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
if not isinstance(required_versions, set):
559
required_versions = set(required_versions)
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
self.check_not_reserved_id(version_id)
615
parents = self.get_parents(version_id)
620
index_memo = self._index.get_position(version_id)
621
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
622
noeol = 'no-eol' in self._index.get_options(version_id)
623
if 'fulltext' == self._index.get_method(version_id):
624
new_content = self.factory.parse_fulltext(data, version_id)
625
if parent is not None:
626
reference_content = self._get_content(parent)
627
old_texts = reference_content.text()
630
new_texts = new_content.text()
631
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
633
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
635
delta = self.factory.parse_line_delta(data, version_id)
636
return parent, sha1, noeol, delta
638
def get_format_signature(self):
639
"""See VersionedFile.get_format_signature()."""
640
if self.factory.annotated:
641
annotated_part = "annotated"
643
annotated_part = "plain"
644
return "knit-%s" % (annotated_part,)
646
def get_graph_with_ghosts(self):
647
"""See VersionedFile.get_graph_with_ghosts()."""
648
graph_items = self._index.get_graph()
649
return dict(graph_items)
651
def get_sha1(self, version_id):
652
return self.get_sha1s([version_id])[0]
654
def get_sha1s(self, version_ids):
655
"""See VersionedFile.get_sha1()."""
656
record_map = self._get_record_map(version_ids)
657
# record entry 2 is the 'digest'.
658
return [record_map[v][2] for v in version_ids]
662
"""See VersionedFile.get_suffixes()."""
663
return [DATA_SUFFIX, INDEX_SUFFIX]
665
def has_ghost(self, version_id):
666
"""True if there is a ghost reference in the file to version_id."""
668
if self.has_version(version_id):
670
# optimisable if needed by memoising the _ghosts set.
671
items = self._index.get_graph()
672
for node, parents in items:
673
for parent in parents:
674
if parent not in self._index._cache:
675
if parent == version_id:
679
def insert_data_stream(self, (format, data_list, reader_callable)):
680
"""Insert knit records from a data stream into this knit.
682
If a version in the stream is already present in this knit, it will not
683
be inserted a second time. It will be checked for consistency with the
684
stored version however, and may cause a KnitCorrupt error to be raised
685
if the data in the stream disagrees with the already stored data.
687
:seealso: get_data_stream
689
if format != self.get_format_signature():
690
trace.mutter('incompatible format signature inserting to %r', self)
691
raise KnitDataStreamIncompatible(
692
format, self.get_format_signature())
694
for version_id, options, length, parents in data_list:
695
if self.has_version(version_id):
696
# First check: the list of parents.
697
my_parents = self.get_parents_with_ghosts(version_id)
698
if my_parents != parents:
699
# XXX: KnitCorrupt is not quite the right exception here.
702
'parents list %r from data stream does not match '
703
'already recorded parents %r for %s'
704
% (parents, my_parents, version_id))
706
# Also check the SHA-1 of the fulltext this content will
708
raw_data = reader_callable(length)
709
my_fulltext_sha1 = self.get_sha1(version_id)
710
df, rec = self._data._parse_record_header(version_id, raw_data)
711
stream_fulltext_sha1 = rec[3]
712
if my_fulltext_sha1 != stream_fulltext_sha1:
713
# Actually, we don't know if it's this knit that's corrupt,
714
# or the data stream we're trying to insert.
716
self.filename, 'sha-1 does not match %s' % version_id)
718
self._add_raw_records(
719
[(version_id, options, parents, length)],
720
reader_callable(length))
723
"""See VersionedFile.versions."""
724
if 'evil' in debug.debug_flags:
725
trace.mutter_callsite(2, "versions scales with size of history")
726
return self._index.get_versions()
728
def has_version(self, version_id):
729
"""See VersionedFile.has_version."""
730
if 'evil' in debug.debug_flags:
731
trace.mutter_callsite(2, "has_version is a LBYL scenario")
732
return self._index.has_version(version_id)
734
__contains__ = has_version
736
def _merge_annotations(self, content, parents, parent_texts={},
737
delta=None, annotated=None,
738
left_matching_blocks=None):
739
"""Merge annotations for content. This is done by comparing
740
the annotations based on changed to the text.
742
if left_matching_blocks is not None:
743
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
747
for parent_id in parents:
748
merge_content = self._get_content(parent_id, parent_texts)
749
if (parent_id == parents[0] and delta_seq is not None):
752
seq = patiencediff.PatienceSequenceMatcher(
753
None, merge_content.text(), content.text())
754
for i, j, n in seq.get_matching_blocks():
757
# this appears to copy (origin, text) pairs across to the
758
# new content for any line that matches the last-checked
760
content._lines[j:j+n] = merge_content._lines[i:i+n]
762
if delta_seq is None:
763
reference_content = self._get_content(parents[0], parent_texts)
764
new_texts = content.text()
765
old_texts = reference_content.text()
766
delta_seq = patiencediff.PatienceSequenceMatcher(
767
None, old_texts, new_texts)
768
return self._make_line_delta(delta_seq, content)
770
def _make_line_delta(self, delta_seq, new_content):
771
"""Generate a line delta from delta_seq and new_content."""
773
for op in delta_seq.get_opcodes():
776
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
779
def _get_components_positions(self, version_ids):
780
"""Produce a map of position data for the components of versions.
