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,
835
"""See VersionedFile.add_lines_with_ghosts()."""
836
self._check_add(version_id, lines)
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):
842
"""See VersionedFile.add_lines."""
843
self._check_add(version_id, lines)
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):
849
"""check that version_id and lines are safe to add."""
850
assert self.writable, "knit is not opened for write"
851
### FIXME escape. RBC 20060228
852
if contains_whitespace(version_id):
853
raise InvalidRevisionId(version_id, self.filename)
854
self.check_not_reserved_id(version_id)
855
# Technically this is a case of Look Before You Leap, but:
856
# - for knits this saves wasted space in the error case
857
# - for packs this avoids dead space in the pack
858
# - it also avoids undetected poisoning attacks.
859
# - its 1.5% of total commit time, so ignore it unless it becomes a
861
if self.has_version(version_id):
862
raise RevisionAlreadyPresent(version_id, self.filename)
863
self._check_lines_not_unicode(lines)
864
self._check_lines_are_lines(lines)
866
def _add(self, version_id, lines, parents, delta, parent_texts,
867
left_matching_blocks, nostore_sha):
868
"""Add a set of lines on top of version specified by parents.
870
If delta is true, compress the text as a line-delta against
873
Any versions not present will be converted into ghosts.
875
# 461 0 6546.0390 43.9100 bzrlib.knit:489(_add)
876
# +400 0 889.4890 418.9790 +bzrlib.knit:192(lower_fulltext)
877
# +461 0 1364.8070 108.8030 +bzrlib.knit:996(add_record)
878
# +461 0 193.3940 41.5720 +bzrlib.knit:898(add_version)
879
# +461 0 134.0590 18.3810 +bzrlib.osutils:361(sha_strings)
880
# +461 0 36.3420 15.4540 +bzrlib.knit:146(make)
881
# +1383 0 8.0370 8.0370 +<len>
882
# +61 0 13.5770 7.9190 +bzrlib.knit:199(lower_line_delta)
883
# +61 0 963.3470 7.8740 +bzrlib.knit:427(_get_content)
884
# +61 0 973.9950 5.2950 +bzrlib.knit:136(line_delta)
885
# +61 0 1918.1800 5.2640 +bzrlib.knit:359(_merge_annotations)
889
if parent_texts is None:
891
for parent in parents:
892
if not self.has_version(parent):
893
ghosts.append(parent)
895
present_parents.append(parent)
897
if delta and not len(present_parents):
900
digest = sha_strings(lines)
901
if nostore_sha == digest:
902
raise errors.ExistingContent
903
text_length = sum(map(len, lines))
906
if lines[-1][-1] != '\n':
907
options.append('no-eol')
908
lines[-1] = lines[-1] + '\n'
910
if len(present_parents) and delta:
911
# To speed the extract of texts the delta chain is limited
912
# to a fixed number of deltas. This should minimize both
913
# I/O and the time spend applying deltas.
914
delta = self._check_should_delta(present_parents)
916
assert isinstance(version_id, str)
917
content = self.factory.make(lines, version_id)
918
if delta or (self.factory.annotated and len(present_parents) > 0):
919
# Merge annotations from parent texts if so is needed.
920
delta_hunks = self._merge_annotations(content, present_parents,
921
parent_texts, delta, self.factory.annotated,
922
left_matching_blocks)
925
options.append('line-delta')
926
store_lines = self.factory.lower_line_delta(delta_hunks)
928
options.append('fulltext')
929
store_lines = self.factory.lower_fulltext(content)
931
access_memo = self._data.add_record(version_id, digest, store_lines)
932
self._index.add_version(version_id, options, access_memo, parents)
933
return digest, text_length, content
935
def check(self, progress_bar=None):
936
"""See VersionedFile.check()."""
938
def _clone_text(self, new_version_id, old_version_id, parents):
939
"""See VersionedFile.clone_text()."""
940
# FIXME RBC 20060228 make fast by only inserting an index with null
942
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
944
def get_lines(self, version_id):
945
"""See VersionedFile.get_lines()."""
946
return self.get_line_list([version_id])[0]
948
def _get_record_map(self, version_ids):
949
"""Produce a dictionary of knit records.
951
The keys are version_ids, the values are tuples of (method, content,
953
method is the way the content should be applied.
954
content is a KnitContent object.
955
digest is the SHA1 digest of this version id after all steps are done
956
next is the build-parent of the version, i.e. the leftmost ancestor.
957
If the method is fulltext, next will be None.
959
position_map = self._get_components_positions(version_ids)
960
# c = component_id, m = method, i_m = index_memo, n = next
961
records = [(c, i_m) for c, (m, i_m, n) in position_map.iteritems()]
963
for component_id, content, digest in \
964
self._data.read_records_iter(records):
965
method, index_memo, next = position_map[component_id]
966
record_map[component_id] = method, content, digest, next
970
def get_text(self, version_id):
971
"""See VersionedFile.get_text"""
972
return self.get_texts([version_id])[0]
974
def get_texts(self, version_ids):
975
return [''.join(l) for l in self.get_line_list(version_ids)]
977
def get_line_list(self, version_ids):
978
"""Return the texts of listed versions as a list of strings."""
