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
65
from itertools import izip, chain
70
from zlib import Z_DEFAULT_COMPRESSION
73
from bzrlib.lazy_import import lazy_import
74
lazy_import(globals(), """
92
from bzrlib.errors import (
98
KnitDataStreamIncompatible,
101
RevisionAlreadyPresent,
103
from bzrlib.tuned_gzip import GzipFile, bytes_to_gzip
104
from bzrlib.osutils import (
110
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
111
from bzrlib.tsort import topo_sort
114
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
117
# TODO: Split out code specific to this format into an associated object.
119
# TODO: Can we put in some kind of value to check that the index and data
120
# files belong together?
122
# TODO: accommodate binaries, perhaps by storing a byte count
124
# TODO: function to check whole file
126
# TODO: atomically append data, then measure backwards from the cursor
127
# position after writing to work out where it was located. we may need to
128
# bypass python file buffering.
130
DATA_SUFFIX = '.knit'
131
INDEX_SUFFIX = '.kndx'
134
class KnitContent(object):
135
"""Content of a knit version to which deltas can be applied."""
138
"""Return a list of (origin, text) tuples."""
139
return list(self.annotate_iter())
141
def line_delta_iter(self, new_lines):
142
"""Generate line-based delta from this content to new_lines."""
143
new_texts = new_lines.text()
144
old_texts = self.text()
145
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
146
for tag, i1, i2, j1, j2 in s.get_opcodes():
149
# ofrom, oto, length, data
150
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
152
def line_delta(self, new_lines):
153
return list(self.line_delta_iter(new_lines))
156
def get_line_delta_blocks(knit_delta, source, target):
157
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
158
target_len = len(target)
161
for s_begin, s_end, t_len, new_text in knit_delta:
162
true_n = s_begin - s_pos
165
# knit deltas do not provide reliable info about whether the
166
# last line of a file matches, due to eol handling.
167
if source[s_pos + n -1] != target[t_pos + n -1]:
170
yield s_pos, t_pos, n
171
t_pos += t_len + true_n
173
n = target_len - t_pos
175
if source[s_pos + n -1] != target[t_pos + n -1]:
178
yield s_pos, t_pos, n
179
yield s_pos + (target_len - t_pos), target_len, 0
182
class AnnotatedKnitContent(KnitContent):
183
"""Annotated content."""
185
def __init__(self, lines):
188
def annotate_iter(self):
189
"""Yield tuples of (origin, text) for each content line."""
190
return iter(self._lines)
192
def strip_last_line_newline(self):
193
line = self._lines[-1][1].rstrip('\n')
194
self._lines[-1] = (self._lines[-1][0], line)
197
return [text for origin, text in self._lines]
200
return AnnotatedKnitContent(self._lines[:])
203
class PlainKnitContent(KnitContent):
204
"""Unannotated content.
206
When annotate[_iter] is called on this content, the same version is reported
207
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
211
def __init__(self, lines, version_id):
213
self._version_id = version_id
215
def annotate_iter(self):
216
"""Yield tuples of (origin, text) for each content line."""
217
for line in self._lines:
218
yield self._version_id, line
221
return PlainKnitContent(self._lines[:], self._version_id)
223
def strip_last_line_newline(self):
224
self._lines[-1] = self._lines[-1].rstrip('\n')
230
class KnitAnnotateFactory(object):
231
"""Factory for creating annotated Content objects."""
235
def make(self, lines, version_id):
236
num_lines = len(lines)
237
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
239
def parse_fulltext(self, content, version_id):
240
"""Convert fulltext to internal representation
242
fulltext content is of the format
243
revid(utf8) plaintext\n
244
internal representation is of the format:
247
# TODO: jam 20070209 The tests expect this to be returned as tuples,
248
# but the code itself doesn't really depend on that.
249
# Figure out a way to not require the overhead of turning the
250
# list back into tuples.
251
lines = [tuple(line.split(' ', 1)) for line in content]
252
return AnnotatedKnitContent(lines)
254
def parse_line_delta_iter(self, lines):
255
return iter(self.parse_line_delta(lines))
257
def parse_line_delta(self, lines, version_id):
258
"""Convert a line based delta into internal representation.
260
line delta is in the form of:
261
intstart intend intcount
263
revid(utf8) newline\n
264
internal representation is
265
(start, end, count, [1..count tuples (revid, newline)])
272
def cache_and_return(line):
273
origin, text = line.split(' ', 1)
274
return cache.setdefault(origin, origin), text
276
# walk through the lines parsing.
278
start, end, count = [int(n) for n in header.split(',')]
279
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
280
result.append((start, end, count, contents))
283
def get_fulltext_content(self, lines):
284
"""Extract just the content lines from a fulltext."""
285
return (line.split(' ', 1)[1] for line in lines)
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):
299
origin, text = next().split(' ', 1)
302
def lower_fulltext(self, content):
303
"""convert a fulltext content record into a serializable form.
305
see parse_fulltext which this inverts.
307
# TODO: jam 20070209 We only do the caching thing to make sure that
308
# the origin is a valid utf-8 line, eventually we could remove it
309
return ['%s %s' % (o, t) for o, t in content._lines]
311
def lower_line_delta(self, delta):
312
"""convert a delta into a serializable form.
314
See parse_line_delta which this inverts.
316
# TODO: jam 20070209 We only do the caching thing to make sure that
317
# the origin is a valid utf-8 line, eventually we could remove it
319
for start, end, c, lines in delta:
320
out.append('%d,%d,%d\n' % (start, end, c))
321
out.extend(origin + ' ' + text
322
for origin, text in lines)
325
def annotate_iter(self, knit, version_id):
326
content = knit._get_content(version_id)
327
return content.annotate_iter()
330
class KnitPlainFactory(object):
331
"""Factory for creating plain Content objects."""
335
def make(self, lines, version_id):
336
return PlainKnitContent(lines, version_id)
338
def parse_fulltext(self, content, version_id):
339
"""This parses an unannotated fulltext.
341
Note that this is not a noop - the internal representation
342
has (versionid, line) - its just a constant versionid.
344
return self.make(content, version_id)
346
def parse_line_delta_iter(self, lines, version_id):
348
num_lines = len(lines)
349
while cur < num_lines:
352
start, end, c = [int(n) for n in header.split(',')]
353
yield start, end, c, lines[cur:cur+c]
356
def parse_line_delta(self, lines, version_id):
357
return list(self.parse_line_delta_iter(lines, version_id))
359
def get_fulltext_content(self, lines):
360
"""Extract just the content lines from a fulltext."""
363
def get_linedelta_content(self, lines):
364
"""Extract just the content from a line delta.
366
This doesn't return all of the extra information stored in a delta.
367
Only the actual content lines.
372
header = header.split(',')
373
count = int(header[2])
374
for i in xrange(count):
377
def lower_fulltext(self, content):
378
return content.text()
380
def lower_line_delta(self, delta):
382
for start, end, c, lines in delta:
383
out.append('%d,%d,%d\n' % (start, end, c))
387
def annotate_iter(self, knit, version_id):
388
return annotate_knit(knit, version_id)
391
def make_empty_knit(transport, relpath):
392
"""Construct a empty knit at the specified location."""
393
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
396
class KnitVersionedFile(VersionedFile):
397
"""Weave-like structure with faster random access.
399
A knit stores a number of texts and a summary of the relationships
400
between them. Texts are identified by a string version-id. Texts
401
are normally stored and retrieved as a series of lines, but can
402
also be passed as single strings.
404
Lines are stored with the trailing newline (if any) included, to
405
avoid special cases for files with no final newline. Lines are
406
composed of 8-bit characters, not unicode. The combination of
407
these approaches should mean any 'binary' file can be safely
408
stored and retrieved.
411
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
412
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
413
create=False, create_parent_dir=False, delay_create=False,
414
dir_mode=None, index=None, access_method=None):
415
"""Construct a knit at location specified by relpath.
417
:param create: If not True, only open an existing knit.
418
:param create_parent_dir: If True, create the parent directory if
419
creating the file fails. (This is used for stores with
420
hash-prefixes that may not exist yet)
421
:param delay_create: The calling code is aware that the knit won't
422
actually be created until the first data is stored.
