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 apply_delta(self, delta, new_version_id):
142
"""Apply delta to this object to become new_version_id."""
143
raise NotImplementedError(self.apply_delta)
145
def line_delta_iter(self, new_lines):
146
"""Generate line-based delta from this content to new_lines."""
147
new_texts = new_lines.text()
148
old_texts = self.text()
149
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
150
for tag, i1, i2, j1, j2 in s.get_opcodes():
153
# ofrom, oto, length, data
154
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
156
def line_delta(self, new_lines):
157
return list(self.line_delta_iter(new_lines))
160
def get_line_delta_blocks(knit_delta, source, target):
161
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
162
target_len = len(target)
165
for s_begin, s_end, t_len, new_text in knit_delta:
166
true_n = s_begin - s_pos
169
# knit deltas do not provide reliable info about whether the
170
# last line of a file matches, due to eol handling.
171
if source[s_pos + n -1] != target[t_pos + n -1]:
174
yield s_pos, t_pos, n
175
t_pos += t_len + true_n
177
n = target_len - t_pos
179
if source[s_pos + n -1] != target[t_pos + n -1]:
182
yield s_pos, t_pos, n
183
yield s_pos + (target_len - t_pos), target_len, 0
186
class AnnotatedKnitContent(KnitContent):
187
"""Annotated content."""
189
def __init__(self, lines):
192
def annotate_iter(self):
193
"""Yield tuples of (origin, text) for each content line."""
194
return iter(self._lines)
196
def apply_delta(self, delta, new_version_id):
197
"""Apply delta to this object to become new_version_id."""
200
for start, end, count, delta_lines in delta:
201
lines[offset+start:offset+end] = delta_lines
202
offset = offset + (start - end) + count
204
def strip_last_line_newline(self):
205
line = self._lines[-1][1].rstrip('\n')
206
self._lines[-1] = (self._lines[-1][0], line)
210
return [text for origin, text in self._lines]
211
except ValueError, e:
212
# most commonly (only?) caused by the internal form of the knit
213
# missing annotation information because of a bug - see thread
215
raise KnitCorrupt(self,
216
"line in annotated knit missing annotation information: %s"
220
return AnnotatedKnitContent(self._lines[:])
223
class PlainKnitContent(KnitContent):
224
"""Unannotated content.
226
When annotate[_iter] is called on this content, the same version is reported
227
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
231
def __init__(self, lines, version_id):
233
self._version_id = version_id
235
def annotate_iter(self):
236
"""Yield tuples of (origin, text) for each content line."""
237
for line in self._lines:
238
yield self._version_id, line
240
def apply_delta(self, delta, new_version_id):
241
"""Apply delta to this object to become new_version_id."""
244
for start, end, count, delta_lines in delta:
245
lines[offset+start:offset+end] = delta_lines
246
offset = offset + (start - end) + count
247
self._version_id = new_version_id
250
return PlainKnitContent(self._lines[:], self._version_id)
252
def strip_last_line_newline(self):
253
self._lines[-1] = self._lines[-1].rstrip('\n')
259
class KnitAnnotateFactory(object):
260
"""Factory for creating annotated Content objects."""
264
def make(self, lines, version_id):
265
num_lines = len(lines)
266
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
268
def parse_fulltext(self, content, version_id):
269
"""Convert fulltext to internal representation
271
fulltext content is of the format
272
revid(utf8) plaintext\n
273
internal representation is of the format:
276
# TODO: jam 20070209 The tests expect this to be returned as tuples,
277
# but the code itself doesn't really depend on that.
278
# Figure out a way to not require the overhead of turning the
279
# list back into tuples.
280
lines = [tuple(line.split(' ', 1)) for line in content]
281
return AnnotatedKnitContent(lines)
283
def parse_line_delta_iter(self, lines):
284
return iter(self.parse_line_delta(lines))
286
def parse_line_delta(self, lines, version_id, plain=False):
287
"""Convert a line based delta into internal representation.
289
line delta is in the form of:
290
intstart intend intcount
292
revid(utf8) newline\n
293
internal representation is
294
(start, end, count, [1..count tuples (revid, newline)])
296
:param plain: If True, the lines are returned as a plain
297
list without annotations, not as a list of (origin, content) tuples, i.e.
298
(start, end, count, [1..count newline])
305
def cache_and_return(line):
306
origin, text = line.split(' ', 1)
307
return cache.setdefault(origin, origin), text
309
# walk through the lines parsing.
310
# Note that the plain test is explicitly pulled out of the
311
# loop to minimise any performance impact
314
start, end, count = [int(n) for n in header.split(',')]
315
contents = [next().split(' ', 1)[1] for i in xrange(count)]
316
result.append((start, end, count, contents))
319
start, end, count = [int(n) for n in header.split(',')]
320
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
321
result.append((start, end, count, contents))
324
def get_fulltext_content(self, lines):
325
"""Extract just the content lines from a fulltext."""
326
return (line.split(' ', 1)[1] for line in lines)
328
def get_linedelta_content(self, lines):
329
"""Extract just the content from a line delta.
331
This doesn't return all of the extra information stored in a delta.
332
Only the actual content lines.
337
header = header.split(',')
338
count = int(header[2])
339
for i in xrange(count):
340
origin, text = next().split(' ', 1)
343
def lower_fulltext(self, content):
344
"""convert a fulltext content record into a serializable form.
346
see parse_fulltext which this inverts.
348
# TODO: jam 20070209 We only do the caching thing to make sure that
349
# the origin is a valid utf-8 line, eventually we could remove it
350
return ['%s %s' % (o, t) for o, t in content._lines]
352
def lower_line_delta(self, delta):
353
"""convert a delta into a serializable form.
355
See parse_line_delta which this inverts.
357
# TODO: jam 20070209 We only do the caching thing to make sure that
358
# the origin is a valid utf-8 line, eventually we could remove it
360
for start, end, c, lines in delta:
361
out.append('%d,%d,%d\n' % (start, end, c))
362
out.extend(origin + ' ' + text
363
for origin, text in lines)
366
def annotate_iter(self, knit, version_id):
367
content = knit._get_content(version_id)
368
return content.annotate_iter()
371
class KnitPlainFactory(object):
372
"""Factory for creating plain Content objects."""
376
def make(self, lines, version_id):
377
return PlainKnitContent(lines, version_id)
379
def parse_fulltext(self, content, version_id):
380
"""This parses an unannotated fulltext.
382
Note that this is not a noop - the internal representation
383
has (versionid, line) - its just a constant versionid.
385
return self.make(content, version_id)
387
def parse_line_delta_iter(self, lines, version_id):
389
num_lines = len(lines)
390
while cur < num_lines:
393
start, end, c = [int(n) for n in header.split(',')]
394
yield start, end, c, lines[cur:cur+c]
397
def parse_line_delta(self, lines, version_id):
398
return list(self.parse_line_delta_iter(lines, version_id))
400
def get_fulltext_content(self, lines):
401
"""Extract just the content lines from a fulltext."""
404
def get_linedelta_content(self, lines):
405
"""Extract just the content from a line delta.
407
This doesn't return all of the extra information stored in a delta.
408
Only the actual content lines.
413
header = header.split(',')
414
count = int(header[2])
415
for i in xrange(count):
418
def lower_fulltext(self, content):
419
return content.text()
421
def lower_line_delta(self, delta):
423
for start, end, c, lines in delta:
424
out.append('%d,%d,%d\n' % (start, end, c))
428
def annotate_iter(self, knit, version_id):
429
return annotate_knit(knit, version_id)
432
def make_empty_knit(transport, relpath):
433
"""Construct a empty knit at the specified location."""
434
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
437
class KnitVersionedFile(VersionedFile):
438
"""Weave-like structure with faster random access.
440
A knit stores a number of texts and a summary of the relationships
441
between them. Texts are identified by a string version-id. Texts
442
are normally stored and retrieved as a series of lines, but can
443
also be passed as single strings.
445
Lines are stored with the trailing newline (if any) included, to
446
avoid special cases for files with no final newline. Lines are
447
composed of 8-bit characters, not unicode. The combination of
448
these approaches should mean any 'binary' file can be safely
449
stored and retrieved.
452
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
453
factory=None, delta=True, create=False, create_parent_dir=False,
454
delay_create=False, dir_mode=None, index=None, access_method=None):
455
"""Construct a knit at location specified by relpath.
457
:param create: If not True, only open an existing knit.
