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
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 _fix_parents(self, version_id, new_parents):
538
"""Fix the parents list for version.
540
This is done by appending a new version to the index
541
with identical data except for the parents list.
542
the parents list must be a superset of the current
545
current_values = self._index._cache[version_id]
546
assert set(current_values[4]).difference(set(new_parents)) == set()
547
self._index.add_version(version_id,
549
(None, current_values[2], current_values[3]),
552
def get_data_stream(self, required_versions):
553
"""Get a data stream for the specified versions.
555
Versions may be returned in any order, not necessarily the order
558
:param required_versions: The exact set of versions to be extracted.
559
Unlike some other knit methods, this is not used to generate a
560
transitive closure, rather it is used precisely as given.
562
:returns: format_signature, list of (version, options, length, parents),
565
required_versions = set([osutils.safe_revision_id(v) for v in
567
# we don't care about inclusions, the caller cares.
568
# but we need to setup a list of records to visit.
569
for version_id in required_versions:
570
if not self.has_version(version_id):
571
raise RevisionNotPresent(version_id, self.filename)
572
# Pick the desired versions out of the index in oldest-to-newest order
574
for version_id in self.versions():
575
if version_id in required_versions:
576
version_list.append(version_id)
578
# create the list of version information for the result
579
copy_queue_records = []
581
result_version_list = []
582
for version_id in version_list:
583
options = self._index.get_options(version_id)
584
parents = self._index.get_parents_with_ghosts(version_id)
585
index_memo = self._index.get_position(version_id)
586
copy_queue_records.append((version_id, index_memo))
587
none, data_pos, data_size = index_memo
588
copy_set.add(version_id)
589
# version, options, length, parents
590
result_version_list.append((version_id, options, data_size,
593
# Read the compressed record data.
595
# From here down to the return should really be logic in the returned
596
# callable -- in a class that adapts read_records_iter_raw to read
599
for (version_id, raw_data), \
600
(version_id2, options, _, parents) in \
601
izip(self._data.read_records_iter_raw(copy_queue_records),
602
result_version_list):
603
assert version_id == version_id2, 'logic error, inconsistent results'
604
raw_datum.append(raw_data)
605
pseudo_file = StringIO(''.join(raw_datum))
608
return pseudo_file.read()
610
return pseudo_file.read(length)
611
return (self.get_format_signature(), result_version_list, read)
613
def _extract_blocks(self, version_id, source, target):
614
if self._index.get_method(version_id) != 'line-delta':
616
parent, sha1, noeol, delta = self.get_delta(version_id)
617
return KnitContent.get_line_delta_blocks(delta, source, target)
619
def get_delta(self, version_id):
620
"""Get a delta for constructing version from some other version."""
621
version_id = osutils.safe_revision_id(version_id)
622
self.check_not_reserved_id(version_id)
623
parents = self.get_parents(version_id)
628
index_memo = self._index.get_position(version_id)
629
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
630
noeol = 'no-eol' in self._index.get_options(version_id)
631
if 'fulltext' == self._index.get_method(version_id):
632
new_content = self.factory.parse_fulltext(data, version_id)
633
if parent is not None:
634
reference_content = self._get_content(parent)
635
old_texts = reference_content.text()
638
new_texts = new_content.text()
639
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
641
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
643
delta = self.factory.parse_line_delta(data, version_id)
644
return parent, sha1, noeol, delta
646
def get_format_signature(self):
647
"""See VersionedFile.get_format_signature()."""
648
if self.factory.annotated:
649
annotated_part = "annotated"
651
annotated_part = "plain"
652
return "knit-%s" % (annotated_part,)
654
def get_graph_with_ghosts(self):
655
"""See VersionedFile.get_graph_with_ghosts()."""
656
graph_items = self._index.get_graph()
657
return dict(graph_items)
659
def get_sha1(self, version_id):
660
return self.get_sha1s([version_id])[0]
662
def get_sha1s(self, version_ids):
663
"""See VersionedFile.get_sha1()."""
664
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
665
record_map = self._get_record_map(version_ids)
666
# record entry 2 is the 'digest'.
667
return [record_map[v][2] for v in version_ids]
671
"""See VersionedFile.get_suffixes()."""
672
return [DATA_SUFFIX, INDEX_SUFFIX]
674
def has_ghost(self, version_id):
675
"""True if there is a ghost reference in the file to version_id."""
676
version_id = osutils.safe_revision_id(version_id)
678
if self.has_version(version_id):
680
# optimisable if needed by memoising the _ghosts set.
681
items = self._index.get_graph()
682
for node, parents in items:
683
for parent in parents:
684
if parent not in self._index._cache:
685
if parent == version_id:
689
def insert_data_stream(self, (format, data_list, reader_callable)):
690
"""Insert knit records from a data stream into this knit.
692
If a version in the stream is already present in this knit, it will not
693
be inserted a second time. It will be checked for consistency with the
694
stored version however, and may cause a KnitCorrupt error to be raised
695
if the data in the stream disagrees with the already stored data.
697
:seealso: get_data_stream
699
if format != self.get_format_signature():
700
trace.mutter('incompatible format signature inserting to %r', self)
701
raise KnitDataStreamIncompatible(
702
format, self.get_format_signature())
704
for version_id, options, length, parents in data_list:
705
if self.has_version(version_id):
706
# First check: the list of parents.
707
my_parents = self.get_parents_with_ghosts(version_id)
708
if my_parents != parents:
709
# XXX: KnitCorrupt is not quite the right exception here.
712
'parents list %r from data stream does not match '
713
'already recorded parents %r for %s'
714
% (parents, my_parents, version_id))
716
# Also check the SHA-1 of the fulltext this content will
718
raw_data = reader_callable(length)
719
my_fulltext_sha1 = self.get_sha1(version_id)
720
df, rec = self._data._parse_record_header(version_id, raw_data)
721
stream_fulltext_sha1 = rec[3]
722
if my_fulltext_sha1 != stream_fulltext_sha1:
723
# Actually, we don't know if it's this knit that's corrupt,
724
# or the data stream we're trying to insert.
726
self.filename, 'sha-1 does not match %s' % version_id)
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
version_id = osutils.safe_revision_id(version_id)
743
return self._index.has_version(version_id)
745
__contains__ = has_version
747
def _merge_annotations(self, content, parents, parent_texts={},
748
delta=None, annotated=None,
749
left_matching_blocks=None):
750
"""Merge annotations for content. This is done by comparing
751
the annotations based on changed to the text.
