131
126
DATA_SUFFIX = '.knit'
132
127
INDEX_SUFFIX = '.kndx'
128
_STREAM_MIN_BUFFER_SIZE = 5*1024*1024
131
class KnitAdapter(object):
132
"""Base class for knit record adaption."""
134
def __init__(self, basis_vf):
135
"""Create an adapter which accesses full texts from basis_vf.
137
:param basis_vf: A versioned file to access basis texts of deltas from.
138
May be None for adapters that do not need to access basis texts.
140
self._data = KnitVersionedFiles(None, None)
141
self._annotate_factory = KnitAnnotateFactory()
142
self._plain_factory = KnitPlainFactory()
143
self._basis_vf = basis_vf
146
class FTAnnotatedToUnannotated(KnitAdapter):
147
"""An adapter from FT annotated knits to unannotated ones."""
149
def get_bytes(self, factory):
150
annotated_compressed_bytes = factory._raw_record
152
self._data._parse_record_unchecked(annotated_compressed_bytes)
153
content = self._annotate_factory.parse_fulltext(contents, rec[1])
154
size, bytes = self._data._record_to_data((rec[1],), rec[3], content.text())
158
class DeltaAnnotatedToUnannotated(KnitAdapter):
159
"""An adapter for deltas from annotated to unannotated."""
161
def get_bytes(self, factory):
162
annotated_compressed_bytes = factory._raw_record
164
self._data._parse_record_unchecked(annotated_compressed_bytes)
165
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
167
contents = self._plain_factory.lower_line_delta(delta)
168
size, bytes = self._data._record_to_data((rec[1],), rec[3], contents)
172
class FTAnnotatedToFullText(KnitAdapter):
173
"""An adapter from FT annotated knits to unannotated ones."""
175
def get_bytes(self, factory):
176
annotated_compressed_bytes = factory._raw_record
178
self._data._parse_record_unchecked(annotated_compressed_bytes)
179
content, delta = self._annotate_factory.parse_record(factory.key[-1],
180
contents, factory._build_details, None)
181
return ''.join(content.text())
184
class DeltaAnnotatedToFullText(KnitAdapter):
185
"""An adapter for deltas from annotated to unannotated."""
187
def get_bytes(self, factory):
188
annotated_compressed_bytes = factory._raw_record
190
self._data._parse_record_unchecked(annotated_compressed_bytes)
191
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
193
compression_parent = factory.parents[0]
194
basis_entry = self._basis_vf.get_record_stream(
195
[compression_parent], 'unordered', True).next()
196
if basis_entry.storage_kind == 'absent':
197
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
198
basis_chunks = basis_entry.get_bytes_as('chunked')
199
basis_lines = osutils.chunks_to_lines(basis_chunks)
200
# Manually apply the delta because we have one annotated content and
202
basis_content = PlainKnitContent(basis_lines, compression_parent)
203
basis_content.apply_delta(delta, rec[1])
204
basis_content._should_strip_eol = factory._build_details[1]
205
return ''.join(basis_content.text())
208
class FTPlainToFullText(KnitAdapter):
209
"""An adapter from FT plain knits to unannotated ones."""
211
def get_bytes(self, factory):
212
compressed_bytes = factory._raw_record
214
self._data._parse_record_unchecked(compressed_bytes)
215
content, delta = self._plain_factory.parse_record(factory.key[-1],
216
contents, factory._build_details, None)
217
return ''.join(content.text())
220
class DeltaPlainToFullText(KnitAdapter):
221
"""An adapter for deltas from annotated to unannotated."""
223
def get_bytes(self, factory):
224
compressed_bytes = factory._raw_record
226
self._data._parse_record_unchecked(compressed_bytes)
227
delta = self._plain_factory.parse_line_delta(contents, rec[1])
228
compression_parent = factory.parents[0]
229
# XXX: string splitting overhead.
230
basis_entry = self._basis_vf.get_record_stream(
231
[compression_parent], 'unordered', True).next()
232
if basis_entry.storage_kind == 'absent':
233
raise errors.RevisionNotPresent(compression_parent, self._basis_vf)
234
basis_chunks = basis_entry.get_bytes_as('chunked')
235
basis_lines = osutils.chunks_to_lines(basis_chunks)
236
basis_content = PlainKnitContent(basis_lines, compression_parent)
237
# Manually apply the delta because we have one annotated content and
239
content, _ = self._plain_factory.parse_record(rec[1], contents,
240
factory._build_details, basis_content)
241
return ''.join(content.text())
244
class KnitContentFactory(ContentFactory):
245
"""Content factory for streaming from knits.
247
:seealso ContentFactory:
250
def __init__(self, key, parents, build_details, sha1, raw_record,
251
annotated, knit=None, network_bytes=None):
252
"""Create a KnitContentFactory for key.
255
:param parents: The parents.
256
:param build_details: The build details as returned from
258
:param sha1: The sha1 expected from the full text of this object.
259
:param raw_record: The bytes of the knit data from disk.
260
:param annotated: True if the raw data is annotated.
261
:param network_bytes: None to calculate the network bytes on demand,
262
not-none if they are already known.
264
ContentFactory.__init__(self)
267
self.parents = parents
268
if build_details[0] == 'line-delta':
273
annotated_kind = 'annotated-'
276
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
277
self._raw_record = raw_record
278
self._network_bytes = network_bytes
279
self._build_details = build_details
282
def _create_network_bytes(self):
283
"""Create a fully serialised network version for transmission."""
284
# storage_kind, key, parents, Noeol, raw_record
285
key_bytes = '\x00'.join(self.key)
286
if self.parents is None:
287
parent_bytes = 'None:'
289
parent_bytes = '\t'.join('\x00'.join(key) for key in self.parents)
290
if self._build_details[1]:
294
network_bytes = "%s\n%s\n%s\n%s%s" % (self.storage_kind, key_bytes,
295
parent_bytes, noeol, self._raw_record)
296
self._network_bytes = network_bytes
298
def get_bytes_as(self, storage_kind):
299
if storage_kind == self.storage_kind:
300
if self._network_bytes is None:
301
self._create_network_bytes()
302
return self._network_bytes
303
if ('-ft-' in self.storage_kind and
304
storage_kind in ('chunked', 'fulltext')):
305
adapter_key = (self.storage_kind, 'fulltext')
306
adapter_factory = adapter_registry.get(adapter_key)
307
adapter = adapter_factory(None)
308
bytes = adapter.get_bytes(self)
309
if storage_kind == 'chunked':
313
if self._knit is not None:
314
# Not redundant with direct conversion above - that only handles
316
if storage_kind == 'chunked':
317
return self._knit.get_lines(self.key[0])
318
elif storage_kind == 'fulltext':
319
return self._knit.get_text(self.key[0])
320
raise errors.UnavailableRepresentation(self.key, storage_kind,
324
class LazyKnitContentFactory(ContentFactory):
325
"""A ContentFactory which can either generate full text or a wire form.
327
:seealso ContentFactory:
330
def __init__(self, key, parents, generator, first):
331
"""Create a LazyKnitContentFactory.
333
:param key: The key of the record.
334
:param parents: The parents of the record.
335
:param generator: A _ContentMapGenerator containing the record for this
337
:param first: Is this the first content object returned from generator?
338
if it is, its storage kind is knit-delta-closure, otherwise it is
339
knit-delta-closure-ref
342
self.parents = parents
344
self._generator = generator
345
self.storage_kind = "knit-delta-closure"
347
self.storage_kind = self.storage_kind + "-ref"
350
def get_bytes_as(self, storage_kind):
351
if storage_kind == self.storage_kind:
353
return self._generator._wire_bytes()
355
# all the keys etc are contained in the bytes returned in the
358
if storage_kind in ('chunked', 'fulltext'):
359
chunks = self._generator._get_one_work(self.key).text()
360
if storage_kind == 'chunked':
363
return ''.join(chunks)
364
raise errors.UnavailableRepresentation(self.key, storage_kind,
368
def knit_delta_closure_to_records(storage_kind, bytes, line_end):
369
"""Convert a network record to a iterator over stream records.
371
:param storage_kind: The storage kind of the record.
372
Must be 'knit-delta-closure'.
373
:param bytes: The bytes of the record on the network.
375
generator = _NetworkContentMapGenerator(bytes, line_end)
376
return generator.get_record_stream()
379
def knit_network_to_record(storage_kind, bytes, line_end):
380
"""Convert a network record to a record object.
382
:param storage_kind: The storage kind of the record.
383
:param bytes: The bytes of the record on the network.
386
line_end = bytes.find('\n', start)
387
key = tuple(bytes[start:line_end].split('\x00'))
389
line_end = bytes.find('\n', start)
390
parent_line = bytes[start:line_end]
391
if parent_line == 'None:':
395
[tuple(segment.split('\x00')) for segment in parent_line.split('\t')
398
noeol = bytes[start] == 'N'
399
if 'ft' in storage_kind:
402
method = 'line-delta'
403
build_details = (method, noeol)
405
raw_record = bytes[start:]
406
annotated = 'annotated' in storage_kind
407
return [KnitContentFactory(key, parents, build_details, None, raw_record,
408
annotated, network_bytes=bytes)]
135
411
class KnitContent(object):
136
"""Content of a knit version to which deltas can be applied."""
412
"""Content of a knit version to which deltas can be applied.
414
This is always stored in memory as a list of lines with \n at the end,
415
plus a flag saying if the final ending is really there or not, because that
416
corresponds to the on-disk knit representation.
138
419
def __init__(self):
139
420
self._should_strip_eol = False
142
"""Return a list of (origin, text) tuples."""
143
return list(self.annotate_iter())
145
422
def apply_delta(self, delta, new_version_id):
146
423
"""Apply delta to this object to become new_version_id."""
147
424
raise NotImplementedError(self.apply_delta)
149
def cleanup_eol(self, copy_on_mutate=True):
150
if self._should_strip_eol:
152
self._lines = self._lines[:]
153
self.strip_last_line_newline()
155
426
def line_delta_iter(self, new_lines):
156
427
"""Generate line-based delta from this content to new_lines."""
157
428
new_texts = new_lines.text()
483
755
out.extend(lines)
486
def annotate_iter(self, knit, version_id):
758
def annotate(self, knit, key):
487
759
annotator = _KnitAnnotator(knit)
488
return iter(annotator.annotate(version_id))
491
def make_empty_knit(transport, relpath):
492
"""Construct a empty knit at the specified location."""
493
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
496
class KnitVersionedFile(VersionedFile):
497
"""Weave-like structure with faster random access.
499
A knit stores a number of texts and a summary of the relationships
500
between them. Texts are identified by a string version-id. Texts
501
are normally stored and retrieved as a series of lines, but can
502
also be passed as single strings.
504
Lines are stored with the trailing newline (if any) included, to
505
avoid special cases for files with no final newline. Lines are
506
composed of 8-bit characters, not unicode. The combination of
507
these approaches should mean any 'binary' file can be safely
508
stored and retrieved.
511
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
512
factory=None, delta=True, create=False, create_parent_dir=False,
513
delay_create=False, dir_mode=None, index=None, access_method=None):
514
"""Construct a knit at location specified by relpath.
516
:param create: If not True, only open an existing knit.
517
:param create_parent_dir: If True, create the parent directory if
518
creating the file fails. (This is used for stores with
519
hash-prefixes that may not exist yet)
520
:param delay_create: The calling code is aware that the knit won't
521
actually be created until the first data is stored.
522
:param index: An index to use for the knit.
760
return annotator.annotate_flat(key)
764
def make_file_factory(annotated, mapper):
765
"""Create a factory for creating a file based KnitVersionedFiles.
767
This is only functional enough to run interface tests, it doesn't try to
768
provide a full pack environment.
770
:param annotated: knit annotations are wanted.
771
:param mapper: The mapper from keys to paths.
773
def factory(transport):
774
index = _KndxIndex(transport, mapper, lambda:None, lambda:True, lambda:True)
775
access = _KnitKeyAccess(transport, mapper)
776
return KnitVersionedFiles(index, access, annotated=annotated)
780
def make_pack_factory(graph, delta, keylength):
781
"""Create a factory for creating a pack based VersionedFiles.
783
This is only functional enough to run interface tests, it doesn't try to
784
provide a full pack environment.
786
:param graph: Store a graph.
787
:param delta: Delta compress contents.
788
:param keylength: How long should keys be.
790
def factory(transport):
791
parents = graph or delta
797
max_delta_chain = 200
800
graph_index = _mod_index.InMemoryGraphIndex(reference_lists=ref_length,
801
key_elements=keylength)
802
stream = transport.open_write_stream('newpack')
803
writer = pack.ContainerWriter(stream.write)
805
index = _KnitGraphIndex(graph_index, lambda:True, parents=parents,
806
deltas=delta, add_callback=graph_index.add_nodes)
807
access = _DirectPackAccess({})
808
access.set_writer(writer, graph_index, (transport, 'newpack'))
809
result = KnitVersionedFiles(index, access,
810
max_delta_chain=max_delta_chain)
811
result.stream = stream
812
result.writer = writer
817
def cleanup_pack_knit(versioned_files):
818
versioned_files.stream.close()
819
versioned_files.writer.end()
822
def _get_total_build_size(self, keys, positions):
823
"""Determine the total bytes to build these keys.
825
(helper function because _KnitGraphIndex and _KndxIndex work the same, but
826
don't inherit from a common base.)
828
:param keys: Keys that we want to build
829
:param positions: dict of {key, (info, index_memo, comp_parent)} (such
830
as returned by _get_components_positions)
831
:return: Number of bytes to build those keys
833
all_build_index_memos = {}
837
for key in build_keys:
838
# This is mostly for the 'stacked' case
839
# Where we will be getting the data from a fallback
840
if key not in positions:
842
_, index_memo, compression_parent = positions[key]
843
all_build_index_memos[key] = index_memo
844
if compression_parent not in all_build_index_memos:
845
next_keys.add(compression_parent)
846
build_keys = next_keys
847
return sum([index_memo[2] for index_memo
848
in all_build_index_memos.itervalues()])
851
class KnitVersionedFiles(VersionedFiles):
852
"""Storage for many versioned files using knit compression.
854
Backend storage is managed by indices and data objects.
856
:ivar _index: A _KnitGraphIndex or similar that can describe the
857
parents, graph, compression and data location of entries in this
858
KnitVersionedFiles. Note that this is only the index for
859
*this* vfs; if there are fallbacks they must be queried separately.
862
def __init__(self, index, data_access, max_delta_chain=200,
863
annotated=False, reload_func=None):
864
"""Create a KnitVersionedFiles with index and data_access.
866
:param index: The index for the knit data.
867
:param data_access: The access object to store and retrieve knit
869
:param max_delta_chain: The maximum number of deltas to permit during
870
insertion. Set to 0 to prohibit the use of deltas.
871
:param annotated: Set to True to cause annotations to be calculated and
872
stored during insertion.
873
:param reload_func: An function that can be called if we think we need
874
to reload the pack listing and try again. See
875
'bzrlib.repofmt.pack_repo.AggregateIndex' for the signature.
524
if access_mode is None:
526
super(KnitVersionedFile, self).__init__(access_mode)
527
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
528
self.transport = transport
529
self.filename = relpath
530
self.factory = factory or KnitAnnotateFactory()
531
self.writable = (access_mode == 'w')
534
self._max_delta_chain = 200
537
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
538
access_mode, create=create, file_mode=file_mode,
539
create_parent_dir=create_parent_dir, delay_create=delay_create,
543
if access_method is None:
544
_access = _KnitAccess(transport, relpath + DATA_SUFFIX, file_mode, dir_mode,
545
((create and not len(self)) and delay_create), create_parent_dir)
547
_access = access_method
548
if create and not len(self) and not delay_create:
550
self._data = _KnitData(_access)
878
self._access = data_access
879
self._max_delta_chain = max_delta_chain
881
self._factory = KnitAnnotateFactory()
883
self._factory = KnitPlainFactory()
884
self._fallback_vfs = []
885
self._reload_func = reload_func
552
887
def __repr__(self):
553
return '%s(%s)' % (self.__class__.__name__,
554
self.transport.abspath(self.filename))
556
def _check_should_delta(self, first_parents):
888
return "%s(%r, %r)" % (
889
self.__class__.__name__,
893
def add_fallback_versioned_files(self, a_versioned_files):
894
"""Add a source of texts for texts not present in this knit.
896
:param a_versioned_files: A VersionedFiles object.
898
self._fallback_vfs.append(a_versioned_files)
900
def add_lines(self, key, parents, lines, parent_texts=None,
901
left_matching_blocks=None, nostore_sha=None, random_id=False,
903
"""See VersionedFiles.add_lines()."""
904
self._index._check_write_ok()
905
self._check_add(key, lines, random_id, check_content)
907
# The caller might pass None if there is no graph data, but kndx
908
# indexes can't directly store that, so we give them
909
# an empty tuple instead.
911
line_bytes = ''.join(lines)
912
return self._add(key, lines, parents,
913
parent_texts, left_matching_blocks, nostore_sha, random_id,
914
line_bytes=line_bytes)
916
def _add_text(self, key, parents, text, nostore_sha=None, random_id=False):
917
"""See VersionedFiles._add_text()."""
918
self._index._check_write_ok()
919
self._check_add(key, None, random_id, check_content=False)
920
if text.__class__ is not str:
921
raise errors.BzrBadParameterUnicode("text")
923
# The caller might pass None if there is no graph data, but kndx
924
# indexes can't directly store that, so we give them
925
# an empty tuple instead.
927
return self._add(key, None, parents,
928
None, None, nostore_sha, random_id,
931
def _add(self, key, lines, parents, parent_texts,
932
left_matching_blocks, nostore_sha, random_id,
934
"""Add a set of lines on top of version specified by parents.
936
Any versions not present will be converted into ghosts.
938
:param lines: A list of strings where each one is a single line (has a
939
single newline at the end of the string) This is now optional
940
(callers can pass None). It is left in its location for backwards
941
compatibility. It should ''.join(lines) must == line_bytes
942
:param line_bytes: A single string containing the content
944
We pass both lines and line_bytes because different routes bring the
945
values to this function. And for memory efficiency, we don't want to
946
have to split/join on-demand.
948
# first thing, if the content is something we don't need to store, find
950
digest = sha_string(line_bytes)
951
if nostore_sha == digest:
952
raise errors.ExistingContent
955
if parent_texts is None:
957
# Do a single query to ascertain parent presence; we only compress
958
# against parents in the same kvf.
959
present_parent_map = self._index.get_parent_map(parents)
960
for parent in parents:
961
if parent in present_parent_map:
962
present_parents.append(parent)
964
# Currently we can only compress against the left most present parent.
965
if (len(present_parents) == 0 or
966
present_parents[0] != parents[0]):
969
# To speed the extract of texts the delta chain is limited
970
# to a fixed number of deltas. This should minimize both
971
# I/O and the time spend applying deltas.
972
delta = self._check_should_delta(present_parents[0])
974
text_length = len(line_bytes)
977
# Note: line_bytes is not modified to add a newline, that is tracked
978
# via the no_eol flag. 'lines' *is* modified, because that is the
979
# general values needed by the Content code.
980
if line_bytes and line_bytes[-1] != '\n':
981
options.append('no-eol')
983
# Copy the existing list, or create a new one
985
lines = osutils.split_lines(line_bytes)
988
# Replace the last line with one that ends in a final newline
989
lines[-1] = lines[-1] + '\n'
991
lines = osutils.split_lines(line_bytes)
993
for element in key[:-1]:
994
if type(element) is not str:
995
raise TypeError("key contains non-strings: %r" % (key,))
997
key = key[:-1] + ('sha1:' + digest,)
998
elif type(key[-1]) is not str:
999
raise TypeError("key contains non-strings: %r" % (key,))
1000
# Knit hunks are still last-element only
1001
version_id = key[-1]
1002
content = self._factory.make(lines, version_id)
1004
# Hint to the content object that its text() call should strip the
1006
content._should_strip_eol = True
1007
if delta or (self._factory.annotated and len(present_parents) > 0):
1008
# Merge annotations from parent texts if needed.
1009
delta_hunks = self._merge_annotations(content, present_parents,
1010
parent_texts, delta, self._factory.annotated,
1011
left_matching_blocks)
1014
options.append('line-delta')
1015
store_lines = self._factory.lower_line_delta(delta_hunks)
1016
size, bytes = self._record_to_data(key, digest,
1019
options.append('fulltext')
1020
# isinstance is slower and we have no hierarchy.
1021
if self._factory.__class__ is KnitPlainFactory:
1022
# Use the already joined bytes saving iteration time in
1024
dense_lines = [line_bytes]
1026
dense_lines.append('\n')
1027
size, bytes = self._record_to_data(key, digest,
1030
# get mixed annotation + content and feed it into the
1032
store_lines = self._factory.lower_fulltext(content)
1033
size, bytes = self._record_to_data(key, digest,
1036
access_memo = self._access.add_raw_records([(key, size)], bytes)[0]
1037
self._index.add_records(
1038
((key, options, access_memo, parents),),
1039
random_id=random_id)
1040
return digest, text_length, content
1042
def annotate(self, key):
1043
"""See VersionedFiles.annotate."""
