self.factory, delta=self.delta, create=True)
404
def _fix_parents(self, version, new_parents):
405
"""Fix the parents list for version.
407
This is done by appending a new version to the index
408
with identical data except for the parents list.
409
the parents list must be a superset of the current
412
current_values = self._index._cache[version]
413
assert set(current_values[4]).difference(set(new_parents)) == set()
414
self._index.add_version(version,
420
def get_delta(self, version_id):
421
"""Get a delta for constructing version from some other version."""
422
if not self.has_version(version_id):
423
raise RevisionNotPresent(version_id, self.filename)
425
parents = self.get_parents(version_id)
430
data_pos, data_size = self._index.get_position(version_id)
431
data, sha1 = self._data.read_records(((version_id, data_pos, data_size),))[version_id]
432
version_idx = self._index.lookup(version_id)
433
noeol = 'no-eol' in self._index.get_options(version_id)
434
if 'fulltext' == self._index.get_method(version_id):
435
new_content = self.factory.parse_fulltext(data, version_idx)
436
if parent is not None:
437
reference_content = self._get_content(parent)
438
old_texts = reference_content.text()
441
new_texts = new_content.text()
442
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
443
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
445
delta = self.factory.parse_line_delta(data, version_idx)
446
return parent, sha1, noeol, delta
448
def get_graph_with_ghosts(self):
449
"""See VersionedFile.get_graph_with_ghosts()."""
450
graph_items = self._index.get_graph()
451
return dict(graph_items)
453
def get_sha1(self, version_id):
454
"""See VersionedFile.get_sha1()."""
455
record_map = self._get_record_map([version_id])
456
method, content, digest, next = record_map[version_id]
461
"""See VersionedFile.get_suffixes()."""
462
return [DATA_SUFFIX, INDEX_SUFFIX]
464
def has_ghost(self, version_id):
465
"""True if there is a ghost reference in the file to version_id."""
467
if self.has_version(version_id):
1142
# Simple heuristic - if the total I/O wold be greater as a delta than
1143
# the originally installed fulltext, we create a new fulltext.
1144
return fulltext_size > delta_size
1146
def _build_details_to_components(self, build_details):
1147
"""Convert a build_details tuple to a position tuple."""
1148
# record_details, access_memo, compression_parent
1149
return build_details[3], build_details[0], build_details[1]
1151
def _get_components_positions(self, keys, allow_missing=False):
1152
"""Produce a map of position data for the components of keys.
469
# optimisable if needed by memoising the _ghosts set.
470
items = self._index.get_graph()
471
for node, parents in items:
472
for parent in parents:
473
if parent not in self._index._cache:
474
if parent == version_id:
479
"""See VersionedFile.versions."""
480
return self._index.get_versions()
482
def has_version(self, version_id):
483
"""See VersionedFile.has_version."""
484
return self._index.has_version(version_id)
486
__contains__ = has_version
488
def _merge_annotations(self, content, parents, parent_texts={},
489
delta=None, annotated=None):
490
"""Merge annotations for content. This is done by comparing
491
the annotations based on changed to the text.
495
for parent_id in parents:
496
merge_content = self._get_content(parent_id, parent_texts)
497
seq = KnitSequenceMatcher(None, merge_content.text(), content.text())
498
if delta_seq is None:
499
# setup a delta seq to reuse.
501
for i, j, n in seq.get_matching_blocks():
504
# this appears to copy (origin, text) pairs across to the new
505
# content for any line that matches the last-checked parent.
506
# FIXME: save the sequence control data for delta compression
507
# against the most relevant parent rather than rediffing.
508
content._lines[j:j+n] = merge_content._lines[i:i+n]
511
reference_content = self._get_content(parents[0], parent_texts)
512
new_texts = content.text()
513
old_texts = reference_content.text()
514
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
515
return self._make_line_delta(delta_seq, content)
517
def _make_line_delta(self, delta_seq, new_content):
518
"""Generate a line delta from delta_seq and new_content."""
520
for op in delta_seq.get_opcodes():
523
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
526
def _get_components_positions(self, version_ids):
527
"""Produce a map of position data for the components of versions.
1154
529
This data is intended to be used for retrieving the knit records.
1156
A dict of key to (record_details, index_memo, next, parents) is
531
A dict of version_id to (method, data_pos, data_size, next) is
1159
* method is the way referenced data should be applied.
1160
* index_memo is the handle to pass to the data access to actually get
1162
* next is the build-parent of the version, or None for fulltexts.
1163
* parents is the version_ids of the parents of this version
1165
:param allow_missing: If True do not raise an error on a missing
1166
component, just ignore it.
533
method is the way referenced data should be applied.
534
data_pos is the position of the data in the knit.
535
data_size is the size of the data in the knit.
536
next is the build-parent of the version, or None for fulltexts.
1168
538
component_data = {}
1169
pending_components = keys
1170
while pending_components:
1171
build_details = self._index.get_build_details(pending_components)
1172
current_components = set(pending_components)
1173
pending_components = set()
1174
for key, details in build_details.iteritems():
1175
(index_memo, compression_parent, parents,
1176
record_details) = details
1177
method = record_details[0]
1178
if compression_parent is not None:
1179
pending_components.add(compression_parent)
1180
component_data[key] = self._build_details_to_components(details)
1181
missing = current_components.difference(build_details)
1182
if missing and not allow_missing:
1183
raise errors.RevisionNotPresent(missing.pop(), self)
539
for version_id in version_ids:
542
while cursor is not None and cursor not in component_data:
543
method = self._index.get_method(cursor)
544
if method == 'fulltext':
547
next = self.get_parents(cursor)[0]
548
data_pos, data_size = self._index.get_position(cursor)
549
component_data[cursor] = (method, data_pos, data_size, next)
1184
551
return component_data
1186
def _get_content(self, key, parent_texts={}):
553
def _get_content(self, version_id, parent_texts={}):
1187
554
"""Returns a content object that makes up the specified
1189
cached_version = parent_texts.get(key, None)
556
if not self.has_version(version_id):
557
raise RevisionNotPresent(version_id, self.filename)
559
cached_version = parent_texts.get(version_id, None)
1190
560
if cached_version is not None:
1191
# Ensure the cache dict is valid.
1192
if not self.get_parent_map([key]):
1193
raise RevisionNotPresent(key, self)
1194
561
return cached_version
1195
generator = _VFContentMapGenerator(self, [key])
1196
return generator._get_content(key)
1198
def get_parent_map(self, keys):
1199
"""Get a map of the graph parents of keys.
1201
:param keys: The keys to look up parents for.
1202
:return: A mapping from keys to parents. Absent keys are absent from
1205
return self._get_parent_map_with_sources(keys)[0]
1207
def _get_parent_map_with_sources(self, keys):
1208
"""Get a map of the parents of keys.
1210
:param keys: The keys to look up parents for.
1211
:return: A tuple. The first element is a mapping from keys to parents.
1212
Absent keys are absent from the mapping. The second element is a
1213
list with the locations each key was found in. The first element
1214
is the in-this-knit parents, the second the first fallback source,
1218
sources = [self._index] + self._immediate_fallback_vfs
1221
for source in sources:
1224
new_result = source.get_parent_map(missing)
1225
source_results.append(new_result)
1226
result.update(new_result)
1227
missing.difference_update(set(new_result))
1228
return result, source_results
1230
def _get_record_map(self, keys, allow_missing=False):
563
text_map, contents_map = self._get_content_maps([version_id])
564
return contents_map[version_id]
566
def _check_versions_present(self, version_ids):
567
"""Check that all specified versions are present."""
568
version_ids = set(version_ids)
569
for r in list(version_ids):
570
if self._index.has_version(r):
571
version_ids.remove(r)
573
raise RevisionNotPresent(list(version_ids)[0], self.filename)
575
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
576
"""See VersionedFile.add_lines_with_ghosts()."""
577
self._check_add(version_id, lines)
578
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
580
def _add_lines(self, version_id, parents, lines, parent_texts):
581
"""See VersionedFile.add_lines."""
582
self._check_add(version_id, lines)
583
self._check_versions_present(parents)
584
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
586
def _check_add(self, version_id, lines):
587
"""check that version_id and lines are safe to add."""
588
assert self.writable, "knit is not opened for write"
589
### FIXME escape. RBC 20060228
590
if contains_whitespace(version_id):
591
raise InvalidRevisionId(version_id, self.filename)
592
if self.has_version(version_id):
593
raise RevisionAlreadyPresent(version_id, self.filename)
594
self._check_lines_not_unicode(lines)
595
self._check_lines_are_lines(lines)
597
def _add(self, version_id, lines, parents, delta, parent_texts):
598
"""Add a set of lines on top of version specified by parents.
600
If delta is true, compress the text as a line-delta against
603
Any versions not present will be converted into ghosts.
605
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
606
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
607
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
608
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
609
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
610
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
611
# +1383 0 8.0370 8.0370 +<len>
612
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
613
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
614
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
615
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
619
if parent_texts is None:
621
for parent in parents:
622
if not self.has_version(parent):
623
ghosts.append(parent)
625
present_parents.append(parent)
627
if delta and not len(present_parents):
630
digest = sha_strings(lines)
633
if lines[-1][-1] != '\n':
634
options.append('no-eol')
635
lines[-1] = lines[-1] + '\n'
637
if len(present_parents) and delta:
638
# To speed the extract of texts the delta chain is limited
639
# to a fixed number of deltas. This should minimize both
640
# I/O and the time spend applying deltas.
642
delta_parents = present_parents
644
parent = delta_parents[0]
645
method = self._index.get_method(parent)
646
if method == 'fulltext':
648
delta_parents = self._index.get_parents(parent)
650
if method == 'line-delta':
653
lines = self.factory.make(lines, version_id)
654
if delta or (self.factory.annotated and len(present_parents) > 0):
655
# Merge annotations from parent texts if so is needed.
656
delta_hunks = self._merge_annotations(lines, present_parents, parent_texts,
657
delta, self.factory.annotated)
660
options.append('line-delta')
661
store_lines = self.factory.lower_line_delta(delta_hunks)
663
options.append('fulltext')
664
store_lines = self.factory.lower_fulltext(lines)
666
where, size = self._data.add_record(version_id, digest, store_lines)
667
self._index.add_version(version_id, options, where, size, parents)
670
def check(self, progress_bar=None):
671
"""See VersionedFile.check()."""
673
def _clone_text(self, new_version_id, old_version_id, parents):
674
"""See VersionedFile.clone_text()."""
675
# FIXME RBC 20060228 make fast by only inserting an index with null
677
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
679
def get_lines(self, version_id):
680
"""See VersionedFile.get_lines()."""
681
return self.get_line_list([version_id])[0]
683
def _get_record_map(self, version_ids):
1231
684
"""Produce a dictionary of knit records.
1233
:return: {key:(record, record_details, digest, next)}
1235
* record: data returned from read_records (a KnitContentobject)
1236
* record_details: opaque information to pass to parse_record
1237
* digest: SHA1 digest of the full text after all steps are done
1238
* next: build-parent of the version, i.e. the leftmost ancestor.
1239
Will be None if the record is not a delta.
1241
:param keys: The keys to build a map for
1242
:param allow_missing: If some records are missing, rather than
1243
error, just return the data that could be generated.
1245
raw_map = self._get_record_map_unparsed(keys,
1246
allow_missing=allow_missing)
1247
return self._raw_map_to_record_map(raw_map)
1249
def _raw_map_to_record_map(self, raw_map):
1250
"""Parse the contents of _get_record_map_unparsed.
1252
:return: see _get_record_map.
1256
data, record_details, next = raw_map[key]
1257
content, digest = self._parse_record(key[-1], data)
1258
result[key] = content, record_details, digest, next
1261
def _get_record_map_unparsed(self, keys, allow_missing=False):
1262
"""Get the raw data for reconstructing keys without parsing it.
1264
:return: A dict suitable for parsing via _raw_map_to_record_map.
1265
key-> raw_bytes, (method, noeol), compression_parent
1267
# This retries the whole request if anything fails. Potentially we
1268
# could be a bit more selective. We could track the keys whose records
1269
# we have successfully found, and then only request the new records
1270
# from there. However, _get_components_positions grabs the whole build
1271
# chain, which means we'll likely try to grab the same records again
1272
# anyway. Also, can the build chains change as part of a pack
1273
# operation? We wouldn't want to end up with a broken chain.
1276
position_map = self._get_components_positions(keys,
1277
allow_missing=allow_missing)
1278
# key = component_id, r = record_details, i_m = index_memo,
1280
records = [(key, i_m) for key, (r, i_m, n)
1281
in position_map.iteritems()]
1282
# Sort by the index memo, so that we request records from the
1283
# same pack file together, and in forward-sorted order
1284
records.sort(key=operator.itemgetter(1))
1286
for key, data in self._read_records_iter_unchecked(records):
1287
(record_details, index_memo, next) = position_map[key]
1288
raw_record_map[key] = data, record_details, next
1289
return raw_record_map
1290
except errors.RetryWithNewPacks, e:
1291
self._access.reload_or_raise(e)
1294
def _split_by_prefix(cls, keys):
1295
"""For the given keys, split them up based on their prefix.
1297
To keep memory pressure somewhat under control, split the
1298
requests back into per-file-id requests, otherwise "bzr co"
1299
extracts the full tree into memory before writing it to disk.
1300
This should be revisited if _get_content_maps() can ever cross
1303
The keys for a given file_id are kept in the same relative order.
1304
Ordering between file_ids is not, though prefix_order will return the
1305
order that the key was first seen.
1307
:param keys: An iterable of key tuples
1308
:return: (split_map, prefix_order)
1309
split_map A dictionary mapping prefix => keys
1310
prefix_order The order that we saw the various prefixes
1312
split_by_prefix = {}
1320
if prefix in split_by_prefix:
1321
split_by_prefix[prefix].append(key)
1323
split_by_prefix[prefix] = [key]
1324
prefix_order.append(prefix)
1325
return split_by_prefix, prefix_order
1327
def _group_keys_for_io(self, keys, non_local_keys, positions,
1328
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1329
"""For the given keys, group them into 'best-sized' requests.
1331
The idea is to avoid making 1 request per file, but to never try to
1332
unpack an entire 1.5GB source tree in a single pass. Also when
1333
possible, we should try to group requests to the same pack file
1336
:return: list of (keys, non_local) tuples that indicate what keys
1337
should be fetched next.