782
This data is intended to be used for retrieving the knit records.
784
A dict of version_id to (method, data_pos, data_size, next) is
786
method is the way referenced data should be applied.
787
data_pos is the position of the data in the knit.
788
data_size is the size of the data in the knit.
789
next is the build-parent of the version, or None for fulltexts.
792
for version_id in version_ids:
795
while cursor is not None and cursor not in component_data:
796
method = self._index.get_method(cursor)
797
if method == 'fulltext':
800
next = self.get_parents(cursor)[0]
801
index_memo = self._index.get_position(cursor)
802
component_data[cursor] = (method, index_memo, next)
804
return component_data
806
def _get_content(self, version_id, parent_texts={}):
807
"""Returns a content object that makes up the specified
809
cached_version = parent_texts.get(version_id, None)
810
if cached_version is not None:
811
if not self.has_version(version_id):
812
raise RevisionNotPresent(version_id, self.filename)
813
return cached_version
815
text_map, contents_map = self._get_content_maps([version_id])
816
return contents_map[version_id]
818
def _check_versions_present(self, version_ids):
819
"""Check that all specified versions are present."""
820
self._index.check_versions_present(version_ids)
822
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
823
nostore_sha, random_id, check_content):
824
"""See VersionedFile.add_lines_with_ghosts()."""
825
self._check_add(version_id, lines, random_id, check_content)
826
return self._add(version_id, lines, parents, self.delta,
827
parent_texts, None, nostore_sha, random_id)
829
def _add_lines(self, version_id, parents, lines, parent_texts,
830
left_matching_blocks, nostore_sha, random_id, check_content):
831
"""See VersionedFile.add_lines."""
832
self._check_add(version_id, lines, random_id, check_content)
833
self._check_versions_present(parents)
834
return self._add(version_id, lines[:], parents, self.delta,
835
parent_texts, left_matching_blocks, nostore_sha, random_id)
837
def _check_add(self, version_id, lines, random_id, check_content):
838
"""check that version_id and lines are safe to add."""
839
if contains_whitespace(version_id):
840
raise InvalidRevisionId(version_id, self.filename)
841
self.check_not_reserved_id(version_id)
842
# Technically this could be avoided if we are happy to allow duplicate
843
# id insertion when other things than bzr core insert texts, but it
844
# seems useful for folk using the knit api directly to have some safety
845
# blanket that we can disable.
846
if not random_id and self.has_version(version_id):
847
raise RevisionAlreadyPresent(version_id, self.filename)
849
self._check_lines_not_unicode(lines)
850
self._check_lines_are_lines(lines)
852
def _add(self, version_id, lines, parents, delta, parent_texts,
853
left_matching_blocks, nostore_sha, random_id):
854
"""Add a set of lines on top of version specified by parents.
856
If delta is true, compress the text as a line-delta against
859
Any versions not present will be converted into ghosts.
861
# first thing, if the content is something we don't need to store, find
863
line_bytes = ''.join(lines)
864
digest = sha_string(line_bytes)
865
if nostore_sha == digest:
866
raise errors.ExistingContent
869
if parent_texts is None:
871
for parent in parents:
872
if self.has_version(parent):
873
present_parents.append(parent)
875
# can only compress against the left most present parent.
877
(len(present_parents) == 0 or
878
present_parents[0] != parents[0])):
881
text_length = len(line_bytes)
884
if lines[-1][-1] != '\n':
885
# copy the contents of lines.
887
options.append('no-eol')
888
lines[-1] = lines[-1] + '\n'
892
# To speed the extract of texts the delta chain is limited
893
# to a fixed number of deltas. This should minimize both
894
# I/O and the time spend applying deltas.
895
delta = self._check_should_delta(present_parents)
897
assert isinstance(version_id, str)
898
content = self.factory.make(lines, version_id)
899
if delta or (self.factory.annotated and len(present_parents) > 0):
900
# Merge annotations from parent texts if needed.
901
delta_hunks = self._merge_annotations(content, present_parents,
902
parent_texts, delta, self.factory.annotated,
903
left_matching_blocks)
906
options.append('line-delta')
907
store_lines = self.factory.lower_line_delta(delta_hunks)
908
size, bytes = self._data._record_to_data(version_id, digest,
911
options.append('fulltext')
912
# isinstance is slower and we have no hierarchy.
913
if self.factory.__class__ == KnitPlainFactory:
914
# Use the already joined bytes saving iteration time in
916
size, bytes = self._data._record_to_data(version_id, digest,
919
# get mixed annotation + content and feed it into the
921
store_lines = self.factory.lower_fulltext(content)
922
size, bytes = self._data._record_to_data(version_id, digest,
925
access_memo = self._data.add_raw_records([size], bytes)[0]
926
self._index.add_versions(
927
((version_id, options, access_memo, parents),),
929
return digest, text_length, content
931
def check(self, progress_bar=None):
932
"""See VersionedFile.check()."""
934
def _clone_text(self, new_version_id, old_version_id, parents):
935
"""See VersionedFile.clone_text()."""
936
# FIXME RBC 20060228 make fast by only inserting an index with null
938
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
940
def get_lines(self, version_id):
941
"""See VersionedFile.get_lines()."""
