175
154
internal representation is of the format:
176
155
(revid, plaintext)
178
# TODO: jam 20070209 The tests expect this to be returned as tuples,
179
# but the code itself doesn't really depend on that.
180
# Figure out a way to not require the overhead of turning the
181
# list back into tuples.
182
lines = [tuple(line.split(' ', 1)) for line in content]
159
origin, text = line.split(' ', 1)
160
lines.append((origin.decode('utf-8'), text))
183
161
return KnitContent(lines)
185
163
def parse_line_delta_iter(self, lines):
186
return iter(self.parse_line_delta(lines))
188
def parse_line_delta(self, lines, version_id):
189
164
"""Convert a line based delta into internal representation.
191
166
line delta is in the form of:
192
167
intstart intend intcount
194
169
revid(utf8) newline\n
195
internal representation is
170
internal represnetation is
196
171
(start, end, count, [1..count tuples (revid, newline)])
203
def cache_and_return(line):
204
origin, text = line.split(' ', 1)
205
return cache.setdefault(origin, origin), text
207
# walk through the lines parsing.
209
start, end, count = [int(n) for n in header.split(',')]
210
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
211
result.append((start, end, count, contents))
214
def get_fulltext_content(self, lines):
215
"""Extract just the content lines from a fulltext."""
216
return (line.split(' ', 1)[1] for line in lines)
218
def get_linedelta_content(self, lines):
219
"""Extract just the content from a line delta.
221
This doesn't return all of the extra information stored in a delta.
222
Only the actual content lines.
227
header = header.split(',')
228
count = int(header[2])
229
for i in xrange(count):
230
origin, text = next().split(' ', 1)
174
header = lines.pop(0)
175
start, end, c = [int(n) for n in header.split(',')]
178
origin, text = lines.pop(0).split(' ', 1)
179
contents.append((origin.decode('utf-8'), text))
180
yield start, end, c, contents
182
def parse_line_delta(self, lines, version):
183
return list(self.parse_line_delta_iter(lines))
233
185
def lower_fulltext(self, content):
234
186
"""convert a fulltext content record into a serializable form.
236
188
see parse_fulltext which this inverts.
238
# TODO: jam 20070209 We only do the caching thing to make sure that
239
# the origin is a valid utf-8 line, eventually we could remove it
240
return ['%s %s' % (o, t) for o, t in content._lines]
190
return ['%s %s' % (o.encode('utf-8'), t) for o, t in content._lines]
242
192
def lower_line_delta(self, delta):
243
193
"""convert a delta into a serializable form.
245
See parse_line_delta which this inverts.
195
See parse_line_delta_iter which this inverts.
247
# TODO: jam 20070209 We only do the caching thing to make sure that
248
# the origin is a valid utf-8 line, eventually we could remove it
250
198
for start, end, c, lines in delta:
251
199
out.append('%d,%d,%d\n' % (start, end, c))
252
out.extend(origin + ' ' + text
253
for origin, text in lines)
200
for origin, text in lines:
201
out.append('%s %s' % (origin.encode('utf-8'), text))
260
208
annotated = False
262
def parse_fulltext(self, content, version_id):
210
def parse_fulltext(self, content, version):
263
211
"""This parses an unannotated fulltext.
265
213
Note that this is not a noop - the internal representation
266
214
has (versionid, line) - its just a constant versionid.
268
return self.make(content, version_id)
216
return self.make(content, version)
270
def parse_line_delta_iter(self, lines, version_id):
272
num_lines = len(lines)
273
while cur < num_lines:
218
def parse_line_delta_iter(self, lines, version):
220
header = lines.pop(0)
276
221
start, end, c = [int(n) for n in header.split(',')]
277
yield start, end, c, zip([version_id] * c, lines[cur:cur+c])
280
def parse_line_delta(self, lines, version_id):
281
return list(self.parse_line_delta_iter(lines, version_id))
283
def get_fulltext_content(self, lines):
284
"""Extract just the content lines from a fulltext."""
287
def get_linedelta_content(self, lines):
288
"""Extract just the content from a line delta.
290
This doesn't return all of the extra information stored in a delta.
291
Only the actual content lines.
296
header = header.split(',')
297
count = int(header[2])
298
for i in xrange(count):
222
yield start, end, c, zip([version] * c, lines[:c])
225
def parse_line_delta(self, lines, version):
226
return list(self.parse_line_delta_iter(lines, version))
301
228
def lower_fulltext(self, content):
302
229
return content.text()
330
257
stored and retrieved.
333
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
334
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
335
create=False, create_parent_dir=False, delay_create=False,
260
def __init__(self, relpath, transport, file_mode=None, access_mode=None, factory=None,
261
basis_knit=None, delta=True, create=False):
337
262
"""Construct a knit at location specified by relpath.
339
264
:param create: If not True, only open an existing knit.
340
:param create_parent_dir: If True, create the parent directory if
341
creating the file fails. (This is used for stores with
342
hash-prefixes that may not exist yet)
343
:param delay_create: The calling code is aware that the knit won't
344
actually be created until the first data is stored.
346
if deprecated_passed(basis_knit):
347
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
348
" deprecated as of bzr 0.9.",
349
DeprecationWarning, stacklevel=2)
350
266
if access_mode is None:
351
267
access_mode = 'w'
352
268
super(KnitVersionedFile, self).__init__(access_mode)
353
269
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
270
assert not basis_knit or isinstance(basis_knit, KnitVersionedFile), \
354
273
self.transport = transport
355
274
self.filename = relpath
275
self.basis_knit = basis_knit
356
276
self.factory = factory or KnitAnnotateFactory()
357
277
self.writable = (access_mode == 'w')
358
278
self.delta = delta
360
self._max_delta_chain = 200
362
280
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
363
access_mode, create=create, file_mode=file_mode,
364
create_parent_dir=create_parent_dir, delay_create=delay_create,
281
access_mode, create=create)
366
282
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
367
access_mode, create=create and not len(self), file_mode=file_mode,
368
create_parent_dir=create_parent_dir, delay_create=delay_create,
372
return '%s(%s)' % (self.__class__.__name__,
373
self.transport.abspath(self.filename))
375
def _check_should_delta(self, first_parents):
376
"""Iterate back through the parent listing, looking for a fulltext.