458
:param create_parent_dir: If True, create the parent directory if
459
creating the file fails. (This is used for stores with
460
hash-prefixes that may not exist yet)
461
:param delay_create: The calling code is aware that the knit won't
462
actually be created until the first data is stored.
463
:param index: An index to use for the knit.
465
if access_mode is None:
467
super(KnitVersionedFile, self).__init__(access_mode)
468
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
469
self.transport = transport
470
self.filename = relpath
471
self.factory = factory or KnitAnnotateFactory()
472
self.writable = (access_mode == 'w')
475
self._max_delta_chain = 200
478
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
479
access_mode, create=create, file_mode=file_mode,
480
create_parent_dir=create_parent_dir, delay_create=delay_create,
484
if access_method is None:
485
_access = _KnitAccess(transport, relpath + DATA_SUFFIX, file_mode, dir_mode,
486
((create and not len(self)) and delay_create), create_parent_dir)
488
_access = access_method
489
if create and not len(self) and not delay_create:
491
self._data = _KnitData(_access)
494
return '%s(%s)' % (self.__class__.__name__,
495
self.transport.abspath(self.filename))
497
def _check_should_delta(self, first_parents):
498
"""Iterate back through the parent listing, looking for a fulltext.
500
This is used when we want to decide whether to add a delta or a new
501
fulltext. It searches for _max_delta_chain parents. When it finds a
502
fulltext parent, it sees if the total size of the deltas leading up to
503
it is large enough to indicate that we want a new full text anyway.
505
Return True if we should create a new delta, False if we should use a
510
delta_parents = first_parents
511
for count in xrange(self._max_delta_chain):
512
parent = delta_parents[0]
513
method = self._index.get_method(parent)
514
index, pos, size = self._index.get_position(parent)
515
if method == 'fulltext':
519
delta_parents = self._index.get_parents(parent)
521
# We couldn't find a fulltext, so we must create a new one
524
return fulltext_size > delta_size
526
def _add_raw_records(self, records, data):
527
"""Add all the records 'records' with data pre-joined in 'data'.
529
:param records: A list of tuples(version_id, options, parents, size).
530
:param data: The data for the records. When it is written, the records
531
are adjusted to have pos pointing into data by the sum of
532
the preceding records sizes.
535
raw_record_sizes = [record[3] for record in records]
536
positions = self._data.add_raw_records(raw_record_sizes, data)
539
for (version_id, options, parents, size), access_memo in zip(
541
index_entries.append((version_id, options, access_memo, parents))
542
if self._data._do_cache:
543
self._data._cache[version_id] = data[offset:offset+size]
545
self._index.add_versions(index_entries)
547
def enable_cache(self):
548
"""Start caching data for this knit"""
549
self._data.enable_cache()
551
def clear_cache(self):
552
"""Clear the data cache only."""
553
self._data.clear_cache()
555
def copy_to(self, name, transport):
556
"""See VersionedFile.copy_to()."""
557
# copy the current index to a temp index to avoid racing with local
559
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
560
self.transport.get(self._index._filename))
562
f = self._data._open_file()
564
transport.put_file(name + DATA_SUFFIX, f)
567
# move the copied index into place
568
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
570
def create_empty(self, name, transport, mode=None):
571
return KnitVersionedFile(name, transport, factory=self.factory,
572
delta=self.delta, create=True)
574
def get_data_stream(self, required_versions):
575
"""Get a data stream for the specified versions.
577
Versions may be returned in any order, not necessarily the order
580
:param required_versions: The exact set of versions to be extracted.
581
Unlike some other knit methods, this is not used to generate a
582
transitive closure, rather it is used precisely as given.
584
:returns: format_signature, list of (version, options, length, parents),
587
if not isinstance(required_versions, set):
588
required_versions = set(required_versions)
589
# we don't care about inclusions, the caller cares.
590
# but we need to setup a list of records to visit.
591
for version_id in required_versions:
592
if not self.has_version(version_id):
593
raise RevisionNotPresent(version_id, self.filename)
594
# Pick the desired versions out of the index in oldest-to-newest order
596
for version_id in self.versions():
597
if version_id in required_versions:
598
version_list.append(version_id)
600
# create the list of version information for the result
601
copy_queue_records = []
603
result_version_list = []
604
for version_id in version_list:
605
options = self._index.get_options(version_id)
606
parents = self._index.get_parents_with_ghosts(version_id)
607
index_memo = self._index.get_position(version_id)
608
copy_queue_records.append((version_id, index_memo))
609
none, data_pos, data_size = index_memo
610
copy_set.add(version_id)
611
# version, options, length, parents
612
result_version_list.append((version_id, options, data_size,
615
# Read the compressed record data.
617
# From here down to the return should really be logic in the returned
618
# callable -- in a class that adapts read_records_iter_raw to read
621
for (version_id, raw_data), \
622
(version_id2, options, _, parents) in \
623
izip(self._data.read_records_iter_raw(copy_queue_records),
624
result_version_list):
625
assert version_id == version_id2, 'logic error, inconsistent results'
626
raw_datum.append(raw_data)
627
pseudo_file = StringIO(''.join(raw_datum))
630
return pseudo_file.read()
632
return pseudo_file.read(length)
633
return (self.get_format_signature(), result_version_list, read)
635
def _extract_blocks(self, version_id, source, target):
636
if self._index.get_method(version_id) != 'line-delta':
638
parent, sha1, noeol, delta = self.get_delta(version_id)
639
return KnitContent.get_line_delta_blocks(delta, source, target)
641
def get_delta(self, version_id):
642
"""Get a delta for constructing version from some other version."""
643
self.check_not_reserved_id(version_id)
644
parents = self.get_parents(version_id)
649
index_memo = self._index.get_position(version_id)
650
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
651
noeol = 'no-eol' in self._index.get_options(version_id)
652
if 'fulltext' == self._index.get_method(version_id):
653
new_content = self.factory.parse_fulltext(data, version_id)
654
if parent is not None:
655
reference_content = self._get_content(parent)
656
old_texts = reference_content.text()
659
new_texts = new_content.text()
660
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
662
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
664
delta = self.factory.parse_line_delta(data, version_id)
665
return parent, sha1, noeol, delta
667
def get_format_signature(self):
668
"""See VersionedFile.get_format_signature()."""
669
if self.factory.annotated:
670
annotated_part = "annotated"
672
annotated_part = "plain"
673
return "knit-%s" % (annotated_part,)
675
def get_graph_with_ghosts(self):
676
"""See VersionedFile.get_graph_with_ghosts()."""
677
graph_items = self._index.get_graph()
678
return dict(graph_items)
680
def get_sha1(self, version_id):
681
return self.get_sha1s([version_id])[0]
683
def get_sha1s(self, version_ids):
684
"""See VersionedFile.get_sha1()."""
685
record_map = self._get_record_map(version_ids)
686
# record entry 2 is the 'digest'.
687
return [record_map[v][2] for v in version_ids]
691
"""See VersionedFile.get_suffixes()."""
692
return [DATA_SUFFIX, INDEX_SUFFIX]
694
def has_ghost(self, version_id):
695
"""True if there is a ghost reference in the file to version_id."""
697
if self.has_version(version_id):
699
# optimisable if needed by memoising the _ghosts set.
700
items = self._index.get_graph()
701
for node, parents in items:
702
for parent in parents:
703
if parent not in self._index._cache:
704
if parent == version_id:
708
def insert_data_stream(self, (format, data_list, reader_callable)):
709
"""Insert knit records from a data stream into this knit.
711
If a version in the stream is already present in this knit, it will not
712
be inserted a second time. It will be checked for consistency with the
713
stored version however, and may cause a KnitCorrupt error to be raised
714
if the data in the stream disagrees with the already stored data.