942
return self.get_line_list([version_id])[0]
944
def _get_record_map(self, version_ids):
945
"""Produce a dictionary of knit records.
947
The keys are version_ids, the values are tuples of (method, content,
949
method is the way the content should be applied.
950
content is a KnitContent object.
951
digest is the SHA1 digest of this version id after all steps are done
952
next is the build-parent of the version, i.e. the leftmost ancestor.
953
If the method is fulltext, next will be None.
955
position_map = self._get_components_positions(version_ids)
956
# c = component_id, m = method, i_m = index_memo, n = next
957
records = [(c, i_m) for c, (m, i_m, n) in position_map.iteritems()]
959
for component_id, content, digest in \
960
self._data.read_records_iter(records):
961
method, index_memo, next = position_map[component_id]
962
record_map[component_id] = method, content, digest, next
966
def get_text(self, version_id):
967
"""See VersionedFile.get_text"""
968
return self.get_texts([version_id])[0]
970
def get_texts(self, version_ids):
971
return [''.join(l) for l in self.get_line_list(version_ids)]
973
def get_line_list(self, version_ids):
974
"""Return the texts of listed versions as a list of strings."""
975
for version_id in version_ids:
976
self.check_not_reserved_id(version_id)
977
text_map, content_map = self._get_content_maps(version_ids)
978
return [text_map[v] for v in version_ids]
980
_get_lf_split_line_list = get_line_list
982
def _get_content_maps(self, version_ids):
983
"""Produce maps of text and KnitContents
985
:return: (text_map, content_map) where text_map contains the texts for
986
the requested versions and content_map contains the KnitContents.
987
Both dicts take version_ids as their keys.
989
record_map = self._get_record_map(version_ids)
994
for version_id in version_ids:
997
while cursor is not None:
998
method, data, digest, next = record_map[cursor]
999
components.append((cursor, method, data, digest))
1000
if cursor in content_map:
1005
for component_id, method, data, digest in reversed(components):
1006
if component_id in content_map:
1007
content = content_map[component_id]
1009
if method == 'fulltext':
1010
assert content is None
1011
content = self.factory.parse_fulltext(data, version_id)
1012
elif method == 'line-delta':
1013
delta = self.factory.parse_line_delta(data, version_id)
1014
content = content.copy()
1015
content._lines = self._apply_delta(content._lines,
1017
content_map[component_id] = content
1019
if 'no-eol' in self._index.get_options(version_id):
1020
content = content.copy()
1021
content.strip_last_line_newline()
1022
final_content[version_id] = content
1024
# digest here is the digest from the last applied component.
1025
text = content.text()
1026
if sha_strings(text) != digest:
1027
raise KnitCorrupt(self.filename,
1028
'sha-1 does not match %s' % version_id)
1030
text_map[version_id] = text
1031
return text_map, final_content
1034
def _apply_delta(lines, delta):
1035
"""Apply delta to lines."""
1038
for start, end, count, delta_lines in delta:
1039
lines[offset+start:offset+end] = delta_lines
1040
offset = offset + (start - end) + count
1043
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1045
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1046
if version_ids is None:
1047
version_ids = self.versions()
1049
pb = progress.DummyProgress()
1050
# we don't care about inclusions, the caller cares.
1051
# but we need to setup a list of records to visit.
1052
# we need version_id, position, length
1053
version_id_records = []
1054
requested_versions = set(version_ids)
1055
# filter for available versions
1056
for version_id in requested_versions:
1057
if not self.has_version(version_id):
1058
raise RevisionNotPresent(version_id, self.filename)
1059
# get a in-component-order queue:
1060
for version_id in self.versions():
1061
if version_id in requested_versions:
1062
index_memo = self._index.get_position(version_id)
1063
version_id_records.append((version_id, index_memo))
1065
total = len(version_id_records)
1066
for version_idx, (version_id, data, sha_value) in \
1067
enumerate(self._data.read_records_iter(version_id_records)):
1068
pb.update('Walking content.', version_idx, total)
1069
method = self._index.get_method(version_id)
1071
assert method in ('fulltext', 'line-delta')
1072
if method == 'fulltext':
1073
line_iterator = self.factory.get_fulltext_content(data)
1075
line_iterator = self.factory.get_linedelta_content(data)
1076
for line in line_iterator:
1079
pb.update('Walking content.', total, total)
1081
def iter_parents(self, version_ids):
1082
"""Iterate through the parents for many version ids.
1084
:param version_ids: An iterable yielding version_ids.
1085
:return: An iterator that yields (version_id, parents). Requested
1086
version_ids not present in the versioned file are simply skipped.
1087
The order is undefined, allowing for different optimisations in
1088
the underlying implementation.
1090
return self._index.iter_parents(version_ids)
1092
def num_versions(self):
1093
"""See VersionedFile.num_versions()."""
1094
return self._index.num_versions()
1096
__len__ = num_versions
1098
def annotate_iter(self, version_id):
1099
"""See VersionedFile.annotate_iter."""
1100
return self.factory.annotate_iter(self, version_id)
1102
def get_parents(self, version_id):
1103
"""See VersionedFile.get_parents."""
1106
# 52554 calls in 1264 872 internal down from 3674
1108
return self._index.get_parents(version_id)
1110
raise RevisionNotPresent(version_id, self.filename)
1112
def get_parents_with_ghosts(self, version_id):
1113
"""See VersionedFile.get_parents."""
