1
# Copyright (C) 2005, 2006, 2007 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""Knit versionedfile implementation.
19
A knit is a versioned file implementation that supports efficient append only
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
end-marker; simply "end VERSION"
28
delta can be line or full contents.a
29
... the 8's there are the index number of the annotation.
30
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
8 e.set('executable', 'yes')
36
8 if elt.get('executable') == 'yes':
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
43
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
44
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
45
09:33 < lifeless> right
46
09:33 < jrydberg> lifeless: the position and size is the range in the data file
49
so the index sequence is the dictionary compressed sequence number used
50
in the deltas to provide line annotation
55
# 10:16 < lifeless> make partial index writes safe
56
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
57
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
59
# move sha1 out of the content so that join is faster at verifying parents
60
# record content length ?
64
from cStringIO import StringIO
66
from itertools import izip, chain
73
from bzrlib.lazy_import import lazy_import
74
lazy_import(globals(), """
89
from bzrlib.errors import (
97
RevisionAlreadyPresent,
99
from bzrlib.tuned_gzip import GzipFile
100
from bzrlib.osutils import (
105
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
106
from bzrlib.tsort import topo_sort
109
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
112
# TODO: Split out code specific to this format into an associated object.
114
# TODO: Can we put in some kind of value to check that the index and data
115
# files belong together?
117
# TODO: accommodate binaries, perhaps by storing a byte count
119
# TODO: function to check whole file
121
# TODO: atomically append data, then measure backwards from the cursor
122
# position after writing to work out where it was located. we may need to
123
# bypass python file buffering.
125
DATA_SUFFIX = '.knit'
126
INDEX_SUFFIX = '.kndx'
129
class KnitContent(object):
130
"""Content of a knit version to which deltas can be applied."""
132
def __init__(self, lines):
135
def annotate_iter(self):
136
"""Yield tuples of (origin, text) for each content line."""
137
return iter(self._lines)
140
"""Return a list of (origin, text) tuples."""
141
return list(self.annotate_iter())
143
def line_delta_iter(self, new_lines):
144
"""Generate line-based delta from this content to new_lines."""
145
new_texts = new_lines.text()
146
old_texts = self.text()
147
s = KnitSequenceMatcher(None, old_texts, new_texts)
148
for tag, i1, i2, j1, j2 in s.get_opcodes():
151
# ofrom, oto, length, data
152
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
154
def line_delta(self, new_lines):
155
return list(self.line_delta_iter(new_lines))
158
return [text for origin, text in self._lines]
161
return KnitContent(self._lines[:])
164
def get_line_delta_blocks(knit_delta, source, target):
165
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
166
target_len = len(target)
169
for s_begin, s_end, t_len, new_text in knit_delta:
170
true_n = s_begin - s_pos
173
# knit deltas do not provide reliable info about whether the
174
# last line of a file matches, due to eol handling.
175
if source[s_pos + n -1] != target[t_pos + n -1]:
178
yield s_pos, t_pos, n
179
t_pos += t_len + true_n
181
n = target_len - t_pos
183
if source[s_pos + n -1] != target[t_pos + n -1]:
186
yield s_pos, t_pos, n
187
yield s_pos + (target_len - t_pos), target_len, 0
190
class _KnitFactory(object):
191
"""Base factory for creating content objects."""
193
def make(self, lines, version_id):
194
num_lines = len(lines)
195
return KnitContent(zip([version_id] * num_lines, lines))
198
class KnitAnnotateFactory(_KnitFactory):
199
"""Factory for creating annotated Content objects."""
203
def parse_fulltext(self, content, version_id):
204
"""Convert fulltext to internal representation
206
fulltext content is of the format
207
revid(utf8) plaintext\n
208
internal representation is of the format:
211
# TODO: jam 20070209 The tests expect this to be returned as tuples,
212
# but the code itself doesn't really depend on that.
213
# Figure out a way to not require the overhead of turning the
214
# list back into tuples.
215
lines = [tuple(line.split(' ', 1)) for line in content]
216
return KnitContent(lines)
218
def parse_line_delta_iter(self, lines):
219
return iter(self.parse_line_delta(lines))
221
def parse_line_delta(self, lines, version_id):
222
"""Convert a line based delta into internal representation.
224
line delta is in the form of:
225
intstart intend intcount
227
revid(utf8) newline\n
228
internal representation is
229
(start, end, count, [1..count tuples (revid, newline)])
236
def cache_and_return(line):
237
origin, text = line.split(' ', 1)
238
return cache.setdefault(origin, origin), text
240
# walk through the lines parsing.
242
start, end, count = [int(n) for n in header.split(',')]
243
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
244
result.append((start, end, count, contents))
247
def get_fulltext_content(self, lines):
248
"""Extract just the content lines from a fulltext."""
249
return (line.split(' ', 1)[1] for line in lines)
251
def get_linedelta_content(self, lines):
252
"""Extract just the content from a line delta.
254
This doesn't return all of the extra information stored in a delta.
255
Only the actual content lines.
260
header = header.split(',')
261
count = int(header[2])
262
for i in xrange(count):
263
origin, text = next().split(' ', 1)
266
def lower_fulltext(self, content):
267
"""convert a fulltext content record into a serializable form.
269
see parse_fulltext which this inverts.
271
# TODO: jam 20070209 We only do the caching thing to make sure that
272
# the origin is a valid utf-8 line, eventually we could remove it
273
return ['%s %s' % (o, t) for o, t in content._lines]
275
def lower_line_delta(self, delta):
276
"""convert a delta into a serializable form.
278
See parse_line_delta which this inverts.
280
# TODO: jam 20070209 We only do the caching thing to make sure that
281
# the origin is a valid utf-8 line, eventually we could remove it
283
for start, end, c, lines in delta:
284
out.append('%d,%d,%d\n' % (start, end, c))
285
out.extend(origin + ' ' + text
286
for origin, text in lines)
290
class KnitPlainFactory(_KnitFactory):
291
"""Factory for creating plain Content objects."""
295
def parse_fulltext(self, content, version_id):
296
"""This parses an unannotated fulltext.
298
Note that this is not a noop - the internal representation
299
has (versionid, line) - its just a constant versionid.
301
return self.make(content, version_id)
303
def parse_line_delta_iter(self, lines, version_id):
305
num_lines = len(lines)
306
while cur < num_lines:
309
start, end, c = [int(n) for n in header.split(',')]
310
yield start, end, c, zip([version_id] * c, lines[cur:cur+c])
313
def parse_line_delta(self, lines, version_id):
314
return list(self.parse_line_delta_iter(lines, version_id))
316
def get_fulltext_content(self, lines):
317
"""Extract just the content lines from a fulltext."""
320
def get_linedelta_content(self, lines):
321
"""Extract just the content from a line delta.
323
This doesn't return all of the extra information stored in a delta.
324
Only the actual content lines.
329
header = header.split(',')
330
count = int(header[2])
331
for i in xrange(count):
334
def lower_fulltext(self, content):
335
return content.text()
337
def lower_line_delta(self, delta):
339
for start, end, c, lines in delta:
340
out.append('%d,%d,%d\n' % (start, end, c))
341
out.extend([text for origin, text in lines])
345
def make_empty_knit(transport, relpath):
346
"""Construct a empty knit at the specified location."""
347
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
350
class KnitVersionedFile(VersionedFile):
351
"""Weave-like structure with faster random access.
353
A knit stores a number of texts and a summary of the relationships
354
between them. Texts are identified by a string version-id. Texts
355
are normally stored and retrieved as a series of lines, but can
356
also be passed as single strings.
358
Lines are stored with the trailing newline (if any) included, to
359
avoid special cases for files with no final newline. Lines are
360
composed of 8-bit characters, not unicode. The combination of
361
these approaches should mean any 'binary' file can be safely
362
stored and retrieved.
365
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
366
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
367
create=False, create_parent_dir=False, delay_create=False,
368
dir_mode=None, index=None, access_method=None):
369
"""Construct a knit at location specified by relpath.
371
:param create: If not True, only open an existing knit.
372
:param create_parent_dir: If True, create the parent directory if
373
creating the file fails. (This is used for stores with
374
hash-prefixes that may not exist yet)
375
:param delay_create: The calling code is aware that the knit won't
376
actually be created until the first data is stored.
377
:param index: An index to use for the knit.