1044
return self._factory.annotate(self, key)
1046
def get_annotator(self):
1047
return _KnitAnnotator(self)
1049
def check(self, progress_bar=None, keys=None):
1050
"""See VersionedFiles.check()."""
1052
return self._logical_check()
1054
# At the moment, check does not extra work over get_record_stream
1055
return self.get_record_stream(keys, 'unordered', True)
1057
def _logical_check(self):
1058
# This doesn't actually test extraction of everything, but that will
1059
# impact 'bzr check' substantially, and needs to be integrated with
1060
# care. However, it does check for the obvious problem of a delta with
1062
keys = self._index.keys()
1063
parent_map = self.get_parent_map(keys)
1065
if self._index.get_method(key) != 'fulltext':
1066
compression_parent = parent_map[key][0]
1067
if compression_parent not in parent_map:
1068
raise errors.KnitCorrupt(self,
1069
"Missing basis parent %s for %s" % (
1070
compression_parent, key))
1071
for fallback_vfs in self._fallback_vfs:
1072
fallback_vfs.check()
1074
def _check_add(self, key, lines, random_id, check_content):
1075
"""check that version_id and lines are safe to add."""
1076
version_id = key[-1]
1077
if version_id is not None:
1078
if contains_whitespace(version_id):
1079
raise InvalidRevisionId(version_id, self)
1080
self.check_not_reserved_id(version_id)
1081
# TODO: If random_id==False and the key is already present, we should
1082
# probably check that the existing content is identical to what is
1083
# being inserted, and otherwise raise an exception. This would make
1084
# the bundle code simpler.
1086
self._check_lines_not_unicode(lines)
1087
self._check_lines_are_lines(lines)
1089
def _check_header(self, key, line):
1090
rec = self._split_header(line)
1091
self._check_header_version(rec, key[-1])
1094
def _check_header_version(self, rec, version_id):
1095
"""Checks the header version on original format knit records.
1097
These have the last component of the key embedded in the record.
1099
if rec[1] != version_id:
1100
raise KnitCorrupt(self,
1101
'unexpected version, wanted %r, got %r' % (version_id, rec[1]))
1103
def _check_should_delta(self, parent):
557
1104
"""Iterate back through the parent listing, looking for a fulltext.
559
1106
This is used when we want to decide whether to add a delta or a new
568
1115
fulltext_size = None
569
delta_parents = first_parents
570
1116
for count in xrange(self._max_delta_chain):
571
parent = delta_parents[0]
572
method = self._index.get_method(parent)
573
index, pos, size = self._index.get_position(parent)
574
if method == 'fulltext':
1118
# Note that this only looks in the index of this particular
1119
# KnitVersionedFiles, not in the fallbacks. This ensures that
1120
# we won't store a delta spanning physical repository
1122
build_details = self._index.get_build_details([parent])
1123
parent_details = build_details[parent]
1124
except (RevisionNotPresent, KeyError), e:
1125
# Some basis is not locally present: always fulltext
1127
index_memo, compression_parent, _, _ = parent_details
1128
_, _, size = index_memo
1129
if compression_parent is None:
575
1130
fulltext_size = size
577
1132
delta_size += size
578
delta_parents = self._index.get_parents(parent)
1133
# We don't explicitly check for presence because this is in an
1134
# inner loop, and if it's missing it'll fail anyhow.
1135
parent = compression_parent
580
1137
# We couldn't find a fulltext, so we must create a new one
1139
# Simple heuristic - if the total I/O wold be greater as a delta than
1140
# the originally installed fulltext, we create a new fulltext.
583
1141
return fulltext_size > delta_size
585
def _add_raw_records(self, records, data):
586
"""Add all the records 'records' with data pre-joined in 'data'.
588
:param records: A list of tuples(version_id, options, parents, size).
589
:param data: The data for the records. When it is written, the records
590
are adjusted to have pos pointing into data by the sum of
591
the preceding records sizes.
594
raw_record_sizes = [record[3] for record in records]
595
positions = self._data.add_raw_records(raw_record_sizes, data)
598
for (version_id, options, parents, size), access_memo in zip(
600
index_entries.append((version_id, options, access_memo, parents))
601
if self._data._do_cache:
602
self._data._cache[version_id] = data[offset:offset+size]
604
self._index.add_versions(index_entries)
606
def enable_cache(self):
607
"""Start caching data for this knit"""
608
self._data.enable_cache()
610
def clear_cache(self):
611
"""Clear the data cache only."""
612
self._data.clear_cache()
614
def copy_to(self, name, transport):
615
"""See VersionedFile.copy_to()."""
616
# copy the current index to a temp index to avoid racing with local
618
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
619
self.transport.get(self._index._filename))
621
f = self._data._open_file()
623
transport.put_file(name + DATA_SUFFIX, f)
626
# move the copied index into place
627
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
629
def create_empty(self, name, transport, mode=None):
630
return KnitVersionedFile(name, transport, factory=self.factory,
631
delta=self.delta, create=True)
633
def get_data_stream(self, required_versions):
634
"""Get a data stream for the specified versions.
636
Versions may be returned in any order, not necessarily the order
637
specified. They are returned in a partial order by compression
638
parent, so that the deltas can be applied as the data stream is
639
inserted; however note that compression parents will not be sent
640
unless they were specifically requested, as the client may already
643
:param required_versions: The exact set of versions to be extracted.
644
Unlike some other knit methods, this is not used to generate a
645
transitive closure, rather it is used precisely as given.
647
:returns: format_signature, list of (version, options, length, parents),
650
required_version_set = frozenset(required_versions)
652
# list of revisions that can just be sent without waiting for their
655
# map from revision to the children based on it
657
# first, read all relevant index data, enough to sort into the right
659
for version_id in required_versions:
660
options = self._index.get_options(version_id)
661
parents = self._index.get_parents_with_ghosts(version_id)
662
index_memo = self._index.get_position(version_id)
663
version_index[version_id] = (index_memo, options, parents)
664
if ('line-delta' in options
665
and parents[0] in required_version_set):
666
# must wait until the parent has been sent
667
deferred.setdefault(parents[0], []). \
670
# either a fulltext, or a delta whose parent the client did
671
# not ask for and presumably already has
672
ready_to_send.append(version_id)
673
# build a list of results to return, plus instructions for data to
675
copy_queue_records = []
676
temp_version_list = []
678
# XXX: pushing and popping lists may be a bit inefficient
679
version_id = ready_to_send.pop(0)
680
(index_memo, options, parents) = version_index[version_id]
681
copy_queue_records.append((version_id, index_memo))
682
none, data_pos, data_size = index_memo
683
temp_version_list.append((version_id, options, data_size,
685
if version_id in deferred:
686
# now we can send all the children of this revision - we could
687
# put them in anywhere, but we hope that sending them soon
688
# after the fulltext will give good locality in the receiver
689
ready_to_send[:0] = deferred.pop(version_id)
690
assert len(deferred) == 0, \
691
"Still have compressed child versions waiting to be sent"
692
# XXX: The stream format is such that we cannot stream it - we have to
693
# know the length of all the data a-priori.
695
result_version_list = []
696
for (version_id, raw_data), \
697
(version_id2, options, _, parents) in \
698
izip(self._data.read_records_iter_raw(copy_queue_records),
700
assert version_id == version_id2, \
701
'logic error, inconsistent results'
702
raw_datum.append(raw_data)
703
result_version_list.append(
704
(version_id, options, len(raw_data), parents))
705
# provide a callback to get data incrementally.
706
pseudo_file = StringIO(''.join(raw_datum))
709
return pseudo_file.read()
711
return pseudo_file.read(length)
712
return (self.get_format_signature(), result_version_list, read)
714
def _extract_blocks(self, version_id, source, target):
715
if self._index.get_method(version_id) != 'line-delta':
717
parent, sha1, noeol, delta = self.get_delta(version_id)
718
return KnitContent.get_line_delta_blocks(delta, source, target)
720
def get_delta(self, version_id):
721
"""Get a delta for constructing version from some other version."""
722
self.check_not_reserved_id(version_id)
723
parents = self.get_parents(version_id)
728
index_memo = self._index.get_position(version_id)
729
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
730
noeol = 'no-eol' in self._index.get_options(version_id)
731
if 'fulltext' == self._index.get_method(version_id):
732
new_content = self.factory.parse_fulltext(data, version_id)
733
if parent is not None:
734
reference_content = self._get_content(parent)
735
old_texts = reference_content.text()
738
new_texts = new_content.text()
739
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
741
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
743
delta = self.factory.parse_line_delta(data, version_id)
744
return parent, sha1, noeol, delta
746
def get_format_signature(self):
747
"""See VersionedFile.get_format_signature()."""
748
if self.factory.annotated:
749
annotated_part = "annotated"
751
annotated_part = "plain"
752
return "knit-%s" % (annotated_part,)
754
def get_graph_with_ghosts(self):
755
"""See VersionedFile.get_graph_with_ghosts()."""
756
graph_items = self._index.get_graph()
757
return dict(graph_items)
759
def get_sha1(self, version_id):
760
return self.get_sha1s([version_id])[0]
762
def get_sha1s(self, version_ids):
763
"""See VersionedFile.get_sha1()."""
764
record_map = self._get_record_map(version_ids)
765
# record entry 2 is the 'digest'.
766
return [record_map[v][2] for v in version_ids]
770
"""See VersionedFile.get_suffixes()."""
771
return [DATA_SUFFIX, INDEX_SUFFIX]
773
def has_ghost(self, version_id):
774
"""True if there is a ghost reference in the file to version_id."""
776
if self.has_version(version_id):
778
# optimisable if needed by memoising the _ghosts set.
779
items = self._index.get_graph()
780
for node, parents in items:
781
for parent in parents:
782
if parent not in self._index._cache:
783
if parent == version_id:
787
def insert_data_stream(self, (format, data_list, reader_callable)):
788
"""Insert knit records from a data stream into this knit.
790
If a version in the stream is already present in this knit, it will not
791
be inserted a second time. It will be checked for consistency with the
792
stored version however, and may cause a KnitCorrupt error to be raised
793
if the data in the stream disagrees with the already stored data.
795
:seealso: get_data_stream
797
if format != self.get_format_signature():
798
if 'knit' in debug.debug_flags:
800
'incompatible format signature inserting to %r', self)
801
source = self._knit_from_datastream(
802
(format, data_list, reader_callable))
806
for version_id, options, length, parents in data_list:
807
if self.has_version(version_id):
808
# First check: the list of parents.
809
my_parents = self.get_parents_with_ghosts(version_id)
810
if tuple(my_parents) != tuple(parents):
811
# XXX: KnitCorrupt is not quite the right exception here.
814
'parents list %r from data stream does not match '
815
'already recorded parents %r for %s'
816
% (parents, my_parents, version_id))
818
# Also check the SHA-1 of the fulltext this content will
820
raw_data = reader_callable(length)
821
my_fulltext_sha1 = self.get_sha1(version_id)
822
df, rec = self._data._parse_record_header(version_id, raw_data)
823
stream_fulltext_sha1 = rec[3]
824
if my_fulltext_sha1 != stream_fulltext_sha1:
825
# Actually, we don't know if it's this knit that's corrupt,
826
# or the data stream we're trying to insert.
828
self.filename, 'sha-1 does not match %s' % version_id)
830
if 'line-delta' in options:
831
# Make sure that this knit record is actually useful: a
832
# line-delta is no use unless we have its parent.
833
# Fetching from a broken repository with this problem
834
# shouldn't break the target repository.
836
# See https://bugs.launchpad.net/bzr/+bug/164443
837
if not self._index.has_version(parents[0]):
840
'line-delta from stream '
843
'missing parent %s\n'
844
'Try running "bzr check" '
845
'on the source repository, and "bzr reconcile" '
847
(version_id, parents[0]))
848
self._add_raw_records(
849
[(version_id, options, parents, length)],
850
reader_callable(length))
852
def _knit_from_datastream(self, (format, data_list, reader_callable)):
853
"""Create a knit object from a data stream.
855
This method exists to allow conversion of data streams that do not
856
match the signature of this knit. Generally it will be slower and use
857
more memory to use this method to insert data, but it will work.
859
:seealso: get_data_stream for details on datastreams.
860
:return: A knit versioned file which can be used to join the datastream
863
if format == "knit-plain":
864
factory = KnitPlainFactory()
865
elif format == "knit-annotated":
866
factory = KnitAnnotateFactory()
868
raise errors.KnitDataStreamUnknown(format)
869
index = _StreamIndex(data_list, self._index)
870
access = _StreamAccess(reader_callable, index, self, factory)
871
return KnitVersionedFile(self.filename, self.transport,
872
factory=factory, index=index, access_method=access)
875
"""See VersionedFile.versions."""
876
if 'evil' in debug.debug_flags:
877
trace.mutter_callsite(2, "versions scales with size of history")
878
return self._index.get_versions()
880
def has_version(self, version_id):
881
"""See VersionedFile.has_version."""
882
if 'evil' in debug.debug_flags:
883
trace.mutter_callsite(2, "has_version is a LBYL scenario")
884
return self._index.has_version(version_id)
886
__contains__ = has_version
888
def _merge_annotations(self, content, parents, parent_texts={},
889
delta=None, annotated=None,
890
left_matching_blocks=None):
891
"""Merge annotations for content. This is done by comparing
892
the annotations based on changed to the text.
894
if left_matching_blocks is not None:
895
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
899
for parent_id in parents:
900
merge_content = self._get_content(parent_id, parent_texts)
901
if (parent_id == parents[0] and delta_seq is not None):
904
seq = patiencediff.PatienceSequenceMatcher(
905
None, merge_content.text(), content.text())
906
for i, j, n in seq.get_matching_blocks():
909
# this appears to copy (origin, text) pairs across to the
910
# new content for any line that matches the last-checked
912
content._lines[j:j+n] = merge_content._lines[i:i+n]
914
if delta_seq is None:
915
reference_content = self._get_content(parents[0], parent_texts)
916
new_texts = content.text()
917
old_texts = reference_content.text()
918
delta_seq = patiencediff.PatienceSequenceMatcher(
919
None, old_texts, new_texts)
920
return self._make_line_delta(delta_seq, content)
922
def _make_line_delta(self, delta_seq, new_content):
923
"""Generate a line delta from delta_seq and new_content."""
925
for op in delta_seq.get_opcodes():
928
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
931
def _get_components_positions(self, version_ids):
932
"""Produce a map of position data for the components of versions.
1143
def _build_details_to_components(self, build_details):
1144
"""Convert a build_details tuple to a position tuple."""
1145
# record_details, access_memo, compression_parent
1146
return build_details[3], build_details[0], build_details[1]
1148
def _get_components_positions(self, keys, allow_missing=False):
1149
"""Produce a map of position data for the components of keys.
934
1151
This data is intended to be used for retrieving the knit records.
936
A dict of version_id to (record_details, index_memo, next, parents) is
1153
A dict of key to (record_details, index_memo, next, parents) is
938
1155
method is the way referenced data should be applied.
939
1156
index_memo is the handle to pass to the data access to actually get the
941
1158
next is the build-parent of the version, or None for fulltexts.
942
1159
parents is the version_ids of the parents of this version
1161
:param allow_missing: If True do not raise an error on a missing component,
944
1164
component_data = {}
945
pending_components = version_ids
1165
pending_components = keys
946
1166
while pending_components:
947
1167
build_details = self._index.get_build_details(pending_components)
948
1168
current_components = set(pending_components)
949
1169
pending_components = set()
950
for version_id, details in build_details.iteritems():
1170
for key, details in build_details.iteritems():
951
1171
(index_memo, compression_parent, parents,
952
1172
record_details) = details
953
1173
method = record_details[0]
954
1174
if compression_parent is not None:
955
1175
pending_components.add(compression_parent)
956
component_data[version_id] = (record_details, index_memo,
1176
component_data[key] = self._build_details_to_components(details)
958
1177
missing = current_components.difference(build_details)
960
raise errors.RevisionNotPresent(missing.pop(), self.filename)
1178
if missing and not allow_missing:
1179
raise errors.RevisionNotPresent(missing.pop(), self)
961
1180
return component_data
963
def _get_content(self, version_id, parent_texts={}):
1182
def _get_content(self, key, parent_texts={}):
964
1183
"""Returns a content object that makes up the specified
966
cached_version = parent_texts.get(version_id, None)
1185
cached_version = parent_texts.get(key, None)
967
1186
if cached_version is not None:
968
if not self.has_version(version_id):
969
raise RevisionNotPresent(version_id, self.filename)
1187
# Ensure the cache dict is valid.
1188
if not self.get_parent_map([key]):
1189
raise RevisionNotPresent(key, self)
970
1190
return cached_version
972
text_map, contents_map = self._get_content_maps([version_id])
973
return contents_map[version_id]
975
def _check_versions_present(self, version_ids):
976
"""Check that all specified versions are present."""
977
self._index.check_versions_present(version_ids)
979
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
980
nostore_sha, random_id, check_content):
981
"""See VersionedFile.add_lines_with_ghosts()."""
982
self._check_add(version_id, lines, random_id, check_content)
983
return self._add(version_id, lines, parents, self.delta,
984
parent_texts, None, nostore_sha, random_id)
986
def _add_lines(self, version_id, parents, lines, parent_texts,
987
left_matching_blocks, nostore_sha, random_id, check_content):
988
"""See VersionedFile.add_lines."""
989
self._check_add(version_id, lines, random_id, check_content)
990
self._check_versions_present(parents)
991
return self._add(version_id, lines[:], parents, self.delta,
992
parent_texts, left_matching_blocks, nostore_sha, random_id)
994
def _check_add(self, version_id, lines, random_id, check_content):
995
"""check that version_id and lines are safe to add."""
996
if contains_whitespace(version_id):
997
raise InvalidRevisionId(version_id, self.filename)
998
self.check_not_reserved_id(version_id)
999
# Technically this could be avoided if we are happy to allow duplicate
1000
# id insertion when other things than bzr core insert texts, but it
1001
# seems useful for folk using the knit api directly to have some safety
1002
# blanket that we can disable.
1003
if not random_id and self.has_version(version_id):
1004
raise RevisionAlreadyPresent(version_id, self.filename)
1006
self._check_lines_not_unicode(lines)
1007
self._check_lines_are_lines(lines)
1009
def _add(self, version_id, lines, parents, delta, parent_texts,
1010
left_matching_blocks, nostore_sha, random_id):
1011
"""Add a set of lines on top of version specified by parents.
1013
If delta is true, compress the text as a line-delta against
1016
Any versions not present will be converted into ghosts.