1339
# TODO: Ideally we would group on 2 factors. We want to extract texts
1340
# from the same pack file together, and we want to extract all
1341
# the texts for a given build-chain together. Ultimately it
1342
# probably needs a better global view.
1343
total_keys = len(keys)
1344
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1345
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1347
cur_non_local = set()
1351
for prefix in prefix_order:
1352
keys = prefix_split_keys[prefix]
1353
non_local = prefix_split_non_local_keys.get(prefix, [])
1355
this_size = self._index._get_total_build_size(keys, positions)
1356
cur_size += this_size
1357
cur_keys.extend(keys)
1358
cur_non_local.update(non_local)
1359
if cur_size > _min_buffer_size:
1360
result.append((cur_keys, cur_non_local))
1361
sizes.append(cur_size)
1363
cur_non_local = set()
1366
result.append((cur_keys, cur_non_local))
1367
sizes.append(cur_size)
1370
def get_record_stream(self, keys, ordering, include_delta_closure):
1371
"""Get a stream of records for keys.
1373
:param keys: The keys to include.
1374
:param ordering: Either 'unordered' or 'topological'. A topologically
1375
sorted stream has compression parents strictly before their
1377
:param include_delta_closure: If True then the closure across any
1378
compression parents will be included (in the opaque data).
1379
:return: An iterator of ContentFactory objects, each of which is only
1380
valid until the iterator is advanced.
1382
# keys might be a generator
1386
if not self._index.has_graph:
1387
# Cannot sort when no graph has been stored.
1388
ordering = 'unordered'
1390
remaining_keys = keys
1393
keys = set(remaining_keys)
1394
for content_factory in self._get_remaining_record_stream(keys,
1395
ordering, include_delta_closure):
1396
remaining_keys.discard(content_factory.key)
1397
yield content_factory
1399
except errors.RetryWithNewPacks, e:
1400
self._access.reload_or_raise(e)
1402
def _get_remaining_record_stream(self, keys, ordering,
1403
include_delta_closure):
1404
"""This function is the 'retry' portion for get_record_stream."""
1405
if include_delta_closure:
1406
positions = self._get_components_positions(keys, allow_missing=True)
1408
build_details = self._index.get_build_details(keys)
1410
# (record_details, access_memo, compression_parent_key)
1411
positions = dict((key, self._build_details_to_components(details))
1412
for key, details in build_details.iteritems())
1413
absent_keys = keys.difference(set(positions))
1414
# There may be more absent keys : if we're missing the basis component
1415
# and are trying to include the delta closure.
1416
# XXX: We should not ever need to examine remote sources because we do
1417
# not permit deltas across versioned files boundaries.
1418
if include_delta_closure:
1419
needed_from_fallback = set()
1420
# Build up reconstructable_keys dict. key:True in this dict means
1421
# the key can be reconstructed.
1422
reconstructable_keys = {}
1426
chain = [key, positions[key][2]]
1428
needed_from_fallback.add(key)
1431
while chain[-1] is not None:
1432
if chain[-1] in reconstructable_keys:
1433
result = reconstructable_keys[chain[-1]]
1437
chain.append(positions[chain[-1]][2])
1439
# missing basis component
1440
needed_from_fallback.add(chain[-1])
1443
for chain_key in chain[:-1]:
1444
reconstructable_keys[chain_key] = result
1446
needed_from_fallback.add(key)
1447
# Double index lookups here : need a unified api ?
1448
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1449
if ordering in ('topological', 'groupcompress'):
1450
if ordering == 'topological':
1451
# Global topological sort
1452
present_keys = tsort.topo_sort(global_map)
1454
present_keys = sort_groupcompress(global_map)
1455
# Now group by source:
1457
current_source = None
1458
for key in present_keys:
1459
for parent_map in parent_maps:
1460
if key in parent_map:
1461
key_source = parent_map
1463
if current_source is not key_source:
1464
source_keys.append((key_source, []))
1465
current_source = key_source
1466
source_keys[-1][1].append(key)
1468
if ordering != 'unordered':
1469
raise AssertionError('valid values for ordering are:'
1470
' "unordered", "groupcompress" or "topological" not: %r'
1472
# Just group by source; remote sources first.
1475
for parent_map in reversed(parent_maps):
1476
source_keys.append((parent_map, []))
1477
for key in parent_map:
1478
present_keys.append(key)
1479
source_keys[-1][1].append(key)
1480
# We have been requested to return these records in an order that
1481
# suits us. So we ask the index to give us an optimally sorted
1483
for source, sub_keys in source_keys:
1484
if source is parent_maps[0]:
1485
# Only sort the keys for this VF
1486
self._index._sort_keys_by_io(sub_keys, positions)
1487
absent_keys = keys - set(global_map)
1488
for key in absent_keys:
1489
yield AbsentContentFactory(key)
1490
# restrict our view to the keys we can answer.
1491
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1492
# XXX: At that point we need to consider the impact of double reads by
1493
# utilising components multiple times.
1494
if include_delta_closure:
1495
# XXX: get_content_maps performs its own index queries; allow state
1497
non_local_keys = needed_from_fallback - absent_keys
1498
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1501
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1504
for record in generator.get_record_stream():
1507
for source, keys in source_keys:
1508
if source is parent_maps[0]:
1509
# this KnitVersionedFiles
1510
records = [(key, positions[key][1]) for key in keys]
1511
for key, raw_data in self._read_records_iter_unchecked(records):
1512
(record_details, index_memo, _) = positions[key]
1513
yield KnitContentFactory(key, global_map[key],
1514
record_details, None, raw_data, self._factory.annotated, None)
1516
vf = self._immediate_fallback_vfs[parent_maps.index(source) - 1]
1517
for record in vf.get_record_stream(keys, ordering,
1518
include_delta_closure):
1521
def get_sha1s(self, keys):
1522
"""See VersionedFiles.get_sha1s()."""
1524
record_map = self._get_record_map(missing, allow_missing=True)
1526
for key, details in record_map.iteritems():
1527
if key not in missing:
1529
# record entry 2 is the 'digest'.
1530
result[key] = details[2]
1531
missing.difference_update(set(result))
1532
for source in self._immediate_fallback_vfs:
1535
new_result = source.get_sha1s(missing)
1536
result.update(new_result)
1537
missing.difference_update(set(new_result))
1540
def insert_record_stream(self, stream):
1541
"""Insert a record stream into this container.
1543
:param stream: A stream of records to insert.
1545
:seealso VersionedFiles.get_record_stream:
1547
def get_adapter(adapter_key):
1549
return adapters[adapter_key]
1551
adapter_factory = adapter_registry.get(adapter_key)
1552
adapter = adapter_factory(self)
1553
adapters[adapter_key] = adapter
1556
if self._factory.annotated:
1557
# self is annotated, we need annotated knits to use directly.
1558
annotated = "annotated-"
1561
# self is not annotated, but we can strip annotations cheaply.
1563
convertibles = set(["knit-annotated-ft-gz"])
1564
if self._max_delta_chain:
1565
delta_types.add("knit-annotated-delta-gz")
1566
convertibles.add("knit-annotated-delta-gz")
1567
# The set of types we can cheaply adapt without needing basis texts.
1568
native_types = set()
1569
if self._max_delta_chain:
1570
native_types.add("knit-%sdelta-gz" % annotated)
1571
delta_types.add("knit-%sdelta-gz" % annotated)
1572
native_types.add("knit-%sft-gz" % annotated)
1573
knit_types = native_types.union(convertibles)
1575
# Buffer all index entries that we can't add immediately because their
1576
# basis parent is missing. We don't buffer all because generating
1577
# annotations may require access to some of the new records. However we
1578
# can't generate annotations from new deltas until their basis parent
1579
# is present anyway, so we get away with not needing an index that
1580
# includes the new keys.
1582
# See <http://launchpad.net/bugs/300177> about ordering of compression
1583
# parents in the records - to be conservative, we insist that all
1584
# parents must be present to avoid expanding to a fulltext.
1586
# key = basis_parent, value = index entry to add
1587
buffered_index_entries = {}
1588
for record in stream:
1589
kind = record.storage_kind
1590
if kind.startswith('knit-') and kind.endswith('-gz'):
1591
# Check that the ID in the header of the raw knit bytes matches
1592
# the record metadata.
1593
raw_data = record._raw_record
1594
df, rec = self._parse_record_header(record.key, raw_data)
1597
parents = record.parents
1598
if record.storage_kind in delta_types:
1599
# TODO: eventually the record itself should track
1600
# compression_parent
1601
compression_parent = parents[0]
1603
compression_parent = None
1604
# Raise an error when a record is missing.
1605
if record.storage_kind == 'absent':
1606
raise RevisionNotPresent([record.key], self)
1607
elif ((record.storage_kind in knit_types)
1608
and (compression_parent is None
1609
or not self._immediate_fallback_vfs
1610
or self._index.has_key(compression_parent)
1611
or not self.has_key(compression_parent))):
1612
# we can insert the knit record literally if either it has no
1613
# compression parent OR we already have its basis in this kvf
1614
# OR the basis is not present even in the fallbacks. In the
1615
# last case it will either turn up later in the stream and all
1616
# will be well, or it won't turn up at all and we'll raise an
1619
# TODO: self.has_key is somewhat redundant with
1620
# self._index.has_key; we really want something that directly
1621
# asks if it's only present in the fallbacks. -- mbp 20081119
1622
if record.storage_kind not in native_types:
1624
adapter_key = (record.storage_kind, "knit-delta-gz")
1625
adapter = get_adapter(adapter_key)
1627
adapter_key = (record.storage_kind, "knit-ft-gz")
1628
adapter = get_adapter(adapter_key)
1629
bytes = adapter.get_bytes(record)
1631
# It's a knit record, it has a _raw_record field (even if
1632
# it was reconstituted from a network stream).
1633
bytes = record._raw_record
1634
options = [record._build_details[0]]
1635
if record._build_details[1]:
1636
options.append('no-eol')
1637
# Just blat it across.
1638
# Note: This does end up adding data on duplicate keys. As
1639
# modern repositories use atomic insertions this should not
1640
# lead to excessive growth in the event of interrupted fetches.
1641
# 'knit' repositories may suffer excessive growth, but as a
1642
# deprecated format this is tolerable. It can be fixed if
1643
# needed by in the kndx index support raising on a duplicate
1644
# add with identical parents and options.
1645
access_memo = self._access.add_raw_records(
1646
[(record.key, len(bytes))], bytes)[0]
1647
index_entry = (record.key, options, access_memo, parents)
1648
if 'fulltext' not in options:
1649
# Not a fulltext, so we need to make sure the compression
1650
# parent will also be present.
1651
# Note that pack backed knits don't need to buffer here
1652
# because they buffer all writes to the transaction level,
1653
# but we don't expose that difference at the index level. If
1654
# the query here has sufficient cost to show up in
1655
# profiling we should do that.
1657
# They're required to be physically in this
1658
# KnitVersionedFiles, not in a fallback.
1659
if not self._index.has_key(compression_parent):
1660
pending = buffered_index_entries.setdefault(
1661
compression_parent, [])
1662
pending.append(index_entry)
1665
self._index.add_records([index_entry])
1666
elif record.storage_kind == 'chunked':
1667
self.add_lines(record.key, parents,
1668
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1670
# Not suitable for direct insertion as a
1671
# delta, either because it's not the right format, or this
1672
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1673
# 0) or because it depends on a base only present in the
1675
self._access.flush()
1677
# Try getting a fulltext directly from the record.
1678
bytes = record.get_bytes_as('fulltext')
1679
except errors.UnavailableRepresentation:
1680
adapter_key = record.storage_kind, 'fulltext'
1681
adapter = get_adapter(adapter_key)
1682
bytes = adapter.get_bytes(record)
1683
lines = split_lines(bytes)
1685
self.add_lines(record.key, parents, lines)
1686
except errors.RevisionAlreadyPresent:
1688
# Add any records whose basis parent is now available.
1690
added_keys = [record.key]
1692
key = added_keys.pop(0)
1693
if key in buffered_index_entries:
1694
index_entries = buffered_index_entries[key]
1695
self._index.add_records(index_entries)
1697
[index_entry[0] for index_entry in index_entries])
1698
del buffered_index_entries[key]
1699
if buffered_index_entries:
1700
# There were index entries buffered at the end of the stream,
1701
# So these need to be added (if the index supports holding such
1702
# entries for later insertion)
1704
for key in buffered_index_entries:
1705
index_entries = buffered_index_entries[key]
1706
all_entries.extend(index_entries)
1707
self._index.add_records(
1708
all_entries, missing_compression_parents=True)
1710
def get_missing_compression_parent_keys(self):
1711
"""Return an iterable of keys of missing compression parents.
1713
Check this after calling insert_record_stream to find out if there are
1714
any missing compression parents. If there are, the records that
1715
depend on them are not able to be inserted safely. For atomic
1716
KnitVersionedFiles built on packs, the transaction should be aborted or
1717
suspended - commit will fail at this point. Nonatomic knits will error
1718
earlier because they have no staging area to put pending entries into.
1720
return self._index.get_missing_compression_parents()
1722
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1723
"""Iterate over the lines in the versioned files from keys.
1725
This may return lines from other keys. Each item the returned
1726
iterator yields is a tuple of a line and a text version that that line
1727
is present in (not introduced in).
1729
Ordering of results is in whatever order is most suitable for the
1730
underlying storage format.
1732
If a progress bar is supplied, it may be used to indicate progress.
1733
The caller is responsible for cleaning up progress bars (because this
1737
* Lines are normalised by the underlying store: they will all have \\n
1739
* Lines are returned in arbitrary order.
1740
* If a requested key did not change any lines (or didn't have any
1741
lines), it may not be mentioned at all in the result.
1743
:param pb: Progress bar supplied by caller.
1744
:return: An iterator over (line, key).
1747
pb = ui.ui_factory.nested_progress_bar()
1753
# we don't care about inclusions, the caller cares.
1754
# but we need to setup a list of records to visit.
1755
# we need key, position, length
1757
build_details = self._index.get_build_details(keys)
1758
for key, details in build_details.iteritems():
1760
key_records.append((key, details[0]))
1761
records_iter = enumerate(self._read_records_iter(key_records))
1762
for (key_idx, (key, data, sha_value)) in records_iter:
1763
pb.update('Walking content', key_idx, total)
1764
compression_parent = build_details[key][1]
1765
if compression_parent is None:
1767
line_iterator = self._factory.get_fulltext_content(data)
1770
line_iterator = self._factory.get_linedelta_content(data)
1771
# Now that we are yielding the data for this key, remove it
1774
# XXX: It might be more efficient to yield (key,
1775
# line_iterator) in the future. However for now, this is a
1776
# simpler change to integrate into the rest of the
1777
# codebase. RBC 20071110
1778
for line in line_iterator:
1781
except errors.RetryWithNewPacks, e:
1782
self._access.reload_or_raise(e)
1783
# If there are still keys we've not yet found, we look in the fallback
1784
# vfs, and hope to find them there. Note that if the keys are found
1785
# but had no changes or no content, the fallback may not return
1787
if keys and not self._immediate_fallback_vfs:
1788
# XXX: strictly the second parameter is meant to be the file id
1789
# but it's not easily accessible here.
1790
raise RevisionNotPresent(keys, repr(self))
1791
for source in self._immediate_fallback_vfs:
1795
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1796
source_keys.add(key)
1798
keys.difference_update(source_keys)
1799
pb.update('Walking content', total, total)
1801
def _make_line_delta(self, delta_seq, new_content):
1802
"""Generate a line delta from delta_seq and new_content."""
