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 (
109
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
110
from bzrlib.tsort import topo_sort
113
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
116
# TODO: Split out code specific to this format into an associated object.
118
# TODO: Can we put in some kind of value to check that the index and data
119
# files belong together?
121
# TODO: accommodate binaries, perhaps by storing a byte count
123
# TODO: function to check whole file
125
# TODO: atomically append data, then measure backwards from the cursor
126
# position after writing to work out where it was located. we may need to
127
# bypass python file buffering.
129
DATA_SUFFIX = '.knit'
130
INDEX_SUFFIX = '.kndx'
133
class KnitContent(object):
134
"""Content of a knit version to which deltas can be applied."""
137
"""Return a list of (origin, text) tuples."""
138
return list(self.annotate_iter())
140
def line_delta_iter(self, new_lines):
141
"""Generate line-based delta from this content to new_lines."""
142
new_texts = new_lines.text()
143
old_texts = self.text()
144
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
145
for tag, i1, i2, j1, j2 in s.get_opcodes():
148
# ofrom, oto, length, data
149
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
151
def line_delta(self, new_lines):
152
return list(self.line_delta_iter(new_lines))
155
def get_line_delta_blocks(knit_delta, source, target):
156
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
157
target_len = len(target)
160
for s_begin, s_end, t_len, new_text in knit_delta:
161
true_n = s_begin - s_pos
164
# knit deltas do not provide reliable info about whether the
165
# last line of a file matches, due to eol handling.
166
if source[s_pos + n -1] != target[t_pos + n -1]:
169
yield s_pos, t_pos, n
170
t_pos += t_len + true_n
172
n = target_len - t_pos
174
if source[s_pos + n -1] != target[t_pos + n -1]:
177
yield s_pos, t_pos, n
178
yield s_pos + (target_len - t_pos), target_len, 0
181
class AnnotatedKnitContent(KnitContent):
182
"""Annotated content."""
184
def __init__(self, lines):
187
def annotate_iter(self):
188
"""Yield tuples of (origin, text) for each content line."""
189
return iter(self._lines)
191
def strip_last_line_newline(self):
192
line = self._lines[-1][1].rstrip('\n')
193
self._lines[-1] = (self._lines[-1][0], line)
196
return [text for origin, text in self._lines]
199
return AnnotatedKnitContent(self._lines[:])
202
class PlainKnitContent(KnitContent):
203
"""Unannotated content.
205
When annotate[_iter] is called on this content, the same version is reported
206
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
210
def __init__(self, lines, version_id):
212
self._version_id = version_id
214
def annotate_iter(self):
215
"""Yield tuples of (origin, text) for each content line."""
216
for line in self._lines:
217
yield self._version_id, line
220
return PlainKnitContent(self._lines[:], self._version_id)
222
def strip_last_line_newline(self):
223
self._lines[-1] = self._lines[-1].rstrip('\n')
229
class KnitAnnotateFactory(object):
230
"""Factory for creating annotated Content objects."""
234
def make(self, lines, version_id):
235
num_lines = len(lines)
236
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
238
def parse_fulltext(self, content, version_id):
239
"""Convert fulltext to internal representation
241
fulltext content is of the format
242
revid(utf8) plaintext\n
243
internal representation is of the format:
246
# TODO: jam 20070209 The tests expect this to be returned as tuples,
247
# but the code itself doesn't really depend on that.
248
# Figure out a way to not require the overhead of turning the
249
# list back into tuples.
250
lines = [tuple(line.split(' ', 1)) for line in content]
251
return AnnotatedKnitContent(lines)
253
def parse_line_delta_iter(self, lines):
254
return iter(self.parse_line_delta(lines))
256
def parse_line_delta(self, lines, version_id):
257
"""Convert a line based delta into internal representation.
259
line delta is in the form of:
260
intstart intend intcount
262
revid(utf8) newline\n
263
internal representation is
264
(start, end, count, [1..count tuples (revid, newline)])
271
def cache_and_return(line):
272
origin, text = line.split(' ', 1)
273
return cache.setdefault(origin, origin), text
275
# walk through the lines parsing.
277
start, end, count = [int(n) for n in header.split(',')]
278
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
279
result.append((start, end, count, contents))
282
def get_fulltext_content(self, lines):
283
"""Extract just the content lines from a fulltext."""
284
return (line.split(' ', 1)[1] for line in lines)
286
def get_linedelta_content(self, lines):
287
"""Extract just the content from a line delta.
289
This doesn't return all of the extra information stored in a delta.
290
Only the actual content lines.
295
header = header.split(',')
296
count = int(header[2])
297
for i in xrange(count):
298
origin, text = next().split(' ', 1)
301
def lower_fulltext(self, content):
302
"""convert a fulltext content record into a serializable form.
304
see parse_fulltext which this inverts.
306
# TODO: jam 20070209 We only do the caching thing to make sure that
307
# the origin is a valid utf-8 line, eventually we could remove it
308
return ['%s %s' % (o, t) for o, t in content._lines]
310
def lower_line_delta(self, delta):
311
"""convert a delta into a serializable form.
313
See parse_line_delta which this inverts.
315
# TODO: jam 20070209 We only do the caching thing to make sure that
316
# the origin is a valid utf-8 line, eventually we could remove it
318
for start, end, c, lines in delta:
319
out.append('%d,%d,%d\n' % (start, end, c))
320
out.extend(origin + ' ' + text
321
for origin, text in lines)
324
def annotate_iter(self, knit, version_id):
325
content = knit._get_content(version_id)
326
return content.annotate_iter()
329
class KnitPlainFactory(object):
330
"""Factory for creating plain Content objects."""
334
def make(self, lines, version_id):
335
return PlainKnitContent(lines, version_id)
337
def parse_fulltext(self, content, version_id):
338
"""This parses an unannotated fulltext.
340
Note that this is not a noop - the internal representation
341
has (versionid, line) - its just a constant versionid.
343
return self.make(content, version_id)
345
def parse_line_delta_iter(self, lines, version_id):
347
num_lines = len(lines)
348
while cur < num_lines:
351
start, end, c = [int(n) for n in header.split(',')]
352
yield start, end, c, lines[cur:cur+c]
355
def parse_line_delta(self, lines, version_id):
356
return list(self.parse_line_delta_iter(lines, version_id))
358
def get_fulltext_content(self, lines):
359
"""Extract just the content lines from a fulltext."""
362
def get_linedelta_content(self, lines):
363
"""Extract just the content from a line delta.
365
This doesn't return all of the extra information stored in a delta.
366
Only the actual content lines.
371
header = header.split(',')
372
count = int(header[2])
373
for i in xrange(count):
376
def lower_fulltext(self, content):
377
return content.text()
379
def lower_line_delta(self, delta):
381
for start, end, c, lines in delta:
382
out.append('%d,%d,%d\n' % (start, end, c))
386
def annotate_iter(self, knit, version_id):
387
return annotate_knit(knit, version_id)
390
def make_empty_knit(transport, relpath):
391
"""Construct a empty knit at the specified location."""
392
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
395
class KnitVersionedFile(VersionedFile):
396
"""Weave-like structure with faster random access.
398
A knit stores a number of texts and a summary of the relationships
399
between them. Texts are identified by a string version-id. Texts
400
are normally stored and retrieved as a series of lines, but can
401
also be passed as single strings.
403
Lines are stored with the trailing newline (if any) included, to
404
avoid special cases for files with no final newline. Lines are
405
composed of 8-bit characters, not unicode. The combination of
406
these approaches should mean any 'binary' file can be safely
407
stored and retrieved.
410
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
411
factory=None, basis_knit=DEPRECATED_PARAMETER, delta=True,
412
create=False, create_parent_dir=False, delay_create=False,
413
dir_mode=None, index=None, access_method=None):
414
"""Construct a knit at location specified by relpath.
416
:param create: If not True, only open an existing knit.
417
:param create_parent_dir: If True, create the parent directory if
418
creating the file fails. (This is used for stores with
419
hash-prefixes that may not exist yet)
420
:param delay_create: The calling code is aware that the knit won't
421
actually be created until the first data is stored.