1018
# first thing, if the content is something we don't need to store, find
1020
line_bytes = ''.join(lines)
1021
digest = sha_string(line_bytes)
1022
if nostore_sha == digest:
1023
raise errors.ExistingContent
1025
present_parents = []
1026
if parent_texts is None:
1028
for parent in parents:
1029
if self.has_version(parent):
1030
present_parents.append(parent)
1032
# can only compress against the left most present parent.
1034
(len(present_parents) == 0 or
1035
present_parents[0] != parents[0])):
1038
text_length = len(line_bytes)
1041
if lines[-1][-1] != '\n':
1042
# copy the contents of lines.
1044
options.append('no-eol')
1045
lines[-1] = lines[-1] + '\n'
1049
# To speed the extract of texts the delta chain is limited
1050
# to a fixed number of deltas. This should minimize both
1051
# I/O and the time spend applying deltas.
1052
delta = self._check_should_delta(present_parents)
1054
assert isinstance(version_id, str)
1055
content = self.factory.make(lines, version_id)
1056
if delta or (self.factory.annotated and len(present_parents) > 0):
1057
# Merge annotations from parent texts if needed.
1058
delta_hunks = self._merge_annotations(content, present_parents,
1059
parent_texts, delta, self.factory.annotated,
1060
left_matching_blocks)
1063
options.append('line-delta')
1064
store_lines = self.factory.lower_line_delta(delta_hunks)
1065
size, bytes = self._data._record_to_data(version_id, digest,
1068
options.append('fulltext')
1069
# isinstance is slower and we have no hierarchy.
1070
if self.factory.__class__ == KnitPlainFactory:
1071
# Use the already joined bytes saving iteration time in
1073
size, bytes = self._data._record_to_data(version_id, digest,
1074
lines, [line_bytes])
1076
# get mixed annotation + content and feed it into the
1078
store_lines = self.factory.lower_fulltext(content)
1079
size, bytes = self._data._record_to_data(version_id, digest,
1082
access_memo = self._data.add_raw_records([size], bytes)[0]
1083
self._index.add_versions(
1084
((version_id, options, access_memo, parents),),
1085
random_id=random_id)
1086
return digest, text_length, content
1088
def check(self, progress_bar=None):
1089
"""See VersionedFile.check()."""
1091
def _clone_text(self, new_version_id, old_version_id, parents):
1092
"""See VersionedFile.clone_text()."""
1093
# FIXME RBC 20060228 make fast by only inserting an index with null
1095
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
1097
def get_lines(self, version_id):
1098
"""See VersionedFile.get_lines()."""
1099
return self.get_line_list([version_id])[0]
1101
def _get_record_map(self, version_ids):
1191
generator = _VFContentMapGenerator(self, [key])
1192
return generator._get_content(key)
1194
def get_known_graph_ancestry(self, keys):
1195
"""Get a KnownGraph instance with the ancestry of keys."""
1196
parent_map, missing_keys = self._index.find_ancestry(keys)
1197
for fallback in self._fallback_vfs:
1198
if not missing_keys:
1200
(f_parent_map, f_missing_keys) = fallback._index.find_ancestry(
1202
parent_map.update(f_parent_map)
1203
missing_keys = f_missing_keys
1204
kg = _mod_graph.KnownGraph(parent_map)
1207
def get_parent_map(self, keys):
1208
"""Get a map of the graph parents of keys.
1210
:param keys: The keys to look up parents for.
1211
:return: A mapping from keys to parents. Absent keys are absent from
1214
return self._get_parent_map_with_sources(keys)[0]
1216
def _get_parent_map_with_sources(self, keys):
1217
"""Get a map of the parents of keys.
1219
:param keys: The keys to look up parents for.
1220
:return: A tuple. The first element is a mapping from keys to parents.
1221
Absent keys are absent from the mapping. The second element is a
1222
list with the locations each key was found in. The first element
1223
is the in-this-knit parents, the second the first fallback source,
1227
sources = [self._index] + self._fallback_vfs
1230
for source in sources:
1233
new_result = source.get_parent_map(missing)
1234
source_results.append(new_result)
1235
result.update(new_result)
1236
missing.difference_update(set(new_result))
1237
return result, source_results
1239
def _get_record_map(self, keys, allow_missing=False):
1102
1240
"""Produce a dictionary of knit records.
1104
:return: {version_id:(record, record_details, digest, next)}
1242
:return: {key:(record, record_details, digest, next)}
1106
data returned from read_records
1244
data returned from read_records (a KnitContentobject)
1108
1246
opaque information to pass to parse_record
1112
1250
build-parent of the version, i.e. the leftmost ancestor.
1113
1251
Will be None if the record is not a delta.
1115
position_map = self._get_components_positions(version_ids)
1116
# c = component_id, r = record_details, i_m = index_memo, n = next
1117
records = [(c, i_m) for c, (r, i_m, n)
1118
in position_map.iteritems()]
1120
for component_id, record, digest in \
1121
self._data.read_records_iter(records):
1122
(record_details, index_memo, next) = position_map[component_id]
1123
record_map[component_id] = record, record_details, digest, next
1127
def get_text(self, version_id):
1128
"""See VersionedFile.get_text"""
1129
return self.get_texts([version_id])[0]
1131
def get_texts(self, version_ids):
1132
return [''.join(l) for l in self.get_line_list(version_ids)]
1134
def get_line_list(self, version_ids):
1135
"""Return the texts of listed versions as a list of strings."""
1136
for version_id in version_ids:
1137
self.check_not_reserved_id(version_id)
1138
text_map, content_map = self._get_content_maps(version_ids)
1139
return [text_map[v] for v in version_ids]
1141
_get_lf_split_line_list = get_line_list
1143
def _get_content_maps(self, version_ids):
1144
"""Produce maps of text and KnitContents
1252
:param keys: The keys to build a map for
1253
:param allow_missing: If some records are missing, rather than
1254
error, just return the data that could be generated.
1256
raw_map = self._get_record_map_unparsed(keys,
1257
allow_missing=allow_missing)
1258
return self._raw_map_to_record_map(raw_map)
1260
def _raw_map_to_record_map(self, raw_map):
1261
"""Parse the contents of _get_record_map_unparsed.
1263
:return: see _get_record_map.
1267
data, record_details, next = raw_map[key]
1268
content, digest = self._parse_record(key[-1], data)
1269
result[key] = content, record_details, digest, next
1272
def _get_record_map_unparsed(self, keys, allow_missing=False):
1273
"""Get the raw data for reconstructing keys without parsing it.
1275
:return: A dict suitable for parsing via _raw_map_to_record_map.
1276
key-> raw_bytes, (method, noeol), compression_parent
1278
# This retries the whole request if anything fails. Potentially we
1279
# could be a bit more selective. We could track the keys whose records
1280
# we have successfully found, and then only request the new records
1281
# from there. However, _get_components_positions grabs the whole build
1282
# chain, which means we'll likely try to grab the same records again
1283
# anyway. Also, can the build chains change as part of a pack
1284
# operation? We wouldn't want to end up with a broken chain.
1287
position_map = self._get_components_positions(keys,
1288
allow_missing=allow_missing)
1289
# key = component_id, r = record_details, i_m = index_memo,
1291
records = [(key, i_m) for key, (r, i_m, n)
1292
in position_map.iteritems()]
1293
# Sort by the index memo, so that we request records from the
1294
# same pack file together, and in forward-sorted order
1295
records.sort(key=operator.itemgetter(1))
1297
for key, data in self._read_records_iter_unchecked(records):
1298
(record_details, index_memo, next) = position_map[key]
1299
raw_record_map[key] = data, record_details, next
1300
return raw_record_map
1301
except errors.RetryWithNewPacks, e:
1302
self._access.reload_or_raise(e)
1305
def _split_by_prefix(cls, keys):
1306
"""For the given keys, split them up based on their prefix.
1308
To keep memory pressure somewhat under control, split the
1309
requests back into per-file-id requests, otherwise "bzr co"
1310
extracts the full tree into memory before writing it to disk.
1311
This should be revisited if _get_content_maps() can ever cross
1314
The keys for a given file_id are kept in the same relative order.
1315
Ordering between file_ids is not, though prefix_order will return the
1316
order that the key was first seen.
1318
:param keys: An iterable of key tuples
1319
:return: (split_map, prefix_order)
1320
split_map A dictionary mapping prefix => keys
1321
prefix_order The order that we saw the various prefixes
1323
split_by_prefix = {}
1331
if prefix in split_by_prefix:
1332
split_by_prefix[prefix].append(key)
1334
split_by_prefix[prefix] = [key]
1335
prefix_order.append(prefix)
1336
return split_by_prefix, prefix_order
1338
def _group_keys_for_io(self, keys, non_local_keys, positions,
1339
_min_buffer_size=_STREAM_MIN_BUFFER_SIZE):
1340
"""For the given keys, group them into 'best-sized' requests.
1342
The idea is to avoid making 1 request per file, but to never try to
1343
unpack an entire 1.5GB source tree in a single pass. Also when
1344
possible, we should try to group requests to the same pack file
1347
:return: list of (keys, non_local) tuples that indicate what keys
1348
should be fetched next.
1350
# TODO: Ideally we would group on 2 factors. We want to extract texts
1351
# from the same pack file together, and we want to extract all
1352
# the texts for a given build-chain together. Ultimately it
1353
# probably needs a better global view.
1354
total_keys = len(keys)
1355
prefix_split_keys, prefix_order = self._split_by_prefix(keys)
1356
prefix_split_non_local_keys, _ = self._split_by_prefix(non_local_keys)
1358
cur_non_local = set()
1362
for prefix in prefix_order:
1363
keys = prefix_split_keys[prefix]
1364
non_local = prefix_split_non_local_keys.get(prefix, [])
1366
this_size = self._index._get_total_build_size(keys, positions)
1367
cur_size += this_size
1368
cur_keys.extend(keys)
1369
cur_non_local.update(non_local)
1370
if cur_size > _min_buffer_size:
1371
result.append((cur_keys, cur_non_local))
1372
sizes.append(cur_size)
1374
cur_non_local = set()
1377
result.append((cur_keys, cur_non_local))
1378
sizes.append(cur_size)
1381
def get_record_stream(self, keys, ordering, include_delta_closure):
1382
"""Get a stream of records for keys.
1384
:param keys: The keys to include.
1385
:param ordering: Either 'unordered' or 'topological'. A topologically
1386
sorted stream has compression parents strictly before their
1388
:param include_delta_closure: If True then the closure across any
1389
compression parents will be included (in the opaque data).
1390
:return: An iterator of ContentFactory objects, each of which is only
1391
valid until the iterator is advanced.
1393
# keys might be a generator
1397
if not self._index.has_graph:
1398
# Cannot sort when no graph has been stored.
1399
ordering = 'unordered'
1401
remaining_keys = keys
1404
keys = set(remaining_keys)
1405
for content_factory in self._get_remaining_record_stream(keys,
1406
ordering, include_delta_closure):
1407
remaining_keys.discard(content_factory.key)
1408
yield content_factory
1410
except errors.RetryWithNewPacks, e:
1411
self._access.reload_or_raise(e)
1413
def _get_remaining_record_stream(self, keys, ordering,
1414
include_delta_closure):
1415
"""This function is the 'retry' portion for get_record_stream."""
1416
if include_delta_closure:
1417
positions = self._get_components_positions(keys, allow_missing=True)
1419
build_details = self._index.get_build_details(keys)
1421
# (record_details, access_memo, compression_parent_key)
1422
positions = dict((key, self._build_details_to_components(details))
1423
for key, details in build_details.iteritems())
1424
absent_keys = keys.difference(set(positions))
1425
# There may be more absent keys : if we're missing the basis component
1426
# and are trying to include the delta closure.
1427
# XXX: We should not ever need to examine remote sources because we do
1428
# not permit deltas across versioned files boundaries.
1429
if include_delta_closure:
1430
needed_from_fallback = set()
1431
# Build up reconstructable_keys dict. key:True in this dict means
1432
# the key can be reconstructed.
1433
reconstructable_keys = {}
1437
chain = [key, positions[key][2]]
1439
needed_from_fallback.add(key)
1442
while chain[-1] is not None:
1443
if chain[-1] in reconstructable_keys:
1444
result = reconstructable_keys[chain[-1]]
1448
chain.append(positions[chain[-1]][2])
1450
# missing basis component
1451
needed_from_fallback.add(chain[-1])
1454
for chain_key in chain[:-1]:
1455
reconstructable_keys[chain_key] = result
1457
needed_from_fallback.add(key)
1458
# Double index lookups here : need a unified api ?
1459
global_map, parent_maps = self._get_parent_map_with_sources(keys)
1460
if ordering in ('topological', 'groupcompress'):
1461
if ordering == 'topological':
1462
# Global topological sort
1463
present_keys = tsort.topo_sort(global_map)
1465
present_keys = sort_groupcompress(global_map)
1466
# Now group by source:
1468
current_source = None
1469
for key in present_keys:
1470
for parent_map in parent_maps:
1471
if key in parent_map:
1472
key_source = parent_map
1474
if current_source is not key_source:
1475
source_keys.append((key_source, []))
1476
current_source = key_source
1477
source_keys[-1][1].append(key)
1479
if ordering != 'unordered':
1480
raise AssertionError('valid values for ordering are:'
1481
' "unordered", "groupcompress" or "topological" not: %r'
1483
# Just group by source; remote sources first.
1486
for parent_map in reversed(parent_maps):
1487
source_keys.append((parent_map, []))
1488
for key in parent_map:
1489
present_keys.append(key)
1490
source_keys[-1][1].append(key)
1491
# We have been requested to return these records in an order that
1492
# suits us. So we ask the index to give us an optimally sorted
1494
for source, sub_keys in source_keys:
1495
if source is parent_maps[0]:
1496
# Only sort the keys for this VF
1497
self._index._sort_keys_by_io(sub_keys, positions)
1498
absent_keys = keys - set(global_map)
1499
for key in absent_keys:
1500
yield AbsentContentFactory(key)
1501
# restrict our view to the keys we can answer.
1502
# XXX: Memory: TODO: batch data here to cap buffered data at (say) 1MB.
1503
# XXX: At that point we need to consider the impact of double reads by
1504
# utilising components multiple times.
1505
if include_delta_closure:
1506
# XXX: get_content_maps performs its own index queries; allow state
1508
non_local_keys = needed_from_fallback - absent_keys
1509
for keys, non_local_keys in self._group_keys_for_io(present_keys,
1512
generator = _VFContentMapGenerator(self, keys, non_local_keys,
1515
for record in generator.get_record_stream():
1518
for source, keys in source_keys:
1519
if source is parent_maps[0]:
1520
# this KnitVersionedFiles
1521
records = [(key, positions[key][1]) for key in keys]
1522
for key, raw_data in self._read_records_iter_unchecked(records):
1523
(record_details, index_memo, _) = positions[key]
1524
yield KnitContentFactory(key, global_map[key],
1525
record_details, None, raw_data, self._factory.annotated, None)
1527
vf = self._fallback_vfs[parent_maps.index(source) - 1]
1528
for record in vf.get_record_stream(keys, ordering,
1529
include_delta_closure):
1532
def get_sha1s(self, keys):
1533
"""See VersionedFiles.get_sha1s()."""
1535
record_map = self._get_record_map(missing, allow_missing=True)
1537
for key, details in record_map.iteritems():
1538
if key not in missing:
1540
# record entry 2 is the 'digest'.
1541
result[key] = details[2]
1542
missing.difference_update(set(result))
1543
for source in self._fallback_vfs:
1546
new_result = source.get_sha1s(missing)
1547
result.update(new_result)
1548
missing.difference_update(set(new_result))
1551
def insert_record_stream(self, stream):
1552
"""Insert a record stream into this container.
1554
:param stream: A stream of records to insert.
1556
:seealso VersionedFiles.get_record_stream:
1558
def get_adapter(adapter_key):
1560
return adapters[adapter_key]
1562
adapter_factory = adapter_registry.get(adapter_key)
1563
adapter = adapter_factory(self)
1564
adapters[adapter_key] = adapter
1567
if self._factory.annotated:
1568
# self is annotated, we need annotated knits to use directly.
1569
annotated = "annotated-"
1572
# self is not annotated, but we can strip annotations cheaply.
1574
convertibles = set(["knit-annotated-ft-gz"])
1575
if self._max_delta_chain:
1576
delta_types.add("knit-annotated-delta-gz")
1577
convertibles.add("knit-annotated-delta-gz")
1578
# The set of types we can cheaply adapt without needing basis texts.
1579
native_types = set()
1580
if self._max_delta_chain:
1581
native_types.add("knit-%sdelta-gz" % annotated)
1582
delta_types.add("knit-%sdelta-gz" % annotated)
1583
native_types.add("knit-%sft-gz" % annotated)
1584
knit_types = native_types.union(convertibles)
1586
# Buffer all index entries that we can't add immediately because their
1587
# basis parent is missing. We don't buffer all because generating
1588
# annotations may require access to some of the new records. However we
1589
# can't generate annotations from new deltas until their basis parent
1590
# is present anyway, so we get away with not needing an index that
1591
# includes the new keys.
1593
# See <http://launchpad.net/bugs/300177> about ordering of compression
1594
# parents in the records - to be conservative, we insist that all
1595
# parents must be present to avoid expanding to a fulltext.
1597
# key = basis_parent, value = index entry to add
1598
buffered_index_entries = {}
1599
for record in stream:
1600
kind = record.storage_kind
1601
if kind.startswith('knit-') and kind.endswith('-gz'):
1602
# Check that the ID in the header of the raw knit bytes matches
1603
# the record metadata.
1604
raw_data = record._raw_record
1605
df, rec = self._parse_record_header(record.key, raw_data)
1608
parents = record.parents
1609
if record.storage_kind in delta_types:
1610
# TODO: eventually the record itself should track
1611
# compression_parent
1612
compression_parent = parents[0]
1614
compression_parent = None
1615
# Raise an error when a record is missing.
1616
if record.storage_kind == 'absent':
1617
raise RevisionNotPresent([record.key], self)
1618
elif ((record.storage_kind in knit_types)
1619
and (compression_parent is None
1620
or not self._fallback_vfs
1621
or self._index.has_key(compression_parent)
1622
or not self.has_key(compression_parent))):
1623
# we can insert the knit record literally if either it has no
1624
# compression parent OR we already have its basis in this kvf
1625
# OR the basis is not present even in the fallbacks. In the
1626
# last case it will either turn up later in the stream and all
1627
# will be well, or it won't turn up at all and we'll raise an
1630
# TODO: self.has_key is somewhat redundant with
1631
# self._index.has_key; we really want something that directly
1632
# asks if it's only present in the fallbacks. -- mbp 20081119
1633
if record.storage_kind not in native_types:
1635
adapter_key = (record.storage_kind, "knit-delta-gz")
1636
adapter = get_adapter(adapter_key)
1638
adapter_key = (record.storage_kind, "knit-ft-gz")
1639
adapter = get_adapter(adapter_key)
1640
bytes = adapter.get_bytes(record)
1642
# It's a knit record, it has a _raw_record field (even if
1643
# it was reconstituted from a network stream).
1644
bytes = record._raw_record
1645
options = [record._build_details[0]]
1646
if record._build_details[1]:
1647
options.append('no-eol')
1648
# Just blat it across.
1649
# Note: This does end up adding data on duplicate keys. As
1650
# modern repositories use atomic insertions this should not
1651
# lead to excessive growth in the event of interrupted fetches.
1652
# 'knit' repositories may suffer excessive growth, but as a
1653
# deprecated format this is tolerable. It can be fixed if
1654
# needed by in the kndx index support raising on a duplicate
1655
# add with identical parents and options.
1656
access_memo = self._access.add_raw_records(
1657
[(record.key, len(bytes))], bytes)[0]
1658
index_entry = (record.key, options, access_memo, parents)
1659
if 'fulltext' not in options:
1660
# Not a fulltext, so we need to make sure the compression
1661
# parent will also be present.
1662
# Note that pack backed knits don't need to buffer here
1663
# because they buffer all writes to the transaction level,
1664
# but we don't expose that difference at the index level. If
1665
# the query here has sufficient cost to show up in
1666
# profiling we should do that.
1668
# They're required to be physically in this
1669
# KnitVersionedFiles, not in a fallback.
1670
if not self._index.has_key(compression_parent):
1671
pending = buffered_index_entries.setdefault(
1672
compression_parent, [])
1673
pending.append(index_entry)
1676
self._index.add_records([index_entry])
1677
elif record.storage_kind == 'chunked':
1678
self.add_lines(record.key, parents,
1679
osutils.chunks_to_lines(record.get_bytes_as('chunked')))
1681
# Not suitable for direct insertion as a
1682
# delta, either because it's not the right format, or this
1683
# KnitVersionedFiles doesn't permit deltas (_max_delta_chain ==
1684
# 0) or because it depends on a base only present in the
1686
self._access.flush()
1688
# Try getting a fulltext directly from the record.
1689
bytes = record.get_bytes_as('fulltext')
1690
except errors.UnavailableRepresentation:
1691
adapter_key = record.storage_kind, 'fulltext'
1692
adapter = get_adapter(adapter_key)
1693
bytes = adapter.get_bytes(record)
1694
lines = split_lines(bytes)
1696
self.add_lines(record.key, parents, lines)
1697
except errors.RevisionAlreadyPresent:
1699
# Add any records whose basis parent is now available.
1701
added_keys = [record.key]
1703
key = added_keys.pop(0)
1704
if key in buffered_index_entries:
1705
index_entries = buffered_index_entries[key]
1706
self._index.add_records(index_entries)
1708
[index_entry[0] for index_entry in index_entries])
1709
del buffered_index_entries[key]
1710
if buffered_index_entries:
1711
# There were index entries buffered at the end of the stream,
1712
# So these need to be added (if the index supports holding such
1713
# entries for later insertion)
1715
for key in buffered_index_entries:
1716
index_entries = buffered_index_entries[key]
1717
all_entries.extend(index_entries)
1718
self._index.add_records(
1719
all_entries, missing_compression_parents=True)
1721
def get_missing_compression_parent_keys(self):
1722
"""Return an iterable of keys of missing compression parents.