1804
for op in delta_seq.get_opcodes():
1805
if op[0] == 'equal':
1807
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1810
def _merge_annotations(self, content, parents, parent_texts={},
1811
delta=None, annotated=None,
1812
left_matching_blocks=None):
1813
"""Merge annotations for content and generate deltas.
1815
This is done by comparing the annotations based on changes to the text
1816
and generating a delta on the resulting full texts. If annotations are
1817
not being created then a simple delta is created.
1819
if left_matching_blocks is not None:
1820
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1824
for parent_key in parents:
1825
merge_content = self._get_content(parent_key, parent_texts)
1826
if (parent_key == parents[0] and delta_seq is not None):
1829
seq = patiencediff.PatienceSequenceMatcher(
1830
None, merge_content.text(), content.text())
1831
for i, j, n in seq.get_matching_blocks():
1834
# this copies (origin, text) pairs across to the new
1835
# content for any line that matches the last-checked
1837
content._lines[j:j+n] = merge_content._lines[i:i+n]
1838
# XXX: Robert says the following block is a workaround for a
1839
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1840
if content._lines and content._lines[-1][1][-1] != '\n':
1841
# The copied annotation was from a line without a trailing EOL,
1842
# reinstate one for the content object, to ensure correct
1844
line = content._lines[-1][1] + '\n'
1845
content._lines[-1] = (content._lines[-1][0], line)
1847
if delta_seq is None:
1848
reference_content = self._get_content(parents[0], parent_texts)
1849
new_texts = content.text()
1850
old_texts = reference_content.text()
1851
delta_seq = patiencediff.PatienceSequenceMatcher(
1852
None, old_texts, new_texts)
1853
return self._make_line_delta(delta_seq, content)
1855
def _parse_record(self, version_id, data):
1856
"""Parse an original format knit record.
1858
These have the last element of the key only present in the stored data.
1860
rec, record_contents = self._parse_record_unchecked(data)
1861
self._check_header_version(rec, version_id)
1862
return record_contents, rec[3]
1864
def _parse_record_header(self, key, raw_data):
1865
"""Parse a record header for consistency.
1867
:return: the header and the decompressor stream.
1868
as (stream, header_record)
1870
df = gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1873
rec = self._check_header(key, df.readline())
1874
except Exception, e:
1875
raise KnitCorrupt(self,
1876
"While reading {%s} got %s(%s)"
1877
% (key, e.__class__.__name__, str(e)))
1880
def _parse_record_unchecked(self, data):
1882
# 4168 calls in 2880 217 internal
1883
# 4168 calls to _parse_record_header in 2121
1884
# 4168 calls to readlines in 330
1885
df = gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1887
record_contents = df.readlines()
1888
except Exception, e:
1889
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1890
(data, e.__class__.__name__, str(e)))
1891
header = record_contents.pop(0)
1892
rec = self._split_header(header)
1893
last_line = record_contents.pop()
1894
if len(record_contents) != int(rec[2]):
1895
raise KnitCorrupt(self,
1896
'incorrect number of lines %s != %s'
1897
' for version {%s} %s'
1898
% (len(record_contents), int(rec[2]),
1899
rec[1], record_contents))
1900
if last_line != 'end %s\n' % rec[1]:
1901
raise KnitCorrupt(self,
1902
'unexpected version end line %r, wanted %r'
1903
% (last_line, rec[1]))
1905
return rec, record_contents
1907
def _read_records_iter(self, records):
1908
"""Read text records from data file and yield result.
1910
The result will be returned in whatever is the fastest to read.
1911
Not by the order requested. Also, multiple requests for the same
1912
record will only yield 1 response.
1914
:param records: A list of (key, access_memo) entries
1915
:return: Yields (key, contents, digest) in the order
1916
read, not the order requested
1921
# XXX: This smells wrong, IO may not be getting ordered right.
1922
needed_records = sorted(set(records), key=operator.itemgetter(1))
1923
if not needed_records:
1926
# The transport optimizes the fetching as well
1927
# (ie, reads continuous ranges.)
1928
raw_data = self._access.get_raw_records(
1929
[index_memo for key, index_memo in needed_records])
1931
for (key, index_memo), data in \
1932
izip(iter(needed_records), raw_data):
1933
content, digest = self._parse_record(key[-1], data)
1934
yield key, content, digest
1936
def _read_records_iter_raw(self, records):
1937
"""Read text records from data file and yield raw data.
1939
This unpacks enough of the text record to validate the id is
1940
as expected but thats all.
1942
Each item the iterator yields is (key, bytes,
1943
expected_sha1_of_full_text).
1945
for key, data in self._read_records_iter_unchecked(records):
1946
# validate the header (note that we can only use the suffix in
1947
# current knit records).
1948
df, rec = self._parse_record_header(key, data)
1950
yield key, data, rec[3]
1952
def _read_records_iter_unchecked(self, records):
1953
"""Read text records from data file and yield raw data.
1955
No validation is done.
1957
Yields tuples of (key, data).
1959
# setup an iterator of the external records:
1960
# uses readv so nice and fast we hope.
1962
# grab the disk data needed.
1963
needed_offsets = [index_memo for key, index_memo
1965
raw_records = self._access.get_raw_records(needed_offsets)
1967
for key, index_memo in records:
1968
data = raw_records.next()
1971
def _record_to_data(self, key, digest, lines, dense_lines=None):
1972
"""Convert key, digest, lines into a raw data block.
1974
:param key: The key of the record. Currently keys are always serialised
1975
using just the trailing component.
1976
:param dense_lines: The bytes of lines but in a denser form. For
1977
instance, if lines is a list of 1000 bytestrings each ending in
1978
\\n, dense_lines may be a list with one line in it, containing all
1979
the 1000's lines and their \\n's. Using dense_lines if it is
1980
already known is a win because the string join to create bytes in
1981
this function spends less time resizing the final string.
1982
:return: (len, a StringIO instance with the raw data ready to read.)
1984
chunks = ["version %s %d %s\n" % (key[-1], len(lines), digest)]
1985
chunks.extend(dense_lines or lines)
1986
chunks.append("end %s\n" % key[-1])
1987
for chunk in chunks:
1988
if type(chunk) is not str:
1989
raise AssertionError(
1990
'data must be plain bytes was %s' % type(chunk))
1991
if lines and lines[-1][-1] != '\n':
1992
raise ValueError('corrupt lines value %r' % lines)
1993
compressed_bytes = tuned_gzip.chunks_to_gzip(chunks)
1994
return len(compressed_bytes), compressed_bytes
1996
def _split_header(self, line):
1999
raise KnitCorrupt(self,
2000
'unexpected number of elements in record header')
2004
"""See VersionedFiles.keys."""
2005
if 'evil' in debug.debug_flags:
2006
trace.mutter_callsite(2, "keys scales with size of history")
2007
sources = [self._index] + self._immediate_fallback_vfs
2009
for source in sources:
2010
result.update(source.keys())
2014
class _ContentMapGenerator(object):
2015
"""Generate texts or expose raw deltas for a set of texts."""
2017
def __init__(self, ordering='unordered'):
2018
self._ordering = ordering
2020
def _get_content(self, key):
2021
"""Get the content object for key."""
2022
# Note that _get_content is only called when the _ContentMapGenerator
2023
# has been constructed with just one key requested for reconstruction.
2024
if key in self.nonlocal_keys:
2025
record = self.get_record_stream().next()
2026
# Create a content object on the fly
2027
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
2028
return PlainKnitContent(lines, record.key)
2030
# local keys we can ask for directly
2031
return self._get_one_work(key)
2033
def get_record_stream(self):
2034
"""Get a record stream for the keys requested during __init__."""
2035
for record in self._work():
2039
"""Produce maps of text and KnitContents as dicts.
686
The keys are version_ids, the values are tuples of (method, content,
688
method is the way the content should be applied.
689
content is a KnitContent object.
690
digest is the SHA1 digest of this version id after all steps are done
691
next is the build-parent of the version, i.e. the leftmost ancestor.
692
If the method is fulltext, next will be None.
694
position_map = self._get_components_positions(version_ids)
695
# c = component_id, m = method, p = position, s = size, n = next
696
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
698
for component_id, content, digest in\
699
self._data.read_records_iter(records):
700
method, position, size, next = position_map[component_id]
701
record_map[component_id] = method, content, digest, next
705
def get_text(self, version_id):
706
"""See VersionedFile.get_text"""
707
return self.get_texts([version_id])[0]
709
def get_texts(self, version_ids):
710
return [''.join(l) for l in self.get_line_list(version_ids)]
712
def get_line_list(self, version_ids):
713
"""Return the texts of listed versions as a list of strings."""
714
text_map, content_map = self._get_content_maps(version_ids)
715
return [text_map[v] for v in version_ids]
717
def _get_content_maps(self, version_ids):
718
"""Produce maps of text and KnitContents
2041
720
:return: (text_map, content_map) where text_map contains the texts for
2042
the requested versions and content_map contains the KnitContents.
721
the requested versions and content_map contains the KnitContents.
722
Both dicts take version_ids as their keys.
2044
# NB: By definition we never need to read remote sources unless texts
2045
# are requested from them: we don't delta across stores - and we
2046
# explicitly do not want to to prevent data loss situations.
2047
if self.global_map is None:
2048
self.global_map = self.vf.get_parent_map(self.keys)
2049
nonlocal_keys = self.nonlocal_keys
2051
missing_keys = set(nonlocal_keys)
2052
# Read from remote versioned file instances and provide to our caller.
2053
for source in self.vf._immediate_fallback_vfs:
2054
if not missing_keys:
2056
# Loop over fallback repositories asking them for texts - ignore
2057
# any missing from a particular fallback.
2058
for record in source.get_record_stream(missing_keys,
2059
self._ordering, True):
2060
if record.storage_kind == 'absent':
2061
# Not in thie particular stream, may be in one of the
2062
# other fallback vfs objects.
2064
missing_keys.remove(record.key)
2067
if self._raw_record_map is None:
2068
raise AssertionError('_raw_record_map should have been filled')
2070
for key in self.keys:
2071
if key in self.nonlocal_keys:
2073
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2076
def _get_one_work(self, requested_key):
2077
# Now, if we have calculated everything already, just return the
2079
if requested_key in self._contents_map:
2080
return self._contents_map[requested_key]
2081
# To simplify things, parse everything at once - code that wants one text
2082
# probably wants them all.
2083
# FUTURE: This function could be improved for the 'extract many' case
2084
# by tracking each component and only doing the copy when the number of
2085
# children than need to apply delta's to it is > 1 or it is part of the
2087
multiple_versions = len(self.keys) != 1
2088
if self._record_map is None:
2089
self._record_map = self.vf._raw_map_to_record_map(
2090
self._raw_record_map)
2091
record_map = self._record_map
2092
# raw_record_map is key:
2093
# Have read and parsed records at this point.
2094
for key in self.keys:
2095
if key in self.nonlocal_keys:
724
for version_id in version_ids:
725
if not self.has_version(version_id):
726
raise RevisionNotPresent(version_id, self.filename)
727
record_map = self._get_record_map(version_ids)
732
for version_id in version_ids:
2100
735
while cursor is not None:
2102
record, record_details, digest, next = record_map[cursor]
2104
raise RevisionNotPresent(cursor, self)
2105
components.append((cursor, record, record_details, digest))
736
method, data, digest, next = record_map[cursor]
737
components.append((cursor, method, data, digest))
738
if cursor in content_map:
2107
if cursor in self._contents_map:
2108
# no need to plan further back
2109
components.append((cursor, None, None, None))
2113
for (component_id, record, record_details,
2114
digest) in reversed(components):
2115
if component_id in self._contents_map:
2116
content = self._contents_map[component_id]
743
for component_id, method, data, digest in reversed(components):
744
if component_id in content_map:
745
content = content_map[component_id]
2118
content, delta = self._factory.parse_record(key[-1],
2119
record, record_details, content,
2120
copy_base_content=multiple_versions)
2121
if multiple_versions:
2122
self._contents_map[component_id] = content
747
version_idx = self._index.lookup(component_id)
748
if method == 'fulltext':
749
assert content is None
750
content = self.factory.parse_fulltext(data, version_idx)
751
elif method == 'line-delta':
752
delta = self.factory.parse_line_delta(data[:],
754
content = content.copy()
755
content._lines = self._apply_delta(content._lines,
757
content_map[component_id] = content
759
if 'no-eol' in self._index.get_options(version_id):
760
content = content.copy()
761
line = content._lines[-1][1].rstrip('\n')
762
content._lines[-1] = (content._lines[-1][0], line)
763
final_content[version_id] = content
2124
765
# digest here is the digest from the last applied component.
2125
766
text = content.text()
2126
actual_sha = sha_strings(text)
2127
if actual_sha != digest:
2128
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2129
if multiple_versions:
2130
return self._contents_map[requested_key]
2134
def _wire_bytes(self):
2135
"""Get the bytes to put on the wire for 'key'.
2137
The first collection of bytes asked for returns the serialised
2138
raw_record_map and the additional details (key, parent) for key.
2139
Subsequent calls return just the additional details (key, parent).
2140
The wire storage_kind given for the first key is 'knit-delta-closure',
2141
For subsequent keys it is 'knit-delta-closure-ref'.
2143
:param key: A key from the content generator.
2144
:return: Bytes to put on the wire.
2147
# kind marker for dispatch on the far side,
2148
lines.append('knit-delta-closure')
2150
if self.vf._factory.annotated:
2151
lines.append('annotated')
2154
# then the list of keys
2155
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2156
if key not in self.nonlocal_keys]))
2157
# then the _raw_record_map in serialised form:
2159
# for each item in the map:
2161
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2162
# one line with method
2163
# one line with noeol
2164
# one line with next ('' for None)
2165
# one line with byte count of the record bytes
2167
for key, (record_bytes, (method, noeol), next) in \
2168
self._raw_record_map.iteritems():
2169
key_bytes = '\x00'.join(key)
2170
parents = self.global_map.get(key, None)
2172
parent_bytes = 'None:'
2174
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2175
method_bytes = method
2181
next_bytes = '\x00'.join(next)
2184
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2185
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2186
len(record_bytes), record_bytes))
2187
map_bytes = ''.join(map_byte_list)
2188
lines.append(map_bytes)
2189
bytes = '\n'.join(lines)
2193
class _VFContentMapGenerator(_ContentMapGenerator):
2194
"""Content map generator reading from a VersionedFiles object."""