753
if left_matching_blocks is not None:
754
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
758
for parent_id in parents:
759
merge_content = self._get_content(parent_id, parent_texts)
760
if (parent_id == parents[0] and delta_seq is not None):
763
seq = patiencediff.PatienceSequenceMatcher(
764
None, merge_content.text(), content.text())
765
for i, j, n in seq.get_matching_blocks():
768
# this appears to copy (origin, text) pairs across to the
769
# new content for any line that matches the last-checked
771
content._lines[j:j+n] = merge_content._lines[i:i+n]
773
if delta_seq is None:
774
reference_content = self._get_content(parents[0], parent_texts)
775
new_texts = content.text()
776
old_texts = reference_content.text()
777
delta_seq = patiencediff.PatienceSequenceMatcher(
778
None, old_texts, new_texts)
779
return self._make_line_delta(delta_seq, content)
781
def _make_line_delta(self, delta_seq, new_content):
782
"""Generate a line delta from delta_seq and new_content."""
784
for op in delta_seq.get_opcodes():
787
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
790
def _get_components_positions(self, version_ids):
791
"""Produce a map of position data for the components of versions.
793
This data is intended to be used for retrieving the knit records.
795
A dict of version_id to (method, data_pos, data_size, next) is
797
method is the way referenced data should be applied.
798
data_pos is the position of the data in the knit.
799
data_size is the size of the data in the knit.
800
next is the build-parent of the version, or None for fulltexts.
803
for version_id in version_ids:
806
while cursor is not None and cursor not in component_data:
807
method = self._index.get_method(cursor)
808
if method == 'fulltext':
811
next = self.get_parents(cursor)[0]
812
index_memo = self._index.get_position(cursor)
813
component_data[cursor] = (method, index_memo, next)
815
return component_data
817
def _get_content(self, version_id, parent_texts={}):
818
"""Returns a content object that makes up the specified
820
cached_version = parent_texts.get(version_id, None)
821
if cached_version is not None:
822
if not self.has_version(version_id):
823
raise RevisionNotPresent(version_id, self.filename)
824
return cached_version
826
text_map, contents_map = self._get_content_maps([version_id])
827
return contents_map[version_id]
829
def _check_versions_present(self, version_ids):
830
"""Check that all specified versions are present."""
831
self._index.check_versions_present(version_ids)
833
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
834
nostore_sha, random_id, check_content):
835
"""See VersionedFile.add_lines_with_ghosts()."""
836
self._check_add(version_id, lines, random_id, check_content)
837
return self._add(version_id, lines, parents, self.delta,
838
parent_texts, None, nostore_sha)
840
def _add_lines(self, version_id, parents, lines, parent_texts,
841
left_matching_blocks, nostore_sha, random_id, check_content):
842
"""See VersionedFile.add_lines."""
843
self._check_add(version_id, lines, random_id, check_content)
844
self._check_versions_present(parents)
845
return self._add(version_id, lines[:], parents, self.delta,
846
parent_texts, left_matching_blocks, nostore_sha)
848
def _check_add(self, version_id, lines, random_id, check_content):
849
"""check that version_id and lines are safe to add."""
850
if contains_whitespace(version_id):
851
raise InvalidRevisionId(version_id, self.filename)
852
self.check_not_reserved_id(version_id)
853
# Technically this could be avoided if we are happy to allow duplicate
854
# id insertion when other things than bzr core insert texts, but it
855
# seems useful for folk using the knit api directly to have some safety
856
# blanket that we can disable.
857
if not random_id and self.has_version(version_id):
858
raise RevisionAlreadyPresent(version_id, self.filename)
860
self._check_lines_not_unicode(lines)
861
self._check_lines_are_lines(lines)
863
def _add(self, version_id, lines, parents, delta, parent_texts,
864
left_matching_blocks, nostore_sha):
865
"""Add a set of lines on top of version specified by parents.
867
If delta is true, compress the text as a line-delta against
870
Any versions not present will be converted into ghosts.
872
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
873
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
874
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
875
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
876
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
877
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
878
# +1383 0 8.0370 8.0370 +<len>
879
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
880
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
881
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
882
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
885
if parent_texts is None:
887
for parent in parents:
888
if self.has_version(parent):
889
present_parents.append(parent)
891
# can only compress against the left most present parent.
893
(len(present_parents) == 0 or
894
present_parents[0] != parents[0])):
897
digest = sha_strings(lines)
898
if nostore_sha == digest:
899
raise errors.ExistingContent
900
text_length = sum(map(len, lines))
903
if lines[-1][-1] != '\n':
904
# copy the contents of lines.
906
options.append('no-eol')
907
lines[-1] = lines[-1] + '\n'
910
# To speed the extract of texts the delta chain is limited
911
# to a fixed number of deltas. This should minimize both
912
# I/O and the time spend applying deltas.
913
delta = self._check_should_delta(present_parents)
915
assert isinstance(version_id, str)
916
content = self.factory.make(lines, version_id)
917
if delta or (self.factory.annotated and len(present_parents) > 0):
918
# Merge annotations from parent texts if needed.
919
delta_hunks = self._merge_annotations(content, present_parents,
920
parent_texts, delta, self.factory.annotated,
921
left_matching_blocks)
924
options.append('line-delta')
925
store_lines = self.factory.lower_line_delta(delta_hunks)
927
options.append('fulltext')
928
store_lines = self.factory.lower_fulltext(content)
930
access_memo = self._data.add_record(version_id, digest, store_lines)
931
self._index.add_version(version_id, options, access_memo, parents)
932
return digest, text_length, content
934
def check(self, progress_bar=None):
935
"""See VersionedFile.check()."""
937
def _clone_text(self, new_version_id, old_version_id, parents):
938
"""See VersionedFile.clone_text()."""
939
# FIXME RBC 20060228 make fast by only inserting an index with null
941
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
943
def get_lines(self, version_id):
944
"""See VersionedFile.get_lines()."""
945
return self.get_line_list([version_id])[0]
947
def _get_record_map(self, version_ids):
948
"""Produce a dictionary of knit records.
950
The keys are version_ids, the values are tuples of (method, content,
952
method is the way the content should be applied.
953
content is a KnitContent object.
954
digest is the SHA1 digest of this version id after all steps are done
955
next is the build-parent of the version, i.e. the leftmost ancestor.
956
If the method is fulltext, next will be None.
958
position_map = self._get_components_positions(version_ids)
959
# c = component_id, m = method, i_m = index_memo, n = next
960
records = [(c, i_m) for c, (m, i_m, n) in position_map.iteritems()]
962
for component_id, content, digest in \
963
self._data.read_records_iter(records):
964
method, index_memo, next = position_map[component_id]
965
record_map[component_id] = method, content, digest, next
969
def get_text(self, version_id):
970
"""See VersionedFile.get_text"""
971
return self.get_texts([version_id])[0]
973
def get_texts(self, version_ids):
974
return [''.join(l) for l in self.get_line_list(version_ids)]
976
def get_line_list(self, version_ids):
977
"""Return the texts of listed versions as a list of strings."""
978
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
979
for version_id in version_ids:
980
self.check_not_reserved_id(version_id)
981
text_map, content_map = self._get_content_maps(version_ids)
982
return [text_map[v] for v in version_ids]
984
_get_lf_split_line_list = get_line_list
986
def _get_content_maps(self, version_ids):
987
"""Produce maps of text and KnitContents
989
:return: (text_map, content_map) where text_map contains the texts for
990
the requested versions and content_map contains the KnitContents.