1724
Check this after calling insert_record_stream to find out if there are
1725
any missing compression parents. If there are, the records that
1726
depend on them are not able to be inserted safely. For atomic
1727
KnitVersionedFiles built on packs, the transaction should be aborted or
1728
suspended - commit will fail at this point. Nonatomic knits will error
1729
earlier because they have no staging area to put pending entries into.
1731
return self._index.get_missing_compression_parents()
1733
def iter_lines_added_or_present_in_keys(self, keys, pb=None):
1734
"""Iterate over the lines in the versioned files from keys.
1736
This may return lines from other keys. Each item the returned
1737
iterator yields is a tuple of a line and a text version that that line
1738
is present in (not introduced in).
1740
Ordering of results is in whatever order is most suitable for the
1741
underlying storage format.
1743
If a progress bar is supplied, it may be used to indicate progress.
1744
The caller is responsible for cleaning up progress bars (because this
1748
* Lines are normalised by the underlying store: they will all have \\n
1750
* Lines are returned in arbitrary order.
1751
* If a requested key did not change any lines (or didn't have any
1752
lines), it may not be mentioned at all in the result.
1754
:param pb: Progress bar supplied by caller.
1755
:return: An iterator over (line, key).
1758
pb = progress.DummyProgress()
1764
# we don't care about inclusions, the caller cares.
1765
# but we need to setup a list of records to visit.
1766
# we need key, position, length
1768
build_details = self._index.get_build_details(keys)
1769
for key, details in build_details.iteritems():
1771
key_records.append((key, details[0]))
1772
records_iter = enumerate(self._read_records_iter(key_records))
1773
for (key_idx, (key, data, sha_value)) in records_iter:
1774
pb.update('Walking content', key_idx, total)
1775
compression_parent = build_details[key][1]
1776
if compression_parent is None:
1778
line_iterator = self._factory.get_fulltext_content(data)
1781
line_iterator = self._factory.get_linedelta_content(data)
1782
# Now that we are yielding the data for this key, remove it
1785
# XXX: It might be more efficient to yield (key,
1786
# line_iterator) in the future. However for now, this is a
1787
# simpler change to integrate into the rest of the
1788
# codebase. RBC 20071110
1789
for line in line_iterator:
1792
except errors.RetryWithNewPacks, e:
1793
self._access.reload_or_raise(e)
1794
# If there are still keys we've not yet found, we look in the fallback
1795
# vfs, and hope to find them there. Note that if the keys are found
1796
# but had no changes or no content, the fallback may not return
1798
if keys and not self._fallback_vfs:
1799
# XXX: strictly the second parameter is meant to be the file id
1800
# but it's not easily accessible here.
1801
raise RevisionNotPresent(keys, repr(self))
1802
for source in self._fallback_vfs:
1806
for line, key in source.iter_lines_added_or_present_in_keys(keys):
1807
source_keys.add(key)
1809
keys.difference_update(source_keys)
1810
pb.update('Walking content', total, total)
1812
def _make_line_delta(self, delta_seq, new_content):
1813
"""Generate a line delta from delta_seq and new_content."""
1815
for op in delta_seq.get_opcodes():
1816
if op[0] == 'equal':
1818
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1821
def _merge_annotations(self, content, parents, parent_texts={},
1822
delta=None, annotated=None,
1823
left_matching_blocks=None):
1824
"""Merge annotations for content and generate deltas.
1826
This is done by comparing the annotations based on changes to the text
1827
and generating a delta on the resulting full texts. If annotations are
1828
not being created then a simple delta is created.
1830
if left_matching_blocks is not None:
1831
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1835
for parent_key in parents:
1836
merge_content = self._get_content(parent_key, parent_texts)
1837
if (parent_key == parents[0] and delta_seq is not None):
1840
seq = patiencediff.PatienceSequenceMatcher(
1841
None, merge_content.text(), content.text())
1842
for i, j, n in seq.get_matching_blocks():
1845
# this copies (origin, text) pairs across to the new
1846
# content for any line that matches the last-checked
1848
content._lines[j:j+n] = merge_content._lines[i:i+n]
1849
# XXX: Robert says the following block is a workaround for a
1850
# now-fixed bug and it can probably be deleted. -- mbp 20080618
1851
if content._lines and content._lines[-1][1][-1] != '\n':
1852
# The copied annotation was from a line without a trailing EOL,
1853
# reinstate one for the content object, to ensure correct
1855
line = content._lines[-1][1] + '\n'
1856
content._lines[-1] = (content._lines[-1][0], line)
1858
if delta_seq is None:
1859
reference_content = self._get_content(parents[0], parent_texts)
1860
new_texts = content.text()
1861
old_texts = reference_content.text()
1862
delta_seq = patiencediff.PatienceSequenceMatcher(
1863
None, old_texts, new_texts)
1864
return self._make_line_delta(delta_seq, content)
1866
def _parse_record(self, version_id, data):
1867
"""Parse an original format knit record.
1869
These have the last element of the key only present in the stored data.
1871
rec, record_contents = self._parse_record_unchecked(data)
1872
self._check_header_version(rec, version_id)
1873
return record_contents, rec[3]
1875
def _parse_record_header(self, key, raw_data):
1876
"""Parse a record header for consistency.
1878
:return: the header and the decompressor stream.
1879
as (stream, header_record)
1881
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(raw_data))
1884
rec = self._check_header(key, df.readline())
1885
except Exception, e:
1886
raise KnitCorrupt(self,
1887
"While reading {%s} got %s(%s)"
1888
% (key, e.__class__.__name__, str(e)))
1891
def _parse_record_unchecked(self, data):
1893
# 4168 calls in 2880 217 internal
1894
# 4168 calls to _parse_record_header in 2121
1895
# 4168 calls to readlines in 330
1896
df = tuned_gzip.GzipFile(mode='rb', fileobj=StringIO(data))
1898
record_contents = df.readlines()
1899
except Exception, e:
1900
raise KnitCorrupt(self, "Corrupt compressed record %r, got %s(%s)" %
1901
(data, e.__class__.__name__, str(e)))
1902
header = record_contents.pop(0)
1903
rec = self._split_header(header)
1904
last_line = record_contents.pop()
1905
if len(record_contents) != int(rec[2]):
1906
raise KnitCorrupt(self,
1907
'incorrect number of lines %s != %s'
1908
' for version {%s} %s'
1909
% (len(record_contents), int(rec[2]),
1910
rec[1], record_contents))
1911
if last_line != 'end %s\n' % rec[1]:
1912
raise KnitCorrupt(self,
1913
'unexpected version end line %r, wanted %r'
1914
% (last_line, rec[1]))
1916
return rec, record_contents
1918
def _read_records_iter(self, records):
1919
"""Read text records from data file and yield result.
1921
The result will be returned in whatever is the fastest to read.
1922
Not by the order requested. Also, multiple requests for the same
1923
record will only yield 1 response.
1924
:param records: A list of (key, access_memo) entries
1925
:return: Yields (key, contents, digest) in the order
1926
read, not the order requested
1931
# XXX: This smells wrong, IO may not be getting ordered right.
1932
needed_records = sorted(set(records), key=operator.itemgetter(1))
1933
if not needed_records:
1936
# The transport optimizes the fetching as well
1937
# (ie, reads continuous ranges.)
1938
raw_data = self._access.get_raw_records(
1939
[index_memo for key, index_memo in needed_records])
1941
for (key, index_memo), data in \
1942
izip(iter(needed_records), raw_data):
1943
content, digest = self._parse_record(key[-1], data)
1944
yield key, content, digest
1946
def _read_records_iter_raw(self, records):
1947
"""Read text records from data file and yield raw data.
1949
This unpacks enough of the text record to validate the id is
1950
as expected but thats all.
1952
Each item the iterator yields is (key, bytes,
1953
expected_sha1_of_full_text).
1955
for key, data in self._read_records_iter_unchecked(records):
1956
# validate the header (note that we can only use the suffix in
1957
# current knit records).
1958
df, rec = self._parse_record_header(key, data)
1960
yield key, data, rec[3]
1962
def _read_records_iter_unchecked(self, records):
1963
"""Read text records from data file and yield raw data.
1965
No validation is done.
1967
Yields tuples of (key, data).
1969
# setup an iterator of the external records:
1970
# uses readv so nice and fast we hope.
1972
# grab the disk data needed.
1973
needed_offsets = [index_memo for key, index_memo
1975
raw_records = self._access.get_raw_records(needed_offsets)
1977
for key, index_memo in records:
1978
data = raw_records.next()
1981
def _record_to_data(self, key, digest, lines, dense_lines=None):
1982
"""Convert key, digest, lines into a raw data block.
1984
:param key: The key of the record. Currently keys are always serialised
1985
using just the trailing component.
1986
:param dense_lines: The bytes of lines but in a denser form. For
1987
instance, if lines is a list of 1000 bytestrings each ending in \n,
1988
dense_lines may be a list with one line in it, containing all the
1989
1000's lines and their \n's. Using dense_lines if it is already
1990
known is a win because the string join to create bytes in this
1991
function spends less time resizing the final string.
1992
:return: (len, a StringIO instance with the raw data ready to read.)
1994
chunks = ["version %s %d %s\n" % (key[-1], len(lines), digest)]
1995
chunks.extend(dense_lines or lines)
1996
chunks.append("end %s\n" % key[-1])
1997
for chunk in chunks:
1998
if type(chunk) is not str:
1999
raise AssertionError(
2000
'data must be plain bytes was %s' % type(chunk))
2001
if lines and lines[-1][-1] != '\n':
2002
raise ValueError('corrupt lines value %r' % lines)
2003
compressed_bytes = tuned_gzip.chunks_to_gzip(chunks)
2004
return len(compressed_bytes), compressed_bytes
2006
def _split_header(self, line):
2009
raise KnitCorrupt(self,
2010
'unexpected number of elements in record header')
2014
"""See VersionedFiles.keys."""
2015
if 'evil' in debug.debug_flags:
2016
trace.mutter_callsite(2, "keys scales with size of history")
2017
sources = [self._index] + self._fallback_vfs
2019
for source in sources:
2020
result.update(source.keys())
2024
class _ContentMapGenerator(object):
2025
"""Generate texts or expose raw deltas for a set of texts."""
2027
def __init__(self, ordering='unordered'):
2028
self._ordering = ordering
2030
def _get_content(self, key):
2031
"""Get the content object for key."""
2032
# Note that _get_content is only called when the _ContentMapGenerator
2033
# has been constructed with just one key requested for reconstruction.
2034
if key in self.nonlocal_keys:
2035
record = self.get_record_stream().next()
2036
# Create a content object on the fly
2037
lines = osutils.chunks_to_lines(record.get_bytes_as('chunked'))
2038
return PlainKnitContent(lines, record.key)
2040
# local keys we can ask for directly
2041
return self._get_one_work(key)
2043
def get_record_stream(self):
2044
"""Get a record stream for the keys requested during __init__."""
2045
for record in self._work():
2049
"""Produce maps of text and KnitContents as dicts.
1146
2051
:return: (text_map, content_map) where text_map contains the texts for
1147
the requested versions and content_map contains the KnitContents.
1148
Both dicts take version_ids as their keys.
2052
the requested versions and content_map contains the KnitContents.
2054
# NB: By definition we never need to read remote sources unless texts
2055
# are requested from them: we don't delta across stores - and we
2056
# explicitly do not want to to prevent data loss situations.
2057
if self.global_map is None:
2058
self.global_map = self.vf.get_parent_map(self.keys)
2059
nonlocal_keys = self.nonlocal_keys
2061
missing_keys = set(nonlocal_keys)
2062
# Read from remote versioned file instances and provide to our caller.
2063
for source in self.vf._fallback_vfs:
2064
if not missing_keys:
2066
# Loop over fallback repositories asking them for texts - ignore
2067
# any missing from a particular fallback.
2068
for record in source.get_record_stream(missing_keys,
2069
self._ordering, True):
2070
if record.storage_kind == 'absent':
2071
# Not in thie particular stream, may be in one of the
2072
# other fallback vfs objects.
2074
missing_keys.remove(record.key)
2077
if self._raw_record_map is None:
2078
raise AssertionError('_raw_record_map should have been filled')
2080
for key in self.keys:
2081
if key in self.nonlocal_keys:
2083
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2086
def _get_one_work(self, requested_key):
2087
# Now, if we have calculated everything already, just return the
2089
if requested_key in self._contents_map:
2090
return self._contents_map[requested_key]
2091
# To simplify things, parse everything at once - code that wants one text
2092
# probably wants them all.
1150
2093
# FUTURE: This function could be improved for the 'extract many' case
1151
2094
# by tracking each component and only doing the copy when the number of
1152
2095
# children than need to apply delta's to it is > 1 or it is part of the
1153
2096
# final output.
1154
version_ids = list(version_ids)
1155
multiple_versions = len(version_ids) != 1
1156
record_map = self._get_record_map(version_ids)
1161
for version_id in version_ids:
2097
multiple_versions = len(self.keys) != 1
2098
if self._record_map is None:
2099
self._record_map = self.vf._raw_map_to_record_map(
2100
self._raw_record_map)
2101
record_map = self._record_map
2102
# raw_record_map is key:
2103
# Have read and parsed records at this point.
2104
for key in self.keys:
2105
if key in self.nonlocal_keys:
1162
2108
components = []
1164
2110
while cursor is not None:
1165
record, record_details, digest, next = record_map[cursor]
2112
record, record_details, digest, next = record_map[cursor]
2114
raise RevisionNotPresent(cursor, self)
1166
2115
components.append((cursor, record, record_details, digest))
1167
if cursor in content_map:
2117
if cursor in self._contents_map:
2118
# no need to plan further back
2119
components.append((cursor, None, None, None))
1172
2123
for (component_id, record, record_details,
1173
2124
digest) in reversed(components):
1174
if component_id in content_map:
1175
content = content_map[component_id]
2125
if component_id in self._contents_map:
2126
content = self._contents_map[component_id]
1177
content, delta = self.factory.parse_record(version_id,
2128
content, delta = self._factory.parse_record(key[-1],
1178
2129
record, record_details, content,
1179
2130
copy_base_content=multiple_versions)
1180
2131
if multiple_versions:
1181
content_map[component_id] = content
1183
content.cleanup_eol(copy_on_mutate=multiple_versions)
1184
final_content[version_id] = content
2132
self._contents_map[component_id] = content
1186
2134
# digest here is the digest from the last applied component.
1187
2135
text = content.text()
1188
2136
actual_sha = sha_strings(text)
1189
2137
if actual_sha != digest:
1190
raise KnitCorrupt(self.filename,
1192
'\n of reconstructed text does not match'
1194
'\n for version %s' %
1195
(actual_sha, digest, version_id))
1196
text_map[version_id] = text
1197
return text_map, final_content
1199
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1201
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1202
if version_ids is None:
1203
version_ids = self.versions()
1205
pb = progress.DummyProgress()
1206
# we don't care about inclusions, the caller cares.
1207
# but we need to setup a list of records to visit.
1208
# we need version_id, position, length
1209
version_id_records = []
1210
requested_versions = set(version_ids)
1211
# filter for available versions
1212
for version_id in requested_versions:
1213
if not self.has_version(version_id):
1214
raise RevisionNotPresent(version_id, self.filename)
1215
# get a in-component-order queue:
1216
for version_id in self.versions():
1217
if version_id in requested_versions:
1218
index_memo = self._index.get_position(version_id)
1219
version_id_records.append((version_id, index_memo))
1221
total = len(version_id_records)
1222
for version_idx, (version_id, data, sha_value) in \
1223
enumerate(self._data.read_records_iter(version_id_records)):
1224
pb.update('Walking content.', version_idx, total)
1225
method = self._index.get_method(version_id)
1227
assert method in ('fulltext', 'line-delta')
1228
if method == 'fulltext':
1229
line_iterator = self.factory.get_fulltext_content(data)
1231
line_iterator = self.factory.get_linedelta_content(data)
1232
# XXX: It might be more efficient to yield (version_id,
1233
# line_iterator) in the future. However for now, this is a simpler
1234
# change to integrate into the rest of the codebase. RBC 20071110
1235
for line in line_iterator:
1236
yield line, version_id
1238
pb.update('Walking content.', total, total)
1240
def iter_parents(self, version_ids):
1241
"""Iterate through the parents for many version ids.
1243
:param version_ids: An iterable yielding version_ids.
1244
:return: An iterator that yields (version_id, parents). Requested
1245
version_ids not present in the versioned file are simply skipped.
1246
The order is undefined, allowing for different optimisations in
1247
the underlying implementation.
1249
return self._index.iter_parents(version_ids)
1251
def num_versions(self):
1252
"""See VersionedFile.num_versions()."""
1253
return self._index.num_versions()
1255
__len__ = num_versions
1257
def annotate_iter(self, version_id):
1258
"""See VersionedFile.annotate_iter."""
1259
return self.factory.annotate_iter(self, version_id)
1261
def get_parents(self, version_id):
1262
"""See VersionedFile.get_parents."""
1265
# 52554 calls in 1264 872 internal down from 3674
1267
return self._index.get_parents(version_id)
1269
raise RevisionNotPresent(version_id, self.filename)
1271
def get_parents_with_ghosts(self, version_id):
1272
"""See VersionedFile.get_parents."""
1274
return self._index.get_parents_with_ghosts(version_id)
1276
raise RevisionNotPresent(version_id, self.filename)
1278
def get_ancestry(self, versions, topo_sorted=True):
1279
"""See VersionedFile.get_ancestry."""
1280
if isinstance(versions, basestring):
1281
versions = [versions]
1284
return self._index.get_ancestry(versions, topo_sorted)
1286
def get_ancestry_with_ghosts(self, versions):
1287
"""See VersionedFile.get_ancestry_with_ghosts."""
1288
if isinstance(versions, basestring):
1289
versions = [versions]
1292
return self._index.get_ancestry_with_ghosts(versions)
1294
def plan_merge(self, ver_a, ver_b):
1295
"""See VersionedFile.plan_merge."""
1296
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1297
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1298
annotated_a = self.annotate(ver_a)
1299
annotated_b = self.annotate(ver_b)
1300
return merge._plan_annotate_merge(annotated_a, annotated_b,
1301
ancestors_a, ancestors_b)
1304
class _KnitComponentFile(object):
1305
"""One of the files used to implement a knit database"""
1307
def __init__(self, transport, filename, mode, file_mode=None,
1308
create_parent_dir=False, dir_mode=None):
1309
self._transport = transport
1310
self._filename = filename
1312
self._file_mode = file_mode
1313
self._dir_mode = dir_mode
1314
self._create_parent_dir = create_parent_dir
1315
self._need_to_create = False
1317
def _full_path(self):
1318
"""Return the full path to this file."""
1319
return self._transport.base + self._filename
1321
def check_header(self, fp):
1322
line = fp.readline()
1324
# An empty file can actually be treated as though the file doesn't
1326
raise errors.NoSuchFile(self._full_path())
1327
if line != self.HEADER:
1328
raise KnitHeaderError(badline=line,
1329
filename=self._transport.abspath(self._filename))
1332
return '%s(%s)' % (self.__class__.__name__, self._filename)
1335
class _KnitIndex(_KnitComponentFile):
1336
"""Manages knit index file.