1115
return self._index.get_parents_with_ghosts(version_id)
1117
raise RevisionNotPresent(version_id, self.filename)
1119
def get_ancestry(self, versions, topo_sorted=True):
1120
"""See VersionedFile.get_ancestry."""
1121
if isinstance(versions, basestring):
1122
versions = [versions]
1125
return self._index.get_ancestry(versions, topo_sorted)
1127
def get_ancestry_with_ghosts(self, versions):
1128
"""See VersionedFile.get_ancestry_with_ghosts."""
1129
if isinstance(versions, basestring):
1130
versions = [versions]
1133
return self._index.get_ancestry_with_ghosts(versions)
1135
def plan_merge(self, ver_a, ver_b):
1136
"""See VersionedFile.plan_merge."""
1137
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1138
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1139
annotated_a = self.annotate(ver_a)
1140
annotated_b = self.annotate(ver_b)
1141
return merge._plan_annotate_merge(annotated_a, annotated_b,
1142
ancestors_a, ancestors_b)
1145
class _KnitComponentFile(object):
1146
"""One of the files used to implement a knit database"""
1148
def __init__(self, transport, filename, mode, file_mode=None,
1149
create_parent_dir=False, dir_mode=None):
1150
self._transport = transport
1151
self._filename = filename
1153
self._file_mode = file_mode
1154
self._dir_mode = dir_mode
1155
self._create_parent_dir = create_parent_dir
1156
self._need_to_create = False
1158
def _full_path(self):
1159
"""Return the full path to this file."""
1160
return self._transport.base + self._filename
1162
def check_header(self, fp):
1163
line = fp.readline()
1165
# An empty file can actually be treated as though the file doesn't
1167
raise errors.NoSuchFile(self._full_path())
1168
if line != self.HEADER:
1169
raise KnitHeaderError(badline=line,
1170
filename=self._transport.abspath(self._filename))
1173
return '%s(%s)' % (self.__class__.__name__, self._filename)
1176
class _KnitIndex(_KnitComponentFile):
1177
"""Manages knit index file.
1179
The index is already kept in memory and read on startup, to enable
1180
fast lookups of revision information. The cursor of the index
1181
file is always pointing to the end, making it easy to append
1184
_cache is a cache for fast mapping from version id to a Index
1187
_history is a cache for fast mapping from indexes to version ids.
1189
The index data format is dictionary compressed when it comes to
1190
parent references; a index entry may only have parents that with a
1191
lover index number. As a result, the index is topological sorted.
1193
Duplicate entries may be written to the index for a single version id
1194
if this is done then the latter one completely replaces the former:
1195
this allows updates to correct version and parent information.
1196
Note that the two entries may share the delta, and that successive
1197
annotations and references MUST point to the first entry.
1199
The index file on disc contains a header, followed by one line per knit
1200
record. The same revision can be present in an index file more than once.
1201
The first occurrence gets assigned a sequence number starting from 0.
1203
The format of a single line is
1204
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1205
REVISION_ID is a utf8-encoded revision id
1206
FLAGS is a comma separated list of flags about the record. Values include
1207
no-eol, line-delta, fulltext.
1208
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1209
that the the compressed data starts at.
1210
LENGTH is the ascii representation of the length of the data file.
1211
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1213
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1214
revision id already in the knit that is a parent of REVISION_ID.
1215
The ' :' marker is the end of record marker.
1218
when a write is interrupted to the index file, it will result in a line
1219
that does not end in ' :'. If the ' :' is not present at the end of a line,
1220
or at the end of the file, then the record that is missing it will be
1221
ignored by the parser.
1223
When writing new records to the index file, the data is preceded by '\n'
1224
to ensure that records always start on new lines even if the last write was
1225
interrupted. As a result its normal for the last line in the index to be
1226
missing a trailing newline. One can be added with no harmful effects.
1229
HEADER = "# bzr knit index 8\n"
1231
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1232
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1234
def _cache_version(self, version_id, options, pos, size, parents):
1235
"""Cache a version record in the history array and index cache.
1237
This is inlined into _load_data for performance. KEEP IN SYNC.
1238
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1241
# only want the _history index to reference the 1st index entry
1243
if version_id not in self._cache:
1244
index = len(self._history)
1245
self._history.append(version_id)
1247
index = self._cache[version_id][5]
1248
self._cache[version_id] = (version_id,
1255
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1256
create_parent_dir=False, delay_create=False, dir_mode=None):
1257
_KnitComponentFile.__init__(self, transport, filename, mode,
1258
file_mode=file_mode,
1259
create_parent_dir=create_parent_dir,
1262
# position in _history is the 'official' index for a revision
1263
# but the values may have come from a newer entry.
1264
# so - wc -l of a knit index is != the number of unique names
1268
fp = self._transport.get(self._filename)
1270
# _load_data may raise NoSuchFile if the target knit is
1272
_load_data(self, fp)
1276
if mode != 'w' or not create:
1279
self._need_to_create = True
1281
self._transport.put_bytes_non_atomic(
1282
self._filename, self.HEADER, mode=self._file_mode)
1284
def get_graph(self):
1285
"""Return a list of the node:parents lists from this knit index."""
