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, plain=False):
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)])
267
:param plain: If True, the lines are returned as a plain
268
list, not as a list of tuples, i.e.
269
(start, end, count, [1..count newline])
276
def cache_and_return(line):
277
origin, text = line.split(' ', 1)
278
return cache.setdefault(origin, origin), text
280
# walk through the lines parsing.
281
# Note that the plain test is explicitly pulled out of the
282
# loop to minimise any performance impact
285
start, end, count = [int(n) for n in header.split(',')]
286
contents = [next().split(' ', 1)[1] for i in xrange(count)]
287
result.append((start, end, count, contents))
290
start, end, count = [int(n) for n in header.split(',')]
291
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
292
result.append((start, end, count, contents))
295
def get_fulltext_content(self, lines):
296
"""Extract just the content lines from a fulltext."""
297
return (line.split(' ', 1)[1] for line in lines)
299
def get_linedelta_content(self, lines):
300
"""Extract just the content from a line delta.
302
This doesn't return all of the extra information stored in a delta.
303
Only the actual content lines.
308
header = header.split(',')
309
count = int(header[2])
310
for i in xrange(count):
311
origin, text = next().split(' ', 1)
314
def lower_fulltext(self, content):
315
"""convert a fulltext content record into a serializable form.
317
see parse_fulltext which this inverts.
319
# TODO: jam 20070209 We only do the caching thing to make sure that
320
# the origin is a valid utf-8 line, eventually we could remove it
321
return ['%s %s' % (o, t) for o, t in content._lines]
323
def lower_line_delta(self, delta):
324
"""convert a delta into a serializable form.
326
See parse_line_delta which this inverts.
328
# TODO: jam 20070209 We only do the caching thing to make sure that
329
# the origin is a valid utf-8 line, eventually we could remove it
331
for start, end, c, lines in delta:
332
out.append('%d,%d,%d\n' % (start, end, c))
333
out.extend(origin + ' ' + text
334
for origin, text in lines)
337
def annotate_iter(self, knit, version_id):
338
content = knit._get_content(version_id)
339
return content.annotate_iter()
342
class KnitPlainFactory(object):
343
"""Factory for creating plain Content objects."""
347
def make(self, lines, version_id):
348
return PlainKnitContent(lines, version_id)
350
def parse_fulltext(self, content, version_id):
351
"""This parses an unannotated fulltext.
353
Note that this is not a noop - the internal representation
354
has (versionid, line) - its just a constant versionid.
356
return self.make(content, version_id)
358
def parse_line_delta_iter(self, lines, version_id):
360
num_lines = len(lines)
361
while cur < num_lines:
364
start, end, c = [int(n) for n in header.split(',')]
365
yield start, end, c, lines[cur:cur+c]
368
def parse_line_delta(self, lines, version_id):
369
return list(self.parse_line_delta_iter(lines, version_id))
371
def get_fulltext_content(self, lines):
372
"""Extract just the content lines from a fulltext."""
375
def get_linedelta_content(self, lines):
376
"""Extract just the content from a line delta.
378
This doesn't return all of the extra information stored in a delta.
379
Only the actual content lines.
384
header = header.split(',')
385
count = int(header[2])
386
for i in xrange(count):
389
def lower_fulltext(self, content):
390
return content.text()
392
def lower_line_delta(self, delta):
394
for start, end, c, lines in delta:
395
out.append('%d,%d,%d\n' % (start, end, c))
399
def annotate_iter(self, knit, version_id):
400
return annotate_knit(knit, version_id)
403
def make_empty_knit(transport, relpath):
404
"""Construct a empty knit at the specified location."""
405
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
408
class KnitVersionedFile(VersionedFile):
409
"""Weave-like structure with faster random access.
411
A knit stores a number of texts and a summary of the relationships
412
between them. Texts are identified by a string version-id. Texts
413
are normally stored and retrieved as a series of lines, but can
414
also be passed as single strings.
416
Lines are stored with the trailing newline (if any) included, to
417
avoid special cases for files with no final newline. Lines are
418
composed of 8-bit characters, not unicode. The combination of
419
these approaches should mean any 'binary' file can be safely
420
stored and retrieved.
423
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
424
factory=None, delta=True, create=False, create_parent_dir=False,
425
delay_create=False, dir_mode=None, index=None, access_method=None):
426
"""Construct a knit at location specified by relpath.
428
:param create: If not True, only open an existing knit.
429
:param create_parent_dir: If True, create the parent directory if
430
creating the file fails. (This is used for stores with
431
hash-prefixes that may not exist yet)
432
:param delay_create: The calling code is aware that the knit won't
433
actually be created until the first data is stored.
434
:param index: An index to use for the knit.
436
if access_mode is None:
438
super(KnitVersionedFile, self).__init__(access_mode)
439
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
440
self.transport = transport
441
self.filename = relpath
442
self.factory = factory or KnitAnnotateFactory()
443
self.writable = (access_mode == 'w')
446
self._max_delta_chain = 200
449
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
450
access_mode, create=create, file_mode=file_mode,
451
create_parent_dir=create_parent_dir, delay_create=delay_create,
455
if access_method is None:
456
_access = _KnitAccess(transport, relpath + DATA_SUFFIX, file_mode, dir_mode,
457
((create and not len(self)) and delay_create), create_parent_dir)
459
_access = access_method
460
if create and not len(self) and not delay_create:
462
self._data = _KnitData(_access)
465
return '%s(%s)' % (self.__class__.__name__,
466
self.transport.abspath(self.filename))
468
def _check_should_delta(self, first_parents):
469
"""Iterate back through the parent listing, looking for a fulltext.
471
This is used when we want to decide whether to add a delta or a new
472
fulltext. It searches for _max_delta_chain parents. When it finds a
473
fulltext parent, it sees if the total size of the deltas leading up to
474
it is large enough to indicate that we want a new full text anyway.
476
Return True if we should create a new delta, False if we should use a
481
delta_parents = first_parents
482
for count in xrange(self._max_delta_chain):
483
parent = delta_parents[0]
484
method = self._index.get_method(parent)
485
index, pos, size = self._index.get_position(parent)
486
if method == 'fulltext':
490
delta_parents = self._index.get_parents(parent)
492
# We couldn't find a fulltext, so we must create a new one
495
return fulltext_size > delta_size
497
def _add_raw_records(self, records, data):
498
"""Add all the records 'records' with data pre-joined in 'data'.
500
:param records: A list of tuples(version_id, options, parents, size).
501
:param data: The data for the records. When it is written, the records
502
are adjusted to have pos pointing into data by the sum of
503
the preceding records sizes.
506
raw_record_sizes = [record[3] for record in records]
507
positions = self._data.add_raw_records(raw_record_sizes, data)
510
for (version_id, options, parents, size), access_memo in zip(
512
index_entries.append((version_id, options, access_memo, parents))
513
if self._data._do_cache:
514
self._data._cache[version_id] = data[offset:offset+size]
516
self._index.add_versions(index_entries)
518
def enable_cache(self):
519
"""Start caching data for this knit"""
520
self._data.enable_cache()
522
def clear_cache(self):
523
"""Clear the data cache only."""
524
self._data.clear_cache()
526
def copy_to(self, name, transport):
527
"""See VersionedFile.copy_to()."""
528
# copy the current index to a temp index to avoid racing with local
530
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
531
self.transport.get(self._index._filename))
533
f = self._data._open_file()
535
transport.put_file(name + DATA_SUFFIX, f)
538
# move the copied index into place
539
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
541
def create_empty(self, name, transport, mode=None):
542
return KnitVersionedFile(name, transport, factory=self.factory,
543
delta=self.delta, create=True)
545
def get_data_stream(self, required_versions):
546
"""Get a data stream for the specified versions.
548
Versions may be returned in any order, not necessarily the order
551
:param required_versions: The exact set of versions to be extracted.