1338
The index is already kept in memory and read on startup, to enable
2138
raise SHA1KnitCorrupt(self, actual_sha, digest, key, text)
2139
if multiple_versions:
2140
return self._contents_map[requested_key]
2144
def _wire_bytes(self):
2145
"""Get the bytes to put on the wire for 'key'.
2147
The first collection of bytes asked for returns the serialised
2148
raw_record_map and the additional details (key, parent) for key.
2149
Subsequent calls return just the additional details (key, parent).
2150
The wire storage_kind given for the first key is 'knit-delta-closure',
2151
For subsequent keys it is 'knit-delta-closure-ref'.
2153
:param key: A key from the content generator.
2154
:return: Bytes to put on the wire.
2157
# kind marker for dispatch on the far side,
2158
lines.append('knit-delta-closure')
2160
if self.vf._factory.annotated:
2161
lines.append('annotated')
2164
# then the list of keys
2165
lines.append('\t'.join(['\x00'.join(key) for key in self.keys
2166
if key not in self.nonlocal_keys]))
2167
# then the _raw_record_map in serialised form:
2169
# for each item in the map:
2171
# 1 line with parents if the key is to be yielded (None: for None, '' for ())
2172
# one line with method
2173
# one line with noeol
2174
# one line with next ('' for None)
2175
# one line with byte count of the record bytes
2177
for key, (record_bytes, (method, noeol), next) in \
2178
self._raw_record_map.iteritems():
2179
key_bytes = '\x00'.join(key)
2180
parents = self.global_map.get(key, None)
2182
parent_bytes = 'None:'
2184
parent_bytes = '\t'.join('\x00'.join(key) for key in parents)
2185
method_bytes = method
2191
next_bytes = '\x00'.join(next)
2194
map_byte_list.append('%s\n%s\n%s\n%s\n%s\n%d\n%s' % (
2195
key_bytes, parent_bytes, method_bytes, noeol_bytes, next_bytes,
2196
len(record_bytes), record_bytes))
2197
map_bytes = ''.join(map_byte_list)
2198
lines.append(map_bytes)
2199
bytes = '\n'.join(lines)
2203
class _VFContentMapGenerator(_ContentMapGenerator):
2204
"""Content map generator reading from a VersionedFiles object."""
2206
def __init__(self, versioned_files, keys, nonlocal_keys=None,
2207
global_map=None, raw_record_map=None, ordering='unordered'):
2208
"""Create a _ContentMapGenerator.
2210
:param versioned_files: The versioned files that the texts are being
2212
:param keys: The keys to produce content maps for.
2213
:param nonlocal_keys: An iterable of keys(possibly intersecting keys)
2214
which are known to not be in this knit, but rather in one of the
2216
:param global_map: The result of get_parent_map(keys) (or a supermap).
2217
This is required if get_record_stream() is to be used.
2218
:param raw_record_map: A unparsed raw record map to use for answering
2221
_ContentMapGenerator.__init__(self, ordering=ordering)
2222
# The vf to source data from
2223
self.vf = versioned_files
2225
self.keys = list(keys)
2226
# Keys known to be in fallback vfs objects
2227
if nonlocal_keys is None:
2228
self.nonlocal_keys = set()
2230
self.nonlocal_keys = frozenset(nonlocal_keys)
2231
# Parents data for keys to be returned in get_record_stream
2232
self.global_map = global_map
2233
# The chunked lists for self.keys in text form
2235
# A cache of KnitContent objects used in extracting texts.
2236
self._contents_map = {}
2237
# All the knit records needed to assemble the requested keys as full
2239
self._record_map = None
2240
if raw_record_map is None:
2241
self._raw_record_map = self.vf._get_record_map_unparsed(keys,
2244
self._raw_record_map = raw_record_map
2245
# the factory for parsing records
2246
self._factory = self.vf._factory
2249
class _NetworkContentMapGenerator(_ContentMapGenerator):
2250
"""Content map generator sourced from a network stream."""
2252
def __init__(self, bytes, line_end):
2253
"""Construct a _NetworkContentMapGenerator from a bytes block."""
2255
self.global_map = {}
2256
self._raw_record_map = {}
2257
self._contents_map = {}
2258
self._record_map = None
2259
self.nonlocal_keys = []
2260
# Get access to record parsing facilities
2261
self.vf = KnitVersionedFiles(None, None)
2264
line_end = bytes.find('\n', start)
2265
line = bytes[start:line_end]
2266
start = line_end + 1
2267
if line == 'annotated':
2268
self._factory = KnitAnnotateFactory()
2270
self._factory = KnitPlainFactory()
2271
# list of keys to emit in get_record_stream
2272
line_end = bytes.find('\n', start)
2273
line = bytes[start:line_end]
2274
start = line_end + 1
2276
tuple(segment.split('\x00')) for segment in line.split('\t')
2278
# now a loop until the end. XXX: It would be nice if this was just a
2279
# bunch of the same records as get_record_stream(..., False) gives, but
2280
# there is a decent sized gap stopping that at the moment.
2284
line_end = bytes.find('\n', start)
2285
key = tuple(bytes[start:line_end].split('\x00'))
2286
start = line_end + 1
2287
# 1 line with parents (None: for None, '' for ())
2288
line_end = bytes.find('\n', start)
2289
line = bytes[start:line_end]
2294
[tuple(segment.split('\x00')) for segment in line.split('\t')
2296
self.global_map[key] = parents
2297
start = line_end + 1
2298
# one line with method
2299
line_end = bytes.find('\n', start)
2300
line = bytes[start:line_end]
2302
start = line_end + 1
2303
# one line with noeol
2304
line_end = bytes.find('\n', start)
2305
line = bytes[start:line_end]
2307
start = line_end + 1
2308
# one line with next ('' for None)
2309
line_end = bytes.find('\n', start)
2310
line = bytes[start:line_end]
2314
next = tuple(bytes[start:line_end].split('\x00'))
2315
start = line_end + 1
2316
# one line with byte count of the record bytes
2317
line_end = bytes.find('\n', start)
2318
line = bytes[start:line_end]
2320
start = line_end + 1
2322
record_bytes = bytes[start:start+count]
2323
start = start + count
2325
self._raw_record_map[key] = (record_bytes, (method, noeol), next)
2327
def get_record_stream(self):
2328
"""Get a record stream for for keys requested by the bytestream."""
2330
for key in self.keys:
2331
yield LazyKnitContentFactory(key, self.global_map[key], self, first)
2334
def _wire_bytes(self):
2338
class _KndxIndex(object):
2339
"""Manages knit index files
2341
The index is kept in memory and read on startup, to enable
1339
2342
fast lookups of revision information. The cursor of the index
1340
2343
file is always pointing to the end, making it easy to append
1503
2544
extra information about the content which needs to be passed to
1504
2545
Factory.parse_record
2547
parent_map = self.get_parent_map(keys)
1507
for version_id in version_ids:
1508
if version_id not in self._cache:
1509
# ghosts are omitted
1511
method = self.get_method(version_id)
1512
parents = self.get_parents_with_ghosts(version_id)
2550
if key not in parent_map:
2552
method = self.get_method(key)
2553
parents = parent_map[key]
1513
2554
if method == 'fulltext':
1514
2555
compression_parent = None
1516
2557
compression_parent = parents[0]
1517
noeol = 'no-eol' in self.get_options(version_id)
1518
index_memo = self.get_position(version_id)
1519
result[version_id] = (index_memo, compression_parent,
2558
noeol = 'no-eol' in self.get_options(key)
2559
index_memo = self.get_position(key)
2560
result[key] = (index_memo, compression_parent,
1520
2561
parents, (method, noeol))
1523
def iter_parents(self, version_ids):
1524
"""Iterate through the parents for many version ids.
1526
:param version_ids: An iterable yielding version_ids.
1527
:return: An iterator that yields (version_id, parents). Requested
1528
version_ids not present in the versioned file are simply skipped.
1529
The order is undefined, allowing for different optimisations in
1530
the underlying implementation.
1532
for version_id in version_ids:
1534
yield version_id, tuple(self.get_parents(version_id))
2564
def get_method(self, key):
2565
"""Return compression method of specified key."""
2566
options = self.get_options(key)
2567
if 'fulltext' in options:
2569
elif 'line-delta' in options:
2572
raise errors.KnitIndexUnknownMethod(self, options)
2574
def get_options(self, key):
2575
"""Return a list representing options.
2579
prefix, suffix = self._split_key(key)
2580
self._load_prefixes([prefix])
2582
return self._kndx_cache[prefix][0][suffix][1]
2584
raise RevisionNotPresent(key, self)
2586
def find_ancestry(self, keys):
2587
"""See CombinedGraphIndex.find_ancestry()"""
2588
prefixes = set(key[:-1] for key in keys)
2589
self._load_prefixes(prefixes)
2592
missing_keys = set()
2593
pending_keys = list(keys)
2594
# This assumes that keys will not reference parents in a different
2595
# prefix, which is accurate so far.
2597
key = pending_keys.pop()
2598
if key in parent_map:
2602
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
2604
missing_keys.add(key)
2606
parent_keys = tuple([prefix + (suffix,)
2607
for suffix in suffix_parents])
2608
parent_map[key] = parent_keys
2609
pending_keys.extend([p for p in parent_keys
2610
if p not in parent_map])
2611
return parent_map, missing_keys
2613
def get_parent_map(self, keys):
2614
"""Get a map of the parents of keys.
2616
:param keys: The keys to look up parents for.
2617
:return: A mapping from keys to parents. Absent keys are absent from
2620
# Parse what we need to up front, this potentially trades off I/O
2621
# locality (.kndx and .knit in the same block group for the same file
2622
# id) for less checking in inner loops.
2623
prefixes = set(key[:-1] for key in keys)
2624
self._load_prefixes(prefixes)
2629
suffix_parents = self._kndx_cache[prefix][0][key[-1]][4]
1535
2630
except KeyError:
1538
def num_versions(self):
1539
return len(self._history)
1541
__len__ = num_versions
1543
def get_versions(self):
1544
"""Get all the versions in the file. not topologically sorted."""
1545
return self._history
1547
def _version_list_to_index(self, versions):
2633
result[key] = tuple(prefix + (suffix,) for
2634
suffix in suffix_parents)
2637
def get_position(self, key):
2638
"""Return details needed to access the version.
2640
:return: a tuple (key, data position, size) to hand to the access
2641
logic to get the record.
2643
prefix, suffix = self._split_key(key)
2644
self._load_prefixes([prefix])
2645
entry = self._kndx_cache[prefix][0][suffix]
2646
return key, entry[2], entry[3]
2648
has_key = _mod_index._has_key_from_parent_map
2650
def _init_index(self, path, extra_lines=[]):
2651
"""Initialize an index."""
2653
sio.write(self.HEADER)
2654
sio.writelines(extra_lines)
2656
self._transport.put_file_non_atomic(path, sio,
2657
create_parent_dir=True)
2658
# self._create_parent_dir)
2659
# mode=self._file_mode,
2660
# dir_mode=self._dir_mode)
2663
"""Get all the keys in the collection.
2665
The keys are not ordered.
2668
# Identify all key prefixes.
2669
# XXX: A bit hacky, needs polish.
2670
if type(self._mapper) is ConstantMapper:
2674
for quoted_relpath in self._transport.iter_files_recursive():
2675
path, ext = os.path.splitext(quoted_relpath)
2677
prefixes = [self._mapper.unmap(path) for path in relpaths]
2678
self._load_prefixes(prefixes)
2679
for prefix in prefixes:
2680
for suffix in self._kndx_cache[prefix][1]:
2681
result.add(prefix + (suffix,))
2684
def _load_prefixes(self, prefixes):
2685
"""Load the indices for prefixes."""
2687
for prefix in prefixes:
2688
if prefix not in self._kndx_cache:
2689
# the load_data interface writes to these variables.
2692
self._filename = prefix
2694
path = self._mapper.map(prefix) + '.kndx'
2695
fp = self._transport.get(path)
2697
# _load_data may raise NoSuchFile if the target knit is
2699
_load_data(self, fp)
2702
self._kndx_cache[prefix] = (self._cache, self._history)
2707
self._kndx_cache[prefix] = ({}, [])
2708
if type(self._mapper) is ConstantMapper:
2709
# preserve behaviour for revisions.kndx etc.
2710
self._init_index(path)
2715
missing_keys = _mod_index._missing_keys_from_parent_map
2717
def _partition_keys(self, keys):
2718
"""Turn keys into a dict of prefix:suffix_list."""
2721
prefix_keys = result.setdefault(key[:-1], [])
2722
prefix_keys.append(key[-1])
2725
def _dictionary_compress(self, keys):
2726
"""Dictionary compress keys.
2728
:param keys: The keys to generate references to.
2729
:return: A string representation of keys. keys which are present are
2730
dictionary compressed, and others are emitted as fulltext with a
1548
2735
result_list = []
1550
for version in versions:
1551
if version in cache:
2736
prefix = keys[0][:-1]
2737
cache = self._kndx_cache[prefix][0]
2739
if key[:-1] != prefix:
2740
# kndx indices cannot refer across partitioned storage.
2741
raise ValueError("mismatched prefixes for %r" % keys)
2742
if key[-1] in cache:
1552
2743
# -- inlined lookup() --
1553
result_list.append(str(cache[version][5]))
2744
result_list.append(str(cache[key[-1]][5]))
1554
2745
# -- end lookup () --
1556
result_list.append('.' + version)
2747
result_list.append('.' + key[-1])
1557
2748
return ' '.join(result_list)
1559
def add_version(self, version_id, options, index_memo, parents):
1560
"""Add a version record to the index."""
1561
self.add_versions(((version_id, options, index_memo, parents),))
1563
def add_versions(self, versions, random_id=False):
1564
"""Add multiple versions to the index.
1566
:param versions: a list of tuples:
1567
(version_id, options, pos, size, parents).
1568
:param random_id: If True the ids being added were randomly generated
1569
and no check for existence will be performed.
1572
orig_history = self._history[:]
1573
orig_cache = self._cache.copy()
1576
for version_id, options, (index, pos, size), parents in versions:
1577
line = "\n%s %s %s %s %s :" % (version_id,
1581
self._version_list_to_index(parents))
1582
assert isinstance(line, str), \
1583
'content must be utf-8 encoded: %r' % (line,)
1585
self._cache_version(version_id, options, pos, size, parents)
1586
if not self._need_to_create:
1587
self._transport.append_bytes(self._filename, ''.join(lines))
1590
sio.write(self.HEADER)
1591
sio.writelines(lines)
1593
self._transport.put_file_non_atomic(self._filename, sio,
1594
create_parent_dir=self._create_parent_dir,
1595
mode=self._file_mode,
1596
dir_mode=self._dir_mode)
1597
self._need_to_create = False
1599
# If any problems happen, restore the original values and re-raise
1600
self._history = orig_history
1601
self._cache = orig_cache
1604
def has_version(self, version_id):
1605
"""True if the version is in the index."""
1606
return version_id in self._cache
1608
def get_position(self, version_id):
1609
"""Return details needed to access the version.
1611
.kndx indices do not support split-out data, so return None for the
1614
:return: a tuple (None, data position, size) to hand to the access
1615
logic to get the record.
1617
entry = self._cache[version_id]
1618
return None, entry[2], entry[3]
1620
def get_method(self, version_id):
1621
"""Return compression method of specified version."""
1623
options = self._cache[version_id][1]
2750
def _reset_cache(self):
2751
# Possibly this should be a LRU cache. A dictionary from key_prefix to
2752
# (cache_dict, history_vector) for parsed kndx files.
2753
self._kndx_cache = {}
2754
self._scope = self._get_scope()
2755
allow_writes = self._allow_writes()
2761
def _sort_keys_by_io(self, keys, positions):
2762
"""Figure out an optimal order to read the records for the given keys.
2764
Sort keys, grouped by index and sorted by position.
2766
:param keys: A list of keys whose records we want to read. This will be
2768
:param positions: A dict, such as the one returned by
2769
_get_components_positions()
2772
def get_sort_key(key):
2773
index_memo = positions[key][1]
2774
# Group by prefix and position. index_memo[0] is the key, so it is
2775
# (file_id, revision_id) and we don't want to sort on revision_id,
2776
# index_memo[1] is the position, and index_memo[2] is the size,
2777
# which doesn't matter for the sort
2778
return index_memo[0][:-1], index_memo[1]
2779
return keys.sort(key=get_sort_key)
2781
_get_total_build_size = _get_total_build_size
2783
def _split_key(self, key):
2784
"""Split key into a prefix and suffix."""
2785
return key[:-1], key[-1]
2788
class _KeyRefs(object):
2790
def __init__(self, track_new_keys=False):
2791
# dict mapping 'key' to 'set of keys referring to that key'
2794
# set remembering all new keys
2795
self.new_keys = set()
2797
self.new_keys = None
2803
self.new_keys.clear()
2805
def add_references(self, key, refs):
2806
# Record the new references
2807
for referenced in refs:
2809
needed_by = self.refs[referenced]
2811
needed_by = self.refs[referenced] = set()
2813
# Discard references satisfied by the new key
2816
def get_new_keys(self):
2817
return self.new_keys
2819
def get_unsatisfied_refs(self):
2820
return self.refs.iterkeys()
2822
def _satisfy_refs_for_key(self, key):
1624
2825
except KeyError:
1625
raise RevisionNotPresent(version_id, self._filename)
1626
if 'fulltext' in options:
1629
if 'line-delta' not in options:
1630
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1633
def get_options(self, version_id):
1634
"""Return a list representing options.
1638
return self._cache[version_id][1]
1640
def get_parents(self, version_id):
1641
"""Return parents of specified version ignoring ghosts."""
1642
return [parent for parent in self._cache[version_id][4]
1643
if parent in self._cache]
1645
def get_parents_with_ghosts(self, version_id):
1646
"""Return parents of specified version with ghosts."""
1647
return self._cache[version_id][4]
1649
def check_versions_present(self, version_ids):
1650
"""Check that all specified versions are present."""
1652
for version_id in version_ids:
1653
if version_id not in cache:
1654
raise RevisionNotPresent(version_id, self._filename)
1657
class KnitGraphIndex(object):
1658
"""A knit index that builds on GraphIndex."""
1660
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
2826
# No keys depended on this key. That's ok.
2829
def add_key(self, key):
2830
# satisfy refs for key, and remember that we've seen this key.
2831
self._satisfy_refs_for_key(key)
2832
if self.new_keys is not None:
2833
self.new_keys.add(key)
2835
def satisfy_refs_for_keys(self, keys):
2837
self._satisfy_refs_for_key(key)
2839
def get_referrers(self):
2841
for referrers in self.refs.itervalues():
2842
result.update(referrers)
2846
class _KnitGraphIndex(object):
2847
"""A KnitVersionedFiles index layered on GraphIndex."""
2849
def __init__(self, graph_index, is_locked, deltas=False, parents=True,
2850
add_callback=None, track_external_parent_refs=False):
1661
2851
"""Construct a KnitGraphIndex on a graph_index.
1663
2853
:param graph_index: An implementation of bzrlib.index.GraphIndex.
2854
:param is_locked: A callback to check whether the object should answer
1664
2856
:param deltas: Allow delta-compressed records.
2857
:param parents: If True, record knits parents, if not do not record
1665
2859
:param add_callback: If not None, allow additions to the index and call
1666
2860
this callback with a list of added GraphIndex nodes:
1667
2861
[(node, value, node_refs), ...]
1668
:param parents: If True, record knits parents, if not do not record
2862
:param is_locked: A callback, returns True if the index is locked and
2864
:param track_external_parent_refs: If True, record all external parent
2865
references parents from added records. These can be retrieved
2866
later by calling get_missing_parents().
2868
self._add_callback = add_callback
1671
2869
self._graph_index = graph_index
1672
2870
self._deltas = deltas
1673
self._add_callback = add_callback
1674
2871
self._parents = parents
1675
2872
if deltas and not parents:
2873
# XXX: TODO: Delta tree and parent graph should be conceptually
1676
2875
raise KnitCorrupt(self, "Cannot do delta compression without "
1677
2876
"parent tracking.")
1679
def _get_entries(self, keys, check_present=False):
1680
"""Get the entries for keys.
1682
:param keys: An iterable of index keys, - 1-tuples.