2196
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2197
global_map=None, raw_record_map=None, ordering='unordered'):
2198
"""Create a _ContentMapGenerator.
2200
:param versioned_files: The versioned files that the texts are being
2202
:param keys: The keys to produce content maps for.
2203
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2204
which are known to not be in this knit, but rather in one of the
2206
:param global_map: The result of get_parent_map(keys) (or a supermap).
2207
This is required if get_record_stream() is to be used.
2208
:param raw_record_map: A unparsed raw record map to use for answering
2211
_ContentMapGenerator.__init__(self, ordering=ordering)
2212
# The vf to source data from
2213
self.vf = versioned_files
2215
self.keys = list(keys)
2216
# Keys known to be in fallback vfs objects
2217
if nonlocal_keys is None:
2218
self.nonlocal_keys = set()
2220
self.nonlocal_keys = frozenset(nonlocal_keys)
2221
# Parents data for keys to be returned in get_record_stream
2222
self.global_map = global_map
2223
# The chunked lists for self.keys in text form
2225
# A cache of KnitContent objects used in extracting texts.
2226
self._contents_map = {}
2227
# All the knit records needed to assemble the requested keys as full
2229
self._record_map = None
2230
if raw_record_map is None:
2231
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2234
self._raw_record_map = raw_record_map
2235
# the factory for parsing records
2236
self._factory = self.vf._factory
2239
class _NetworkContentMapGenerator(_ContentMapGenerator):
2240
"""Content map generator sourced from a network stream."""
2242
def __init__(self, bytes, line_end):
2243
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2245
self.global_map = {}
2246
self._raw_record_map = {}
2247
self._contents_map = {}
2248
self._record_map = None
2249
self.nonlocal_keys = []
2250
# Get access to record parsing facilities
2251
self.vf = KnitVersionedFiles(None, None)
2254
line_end = bytes.find('\n', start)
2255
line = bytes[start:line_end]
2256
start = line_end + 1
2257
if line == 'annotated':
2258
self._factory = KnitAnnotateFactory()
2260
self._factory = KnitPlainFactory()
2261
# list of keys to emit in get_record_stream
2262
line_end = bytes.find('\n', start)
2263
line = bytes[start:line_end]
2264
start = line_end + 1
2266
tuple(segment.split('\x00')) for segment in line.split('\t')
2268
# now a loop until the end. XXX: It would be nice if this was just a
2269
# bunch of the same records as get_record_stream(..., False) gives, but
2270
# there is a decent sized gap stopping that at the moment.
2274
line_end = bytes.find('\n', start)
2275
key = tuple(bytes[start:line_end].split('\x00'))
2276
start = line_end + 1
2277
# 1 line with parents (None: for None, '' for ())
2278
line_end = bytes.find('\n', start)
2279
line = bytes[start:line_end]
2284
[tuple(segment.split('\x00')) for segment in line.split('\t')
2286
self.global_map[key] = parents
2287
start = line_end + 1
2288
# one line with method
2289
line_end = bytes.find('\n', start)
2290
line = bytes[start:line_end]
2292
start = line_end + 1
2293
# one line with noeol
2294
line_end = bytes.find('\n', start)
2295
line = bytes[start:line_end]
2297
start = line_end + 1
2298
# one line with next ('' for None)
2299
line_end = bytes.find('\n', start)
2300
line = bytes[start:line_end]
2304
next = tuple(bytes[start:line_end].split('\x00'))
2305
start = line_end + 1
2306
# one line with byte count of the record bytes
2307
line_end = bytes.find('\n', start)
2308
line = bytes[start:line_end]
2310
start = line_end + 1
2312
record_bytes = bytes[start:start+count]
2313
start = start + count
2315
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2317
def get_record_stream(self):
2318
"""Get a record stream for for keys requested by the bytestream."""
2320
for key in self.keys:
2321
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2324
def _wire_bytes(self):
2328
class _KndxIndex(object):
2329
"""Manages knit index files
2331
The index is kept in memory and read on startup, to enable
767
if sha_strings(text) != digest:
768
raise KnitCorrupt(self.filename,
769
'sha-1 does not match %s' % version_id)
771
text_map[version_id] = text
772
return text_map, final_content
774
def iter_lines_added_or_present_in_versions(self, version_ids=None):
775
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
776
if version_ids is None:
777
version_ids = self.versions()
778
# we don't care about inclusions, the caller cares.
779
# but we need to setup a list of records to visit.
780
# we need version_id, position, length
781
version_id_records = []
782
requested_versions = list(version_ids)
783
# filter for available versions
784
for version_id in requested_versions:
785
if not self.has_version(version_id):
786
raise RevisionNotPresent(version_id, self.filename)
787
# get a in-component-order queue:
789
for version_id in self.versions():
790
if version_id in requested_versions:
791
version_ids.append(version_id)
792
data_pos, length = self._index.get_position(version_id)
793
version_id_records.append((version_id, data_pos, length))
795
pb = bzrlib.ui.ui_factory.nested_progress_bar()
797
total = len(version_id_records)
799
pb.update('Walking content.', count, total)
800
for version_id, data, sha_value in \
801
self._data.read_records_iter(version_id_records):
802
pb.update('Walking content.', count, total)
803
method = self._index.get_method(version_id)
804
version_idx = self._index.lookup(version_id)
805
assert method in ('fulltext', 'line-delta')
806
if method == 'fulltext':
807
content = self.factory.parse_fulltext(data, version_idx)
808
for line in content.text():
811
delta = self.factory.parse_line_delta(data, version_idx)
812
for start, end, count, lines in delta:
813
for origin, line in lines:
816
pb.update('Walking content.', total, total)
819
pb.update('Walking content.', total, total)
823
def num_versions(self):
824
"""See VersionedFile.num_versions()."""
825
return self._index.num_versions()
827
__len__ = num_versions
829
def annotate_iter(self, version_id):
830
"""See VersionedFile.annotate_iter."""
831
content = self._get_content(version_id)
832
for origin, text in content.annotate_iter():
835
def get_parents(self, version_id):
836
"""See VersionedFile.get_parents."""
839
# 52554 calls in 1264 872 internal down from 3674
841
return self._index.get_parents(version_id)
843
raise RevisionNotPresent(version_id, self.filename)
845
def get_parents_with_ghosts(self, version_id):
846
"""See VersionedFile.get_parents."""
848
return self._index.get_parents_with_ghosts(version_id)
850
raise RevisionNotPresent(version_id, self.filename)
852
def get_ancestry(self, versions):
853
"""See VersionedFile.get_ancestry."""
854
if isinstance(versions, basestring):
855
versions = [versions]
858
self._check_versions_present(versions)
859
return self._index.get_ancestry(versions)
861
def get_ancestry_with_ghosts(self, versions):
862
"""See VersionedFile.get_ancestry_with_ghosts."""
863
if isinstance(versions, basestring):
864
versions = [versions]
867
self._check_versions_present(versions)
868
return self._index.get_ancestry_with_ghosts(versions)
870
#@deprecated_method(zero_eight)
871
def walk(self, version_ids):
872
"""See VersionedFile.walk."""
873
# We take the short path here, and extract all relevant texts
874
# and put them in a weave and let that do all the work. Far
875
# from optimal, but is much simpler.
876
# FIXME RB 20060228 this really is inefficient!
877
from bzrlib.weave import Weave
879
w = Weave(self.filename)
880
ancestry = self.get_ancestry(version_ids)
881
sorted_graph = topo_sort(self._index.get_graph())
882
version_list = [vid for vid in sorted_graph if vid in ancestry]
884
for version_id in version_list:
885
lines = self.get_lines(version_id)
886
w.add_lines(version_id, self.get_parents(version_id), lines)
888
for lineno, insert_id, dset, line in w.walk(version_ids):
889
yield lineno, insert_id, dset, line
891
def plan_merge(self, ver_a, ver_b):
892
"""See VersionedFile.plan_merge."""
893
ancestors_b = set(self.get_ancestry(ver_b))
894
def status_a(revision, text):
895
if revision in ancestors_b:
896
return 'killed-b', text
900
ancestors_a = set(self.get_ancestry(ver_a))
901
def status_b(revision, text):
902
if revision in ancestors_a:
903
return 'killed-a', text
907
annotated_a = self.annotate(ver_a)
908
annotated_b = self.annotate(ver_b)
909
plain_a = [t for (a, t) in annotated_a]
910
plain_b = [t for (a, t) in annotated_b]
911
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
914
for ai, bi, l in blocks:
915
# process all mismatched sections
916
# (last mismatched section is handled because blocks always
917
# includes a 0-length last block)
918
for revision, text in annotated_a[a_cur:ai]:
919
yield status_a(revision, text)
920
for revision, text in annotated_b[b_cur:bi]:
921
yield status_b(revision, text)
923
# and now the matched section
926
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
927
assert text_a == text_b
928
yield "unchanged", text_a
931
class _KnitComponentFile(object):
932
"""One of the files used to implement a knit database"""
934
def __init__(self, transport, filename, mode, file_mode=None):
935
self._transport = transport
936
self._filename = filename
938
self._file_mode=file_mode
940
def write_header(self):
941
if self._transport.append(self._filename, StringIO(self.HEADER),
942
mode=self._file_mode):
943
raise KnitCorrupt(self._filename, 'misaligned after writing header')
945
def check_header(self, fp):
947
if line != self.HEADER:
948
raise KnitHeaderError(badline=line)
951
"""Commit is a nop."""
954
return '%s(%s)' % (self.__class__.__name__, self._filename)
957
class _KnitIndex(_KnitComponentFile):
958
"""Manages knit index file.
960
The index is already kept in memory and read on startup, to enable
2332
961
fast lookups of revision information. The cursor of the index
2333
962
file is always pointing to the end, making it easy to append
2345
974
Duplicate entries may be written to the index for a single version id
2346
975
if this is done then the latter one completely replaces the former:
2347
this allows updates to correct version and parent information.
976
this allows updates to correct version and parent information.
2348
977
Note that the two entries may share the delta, and that successive
2349
978
annotations and references MUST point to the first entry.
2351
980
The index file on disc contains a header, followed by one line per knit
2352
981
record. The same revision can be present in an index file more than once.
2353
The first occurrence gets assigned a sequence number starting from 0.
982
The first occurrence gets assigned a sequence number starting from 0.
2355
984
The format of a single line is
2356
985
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
2357
986
REVISION_ID is a utf8-encoded revision id
2358
FLAGS is a comma separated list of flags about the record. Values include
987
FLAGS is a comma separated list of flags about the record. Values include
2359
988
no-eol, line-delta, fulltext.
2360
989
BYTE_OFFSET is the ascii representation of the byte offset in the data file
2361
that the compressed data starts at.
990
that the the compressed data starts at.
2362
991
LENGTH is the ascii representation of the length of the data file.
2363
992
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
2365
994
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
2366
995
revision id already in the knit that is a parent of REVISION_ID.
2367
996
The ' :' marker is the end of record marker.
2370
when a write is interrupted to the index file, it will result in a line
2371
that does not end in ' :'. If the ' :' is not present at the end of a line,
2372
or at the end of the file, then the record that is missing it will be
2373
ignored by the parser.
999
when a write is interrupted to the index file, it will result in a line that
1000
does not end in ' :'. If the ' :' is not present at the end of a line, or at
1001
the end of the file, then the record that is missing it will be ignored by
2375
1004
When writing new records to the index file, the data is preceded by '\n'
2376
1005
to ensure that records always start on new lines even if the last write was
2377
1006
interrupted. As a result its normal for the last line in the index to be
2378
1007
missing a trailing newline. One can be added with no harmful effects.
2380
:ivar _kndx_cache: dict from prefix to the old state of KnitIndex objects,
2381
where prefix is e.g. the (fileid,) for .texts instances or () for
2382
constant-mapped things like .revisions, and the old state is
2383
tuple(history_vector, cache_dict). This is used to prevent having an
2384
ABI change with the C extension that reads .kndx files.
2387
1010
HEADER = "# bzr knit index 8\n"
2389
def __init__(self, transport, mapper, get_scope, allow_writes, is_locked):
2390
"""Create a _KndxIndex on transport using mapper."""
2391
self._transport = transport
2392
self._mapper = mapper
2393
self._get_scope = get_scope
2394
self._allow_writes = allow_writes
2395
self._is_locked = is_locked
2397
self.has_graph = True
2399
def add_records(self, records, random_id=False, missing_compression_parents=False):
2400
"""Add multiple records to the index.
2402
:param records: a list of tuples:
2403
(key, options, access_memo, parents).
2404
:param random_id: If True the ids being added were randomly generated
2405
and no check for existence will be performed.
2406
:param missing_compression_parents: If True the records being added are
2407
only compressed against texts already in the index (or inside
2408
records). If False the records all refer to unavailable texts (or
2409
texts inside records) as compression parents.
2411
if missing_compression_parents:
2412
# It might be nice to get the edge of the records. But keys isn't
2414
keys = sorted(record[0] for record in records)
2415
raise errors.RevisionNotPresent(keys, self)
2417
for record in records:
2420
path = self._mapper.map(key) + '.kndx'
2421
path_keys = paths.setdefault(path, (prefix, []))
2422
path_keys[1].append(record)
2423
for path in sorted(paths):
2424
prefix, path_keys = paths[path]
2425
self._load_prefixes([prefix])
2427
orig_history = self._kndx_cache[prefix][1][:]
2428
orig_cache = self._kndx_cache[prefix][0].copy()
2431
for key, options, (_, pos, size), parents in path_keys:
2433
# kndx indices cannot be parentless.
2435
line = "\n%s %s %s %s %s :" % (
2436
key[-1], ','.join(options), pos, size,
2437
self._dictionary_compress(parents))
2438
if type(line) is not str:
2439
raise AssertionError(
2440
'data must be utf8 was %s' % type(line))
2442
self._cache_key(key, options, pos, size, parents)
2443
if len(orig_history):
2444
self._transport.append_bytes(path, ''.join(lines))
2446
self._init_index(path, lines)
2448
# If any problems happen, restore the original values and re-raise
2449
self._kndx_cache[prefix] = (orig_cache, orig_history)
2452
def scan_unvalidated_index(self, graph_index):
2453
"""See _KnitGraphIndex.scan_unvalidated_index."""