423
:param index: An index to use for the knit.
425
if deprecated_passed(basis_knit):
426
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
427
" deprecated as of bzr 0.9.",
428
DeprecationWarning, stacklevel=2)
429
if access_mode is None:
431
super(KnitVersionedFile, self).__init__(access_mode)
432
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
433
self.transport = transport
434
self.filename = relpath
435
self.factory = factory or KnitAnnotateFactory()
436
self.writable = (access_mode == 'w')
439
self._max_delta_chain = 200
442
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
443
access_mode, create=create, file_mode=file_mode,
444
create_parent_dir=create_parent_dir, delay_create=delay_create,
448
if access_method is None:
449
_access = _KnitAccess(transport, relpath + DATA_SUFFIX, file_mode, dir_mode,
450
((create and not len(self)) and delay_create), create_parent_dir)
452
_access = access_method
453
if create and not len(self) and not delay_create:
455
self._data = _KnitData(_access)
458
return '%s(%s)' % (self.__class__.__name__,
459
self.transport.abspath(self.filename))
461
def _check_should_delta(self, first_parents):
462
"""Iterate back through the parent listing, looking for a fulltext.
464
This is used when we want to decide whether to add a delta or a new
465
fulltext. It searches for _max_delta_chain parents. When it finds a
466
fulltext parent, it sees if the total size of the deltas leading up to
467
it is large enough to indicate that we want a new full text anyway.
469
Return True if we should create a new delta, False if we should use a
474
delta_parents = first_parents
475
for count in xrange(self._max_delta_chain):
476
parent = delta_parents[0]
477
method = self._index.get_method(parent)
478
index, pos, size = self._index.get_position(parent)
479
if method == 'fulltext':
483
delta_parents = self._index.get_parents(parent)
485
# We couldn't find a fulltext, so we must create a new one
488
return fulltext_size > delta_size
490
def _add_raw_records(self, records, data):
491
"""Add all the records 'records' with data pre-joined in 'data'.
493
:param records: A list of tuples(version_id, options, parents, size).
494
:param data: The data for the records. When it is written, the records
495
are adjusted to have pos pointing into data by the sum of
496
the preceding records sizes.
499
raw_record_sizes = [record[3] for record in records]
500
positions = self._data.add_raw_records(raw_record_sizes, data)
503
for (version_id, options, parents, size), access_memo in zip(
505
index_entries.append((version_id, options, access_memo, parents))
506
if self._data._do_cache:
507
self._data._cache[version_id] = data[offset:offset+size]
509
self._index.add_versions(index_entries)
511
def enable_cache(self):
512
"""Start caching data for this knit"""
513
self._data.enable_cache()
515
def clear_cache(self):
516
"""Clear the data cache only."""
517
self._data.clear_cache()
519
def copy_to(self, name, transport):
520
"""See VersionedFile.copy_to()."""
521
# copy the current index to a temp index to avoid racing with local
523
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
524
self.transport.get(self._index._filename))
526
f = self._data._open_file()
528
transport.put_file(name + DATA_SUFFIX, f)
531
# move the copied index into place
532
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
534
def create_empty(self, name, transport, mode=None):
535
return KnitVersionedFile(name, transport, factory=self.factory,
536
delta=self.delta, create=True)
538
def get_data_stream(self, required_versions):
539
"""Get a data stream for the specified versions.
541
Versions may be returned in any order, not necessarily the order
544
:param required_versions: The exact set of versions to be extracted.
545
Unlike some other knit methods, this is not used to generate a
546
transitive closure, rather it is used precisely as given.
548
:returns: format_signature, list of (version, options, length, parents),
551
required_versions = set([osutils.safe_revision_id(v) for v in
553
# we don't care about inclusions, the caller cares.
554
# but we need to setup a list of records to visit.
555
for version_id in required_versions:
556
if not self.has_version(version_id):
557
raise RevisionNotPresent(version_id, self.filename)
558
# Pick the desired versions out of the index in oldest-to-newest order
560
for version_id in self.versions():
561
if version_id in required_versions:
562
version_list.append(version_id)
564
# create the list of version information for the result
565
copy_queue_records = []
567
result_version_list = []
568
for version_id in version_list:
569
options = self._index.get_options(version_id)
570
parents = self._index.get_parents_with_ghosts(version_id)
571
index_memo = self._index.get_position(version_id)
572
copy_queue_records.append((version_id, index_memo))
573
none, data_pos, data_size = index_memo
574
copy_set.add(version_id)
575
# version, options, length, parents
576
result_version_list.append((version_id, options, data_size,
579
# Read the compressed record data.
581
# From here down to the return should really be logic in the returned
582
# callable -- in a class that adapts read_records_iter_raw to read
585
for (version_id, raw_data), \
586
(version_id2, options, _, parents) in \
587
izip(self._data.read_records_iter_raw(copy_queue_records),
588
result_version_list):
589
assert version_id == version_id2, 'logic error, inconsistent results'
590
raw_datum.append(raw_data)
591
pseudo_file = StringIO(''.join(raw_datum))
594
return pseudo_file.read()
596
return pseudo_file.read(length)
597
return (self.get_format_signature(), result_version_list, read)
599
def _extract_blocks(self, version_id, source, target):
600
if self._index.get_method(version_id) != 'line-delta':
602
parent, sha1, noeol, delta = self.get_delta(version_id)
603
return KnitContent.get_line_delta_blocks(delta, source, target)
605
def get_delta(self, version_id):
606
"""Get a delta for constructing version from some other version."""
607
version_id = osutils.safe_revision_id(version_id)
608
self.check_not_reserved_id(version_id)
609
parents = self.get_parents(version_id)
614
index_memo = self._index.get_position(version_id)
615
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
616
noeol = 'no-eol' in self._index.get_options(version_id)
617
if 'fulltext' == self._index.get_method(version_id):
618
new_content = self.factory.parse_fulltext(data, version_id)
619
if parent is not None:
620
reference_content = self._get_content(parent)
621
old_texts = reference_content.text()
624
new_texts = new_content.text()
625
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
627
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
629
delta = self.factory.parse_line_delta(data, version_id)
630
return parent, sha1, noeol, delta
632
def get_format_signature(self):
633
"""See VersionedFile.get_format_signature()."""
634
if self.factory.annotated:
635
annotated_part = "annotated"
637
annotated_part = "plain"
638
return "knit-%s" % (annotated_part,)
640
def get_graph_with_ghosts(self):
641
"""See VersionedFile.get_graph_with_ghosts()."""
642
graph_items = self._index.get_graph()
643
return dict(graph_items)
645
def get_sha1(self, version_id):
646
return self.get_sha1s([version_id])[0]
648
def get_sha1s(self, version_ids):
649
"""See VersionedFile.get_sha1()."""
650
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
651
record_map = self._get_record_map(version_ids)
652
# record entry 2 is the 'digest'.
653
return [record_map[v][2] for v in version_ids]
657
"""See VersionedFile.get_suffixes()."""
658
return [DATA_SUFFIX, INDEX_SUFFIX]
660
def has_ghost(self, version_id):
661
"""True if there is a ghost reference in the file to version_id."""
662
version_id = osutils.safe_revision_id(version_id)
664
if self.has_version(version_id):
666
# optimisable if needed by memoising the _ghosts set.
667
items = self._index.get_graph()
668
for node, parents in items:
669
for parent in parents:
670
if parent not in self._index._cache:
671
if parent == version_id:
675
def insert_data_stream(self, (format, data_list, reader_callable)):
676
"""Insert knit records from a data stream into this knit.
678
If a version in the stream is already present in this knit, it will not
679
be inserted a second time. It will be checked for consistency with the
680
stored version however, and may cause a KnitCorrupt error to be raised
681
if the data in the stream disagrees with the already stored data.