422
:param index: An index to use for the knit.
424
if deprecated_passed(basis_knit):
425
warnings.warn("KnitVersionedFile.__(): The basis_knit parameter is"
426
" deprecated as of bzr 0.9.",
427
DeprecationWarning, stacklevel=2)
428
if access_mode is None:
430
super(KnitVersionedFile, self).__init__(access_mode)
431
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
432
self.transport = transport
433
self.filename = relpath
434
self.factory = factory or KnitAnnotateFactory()
435
self.writable = (access_mode == 'w')
438
self._max_delta_chain = 200
441
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
442
access_mode, create=create, file_mode=file_mode,
443
create_parent_dir=create_parent_dir, delay_create=delay_create,
447
if access_method is None:
448
_access = _KnitAccess(transport, relpath + DATA_SUFFIX, file_mode, dir_mode,
449
((create and not len(self)) and delay_create), create_parent_dir)
451
_access = access_method
452
if create and not len(self) and not delay_create:
454
self._data = _KnitData(_access)
457
return '%s(%s)' % (self.__class__.__name__,
458
self.transport.abspath(self.filename))
460
def _check_should_delta(self, first_parents):
461
"""Iterate back through the parent listing, looking for a fulltext.
463
This is used when we want to decide whether to add a delta or a new
464
fulltext. It searches for _max_delta_chain parents. When it finds a
465
fulltext parent, it sees if the total size of the deltas leading up to
466
it is large enough to indicate that we want a new full text anyway.
468
Return True if we should create a new delta, False if we should use a
473
delta_parents = first_parents
474
for count in xrange(self._max_delta_chain):
475
parent = delta_parents[0]
476
method = self._index.get_method(parent)
477
index, pos, size = self._index.get_position(parent)
478
if method == 'fulltext':
482
delta_parents = self._index.get_parents(parent)
484
# We couldn't find a fulltext, so we must create a new one
487
return fulltext_size > delta_size
489
def _add_raw_records(self, records, data):
490
"""Add all the records 'records' with data pre-joined in 'data'.
492
:param records: A list of tuples(version_id, options, parents, size).
493
:param data: The data for the records. When it is written, the records
494
are adjusted to have pos pointing into data by the sum of
495
the preceding records sizes.
498
raw_record_sizes = [record[3] for record in records]
499
positions = self._data.add_raw_records(raw_record_sizes, data)
502
for (version_id, options, parents, size), access_memo in zip(
504
index_entries.append((version_id, options, access_memo, parents))
505
if self._data._do_cache:
506
self._data._cache[version_id] = data[offset:offset+size]
508
self._index.add_versions(index_entries)
510
def enable_cache(self):
511
"""Start caching data for this knit"""
512
self._data.enable_cache()
514
def clear_cache(self):
515
"""Clear the data cache only."""
516
self._data.clear_cache()
518
def copy_to(self, name, transport):
519
"""See VersionedFile.copy_to()."""
520
# copy the current index to a temp index to avoid racing with local
522
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
523
self.transport.get(self._index._filename))
525
f = self._data._open_file()
527
transport.put_file(name + DATA_SUFFIX, f)
530
# move the copied index into place
531
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
533
def create_empty(self, name, transport, mode=None):
534
return KnitVersionedFile(name, transport, factory=self.factory,
535
delta=self.delta, create=True)
537
def get_data_stream(self, required_versions):
538
"""Get a data stream for the specified versions.
540
Versions may be returned in any order, not necessarily the order
543
:param required_versions: The exact set of versions to be extracted.
544
Unlike some other knit methods, this is not used to generate a
545
transitive closure, rather it is used precisely as given.
547
:returns: format_signature, list of (version, options, length, parents),
550
required_versions = set([osutils.safe_revision_id(v) for v in
552
# we don't care about inclusions, the caller cares.
553
# but we need to setup a list of records to visit.
554
for version_id in required_versions:
555
if not self.has_version(version_id):
556
raise RevisionNotPresent(version_id, self.filename)
557
# Pick the desired versions out of the index in oldest-to-newest order
559
for version_id in self.versions():
560
if version_id in required_versions:
561
version_list.append(version_id)
563
# create the list of version information for the result
564
copy_queue_records = []
566
result_version_list = []
567
for version_id in version_list:
568
options = self._index.get_options(version_id)
569
parents = self._index.get_parents_with_ghosts(version_id)
570
index_memo = self._index.get_position(version_id)
571
copy_queue_records.append((version_id, index_memo))
572
none, data_pos, data_size = index_memo
573
copy_set.add(version_id)
574
# version, options, length, parents
575
result_version_list.append((version_id, options, data_size,
578
# Read the compressed record data.
580
# From here down to the return should really be logic in the returned
581
# callable -- in a class that adapts read_records_iter_raw to read
584
for (version_id, raw_data), \
585
(version_id2, options, _, parents) in \
586
izip(self._data.read_records_iter_raw(copy_queue_records),
587
result_version_list):
588
assert version_id == version_id2, 'logic error, inconsistent results'
589
raw_datum.append(raw_data)
590
pseudo_file = StringIO(''.join(raw_datum))
593
return pseudo_file.read()
595
return pseudo_file.read(length)
596
return (self.get_format_signature(), result_version_list, read)
598
def _extract_blocks(self, version_id, source, target):
599
if self._index.get_method(version_id) != 'line-delta':
601
parent, sha1, noeol, delta = self.get_delta(version_id)
602
return KnitContent.get_line_delta_blocks(delta, source, target)
604
def get_delta(self, version_id):
605
"""Get a delta for constructing version from some other version."""
606
version_id = osutils.safe_revision_id(version_id)
607
self.check_not_reserved_id(version_id)
608
parents = self.get_parents(version_id)
613
index_memo = self._index.get_position(version_id)
614
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
615
noeol = 'no-eol' in self._index.get_options(version_id)
616
if 'fulltext' == self._index.get_method(version_id):
617
new_content = self.factory.parse_fulltext(data, version_id)
618
if parent is not None:
619
reference_content = self._get_content(parent)
620
old_texts = reference_content.text()
623
new_texts = new_content.text()
624
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
626
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
628
delta = self.factory.parse_line_delta(data, version_id)
629
return parent, sha1, noeol, delta
631
def get_format_signature(self):
632
"""See VersionedFile.get_format_signature()."""
633
if self.factory.annotated:
634
annotated_part = "annotated"
636
annotated_part = "plain"
637
return "knit-%s" % (annotated_part,)
639
def get_graph_with_ghosts(self):
640
"""See VersionedFile.get_graph_with_ghosts()."""
641
graph_items = self._index.get_graph()
642
return dict(graph_items)
644
def get_sha1(self, version_id):
645
return self.get_sha1s([version_id])[0]
647
def get_sha1s(self, version_ids):
648
"""See VersionedFile.get_sha1()."""
649
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
650
record_map = self._get_record_map(version_ids)
651
# record entry 2 is the 'digest'.
652
return [record_map[v][2] for v in version_ids]
656
"""See VersionedFile.get_suffixes()."""
657
return [DATA_SUFFIX, INDEX_SUFFIX]
659
def has_ghost(self, version_id):
660
"""True if there is a ghost reference in the file to version_id."""
661
version_id = osutils.safe_revision_id(version_id)
663
if self.has_version(version_id):
665
# optimisable if needed by memoising the _ghosts set.
666
items = self._index.get_graph()
667
for node, parents in items:
668
for parent in parents:
669
if parent not in self._index._cache:
670
if parent == version_id:
674
def insert_data_stream(self, (format, data_list, reader_callable)):
675
"""Insert knit records from a data stream into this knit.
677
If a version in the stream is already present in this knit, it will not
678
be inserted a second time. It will be checked for consistency with the
679
stored version however, and may cause a KnitCorrupt error to be raised
680
if the data in the stream disagrees with the already stored data.