979
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
980
for version_id in version_ids:
981
self.check_not_reserved_id(version_id)
982
text_map, content_map = self._get_content_maps(version_ids)
983
return [text_map[v] for v in version_ids]
985
_get_lf_split_line_list = get_line_list
987
def _get_content_maps(self, version_ids):
988
"""Produce maps of text and KnitContents
990
:return: (text_map, content_map) where text_map contains the texts for
991
the requested versions and content_map contains the KnitContents.
992
Both dicts take version_ids as their keys.
994
for version_id in version_ids:
995
if not self.has_version(version_id):
996
raise RevisionNotPresent(version_id, self.filename)
997
record_map = self._get_record_map(version_ids)
1002
for version_id in version_ids:
1005
while cursor is not None:
1006
method, data, digest, next = record_map[cursor]
1007
components.append((cursor, method, data, digest))
1008
if cursor in content_map:
1013
for component_id, method, data, digest in reversed(components):
1014
if component_id in content_map:
1015
content = content_map[component_id]
1017
if method == 'fulltext':
1018
assert content is None
1019
content = self.factory.parse_fulltext(data, version_id)
1020
elif method == 'line-delta':
1021
delta = self.factory.parse_line_delta(data, version_id)
1022
content = content.copy()
1023
content._lines = self._apply_delta(content._lines,
1025
content_map[component_id] = content
1027
if 'no-eol' in self._index.get_options(version_id):
1028
content = content.copy()
1029
content.strip_last_line_newline()
1030
final_content[version_id] = content
1032
# digest here is the digest from the last applied component.
1033
text = content.text()
1034
if sha_strings(text) != digest:
1035
raise KnitCorrupt(self.filename,
1036
'sha-1 does not match %s' % version_id)
1038
text_map[version_id] = text
1039
return text_map, final_content
1041
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1043
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1044
if version_ids is None:
1045
version_ids = self.versions()
1047
version_ids = [osutils.safe_revision_id(v) for v in version_ids]
1049
pb = progress.DummyProgress()
1050
# we don't care about inclusions, the caller cares.
1051
# but we need to setup a list of records to visit.
1052
# we need version_id, position, length
1053
version_id_records = []
1054
requested_versions = set(version_ids)
1055
# filter for available versions
1056
for version_id in requested_versions:
1057
if not self.has_version(version_id):
1058
raise RevisionNotPresent(version_id, self.filename)
1059
# get a in-component-order queue:
1060
for version_id in self.versions():
1061
if version_id in requested_versions:
1062
index_memo = self._index.get_position(version_id)
1063
version_id_records.append((version_id, index_memo))
1065
total = len(version_id_records)
1066
for version_idx, (version_id, data, sha_value) in \
1067
enumerate(self._data.read_records_iter(version_id_records)):
1068
pb.update('Walking content.', version_idx, total)
1069
method = self._index.get_method(version_id)
1071
assert method in ('fulltext', 'line-delta')
1072
if method == 'fulltext':
1073
line_iterator = self.factory.get_fulltext_content(data)
1075
line_iterator = self.factory.get_linedelta_content(data)
1076
for line in line_iterator:
1079
pb.update('Walking content.', total, total)
1081
def iter_parents(self, version_ids):
1082
"""Iterate through the parents for many version ids.
1084
:param version_ids: An iterable yielding version_ids.
1085
:return: An iterator that yields (version_id, parents). Requested
1086
version_ids not present in the versioned file are simply skipped.
1087
The order is undefined, allowing for different optimisations in
1088
the underlying implementation.
1090
version_ids = [osutils.safe_revision_id(version_id) for
1091
version_id in version_ids]
1092
return self._index.iter_parents(version_ids)
1094
def num_versions(self):
1095
"""See VersionedFile.num_versions()."""
1096
return self._index.num_versions()
1098
__len__ = num_versions
1100
def annotate_iter(self, version_id):
1101
"""See VersionedFile.annotate_iter."""
1102
version_id = osutils.safe_revision_id(version_id)
1103
return self.factory.annotate_iter(self, version_id)
1105
def get_parents(self, version_id):
1106
"""See VersionedFile.get_parents."""
1109
# 52554 calls in 1264 872 internal down from 3674
1110
version_id = osutils.safe_revision_id(version_id)
1112
return self._index.get_parents(version_id)
1114
raise RevisionNotPresent(version_id, self.filename)
1116
def get_parents_with_ghosts(self, version_id):
1117
"""See VersionedFile.get_parents."""
1118
version_id = osutils.safe_revision_id(version_id)
1120
return self._index.get_parents_with_ghosts(version_id)
1122
raise RevisionNotPresent(version_id, self.filename)
1124
def get_ancestry(self, versions, topo_sorted=True):
1125
"""See VersionedFile.get_ancestry."""
1126
if isinstance(versions, basestring):
1127
versions = [versions]
1130
versions = [osutils.safe_revision_id(v) for v in versions]
1131
return self._index.get_ancestry(versions, topo_sorted)
1133
def get_ancestry_with_ghosts(self, versions):
1134
"""See VersionedFile.get_ancestry_with_ghosts."""