683
:seealso: get_data_stream
685
if format != self.get_format_signature():
686
trace.mutter('incompatible format signature inserting to %r', self)
687
raise KnitDataStreamIncompatible(
688
format, self.get_format_signature())
690
for version_id, options, length, parents in data_list:
691
if self.has_version(version_id):
692
# First check: the list of parents.
693
my_parents = self.get_parents_with_ghosts(version_id)
694
if my_parents != parents:
695
# XXX: KnitCorrupt is not quite the right exception here.
698
'parents list %r from data stream does not match '
699
'already recorded parents %r for %s'
700
% (parents, my_parents, version_id))
702
# Also check the SHA-1 of the fulltext this content will
704
raw_data = reader_callable(length)
705
my_fulltext_sha1 = self.get_sha1(version_id)
706
df, rec = self._data._parse_record_header(version_id, raw_data)
707
stream_fulltext_sha1 = rec[3]
708
if my_fulltext_sha1 != stream_fulltext_sha1:
709
# Actually, we don't know if it's this knit that's corrupt,
710
# or the data stream we're trying to insert.
712
self.filename, 'sha-1 does not match %s' % version_id)
714
self._add_raw_records(
715
[(version_id, options, parents, length)],
716
reader_callable(length))
719
"""See VersionedFile.versions."""
720
if 'evil' in debug.debug_flags:
721
trace.mutter_callsite(2, "versions scales with size of history")
722
return self._index.get_versions()
724
def has_version(self, version_id):
725
"""See VersionedFile.has_version."""
726
if 'evil' in debug.debug_flags:
727
trace.mutter_callsite(2, "has_version is a LBYL scenario")
728
version_id = osutils.safe_revision_id(version_id)
729
return self._index.has_version(version_id)
731
__contains__ = has_version
733
def _merge_annotations(self, content, parents, parent_texts={},
734
delta=None, annotated=None,
735
left_matching_blocks=None):
736
"""Merge annotations for content. This is done by comparing
737
the annotations based on changed to the text.
739
if left_matching_blocks is not None:
740
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
744
for parent_id in parents:
745
merge_content = self._get_content(parent_id, parent_texts)
746
if (parent_id == parents[0] and delta_seq is not None):
749
seq = patiencediff.PatienceSequenceMatcher(
750
None, merge_content.text(), content.text())
751
for i, j, n in seq.get_matching_blocks():
754
# this appears to copy (origin, text) pairs across to the
755
# new content for any line that matches the last-checked
757
content._lines[j:j+n] = merge_content._lines[i:i+n]
759
if delta_seq is None:
760
reference_content = self._get_content(parents[0], parent_texts)
761
new_texts = content.text()
762
old_texts = reference_content.text()
763
delta_seq = patiencediff.PatienceSequenceMatcher(
764
None, old_texts, new_texts)
765
return self._make_line_delta(delta_seq, content)
767
def _make_line_delta(self, delta_seq, new_content):
768
"""Generate a line delta from delta_seq and new_content."""
770
for op in delta_seq.get_opcodes():
773
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
776
def _get_components_positions(self, version_ids):
777
"""Produce a map of position data for the components of versions.
779
This data is intended to be used for retrieving the knit records.
781
A dict of version_id to (method, data_pos, data_size, next) is
783
method is the way referenced data should be applied.
784
data_pos is the position of the data in the knit.
785
data_size is the size of the data in the knit.
786
next is the build-parent of the version, or None for fulltexts.
789
for version_id in version_ids:
792
while cursor is not None and cursor not in component_data:
793
method = self._index.get_method(cursor)
794
if method == 'fulltext':
797
next = self.get_parents(cursor)[0]
798
index_memo = self._index.get_position(cursor)
799
component_data[cursor] = (method, index_memo, next)
801
return component_data
803
def _get_content(self, version_id, parent_texts={}):
804
"""Returns a content object that makes up the specified
806
cached_version = parent_texts.get(version_id, None)
807
if cached_version is not None:
808
if not self.has_version(version_id):
809
raise RevisionNotPresent(version_id, self.filename)
810
return cached_version
812
text_map, contents_map = self._get_content_maps([version_id])
813
return contents_map[version_id]
815
def _check_versions_present(self, version_ids):
816
"""Check that all specified versions are present."""
817
self._index.check_versions_present(version_ids)
819
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
820
nostore_sha, random_id, check_content):
821
"""See VersionedFile.add_lines_with_ghosts()."""
822
self._check_add(version_id, lines, random_id, check_content)
823
return self._add(version_id, lines, parents, self.delta,
824
parent_texts, None, nostore_sha, random_id)
826
def _add_lines(self, version_id, parents, lines, parent_texts,
827
left_matching_blocks, nostore_sha, random_id, check_content):
828
"""See VersionedFile.add_lines."""
829
self._check_add(version_id, lines, random_id, check_content)
830
self._check_versions_present(parents)
831
return self._add(version_id, lines[:], parents, self.delta,
832
parent_texts, left_matching_blocks, nostore_sha, random_id)
834
def _check_add(self, version_id, lines, random_id, check_content):
835
"""check that version_id and lines are safe to add."""
836
if contains_whitespace(version_id):
837
raise InvalidRevisionId(version_id, self.filename)
838
self.check_not_reserved_id(version_id)
839
# Technically this could be avoided if we are happy to allow duplicate
840
# id insertion when other things than bzr core insert texts, but it
841
# seems useful for folk using the knit api directly to have some safety
842
# blanket that we can disable.
843
if not random_id and self.has_version(version_id):
844
raise RevisionAlreadyPresent(version_id, self.filename)
846
self._check_lines_not_unicode(lines)
847
self._check_lines_are_lines(lines)
849
def _add(self, version_id, lines, parents, delta, parent_texts,
850
left_matching_blocks, nostore_sha, random_id):
851
"""Add a set of lines on top of version specified by parents.
853
If delta is true, compress the text as a line-delta against
856
Any versions not present will be converted into ghosts.
858
# first thing, if the content is something we don't need to store, find
860
line_bytes = ''.join(lines)
861
digest = sha_string(line_bytes)
862
if nostore_sha == digest:
863
raise errors.ExistingContent
866
if parent_texts is None:
868
for parent in parents:
869
if self.has_version(parent):
870
present_parents.append(parent)
872
# can only compress against the left most present parent.
874
(len(present_parents) == 0 or
875
present_parents[0] != parents[0])):
878
text_length = len(line_bytes)
881
if lines[-1][-1] != '\n':
882
# copy the contents of lines.
884
options.append('no-eol')
885
lines[-1] = lines[-1] + '\n'
888
# To speed the extract of texts the delta chain is limited
889
# to a fixed number of deltas. This should minimize both
890
# I/O and the time spend applying deltas.
891
delta = self._check_should_delta(present_parents)
893
assert isinstance(version_id, str)
894
content = self.factory.make(lines, version_id)
895
if delta or (self.factory.annotated and len(present_parents) > 0):
896
# Merge annotations from parent texts if needed.
897
delta_hunks = self._merge_annotations(content, present_parents,
898
parent_texts, delta, self.factory.annotated,
899
left_matching_blocks)
902
options.append('line-delta')
903
store_lines = self.factory.lower_line_delta(delta_hunks)
904
size, bytes = self._data._record_to_data(version_id, digest,
907
options.append('fulltext')
908
# get mixed annotation + content and feed it into the
910
store_lines = self.factory.lower_fulltext(content)
911
size, bytes = self._data._record_to_data(version_id, digest,
914
access_memo = self._data.add_raw_records([size], bytes)[0]
915
self._index.add_versions(
916
((version_id, options, access_memo, parents),),
918
return digest, text_length, content
920
def check(self, progress_bar=None):
921
"""See VersionedFile.check()."""
923
def _clone_text(self, new_version_id, old_version_id, parents):
924
"""See VersionedFile.clone_text()."""
925
# FIXME RBC 20060228 make fast by only inserting an index with null
927
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
929
def get_lines(self, version_id):
930
"""See VersionedFile.get_lines()."""
931
return self.get_line_list([version_id])[0]
933
def _get_record_map(self, version_ids):
934
"""Produce a dictionary of knit records.
936
The keys are version_ids, the values are tuples of (method, content,
938
method is the way the content should be applied.
939
content is a KnitContent object.
940
digest is the SHA1 digest of this version id after all steps are done
941
next is the build-parent of the version, i.e. the leftmost ancestor.
942
If the method is fulltext, next will be None.
