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
if 'line-delta' in options:
719
# Make sure that this knit record is actually useful: a
720
# line-delta is no use unless we have its parent.
721
# Fetching from a broken repository with this problem
722
# shouldn't break the target repository.
723
if not self._index.has_version(parents[0]):
726
'line-delta from stream references '
727
'missing parent %s' % parents[0])
728
self._add_raw_records(
729
[(version_id, options, parents, length)],
730
reader_callable(length))
733
"""See VersionedFile.versions."""
734
if 'evil' in debug.debug_flags:
735
trace.mutter_callsite(2, "versions scales with size of history")
736
return self._index.get_versions()
738
def has_version(self, version_id):
739
"""See VersionedFile.has_version."""
740
if 'evil' in debug.debug_flags:
741
trace.mutter_callsite(2, "has_version is a LBYL scenario")
742
return self._index.has_version(version_id)
744
__contains__ = has_version
746
def _merge_annotations(self, content, parents, parent_texts={},
747
delta=None, annotated=None,
748
left_matching_blocks=None):
749
"""Merge annotations for content. This is done by comparing
750
the annotations based on changed to the text.
752
if left_matching_blocks is not None:
753
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
757
for parent_id in parents:
758
merge_content = self._get_content(parent_id, parent_texts)
759
if (parent_id == parents[0] and delta_seq is not None):
762
seq = patiencediff.PatienceSequenceMatcher(
763
None, merge_content.text(), content.text())
764
for i, j, n in seq.get_matching_blocks():
767
# this appears to copy (origin, text) pairs across to the
768
# new content for any line that matches the last-checked
770
content._lines[j:j+n] = merge_content._lines[i:i+n]
772
if delta_seq is None:
773
reference_content = self._get_content(parents[0], parent_texts)
774
new_texts = content.text()
775
old_texts = reference_content.text()
776
delta_seq = patiencediff.PatienceSequenceMatcher(
777
None, old_texts, new_texts)
778
return self._make_line_delta(delta_seq, content)
780
def _make_line_delta(self, delta_seq, new_content):
781
"""Generate a line delta from delta_seq and new_content."""
783
for op in delta_seq.get_opcodes():
786
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
789
def _get_components_positions(self, version_ids):
790
"""Produce a map of position data for the components of versions.
792
This data is intended to be used for retrieving the knit records.
794
A dict of version_id to (method, data_pos, data_size, next) is
796
method is the way referenced data should be applied.
797
data_pos is the position of the data in the knit.
798
data_size is the size of the data in the knit.
799
next is the build-parent of the version, or None for fulltexts.
802
for version_id in version_ids:
805
while cursor is not None and cursor not in component_data:
806
method = self._index.get_method(cursor)
807
if method == 'fulltext':
810
next = self.get_parents(cursor)[0]
811
index_memo = self._index.get_position(cursor)
812
component_data[cursor] = (method, index_memo, next)
814
return component_data
816
def _get_content(self, version_id, parent_texts={}):
817
"""Returns a content object that makes up the specified
819
cached_version = parent_texts.get(version_id, None)
820
if cached_version is not None:
821
if not self.has_version(version_id):
822
raise RevisionNotPresent(version_id, self.filename)
823
return cached_version
825
text_map, contents_map = self._get_content_maps([version_id])
826
return contents_map[version_id]
828
def _check_versions_present(self, version_ids):
829
"""Check that all specified versions are present."""
830
self._index.check_versions_present(version_ids)
832
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
833
nostore_sha, random_id, check_content):
834
"""See VersionedFile.add_lines_with_ghosts()."""
835
self._check_add(version_id, lines, random_id, check_content)
836
return self._add(version_id, lines, parents, self.delta,
837
parent_texts, None, nostore_sha, random_id)
839
def _add_lines(self, version_id, parents, lines, parent_texts,
840
left_matching_blocks, nostore_sha, random_id, check_content):
841
"""See VersionedFile.add_lines."""
842
self._check_add(version_id, lines, random_id, check_content)
843
self._check_versions_present(parents)
844
return self._add(version_id, lines[:], parents, self.delta,
845
parent_texts, left_matching_blocks, nostore_sha, random_id)
847
def _check_add(self, version_id, lines, random_id, check_content):
848
"""check that version_id and lines are safe to add."""
849
if contains_whitespace(version_id):
850
raise InvalidRevisionId(version_id, self.filename)
851
self.check_not_reserved_id(version_id)
852
# Technically this could be avoided if we are happy to allow duplicate
853
# id insertion when other things than bzr core insert texts, but it
854
# seems useful for folk using the knit api directly to have some safety
855
# blanket that we can disable.
856
if not random_id and self.has_version(version_id):
857
raise RevisionAlreadyPresent(version_id, self.filename)
859
self._check_lines_not_unicode(lines)
860
self._check_lines_are_lines(lines)
862
def _add(self, version_id, lines, parents, delta, parent_texts,
863
left_matching_blocks, nostore_sha, random_id):
864
"""Add a set of lines on top of version specified by parents.
866
If delta is true, compress the text as a line-delta against
869
Any versions not present will be converted into ghosts.
871
# first thing, if the content is something we don't need to store, find
873
line_bytes = ''.join(lines)
874
digest = sha_string(line_bytes)
875
if nostore_sha == digest:
876
raise errors.ExistingContent
879
if parent_texts is None:
881
for parent in parents:
882
if self.has_version(parent):
883
present_parents.append(parent)
885
# can only compress against the left most present parent.
887
(len(present_parents) == 0 or
888
present_parents[0] != parents[0])):
891
text_length = len(line_bytes)
894
if lines[-1][-1] != '\n':
895
# copy the contents of lines.
897
options.append('no-eol')
898
lines[-1] = lines[-1] + '\n'
902
# To speed the extract of texts the delta chain is limited
903
# to a fixed number of deltas. This should minimize both
904
# I/O and the time spend applying deltas.
905
delta = self._check_should_delta(present_parents)
907
assert isinstance(version_id, str)
908
content = self.factory.make(lines, version_id)
909
if delta or (self.factory.annotated and len(present_parents) > 0):
910
# Merge annotations from parent texts if needed.
911
delta_hunks = self._merge_annotations(content, present_parents,
912
parent_texts, delta, self.factory.annotated,
913
left_matching_blocks)
916
options.append('line-delta')
917
store_lines = self.factory.lower_line_delta(delta_hunks)
918
size, bytes = self._data._record_to_data(version_id, digest,
921
options.append('fulltext')
922
# isinstance is slower and we have no hierarchy.
923
if self.factory.__class__ == KnitPlainFactory:
924
# Use the already joined bytes saving iteration time in
926
size, bytes = self._data._record_to_data(version_id, digest,
929
# get mixed annotation + content and feed it into the
931
store_lines = self.factory.lower_fulltext(content)
932
size, bytes = self._data._record_to_data(version_id, digest,
935
access_memo = self._data.add_raw_records([size], bytes)[0]
936
self._index.add_versions(
937
((version_id, options, access_memo, parents),),
939
return digest, text_length, content
941
def check(self, progress_bar=None):
942
"""See VersionedFile.check()."""