552
Unlike some other knit methods, this is not used to generate a
553
transitive closure, rather it is used precisely as given.
555
:returns: format_signature, list of (version, options, length, parents),
558
required_versions = set([osutils.safe_revision_id(v) for v in
560
# we don't care about inclusions, the caller cares.
561
# but we need to setup a list of records to visit.
562
for version_id in required_versions:
563
if not self.has_version(version_id):
564
raise RevisionNotPresent(version_id, self.filename)
565
# Pick the desired versions out of the index in oldest-to-newest order
567
for version_id in self.versions():
568
if version_id in required_versions:
569
version_list.append(version_id)
571
# create the list of version information for the result
572
copy_queue_records = []
574
result_version_list = []
575
for version_id in version_list:
576
options = self._index.get_options(version_id)
577
parents = self._index.get_parents_with_ghosts(version_id)
578
index_memo = self._index.get_position(version_id)
579
copy_queue_records.append((version_id, index_memo))
580
none, data_pos, data_size = index_memo
581
copy_set.add(version_id)
582
# version, options, length, parents
583
result_version_list.append((version_id, options, data_size,
586
# Read the compressed record data.
588
# From here down to the return should really be logic in the returned
589
# callable -- in a class that adapts read_records_iter_raw to read
592
for (version_id, raw_data), \
593
(version_id2, options, _, parents) in \
594
izip(self._data.read_records_iter_raw(copy_queue_records),
595
result_version_list):
596
assert version_id == version_id2, 'logic error, inconsistent results'
597
raw_datum.append(raw_data)
598
pseudo_file = StringIO(''.join(raw_datum))
601
return pseudo_file.read()
603
return pseudo_file.read(length)
604
return (self.get_format_signature(), result_version_list, read)
606
def _extract_blocks(self, version_id, source, target):
607
if self._index.get_method(version_id) != 'line-delta':
609
parent, sha1, noeol, delta = self.get_delta(version_id)
610
return KnitContent.get_line_delta_blocks(delta, source, target)
612
def get_delta(self, version_id):
613
"""Get a delta for constructing version from some other version."""
614
version_id = osutils.safe_revision_id(version_id)
615
self.check_not_reserved_id(version_id)
616
parents = self.get_parents(version_id)
621
index_memo = self._index.get_position(version_id)
622
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
623
noeol = 'no-eol' in self._index.get_options(version_id)
624
if 'fulltext' == self._index.get_method(version_id):
625
new_content = self.factory.parse_fulltext(data, version_id)
626
if parent is not None:
627
reference_content = self._get_content(parent)
628
old_texts = reference_content.text()
631
new_texts = new_content.text()
632
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
634
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
636
delta = self.factory.parse_line_delta(data, version_id)
637
return parent, sha1, noeol, delta
639
def get_format_signature(self):
640
"""See VersionedFile.get_format_signature()."""
641
if self.factory.annotated:
642
annotated_part = "annotated"
644
annotated_part = "plain"
645
return "knit-%s" % (annotated_part,)
647
def get_graph_with_ghosts(self):
648
"""See VersionedFile.get_graph_with_ghosts()."""
649
graph_items = self._index.get_graph()
650
return dict(graph_items)
652
def get_sha1(self, version_id):
653
return self.get_sha1s([version_id])[0]
655
def get_sha1s(self, version_ids):
656
"""See VersionedFile.get_sha1()."""
657
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
658
record_map = self._get_record_map(version_ids)
659
# record entry 2 is the 'digest'.
660
return [record_map[v][2] for v in version_ids]
664
"""See VersionedFile.get_suffixes()."""
665
return [DATA_SUFFIX, INDEX_SUFFIX]
667
def has_ghost(self, version_id):
668
"""True if there is a ghost reference in the file to version_id."""
669
version_id = osutils.safe_revision_id(version_id)
671
if self.has_version(version_id):
673
# optimisable if needed by memoising the _ghosts set.
674
items = self._index.get_graph()
675
for node, parents in items:
676
for parent in parents:
677
if parent not in self._index._cache:
678
if parent == version_id:
682
def insert_data_stream(self, (format, data_list, reader_callable)):
683
"""Insert knit records from a data stream into this knit.
685
If a version in the stream is already present in this knit, it will not
686
be inserted a second time. It will be checked for consistency with the
687
stored version however, and may cause a KnitCorrupt error to be raised
688
if the data in the stream disagrees with the already stored data.
690
:seealso: get_data_stream
692
if format != self.get_format_signature():
693
trace.mutter('incompatible format signature inserting to %r', self)
694
raise KnitDataStreamIncompatible(
695
format, self.get_format_signature())
697
for version_id, options, length, parents in data_list:
698
if self.has_version(version_id):
699
# First check: the list of parents.
700
my_parents = self.get_parents_with_ghosts(version_id)
701
if my_parents != parents:
702
# XXX: KnitCorrupt is not quite the right exception here.
705
'parents list %r from data stream does not match '
706
'already recorded parents %r for %s'
707
% (parents, my_parents, version_id))
709
# Also check the SHA-1 of the fulltext this content will
711
raw_data = reader_callable(length)
712
my_fulltext_sha1 = self.get_sha1(version_id)
713
df, rec = self._data._parse_record_header(version_id, raw_data)
714
stream_fulltext_sha1 = rec[3]
715
if my_fulltext_sha1 != stream_fulltext_sha1:
716
# Actually, we don't know if it's this knit that's corrupt,
717
# or the data stream we're trying to insert.
719
self.filename, 'sha-1 does not match %s' % version_id)
721
self._add_raw_records(
722
[(version_id, options, parents, length)],
723
reader_callable(length))
726
"""See VersionedFile.versions."""
727
if 'evil' in debug.debug_flags:
728
trace.mutter_callsite(2, "versions scales with size of history")
729
return self._index.get_versions()
731
def has_version(self, version_id):
732
"""See VersionedFile.has_version."""
733
if 'evil' in debug.debug_flags:
734
trace.mutter_callsite(2, "has_version is a LBYL scenario")
735
version_id = osutils.safe_revision_id(version_id)
736
return self._index.has_version(version_id)
738
__contains__ = has_version
740
def _merge_annotations(self, content, parents, parent_texts={},
741
delta=None, annotated=None,
742
left_matching_blocks=None):
743
"""Merge annotations for content. This is done by comparing
744
the annotations based on changed to the text.
746
if left_matching_blocks is not None:
747
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
751
for parent_id in parents:
752
merge_content = self._get_content(parent_id, parent_texts)
753
if (parent_id == parents[0] and delta_seq is not None):
756
seq = patiencediff.PatienceSequenceMatcher(
757
None, merge_content.text(), content.text())
758
for i, j, n in seq.get_matching_blocks():
761
# this appears to copy (origin, text) pairs across to the
762
# new content for any line that matches the last-checked
764
content._lines[j:j+n] = merge_content._lines[i:i+n]
766
if delta_seq is None:
767
reference_content = self._get_content(parents[0], parent_texts)
768
new_texts = content.text()
769
old_texts = reference_content.text()
770
delta_seq = patiencediff.PatienceSequenceMatcher(
771
None, old_texts, new_texts)
772
return self._make_line_delta(delta_seq, content)
774
def _make_line_delta(self, delta_seq, new_content):
775
"""Generate a line delta from delta_seq and new_content."""
777
for op in delta_seq.get_opcodes():
780
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
783
def _get_components_positions(self, version_ids):
784
"""Produce a map of position data for the components of versions.
786
This data is intended to be used for retrieving the knit records.
788
A dict of version_id to (method, data_pos, data_size, next) is
790
method is the way referenced data should be applied.
791
data_pos is the position of the data in the knit.
792
data_size is the size of the data in the knit.
793
next is the build-parent of the version, or None for fulltexts.