944
position_map = self._get_components_positions(version_ids)
945
# c = component_id, m = method, i_m = index_memo, n = next
946
records = [(c, i_m) for c, (m, i_m, n) in position_map.iteritems()]
948
for component_id, content, digest in \
949
self._data.read_records_iter(records):
950
method, index_memo, next = position_map[component_id]
951
record_map[component_id] = method, content, digest, next
955
def get_text(self, version_id):
956
"""See VersionedFile.get_text"""
957
return self.get_texts([version_id])[0]
959
def get_texts(self, version_ids):
960
return [''.join(l) for l in self.get_line_list(version_ids)]
962
def get_line_list(self, version_ids):
963
"""Return the texts of listed versions as a list of strings."""
964
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
965
for version_id in version_ids:
966
self.check_not_reserved_id(version_id)
967
text_map, content_map = self._get_content_maps(version_ids)
968
return [text_map[v] for v in version_ids]
970
_get_lf_split_line_list = get_line_list
972
def _get_content_maps(self, version_ids):
973
"""Produce maps of text and KnitContents
975
:return: (text_map, content_map) where text_map contains the texts for
976
the requested versions and content_map contains the KnitContents.
977
Both dicts take version_ids as their keys.
979
for version_id in version_ids:
980
if not self.has_version(version_id):
981
raise RevisionNotPresent(version_id, self.filename)
982
record_map = self._get_record_map(version_ids)
987
for version_id in version_ids:
990
while cursor is not None:
991
method, data, digest, next = record_map[cursor]
992
components.append((cursor, method, data, digest))
993
if cursor in content_map:
998
for component_id, method, data, digest in reversed(components):
999
if component_id in content_map:
1000
content = content_map[component_id]
1002
if method == 'fulltext':
1003
assert content is None
1004
content = self.factory.parse_fulltext(data, version_id)
1005
elif method == 'line-delta':
1006
delta = self.factory.parse_line_delta(data, version_id)
1007
content = content.copy()
1008
content._lines = self._apply_delta(content._lines,
1010
content_map[component_id] = content
1012
if 'no-eol' in self._index.get_options(version_id):
1013
content = content.copy()
1014
content.strip_last_line_newline()
1015
final_content[version_id] = content
1017
# digest here is the digest from the last applied component.
1018
text = content.text()
1019
if sha_strings(text) != digest:
1020
raise KnitCorrupt(self.filename,
1021
'sha-1 does not match %s' % version_id)
1023
text_map[version_id] = text
1024
return text_map, final_content
1027
def _apply_delta(lines, delta):
1028
"""Apply delta to lines."""
1031
for start, end, count, delta_lines in delta:
1032
lines[offset+start:offset+end] = delta_lines
1033
offset = offset + (start - end) + count
1036
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1038
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1039
if version_ids is None:
1040
version_ids = self.versions()
1042
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
1044
pb = progress.DummyProgress()
1045
# we don't care about inclusions, the caller cares.
1046
# but we need to setup a list of records to visit.
1047
# we need version_id, position, length
1048
version_id_records = []
1049
requested_versions = set(version_ids)
1050
# filter for available versions
1051
for version_id in requested_versions:
1052
if not self.has_version(version_id):
1053
raise RevisionNotPresent(version_id, self.filename)
1054
# get a in-component-order queue:
1055
for version_id in self.versions():
1056
if version_id in requested_versions:
1057
index_memo = self._index.get_position(version_id)
1058
version_id_records.append((version_id, index_memo))
1060
total = len(version_id_records)
1061
for version_idx, (version_id, data, sha_value) in \
1062
enumerate(self._data.read_records_iter(version_id_records)):
1063
pb.update('Walking content.', version_idx, total)
1064
method = self._index.get_method(version_id)
1066
assert method in ('fulltext', 'line-delta')
1067
if method == 'fulltext':
1068
line_iterator = self.factory.get_fulltext_content(data)
1070
line_iterator = self.factory.get_linedelta_content(data)
1071
for line in line_iterator:
1074
pb.update('Walking content.', total, total)
1076
def iter_parents(self, version_ids):
1077
"""Iterate through the parents for many version ids.
1079
:param version_ids: An iterable yielding version_ids.
1080
:return: An iterator that yields (version_id, parents). Requested
1081
version_ids not present in the versioned file are simply skipped.
1082
The order is undefined, allowing for different optimisations in
1083
the underlying implementation.
1085
version_ids = [osutils.safe_revision_id(version_id) for
1086
version_id in version_ids]
1087
return self._index.iter_parents(version_ids)
1089
def num_versions(self):
1090
"""See VersionedFile.num_versions()."""
1091
return self._index.num_versions()
1093
__len__ = num_versions
1095
def annotate_iter(self, version_id):
1096
"""See VersionedFile.annotate_iter."""
1097
version_id = osutils.safe_revision_id(version_id)
1098
return self.factory.annotate_iter(self, version_id)
1100
def get_parents(self, version_id):
1101
"""See VersionedFile.get_parents."""
1104
# 52554 calls in 1264 872 internal down from 3674
1105
version_id = osutils.safe_revision_id(version_id)
1107
return self._index.get_parents(version_id)
1109
raise RevisionNotPresent(version_id, self.filename)
1111
def get_parents_with_ghosts(self, version_id):
1112
"""See VersionedFile.get_parents."""
1113
version_id = osutils.safe_revision_id(version_id)
1115
return self._index.get_parents_with_ghosts(version_id)
1117
raise RevisionNotPresent(version_id, self.filename)
1119
def get_ancestry(self, versions, topo_sorted=True):
1120
"""See VersionedFile.get_ancestry."""
1121
if isinstance(versions, basestring):
1122
versions = [versions]
1125
versions = [osutils.safe_revision_id(v) for v in versions]
1126
return self._index.get_ancestry(versions, topo_sorted)
1128
def get_ancestry_with_ghosts(self, versions):
1129
"""See VersionedFile.get_ancestry_with_ghosts."""
1130
if isinstance(versions, basestring):
1131
versions = [versions]
1134
versions = [osutils.safe_revision_id(v) for v in versions]
1135
return self._index.get_ancestry_with_ghosts(versions)
1137
def plan_merge(self, ver_a, ver_b):
1138
"""See VersionedFile.plan_merge."""
1139
ver_a = osutils.safe_revision_id(ver_a)
1140
ver_b = osutils.safe_revision_id(ver_b)
1141
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1143
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1144
annotated_a = self.annotate(ver_a)
1145
annotated_b = self.annotate(ver_b)
1146
return merge._plan_annotate_merge(annotated_a, annotated_b,
1147
ancestors_a, ancestors_b)
1150
class _KnitComponentFile(object):
1151
"""One of the files used to implement a knit database"""
1153
def __init__(self, transport, filename, mode, file_mode=None,
1154
create_parent_dir=False, dir_mode=None):
1155
self._transport = transport
1156
self._filename = filename
1158
self._file_mode = file_mode
1159
self._dir_mode = dir_mode
1160
self._create_parent_dir = create_parent_dir
1161
self._need_to_create = False
1163
def _full_path(self):
1164
"""Return the full path to this file."""
1165
return self._transport.base + self._filename
1167
def check_header(self, fp):
1168
line = fp.readline()
1170
# An empty file can actually be treated as though the file doesn't
1172
raise errors.NoSuchFile(self._full_path())
1173
if line != self.HEADER:
1174
raise KnitHeaderError(badline=line,
1175
filename=self._transport.abspath(self._filename))
1178
return '%s(%s)' % (self.__class__.__name__, self._filename)
1181
class _KnitIndex(_KnitComponentFile):
1182
"""Manages knit index file.
1184
The index is already kept in memory and read on startup, to enable
1185
fast lookups of revision information. The cursor of the index
1186
file is always pointing to the end, making it easy to append
1189
_cache is a cache for fast mapping from version id to a Index
1192
_history is a cache for fast mapping from indexes to version ids.
1194
The index data format is dictionary compressed when it comes to
1195
parent references; a index entry may only have parents that with a
1196
lover index number. As a result, the index is topological sorted.
1198
Duplicate entries may be written to the index for a single version id
1199
if this is done then the latter one completely replaces the former:
1200
this allows updates to correct version and parent information.