1687
for node in self._graph_index.iter_entries(keys):
1689
found_keys.add(node[1])
1691
# adapt parentless index to the rest of the code.
1692
for node in self._graph_index.iter_entries(keys):
1693
yield node[0], node[1], node[2], ()
1694
found_keys.add(node[1])
1696
missing_keys = keys.difference(found_keys)
1698
raise RevisionNotPresent(missing_keys.pop(), self)
1700
def _present_keys(self, version_ids):
1702
node[1] for node in self._get_entries(version_ids)])
1704
def _parentless_ancestry(self, versions):
1705
"""Honour the get_ancestry API for parentless knit indices."""
1706
wanted_keys = self._version_ids_to_keys(versions)
1707
present_keys = self._present_keys(wanted_keys)
1708
missing = set(wanted_keys).difference(present_keys)
1710
raise RevisionNotPresent(missing.pop(), self)
1711
return list(self._keys_to_version_ids(present_keys))
1713
def get_ancestry(self, versions, topo_sorted=True):
1714
"""See VersionedFile.get_ancestry."""
1715
if not self._parents:
1716
return self._parentless_ancestry(versions)
1717
# XXX: This will do len(history) index calls - perhaps
1718
# it should be altered to be a index core feature?
1719
# get a graph of all the mentioned versions:
1722
versions = self._version_ids_to_keys(versions)
1723
pending = set(versions)
1725
# get all pending nodes
1726
this_iteration = pending
1727
new_nodes = self._get_entries(this_iteration)
1730
for (index, key, value, node_refs) in new_nodes:
1731
# dont ask for ghosties - otherwise
1732
# we we can end up looping with pending
1733
# being entirely ghosted.
1734
graph[key] = [parent for parent in node_refs[0]
1735
if parent not in ghosts]
1737
for parent in graph[key]:
1738
# dont examine known nodes again
1743
ghosts.update(this_iteration.difference(found))
1744
if versions.difference(graph):
1745
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1747
result_keys = topo_sort(graph.items())
1749
result_keys = graph.iterkeys()
1750
return [key[0] for key in result_keys]
1752
def get_ancestry_with_ghosts(self, versions):
1753
"""See VersionedFile.get_ancestry."""
1754
if not self._parents:
1755
return self._parentless_ancestry(versions)
1756
# XXX: This will do len(history) index calls - perhaps
1757
# it should be altered to be a index core feature?
1758
# get a graph of all the mentioned versions:
1760
versions = self._version_ids_to_keys(versions)
1761
pending = set(versions)
1763
# get all pending nodes
1764
this_iteration = pending
1765
new_nodes = self._get_entries(this_iteration)
1767
for (index, key, value, node_refs) in new_nodes:
1768
graph[key] = node_refs[0]
1770
for parent in graph[key]:
1771
# dont examine known nodes again
1775
missing_versions = this_iteration.difference(graph)
1776
missing_needed = versions.intersection(missing_versions)
1778
raise RevisionNotPresent(missing_needed.pop(), self)
1779
for missing_version in missing_versions:
1780
# add a key, no parents
1781
graph[missing_version] = []
1782
pending.discard(missing_version) # don't look for it
1783
result_keys = topo_sort(graph.items())
1784
return [key[0] for key in result_keys]
1786
def get_build_details(self, version_ids):
1787
"""Get the method, index_memo and compression parent for version_ids.
1789
Ghosts are omitted from the result.
1791
:param version_ids: An iterable of version_ids.
1792
:return: A dict of version_id:(index_memo, compression_parent,
1793
parents, record_details).
1795
opaque structure to pass to read_records to extract the raw
1798
Content that this record is built upon, may be None
1800
Logical parents of this node
1802
extra information about the content which needs to be passed to
1803
Factory.parse_record
1806
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
1807
for entry in entries:
1808
version_id = self._keys_to_version_ids((entry[1],))[0]
1809
if not self._parents:
1812
parents = self._keys_to_version_ids(entry[3][0])
1813
if not self._deltas:
1814
compression_parent = None
1816
compression_parent_key = self._compression_parent(entry)
1817
if compression_parent_key:
1818
compression_parent = self._keys_to_version_ids(
1819
(compression_parent_key,))[0]
1821
compression_parent = None
1822
noeol = (entry[2][0] == 'N')
1823
if compression_parent:
1824
method = 'line-delta'
1827
result[version_id] = (self._node_to_position(entry),
1828
compression_parent, parents,
1832
def _compression_parent(self, an_entry):
1833
# return the key that an_entry is compressed against, or None
1834
# Grab the second parent list (as deltas implies parents currently)
1835
compression_parents = an_entry[3][1]
1836
if not compression_parents:
1838
assert len(compression_parents) == 1
1839
return compression_parents[0]
1841
def _get_method(self, node):
1842
if not self._deltas:
1844
if self._compression_parent(node):
1849
def get_graph(self):
1850
"""Return a list of the node:parents lists from this knit index."""
1851
if not self._parents:
1852
return [(key, ()) for key in self.get_versions()]
1854
for index, key, value, refs in self._graph_index.iter_all_entries():
1855
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1858
def iter_parents(self, version_ids):
1859
"""Iterate through the parents for many version ids.
1861
:param version_ids: An iterable yielding version_ids.
1862
:return: An iterator that yields (version_id, parents). Requested
1863
version_ids not present in the versioned file are simply skipped.
1864
The order is undefined, allowing for different optimisations in
1865
the underlying implementation.
1868
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1870
present_parents = set()
1871
for node in all_nodes:
1872
all_parents.update(node[3][0])
1873
# any node we are querying must be present
1874
present_parents.add(node[1])
1875
unknown_parents = all_parents.difference(present_parents)
1876
present_parents.update(self._present_keys(unknown_parents))
1877
for node in all_nodes:
1879
for parent in node[3][0]:
1880
if parent in present_parents:
1881
parents.append(parent[0])
1882
yield node[1][0], tuple(parents)
1884
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1885
yield node[1][0], ()
1887
def num_versions(self):
1888
return len(list(self._graph_index.iter_all_entries()))
1890
__len__ = num_versions
1892
def get_versions(self):
1893
"""Get all the versions in the file. not topologically sorted."""
1894
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1896
def has_version(self, version_id):
1897
"""True if the version is in the index."""
1898
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1900
def _keys_to_version_ids(self, keys):
1901
return tuple(key[0] for key in keys)
1903
def get_position(self, version_id):
1904
"""Return details needed to access the version.
1906
:return: a tuple (index, data position, size) to hand to the access
1907
logic to get the record.
1909
node = self._get_node(version_id)
1910
return self._node_to_position(node)
1912
def _node_to_position(self, node):
1913
"""Convert an index value to position details."""
1914
bits = node[2][1:].split(' ')
1915
return node[0], int(bits[0]), int(bits[1])
1917
def get_method(self, version_id):
1918
"""Return compression method of specified version."""
1919
return self._get_method(self._get_node(version_id))
1921
def _get_node(self, version_id):
1923
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1925
raise RevisionNotPresent(version_id, self)
1927
def get_options(self, version_id):
1928
"""Return a list representing options.
1932
node = self._get_node(version_id)
1933
options = [self._get_method(node)]
1934
if node[2][0] == 'N':
1935
options.append('no-eol')
1938
def get_parents(self, version_id):
1939
"""Return parents of specified version ignoring ghosts."""
1940
parents = list(self.iter_parents([version_id]))
1943
raise errors.RevisionNotPresent(version_id, self)
1944
return parents[0][1]
1946
def get_parents_with_ghosts(self, version_id):
1947
"""Return parents of specified version with ghosts."""
1948
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1949
check_present=True))
1950
if not self._parents:
1952
return self._keys_to_version_ids(nodes[0][3][0])
1954
def check_versions_present(self, version_ids):
1955
"""Check that all specified versions are present."""
1956
keys = self._version_ids_to_keys(version_ids)
1957
present = self._present_keys(keys)
1958
missing = keys.difference(present)
1960
raise RevisionNotPresent(missing.pop(), self)
1962
def add_version(self, version_id, options, access_memo, parents):
1963
"""Add a version record to the index."""
1964
return self.add_versions(((version_id, options, access_memo, parents),))
1966
def add_versions(self, versions, random_id=False):
1967
"""Add multiple versions to the index.
2877
self.has_graph = parents
2878
self._is_locked = is_locked
2879
self._missing_compression_parents = set()
2880
if track_external_parent_refs:
2881
self._key_dependencies = _KeyRefs()
2883
self._key_dependencies = None
2886
return "%s(%r)" % (self.__class__.__name__, self._graph_index)
2888
def add_records(self, records, random_id=False,
2889
missing_compression_parents=False):
2890
"""Add multiple records to the index.
1969
2892
This function does not insert data into the Immutable GraphIndex
1970
2893
backing the KnitGraphIndex, instead it prepares data for insertion by
1971
2894
the caller and checks that it is safe to insert then calls
1972
2895
self._add_callback with the prepared GraphIndex nodes.
1974
:param versions: a list of tuples:
1975
(version_id, options, pos, size, parents).
2897
:param records: a list of tuples:
2898
(key, options, access_memo, parents).
1976
2899
:param random_id: If True the ids being added were randomly generated
1977
2900
and no check for existence will be performed.
2901
:param missing_compression_parents: If True the records being added are
2902
only compressed against texts already in the index (or inside
2903
records). If False the records all refer to unavailable texts (or
2904
texts inside records) as compression parents.
1979
2906
if not self._add_callback:
1980
2907
raise errors.ReadOnlyError(self)
1981
2908
# we hope there are no repositories with inconsistent parentage
1986
for (version_id, options, access_memo, parents) in versions:
2912
compression_parents = set()
2913
key_dependencies = self._key_dependencies
2914
for (key, options, access_memo, parents) in records:
2916
parents = tuple(parents)
2917
if key_dependencies is not None:
2918
key_dependencies.add_references(key, parents)
1987
2919
index, pos, size = access_memo
1988
key = (version_id, )
1989
parents = tuple((parent, ) for parent in parents)
1990
2920
if 'no-eol' in options:
2024
2963
for key, (value, node_refs) in keys.iteritems():
2025
2964
result.append((key, value))
2026
2965
self._add_callback(result)
2028
def _version_ids_to_keys(self, version_ids):
2029
return set((version_id, ) for version_id in version_ids)
2032
class _KnitAccess(object):
2033
"""Access to knit records in a .knit file."""
2035
def __init__(self, transport, filename, _file_mode, _dir_mode,
2036
_need_to_create, _create_parent_dir):
2037
"""Create a _KnitAccess for accessing and inserting data.
2039
:param transport: The transport the .knit is located on.
2040
:param filename: The filename of the .knit.
2966
if missing_compression_parents:
2967
# This may appear to be incorrect (it does not check for
2968
# compression parents that are in the existing graph index),
2969
# but such records won't have been buffered, so this is
2970
# actually correct: every entry when
2971
# missing_compression_parents==True either has a missing parent, or
2972
# a parent that is one of the keys in records.
2973
compression_parents.difference_update(keys)
2974
self._missing_compression_parents.update(compression_parents)
2975
# Adding records may have satisfied missing compression parents.
2976
self._missing_compression_parents.difference_update(keys)
2978
def scan_unvalidated_index(self, graph_index):
2979
"""Inform this _KnitGraphIndex that there is an unvalidated index.
2981
This allows this _KnitGraphIndex to keep track of any missing
2982
compression parents we may want to have filled in to make those
2985
:param graph_index: A GraphIndex
2988
new_missing = graph_index.external_references(ref_list_num=1)
2989
new_missing.difference_update(self.get_parent_map(new_missing))
2990
self._missing_compression_parents.update(new_missing)
2991
if self._key_dependencies is not None:
2992
# Add parent refs from graph_index (and discard parent refs that
2993
# the graph_index has).
2994
for node in graph_index.iter_all_entries():
2995
self._key_dependencies.add_references(node[1], node[3][0])
2997
def get_missing_compression_parents(self):
2998
"""Return the keys of missing compression parents.
3000
Missing compression parents occur when a record stream was missing
3001
basis texts, or a index was scanned that had missing basis texts.
3003
return frozenset(self._missing_compression_parents)
3005
def get_missing_parents(self):
3006
"""Return the keys of missing parents."""
3007
# If updating this, you should also update
3008
# groupcompress._GCGraphIndex.get_missing_parents
3009
# We may have false positives, so filter those out.
3010
self._key_dependencies.satisfy_refs_for_keys(
3011
self.get_parent_map(self._key_dependencies.get_unsatisfied_refs()))
3012
return frozenset(self._key_dependencies.get_unsatisfied_refs())
3014
def _check_read(self):
3015
"""raise if reads are not permitted."""
3016
if not self._is_locked():
3017
raise errors.ObjectNotLocked(self)
3019
def _check_write_ok(self):
3020
"""Assert if writes are not permitted."""
3021
if not self._is_locked():
3022
raise errors.ObjectNotLocked(self)
3024
def _compression_parent(self, an_entry):
3025
# return the key that an_entry is compressed against, or None
3026
# Grab the second parent list (as deltas implies parents currently)
3027
compression_parents = an_entry[3][1]
3028
if not compression_parents:
3030
if len(compression_parents) != 1:
3031
raise AssertionError(
3032
"Too many compression parents: %r" % compression_parents)
3033
return compression_parents[0]
3035
def get_build_details(self, keys):
3036
"""Get the method, index_memo and compression parent for version_ids.
3038
Ghosts are omitted from the result.
3040
:param keys: An iterable of keys.
3041
:return: A dict of key:
3042
(index_memo, compression_parent, parents, record_details).
3044
opaque structure to pass to read_records to extract the raw
3047
Content that this record is built upon, may be None
3049
Logical parents of this node
3051
extra information about the content which needs to be passed to
3052
Factory.parse_record
3056
entries = self._get_entries(keys, False)
3057
for entry in entries:
3059
if not self._parents:
3062
parents = entry[3][0]
3063
if not self._deltas:
3064
compression_parent_key = None
3066
compression_parent_key = self._compression_parent(entry)
3067
noeol = (entry[2][0] == 'N')
3068
if compression_parent_key:
3069
method = 'line-delta'
3072
result[key] = (self._node_to_position(entry),
3073
compression_parent_key, parents,
3077
def _get_entries(self, keys, check_present=False):
3078
"""Get the entries for keys.
3080
:param keys: An iterable of index key tuples.
3085
for node in self._graph_index.iter_entries(keys):
3087
found_keys.add(node[1])
3089
# adapt parentless index to the rest of the code.
3090
for node in self._graph_index.iter_entries(keys):
3091
yield node[0], node[1], node[2], ()
3092
found_keys.add(node[1])
3094
missing_keys = keys.difference(found_keys)
3096
raise RevisionNotPresent(missing_keys.pop(), self)
3098
def get_method(self, key):
3099
"""Return compression method of specified key."""
3100
return self._get_method(self._get_node(key))
3102
def _get_method(self, node):
3103
if not self._deltas:
3105
if self._compression_parent(node):
3110
def _get_node(self, key):
3112
return list(self._get_entries([key]))[0]
3114
raise RevisionNotPresent(key, self)
3116
def get_options(self, key):
3117
"""Return a list representing options.
3121
node = self._get_node(key)
3122
options = [self._get_method(node)]
3123
if node[2][0] == 'N':
3124
options.append('no-eol')
3127
def find_ancestry(self, keys):
3128
"""See CombinedGraphIndex.find_ancestry()"""
3129
return self._graph_index.find_ancestry(keys, 0)
3131
def get_parent_map(self, keys):
3132
"""Get a map of the parents of keys.
3134
:param keys: The keys to look up parents for.
3135
:return: A mapping from keys to parents. Absent keys are absent from
3139
nodes = self._get_entries(keys)
3143
result[node[1]] = node[3][0]
3146
result[node[1]] = None
3149
def get_position(self, key):
3150
"""Return details needed to access the version.
3152
:return: a tuple (index, data position, size) to hand to the access
3153
logic to get the record.
3155
node = self._get_node(key)
3156
return self._node_to_position(node)
3158
has_key = _mod_index._has_key_from_parent_map
3161
"""Get all the keys in the collection.
3163
The keys are not ordered.
3166
return [node[1] for node in self._graph_index.iter_all_entries()]
3168
missing_keys = _mod_index._missing_keys_from_parent_map
3170
def _node_to_position(self, node):
3171
"""Convert an index value to position details."""
3172
bits = node[2][1:].split(' ')
3173
return node[0], int(bits[0]), int(bits[1])
3175
def _sort_keys_by_io(self, keys, positions):
3176
"""Figure out an optimal order to read the records for the given keys.
3178
Sort keys, grouped by index and sorted by position.
3180
:param keys: A list of keys whose records we want to read. This will be
3182
:param positions: A dict, such as the one returned by
3183
_get_components_positions()
3186
def get_index_memo(key):
3187
# index_memo is at offset [1]. It is made up of (GraphIndex,
3188
# position, size). GI is an object, which will be unique for each
3189
# pack file. This causes us to group by pack file, then sort by
3190
# position. Size doesn't matter, but it isn't worth breaking up the
3192
return positions[key][1]
3193
return keys.sort(key=get_index_memo)
3195
_get_total_build_size = _get_total_build_size
3198
class _KnitKeyAccess(object):
3199
"""Access to records in .knit files."""
3201
def __init__(self, transport, mapper):
3202
"""Create a _KnitKeyAccess with transport and mapper.
3204
:param transport: The transport the access object is rooted at.
3205
:param mapper: The mapper used to map keys to .knit files.
2042
3207
self._transport = transport
2043
self._filename = filename
2044
self._file_mode = _file_mode
2045
self._dir_mode = _dir_mode
2046
self._need_to_create = _need_to_create
2047
self._create_parent_dir = _create_parent_dir
3208
self._mapper = mapper
2049
def add_raw_records(self, sizes, raw_data):
3210
def add_raw_records(self, key_sizes, raw_data):
2050
3211
"""Add raw knit bytes to a storage area.
2052
The data is spooled to whereever the access method is storing data.
3213
The data is spooled to the container writer in one bytes-record per
2054
:param sizes: An iterable containing the size of each raw data segment.
3216
:param sizes: An iterable of tuples containing the key and size of each
2055
3218
:param raw_data: A bytestring containing the data.
2056
:return: A list of memos to retrieve the record later. Each memo is a
2057
tuple - (index, pos, length), where the index field is always None
2058
for the .knit access method.
3219
:return: A list of memos to retrieve the record later. Each memo is an
3220
opaque index memo. For _KnitKeyAccess the memo is (key, pos,
3221
length), where the key is the record key.
2060
assert type(raw_data) == str, \
2061
'data must be plain bytes was %s' % type(raw_data)
2062
if not self._need_to_create:
2063
base = self._transport.append_bytes(self._filename, raw_data)
2065
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2066
create_parent_dir=self._create_parent_dir,
2067
mode=self._file_mode,
2068
dir_mode=self._dir_mode)
2069
self._need_to_create = False
3223
if type(raw_data) is not str:
3224
raise AssertionError(
3225
'data must be plain bytes was %s' % type(raw_data))
2073
result.append((None, base, size))
3228
# TODO: This can be tuned for writing to sftp and other servers where
3229
# append() is relatively expensive by grouping the writes to each key
3231
for key, size in key_sizes:
3232
path = self._mapper.map(key)
3234
base = self._transport.append_bytes(path + '.knit',
3235
raw_data[offset:offset+size])
3236
except errors.NoSuchFile:
3237
self._transport.mkdir(osutils.dirname(path))
3238
base = self._transport.append_bytes(path + '.knit',
3239
raw_data[offset:offset+size])
3243
result.append((key, base, size))
2078
"""IFF this data access has its own storage area, initialise it.
2082
self._transport.put_bytes_non_atomic(self._filename, '',
2083
mode=self._file_mode)
2085
def open_file(self):
2086
"""IFF this data access can be represented as a single file, open it.
2088
For knits that are not mapped to a single file on disk this will
2091
:return: None or a file handle.
2094
return self._transport.get(self._filename)
3247
"""Flush pending writes on this access object.
3249
For .knit files this is a no-op.
2099
3253
def get_raw_records(self, memos_for_retrieval):
2100
3254
"""Get the raw bytes for a records.