944
def _clone_text(self, new_version_id, old_version_id, parents):
945
"""See VersionedFile.clone_text()."""
946
# FIXME RBC 20060228 make fast by only inserting an index with null
948
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
950
def get_lines(self, version_id):
951
"""See VersionedFile.get_lines()."""
952
return self.get_line_list([version_id])[0]
954
def _get_record_map(self, version_ids):
955
"""Produce a dictionary of knit records.
957
The keys are version_ids, the values are tuples of (method, content,
959
method is the way the content should be applied.
960
content is a KnitContent object.
961
digest is the SHA1 digest of this version id after all steps are done
962
next is the build-parent of the version, i.e. the leftmost ancestor.
963
If the method is fulltext, next will be None.
965
position_map = self._get_components_positions(version_ids)
966
# c = component_id, m = method, i_m = index_memo, n = next
967
records = [(c, i_m) for c, (m, i_m, n) in position_map.iteritems()]
969
for component_id, content, digest in \
970
self._data.read_records_iter(records):
971
method, index_memo, next = position_map[component_id]
972
record_map[component_id] = method, content, digest, next
976
def get_text(self, version_id):
977
"""See VersionedFile.get_text"""
978
return self.get_texts([version_id])[0]
980
def get_texts(self, version_ids):
981
return [''.join(l) for l in self.get_line_list(version_ids)]
983
def get_line_list(self, version_ids):
984
"""Return the texts of listed versions as a list of strings."""
985
for version_id in version_ids:
986
self.check_not_reserved_id(version_id)
987
text_map, content_map = self._get_content_maps(version_ids)
988
return [text_map[v] for v in version_ids]
990
_get_lf_split_line_list = get_line_list
992
def _get_content_maps(self, version_ids):
993
"""Produce maps of text and KnitContents
995
:return: (text_map, content_map) where text_map contains the texts for
996
the requested versions and content_map contains the KnitContents.
997
Both dicts take version_ids as their keys.
999
record_map = self._get_record_map(version_ids)
1004
for version_id in version_ids:
1007
while cursor is not None:
1008
method, data, digest, next = record_map[cursor]
1009
components.append((cursor, method, data, digest))
1010
if cursor in content_map:
1015
for component_id, method, data, digest in reversed(components):
1016
if component_id in content_map:
1017
content = content_map[component_id]
1019
if method == 'fulltext':
1020
assert content is None
1021
content = self.factory.parse_fulltext(data, version_id)
1022
elif method == 'line-delta':
1023
delta = self.factory.parse_line_delta(data, version_id)
1024
content = content.copy()
1025
content._lines = self._apply_delta(content._lines,
1027
content_map[component_id] = content
1029
if 'no-eol' in self._index.get_options(version_id):
1030
content = content.copy()
1031
content.strip_last_line_newline()
1032
final_content[version_id] = content
1034
# digest here is the digest from the last applied component.
1035
text = content.text()
1036
if sha_strings(text) != digest:
1037
raise KnitCorrupt(self.filename,
1038
'sha-1 does not match %s' % version_id)
1040
text_map[version_id] = text
1041
return text_map, final_content
1044
def _apply_delta(lines, delta):
1045
"""Apply delta to lines."""
1048
for start, end, count, delta_lines in delta:
1049
lines[offset+start:offset+end] = delta_lines
1050
offset = offset + (start - end) + count
1053
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1055
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1056
if version_ids is None:
1057
version_ids = self.versions()
1059
pb = progress.DummyProgress()
1060
# we don't care about inclusions, the caller cares.
1061
# but we need to setup a list of records to visit.
1062
# we need version_id, position, length
1063
version_id_records = []
1064
requested_versions = set(version_ids)
1065
# filter for available versions
1066
for version_id in requested_versions:
1067
if not self.has_version(version_id):
1068
raise RevisionNotPresent(version_id, self.filename)
1069
# get a in-component-order queue:
1070
for version_id in self.versions():
1071
if version_id in requested_versions:
1072
index_memo = self._index.get_position(version_id)
1073
version_id_records.append((version_id, index_memo))
1075
total = len(version_id_records)
1076
for version_idx, (version_id, data, sha_value) in \
1077
enumerate(self._data.read_records_iter(version_id_records)):
1078
pb.update('Walking content.', version_idx, total)
1079
method = self._index.get_method(version_id)
1081
assert method in ('fulltext', 'line-delta')
1082
if method == 'fulltext':
1083
line_iterator = self.factory.get_fulltext_content(data)
1085
line_iterator = self.factory.get_linedelta_content(data)
1086
for line in line_iterator:
1089
pb.update('Walking content.', total, total)
1091
def iter_parents(self, version_ids):
1092
"""Iterate through the parents for many version ids.
1094
:param version_ids: An iterable yielding version_ids.
1095
:return: An iterator that yields (version_id, parents). Requested
1096
version_ids not present in the versioned file are simply skipped.
1097
The order is undefined, allowing for different optimisations in
1098
the underlying implementation.
1100
return self._index.iter_parents(version_ids)
1102
def num_versions(self):
1103
"""See VersionedFile.num_versions()."""
1104
return self._index.num_versions()
1106
__len__ = num_versions
1108
def annotate_iter(self, version_id):
1109
"""See VersionedFile.annotate_iter."""
1110
return self.factory.annotate_iter(self, version_id)
1112
def get_parents(self, version_id):
1113
"""See VersionedFile.get_parents."""