1201
Note that the two entries may share the delta, and that successive
1202
annotations and references MUST point to the first entry.
1204
The index file on disc contains a header, followed by one line per knit
1205
record. The same revision can be present in an index file more than once.
1206
The first occurrence gets assigned a sequence number starting from 0.
1208
The format of a single line is
1209
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1210
REVISION_ID is a utf8-encoded revision id
1211
FLAGS is a comma separated list of flags about the record. Values include
1212
no-eol, line-delta, fulltext.
1213
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1214
that the the compressed data starts at.
1215
LENGTH is the ascii representation of the length of the data file.
1216
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1218
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1219
revision id already in the knit that is a parent of REVISION_ID.
1220
The ' :' marker is the end of record marker.
1223
when a write is interrupted to the index file, it will result in a line
1224
that does not end in ' :'. If the ' :' is not present at the end of a line,
1225
or at the end of the file, then the record that is missing it will be
1226
ignored by the parser.
1228
When writing new records to the index file, the data is preceded by '\n'
1229
to ensure that records always start on new lines even if the last write was
1230
interrupted. As a result its normal for the last line in the index to be
1231
missing a trailing newline. One can be added with no harmful effects.
1234
HEADER = "# bzr knit index 8\n"
1236
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1237
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1239
def _cache_version(self, version_id, options, pos, size, parents):
1240
"""Cache a version record in the history array and index cache.
1242
This is inlined into _load_data for performance. KEEP IN SYNC.
1243
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1246
# only want the _history index to reference the 1st index entry
1248
if version_id not in self._cache:
1249
index = len(self._history)
1250
self._history.append(version_id)
1252
index = self._cache[version_id][5]
1253
self._cache[version_id] = (version_id,
1260
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1261
create_parent_dir=False, delay_create=False, dir_mode=None):
1262
_KnitComponentFile.__init__(self, transport, filename, mode,
1263
file_mode=file_mode,
1264
create_parent_dir=create_parent_dir,
1267
# position in _history is the 'official' index for a revision
1268
# but the values may have come from a newer entry.
1269
# so - wc -l of a knit index is != the number of unique names
1273
fp = self._transport.get(self._filename)
1275
# _load_data may raise NoSuchFile if the target knit is
1277
_load_data(self, fp)
1281
if mode != 'w' or not create:
1284
self._need_to_create = True
1286
self._transport.put_bytes_non_atomic(
1287
self._filename, self.HEADER, mode=self._file_mode)
1289
def get_graph(self):
1290
"""Return a list of the node:parents lists from this knit index."""
1291
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1293
def get_ancestry(self, versions, topo_sorted=True):
1294
"""See VersionedFile.get_ancestry."""
1295
# get a graph of all the mentioned versions:
1297
pending = set(versions)
1300
version = pending.pop()
1303
parents = [p for p in cache[version][4] if p in cache]
1305
raise RevisionNotPresent(version, self._filename)
1306
# if not completed and not a ghost
1307
pending.update([p for p in parents if p not in graph])
1308
graph[version] = parents
1311
return topo_sort(graph.items())
1313
def get_ancestry_with_ghosts(self, versions):
1314
"""See VersionedFile.get_ancestry_with_ghosts."""
1315
# get a graph of all the mentioned versions:
1316
self.check_versions_present(versions)
1319
pending = set(versions)
1321
version = pending.pop()
1323
parents = cache[version][4]
1329
pending.update([p for p in parents if p not in graph])
1330
graph[version] = parents
1331
return topo_sort(graph.items())
1333
def iter_parents(self, version_ids):
1334
"""Iterate through the parents for many version ids.
1336
:param version_ids: An iterable yielding version_ids.
1337
:return: An iterator that yields (version_id, parents). Requested
1338
version_ids not present in the versioned file are simply skipped.
1339
The order is undefined, allowing for different optimisations in
1340
the underlying implementation.
1342
for version_id in version_ids:
1344
yield version_id, tuple(self.get_parents(version_id))
1348
def num_versions(self):
1349
return len(self._history)
1351
__len__ = num_versions
1353
def get_versions(self):
1354
"""Get all the versions in the file. not topologically sorted."""
1355
return self._history
1357
def _version_list_to_index(self, versions):
1360
for version in versions:
1361
if version in cache:
1362
# -- inlined lookup() --
1363
result_list.append(str(cache[version][5]))
1364
# -- end lookup () --
1366
result_list.append('.' + version)
1367
return ' '.join(result_list)
1369
def add_version(self, version_id, options, index_memo, parents):
1370
"""Add a version record to the index."""
1371
self.add_versions(((version_id, options, index_memo, parents),))
1373
def add_versions(self, versions, random_id=False):
1374
"""Add multiple versions to the index.
1376
:param versions: a list of tuples:
1377
(version_id, options, pos, size, parents).
1378
:param random_id: If True the ids being added were randomly generated
1379
and no check for existence will be performed.
1382
orig_history = self._history[:]
1383
orig_cache = self._cache.copy()
1386
for version_id, options, (index, pos, size), parents in versions:
1387
line = "\n%s %s %s %s %s :" % (version_id,
1391
self._version_list_to_index(parents))
1392
assert isinstance(line, str), \
1393
'content must be utf-8 encoded: %r' % (line,)
1395
self._cache_version(version_id, options, pos, size, parents)
1396
if not self._need_to_create:
1397
self._transport.append_bytes(self._filename, ''.join(lines))
1400
sio.write(self.HEADER)
1401
sio.writelines(lines)
1403
self._transport.put_file_non_atomic(self._filename, sio,
1404
create_parent_dir=self._create_parent_dir,
1405
mode=self._file_mode,
1406
dir_mode=self._dir_mode)
1407
self._need_to_create = False
1409
# If any problems happen, restore the original values and re-raise
1410
self._history = orig_history
1411
self._cache = orig_cache
1414
def has_version(self, version_id):
1415
"""True if the version is in the index."""
1416
return version_id in self._cache
1418
def get_position(self, version_id):
1419
"""Return details needed to access the version.
1421
.kndx indices do not support split-out data, so return None for the
1424
:return: a tuple (None, data position, size) to hand to the access
1425
logic to get the record.
1427
entry = self._cache[version_id]
1428
return None, entry[2], entry[3]
1430
def get_method(self, version_id):
1431
"""Return compression method of specified version."""
1432
options = self._cache[version_id][1]
1433
if 'fulltext' in options:
1436
if 'line-delta' not in options:
1437
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1440
def get_options(self, version_id):
1441
"""Return a string represention options.
1445
return self._cache[version_id][1]
1447
def get_parents(self, version_id):
1448
"""Return parents of specified version ignoring ghosts."""
1449
return [parent for parent in self._cache[version_id][4]
1450
if parent in self._cache]
1452
def get_parents_with_ghosts(self, version_id):
1453
"""Return parents of specified version with ghosts."""
1454
return self._cache[version_id][4]
1456
def check_versions_present(self, version_ids):
1457
"""Check that all specified versions are present."""
1459
for version_id in version_ids:
1460
if version_id not in cache:
1461
raise RevisionNotPresent(version_id, self._filename)
1464
class KnitGraphIndex(object):
1465
"""A knit index that builds on GraphIndex."""
1467
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1468
"""Construct a KnitGraphIndex on a graph_index.
1470
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1471
:param deltas: Allow delta-compressed records.
1472
:param add_callback: If not None, allow additions to the index and call
1473
this callback with a list of added GraphIndex nodes:
1474
[(node, value, node_refs), ...]
1475
:param parents: If True, record knits parents, if not do not record
1478
self._graph_index = graph_index
1479
self._deltas = deltas
1480
self._add_callback = add_callback
1481
self._parents = parents
1482
if deltas and not parents:
1483
raise KnitCorrupt(self, "Cannot do delta compression without "
1486
def _get_entries(self, keys, check_present=False):
1487
"""Get the entries for keys.
1489
:param keys: An iterable of index keys, - 1-tuples.
