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(), """
94
from bzrlib.errors import (
102
RevisionAlreadyPresent,
104
from bzrlib.tuned_gzip import GzipFile, bytes_to_gzip
105
from bzrlib.osutils import (
111
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
112
from bzrlib.tsort import topo_sort
115
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
118
# TODO: Split out code specific to this format into an associated object.
120
# TODO: Can we put in some kind of value to check that the index and data
121
# files belong together?
123
# TODO: accommodate binaries, perhaps by storing a byte count
125
# TODO: function to check whole file
127
# TODO: atomically append data, then measure backwards from the cursor
128
# position after writing to work out where it was located. we may need to
129
# bypass python file buffering.
131
DATA_SUFFIX = '.knit'
132
INDEX_SUFFIX = '.kndx'
135
class KnitContent(object):
136
"""Content of a knit version to which deltas can be applied."""
139
"""Return a list of (origin, text) tuples."""
140
return list(self.annotate_iter())
142
def apply_delta(self, delta, new_version_id):
143
"""Apply delta to this object to become new_version_id."""
144
raise NotImplementedError(self.apply_delta)
146
def line_delta_iter(self, new_lines):
147
"""Generate line-based delta from this content to new_lines."""
148
new_texts = new_lines.text()
149
old_texts = self.text()
150
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
151
for tag, i1, i2, j1, j2 in s.get_opcodes():
154
# ofrom, oto, length, data
155
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
157
def line_delta(self, new_lines):
158
return list(self.line_delta_iter(new_lines))
161
def get_line_delta_blocks(knit_delta, source, target):
162
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
163
target_len = len(target)
166
for s_begin, s_end, t_len, new_text in knit_delta:
167
true_n = s_begin - s_pos
170
# knit deltas do not provide reliable info about whether the
171
# last line of a file matches, due to eol handling.
172
if source[s_pos + n -1] != target[t_pos + n -1]:
175
yield s_pos, t_pos, n
176
t_pos += t_len + true_n
178
n = target_len - t_pos
180
if source[s_pos + n -1] != target[t_pos + n -1]:
183
yield s_pos, t_pos, n
184
yield s_pos + (target_len - t_pos), target_len, 0
187
class AnnotatedKnitContent(KnitContent):
188
"""Annotated content."""
190
def __init__(self, lines):
193
def annotate_iter(self):
194
"""Yield tuples of (origin, text) for each content line."""
195
return iter(self._lines)
197
def apply_delta(self, delta, new_version_id):
198
"""Apply delta to this object to become new_version_id."""
201
for start, end, count, delta_lines in delta:
202
lines[offset+start:offset+end] = delta_lines
203
offset = offset + (start - end) + count
205
def strip_last_line_newline(self):
206
line = self._lines[-1][1].rstrip('\n')
207
self._lines[-1] = (self._lines[-1][0], line)
211
return [text for origin, text in self._lines]
212
except ValueError, e:
213
# most commonly (only?) caused by the internal form of the knit
214
# missing annotation information because of a bug - see thread
216
raise KnitCorrupt(self,
217
"line in annotated knit missing annotation information: %s"
221
return AnnotatedKnitContent(self._lines[:])
224
class PlainKnitContent(KnitContent):
225
"""Unannotated content.
227
When annotate[_iter] is called on this content, the same version is reported
228
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
232
def __init__(self, lines, version_id):
234
self._version_id = version_id
236
def annotate_iter(self):
237
"""Yield tuples of (origin, text) for each content line."""
238
for line in self._lines:
239
yield self._version_id, line
241
def apply_delta(self, delta, new_version_id):
242
"""Apply delta to this object to become new_version_id."""
245
for start, end, count, delta_lines in delta:
246
lines[offset+start:offset+end] = delta_lines
247
offset = offset + (start - end) + count
248
self._version_id = new_version_id
251
return PlainKnitContent(self._lines[:], self._version_id)
253
def strip_last_line_newline(self):
254
self._lines[-1] = self._lines[-1].rstrip('\n')
260
class KnitAnnotateFactory(object):
261
"""Factory for creating annotated Content objects."""
265
def make(self, lines, version_id):
266
num_lines = len(lines)
267
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
269
def parse_fulltext(self, content, version_id):
270
"""Convert fulltext to internal representation
272
fulltext content is of the format
273
revid(utf8) plaintext\n
274
internal representation is of the format:
277
# TODO: jam 20070209 The tests expect this to be returned as tuples,
278
# but the code itself doesn't really depend on that.
279
# Figure out a way to not require the overhead of turning the
280
# list back into tuples.
281
lines = [tuple(line.split(' ', 1)) for line in content]
282
return AnnotatedKnitContent(lines)
284
def parse_line_delta_iter(self, lines):
285
return iter(self.parse_line_delta(lines))
287
def parse_line_delta(self, lines, version_id, plain=False):
288
"""Convert a line based delta into internal representation.