1286
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1288
def get_ancestry(self, versions, topo_sorted=True):
1289
"""See VersionedFile.get_ancestry."""
1290
# get a graph of all the mentioned versions:
1292
pending = set(versions)
1295
version = pending.pop()
1298
parents = [p for p in cache[version][4] if p in cache]
1300
raise RevisionNotPresent(version, self._filename)
1301
# if not completed and not a ghost
1302
pending.update([p for p in parents if p not in graph])
1303
graph[version] = parents
1306
return topo_sort(graph.items())
1308
def get_ancestry_with_ghosts(self, versions):
1309
"""See VersionedFile.get_ancestry_with_ghosts."""
1310
# get a graph of all the mentioned versions:
1311
self.check_versions_present(versions)
1314
pending = set(versions)
1316
version = pending.pop()
1318
parents = cache[version][4]
1324
pending.update([p for p in parents if p not in graph])
1325
graph[version] = parents
1326
return topo_sort(graph.items())
1328
def iter_parents(self, version_ids):
1329
"""Iterate through the parents for many version ids.
1331
:param version_ids: An iterable yielding version_ids.
1332
:return: An iterator that yields (version_id, parents). Requested
1333
version_ids not present in the versioned file are simply skipped.
1334
The order is undefined, allowing for different optimisations in
1335
the underlying implementation.
1337
for version_id in version_ids:
1339
yield version_id, tuple(self.get_parents(version_id))
1343
def num_versions(self):
1344
return len(self._history)
1346
__len__ = num_versions
1348
def get_versions(self):
1349
"""Get all the versions in the file. not topologically sorted."""
1350
return self._history
1352
def _version_list_to_index(self, versions):
1355
for version in versions:
1356
if version in cache:
1357
# -- inlined lookup() --
1358
result_list.append(str(cache[version][5]))
1359
# -- end lookup () --
1361
result_list.append('.' + version)
1362
return ' '.join(result_list)
1364
def add_version(self, version_id, options, index_memo, parents):
1365
"""Add a version record to the index."""
1366
self.add_versions(((version_id, options, index_memo, parents),))
1368
def add_versions(self, versions, random_id=False):
1369
"""Add multiple versions to the index.
1371
:param versions: a list of tuples:
1372
(version_id, options, pos, size, parents).
1373
:param random_id: If True the ids being added were randomly generated
1374
and no check for existence will be performed.
1377
orig_history = self._history[:]
1378
orig_cache = self._cache.copy()
1381
for version_id, options, (index, pos, size), parents in versions:
1382
line = "\n%s %s %s %s %s :" % (version_id,
1386
self._version_list_to_index(parents))
1387
assert isinstance(line, str), \
1388
'content must be utf-8 encoded: %r' % (line,)
1390
self._cache_version(version_id, options, pos, size, parents)
1391
if not self._need_to_create:
1392
self._transport.append_bytes(self._filename, ''.join(lines))
1395
sio.write(self.HEADER)
1396
sio.writelines(lines)
1398
self._transport.put_file_non_atomic(self._filename, sio,
1399
create_parent_dir=self._create_parent_dir,
1400
mode=self._file_mode,
1401
dir_mode=self._dir_mode)
1402
self._need_to_create = False
1404
# If any problems happen, restore the original values and re-raise
1405
self._history = orig_history
1406
self._cache = orig_cache
1409
def has_version(self, version_id):
1410
"""True if the version is in the index."""
1411
return version_id in self._cache
1413
def get_position(self, version_id):
1414
"""Return details needed to access the version.
1416
.kndx indices do not support split-out data, so return None for the
1419
:return: a tuple (None, data position, size) to hand to the access
1420
logic to get the record.
1422
entry = self._cache[version_id]
1423
return None, entry[2], entry[3]
1425
def get_method(self, version_id):
1426
"""Return compression method of specified version."""
1428
options = self._cache[version_id][1]
1430
raise RevisionNotPresent(version_id, self._filename)
1431
if 'fulltext' in options:
1434
if 'line-delta' not in options:
1435
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1438
def get_options(self, version_id):
1439
"""Return a string represention options.
1443
return self._cache[version_id][1]
1445
def get_parents(self, version_id):
1446
"""Return parents of specified version ignoring ghosts."""
1447
return [parent for parent in self._cache[version_id][4]
1448
if parent in self._cache]
1450
def get_parents_with_ghosts(self, version_id):
1451
"""Return parents of specified version with ghosts."""
1452
return self._cache[version_id][4]
1454
def check_versions_present(self, version_ids):
1455
"""Check that all specified versions are present."""
1457
for version_id in version_ids:
1458
if version_id not in cache:
1459
raise RevisionNotPresent(version_id, self._filename)
1462
class KnitGraphIndex(object):
1463
"""A knit index that builds on GraphIndex."""
1465
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1466
"""Construct a KnitGraphIndex on a graph_index.
1468
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1469
:param deltas: Allow delta-compressed records.
1470
:param add_callback: If not None, allow additions to the index and call
1471
this callback with a list of added GraphIndex nodes:
1472
[(node, value, node_refs), ...]