1116
# 52554 calls in 1264 872 internal down from 3674
1118
return self._index.get_parents(version_id)
1120
raise RevisionNotPresent(version_id, self.filename)
1122
def get_parents_with_ghosts(self, version_id):
1123
"""See VersionedFile.get_parents."""
1125
return self._index.get_parents_with_ghosts(version_id)
1127
raise RevisionNotPresent(version_id, self.filename)
1129
def get_ancestry(self, versions, topo_sorted=True):
1130
"""See VersionedFile.get_ancestry."""
1131
if isinstance(versions, basestring):
1132
versions = [versions]
1135
return self._index.get_ancestry(versions, topo_sorted)
1137
def get_ancestry_with_ghosts(self, versions):
1138
"""See VersionedFile.get_ancestry_with_ghosts."""
1139
if isinstance(versions, basestring):
1140
versions = [versions]
1143
return self._index.get_ancestry_with_ghosts(versions)
1145
def plan_merge(self, ver_a, ver_b):
1146
"""See VersionedFile.plan_merge."""
1147
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1148
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1149
annotated_a = self.annotate(ver_a)
1150
annotated_b = self.annotate(ver_b)
1151
return merge._plan_annotate_merge(annotated_a, annotated_b,
1152
ancestors_a, ancestors_b)
1155
class _KnitComponentFile(object):
1156
"""One of the files used to implement a knit database"""
1158
def __init__(self, transport, filename, mode, file_mode=None,
1159
create_parent_dir=False, dir_mode=None):
1160
self._transport = transport
1161
self._filename = filename
1163
self._file_mode = file_mode
1164
self._dir_mode = dir_mode
1165
self._create_parent_dir = create_parent_dir
1166
self._need_to_create = False
1168
def _full_path(self):
1169
"""Return the full path to this file."""
1170
return self._transport.base + self._filename
1172
def check_header(self, fp):
1173
line = fp.readline()
1175
# An empty file can actually be treated as though the file doesn't
1177
raise errors.NoSuchFile(self._full_path())
1178
if line != self.HEADER:
1179
raise KnitHeaderError(badline=line,
1180
filename=self._transport.abspath(self._filename))
1183
return '%s(%s)' % (self.__class__.__name__, self._filename)
1186
class _KnitIndex(_KnitComponentFile):
1187
"""Manages knit index file.
1189
The index is already kept in memory and read on startup, to enable
1190
fast lookups of revision information. The cursor of the index
1191
file is always pointing to the end, making it easy to append
1194
_cache is a cache for fast mapping from version id to a Index
1197
_history is a cache for fast mapping from indexes to version ids.
1199
The index data format is dictionary compressed when it comes to
1200
parent references; a index entry may only have parents that with a
1201
lover index number. As a result, the index is topological sorted.
1203
Duplicate entries may be written to the index for a single version id
1204
if this is done then the latter one completely replaces the former:
1205
this allows updates to correct version and parent information.
1206
Note that the two entries may share the delta, and that successive
1207
annotations and references MUST point to the first entry.
1209
The index file on disc contains a header, followed by one line per knit
1210
record. The same revision can be present in an index file more than once.
1211
The first occurrence gets assigned a sequence number starting from 0.
1213
The format of a single line is
1214
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1215
REVISION_ID is a utf8-encoded revision id
1216
FLAGS is a comma separated list of flags about the record. Values include
1217
no-eol, line-delta, fulltext.
1218
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1219
that the the compressed data starts at.
1220
LENGTH is the ascii representation of the length of the data file.
1221
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1223
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1224
revision id already in the knit that is a parent of REVISION_ID.
1225
The ' :' marker is the end of record marker.
1228
when a write is interrupted to the index file, it will result in a line
1229
that does not end in ' :'. If the ' :' is not present at the end of a line,
1230
or at the end of the file, then the record that is missing it will be
1231
ignored by the parser.
1233
When writing new records to the index file, the data is preceded by '\n'
1234
to ensure that records always start on new lines even if the last write was
1235
interrupted. As a result its normal for the last line in the index to be
1236
missing a trailing newline. One can be added with no harmful effects.
1239
HEADER = "# bzr knit index 8\n"
1241
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1242
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1244
def _cache_version(self, version_id, options, pos, size, parents):
1245
"""Cache a version record in the history array and index cache.
1247
This is inlined into _load_data for performance. KEEP IN SYNC.
1248
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1251
# only want the _history index to reference the 1st index entry
1253
if version_id not in self._cache:
1254
index = len(self._history)
1255
self._history.append(version_id)
1257
index = self._cache[version_id][5]
1258
self._cache[version_id] = (version_id,
1265
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1266
create_parent_dir=False, delay_create=False, dir_mode=None):
1267
_KnitComponentFile.__init__(self, transport, filename, mode,
1268
file_mode=file_mode,
1269
create_parent_dir=create_parent_dir,
1272
# position in _history is the 'official' index for a revision
1273
# but the values may have come from a newer entry.
1274
# so - wc -l of a knit index is != the number of unique names
1278
fp = self._transport.get(self._filename)
1280
# _load_data may raise NoSuchFile if the target knit is
1282
_load_data(self, fp)
1286
if mode != 'w' or not create:
1289
self._need_to_create = True
1291
self._transport.put_bytes_non_atomic(
1292
self._filename, self.HEADER, mode=self._file_mode)
1294
def get_graph(self):
1295
"""Return a list of the node:parents lists from this knit index."""
1296
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1298
def get_ancestry(self, versions, topo_sorted=True):
1299
"""See VersionedFile.get_ancestry."""
1300
# get a graph of all the mentioned versions:
1302
pending = set(versions)
1305
version = pending.pop()
1308
parents = [p for p in cache[version][4] if p in cache]
1310
raise RevisionNotPresent(version, self._filename)
1311
# if not completed and not a ghost
1312
pending.update([p for p in parents if p not in graph])
1313
graph[version] = parents
1316
return topo_sort(graph.items())
1318
def get_ancestry_with_ghosts(self, versions):
1319
"""See VersionedFile.get_ancestry_with_ghosts."""