2454
# Because kndx files do not support atomic insertion via separate index
2455
# files, they do not support this method.
2456
raise NotImplementedError(self.scan_unvalidated_index)
2458
def get_missing_compression_parents(self):
2459
"""See _KnitGraphIndex.get_missing_compression_parents."""
2460
# Because kndx files do not support atomic insertion via separate index
2461
# files, they do not support this method.
2462
raise NotImplementedError(self.get_missing_compression_parents)
2464
def _cache_key(self, key, options, pos, size, parent_keys):
1012
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1013
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1015
def _cache_version(self, version_id, options, pos, size, parents):
2465
1016
"""Cache a version record in the history array and index cache.
2467
This is inlined into _load_data for performance. KEEP IN SYNC.
1018
This is inlined into __init__ for performance. KEEP IN SYNC.
2468
1019
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
2472
version_id = key[-1]
2473
# last-element only for compatibilty with the C load_data.
2474
parents = tuple(parent[-1] for parent in parent_keys)
2475
for parent in parent_keys:
2476
if parent[:-1] != prefix:
2477
raise ValueError("mismatched prefixes for %r, %r" % (
2479
cache, history = self._kndx_cache[prefix]
2480
1022
# only want the _history index to reference the 1st index entry
2481
1023
# for version_id
2482
if version_id not in cache:
2483
index = len(history)
2484
history.append(version_id)
1024
if version_id not in self._cache:
1025
index = len(self._history)
1026
self._history.append(version_id)
2486
index = cache[version_id][5]
2487
cache[version_id] = (version_id,
1028
index = self._cache[version_id][5]
1029
self._cache[version_id] = (version_id,
2494
def check_header(self, fp):
2495
line = fp.readline()
2497
# An empty file can actually be treated as though the file doesn't
2499
raise errors.NoSuchFile(self)
2500
if line != self.HEADER:
2501
raise KnitHeaderError(badline=line, filename=self)
2503
def _check_read(self):
2504
if not self._is_locked():
2505
raise errors.ObjectNotLocked(self)
2506
if self._get_scope() != self._scope:
2509
def _check_write_ok(self):
2510
"""Assert if not writes are permitted."""
2511
if not self._is_locked():
2512
raise errors.ObjectNotLocked(self)
2513
if self._get_scope() != self._scope:
2515
if self._mode != 'w':
2516
raise errors.ReadOnlyObjectDirtiedError(self)
2518
def get_build_details(self, keys):
2519
"""Get the method, index_memo and compression parent for keys.
2521
Ghosts are omitted from the result.
2523
:param keys: An iterable of keys.
2524
:return: A dict of key:(index_memo, compression_parent, parents,
2527
opaque structure to pass to read_records to extract the raw
2530
Content that this record is built upon, may be None
2532
Logical parents of this node
2534
extra information about the content which needs to be passed to
2535
Factory.parse_record
2537
parent_map = self.get_parent_map(keys)
2540
if key not in parent_map:
2542
method = self.get_method(key)
2543
parents = parent_map[key]
2544
if method == 'fulltext':
2545
compression_parent = None
2547
compression_parent = parents[0]
2548
noeol = 'no-eol' in self.get_options(key)
2549
index_memo = self.get_position(key)
2550
result[key] = (index_memo, compression_parent,
2551
parents, (method, noeol))
2554
def get_method(self, key):
2555
"""Return compression method of specified key."""
2556
options = self.get_options(key)
2557
if 'fulltext' in options:
2559
elif 'line-delta' in options:
2562
raise errors.KnitIndexUnknownMethod(self, options)
2564
def get_options(self, key):
2565
"""Return a list representing options.
2569
prefix, suffix = self._split_key(key)
2570
self._load_prefixes([prefix])
1036
def __init__(self, transport, filename, mode, create=False, file_mode=None):
1037
_KnitComponentFile.__init__(self, transport, filename, mode, file_mode)
1039
# position in _history is the 'official' index for a revision
1040
# but the values may have come from a newer entry.
1041
# so - wc -l of a knit index is != the number of unique names
1044
pb = bzrlib.ui.ui_factory.nested_progress_bar()
2572
return self._kndx_cache[prefix][0][suffix][1]
2574
raise RevisionNotPresent(key, self)
2576
def find_ancestry(self, keys):
2577
"""See CombinedGraphIndex.find_ancestry()"""
2578
prefixes = set(key[:-1] for key in keys)
2579
self._load_prefixes(prefixes)
2582
missing_keys = set()
2583
pending_keys = list(keys)
2584
# This assumes that keys will not reference parents in a different
2585
# prefix, which is accurate so far.
2587
key = pending_keys.pop()
2588
if key in parent_map:
2592
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2594
missing_keys.add(key)
1049
pb.update('read knit index', count, total)
1050
fp = self._transport.get(self._filename)
1052
self.check_header(fp)
1053
# readlines reads the whole file at once:
1054
# bad for transports like http, good for local disk
1055
# we save 60 ms doing this one change (
1056
# from calling readline each time to calling
1058
# probably what we want for nice behaviour on
1059
# http is a incremental readlines that yields, or
1060
# a check for local vs non local indexes,
1061
for l in fp.readlines():
1063
if len(rec) < 5 or rec[-1] != ':':
1065
# FIXME: in the future we should determine if its a
1066
# short write - and ignore it
1067
# or a different failure, and raise. RBC 20060407
1071
#pb.update('read knit index', count, total)
1072
# See self._parse_parents
1074
for value in rec[4:-1]:
1076
# uncompressed reference
1077
parents.append(value[1:])
1079
# this is 15/4000ms faster than isinstance,
1081
# this function is called thousands of times a
1082
# second so small variations add up.
1083
assert value.__class__ is str
1084
parents.append(self._history[int(value)])
1085
# end self._parse_parents
1086
# self._cache_version(rec[0],
1087
# rec[1].split(','),
1091
# --- self._cache_version
1092
# only want the _history index to reference the 1st
1093
# index entry for version_id
1095
if version_id not in self._cache:
1096
index = len(self._history)
1097
self._history.append(version_id)
1099
index = self._cache[version_id][5]
1100
self._cache[version_id] = (version_id,
1106
# --- self._cache_version
1109
except NoSuchFile, e:
1110
if mode != 'w' or not create:
1114
pb.update('read knit index', total, total)
1117
def _parse_parents(self, compressed_parents):
1118
"""convert a list of string parent values into version ids.
1120
ints are looked up in the index.
1121
.FOO values are ghosts and converted in to FOO.
1123
NOTE: the function is retained here for clarity, and for possible
1124
use in partial index reads. However bulk processing now has
1125
it inlined in __init__ for inner-loop optimisation.
1128
for value in compressed_parents:
1129
if value[-1] == '.':
1130
# uncompressed reference
1131
result.append(value[1:])
2596
parent_keys = tuple([prefix + (suffix,)
2597
for suffix in suffix_parents])
2598
parent_map[key] = parent_keys
2599
pending_keys.extend([p for p in parent_keys
2600
if p not in parent_map])
2601
return parent_map, missing_keys
2603
def get_parent_map(self, keys):
2604
"""Get a map of the parents of keys.
2606
:param keys: The keys to look up parents for.
2607
:return: A mapping from keys to parents. Absent keys are absent from
2610
# Parse what we need to up front, this potentially trades off I/O
2611
# locality (.kndx and .knit in the same block group for the same file
2612
# id) for less checking in inner loops.
2613
prefixes = set(key[:-1] for key in keys)
2614
self._load_prefixes(prefixes)
1133
# this is 15/4000ms faster than isinstance,
1134
# this function is called thousands of times a
1135
# second so small variations add up.
1136
assert value.__class__ is str
1137
result.append(self._history[int(value)])
1140
def get_graph(self):
1142
for version_id, index in self._cache.iteritems():
1143
graph.append((version_id, index[4]))
1146
def get_ancestry(self, versions):
1147
"""See VersionedFile.get_ancestry."""
1148
# get a graph of all the mentioned versions:
1150
pending = set(versions)
1152
version = pending.pop()
1153
parents = self._cache[version][4]
1154
# got the parents ok
1156
parents = [parent for parent in parents if parent in self._cache]
1157
for parent in parents:
1158
# if not completed and not a ghost
1159
if parent not in graph:
1161
graph[version] = parents
1162
return topo_sort(graph.items())
1164
def get_ancestry_with_ghosts(self, versions):
1165
"""See VersionedFile.get_ancestry_with_ghosts."""
1166
# get a graph of all the mentioned versions:
1168
pending = set(versions)
1170
version = pending.pop()
2619
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
1172
parents = self._cache[version][4]
2620
1173
except KeyError:
2623
result[key] = tuple(prefix + (suffix,) for
2624
suffix in suffix_parents)
2627
def get_position(self, key):
2628
"""Return details needed to access the version.
2630
:return: a tuple (key, data position, size) to hand to the access
2631
logic to get the record.
2633
prefix, suffix = self._split_key(key)
2634
self._load_prefixes([prefix])
2635
entry = self._kndx_cache[prefix][0][suffix]
2636
return key, entry[2], entry[3]
2638
has_key = _mod_index._has_key_from_parent_map
2640
def _init_index(self, path, extra_lines=[]):
2641
"""Initialize an index."""
2643
sio.write(self.HEADER)
2644
sio.writelines(extra_lines)
2646
self._transport.put_file_non_atomic(path, sio,
2647
create_parent_dir=True)
2648
# self._create_parent_dir)
2649
# mode=self._file_mode,
2650
# dir_mode=self._dir_mode)
2653
"""Get all the keys in the collection.
2655
The keys are not ordered.
2658
# Identify all key prefixes.
2659
# XXX: A bit hacky, needs polish.
2660
if type(self._mapper) is ConstantMapper:
2664
for quoted_relpath in self._transport.iter_files_recursive():
2665
path, ext = os.path.splitext(quoted_relpath)
2667
prefixes = [self._mapper.unmap(path) for path in relpaths]
2668
self._load_prefixes(prefixes)
2669
for prefix in prefixes:
2670
for suffix in self._kndx_cache[prefix][1]:
2671
result.add(prefix + (suffix,))
2674
def _load_prefixes(self, prefixes):
2675
"""Load the indices for prefixes."""
2677
for prefix in prefixes:
2678
if prefix not in self._kndx_cache:
2679
# the load_data interface writes to these variables.
2682
self._filename = prefix
2684
path = self._mapper.map(prefix) + '.kndx'
2685
fp = self._transport.get(path)
2687
# _load_data may raise NoSuchFile if the target knit is
2689
_load_data(self, fp)
2692
self._kndx_cache[prefix] = (self._cache, self._history)
2697
self._kndx_cache[prefix] = ({}, [])
2698
if type(self._mapper) is ConstantMapper:
2699
# preserve behaviour for revisions.kndx etc.
2700
self._init_index(path)
2705
missing_keys = _mod_index._missing_keys_from_parent_map
2707
def _partition_keys(self, keys):
2708
"""Turn keys into a dict of prefix:suffix_list."""
2711
prefix_keys = result.setdefault(key[:-1], [])
2712
prefix_keys.append(key[-1])
2715
def _dictionary_compress(self, keys):
2716
"""Dictionary compress keys.
2718
:param keys: The keys to generate references to.
2719
:return: A string representation of keys. keys which are present are
2720
dictionary compressed, and others are emitted as fulltext with a
1178
# got the parents ok
1179
for parent in parents:
1180
if parent not in graph:
1182
graph[version] = parents
1183
return topo_sort(graph.items())
1185
def num_versions(self):
1186
return len(self._history)
1188
__len__ = num_versions
1190
def get_versions(self):
1191
return self._history
1193
def idx_to_name(self, idx):
1194
return self._history[idx]
1196
def lookup(self, version_id):
1197
assert version_id in self._cache
1198
return self._cache[version_id][5]
1200
def _version_list_to_index(self, versions):
2725
1201
result_list = []
2726
prefix = keys[0][:-1]
2727
cache = self._kndx_cache[prefix][0]
2729
if key[:-1] != prefix:
2730
# kndx indices cannot refer across partitioned storage.
2731
raise ValueError("mismatched prefixes for %r" % keys)
2732
if key[-1] in cache:
1202
for version in versions:
1203
if version in self._cache:
2733
1204
# -- inlined lookup() --
2734
result_list.append(str(cache[key[-1]][5]))
1205
result_list.append(str(self._cache[version][5]))
2735
1206
# -- end lookup () --
2737
result_list.append('.' + key[-1])
1208
result_list.append('.' + version.encode('utf-8'))
2738
1209
return ' '.join(result_list)
2740
def _reset_cache(self):
2741
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2742
# (cache_dict, history_vector) for parsed kndx files.
2743
self._kndx_cache = {}
2744
self._scope = self._get_scope()
2745
allow_writes = self._allow_writes()
2751
def _sort_keys_by_io(self, keys, positions):
2752
"""Figure out an optimal order to read the records for the given keys.
2754
Sort keys, grouped by index and sorted by position.
2756
:param keys: A list of keys whose records we want to read. This will be
2758
:param positions: A dict, such as the one returned by
2759
_get_components_positions()
2762
def get_sort_key(key):
2763
index_memo = positions[key][1]
2764
# Group by prefix and position. index_memo[0] is the key, so it is
2765
# (file_id, revision_id) and we don't want to sort on revision_id,
2766
# index_memo[1] is the position, and index_memo[2] is the size,
2767
# which doesn't matter for the sort
2768
return index_memo[0][:-1], index_memo[1]
2769
return keys.sort(key=get_sort_key)
2771
_get_total_build_size = _get_total_build_size
2773
def _split_key(self, key):
2774
"""Split key into a prefix and suffix."""
2775
return key[:-1], key[-1]
2778
class _KnitGraphIndex(object):
2779
"""A KnitVersionedFiles index layered on GraphIndex."""
2781
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2782
add_callback=None, track_external_parent_refs=False):
2783
"""Construct a KnitGraphIndex on a graph_index.
2785
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2786
:param is_locked: A callback to check whether the object should answer
2788
:param deltas: Allow delta-compressed records.
2789
:param parents: If True, record knits parents, if not do not record
2791
:param add_callback: If not None, allow additions to the index and call
2792
this callback with a list of added GraphIndex nodes:
2793
[(node, value, node_refs), ...]
2794
:param is_locked: A callback, returns True if the index is locked and
2796
:param track_external_parent_refs: If True, record all external parent
2797
references parents from added records. These can be retrieved
2798
later by calling get_missing_parents().