378
This is used when we want to decide whether to add a delta or a new
379
fulltext. It searches for _max_delta_chain parents. When it finds a
380
fulltext parent, it sees if the total size of the deltas leading up to
381
it is large enough to indicate that we want a new full text anyway.
383
Return True if we should create a new delta, False if we should use a
388
delta_parents = first_parents
389
for count in xrange(self._max_delta_chain):
390
parent = delta_parents[0]
391
method = self._index.get_method(parent)
392
pos, size = self._index.get_position(parent)
393
if method == 'fulltext':
397
delta_parents = self._index.get_parents(parent)
399
# We couldn't find a fulltext, so we must create a new one
402
return fulltext_size > delta_size
283
access_mode, create=not len(self.versions()))
404
285
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
405
286
"""See VersionedFile._add_delta()."""
631
477
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
632
478
return diff_hunks
634
def _get_components_positions(self, version_ids):
635
"""Produce a map of position data for the components of versions.
637
This data is intended to be used for retrieving the knit records.
639
A dict of version_id to (method, data_pos, data_size, next) is
641
method is the way referenced data should be applied.
642
data_pos is the position of the data in the knit.
643
data_size is the size of the data in the knit.
644
next is the build-parent of the version, or None for fulltexts.
480
def _get_components(self, version_id):
481
"""Return a list of (version_id, method, data) tuples that
482
makes up version specified by version_id of the knit.
484
The components should be applied in the order of the returned
487
The basis knit will be used to the largest extent possible
488
since it is assumed that accesses to it is faster.
647
for version_id in version_ids:
650
while cursor is not None and cursor not in component_data:
651
method = self._index.get_method(cursor)
652
if method == 'fulltext':
655
next = self.get_parents(cursor)[0]
656
data_pos, data_size = self._index.get_position(cursor)
657
component_data[cursor] = (method, data_pos, data_size, next)
659
return component_data
490
# needed_revisions holds a list of (method, version_id) of
491
# versions that is needed to be fetched to construct the final
492
# version of the file.
494
# basis_revisions is a list of versions that needs to be
495
# fetched but exists in the basis knit.
497
basis = self.basis_knit
504
if basis and basis._index.has_version(cursor):
506
basis_versions.append(cursor)
507
method = picked_knit._index.get_method(cursor)
508
needed_versions.append((method, cursor))
509
if method == 'fulltext':
511
cursor = picked_knit.get_parents(cursor)[0]
516
for comp_id in basis_versions:
517
data_pos, data_size = basis._index.get_data_position(comp_id)
518
records.append((piece_id, data_pos, data_size))
519
components.update(basis._data.read_records(records))
522
for comp_id in [vid for method, vid in needed_versions
523
if vid not in basis_versions]:
524
data_pos, data_size = self._index.get_position(comp_id)
525
records.append((comp_id, data_pos, data_size))
526
components.update(self._data.read_records(records))
528
# get_data_records returns a mapping with the version id as
529
# index and the value as data. The order the components need
530
# to be applied is held by needed_versions (reversed).
532
for method, comp_id in reversed(needed_versions):
533
out.append((comp_id, method, components[comp_id]))
661
537
def _get_content(self, version_id, parent_texts={}):
662
538
"""Returns a content object that makes up the specified
668
544
if cached_version is not None:
669
545
return cached_version
671
text_map, contents_map = self._get_content_maps([version_id])
672
return contents_map[version_id]
547
if self.basis_knit and version_id in self.basis_knit:
548
return self.basis_knit._get_content(version_id)
551
components = self._get_components(version_id)
552
for component_id, method, (data, digest) in components:
553
version_idx = self._index.lookup(component_id)
554
if method == 'fulltext':
555
assert content is None
556
content = self.factory.parse_fulltext(data, version_idx)
557
elif method == 'line-delta':
558
delta = self.factory.parse_line_delta(data, version_idx)
559
content._lines = self._apply_delta(content._lines, delta)
561
if 'no-eol' in self._index.get_options(version_id):
562
line = content._lines[-1][1].rstrip('\n')
563
content._lines[-1] = (content._lines[-1][0], line)
565
if sha_strings(content.text()) != digest:
566
import pdb;pdb.set_trace()
567
raise KnitCorrupt(self.filename, 'sha-1 does not match %s' % version_id)
674
571
def _check_versions_present(self, version_ids):
675
572
"""Check that all specified versions are present."""
676
self._index.check_versions_present(version_ids)
573
version_ids = set(version_ids)
574
for r in list(version_ids):
575
if self._index.has_version(r):
576
version_ids.remove(r)
578
raise RevisionNotPresent(list(version_ids)[0], self.filename)
678
580
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
679
581
"""See VersionedFile.add_lines_with_ghosts()."""
768
680
def _clone_text(self, new_version_id, old_version_id, parents):
769
681
"""See VersionedFile.clone_text()."""
770
# FIXME RBC 20060228 make fast by only inserting an index with null
682
# FIXME RBC 20060228 make fast by only inserting an index with null delta.
772
683
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
774
685
def get_lines(self, version_id):
775
686
"""See VersionedFile.get_lines()."""
776
return self.get_line_list([version_id])[0]
778
def _get_record_map(self, version_ids):
779
"""Produce a dictionary of knit records.