991
Both dicts take version_ids as their keys.
993
for version_id in version_ids:
994
if not self.has_version(version_id):
995
raise RevisionNotPresent(version_id, self.filename)
996
record_map = self._get_record_map(version_ids)
1001
for version_id in version_ids:
1004
while cursor is not None:
1005
method, data, digest, next = record_map[cursor]
1006
components.append((cursor, method, data, digest))
1007
if cursor in content_map:
1012
for component_id, method, data, digest in reversed(components):
1013
if component_id in content_map:
1014
content = content_map[component_id]
1016
if method == 'fulltext':
1017
assert content is None
1018
content = self.factory.parse_fulltext(data, version_id)
1019
elif method == 'line-delta':
1020
delta = self.factory.parse_line_delta(data, version_id)
1021
content = content.copy()
1022
content._lines = self._apply_delta(content._lines,
1024
content_map[component_id] = content
1026
if 'no-eol' in self._index.get_options(version_id):
1027
content = content.copy()
1028
content.strip_last_line_newline()
1029
final_content[version_id] = content
1031
# digest here is the digest from the last applied component.
1032
text = content.text()
1033
if sha_strings(text) != digest:
1034
raise KnitCorrupt(self.filename,
1035
'sha-1 does not match %s' % version_id)
1037
text_map[version_id] = text
1038
return text_map, final_content
1040
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1042
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1043
if version_ids is None:
1044
version_ids = self.versions()
1046
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
1048
pb = progress.DummyProgress()
1049
# we don't care about inclusions, the caller cares.
1050
# but we need to setup a list of records to visit.
1051
# we need version_id, position, length
1052
version_id_records = []
1053
requested_versions = set(version_ids)
1054
# filter for available versions
1055
for version_id in requested_versions:
1056
if not self.has_version(version_id):
1057
raise RevisionNotPresent(version_id, self.filename)
1058
# get a in-component-order queue:
1059
for version_id in self.versions():
1060
if version_id in requested_versions:
1061
index_memo = self._index.get_position(version_id)
1062
version_id_records.append((version_id, index_memo))
1064
total = len(version_id_records)
1065
for version_idx, (version_id, data, sha_value) in \
1066
enumerate(self._data.read_records_iter(version_id_records)):
1067
pb.update('Walking content.', version_idx, total)
1068
method = self._index.get_method(version_id)
1070
assert method in ('fulltext', 'line-delta')
1071
if method == 'fulltext':
1072
line_iterator = self.factory.get_fulltext_content(data)
1074
line_iterator = self.factory.get_linedelta_content(data)
1075
for line in line_iterator:
1078
pb.update('Walking content.', total, total)
1080
def iter_parents(self, version_ids):
1081
"""Iterate through the parents for many version ids.
1083
:param version_ids: An iterable yielding version_ids.
1084
:return: An iterator that yields (version_id, parents). Requested
1085
version_ids not present in the versioned file are simply skipped.
1086
The order is undefined, allowing for different optimisations in
1087
the underlying implementation.
1089
version_ids = [osutils.safe_revision_id(version_id) for
1090
version_id in version_ids]
1091
return self._index.iter_parents(version_ids)
1093
def num_versions(self):
1094
"""See VersionedFile.num_versions()."""
1095
return self._index.num_versions()
1097
__len__ = num_versions
1099
def annotate_iter(self, version_id):
1100
"""See VersionedFile.annotate_iter."""
1101
version_id = osutils.safe_revision_id(version_id)
1102
return self.factory.annotate_iter(self, version_id)
1104
def get_parents(self, version_id):
1105
"""See VersionedFile.get_parents."""
1108
# 52554 calls in 1264 872 internal down from 3674
1109
version_id = osutils.safe_revision_id(version_id)
1111
return self._index.get_parents(version_id)
1113
raise RevisionNotPresent(version_id, self.filename)
1115
def get_parents_with_ghosts(self, version_id):
1116
"""See VersionedFile.get_parents."""
1117
version_id = osutils.safe_revision_id(version_id)
1119
return self._index.get_parents_with_ghosts(version_id)
1121
raise RevisionNotPresent(version_id, self.filename)
1123
def get_ancestry(self, versions, topo_sorted=True):
1124
"""See VersionedFile.get_ancestry."""
1125
if isinstance(versions, basestring):
1126
versions = [versions]
1129
versions = [osutils.safe_revision_id(v) for v in versions]
1130
return self._index.get_ancestry(versions, topo_sorted)
1132
def get_ancestry_with_ghosts(self, versions):
1133
"""See VersionedFile.get_ancestry_with_ghosts."""
1134
if isinstance(versions, basestring):
1135
versions = [versions]
1138
versions = [osutils.safe_revision_id(v) for v in versions]
1139
return self._index.get_ancestry_with_ghosts(versions)
1141
def plan_merge(self, ver_a, ver_b):
1142
"""See VersionedFile.plan_merge."""
1143
ver_a = osutils.safe_revision_id(ver_a)
1144
ver_b = osutils.safe_revision_id(ver_b)
1145
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1147
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1148
annotated_a = self.annotate(ver_a)
1149
annotated_b = self.annotate(ver_b)
1150
return merge._plan_annotate_merge(annotated_a, annotated_b,
1151
ancestors_a, ancestors_b)
1154
class _KnitComponentFile(object):
1155
"""One of the files used to implement a knit database"""
1157
def __init__(self, transport, filename, mode, file_mode=None,
1158
create_parent_dir=False, dir_mode=None):
1159
self._transport = transport
1160
self._filename = filename
1162
self._file_mode = file_mode
1163
self._dir_mode = dir_mode
1164
self._create_parent_dir = create_parent_dir
1165
self._need_to_create = False
1167
def _full_path(self):
1168
"""Return the full path to this file."""
1169
return self._transport.base + self._filename
1171
def check_header(self, fp):
1172
line = fp.readline()
1174
# An empty file can actually be treated as though the file doesn't
1176
raise errors.NoSuchFile(self._full_path())
1177
if line != self.HEADER:
1178
raise KnitHeaderError(badline=line,
1179
filename=self._transport.abspath(self._filename))
1182
return '%s(%s)' % (self.__class__.__name__, self._filename)
1185
class _KnitIndex(_KnitComponentFile):
1186
"""Manages knit index file.
1188
The index is already kept in memory and read on startup, to enable
1189
fast lookups of revision information. The cursor of the index
1190
file is always pointing to the end, making it easy to append
1193
_cache is a cache for fast mapping from version id to a Index
1196
_history is a cache for fast mapping from indexes to version ids.
1198
The index data format is dictionary compressed when it comes to
1199
parent references; a index entry may only have parents that with a
1200
lover index number. As a result, the index is topological sorted.