1473
:param parents: If True, record knits parents, if not do not record
1476
self._graph_index = graph_index
1477
self._deltas = deltas
1478
self._add_callback = add_callback
1479
self._parents = parents
1480
if deltas and not parents:
1481
raise KnitCorrupt(self, "Cannot do delta compression without "
1484
def _get_entries(self, keys, check_present=False):
1485
"""Get the entries for keys.
1487
:param keys: An iterable of index keys, - 1-tuples.
1492
for node in self._graph_index.iter_entries(keys):
1494
found_keys.add(node[1])
1496
# adapt parentless index to the rest of the code.
1497
for node in self._graph_index.iter_entries(keys):
1498
yield node[0], node[1], node[2], ()
1499
found_keys.add(node[1])
1501
missing_keys = keys.difference(found_keys)
1503
raise RevisionNotPresent(missing_keys.pop(), self)
1505
def _present_keys(self, version_ids):
1507
node[1] for node in self._get_entries(version_ids)])
1509
def _parentless_ancestry(self, versions):
1510
"""Honour the get_ancestry API for parentless knit indices."""
1511
wanted_keys = self._version_ids_to_keys(versions)
1512
present_keys = self._present_keys(wanted_keys)
1513
missing = set(wanted_keys).difference(present_keys)
1515
raise RevisionNotPresent(missing.pop(), self)
1516
return list(self._keys_to_version_ids(present_keys))
1518
def get_ancestry(self, versions, topo_sorted=True):
1519
"""See VersionedFile.get_ancestry."""
1520
if not self._parents:
1521
return self._parentless_ancestry(versions)
1522
# XXX: This will do len(history) index calls - perhaps
1523
# it should be altered to be a index core feature?
1524
# get a graph of all the mentioned versions:
1527
versions = self._version_ids_to_keys(versions)
1528
pending = set(versions)
1530
# get all pending nodes
1531
this_iteration = pending
1532
new_nodes = self._get_entries(this_iteration)
1535
for (index, key, value, node_refs) in new_nodes:
1536
# dont ask for ghosties - otherwise
1537
# we we can end up looping with pending
1538
# being entirely ghosted.
1539
graph[key] = [parent for parent in node_refs[0]
1540
if parent not in ghosts]
1542
for parent in graph[key]:
1543
# dont examine known nodes again
1548
ghosts.update(this_iteration.difference(found))
1549
if versions.difference(graph):
1550
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1552
result_keys = topo_sort(graph.items())
1554
result_keys = graph.iterkeys()
1555
return [key[0] for key in result_keys]
1557
def get_ancestry_with_ghosts(self, versions):
1558
"""See VersionedFile.get_ancestry."""
1559
if not self._parents:
1560
return self._parentless_ancestry(versions)
1561
# XXX: This will do len(history) index calls - perhaps
1562
# it should be altered to be a index core feature?
1563
# get a graph of all the mentioned versions:
1565
versions = self._version_ids_to_keys(versions)
1566
pending = set(versions)
1568
# get all pending nodes
1569
this_iteration = pending
1570
new_nodes = self._get_entries(this_iteration)
1572
for (index, key, value, node_refs) in new_nodes:
1573
graph[key] = node_refs[0]
1575
for parent in graph[key]:
1576
# dont examine known nodes again
1580
missing_versions = this_iteration.difference(graph)
1581
missing_needed = versions.intersection(missing_versions)
1583
raise RevisionNotPresent(missing_needed.pop(), self)
1584
for missing_version in missing_versions:
1585
# add a key, no parents
1586
graph[missing_version] = []
1587
pending.discard(missing_version) # don't look for it
1588
result_keys = topo_sort(graph.items())
1589
return [key[0] for key in result_keys]
1591
def get_graph(self):
1592
"""Return a list of the node:parents lists from this knit index."""
1593
if not self._parents:
1594
return [(key, ()) for key in self.get_versions()]
1596
for index, key, value, refs in self._graph_index.iter_all_entries():
1597
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1600
def iter_parents(self, version_ids):
1601
"""Iterate through the parents for many version ids.
1603
:param version_ids: An iterable yielding version_ids.
1604
:return: An iterator that yields (version_id, parents). Requested
1605
version_ids not present in the versioned file are simply skipped.
1606
The order is undefined, allowing for different optimisations in
1607
the underlying implementation.
1610
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1612
present_parents = set()
1613
for node in all_nodes:
1614
all_parents.update(node[3][0])
1615
# any node we are querying must be present
1616
present_parents.add(node[1])
1617
unknown_parents = all_parents.difference(present_parents)
1618
present_parents.update(self._present_keys(unknown_parents))
1619
for node in all_nodes:
1621
for parent in node[3][0]:
1622
if parent in present_parents:
1623
parents.append(parent[0])
1624
yield node[1][0], tuple(parents)
1626
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1627
yield node[1][0], ()
1629
def num_versions(self):
1630
return len(list(self._graph_index.iter_all_entries()))
1632
__len__ = num_versions
1634
def get_versions(self):
1635
"""Get all the versions in the file. not topologically sorted."""
1636
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1638
def has_version(self, version_id):
1639
"""True if the version is in the index."""
1640
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1642
def _keys_to_version_ids(self, keys):
1643
return tuple(key[0] for key in keys)
1645
def get_position(self, version_id):
1646
"""Return details needed to access the version.
1648
:return: a tuple (index, data position, size) to hand to the access
1649
logic to get the record.