1494
for node in self._graph_index.iter_entries(keys):
1496
found_keys.add(node[1])
1498
# adapt parentless index to the rest of the code.
1499
for node in self._graph_index.iter_entries(keys):
1500
yield node[0], node[1], node[2], ()
1501
found_keys.add(node[1])
1503
missing_keys = keys.difference(found_keys)
1505
raise RevisionNotPresent(missing_keys.pop(), self)
1507
def _present_keys(self, version_ids):
1509
node[1] for node in self._get_entries(version_ids)])
1511
def _parentless_ancestry(self, versions):
1512
"""Honour the get_ancestry API for parentless knit indices."""
1513
wanted_keys = self._version_ids_to_keys(versions)
1514
present_keys = self._present_keys(wanted_keys)
1515
missing = set(wanted_keys).difference(present_keys)
1517
raise RevisionNotPresent(missing.pop(), self)
1518
return list(self._keys_to_version_ids(present_keys))
1520
def get_ancestry(self, versions, topo_sorted=True):
1521
"""See VersionedFile.get_ancestry."""
1522
if not self._parents:
1523
return self._parentless_ancestry(versions)
1524
# XXX: This will do len(history) index calls - perhaps
1525
# it should be altered to be a index core feature?
1526
# get a graph of all the mentioned versions:
1529
versions = self._version_ids_to_keys(versions)
1530
pending = set(versions)
1532
# get all pending nodes
1533
this_iteration = pending
1534
new_nodes = self._get_entries(this_iteration)
1537
for (index, key, value, node_refs) in new_nodes:
1538
# dont ask for ghosties - otherwise
1539
# we we can end up looping with pending
1540
# being entirely ghosted.
1541
graph[key] = [parent for parent in node_refs[0]
1542
if parent not in ghosts]
1544
for parent in graph[key]:
1545
# dont examine known nodes again
1550
ghosts.update(this_iteration.difference(found))
1551
if versions.difference(graph):
1552
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1554
result_keys = topo_sort(graph.items())
1556
result_keys = graph.iterkeys()
1557
return [key[0] for key in result_keys]
1559
def get_ancestry_with_ghosts(self, versions):
1560
"""See VersionedFile.get_ancestry."""
1561
if not self._parents:
1562
return self._parentless_ancestry(versions)
1563
# XXX: This will do len(history) index calls - perhaps
1564
# it should be altered to be a index core feature?
1565
# get a graph of all the mentioned versions:
1567
versions = self._version_ids_to_keys(versions)
1568
pending = set(versions)
1570
# get all pending nodes
1571
this_iteration = pending
1572
new_nodes = self._get_entries(this_iteration)
1574
for (index, key, value, node_refs) in new_nodes:
1575
graph[key] = node_refs[0]
1577
for parent in graph[key]:
1578
# dont examine known nodes again
1582
missing_versions = this_iteration.difference(graph)
1583
missing_needed = versions.intersection(missing_versions)
1585
raise RevisionNotPresent(missing_needed.pop(), self)
1586
for missing_version in missing_versions:
1587
# add a key, no parents
1588
graph[missing_version] = []
1589
pending.discard(missing_version) # don't look for it
1590
result_keys = topo_sort(graph.items())
1591
return [key[0] for key in result_keys]
1593
def get_graph(self):
1594
"""Return a list of the node:parents lists from this knit index."""
1595
if not self._parents:
1596
return [(key, ()) for key in self.get_versions()]
1598
for index, key, value, refs in self._graph_index.iter_all_entries():
1599
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1602
def iter_parents(self, version_ids):
1603
"""Iterate through the parents for many version ids.
1605
:param version_ids: An iterable yielding version_ids.
1606
:return: An iterator that yields (version_id, parents). Requested
1607
version_ids not present in the versioned file are simply skipped.
1608
The order is undefined, allowing for different optimisations in
1609
the underlying implementation.
1612
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1614
present_parents = set()
1615
for node in all_nodes:
1616
all_parents.update(node[3][0])
1617
# any node we are querying must be present
1618
present_parents.add(node[1])
1619
unknown_parents = all_parents.difference(present_parents)
1620
present_parents.update(self._present_keys(unknown_parents))
1621
for node in all_nodes:
1623
for parent in node[3][0]:
1624
if parent in present_parents:
1625
parents.append(parent[0])
1626
yield node[1][0], tuple(parents)
1628
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1629
yield node[1][0], ()
1631
def num_versions(self):
1632
return len(list(self._graph_index.iter_all_entries()))
1634
__len__ = num_versions
1636
def get_versions(self):
1637
"""Get all the versions in the file. not topologically sorted."""
1638
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1640
def has_version(self, version_id):
1641
"""True if the version is in the index."""
1642
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1644
def _keys_to_version_ids(self, keys):
1645
return tuple(key[0] for key in keys)
1647
def get_position(self, version_id):
1648
"""Return details needed to access the version.
1650
:return: a tuple (index, data position, size) to hand to the access
1651
logic to get the record.
1653
node = self._get_node(version_id)
1654
bits = node[2][1:].split(' ')
1655
return node[0], int(bits[0]), int(bits[1])
1657
def get_method(self, version_id):
1658
"""Return compression method of specified version."""
1659
if not self._deltas:
1661
return self._parent_compression(self._get_node(version_id)[3][1])
1663
def _parent_compression(self, reference_list):
1664
# use the second reference list to decide if this is delta'd or not.
1665
if len(reference_list):
1670
def _get_node(self, version_id):
1671
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1673
def get_options(self, version_id):
1674
"""Return a string represention options.
1678
node = self._get_node(version_id)
1679
if not self._deltas:
1680
options = ['fulltext']
1682
options = [self._parent_compression(node[3][1])]
1683
if node[2][0] == 'N':
1684
options.append('no-eol')
1687
def get_parents(self, version_id):
1688
"""Return parents of specified version ignoring ghosts."""
1689
parents = list(self.iter_parents([version_id]))
1692
raise errors.RevisionNotPresent(version_id, self)
1693
return parents[0][1]
1695
def get_parents_with_ghosts(self, version_id):
1696
"""Return parents of specified version with ghosts."""
1697
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1698
check_present=True))
1699
if not self._parents:
1701
return self._keys_to_version_ids(nodes[0][3][0])
1703
def check_versions_present(self, version_ids):
1704
"""Check that all specified versions are present."""
1705
keys = self._version_ids_to_keys(version_ids)
1706
present = self._present_keys(keys)
1707
missing = keys.difference(present)
1709
raise RevisionNotPresent(missing.pop(), self)
1711
def add_version(self, version_id, options, access_memo, parents):
1712
"""Add a version record to the index."""
1713
return self.add_versions(((version_id, options, access_memo, parents),))
1715
def add_versions(self, versions, random_id=False):
1716
"""Add multiple versions to the index.
1718
This function does not insert data into the Immutable GraphIndex
1719
backing the KnitGraphIndex, instead it prepares data for insertion by
1720
the caller and checks that it is safe to insert then calls
1721
self._add_callback with the prepared GraphIndex nodes.
1723
:param versions: a list of tuples:
1724
(version_id, options, pos, size, parents).
1725
:param random_id: If True the ids being added were randomly generated
1726
and no check for existence will be performed.
1728
if not self._add_callback:
1729
raise errors.ReadOnlyError(self)
1730
# we hope there are no repositories with inconsistent parentage
1735
for (version_id, options, access_memo, parents) in versions:
1736
index, pos, size = access_memo
1737
key = (version_id, )
1738
parents = tuple((parent, ) for parent in parents)
1739
if 'no-eol' in options:
1743
value += "%d %d" % (pos, size)
1744
if not self._deltas:
1745
if 'line-delta' in options:
1746
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1749
if 'line-delta' in options:
1750
node_refs = (parents, (parents[0],))
1752
node_refs = (parents, ())
1754
node_refs = (parents, )
1757
raise KnitCorrupt(self, "attempt to add node with parents "
1758
"in parentless index.")
1760
keys[key] = (value, node_refs)
1762
present_nodes = self._get_entries(keys)
1763
for (index, key, value, node_refs) in present_nodes:
1764
if (value, node_refs) != keys[key]:
1765
raise KnitCorrupt(self, "inconsistent details in add_versions"
1766
": %s %s" % ((value, node_refs), keys[key]))
1770
for key, (value, node_refs) in keys.iteritems():
1771
result.append((key, value, node_refs))
1773
for key, (value, node_refs) in keys.iteritems():
1774
result.append((key, value))
1775
self._add_callback(result)
1777
def _version_ids_to_keys(self, version_ids):
1778
return set((version_id, ) for version_id in version_ids)
1781
class _KnitAccess(object):
1782
"""Access to knit records in a .knit file."""