716
:seealso: get_data_stream
718
if format != self.get_format_signature():
719
trace.mutter('incompatible format signature inserting to %r', self)
720
raise KnitDataStreamIncompatible(
721
format, self.get_format_signature())
723
for version_id, options, length, parents in data_list:
724
if self.has_version(version_id):
725
# First check: the list of parents.
726
my_parents = self.get_parents_with_ghosts(version_id)
727
if my_parents != parents:
728
# XXX: KnitCorrupt is not quite the right exception here.
731
'parents list %r from data stream does not match '
732
'already recorded parents %r for %s'
733
% (parents, my_parents, version_id))
735
# Also check the SHA-1 of the fulltext this content will
737
raw_data = reader_callable(length)
738
my_fulltext_sha1 = self.get_sha1(version_id)
739
df, rec = self._data._parse_record_header(version_id, raw_data)
740
stream_fulltext_sha1 = rec[3]
741
if my_fulltext_sha1 != stream_fulltext_sha1:
742
# Actually, we don't know if it's this knit that's corrupt,
743
# or the data stream we're trying to insert.
745
self.filename, 'sha-1 does not match %s' % version_id)
747
if 'line-delta' in options:
748
# Make sure that this knit record is actually useful: a
749
# line-delta is no use unless we have its parent.
750
# Fetching from a broken repository with this problem
751
# shouldn't break the target repository.
752
if not self._index.has_version(parents[0]):
755
'line-delta from stream references '
756
'missing parent %s' % parents[0])
757
self._add_raw_records(
758
[(version_id, options, parents, length)],
759
reader_callable(length))
762
"""See VersionedFile.versions."""
763
if 'evil' in debug.debug_flags:
764
trace.mutter_callsite(2, "versions scales with size of history")
765
return self._index.get_versions()
767
def has_version(self, version_id):
768
"""See VersionedFile.has_version."""
769
if 'evil' in debug.debug_flags:
770
trace.mutter_callsite(2, "has_version is a LBYL scenario")
771
return self._index.has_version(version_id)
773
__contains__ = has_version
775
def _merge_annotations(self, content, parents, parent_texts={},
776
delta=None, annotated=None,
777
left_matching_blocks=None):
778
"""Merge annotations for content. This is done by comparing
779
the annotations based on changed to the text.
781
if left_matching_blocks is not None:
782
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
786
for parent_id in parents:
787
merge_content = self._get_content(parent_id, parent_texts)
788
if (parent_id == parents[0] and delta_seq is not None):
791
seq = patiencediff.PatienceSequenceMatcher(
792
None, merge_content.text(), content.text())
793
for i, j, n in seq.get_matching_blocks():
796
# this appears to copy (origin, text) pairs across to the
797
# new content for any line that matches the last-checked
799
content._lines[j:j+n] = merge_content._lines[i:i+n]
801
if delta_seq is None:
802
reference_content = self._get_content(parents[0], parent_texts)
803
new_texts = content.text()
804
old_texts = reference_content.text()
805
delta_seq = patiencediff.PatienceSequenceMatcher(
806
None, old_texts, new_texts)
807
return self._make_line_delta(delta_seq, content)
809
def _make_line_delta(self, delta_seq, new_content):
810
"""Generate a line delta from delta_seq and new_content."""
812
for op in delta_seq.get_opcodes():
815
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
818
def _get_components_positions(self, version_ids):
819
"""Produce a map of position data for the components of versions.
821
This data is intended to be used for retrieving the knit records.
823
A dict of version_id to (method, data_pos, data_size, next) is
825
method is the way referenced data should be applied.
826
data_pos is the position of the data in the knit.
827
data_size is the size of the data in the knit.
828
next is the build-parent of the version, or None for fulltexts.
831
for version_id in version_ids:
834
while cursor is not None and cursor not in component_data:
835
method = self._index.get_method(cursor)
836
if method == 'fulltext':
839
next = self.get_parents_with_ghosts(cursor)[0]
840
index_memo = self._index.get_position(cursor)
841
component_data[cursor] = (method, index_memo, next)
843
return component_data
845
def _get_content(self, version_id, parent_texts={}):
846
"""Returns a content object that makes up the specified
848
cached_version = parent_texts.get(version_id, None)
849
if cached_version is not None:
850
if not self.has_version(version_id):
851
raise RevisionNotPresent(version_id, self.filename)
852
return cached_version
854
text_map, contents_map = self._get_content_maps([version_id])
855
return contents_map[version_id]
857
def _check_versions_present(self, version_ids):
858
"""Check that all specified versions are present."""
859
self._index.check_versions_present(version_ids)
861
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
862
nostore_sha, random_id, check_content):
863
"""See VersionedFile.add_lines_with_ghosts()."""
864
self._check_add(version_id, lines, random_id, check_content)
865
return self._add(version_id, lines, parents, self.delta,
866
parent_texts, None, nostore_sha, random_id)
868
def _add_lines(self, version_id, parents, lines, parent_texts,
869
left_matching_blocks, nostore_sha, random_id, check_content):
870
"""See VersionedFile.add_lines."""
871
self._check_add(version_id, lines, random_id, check_content)
872
self._check_versions_present(parents)
873
return self._add(version_id, lines[:], parents, self.delta,
874
parent_texts, left_matching_blocks, nostore_sha, random_id)
876
def _check_add(self, version_id, lines, random_id, check_content):
877
"""check that version_id and lines are safe to add."""
878
if contains_whitespace(version_id):
879
raise InvalidRevisionId(version_id, self.filename)
880
self.check_not_reserved_id(version_id)
881
# Technically this could be avoided if we are happy to allow duplicate
882
# id insertion when other things than bzr core insert texts, but it
883
# seems useful for folk using the knit api directly to have some safety
884
# blanket that we can disable.
885
if not random_id and self.has_version(version_id):
886
raise RevisionAlreadyPresent(version_id, self.filename)
888
self._check_lines_not_unicode(lines)
889
self._check_lines_are_lines(lines)
891
def _add(self, version_id, lines, parents, delta, parent_texts,
892
left_matching_blocks, nostore_sha, random_id):
893
"""Add a set of lines on top of version specified by parents.
895
If delta is true, compress the text as a line-delta against
898
Any versions not present will be converted into ghosts.
900
# first thing, if the content is something we don't need to store, find
902
line_bytes = ''.join(lines)
903
digest = sha_string(line_bytes)
904
if nostore_sha == digest:
905
raise errors.ExistingContent
908
if parent_texts is None:
910
for parent in parents:
911
if self.has_version(parent):
912
present_parents.append(parent)
914
# can only compress against the left most present parent.
916
(len(present_parents) == 0 or
917
present_parents[0] != parents[0])):
920
text_length = len(line_bytes)
923
if lines[-1][-1] != '\n':
924
# copy the contents of lines.
926
options.append('no-eol')
927
lines[-1] = lines[-1] + '\n'
931
# To speed the extract of texts the delta chain is limited
932
# to a fixed number of deltas. This should minimize both
933
# I/O and the time spend applying deltas.
934
delta = self._check_should_delta(present_parents)
936
assert isinstance(version_id, str)
937
content = self.factory.make(lines, version_id)
938
if delta or (self.factory.annotated and len(present_parents) > 0):
939
# Merge annotations from parent texts if needed.
940
delta_hunks = self._merge_annotations(content, present_parents,
941
parent_texts, delta, self.factory.annotated,
942
left_matching_blocks)
945
options.append('line-delta')
946
store_lines = self.factory.lower_line_delta(delta_hunks)
947
size, bytes = self._data._record_to_data(version_id, digest,
950
options.append('fulltext')
951
# isinstance is slower and we have no hierarchy.
952
if self.factory.__class__ == KnitPlainFactory:
953
# Use the already joined bytes saving iteration time in
955
size, bytes = self._data._record_to_data(version_id, digest,
958
# get mixed annotation + content and feed it into the
960
store_lines = self.factory.lower_fulltext(content)
961
size, bytes = self._data._record_to_data(version_id, digest,
964
access_memo = self._data.add_raw_records([size], bytes)[0]
965
self._index.add_versions(
966
((version_id, options, access_memo, parents),),
968
return digest, text_length, content
970
def check(self, progress_bar=None):
971
"""See VersionedFile.check()."""
973
def _clone_text(self, new_version_id, old_version_id, parents):
974
"""See VersionedFile.clone_text()."""
975
# FIXME RBC 20060228 make fast by only inserting an index with null
977
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
979
def get_lines(self, version_id):
980
"""See VersionedFile.get_lines()."""
981
return self.get_line_list([version_id])[0]
983
def _get_record_map(self, version_ids):
984
"""Produce a dictionary of knit records.
986
The keys are version_ids, the values are tuples of (method, content,
988
method is the way the content should be applied.
989
content is a KnitContent object.
990
digest is the SHA1 digest of this version id after all steps are done
991
next is the build-parent of the version, i.e. the leftmost ancestor.
992
If the method is fulltext, next will be None.