1135
if isinstance(versions, basestring):
1136
versions = [versions]
1139
versions = [osutils.safe_revision_id(v) for v in versions]
1140
return self._index.get_ancestry_with_ghosts(versions)
1142
def plan_merge(self, ver_a, ver_b):
1143
"""See VersionedFile.plan_merge."""
1144
ver_a = osutils.safe_revision_id(ver_a)
1145
ver_b = osutils.safe_revision_id(ver_b)
1146
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):
1379
"""Add multiple versions to the index.
1381
:param versions: a list of tuples:
1382
(version_id, options, pos, size, parents).
1385
orig_history = self._history[:]
1386
orig_cache = self._cache.copy()
1389
for version_id, options, (index, pos, size), parents in versions:
1390
line = "\n%s %s %s %s %s :" % (version_id,
1394
self._version_list_to_index(parents))
1395
assert isinstance(line, str), \
1396
'content must be utf-8 encoded: %r' % (line,)
1398
self._cache_version(version_id, options, pos, size, parents)
1399
if not self._need_to_create:
1400
self._transport.append_bytes(self._filename, ''.join(lines))
1403
sio.write(self.HEADER)
1404
sio.writelines(lines)
1406
self._transport.put_file_non_atomic(self._filename, sio,
1407
create_parent_dir=self._create_parent_dir,
1408
mode=self._file_mode,
1409
dir_mode=self._dir_mode)
1410
self._need_to_create = False
1412
# If any problems happen, restore the original values and re-raise
1413
self._history = orig_history
1414
self._cache = orig_cache
1417
def has_version(self, version_id):
1418
"""True if the version is in the index."""
1419
return version_id in self._cache
1421
def get_position(self, version_id):
1422
"""Return details needed to access the version.
1424
.kndx indices do not support split-out data, so return None for the
1427
:return: a tuple (None, data position, size) to hand to the access
1428
logic to get the record.
1430
entry = self._cache[version_id]
1431
return None, entry[2], entry[3]
1433
def get_method(self, version_id):
1434
"""Return compression method of specified version."""
1435
options = self._cache[version_id][1]
1436
if 'fulltext' in options:
1439
if 'line-delta' not in options:
1440
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1443
def get_options(self, version_id):
1444
"""Return a string represention options.
1448
return self._cache[version_id][1]
1450
def get_parents(self, version_id):
1451
"""Return parents of specified version ignoring ghosts."""
1452
return [parent for parent in self._cache[version_id][4]
1453
if parent in self._cache]
1455
def get_parents_with_ghosts(self, version_id):
1456
"""Return parents of specified version with ghosts."""
1457
return self._cache[version_id][4]
1459
def check_versions_present(self, version_ids):
1460
"""Check that all specified versions are present."""
1462
for version_id in version_ids:
1463
if version_id not in cache:
1464
raise RevisionNotPresent(version_id, self._filename)
1467
class KnitGraphIndex(object):
1468
"""A knit index that builds on GraphIndex."""
1470
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1471
"""Construct a KnitGraphIndex on a graph_index.
1473
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1474
:param deltas: Allow delta-compressed records.
1475
:param add_callback: If not None, allow additions to the index and call
1476
this callback with a list of added GraphIndex nodes:
1477
[(node, value, node_refs), ...]
1478
:param parents: If True, record knits parents, if not do not record
1481
self._graph_index = graph_index
1482
self._deltas = deltas
1483
self._add_callback = add_callback
1484
self._parents = parents
1485
if deltas and not parents:
1486
raise KnitCorrupt(self, "Cannot do delta compression without "
1489
def _get_entries(self, keys, check_present=False):
1490
"""Get the entries for keys.
1492
:param keys: An iterable of index keys, - 1-tuples.
1497
for node in self._graph_index.iter_entries(keys):
1499
found_keys.add(node[1])
1501
# adapt parentless index to the rest of the code.
1502
for node in self._graph_index.iter_entries(keys):
1503
yield node[0], node[1], node[2], ()
1504
found_keys.add(node[1])
1506
missing_keys = keys.difference(found_keys)
1508
raise RevisionNotPresent(missing_keys.pop(), self)
1510
def _present_keys(self, version_ids):
1512
node[1] for node in self._get_entries(version_ids)])
1514
def _parentless_ancestry(self, versions):
1515
"""Honour the get_ancestry API for parentless knit indices."""
1516
wanted_keys = self._version_ids_to_keys(versions)
1517
present_keys = self._present_keys(wanted_keys)
1518
missing = set(wanted_keys).difference(present_keys)
1520
raise RevisionNotPresent(missing.pop(), self)
1521
return list(self._keys_to_version_ids(present_keys))
1523
def get_ancestry(self, versions, topo_sorted=True):
1524
"""See VersionedFile.get_ancestry."""