2102
:param memos_for_retrieval: An iterable containing the (index, pos,
2103
length) memo for retrieving the bytes. The .knit method ignores
2104
the index as there is always only a single file.
3256
:param memos_for_retrieval: An iterable containing the access memo for
3257
retrieving the bytes.
2105
3258
:return: An iterator over the bytes of the records.
2107
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2108
for pos, data in self._transport.readv(self._filename, read_vector):
2112
class _PackAccess(object):
2113
"""Access to knit records via a collection of packs."""
2115
def __init__(self, index_to_packs, writer=None):
2116
"""Create a _PackAccess object.
3260
# first pass, group into same-index request to minimise readv's issued.
3262
current_prefix = None
3263
for (key, offset, length) in memos_for_retrieval:
3264
if current_prefix == key[:-1]:
3265
current_list.append((offset, length))
3267
if current_prefix is not None:
3268
request_lists.append((current_prefix, current_list))
3269
current_prefix = key[:-1]
3270
current_list = [(offset, length)]
3271
# handle the last entry
3272
if current_prefix is not None:
3273
request_lists.append((current_prefix, current_list))
3274
for prefix, read_vector in request_lists:
3275
path = self._mapper.map(prefix) + '.knit'
3276
for pos, data in self._transport.readv(path, read_vector):
3280
class _DirectPackAccess(object):
3281
"""Access to data in one or more packs with less translation."""
3283
def __init__(self, index_to_packs, reload_func=None, flush_func=None):
3284
"""Create a _DirectPackAccess object.
2118
3286
:param index_to_packs: A dict mapping index objects to the transport
2119
3287
and file names for obtaining data.
2120
:param writer: A tuple (pack.ContainerWriter, write_index) which
2121
contains the pack to write, and the index that reads from it will
3288
:param reload_func: A function to call if we determine that the pack
3289
files have moved and we need to reload our caches. See
3290
bzrlib.repo_fmt.pack_repo.AggregateIndex for more details.
2125
self.container_writer = writer[0]
2126
self.write_index = writer[1]
2128
self.container_writer = None
2129
self.write_index = None
2130
self.indices = index_to_packs
3292
self._container_writer = None
3293
self._write_index = None
3294
self._indices = index_to_packs
3295
self._reload_func = reload_func
3296
self._flush_func = flush_func
2132
def add_raw_records(self, sizes, raw_data):
3298
def add_raw_records(self, key_sizes, raw_data):
2133
3299
"""Add raw knit bytes to a storage area.
2135
3301
The data is spooled to the container writer in one bytes-record per
2138
:param sizes: An iterable containing the size of each raw data segment.
3304
:param sizes: An iterable of tuples containing the key and size of each
2139
3306
:param raw_data: A bytestring containing the data.
2140
:return: A list of memos to retrieve the record later. Each memo is a
2141
tuple - (index, pos, length), where the index field is the
2142
write_index object supplied to the PackAccess object.
3307
:return: A list of memos to retrieve the record later. Each memo is an
3308
opaque index memo. For _DirectPackAccess the memo is (index, pos,
3309
length), where the index field is the write_index object supplied
3310
to the PackAccess object.
2144
assert type(raw_data) == str, \
2145
'data must be plain bytes was %s' % type(raw_data)
3312
if type(raw_data) is not str:
3313
raise AssertionError(
3314
'data must be plain bytes was %s' % type(raw_data))
2149
p_offset, p_length = self.container_writer.add_bytes_record(
3317
for key, size in key_sizes:
3318
p_offset, p_length = self._container_writer.add_bytes_record(
2150
3319
raw_data[offset:offset+size], [])
2152
result.append((self.write_index, p_offset, p_length))
3321
result.append((self._write_index, p_offset, p_length))
2156
"""Pack based knits do not get individually created."""
3325
"""Flush pending writes on this access object.
3327
This will flush any buffered writes to a NewPack.
3329
if self._flush_func is not None:
2158
3332
def get_raw_records(self, memos_for_retrieval):
2159
3333
"""Get the raw bytes for a records.
2161
:param memos_for_retrieval: An iterable containing the (index, pos,
3335
:param memos_for_retrieval: An iterable containing the (index, pos,
2162
3336
length) memo for retrieving the bytes. The Pack access method
2163
3337
looks up the pack to use for a given record in its index_to_pack
2179
3353
if current_index is not None:
2180
3354
request_lists.append((current_index, current_list))
2181
3355
for index, offsets in request_lists:
2182
transport, path = self.indices[index]
2183
reader = pack.make_readv_reader(transport, path, offsets)
2184
for names, read_func in reader.iter_records():
2185
yield read_func(None)
2187
def open_file(self):
2188
"""Pack based knits have no single file."""
2191
def set_writer(self, writer, index, (transport, packname)):
3357
transport, path = self._indices[index]
3359
# A KeyError here indicates that someone has triggered an index
3360
# reload, and this index has gone missing, we need to start
3362
if self._reload_func is None:
3363
# If we don't have a _reload_func there is nothing that can
3366
raise errors.RetryWithNewPacks(index,
3367
reload_occurred=True,
3368
exc_info=sys.exc_info())
3370
reader = pack.make_readv_reader(transport, path, offsets)
3371
for names, read_func in reader.iter_records():
3372
yield read_func(None)
3373
except errors.NoSuchFile:
3374
# A NoSuchFile error indicates that a pack file has gone
3375
# missing on disk, we need to trigger a reload, and start over.
3376
if self._reload_func is None:
3378
raise errors.RetryWithNewPacks(transport.abspath(path),
3379
reload_occurred=False,
3380
exc_info=sys.exc_info())
3382
def set_writer(self, writer, index, transport_packname):
2192
3383
"""Set a writer to use for adding data."""
2193
3384
if index is not None:
2194
self.indices[index] = (transport, packname)
2195
self.container_writer = writer
2196
self.write_index = index
2199
class _StreamAccess(object):
2200
"""A Knit Access object that provides data from a datastream.
2202
It also provides a fallback to present as unannotated data, annotated data
2203
from a *backing* access object.
2205
This is triggered by a index_memo which is pointing to a different index
2206
than this was constructed with, and is used to allow extracting full
2207
unannotated texts for insertion into annotated knits.
2210
def __init__(self, reader_callable, stream_index, backing_knit,
2212
"""Create a _StreamAccess object.
2214
:param reader_callable: The reader_callable from the datastream.
2215
This is called to buffer all the data immediately, for
2217
:param stream_index: The index the data stream this provides access to
2218
which will be present in native index_memo's.
2219
:param backing_knit: The knit object that will provide access to
2220
annotated texts which are not available in the stream, so as to
2221
create unannotated texts.
2222
:param orig_factory: The original content factory used to generate the
2223
stream. This is used for checking whether the thunk code for
2224
supporting _copy_texts will generate the correct form of data.
2226
self.data = reader_callable(None)
2227
self.stream_index = stream_index
2228
self.backing_knit = backing_knit
2229
self.orig_factory = orig_factory
2231
def get_raw_records(self, memos_for_retrieval):
2232
"""Get the raw bytes for a records.
2234
:param memos_for_retrieval: An iterable containing the (thunk_flag,
2235
index, start, end) memo for retrieving the bytes.
2236
:return: An iterator over the bytes of the records.
2238
# use a generator for memory friendliness
2239
for thunk_flag, version_id, start, end in memos_for_retrieval:
2240
if version_id is self.stream_index:
2241
yield self.data[start:end]
2243
# we have been asked to thunk. This thunking only occurs when
2244
# we are obtaining plain texts from an annotated backing knit
2245
# so that _copy_texts will work.
2246
# We could improve performance here by scanning for where we need
2247
# to do this and using get_line_list, then interleaving the output
2248
# as desired. However, for now, this is sufficient.
2249
if self.orig_factory.__class__ != KnitPlainFactory:
2250
raise errors.KnitCorrupt(
2251
self, 'Bad thunk request %r' % version_id)
2252
lines = self.backing_knit.get_lines(version_id)
2253
line_bytes = ''.join(lines)
2254
digest = sha_string(line_bytes)
2256
if lines[-1][-1] != '\n':
2257
lines[-1] = lines[-1] + '\n'
2259
orig_options = list(self.backing_knit._index.get_options(version_id))
2260
if 'fulltext' not in orig_options:
2261
if 'line-delta' not in orig_options:
2262
raise errors.KnitCorrupt(self,
2263
'Unknown compression method %r' % orig_options)
2264
orig_options.remove('line-delta')
2265
orig_options.append('fulltext')
2266
# We want plain data, because we expect to thunk only to allow text
2268
size, bytes = self.backing_knit._data._record_to_data(version_id,
2269
digest, lines, line_bytes)
2273
class _StreamIndex(object):
2274
"""A Knit Index object that uses the data map from a datastream."""
2276
def __init__(self, data_list, backing_index):
2277
"""Create a _StreamIndex object.
2279
:param data_list: The data_list from the datastream.
2280
:param backing_index: The index which will supply values for nodes
2281
referenced outside of this stream.
2283
self.data_list = data_list
2284
self.backing_index = backing_index
2285
self._by_version = {}
2287
for key, options, length, parents in data_list:
2288
self._by_version[key] = options, (pos, pos + length), parents
2291
def get_ancestry(self, versions, topo_sorted):
2292
"""Get an ancestry list for versions."""
2294
# Not needed for basic joins
2295
raise NotImplementedError(self.get_ancestry)
2296
# get a graph of all the mentioned versions:
2297
# Little ugly - basically copied from KnitIndex, but don't want to
2298
# accidentally incorporate too much of that index's code.
2300
pending = set(versions)
2301
cache = self._by_version
2303
version = pending.pop()
2306
parents = [p for p in cache[version][2] if p in cache]
2308
raise RevisionNotPresent(version, self)
2309
# if not completed and not a ghost
2310
pending.update([p for p in parents if p not in ancestry])
2311
ancestry.add(version)
2312
return list(ancestry)
2314
def get_build_details(self, version_ids):
2315
"""Get the method, index_memo and compression parent for version_ids.
2317
Ghosts are omitted from the result.
2319
:param version_ids: An iterable of version_ids.
2320
:return: A dict of version_id:(index_memo, compression_parent,
2321
parents, record_details).
2323
opaque structure to pass to read_records to extract the raw
2326
Content that this record is built upon, may be None
2328
Logical parents of this node
2330
extra information about the content which needs to be passed to
2331
Factory.parse_record
2334
for version_id in version_ids:
2336
method = self.get_method(version_id)
2337
except errors.RevisionNotPresent:
2338
# ghosts are omitted
2340
parent_ids = self.get_parents_with_ghosts(version_id)
2341
noeol = ('no-eol' in self.get_options(version_id))
2342
if method == 'fulltext':
2343
compression_parent = None
2345
compression_parent = parent_ids[0]
2346
index_memo = self.get_position(version_id)
2347
result[version_id] = (index_memo, compression_parent,
2348
parent_ids, (method, noeol))
2351
def get_method(self, version_id):
2352
"""Return compression method of specified version."""
2354
options = self._by_version[version_id][0]
2356
# Strictly speaking this should check in the backing knit, but
2357
# until we have a test to discriminate, this will do.
2358
return self.backing_index.get_method(version_id)
2359
if 'fulltext' in options:
2361
elif 'line-delta' in options:
2364
raise errors.KnitIndexUnknownMethod(self, options)
2366
def get_options(self, version_id):
2367
"""Return a list representing options.
2372
return self._by_version[version_id][0]
2374
return self.backing_index.get_options(version_id)
2376
def get_parents_with_ghosts(self, version_id):
2377
"""Return parents of specified version with ghosts."""
2379
return self._by_version[version_id][2]
2381
return self.backing_index.get_parents_with_ghosts(version_id)
2383
def get_position(self, version_id):
2384
"""Return details needed to access the version.
2386
_StreamAccess has the data as a big array, so we return slice
2387
coordinates into that (as index_memo's are opaque outside the
2388
index and matching access class).
2390
:return: a tuple (thunk_flag, index, start, end). If thunk_flag is
2391
False, index will be self, otherwise it will be a version id.
2394
start, end = self._by_version[version_id][1]
2395
return False, self, start, end
2397
# Signal to the access object to handle this from the backing knit.
2398
return (True, version_id, None, None)
2400
def get_versions(self):
2401
"""Get all the versions in the stream."""
2402
return self._by_version.keys()
2404
def iter_parents(self, version_ids):
2405
"""Iterate through the parents for many version ids.
2407
:param version_ids: An iterable yielding version_ids.
2408
:return: An iterator that yields (version_id, parents). Requested
2409
version_ids not present in the versioned file are simply skipped.
2410
The order is undefined, allowing for different optimisations in
2411
the underlying implementation.
2414
for version in version_ids:
2416
result.append((version, self._by_version[version][2]))
2422
class _KnitData(object):
2423
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2425
The KnitData class provides the logic for parsing and using knit records,
2426
making use of an access method for the low level read and write operations.
2429
def __init__(self, access):
2430
"""Create a KnitData object.
2432
:param access: The access method to use. Access methods such as
2433
_KnitAccess manage the insertion of raw records and the subsequent
2434
retrieval of the same.
2436
self._access = access
2437
self._checked = False
2438
# TODO: jam 20060713 conceptually, this could spill to disk
2439
# if the cached size gets larger than a certain amount
2440
# but it complicates the model a bit, so for now just use
2441
# a simple dictionary
2443
self._do_cache = False
2445
def enable_cache(self):
2446
"""Enable caching of reads."""
2447
self._do_cache = True
2449
def clear_cache(self):
2450
"""Clear the record cache."""
2451
self._do_cache = False
2454
def _open_file(self):
2455
return self._access.open_file()
2457
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2458
"""Convert version_id, digest, lines into a raw data block.
2460
:param dense_lines: The bytes of lines but in a denser form. For
2461
instance, if lines is a list of 1000 bytestrings each ending in \n,
2462
dense_lines may be a list with one line in it, containing all the
2463
1000's lines and their \n's. Using dense_lines if it is already
2464
known is a win because the string join to create bytes in this
2465
function spends less time resizing the final string.
2466
:return: (len, a StringIO instance with the raw data ready to read.)
2468
# Note: using a string copy here increases memory pressure with e.g.
2469
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2470
# when doing the initial commit of a mozilla tree. RBC 20070921
2471
bytes = ''.join(chain(
2472
["version %s %d %s\n" % (version_id,
2475
dense_lines or lines,
2476
["end %s\n" % version_id]))
2477
assert bytes.__class__ == str
2478
compressed_bytes = bytes_to_gzip(bytes)
2479
return len(compressed_bytes), compressed_bytes
2481
def add_raw_records(self, sizes, raw_data):
2482
"""Append a prepared record to the data file.
2484
:param sizes: An iterable containing the size of each raw data segment.
2485
:param raw_data: A bytestring containing the data.
2486
:return: a list of index data for the way the data was stored.
2487
See the access method add_raw_records documentation for more
2490
return self._access.add_raw_records(sizes, raw_data)
2492
def _parse_record_header(self, version_id, raw_data):
2493
"""Parse a record header for consistency.
2495
:return: the header and the decompressor stream.
2496
as (stream, header_record)
2498
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2500
rec = self._check_header(version_id, df.readline())
2501
except Exception, e:
2502
raise KnitCorrupt(self._access,
2503
"While reading {%s} got %s(%s)"
2504
% (version_id, e.__class__.__name__, str(e)))
2507
def _check_header(self, version_id, line):
2510
raise KnitCorrupt(self._access,
2511
'unexpected number of elements in record header')
2512
if rec[1] != version_id:
2513
raise KnitCorrupt(self._access,
2514
'unexpected version, wanted %r, got %r'
2515
% (version_id, rec[1]))
2518
def _parse_record(self, version_id, data):
2520
# 4168 calls in 2880 217 internal
2521
# 4168 calls to _parse_record_header in 2121
2522
# 4168 calls to readlines in 330
2523
df = GzipFile(mode='rb', fileobj=StringIO(data))
2526
record_contents = df.readlines()
2527
except Exception, e:
2528
raise KnitCorrupt(self._access,
2529
"While reading {%s} got %s(%s)"
2530
% (version_id, e.__class__.__name__, str(e)))
2531
header = record_contents.pop(0)
2532
rec = self._check_header(version_id, header)
2534
last_line = record_contents.pop()
2535
if len(record_contents) != int(rec[2]):
2536
raise KnitCorrupt(self._access,
2537
'incorrect number of lines %s != %s'
2539
% (len(record_contents), int(rec[2]),
2541
if last_line != 'end %s\n' % rec[1]:
2542
raise KnitCorrupt(self._access,
2543
'unexpected version end line %r, wanted %r'
2544
% (last_line, version_id))
2546
return record_contents, rec[3]
2548
def read_records_iter_raw(self, records):
2549
"""Read text records from data file and yield raw data.
2551
This unpacks enough of the text record to validate the id is
2552
as expected but thats all.
2554
# setup an iterator of the external records:
2555
# uses readv so nice and fast we hope.
2557
# grab the disk data needed.
2559
# Don't check _cache if it is empty
2560
needed_offsets = [index_memo for version_id, index_memo
2562
if version_id not in self._cache]
2564
needed_offsets = [index_memo for version_id, index_memo
2567
raw_records = self._access.get_raw_records(needed_offsets)
2569
for version_id, index_memo in records:
2570
if version_id in self._cache:
2571
# This data has already been validated
2572
data = self._cache[version_id]
2574
data = raw_records.next()
2576
self._cache[version_id] = data
2578
# validate the header
2579
df, rec = self._parse_record_header(version_id, data)
2581
yield version_id, data
2583
def read_records_iter(self, records):
2584
"""Read text records from data file and yield result.
2586
The result will be returned in whatever is the fastest to read.
2587
Not by the order requested. Also, multiple requests for the same
2588
record will only yield 1 response.
2589
:param records: A list of (version_id, pos, len) entries
2590
:return: Yields (version_id, contents, digest) in the order
2591
read, not the order requested
2597
# Skip records we have alread seen
2598
yielded_records = set()
2599
needed_records = set()
2600
for record in records:
2601
if record[0] in self._cache:
2602
if record[0] in yielded_records:
2604
yielded_records.add(record[0])
2605
data = self._cache[record[0]]
2606
content, digest = self._parse_record(record[0], data)
2607
yield (record[0], content, digest)
2609
needed_records.add(record)
2610
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2612
needed_records = sorted(set(records), key=operator.itemgetter(1))
2614
if not needed_records:
2617
# The transport optimizes the fetching as well
2618
# (ie, reads continuous ranges.)
2619
raw_data = self._access.get_raw_records(
2620
[index_memo for version_id, index_memo in needed_records])
2622
for (version_id, index_memo), data in \
2623
izip(iter(needed_records), raw_data):
2624
content, digest = self._parse_record(version_id, data)
2626
self._cache[version_id] = data
2627
yield version_id, content, digest
2629
def read_records(self, records):
2630
"""Read records into a dictionary."""
2632
for record_id, content, digest in \
2633
self.read_records_iter(records):
2634
components[record_id] = (content, digest)
2638
class InterKnit(InterVersionedFile):
2639
"""Optimised code paths for knit to knit operations."""
2641
_matching_file_from_factory = KnitVersionedFile
2642
_matching_file_to_factory = KnitVersionedFile
2645
def is_compatible(source, target):
2646
"""Be compatible with knits. """
2648
return (isinstance(source, KnitVersionedFile) and
2649
isinstance(target, KnitVersionedFile))
2650
except AttributeError:
2653
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2654
"""Copy texts to the target by extracting and adding them one by one.
2656
see join() for the parameter definitions.
2658
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2659
graph = self.source.get_graph(version_ids)
2660
order = topo_sort(graph.items())
2662
def size_of_content(content):
2663
return sum(len(line) for line in content.text())
2664
# Cache at most 10MB of parent texts
2665
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2666
compute_size=size_of_content)
2667
# TODO: jam 20071116 It would be nice to have a streaming interface to
2668
# get multiple texts from a source. The source could be smarter
2669
# about how it handled intermediate stages.
2670
# get_line_list() or make_mpdiffs() seem like a possibility, but
2671
# at the moment they extract all full texts into memory, which
2672
# causes us to store more than our 3x fulltext goal.
2673
# Repository.iter_files_bytes() may be another possibility
2674
to_process = [version for version in order
2675
if version not in self.target]
2676
total = len(to_process)
2677
pb = ui.ui_factory.nested_progress_bar()
2679
for index, version in enumerate(to_process):
2680
pb.update('Converting versioned data', index, total)
2681
sha1, num_bytes, parent_text = self.target.add_lines(version,
2682
self.source.get_parents_with_ghosts(version),
2683
self.source.get_lines(version),
2684
parent_texts=parent_cache)
2685
parent_cache[version] = parent_text
2690
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2691
"""See InterVersionedFile.join."""