994
position_map = self._get_components_positions(version_ids)
995
# c = component_id, m = method, i_m = index_memo, n = next
996
records = [(c, i_m) for c, (m, i_m, n) in position_map.iteritems()]
998
for component_id, content, digest in \
999
self._data.read_records_iter(records):
1000
method, index_memo, next = position_map[component_id]
1001
record_map[component_id] = method, content, digest, next
1005
def get_text(self, version_id):
1006
"""See VersionedFile.get_text"""
1007
return self.get_texts([version_id])[0]
1009
def get_texts(self, version_ids):
1010
return [''.join(l) for l in self.get_line_list(version_ids)]
1012
def get_line_list(self, version_ids):
1013
"""Return the texts of listed versions as a list of strings."""
1014
for version_id in version_ids:
1015
self.check_not_reserved_id(version_id)
1016
text_map, content_map = self._get_content_maps(version_ids)
1017
return [text_map[v] for v in version_ids]
1019
_get_lf_split_line_list = get_line_list
1021
def _get_content_maps(self, version_ids):
1022
"""Produce maps of text and KnitContents
1024
:return: (text_map, content_map) where text_map contains the texts for
1025
the requested versions and content_map contains the KnitContents.
1026
Both dicts take version_ids as their keys.
1028
# FUTURE: This function could be improved for the 'extract many' case
1029
# by tracking each component and only doing the copy when the number of
1030
# children than need to apply delta's to it is > 1 or it is part of the
1032
version_ids = list(version_ids)
1033
multiple_versions = len(version_ids) != 1
1034
record_map = self._get_record_map(version_ids)
1039
for version_id in version_ids:
1042
while cursor is not None:
1043
method, data, digest, next = record_map[cursor]
1044
components.append((cursor, method, data, digest))
1045
if cursor in content_map:
1050
for component_id, method, data, digest in reversed(components):
1051
if component_id in content_map:
1052
content = content_map[component_id]
1054
if method == 'fulltext':
1055
assert content is None
1056
content = self.factory.parse_fulltext(data, version_id)
1057
elif method == 'line-delta':
1058
delta = self.factory.parse_line_delta(data, version_id)
1059
if multiple_versions:
1060
# only doing this when we want multiple versions
1061
# output avoids list copies - which reference and
1062
# dereference many strings.
1063
content = content.copy()
1064
content.apply_delta(delta, version_id)
1065
if multiple_versions:
1066
content_map[component_id] = content
1068
if 'no-eol' in self._index.get_options(version_id):
1069
if multiple_versions:
1070
content = content.copy()
1071
content.strip_last_line_newline()
1072
final_content[version_id] = content
1074
# digest here is the digest from the last applied component.
1075
text = content.text()
1076
actual_sha = sha_strings(text)
1077
if actual_sha != digest:
1078
raise KnitCorrupt(self.filename,
1080
'\n of reconstructed text does not match'
1082
'\n for version %s' %
1083
(actual_sha, digest, version_id))
1084
text_map[version_id] = text
1085
return text_map, final_content
1087
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1089
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1090
if version_ids is None:
1091
version_ids = self.versions()
1093
pb = progress.DummyProgress()
1094
# we don't care about inclusions, the caller cares.
1095
# but we need to setup a list of records to visit.
1096
# we need version_id, position, length
1097
version_id_records = []
1098
requested_versions = set(version_ids)
1099
# filter for available versions
1100
for version_id in requested_versions:
1101
if not self.has_version(version_id):
1102
raise RevisionNotPresent(version_id, self.filename)
1103
# get a in-component-order queue:
1104
for version_id in self.versions():
1105
if version_id in requested_versions:
1106
index_memo = self._index.get_position(version_id)
1107
version_id_records.append((version_id, index_memo))
1109
total = len(version_id_records)
1110
for version_idx, (version_id, data, sha_value) in \
1111
enumerate(self._data.read_records_iter(version_id_records)):
1112
pb.update('Walking content.', version_idx, total)
1113
method = self._index.get_method(version_id)
1115
assert method in ('fulltext', 'line-delta')
1116
if method == 'fulltext':
1117
line_iterator = self.factory.get_fulltext_content(data)
1119
line_iterator = self.factory.get_linedelta_content(data)
1120
# XXX: It might be more efficient to yield (version_id,
1121
# line_iterator) in the future. However for now, this is a simpler
1122
# change to integrate into the rest of the codebase. RBC 20071110
1123
for line in line_iterator:
1124
yield line, version_id
1126
pb.update('Walking content.', total, total)
1128
def iter_parents(self, version_ids):
1129
"""Iterate through the parents for many version ids.
1131
:param version_ids: An iterable yielding version_ids.
1132
:return: An iterator that yields (version_id, parents). Requested
1133
version_ids not present in the versioned file are simply skipped.
1134
The order is undefined, allowing for different optimisations in
1135
the underlying implementation.
1137
return self._index.iter_parents(version_ids)
1139
def num_versions(self):
1140
"""See VersionedFile.num_versions()."""
1141
return self._index.num_versions()
1143
__len__ = num_versions
1145
def annotate_iter(self, version_id):
1146
"""See VersionedFile.annotate_iter."""
1147
return self.factory.annotate_iter(self, version_id)
1149
def get_parents(self, version_id):
1150
"""See VersionedFile.get_parents."""
1153
# 52554 calls in 1264 872 internal down from 3674
1155
return self._index.get_parents(version_id)
1157
raise RevisionNotPresent(version_id, self.filename)
1159
def get_parents_with_ghosts(self, version_id):
1160
"""See VersionedFile.get_parents."""
1162
return self._index.get_parents_with_ghosts(version_id)
1164
raise RevisionNotPresent(version_id, self.filename)
1166
def get_ancestry(self, versions, topo_sorted=True):
1167
"""See VersionedFile.get_ancestry."""
1168
if isinstance(versions, basestring):
1169
versions = [versions]
1172
return self._index.get_ancestry(versions, topo_sorted)
1174
def get_ancestry_with_ghosts(self, versions):
1175
"""See VersionedFile.get_ancestry_with_ghosts."""
1176
if isinstance(versions, basestring):
1177
versions = [versions]
1180
return self._index.get_ancestry_with_ghosts(versions)
1182
def plan_merge(self, ver_a, ver_b):
1183
"""See VersionedFile.plan_merge."""
1184
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1185
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1186
annotated_a = self.annotate(ver_a)
1187
annotated_b = self.annotate(ver_b)
1188
return merge._plan_annotate_merge(annotated_a, annotated_b,
1189
ancestors_a, ancestors_b)
1192
class _KnitComponentFile(object):
1193
"""One of the files used to implement a knit database"""
1195
def __init__(self, transport, filename, mode, file_mode=None,
1196
create_parent_dir=False, dir_mode=None):
1197
self._transport = transport
1198
self._filename = filename
1200
self._file_mode = file_mode
1201
self._dir_mode = dir_mode
1202
self._create_parent_dir = create_parent_dir
1203
self._need_to_create = False
1205
def _full_path(self):
1206
"""Return the full path to this file."""
1207
return self._transport.base + self._filename
1209
def check_header(self, fp):
1210
line = fp.readline()
1212
# An empty file can actually be treated as though the file doesn't
1214
raise errors.NoSuchFile(self._full_path())
1215
if line != self.HEADER:
1216
raise KnitHeaderError(badline=line,
1217
filename=self._transport.abspath(self._filename))
1220
return '%s(%s)' % (self.__class__.__name__, self._filename)
1223
class _KnitIndex(_KnitComponentFile):
1224
"""Manages knit index file.
1226
The index is already kept in memory and read on startup, to enable
1227
fast lookups of revision information. The cursor of the index
1228
file is always pointing to the end, making it easy to append
1231
_cache is a cache for fast mapping from version id to a Index
1234
_history is a cache for fast mapping from indexes to version ids.
1236
The index data format is dictionary compressed when it comes to
1237
parent references; a index entry may only have parents that with a
1238
lover index number. As a result, the index is topological sorted.
1240
Duplicate entries may be written to the index for a single version id
1241
if this is done then the latter one completely replaces the former:
1242
this allows updates to correct version and parent information.
1243
Note that the two entries may share the delta, and that successive
1244
annotations and references MUST point to the first entry.
1246
The index file on disc contains a header, followed by one line per knit
1247
record. The same revision can be present in an index file more than once.
1248
The first occurrence gets assigned a sequence number starting from 0.
1250
The format of a single line is
1251
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1252
REVISION_ID is a utf8-encoded revision id
1253
FLAGS is a comma separated list of flags about the record. Values include
1254
no-eol, line-delta, fulltext.
1255
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1256
that the the compressed data starts at.