379
if deprecated_passed(basis_knit):
380
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
381
" deprecated as of bzr 0.9.",
382
DeprecationWarning, stacklevel=2)
383
if access_mode is None:
385
super(KnitVersionedFile, self).__init__(access_mode)
386
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
387
self.transport = transport
388
self.filename = relpath
389
self.factory = factory or KnitAnnotateFactory()
390
self.writable = (access_mode == 'w')
393
self._max_delta_chain = 200
396
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
397
access_mode, create=create, file_mode=file_mode,
398
create_parent_dir=create_parent_dir, delay_create=delay_create,
402
if access_method is None:
403
_access = _KnitAccess(transport, relpath + DATA_SUFFIX, file_mode, dir_mode,
404
((create and not len(self)) and delay_create), create_parent_dir)
406
_access = access_method
407
if create and not len(self) and not delay_create:
409
self._data = _KnitData(_access)
412
return '%s(%s)' % (self.__class__.__name__,
413
self.transport.abspath(self.filename))
415
def _check_should_delta(self, first_parents):
416
"""Iterate back through the parent listing, looking for a fulltext.
418
This is used when we want to decide whether to add a delta or a new
419
fulltext. It searches for _max_delta_chain parents. When it finds a
420
fulltext parent, it sees if the total size of the deltas leading up to
421
it is large enough to indicate that we want a new full text anyway.
423
Return True if we should create a new delta, False if we should use a
428
delta_parents = first_parents
429
for count in xrange(self._max_delta_chain):
430
parent = delta_parents[0]
431
method = self._index.get_method(parent)
432
index, pos, size = self._index.get_position(parent)
433
if method == 'fulltext':
437
delta_parents = self._index.get_parents(parent)
439
# We couldn't find a fulltext, so we must create a new one
442
return fulltext_size > delta_size
444
def _add_delta(self, version_id, parents, delta_parent, sha1, noeol, delta):
445
"""See VersionedFile._add_delta()."""
446
self._check_add(version_id, []) # should we check the lines ?
447
self._check_versions_present(parents)
451
for parent in parents:
452
if not self.has_version(parent):
453
ghosts.append(parent)
455
present_parents.append(parent)
457
if delta_parent is None:
458
# reconstitute as full text.
459
assert len(delta) == 1 or len(delta) == 0
461
assert delta[0][0] == 0
462
assert delta[0][1] == 0, delta[0][1]
463
return super(KnitVersionedFile, self)._add_delta(version_id,
474
options.append('no-eol')
476
if delta_parent is not None:
477
# determine the current delta chain length.
478
# To speed the extract of texts the delta chain is limited
479
# to a fixed number of deltas. This should minimize both
480
# I/O and the time spend applying deltas.
481
# The window was changed to a maximum of 200 deltas, but also added
482
# was a check that the total compressed size of the deltas is
483
# smaller than the compressed size of the fulltext.
484
if not self._check_should_delta([delta_parent]):
485
# We don't want a delta here, just do a normal insertion.
486
return super(KnitVersionedFile, self)._add_delta(version_id,
493
options.append('line-delta')
494
store_lines = self.factory.lower_line_delta(delta)
496
access_memo = self._data.add_record(version_id, digest, store_lines)
497
self._index.add_version(version_id, options, access_memo, parents)
499
def _add_raw_records(self, records, data):
500
"""Add all the records 'records' with data pre-joined in 'data'.
502
:param records: A list of tuples(version_id, options, parents, size).
503
:param data: The data for the records. When it is written, the records
504
are adjusted to have pos pointing into data by the sum of
505
the preceding records sizes.
508
raw_record_sizes = [record[3] for record in records]
509
positions = self._data.add_raw_records(raw_record_sizes, data)
512
for (version_id, options, parents, size), access_memo in zip(
514
index_entries.append((version_id, options, access_memo, parents))
515
if self._data._do_cache:
516
self._data._cache[version_id] = data[offset:offset+size]
518
self._index.add_versions(index_entries)
520
def enable_cache(self):
521
"""Start caching data for this knit"""
522
self._data.enable_cache()
524
def clear_cache(self):
525
"""Clear the data cache only."""
526
self._data.clear_cache()
528
def copy_to(self, name, transport):
529
"""See VersionedFile.copy_to()."""
530
# copy the current index to a temp index to avoid racing with local
532
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
533
self.transport.get(self._index._filename))
535
f = self._data._open_file()
537
transport.put_file(name + DATA_SUFFIX, f)
540
# move the copied index into place
541
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
543
def create_empty(self, name, transport, mode=None):
544
return KnitVersionedFile(name, transport, factory=self.factory,
545
delta=self.delta, create=True)
547
def _fix_parents(self, version_id, new_parents):
548
"""Fix the parents list for version.
550
This is done by appending a new version to the index
551
with identical data except for the parents list.
552
the parents list must be a superset of the current
555
current_values = self._index._cache[version_id]
556
assert set(current_values[4]).difference(set(new_parents)) == set()
557
self._index.add_version(version_id,
559
(None, current_values[2], current_values[3]),
562
def _extract_blocks(self, version_id, source, target):
563
if self._index.get_method(version_id) != 'line-delta':
565
parent, sha1, noeol, delta = self.get_delta(version_id)
566
return KnitContent.get_line_delta_blocks(delta, source, target)
568
def get_delta(self, version_id):
569
"""Get a delta for constructing version from some other version."""
570
version_id = osutils.safe_revision_id(version_id)
571
self.check_not_reserved_id(version_id)
572
if not self.has_version(version_id):
573
raise RevisionNotPresent(version_id, self.filename)
575
parents = self.get_parents(version_id)
580
index_memo = self._index.get_position(version_id)
581
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
582
noeol = 'no-eol' in self._index.get_options(version_id)
583
if 'fulltext' == self._index.get_method(version_id):
584
new_content = self.factory.parse_fulltext(data, version_id)
585
if parent is not None:
586
reference_content = self._get_content(parent)
587
old_texts = reference_content.text()
590
new_texts = new_content.text()
591
delta_seq = KnitSequenceMatcher(None, old_texts, new_texts)
592
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
594
delta = self.factory.parse_line_delta(data, version_id)
595
return parent, sha1, noeol, delta
597
def get_graph_with_ghosts(self):
598
"""See VersionedFile.get_graph_with_ghosts()."""
599
graph_items = self._index.get_graph()
600
return dict(graph_items)
602
def get_sha1(self, version_id):
603
return self.get_sha1s([version_id])[0]
605
def get_sha1s(self, version_ids):
606
"""See VersionedFile.get_sha1()."""
607
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
608
record_map = self._get_record_map(version_ids)
609
# record entry 2 is the 'digest'.
610
return [record_map[v][2] for v in version_ids]
614
"""See VersionedFile.get_suffixes()."""
615
return [DATA_SUFFIX, INDEX_SUFFIX]
617
def has_ghost(self, version_id):
618
"""True if there is a ghost reference in the file to version_id."""
619
version_id = osutils.safe_revision_id(version_id)
621
if self.has_version(version_id):
623
# optimisable if needed by memoising the _ghosts set.
624
items = self._index.get_graph()
625
for node, parents in items:
626
for parent in parents:
627
if parent not in self._index._cache:
628
if parent == version_id:
633
"""See VersionedFile.versions."""
634
return self._index.get_versions()
636
def has_version(self, version_id):
637
"""See VersionedFile.has_version."""
638
version_id = osutils.safe_revision_id(version_id)
639
return self._index.has_version(version_id)
641
__contains__ = has_version
643
def _merge_annotations(self, content, parents, parent_texts={},
644
delta=None, annotated=None,
645
left_matching_blocks=None):
646
"""Merge annotations for content. This is done by comparing
647
the annotations based on changed to the text.
649
if left_matching_blocks is not None:
650
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
654
for parent_id in parents:
655
merge_content = self._get_content(parent_id, parent_texts)
656
if (parent_id == parents[0] and delta_seq is not None):
659
seq = patiencediff.PatienceSequenceMatcher(
660
None, merge_content.text(), content.text())
661
for i, j, n in seq.get_matching_blocks():
664
# this appears to copy (origin, text) pairs across to the
665
# new content for any line that matches the last-checked
667
content._lines[j:j+n] = merge_content._lines[i:i+n]
669
if delta_seq is None:
670
reference_content = self._get_content(parents[0], parent_texts)
671
new_texts = content.text()
672
old_texts = reference_content.text()
673
delta_seq = patiencediff.PatienceSequenceMatcher(
674
None, old_texts, new_texts)
675
return self._make_line_delta(delta_seq, content)
677
def _make_line_delta(self, delta_seq, new_content):
678
"""Generate a line delta from delta_seq and new_content."""
680
for op in delta_seq.get_opcodes():
683
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
686
def _get_components_positions(self, version_ids):
687
"""Produce a map of position data for the components of versions.
689
This data is intended to be used for retrieving the knit records.
691
A dict of version_id to (method, data_pos, data_size, next) is
693
method is the way referenced data should be applied.