796
for version_id in version_ids:
799
while cursor is not None and cursor not in component_data:
800
method = self._index.get_method(cursor)
801
if method == 'fulltext':
804
next = self.get_parents(cursor)[0]
805
index_memo = self._index.get_position(cursor)
806
component_data[cursor] = (method, index_memo, next)
808
return component_data
810
def _get_content(self, version_id, parent_texts={}):
811
"""Returns a content object that makes up the specified
813
cached_version = parent_texts.get(version_id, None)
814
if cached_version is not None:
815
if not self.has_version(version_id):
816
raise RevisionNotPresent(version_id, self.filename)
817
return cached_version
819
text_map, contents_map = self._get_content_maps([version_id])
820
return contents_map[version_id]
822
def _check_versions_present(self, version_ids):
823
"""Check that all specified versions are present."""
824
self._index.check_versions_present(version_ids)
826
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
827
nostore_sha, random_id, check_content):
828
"""See VersionedFile.add_lines_with_ghosts()."""
829
self._check_add(version_id, lines, random_id, check_content)
830
return self._add(version_id, lines, parents, self.delta,
831
parent_texts, None, nostore_sha, random_id)
833
def _add_lines(self, version_id, parents, lines, parent_texts,
834
left_matching_blocks, nostore_sha, random_id, check_content):
835
"""See VersionedFile.add_lines."""
836
self._check_add(version_id, lines, random_id, check_content)
837
self._check_versions_present(parents)
838
return self._add(version_id, lines[:], parents, self.delta,
839
parent_texts, left_matching_blocks, nostore_sha, random_id)
841
def _check_add(self, version_id, lines, random_id, check_content):
842
"""check that version_id and lines are safe to add."""
843
if contains_whitespace(version_id):
844
raise InvalidRevisionId(version_id, self.filename)
845
self.check_not_reserved_id(version_id)
846
# Technically this could be avoided if we are happy to allow duplicate
847
# id insertion when other things than bzr core insert texts, but it
848
# seems useful for folk using the knit api directly to have some safety
849
# blanket that we can disable.
850
if not random_id and self.has_version(version_id):
851
raise RevisionAlreadyPresent(version_id, self.filename)
853
self._check_lines_not_unicode(lines)
854
self._check_lines_are_lines(lines)
856
def _add(self, version_id, lines, parents, delta, parent_texts,
857
left_matching_blocks, nostore_sha, random_id):
858
"""Add a set of lines on top of version specified by parents.
860
If delta is true, compress the text as a line-delta against
863
Any versions not present will be converted into ghosts.
865
# first thing, if the content is something we don't need to store, find
867
line_bytes = ''.join(lines)
868
digest = sha_string(line_bytes)
869
if nostore_sha == digest:
870
raise errors.ExistingContent
873
if parent_texts is None:
875
for parent in parents:
876
if self.has_version(parent):
877
present_parents.append(parent)
879
# can only compress against the left most present parent.
881
(len(present_parents) == 0 or
882
present_parents[0] != parents[0])):
885
text_length = len(line_bytes)
888
if lines[-1][-1] != '\n':
889
# copy the contents of lines.
891
options.append('no-eol')
892
lines[-1] = lines[-1] + '\n'
895
# To speed the extract of texts the delta chain is limited
896
# to a fixed number of deltas. This should minimize both
897
# I/O and the time spend applying deltas.
898
delta = self._check_should_delta(present_parents)
900
assert isinstance(version_id, str)
901
content = self.factory.make(lines, version_id)
902
if delta or (self.factory.annotated and len(present_parents) > 0):
903
# Merge annotations from parent texts if needed.
904
delta_hunks = self._merge_annotations(content, present_parents,
905
parent_texts, delta, self.factory.annotated,
906
left_matching_blocks)
909
options.append('line-delta')
910
store_lines = self.factory.lower_line_delta(delta_hunks)
911
size, bytes = self._data._record_to_data(version_id, digest,
914
options.append('fulltext')
915
# get mixed annotation + content and feed it into the
917
store_lines = self.factory.lower_fulltext(content)
918
size, bytes = self._data._record_to_data(version_id, digest,
921
access_memo = self._data.add_raw_records([size], bytes)[0]
922
self._index.add_versions(
923
((version_id, options, access_memo, parents),),
925
return digest, text_length, content
927
def check(self, progress_bar=None):
928
"""See VersionedFile.check()."""
930
def _clone_text(self, new_version_id, old_version_id, parents):
931
"""See VersionedFile.clone_text()."""
932
# FIXME RBC 20060228 make fast by only inserting an index with null
934
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
936
def get_lines(self, version_id):
937
"""See VersionedFile.get_lines()."""
938
return self.get_line_list([version_id])[0]
940
def _get_record_map(self, version_ids):
941
"""Produce a dictionary of knit records.
943
The keys are version_ids, the values are tuples of (method, content,
945
method is the way the content should be applied.
946
content is a KnitContent object.
947
digest is the SHA1 digest of this version id after all steps are done
948
next is the build-parent of the version, i.e. the leftmost ancestor.
949
If the method is fulltext, next will be None.
951
position_map = self._get_components_positions(version_ids)
952
# c = component_id, m = method, i_m = index_memo, n = next
953
records = [(c, i_m) for c, (m, i_m, n) in position_map.iteritems()]
955
for component_id, content, digest in \
956
self._data.read_records_iter(records):
957
method, index_memo, next = position_map[component_id]
958
record_map[component_id] = method, content, digest, next
962
def get_text(self, version_id):
963
"""See VersionedFile.get_text"""
964
return self.get_texts([version_id])[0]
966
def get_texts(self, version_ids):
967
return [''.join(l) for l in self.get_line_list(version_ids)]
969
def get_line_list(self, version_ids):
970
"""Return the texts of listed versions as a list of strings."""
971
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
972
for version_id in version_ids:
973
self.check_not_reserved_id(version_id)
974
text_map, content_map = self._get_content_maps(version_ids)
975
return [text_map[v] for v in version_ids]
977
_get_lf_split_line_list = get_line_list
979
def _get_content_maps(self, version_ids):
980
"""Produce maps of text and KnitContents
982
:return: (text_map, content_map) where text_map contains the texts for
983
the requested versions and content_map contains the KnitContents.
984
Both dicts take version_ids as their keys.
986
record_map = self._get_record_map(version_ids)
991
for version_id in version_ids:
994
while cursor is not None:
995
method, data, digest, next = record_map[cursor]
996
components.append((cursor, method, data, digest))
997
if cursor in content_map:
1002
for component_id, method, data, digest in reversed(components):
1003
if component_id in content_map:
1004
content = content_map[component_id]
1006
if method == 'fulltext':
1007
assert content is None
1008
content = self.factory.parse_fulltext(data, version_id)
1009
elif method == 'line-delta':
1010
delta = self.factory.parse_line_delta(data, version_id)
1011
content = content.copy()
1012
content._lines = self._apply_delta(content._lines,
1014
content_map[component_id] = content
1016
if 'no-eol' in self._index.get_options(version_id):
1017
content = content.copy()
1018
content.strip_last_line_newline()
1019
final_content[version_id] = content
1021
# digest here is the digest from the last applied component.
1022
text = content.text()
1023
if sha_strings(text) != digest:
1024
raise KnitCorrupt(self.filename,
1025
'sha-1 does not match %s' % version_id)
1027
text_map[version_id] = text
1028
return text_map, final_content
1031
def _apply_delta(lines, delta):
1032
"""Apply delta to lines."""