2800
self._add_callback = add_callback
2801
self._graph_index = graph_index
2802
self._deltas = deltas
2803
self._parents = parents
2804
if deltas and not parents:
2805
# XXX: TODO: Delta tree and parent graph should be conceptually
2807
raise KnitCorrupt(self, "Cannot do delta compression without "
2809
self.has_graph = parents
2810
self._is_locked = is_locked
2811
self._missing_compression_parents = set()
2812
if track_external_parent_refs:
2813
self._key_dependencies = _KeyRefs()
2815
self._key_dependencies = None
2818
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2820
def add_records(self, records, random_id=False,
2821
missing_compression_parents=False):
2822
"""Add multiple records to the index.
2824
This function does not insert data into the Immutable GraphIndex
2825
backing the KnitGraphIndex, instead it prepares data for insertion by
2826
the caller and checks that it is safe to insert then calls
2827
self._add_callback with the prepared GraphIndex nodes.
2829
:param records: a list of tuples:
2830
(key, options, access_memo, parents).
2831
:param random_id: If True the ids being added were randomly generated
2832
and no check for existence will be performed.
2833
:param missing_compression_parents: If True the records being added are
2834
only compressed against texts already in the index (or inside
2835
records). If False the records all refer to unavailable texts (or
2836
texts inside records) as compression parents.
2838
if not self._add_callback:
2839
raise errors.ReadOnlyError(self)
2840
# we hope there are no repositories with inconsistent parentage
2844
compression_parents = set()
2845
key_dependencies = self._key_dependencies
2846
for (key, options, access_memo, parents) in records:
2848
parents = tuple(parents)
2849
if key_dependencies is not None:
2850
key_dependencies.add_references(key, parents)
2851
index, pos, size = access_memo
2852
if 'no-eol' in options:
2856
value += "%d %d" % (pos, size)
2857
if not self._deltas:
2858
if 'line-delta' in options:
2859
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2862
if 'line-delta' in options:
2863
node_refs = (parents, (parents[0],))
2864
if missing_compression_parents:
2865
compression_parents.add(parents[0])
2867
node_refs = (parents, ())
2869
node_refs = (parents, )
2872
raise KnitCorrupt(self, "attempt to add node with parents "
2873
"in parentless index.")
2875
keys[key] = (value, node_refs)
2878
present_nodes = self._get_entries(keys)
2879
for (index, key, value, node_refs) in present_nodes:
2880
parents = node_refs[:1]
2881
# Sometimes these are passed as a list rather than a tuple
2882
passed = static_tuple.as_tuples(keys[key])
2883
passed_parents = passed[1][:1]
2884
if (value[0] != keys[key][0][0] or
2885
parents != passed_parents):
2886
node_refs = static_tuple.as_tuples(node_refs)
2887
raise KnitCorrupt(self, "inconsistent details in add_records"
2888
": %s %s" % ((value, node_refs), passed))
2892
for key, (value, node_refs) in keys.iteritems():
2893
result.append((key, value, node_refs))
2895
for key, (value, node_refs) in keys.iteritems():
2896
result.append((key, value))
2897
self._add_callback(result)
2898
if missing_compression_parents:
2899
# This may appear to be incorrect (it does not check for
2900
# compression parents that are in the existing graph index),
2901
# but such records won't have been buffered, so this is
2902
# actually correct: every entry when
2903
# missing_compression_parents==True either has a missing parent, or
2904
# a parent that is one of the keys in records.
2905
compression_parents.difference_update(keys)
2906
self._missing_compression_parents.update(compression_parents)
2907
# Adding records may have satisfied missing compression parents.
2908
self._missing_compression_parents.difference_update(keys)
2910
def scan_unvalidated_index(self, graph_index):
2911
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2913
This allows this _KnitGraphIndex to keep track of any missing
2914
compression parents we may want to have filled in to make those
2917
:param graph_index: A GraphIndex
2920
new_missing = graph_index.external_references(ref_list_num=1)
2921
new_missing.difference_update(self.get_parent_map(new_missing))
2922
self._missing_compression_parents.update(new_missing)
2923
if self._key_dependencies is not None:
2924
# Add parent refs from graph_index (and discard parent refs that
2925
# the graph_index has).
2926
for node in graph_index.iter_all_entries():
2927
self._key_dependencies.add_references(node[1], node[3][0])
2929
def get_missing_compression_parents(self):
2930
"""Return the keys of missing compression parents.
2932
Missing compression parents occur when a record stream was missing
2933
basis texts, or a index was scanned that had missing basis texts.
2935
return frozenset(self._missing_compression_parents)
2937
def get_missing_parents(self):
2938
"""Return the keys of missing parents."""
2939
# If updating this, you should also update
2940
# groupcompress._GCGraphIndex.get_missing_parents
2941
# We may have false positives, so filter those out.
2942
self._key_dependencies.satisfy_refs_for_keys(
2943
self.get_parent_map(self._key_dependencies.get_unsatisfied_refs()))
2944
return frozenset(self._key_dependencies.get_unsatisfied_refs())
2946
def _check_read(self):
2947
"""raise if reads are not permitted."""
2948
if not self._is_locked():
2949
raise errors.ObjectNotLocked(self)
2951
def _check_write_ok(self):
2952
"""Assert if writes are not permitted."""
2953
if not self._is_locked():
2954
raise errors.ObjectNotLocked(self)
2956
def _compression_parent(self, an_entry):
2957
# return the key that an_entry is compressed against, or None
2958
# Grab the second parent list (as deltas implies parents currently)
2959
compression_parents = an_entry[3][1]
2960
if not compression_parents:
2962
if len(compression_parents) != 1:
2963
raise AssertionError(
2964
"Too many compression parents: %r" % compression_parents)
2965
return compression_parents[0]
2967
def get_build_details(self, keys):
2968
"""Get the method, index_memo and compression parent for version_ids.
2970
Ghosts are omitted from the result.
2972
:param keys: An iterable of keys.
2973
:return: A dict of key:
2974
(index_memo, compression_parent, parents, record_details).
2976
opaque structure to pass to read_records to extract the raw
2979
Content that this record is built upon, may be None
2981
Logical parents of this node
2983
extra information about the content which needs to be passed to
2984
Factory.parse_record
2988
entries = self._get_entries(keys, False)
2989
for entry in entries:
2991
if not self._parents:
2994
parents = entry[3][0]
2995
if not self._deltas:
2996
compression_parent_key = None
2998
compression_parent_key = self._compression_parent(entry)
2999
noeol = (entry[2][0] == 'N')
3000
if compression_parent_key:
3001
method = 'line-delta'
3004
result[key] = (self._node_to_position(entry),
3005
compression_parent_key, parents,
3009
def _get_entries(self, keys, check_present=False):
3010
"""Get the entries for keys.
3012
:param keys: An iterable of index key tuples.
3017
for node in self._graph_index.iter_entries(keys):
3019
found_keys.add(node[1])
3021
# adapt parentless index to the rest of the code.
3022
for node in self._graph_index.iter_entries(keys):
3023
yield node[0], node[1], node[2], ()
3024
found_keys.add(node[1])
3026
missing_keys = keys.difference(found_keys)
3028
raise RevisionNotPresent(missing_keys.pop(), self)
3030
def get_method(self, key):
3031
"""Return compression method of specified key."""
3032
return self._get_method(self._get_node(key))
3034
def _get_method(self, node):
3035
if not self._deltas:
1211
def add_version(self, version_id, options, pos, size, parents):
1212
"""Add a version record to the index."""
1213
self.add_versions(((version_id, options, pos, size, parents),))
1215
def add_versions(self, versions):
1216
"""Add multiple versions to the index.
1218
:param versions: a list of tuples:
1219
(version_id, options, pos, size, parents).
1222
for version_id, options, pos, size, parents in versions:
1223
line = "\n%s %s %s %s %s :" % (version_id.encode('utf-8'),
1227
self._version_list_to_index(parents))
1228
assert isinstance(line, str), \
1229
'content must be utf-8 encoded: %r' % (line,)
1231
self._transport.append(self._filename, StringIO(''.join(lines)))
1232
# cache after writing, so that a failed write leads to missing cache
1233
# entries not extra ones. XXX TODO: RBC 20060502 in the event of a
1234
# failure, reload the index or flush it or some such, to prevent
1235
# writing records that did complete twice.
1236
for version_id, options, pos, size, parents in versions:
1237
self._cache_version(version_id, options, pos, size, parents)
1239
def has_version(self, version_id):
1240
"""True if the version is in the index."""
1241
return self._cache.has_key(version_id)
1243
def get_position(self, version_id):
1244
"""Return data position and size of specified version."""
1245
return (self._cache[version_id][2], \
1246
self._cache[version_id][3])
1248
def get_method(self, version_id):
1249
"""Return compression method of specified version."""
1250
options = self._cache[version_id][1]
1251
if 'fulltext' in options:
3036
1252
return 'fulltext'
3037
if self._compression_parent(node):
1254
assert 'line-delta' in options
3038
1255
return 'line-delta'
3042
def _get_node(self, key):
3044
return list(self._get_entries([key]))[0]
3046
raise RevisionNotPresent(key, self)
3048
def get_options(self, key):
3049
"""Return a list representing options.
3053
node = self._get_node(key)
3054
options = [self._get_method(node)]
3055
if node[2][0] == 'N':
3056
options.append('no-eol')
3059
def find_ancestry(self, keys):
3060
"""See CombinedGraphIndex.find_ancestry()"""
3061
return self._graph_index.find_ancestry(keys, 0)
3063
def get_parent_map(self, keys):
3064
"""Get a map of the parents of keys.
3066
:param keys: The keys to look up parents for.
3067
:return: A mapping from keys to parents. Absent keys are absent from
3071
nodes = self._get_entries(keys)
3075
result[node[1]] = node[3][0]
3078
result[node[1]] = None
3081
def get_position(self, key):
3082
"""Return details needed to access the version.
3084
:return: a tuple (index, data position, size) to hand to the access
3085
logic to get the record.
3087
node = self._get_node(key)
3088
return self._node_to_position(node)
3090
has_key = _mod_index._has_key_from_parent_map
3093
"""Get all the keys in the collection.
3095
The keys are not ordered.
3098
return [node[1] for node in self._graph_index.iter_all_entries()]
3100
missing_keys = _mod_index._missing_keys_from_parent_map
3102
def _node_to_position(self, node):
3103
"""Convert an index value to position details."""
3104
bits = node[2][1:].split(' ')
3105
return node[0], int(bits[0]), int(bits[1])
3107
def _sort_keys_by_io(self, keys, positions):
3108
"""Figure out an optimal order to read the records for the given keys.
3110
Sort keys, grouped by index and sorted by position.
3112
:param keys: A list of keys whose records we want to read. This will be
3114
:param positions: A dict, such as the one returned by
3115
_get_components_positions()
3118
def get_index_memo(key):
3119
# index_memo is at offset [1]. It is made up of (GraphIndex,
3120
# position, size). GI is an object, which will be unique for each
3121
# pack file. This causes us to group by pack file, then sort by
3122
# position. Size doesn't matter, but it isn't worth breaking up the
3124
return positions[key][1]
3125
return keys.sort(key=get_index_memo)
3127
_get_total_build_size = _get_total_build_size
3130
class _KnitKeyAccess(object):
3131
"""Access to records in .knit files."""
3133
def __init__(self, transport, mapper):
3134
"""Create a _KnitKeyAccess with transport and mapper.
3136
:param transport: The transport the access object is rooted at.
3137
:param mapper: The mapper used to map keys to .knit files.
3139
self._transport = transport
3140
self._mapper = mapper
3142
def add_raw_records(self, key_sizes, raw_data):
3143
"""Add raw knit bytes to a storage area.
3145
The data is spooled to the container writer in one bytes-record per
3148
:param sizes: An iterable of tuples containing the key and size of each
3150
:param raw_data: A bytestring containing the data.
3151
:return: A list of memos to retrieve the record later. Each memo is an
3152
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3153
length), where the key is the record key.
3155
if type(raw_data) is not str:
3156
raise AssertionError(
3157
'data must be plain bytes was %s' % type(raw_data))
3160
# TODO: This can be tuned for writing to sftp and other servers where
3161
# append() is relatively expensive by grouping the writes to each key
3163
for key, size in key_sizes:
3164
path = self._mapper.map(key)
3166
base = self._transport.append_bytes(path + '.knit',
3167
raw_data[offset:offset+size])
3168
except errors.NoSuchFile:
3169
self._transport.mkdir(osutils.dirname(path))
3170
base = self._transport.append_bytes(path + '.knit',
3171
raw_data[offset:offset+size])
3175
result.append((key, base, size))
3179
"""Flush pending writes on this access object.
3181
For .knit files this is a no-op.
1257
def get_options(self, version_id):
1258
return self._cache[version_id][1]
1260
def get_parents(self, version_id):
1261
"""Return parents of specified version ignoring ghosts."""
1262
return [parent for parent in self._cache[version_id][4]
1263
if parent in self._cache]
1265
def get_parents_with_ghosts(self, version_id):
1266
"""Return parents of specified version with ghosts."""
1267
return self._cache[version_id][4]
1269
def check_versions_present(self, version_ids):
1270
"""Check that all specified versions are present."""
1271
version_ids = set(version_ids)
1272
for version_id in list(version_ids):
1273
if version_id in self._cache:
1274
version_ids.remove(version_id)
1276
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1279
class _KnitData(_KnitComponentFile):
1280
"""Contents of the knit data file"""
1282
HEADER = "# bzr knit data 8\n"
1284
def __init__(self, transport, filename, mode, create=False, file_mode=None):
1285
_KnitComponentFile.__init__(self, transport, filename, mode)
1286
self._checked = False
1288
self._transport.put(self._filename, StringIO(''), mode=file_mode)
1290
def clear_cache(self):
1291
"""Clear the record cache."""
3185
def get_raw_records(self, memos_for_retrieval):
3186
"""Get the raw bytes for a records.
3188
:param memos_for_retrieval: An iterable containing the access memo for
3189
retrieving the bytes.
3190
:return: An iterator over the bytes of the records.
3192
# first pass, group into same-index request to minimise readv's issued.
3194
current_prefix = None
3195
for (key, offset, length) in memos_for_retrieval:
3196
if current_prefix == key[:-1]:
3197
current_list.append((offset, length))
3199
if current_prefix is not None:
3200
request_lists.append((current_prefix, current_list))
3201
current_prefix = key[:-1]
3202
current_list = [(offset, length)]
3203
# handle the last entry
3204
if current_prefix is not None:
3205
request_lists.append((current_prefix, current_list))
3206
for prefix, read_vector in request_lists:
3207
path = self._mapper.map(prefix) + '.knit'
3208
for pos, data in self._transport.readv(path, read_vector):
3212
class _DirectPackAccess(object):
3213
"""Access to data in one or more packs with less translation."""