1202
Duplicate entries may be written to the index for a single version id
1203
if this is done then the latter one completely replaces the former:
1204
this allows updates to correct version and parent information.
1205
Note that the two entries may share the delta, and that successive
1206
annotations and references MUST point to the first entry.
1208
The index file on disc contains a header, followed by one line per knit
1209
record. The same revision can be present in an index file more than once.
1210
The first occurrence gets assigned a sequence number starting from 0.
1212
The format of a single line is
1213
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1214
REVISION_ID is a utf8-encoded revision id
1215
FLAGS is a comma separated list of flags about the record. Values include
1216
no-eol, line-delta, fulltext.
1217
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1218
that the the compressed data starts at.
1219
LENGTH is the ascii representation of the length of the data file.
1220
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1222
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1223
revision id already in the knit that is a parent of REVISION_ID.
1224
The ' :' marker is the end of record marker.
1227
when a write is interrupted to the index file, it will result in a line
1228
that does not end in ' :'. If the ' :' is not present at the end of a line,
1229
or at the end of the file, then the record that is missing it will be
1230
ignored by the parser.
1232
When writing new records to the index file, the data is preceded by '\n'
1233
to ensure that records always start on new lines even if the last write was
1234
interrupted. As a result its normal for the last line in the index to be
1235
missing a trailing newline. One can be added with no harmful effects.
1238
HEADER = "# bzr knit index 8\n"
1240
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1241
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1243
def _cache_version(self, version_id, options, pos, size, parents):
1244
"""Cache a version record in the history array and index cache.
1246
This is inlined into _load_data for performance. KEEP IN SYNC.
1247
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1250
# only want the _history index to reference the 1st index entry
1252
if version_id not in self._cache:
1253
index = len(self._history)
1254
self._history.append(version_id)
1256
index = self._cache[version_id][5]
1257
self._cache[version_id] = (version_id,
1264
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1265
create_parent_dir=False, delay_create=False, dir_mode=None):
1266
_KnitComponentFile.__init__(self, transport, filename, mode,
1267
file_mode=file_mode,
1268
create_parent_dir=create_parent_dir,
1271
# position in _history is the 'official' index for a revision
1272
# but the values may have come from a newer entry.
1273
# so - wc -l of a knit index is != the number of unique names
1277
fp = self._transport.get(self._filename)
1279
# _load_data may raise NoSuchFile if the target knit is
1281
_load_data(self, fp)
1285
if mode != 'w' or not create:
1288
self._need_to_create = True
1290
self._transport.put_bytes_non_atomic(
1291
self._filename, self.HEADER, mode=self._file_mode)
1293
def get_graph(self):
1294
"""Return a list of the node:parents lists from this knit index."""
1295
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1297
def get_ancestry(self, versions, topo_sorted=True):
1298
"""See VersionedFile.get_ancestry."""
1299
# get a graph of all the mentioned versions:
1301
pending = set(versions)
1304
version = pending.pop()
1307
parents = [p for p in cache[version][4] if p in cache]
1309
raise RevisionNotPresent(version, self._filename)
1310
# if not completed and not a ghost
1311
pending.update([p for p in parents if p not in graph])
1312
graph[version] = parents
1315
return topo_sort(graph.items())
1317
def get_ancestry_with_ghosts(self, versions):
1318
"""See VersionedFile.get_ancestry_with_ghosts."""
1319
# get a graph of all the mentioned versions:
1320
self.check_versions_present(versions)
1323
pending = set(versions)
1325
version = pending.pop()
1327
parents = cache[version][4]
1333
pending.update([p for p in parents if p not in graph])
1334
graph[version] = parents
1335
return topo_sort(graph.items())
1337
def iter_parents(self, version_ids):
1338
"""Iterate through the parents for many version ids.
1340
:param version_ids: An iterable yielding version_ids.
1341
:return: An iterator that yields (version_id, parents). Requested
1342
version_ids not present in the versioned file are simply skipped.
1343
The order is undefined, allowing for different optimisations in
1344
the underlying implementation.
1346
for version_id in version_ids:
1348
yield version_id, tuple(self.get_parents(version_id))
1352
def num_versions(self):
1353
return len(self._history)
1355
__len__ = num_versions
1357
def get_versions(self):
1358
"""Get all the versions in the file. not topologically sorted."""
1359
return self._history
1361
def _version_list_to_index(self, versions):
1364
for version in versions:
1365
if version in cache:
1366
# -- inlined lookup() --
1367
result_list.append(str(cache[version][5]))
1368
# -- end lookup () --
1370
result_list.append('.' + version)
1371
return ' '.join(result_list)
1373
def add_version(self, version_id, options, index_memo, parents):
1374
"""Add a version record to the index."""
1375
self.add_versions(((version_id, options, index_memo, parents),))
1377
def add_versions(self, versions):
1378
"""Add multiple versions to the index.
1380
:param versions: a list of tuples:
1381
(version_id, options, pos, size, parents).
1384
orig_history = self._history[:]
1385
orig_cache = self._cache.copy()
1388
for version_id, options, (index, pos, size), parents in versions:
1389
line = "\n%s %s %s %s %s :" % (version_id,
1393
self._version_list_to_index(parents))
1394
assert isinstance(line, str), \
1395
'content must be utf-8 encoded: %r' % (line,)
1397
self._cache_version(version_id, options, pos, size, parents)
1398
if not self._need_to_create:
1399
self._transport.append_bytes(self._filename, ''.join(lines))
1402
sio.write(self.HEADER)
1403
sio.writelines(lines)
1405
self._transport.put_file_non_atomic(self._filename, sio,
1406
create_parent_dir=self._create_parent_dir,
1407
mode=self._file_mode,
1408
dir_mode=self._dir_mode)
1409
self._need_to_create = False
1411
# If any problems happen, restore the original values and re-raise
1412
self._history = orig_history
1413
self._cache = orig_cache
1416
def has_version(self, version_id):
1417
"""True if the version is in the index."""
1418
return version_id in self._cache
1420
def get_position(self, version_id):
1421
"""Return details needed to access the version.
1423
.kndx indices do not support split-out data, so return None for the
1426
:return: a tuple (None, data position, size) to hand to the access
1427
logic to get the record.
1429
entry = self._cache[version_id]
1430
return None, entry[2], entry[3]
1432
def get_method(self, version_id):
1433
"""Return compression method of specified version."""
1434
options = self._cache[version_id][1]
1435
if 'fulltext' in options:
1438
if 'line-delta' not in options:
1439
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1442
def get_options(self, version_id):
1443
"""Return a string represention options.
1447
return self._cache[version_id][1]
1449
def get_parents(self, version_id):
1450
"""Return parents of specified version ignoring ghosts."""
1451
return [parent for parent in self._cache[version_id][4]
1452
if parent in self._cache]
1454
def get_parents_with_ghosts(self, version_id):
1455
"""Return parents of specified version with ghosts."""