1257
LENGTH is the ascii representation of the length of the data file.
1258
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1260
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1261
revision id already in the knit that is a parent of REVISION_ID.
1262
The ' :' marker is the end of record marker.
1265
when a write is interrupted to the index file, it will result in a line
1266
that does not end in ' :'. If the ' :' is not present at the end of a line,
1267
or at the end of the file, then the record that is missing it will be
1268
ignored by the parser.
1270
When writing new records to the index file, the data is preceded by '\n'
1271
to ensure that records always start on new lines even if the last write was
1272
interrupted. As a result its normal for the last line in the index to be
1273
missing a trailing newline. One can be added with no harmful effects.
1276
HEADER = "# bzr knit index 8\n"
1278
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1279
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1281
def _cache_version(self, version_id, options, pos, size, parents):
1282
"""Cache a version record in the history array and index cache.
1284
This is inlined into _load_data for performance. KEEP IN SYNC.
1285
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1288
# only want the _history index to reference the 1st index entry
1290
if version_id not in self._cache:
1291
index = len(self._history)
1292
self._history.append(version_id)
1294
index = self._cache[version_id][5]
1295
self._cache[version_id] = (version_id,
1302
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1303
create_parent_dir=False, delay_create=False, dir_mode=None):
1304
_KnitComponentFile.__init__(self, transport, filename, mode,
1305
file_mode=file_mode,
1306
create_parent_dir=create_parent_dir,
1309
# position in _history is the 'official' index for a revision
1310
# but the values may have come from a newer entry.
1311
# so - wc -l of a knit index is != the number of unique names
1315
fp = self._transport.get(self._filename)
1317
# _load_data may raise NoSuchFile if the target knit is
1319
_load_data(self, fp)
1323
if mode != 'w' or not create:
1326
self._need_to_create = True
1328
self._transport.put_bytes_non_atomic(
1329
self._filename, self.HEADER, mode=self._file_mode)
1331
def get_graph(self):
1332
"""Return a list of the node:parents lists from this knit index."""
1333
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1335
def get_ancestry(self, versions, topo_sorted=True):
1336
"""See VersionedFile.get_ancestry."""
1337
# get a graph of all the mentioned versions:
1339
pending = set(versions)
1342
version = pending.pop()
1345
parents = [p for p in cache[version][4] if p in cache]
1347
raise RevisionNotPresent(version, self._filename)
1348
# if not completed and not a ghost
1349
pending.update([p for p in parents if p not in graph])
1350
graph[version] = parents
1353
return topo_sort(graph.items())
1355
def get_ancestry_with_ghosts(self, versions):
1356
"""See VersionedFile.get_ancestry_with_ghosts."""
1357
# get a graph of all the mentioned versions:
1358
self.check_versions_present(versions)
1361
pending = set(versions)
1363
version = pending.pop()
1365
parents = cache[version][4]
1371
pending.update([p for p in parents if p not in graph])
1372
graph[version] = parents
1373
return topo_sort(graph.items())
1375
def iter_parents(self, version_ids):
1376
"""Iterate through the parents for many version ids.
1378
:param version_ids: An iterable yielding version_ids.
1379
:return: An iterator that yields (version_id, parents). Requested
1380
version_ids not present in the versioned file are simply skipped.
1381
The order is undefined, allowing for different optimisations in
1382
the underlying implementation.
1384
for version_id in version_ids:
1386
yield version_id, tuple(self.get_parents(version_id))
1390
def num_versions(self):
1391
return len(self._history)
1393
__len__ = num_versions
1395
def get_versions(self):
1396
"""Get all the versions in the file. not topologically sorted."""
1397
return self._history
1399
def _version_list_to_index(self, versions):
1402
for version in versions:
1403
if version in cache:
1404
# -- inlined lookup() --
1405
result_list.append(str(cache[version][5]))
1406
# -- end lookup () --
1408
result_list.append('.' + version)
1409
return ' '.join(result_list)
1411
def add_version(self, version_id, options, index_memo, parents):
1412
"""Add a version record to the index."""
1413
self.add_versions(((version_id, options, index_memo, parents),))
1415
def add_versions(self, versions, random_id=False):
1416
"""Add multiple versions to the index.
1418
:param versions: a list of tuples:
1419
(version_id, options, pos, size, parents).
1420
:param random_id: If True the ids being added were randomly generated
1421
and no check for existence will be performed.
1424
orig_history = self._history[:]
1425
orig_cache = self._cache.copy()
1428
for version_id, options, (index, pos, size), parents in versions:
1429
line = "\n%s %s %s %s %s :" % (version_id,
1433
self._version_list_to_index(parents))
1434
assert isinstance(line, str), \
1435
'content must be utf-8 encoded: %r' % (line,)
1437
self._cache_version(version_id, options, pos, size, parents)
1438
if not self._need_to_create:
1439
self._transport.append_bytes(self._filename, ''.join(lines))
1442
sio.write(self.HEADER)
1443
sio.writelines(lines)
1445
self._transport.put_file_non_atomic(self._filename, sio,
1446
create_parent_dir=self._create_parent_dir,
1447
mode=self._file_mode,
1448
dir_mode=self._dir_mode)
1449
self._need_to_create = False
1451
# If any problems happen, restore the original values and re-raise
1452
self._history = orig_history
1453
self._cache = orig_cache
1456
def has_version(self, version_id):
1457
"""True if the version is in the index."""
1458
return version_id in self._cache
1460
def get_position(self, version_id):
1461
"""Return details needed to access the version.
1463
.kndx indices do not support split-out data, so return None for the
1466
:return: a tuple (None, data position, size) to hand to the access
1467
logic to get the record.
1469
entry = self._cache[version_id]
1470
return None, entry[2], entry[3]
1472
def get_method(self, version_id):
1473
"""Return compression method of specified version."""
1475
options = self._cache[version_id][1]
1477
raise RevisionNotPresent(version_id, self._filename)
1478
if 'fulltext' in options:
1481
if 'line-delta' not in options:
1482
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1485
def get_options(self, version_id):
1486
"""Return a string represention options.
1490
return self._cache[version_id][1]
1492
def get_parents(self, version_id):
1493
"""Return parents of specified version ignoring ghosts."""
1494
return [parent for parent in self._cache[version_id][4]
1495
if parent in self._cache]
1497
def get_parents_with_ghosts(self, version_id):
1498
"""Return parents of specified version with ghosts."""
1499
return self._cache[version_id][4]
1501
def check_versions_present(self, version_ids):
1502
"""Check that all specified versions are present."""
1504
for version_id in version_ids:
1505
if version_id not in cache:
1506
raise RevisionNotPresent(version_id, self._filename)
1509
class KnitGraphIndex(object):
1510
"""A knit index that builds on GraphIndex."""
1512
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1513
"""Construct a KnitGraphIndex on a graph_index.
1515
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1516
:param deltas: Allow delta-compressed records.
1517
:param add_callback: If not None, allow additions to the index and call
1518
this callback with a list of added GraphIndex nodes:
1519
[(node, value, node_refs), ...]
1520
:param parents: If True, record knits parents, if not do not record
1523
self._graph_index = graph_index
1524
self._deltas = deltas
1525
self._add_callback = add_callback
1526
self._parents = parents
1527
if deltas and not parents:
1528
raise KnitCorrupt(self, "Cannot do delta compression without "
1531
def _get_entries(self, keys, check_present=False):
1532
"""Get the entries for keys.
1534
:param keys: An iterable of index keys, - 1-tuples.
1539
for node in self._graph_index.iter_entries(keys):
1541
found_keys.add(node[1])
1543
# adapt parentless index to the rest of the code.
1544
for node in self._graph_index.iter_entries(keys):
1545
yield node[0], node[1], node[2], ()
1546
found_keys.add(node[1])
1548
missing_keys = keys.difference(found_keys)
1550
raise RevisionNotPresent(missing_keys.pop(), self)
1552
def _present_keys(self, version_ids):
1554
node[1] for node in self._get_entries(version_ids)])
1556
def _parentless_ancestry(self, versions):
1557
"""Honour the get_ancestry API for parentless knit indices."""
1558
wanted_keys = self._version_ids_to_keys(versions)
1559
present_keys = self._present_keys(wanted_keys)
1560
missing = set(wanted_keys).difference(present_keys)
1562
raise RevisionNotPresent(missing.pop(), self)
1563
return list(self._keys_to_version_ids(present_keys))
1565
def get_ancestry(self, versions, topo_sorted=True):
1566
"""See VersionedFile.get_ancestry."""