682
:seealso: get_data_stream
684
if format != self.get_format_signature():
685
trace.mutter('incompatible format signature inserting to %r', self)
686
raise KnitDataStreamIncompatible(
687
format, self.get_format_signature())
689
for version_id, options, length, parents in data_list:
690
if self.has_version(version_id):
691
# First check: the list of parents.
692
my_parents = self.get_parents_with_ghosts(version_id)
693
if my_parents != parents:
694
# XXX: KnitCorrupt is not quite the right exception here.
697
'parents list %r from data stream does not match '
698
'already recorded parents %r for %s'
699
% (parents, my_parents, version_id))
701
# Also check the SHA-1 of the fulltext this content will
703
raw_data = reader_callable(length)
704
my_fulltext_sha1 = self.get_sha1(version_id)
705
df, rec = self._data._parse_record_header(version_id, raw_data)
706
stream_fulltext_sha1 = rec[3]
707
if my_fulltext_sha1 != stream_fulltext_sha1:
708
# Actually, we don't know if it's this knit that's corrupt,
709
# or the data stream we're trying to insert.
711
self.filename, 'sha-1 does not match %s' % version_id)
713
self._add_raw_records(
714
[(version_id, options, parents, length)],
715
reader_callable(length))
718
"""See VersionedFile.versions."""
719
if 'evil' in debug.debug_flags:
720
trace.mutter_callsite(2, "versions scales with size of history")
721
return self._index.get_versions()
723
def has_version(self, version_id):
724
"""See VersionedFile.has_version."""
725
if 'evil' in debug.debug_flags:
726
trace.mutter_callsite(2, "has_version is a LBYL scenario")
727
version_id = osutils.safe_revision_id(version_id)
728
return self._index.has_version(version_id)
730
__contains__ = has_version
732
def _merge_annotations(self, content, parents, parent_texts={},
733
delta=None, annotated=None,
734
left_matching_blocks=None):
735
"""Merge annotations for content. This is done by comparing
736
the annotations based on changed to the text.
738
if left_matching_blocks is not None:
739
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
743
for parent_id in parents:
744
merge_content = self._get_content(parent_id, parent_texts)
745
if (parent_id == parents[0] and delta_seq is not None):
748
seq = patiencediff.PatienceSequenceMatcher(
749
None, merge_content.text(), content.text())
750
for i, j, n in seq.get_matching_blocks():
753
# this appears to copy (origin, text) pairs across to the
754
# new content for any line that matches the last-checked
756
content._lines[j:j+n] = merge_content._lines[i:i+n]
758
if delta_seq is None:
759
reference_content = self._get_content(parents[0], parent_texts)
760
new_texts = content.text()
761
old_texts = reference_content.text()
762
delta_seq = patiencediff.PatienceSequenceMatcher(
763
None, old_texts, new_texts)
764
return self._make_line_delta(delta_seq, content)
766
def _make_line_delta(self, delta_seq, new_content):
767
"""Generate a line delta from delta_seq and new_content."""
769
for op in delta_seq.get_opcodes():
772
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
775
def _get_components_positions(self, version_ids):
776
"""Produce a map of position data for the components of versions.
778
This data is intended to be used for retrieving the knit records.
780
A dict of version_id to (method, data_pos, data_size, next) is
782
method is the way referenced data should be applied.
783
data_pos is the position of the data in the knit.
784
data_size is the size of the data in the knit.
785
next is the build-parent of the version, or None for fulltexts.
788
for version_id in version_ids:
791
while cursor is not None and cursor not in component_data:
792
method = self._index.get_method(cursor)
793
if method == 'fulltext':
796
next = self.get_parents(cursor)[0]
797
index_memo = self._index.get_position(cursor)
798
component_data[cursor] = (method, index_memo, next)
800
return component_data
802
def _get_content(self, version_id, parent_texts={}):
803
"""Returns a content object that makes up the specified
805
cached_version = parent_texts.get(version_id, None)
806
if cached_version is not None:
807
if not self.has_version(version_id):
808
raise RevisionNotPresent(version_id, self.filename)
809
return cached_version
811
text_map, contents_map = self._get_content_maps([version_id])
812
return contents_map[version_id]
814
def _check_versions_present(self, version_ids):
815
"""Check that all specified versions are present."""
816
self._index.check_versions_present(version_ids)
818
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
819
nostore_sha, random_id, check_content):
820
"""See VersionedFile.add_lines_with_ghosts()."""
821
self._check_add(version_id, lines, random_id, check_content)
822
return self._add(version_id, lines, parents, self.delta,
823
parent_texts, None, nostore_sha, random_id)
825
def _add_lines(self, version_id, parents, lines, parent_texts,
826
left_matching_blocks, nostore_sha, random_id, check_content):
827
"""See VersionedFile.add_lines."""
828
self._check_add(version_id, lines, random_id, check_content)
829
self._check_versions_present(parents)
830
return self._add(version_id, lines[:], parents, self.delta,
831
parent_texts, left_matching_blocks, nostore_sha, random_id)
833
def _check_add(self, version_id, lines, random_id, check_content):
834
"""check that version_id and lines are safe to add."""
835
if contains_whitespace(version_id):
836
raise InvalidRevisionId(version_id, self.filename)
837
self.check_not_reserved_id(version_id)
838
# Technically this could be avoided if we are happy to allow duplicate
839
# id insertion when other things than bzr core insert texts, but it
840
# seems useful for folk using the knit api directly to have some safety
841
# blanket that we can disable.
842
if not random_id and self.has_version(version_id):
843
raise RevisionAlreadyPresent(version_id, self.filename)
845
self._check_lines_not_unicode(lines)
846
self._check_lines_are_lines(lines)
848
def _add(self, version_id, lines, parents, delta, parent_texts,
849
left_matching_blocks, nostore_sha, random_id):
850
"""Add a set of lines on top of version specified by parents.
852
If delta is true, compress the text as a line-delta against
855
Any versions not present will be converted into ghosts.
857
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
858
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
859
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
860
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
861
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
862
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
863
# +1383 0 8.0370 8.0370 +<len>
864
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
865
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
866
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
867
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
870
if parent_texts is None:
872
for parent in parents:
873
if self.has_version(parent):
874
present_parents.append(parent)
876
# can only compress against the left most present parent.
878
(len(present_parents) == 0 or
879
present_parents[0] != parents[0])):
882
digest = sha_strings(lines)
883
if nostore_sha == digest:
884
raise errors.ExistingContent
885
text_length = sum(map(len, lines))
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)
912
options.append('fulltext')
913
store_lines = self.factory.lower_fulltext(content)
915
access_memo = self._data.add_record(version_id, digest, store_lines)
916
self._index.add_versions(
917
((version_id, options, access_memo, parents),), random_id=random_id)
918
return digest, text_length, content
920
def check(self, progress_bar=None):
921
"""See VersionedFile.check()."""
923
def _clone_text(self, new_version_id, old_version_id, parents):
924
"""See VersionedFile.clone_text()."""
925
# FIXME RBC 20060228 make fast by only inserting an index with null
927
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
929
def get_lines(self, version_id):
930
"""See VersionedFile.get_lines()."""
931
return self.get_line_list([version_id])[0]
933
def _get_record_map(self, version_ids):
934
"""Produce a dictionary of knit records.
936
The keys are version_ids, the values are tuples of (method, content,
938
method is the way the content should be applied.
939
content is a KnitContent object.
940
digest is the SHA1 digest of this version id after all steps are done
941
next is the build-parent of the version, i.e. the leftmost ancestor.
942
If the method is fulltext, next will be None.
944
position_map = self._get_components_positions(version_ids)
945
# c = component_id, m = method, i_m = index_memo, n = next
946
records = [(c, i_m) for c, (m, i_m, n) in position_map.iteritems()]
948
for component_id, content, digest in \
949
self._data.read_records_iter(records):
950
method, index_memo, next = position_map[component_id]
951
record_map[component_id] = method, content, digest, next
955
def get_text(self, version_id):
956
"""See VersionedFile.get_text"""
957
return self.get_texts([version_id])[0]
959
def get_texts(self, version_ids):
960
return [''.join(l) for l in self.get_line_list(version_ids)]
962
def get_line_list(self, version_ids):
963
"""Return the texts of listed versions as a list of strings."""