781
The keys are version_ids, the values are tuples of (method, content,
783
method is the way the content should be applied.
784
content is a KnitContent object.
785
digest is the SHA1 digest of this version id after all steps are done
786
next is the build-parent of the version, i.e. the leftmost ancestor.
787
If the method is fulltext, next will be None.
789
position_map = self._get_components_positions(version_ids)
790
# c = component_id, m = method, p = position, s = size, n = next
791
records = [(c, p, s) for c, (m, p, s, n) in position_map.iteritems()]
793
for component_id, content, digest in \
794
self._data.read_records_iter(records):
795
method, position, size, next = position_map[component_id]
796
record_map[component_id] = method, content, digest, next
800
def get_text(self, version_id):
801
"""See VersionedFile.get_text"""
802
return self.get_texts([version_id])[0]
804
def get_texts(self, version_ids):
805
return [''.join(l) for l in self.get_line_list(version_ids)]
807
def get_line_list(self, version_ids):
808
"""Return the texts of listed versions as a list of strings."""
809
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
810
for version_id in version_ids:
811
self.check_not_reserved_id(version_id)
812
text_map, content_map = self._get_content_maps(version_ids)
813
return [text_map[v] for v in version_ids]
815
def _get_content_maps(self, version_ids):
816
"""Produce maps of text and KnitContents
818
:return: (text_map, content_map) where text_map contains the texts for
819
the requested versions and content_map contains the KnitContents.
820
Both dicts take version_ids as their keys.
822
for version_id in version_ids:
823
if not self.has_version(version_id):
824
raise RevisionNotPresent(version_id, self.filename)
825
record_map = self._get_record_map(version_ids)
830
for version_id in version_ids:
833
while cursor is not None:
834
method, data, digest, next = record_map[cursor]
835
components.append((cursor, method, data, digest))
836
if cursor in content_map:
841
for component_id, method, data, digest in reversed(components):
842
if component_id in content_map:
843
content = content_map[component_id]
845
if method == 'fulltext':
846
assert content is None
847
content = self.factory.parse_fulltext(data, version_id)
848
elif method == 'line-delta':
849
delta = self.factory.parse_line_delta(data, version_id)
850
content = content.copy()
851
content._lines = self._apply_delta(content._lines,
853
content_map[component_id] = content
855
if 'no-eol' in self._index.get_options(version_id):
856
content = content.copy()
857
line = content._lines[-1][1].rstrip('\n')
858
content._lines[-1] = (content._lines[-1][0], line)
859
final_content[version_id] = content
861
# digest here is the digest from the last applied component.
862
text = content.text()
863
if sha_strings(text) != digest:
864
raise KnitCorrupt(self.filename,
865
'sha-1 does not match %s' % version_id)
867
text_map[version_id] = text
868
return text_map, final_content
870
def iter_lines_added_or_present_in_versions(self, version_ids=None,
687
return self._get_content(version_id).text()
689
def iter_lines_added_or_present_in_versions(self, version_ids=None):
872
690
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
873
691
if version_ids is None:
874
692
version_ids = self.versions()
876
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
878
pb = progress.DummyProgress()
879
# we don't care about inclusions, the caller cares.
693
# we dont care about inclusions, the caller cares.
880
694
# but we need to setup a list of records to visit.
881
695
# we need version_id, position, length
882
696
version_id_records = []
883
requested_versions = set(version_ids)
697
requested_versions = list(version_ids)
884
698
# filter for available versions
885
699
for version_id in requested_versions:
886
700
if not self.has_version(version_id):
887
701
raise RevisionNotPresent(version_id, self.filename)
888
702
# get a in-component-order queue:
889
704
for version_id in self.versions():
890
705
if version_id in requested_versions:
706
version_ids.append(version_id)
891
707
data_pos, length = self._index.get_position(version_id)
892
708
version_id_records.append((version_id, data_pos, length))
710
pb = bzrlib.ui.ui_factory.nested_progress_bar()
894
712
total = len(version_id_records)
895
for version_idx, (version_id, data, sha_value) in \
896
enumerate(self._data.read_records_iter(version_id_records)):
897
pb.update('Walking content.', version_idx, total)
898
method = self._index.get_method(version_id)
900
assert method in ('fulltext', 'line-delta')
901
if method == 'fulltext':
902
line_iterator = self.factory.get_fulltext_content(data)
904
line_iterator = self.factory.get_linedelta_content(data)
905
for line in line_iterator:
908
pb.update('Walking content.', total, total)
714
pb.update('Walking content.', count, total)
715
for version_id, data, sha_value in \
716
self._data.read_records_iter(version_id_records):
717
pb.update('Walking content.', count, total)
718
method = self._index.get_method(version_id)
719
version_idx = self._index.lookup(version_id)
720
assert method in ('fulltext', 'line-delta')
721
if method == 'fulltext':
722
content = self.factory.parse_fulltext(data, version_idx)
723
for line in content.text():
726
delta = self.factory.parse_line_delta(data, version_idx)
727
for start, end, count, lines in delta:
728
for origin, line in lines:
731
pb.update('Walking content.', total, total)
734
pb.update('Walking content.', total, total)
910
738
def num_versions(self):
911
739
"""See VersionedFile.num_versions()."""
916
744
def annotate_iter(self, version_id):
917
745
"""See VersionedFile.annotate_iter."""
918
version_id = osutils.safe_revision_id(version_id)
919
746
content = self._get_content(version_id)
920
747
for origin, text in content.annotate_iter():
921
748
yield origin, text
923
750
def get_parents(self, version_id):
924
751
"""See VersionedFile.get_parents."""