1456
return self._cache[version_id][4]
1458
def check_versions_present(self, version_ids):
1459
"""Check that all specified versions are present."""
1461
for version_id in version_ids:
1462
if version_id not in cache:
1463
raise RevisionNotPresent(version_id, self._filename)
1466
class KnitGraphIndex(object):
1467
"""A knit index that builds on GraphIndex."""
1469
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1470
"""Construct a KnitGraphIndex on a graph_index.
1472
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1473
:param deltas: Allow delta-compressed records.
1474
:param add_callback: If not None, allow additions to the index and call
1475
this callback with a list of added GraphIndex nodes:
1476
[(node, value, node_refs), ...]
1477
:param parents: If True, record knits parents, if not do not record
1480
self._graph_index = graph_index
1481
self._deltas = deltas
1482
self._add_callback = add_callback
1483
self._parents = parents
1484
if deltas and not parents:
1485
raise KnitCorrupt(self, "Cannot do delta compression without "
1488
def _get_entries(self, keys, check_present=False):
1489
"""Get the entries for keys.
1491
:param keys: An iterable of index keys, - 1-tuples.
1496
for node in self._graph_index.iter_entries(keys):
1498
found_keys.add(node[1])
1500
# adapt parentless index to the rest of the code.
1501
for node in self._graph_index.iter_entries(keys):
1502
yield node[0], node[1], node[2], ()
1503
found_keys.add(node[1])
1505
missing_keys = keys.difference(found_keys)
1507
raise RevisionNotPresent(missing_keys.pop(), self)
1509
def _present_keys(self, version_ids):
1511
node[1] for node in self._get_entries(version_ids)])
1513
def _parentless_ancestry(self, versions):
1514
"""Honour the get_ancestry API for parentless knit indices."""
1515
wanted_keys = self._version_ids_to_keys(versions)
1516
present_keys = self._present_keys(wanted_keys)
1517
missing = set(wanted_keys).difference(present_keys)
1519
raise RevisionNotPresent(missing.pop(), self)
1520
return list(self._keys_to_version_ids(present_keys))
1522
def get_ancestry(self, versions, topo_sorted=True):
1523
"""See VersionedFile.get_ancestry."""
1524
if not self._parents:
1525
return self._parentless_ancestry(versions)
1526
# XXX: This will do len(history) index calls - perhaps
1527
# it should be altered to be a index core feature?
1528
# get a graph of all the mentioned versions:
1531
versions = self._version_ids_to_keys(versions)
1532
pending = set(versions)
1534
# get all pending nodes
1535
this_iteration = pending
1536
new_nodes = self._get_entries(this_iteration)
1539
for (index, key, value, node_refs) in new_nodes:
1540
# dont ask for ghosties - otherwise
1541
# we we can end up looping with pending
1542
# being entirely ghosted.
1543
graph[key] = [parent for parent in node_refs[0]
1544
if parent not in ghosts]
1546
for parent in graph[key]:
1547
# dont examine known nodes again
1552
ghosts.update(this_iteration.difference(found))
1553
if versions.difference(graph):
1554
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1556
result_keys = topo_sort(graph.items())
1558
result_keys = graph.iterkeys()
1559
return [key[0] for key in result_keys]
1561
def get_ancestry_with_ghosts(self, versions):
1562
"""See VersionedFile.get_ancestry."""
1563
if not self._parents:
1564
return self._parentless_ancestry(versions)
1565
# XXX: This will do len(history) index calls - perhaps
1566
# it should be altered to be a index core feature?
1567
# get a graph of all the mentioned versions:
1569
versions = self._version_ids_to_keys(versions)
1570
pending = set(versions)
1572
# get all pending nodes
1573
this_iteration = pending
1574
new_nodes = self._get_entries(this_iteration)
1576
for (index, key, value, node_refs) in new_nodes:
1577
graph[key] = node_refs[0]
1579
for parent in graph[key]:
1580
# dont examine known nodes again
1584
missing_versions = this_iteration.difference(graph)
1585
missing_needed = versions.intersection(missing_versions)
1587
raise RevisionNotPresent(missing_needed.pop(), self)
1588
for missing_version in missing_versions:
1589
# add a key, no parents
1590
graph[missing_version] = []
1591
pending.discard(missing_version) # don't look for it
1592
result_keys = topo_sort(graph.items())
1593
return [key[0] for key in result_keys]
1595
def get_graph(self):
1596
"""Return a list of the node:parents lists from this knit index."""
1597
if not self._parents:
1598
return [(key, ()) for key in self.get_versions()]
1600
for index, key, value, refs in self._graph_index.iter_all_entries():
1601
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1604
def iter_parents(self, version_ids):
1605
"""Iterate through the parents for many version ids.
1607
:param version_ids: An iterable yielding version_ids.
1608
:return: An iterator that yields (version_id, parents). Requested
1609
version_ids not present in the versioned file are simply skipped.
1610
The order is undefined, allowing for different optimisations in
1611
the underlying implementation.
1614
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1616
present_parents = set()
1617
for node in all_nodes:
1618
all_parents.update(node[3][0])
1619
# any node we are querying must be present
1620
present_parents.add(node[1])
1621
unknown_parents = all_parents.difference(present_parents)
1622
present_parents.update(self._present_keys(unknown_parents))
1623
for node in all_nodes:
1625
for parent in node[3][0]:
1626
if parent in present_parents:
1627
parents.append(parent[0])
1628
yield node[1][0], tuple(parents)
1630
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1631
yield node[1][0], ()
1633
def num_versions(self):
1634
return len(list(self._graph_index.iter_all_entries()))
1636
__len__ = num_versions
1638
def get_versions(self):
1639
"""Get all the versions in the file. not topologically sorted."""
1640
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1642
def has_version(self, version_id):
1643
"""True if the version is in the index."""
1644
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1646
def _keys_to_version_ids(self, keys):
1647
return tuple(key[0] for key in keys)
1649
def get_position(self, version_id):
1650
"""Return details needed to access the version.
1652
:return: a tuple (index, data position, size) to hand to the access
1653
logic to get the record.
1655
node = self._get_node(version_id)
1656
bits = node[2][1:].split(' ')
1657
return node[0], int(bits[0]), int(bits[1])
1659
def get_method(self, version_id):
1660
"""Return compression method of specified version."""
1661
if not self._deltas:
1663
return self._parent_compression(self._get_node(version_id)[3][1])
1665
def _parent_compression(self, reference_list):
1666
# use the second reference list to decide if this is delta'd or not.
1667
if len(reference_list):
1672
def _get_node(self, version_id):
1673
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1675
def get_options(self, version_id):
1676
"""Return a string represention options.
1680
node = self._get_node(version_id)
1681
if not self._deltas:
1682
options = ['fulltext']
1684
options = [self._parent_compression(node[3][1])]
1685
if node[2][0] == 'N':
1686
options.append('no-eol')
1689
def get_parents(self, version_id):
1690
"""Return parents of specified version ignoring ghosts."""