1525
if not self._parents:
1526
return self._parentless_ancestry(versions)
1527
# XXX: This will do len(history) index calls - perhaps
1528
# it should be altered to be a index core feature?
1529
# get a graph of all the mentioned versions:
1532
versions = self._version_ids_to_keys(versions)
1533
pending = set(versions)
1535
# get all pending nodes
1536
this_iteration = pending
1537
new_nodes = self._get_entries(this_iteration)
1540
for (index, key, value, node_refs) in new_nodes:
1541
# dont ask for ghosties - otherwise
1542
# we we can end up looping with pending
1543
# being entirely ghosted.
1544
graph[key] = [parent for parent in node_refs[0]
1545
if parent not in ghosts]
1547
for parent in graph[key]:
1548
# dont examine known nodes again
1553
ghosts.update(this_iteration.difference(found))
1554
if versions.difference(graph):
1555
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1557
result_keys = topo_sort(graph.items())
1559
result_keys = graph.iterkeys()
1560
return [key[0] for key in result_keys]
1562
def get_ancestry_with_ghosts(self, versions):
1563
"""See VersionedFile.get_ancestry."""
1564
if not self._parents:
1565
return self._parentless_ancestry(versions)
1566
# XXX: This will do len(history) index calls - perhaps
1567
# it should be altered to be a index core feature?
1568
# get a graph of all the mentioned versions:
1570
versions = self._version_ids_to_keys(versions)
1571
pending = set(versions)
1573
# get all pending nodes
1574
this_iteration = pending
1575
new_nodes = self._get_entries(this_iteration)
1577
for (index, key, value, node_refs) in new_nodes:
1578
graph[key] = node_refs[0]
1580
for parent in graph[key]:
1581
# dont examine known nodes again
1585
missing_versions = this_iteration.difference(graph)
1586
missing_needed = versions.intersection(missing_versions)
1588
raise RevisionNotPresent(missing_needed.pop(), self)
1589
for missing_version in missing_versions:
1590
# add a key, no parents
1591
graph[missing_version] = []
1592
pending.discard(missing_version) # don't look for it
1593
result_keys = topo_sort(graph.items())
1594
return [key[0] for key in result_keys]
1596
def get_graph(self):
1597
"""Return a list of the node:parents lists from this knit index."""
1598
if not self._parents:
1599
return [(key, ()) for key in self.get_versions()]
1601
for index, key, value, refs in self._graph_index.iter_all_entries():
1602
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1605
def iter_parents(self, version_ids):
1606
"""Iterate through the parents for many version ids.
1608
:param version_ids: An iterable yielding version_ids.
1609
:return: An iterator that yields (version_id, parents). Requested
1610
version_ids not present in the versioned file are simply skipped.
1611
The order is undefined, allowing for different optimisations in
1612
the underlying implementation.
1615
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1617
present_parents = set()
1618
for node in all_nodes:
1619
all_parents.update(node[3][0])
1620
# any node we are querying must be present
1621
present_parents.add(node[1])
1622
unknown_parents = all_parents.difference(present_parents)
1623
present_parents.update(self._present_keys(unknown_parents))
1624
for node in all_nodes:
1626
for parent in node[3][0]:
1627
if parent in present_parents:
1628
parents.append(parent[0])
1629
yield node[1][0], tuple(parents)
1631
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1632
yield node[1][0], ()
1634
def num_versions(self):
1635
return len(list(self._graph_index.iter_all_entries()))
1637
__len__ = num_versions
1639
def get_versions(self):
1640
"""Get all the versions in the file. not topologically sorted."""
1641
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1643
def has_version(self, version_id):
1644
"""True if the version is in the index."""
1645
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1647
def _keys_to_version_ids(self, keys):
1648
return tuple(key[0] for key in keys)
1650
def get_position(self, version_id):
1651
"""Return details needed to access the version.
1653
:return: a tuple (index, data position, size) to hand to the access
1654
logic to get the record.
1656
node = self._get_node(version_id)
1657
bits = node[2][1:].split(' ')
1658
return node[0], int(bits[0]), int(bits[1])
1660
def get_method(self, version_id):
1661
"""Return compression method of specified version."""
1662
if not self._deltas:
1664
return self._parent_compression(self._get_node(version_id)[3][1])
1666
def _parent_compression(self, reference_list):
1667
# use the second reference list to decide if this is delta'd or not.
1668
if len(reference_list):
1673
def _get_node(self, version_id):
1674
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1676
def get_options(self, version_id):
1677
"""Return a string represention options.
1681
node = self._get_node(version_id)
1682
if not self._deltas:
1683
options = ['fulltext']
1685
options = [self._parent_compression(node[3][1])]
1686
if node[2][0] == 'N':
1687
options.append('no-eol')
1690
def get_parents(self, version_id):
1691
"""Return parents of specified version ignoring ghosts."""
1692
parents = list(self.iter_parents([version_id]))
1695
raise errors.RevisionNotPresent(version_id, self)
1696
return parents[0][1]
1698
def get_parents_with_ghosts(self, version_id):
1699
"""Return parents of specified version with ghosts."""
1700
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1701
check_present=True))
1702
if not self._parents:
1704
return self._keys_to_version_ids(nodes[0][3][0])
1706
def check_versions_present(self, version_ids):
1707
"""Check that all specified versions are present."""
1708
keys = self._version_ids_to_keys(version_ids)
1709
present = self._present_keys(keys)
1710
missing = keys.difference(present)
1712
raise RevisionNotPresent(missing.pop(), self)
1714
def add_version(self, version_id, options, access_memo, parents):
1715
"""Add a version record to the index."""