1035
for start, end, count, delta_lines in delta:
1036
lines[offset+start:offset+end] = delta_lines
1037
offset = offset + (start - end) + count
1040
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1042
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1043
if version_ids is None:
1044
version_ids = self.versions()
1046
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
1048
pb = progress.DummyProgress()
1049
# we don't care about inclusions, the caller cares.
1050
# but we need to setup a list of records to visit.
1051
# we need version_id, position, length
1052
version_id_records = []
1053
requested_versions = set(version_ids)
1054
# filter for available versions
1055
for version_id in requested_versions:
1056
if not self.has_version(version_id):
1057
raise RevisionNotPresent(version_id, self.filename)
1058
# get a in-component-order queue:
1059
for version_id in self.versions():
1060
if version_id in requested_versions:
1061
index_memo = self._index.get_position(version_id)
1062
version_id_records.append((version_id, index_memo))
1064
total = len(version_id_records)
1065
for version_idx, (version_id, data, sha_value) in \
1066
enumerate(self._data.read_records_iter(version_id_records)):
1067
pb.update('Walking content.', version_idx, total)
1068
method = self._index.get_method(version_id)
1070
assert method in ('fulltext', 'line-delta')
1071
if method == 'fulltext':
1072
line_iterator = self.factory.get_fulltext_content(data)
1074
line_iterator = self.factory.get_linedelta_content(data)
1075
for line in line_iterator:
1078
pb.update('Walking content.', total, total)
1080
def iter_parents(self, version_ids):
1081
"""Iterate through the parents for many version ids.
1083
:param version_ids: An iterable yielding version_ids.
1084
:return: An iterator that yields (version_id, parents). Requested
1085
version_ids not present in the versioned file are simply skipped.
1086
The order is undefined, allowing for different optimisations in
1087
the underlying implementation.
1089
version_ids = [osutils.safe_revision_id(version_id) for
1090
version_id in version_ids]
1091
return self._index.iter_parents(version_ids)
1093
def num_versions(self):
1094
"""See VersionedFile.num_versions()."""
1095
return self._index.num_versions()
1097
__len__ = num_versions
1099
def annotate_iter(self, version_id):
1100
"""See VersionedFile.annotate_iter."""
1101
version_id = osutils.safe_revision_id(version_id)
1102
return self.factory.annotate_iter(self, version_id)
1104
def get_parents(self, version_id):
1105
"""See VersionedFile.get_parents."""
1108
# 52554 calls in 1264 872 internal down from 3674
1109
version_id = osutils.safe_revision_id(version_id)
1111
return self._index.get_parents(version_id)
1113
raise RevisionNotPresent(version_id, self.filename)
1115
def get_parents_with_ghosts(self, version_id):
1116
"""See VersionedFile.get_parents."""
1117
version_id = osutils.safe_revision_id(version_id)
1119
return self._index.get_parents_with_ghosts(version_id)
1121
raise RevisionNotPresent(version_id, self.filename)
1123
def get_ancestry(self, versions, topo_sorted=True):
1124
"""See VersionedFile.get_ancestry."""
1125
if isinstance(versions, basestring):
1126
versions = [versions]
1129
versions = [osutils.safe_revision_id(v) for v in versions]
1130
return self._index.get_ancestry(versions, topo_sorted)
1132
def get_ancestry_with_ghosts(self, versions):
1133
"""See VersionedFile.get_ancestry_with_ghosts."""
1134
if isinstance(versions, basestring):
1135
versions = [versions]
1138
versions = [osutils.safe_revision_id(v) for v in versions]
1139
return self._index.get_ancestry_with_ghosts(versions)
1141
def plan_merge(self, ver_a, ver_b):
1142
"""See VersionedFile.plan_merge."""
1143
ver_a = osutils.safe_revision_id(ver_a)
1144
ver_b = osutils.safe_revision_id(ver_b)
1145
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1147
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1148
annotated_a = self.annotate(ver_a)
1149
annotated_b = self.annotate(ver_b)
1150
return merge._plan_annotate_merge(annotated_a, annotated_b,
1151
ancestors_a, ancestors_b)
1154
class _KnitComponentFile(object):
1155
"""One of the files used to implement a knit database"""
1157
def __init__(self, transport, filename, mode, file_mode=None,
1158
create_parent_dir=False, dir_mode=None):
1159
self._transport = transport
1160
self._filename = filename
1162
self._file_mode = file_mode
1163
self._dir_mode = dir_mode
1164
self._create_parent_dir = create_parent_dir
1165
self._need_to_create = False
1167
def _full_path(self):
1168
"""Return the full path to this file."""
1169
return self._transport.base + self._filename
1171
def check_header(self, fp):
1172
line = fp.readline()
1174
# An empty file can actually be treated as though the file doesn't
1176
raise errors.NoSuchFile(self._full_path())
1177
if line != self.HEADER:
1178
raise KnitHeaderError(badline=line,
1179
filename=self._transport.abspath(self._filename))
1182
return '%s(%s)' % (self.__class__.__name__, self._filename)
1185
class _KnitIndex(_KnitComponentFile):
1186
"""Manages knit index file.
1188
The index is already kept in memory and read on startup, to enable
1189
fast lookups of revision information. The cursor of the index
1190
file is always pointing to the end, making it easy to append
1193
_cache is a cache for fast mapping from version id to a Index
1196
_history is a cache for fast mapping from indexes to version ids.
1198
The index data format is dictionary compressed when it comes to
1199
parent references; a index entry may only have parents that with a
1200
lover index number. As a result, the index is topological sorted.
1202
Duplicate entries may be written to the index for a single version id
1203
if this is done then the latter one completely replaces the former:
1204
this allows updates to correct version and parent information.
1205
Note that the two entries may share the delta, and that successive
1206
annotations and references MUST point to the first entry.
1208
The index file on disc contains a header, followed by one line per knit
1209
record. The same revision can be present in an index file more than once.
1210
The first occurrence gets assigned a sequence number starting from 0.
1212
The format of a single line is
1213
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1214
REVISION_ID is a utf8-encoded revision id
1215
FLAGS is a comma separated list of flags about the record. Values include
1216
no-eol, line-delta, fulltext.
1217
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1218
that the the compressed data starts at.
1219
LENGTH is the ascii representation of the length of the data file.
1220
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1222
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1223
revision id already in the knit that is a parent of REVISION_ID.
1224
The ' :' marker is the end of record marker.
1227
when a write is interrupted to the index file, it will result in a line
1228
that does not end in ' :'. If the ' :' is not present at the end of a line,
1229
or at the end of the file, then the record that is missing it will be
1230
ignored by the parser.
1232
When writing new records to the index file, the data is preceded by '\n'
1233
to ensure that records always start on new lines even if the last write was
1234
interrupted. As a result its normal for the last line in the index to be
1235
missing a trailing newline. One can be added with no harmful effects.
1238
HEADER = "# bzr knit index 8\n"
1240
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1241
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1243
def _cache_version(self, version_id, options, pos, size, parents):
1244
"""Cache a version record in the history array and index cache.
1246
This is inlined into _load_data for performance. KEEP IN SYNC.
1247
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1250
# only want the _history index to reference the 1st index entry
1252
if version_id not in self._cache:
1253
index = len(self._history)
1254
self._history.append(version_id)
1256
index = self._cache[version_id][5]
1257
self._cache[version_id] = (version_id,
1264
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1265
create_parent_dir=False, delay_create=False, dir_mode=None):
1266
_KnitComponentFile.__init__(self, transport, filename, mode,
1267
file_mode=file_mode,
1268
create_parent_dir=create_parent_dir,
1271
# position in _history is the 'official' index for a revision
1272
# but the values may have come from a newer entry.
1273
# so - wc -l of a knit index is != the number of unique names
1277
fp = self._transport.get(self._filename)
1279
# _load_data may raise NoSuchFile if the target knit is
1281
_load_data(self, fp)
1285
if mode != 'w' or not create:
1288
self._need_to_create = True
1290
self._transport.put_bytes_non_atomic(
1291
self._filename, self.HEADER, mode=self._file_mode)
1293
def get_graph(self):
1294
"""Return a list of the node:parents lists from this knit index."""