1320
# get a graph of all the mentioned versions:
1321
self.check_versions_present(versions)
1324
pending = set(versions)
1326
version = pending.pop()
1328
parents = cache[version][4]
1334
pending.update([p for p in parents if p not in graph])
1335
graph[version] = parents
1336
return topo_sort(graph.items())
1338
def iter_parents(self, version_ids):
1339
"""Iterate through the parents for many version ids.
1341
:param version_ids: An iterable yielding version_ids.
1342
:return: An iterator that yields (version_id, parents). Requested
1343
version_ids not present in the versioned file are simply skipped.
1344
The order is undefined, allowing for different optimisations in
1345
the underlying implementation.
1347
for version_id in version_ids:
1349
yield version_id, tuple(self.get_parents(version_id))
1353
def num_versions(self):
1354
return len(self._history)
1356
__len__ = num_versions
1358
def get_versions(self):
1359
"""Get all the versions in the file. not topologically sorted."""
1360
return self._history
1362
def _version_list_to_index(self, versions):
1365
for version in versions:
1366
if version in cache:
1367
# -- inlined lookup() --
1368
result_list.append(str(cache[version][5]))
1369
# -- end lookup () --
1371
result_list.append('.' + version)
1372
return ' '.join(result_list)
1374
def add_version(self, version_id, options, index_memo, parents):
1375
"""Add a version record to the index."""
1376
self.add_versions(((version_id, options, index_memo, parents),))
1378
def add_versions(self, versions, random_id=False):
1379
"""Add multiple versions to the index.
1381
:param versions: a list of tuples:
1382
(version_id, options, pos, size, parents).
1383
:param random_id: If True the ids being added were randomly generated
1384
and no check for existence will be performed.
1387
orig_history = self._history[:]
1388
orig_cache = self._cache.copy()
1391
for version_id, options, (index, pos, size), parents in versions:
1392
line = "\n%s %s %s %s %s :" % (version_id,
1396
self._version_list_to_index(parents))
1397
assert isinstance(line, str), \
1398
'content must be utf-8 encoded: %r' % (line,)
1400
self._cache_version(version_id, options, pos, size, parents)
1401
if not self._need_to_create:
1402
self._transport.append_bytes(self._filename, ''.join(lines))
1405
sio.write(self.HEADER)
1406
sio.writelines(lines)
1408
self._transport.put_file_non_atomic(self._filename, sio,
1409
create_parent_dir=self._create_parent_dir,
1410
mode=self._file_mode,
1411
dir_mode=self._dir_mode)
1412
self._need_to_create = False
1414
# If any problems happen, restore the original values and re-raise
1415
self._history = orig_history
1416
self._cache = orig_cache
1419
def has_version(self, version_id):
1420
"""True if the version is in the index."""
1421
return version_id in self._cache
1423
def get_position(self, version_id):
1424
"""Return details needed to access the version.
1426
.kndx indices do not support split-out data, so return None for the
1429
:return: a tuple (None, data position, size) to hand to the access
1430
logic to get the record.
1432
entry = self._cache[version_id]
1433
return None, entry[2], entry[3]
1435
def get_method(self, version_id):
1436
"""Return compression method of specified version."""
1438
options = self._cache[version_id][1]
1440
raise RevisionNotPresent(version_id, self._filename)
1441
if 'fulltext' in options:
1444
if 'line-delta' not in options:
1445
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1448
def get_options(self, version_id):
1449
"""Return a string represention options.
1453
return self._cache[version_id][1]
1455
def get_parents(self, version_id):
1456
"""Return parents of specified version ignoring ghosts."""
1457
return [parent for parent in self._cache[version_id][4]
1458
if parent in self._cache]
1460
def get_parents_with_ghosts(self, version_id):
1461
"""Return parents of specified version with ghosts."""
1462
return self._cache[version_id][4]
1464
def check_versions_present(self, version_ids):
1465
"""Check that all specified versions are present."""
1467
for version_id in version_ids:
1468
if version_id not in cache:
1469
raise RevisionNotPresent(version_id, self._filename)
1472
class KnitGraphIndex(object):
1473
"""A knit index that builds on GraphIndex."""
1475
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1476
"""Construct a KnitGraphIndex on a graph_index.
1478
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1479
:param deltas: Allow delta-compressed records.
1480
:param add_callback: If not None, allow additions to the index and call
1481
this callback with a list of added GraphIndex nodes:
1482
[(node, value, node_refs), ...]
1483
:param parents: If True, record knits parents, if not do not record
1486
self._graph_index = graph_index
1487
self._deltas = deltas
1488
self._add_callback = add_callback
1489
self._parents = parents
1490
if deltas and not parents:
1491
raise KnitCorrupt(self, "Cannot do delta compression without "
1494
def _get_entries(self, keys, check_present=False):
1495
"""Get the entries for keys.
1497
:param keys: An iterable of index keys, - 1-tuples.
1502
for node in self._graph_index.iter_entries(keys):
1504
found_keys.add(node[1])
1506
# adapt parentless index to the rest of the code.
1507
for node in self._graph_index.iter_entries(keys):
1508
yield node[0], node[1], node[2], ()
1509
found_keys.add(node[1])
1511
missing_keys = keys.difference(found_keys)
1513
raise RevisionNotPresent(missing_keys.pop(), self)
1515
def _present_keys(self, version_ids):
1517
node[1] for node in self._get_entries(version_ids)])
1519
def _parentless_ancestry(self, versions):
1520
"""Honour the get_ancestry API for parentless knit indices."""
1521
wanted_keys = self._version_ids_to_keys(versions)
1522
present_keys = self._present_keys(wanted_keys)
1523
missing = set(wanted_keys).difference(present_keys)
1525
raise RevisionNotPresent(missing.pop(), self)
1526
return list(self._keys_to_version_ids(present_keys))
1528
def get_ancestry(self, versions, topo_sorted=True):
1529
"""See VersionedFile.get_ancestry."""