1716
return self.add_versions(((version_id, options, access_memo, parents),))
1718
def add_versions(self, versions):
1719
"""Add multiple versions to the index.
1721
This function does not insert data into the Immutable GraphIndex
1722
backing the KnitGraphIndex, instead it prepares data for insertion by
1723
the caller and checks that it is safe to insert then calls
1724
self._add_callback with the prepared GraphIndex nodes.
1726
:param versions: a list of tuples:
1727
(version_id, options, pos, size, parents).
1729
if not self._add_callback:
1730
raise errors.ReadOnlyError(self)
1731
# we hope there are no repositories with inconsistent parentage
1736
for (version_id, options, access_memo, parents) in versions:
1737
index, pos, size = access_memo
1738
key = (version_id, )
1739
parents = tuple((parent, ) for parent in parents)
1740
if 'no-eol' in options:
1744
value += "%d %d" % (pos, size)
1745
if not self._deltas:
1746
if 'line-delta' in options:
1747
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1750
if 'line-delta' in options:
1751
node_refs = (parents, (parents[0],))
1753
node_refs = (parents, ())
1755
node_refs = (parents, )
1758
raise KnitCorrupt(self, "attempt to add node with parents "
1759
"in parentless index.")
1761
keys[key] = (value, node_refs)
1762
present_nodes = self._get_entries(keys)
1763
for (index, key, value, node_refs) in present_nodes:
1764
if (value, node_refs) != keys[key]:
1765
raise KnitCorrupt(self, "inconsistent details in add_versions"
1766
": %s %s" % ((value, node_refs), keys[key]))
1770
for key, (value, node_refs) in keys.iteritems():
1771
result.append((key, value, node_refs))
1773
for key, (value, node_refs) in keys.iteritems():
1774
result.append((key, value))
1775
self._add_callback(result)
1777
def _version_ids_to_keys(self, version_ids):
1778
return set((version_id, ) for version_id in version_ids)
1781
class _KnitAccess(object):
1782
"""Access to knit records in a .knit file."""
1784
def __init__(self, transport, filename, _file_mode, _dir_mode,
1785
_need_to_create, _create_parent_dir):
1786
"""Create a _KnitAccess for accessing and inserting data.
1788
:param transport: The transport the .knit is located on.
1789
:param filename: The filename of the .knit.
1791
self._transport = transport
1792
self._filename = filename
1793
self._file_mode = _file_mode
1794
self._dir_mode = _dir_mode
1795
self._need_to_create = _need_to_create
1796
self._create_parent_dir = _create_parent_dir
1798
def add_raw_records(self, sizes, raw_data):
1799
"""Add raw knit bytes to a storage area.
1801
The data is spooled to whereever the access method is storing data.
1803
:param sizes: An iterable containing the size of each raw data segment.
1804
:param raw_data: A bytestring containing the data.
1805
:return: A list of memos to retrieve the record later. Each memo is a
1806
tuple - (index, pos, length), where the index field is always None
1807
for the .knit access method.
1809
assert type(raw_data) == str, \
1810
'data must be plain bytes was %s' % type(raw_data)
1811
if not self._need_to_create:
1812
base = self._transport.append_bytes(self._filename, raw_data)
1814
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1815
create_parent_dir=self._create_parent_dir,
1816
mode=self._file_mode,
1817
dir_mode=self._dir_mode)
1818
self._need_to_create = False
1822
result.append((None, base, size))
1827
"""IFF this data access has its own storage area, initialise it.
1831
self._transport.put_bytes_non_atomic(self._filename, '',
1832
mode=self._file_mode)
1834
def open_file(self):
1835
"""IFF this data access can be represented as a single file, open it.
1837
For knits that are not mapped to a single file on disk this will
1840
:return: None or a file handle.
1843
return self._transport.get(self._filename)
1848
def get_raw_records(self, memos_for_retrieval):
1849
"""Get the raw bytes for a records.
1851
:param memos_for_retrieval: An iterable containing the (index, pos,
1852
length) memo for retrieving the bytes. The .knit method ignores
1853
the index as there is always only a single file.
1854
:return: An iterator over the bytes of the records.
1856
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
1857
for pos, data in self._transport.readv(self._filename, read_vector):
1861
class _PackAccess(object):
1862
"""Access to knit records via a collection of packs."""
1864
def __init__(self, index_to_packs, writer=None):
1865
"""Create a _PackAccess object.
1867
:param index_to_packs: A dict mapping index objects to the transport
1868
and file names for obtaining data.
1869
:param writer: A tuple (pack.ContainerWriter, write_index) which
1870
contains the pack to write, and the index that reads from it will
1874
self.container_writer = writer[0]
1875
self.write_index = writer[1]
1877
self.container_writer = None
1878
self.write_index = None
1879
self.indices = index_to_packs
1881
def add_raw_records(self, sizes, raw_data):
1882
"""Add raw knit bytes to a storage area.
1884
The data is spooled to the container writer in one bytes-record per
1887
:param sizes: An iterable containing the size of each raw data segment.
1888
:param raw_data: A bytestring containing the data.
1889
:return: A list of memos to retrieve the record later. Each memo is a
1890
tuple - (index, pos, length), where the index field is the
1891
write_index object supplied to the PackAccess object.