3215
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
3216
"""Create a _DirectPackAccess object.
3218
:param index_to_packs: A dict mapping index objects to the transport
3219
and file names for obtaining data.
3220
:param reload_func: A function to call if we determine that the pack
3221
files have moved and we need to reload our caches. See
3222
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
3224
self._container_writer = None
3225
self._write_index = None
3226
self._indices = index_to_packs
3227
self._reload_func = reload_func
3228
self._flush_func = flush_func
3230
def add_raw_records(self, key_sizes, raw_data):
3231
"""Add raw knit bytes to a storage area.
3233
The data is spooled to the container writer in one bytes-record per
3236
:param sizes: An iterable of tuples containing the key and size of each
3238
:param raw_data: A bytestring containing the data.
3239
:return: A list of memos to retrieve the record later. Each memo is an
3240
opaque index memo. For _DirectPackAccess the memo is (index, pos,
3241
length), where the index field is the write_index object supplied
3242
to the PackAccess object.
3244
if type(raw_data) is not str:
3245
raise AssertionError(
3246
'data must be plain bytes was %s' % type(raw_data))
3249
for key, size in key_sizes:
3250
p_offset, p_length = self._container_writer.add_bytes_record(
3251
raw_data[offset:offset+size], [])
3253
result.append((self._write_index, p_offset, p_length))
3257
"""Flush pending writes on this access object.
3259
This will flush any buffered writes to a NewPack.
3261
if self._flush_func is not None:
1294
def _open_file(self):
1296
return self._transport.get(self._filename)
1301
def _record_to_data(self, version_id, digest, lines):
1302
"""Convert version_id, digest, lines into a raw data block.
1304
:return: (len, a StringIO instance with the raw data ready to read.)
1307
data_file = GzipFile(None, mode='wb', fileobj=sio)
1308
data_file.writelines(chain(
1309
["version %s %d %s\n" % (version_id.encode('utf-8'),
1313
["end %s\n" % version_id.encode('utf-8')]))
1320
def add_raw_record(self, raw_data):
1321
"""Append a prepared record to the data file.
1323
:return: the offset in the data file raw_data was written.
1325
assert isinstance(raw_data, str), 'data must be plain bytes'
1326
return self._transport.append(self._filename, StringIO(raw_data))
1328
def add_record(self, version_id, digest, lines):
1329
"""Write new text record to disk. Returns the position in the
1330
file where it was written."""
1331
size, sio = self._record_to_data(version_id, digest, lines)
1333
start_pos = self._transport.append(self._filename, sio)
1334
return start_pos, size
1336
def _parse_record_header(self, version_id, raw_data):
1337
"""Parse a record header for consistency.
1339
:return: the header and the decompressor stream.
1340
as (stream, header_record)
1342
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1343
rec = df.readline().split()
1345
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
1346
if rec[1].decode('utf-8')!= version_id:
1347
raise KnitCorrupt(self._filename,
1348
'unexpected version, wanted %r, got %r' % (
1349
version_id, rec[1]))
1352
def _parse_record(self, version_id, data):
1354
# 4168 calls in 2880 217 internal
1355
# 4168 calls to _parse_record_header in 2121
1356
# 4168 calls to readlines in 330
1357
df, rec = self._parse_record_header(version_id, data)
1358
record_contents = df.readlines()
1359
l = record_contents.pop()
1360
assert len(record_contents) == int(rec[2])
1361
if l.decode('utf-8') != 'end %s\n' % version_id:
1362
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
1365
return record_contents, rec[3]
1367
def read_records_iter_raw(self, records):
1368
"""Read text records from data file and yield raw data.
1370
This unpacks enough of the text record to validate the id is
1371
as expected but thats all.
1373
It will actively recompress currently cached records on the
1374
basis that that is cheaper than I/O activity.
1376
# setup an iterator of the external records:
1377
# uses readv so nice and fast we hope.
1379
# grab the disk data needed.
1380
raw_records = self._transport.readv(self._filename,
1381
[(pos, size) for version_id, pos, size in records])
1383
for version_id, pos, size in records:
1384
pos, data = raw_records.next()
1385
# validate the header
1386
df, rec = self._parse_record_header(version_id, data)
1388
yield version_id, data
1390
def read_records_iter(self, records):
1391
"""Read text records from data file and yield result.
1393
Each passed record is a tuple of (version_id, pos, len) and
1394
will be read in the given order. Yields (version_id,
1397
if len(records) == 0:
1400
# 60890 calls for 4168 extractions in 5045, 683 internal.
1401
# 4168 calls to readv in 1411
1402
# 4168 calls to parse_record in 2880
1404
# Get unique records, sorted by position
1405
needed_records = sorted(set(records), key=operator.itemgetter(1))
1407
# We take it that the transport optimizes the fetching as good
1408
# as possible (ie, reads continuous ranges.)
1409
response = self._transport.readv(self._filename,
1410
[(pos, size) for version_id, pos, size in needed_records])
1413
for (record_id, pos, size), (pos, data) in \
1414
izip(iter(needed_records), response):
1415
content, digest = self._parse_record(record_id, data)
1416
record_map[record_id] = (digest, content)
1418
for version_id, pos, size in records:
1419
digest, content = record_map[version_id]
1420
yield version_id, content, digest
1422
def read_records(self, records):
1423
"""Read records into a dictionary."""
1425
for record_id, content, digest in self.read_records_iter(records):
1426
components[record_id] = (content, digest)
1430
class InterKnit(InterVersionedFile):
1431
"""Optimised code paths for knit to knit operations."""
1433
_matching_file_from_factory = KnitVersionedFile
1434
_matching_file_to_factory = KnitVersionedFile
1437
def is_compatible(source, target):
1438
"""Be compatible with knits. """
1440
return (isinstance(source, KnitVersionedFile) and
1441
isinstance(target, KnitVersionedFile))
1442
except AttributeError:
1445
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1446
"""See InterVersionedFile.join."""
1447
assert isinstance(self.source, KnitVersionedFile)
1448
assert isinstance(self.target, KnitVersionedFile)
1450
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1455
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1457
version_ids = list(version_ids)
1458
if None in version_ids:
1459
version_ids.remove(None)
1461
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1462
this_versions = set(self.target._index.get_versions())
1463
needed_versions = self.source_ancestry - this_versions
1464
cross_check_versions = self.source_ancestry.intersection(this_versions)
1465
mismatched_versions = set()
1466
for version in cross_check_versions:
1467
# scan to include needed parents.
1468
n1 = set(self.target.get_parents_with_ghosts(version))
1469
n2 = set(self.source.get_parents_with_ghosts(version))
1471
# FIXME TEST this check for cycles being introduced works
1472
# the logic is we have a cycle if in our graph we are an
1473
# ancestor of any of the n2 revisions.
1479
parent_ancestors = self.source.get_ancestry(parent)
1480
if version in parent_ancestors:
1481
raise errors.GraphCycleError([parent, version])
1482
# ensure this parent will be available later.
1483
new_parents = n2.difference(n1)
1484
needed_versions.update(new_parents.difference(this_versions))
1485
mismatched_versions.add(version)
1487
if not needed_versions and not mismatched_versions:
1489
full_list = topo_sort(self.source.get_graph())
1491
version_list = [i for i in full_list if (not self.target.has_version(i)
1492
and i in needed_versions)]
1496
copy_queue_records = []
1498
for version_id in version_list:
1499
options = self.source._index.get_options(version_id)
1500
parents = self.source._index.get_parents_with_ghosts(version_id)
1501
# check that its will be a consistent copy:
1502
for parent in parents:
1503
# if source has the parent, we must :
1504
# * already have it or
1505
# * have it scheduled already
1506
# otherwise we don't care
1507
assert (self.target.has_version(parent) or
1508
parent in copy_set or
1509
not self.source.has_version(parent))
1510
data_pos, data_size = self.source._index.get_position(version_id)
1511
copy_queue_records.append((version_id, data_pos, data_size))
1512
copy_queue.append((version_id, options, parents))
1513
copy_set.add(version_id)
1515
# data suck the join:
1517
total = len(version_list)
1520
for (version_id, raw_data), \
1521
(version_id2, options, parents) in \
1522
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1524
assert version_id == version_id2, 'logic error, inconsistent results'
1526
pb.update("Joining knit", count, total)
1527
raw_records.append((version_id, options, parents, len(raw_data)))
1528
raw_datum.append(raw_data)
1529
self.target._add_raw_records(raw_records, ''.join(raw_datum))
1531
for version in mismatched_versions:
1532
# FIXME RBC 20060309 is this needed?
1533
n1 = set(self.target.get_parents_with_ghosts(version))
1534
n2 = set(self.source.get_parents_with_ghosts(version))
1535
# write a combined record to our history preserving the current
1536
# parents as first in the list
1537
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1538
self.target.fix_parents(version, new_parents)
1544
InterVersionedFile.register_optimiser(InterKnit)
1547
class WeaveToKnit(InterVersionedFile):
1548
"""Optimised code paths for weave to knit operations."""
1550
_matching_file_from_factory = bzrlib.weave.WeaveFile
1551
_matching_file_to_factory = KnitVersionedFile
1554
def is_compatible(source, target):
1555
"""Be compatible with weaves to knits."""
1557
return (isinstance(source, bzrlib.weave.Weave) and
1558
isinstance(target, KnitVersionedFile))
1559
except AttributeError:
1562
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1563
"""See InterVersionedFile.join."""
1564
assert isinstance(self.source, bzrlib.weave.Weave)
1565
assert isinstance(self.target, KnitVersionedFile)
1567
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1572
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1574
version_ids = list(version_ids)
1576
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1577
this_versions = set(self.target._index.get_versions())
1578
needed_versions = self.source_ancestry - this_versions
1579
cross_check_versions = self.source_ancestry.intersection(this_versions)
1580
mismatched_versions = set()
1581
for version in cross_check_versions:
1582
# scan to include needed parents.
1583
n1 = set(self.target.get_parents_with_ghosts(version))
1584
n2 = set(self.source.get_parents(version))
1585
# if all of n2's parents are in n1, then its fine.
1586
if n2.difference(n1):
1587
# FIXME TEST this check for cycles being introduced works
1588
# the logic is we have a cycle if in our graph we are an
1589
# ancestor of any of the n2 revisions.
1595
parent_ancestors = self.source.get_ancestry(parent)
1596
if version in parent_ancestors:
1597
raise errors.GraphCycleError([parent, version])
1598
# ensure this parent will be available later.
1599
new_parents = n2.difference(n1)
1600
needed_versions.update(new_parents.difference(this_versions))
1601
mismatched_versions.add(version)
1603
if not needed_versions and not mismatched_versions:
1605
full_list = topo_sort(self.source.get_graph())
1607
version_list = [i for i in full_list if (not self.target.has_version(i)
1608
and i in needed_versions)]
1612
total = len(version_list)
1613
for version_id in version_list:
1614
pb.update("Converting to knit", count, total)
1615
parents = self.source.get_parents(version_id)
1616
# check that its will be a consistent copy:
1617
for parent in parents:
1618
# if source has the parent, we must already have it
1619
assert (self.target.has_version(parent))
1620
self.target.add_lines(
1621
version_id, parents, self.source.get_lines(version_id))
1624
for version in mismatched_versions:
1625
# FIXME RBC 20060309 is this needed?
1626
n1 = set(self.target.get_parents_with_ghosts(version))
1627
n2 = set(self.source.get_parents(version))
1628
# write a combined record to our history preserving the current
1629
# parents as first in the list
1630
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1631
self.target.fix_parents(version, new_parents)
1637
InterVersionedFile.register_optimiser(WeaveToKnit)
1640
class KnitSequenceMatcher(difflib.SequenceMatcher):
1641
"""Knit tuned sequence matcher.
1643
This is based on profiling of difflib which indicated some improvements
1644
for our usage pattern.
1647
def find_longest_match(self, alo, ahi, blo, bhi):
1648
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
1650
If isjunk is not defined:
1652
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
1653
alo <= i <= i+k <= ahi
1654
blo <= j <= j+k <= bhi
1655
and for all (i',j',k') meeting those conditions,
1658
and if i == i', j <= j'
1660
In other words, of all maximal matching blocks, return one that
1661
starts earliest in a, and of all those maximal matching blocks that
1662
start earliest in a, return the one that starts earliest in b.
1664
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
1665
>>> s.find_longest_match(0, 5, 0, 9)
1668
If isjunk is defined, first the longest matching block is
1669
determined as above, but with the additional restriction that no
1670
junk element appears in the block. Then that block is extended as
1671
far as possible by matching (only) junk elements on both sides. So
1672
the resulting block never matches on junk except as identical junk
1673
happens to be adjacent to an "interesting" match.
1675
Here's the same example as before, but considering blanks to be
1676
junk. That prevents " abcd" from matching the " abcd" at the tail
1677
end of the second sequence directly. Instead only the "abcd" can
1678
match, and matches the leftmost "abcd" in the second sequence:
1680
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
1681
>>> s.find_longest_match(0, 5, 0, 9)
1684
If no blocks match, return (alo, blo, 0).
1686
>>> s = SequenceMatcher(None, "ab", "c")
1687
>>> s.find_longest_match(0, 2, 0, 1)
1691
# CAUTION: stripping common prefix or suffix would be incorrect.
1695
# Longest matching block is "ab", but if common prefix is
1696
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
1697
# strip, so ends up claiming that ab is changed to acab by
1698
# inserting "ca" in the middle. That's minimal but unintuitive:
1699
# "it's obvious" that someone inserted "ac" at the front.
1700
# Windiff ends up at the same place as diff, but by pairing up
1701
# the unique 'b's and then matching the first two 'a's.
1703
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
1704
besti, bestj, bestsize = alo, blo, 0
1705
# find longest junk-free match
1706
# during an iteration of the loop, j2len[j] = length of longest
1707
# junk-free match ending with a[i-1] and b[j]
1711
for i in xrange(alo, ahi):
1712
# look at all instances of a[i] in b; note that because
1713
# b2j has no junk keys, the loop is skipped if a[i] is junk
1714
j2lenget = j2len.get
3264
def get_raw_records(self, memos_for_retrieval):
3265
"""Get the raw bytes for a records.
1717
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
1718
# following improvement
1719
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
1720
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
1721
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
1723
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
1724
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
1725
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
3267
:param memos_for_retrieval: An iterable containing the (index, pos,
3268
length) memo for retrieving the bytes. The Pack access method
3269
looks up the pack to use for a given record in its index_to_pack
3271
:return: An iterator over the bytes of the records.