1295
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1297
def get_ancestry(self, versions, topo_sorted=True):
1298
"""See VersionedFile.get_ancestry."""
1299
# get a graph of all the mentioned versions:
1301
pending = set(versions)
1304
version = pending.pop()
1307
parents = [p for p in cache[version][4] if p in cache]
1309
raise RevisionNotPresent(version, self._filename)
1310
# if not completed and not a ghost
1311
pending.update([p for p in parents if p not in graph])
1312
graph[version] = parents
1315
return topo_sort(graph.items())
1317
def get_ancestry_with_ghosts(self, versions):
1318
"""See VersionedFile.get_ancestry_with_ghosts."""
1319
# get a graph of all the mentioned versions:
1320
self.check_versions_present(versions)
1323
pending = set(versions)
1325
version = pending.pop()
1327
parents = cache[version][4]
1333
pending.update([p for p in parents if p not in graph])
1334
graph[version] = parents
1335
return topo_sort(graph.items())
1337
def iter_parents(self, version_ids):
1338
"""Iterate through the parents for many version ids.
1340
:param version_ids: An iterable yielding version_ids.
1341
:return: An iterator that yields (version_id, parents). Requested
1342
version_ids not present in the versioned file are simply skipped.
1343
The order is undefined, allowing for different optimisations in
1344
the underlying implementation.
1346
for version_id in version_ids:
1348
yield version_id, tuple(self.get_parents(version_id))
1352
def num_versions(self):
1353
return len(self._history)
1355
__len__ = num_versions
1357
def get_versions(self):
1358
"""Get all the versions in the file. not topologically sorted."""
1359
return self._history
1361
def _version_list_to_index(self, versions):
1364
for version in versions:
1365
if version in cache:
1366
# -- inlined lookup() --
1367
result_list.append(str(cache[version][5]))
1368
# -- end lookup () --
1370
result_list.append('.' + version)
1371
return ' '.join(result_list)
1373
def add_version(self, version_id, options, index_memo, parents):
1374
"""Add a version record to the index."""
1375
self.add_versions(((version_id, options, index_memo, parents),))
1377
def add_versions(self, versions, random_id=False):
1378
"""Add multiple versions to the index.
1380
:param versions: a list of tuples:
1381
(version_id, options, pos, size, parents).
1382
:param random_id: If True the ids being added were randomly generated
1383
and no check for existence will be performed.
1386
orig_history = self._history[:]
1387
orig_cache = self._cache.copy()
1390
for version_id, options, (index, pos, size), parents in versions:
1391
line = "\n%s %s %s %s %s :" % (version_id,
1395
self._version_list_to_index(parents))
1396
assert isinstance(line, str), \
1397
'content must be utf-8 encoded: %r' % (line,)
1399
self._cache_version(version_id, options, pos, size, parents)
1400
if not self._need_to_create:
1401
self._transport.append_bytes(self._filename, ''.join(lines))
1404
sio.write(self.HEADER)
1405
sio.writelines(lines)
1407
self._transport.put_file_non_atomic(self._filename, sio,
1408
create_parent_dir=self._create_parent_dir,
1409
mode=self._file_mode,
1410
dir_mode=self._dir_mode)
1411
self._need_to_create = False
1413
# If any problems happen, restore the original values and re-raise
1414
self._history = orig_history
1415
self._cache = orig_cache
1418
def has_version(self, version_id):
1419
"""True if the version is in the index."""
1420
return version_id in self._cache
1422
def get_position(self, version_id):
1423
"""Return details needed to access the version.
1425
.kndx indices do not support split-out data, so return None for the
1428
:return: a tuple (None, data position, size) to hand to the access
1429
logic to get the record.
1431
entry = self._cache[version_id]
1432
return None, entry[2], entry[3]
1434
def get_method(self, version_id):
1435
"""Return compression method of specified version."""
1437
options = self._cache[version_id][1]
1439
raise RevisionNotPresent(version_id, self._filename)
1440
if 'fulltext' in options:
1443
if 'line-delta' not in options:
1444
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1447
def get_options(self, version_id):
1448
"""Return a string represention options.
1452
return self._cache[version_id][1]
1454
def get_parents(self, version_id):
1455
"""Return parents of specified version ignoring ghosts."""
1456
return [parent for parent in self._cache[version_id][4]
1457
if parent in self._cache]
1459
def get_parents_with_ghosts(self, version_id):
1460
"""Return parents of specified version with ghosts."""
1461
return self._cache[version_id][4]
1463
def check_versions_present(self, version_ids):
1464
"""Check that all specified versions are present."""
1466
for version_id in version_ids:
1467
if version_id not in cache:
1468
raise RevisionNotPresent(version_id, self._filename)
1471
class KnitGraphIndex(object):
1472
"""A knit index that builds on GraphIndex."""
1474
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1475
"""Construct a KnitGraphIndex on a graph_index.
1477
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1478
:param deltas: Allow delta-compressed records.
1479
:param add_callback: If not None, allow additions to the index and call
1480
this callback with a list of added GraphIndex nodes:
1481
[(node, value, node_refs), ...]
1482
:param parents: If True, record knits parents, if not do not record
1485
self._graph_index = graph_index
1486
self._deltas = deltas
1487
self._add_callback = add_callback
1488
self._parents = parents
1489
if deltas and not parents:
1490
raise KnitCorrupt(self, "Cannot do delta compression without "
1493
def _get_entries(self, keys, check_present=False):
1494
"""Get the entries for keys.
1496
:param keys: An iterable of index keys, - 1-tuples.
1501
for node in self._graph_index.iter_entries(keys):
1503
found_keys.add(node[1])
1505
# adapt parentless index to the rest of the code.
1506
for node in self._graph_index.iter_entries(keys):
1507
yield node[0], node[1], node[2], ()
1508
found_keys.add(node[1])
1510
missing_keys = keys.difference(found_keys)
1512
raise RevisionNotPresent(missing_keys.pop(), self)
1514
def _present_keys(self, version_ids):
1516
node[1] for node in self._get_entries(version_ids)])
1518
def _parentless_ancestry(self, versions):
1519
"""Honour the get_ancestry API for parentless knit indices."""
1520
wanted_keys = self._version_ids_to_keys(versions)
1521
present_keys = self._present_keys(wanted_keys)
1522
missing = set(wanted_keys).difference(present_keys)
1524
raise RevisionNotPresent(missing.pop(), self)
1525
return list(self._keys_to_version_ids(present_keys))
1527
def get_ancestry(self, versions, topo_sorted=True):
1528
"""See VersionedFile.get_ancestry."""
1529
if not self._parents:
1530
return self._parentless_ancestry(versions)
1531
# XXX: This will do len(history) index calls - perhaps
1532
# it should be altered to be a index core feature?
1533
# get a graph of all the mentioned versions:
1536
versions = self._version_ids_to_keys(versions)
1537
pending = set(versions)
1539
# get all pending nodes
1540
this_iteration = pending
1541
new_nodes = self._get_entries(this_iteration)
1544
for (index, key, value, node_refs) in new_nodes:
1545
# dont ask for ghosties - otherwise
1546
# we we can end up looping with pending
1547
# being entirely ghosted.
1548
graph[key] = [parent for parent in node_refs[0]
1549
if parent not in ghosts]
1551
for parent in graph[key]:
1552
# dont examine known nodes again
1557
ghosts.update(this_iteration.difference(found))
1558
if versions.difference(graph):
1559
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1561
result_keys = topo_sort(graph.items())
1563
result_keys = graph.iterkeys()
1564
return [key[0] for key in result_keys]
1566
def get_ancestry_with_ghosts(self, versions):
1567
"""See VersionedFile.get_ancestry."""