1893
assert type(raw_data) == str, \
1894
'data must be plain bytes was %s' % type(raw_data)
1898
p_offset, p_length = self.container_writer.add_bytes_record(
1899
raw_data[offset:offset+size], [])
1901
result.append((self.write_index, p_offset, p_length))
1905
"""Pack based knits do not get individually created."""
1907
def get_raw_records(self, memos_for_retrieval):
1908
"""Get the raw bytes for a records.
1910
:param memos_for_retrieval: An iterable containing the (index, pos,
1911
length) memo for retrieving the bytes. The Pack access method
1912
looks up the pack to use for a given record in its index_to_pack
1914
:return: An iterator over the bytes of the records.
1916
# first pass, group into same-index requests
1918
current_index = None
1919
for (index, offset, length) in memos_for_retrieval:
1920
if current_index == index:
1921
current_list.append((offset, length))
1923
if current_index is not None:
1924
request_lists.append((current_index, current_list))
1925
current_index = index
1926
current_list = [(offset, length)]
1927
# handle the last entry
1928
if current_index is not None:
1929
request_lists.append((current_index, current_list))
1930
for index, offsets in request_lists:
1931
transport, path = self.indices[index]
1932
reader = pack.make_readv_reader(transport, path, offsets)
1933
for names, read_func in reader.iter_records():
1934
yield read_func(None)
1936
def open_file(self):
1937
"""Pack based knits have no single file."""
1940
def set_writer(self, writer, index, (transport, packname)):
1941
"""Set a writer to use for adding data."""
1942
self.indices[index] = (transport, packname)
1943
self.container_writer = writer
1944
self.write_index = index
1947
class _KnitData(object):
1948
"""Manage extraction of data from a KnitAccess, caching and decompressing.
1950
The KnitData class provides the logic for parsing and using knit records,
1951
making use of an access method for the low level read and write operations.
1954
def __init__(self, access):
1955
"""Create a KnitData object.
1957
:param access: The access method to use. Access methods such as
1958
_KnitAccess manage the insertion of raw records and the subsequent
1959
retrieval of the same.
1961
self._access = access
1962
self._checked = False
1963
# TODO: jam 20060713 conceptually, this could spill to disk
1964
# if the cached size gets larger than a certain amount
1965
# but it complicates the model a bit, so for now just use
1966
# a simple dictionary
1968
self._do_cache = False
1970
def enable_cache(self):
1971
"""Enable caching of reads."""
1972
self._do_cache = True
1974
def clear_cache(self):
1975
"""Clear the record cache."""
1976
self._do_cache = False
1979
def _open_file(self):
1980
return self._access.open_file()
1982
def _record_to_data(self, version_id, digest, lines):
1983
"""Convert version_id, digest, lines into a raw data block.
1985
:return: (len, a StringIO instance with the raw data ready to read.)
1988
data_file = GzipFile(None, mode='wb', fileobj=sio,
1989
compresslevel=Z_DEFAULT_COMPRESSION)
1991
assert isinstance(version_id, str)
1992
data_file.writelines(chain(
1993
["version %s %d %s\n" % (version_id,
1997
["end %s\n" % version_id]))
2004
def add_raw_records(self, sizes, raw_data):
2005
"""Append a prepared record to the data file.
2007
:param sizes: An iterable containing the size of each raw data segment.
2008
:param raw_data: A bytestring containing the data.
2009
:return: a list of index data for the way the data was stored.
2010
See the access method add_raw_records documentation for more
2013
return self._access.add_raw_records(sizes, raw_data)
2015
def add_record(self, version_id, digest, lines):
2016
"""Write new text record to disk.
2018
Returns index data for retrieving it later, as per add_raw_records.
2020
size, sio = self._record_to_data(version_id, digest, lines)
2021
result = self.add_raw_records([size], sio.getvalue())
2023
self._cache[version_id] = sio.getvalue()
2026
def _parse_record_header(self, version_id, raw_data):
2027
"""Parse a record header for consistency.
2029
:return: the header and the decompressor stream.
2030
as (stream, header_record)
2032
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2034
rec = self._check_header(version_id, df.readline())
2035
except Exception, e:
2036
raise KnitCorrupt(self._access,
2037
"While reading {%s} got %s(%s)"
2038
% (version_id, e.__class__.__name__, str(e)))
2041
def _check_header(self, version_id, line):
2044
raise KnitCorrupt(self._access,
2045
'unexpected number of elements in record header')
2046
if rec[1] != version_id:
2047
raise KnitCorrupt(self._access,
2048
'unexpected version, wanted %r, got %r'
2049
% (version_id, rec[1]))
2052
def _parse_record(self, version_id, data):
2054
# 4168 calls in 2880 217 internal
2055
# 4168 calls to _parse_record_header in 2121
2056
# 4168 calls to readlines in 330
2057
df = GzipFile(mode='rb', fileobj=StringIO(data))
2060
record_contents = df.readlines()
2061
except Exception, e:
2062
raise KnitCorrupt(self._access,
2063
"While reading {%s} got %s(%s)"
2064
% (version_id, e.__class__.__name__, str(e)))
2065
header = record_contents.pop(0)
2066
rec = self._check_header(version_id, header)
2068
last_line = record_contents.pop()
2069
if len(record_contents) != int(rec[2]):
2070
raise KnitCorrupt(self._access,
2071
'incorrect number of lines %s != %s'
2073
% (len(record_contents), int(rec[2]),
2075
if last_line != 'end %s\n' % rec[1]:
2076
raise KnitCorrupt(self._access,
2077
'unexpected version end line %r, wanted %r'
2078
% (last_line, version_id))
2080
return record_contents, rec[3]
2082
def read_records_iter_raw(self, records):
2083
"""Read text records from data file and yield raw data.