694
data_pos is the position of the data in the knit.
695
data_size is the size of the data in the knit.
696
next is the build-parent of the version, or None for fulltexts.
699
for version_id in version_ids:
702
while cursor is not None and cursor not in component_data:
703
method = self._index.get_method(cursor)
704
if method == 'fulltext':
707
next = self.get_parents(cursor)[0]
708
index_memo = self._index.get_position(cursor)
709
component_data[cursor] = (method, index_memo, next)
711
return component_data
713
def _get_content(self, version_id, parent_texts={}):
714
"""Returns a content object that makes up the specified
716
if not self.has_version(version_id):
717
raise RevisionNotPresent(version_id, self.filename)
719
cached_version = parent_texts.get(version_id, None)
720
if cached_version is not None:
721
return cached_version
723
text_map, contents_map = self._get_content_maps([version_id])
724
return contents_map[version_id]
726
def _check_versions_present(self, version_ids):
727
"""Check that all specified versions are present."""
728
self._index.check_versions_present(version_ids)
730
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts):
731
"""See VersionedFile.add_lines_with_ghosts()."""
732
self._check_add(version_id, lines)
733
return self._add(version_id, lines[:], parents, self.delta, parent_texts)
735
def _add_lines(self, version_id, parents, lines, parent_texts,
736
left_matching_blocks=None):
737
"""See VersionedFile.add_lines."""
738
self._check_add(version_id, lines)
739
self._check_versions_present(parents)
740
return self._add(version_id, lines[:], parents, self.delta,
741
parent_texts, left_matching_blocks)
743
def _check_add(self, version_id, lines):
744
"""check that version_id and lines are safe to add."""
745
assert self.writable, "knit is not opened for write"
746
### FIXME escape. RBC 20060228
747
if contains_whitespace(version_id):
748
raise InvalidRevisionId(version_id, self.filename)
749
self.check_not_reserved_id(version_id)
750
if self.has_version(version_id):
751
raise RevisionAlreadyPresent(version_id, self.filename)
752
self._check_lines_not_unicode(lines)
753
self._check_lines_are_lines(lines)
755
def _add(self, version_id, lines, parents, delta, parent_texts,
756
left_matching_blocks=None):
757
"""Add a set of lines on top of version specified by parents.
759
If delta is true, compress the text as a line-delta against
762
Any versions not present will be converted into ghosts.
764
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
765
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
766
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
767
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
768
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
769
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
770
# +1383 0 8.0370 8.0370 +<len>
771
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
772
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
773
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
774
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
778
if parent_texts is None:
780
for parent in parents:
781
if not self.has_version(parent):
782
ghosts.append(parent)
784
present_parents.append(parent)
786
if delta and not len(present_parents):
789
digest = sha_strings(lines)
792
if lines[-1][-1] != '\n':
793
options.append('no-eol')
794
lines[-1] = lines[-1] + '\n'
796
if len(present_parents) and delta:
797
# To speed the extract of texts the delta chain is limited
798
# to a fixed number of deltas. This should minimize both
799
# I/O and the time spend applying deltas.
800
delta = self._check_should_delta(present_parents)
802
assert isinstance(version_id, str)
803
lines = self.factory.make(lines, version_id)
804
if delta or (self.factory.annotated and len(present_parents) > 0):
805
# Merge annotations from parent texts if so is needed.
806
delta_hunks = self._merge_annotations(lines, present_parents,
807
parent_texts, delta, self.factory.annotated,
808
left_matching_blocks)
811
options.append('line-delta')
812
store_lines = self.factory.lower_line_delta(delta_hunks)
814
options.append('fulltext')
815
store_lines = self.factory.lower_fulltext(lines)
817
access_memo = self._data.add_record(version_id, digest, store_lines)
818
self._index.add_version(version_id, options, access_memo, parents)
821
def check(self, progress_bar=None):
822
"""See VersionedFile.check()."""
824
def _clone_text(self, new_version_id, old_version_id, parents):
825
"""See VersionedFile.clone_text()."""
826
# FIXME RBC 20060228 make fast by only inserting an index with null
828
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
830
def get_lines(self, version_id):
831
"""See VersionedFile.get_lines()."""
832
return self.get_line_list([version_id])[0]
834
def _get_record_map(self, version_ids):
835
"""Produce a dictionary of knit records.
837
The keys are version_ids, the values are tuples of (method, content,
839
method is the way the content should be applied.
840
content is a KnitContent object.
841
digest is the SHA1 digest of this version id after all steps are done
842
next is the build-parent of the version, i.e. the leftmost ancestor.
843
If the method is fulltext, next will be None.
845
position_map = self._get_components_positions(version_ids)
846
# c = component_id, m = method, i_m = index_memo, n = next
847
records = [(c, i_m) for c, (m, i_m, n) in position_map.iteritems()]
849
for component_id, content, digest in \
850
self._data.read_records_iter(records):
851
method, index_memo, next = position_map[component_id]
852
record_map[component_id] = method, content, digest, next
856
def get_text(self, version_id):
857
"""See VersionedFile.get_text"""
858
return self.get_texts([version_id])[0]
860
def get_texts(self, version_ids):
861
return [''.join(l) for l in self.get_line_list(version_ids)]
863
def get_line_list(self, version_ids):
864
"""Return the texts of listed versions as a list of strings."""
865
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
866
for version_id in version_ids:
867
self.check_not_reserved_id(version_id)
868
text_map, content_map = self._get_content_maps(version_ids)
869
return [text_map[v] for v in version_ids]
871
_get_lf_split_line_list = get_line_list
873
def _get_content_maps(self, version_ids):
874
"""Produce maps of text and KnitContents
876
:return: (text_map, content_map) where text_map contains the texts for
877
the requested versions and content_map contains the KnitContents.
878
Both dicts take version_ids as their keys.
880
for version_id in version_ids:
881
if not self.has_version(version_id):
882
raise RevisionNotPresent(version_id, self.filename)
883
record_map = self._get_record_map(version_ids)
888
for version_id in version_ids:
891
while cursor is not None:
892
method, data, digest, next = record_map[cursor]
893
components.append((cursor, method, data, digest))
894
if cursor in content_map:
899
for component_id, method, data, digest in reversed(components):
900
if component_id in content_map:
901
content = content_map[component_id]
903
if method == 'fulltext':
904
assert content is None
905
content = self.factory.parse_fulltext(data, version_id)
906
elif method == 'line-delta':
907
delta = self.factory.parse_line_delta(data, version_id)
908
content = content.copy()
909
content._lines = self._apply_delta(content._lines,
911
content_map[component_id] = content
913
if 'no-eol' in self._index.get_options(version_id):
914
content = content.copy()
915
line = content._lines[-1][1].rstrip('\n')
916
content._lines[-1] = (content._lines[-1][0], line)
917
final_content[version_id] = content
919
# digest here is the digest from the last applied component.
920
text = content.text()
921
if sha_strings(text) != digest:
922
raise KnitCorrupt(self.filename,
923
'sha-1 does not match %s' % version_id)
925
text_map[version_id] = text
926
return text_map, final_content
928
def iter_lines_added_or_present_in_versions(self, version_ids=None,
930
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
931
if version_ids is None:
932
version_ids = self.versions()
934
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
936
pb = progress.DummyProgress()
937
# we don't care about inclusions, the caller cares.
938
# but we need to setup a list of records to visit.
939
# we need version_id, position, length
940
version_id_records = []
941
requested_versions = set(version_ids)
942
# filter for available versions
943
for version_id in requested_versions:
944
if not self.has_version(version_id):
945
raise RevisionNotPresent(version_id, self.filename)
946
# get a in-component-order queue:
947
for version_id in self.versions():
948
if version_id in requested_versions:
949
index_memo = self._index.get_position(version_id)
950
version_id_records.append((version_id, index_memo))
952
total = len(version_id_records)
953
for version_idx, (version_id, data, sha_value) in \
954
enumerate(self._data.read_records_iter(version_id_records)):
955
pb.update('Walking content.', version_idx, total)
956
method = self._index.get_method(version_id)
958
assert method in ('fulltext', 'line-delta')
959
if method == 'fulltext':
960
line_iterator = self.factory.get_fulltext_content(data)
962
line_iterator = self.factory.get_linedelta_content(data)
963
for line in line_iterator:
966
pb.update('Walking content.', total, total)
968
def iter_parents(self, version_ids):
969
"""Iterate through the parents for many version ids.
971
:param version_ids: An iterable yielding version_ids.
972
:return: An iterator that yields (version_id, parents). Requested
973
version_ids not present in the versioned file are simply skipped.