1568
if not self._parents:
1569
return self._parentless_ancestry(versions)
1570
# XXX: This will do len(history) index calls - perhaps
1571
# it should be altered to be a index core feature?
1572
# 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)
1581
for (index, key, value, node_refs) in new_nodes:
1582
graph[key] = node_refs[0]
1584
for parent in graph[key]:
1585
# dont examine known nodes again
1589
missing_versions = this_iteration.difference(graph)
1590
missing_needed = versions.intersection(missing_versions)
1592
raise RevisionNotPresent(missing_needed.pop(), self)
1593
for missing_version in missing_versions:
1594
# add a key, no parents
1595
graph[missing_version] = []
1596
pending.discard(missing_version) # don't look for it
1597
result_keys = topo_sort(graph.items())
1598
return [key[0] for key in result_keys]
1600
def get_graph(self):
1601
"""Return a list of the node:parents lists from this knit index."""
1602
if not self._parents:
1603
return [(key, ()) for key in self.get_versions()]
1605
for index, key, value, refs in self._graph_index.iter_all_entries():
1606
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1609
def iter_parents(self, version_ids):
1610
"""Iterate through the parents for many version ids.
1612
:param version_ids: An iterable yielding version_ids.
1613
:return: An iterator that yields (version_id, parents). Requested
1614
version_ids not present in the versioned file are simply skipped.
1615
The order is undefined, allowing for different optimisations in
1616
the underlying implementation.
1619
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1621
present_parents = set()
1622
for node in all_nodes:
1623
all_parents.update(node[3][0])
1624
# any node we are querying must be present
1625
present_parents.add(node[1])
1626
unknown_parents = all_parents.difference(present_parents)
1627
present_parents.update(self._present_keys(unknown_parents))
1628
for node in all_nodes:
1630
for parent in node[3][0]:
1631
if parent in present_parents:
1632
parents.append(parent[0])
1633
yield node[1][0], tuple(parents)
1635
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1636
yield node[1][0], ()
1638
def num_versions(self):
1639
return len(list(self._graph_index.iter_all_entries()))
1641
__len__ = num_versions
1643
def get_versions(self):
1644
"""Get all the versions in the file. not topologically sorted."""
1645
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1647
def has_version(self, version_id):
1648
"""True if the version is in the index."""
1649
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1651
def _keys_to_version_ids(self, keys):
1652
return tuple(key[0] for key in keys)
1654
def get_position(self, version_id):
1655
"""Return details needed to access the version.
1657
:return: a tuple (index, data position, size) to hand to the access
1658
logic to get the record.
1660
node = self._get_node(version_id)
1661
bits = node[2][1:].split(' ')
1662
return node[0], int(bits[0]), int(bits[1])
1664
def get_method(self, version_id):
1665
"""Return compression method of specified version."""
1666
if not self._deltas:
1668
return self._parent_compression(self._get_node(version_id)[3][1])
1670
def _parent_compression(self, reference_list):
1671
# use the second reference list to decide if this is delta'd or not.
1672
if len(reference_list):
1677
def _get_node(self, version_id):
1679
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1681
raise RevisionNotPresent(version_id, self)
1683
def get_options(self, version_id):
1684
"""Return a string represention options.
1688
node = self._get_node(version_id)
1689
if not self._deltas:
1690
options = ['fulltext']
1692
options = [self._parent_compression(node[3][1])]
1693
if node[2][0] == 'N':
1694
options.append('no-eol')
1697
def get_parents(self, version_id):
1698
"""Return parents of specified version ignoring ghosts."""
1699
parents = list(self.iter_parents([version_id]))
1702
raise errors.RevisionNotPresent(version_id, self)
1703
return parents[0][1]
1705
def get_parents_with_ghosts(self, version_id):
1706
"""Return parents of specified version with ghosts."""
1707
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1708
check_present=True))
1709
if not self._parents:
1711
return self._keys_to_version_ids(nodes[0][3][0])
1713
def check_versions_present(self, version_ids):
1714
"""Check that all specified versions are present."""
1715
keys = self._version_ids_to_keys(version_ids)
1716
present = self._present_keys(keys)
1717
missing = keys.difference(present)
1719
raise RevisionNotPresent(missing.pop(), self)
1721
def add_version(self, version_id, options, access_memo, parents):
1722
"""Add a version record to the index."""
1723
return self.add_versions(((version_id, options, access_memo, parents),))
1725
def add_versions(self, versions, random_id=False):
1726
"""Add multiple versions to the index.
1728
This function does not insert data into the Immutable GraphIndex
1729
backing the KnitGraphIndex, instead it prepares data for insertion by
1730
the caller and checks that it is safe to insert then calls
1731
self._add_callback with the prepared GraphIndex nodes.
1733
:param versions: a list of tuples:
1734
(version_id, options, pos, size, parents).
1735
:param random_id: If True the ids being added were randomly generated
1736
and no check for existence will be performed.
1738
if not self._add_callback:
1739
raise errors.ReadOnlyError(self)
1740
# we hope there are no repositories with inconsistent parentage
1745
for (version_id, options, access_memo, parents) in versions:
1746
index, pos, size = access_memo
1747
key = (version_id, )
1748
parents = tuple((parent, ) for parent in parents)
1749
if 'no-eol' in options:
1753
value += "%d %d" % (pos, size)
1754
if not self._deltas:
1755
if 'line-delta' in options:
1756
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1759
if 'line-delta' in options:
1760
node_refs = (parents, (parents[0],))
1762
node_refs = (parents, ())
1764
node_refs = (parents, )
1767
raise KnitCorrupt(self, "attempt to add node with parents "
1768
"in parentless index.")
1770
keys[key] = (value, node_refs)
1772
present_nodes = self._get_entries(keys)
1773
for (index, key, value, node_refs) in present_nodes:
1774
if (value, node_refs) != keys[key]:
1775
raise KnitCorrupt(self, "inconsistent details in add_versions"
1776
": %s %s" % ((value, node_refs), keys[key]))
1780
for key, (value, node_refs) in keys.iteritems():
1781
result.append((key, value, node_refs))
1783
for key, (value, node_refs) in keys.iteritems():
1784
result.append((key, value))
1785
self._add_callback(result)
1787
def _version_ids_to_keys(self, version_ids):
1788
return set((version_id, ) for version_id in version_ids)
1791
class _KnitAccess(object):
1792
"""Access to knit records in a .knit file."""
1794
def __init__(self, transport, filename, _file_mode, _dir_mode,
1795
_need_to_create, _create_parent_dir):
1796
"""Create a _KnitAccess for accessing and inserting data.
1798
:param transport: The transport the .knit is located on.
1799
:param filename: The filename of the .knit.
1801
self._transport = transport
1802
self._filename = filename
1803
self._file_mode = _file_mode
1804
self._dir_mode = _dir_mode
1805
self._need_to_create = _need_to_create
1806
self._create_parent_dir = _create_parent_dir
1808
def add_raw_records(self, sizes, raw_data):
1809
"""Add raw knit bytes to a storage area.
1811
The data is spooled to whereever the access method is storing data.
1813
:param sizes: An iterable containing the size of each raw data segment.
1814
:param raw_data: A bytestring containing the data.
1815
:return: A list of memos to retrieve the record later. Each memo is a
1816
tuple - (index, pos, length), where the index field is always None
1817
for the .knit access method.
1819
assert type(raw_data) == str, \
1820
'data must be plain bytes was %s' % type(raw_data)
1821
if not self._need_to_create:
1822
base = self._transport.append_bytes(self._filename, raw_data)
1824
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1825
create_parent_dir=self._create_parent_dir,
1826
mode=self._file_mode,
1827
dir_mode=self._dir_mode)
1828
self._need_to_create = False
1832
result.append((None, base, size))
1837
"""IFF this data access has its own storage area, initialise it.