2085
This unpacks enough of the text record to validate the id is
2086
as expected but thats all.
2088
# setup an iterator of the external records:
2089
# uses readv so nice and fast we hope.
2091
# grab the disk data needed.
2093
# Don't check _cache if it is empty
2094
needed_offsets = [index_memo for version_id, index_memo
2096
if version_id not in self._cache]
2098
needed_offsets = [index_memo for version_id, index_memo
2101
raw_records = self._access.get_raw_records(needed_offsets)
2103
for version_id, index_memo in records:
2104
if version_id in self._cache:
2105
# This data has already been validated
2106
data = self._cache[version_id]
2108
data = raw_records.next()
2110
self._cache[version_id] = data
2112
# validate the header
2113
df, rec = self._parse_record_header(version_id, data)
2115
yield version_id, data
2117
def read_records_iter(self, records):
2118
"""Read text records from data file and yield result.
2120
The result will be returned in whatever is the fastest to read.
2121
Not by the order requested. Also, multiple requests for the same
2122
record will only yield 1 response.
2123
:param records: A list of (version_id, pos, len) entries
2124
:return: Yields (version_id, contents, digest) in the order
2125
read, not the order requested
2131
# Skip records we have alread seen
2132
yielded_records = set()
2133
needed_records = set()
2134
for record in records:
2135
if record[0] in self._cache:
2136
if record[0] in yielded_records:
2138
yielded_records.add(record[0])
2139
data = self._cache[record[0]]
2140
content, digest = self._parse_record(record[0], data)
2141
yield (record[0], content, digest)
2143
needed_records.add(record)
2144
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2146
needed_records = sorted(set(records), key=operator.itemgetter(1))
2148
if not needed_records:
2151
# The transport optimizes the fetching as well
2152
# (ie, reads continuous ranges.)
2153
raw_data = self._access.get_raw_records(
2154
[index_memo for version_id, index_memo in needed_records])
2156
for (version_id, index_memo), data in \
2157
izip(iter(needed_records), raw_data):
2158
content, digest = self._parse_record(version_id, data)
2160
self._cache[version_id] = data
2161
yield version_id, content, digest
2163
def read_records(self, records):
2164
"""Read records into a dictionary."""
2166
for record_id, content, digest in \
2167
self.read_records_iter(records):
2168
components[record_id] = (content, digest)
2172
class InterKnit(InterVersionedFile):
2173
"""Optimised code paths for knit to knit operations."""
2175
_matching_file_from_factory = KnitVersionedFile
2176
_matching_file_to_factory = KnitVersionedFile
2179
def is_compatible(source, target):
2180
"""Be compatible with knits. """
2182
return (isinstance(source, KnitVersionedFile) and
2183
isinstance(target, KnitVersionedFile))
2184
except AttributeError:
2187
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2188
"""See InterVersionedFile.join."""
2189
assert isinstance(self.source, KnitVersionedFile)
2190
assert isinstance(self.target, KnitVersionedFile)
2192
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2197
pb = ui.ui_factory.nested_progress_bar()
2199
version_ids = list(version_ids)
2200
if None in version_ids:
2201
version_ids.remove(None)
2203
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2204
this_versions = set(self.target._index.get_versions())
2205
needed_versions = self.source_ancestry - this_versions
2206
cross_check_versions = self.source_ancestry.intersection(this_versions)
2207
mismatched_versions = set()
2208
for version in cross_check_versions:
2209
# scan to include needed parents.
2210
n1 = set(self.target.get_parents_with_ghosts(version))
2211
n2 = set(self.source.get_parents_with_ghosts(version))
2213
# FIXME TEST this check for cycles being introduced works
2214
# the logic is we have a cycle if in our graph we are an
2215
# ancestor of any of the n2 revisions.
2221
parent_ancestors = self.source.get_ancestry(parent)
2222
if version in parent_ancestors:
2223
raise errors.GraphCycleError([parent, version])