974
The order is undefined, allowing for different optimisations in
975
the underlying implementation.
977
version_ids = [osutils.safe_revision_id(version_id) for
978
version_id in version_ids]
979
return self._index.iter_parents(version_ids)
981
def num_versions(self):
982
"""See VersionedFile.num_versions()."""
983
return self._index.num_versions()
985
__len__ = num_versions
987
def annotate_iter(self, version_id):
988
"""See VersionedFile.annotate_iter."""
989
version_id = osutils.safe_revision_id(version_id)
990
content = self._get_content(version_id)
991
for origin, text in content.annotate_iter():
994
def get_parents(self, version_id):
995
"""See VersionedFile.get_parents."""
998
# 52554 calls in 1264 872 internal down from 3674
999
version_id = osutils.safe_revision_id(version_id)
1001
return self._index.get_parents(version_id)
1003
raise RevisionNotPresent(version_id, self.filename)
1005
def get_parents_with_ghosts(self, version_id):
1006
"""See VersionedFile.get_parents."""
1007
version_id = osutils.safe_revision_id(version_id)
1009
return self._index.get_parents_with_ghosts(version_id)
1011
raise RevisionNotPresent(version_id, self.filename)
1013
def get_ancestry(self, versions, topo_sorted=True):
1014
"""See VersionedFile.get_ancestry."""
1015
if isinstance(versions, basestring):
1016
versions = [versions]
1019
versions = [osutils.safe_revision_id(v) for v in versions]
1020
return self._index.get_ancestry(versions, topo_sorted)
1022
def get_ancestry_with_ghosts(self, versions):
1023
"""See VersionedFile.get_ancestry_with_ghosts."""
1024
if isinstance(versions, basestring):
1025
versions = [versions]
1028
versions = [osutils.safe_revision_id(v) for v in versions]
1029
return self._index.get_ancestry_with_ghosts(versions)
1031
def plan_merge(self, ver_a, ver_b):
1032
"""See VersionedFile.plan_merge."""
1033
ver_a = osutils.safe_revision_id(ver_a)
1034
ver_b = osutils.safe_revision_id(ver_b)
1035
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1037
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1038
annotated_a = self.annotate(ver_a)
1039
annotated_b = self.annotate(ver_b)
1040
return merge._plan_annotate_merge(annotated_a, annotated_b,
1041
ancestors_a, ancestors_b)
1044
class _KnitComponentFile(object):
1045
"""One of the files used to implement a knit database"""
1047
def __init__(self, transport, filename, mode, file_mode=None,
1048
create_parent_dir=False, dir_mode=None):
1049
self._transport = transport
1050
self._filename = filename
1052
self._file_mode = file_mode
1053
self._dir_mode = dir_mode
1054
self._create_parent_dir = create_parent_dir
1055
self._need_to_create = False
1057
def _full_path(self):
1058
"""Return the full path to this file."""
1059
return self._transport.base + self._filename
1061
def check_header(self, fp):
1062
line = fp.readline()
1064
# An empty file can actually be treated as though the file doesn't
1066
raise errors.NoSuchFile(self._full_path())
1067
if line != self.HEADER:
1068
raise KnitHeaderError(badline=line,
1069
filename=self._transport.abspath(self._filename))
1072
return '%s(%s)' % (self.__class__.__name__, self._filename)
1075
class _KnitIndex(_KnitComponentFile):
1076
"""Manages knit index file.
1078
The index is already kept in memory and read on startup, to enable
1079
fast lookups of revision information. The cursor of the index
1080
file is always pointing to the end, making it easy to append
1083
_cache is a cache for fast mapping from version id to a Index
1086
_history is a cache for fast mapping from indexes to version ids.
1088
The index data format is dictionary compressed when it comes to
1089
parent references; a index entry may only have parents that with a
1090
lover index number. As a result, the index is topological sorted.
1092
Duplicate entries may be written to the index for a single version id
1093
if this is done then the latter one completely replaces the former:
1094
this allows updates to correct version and parent information.
1095
Note that the two entries may share the delta, and that successive
1096
annotations and references MUST point to the first entry.
1098
The index file on disc contains a header, followed by one line per knit
1099
record. The same revision can be present in an index file more than once.
1100
The first occurrence gets assigned a sequence number starting from 0.
1102
The format of a single line is
1103
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1104
REVISION_ID is a utf8-encoded revision id
1105
FLAGS is a comma separated list of flags about the record. Values include
1106
no-eol, line-delta, fulltext.
1107
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1108
that the the compressed data starts at.
1109
LENGTH is the ascii representation of the length of the data file.
1110
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1112
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1113
revision id already in the knit that is a parent of REVISION_ID.
1114
The ' :' marker is the end of record marker.
1117
when a write is interrupted to the index file, it will result in a line
1118
that does not end in ' :'. If the ' :' is not present at the end of a line,
1119
or at the end of the file, then the record that is missing it will be
1120
ignored by the parser.
1122
When writing new records to the index file, the data is preceded by '\n'
1123
to ensure that records always start on new lines even if the last write was
1124
interrupted. As a result its normal for the last line in the index to be
1125
missing a trailing newline. One can be added with no harmful effects.
1128
HEADER = "# bzr knit index 8\n"
1130
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1131
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1133
def _cache_version(self, version_id, options, pos, size, parents):
1134
"""Cache a version record in the history array and index cache.
1136
This is inlined into _load_data for performance. KEEP IN SYNC.
1137
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1140
# only want the _history index to reference the 1st index entry
1142
if version_id not in self._cache:
1143
index = len(self._history)
1144
self._history.append(version_id)
1146
index = self._cache[version_id][5]
1147
self._cache[version_id] = (version_id,
1154
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1155
create_parent_dir=False, delay_create=False, dir_mode=None):
1156
_KnitComponentFile.__init__(self, transport, filename, mode,
1157
file_mode=file_mode,
1158
create_parent_dir=create_parent_dir,
1161
# position in _history is the 'official' index for a revision
1162
# but the values may have come from a newer entry.
1163
# so - wc -l of a knit index is != the number of unique names
1167
fp = self._transport.get(self._filename)
1169
# _load_data may raise NoSuchFile if the target knit is
1171
_load_data(self, fp)
1175
if mode != 'w' or not create:
1178
self._need_to_create = True
1180
self._transport.put_bytes_non_atomic(
1181
self._filename, self.HEADER, mode=self._file_mode)
1183
def get_graph(self):
1184
"""Return a list of the node:parents lists from this knit index."""
1185
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1187
def get_ancestry(self, versions, topo_sorted=True):
1188
"""See VersionedFile.get_ancestry."""
1189
# get a graph of all the mentioned versions:
1191
pending = set(versions)
1194
version = pending.pop()
1197
parents = [p for p in cache[version][4] if p in cache]
1199
raise RevisionNotPresent(version, self._filename)
1200
# if not completed and not a ghost
1201
pending.update([p for p in parents if p not in graph])
1202
graph[version] = parents
1205
return topo_sort(graph.items())
1207
def get_ancestry_with_ghosts(self, versions):
1208
"""See VersionedFile.get_ancestry_with_ghosts."""
1209
# get a graph of all the mentioned versions:
1210
self.check_versions_present(versions)
1213
pending = set(versions)
1215
version = pending.pop()
1217
parents = cache[version][4]
1223
pending.update([p for p in parents if p not in graph])
1224
graph[version] = parents
1225
return topo_sort(graph.items())
1227
def iter_parents(self, version_ids):
1228
"""Iterate through the parents for many version ids.
1230
:param version_ids: An iterable yielding version_ids.
1231
:return: An iterator that yields (version_id, parents). Requested
1232
version_ids not present in the versioned file are simply skipped.
1233
The order is undefined, allowing for different optimisations in
1234
the underlying implementation.
1236
for version_id in version_ids:
1238
yield version_id, tuple(self.get_parents(version_id))
1242
def num_versions(self):
1243
return len(self._history)
1245
__len__ = num_versions
1247
def get_versions(self):
1248
"""Get all the versions in the file. not topologically sorted."""
1249
return self._history
1251
def _version_list_to_index(self, versions):
1254
for version in versions:
1255
if version in cache:
1256
# -- inlined lookup() --
1257
result_list.append(str(cache[version][5]))
1258
# -- end lookup () --
1260
result_list.append('.' + version)
1261
return ' '.join(result_list)
1263
def add_version(self, version_id, options, index_memo, parents):
1264
"""Add a version record to the index."""
1265
self.add_versions(((version_id, options, index_memo, parents),))
1267
def add_versions(self, versions):
1268
"""Add multiple versions to the index.
1270
:param versions: a list of tuples:
1271
(version_id, options, pos, size, parents).