927
# 52554 calls in 1264 872 internal down from 3674
928
version_id = osutils.safe_revision_id(version_id)
930
return self._index.get_parents(version_id)
932
raise RevisionNotPresent(version_id, self.filename)
752
self._check_versions_present([version_id])
753
return list(self._index.get_parents(version_id))
934
755
def get_parents_with_ghosts(self, version_id):
935
756
"""See VersionedFile.get_parents."""
936
version_id = osutils.safe_revision_id(version_id)
938
return self._index.get_parents_with_ghosts(version_id)
940
raise RevisionNotPresent(version_id, self.filename)
757
self._check_versions_present([version_id])
758
return list(self._index.get_parents_with_ghosts(version_id))
942
def get_ancestry(self, versions, topo_sorted=True):
760
def get_ancestry(self, versions):
943
761
"""See VersionedFile.get_ancestry."""
944
762
if isinstance(versions, basestring):
945
763
versions = [versions]
948
versions = [osutils.safe_revision_id(v) for v in versions]
949
return self._index.get_ancestry(versions, topo_sorted)
766
self._check_versions_present(versions)
767
return self._index.get_ancestry(versions)
951
769
def get_ancestry_with_ghosts(self, versions):
952
770
"""See VersionedFile.get_ancestry_with_ghosts."""
978
796
for lineno, insert_id, dset, line in w.walk(version_ids):
979
797
yield lineno, insert_id, dset, line
981
def plan_merge(self, ver_a, ver_b):
982
"""See VersionedFile.plan_merge."""
983
ver_a = osutils.safe_revision_id(ver_a)
984
ver_b = osutils.safe_revision_id(ver_b)
985
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
986
def status_a(revision, text):
987
if revision in ancestors_b:
988
return 'killed-b', text
992
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
993
def status_b(revision, text):
994
if revision in ancestors_a:
995
return 'killed-a', text
999
annotated_a = self.annotate(ver_a)
1000
annotated_b = self.annotate(ver_b)
1001
plain_a = [t for (a, t) in annotated_a]
1002
plain_b = [t for (a, t) in annotated_b]
1003
blocks = KnitSequenceMatcher(None, plain_a, plain_b).get_matching_blocks()
1006
for ai, bi, l in blocks:
1007
# process all mismatched sections
1008
# (last mismatched section is handled because blocks always
1009
# includes a 0-length last block)
1010
for revision, text in annotated_a[a_cur:ai]:
1011
yield status_a(revision, text)
1012
for revision, text in annotated_b[b_cur:bi]:
1013
yield status_b(revision, text)
1015
# and now the matched section
1018
for text_a, text_b in zip(plain_a[ai:a_cur], plain_b[bi:b_cur]):
1019
assert text_a == text_b
1020
yield "unchanged", text_a
1023
800
class _KnitComponentFile(object):
1024
801
"""One of the files used to implement a knit database"""
1026
def __init__(self, transport, filename, mode, file_mode=None,
1027
create_parent_dir=False, dir_mode=None):
803
def __init__(self, transport, filename, mode):
1028
804
self._transport = transport
1029
805
self._filename = filename
1030
806
self._mode = mode
1031
self._file_mode = file_mode
1032
self._dir_mode = dir_mode
1033
self._create_parent_dir = create_parent_dir
1034
self._need_to_create = False
1036
def _full_path(self):
1037
"""Return the full path to this file."""
1038
return self._transport.base + self._filename
808
def write_header(self):
809
if self._transport.append(self._filename, StringIO(self.HEADER)):
810
raise KnitCorrupt(self._filename, 'misaligned after writing header')
1040
812
def check_header(self, fp):
1041
line = fp.readline()
1043
# An empty file can actually be treated as though the file doesn't
1045
raise errors.NoSuchFile(self._full_path())
813
line = fp.read(len(self.HEADER))
1046
814
if line != self.HEADER:
1047
raise KnitHeaderError(badline=line,
1048
filename=self._transport.abspath(self._filename))
815
raise KnitHeaderError(badline=line)
1050
817
def commit(self):
1051
818
"""Commit is a nop."""
1076
843
this allows updates to correct version and parent information.
1077
844
Note that the two entries may share the delta, and that successive
1078
845
annotations and references MUST point to the first entry.
1080
The index file on disc contains a header, followed by one line per knit
1081
record. The same revision can be present in an index file more than once.
1082
The first occurrence gets assigned a sequence number starting from 0.
1084
The format of a single line is
1085
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1086
REVISION_ID is a utf8-encoded revision id
1087
FLAGS is a comma separated list of flags about the record. Values include
1088
no-eol, line-delta, fulltext.
1089
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1090
that the the compressed data starts at.
1091
LENGTH is the ascii representation of the length of the data file.
1092
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1094
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1095
revision id already in the knit that is a parent of REVISION_ID.
1096
The ' :' marker is the end of record marker.
1099
when a write is interrupted to the index file, it will result in a line
1100
that does not end in ' :'. If the ' :' is not present at the end of a line,
1101
or at the end of the file, then the record that is missing it will be
1102
ignored by the parser.
1104
When writing new records to the index file, the data is preceded by '\n'
1105
to ensure that records always start on new lines even if the last write was
1106
interrupted. As a result its normal for the last line in the index to be
1107
missing a trailing newline. One can be added with no harmful effects.
1110
HEADER = "# bzr knit index 8\n"
848
HEADER = "# bzr knit index 7\n"
1112
850
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1113
851
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1115
853
def _cache_version(self, version_id, options, pos, size, parents):
1116
854
"""Cache a version record in the history array and index cache.
1118
This is inlined into _load_data for performance. KEEP IN SYNC.
856
This is inlined into __init__ for performance. KEEP IN SYNC.