1841
self._transport.put_bytes_non_atomic(self._filename, '',
1842
mode=self._file_mode)
1844
def open_file(self):
1845
"""IFF this data access can be represented as a single file, open it.
1847
For knits that are not mapped to a single file on disk this will
1850
:return: None or a file handle.
1853
return self._transport.get(self._filename)
1858
def get_raw_records(self, memos_for_retrieval):
1859
"""Get the raw bytes for a records.
1861
:param memos_for_retrieval: An iterable containing the (index, pos,
1862
length) memo for retrieving the bytes. The .knit method ignores
1863
the index as there is always only a single file.
1864
:return: An iterator over the bytes of the records.
1866
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
1867
for pos, data in self._transport.readv(self._filename, read_vector):
1871
class _PackAccess(object):
1872
"""Access to knit records via a collection of packs."""
1874
def __init__(self, index_to_packs, writer=None):
1875
"""Create a _PackAccess object.
1877
:param index_to_packs: A dict mapping index objects to the transport
1878
and file names for obtaining data.
1879
:param writer: A tuple (pack.ContainerWriter, write_index) which
1880
contains the pack to write, and the index that reads from it will
1884
self.container_writer = writer[0]
1885
self.write_index = writer[1]
1887
self.container_writer = None
1888
self.write_index = None
1889
self.indices = index_to_packs
1891
def add_raw_records(self, sizes, raw_data):
1892
"""Add raw knit bytes to a storage area.
1894
The data is spooled to the container writer in one bytes-record per
1897
:param sizes: An iterable containing the size of each raw data segment.
1898
:param raw_data: A bytestring containing the data.
1899
:return: A list of memos to retrieve the record later. Each memo is a
1900
tuple - (index, pos, length), where the index field is the
1901
write_index object supplied to the PackAccess object.
1903
assert type(raw_data) == str, \
1904
'data must be plain bytes was %s' % type(raw_data)
1908
p_offset, p_length = self.container_writer.add_bytes_record(
1909
raw_data[offset:offset+size], [])
1911
result.append((self.write_index, p_offset, p_length))
1915
"""Pack based knits do not get individually created."""
1917
def get_raw_records(self, memos_for_retrieval):
1918
"""Get the raw bytes for a records.
1920
:param memos_for_retrieval: An iterable containing the (index, pos,
1921
length) memo for retrieving the bytes. The Pack access method
1922
looks up the pack to use for a given record in its index_to_pack
1924
:return: An iterator over the bytes of the records.
1926
# first pass, group into same-index requests
1928
current_index = None
1929
for (index, offset, length) in memos_for_retrieval:
1930
if current_index == index:
1931
current_list.append((offset, length))
1933
if current_index is not None:
1934
request_lists.append((current_index, current_list))
1935
current_index = index
1936
current_list = [(offset, length)]
1937
# handle the last entry
1938
if current_index is not None:
1939
request_lists.append((current_index, current_list))
1940
for index, offsets in request_lists:
1941
transport, path = self.indices[index]
1942
reader = pack.make_readv_reader(transport, path, offsets)
1943
for names, read_func in reader.iter_records():
1944
yield read_func(None)
1946
def open_file(self):
1947
"""Pack based knits have no single file."""
1950
def set_writer(self, writer, index, (transport, packname)):
1951
"""Set a writer to use for adding data."""
1952
self.indices[index] = (transport, packname)
1953
self.container_writer = writer
1954
self.write_index = index
1957
class _KnitData(object):
1958
"""Manage extraction of data from a KnitAccess, caching and decompressing.
1960
The KnitData class provides the logic for parsing and using knit records,
1961
making use of an access method for the low level read and write operations.
1964
def __init__(self, access):
1965
"""Create a KnitData object.
1967
:param access: The access method to use. Access methods such as
1968
_KnitAccess manage the insertion of raw records and the subsequent
1969
retrieval of the same.
1971
self._access = access
1972
self._checked = False
1973
# TODO: jam 20060713 conceptually, this could spill to disk
1974
# if the cached size gets larger than a certain amount
1975
# but it complicates the model a bit, so for now just use
1976
# a simple dictionary
1978
self._do_cache = False
1980
def enable_cache(self):
1981
"""Enable caching of reads."""
1982
self._do_cache = True
1984
def clear_cache(self):
1985
"""Clear the record cache."""
1986
self._do_cache = False
1989
def _open_file(self):
1990
return self._access.open_file()
1992
def _record_to_data(self, version_id, digest, lines):
1993
"""Convert version_id, digest, lines into a raw data block.
1995
:return: (len, a StringIO instance with the raw data ready to read.)
1997
bytes = (''.join(chain(
1998
["version %s %d %s\n" % (version_id,
2002
["end %s\n" % version_id])))
2003
assert bytes.__class__ == str
2004
compressed_bytes = bytes_to_gzip(bytes)
2005
return len(compressed_bytes), compressed_bytes
2007
def add_raw_records(self, sizes, raw_data):
2008
"""Append a prepared record to the data file.
2010
:param sizes: An iterable containing the size of each raw data segment.
2011
:param raw_data: A bytestring containing the data.
2012
:return: a list of index data for the way the data was stored.
2013
See the access method add_raw_records documentation for more
2016
return self._access.add_raw_records(sizes, raw_data)
2018
def _parse_record_header(self, version_id, raw_data):
2019
"""Parse a record header for consistency.
2021
:return: the header and the decompressor stream.
2022
as (stream, header_record)
2024
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2026
rec = self._check_header(version_id, df.readline())
2027
except Exception, e:
2028
raise KnitCorrupt(self._access,
2029
"While reading {%s} got %s(%s)"
2030
% (version_id, e.__class__.__name__, str(e)))
2033
def _check_header(self, version_id, line):
2036
raise KnitCorrupt(self._access,
2037
'unexpected number of elements in record header')
2038
if rec[1] != version_id:
2039
raise KnitCorrupt(self._access,
2040
'unexpected version, wanted %r, got %r'
2041
% (version_id, rec[1]))
2044
def _parse_record(self, version_id, data):
2046
# 4168 calls in 2880 217 internal
2047
# 4168 calls to _parse_record_header in 2121
2048
# 4168 calls to readlines in 330
2049
df = GzipFile(mode='rb', fileobj=StringIO(data))
2052
record_contents = df.readlines()
2053
except Exception, e:
2054
raise KnitCorrupt(self._access,
2055
"While reading {%s} got %s(%s)"
2056
% (version_id, e.__class__.__name__, str(e)))
2057
header = record_contents.pop(0)
2058
rec = self._check_header(version_id, header)
2060
last_line = record_contents.pop()
2061
if len(record_contents) != int(rec[2]):
2062
raise KnitCorrupt(self._access,
2063
'incorrect number of lines %s != %s'
2065
% (len(record_contents), int(rec[2]),
2067
if last_line != 'end %s\n' % rec[1]:
2068
raise KnitCorrupt(self._access,
2069
'unexpected version end line %r, wanted %r'
2070
% (last_line, version_id))
2072
return record_contents, rec[3]
2074
def read_records_iter_raw(self, records):
2075
"""Read text records from data file and yield raw data.
2077
This unpacks enough of the text record to validate the id is
2078
as expected but thats all.
2080
# setup an iterator of the external records:
2081
# uses readv so nice and fast we hope.
2083
# grab the disk data needed.
2085
# Don't check _cache if it is empty
2086
needed_offsets = [index_memo for version_id, index_memo
2088
if version_id not in self._cache]
2090
needed_offsets = [index_memo for version_id, index_memo
2093
raw_records = self._access.get_raw_records(needed_offsets)
2095
for version_id, index_memo in records:
2096
if version_id in self._cache:
2097
# This data has already been validated
2098
data = self._cache[version_id]
2100
data = raw_records.next()
2102
self._cache[version_id] = data
2104
# validate the header
2105
df, rec = self._parse_record_header(version_id, data)
2107
yield version_id, data
2109
def read_records_iter(self, records):
2110
"""Read text records from data file and yield result.