964
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
965
for version_id in version_ids:
966
self.check_not_reserved_id(version_id)
967
text_map, content_map = self._get_content_maps(version_ids)
968
return [text_map[v] for v in version_ids]
970
_get_lf_split_line_list = get_line_list
972
def _get_content_maps(self, version_ids):
973
"""Produce maps of text and KnitContents
975
:return: (text_map, content_map) where text_map contains the texts for
976
the requested versions and content_map contains the KnitContents.
977
Both dicts take version_ids as their keys.
979
for version_id in version_ids:
980
if not self.has_version(version_id):
981
raise RevisionNotPresent(version_id, self.filename)
982
record_map = self._get_record_map(version_ids)
987
for version_id in version_ids:
990
while cursor is not None:
991
method, data, digest, next = record_map[cursor]
992
components.append((cursor, method, data, digest))
993
if cursor in content_map:
998
for component_id, method, data, digest in reversed(components):
999
if component_id in content_map:
1000
content = content_map[component_id]
1002
if method == 'fulltext':
1003
assert content is None
1004
content = self.factory.parse_fulltext(data, version_id)
1005
elif method == 'line-delta':
1006
delta = self.factory.parse_line_delta(data, version_id)
1007
content = content.copy()
1008
content._lines = self._apply_delta(content._lines,
1010
content_map[component_id] = content
1012
if 'no-eol' in self._index.get_options(version_id):
1013
content = content.copy()
1014
content.strip_last_line_newline()
1015
final_content[version_id] = content
1017
# digest here is the digest from the last applied component.
1018
text = content.text()
1019
if sha_strings(text) != digest:
1020
raise KnitCorrupt(self.filename,
1021
'sha-1 does not match %s' % version_id)
1023
text_map[version_id] = text
1024
return text_map, final_content
1027
def _apply_delta(lines, delta):
1028
"""Apply delta to lines."""
1031
for start, end, count, delta_lines in delta:
1032
lines[offset+start:offset+end] = delta_lines
1033
offset = offset + (start - end) + count
1036
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1038
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1039
if version_ids is None:
1040
version_ids = self.versions()
1042
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
1044
pb = progress.DummyProgress()
1045
# we don't care about inclusions, the caller cares.
1046
# but we need to setup a list of records to visit.
1047
# we need version_id, position, length
1048
version_id_records = []
1049
requested_versions = set(version_ids)
1050
# filter for available versions
1051
for version_id in requested_versions:
1052
if not self.has_version(version_id):
1053
raise RevisionNotPresent(version_id, self.filename)
1054
# get a in-component-order queue:
1055
for version_id in self.versions():
1056
if version_id in requested_versions:
1057
index_memo = self._index.get_position(version_id)
1058
version_id_records.append((version_id, index_memo))
1060
total = len(version_id_records)
1061
for version_idx, (version_id, data, sha_value) in \
1062
enumerate(self._data.read_records_iter(version_id_records)):
1063
pb.update('Walking content.', version_idx, total)
1064
method = self._index.get_method(version_id)
1066
assert method in ('fulltext', 'line-delta')
1067
if method == 'fulltext':
1068
line_iterator = self.factory.get_fulltext_content(data)
1070
line_iterator = self.factory.get_linedelta_content(data)
1071
for line in line_iterator:
1074
pb.update('Walking content.', total, total)
1076
def iter_parents(self, version_ids):
1077
"""Iterate through the parents for many version ids.
1079
:param version_ids: An iterable yielding version_ids.
1080
:return: An iterator that yields (version_id, parents). Requested
1081
version_ids not present in the versioned file are simply skipped.
1082
The order is undefined, allowing for different optimisations in
1083
the underlying implementation.
1085
version_ids = [osutils.safe_revision_id(version_id) for
1086
version_id in version_ids]
1087
return self._index.iter_parents(version_ids)
1089
def num_versions(self):
1090
"""See VersionedFile.num_versions()."""
1091
return self._index.num_versions()
1093
__len__ = num_versions
1095
def annotate_iter(self, version_id):
1096
"""See VersionedFile.annotate_iter."""
1097
version_id = osutils.safe_revision_id(version_id)
1098
return self.factory.annotate_iter(self, version_id)
1100
def get_parents(self, version_id):
1101
"""See VersionedFile.get_parents."""
1104
# 52554 calls in 1264 872 internal down from 3674
1105
version_id = osutils.safe_revision_id(version_id)
1107
return self._index.get_parents(version_id)
1109
raise RevisionNotPresent(version_id, self.filename)
1111
def get_parents_with_ghosts(self, version_id):
1112
"""See VersionedFile.get_parents."""
1113
version_id = osutils.safe_revision_id(version_id)
1115
return self._index.get_parents_with_ghosts(version_id)
1117
raise RevisionNotPresent(version_id, self.filename)
1119
def get_ancestry(self, versions, topo_sorted=True):
1120
"""See VersionedFile.get_ancestry."""
1121
if isinstance(versions, basestring):
1122
versions = [versions]
1125
versions = [osutils.safe_revision_id(v) for v in versions]
1126
return self._index.get_ancestry(versions, topo_sorted)
1128
def get_ancestry_with_ghosts(self, versions):
1129
"""See VersionedFile.get_ancestry_with_ghosts."""
1130
if isinstance(versions, basestring):
1131
versions = [versions]
1134
versions = [osutils.safe_revision_id(v) for v in versions]
1135
return self._index.get_ancestry_with_ghosts(versions)
1137
def plan_merge(self, ver_a, ver_b):
1138
"""See VersionedFile.plan_merge."""
1139
ver_a = osutils.safe_revision_id(ver_a)
1140
ver_b = osutils.safe_revision_id(ver_b)
1141
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1143
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1144
annotated_a = self.annotate(ver_a)
1145
annotated_b = self.annotate(ver_b)
1146
return merge._plan_annotate_merge(annotated_a, annotated_b,
1147
ancestors_a, ancestors_b)
1150
class _KnitComponentFile(object):
1151
"""One of the files used to implement a knit database"""
1153
def __init__(self, transport, filename, mode, file_mode=None,
1154
create_parent_dir=False, dir_mode=None):
1155
self._transport = transport
1156
self._filename = filename
1158
self._file_mode = file_mode
1159
self._dir_mode = dir_mode
1160
self._create_parent_dir = create_parent_dir
1161
self._need_to_create = False
1163
def _full_path(self):
1164
"""Return the full path to this file."""
1165
return self._transport.base + self._filename
1167
def check_header(self, fp):
1168
line = fp.readline()
1170
# An empty file can actually be treated as though the file doesn't
1172
raise errors.NoSuchFile(self._full_path())
1173
if line != self.HEADER:
1174
raise KnitHeaderError(badline=line,
1175
filename=self._transport.abspath(self._filename))
1178
return '%s(%s)' % (self.__class__.__name__, self._filename)
1181
class _KnitIndex(_KnitComponentFile):
1182
"""Manages knit index file.
1184
The index is already kept in memory and read on startup, to enable
1185
fast lookups of revision information. The cursor of the index
1186
file is always pointing to the end, making it easy to append
1189
_cache is a cache for fast mapping from version id to a Index
1192
_history is a cache for fast mapping from indexes to version ids.
1194
The index data format is dictionary compressed when it comes to
1195
parent references; a index entry may only have parents that with a
1196
lover index number. As a result, the index is topological sorted.
1198
Duplicate entries may be written to the index for a single version id
1199
if this is done then the latter one completely replaces the former:
1200
this allows updates to correct version and parent information.
1201
Note that the two entries may share the delta, and that successive
1202
annotations and references MUST point to the first entry.
1204
The index file on disc contains a header, followed by one line per knit
1205
record. The same revision can be present in an index file more than once.
1206
The first occurrence gets assigned a sequence number starting from 0.
1208
The format of a single line is
1209
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1210
REVISION_ID is a utf8-encoded revision id
1211
FLAGS is a comma separated list of flags about the record. Values include
1212
no-eol, line-delta, fulltext.
1213
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1214
that the the compressed data starts at.
1215
LENGTH is the ascii representation of the length of the data file.