1119
857
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1122
860
# only want the _history index to reference the 1st index entry
1123
861
# for version_id
1124
862
if version_id not in self._cache:
1125
index = len(self._history)
1126
863
self._history.append(version_id)
1128
index = self._cache[version_id][5]
1129
self._cache[version_id] = (version_id,
864
self._cache[version_id] = (version_id, options, pos, size, parents)
1136
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1137
create_parent_dir=False, delay_create=False, dir_mode=None):
1138
_KnitComponentFile.__init__(self, transport, filename, mode,
1139
file_mode=file_mode,
1140
create_parent_dir=create_parent_dir,
866
def __init__(self, transport, filename, mode, create=False):
867
_KnitComponentFile.__init__(self, transport, filename, mode)
1142
868
self._cache = {}
1143
869
# position in _history is the 'official' index for a revision
1144
870
# but the values may have come from a newer entry.
1145
# so - wc -l of a knit index is != the number of unique names
871
# so - wc -l of a knit index is != the number of uniqe names
1147
873
self._history = []
874
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1149
fp = self._transport.get(self._filename)
1151
# _load_data may raise NoSuchFile if the target knit is
1153
_load_data(self, fp)
1157
if mode != 'w' or not create:
1160
self._need_to_create = True
879
pb.update('read knit index', count, total)
880
fp = self._transport.get(self._filename)
881
self.check_header(fp)
882
# readlines reads the whole file at once:
883
# bad for transports like http, good for local disk
884
# we save 60 ms doing this one change (
885
# from calling readline each time to calling
887
# probably what we want for nice behaviour on
888
# http is a incremental readlines that yields, or
889
# a check for local vs non local indexes,
890
for l in fp.readlines():
894
#pb.update('read knit index', count, total)
895
# See self._parse_parents
897
for value in rec[4:]:
899
# uncompressed reference
900
parents.append(value[1:])
902
# this is 15/4000ms faster than isinstance,
904
# this function is called thousands of times a
905
# second so small variations add up.
906
assert value.__class__ is str
907
parents.append(self._history[int(value)])
908
# end self._parse_parents
909
# self._cache_version(rec[0],
914
# --- self._cache_version
915
# only want the _history index to reference the 1st
916
# index entry for version_id
918
if version_id not in self._cache:
919
self._history.append(version_id)
920
self._cache[version_id] = (version_id,
925
# --- self._cache_version
926
except NoSuchFile, e:
927
if mode != 'w' or not create:
931
pb.update('read knit index', total, total)
934
def _parse_parents(self, compressed_parents):
935
"""convert a list of string parent values into version ids.
937
ints are looked up in the index.
938
.FOO values are ghosts and converted in to FOO.
940
NOTE: the function is retained here for clarity, and for possible
941
use in partial index reads. However bulk processing now has
942
it inlined in __init__ for inner-loop optimisation.
945
for value in compressed_parents:
947
# uncompressed reference
948
result.append(value[1:])
1162
self._transport.put_bytes_non_atomic(
1163
self._filename, self.HEADER, mode=self._file_mode)
950
# this is 15/4000ms faster than isinstance,
951
# this function is called thousands of times a
952
# second so small variations add up.
953
assert value.__class__ is str
954
result.append(self._history[int(value)])
1165
957
def get_graph(self):
1166
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
959
for version_id, index in self._cache.iteritems():
960
graph.append((version_id, index[4]))
1168
def get_ancestry(self, versions, topo_sorted=True):
963
def get_ancestry(self, versions):
1169
964
"""See VersionedFile.get_ancestry."""
1170
965
# get a graph of all the mentioned versions:
1172
967
pending = set(versions)
1175
969
version = pending.pop()
970
parents = self._cache[version][4]
1178
parents = [p for p in cache[version][4] if p in cache]
1180
raise RevisionNotPresent(version, self._filename)
1181
# if not completed and not a ghost
1182
pending.update([p for p in parents if p not in graph])
973
parents = [parent for parent in parents if parent in self._cache]
974
for parent in parents:
975
# if not completed and not a ghost
976
if parent not in graph:
1183
978
graph[version] = parents
1186
979
return topo_sort(graph.items())
1188
981
def get_ancestry_with_ghosts(self, versions):
1189
982
"""See VersionedFile.get_ancestry_with_ghosts."""
1190
983
# get a graph of all the mentioned versions:
1191
self.check_versions_present(versions)
1194
985
pending = set(versions)
1196
987
version = pending.pop()
1198
parents = cache[version][4]
989
parents = self._cache[version][4]
1199
990
except KeyError:
1200
991
# ghost, fake it
1201
992
graph[version] = []
1204
pending.update([p for p in parents if p not in graph])
996
for parent in parents:
997
if parent not in graph:
1205
999
graph[version] = parents
1206
1000
return topo_sort(graph.items())
1219
1013
def lookup(self, version_id):
1220
1014
assert version_id in self._cache
1221
return self._cache[version_id][5]
1015
return self._history.index(version_id)
1223
1017
def _version_list_to_index(self, versions):
1224
1018
result_list = []
1226
1019
for version in versions:
1227
if version in cache:
1228
# -- inlined lookup() --
1229
result_list.append(str(cache[version][5]))
1230
# -- end lookup () --
1020
if version in self._cache:
1021
result_list.append(str(self._history.index(version)))
1232
result_list.append('.' + version)
1023
result_list.append('.' + version.encode('utf-8'))
1233
1024
return ' '.join(result_list)
1235
1026
def add_version(self, version_id, options, pos, size, parents):
1236
1027
"""Add a version record to the index."""
1237
self.add_versions(((version_id, options, pos, size, parents),))
1239
def add_versions(self, versions):
1240
"""Add multiple versions to the index.
1242
:param versions: a list of tuples:
1243
(version_id, options, pos, size, parents).