1691
parents = list(self.iter_parents([version_id]))
1694
raise errors.RevisionNotPresent(version_id, self)
1695
return parents[0][1]
1697
def get_parents_with_ghosts(self, version_id):
1698
"""Return parents of specified version with ghosts."""
1699
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1700
check_present=True))
1701
if not self._parents:
1703
return self._keys_to_version_ids(nodes[0][3][0])
1705
def check_versions_present(self, version_ids):
1706
"""Check that all specified versions are present."""
1707
keys = self._version_ids_to_keys(version_ids)
1708
present = self._present_keys(keys)
1709
missing = keys.difference(present)
1711
raise RevisionNotPresent(missing.pop(), self)
1713
def add_version(self, version_id, options, access_memo, parents):
1714
"""Add a version record to the index."""
1715
return self.add_versions(((version_id, options, access_memo, parents),))
1717
def add_versions(self, versions):
1718
"""Add multiple versions to the index.
1720
This function does not insert data into the Immutable GraphIndex
1721
backing the KnitGraphIndex, instead it prepares data for insertion by
1722
the caller and checks that it is safe to insert then calls
1723
self._add_callback with the prepared GraphIndex nodes.
1725
:param versions: a list of tuples:
1726
(version_id, options, pos, size, parents).
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)
1761
present_nodes = self._get_entries(keys)
1762
for (index, key, value, node_refs) in present_nodes:
1763
if (value, node_refs) != keys[key]:
1764
raise KnitCorrupt(self, "inconsistent details in add_versions"
1765
": %s %s" % ((value, node_refs), keys[key]))
1769
for key, (value, node_refs) in keys.iteritems():
1770
result.append((key, value, node_refs))
1772
for key, (value, node_refs) in keys.iteritems():
1773
result.append((key, value))
1774
self._add_callback(result)
1776
def _version_ids_to_keys(self, version_ids):
1777
return set((version_id, ) for version_id in version_ids)
1780
class _KnitAccess(object):
1781
"""Access to knit records in a .knit file."""
1783
def __init__(self, transport, filename, _file_mode, _dir_mode,
1784
_need_to_create, _create_parent_dir):
1785
"""Create a _KnitAccess for accessing and inserting data.
1787
:param transport: The transport the .knit is located on.
1788
:param filename: The filename of the .knit.
1790
self._transport = transport
1791
self._filename = filename
1792
self._file_mode = _file_mode
1793
self._dir_mode = _dir_mode
1794
self._need_to_create = _need_to_create
1795
self._create_parent_dir = _create_parent_dir
1797
def add_raw_records(self, sizes, raw_data):
1798
"""Add raw knit bytes to a storage area.
1800
The data is spooled to whereever the access method is storing data.
1802
:param sizes: An iterable containing the size of each raw data segment.
1803
:param raw_data: A bytestring containing the data.
1804
:return: A list of memos to retrieve the record later. Each memo is a
1805
tuple - (index, pos, length), where the index field is always None
1806
for the .knit access method.
1808
assert type(raw_data) == str, \
1809
'data must be plain bytes was %s' % type(raw_data)
1810
if not self._need_to_create:
1811
base = self._transport.append_bytes(self._filename, raw_data)
1813
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1814
create_parent_dir=self._create_parent_dir,
1815
mode=self._file_mode,
1816
dir_mode=self._dir_mode)
1817
self._need_to_create = False
1821
result.append((None, base, size))
1826
"""IFF this data access has its own storage area, initialise it.
1830
self._transport.put_bytes_non_atomic(self._filename, '',
1831
mode=self._file_mode)
1833
def open_file(self):
1834
"""IFF this data access can be represented as a single file, open it.
1836
For knits that are not mapped to a single file on disk this will
1839
:return: None or a file handle.
1842
return self._transport.get(self._filename)
1847
def get_raw_records(self, memos_for_retrieval):
1848
"""Get the raw bytes for a records.
1850
:param memos_for_retrieval: An iterable containing the (index, pos,
1851
length) memo for retrieving the bytes. The .knit method ignores
1852
the index as there is always only a single file.
1853
:return: An iterator over the bytes of the records.
1855
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
1856
for pos, data in self._transport.readv(self._filename, read_vector):
1860
class _PackAccess(object):
1861
"""Access to knit records via a collection of packs."""
1863
def __init__(self, index_to_packs, writer=None):
1864
"""Create a _PackAccess object.
1866
:param index_to_packs: A dict mapping index objects to the transport
1867
and file names for obtaining data.
1868
:param writer: A tuple (pack.ContainerWriter, write_index) which
1869
contains the pack to write, and the index that reads from it will
1873
self.container_writer = writer[0]
1874
self.write_index = writer[1]
1876
self.container_writer = None
1877
self.write_index = None
1878
self.indices = index_to_packs
1880
def add_raw_records(self, sizes, raw_data):
1881
"""Add raw knit bytes to a storage area.
1883
The data is spooled to the container writer in one bytes-record per
1886
:param sizes: An iterable containing the size of each raw data segment.
1887
:param raw_data: A bytestring containing the data.
1888
:return: A list of memos to retrieve the record later. Each memo is a
1889
tuple - (index, pos, length), where the index field is the
1890
write_index object supplied to the PackAccess object.
1892
assert type(raw_data) == str, \
1893
'data must be plain bytes was %s' % type(raw_data)
1897
p_offset, p_length = self.container_writer.add_bytes_record(
1898
raw_data[offset:offset+size], [])
1900
result.append((self.write_index, p_offset, p_length))
1904
"""Pack based knits do not get individually created."""
1906
def get_raw_records(self, memos_for_retrieval):
1907
"""Get the raw bytes for a records.
1909
:param memos_for_retrieval: An iterable containing the (index, pos,
1910
length) memo for retrieving the bytes. The Pack access method
1911
looks up the pack to use for a given record in its index_to_pack
1913
:return: An iterator over the bytes of the records.
1915
# first pass, group into same-index requests
1917
current_index = None
1918
for (index, offset, length) in memos_for_retrieval:
1919
if current_index == index:
1920
current_list.append((offset, length))
1922
if current_index is not None:
1923
request_lists.append((current_index, current_list))
1924
current_index = index
1925
current_list = [(offset, length)]
1926
# handle the last entry
1927
if current_index is not None:
1928
request_lists.append((current_index, current_list))
1929
for index, offsets in request_lists:
1930
transport, path = self.indices[index]
1931
reader = pack.make_readv_reader(transport, path, offsets)
1932
for names, read_func in reader.iter_records():
1933
yield read_func(None)
1935
def open_file(self):
1936
"""Pack based knits have no single file."""
1939
def set_writer(self, writer, index, (transport, packname)):
1940
"""Set a writer to use for adding data."""
1941
self.indices[index] = (transport, packname)
1942
self.container_writer = writer
1943
self.write_index = index
1946
class _KnitData(object):
1947
"""Manage extraction of data from a KnitAccess, caching and decompressing.