2112
The result will be returned in whatever is the fastest to read.
2113
Not by the order requested. Also, multiple requests for the same
2114
record will only yield 1 response.
2115
:param records: A list of (version_id, pos, len) entries
2116
:return: Yields (version_id, contents, digest) in the order
2117
read, not the order requested
2123
# Skip records we have alread seen
2124
yielded_records = set()
2125
needed_records = set()
2126
for record in records:
2127
if record[0] in self._cache:
2128
if record[0] in yielded_records:
2130
yielded_records.add(record[0])
2131
data = self._cache[record[0]]
2132
content, digest = self._parse_record(record[0], data)
2133
yield (record[0], content, digest)
2135
needed_records.add(record)
2136
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2138
needed_records = sorted(set(records), key=operator.itemgetter(1))
2140
if not needed_records:
2143
# The transport optimizes the fetching as well
2144
# (ie, reads continuous ranges.)
2145
raw_data = self._access.get_raw_records(
2146
[index_memo for version_id, index_memo in needed_records])
2148
for (version_id, index_memo), data in \
2149
izip(iter(needed_records), raw_data):
2150
content, digest = self._parse_record(version_id, data)
2152
self._cache[version_id] = data
2153
yield version_id, content, digest
2155
def read_records(self, records):
2156
"""Read records into a dictionary."""
2158
for record_id, content, digest in \
2159
self.read_records_iter(records):
2160
components[record_id] = (content, digest)
2164
class InterKnit(InterVersionedFile):
2165
"""Optimised code paths for knit to knit operations."""
2167
_matching_file_from_factory = KnitVersionedFile
2168
_matching_file_to_factory = KnitVersionedFile
2171
def is_compatible(source, target):
2172
"""Be compatible with knits. """
2174
return (isinstance(source, KnitVersionedFile) and
2175
isinstance(target, KnitVersionedFile))
2176
except AttributeError:
2179
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2180
"""See InterVersionedFile.join."""
2181
assert isinstance(self.source, KnitVersionedFile)
2182
assert isinstance(self.target, KnitVersionedFile)
2184
# If the source and target are mismatched w.r.t. annotations vs
2185
# plain, the data needs to be converted accordingly
2186
if self.source.factory.annotated == self.target.factory.annotated:
2188
elif self.source.factory.annotated:
2189
converter = self._anno_to_plain_converter
2191
# We're converting from a plain to an annotated knit. This requires
2192
# building the annotations from scratch. The generic join code
2193
# handles this implicitly so we delegate to it.
2194
return super(InterKnit, self).join(pb, msg, version_ids,
2197
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2201
pb = ui.ui_factory.nested_progress_bar()
2203
version_ids = list(version_ids)
2204
if None in version_ids:
2205
version_ids.remove(None)
2207
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2208
this_versions = set(self.target._index.get_versions())
2209
# XXX: For efficiency we should not look at the whole index,
2210
# we only need to consider the referenced revisions - they
2211
# must all be present, or the method must be full-text.
2212
# TODO, RBC 20070919
2213
needed_versions = self.source_ancestry - this_versions
2215
if not needed_versions:
2217
full_list = topo_sort(self.source.get_graph())
2219
version_list = [i for i in full_list if (not self.target.has_version(i)
2220
and i in needed_versions)]
2224
copy_queue_records = []
2226
for version_id in version_list:
2227
options = self.source._index.get_options(version_id)
2228
parents = self.source._index.get_parents_with_ghosts(version_id)
2229
# check that its will be a consistent copy:
2230
for parent in parents:
2231
# if source has the parent, we must :
2232
# * already have it or
2233
# * have it scheduled already
2234
# otherwise we don't care
2235
assert (self.target.has_version(parent) or
2236
parent in copy_set or
2237
not self.source.has_version(parent))
2238
index_memo = self.source._index.get_position(version_id)
2239
copy_queue_records.append((version_id, index_memo))
2240
copy_queue.append((version_id, options, parents))
2241
copy_set.add(version_id)
2243
# data suck the join:
2245
total = len(version_list)
2248
for (version_id, raw_data), \
2249
(version_id2, options, parents) in \
2250
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2252
assert version_id == version_id2, 'logic error, inconsistent results'
2254
pb.update("Joining knit", count, total)
2256
size, raw_data = converter(raw_data, version_id, options,
2259
size = len(raw_data)
2260
raw_records.append((version_id, options, parents, size))
2261
raw_datum.append(raw_data)
2262
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2267
def _anno_to_plain_converter(self, raw_data, version_id, options,
2269
"""Convert annotated content to plain content."""
2270
data, digest = self.source._data._parse_record(version_id, raw_data)
2271
if 'fulltext' in options:
2272
content = self.source.factory.parse_fulltext(data, version_id)
2273
lines = self.target.factory.lower_fulltext(content)
2275
delta = self.source.factory.parse_line_delta(data, version_id,
2277
lines = self.target.factory.lower_line_delta(delta)
2278
return self.target._data._record_to_data(version_id, digest, lines)
2281
InterVersionedFile.register_optimiser(InterKnit)
2284
class WeaveToKnit(InterVersionedFile):
2285
"""Optimised code paths for weave to knit operations."""
2287
_matching_file_from_factory = bzrlib.weave.WeaveFile
2288
_matching_file_to_factory = KnitVersionedFile
2291
def is_compatible(source, target):
2292
"""Be compatible with weaves to knits."""
2294
return (isinstance(source, bzrlib.weave.Weave) and
2295
isinstance(target, KnitVersionedFile))
2296
except AttributeError:
2299
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2300
"""See InterVersionedFile.join."""
2301
assert isinstance(self.source, bzrlib.weave.Weave)
2302
assert isinstance(self.target, KnitVersionedFile)
2304
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2309
pb = ui.ui_factory.nested_progress_bar()
2311
version_ids = list(version_ids)
2313
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2314
this_versions = set(self.target._index.get_versions())
2315
needed_versions = self.source_ancestry - this_versions
2317
if not needed_versions:
2319
full_list = topo_sort(self.source.get_graph())
2321
version_list = [i for i in full_list if (not self.target.has_version(i)
2322
and i in needed_versions)]
2326
total = len(version_list)
2327
for version_id in version_list:
2328
pb.update("Converting to knit", count, total)
2329
parents = self.source.get_parents(version_id)
2330
# check that its will be a consistent copy:
2331
for parent in parents:
2332
# if source has the parent, we must already have it
2333
assert (self.target.has_version(parent))
2334
self.target.add_lines(
2335
version_id, parents, self.source.get_lines(version_id))
2342
InterVersionedFile.register_optimiser(WeaveToKnit)
2345
# Deprecated, use PatienceSequenceMatcher instead
2346
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2349
def annotate_knit(knit, revision_id):
2350
"""Annotate a knit with no cached annotations.
2352
This implementation is for knits with no cached annotations.
2353
It will work for knits with cached annotations, but this is not
2356
ancestry = knit.get_ancestry(revision_id)
2357
fulltext = dict(zip(ancestry, knit.get_line_list(ancestry)))
2359
for candidate in ancestry:
2360
if candidate in annotations:
2362
parents = knit.get_parents(candidate)
2363
if len(parents) == 0:
2365
elif knit._index.get_method(candidate) != 'line-delta':
2368
parent, sha1, noeol, delta = knit.get_delta(candidate)
2369
blocks = KnitContent.get_line_delta_blocks(delta,
2370
fulltext[parents[0]], fulltext[candidate])
2371
annotations[candidate] = list(annotate.reannotate([annotations[p]
2372
for p in parents], fulltext[candidate], candidate, blocks))
2373
return iter(annotations[revision_id])
2377
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2379
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