1246
orig_history = self._history[:]
1247
orig_cache = self._cache.copy()
1250
for version_id, options, pos, size, parents in versions:
1251
line = "\n%s %s %s %s %s :" % (version_id,
1255
self._version_list_to_index(parents))
1256
assert isinstance(line, str), \
1257
'content must be utf-8 encoded: %r' % (line,)
1259
self._cache_version(version_id, options, pos, size, parents)
1260
if not self._need_to_create:
1261
self._transport.append_bytes(self._filename, ''.join(lines))
1264
sio.write(self.HEADER)
1265
sio.writelines(lines)
1267
self._transport.put_file_non_atomic(self._filename, sio,
1268
create_parent_dir=self._create_parent_dir,
1269
mode=self._file_mode,
1270
dir_mode=self._dir_mode)
1271
self._need_to_create = False
1273
# If any problems happen, restore the original values and re-raise
1274
self._history = orig_history
1275
self._cache = orig_cache
1028
self._cache_version(version_id, options, pos, size, parents)
1030
content = "%s %s %s %s %s\n" % (version_id.encode('utf-8'),
1034
self._version_list_to_index(parents))
1035
assert isinstance(content, str), 'content must be utf-8 encoded'
1036
self._transport.append(self._filename, StringIO(content))
1278
1038
def has_version(self, version_id):
1279
1039
"""True if the version is in the index."""
1280
return version_id in self._cache
1040
return self._cache.has_key(version_id)
1282
1042
def get_position(self, version_id):
1283
1043
"""Return data position and size of specified version."""
1284
entry = self._cache[version_id]
1285
return entry[2], entry[3]
1044
return (self._cache[version_id][2], \
1045
self._cache[version_id][3])
1287
1047
def get_method(self, version_id):
1288
1048
"""Return compression method of specified version."""
1303
1062
if parent in self._cache]
1305
1064
def get_parents_with_ghosts(self, version_id):
1306
"""Return parents of specified version with ghosts."""
1065
"""Return parents of specified version wth ghosts."""
1307
1066
return self._cache[version_id][4]
1309
1068
def check_versions_present(self, version_ids):
1310
1069
"""Check that all specified versions are present."""
1312
for version_id in version_ids:
1313
if version_id not in cache:
1314
raise RevisionNotPresent(version_id, self._filename)
1070
version_ids = set(version_ids)
1071
for version_id in list(version_ids):
1072
if version_id in self._cache:
1073
version_ids.remove(version_id)
1075
raise RevisionNotPresent(list(version_ids)[0], self.filename)
1317
1078
class _KnitData(_KnitComponentFile):
1318
1079
"""Contents of the knit data file"""
1320
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1321
create_parent_dir=False, delay_create=False,
1323
_KnitComponentFile.__init__(self, transport, filename, mode,
1324
file_mode=file_mode,
1325
create_parent_dir=create_parent_dir,
1081
HEADER = "# bzr knit data 7\n"
1083
def __init__(self, transport, filename, mode, create=False):
1084
_KnitComponentFile.__init__(self, transport, filename, mode)
1327
1086
self._checked = False
1328
# TODO: jam 20060713 conceptually, this could spill to disk
1329
# if the cached size gets larger than a certain amount
1330
# but it complicates the model a bit, so for now just use
1331
# a simple dictionary
1333
self._do_cache = False
1336
self._need_to_create = create
1338
self._transport.put_bytes_non_atomic(self._filename, '',
1339
mode=self._file_mode)
1341
def enable_cache(self):
1342
"""Enable caching of reads."""
1343
self._do_cache = True
1088
self._transport.put(self._filename, StringIO(''))
1345
1091
def clear_cache(self):
1346
1092
"""Clear the record cache."""
1347
self._do_cache = False
1350
1095
def _open_file(self):
1352
return self._transport.get(self._filename)
1096
if self._file is None:
1098
self._file = self._transport.get(self._filename)
1357
1103
def _record_to_data(self, version_id, digest, lines):
1358
1104
"""Convert version_id, digest, lines into a raw data block.
1376
1120
return length, sio
1378
1122
def add_raw_record(self, raw_data):
1379
"""Append a prepared record to the data file.
1381
:return: the offset in the data file raw_data was written.
1123
"""Append a prepared record to the data file."""
1383
1124
assert isinstance(raw_data, str), 'data must be plain bytes'
1384
if not self._need_to_create:
1385
return self._transport.append_bytes(self._filename, raw_data)
1387
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1388
create_parent_dir=self._create_parent_dir,
1389
mode=self._file_mode,
1390
dir_mode=self._dir_mode)
1391
self._need_to_create = False
1125
start_pos = self._transport.append(self._filename, StringIO(raw_data))
1126
return start_pos, len(raw_data)
1394
1128
def add_record(self, version_id, digest, lines):
1395
1129
"""Write new text record to disk. Returns the position in the
1396
1130
file where it was written."""