1567
if not self._parents:
1568
return self._parentless_ancestry(versions)
1569
# XXX: This will do len(history) index calls - perhaps
1570
# it should be altered to be a index core feature?
1571
# get a graph of all the mentioned versions:
1574
versions = self._version_ids_to_keys(versions)
1575
pending = set(versions)
1577
# get all pending nodes
1578
this_iteration = pending
1579
new_nodes = self._get_entries(this_iteration)
1582
for (index, key, value, node_refs) in new_nodes:
1583
# dont ask for ghosties - otherwise
1584
# we we can end up looping with pending
1585
# being entirely ghosted.
1586
graph[key] = [parent for parent in node_refs[0]
1587
if parent not in ghosts]
1589
for parent in graph[key]:
1590
# dont examine known nodes again
1595
ghosts.update(this_iteration.difference(found))
1596
if versions.difference(graph):
1597
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1599
result_keys = topo_sort(graph.items())
1601
result_keys = graph.iterkeys()
1602
return [key[0] for key in result_keys]
1604
def get_ancestry_with_ghosts(self, versions):
1605
"""See VersionedFile.get_ancestry."""
1606
if not self._parents:
1607
return self._parentless_ancestry(versions)
1608
# XXX: This will do len(history) index calls - perhaps
1609
# it should be altered to be a index core feature?
1610
# get a graph of all the mentioned versions:
1612
versions = self._version_ids_to_keys(versions)
1613
pending = set(versions)
1615
# get all pending nodes
1616
this_iteration = pending
1617
new_nodes = self._get_entries(this_iteration)
1619
for (index, key, value, node_refs) in new_nodes:
1620
graph[key] = node_refs[0]
1622
for parent in graph[key]:
1623
# dont examine known nodes again
1627
missing_versions = this_iteration.difference(graph)
1628
missing_needed = versions.intersection(missing_versions)
1630
raise RevisionNotPresent(missing_needed.pop(), self)
1631
for missing_version in missing_versions:
1632
# add a key, no parents
1633
graph[missing_version] = []
1634
pending.discard(missing_version) # don't look for it
1635
result_keys = topo_sort(graph.items())
1636
return [key[0] for key in result_keys]
1638
def get_graph(self):
1639
"""Return a list of the node:parents lists from this knit index."""
1640
if not self._parents:
1641
return [(key, ()) for key in self.get_versions()]
1643
for index, key, value, refs in self._graph_index.iter_all_entries():
1644
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1647
def iter_parents(self, version_ids):
1648
"""Iterate through the parents for many version ids.
1650
:param version_ids: An iterable yielding version_ids.
1651
:return: An iterator that yields (version_id, parents). Requested
1652
version_ids not present in the versioned file are simply skipped.
1653
The order is undefined, allowing for different optimisations in
1654
the underlying implementation.
1657
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1659
present_parents = set()
1660
for node in all_nodes:
1661
all_parents.update(node[3][0])
1662
# any node we are querying must be present
1663
present_parents.add(node[1])
1664
unknown_parents = all_parents.difference(present_parents)
1665
present_parents.update(self._present_keys(unknown_parents))
1666
for node in all_nodes:
1668
for parent in node[3][0]:
1669
if parent in present_parents:
1670
parents.append(parent[0])
1671
yield node[1][0], tuple(parents)
1673
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1674
yield node[1][0], ()
1676
def num_versions(self):
1677
return len(list(self._graph_index.iter_all_entries()))
1679
__len__ = num_versions
1681
def get_versions(self):
1682
"""Get all the versions in the file. not topologically sorted."""
1683
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1685
def has_version(self, version_id):
1686
"""True if the version is in the index."""
1687
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1689
def _keys_to_version_ids(self, keys):
1690
return tuple(key[0] for key in keys)
1692
def get_position(self, version_id):
1693
"""Return details needed to access the version.
1695
:return: a tuple (index, data position, size) to hand to the access
1696
logic to get the record.
1698
node = self._get_node(version_id)
1699
bits = node[2][1:].split(' ')
1700
return node[0], int(bits[0]), int(bits[1])
1702
def get_method(self, version_id):
1703
"""Return compression method of specified version."""
1704
if not self._deltas:
1706
return self._parent_compression(self._get_node(version_id)[3][1])
1708
def _parent_compression(self, reference_list):
1709
# use the second reference list to decide if this is delta'd or not.
1710
if len(reference_list):
1715
def _get_node(self, version_id):
1717
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1719
raise RevisionNotPresent(version_id, self)
1721
def get_options(self, version_id):
1722
"""Return a string represention options.
1726
node = self._get_node(version_id)
1727
if not self._deltas:
1728
options = ['fulltext']
1730
options = [self._parent_compression(node[3][1])]
1731
if node[2][0] == 'N':
1732
options.append('no-eol')
1735
def get_parents(self, version_id):
1736
"""Return parents of specified version ignoring ghosts."""
1737
parents = list(self.iter_parents([version_id]))
1740
raise errors.RevisionNotPresent(version_id, self)
1741
return parents[0][1]
1743
def get_parents_with_ghosts(self, version_id):
1744
"""Return parents of specified version with ghosts."""
1745
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1746
check_present=True))
1747
if not self._parents:
1749
return self._keys_to_version_ids(nodes[0][3][0])
1751
def check_versions_present(self, version_ids):
1752
"""Check that all specified versions are present."""
1753
keys = self._version_ids_to_keys(version_ids)
1754
present = self._present_keys(keys)
1755
missing = keys.difference(present)
1757
raise RevisionNotPresent(missing.pop(), self)
1759
def add_version(self, version_id, options, access_memo, parents):
1760
"""Add a version record to the index."""
1761
return self.add_versions(((version_id, options, access_memo, parents),))
1763
def add_versions(self, versions, random_id=False):
1764
"""Add multiple versions to the index.
1766
This function does not insert data into the Immutable GraphIndex
1767
backing the KnitGraphIndex, instead it prepares data for insertion by
1768
the caller and checks that it is safe to insert then calls
1769
self._add_callback with the prepared GraphIndex nodes.
1771
:param versions: a list of tuples:
1772
(version_id, options, pos, size, parents).
1773
:param random_id: If True the ids being added were randomly generated
1774
and no check for existence will be performed.
1776
if not self._add_callback:
1777
raise errors.ReadOnlyError(self)
1778
# we hope there are no repositories with inconsistent parentage
1783
for (version_id, options, access_memo, parents) in versions:
1784
index, pos, size = access_memo
1785
key = (version_id, )
1786
parents = tuple((parent, ) for parent in parents)
1787
if 'no-eol' in options:
1791
value += "%d %d" % (pos, size)
1792
if not self._deltas:
1793
if 'line-delta' in options:
1794
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1797
if 'line-delta' in options:
1798
node_refs = (parents, (parents[0],))
1800
node_refs = (parents, ())
1802
node_refs = (parents, )
1805
raise KnitCorrupt(self, "attempt to add node with parents "
1806
"in parentless index.")
1808
keys[key] = (value, node_refs)
1810
present_nodes = self._get_entries(keys)
1811
for (index, key, value, node_refs) in present_nodes:
1812
if (value, node_refs) != keys[key]:
1813
raise KnitCorrupt(self, "inconsistent details in add_versions"
1814
": %s %s" % ((value, node_refs), keys[key]))
1818
for key, (value, node_refs) in keys.iteritems():
1819
result.append((key, value, node_refs))
1821
for key, (value, node_refs) in keys.iteritems():
1822
result.append((key, value))
1823
self._add_callback(result)
1825
def _version_ids_to_keys(self, version_ids):
1826
return set((version_id, ) for version_id in version_ids)
1829
class _KnitAccess(object):
1830
"""Access to knit records in a .knit file."""
1832
def __init__(self, transport, filename, _file_mode, _dir_mode,
1833
_need_to_create, _create_parent_dir):
1834
"""Create a _KnitAccess for accessing and inserting data.
1836
:param transport: The transport the .knit is located on.
1837
:param filename: The filename of the .knit.
1839
self._transport = transport
1840
self._filename = filename
1841
self._file_mode = _file_mode
1842
self._dir_mode = _dir_mode
1843
self._need_to_create = _need_to_create
1844
self._create_parent_dir = _create_parent_dir
1846
def add_raw_records(self, sizes, raw_data):
1847
"""Add raw knit bytes to a storage area.
1849
The data is spooled to whereever the access method is storing data.