1651
node = self._get_node(version_id)
1652
bits = node[2][1:].split(' ')
1653
return node[0], int(bits[0]), int(bits[1])
1655
def get_method(self, version_id):
1656
"""Return compression method of specified version."""
1657
if not self._deltas:
1659
return self._parent_compression(self._get_node(version_id)[3][1])
1661
def _parent_compression(self, reference_list):
1662
# use the second reference list to decide if this is delta'd or not.
1663
if len(reference_list):
1668
def _get_node(self, version_id):
1670
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1672
raise RevisionNotPresent(version_id, self)
1674
def get_options(self, version_id):
1675
"""Return a string represention options.
1679
node = self._get_node(version_id)
1680
if not self._deltas:
1681
options = ['fulltext']
1683
options = [self._parent_compression(node[3][1])]
1684
if node[2][0] == 'N':
1685
options.append('no-eol')
1688
def get_parents(self, version_id):
1689
"""Return parents of specified version ignoring ghosts."""
1690
parents = list(self.iter_parents([version_id]))
1693
raise errors.RevisionNotPresent(version_id, self)
1694
return parents[0][1]
1696
def get_parents_with_ghosts(self, version_id):
1697
"""Return parents of specified version with ghosts."""
1698
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1699
check_present=True))
1700
if not self._parents:
1702
return self._keys_to_version_ids(nodes[0][3][0])
1704
def check_versions_present(self, version_ids):
1705
"""Check that all specified versions are present."""
1706
keys = self._version_ids_to_keys(version_ids)
1707
present = self._present_keys(keys)
1708
missing = keys.difference(present)
1710
raise RevisionNotPresent(missing.pop(), self)
1712
def add_version(self, version_id, options, access_memo, parents):
1713
"""Add a version record to the index."""
1714
return self.add_versions(((version_id, options, access_memo, parents),))
1716
def add_versions(self, versions, random_id=False):
1717
"""Add multiple versions to the index.
1719
This function does not insert data into the Immutable GraphIndex
1720
backing the KnitGraphIndex, instead it prepares data for insertion by
1721
the caller and checks that it is safe to insert then calls
1722
self._add_callback with the prepared GraphIndex nodes.
1724
:param versions: a list of tuples:
1725
(version_id, options, pos, size, parents).
1726
:param random_id: If True the ids being added were randomly generated
1727
and no check for existence will be performed.
1729
if not self._add_callback:
1730
raise errors.ReadOnlyError(self)
1731
# we hope there are no repositories with inconsistent parentage
1736
for (version_id, options, access_memo, parents) in versions:
1737
index, pos, size = access_memo
1738
key = (version_id, )
1739
parents = tuple((parent, ) for parent in parents)
1740
if 'no-eol' in options:
1744
value += "%d %d" % (pos, size)
1745
if not self._deltas:
1746
if 'line-delta' in options:
1747
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1750
if 'line-delta' in options:
1751
node_refs = (parents, (parents[0],))
1753
node_refs = (parents, ())
1755
node_refs = (parents, )
1758
raise KnitCorrupt(self, "attempt to add node with parents "
1759
"in parentless index.")
1761
keys[key] = (value, node_refs)
1763
present_nodes = self._get_entries(keys)
1764
for (index, key, value, node_refs) in present_nodes:
1765
if (value, node_refs) != keys[key]:
1766
raise KnitCorrupt(self, "inconsistent details in add_versions"
1767
": %s %s" % ((value, node_refs), keys[key]))
1771
for key, (value, node_refs) in keys.iteritems():
1772
result.append((key, value, node_refs))
1774
for key, (value, node_refs) in keys.iteritems():
1775
result.append((key, value))
1776
self._add_callback(result)
1778
def _version_ids_to_keys(self, version_ids):
1779
return set((version_id, ) for version_id in version_ids)
1782
class _KnitAccess(object):
1783
"""Access to knit records in a .knit file."""
1785
def __init__(self, transport, filename, _file_mode, _dir_mode,
1786
_need_to_create, _create_parent_dir):
1787
"""Create a _KnitAccess for accessing and inserting data.
1789
:param transport: The transport the .knit is located on.
1790
:param filename: The filename of the .knit.
1792
self._transport = transport
1793
self._filename = filename
1794
self._file_mode = _file_mode
1795
self._dir_mode = _dir_mode
1796
self._need_to_create = _need_to_create
1797
self._create_parent_dir = _create_parent_dir
1799
def add_raw_records(self, sizes, raw_data):
1800
"""Add raw knit bytes to a storage area.
1802
The data is spooled to whereever the access method is storing data.
1804
:param sizes: An iterable containing the size of each raw data segment.
1805
:param raw_data: A bytestring containing the data.
1806
:return: A list of memos to retrieve the record later. Each memo is a
1807
tuple - (index, pos, length), where the index field is always None
1808
for the .knit access method.
1810
assert type(raw_data) == str, \
1811
'data must be plain bytes was %s' % type(raw_data)
1812
if not self._need_to_create:
1813
base = self._transport.append_bytes(self._filename, raw_data)
1815
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1816
create_parent_dir=self._create_parent_dir,
1817
mode=self._file_mode,
1818
dir_mode=self._dir_mode)
1819
self._need_to_create = False
1823
result.append((None, base, size))
1828
"""IFF this data access has its own storage area, initialise it.