1397
1131
size, sio = self._record_to_data(version_id, digest, lines)
1133
self._records[version_id] = (digest, lines)
1398
1134
# write to disk
1399
if not self._need_to_create:
1400
start_pos = self._transport.append_file(self._filename, sio)
1402
self._transport.put_file_non_atomic(self._filename, sio,
1403
create_parent_dir=self._create_parent_dir,
1404
mode=self._file_mode,
1405
dir_mode=self._dir_mode)
1406
self._need_to_create = False
1409
self._cache[version_id] = sio.getvalue()
1135
start_pos = self._transport.append(self._filename, sio)
1410
1136
return start_pos, size
1412
1138
def _parse_record_header(self, version_id, raw_data):
1416
1142
as (stream, header_record)
1418
1144
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1420
rec = self._check_header(version_id, df.readline())
1421
except Exception, e:
1422
raise KnitCorrupt(self._filename,
1423
"While reading {%s} got %s(%s)"
1424
% (version_id, e.__class__.__name__, str(e)))
1145
rec = df.readline().split()
1147
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
1148
if rec[1].decode('utf-8')!= version_id:
1149
raise KnitCorrupt(self._filename,
1150
'unexpected version, wanted %r, got %r' % (
1151
version_id, rec[1]))
1427
def _check_header(self, version_id, line):
1430
raise KnitCorrupt(self._filename,
1431
'unexpected number of elements in record header')
1432
if rec[1] != version_id:
1433
raise KnitCorrupt(self._filename,
1434
'unexpected version, wanted %r, got %r'
1435
% (version_id, rec[1]))
1438
1154
def _parse_record(self, version_id, data):
1440
# 4168 calls in 2880 217 internal
1441
# 4168 calls to _parse_record_header in 2121
1442
# 4168 calls to readlines in 330
1443
df = GzipFile(mode='rb', fileobj=StringIO(data))
1446
record_contents = df.readlines()
1447
except Exception, e:
1448
raise KnitCorrupt(self._filename,
1449
"While reading {%s} got %s(%s)"
1450
% (version_id, e.__class__.__name__, str(e)))
1451
header = record_contents.pop(0)
1452
rec = self._check_header(version_id, header)
1454
last_line = record_contents.pop()
1455
if len(record_contents) != int(rec[2]):
1456
raise KnitCorrupt(self._filename,
1457
'incorrect number of lines %s != %s'
1459
% (len(record_contents), int(rec[2]),
1461
if last_line != 'end %s\n' % rec[1]:
1462
raise KnitCorrupt(self._filename,
1463
'unexpected version end line %r, wanted %r'
1464
% (last_line, version_id))
1155
df, rec = self._parse_record_header(version_id, data)
1157
record_contents = self._read_record_contents(df, lines)
1159
if l.decode('utf-8') != 'end %s\n' % version_id:
1160
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
1466
1163
return record_contents, rec[3]
1165
def _read_record_contents(self, df, record_lines):
1166
"""Read and return n lines from datafile."""
1168
for i in range(record_lines):
1169
r.append(df.readline())
1468
1172
def read_records_iter_raw(self, records):
1469
1173
"""Read text records from data file and yield raw data.
1471
1175
This unpacks enough of the text record to validate the id is
1472
1176
as expected but thats all.
1178
It will actively recompress currently cached records on the
1179
basis that that is cheaper than I/O activity.
1182
for version_id, pos, size in records:
1183
if version_id not in self._records:
1184
needed_records.append((version_id, pos, size))
1474
1186
# setup an iterator of the external records:
1475
1187
# uses readv so nice and fast we hope.
1188
if len(needed_records):
1477
1189
# grab the disk data needed.
1479
# Don't check _cache if it is empty
1480
needed_offsets = [(pos, size) for version_id, pos, size
1482
if version_id not in self._cache]
1484
needed_offsets = [(pos, size) for version_id, pos, size
1487
raw_records = self._transport.readv(self._filename, needed_offsets)
1190
raw_records = self._transport.readv(self._filename,
1191
[(pos, size) for version_id, pos, size in needed_records])
1489
1193
for version_id, pos, size in records:
1490
if version_id in self._cache:
1491
# This data has already been validated
1492
data = self._cache[version_id]
1194
if version_id in self._records:
1195
# compress a new version
1196
size, sio = self._record_to_data(version_id,
1197
self._records[version_id][0],
1198
self._records[version_id][1])
1199
yield version_id, sio.getvalue()
1494
1201
pos, data = raw_records.next()
1496
self._cache[version_id] = data
1498
1202
# validate the header
1499
1203
df, rec = self._parse_record_header(version_id, data)
1501
yield version_id, data
1205
yield version_id, data
1503
1208
def read_records_iter(self, records):
1504
1209
"""Read text records from data file and yield result.
1506
The result will be returned in whatever is the fastest to read.
1507
Not by the order requested. Also, multiple requests for the same
1508
record will only yield 1 response.
1509
:param records: A list of (version_id, pos, len) entries
1510
:return: Yields (version_id, contents, digest) in the order
1511
read, not the order requested
1211
Each passed record is a tuple of (version_id, pos, len) and
1212
will be read in the given order. Yields (version_id,
1517
# Skip records we have alread seen
1518
yielded_records = set()
1519
needed_records = set()
1520
for record in records:
1521
if record[0] in self._cache:
1522
if record[0] in yielded_records:
1524
yielded_records.add(record[0])
1525
data = self._cache[record[0]]
1526
content, digest = self._parse_record(record[0], data)
1527
yield (record[0], content, digest)
1529
needed_records.add(record)
1530
needed_records = sorted(needed_records, key=operator.itemgetter(1))
1532
needed_records = sorted(set(records), key=operator.itemgetter(1))
1534
if not needed_records:
1537
# The transport optimizes the fetching as well
1538
# (ie, reads continuous ranges.)
1539
readv_response = self._transport.readv(self._filename,
1540
[(pos, size) for version_id, pos, size in needed_records])
1542
for (version_id, pos, size), (pos, data) in \
1543
izip(iter(needed_records), readv_response):
1544
content, digest = self._parse_record(version_id, data)
1546
self._cache[version_id] = data
1547
yield version_id, content, digest
1217
for version_id, pos, size in records:
1218
if version_id not in self._records:
1219
needed_records.append((version_id, pos, size))
1221
if len(needed_records):
1222
# We take it that the transport optimizes the fetching as good
1223
# as possible (ie, reads continous ranges.)
1224
response = self._transport.readv(self._filename,
1225
[(pos, size) for version_id, pos, size in needed_records])
1227
for (record_id, pos, size), (pos, data) in izip(iter(needed_records), response):
1228
content, digest = self._parse_record(record_id, data)
1229
self._records[record_id] = (digest, content)
1231
for version_id, pos, size in records:
1232
yield version_id, copy(self._records[version_id][1]), copy(self._records[version_id][0])
1549
1234
def read_records(self, records):
1550
1235
"""Read records into a dictionary."""