1530
if not self._parents:
1531
return self._parentless_ancestry(versions)
1532
# XXX: This will do len(history) index calls - perhaps
1533
# it should be altered to be a index core feature?
1534
# get a graph of all the mentioned versions:
1537
versions = self._version_ids_to_keys(versions)
1538
pending = set(versions)
1540
# get all pending nodes
1541
this_iteration = pending
1542
new_nodes = self._get_entries(this_iteration)
1545
for (index, key, value, node_refs) in new_nodes:
1546
# dont ask for ghosties - otherwise
1547
# we we can end up looping with pending
1548
# being entirely ghosted.
1549
graph[key] = [parent for parent in node_refs[0]
1550
if parent not in ghosts]
1552
for parent in graph[key]:
1553
# dont examine known nodes again
1558
ghosts.update(this_iteration.difference(found))
1559
if versions.difference(graph):
1560
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1562
result_keys = topo_sort(graph.items())
1564
result_keys = graph.iterkeys()
1565
return [key[0] for key in result_keys]
1567
def get_ancestry_with_ghosts(self, versions):
1568
"""See VersionedFile.get_ancestry."""
1569
if not self._parents:
1570
return self._parentless_ancestry(versions)
1571
# XXX: This will do len(history) index calls - perhaps
1572
# it should be altered to be a index core feature?
1573
# get a graph of all the mentioned versions:
1575
versions = self._version_ids_to_keys(versions)
1576
pending = set(versions)
1578
# get all pending nodes
1579
this_iteration = pending
1580
new_nodes = self._get_entries(this_iteration)
1582
for (index, key, value, node_refs) in new_nodes:
1583
graph[key] = node_refs[0]
1585
for parent in graph[key]:
1586
# dont examine known nodes again
1590
missing_versions = this_iteration.difference(graph)
1591
missing_needed = versions.intersection(missing_versions)
1593
raise RevisionNotPresent(missing_needed.pop(), self)
1594
for missing_version in missing_versions:
1595
# add a key, no parents
1596
graph[missing_version] = []
1597
pending.discard(missing_version) # don't look for it
1598
result_keys = topo_sort(graph.items())
1599
return [key[0] for key in result_keys]
1601
def get_graph(self):
1602
"""Return a list of the node:parents lists from this knit index."""
1603
if not self._parents:
1604
return [(key, ()) for key in self.get_versions()]
1606
for index, key, value, refs in self._graph_index.iter_all_entries():
1607
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1610
def iter_parents(self, version_ids):
1611
"""Iterate through the parents for many version ids.
1613
:param version_ids: An iterable yielding version_ids.
1614
:return: An iterator that yields (version_id, parents). Requested
1615
version_ids not present in the versioned file are simply skipped.
1616
The order is undefined, allowing for different optimisations in
1617
the underlying implementation.
1620
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1622
present_parents = set()
1623
for node in all_nodes:
1624
all_parents.update(node[3][0])
1625
# any node we are querying must be present
1626
present_parents.add(node[1])
1627
unknown_parents = all_parents.difference(present_parents)
1628
present_parents.update(self._present_keys(unknown_parents))
1629
for node in all_nodes:
1631
for parent in node[3][0]:
1632
if parent in present_parents:
1633
parents.append(parent[0])
1634
yield node[1][0], tuple(parents)
1636
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1637
yield node[1][0], ()
1639
def num_versions(self):
1640
return len(list(self._graph_index.iter_all_entries()))
1642
__len__ = num_versions
1644
def get_versions(self):
1645
"""Get all the versions in the file. not topologically sorted."""
1646
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1648
def has_version(self, version_id):
1649
"""True if the version is in the index."""
1650
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1652
def _keys_to_version_ids(self, keys):
1653
return tuple(key[0] for key in keys)
1655
def get_position(self, version_id):
1656
"""Return details needed to access the version.
1658
:return: a tuple (index, data position, size) to hand to the access
1659
logic to get the record.
1661
node = self._get_node(version_id)
1662
bits = node[2][1:].split(' ')
1663
return node[0], int(bits[0]), int(bits[1])
1665
def get_method(self, version_id):
1666
"""Return compression method of specified version."""
1667
if not self._deltas:
1669
return self._parent_compression(self._get_node(version_id)[3][1])
1671
def _parent_compression(self, reference_list):
1672
# use the second reference list to decide if this is delta'd or not.
1673
if len(reference_list):
1678
def _get_node(self, version_id):
1680
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1682
raise RevisionNotPresent(version_id, self)
1684
def get_options(self, version_id):
1685
"""Return a string represention options.
1689
node = self._get_node(version_id)
1690
if not self._deltas:
1691
options = ['fulltext']
1693
options = [self._parent_compression(node[3][1])]
1694
if node[2][0] == 'N':
1695
options.append('no-eol')
1698
def get_parents(self, version_id):
1699
"""Return parents of specified version ignoring ghosts."""
1700
parents = list(self.iter_parents([version_id]))
1703
raise errors.RevisionNotPresent(version_id, self)
1704
return parents[0][1]
1706
def get_parents_with_ghosts(self, version_id):
1707
"""Return parents of specified version with ghosts."""
1708
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1709
check_present=True))
1710
if not self._parents:
1712
return self._keys_to_version_ids(nodes[0][3][0])
1714
def check_versions_present(self, version_ids):
1715
"""Check that all specified versions are present."""
1716
keys = self._version_ids_to_keys(version_ids)
1717
present = self._present_keys(keys)
1718
missing = keys.difference(present)
1720
raise RevisionNotPresent(missing.pop(), self)
1722
def add_version(self, version_id, options, access_memo, parents):
1723
"""Add a version record to the index."""
1724
return self.add_versions(((version_id, options, access_memo, parents),))
1726
def add_versions(self, versions, random_id=False):
1727
"""Add multiple versions to the index.
1729
This function does not insert data into the Immutable GraphIndex
1730
backing the KnitGraphIndex, instead it prepares data for insertion by
1731
the caller and checks that it is safe to insert then calls
1732
self._add_callback with the prepared GraphIndex nodes.
1734
:param versions: a list of tuples:
1735
(version_id, options, pos, size, parents).
1736
:param random_id: If True the ids being added were randomly generated
1737
and no check for existence will be performed.
1739
if not self._add_callback:
1740
raise errors.ReadOnlyError(self)
1741
# we hope there are no repositories with inconsistent parentage
1746
for (version_id, options, access_memo, parents) in versions:
1747
index, pos, size = access_memo
1748
key = (version_id, )
1749
parents = tuple((parent, ) for parent in parents)
1750
if 'no-eol' in options:
1754
value += "%d %d" % (pos, size)
1755
if not self._deltas:
1756
if 'line-delta' in options:
1757
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1760
if 'line-delta' in options:
1761
node_refs = (parents, (parents[0],))
1763
node_refs = (parents, ())
1765
node_refs = (parents, )
1768
raise KnitCorrupt(self, "attempt to add node with parents "
1769
"in parentless index.")