1216
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1218
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1219
revision id already in the knit that is a parent of REVISION_ID.
1220
The ' :' marker is the end of record marker.
1223
when a write is interrupted to the index file, it will result in a line
1224
that does not end in ' :'. If the ' :' is not present at the end of a line,
1225
or at the end of the file, then the record that is missing it will be
1226
ignored by the parser.
1228
When writing new records to the index file, the data is preceded by '\n'
1229
to ensure that records always start on new lines even if the last write was
1230
interrupted. As a result its normal for the last line in the index to be
1231
missing a trailing newline. One can be added with no harmful effects.
1234
HEADER = "# bzr knit index 8\n"
1236
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1237
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1239
def _cache_version(self, version_id, options, pos, size, parents):
1240
"""Cache a version record in the history array and index cache.
1242
This is inlined into _load_data for performance. KEEP IN SYNC.
1243
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1246
# only want the _history index to reference the 1st index entry
1248
if version_id not in self._cache:
1249
index = len(self._history)
1250
self._history.append(version_id)
1252
index = self._cache[version_id][5]
1253
self._cache[version_id] = (version_id,
1260
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1261
create_parent_dir=False, delay_create=False, dir_mode=None):
1262
_KnitComponentFile.__init__(self, transport, filename, mode,
1263
file_mode=file_mode,
1264
create_parent_dir=create_parent_dir,
1267
# position in _history is the 'official' index for a revision
1268
# but the values may have come from a newer entry.
1269
# so - wc -l of a knit index is != the number of unique names
1273
fp = self._transport.get(self._filename)
1275
# _load_data may raise NoSuchFile if the target knit is
1277
_load_data(self, fp)
1281
if mode != 'w' or not create:
1284
self._need_to_create = True
1286
self._transport.put_bytes_non_atomic(
1287
self._filename, self.HEADER, mode=self._file_mode)
1289
def get_graph(self):
1290
"""Return a list of the node:parents lists from this knit index."""
1291
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1293
def get_ancestry(self, versions, topo_sorted=True):
1294
"""See VersionedFile.get_ancestry."""
1295
# get a graph of all the mentioned versions:
1297
pending = set(versions)
1300
version = pending.pop()
1303
parents = [p for p in cache[version][4] if p in cache]
1305
raise RevisionNotPresent(version, self._filename)
1306
# if not completed and not a ghost
1307
pending.update([p for p in parents if p not in graph])
1308
graph[version] = parents
1311
return topo_sort(graph.items())
1313
def get_ancestry_with_ghosts(self, versions):
1314
"""See VersionedFile.get_ancestry_with_ghosts."""
1315
# get a graph of all the mentioned versions:
1316
self.check_versions_present(versions)
1319
pending = set(versions)
1321
version = pending.pop()
1323
parents = cache[version][4]
1329
pending.update([p for p in parents if p not in graph])
1330
graph[version] = parents
1331
return topo_sort(graph.items())
1333
def iter_parents(self, version_ids):
1334
"""Iterate through the parents for many version ids.
1336
:param version_ids: An iterable yielding version_ids.
1337
:return: An iterator that yields (version_id, parents). Requested
1338
version_ids not present in the versioned file are simply skipped.
1339
The order is undefined, allowing for different optimisations in
1340
the underlying implementation.
1342
for version_id in version_ids:
1344
yield version_id, tuple(self.get_parents(version_id))
1348
def num_versions(self):
1349
return len(self._history)
1351
__len__ = num_versions
1353
def get_versions(self):
1354
"""Get all the versions in the file. not topologically sorted."""
1355
return self._history
1357
def _version_list_to_index(self, versions):
1360
for version in versions:
1361
if version in cache:
1362
# -- inlined lookup() --
1363
result_list.append(str(cache[version][5]))
1364
# -- end lookup () --
1366
result_list.append('.' + version)
1367
return ' '.join(result_list)
1369
def add_version(self, version_id, options, index_memo, parents):
1370
"""Add a version record to the index."""
1371
self.add_versions(((version_id, options, index_memo, parents),))
1373
def add_versions(self, versions, random_id=False):
1374
"""Add multiple versions to the index.
1376
:param versions: a list of tuples:
1377
(version_id, options, pos, size, parents).
1378
:param random_id: If True the ids being added were randomly generated
1379
and no check for existence will be performed.
1382
orig_history = self._history[:]
1383
orig_cache = self._cache.copy()
1386
for version_id, options, (index, pos, size), parents in versions:
1387
line = "\n%s %s %s %s %s :" % (version_id,
1391
self._version_list_to_index(parents))
1392
assert isinstance(line, str), \
1393
'content must be utf-8 encoded: %r' % (line,)
1395
self._cache_version(version_id, options, pos, size, parents)
1396
if not self._need_to_create:
1397
self._transport.append_bytes(self._filename, ''.join(lines))
1400
sio.write(self.HEADER)
1401
sio.writelines(lines)
1403
self._transport.put_file_non_atomic(self._filename, sio,
1404
create_parent_dir=self._create_parent_dir,
1405
mode=self._file_mode,
1406
dir_mode=self._dir_mode)
1407
self._need_to_create = False
1409
# If any problems happen, restore the original values and re-raise
1410
self._history = orig_history
1411
self._cache = orig_cache
1414
def has_version(self, version_id):
1415
"""True if the version is in the index."""
1416
return version_id in self._cache
1418
def get_position(self, version_id):
1419
"""Return details needed to access the version.
1421
.kndx indices do not support split-out data, so return None for the
1424
:return: a tuple (None, data position, size) to hand to the access
1425
logic to get the record.
1427
entry = self._cache[version_id]
1428
return None, entry[2], entry[3]
1430
def get_method(self, version_id):
1431
"""Return compression method of specified version."""
1432
options = self._cache[version_id][1]
1433
if 'fulltext' in options:
1436
if 'line-delta' not in options:
1437
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1440
def get_options(self, version_id):
1441
"""Return a string represention options.
1445
return self._cache[version_id][1]
1447
def get_parents(self, version_id):
1448
"""Return parents of specified version ignoring ghosts."""
1449
return [parent for parent in self._cache[version_id][4]
1450
if parent in self._cache]
1452
def get_parents_with_ghosts(self, version_id):
1453
"""Return parents of specified version with ghosts."""
1454
return self._cache[version_id][4]
1456
def check_versions_present(self, version_ids):
1457
"""Check that all specified versions are present."""
1459
for version_id in version_ids:
1460
if version_id not in cache:
1461
raise RevisionNotPresent(version_id, self._filename)
1464
class KnitGraphIndex(object):
1465
"""A knit index that builds on GraphIndex."""
1467
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1468
"""Construct a KnitGraphIndex on a graph_index.
1470
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1471
:param deltas: Allow delta-compressed records.
1472
:param add_callback: If not None, allow additions to the index and call
1473
this callback with a list of added GraphIndex nodes:
1474
[(node, value, node_refs), ...]
1475
:param parents: If True, record knits parents, if not do not record
1478
self._graph_index = graph_index
1479
self._deltas = deltas
1480
self._add_callback = add_callback
1481
self._parents = parents
1482
if deltas and not parents:
1483
raise KnitCorrupt(self, "Cannot do delta compression without "
1486
def _get_entries(self, keys, check_present=False):
1487
"""Get the entries for keys.
1489
:param keys: An iterable of index keys, - 1-tuples.
1494
for node in self._graph_index.iter_entries(keys):
1496
found_keys.add(node[1])
1498
# adapt parentless index to the rest of the code.
1499
for node in self._graph_index.iter_entries(keys):
1500
yield node[0], node[1], node[2], ()
1501
found_keys.add(node[1])
1503
missing_keys = keys.difference(found_keys)
1505
raise RevisionNotPresent(missing_keys.pop(), self)
1507
def _present_keys(self, version_ids):
1509
node[1] for node in self._get_entries(version_ids)])
1511
def _parentless_ancestry(self, versions):
1512
"""Honour the get_ancestry API for parentless knit indices."""