1551
1236
components = {}
1552
for record_id, content, digest in \
1553
self.read_records_iter(records):
1237
for record_id, content, digest in self.read_records_iter(records):
1554
1238
components[record_id] = (content, digest)
1555
1239
return components
1672
1363
InterVersionedFile.register_optimiser(InterKnit)
1675
class WeaveToKnit(InterVersionedFile):
1676
"""Optimised code paths for weave to knit operations."""
1678
_matching_file_from_factory = bzrlib.weave.WeaveFile
1679
_matching_file_to_factory = KnitVersionedFile
1682
def is_compatible(source, target):
1683
"""Be compatible with weaves to knits."""
1685
return (isinstance(source, bzrlib.weave.Weave) and
1686
isinstance(target, KnitVersionedFile))
1687
except AttributeError:
1690
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1691
"""See InterVersionedFile.join."""
1692
assert isinstance(self.source, bzrlib.weave.Weave)
1693
assert isinstance(self.target, KnitVersionedFile)
1695
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1700
pb = ui.ui_factory.nested_progress_bar()
1702
version_ids = list(version_ids)
1704
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1705
this_versions = set(self.target._index.get_versions())
1706
needed_versions = self.source_ancestry - this_versions
1707
cross_check_versions = self.source_ancestry.intersection(this_versions)
1708
mismatched_versions = set()
1709
for version in cross_check_versions:
1710
# scan to include needed parents.
1711
n1 = set(self.target.get_parents_with_ghosts(version))
1712
n2 = set(self.source.get_parents(version))
1713
# if all of n2's parents are in n1, then its fine.
1714
if n2.difference(n1):
1715
# FIXME TEST this check for cycles being introduced works
1716
# the logic is we have a cycle if in our graph we are an
1717
# ancestor of any of the n2 revisions.
1723
parent_ancestors = self.source.get_ancestry(parent)
1724
if version in parent_ancestors:
1725
raise errors.GraphCycleError([parent, version])
1726
# ensure this parent will be available later.
1727
new_parents = n2.difference(n1)
1728
needed_versions.update(new_parents.difference(this_versions))
1729
mismatched_versions.add(version)
1731
if not needed_versions and not mismatched_versions:
1733
full_list = topo_sort(self.source.get_graph())
1735
version_list = [i for i in full_list if (not self.target.has_version(i)
1736
and i in needed_versions)]
1740
total = len(version_list)
1741
for version_id in version_list:
1742
pb.update("Converting to knit", count, total)
1743
parents = self.source.get_parents(version_id)
1744
# check that its will be a consistent copy:
1745
for parent in parents:
1746
# if source has the parent, we must already have it
1747
assert (self.target.has_version(parent))
1748
self.target.add_lines(
1749
version_id, parents, self.source.get_lines(version_id))
1752
for version in mismatched_versions:
1753
# FIXME RBC 20060309 is this needed?
1754
n1 = set(self.target.get_parents_with_ghosts(version))
1755
n2 = set(self.source.get_parents(version))
1756
# write a combined record to our history preserving the current
1757
# parents as first in the list
1758
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1759
self.target.fix_parents(version, new_parents)
1765
InterVersionedFile.register_optimiser(WeaveToKnit)
1768
class KnitSequenceMatcher(difflib.SequenceMatcher):
1366
# make GzipFile faster:
1368
class GzipFile(gzip.GzipFile):
1369
"""Knit tuned version of GzipFile.
1371
This is based on the following lsprof stats:
1372
python 2.4 stock GzipFile write:
1373
58971 0 5644.3090 2721.4730 gzip:193(write)
1374
+58971 0 1159.5530 1159.5530 +<built-in method compress>
1375
+176913 0 987.0320 987.0320 +<len>
1376
+58971 0 423.1450 423.1450 +<zlib.crc32>
1377
+58971 0 353.1060 353.1060 +<method 'write' of 'cStringIO.
1379
tuned GzipFile write:
1380
58971 0 4477.2590 2103.1120 bzrlib.knit:1250(write)
1381
+58971 0 1297.7620 1297.7620 +<built-in method compress>
1382
+58971 0 406.2160 406.2160 +<zlib.crc32>
1383
+58971 0 341.9020 341.9020 +<method 'write' of 'cStringIO.
1385
+58971 0 328.2670 328.2670 +<len>
1388
Yes, its only 1.6 seconds, but they add up.
1391
def write(self, data):
1392
if self.mode != gzip.WRITE:
1394
raise IOError(errno.EBADF, "write() on read-only GzipFile object")
1396
if self.fileobj is None:
1397
raise ValueError, "write() on closed GzipFile object"
1398
data_len = len(data)
1400
self.size = self.size + data_len
1401
self.crc = zlib.crc32(data, self.crc)
1402
self.fileobj.write( self.compress.compress(data) )
1403
self.offset += data_len
1405
def writelines(self, lines):
1406
# profiling indicated a significant overhead
1407
# calling write for each line.
1408
# this batch call is a lot faster :).
1409
# (4 seconds to 1 seconds for the sample upgrades I was testing).
1410
self.write(''.join(lines))
1413
class SequenceMatcher(difflib.SequenceMatcher):
1769
1414
"""Knit tuned sequence matcher.
1771
1416
This is based on profiling of difflib which indicated some improvements
added
removed
bzrlib/knit.py