2224
# ensure this parent will be available later.
2225
new_parents = n2.difference(n1)
2226
needed_versions.update(new_parents.difference(this_versions))
2227
mismatched_versions.add(version)
2229
if not needed_versions and not mismatched_versions:
2231
full_list = topo_sort(self.source.get_graph())
2233
version_list = [i for i in full_list if (not self.target.has_version(i)
2234
and i in needed_versions)]
2238
copy_queue_records = []
2240
for version_id in version_list:
2241
options = self.source._index.get_options(version_id)
2242
parents = self.source._index.get_parents_with_ghosts(version_id)
2243
# check that its will be a consistent copy:
2244
for parent in parents:
2245
# if source has the parent, we must :
2246
# * already have it or
2247
# * have it scheduled already
2248
# otherwise we don't care
2249
assert (self.target.has_version(parent) or
2250
parent in copy_set or
2251
not self.source.has_version(parent))
2252
index_memo = self.source._index.get_position(version_id)
2253
copy_queue_records.append((version_id, index_memo))
2254
copy_queue.append((version_id, options, parents))
2255
copy_set.add(version_id)
2257
# data suck the join:
2259
total = len(version_list)
2262
for (version_id, raw_data), \
2263
(version_id2, options, parents) in \
2264
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2266
assert version_id == version_id2, 'logic error, inconsistent results'
2268
pb.update("Joining knit", count, total)
2269
raw_records.append((version_id, options, parents, len(raw_data)))
2270
raw_datum.append(raw_data)
2271
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2273
for version in mismatched_versions:
2274
# FIXME RBC 20060309 is this needed?
2275
n1 = set(self.target.get_parents_with_ghosts(version))
2276
n2 = set(self.source.get_parents_with_ghosts(version))
2277
# write a combined record to our history preserving the current
2278
# parents as first in the list
2279
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2280
self.target.fix_parents(version, new_parents)
2286
InterVersionedFile.register_optimiser(InterKnit)
2289
class WeaveToKnit(InterVersionedFile):
2290
"""Optimised code paths for weave to knit operations."""
2292
_matching_file_from_factory = bzrlib.weave.WeaveFile
2293
_matching_file_to_factory = KnitVersionedFile
2296
def is_compatible(source, target):
2297
"""Be compatible with weaves to knits."""
2299
return (isinstance(source, bzrlib.weave.Weave) and
2300
isinstance(target, KnitVersionedFile))
2301
except AttributeError:
2304
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2305
"""See InterVersionedFile.join."""
2306
assert isinstance(self.source, bzrlib.weave.Weave)
2307
assert isinstance(self.target, KnitVersionedFile)
2309
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2314
pb = ui.ui_factory.nested_progress_bar()
2316
version_ids = list(version_ids)
2318
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2319
this_versions = set(self.target._index.get_versions())
2320
needed_versions = self.source_ancestry - this_versions
2321
cross_check_versions = self.source_ancestry.intersection(this_versions)
2322
mismatched_versions = set()
2323
for version in cross_check_versions:
2324
# scan to include needed parents.
2325
n1 = set(self.target.get_parents_with_ghosts(version))
2326
n2 = set(self.source.get_parents(version))
2327
# if all of n2's parents are in n1, then its fine.
2328
if n2.difference(n1):
2329
# FIXME TEST this check for cycles being introduced works
2330
# the logic is we have a cycle if in our graph we are an
2331
# ancestor of any of the n2 revisions.
2337
parent_ancestors = self.source.get_ancestry(parent)
2338
if version in parent_ancestors:
2339
raise errors.GraphCycleError([parent, version])
2340
# ensure this parent will be available later.
2341
new_parents = n2.difference(n1)
2342
needed_versions.update(new_parents.difference(this_versions))
2343
mismatched_versions.add(version)
2345
if not needed_versions and not mismatched_versions:
2347
full_list = topo_sort(self.source.get_graph())
2349
version_list = [i for i in full_list if (not self.target.has_version(i)
2350
and i in needed_versions)]
2354
total = len(version_list)
2355
for version_id in version_list:
2356
pb.update("Converting to knit", count, total)
2357
parents = self.source.get_parents(version_id)
2358
# check that its will be a consistent copy:
2359
for parent in parents:
2360
# if source has the parent, we must already have it
2361
assert (self.target.has_version(parent))
2362
self.target.add_lines(
2363
version_id, parents, self.source.get_lines(version_id))
2366
for version in mismatched_versions:
2367
# FIXME RBC 20060309 is this needed?
2368
n1 = set(self.target.get_parents_with_ghosts(version))
2369
n2 = set(self.source.get_parents(version))
2370
# write a combined record to our history preserving the current
2371
# parents as first in the list
2372
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
2373
self.target.fix_parents(version, new_parents)
2379
InterVersionedFile.register_optimiser(WeaveToKnit)
2382
# Deprecated, use PatienceSequenceMatcher instead
2383
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2386
def annotate_knit(knit, revision_id):
2387
"""Annotate a knit with no cached annotations.
2389
This implementation is for knits with no cached annotations.
2390
It will work for knits with cached annotations, but this is not
2393
ancestry = knit.get_ancestry(revision_id)
2394
fulltext = dict(zip(ancestry, knit.get_line_list(ancestry)))
2396
for candidate in ancestry:
2397
if candidate in annotations:
2399
parents = knit.get_parents(candidate)
2400
if len(parents) == 0:
2402
elif knit._index.get_method(candidate) != 'line-delta':
2405
parent, sha1, noeol, delta = knit.get_delta(candidate)
2406
blocks = KnitContent.get_line_delta_blocks(delta,
2407
fulltext[parents[0]], fulltext[candidate])
2408
annotations[candidate] = list(annotate.reannotate([annotations[p]
2409
for p in parents], fulltext[candidate], candidate, blocks))
2410
return iter(annotations[revision_id])
2414
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2416
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