1949
The KnitData class provides the logic for parsing and using knit records,
1950
making use of an access method for the low level read and write operations.
1953
def __init__(self, access):
1954
"""Create a KnitData object.
1956
:param access: The access method to use. Access methods such as
1957
_KnitAccess manage the insertion of raw records and the subsequent
1958
retrieval of the same.
1960
self._access = access
1961
self._checked = False
1962
# TODO: jam 20060713 conceptually, this could spill to disk
1963
# if the cached size gets larger than a certain amount
1964
# but it complicates the model a bit, so for now just use
1965
# a simple dictionary
1967
self._do_cache = False
1969
def enable_cache(self):
1970
"""Enable caching of reads."""
1971
self._do_cache = True
1973
def clear_cache(self):
1974
"""Clear the record cache."""
1975
self._do_cache = False
1978
def _open_file(self):
1979
return self._access.open_file()
1981
def _record_to_data(self, version_id, digest, lines):
1982
"""Convert version_id, digest, lines into a raw data block.
1984
:return: (len, a StringIO instance with the raw data ready to read.)
1987
data_file = GzipFile(None, mode='wb', fileobj=sio,
1988
compresslevel=Z_DEFAULT_COMPRESSION)
1990
assert isinstance(version_id, str)
1991
data_file.writelines(chain(
1992
["version %s %d %s\n" % (version_id,
1996
["end %s\n" % version_id]))
2003
def add_raw_records(self, sizes, raw_data):
2004
"""Append a prepared record to the data file.
2006
:param sizes: An iterable containing the size of each raw data segment.
2007
:param raw_data: A bytestring containing the data.
2008
:return: a list of index data for the way the data was stored.
2009
See the access method add_raw_records documentation for more
2012
return self._access.add_raw_records(sizes, raw_data)
2014
def add_record(self, version_id, digest, lines):
2015
"""Write new text record to disk.
2017
Returns index data for retrieving it later, as per add_raw_records.
2019
size, sio = self._record_to_data(version_id, digest, lines)
2020
result = self.add_raw_records([size], sio.getvalue())
2022
self._cache[version_id] = sio.getvalue()
2025
def _parse_record_header(self, version_id, raw_data):
2026
"""Parse a record header for consistency.
2028
:return: the header and the decompressor stream.
2029
as (stream, header_record)
2031
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2033
rec = self._check_header(version_id, df.readline())
2034
except Exception, e:
2035
raise KnitCorrupt(self._access,
2036
"While reading {%s} got %s(%s)"
2037
% (version_id, e.__class__.__name__, str(e)))
2040
def _check_header(self, version_id, line):
2043
raise KnitCorrupt(self._access,
2044
'unexpected number of elements in record header')
2045
if rec[1] != version_id:
2046
raise KnitCorrupt(self._access,
2047
'unexpected version, wanted %r, got %r'
2048
% (version_id, rec[1]))
2051
def _parse_record(self, version_id, data):
2053
# 4168 calls in 2880 217 internal
2054
# 4168 calls to _parse_record_header in 2121
2055
# 4168 calls to readlines in 330
2056
df = GzipFile(mode='rb', fileobj=StringIO(data))
2059
record_contents = df.readlines()
2060
except Exception, e:
2061
raise KnitCorrupt(self._access,
2062
"While reading {%s} got %s(%s)"
2063
% (version_id, e.__class__.__name__, str(e)))
2064
header = record_contents.pop(0)
2065
rec = self._check_header(version_id, header)
2067
last_line = record_contents.pop()
2068
if len(record_contents) != int(rec[2]):
2069
raise KnitCorrupt(self._access,
2070
'incorrect number of lines %s != %s'
2072
% (len(record_contents), int(rec[2]),
2074
if last_line != 'end %s\n' % rec[1]:
2075
raise KnitCorrupt(self._access,
2076
'unexpected version end line %r, wanted %r'
2077
% (last_line, version_id))
2079
return record_contents, rec[3]
2081
def read_records_iter_raw(self, records):
2082
"""Read text records from data file and yield raw data.
2084
This unpacks enough of the text record to validate the id is
2085
as expected but thats all.
2087
# setup an iterator of the external records:
2088
# uses readv so nice and fast we hope.
2090
# grab the disk data needed.
2092
# Don't check _cache if it is empty
2093
needed_offsets = [index_memo for version_id, index_memo
2095
if version_id not in self._cache]
2097
needed_offsets = [index_memo for version_id, index_memo
2100
raw_records = self._access.get_raw_records(needed_offsets)
2102
for version_id, index_memo in records:
2103
if version_id in self._cache:
2104
# This data has already been validated
2105
data = self._cache[version_id]
2107
data = raw_records.next()
2109
self._cache[version_id] = data
2111
# validate the header
2112
df, rec = self._parse_record_header(version_id, data)
2114
yield version_id, data
2116
def read_records_iter(self, records):
2117
"""Read text records from data file and yield result.
2119
The result will be returned in whatever is the fastest to read.
2120
Not by the order requested. Also, multiple requests for the same
2121
record will only yield 1 response.
2122
:param records: A list of (version_id, pos, len) entries
2123
:return: Yields (version_id, contents, digest) in the order
2124
read, not the order requested
2130
# Skip records we have alread seen
2131
yielded_records = set()
2132
needed_records = set()
2133
for record in records:
2134
if record[0] in self._cache:
2135
if record[0] in yielded_records:
2137
yielded_records.add(record[0])
2138
data = self._cache[record[0]]
2139
content, digest = self._parse_record(record[0], data)
2140
yield (record[0], content, digest)
2142
needed_records.add(record)
2143
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2145
needed_records = sorted(set(records), key=operator.itemgetter(1))
2147
if not needed_records:
2150
# The transport optimizes the fetching as well
2151
# (ie, reads continuous ranges.)
2152
raw_data = self._access.get_raw_records(
2153
[index_memo for version_id, index_memo in needed_records])
2155
for (version_id, index_memo), data in \
2156
izip(iter(needed_records), raw_data):
2157
content, digest = self._parse_record(version_id, data)
2159
self._cache[version_id] = data
2160
yield version_id, content, digest
2162
def read_records(self, records):
2163
"""Read records into a dictionary."""
2165
for record_id, content, digest in \
2166
self.read_records_iter(records):
2167
components[record_id] = (content, digest)
2171
class InterKnit(InterVersionedFile):
2172
"""Optimised code paths for knit to knit operations."""
2174
_matching_file_from_factory = KnitVersionedFile
2175
_matching_file_to_factory = KnitVersionedFile
2178
def is_compatible(source, target):
2179
"""Be compatible with knits. """
2181
return (isinstance(source, KnitVersionedFile) and
2182
isinstance(target, KnitVersionedFile))
2183
except AttributeError:
2186
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2187
"""See InterVersionedFile.join."""