1851
:param sizes: An iterable containing the size of each raw data segment.
1852
:param raw_data: A bytestring containing the data.
1853
:return: A list of memos to retrieve the record later. Each memo is a
1854
tuple - (index, pos, length), where the index field is always None
1855
for the .knit access method.
1857
assert type(raw_data) == str, \
1858
'data must be plain bytes was %s' % type(raw_data)
1859
if not self._need_to_create:
1860
base = self._transport.append_bytes(self._filename, raw_data)
1862
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1863
create_parent_dir=self._create_parent_dir,
1864
mode=self._file_mode,
1865
dir_mode=self._dir_mode)
1866
self._need_to_create = False
1870
result.append((None, base, size))
1875
"""IFF this data access has its own storage area, initialise it.
1879
self._transport.put_bytes_non_atomic(self._filename, '',
1880
mode=self._file_mode)
1882
def open_file(self):
1883
"""IFF this data access can be represented as a single file, open it.
1885
For knits that are not mapped to a single file on disk this will
1888
:return: None or a file handle.
1891
return self._transport.get(self._filename)
1896
def get_raw_records(self, memos_for_retrieval):
1897
"""Get the raw bytes for a records.
1899
:param memos_for_retrieval: An iterable containing the (index, pos,
1900
length) memo for retrieving the bytes. The .knit method ignores
1901
the index as there is always only a single file.
1902
:return: An iterator over the bytes of the records.
1904
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
1905
for pos, data in self._transport.readv(self._filename, read_vector):
1909
class _PackAccess(object):
1910
"""Access to knit records via a collection of packs."""
1912
def __init__(self, index_to_packs, writer=None):
1913
"""Create a _PackAccess object.
1915
:param index_to_packs: A dict mapping index objects to the transport
1916
and file names for obtaining data.
1917
:param writer: A tuple (pack.ContainerWriter, write_index) which
1918
contains the pack to write, and the index that reads from it will
1922
self.container_writer = writer[0]
1923
self.write_index = writer[1]
1925
self.container_writer = None
1926
self.write_index = None
1927
self.indices = index_to_packs
1929
def add_raw_records(self, sizes, raw_data):
1930
"""Add raw knit bytes to a storage area.
1932
The data is spooled to the container writer in one bytes-record per
1935
:param sizes: An iterable containing the size of each raw data segment.
1936
:param raw_data: A bytestring containing the data.
1937
:return: A list of memos to retrieve the record later. Each memo is a
1938
tuple - (index, pos, length), where the index field is the
1939
write_index object supplied to the PackAccess object.
1941
assert type(raw_data) == str, \
1942
'data must be plain bytes was %s' % type(raw_data)
1946
p_offset, p_length = self.container_writer.add_bytes_record(
1947
raw_data[offset:offset+size], [])
1949
result.append((self.write_index, p_offset, p_length))
1953
"""Pack based knits do not get individually created."""
1955
def get_raw_records(self, memos_for_retrieval):
1956
"""Get the raw bytes for a records.
1958
:param memos_for_retrieval: An iterable containing the (index, pos,
1959
length) memo for retrieving the bytes. The Pack access method
1960
looks up the pack to use for a given record in its index_to_pack
1962
:return: An iterator over the bytes of the records.
1964
# first pass, group into same-index requests
1966
current_index = None
1967
for (index, offset, length) in memos_for_retrieval:
1968
if current_index == index:
1969
current_list.append((offset, length))
1971
if current_index is not None:
1972
request_lists.append((current_index, current_list))
1973
current_index = index
1974
current_list = [(offset, length)]
1975
# handle the last entry
1976
if current_index is not None:
1977
request_lists.append((current_index, current_list))
1978
for index, offsets in request_lists:
1979
transport, path = self.indices[index]
1980
reader = pack.make_readv_reader(transport, path, offsets)
1981
for names, read_func in reader.iter_records():
1982
yield read_func(None)
1984
def open_file(self):
1985
"""Pack based knits have no single file."""
1988
def set_writer(self, writer, index, (transport, packname)):
1989
"""Set a writer to use for adding data."""
1990
if index is not None:
1991
self.indices[index] = (transport, packname)
1992
self.container_writer = writer
1993
self.write_index = index
1996
class _KnitData(object):
1997
"""Manage extraction of data from a KnitAccess, caching and decompressing.
1999
The KnitData class provides the logic for parsing and using knit records,
2000
making use of an access method for the low level read and write operations.
2003
def __init__(self, access):
2004
"""Create a KnitData object.
2006
:param access: The access method to use. Access methods such as
2007
_KnitAccess manage the insertion of raw records and the subsequent
2008
retrieval of the same.
2010
self._access = access
2011
self._checked = False
2012
# TODO: jam 20060713 conceptually, this could spill to disk
2013
# if the cached size gets larger than a certain amount
2014
# but it complicates the model a bit, so for now just use
2015
# a simple dictionary
2017
self._do_cache = False
2019
def enable_cache(self):
2020
"""Enable caching of reads."""
2021
self._do_cache = True
2023
def clear_cache(self):
2024
"""Clear the record cache."""
2025
self._do_cache = False
2028
def _open_file(self):
2029
return self._access.open_file()
2031
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2032
"""Convert version_id, digest, lines into a raw data block.
2034
:param dense_lines: The bytes of lines but in a denser form. For
2035
instance, if lines is a list of 1000 bytestrings each ending in \n,
2036
dense_lines may be a list with one line in it, containing all the
2037
1000's lines and their \n's. Using dense_lines if it is already
2038
known is a win because the string join to create bytes in this
2039
function spends less time resizing the final string.
2040
:return: (len, a StringIO instance with the raw data ready to read.)
2042
# Note: using a string copy here increases memory pressure with e.g.
2043
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2044
# when doing the initial commit of a mozilla tree. RBC 20070921
2045
bytes = ''.join(chain(
2046
["version %s %d %s\n" % (version_id,
2049
dense_lines or lines,
2050
["end %s\n" % version_id]))
2051
assert bytes.__class__ == str
2052
compressed_bytes = bytes_to_gzip(bytes)
2053
return len(compressed_bytes), compressed_bytes
2055
def add_raw_records(self, sizes, raw_data):
2056
"""Append a prepared record to the data file.
2058
:param sizes: An iterable containing the size of each raw data segment.
2059
:param raw_data: A bytestring containing the data.
2060
:return: a list of index data for the way the data was stored.
2061
See the access method add_raw_records documentation for more
2064
return self._access.add_raw_records(sizes, raw_data)
2066
def _parse_record_header(self, version_id, raw_data):
2067
"""Parse a record header for consistency.
2069
:return: the header and the decompressor stream.
2070
as (stream, header_record)
2072
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2074
rec = self._check_header(version_id, df.readline())
2075
except Exception, e:
2076
raise KnitCorrupt(self._access,
2077
"While reading {%s} got %s(%s)"
2078
% (version_id, e.__class__.__name__, str(e)))
2081
def _check_header(self, version_id, line):
2084
raise KnitCorrupt(self._access,
2085
'unexpected number of elements in record header')
2086
if rec[1] != version_id:
2087
raise KnitCorrupt(self._access,
2088
'unexpected version, wanted %r, got %r'
2089
% (version_id, rec[1]))
2092
def _parse_record(self, version_id, data):
2094
# 4168 calls in 2880 217 internal
2095
# 4168 calls to _parse_record_header in 2121
2096
# 4168 calls to readlines in 330
2097
df = GzipFile(mode='rb', fileobj=StringIO(data))
2100
record_contents = df.readlines()
2101
except Exception, e:
2102
raise KnitCorrupt(self._access,
2103
"While reading {%s} got %s(%s)"
2104
% (version_id, e.__class__.__name__, str(e)))
2105
header = record_contents.pop(0)
2106
rec = self._check_header(version_id, header)
2108
last_line = record_contents.pop()
2109
if len(record_contents) != int(rec[2]):
2110
raise KnitCorrupt(self._access,
2111
'incorrect number of lines %s != %s'
2113
% (len(record_contents), int(rec[2]),
2115
if last_line != 'end %s\n' % rec[1]:
2116
raise KnitCorrupt(self._access,
2117
'unexpected version end line %r, wanted %r'
2118
% (last_line, version_id))
2120
return record_contents, rec[3]
2122
def read_records_iter_raw(self, records):
2123
"""Read text records from data file and yield raw data.
2125
This unpacks enough of the text record to validate the id is
2126
as expected but thats all.