1274
orig_history = self._history[:]
1275
orig_cache = self._cache.copy()
1278
for version_id, options, (index, pos, size), parents in versions:
1279
line = "\n%s %s %s %s %s :" % (version_id,
1283
self._version_list_to_index(parents))
1284
assert isinstance(line, str), \
1285
'content must be utf-8 encoded: %r' % (line,)
1287
self._cache_version(version_id, options, pos, size, parents)
1288
if not self._need_to_create:
1289
self._transport.append_bytes(self._filename, ''.join(lines))
1292
sio.write(self.HEADER)
1293
sio.writelines(lines)
1295
self._transport.put_file_non_atomic(self._filename, sio,
1296
create_parent_dir=self._create_parent_dir,
1297
mode=self._file_mode,
1298
dir_mode=self._dir_mode)
1299
self._need_to_create = False
1301
# If any problems happen, restore the original values and re-raise
1302
self._history = orig_history
1303
self._cache = orig_cache
1306
def has_version(self, version_id):
1307
"""True if the version is in the index."""
1308
return version_id in self._cache
1310
def get_position(self, version_id):
1311
"""Return details needed to access the version.
1313
.kndx indices do not support split-out data, so return None for the
1316
:return: a tuple (None, data position, size) to hand to the access
1317
logic to get the record.
1319
entry = self._cache[version_id]
1320
return None, entry[2], entry[3]
1322
def get_method(self, version_id):
1323
"""Return compression method of specified version."""
1324
options = self._cache[version_id][1]
1325
if 'fulltext' in options:
1328
if 'line-delta' not in options:
1329
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1332
def get_options(self, version_id):
1333
"""Return a string represention options.
1337
return self._cache[version_id][1]
1339
def get_parents(self, version_id):
1340
"""Return parents of specified version ignoring ghosts."""
1341
return [parent for parent in self._cache[version_id][4]
1342
if parent in self._cache]
1344
def get_parents_with_ghosts(self, version_id):
1345
"""Return parents of specified version with ghosts."""
1346
return self._cache[version_id][4]
1348
def check_versions_present(self, version_ids):
1349
"""Check that all specified versions are present."""
1351
for version_id in version_ids:
1352
if version_id not in cache:
1353
raise RevisionNotPresent(version_id, self._filename)
1356
class KnitGraphIndex(object):
1357
"""A knit index that builds on GraphIndex."""
1359
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1360
"""Construct a KnitGraphIndex on a graph_index.
1362
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1363
:param deltas: Allow delta-compressed records.
1364
:param add_callback: If not None, allow additions to the index and call
1365
this callback with a list of added GraphIndex nodes:
1366
[(node, value, node_refs), ...]
1367
:param parents: If True, record knits parents, if not do not record
1370
self._graph_index = graph_index
1371
self._deltas = deltas
1372
self._add_callback = add_callback
1373
self._parents = parents
1374
if deltas and not parents:
1375
raise KnitCorrupt(self, "Cannot do delta compression without "
1378
def _get_entries(self, keys, check_present=False):
1379
"""Get the entries for keys.
1381
:param keys: An iterable of index keys, - 1-tuples.
1386
for node in self._graph_index.iter_entries(keys):
1388
found_keys.add(node[1])
1390
# adapt parentless index to the rest of the code.
1391
for node in self._graph_index.iter_entries(keys):
1392
yield node[0], node[1], node[2], ()
1393
found_keys.add(node[1])
1395
missing_keys = keys.difference(found_keys)
1397
raise RevisionNotPresent(missing_keys.pop(), self)
1399
def _present_keys(self, version_ids):
1401
node[1] for node in self._get_entries(version_ids)])
1403
def _parentless_ancestry(self, versions):
1404
"""Honour the get_ancestry API for parentless knit indices."""
1405
wanted_keys = self._version_ids_to_keys(versions)
1406
present_keys = self._present_keys(wanted_keys)
1407
missing = set(wanted_keys).difference(present_keys)
1409
raise RevisionNotPresent(missing.pop(), self)
1410
return list(self._keys_to_version_ids(present_keys))
1412
def get_ancestry(self, versions, topo_sorted=True):
1413
"""See VersionedFile.get_ancestry."""
1414
if not self._parents:
1415
return self._parentless_ancestry(versions)
1416
# XXX: This will do len(history) index calls - perhaps
1417
# it should be altered to be a index core feature?
1418
# get a graph of all the mentioned versions:
1421
versions = self._version_ids_to_keys(versions)
1422
pending = set(versions)
1424
# get all pending nodes
1425
this_iteration = pending
1426
new_nodes = self._get_entries(this_iteration)
1429
for (index, key, value, node_refs) in new_nodes:
1430
# dont ask for ghosties - otherwise
1431
# we we can end up looping with pending
1432
# being entirely ghosted.
1433
graph[key] = [parent for parent in node_refs[0]
1434
if parent not in ghosts]
1436
for parent in graph[key]:
1437
# dont examine known nodes again
1442
ghosts.update(this_iteration.difference(found))
1443
if versions.difference(graph):
1444
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1446
result_keys = topo_sort(graph.items())
1448
result_keys = graph.iterkeys()
1449
return [key[0] for key in result_keys]
1451
def get_ancestry_with_ghosts(self, versions):
1452
"""See VersionedFile.get_ancestry."""
1453
if not self._parents:
1454
return self._parentless_ancestry(versions)
1455
# XXX: This will do len(history) index calls - perhaps
1456
# it should be altered to be a index core feature?
1457
# get a graph of all the mentioned versions:
1459
versions = self._version_ids_to_keys(versions)
1460
pending = set(versions)
1462
# get all pending nodes
1463
this_iteration = pending
1464
new_nodes = self._get_entries(this_iteration)
1466
for (index, key, value, node_refs) in new_nodes:
1467
graph[key] = node_refs[0]
1469
for parent in graph[key]:
1470
# dont examine known nodes again
1474
missing_versions = this_iteration.difference(graph)
1475
missing_needed = versions.intersection(missing_versions)
1477
raise RevisionNotPresent(missing_needed.pop(), self)
1478
for missing_version in missing_versions:
1479
# add a key, no parents
1480
graph[missing_version] = []
1481
pending.discard(missing_version) # don't look for it
1482
result_keys = topo_sort(graph.items())
1483
return [key[0] for key in result_keys]
1485
def get_graph(self):
1486
"""Return a list of the node:parents lists from this knit index."""
1487
if not self._parents:
1488
return [(key, ()) for key in self.get_versions()]
1490
for index, key, value, refs in self._graph_index.iter_all_entries():
1491
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1494
def iter_parents(self, version_ids):
1495
"""Iterate through the parents for many version ids.
1497
:param version_ids: An iterable yielding version_ids.
1498
:return: An iterator that yields (version_id, parents). Requested
1499
version_ids not present in the versioned file are simply skipped.
1500
The order is undefined, allowing for different optimisations in
1501
the underlying implementation.
1504
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1506
present_parents = set()
1507
for node in all_nodes:
1508
all_parents.update(node[3][0])
1509
# any node we are querying must be present
1510
present_parents.add(node[1])
1511
unknown_parents = all_parents.difference(present_parents)
1512
present_parents.update(self._present_keys(unknown_parents))
1513
for node in all_nodes:
1515
for parent in node[3][0]:
1516
if parent in present_parents:
1517
parents.append(parent[0])
1518
yield node[1][0], tuple(parents)
1520
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1521
yield node[1][0], ()
1523
def num_versions(self):
1524
return len(list(self._graph_index.iter_all_entries()))
1526
__len__ = num_versions
1528
def get_versions(self):
1529
"""Get all the versions in the file. not topologically sorted."""
1530
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1532
def has_version(self, version_id):
1533
"""True if the version is in the index."""
1534
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1536
def _keys_to_version_ids(self, keys):
1537
return tuple(key[0] for key in keys)
1539
def get_position(self, version_id):
1540
"""Return details needed to access the version.
1542
:return: a tuple (index, data position, size) to hand to the access
1543
logic to get the record.
1545
node = self._get_node(version_id)
1546
bits = node[2][1:].split(' ')
1547
return node[0], int(bits[0]), int(bits[1])
1549
def get_method(self, version_id):
1550
"""Return compression method of specified version."""
1551
if not self._deltas:
1553
return self._parent_compression(self._get_node(version_id)[3][1])
1555
def _parent_compression(self, reference_list):
1556
# use the second reference list to decide if this is delta'd or not.
1557
if len(reference_list):
1562
def _get_node(self, version_id):
1563
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1565
def get_options(self, version_id):
1566
"""Return a string represention options.