1771
keys[key] = (value, node_refs)
1773
present_nodes = self._get_entries(keys)
1774
for (index, key, value, node_refs) in present_nodes:
1775
if (value, node_refs) != keys[key]:
1776
raise KnitCorrupt(self, "inconsistent details in add_versions"
1777
": %s %s" % ((value, node_refs), keys[key]))
1781
for key, (value, node_refs) in keys.iteritems():
1782
result.append((key, value, node_refs))
1784
for key, (value, node_refs) in keys.iteritems():
1785
result.append((key, value))
1786
self._add_callback(result)
1788
def _version_ids_to_keys(self, version_ids):
1789
return set((version_id, ) for version_id in version_ids)
1792
class _KnitAccess(object):
1793
"""Access to knit records in a .knit file."""
1795
def __init__(self, transport, filename, _file_mode, _dir_mode,
1796
_need_to_create, _create_parent_dir):
1797
"""Create a _KnitAccess for accessing and inserting data.
1799
:param transport: The transport the .knit is located on.
1800
:param filename: The filename of the .knit.
1802
self._transport = transport
1803
self._filename = filename
1804
self._file_mode = _file_mode
1805
self._dir_mode = _dir_mode
1806
self._need_to_create = _need_to_create
1807
self._create_parent_dir = _create_parent_dir
1809
def add_raw_records(self, sizes, raw_data):
1810
"""Add raw knit bytes to a storage area.
1812
The data is spooled to whereever the access method is storing data.
1814
:param sizes: An iterable containing the size of each raw data segment.
1815
:param raw_data: A bytestring containing the data.
1816
:return: A list of memos to retrieve the record later. Each memo is a
1817
tuple - (index, pos, length), where the index field is always None
1818
for the .knit access method.
1820
assert type(raw_data) == str, \
1821
'data must be plain bytes was %s' % type(raw_data)
1822
if not self._need_to_create:
1823
base = self._transport.append_bytes(self._filename, raw_data)
1825
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1826
create_parent_dir=self._create_parent_dir,
1827
mode=self._file_mode,
1828
dir_mode=self._dir_mode)
1829
self._need_to_create = False
1833
result.append((None, base, size))
1838
"""IFF this data access has its own storage area, initialise it.
1842
self._transport.put_bytes_non_atomic(self._filename, '',
1843
mode=self._file_mode)
1845
def open_file(self):
1846
"""IFF this data access can be represented as a single file, open it.
1848
For knits that are not mapped to a single file on disk this will
1851
:return: None or a file handle.
1854
return self._transport.get(self._filename)
1859
def get_raw_records(self, memos_for_retrieval):
1860
"""Get the raw bytes for a records.
1862
:param memos_for_retrieval: An iterable containing the (index, pos,
1863
length) memo for retrieving the bytes. The .knit method ignores
1864
the index as there is always only a single file.
1865
:return: An iterator over the bytes of the records.
1867
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
1868
for pos, data in self._transport.readv(self._filename, read_vector):
1872
class _PackAccess(object):
1873
"""Access to knit records via a collection of packs."""
1875
def __init__(self, index_to_packs, writer=None):
1876
"""Create a _PackAccess object.
1878
:param index_to_packs: A dict mapping index objects to the transport
1879
and file names for obtaining data.
1880
:param writer: A tuple (pack.ContainerWriter, write_index) which
1881
contains the pack to write, and the index that reads from it will
1885
self.container_writer = writer[0]
1886
self.write_index = writer[1]
1888
self.container_writer = None
1889
self.write_index = None
1890
self.indices = index_to_packs
1892
def add_raw_records(self, sizes, raw_data):
1893
"""Add raw knit bytes to a storage area.
1895
The data is spooled to the container writer in one bytes-record per
1898
:param sizes: An iterable containing the size of each raw data segment.
1899
:param raw_data: A bytestring containing the data.
1900
:return: A list of memos to retrieve the record later. Each memo is a
1901
tuple - (index, pos, length), where the index field is the
1902
write_index object supplied to the PackAccess object.
1904
assert type(raw_data) == str, \
1905
'data must be plain bytes was %s' % type(raw_data)
1909
p_offset, p_length = self.container_writer.add_bytes_record(
1910
raw_data[offset:offset+size], [])
1912
result.append((self.write_index, p_offset, p_length))
1916
"""Pack based knits do not get individually created."""
1918
def get_raw_records(self, memos_for_retrieval):
1919
"""Get the raw bytes for a records.
1921
:param memos_for_retrieval: An iterable containing the (index, pos,
1922
length) memo for retrieving the bytes. The Pack access method
1923
looks up the pack to use for a given record in its index_to_pack
1925
:return: An iterator over the bytes of the records.
1927
# first pass, group into same-index requests
1929
current_index = None
1930
for (index, offset, length) in memos_for_retrieval:
1931
if current_index == index:
1932
current_list.append((offset, length))
1934
if current_index is not None:
1935
request_lists.append((current_index, current_list))
1936
current_index = index
1937
current_list = [(offset, length)]
1938
# handle the last entry
1939
if current_index is not None:
1940
request_lists.append((current_index, current_list))
1941
for index, offsets in request_lists:
1942
transport, path = self.indices[index]
1943
reader = pack.make_readv_reader(transport, path, offsets)
1944
for names, read_func in reader.iter_records():
1945
yield read_func(None)
1947
def open_file(self):
1948
"""Pack based knits have no single file."""
1951
def set_writer(self, writer, index, (transport, packname)):
1952
"""Set a writer to use for adding data."""
1953
self.indices[index] = (transport, packname)
1954
self.container_writer = writer
1955
self.write_index = index
1958
class _KnitData(object):
1959
"""Manage extraction of data from a KnitAccess, caching and decompressing.
1961
The KnitData class provides the logic for parsing and using knit records,
1962
making use of an access method for the low level read and write operations.
1965
def __init__(self, access):
1966
"""Create a KnitData object.