1784
def __init__(self, transport, filename, _file_mode, _dir_mode,
1785
_need_to_create, _create_parent_dir):
1786
"""Create a _KnitAccess for accessing and inserting data.
1788
:param transport: The transport the .knit is located on.
1789
:param filename: The filename of the .knit.
1791
self._transport = transport
1792
self._filename = filename
1793
self._file_mode = _file_mode
1794
self._dir_mode = _dir_mode
1795
self._need_to_create = _need_to_create
1796
self._create_parent_dir = _create_parent_dir
1798
def add_raw_records(self, sizes, raw_data):
1799
"""Add raw knit bytes to a storage area.
1801
The data is spooled to whereever the access method is storing data.
1803
:param sizes: An iterable containing the size of each raw data segment.
1804
:param raw_data: A bytestring containing the data.
1805
:return: A list of memos to retrieve the record later. Each memo is a
1806
tuple - (index, pos, length), where the index field is always None
1807
for the .knit access method.
1809
assert type(raw_data) == str, \
1810
'data must be plain bytes was %s' % type(raw_data)
1811
if not self._need_to_create:
1812
base = self._transport.append_bytes(self._filename, raw_data)
1814
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1815
create_parent_dir=self._create_parent_dir,
1816
mode=self._file_mode,
1817
dir_mode=self._dir_mode)
1818
self._need_to_create = False
1822
result.append((None, base, size))
1827
"""IFF this data access has its own storage area, initialise it.
1831
self._transport.put_bytes_non_atomic(self._filename, '',
1832
mode=self._file_mode)
1834
def open_file(self):
1835
"""IFF this data access can be represented as a single file, open it.
1837
For knits that are not mapped to a single file on disk this will
1840
:return: None or a file handle.
1843
return self._transport.get(self._filename)
1848
def get_raw_records(self, memos_for_retrieval):
1849
"""Get the raw bytes for a records.
1851
:param memos_for_retrieval: An iterable containing the (index, pos,
1852
length) memo for retrieving the bytes. The .knit method ignores
1853
the index as there is always only a single file.
1854
:return: An iterator over the bytes of the records.
1856
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
1857
for pos, data in self._transport.readv(self._filename, read_vector):
1861
class _PackAccess(object):
1862
"""Access to knit records via a collection of packs."""
1864
def __init__(self, index_to_packs, writer=None):
1865
"""Create a _PackAccess object.
1867
:param index_to_packs: A dict mapping index objects to the transport
1868
and file names for obtaining data.
1869
:param writer: A tuple (pack.ContainerWriter, write_index) which
1870
contains the pack to write, and the index that reads from it will
1874
self.container_writer = writer[0]
1875
self.write_index = writer[1]
1877
self.container_writer = None
1878
self.write_index = None
1879
self.indices = index_to_packs
1881
def add_raw_records(self, sizes, raw_data):
1882
"""Add raw knit bytes to a storage area.
1884
The data is spooled to the container writer in one bytes-record per
1887
:param sizes: An iterable containing the size of each raw data segment.
1888
:param raw_data: A bytestring containing the data.
1889
:return: A list of memos to retrieve the record later. Each memo is a
1890
tuple - (index, pos, length), where the index field is the
1891
write_index object supplied to the PackAccess object.
1893
assert type(raw_data) == str, \
1894
'data must be plain bytes was %s' % type(raw_data)
1898
p_offset, p_length = self.container_writer.add_bytes_record(
1899
raw_data[offset:offset+size], [])
1901
result.append((self.write_index, p_offset, p_length))
1905
"""Pack based knits do not get individually created."""
1907
def get_raw_records(self, memos_for_retrieval):
1908
"""Get the raw bytes for a records.
1910
:param memos_for_retrieval: An iterable containing the (index, pos,
1911
length) memo for retrieving the bytes. The Pack access method
1912
looks up the pack to use for a given record in its index_to_pack
1914
:return: An iterator over the bytes of the records.
1916
# first pass, group into same-index requests
1918
current_index = None
1919
for (index, offset, length) in memos_for_retrieval:
1920
if current_index == index:
1921
current_list.append((offset, length))
1923
if current_index is not None:
1924
request_lists.append((current_index, current_list))
1925
current_index = index
1926
current_list = [(offset, length)]
1927
# handle the last entry
1928
if current_index is not None:
1929
request_lists.append((current_index, current_list))
1930
for index, offsets in request_lists:
1931
transport, path = self.indices[index]
1932
reader = pack.make_readv_reader(transport, path, offsets)
1933
for names, read_func in reader.iter_records():
1934
yield read_func(None)
1936
def open_file(self):
1937
"""Pack based knits have no single file."""
1940
def set_writer(self, writer, index, (transport, packname)):
1941
"""Set a writer to use for adding data."""
1942
self.indices[index] = (transport, packname)
1943
self.container_writer = writer
1944
self.write_index = index
1947
class _KnitData(object):
1948
"""Manage extraction of data from a KnitAccess, caching and decompressing.
1950
The KnitData class provides the logic for parsing and using knit records,
1951
making use of an access method for the low level read and write operations.
1954
def __init__(self, access):
1955
"""Create a KnitData object.
1957
:param access: The access method to use. Access methods such as
1958
_KnitAccess manage the insertion of raw records and the subsequent
1959
retrieval of the same.
1961
self._access = access
1962
self._checked = False
1963
# TODO: jam 20060713 conceptually, this could spill to disk
1964
# if the cached size gets larger than a certain amount
1965
# but it complicates the model a bit, so for now just use
1966
# a simple dictionary
1968
self._do_cache = False
1970
def enable_cache(self):
1971
"""Enable caching of reads."""
1972
self._do_cache = True
1974
def clear_cache(self):
1975
"""Clear the record cache."""
1976
self._do_cache = False
1979
def _open_file(self):
1980
return self._access.open_file()
1982
def _record_to_data(self, version_id, digest, lines):
1983
"""Convert version_id, digest, lines into a raw data block.
1985
:return: (len, a StringIO instance with the raw data ready to read.)
1987
bytes = (''.join(chain(
1988
["version %s %d %s\n" % (version_id,
1992
["end %s\n" % version_id])))
1993
assert bytes.__class__ == str
1994
compressed_bytes = bytes_to_gzip(bytes)
1995
return len(compressed_bytes), compressed_bytes
1997
def add_raw_records(self, sizes, raw_data):
1998
"""Append a prepared record to the data file.
2000
:param sizes: An iterable containing the size of each raw data segment.
2001
:param raw_data: A bytestring containing the data.
2002
:return: a list of index data for the way the data was stored.
2003
See the access method add_raw_records documentation for more
2006
return self._access.add_raw_records(sizes, raw_data)
2008
def _parse_record_header(self, version_id, raw_data):
2009
"""Parse a record header for consistency.
2011
:return: the header and the decompressor stream.
2012
as (stream, header_record)
2014
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2016
rec = self._check_header(version_id, df.readline())
2017
except Exception, e:
2018
raise KnitCorrupt(self._access,
2019
"While reading {%s} got %s(%s)"
2020
% (version_id, e.__class__.__name__, str(e)))
2023
def _check_header(self, version_id, line):
2026
raise KnitCorrupt(self._access,
2027
'unexpected number of elements in record header')
2028
if rec[1] != version_id:
2029
raise KnitCorrupt(self._access,
2030
'unexpected version, wanted %r, got %r'
2031
% (version_id, rec[1]))
2034
def _parse_record(self, version_id, data):
2036
# 4168 calls in 2880 217 internal
2037
# 4168 calls to _parse_record_header in 2121
2038
# 4168 calls to readlines in 330
2039
df = GzipFile(mode='rb', fileobj=StringIO(data))
2042
record_contents = df.readlines()
2043
except Exception, e:
2044
raise KnitCorrupt(self._access,
2045
"While reading {%s} got %s(%s)"
2046
% (version_id, e.__class__.__name__, str(e)))
2047
header = record_contents.pop(0)
2048
rec = self._check_header(version_id, header)
2050
last_line = record_contents.pop()
2051
if len(record_contents) != int(rec[2]):
2052
raise KnitCorrupt(self._access,
2053
'incorrect number of lines %s != %s'
2055
% (len(record_contents), int(rec[2]),
2057
if last_line != 'end %s\n' % rec[1]:
2058
raise KnitCorrupt(self._access,
2059
'unexpected version end line %r, wanted %r'
2060
% (last_line, version_id))
2062
return record_contents, rec[3]
2064
def read_records_iter_raw(self, records):
2065
"""Read text records from data file and yield raw data.