1570
node = self._get_node(version_id)
1571
if not self._deltas:
1572
options = ['fulltext']
1574
options = [self._parent_compression(node[3][1])]
1575
if node[2][0] == 'N':
1576
options.append('no-eol')
1579
def get_parents(self, version_id):
1580
"""Return parents of specified version ignoring ghosts."""
1581
parents = list(self.iter_parents([version_id]))
1584
raise errors.RevisionNotPresent(version_id, self)
1585
return parents[0][1]
1587
def get_parents_with_ghosts(self, version_id):
1588
"""Return parents of specified version with ghosts."""
1589
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1590
check_present=True))
1591
if not self._parents:
1593
return self._keys_to_version_ids(nodes[0][3][0])
1595
def check_versions_present(self, version_ids):
1596
"""Check that all specified versions are present."""
1597
keys = self._version_ids_to_keys(version_ids)
1598
present = self._present_keys(keys)
1599
missing = keys.difference(present)
1601
raise RevisionNotPresent(missing.pop(), self)
1603
def add_version(self, version_id, options, access_memo, parents):
1604
"""Add a version record to the index."""
1605
return self.add_versions(((version_id, options, access_memo, parents),))
1607
def add_versions(self, versions):
1608
"""Add multiple versions to the index.
1610
This function does not insert data into the Immutable GraphIndex
1611
backing the KnitGraphIndex, instead it prepares data for insertion by
1612
the caller and checks that it is safe to insert then calls
1613
self._add_callback with the prepared GraphIndex nodes.
1615
:param versions: a list of tuples:
1616
(version_id, options, pos, size, parents).
1618
if not self._add_callback:
1619
raise errors.ReadOnlyError(self)
1620
# we hope there are no repositories with inconsistent parentage
1625
for (version_id, options, access_memo, parents) in versions:
1626
index, pos, size = access_memo
1627
key = (version_id, )
1628
parents = tuple((parent, ) for parent in parents)
1629
if 'no-eol' in options:
1633
value += "%d %d" % (pos, size)
1634
if not self._deltas:
1635
if 'line-delta' in options:
1636
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1639
if 'line-delta' in options:
1640
node_refs = (parents, (parents[0],))
1642
node_refs = (parents, ())
1644
node_refs = (parents, )
1647
raise KnitCorrupt(self, "attempt to add node with parents "
1648
"in parentless index.")
1650
keys[key] = (value, node_refs)
1651
present_nodes = self._get_entries(keys)
1652
for (index, key, value, node_refs) in present_nodes:
1653
if (value, node_refs) != keys[key]:
1654
raise KnitCorrupt(self, "inconsistent details in add_versions"
1655
": %s %s" % ((value, node_refs), keys[key]))
1659
for key, (value, node_refs) in keys.iteritems():
1660
result.append((key, value, node_refs))
1662
for key, (value, node_refs) in keys.iteritems():
1663
result.append((key, value))
1664
self._add_callback(result)
1666
def _version_ids_to_keys(self, version_ids):
1667
return set((version_id, ) for version_id in version_ids)
1670
class _KnitAccess(object):
1671
"""Access to knit records in a .knit file."""
1673
def __init__(self, transport, filename, _file_mode, _dir_mode,
1674
_need_to_create, _create_parent_dir):
1675
"""Create a _KnitAccess for accessing and inserting data.
1677
:param transport: The transport the .knit is located on.
1678
:param filename: The filename of the .knit.
1680
self._transport = transport
1681
self._filename = filename
1682
self._file_mode = _file_mode
1683
self._dir_mode = _dir_mode
1684
self._need_to_create = _need_to_create
1685
self._create_parent_dir = _create_parent_dir
1687
def add_raw_records(self, sizes, raw_data):
1688
"""Add raw knit bytes to a storage area.
1690
The data is spooled to whereever the access method is storing data.
1692
:param sizes: An iterable containing the size of each raw data segment.
1693
:param raw_data: A bytestring containing the data.
1694
:return: A list of memos to retrieve the record later. Each memo is a
1695
tuple - (index, pos, length), where the index field is always None
1696
for the .knit access method.
1698
assert type(raw_data) == str, \
1699
'data must be plain bytes was %s' % type(raw_data)
1700
if not self._need_to_create:
1701
base = self._transport.append_bytes(self._filename, raw_data)
1703
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1704
create_parent_dir=self._create_parent_dir,
1705
mode=self._file_mode,
1706
dir_mode=self._dir_mode)
1707
self._need_to_create = False
1711
result.append((None, base, size))
1716
"""IFF this data access has its own storage area, initialise it.
1720
self._transport.put_bytes_non_atomic(self._filename, '',
1721
mode=self._file_mode)
1723
def open_file(self):
1724
"""IFF this data access can be represented as a single file, open it.
1726
For knits that are not mapped to a single file on disk this will
1729
:return: None or a file handle.
1732
return self._transport.get(self._filename)
1737
def get_raw_records(self, memos_for_retrieval):
1738
"""Get the raw bytes for a records.
1740
:param memos_for_retrieval: An iterable containing the (index, pos,
1741
length) memo for retrieving the bytes. The .knit method ignores
1742
the index as there is always only a single file.
1743
:return: An iterator over the bytes of the records.
1745
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
1746
for pos, data in self._transport.readv(self._filename, read_vector):
1750
class _PackAccess(object):
1751
"""Access to knit records via a collection of packs."""
1753
def __init__(self, index_to_packs, writer=None):
1754
"""Create a _PackAccess object.
1756
:param index_to_packs: A dict mapping index objects to the transport
1757
and file names for obtaining data.
1758
:param writer: A tuple (pack.ContainerWriter, write_index) which
1759
contains the pack to write, and the index that reads from it will
1763
self.container_writer = writer[0]
1764
self.write_index = writer[1]
1766
self.container_writer = None
1767
self.write_index = None
1768
self.indices = index_to_packs
1770
def add_raw_records(self, sizes, raw_data):
1771
"""Add raw knit bytes to a storage area.
1773
The data is spooled to the container writer in one bytes-record per
1776
:param sizes: An iterable containing the size of each raw data segment.
1777
:param raw_data: A bytestring containing the data.
1778
:return: A list of memos to retrieve the record later. Each memo is a
1779
tuple - (index, pos, length), where the index field is the
1780
write_index object supplied to the PackAccess object.
1782
assert type(raw_data) == str, \
1783
'data must be plain bytes was %s' % type(raw_data)
1787
p_offset, p_length = self.container_writer.add_bytes_record(
1788
raw_data[offset:offset+size], [])
1790
result.append((self.write_index, p_offset, p_length))
1794
"""Pack based knits do not get individually created."""
1796
def get_raw_records(self, memos_for_retrieval):
1797
"""Get the raw bytes for a records.
1799
:param memos_for_retrieval: An iterable containing the (index, pos,
1800
length) memo for retrieving the bytes. The Pack access method
1801
looks up the pack to use for a given record in its index_to_pack
1803
:return: An iterator over the bytes of the records.
1805
# first pass, group into same-index requests
1807
current_index = None
1808
for (index, offset, length) in memos_for_retrieval:
1809
if current_index == index:
1810
current_list.append((offset, length))
1812
if current_index is not None:
1813
request_lists.append((current_index, current_list))
1814
current_index = index
1815
current_list = [(offset, length)]
1816
# handle the last entry
1817
if current_index is not None:
1818
request_lists.append((current_index, current_list))
1819
for index, offsets in request_lists:
1820
transport, path = self.indices[index]
1821
reader = pack.make_readv_reader(transport, path, offsets)
1822
for names, read_func in reader.iter_records():
1823
yield read_func(None)
1825
def open_file(self):
1826
"""Pack based knits have no single file."""
1829
def set_writer(self, writer, index, (transport, packname)):
1830
"""Set a writer to use for adding data."""
1831
self.indices[index] = (transport, packname)
1832
self.container_writer = writer
1833
self.write_index = index
1836
class _KnitData(object):
1837
"""Manage extraction of data from a KnitAccess, caching and decompressing.
1839
The KnitData class provides the logic for parsing and using knit records,
1840
making use of an access method for the low level read and write operations.
1843
def __init__(self, access):
1844
"""Create a KnitData object.
1846
:param access: The access method to use. Access methods such as
1847
_KnitAccess manage the insertion of raw records and the subsequent
1848
retrieval of the same.
1850
self._access = access
1851
self._checked = False
1852
# TODO: jam 20060713 conceptually, this could spill to disk
1853
# if the cached size gets larger than a certain amount
1854
# but it complicates the model a bit, so for now just use
1855
# a simple dictionary
1857
self._do_cache = False
1859
def enable_cache(self):
1860
"""Enable caching of reads."""