1513
wanted_keys = self._version_ids_to_keys(versions)
1514
present_keys = self._present_keys(wanted_keys)
1515
missing = set(wanted_keys).difference(present_keys)
1517
raise RevisionNotPresent(missing.pop(), self)
1518
return list(self._keys_to_version_ids(present_keys))
1520
def get_ancestry(self, versions, topo_sorted=True):
1521
"""See VersionedFile.get_ancestry."""
1522
if not self._parents:
1523
return self._parentless_ancestry(versions)
1524
# XXX: This will do len(history) index calls - perhaps
1525
# it should be altered to be a index core feature?
1526
# get a graph of all the mentioned versions:
1529
versions = self._version_ids_to_keys(versions)
1530
pending = set(versions)
1532
# get all pending nodes
1533
this_iteration = pending
1534
new_nodes = self._get_entries(this_iteration)
1537
for (index, key, value, node_refs) in new_nodes:
1538
# dont ask for ghosties - otherwise
1539
# we we can end up looping with pending
1540
# being entirely ghosted.
1541
graph[key] = [parent for parent in node_refs[0]
1542
if parent not in ghosts]
1544
for parent in graph[key]:
1545
# dont examine known nodes again
1550
ghosts.update(this_iteration.difference(found))
1551
if versions.difference(graph):
1552
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1554
result_keys = topo_sort(graph.items())
1556
result_keys = graph.iterkeys()
1557
return [key[0] for key in result_keys]
1559
def get_ancestry_with_ghosts(self, versions):
1560
"""See VersionedFile.get_ancestry."""
1561
if not self._parents:
1562
return self._parentless_ancestry(versions)
1563
# XXX: This will do len(history) index calls - perhaps
1564
# it should be altered to be a index core feature?
1565
# get a graph of all the mentioned versions:
1567
versions = self._version_ids_to_keys(versions)
1568
pending = set(versions)
1570
# get all pending nodes
1571
this_iteration = pending
1572
new_nodes = self._get_entries(this_iteration)
1574
for (index, key, value, node_refs) in new_nodes:
1575
graph[key] = node_refs[0]
1577
for parent in graph[key]:
1578
# dont examine known nodes again
1582
missing_versions = this_iteration.difference(graph)
1583
missing_needed = versions.intersection(missing_versions)
1585
raise RevisionNotPresent(missing_needed.pop(), self)
1586
for missing_version in missing_versions:
1587
# add a key, no parents
1588
graph[missing_version] = []
1589
pending.discard(missing_version) # don't look for it
1590
result_keys = topo_sort(graph.items())
1591
return [key[0] for key in result_keys]
1593
def get_graph(self):
1594
"""Return a list of the node:parents lists from this knit index."""
1595
if not self._parents:
1596
return [(key, ()) for key in self.get_versions()]
1598
for index, key, value, refs in self._graph_index.iter_all_entries():
1599
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1602
def iter_parents(self, version_ids):
1603
"""Iterate through the parents for many version ids.
1605
:param version_ids: An iterable yielding version_ids.
1606
:return: An iterator that yields (version_id, parents). Requested
1607
version_ids not present in the versioned file are simply skipped.
1608
The order is undefined, allowing for different optimisations in
1609
the underlying implementation.
1612
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1614
present_parents = set()
1615
for node in all_nodes:
1616
all_parents.update(node[3][0])
1617
# any node we are querying must be present
1618
present_parents.add(node[1])
1619
unknown_parents = all_parents.difference(present_parents)
1620
present_parents.update(self._present_keys(unknown_parents))
1621
for node in all_nodes:
1623
for parent in node[3][0]:
1624
if parent in present_parents:
1625
parents.append(parent[0])
1626
yield node[1][0], tuple(parents)
1628
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1629
yield node[1][0], ()
1631
def num_versions(self):
1632
return len(list(self._graph_index.iter_all_entries()))
1634
__len__ = num_versions
1636
def get_versions(self):
1637
"""Get all the versions in the file. not topologically sorted."""
1638
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1640
def has_version(self, version_id):
1641
"""True if the version is in the index."""
1642
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1644
def _keys_to_version_ids(self, keys):
1645
return tuple(key[0] for key in keys)
1647
def get_position(self, version_id):
1648
"""Return details needed to access the version.
1650
:return: a tuple (index, data position, size) to hand to the access
1651
logic to get the record.
1653
node = self._get_node(version_id)
1654
bits = node[2][1:].split(' ')
1655
return node[0], int(bits[0]), int(bits[1])
1657
def get_method(self, version_id):
1658
"""Return compression method of specified version."""
1659
if not self._deltas:
1661
return self._parent_compression(self._get_node(version_id)[3][1])
1663
def _parent_compression(self, reference_list):
1664
# use the second reference list to decide if this is delta'd or not.
1665
if len(reference_list):
1670
def _get_node(self, version_id):
1671
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1673
def get_options(self, version_id):
1674
"""Return a string represention options.
1678
node = self._get_node(version_id)
1679
if not self._deltas:
1680
options = ['fulltext']
1682
options = [self._parent_compression(node[3][1])]
1683
if node[2][0] == 'N':
1684
options.append('no-eol')
1687
def get_parents(self, version_id):
1688
"""Return parents of specified version ignoring ghosts."""
1689
parents = list(self.iter_parents([version_id]))
1692
raise errors.RevisionNotPresent(version_id, self)
1693
return parents[0][1]
1695
def get_parents_with_ghosts(self, version_id):
1696
"""Return parents of specified version with ghosts."""
1697
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1698
check_present=True))
1699
if not self._parents:
1701
return self._keys_to_version_ids(nodes[0][3][0])
1703
def check_versions_present(self, version_ids):
1704
"""Check that all specified versions are present."""
1705
keys = self._version_ids_to_keys(version_ids)
1706
present = self._present_keys(keys)
1707
missing = keys.difference(present)
1709
raise RevisionNotPresent(missing.pop(), self)
1711
def add_version(self, version_id, options, access_memo, parents):
1712
"""Add a version record to the index."""
1713
return self.add_versions(((version_id, options, access_memo, parents),))
1715
def add_versions(self, versions, random_id=False):
1716
"""Add multiple versions to the index.
1718
This function does not insert data into the Immutable GraphIndex
1719
backing the KnitGraphIndex, instead it prepares data for insertion by
1720
the caller and checks that it is safe to insert then calls
1721
self._add_callback with the prepared GraphIndex nodes.
1723
:param versions: a list of tuples:
1724
(version_id, options, pos, size, parents).
1725
:param random_id: If True the ids being added were randomly generated
1726
and no check for existence will be performed.
1728
if not self._add_callback:
1729
raise errors.ReadOnlyError(self)
1730
# we hope there are no repositories with inconsistent parentage
1735
for (version_id, options, access_memo, parents) in versions:
1736
index, pos, size = access_memo
1737
key = (version_id, )
1738
parents = tuple((parent, ) for parent in parents)
1739
if 'no-eol' in options:
1743
value += "%d %d" % (pos, size)
1744
if not self._deltas:
1745
if 'line-delta' in options:
1746
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1749
if 'line-delta' in options:
1750
node_refs = (parents, (parents[0],))
1752
node_refs = (parents, ())
1754
node_refs = (parents, )
1757
raise KnitCorrupt(self, "attempt to add node with parents "
1758
"in parentless index.")
1760
keys[key] = (value, node_refs)
1762
present_nodes = self._get_entries(keys)
1763
for (index, key, value, node_refs) in present_nodes:
1764
if (value, node_refs) != keys[key]:
1765
raise KnitCorrupt(self, "inconsistent details in add_versions"
1766
": %s %s" % ((value, node_refs), keys[key]))
1770
for key, (value, node_refs) in keys.iteritems():
1771
result.append((key, value, node_refs))
1773
for key, (value, node_refs) in keys.iteritems():
1774
result.append((key, value))
1775
self._add_callback(result)
1777
def _version_ids_to_keys(self, version_ids):
1778
return set((version_id, ) for version_id in version_ids)
1781
class _KnitAccess(object):
1782
"""Access to knit records in a .knit file."""
1784
def __init__(self, transport, filename, _file_mode, _dir_mode,
1785
_need_to_create, _create_parent_dir):
1786
"""Create a _KnitAccess for accessing and inserting data.