1968
:param access: The access method to use. Access methods such as
1969
_KnitAccess manage the insertion of raw records and the subsequent
1970
retrieval of the same.
1972
self._access = access
1973
self._checked = False
1974
# TODO: jam 20060713 conceptually, this could spill to disk
1975
# if the cached size gets larger than a certain amount
1976
# but it complicates the model a bit, so for now just use
1977
# a simple dictionary
1979
self._do_cache = False
1981
def enable_cache(self):
1982
"""Enable caching of reads."""
1983
self._do_cache = True
1985
def clear_cache(self):
1986
"""Clear the record cache."""
1987
self._do_cache = False
1990
def _open_file(self):
1991
return self._access.open_file()
1993
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
1994
"""Convert version_id, digest, lines into a raw data block.
1996
:param dense_lines: The bytes of lines but in a denser form. For
1997
instance, if lines is a list of 1000 bytestrings each ending in \n,
1998
dense_lines may be a list with one line in it, containing all the
1999
1000's lines and their \n's. Using dense_lines if it is already
2000
known is a win because the string join to create bytes in this
2001
function spends less time resizing the final string.
2002
:return: (len, a StringIO instance with the raw data ready to read.)
2004
# Note: using a string copy here increases memory pressure with e.g.
2005
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2006
# when doing the initial commit of a mozilla tree. RBC 20070921
2007
bytes = ''.join(chain(
2008
["version %s %d %s\n" % (version_id,
2011
dense_lines or lines,
2012
["end %s\n" % version_id]))
2013
assert bytes.__class__ == str
2014
compressed_bytes = bytes_to_gzip(bytes)
2015
return len(compressed_bytes), compressed_bytes
2017
def add_raw_records(self, sizes, raw_data):
2018
"""Append a prepared record to the data file.
2020
:param sizes: An iterable containing the size of each raw data segment.
2021
:param raw_data: A bytestring containing the data.
2022
:return: a list of index data for the way the data was stored.
2023
See the access method add_raw_records documentation for more
2026
return self._access.add_raw_records(sizes, raw_data)
2028
def _parse_record_header(self, version_id, raw_data):
2029
"""Parse a record header for consistency.
2031
:return: the header and the decompressor stream.
2032
as (stream, header_record)
2034
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2036
rec = self._check_header(version_id, df.readline())
2037
except Exception, e:
2038
raise KnitCorrupt(self._access,
2039
"While reading {%s} got %s(%s)"
2040
% (version_id, e.__class__.__name__, str(e)))
2043
def _check_header(self, version_id, line):
2046
raise KnitCorrupt(self._access,
2047
'unexpected number of elements in record header')
2048
if rec[1] != version_id:
2049
raise KnitCorrupt(self._access,
2050
'unexpected version, wanted %r, got %r'
2051
% (version_id, rec[1]))
2054
def _parse_record(self, version_id, data):
2056
# 4168 calls in 2880 217 internal
2057
# 4168 calls to _parse_record_header in 2121
2058
# 4168 calls to readlines in 330
2059
df = GzipFile(mode='rb', fileobj=StringIO(data))
2062
record_contents = df.readlines()
2063
except Exception, e:
2064
raise KnitCorrupt(self._access,
2065
"While reading {%s} got %s(%s)"
2066
% (version_id, e.__class__.__name__, str(e)))
2067
header = record_contents.pop(0)
2068
rec = self._check_header(version_id, header)
2070
last_line = record_contents.pop()
2071
if len(record_contents) != int(rec[2]):
2072
raise KnitCorrupt(self._access,
2073
'incorrect number of lines %s != %s'
2075
% (len(record_contents), int(rec[2]),
2077
if last_line != 'end %s\n' % rec[1]:
2078
raise KnitCorrupt(self._access,
2079
'unexpected version end line %r, wanted %r'
2080
% (last_line, version_id))
2082
return record_contents, rec[3]
2084
def read_records_iter_raw(self, records):
2085
"""Read text records from data file and yield raw data.
2087
This unpacks enough of the text record to validate the id is
2088
as expected but thats all.
2090
# setup an iterator of the external records:
2091
# uses readv so nice and fast we hope.
2093
# grab the disk data needed.
2095
# Don't check _cache if it is empty
2096
needed_offsets = [index_memo for version_id, index_memo
2098
if version_id not in self._cache]
2100
needed_offsets = [index_memo for version_id, index_memo
2103
raw_records = self._access.get_raw_records(needed_offsets)
2105
for version_id, index_memo in records:
2106
if version_id in self._cache:
2107
# This data has already been validated
2108
data = self._cache[version_id]
2110
data = raw_records.next()
2112
self._cache[version_id] = data
2114
# validate the header
2115
df, rec = self._parse_record_header(version_id, data)
2117
yield version_id, data
2119
def read_records_iter(self, records):
2120
"""Read text records from data file and yield result.
2122
The result will be returned in whatever is the fastest to read.
2123
Not by the order requested. Also, multiple requests for the same
2124
record will only yield 1 response.
2125
:param records: A list of (version_id, pos, len) entries
2126
:return: Yields (version_id, contents, digest) in the order
2127
read, not the order requested
2133
# Skip records we have alread seen
2134
yielded_records = set()
2135
needed_records = set()
2136
for record in records:
2137
if record[0] in self._cache:
2138
if record[0] in yielded_records:
2140
yielded_records.add(record[0])
2141
data = self._cache[record[0]]
2142
content, digest = self._parse_record(record[0], data)
2143
yield (record[0], content, digest)
2145
needed_records.add(record)
2146
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2148
needed_records = sorted(set(records), key=operator.itemgetter(1))
2150
if not needed_records:
2153
# The transport optimizes the fetching as well
2154
# (ie, reads continuous ranges.)
2155
raw_data = self._access.get_raw_records(
2156
[index_memo for version_id, index_memo in needed_records])
2158
for (version_id, index_memo), data in \
2159
izip(iter(needed_records), raw_data):
2160
content, digest = self._parse_record(version_id, data)
2162
self._cache[version_id] = data
2163
yield version_id, content, digest
2165
def read_records(self, records):
2166
"""Read records into a dictionary."""
2168
for record_id, content, digest in \
2169
self.read_records_iter(records):
2170
components[record_id] = (content, digest)
2174
class InterKnit(InterVersionedFile):
2175
"""Optimised code paths for knit to knit operations."""