2067
This unpacks enough of the text record to validate the id is
2068
as expected but thats all.
2070
# setup an iterator of the external records:
2071
# uses readv so nice and fast we hope.
2073
# grab the disk data needed.
2075
# Don't check _cache if it is empty
2076
needed_offsets = [index_memo for version_id, index_memo
2078
if version_id not in self._cache]
2080
needed_offsets = [index_memo for version_id, index_memo
2083
raw_records = self._access.get_raw_records(needed_offsets)
2085
for version_id, index_memo in records:
2086
if version_id in self._cache:
2087
# This data has already been validated
2088
data = self._cache[version_id]
2090
data = raw_records.next()
2092
self._cache[version_id] = data
2094
# validate the header
2095
df, rec = self._parse_record_header(version_id, data)
2097
yield version_id, data
2099
def read_records_iter(self, records):
2100
"""Read text records from data file and yield result.
2102
The result will be returned in whatever is the fastest to read.
2103
Not by the order requested. Also, multiple requests for the same
2104
record will only yield 1 response.
2105
:param records: A list of (version_id, pos, len) entries
2106
:return: Yields (version_id, contents, digest) in the order
2107
read, not the order requested
2113
# Skip records we have alread seen
2114
yielded_records = set()
2115
needed_records = set()
2116
for record in records:
2117
if record[0] in self._cache:
2118
if record[0] in yielded_records:
2120
yielded_records.add(record[0])
2121
data = self._cache[record[0]]
2122
content, digest = self._parse_record(record[0], data)
2123
yield (record[0], content, digest)
2125
needed_records.add(record)
2126
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2128
needed_records = sorted(set(records), key=operator.itemgetter(1))
2130
if not needed_records:
2133
# The transport optimizes the fetching as well
2134
# (ie, reads continuous ranges.)
2135
raw_data = self._access.get_raw_records(
2136
[index_memo for version_id, index_memo in needed_records])
2138
for (version_id, index_memo), data in \
2139
izip(iter(needed_records), raw_data):
2140
content, digest = self._parse_record(version_id, data)
2142
self._cache[version_id] = data
2143
yield version_id, content, digest
2145
def read_records(self, records):
2146
"""Read records into a dictionary."""
2148
for record_id, content, digest in \
2149
self.read_records_iter(records):
2150
components[record_id] = (content, digest)
2154
class InterKnit(InterVersionedFile):
2155
"""Optimised code paths for knit to knit operations."""
2157
_matching_file_from_factory = KnitVersionedFile
2158
_matching_file_to_factory = KnitVersionedFile
2161
def is_compatible(source, target):
2162
"""Be compatible with knits. """
2164
return (isinstance(source, KnitVersionedFile) and
2165
isinstance(target, KnitVersionedFile))
2166
except AttributeError:
2169
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2170
"""See InterVersionedFile.join."""
2171
assert isinstance(self.source, KnitVersionedFile)
2172
assert isinstance(self.target, KnitVersionedFile)
2174
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2179
pb = ui.ui_factory.nested_progress_bar()
2181
version_ids = list(version_ids)
2182
if None in version_ids:
2183
version_ids.remove(None)
2185
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2186
this_versions = set(self.target._index.get_versions())
2187
# XXX: For efficiency we should not look at the whole index,
2188
# we only need to consider the referenced revisions - they
2189
# must all be present, or the method must be full-text.
2190
# TODO, RBC 20070919
2191
needed_versions = self.source_ancestry - this_versions
2193
if not needed_versions:
2195
full_list = topo_sort(self.source.get_graph())
2197
version_list = [i for i in full_list if (not self.target.has_version(i)
2198
and i in needed_versions)]
2202
copy_queue_records = []
2204
for version_id in version_list:
2205
options = self.source._index.get_options(version_id)
2206
parents = self.source._index.get_parents_with_ghosts(version_id)
2207
# check that its will be a consistent copy:
2208
for parent in parents:
2209
# if source has the parent, we must :
2210
# * already have it or
2211
# * have it scheduled already
2212
# otherwise we don't care
2213
assert (self.target.has_version(parent) or
2214
parent in copy_set or
2215
not self.source.has_version(parent))
2216
index_memo = self.source._index.get_position(version_id)
2217
copy_queue_records.append((version_id, index_memo))
2218
copy_queue.append((version_id, options, parents))
2219
copy_set.add(version_id)
2221
# data suck the join:
2223
total = len(version_list)
2226
for (version_id, raw_data), \
2227
(version_id2, options, parents) in \
2228
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2230
assert version_id == version_id2, 'logic error, inconsistent results'
2232
pb.update("Joining knit", count, total)
2233
raw_records.append((version_id, options, parents, len(raw_data)))
2234
raw_datum.append(raw_data)
2235
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2241
InterVersionedFile.register_optimiser(InterKnit)
2244
class WeaveToKnit(InterVersionedFile):
2245
"""Optimised code paths for weave to knit operations."""
2247
_matching_file_from_factory = bzrlib.weave.WeaveFile
2248
_matching_file_to_factory = KnitVersionedFile
2251
def is_compatible(source, target):
2252
"""Be compatible with weaves to knits."""
2254
return (isinstance(source, bzrlib.weave.Weave) and
2255
isinstance(target, KnitVersionedFile))
2256
except AttributeError:
2259
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2260
"""See InterVersionedFile.join."""
2261
assert isinstance(self.source, bzrlib.weave.Weave)
2262
assert isinstance(self.target, KnitVersionedFile)
2264
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2269
pb = ui.ui_factory.nested_progress_bar()
2271
version_ids = list(version_ids)
2273
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2274
this_versions = set(self.target._index.get_versions())
2275
needed_versions = self.source_ancestry - this_versions
2277
if not needed_versions:
2279
full_list = topo_sort(self.source.get_graph())
2281
version_list = [i for i in full_list if (not self.target.has_version(i)
2282
and i in needed_versions)]
2286
total = len(version_list)
2287
for version_id in version_list:
2288
pb.update("Converting to knit", count, total)
2289
parents = self.source.get_parents(version_id)
2290
# check that its will be a consistent copy:
2291
for parent in parents:
2292
# if source has the parent, we must already have it
2293
assert (self.target.has_version(parent))
2294
self.target.add_lines(
2295
version_id, parents, self.source.get_lines(version_id))
2302
InterVersionedFile.register_optimiser(WeaveToKnit)
2305
# Deprecated, use PatienceSequenceMatcher instead
2306
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2309
def annotate_knit(knit, revision_id):
2310
"""Annotate a knit with no cached annotations.
2312
This implementation is for knits with no cached annotations.
2313
It will work for knits with cached annotations, but this is not
2316
ancestry = knit.get_ancestry(revision_id)
2317
fulltext = dict(zip(ancestry, knit.get_line_list(ancestry)))
2319
for candidate in ancestry:
2320
if candidate in annotations:
2322
parents = knit.get_parents(candidate)
2323
if len(parents) == 0:
2325
elif knit._index.get_method(candidate) != 'line-delta':
2328
parent, sha1, noeol, delta = knit.get_delta(candidate)
2329
blocks = KnitContent.get_line_delta_blocks(delta,
2330
fulltext[parents[0]], fulltext[candidate])
2331
annotations[candidate] = list(annotate.reannotate([annotations[p]
2332
for p in parents], fulltext[candidate], candidate, blocks))
2333
return iter(annotations[revision_id])
2337
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2339
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