2692
assert isinstance(self.source, KnitVersionedFile)
2693
assert isinstance(self.target, KnitVersionedFile)
2695
# If the source and target are mismatched w.r.t. annotations vs
2696
# plain, the data needs to be converted accordingly
2697
if self.source.factory.annotated == self.target.factory.annotated:
2699
elif self.source.factory.annotated:
2700
converter = self._anno_to_plain_converter
2702
# We're converting from a plain to an annotated knit. Copy them
2703
# across by full texts.
2704
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2706
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2710
pb = ui.ui_factory.nested_progress_bar()
2712
version_ids = list(version_ids)
2713
if None in version_ids:
2714
version_ids.remove(None)
2716
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2718
this_versions = set(self.target._index.get_versions())
2719
# XXX: For efficiency we should not look at the whole index,
2720
# we only need to consider the referenced revisions - they
2721
# must all be present, or the method must be full-text.
2722
# TODO, RBC 20070919
2723
needed_versions = self.source_ancestry - this_versions
2725
if not needed_versions:
2727
full_list = topo_sort(self.source.get_graph())
2729
version_list = [i for i in full_list if (not self.target.has_version(i)
2730
and i in needed_versions)]
2734
copy_queue_records = []
2736
for version_id in version_list:
2737
options = self.source._index.get_options(version_id)
2738
parents = self.source._index.get_parents_with_ghosts(version_id)
2739
# check that its will be a consistent copy:
2740
for parent in parents:
2741
# if source has the parent, we must :
2742
# * already have it or
2743
# * have it scheduled already
2744
# otherwise we don't care
2745
assert (self.target.has_version(parent) or
2746
parent in copy_set or
2747
not self.source.has_version(parent))
2748
index_memo = self.source._index.get_position(version_id)
2749
copy_queue_records.append((version_id, index_memo))
2750
copy_queue.append((version_id, options, parents))
2751
copy_set.add(version_id)
2753
# data suck the join:
2755
total = len(version_list)
2758
for (version_id, raw_data), \
2759
(version_id2, options, parents) in \
2760
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2762
assert version_id == version_id2, 'logic error, inconsistent results'
2764
pb.update("Joining knit", count, total)
2766
size, raw_data = converter(raw_data, version_id, options,
2769
size = len(raw_data)
2770
raw_records.append((version_id, options, parents, size))
2771
raw_datum.append(raw_data)
2772
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2777
def _anno_to_plain_converter(self, raw_data, version_id, options,
2779
"""Convert annotated content to plain content."""
2780
data, digest = self.source._data._parse_record(version_id, raw_data)
2781
if 'fulltext' in options:
2782
content = self.source.factory.parse_fulltext(data, version_id)
2783
lines = self.target.factory.lower_fulltext(content)
2785
delta = self.source.factory.parse_line_delta(data, version_id,
2787
lines = self.target.factory.lower_line_delta(delta)
2788
return self.target._data._record_to_data(version_id, digest, lines)
2791
InterVersionedFile.register_optimiser(InterKnit)
2794
class WeaveToKnit(InterVersionedFile):
2795
"""Optimised code paths for weave to knit operations."""
2797
_matching_file_from_factory = bzrlib.weave.WeaveFile
2798
_matching_file_to_factory = KnitVersionedFile
2801
def is_compatible(source, target):
2802
"""Be compatible with weaves to knits."""
2804
return (isinstance(source, bzrlib.weave.Weave) and
2805
isinstance(target, KnitVersionedFile))
2806
except AttributeError:
2809
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2810
"""See InterVersionedFile.join."""
2811
assert isinstance(self.source, bzrlib.weave.Weave)
2812
assert isinstance(self.target, KnitVersionedFile)
2814
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2819
pb = ui.ui_factory.nested_progress_bar()
2821
version_ids = list(version_ids)
2823
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2824
this_versions = set(self.target._index.get_versions())
2825
needed_versions = self.source_ancestry - this_versions
2827
if not needed_versions:
2829
full_list = topo_sort(self.source.get_graph())
2831
version_list = [i for i in full_list if (not self.target.has_version(i)
2832
and i in needed_versions)]
2836
total = len(version_list)
2837
for version_id in version_list:
2838
pb.update("Converting to knit", count, total)
2839
parents = self.source.get_parents(version_id)
2840
# check that its will be a consistent copy:
2841
for parent in parents:
2842
# if source has the parent, we must already have it
2843
assert (self.target.has_version(parent))
2844
self.target.add_lines(
2845
version_id, parents, self.source.get_lines(version_id))
2852
InterVersionedFile.register_optimiser(WeaveToKnit)
3385
self._indices[index] = transport_packname
3386
self._container_writer = writer
3387
self._write_index = index
3389
def reload_or_raise(self, retry_exc):
3390
"""Try calling the reload function, or re-raise the original exception.
3392
This should be called after _DirectPackAccess raises a
3393
RetryWithNewPacks exception. This function will handle the common logic
3394
of determining when the error is fatal versus being temporary.
3395
It will also make sure that the original exception is raised, rather
3396
than the RetryWithNewPacks exception.
3398
If this function returns, then the calling function should retry
3399
whatever operation was being performed. Otherwise an exception will
3402
:param retry_exc: A RetryWithNewPacks exception.
3405
if self._reload_func is None:
3407
elif not self._reload_func():
3408
# The reload claimed that nothing changed
3409
if not retry_exc.reload_occurred:
3410
# If there wasn't an earlier reload, then we really were
3411
# expecting to find changes. We didn't find them, so this is a
3415
exc_class, exc_value, exc_traceback = retry_exc.exc_info
3416
raise exc_class, exc_value, exc_traceback
2855
3419
# Deprecated, use PatienceSequenceMatcher instead
2970
3464
parents to create an annotation, but only need 1 parent to generate the
2973
:return: A list of (revision_id, index_memo) records, suitable for
2974
passing to read_records_iter to start reading in the raw data fro/
3467
:return: A list of (key, index_memo) records, suitable for
3468
passing to read_records_iter to start reading in the raw data from
2977
if revision_id in self._annotated_lines:
2980
pending = set([revision_id])
3471
pending = set([key])
3474
self._num_needed_children[key] = 1
2985
3476
# get all pending nodes
2987
3477
this_iteration = pending
2988
build_details = self._knit._index.get_build_details(this_iteration)
3478
build_details = self._vf._index.get_build_details(this_iteration)
2989
3479
self._all_build_details.update(build_details)
2990
# new_nodes = self._knit._index._get_entries(this_iteration)
3480
# new_nodes = self._vf._index._get_entries(this_iteration)
2991
3481
pending = set()
2992
for rev_id, details in build_details.iteritems():
2993
(index_memo, compression_parent, parents,
3482
for key, details in build_details.iteritems():
3483
(index_memo, compression_parent, parent_keys,
2994
3484
record_details) = details
2995
self._revision_id_graph[rev_id] = parents
2996
records.append((rev_id, index_memo))
3485
self._parent_map[key] = parent_keys
3486
self._heads_provider = None
3487
records.append((key, index_memo))
2997
3488
# Do we actually need to check _annotated_lines?
2998
pending.update(p for p in parents
2999
if p not in self._all_build_details)
3489
pending.update([p for p in parent_keys
3490
if p not in self._all_build_details])
3492
for parent_key in parent_keys:
3493
if parent_key in self._num_needed_children:
3494
self._num_needed_children[parent_key] += 1
3496
self._num_needed_children[parent_key] = 1
3000
3497
if compression_parent:
3001
self._compression_children.setdefault(compression_parent,
3004
for parent in parents:
3005
self._annotate_children.setdefault(parent,
3007
num_gens = generation - kept_generation
3008
if ((num_gens >= self._generations_until_keep)
3009
and len(parents) > 1):
3010
kept_generation = generation
3011
self._nodes_to_keep_annotations.add(rev_id)
3498
if compression_parent in self._num_compression_children:
3499
self._num_compression_children[compression_parent] += 1
3501
self._num_compression_children[compression_parent] = 1
3013
3503
missing_versions = this_iteration.difference(build_details.keys())
3014
self._ghosts.update(missing_versions)
3015
for missing_version in missing_versions:
3016
# add a key, no parents
3017
self._revision_id_graph[missing_version] = ()
3018
pending.discard(missing_version) # don't look for it
3019
# XXX: This should probably be a real exception, as it is a data
3021
assert not self._ghosts.intersection(self._compression_children), \
3022
"We cannot have nodes which have a compression parent of a ghost."
3023
# Cleanout anything that depends on a ghost so that we don't wait for
3024
# the ghost to show up
3025
for node in self._ghosts:
3026
if node in self._annotate_children:
3027
# We won't be building this node
3028
del self._annotate_children[node]
3504
if missing_versions:
3505
for key in missing_versions:
3506
if key in self._parent_map and key in self._text_cache:
3507
# We already have this text ready, we just need to
3508
# yield it later so we get it annotated
3510
parent_keys = self._parent_map[key]
3511
for parent_key in parent_keys:
3512
if parent_key in self._num_needed_children:
3513
self._num_needed_children[parent_key] += 1
3515
self._num_needed_children[parent_key] = 1
3516
pending.update([p for p in parent_keys
3517
if p not in self._all_build_details])
3519
raise errors.RevisionNotPresent(key, self._vf)
3029
3520
# Generally we will want to read the records in reverse order, because
3030
3521
# we find the parent nodes after the children
3031
3522
records.reverse()
3034
def _annotate_records(self, records):
3035
"""Build the annotations for the listed records."""
3523
return records, ann_keys
3525
def _get_needed_texts(self, key, pb=None):
3526
# if True or len(self._vf._fallback_vfs) > 0:
3527
if len(self._vf._fallback_vfs) > 0:
3528
# If we have fallbacks, go to the generic path
3529
for v in annotate.Annotator._get_needed_texts(self, key, pb=pb):
3534
records, ann_keys = self._get_build_graph(key)
3535
for idx, (sub_key, text, num_lines) in enumerate(
3536
self._extract_texts(records)):
3538
pb.update('annotating', idx, len(records))
3539
yield sub_key, text, num_lines
3540
for sub_key in ann_keys:
3541
text = self._text_cache[sub_key]
3542
num_lines = len(text) # bad assumption
3543
yield sub_key, text, num_lines
3545
except errors.RetryWithNewPacks, e:
3546
self._vf._access.reload_or_raise(e)
3547
# The cached build_details are no longer valid
3548
self._all_build_details.clear()
3550
def _cache_delta_blocks(self, key, compression_parent, delta, lines):
3551
parent_lines = self._text_cache[compression_parent]
3552
blocks = list(KnitContent.get_line_delta_blocks(delta, parent_lines, lines))
3553
self._matching_blocks[(key, compression_parent)] = blocks
3555
def _expand_record(self, key, parent_keys, compression_parent, record,
3558
if compression_parent:
3559
if compression_parent not in self._content_objects:
3560
# Waiting for the parent
3561
self._pending_deltas.setdefault(compression_parent, []).append(
3562
(key, parent_keys, record, record_details))
3564
# We have the basis parent, so expand the delta
3565
num = self._num_compression_children[compression_parent]
3568
base_content = self._content_objects.pop(compression_parent)
3569
self._num_compression_children.pop(compression_parent)
3571
self._num_compression_children[compression_parent] = num
3572
base_content = self._content_objects[compression_parent]
3573
# It is tempting to want to copy_base_content=False for the last
3574
# child object. However, whenever noeol=False,
3575
# self._text_cache[parent_key] is content._lines. So mutating it
3576
# gives very bad results.
3577
# The alternative is to copy the lines into text cache, but then we
3578
# are copying anyway, so just do it here.
3579
content, delta = self._vf._factory.parse_record(
3580
key, record, record_details, base_content,
3581
copy_base_content=True)
3584
content, _ = self._vf._factory.parse_record(
3585
key, record, record_details, None)
3586
if self._num_compression_children.get(key, 0) > 0:
3587
self._content_objects[key] = content
3588
lines = content.text()
3589
self._text_cache[key] = lines
3590
if delta is not None:
3591
self._cache_delta_blocks(key, compression_parent, delta, lines)
3594
def _get_parent_annotations_and_matches(self, key, text, parent_key):
3595
"""Get the list of annotations for the parent, and the matching lines.
3597
:param text: The opaque value given by _get_needed_texts
3598
:param parent_key: The key for the parent text
3599
:return: (parent_annotations, matching_blocks)
3600
parent_annotations is a list as long as the number of lines in
3602
matching_blocks is a list of (parent_idx, text_idx, len) tuples
3603
indicating which lines match between the two texts
3605
block_key = (key, parent_key)
3606
if block_key in self._matching_blocks:
3607
blocks = self._matching_blocks.pop(block_key)
3608
parent_annotations = self._annotations_cache[parent_key]
3609
return parent_annotations, blocks
3610
return annotate.Annotator._get_parent_annotations_and_matches(self,
3611
key, text, parent_key)
3613
def _process_pending(self, key):
3614
"""The content for 'key' was just processed.
3616
Determine if there is any more pending work to be processed.
3619
if key in self._pending_deltas:
3620
compression_parent = key
3621
children = self._pending_deltas.pop(key)
3622
for child_key, parent_keys, record, record_details in children:
3623
lines = self._expand_record(child_key, parent_keys,
3625
record, record_details)
3626
if self._check_ready_for_annotations(child_key, parent_keys):
3627
to_return.append(child_key)
3628
# Also check any children that are waiting for this parent to be
3630
if key in self._pending_annotation:
3631
children = self._pending_annotation.pop(key)
3632
to_return.extend([c for c, p_keys in children
3633
if self._check_ready_for_annotations(c, p_keys)])
3636
def _check_ready_for_annotations(self, key, parent_keys):
3637
"""return true if this text is ready to be yielded.
3639
Otherwise, this will return False, and queue the text into
3640
self._pending_annotation
3642
for parent_key in parent_keys:
3643
if parent_key not in self._annotations_cache:
3644
# still waiting on at least one parent text, so queue it up
3645
# Note that if there are multiple parents, we need to wait
3647
self._pending_annotation.setdefault(parent_key,
3648
[]).append((key, parent_keys))
3652
def _extract_texts(self, records):
3653
"""Extract the various texts needed based on records"""
3036
3654
# We iterate in the order read, rather than a strict order requested
3037
3655
# However, process what we can, and put off to the side things that
3038
3656
# still need parents, cleaning them up when those parents are
3040
for (rev_id, record,
3041
digest) in self._knit._data.read_records_iter(records):
3042
if rev_id in self._annotated_lines:
3659
# 1) As 'records' are read, see if we can expand these records into
3660
# Content objects (and thus lines)
3661
# 2) If a given line-delta is waiting on its compression parent, it
3662
# gets queued up into self._pending_deltas, otherwise we expand
3663
# it, and put it into self._text_cache and self._content_objects
3664
# 3) If we expanded the text, we will then check to see if all
3665
# parents have also been processed. If so, this text gets yielded,
3666
# else this record gets set aside into pending_annotation
3667
# 4) Further, if we expanded the text in (2), we will then check to
3668
# see if there are any children in self._pending_deltas waiting to
3669
# also be processed. If so, we go back to (2) for those
3670
# 5) Further again, if we yielded the text, we can then check if that
3671
# 'unlocks' any of the texts in pending_annotations, which should
3672
# then get yielded as well
3673
# Note that both steps 4 and 5 are 'recursive' in that unlocking one
3674
# compression child could unlock yet another, and yielding a fulltext
3675
# will also 'unlock' the children that are waiting on that annotation.
3676
# (Though also, unlocking 1 parent's fulltext, does not unlock a child
3677
# if other parents are also waiting.)
3678
# We want to yield content before expanding child content objects, so
3679
# that we know when we can re-use the content lines, and the annotation
3680
# code can know when it can stop caching fulltexts, as well.
3682
# Children that are missing their compression parent
3684
for (key, record, digest) in self._vf._read_records_iter(records):
3686
details = self._all_build_details[key]
3687
(_, compression_parent, parent_keys, record_details) = details
3688
lines = self._expand_record(key, parent_keys, compression_parent,
3689
record, record_details)
3691
# Pending delta should be queued up
3044
parent_ids = self._revision_id_graph[rev_id]
3045
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3046
details = self._all_build_details[rev_id]
3047
(index_memo, compression_parent, parents,
3048
record_details) = details
3049
nodes_to_annotate = []
3050
# TODO: Remove the punning between compression parents, and
3051
# parent_ids, we should be able to do this without assuming
3053
if len(parent_ids) == 0:
3054
# There are no parents for this node, so just add it
3055
# TODO: This probably needs to be decoupled
3056
assert compression_parent is None
3057
fulltext_content, delta = self._knit.factory.parse_record(
3058
rev_id, record, record_details, None)
3059
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3060
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3061
parent_ids, left_matching_blocks=None))
3063
child = (rev_id, parent_ids, record)
3064
# Check if all the parents are present
3065
self._check_parents(child, nodes_to_annotate)
3066
while nodes_to_annotate:
3067
# Should we use a queue here instead of a stack?
3068
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3069
(index_memo, compression_parent, parents,
3070
record_details) = self._all_build_details[rev_id]
3071
if compression_parent is not None:
3072
comp_children = self._compression_children[compression_parent]
3073
assert rev_id in comp_children
3074
# If there is only 1 child, it is safe to reuse this
3076
reuse_content = (len(comp_children) == 1
3077
and compression_parent not in
3078
self._nodes_to_keep_annotations)
3080
# Remove it from the cache since it will be changing
3081
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3082
# Make sure to copy the fulltext since it might be
3084
parent_fulltext = list(parent_fulltext_content.text())
3086
parent_fulltext_content = self._fulltext_contents[compression_parent]
3087
parent_fulltext = parent_fulltext_content.text()
3088
comp_children.remove(rev_id)
3089
fulltext_content, delta = self._knit.factory.parse_record(
3090
rev_id, record, record_details,
3091
parent_fulltext_content,
3092
copy_base_content=(not reuse_content))
3093
fulltext = self._add_fulltext_content(rev_id,
3095
blocks = KnitContent.get_line_delta_blocks(delta,
3096
parent_fulltext, fulltext)
3098
fulltext_content = self._knit.factory.parse_fulltext(
3100
fulltext = self._add_fulltext_content(rev_id,
3103
nodes_to_annotate.extend(
3104
self._add_annotation(rev_id, fulltext, parent_ids,
3105
left_matching_blocks=blocks))
3107
def _get_heads_provider(self):
3108
"""Create a heads provider for resolving ancestry issues."""
3109
if self._heads_provider is not None:
3110
return self._heads_provider
3111
parent_provider = _mod_graph.DictParentsProvider(
3112
self._revision_id_graph)
3113
graph_obj = _mod_graph.Graph(parent_provider)
3114
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3115
self._heads_provider = head_cache
3118
def annotate(self, revision_id):
3119
"""Return the annotated fulltext at the given revision.
3121
:param revision_id: The revision id for this file
3123
records = self._get_build_graph(revision_id)
3124
if revision_id in self._ghosts:
3125
raise errors.RevisionNotPresent(revision_id, self._knit)
3126
self._annotate_records(records)
3127
return self._annotated_lines[revision_id]
3693
# At this point, we may be able to yield this content, if all
3694
# parents are also finished
3695
yield_this_text = self._check_ready_for_annotations(key,
3698
# All parents present
3699
yield key, lines, len(lines)
3700
to_process = self._process_pending(key)
3702
this_process = to_process
3704
for key in this_process:
3705
lines = self._text_cache[key]
3706
yield key, lines, len(lines)
3707
to_process.extend(self._process_pending(key))
3131
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3710
from bzrlib._knit_load_data_pyx import _load_data_c as _load_data
3711
except ImportError, e:
3712
osutils.failed_to_load_extension(e)
3133
3713
from bzrlib._knit_load_data_py import _load_data_py as _load_data