2177
_matching_file_from_factory = KnitVersionedFile
2178
_matching_file_to_factory = KnitVersionedFile
2181
def is_compatible(source, target):
2182
"""Be compatible with knits. """
2184
return (isinstance(source, KnitVersionedFile) and
2185
isinstance(target, KnitVersionedFile))
2186
except AttributeError:
2189
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2190
"""See InterVersionedFile.join."""
2191
assert isinstance(self.source, KnitVersionedFile)
2192
assert isinstance(self.target, KnitVersionedFile)
2194
# If the source and target are mismatched w.r.t. annotations vs
2195
# plain, the data needs to be converted accordingly
2196
if self.source.factory.annotated == self.target.factory.annotated:
2198
elif self.source.factory.annotated:
2199
converter = self._anno_to_plain_converter
2201
# We're converting from a plain to an annotated knit. This requires
2202
# building the annotations from scratch. The generic join code
2203
# handles this implicitly so we delegate to it.
2204
return super(InterKnit, self).join(pb, msg, version_ids,
2207
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2211
pb = ui.ui_factory.nested_progress_bar()
2213
version_ids = list(version_ids)
2214
if None in version_ids:
2215
version_ids.remove(None)
2217
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2218
this_versions = set(self.target._index.get_versions())
2219
# XXX: For efficiency we should not look at the whole index,
2220
# we only need to consider the referenced revisions - they
2221
# must all be present, or the method must be full-text.
2222
# TODO, RBC 20070919
2223
needed_versions = self.source_ancestry - this_versions
2225
if not needed_versions:
2227
full_list = topo_sort(self.source.get_graph())
2229
version_list = [i for i in full_list if (not self.target.has_version(i)
2230
and i in needed_versions)]
2234
copy_queue_records = []
2236
for version_id in version_list:
2237
options = self.source._index.get_options(version_id)
2238
parents = self.source._index.get_parents_with_ghosts(version_id)
2239
# check that its will be a consistent copy:
2240
for parent in parents:
2241
# if source has the parent, we must :
2242
# * already have it or
2243
# * have it scheduled already
2244
# otherwise we don't care
2245
assert (self.target.has_version(parent) or
2246
parent in copy_set or
2247
not self.source.has_version(parent))
2248
index_memo = self.source._index.get_position(version_id)
2249
copy_queue_records.append((version_id, index_memo))
2250
copy_queue.append((version_id, options, parents))
2251
copy_set.add(version_id)
2253
# data suck the join:
2255
total = len(version_list)
2258
for (version_id, raw_data), \
2259
(version_id2, options, parents) in \
2260
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2262
assert version_id == version_id2, 'logic error, inconsistent results'
2264
pb.update("Joining knit", count, total)
2266
size, raw_data = converter(raw_data, version_id, options,
2269
size = len(raw_data)
2270
raw_records.append((version_id, options, parents, size))
2271
raw_datum.append(raw_data)
2272
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2277
def _anno_to_plain_converter(self, raw_data, version_id, options,
2279
"""Convert annotated content to plain content."""
2280
data, digest = self.source._data._parse_record(version_id, raw_data)
2281
if 'fulltext' in options:
2282
content = self.source.factory.parse_fulltext(data, version_id)
2283
lines = self.target.factory.lower_fulltext(content)
2285
delta = self.source.factory.parse_line_delta(data, version_id,
2287
lines = self.target.factory.lower_line_delta(delta)
2288
return self.target._data._record_to_data(version_id, digest, lines)
2291
InterVersionedFile.register_optimiser(InterKnit)
2294
class WeaveToKnit(InterVersionedFile):
2295
"""Optimised code paths for weave to knit operations."""
2297
_matching_file_from_factory = bzrlib.weave.WeaveFile
2298
_matching_file_to_factory = KnitVersionedFile
2301
def is_compatible(source, target):
2302
"""Be compatible with weaves to knits."""
2304
return (isinstance(source, bzrlib.weave.Weave) and
2305
isinstance(target, KnitVersionedFile))
2306
except AttributeError:
2309
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2310
"""See InterVersionedFile.join."""
2311
assert isinstance(self.source, bzrlib.weave.Weave)
2312
assert isinstance(self.target, KnitVersionedFile)
2314
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2319
pb = ui.ui_factory.nested_progress_bar()
2321
version_ids = list(version_ids)
2323
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2324
this_versions = set(self.target._index.get_versions())
2325
needed_versions = self.source_ancestry - this_versions
2327
if not needed_versions:
2329
full_list = topo_sort(self.source.get_graph())
2331
version_list = [i for i in full_list if (not self.target.has_version(i)
2332
and i in needed_versions)]
2336
total = len(version_list)
2337
for version_id in version_list:
2338
pb.update("Converting to knit", count, total)
2339
parents = self.source.get_parents(version_id)
2340
# check that its will be a consistent copy:
2341
for parent in parents:
2342
# if source has the parent, we must already have it
2343
assert (self.target.has_version(parent))
2344
self.target.add_lines(
2345
version_id, parents, self.source.get_lines(version_id))
2352
InterVersionedFile.register_optimiser(WeaveToKnit)
2355
# Deprecated, use PatienceSequenceMatcher instead
2356
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2359
def annotate_knit(knit, revision_id):
2360
"""Annotate a knit with no cached annotations.
2362
This implementation is for knits with no cached annotations.
2363
It will work for knits with cached annotations, but this is not
2366
ancestry = knit.get_ancestry(revision_id)
2367
fulltext = dict(zip(ancestry, knit.get_line_list(ancestry)))
2369
for candidate in ancestry:
2370
if candidate in annotations:
2372
parents = knit.get_parents(candidate)
2373
if len(parents) == 0:
2375
elif knit._index.get_method(candidate) != 'line-delta':
2378
parent, sha1, noeol, delta = knit.get_delta(candidate)
2379
blocks = KnitContent.get_line_delta_blocks(delta,
2380
fulltext[parents[0]], fulltext[candidate])
2381
annotations[candidate] = list(annotate.reannotate([annotations[p]
2382
for p in parents], fulltext[candidate], candidate, blocks))
2383
return iter(annotations[revision_id])
2387
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2389
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