1788
:param transport: The transport the .knit is located on.
1789
:param filename: The filename of the .knit.
1791
self._transport = transport
1792
self._filename = filename
1793
self._file_mode = _file_mode
1794
self._dir_mode = _dir_mode
1795
self._need_to_create = _need_to_create
1796
self._create_parent_dir = _create_parent_dir
1798
def add_raw_records(self, sizes, raw_data):
1799
"""Add raw knit bytes to a storage area.
1801
The data is spooled to whereever the access method is storing data.
1803
:param sizes: An iterable containing the size of each raw data segment.
1804
:param raw_data: A bytestring containing the data.
1805
:return: A list of memos to retrieve the record later. Each memo is a
1806
tuple - (index, pos, length), where the index field is always None
1807
for the .knit access method.
1809
assert type(raw_data) == str, \
1810
'data must be plain bytes was %s' % type(raw_data)
1811
if not self._need_to_create:
1812
base = self._transport.append_bytes(self._filename, raw_data)
1814
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1815
create_parent_dir=self._create_parent_dir,
1816
mode=self._file_mode,
1817
dir_mode=self._dir_mode)
1818
self._need_to_create = False
1822
result.append((None, base, size))
1827
"""IFF this data access has its own storage area, initialise it.
1831
self._transport.put_bytes_non_atomic(self._filename, '',
1832
mode=self._file_mode)
1834
def open_file(self):
1835
"""IFF this data access can be represented as a single file, open it.
1837
For knits that are not mapped to a single file on disk this will
1840
:return: None or a file handle.
1843
return self._transport.get(self._filename)
1848
def get_raw_records(self, memos_for_retrieval):
1849
"""Get the raw bytes for a records.
1851
:param memos_for_retrieval: An iterable containing the (index, pos,
1852
length) memo for retrieving the bytes. The .knit method ignores
1853
the index as there is always only a single file.
1854
:return: An iterator over the bytes of the records.
1856
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
1857
for pos, data in self._transport.readv(self._filename, read_vector):
1861
class _PackAccess(object):
1862
"""Access to knit records via a collection of packs."""
1864
def __init__(self, index_to_packs, writer=None):
1865
"""Create a _PackAccess object.
1867
:param index_to_packs: A dict mapping index objects to the transport
1868
and file names for obtaining data.
1869
:param writer: A tuple (pack.ContainerWriter, write_index) which
1870
contains the pack to write, and the index that reads from it will
1874
self.container_writer = writer[0]
1875
self.write_index = writer[1]
1877
self.container_writer = None
1878
self.write_index = None
1879
self.indices = index_to_packs
1881
def add_raw_records(self, sizes, raw_data):
1882
"""Add raw knit bytes to a storage area.
1884
The data is spooled to the container writer in one bytes-record per
1887
:param sizes: An iterable containing the size of each raw data segment.
1888
:param raw_data: A bytestring containing the data.
1889
:return: A list of memos to retrieve the record later. Each memo is a
1890
tuple - (index, pos, length), where the index field is the
1891
write_index object supplied to the PackAccess object.
1893
assert type(raw_data) == str, \
1894
'data must be plain bytes was %s' % type(raw_data)
1898
p_offset, p_length = self.container_writer.add_bytes_record(
1899
raw_data[offset:offset+size], [])
1901
result.append((self.write_index, p_offset, p_length))
1905
"""Pack based knits do not get individually created."""
1907
def get_raw_records(self, memos_for_retrieval):
1908
"""Get the raw bytes for a records.
1910
:param memos_for_retrieval: An iterable containing the (index, pos,
1911
length) memo for retrieving the bytes. The Pack access method
1912
looks up the pack to use for a given record in its index_to_pack
1914
:return: An iterator over the bytes of the records.
1916
# first pass, group into same-index requests
1918
current_index = None
1919
for (index, offset, length) in memos_for_retrieval:
1920
if current_index == index:
1921
current_list.append((offset, length))
1923
if current_index is not None:
1924
request_lists.append((current_index, current_list))
1925
current_index = index
1926
current_list = [(offset, length)]
1927
# handle the last entry
1928
if current_index is not None:
1929
request_lists.append((current_index, current_list))
1930
for index, offsets in request_lists:
1931
transport, path = self.indices[index]
1932
reader = pack.make_readv_reader(transport, path, offsets)
1933
for names, read_func in reader.iter_records():
1934
yield read_func(None)
1936
def open_file(self):
1937
"""Pack based knits have no single file."""
1940
def set_writer(self, writer, index, (transport, packname)):
1941
"""Set a writer to use for adding data."""
1942
self.indices[index] = (transport, packname)
1943
self.container_writer = writer
1944
self.write_index = index
1947
class _KnitData(object):
1948
"""Manage extraction of data from a KnitAccess, caching and decompressing.
1950
The KnitData class provides the logic for parsing and using knit records,
1951
making use of an access method for the low level read and write operations.
1954
def __init__(self, access):
1955
"""Create a KnitData object.
1957
:param access: The access method to use. Access methods such as
1958
_KnitAccess manage the insertion of raw records and the subsequent
1959
retrieval of the same.
1961
self._access = access
1962
self._checked = False
1963
# TODO: jam 20060713 conceptually, this could spill to disk
1964
# if the cached size gets larger than a certain amount
1965
# but it complicates the model a bit, so for now just use
1966
# a simple dictionary
1968
self._do_cache = False
1970
def enable_cache(self):
1971
"""Enable caching of reads."""
1972
self._do_cache = True
1974
def clear_cache(self):
1975
"""Clear the record cache."""
1976
self._do_cache = False
1979
def _open_file(self):
1980
return self._access.open_file()
1982
def _record_to_data(self, version_id, digest, lines):
1983
"""Convert version_id, digest, lines into a raw data block.
1985
:return: (len, a StringIO instance with the raw data ready to read.)
1987
bytes = (''.join(chain(
1988
["version %s %d %s\n" % (version_id,
1992
["end %s\n" % version_id])))
1993
assert bytes.__class__ == str
1994
compressed_bytes = bytes_to_gzip(bytes)
1995
return len(compressed_bytes), compressed_bytes
1997
def add_raw_records(self, sizes, raw_data):
1998
"""Append a prepared record to the data file.
2000
:param sizes: An iterable containing the size of each raw data segment.
2001
:param raw_data: A bytestring containing the data.
2002
:return: a list of index data for the way the data was stored.
2003
See the access method add_raw_records documentation for more
2006
return self._access.add_raw_records(sizes, raw_data)
2008
def add_record(self, version_id, digest, lines):
2009
"""Write new text record to disk.
2011
Returns index data for retrieving it later, as per add_raw_records.
2013
size, bytes = self._record_to_data(version_id, digest, lines)
2014
result = self.add_raw_records([size], bytes)
2016
self._cache[version_id] = bytes
2019
def _parse_record_header(self, version_id, raw_data):
2020
"""Parse a record header for consistency.
2022
:return: the header and the decompressor stream.
2023
as (stream, header_record)
2025
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2027
rec = self._check_header(version_id, df.readline())
2028
except Exception, e:
2029
raise KnitCorrupt(self._access,
2030
"While reading {%s} got %s(%s)"
2031
% (version_id, e.__class__.__name__, str(e)))
2034
def _check_header(self, version_id, line):
2037
raise KnitCorrupt(self._access,
2038
'unexpected number of elements in record header')
2039
if rec[1] != version_id:
2040
raise KnitCorrupt(self._access,
2041
'unexpected version, wanted %r, got %r'
2042
% (version_id, rec[1]))
2045
def _parse_record(self, version_id, data):
2047
# 4168 calls in 2880 217 internal
2048
# 4168 calls to _parse_record_header in 2121
2049
# 4168 calls to readlines in 330
2050
df = GzipFile(mode='rb', fileobj=StringIO(data))
2053
record_contents = df.readlines()
2054
except Exception, e:
2055
raise KnitCorrupt(self._access,
2056
"While reading {%s} got %s(%s)"
2057
% (version_id, e.__class__.__name__, str(e)))
2058
header = record_contents.pop(0)
2059
rec = self._check_header(version_id, header)
2061
last_line = record_contents.pop()
2062
if len(record_contents) != int(rec[2]):
2063
raise KnitCorrupt(self._access,
2064
'incorrect number of lines %s != %s'
2066
% (len(record_contents), int(rec[2]),
2068
if last_line != 'end %s\n' % rec[1]:
2069
raise KnitCorrupt(self._access,
2070
'unexpected version end line %r, wanted %r'
2071
% (last_line, version_id))
2073
return record_contents, rec[3]
2075
def read_records_iter_raw(self, records):
2076
"""Read text records from data file and yield raw data.