1832
self._transport.put_bytes_non_atomic(self._filename, '',
1833
mode=self._file_mode)
1835
def open_file(self):
1836
"""IFF this data access can be represented as a single file, open it.
1838
For knits that are not mapped to a single file on disk this will
1841
:return: None or a file handle.
1844
return self._transport.get(self._filename)
1849
def get_raw_records(self, memos_for_retrieval):
1850
"""Get the raw bytes for a records.
1852
:param memos_for_retrieval: An iterable containing the (index, pos,
1853
length) memo for retrieving the bytes. The .knit method ignores
1854
the index as there is always only a single file.
1855
:return: An iterator over the bytes of the records.
1857
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
1858
for pos, data in self._transport.readv(self._filename, read_vector):
1862
class _PackAccess(object):
1863
"""Access to knit records via a collection of packs."""
1865
def __init__(self, index_to_packs, writer=None):
1866
"""Create a _PackAccess object.
1868
:param index_to_packs: A dict mapping index objects to the transport
1869
and file names for obtaining data.
1870
:param writer: A tuple (pack.ContainerWriter, write_index) which
1871
contains the pack to write, and the index that reads from it will
1875
self.container_writer = writer[0]
1876
self.write_index = writer[1]
1878
self.container_writer = None
1879
self.write_index = None
1880
self.indices = index_to_packs
1882
def add_raw_records(self, sizes, raw_data):
1883
"""Add raw knit bytes to a storage area.
1885
The data is spooled to the container writer in one bytes-record per
1888
:param sizes: An iterable containing the size of each raw data segment.
1889
:param raw_data: A bytestring containing the data.
1890
:return: A list of memos to retrieve the record later. Each memo is a
1891
tuple - (index, pos, length), where the index field is the
1892
write_index object supplied to the PackAccess object.
1894
assert type(raw_data) == str, \
1895
'data must be plain bytes was %s' % type(raw_data)
1899
p_offset, p_length = self.container_writer.add_bytes_record(
1900
raw_data[offset:offset+size], [])
1902
result.append((self.write_index, p_offset, p_length))
1906
"""Pack based knits do not get individually created."""
1908
def get_raw_records(self, memos_for_retrieval):
1909
"""Get the raw bytes for a records.
1911
:param memos_for_retrieval: An iterable containing the (index, pos,
1912
length) memo for retrieving the bytes. The Pack access method
1913
looks up the pack to use for a given record in its index_to_pack
1915
:return: An iterator over the bytes of the records.
1917
# first pass, group into same-index requests
1919
current_index = None
1920
for (index, offset, length) in memos_for_retrieval:
1921
if current_index == index:
1922
current_list.append((offset, length))
1924
if current_index is not None:
1925
request_lists.append((current_index, current_list))
1926
current_index = index
1927
current_list = [(offset, length)]
1928
# handle the last entry
1929
if current_index is not None:
1930
request_lists.append((current_index, current_list))
1931
for index, offsets in request_lists:
1932
transport, path = self.indices[index]
1933
reader = pack.make_readv_reader(transport, path, offsets)
1934
for names, read_func in reader.iter_records():
1935
yield read_func(None)
1937
def open_file(self):
1938
"""Pack based knits have no single file."""
1941
def set_writer(self, writer, index, (transport, packname)):
1942
"""Set a writer to use for adding data."""
1943
self.indices[index] = (transport, packname)
1944
self.container_writer = writer
1945
self.write_index = index
1948
class _KnitData(object):
1949
"""Manage extraction of data from a KnitAccess, caching and decompressing.
1951
The KnitData class provides the logic for parsing and using knit records,
1952
making use of an access method for the low level read and write operations.
1955
def __init__(self, access):
1956
"""Create a KnitData object.
1958
:param access: The access method to use. Access methods such as
1959
_KnitAccess manage the insertion of raw records and the subsequent
1960
retrieval of the same.
1962
self._access = access
1963
self._checked = False
1964
# TODO: jam 20060713 conceptually, this could spill to disk
1965
# if the cached size gets larger than a certain amount
1966
# but it complicates the model a bit, so for now just use
1967
# a simple dictionary
1969
self._do_cache = False
1971
def enable_cache(self):
1972
"""Enable caching of reads."""
1973
self._do_cache = True
1975
def clear_cache(self):
1976
"""Clear the record cache."""
1977
self._do_cache = False
1980
def _open_file(self):
1981
return self._access.open_file()
1983
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
1984
"""Convert version_id, digest, lines into a raw data block.
1986
:param dense_lines: The bytes of lines but in a denser form. For
1987
instance, if lines is a list of 1000 bytestrings each ending in \n,
1988
dense_lines may be a list with one line in it, containing all the
1989
1000's lines and their \n's. Using dense_lines if it is already
1990
known is a win because the string join to create bytes in this
1991
function spends less time resizing the final string.
1992
:return: (len, a StringIO instance with the raw data ready to read.)
1994
# Note: using a string copy here increases memory pressure with e.g.
1995
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
1996
# when doing the initial commit of a mozilla tree. RBC 20070921
1997
bytes = ''.join(chain(
1998
["version %s %d %s\n" % (version_id,
2001
dense_lines or lines,
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