1861
self._do_cache = True
1863
def clear_cache(self):
1864
"""Clear the record cache."""
1865
self._do_cache = False
1868
def _open_file(self):
1869
return self._access.open_file()
1871
def _record_to_data(self, version_id, digest, lines):
1872
"""Convert version_id, digest, lines into a raw data block.
1874
:return: (len, a StringIO instance with the raw data ready to read.)
1877
data_file = GzipFile(None, mode='wb', fileobj=sio)
1879
assert isinstance(version_id, str)
1880
data_file.writelines(chain(
1881
["version %s %d %s\n" % (version_id,
1885
["end %s\n" % version_id]))
1892
def add_raw_records(self, sizes, raw_data):
1893
"""Append a prepared record to the data file.
1895
:param sizes: An iterable containing the size of each raw data segment.
1896
:param raw_data: A bytestring containing the data.
1897
:return: a list of index data for the way the data was stored.
1898
See the access method add_raw_records documentation for more
1901
return self._access.add_raw_records(sizes, raw_data)
1903
def add_record(self, version_id, digest, lines):
1904
"""Write new text record to disk.
1906
Returns index data for retrieving it later, as per add_raw_records.
1908
size, sio = self._record_to_data(version_id, digest, lines)
1909
result = self.add_raw_records([size], sio.getvalue())
1911
self._cache[version_id] = sio.getvalue()
1914
def _parse_record_header(self, version_id, raw_data):
1915
"""Parse a record header for consistency.
1917
:return: the header and the decompressor stream.
1918
as (stream, header_record)
1920
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
1922
rec = self._check_header(version_id, df.readline())
1923
except Exception, e:
1924
raise KnitCorrupt(self._access,
1925
"While reading {%s} got %s(%s)"
1926
% (version_id, e.__class__.__name__, str(e)))
1929
def _check_header(self, version_id, line):
1932
raise KnitCorrupt(self._access,
1933
'unexpected number of elements in record header')
1934
if rec[1] != version_id:
1935
raise KnitCorrupt(self._access,
1936
'unexpected version, wanted %r, got %r'
1937
% (version_id, rec[1]))
1940
def _parse_record(self, version_id, data):
1942
# 4168 calls in 2880 217 internal
1943
# 4168 calls to _parse_record_header in 2121
1944
# 4168 calls to readlines in 330
1945
df = GzipFile(mode='rb', fileobj=StringIO(data))
1948
record_contents = df.readlines()
1949
except Exception, e:
1950
raise KnitCorrupt(self._access,
1951
"While reading {%s} got %s(%s)"
1952
% (version_id, e.__class__.__name__, str(e)))
1953
header = record_contents.pop(0)
1954
rec = self._check_header(version_id, header)
1956
last_line = record_contents.pop()
1957
if len(record_contents) != int(rec[2]):
1958
raise KnitCorrupt(self._access,
1959
'incorrect number of lines %s != %s'
1961
% (len(record_contents), int(rec[2]),
1963
if last_line != 'end %s\n' % rec[1]:
1964
raise KnitCorrupt(self._access,
1965
'unexpected version end line %r, wanted %r'
1966
% (last_line, version_id))
1968
return record_contents, rec[3]
1970
def read_records_iter_raw(self, records):
1971
"""Read text records from data file and yield raw data.
1973
This unpacks enough of the text record to validate the id is
1974
as expected but thats all.
1976
# setup an iterator of the external records:
1977
# uses readv so nice and fast we hope.
1979
# grab the disk data needed.
1981
# Don't check _cache if it is empty
1982
needed_offsets = [index_memo for version_id, index_memo
1984
if version_id not in self._cache]
1986
needed_offsets = [index_memo for version_id, index_memo
1989
raw_records = self._access.get_raw_records(needed_offsets)
1991
for version_id, index_memo in records:
1992
if version_id in self._cache:
1993
# This data has already been validated
1994
data = self._cache[version_id]
1996
data = raw_records.next()
1998
self._cache[version_id] = data
2000
# validate the header
2001
df, rec = self._parse_record_header(version_id, data)
2003
yield version_id, data
2005
def read_records_iter(self, records):
2006
"""Read text records from data file and yield result.
2008
The result will be returned in whatever is the fastest to read.
2009
Not by the order requested. Also, multiple requests for the same
2010
record will only yield 1 response.
2011
:param records: A list of (version_id, pos, len) entries
2012
:return: Yields (version_id, contents, digest) in the order
2013
read, not the order requested
2019
# Skip records we have alread seen
2020
yielded_records = set()
2021
needed_records = set()
2022
for record in records:
2023
if record[0] in self._cache:
2024
if record[0] in yielded_records:
2026
yielded_records.add(record[0])
2027
data = self._cache[record[0]]
2028
content, digest = self._parse_record(record[0], data)
2029
yield (record[0], content, digest)
2031
needed_records.add(record)
2032
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2034
needed_records = sorted(set(records), key=operator.itemgetter(1))
2036
if not needed_records:
2039
# The transport optimizes the fetching as well
2040
# (ie, reads continuous ranges.)
2041
raw_data = self._access.get_raw_records(
2042
[index_memo for version_id, index_memo in needed_records])
2044
for (version_id, index_memo), data in \
2045
izip(iter(needed_records), raw_data):
2046
content, digest = self._parse_record(version_id, data)
2048
self._cache[version_id] = data
2049
yield version_id, content, digest
2051
def read_records(self, records):
2052
"""Read records into a dictionary."""
2054
for record_id, content, digest in \
2055
self.read_records_iter(records):
2056
components[record_id] = (content, digest)
2060
class InterKnit(InterVersionedFile):
2061
"""Optimised code paths for knit to knit operations."""
2063
_matching_file_from_factory = KnitVersionedFile
2064
_matching_file_to_factory = KnitVersionedFile
2067
def is_compatible(source, target):
2068
"""Be compatible with knits. """
2070
return (isinstance(source, KnitVersionedFile) and
2071
isinstance(target, KnitVersionedFile))
2072
except AttributeError:
2075
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2076
"""See InterVersionedFile.join."""
2077
assert isinstance(self.source, KnitVersionedFile)
2078
assert isinstance(self.target, KnitVersionedFile)
2080
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2085
pb = ui.ui_factory.nested_progress_bar()
2087
version_ids = list(version_ids)
2088
if None in version_ids:
2089
version_ids.remove(None)
2091
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2092
this_versions = set(self.target._index.get_versions())
2093
needed_versions = self.source_ancestry - this_versions
2094
cross_check_versions = self.source_ancestry.intersection(this_versions)
2095
mismatched_versions = set()
2096
for version in cross_check_versions:
2097
# scan to include needed parents.
2098
n1 = set(self.target.get_parents_with_ghosts(version))
2099
n2 = set(self.source.get_parents_with_ghosts(version))
2101
# FIXME TEST this check for cycles being introduced works
2102
# the logic is we have a cycle if in our graph we are an
2103
# ancestor of any of the n2 revisions.
2109
parent_ancestors = self.source.get_ancestry(parent)
2110
if version in parent_ancestors:
2111
raise errors.GraphCycleError([parent, version])
2112
# ensure this parent will be available later.
2113
new_parents = n2.difference(n1)
2114
needed_versions.update(new_parents.difference(this_versions))
2115
mismatched_versions.add(version)
2117
if not needed_versions and not mismatched_versions:
2119
full_list = topo_sort(self.source.get_graph())
2121
version_list = [i for i in full_list if (not self.target.has_version(i)
2122
and i in needed_versions)]
2126
copy_queue_records = []
2128
for version_id in version_list:
2129
options = self.source._index.get_options(version_id)
2130
parents = self.source._index.get_parents_with_ghosts(version_id)
2131
# check that its will be a consistent copy:
2132
for parent in parents:
2133
# if source has the parent, we must :
2134
# * already have it or
2135
# * have it scheduled already
2136
# otherwise we don't care
2137
assert (self.target.has_version(parent) or
2138
parent in copy_set or
2139
not self.source.has_version(parent))
2140
index_memo = self.source._index.get_position(version_id)
2141
copy_queue_records.append((version_id, index_memo))
2142
copy_queue.append((version_id, options, parents))
2143
copy_set.add(version_id)
2145
# data suck the join:
2147
total = len(version_list)
2150
for (version_id, raw_data), \
2151
(version_id2, options, parents) in \
2152
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2154
assert version_id == version_id2, 'logic error, inconsistent results'
2156
pb.update("Joining knit", count, total)
2157
raw_records.append((version_id, options, parents, len(raw_data)))
2158
raw_datum.append(raw_data)
2159
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2161
for version in mismatched_versions:
2162
# FIXME RBC 20060309 is this needed?
2163
n1 = set(self.target.get_parents_with_ghosts(version))
2164
n2 = set(self.source.get_parents_with_ghosts(version))
2165
# write a combined record to our history preserving the current
2166
# parents as first in the list
2167
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2168
self.target.fix_parents(version, new_parents)
2174
InterVersionedFile.register_optimiser(InterKnit)
2177
class WeaveToKnit(InterVersionedFile):
2178
"""Optimised code paths for weave to knit operations."""