3273
# first pass, group into same-index requests
3275
current_index = None
3276
for (index, offset, length) in memos_for_retrieval:
3277
if current_index == index:
3278
current_list.append((offset, length))
3280
if current_index is not None:
3281
request_lists.append((current_index, current_list))
3282
current_index = index
3283
current_list = [(offset, length)]
3284
# handle the last entry
3285
if current_index is not None:
3286
request_lists.append((current_index, current_list))
3287
for index, offsets in request_lists:
3289
transport, path = self._indices[index]
3290
1729
except KeyError:
3291
# A KeyError here indicates that someone has triggered an index
3292
# reload, and this index has gone missing, we need to start
3294
if self._reload_func is None:
3295
# If we don't have a _reload_func there is nothing that can
3298
raise errors.RetryWithNewPacks(index,
3299
reload_occurred=True,
3300
exc_info=sys.exc_info())
3302
reader = pack.make_readv_reader(transport, path, offsets)
3303
for names, read_func in reader.iter_records():
3304
yield read_func(None)
3305
except errors.NoSuchFile:
3306
# A NoSuchFile error indicates that a pack file has gone
3307
# missing on disk, we need to trigger a reload, and start over.
3308
if self._reload_func is None:
3310
raise errors.RetryWithNewPacks(transport.abspath(path),
3311
reload_occurred=False,
3312
exc_info=sys.exc_info())
3314
def set_writer(self, writer, index, transport_packname):
3315
"""Set a writer to use for adding data."""
3316
if index is not None:
3317
self._indices[index] = transport_packname
3318
self._container_writer = writer
3319
self._write_index = index
3321
def reload_or_raise(self, retry_exc):
3322
"""Try calling the reload function, or re-raise the original exception.
3324
This should be called after _DirectPackAccess raises a
3325
RetryWithNewPacks exception. This function will handle the common logic
3326
of determining when the error is fatal versus being temporary.
3327
It will also make sure that the original exception is raised, rather
3328
than the RetryWithNewPacks exception.
3330
If this function returns, then the calling function should retry
3331
whatever operation was being performed. Otherwise an exception will
3334
:param retry_exc: A RetryWithNewPacks exception.
3337
if self._reload_func is None:
3339
elif not self._reload_func():
3340
# The reload claimed that nothing changed
3341
if not retry_exc.reload_occurred:
3342
# If there wasn't an earlier reload, then we really were
3343
# expecting to find changes. We didn't find them, so this is a
3347
exc_class, exc_value, exc_traceback = retry_exc.exc_info
3348
raise exc_class, exc_value, exc_traceback
3351
def annotate_knit(knit, revision_id):
3352
"""Annotate a knit with no cached annotations.
3354
This implementation is for knits with no cached annotations.
3355
It will work for knits with cached annotations, but this is not
3358
annotator = _KnitAnnotator(knit)
3359
return iter(annotator.annotate_flat(revision_id))
3362
class _KnitAnnotator(annotate.Annotator):
3363
"""Build up the annotations for a text."""
3365
def __init__(self, vf):
3366
annotate.Annotator.__init__(self, vf)
3368
# TODO: handle Nodes which cannot be extracted
3369
# self._ghosts = set()
3371
# Map from (key, parent_key) => matching_blocks, should be 'use once'
3372
self._matching_blocks = {}
3374
# KnitContent objects
3375
self._content_objects = {}
3376
# The number of children that depend on this fulltext content object
3377
self._num_compression_children = {}
3378
# Delta records that need their compression parent before they can be
3380
self._pending_deltas = {}
3381
# Fulltext records that are waiting for their parents fulltexts before
3382
# they can be yielded for annotation
3383
self._pending_annotation = {}
3385
self._all_build_details = {}
3387
def _get_build_graph(self, key):
3388
"""Get the graphs for building texts and annotations.
3390
The data you need for creating a full text may be different than the
3391
data you need to annotate that text. (At a minimum, you need both
3392
parents to create an annotation, but only need 1 parent to generate the
3395
:return: A list of (key, index_memo) records, suitable for
3396
passing to read_records_iter to start reading in the raw data from
3399
pending = set([key])
3402
self._num_needed_children[key] = 1
3404
# get all pending nodes
3405
this_iteration = pending
3406
build_details = self._vf._index.get_build_details(this_iteration)
3407
self._all_build_details.update(build_details)
3408
# new_nodes = self._vf._index._get_entries(this_iteration)
3410
for key, details in build_details.iteritems():
3411
(index_memo, compression_parent, parent_keys,
3412
record_details) = details
3413
self._parent_map[key] = parent_keys
3414
self._heads_provider = None
3415
records.append((key, index_memo))
3416
# Do we actually need to check _annotated_lines?
3417
pending.update([p for p in parent_keys
3418
if p not in self._all_build_details])
3420
for parent_key in parent_keys:
3421
if parent_key in self._num_needed_children:
3422
self._num_needed_children[parent_key] += 1
3424
self._num_needed_children[parent_key] = 1
3425
if compression_parent:
3426
if compression_parent in self._num_compression_children:
3427
self._num_compression_children[compression_parent] += 1
3429
self._num_compression_children[compression_parent] = 1
3431
missing_versions = this_iteration.difference(build_details.keys())
3432
if missing_versions:
3433
for key in missing_versions:
3434
if key in self._parent_map and key in self._text_cache:
3435
# We already have this text ready, we just need to
3436
# yield it later so we get it annotated
3438
parent_keys = self._parent_map[key]
3439
for parent_key in parent_keys:
3440
if parent_key in self._num_needed_children:
3441
self._num_needed_children[parent_key] += 1
3443
self._num_needed_children[parent_key] = 1
3444
pending.update([p for p in parent_keys
3445
if p not in self._all_build_details])
3447
raise errors.RevisionNotPresent(key, self._vf)
3448
# Generally we will want to read the records in reverse order, because
3449
# we find the parent nodes after the children
3451
return records, ann_keys
3453
def _get_needed_texts(self, key, pb=None):
3454
# if True or len(self._vf._immediate_fallback_vfs) > 0:
3455
if len(self._vf._immediate_fallback_vfs) > 0:
3456
# If we have fallbacks, go to the generic path
3457
for v in annotate.Annotator._get_needed_texts(self, key, pb=pb):
3462
records, ann_keys = self._get_build_graph(key)
3463
for idx, (sub_key, text, num_lines) in enumerate(
3464
self._extract_texts(records)):
3466
pb.update('annotating', idx, len(records))
3467
yield sub_key, text, num_lines
3468
for sub_key in ann_keys:
3469
text = self._text_cache[sub_key]
3470
num_lines = len(text) # bad assumption
3471
yield sub_key, text, num_lines
3473
except errors.RetryWithNewPacks, e:
3474
self._vf._access.reload_or_raise(e)
3475
# The cached build_details are no longer valid
3476
self._all_build_details.clear()
3478
def _cache_delta_blocks(self, key, compression_parent, delta, lines):
3479
parent_lines = self._text_cache[compression_parent]
3480
blocks = list(KnitContent.get_line_delta_blocks(delta, parent_lines, lines))
3481
self._matching_blocks[(key, compression_parent)] = blocks
3483
def _expand_record(self, key, parent_keys, compression_parent, record,
3486
if compression_parent:
3487
if compression_parent not in self._content_objects:
3488
# Waiting for the parent
3489
self._pending_deltas.setdefault(compression_parent, []).append(
3490
(key, parent_keys, record, record_details))
3492
# We have the basis parent, so expand the delta
3493
num = self._num_compression_children[compression_parent]
3496
base_content = self._content_objects.pop(compression_parent)
3497
self._num_compression_children.pop(compression_parent)
3499
self._num_compression_children[compression_parent] = num
3500
base_content = self._content_objects[compression_parent]
3501
# It is tempting to want to copy_base_content=False for the last
3502
# child object. However, whenever noeol=False,
3503
# self._text_cache[parent_key] is content._lines. So mutating it
3504
# gives very bad results.
3505
# The alternative is to copy the lines into text cache, but then we
3506
# are copying anyway, so just do it here.
3507
content, delta = self._vf._factory.parse_record(
3508
key, record, record_details, base_content,
3509
copy_base_content=True)
3512
content, _ = self._vf._factory.parse_record(
3513
key, record, record_details, None)
3514
if self._num_compression_children.get(key, 0) > 0:
3515
self._content_objects[key] = content
3516
lines = content.text()
3517
self._text_cache[key] = lines
3518
if delta is not None:
3519
self._cache_delta_blocks(key, compression_parent, delta, lines)
3522
def _get_parent_annotations_and_matches(self, key, text, parent_key):
3523
"""Get the list of annotations for the parent, and the matching lines.
3525
:param text: The opaque value given by _get_needed_texts
3526
:param parent_key: The key for the parent text
3527
:return: (parent_annotations, matching_blocks)
3528
parent_annotations is a list as long as the number of lines in
3530
matching_blocks is a list of (parent_idx, text_idx, len) tuples
3531
indicating which lines match between the two texts
3533
block_key = (key, parent_key)
3534
if block_key in self._matching_blocks:
3535
blocks = self._matching_blocks.pop(block_key)
3536
parent_annotations = self._annotations_cache[parent_key]
3537
return parent_annotations, blocks
3538
return annotate.Annotator._get_parent_annotations_and_matches(self,
3539
key, text, parent_key)
3541
def _process_pending(self, key):
3542
"""The content for 'key' was just processed.
3544
Determine if there is any more pending work to be processed.
3547
if key in self._pending_deltas:
3548
compression_parent = key
3549
children = self._pending_deltas.pop(key)
3550
for child_key, parent_keys, record, record_details in children:
3551
lines = self._expand_record(child_key, parent_keys,
3553
record, record_details)
3554
if self._check_ready_for_annotations(child_key, parent_keys):
3555
to_return.append(child_key)
3556
# Also check any children that are waiting for this parent to be
3558
if key in self._pending_annotation:
3559
children = self._pending_annotation.pop(key)
3560
to_return.extend([c for c, p_keys in children
3561
if self._check_ready_for_annotations(c, p_keys)])
3564
def _check_ready_for_annotations(self, key, parent_keys):
3565
"""return true if this text is ready to be yielded.
3567
Otherwise, this will return False, and queue the text into
3568
self._pending_annotation
3570
for parent_key in parent_keys:
3571
if parent_key not in self._annotations_cache:
3572
# still waiting on at least one parent text, so queue it up
3573
# Note that if there are multiple parents, we need to wait
3575
self._pending_annotation.setdefault(parent_key,
3576
[]).append((key, parent_keys))
3580
def _extract_texts(self, records):
3581
"""Extract the various texts needed based on records"""
3582
# We iterate in the order read, rather than a strict order requested
3583
# However, process what we can, and put off to the side things that
3584
# still need parents, cleaning them up when those parents are
3587
# 1) As 'records' are read, see if we can expand these records into
3588
# Content objects (and thus lines)
3589
# 2) If a given line-delta is waiting on its compression parent, it
3590
# gets queued up into self._pending_deltas, otherwise we expand
3591
# it, and put it into self._text_cache and self._content_objects
3592
# 3) If we expanded the text, we will then check to see if all
3593
# parents have also been processed. If so, this text gets yielded,
3594
# else this record gets set aside into pending_annotation
3595
# 4) Further, if we expanded the text in (2), we will then check to
3596
# see if there are any children in self._pending_deltas waiting to
3597
# also be processed. If so, we go back to (2) for those
3598
# 5) Further again, if we yielded the text, we can then check if that
3599
# 'unlocks' any of the texts in pending_annotations, which should
3600
# then get yielded as well
3601
# Note that both steps 4 and 5 are 'recursive' in that unlocking one
3602
# compression child could unlock yet another, and yielding a fulltext
3603
# will also 'unlock' the children that are waiting on that annotation.
3604
# (Though also, unlocking 1 parent's fulltext, does not unlock a child
3605
# if other parents are also waiting.)
3606
# We want to yield content before expanding child content objects, so
3607
# that we know when we can re-use the content lines, and the annotation
3608
# code can know when it can stop caching fulltexts, as well.
3610
# Children that are missing their compression parent
3612
for (key, record, digest) in self._vf._read_records_iter(records):
3614
details = self._all_build_details[key]
3615
(_, compression_parent, parent_keys, record_details) = details
3616
lines = self._expand_record(key, parent_keys, compression_parent,
3617
record, record_details)
3619
# Pending delta should be queued up
3621
# At this point, we may be able to yield this content, if all
3622
# parents are also finished
3623
yield_this_text = self._check_ready_for_annotations(key,
3626
# All parents present
3627
yield key, lines, len(lines)
3628
to_process = self._process_pending(key)
3630
this_process = to_process
3632
for key in this_process:
3633
lines = self._text_cache[key]
3634
yield key, lines, len(lines)
3635
to_process.extend(self._process_pending(key))
3638
from bzrlib._knit_load_data_pyx import _load_data_c as _load_data
3639
except ImportError, e:
3640
osutils.failed_to_load_extension(e)
3641
from bzrlib._knit_load_data_py import _load_data_py as _load_data
1737
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
1739
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
1742
# Extend the best by non-junk elements on each end. In particular,
1743
# "popular" non-junk elements aren't in b2j, which greatly speeds
1744
# the inner loop above, but also means "the best" match so far
1745
# doesn't contain any junk *or* popular non-junk elements.
1746
while besti > alo and bestj > blo and \
1747
not isbjunk(b[bestj-1]) and \
1748
a[besti-1] == b[bestj-1]:
1749
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1750
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1751
not isbjunk(b[bestj+bestsize]) and \
1752
a[besti+bestsize] == b[bestj+bestsize]:
1755
# Now that we have a wholly interesting match (albeit possibly
1756
# empty!), we may as well suck up the matching junk on each
1757
# side of it too. Can't think of a good reason not to, and it
1758
# saves post-processing the (possibly considerable) expense of
1759
# figuring out what to do with it. In the case of an empty
1760
# interesting match, this is clearly the right thing to do,
1761
# because no other kind of match is possible in the regions.
1762
while besti > alo and bestj > blo and \
1763
isbjunk(b[bestj-1]) and \
1764
a[besti-1] == b[bestj-1]:
1765
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
1766
while besti+bestsize < ahi and bestj+bestsize < bhi and \
1767
isbjunk(b[bestj+bestsize]) and \
1768
a[besti+bestsize] == b[bestj+bestsize]:
1769
bestsize = bestsize + 1
1771
return besti, bestj, bestsize
added
removed
bzrlib/knit.py