2078
This unpacks enough of the text record to validate the id is
2079
as expected but thats all.
2081
# setup an iterator of the external records:
2082
# uses readv so nice and fast we hope.
2084
# grab the disk data needed.
2086
# Don't check _cache if it is empty
2087
needed_offsets = [index_memo for version_id, index_memo
2089
if version_id not in self._cache]
2091
needed_offsets = [index_memo for version_id, index_memo
2094
raw_records = self._access.get_raw_records(needed_offsets)
2096
for version_id, index_memo in records:
2097
if version_id in self._cache:
2098
# This data has already been validated
2099
data = self._cache[version_id]
2101
data = raw_records.next()
2103
self._cache[version_id] = data
2105
# validate the header
2106
df, rec = self._parse_record_header(version_id, data)
2108
yield version_id, data
2110
def read_records_iter(self, records):
2111
"""Read text records from data file and yield result.
2113
The result will be returned in whatever is the fastest to read.
2114
Not by the order requested. Also, multiple requests for the same
2115
record will only yield 1 response.
2116
:param records: A list of (version_id, pos, len) entries
2117
:return: Yields (version_id, contents, digest) in the order
2118
read, not the order requested
2124
# Skip records we have alread seen
2125
yielded_records = set()
2126
needed_records = set()
2127
for record in records:
2128
if record[0] in self._cache:
2129
if record[0] in yielded_records:
2131
yielded_records.add(record[0])
2132
data = self._cache[record[0]]
2133
content, digest = self._parse_record(record[0], data)
2134
yield (record[0], content, digest)
2136
needed_records.add(record)
2137
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2139
needed_records = sorted(set(records), key=operator.itemgetter(1))
2141
if not needed_records:
2144
# The transport optimizes the fetching as well
2145
# (ie, reads continuous ranges.)
2146
raw_data = self._access.get_raw_records(
2147
[index_memo for version_id, index_memo in needed_records])
2149
for (version_id, index_memo), data in \
2150
izip(iter(needed_records), raw_data):
2151
content, digest = self._parse_record(version_id, data)
2153
self._cache[version_id] = data
2154
yield version_id, content, digest
2156
def read_records(self, records):
2157
"""Read records into a dictionary."""
2159
for record_id, content, digest in \
2160
self.read_records_iter(records):
2161
components[record_id] = (content, digest)
2165
class InterKnit(InterVersionedFile):
2166
"""Optimised code paths for knit to knit operations."""
2168
_matching_file_from_factory = KnitVersionedFile
2169
_matching_file_to_factory = KnitVersionedFile
2172
def is_compatible(source, target):
2173
"""Be compatible with knits. """
2175
return (isinstance(source, KnitVersionedFile) and
2176
isinstance(target, KnitVersionedFile))
2177
except AttributeError:
2180
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2181
"""See InterVersionedFile.join."""
2182
assert isinstance(self.source, KnitVersionedFile)
2183
assert isinstance(self.target, KnitVersionedFile)
2185
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2190
pb = ui.ui_factory.nested_progress_bar()
2192
version_ids = list(version_ids)
2193
if None in version_ids:
2194
version_ids.remove(None)
2196
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2197
this_versions = set(self.target._index.get_versions())
2198
# XXX: For efficiency we should not look at the whole index,
2199
# we only need to consider the referenced revisions - they
2200
# must all be present, or the method must be full-text.
2201
# TODO, RBC 20070919
2202
needed_versions = self.source_ancestry - this_versions
2204
if not needed_versions:
2206
full_list = topo_sort(self.source.get_graph())
2208
version_list = [i for i in full_list if (not self.target.has_version(i)
2209
and i in needed_versions)]
2213
copy_queue_records = []
2215
for version_id in version_list:
2216
options = self.source._index.get_options(version_id)
2217
parents = self.source._index.get_parents_with_ghosts(version_id)
2218
# check that its will be a consistent copy:
2219
for parent in parents:
2220
# if source has the parent, we must :
2221
# * already have it or
2222
# * have it scheduled already
2223
# otherwise we don't care
2224
assert (self.target.has_version(parent) or
2225
parent in copy_set or
2226
not self.source.has_version(parent))
2227
index_memo = self.source._index.get_position(version_id)
2228
copy_queue_records.append((version_id, index_memo))
2229
copy_queue.append((version_id, options, parents))
2230
copy_set.add(version_id)
2232
# data suck the join:
2234
total = len(version_list)
2237
for (version_id, raw_data), \
2238
(version_id2, options, parents) in \
2239
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2241
assert version_id == version_id2, 'logic error, inconsistent results'
2243
pb.update("Joining knit", count, total)
2244
raw_records.append((version_id, options, parents, len(raw_data)))
2245
raw_datum.append(raw_data)
2246
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2252
InterVersionedFile.register_optimiser(InterKnit)
2255
class WeaveToKnit(InterVersionedFile):
2256
"""Optimised code paths for weave to knit operations."""
2258
_matching_file_from_factory = bzrlib.weave.WeaveFile
2259
_matching_file_to_factory = KnitVersionedFile
2262
def is_compatible(source, target):
2263
"""Be compatible with weaves to knits."""
2265
return (isinstance(source, bzrlib.weave.Weave) and
2266
isinstance(target, KnitVersionedFile))
2267
except AttributeError:
2270
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2271
"""See InterVersionedFile.join."""
2272
assert isinstance(self.source, bzrlib.weave.Weave)
2273
assert isinstance(self.target, KnitVersionedFile)
2275
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2280
pb = ui.ui_factory.nested_progress_bar()
2282
version_ids = list(version_ids)
2284
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2285
this_versions = set(self.target._index.get_versions())
2286
needed_versions = self.source_ancestry - this_versions
2288
if not needed_versions:
2290
full_list = topo_sort(self.source.get_graph())
2292
version_list = [i for i in full_list if (not self.target.has_version(i)
2293
and i in needed_versions)]
2297
total = len(version_list)
2298
for version_id in version_list:
2299
pb.update("Converting to knit", count, total)
2300
parents = self.source.get_parents(version_id)
2301
# check that its will be a consistent copy:
2302
for parent in parents:
2303
# if source has the parent, we must already have it
2304
assert (self.target.has_version(parent))
2305
self.target.add_lines(
2306
version_id, parents, self.source.get_lines(version_id))
2313
InterVersionedFile.register_optimiser(WeaveToKnit)
2316
# Deprecated, use PatienceSequenceMatcher instead
2317
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2320
def annotate_knit(knit, revision_id):
2321
"""Annotate a knit with no cached annotations.
2323
This implementation is for knits with no cached annotations.
2324
It will work for knits with cached annotations, but this is not
2327
ancestry = knit.get_ancestry(revision_id)
2328
fulltext = dict(zip(ancestry, knit.get_line_list(ancestry)))
2330
for candidate in ancestry:
2331
if candidate in annotations:
2333
parents = knit.get_parents(candidate)
2334
if len(parents) == 0:
2336
elif knit._index.get_method(candidate) != 'line-delta':
2339
parent, sha1, noeol, delta = knit.get_delta(candidate)
2340
blocks = KnitContent.get_line_delta_blocks(delta,
2341
fulltext[parents[0]], fulltext[candidate])
2342
annotations[candidate] = list(annotate.reannotate([annotations[p]
2343
for p in parents], fulltext[candidate], candidate, blocks))
2344
return iter(annotations[revision_id])
2348
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2350
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