2180
_matching_file_from_factory = bzrlib.weave.WeaveFile
2181
_matching_file_to_factory = KnitVersionedFile
2184
def is_compatible(source, target):
2185
"""Be compatible with weaves to knits."""
2187
return (isinstance(source, bzrlib.weave.Weave) and
2188
isinstance(target, KnitVersionedFile))
2189
except AttributeError:
2192
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2193
"""See InterVersionedFile.join."""
2194
assert isinstance(self.source, bzrlib.weave.Weave)
2195
assert isinstance(self.target, KnitVersionedFile)
2197
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2202
pb = ui.ui_factory.nested_progress_bar()
2204
version_ids = list(version_ids)
2206
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2207
this_versions = set(self.target._index.get_versions())
2208
needed_versions = self.source_ancestry - this_versions
2209
cross_check_versions = self.source_ancestry.intersection(this_versions)
2210
mismatched_versions = set()
2211
for version in cross_check_versions:
2212
# scan to include needed parents.
2213
n1 = set(self.target.get_parents_with_ghosts(version))
2214
n2 = set(self.source.get_parents(version))
2215
# if all of n2's parents are in n1, then its fine.
2216
if n2.difference(n1):
2217
# FIXME TEST this check for cycles being introduced works
2218
# the logic is we have a cycle if in our graph we are an
2219
# ancestor of any of the n2 revisions.
2225
parent_ancestors = self.source.get_ancestry(parent)
2226
if version in parent_ancestors:
2227
raise errors.GraphCycleError([parent, version])
2228
# ensure this parent will be available later.
2229
new_parents = n2.difference(n1)
2230
needed_versions.update(new_parents.difference(this_versions))
2231
mismatched_versions.add(version)
2233
if not needed_versions and not mismatched_versions:
2235
full_list = topo_sort(self.source.get_graph())
2237
version_list = [i for i in full_list if (not self.target.has_version(i)
2238
and i in needed_versions)]
2242
total = len(version_list)
2243
for version_id in version_list:
2244
pb.update("Converting to knit", count, total)
2245
parents = self.source.get_parents(version_id)
2246
# check that its will be a consistent copy:
2247
for parent in parents:
2248
# if source has the parent, we must already have it
2249
assert (self.target.has_version(parent))
2250
self.target.add_lines(
2251
version_id, parents, self.source.get_lines(version_id))
2254
for version in mismatched_versions:
2255
# FIXME RBC 20060309 is this needed?
2256
n1 = set(self.target.get_parents_with_ghosts(version))
2257
n2 = set(self.source.get_parents(version))
2258
# write a combined record to our history preserving the current
2259
# parents as first in the list
2260
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2261
self.target.fix_parents(version, new_parents)
2267
InterVersionedFile.register_optimiser(WeaveToKnit)
2270
class KnitSequenceMatcher(difflib.SequenceMatcher):
2271
"""Knit tuned sequence matcher.
2273
This is based on profiling of difflib which indicated some improvements
2274
for our usage pattern.
2277
def find_longest_match(self, alo, ahi, blo, bhi):
2278
"""Find longest matching block in a[alo:ahi] and b[blo:bhi].
2280
If isjunk is not defined:
2282
Return (i,j,k) such that a[i:i+k] is equal to b[j:j+k], where
2283
alo <= i <= i+k <= ahi
2284
blo <= j <= j+k <= bhi
2285
and for all (i',j',k') meeting those conditions,
2288
and if i == i', j <= j'
2290
In other words, of all maximal matching blocks, return one that
2291
starts earliest in a, and of all those maximal matching blocks that
2292
start earliest in a, return the one that starts earliest in b.
2294
>>> s = SequenceMatcher(None, " abcd", "abcd abcd")
2295
>>> s.find_longest_match(0, 5, 0, 9)
2298
If isjunk is defined, first the longest matching block is
2299
determined as above, but with the additional restriction that no
2300
junk element appears in the block. Then that block is extended as
2301
far as possible by matching (only) junk elements on both sides. So
2302
the resulting block never matches on junk except as identical junk
2303
happens to be adjacent to an "interesting" match.
2305
Here's the same example as before, but considering blanks to be
2306
junk. That prevents " abcd" from matching the " abcd" at the tail
2307
end of the second sequence directly. Instead only the "abcd" can
2308
match, and matches the leftmost "abcd" in the second sequence:
2310
>>> s = SequenceMatcher(lambda x: x==" ", " abcd", "abcd abcd")
2311
>>> s.find_longest_match(0, 5, 0, 9)
2314
If no blocks match, return (alo, blo, 0).
2316
>>> s = SequenceMatcher(None, "ab", "c")
2317
>>> s.find_longest_match(0, 2, 0, 1)
2321
# CAUTION: stripping common prefix or suffix would be incorrect.
2325
# Longest matching block is "ab", but if common prefix is
2326
# stripped, it's "a" (tied with "b"). UNIX(tm) diff does so
2327
# strip, so ends up claiming that ab is changed to acab by
2328
# inserting "ca" in the middle. That's minimal but unintuitive:
2329
# "it's obvious" that someone inserted "ac" at the front.
2330
# Windiff ends up at the same place as diff, but by pairing up
2331
# the unique 'b's and then matching the first two 'a's.
2333
a, b, b2j, isbjunk = self.a, self.b, self.b2j, self.isbjunk
2334
besti, bestj, bestsize = alo, blo, 0
2335
# find longest junk-free match
2336
# during an iteration of the loop, j2len[j] = length of longest
2337
# junk-free match ending with a[i-1] and b[j]
2341
for i in xrange(alo, ahi):
2342
# look at all instances of a[i] in b; note that because
2343
# b2j has no junk keys, the loop is skipped if a[i] is junk
2344
j2lenget = j2len.get
2347
# changing b2j.get(a[i], nothing) to a try:KeyError pair produced the
2348
# following improvement
2349
# 704 0 4650.5320 2620.7410 bzrlib.knit:1336(find_longest_match)
2350
# +326674 0 1655.1210 1655.1210 +<method 'get' of 'dict' objects>
2351
# +76519 0 374.6700 374.6700 +<method 'has_key' of 'dict' objects>
2353
# 704 0 3733.2820 2209.6520 bzrlib.knit:1336(find_longest_match)
2354
# +211400 0 1147.3520 1147.3520 +<method 'get' of 'dict' objects>
2355
# +76519 0 376.2780 376.2780 +<method 'has_key' of 'dict' objects>
2367
k = newj2len[j] = 1 + j2lenget(-1 + j, 0)
2369
besti, bestj, bestsize = 1 + i-k, 1 + j-k, k
2372
# Extend the best by non-junk elements on each end. In particular,
2373
# "popular" non-junk elements aren't in b2j, which greatly speeds
2374
# the inner loop above, but also means "the best" match so far
2375
# doesn't contain any junk *or* popular non-junk elements.
2376
while besti > alo and bestj > blo and \
2377
not isbjunk(b[bestj-1]) and \
2378
a[besti-1] == b[bestj-1]:
2379
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2380
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2381
not isbjunk(b[bestj+bestsize]) and \
2382
a[besti+bestsize] == b[bestj+bestsize]:
2385
# Now that we have a wholly interesting match (albeit possibly
2386
# empty!), we may as well suck up the matching junk on each
2387
# side of it too. Can't think of a good reason not to, and it
2388
# saves post-processing the (possibly considerable) expense of
2389
# figuring out what to do with it. In the case of an empty
2390
# interesting match, this is clearly the right thing to do,
2391
# because no other kind of match is possible in the regions.
2392
while besti > alo and bestj > blo and \
2393
isbjunk(b[bestj-1]) and \
2394
a[besti-1] == b[bestj-1]:
2395
besti, bestj, bestsize = besti-1, bestj-1, bestsize+1
2396
while besti+bestsize < ahi and bestj+bestsize < bhi and \
2397
isbjunk(b[bestj+bestsize]) and \
2398
a[besti+bestsize] == b[bestj+bestsize]:
2399
bestsize = bestsize + 1
2401
return besti, bestj, bestsize
2405
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2407
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