2128
# setup an iterator of the external records:
2129
# uses readv so nice and fast we hope.
2131
# grab the disk data needed.
2133
# Don't check _cache if it is empty
2134
needed_offsets = [index_memo for version_id, index_memo
2136
if version_id not in self._cache]
2138
needed_offsets = [index_memo for version_id, index_memo
2141
raw_records = self._access.get_raw_records(needed_offsets)
2143
for version_id, index_memo in records:
2144
if version_id in self._cache:
2145
# This data has already been validated
2146
data = self._cache[version_id]
2148
data = raw_records.next()
2150
self._cache[version_id] = data
2152
# validate the header
2153
df, rec = self._parse_record_header(version_id, data)
2155
yield version_id, data
2157
def read_records_iter(self, records):
2158
"""Read text records from data file and yield result.
2160
The result will be returned in whatever is the fastest to read.
2161
Not by the order requested. Also, multiple requests for the same
2162
record will only yield 1 response.
2163
:param records: A list of (version_id, pos, len) entries
2164
:return: Yields (version_id, contents, digest) in the order
2165
read, not the order requested
2171
# Skip records we have alread seen
2172
yielded_records = set()
2173
needed_records = set()
2174
for record in records:
2175
if record[0] in self._cache:
2176
if record[0] in yielded_records:
2178
yielded_records.add(record[0])
2179
data = self._cache[record[0]]
2180
content, digest = self._parse_record(record[0], data)
2181
yield (record[0], content, digest)
2183
needed_records.add(record)
2184
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2186
needed_records = sorted(set(records), key=operator.itemgetter(1))
2188
if not needed_records:
2191
# The transport optimizes the fetching as well
2192
# (ie, reads continuous ranges.)
2193
raw_data = self._access.get_raw_records(
2194
[index_memo for version_id, index_memo in needed_records])
2196
for (version_id, index_memo), data in \
2197
izip(iter(needed_records), raw_data):
2198
content, digest = self._parse_record(version_id, data)
2200
self._cache[version_id] = data
2201
yield version_id, content, digest
2203
def read_records(self, records):
2204
"""Read records into a dictionary."""
2206
for record_id, content, digest in \
2207
self.read_records_iter(records):
2208
components[record_id] = (content, digest)
2212
class InterKnit(InterVersionedFile):
2213
"""Optimised code paths for knit to knit operations."""
2215
_matching_file_from_factory = KnitVersionedFile
2216
_matching_file_to_factory = KnitVersionedFile
2219
def is_compatible(source, target):
2220
"""Be compatible with knits. """
2222
return (isinstance(source, KnitVersionedFile) and
2223
isinstance(target, KnitVersionedFile))
2224
except AttributeError:
2227
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2228
"""See InterVersionedFile.join."""
2229
assert isinstance(self.source, KnitVersionedFile)
2230
assert isinstance(self.target, KnitVersionedFile)
2232
# If the source and target are mismatched w.r.t. annotations vs
2233
# plain, the data needs to be converted accordingly
2234
if self.source.factory.annotated == self.target.factory.annotated:
2236
elif self.source.factory.annotated:
2237
converter = self._anno_to_plain_converter
2239
# We're converting from a plain to an annotated knit. This requires
2240
# building the annotations from scratch. The generic join code
2241
# handles this implicitly so we delegate to it.
2242
return super(InterKnit, self).join(pb, msg, version_ids,
2245
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2249
pb = ui.ui_factory.nested_progress_bar()
2251
version_ids = list(version_ids)
2252
if None in version_ids:
2253
version_ids.remove(None)
2255
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2256
this_versions = set(self.target._index.get_versions())
2257
# XXX: For efficiency we should not look at the whole index,
2258
# we only need to consider the referenced revisions - they
2259
# must all be present, or the method must be full-text.
2260
# TODO, RBC 20070919
2261
needed_versions = self.source_ancestry - this_versions
2263
if not needed_versions:
2265
full_list = topo_sort(self.source.get_graph())
2267
version_list = [i for i in full_list if (not self.target.has_version(i)
2268
and i in needed_versions)]
2272
copy_queue_records = []
2274
for version_id in version_list:
2275
options = self.source._index.get_options(version_id)
2276
parents = self.source._index.get_parents_with_ghosts(version_id)
2277
# check that its will be a consistent copy:
2278
for parent in parents:
2279
# if source has the parent, we must :
2280
# * already have it or
2281
# * have it scheduled already
2282
# otherwise we don't care
2283
assert (self.target.has_version(parent) or
2284
parent in copy_set or
2285
not self.source.has_version(parent))
2286
index_memo = self.source._index.get_position(version_id)
2287
copy_queue_records.append((version_id, index_memo))
2288
copy_queue.append((version_id, options, parents))
2289
copy_set.add(version_id)
2291
# data suck the join:
2293
total = len(version_list)
2296
for (version_id, raw_data), \
2297
(version_id2, options, parents) in \
2298
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2300
assert version_id == version_id2, 'logic error, inconsistent results'
2302
pb.update("Joining knit", count, total)
2304
size, raw_data = converter(raw_data, version_id, options,
2307
size = len(raw_data)
2308
raw_records.append((version_id, options, parents, size))
2309
raw_datum.append(raw_data)
2310
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2315
def _anno_to_plain_converter(self, raw_data, version_id, options,
2317
"""Convert annotated content to plain content."""
2318
data, digest = self.source._data._parse_record(version_id, raw_data)
2319
if 'fulltext' in options:
2320
content = self.source.factory.parse_fulltext(data, version_id)
2321
lines = self.target.factory.lower_fulltext(content)
2323
delta = self.source.factory.parse_line_delta(data, version_id,
2325
lines = self.target.factory.lower_line_delta(delta)
2326
return self.target._data._record_to_data(version_id, digest, lines)
2329
InterVersionedFile.register_optimiser(InterKnit)
2332
class WeaveToKnit(InterVersionedFile):
2333
"""Optimised code paths for weave to knit operations."""
2335
_matching_file_from_factory = bzrlib.weave.WeaveFile
2336
_matching_file_to_factory = KnitVersionedFile
2339
def is_compatible(source, target):
2340
"""Be compatible with weaves to knits."""
2342
return (isinstance(source, bzrlib.weave.Weave) and
2343
isinstance(target, KnitVersionedFile))
2344
except AttributeError:
2347
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2348
"""See InterVersionedFile.join."""
2349
assert isinstance(self.source, bzrlib.weave.Weave)
2350
assert isinstance(self.target, KnitVersionedFile)
2352
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2357
pb = ui.ui_factory.nested_progress_bar()
2359
version_ids = list(version_ids)
2361
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2362
this_versions = set(self.target._index.get_versions())
2363
needed_versions = self.source_ancestry - this_versions
2365
if not needed_versions:
2367
full_list = topo_sort(self.source.get_graph())
2369
version_list = [i for i in full_list if (not self.target.has_version(i)
2370
and i in needed_versions)]
2374
total = len(version_list)
2375
for version_id in version_list:
2376
pb.update("Converting to knit", count, total)
2377
parents = self.source.get_parents(version_id)
2378
# check that its will be a consistent copy:
2379
for parent in parents:
2380
# if source has the parent, we must already have it
2381
assert (self.target.has_version(parent))
2382
self.target.add_lines(
2383
version_id, parents, self.source.get_lines(version_id))
2390
InterVersionedFile.register_optimiser(WeaveToKnit)
2393
# Deprecated, use PatienceSequenceMatcher instead
2394
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2397
def annotate_knit(knit, revision_id):
2398
"""Annotate a knit with no cached annotations.
2400
This implementation is for knits with no cached annotations.
2401
It will work for knits with cached annotations, but this is not
2404
ancestry = knit.get_ancestry(revision_id)
2405
fulltext = dict(zip(ancestry, knit.get_line_list(ancestry)))
2407
for candidate in ancestry:
2408
if candidate in annotations:
2410
parents = knit.get_parents(candidate)
2411
if len(parents) == 0:
2413
elif knit._index.get_method(candidate) != 'line-delta':
2416
parent, sha1, noeol, delta = knit.get_delta(candidate)
2417
blocks = KnitContent.get_line_delta_blocks(delta,
2418
fulltext[parents[0]], fulltext[candidate])
2419
annotations[candidate] = list(annotate.reannotate([annotations[p]
2420
for p in parents], fulltext[candidate], candidate, blocks))
2421
return iter(annotations[revision_id])
2425
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2427
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