2188
assert isinstance(self.source, KnitVersionedFile)
2189
assert isinstance(self.target, KnitVersionedFile)
2191
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2196
pb = ui.ui_factory.nested_progress_bar()
2198
version_ids = list(version_ids)
2199
if None in version_ids:
2200
version_ids.remove(None)
2202
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2203
this_versions = set(self.target._index.get_versions())
2204
needed_versions = self.source_ancestry - this_versions
2205
cross_check_versions = self.source_ancestry.intersection(this_versions)
2206
mismatched_versions = set()
2207
for version in cross_check_versions:
2208
# scan to include needed parents.
2209
n1 = set(self.target.get_parents_with_ghosts(version))
2210
n2 = set(self.source.get_parents_with_ghosts(version))
2212
# FIXME TEST this check for cycles being introduced works
2213
# the logic is we have a cycle if in our graph we are an
2214
# ancestor of any of the n2 revisions.
2220
parent_ancestors = self.source.get_ancestry(parent)
2221
if version in parent_ancestors:
2222
raise errors.GraphCycleError([parent, version])
2223
# ensure this parent will be available later.
2224
new_parents = n2.difference(n1)
2225
needed_versions.update(new_parents.difference(this_versions))
2226
mismatched_versions.add(version)
2228
if not needed_versions and not mismatched_versions:
2230
full_list = topo_sort(self.source.get_graph())
2232
version_list = [i for i in full_list if (not self.target.has_version(i)
2233
and i in needed_versions)]
2237
copy_queue_records = []
2239
for version_id in version_list:
2240
options = self.source._index.get_options(version_id)
2241
parents = self.source._index.get_parents_with_ghosts(version_id)
2242
# check that its will be a consistent copy:
2243
for parent in parents:
2244
# if source has the parent, we must :
2245
# * already have it or
2246
# * have it scheduled already
2247
# otherwise we don't care
2248
assert (self.target.has_version(parent) or
2249
parent in copy_set or
2250
not self.source.has_version(parent))
2251
index_memo = self.source._index.get_position(version_id)
2252
copy_queue_records.append((version_id, index_memo))
2253
copy_queue.append((version_id, options, parents))
2254
copy_set.add(version_id)
2256
# data suck the join:
2258
total = len(version_list)
2261
for (version_id, raw_data), \
2262
(version_id2, options, parents) in \
2263
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2265
assert version_id == version_id2, 'logic error, inconsistent results'
2267
pb.update("Joining knit", count, total)
2268
raw_records.append((version_id, options, parents, len(raw_data)))
2269
raw_datum.append(raw_data)
2270
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2272
for version in mismatched_versions:
2273
# FIXME RBC 20060309 is this needed?
2274
n1 = set(self.target.get_parents_with_ghosts(version))
2275
n2 = set(self.source.get_parents_with_ghosts(version))
2276
# write a combined record to our history preserving the current
2277
# parents as first in the list
2278
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2279
self.target.fix_parents(version, new_parents)
2285
InterVersionedFile.register_optimiser(InterKnit)
2288
class WeaveToKnit(InterVersionedFile):
2289
"""Optimised code paths for weave to knit operations."""
2291
_matching_file_from_factory = bzrlib.weave.WeaveFile
2292
_matching_file_to_factory = KnitVersionedFile
2295
def is_compatible(source, target):
2296
"""Be compatible with weaves to knits."""
2298
return (isinstance(source, bzrlib.weave.Weave) and
2299
isinstance(target, KnitVersionedFile))
2300
except AttributeError:
2303
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2304
"""See InterVersionedFile.join."""
2305
assert isinstance(self.source, bzrlib.weave.Weave)
2306
assert isinstance(self.target, KnitVersionedFile)
2308
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2313
pb = ui.ui_factory.nested_progress_bar()
2315
version_ids = list(version_ids)
2317
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2318
this_versions = set(self.target._index.get_versions())
2319
needed_versions = self.source_ancestry - this_versions
2320
cross_check_versions = self.source_ancestry.intersection(this_versions)
2321
mismatched_versions = set()
2322
for version in cross_check_versions:
2323
# scan to include needed parents.
2324
n1 = set(self.target.get_parents_with_ghosts(version))
2325
n2 = set(self.source.get_parents(version))
2326
# if all of n2's parents are in n1, then its fine.
2327
if n2.difference(n1):
2328
# FIXME TEST this check for cycles being introduced works
2329
# the logic is we have a cycle if in our graph we are an
2330
# ancestor of any of the n2 revisions.
2336
parent_ancestors = self.source.get_ancestry(parent)
2337
if version in parent_ancestors:
2338
raise errors.GraphCycleError([parent, version])
2339
# ensure this parent will be available later.
2340
new_parents = n2.difference(n1)
2341
needed_versions.update(new_parents.difference(this_versions))
2342
mismatched_versions.add(version)
2344
if not needed_versions and not mismatched_versions:
2346
full_list = topo_sort(self.source.get_graph())
2348
version_list = [i for i in full_list if (not self.target.has_version(i)
2349
and i in needed_versions)]
2353
total = len(version_list)
2354
for version_id in version_list:
2355
pb.update("Converting to knit", count, total)
2356
parents = self.source.get_parents(version_id)
2357
# check that its will be a consistent copy:
2358
for parent in parents:
2359
# if source has the parent, we must already have it
2360
assert (self.target.has_version(parent))
2361
self.target.add_lines(
2362
version_id, parents, self.source.get_lines(version_id))
2365
for version in mismatched_versions:
2366
# FIXME RBC 20060309 is this needed?
2367
n1 = set(self.target.get_parents_with_ghosts(version))
2368
n2 = set(self.source.get_parents(version))
2369
# write a combined record to our history preserving the current
2370
# parents as first in the list
2371
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2372
self.target.fix_parents(version, new_parents)
2378
InterVersionedFile.register_optimiser(WeaveToKnit)
2381
# Deprecated, use PatienceSequenceMatcher instead
2382
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2385
def annotate_knit(knit, revision_id):
2386
"""Annotate a knit with no cached annotations.
2388
This implementation is for knits with no cached annotations.
2389
It will work for knits with cached annotations, but this is not
2392
ancestry = knit.get_ancestry(revision_id)
2393
fulltext = dict(zip(ancestry, knit.get_line_list(ancestry)))
2395
for candidate in ancestry:
2396
if candidate in annotations:
2398
parents = knit.get_parents(candidate)
2399
if len(parents) == 0:
2401
elif knit._index.get_method(candidate) != 'line-delta':
2404
parent, sha1, noeol, delta = knit.get_delta(candidate)
2405
blocks = KnitContent.get_line_delta_blocks(delta,
2406
fulltext[parents[0]], fulltext[candidate])
2407
annotations[candidate] = list(annotate.reannotate([annotations[p]
2408
for p in parents], fulltext[candidate], candidate, blocks))
2409
return iter(annotations[revision_id])
2413
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2415
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