290
line delta is in the form of:
291
intstart intend intcount
293
revid(utf8) newline\n
294
internal representation is
295
(start, end, count, [1..count tuples (revid, newline)])
297
:param plain: If True, the lines are returned as a plain
298
list without annotations, not as a list of (origin, content) tuples, i.e.
299
(start, end, count, [1..count newline])
306
def cache_and_return(line):
307
origin, text = line.split(' ', 1)
308
return cache.setdefault(origin, origin), text
310
# walk through the lines parsing.
311
# Note that the plain test is explicitly pulled out of the
312
# loop to minimise any performance impact
315
start, end, count = [int(n) for n in header.split(',')]
316
contents = [next().split(' ', 1)[1] for i in xrange(count)]
317
result.append((start, end, count, contents))
320
start, end, count = [int(n) for n in header.split(',')]
321
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
322
result.append((start, end, count, contents))
325
def get_fulltext_content(self, lines):
326
"""Extract just the content lines from a fulltext."""
327
return (line.split(' ', 1)[1] for line in lines)
329
def get_linedelta_content(self, lines):
330
"""Extract just the content from a line delta.
332
This doesn't return all of the extra information stored in a delta.
333
Only the actual content lines.
338
header = header.split(',')
339
count = int(header[2])
340
for i in xrange(count):
341
origin, text = next().split(' ', 1)
344
def lower_fulltext(self, content):
345
"""convert a fulltext content record into a serializable form.
347
see parse_fulltext which this inverts.
349
# TODO: jam 20070209 We only do the caching thing to make sure that
350
# the origin is a valid utf-8 line, eventually we could remove it
351
return ['%s %s' % (o, t) for o, t in content._lines]
353
def lower_line_delta(self, delta):
354
"""convert a delta into a serializable form.
356
See parse_line_delta which this inverts.
358
# TODO: jam 20070209 We only do the caching thing to make sure that
359
# the origin is a valid utf-8 line, eventually we could remove it
361
for start, end, c, lines in delta:
362
out.append('%d,%d,%d\n' % (start, end, c))
363
out.extend(origin + ' ' + text
364
for origin, text in lines)
367
def annotate_iter(self, knit, version_id):
368
content = knit._get_content(version_id)
369
return content.annotate_iter()
372
class KnitPlainFactory(object):
373
"""Factory for creating plain Content objects."""
377
def make(self, lines, version_id):
378
return PlainKnitContent(lines, version_id)
380
def parse_fulltext(self, content, version_id):
381
"""This parses an unannotated fulltext.
383
Note that this is not a noop - the internal representation
384
has (versionid, line) - its just a constant versionid.
386
return self.make(content, version_id)
388
def parse_line_delta_iter(self, lines, version_id):
390
num_lines = len(lines)
391
while cur < num_lines:
394
start, end, c = [int(n) for n in header.split(',')]
395
yield start, end, c, lines[cur:cur+c]
398
def parse_line_delta(self, lines, version_id):
399
return list(self.parse_line_delta_iter(lines, version_id))
401
def get_fulltext_content(self, lines):
402
"""Extract just the content lines from a fulltext."""
405
def get_linedelta_content(self, lines):
406
"""Extract just the content from a line delta.
408
This doesn't return all of the extra information stored in a delta.
409
Only the actual content lines.
414
header = header.split(',')
415
count = int(header[2])
416
for i in xrange(count):
419
def lower_fulltext(self, content):
420
return content.text()
422
def lower_line_delta(self, delta):
424
for start, end, c, lines in delta:
425
out.append('%d,%d,%d\n' % (start, end, c))
429
def annotate_iter(self, knit, version_id):
430
annotator = _KnitAnnotator(knit)
431
return iter(annotator.get_annotated_lines(version_id))
434
def make_empty_knit(transport, relpath):
435
"""Construct a empty knit at the specified location."""
436
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
439
class KnitVersionedFile(VersionedFile):
440
"""Weave-like structure with faster random access.
442
A knit stores a number of texts and a summary of the relationships
443
between them. Texts are identified by a string version-id. Texts
444
are normally stored and retrieved as a series of lines, but can
445
also be passed as single strings.
447
Lines are stored with the trailing newline (if any) included, to
448
avoid special cases for files with no final newline. Lines are
449
composed of 8-bit characters, not unicode. The combination of
450
these approaches should mean any 'binary' file can be safely
451
stored and retrieved.
454
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
455
factory=None, delta=True, create=False, create_parent_dir=False,
456
delay_create=False, dir_mode=None, index=None, access_method=None):
457
"""Construct a knit at location specified by relpath.
459
:param create: If not True, only open an existing knit.
460
:param create_parent_dir: If True, create the parent directory if
461
creating the file fails. (This is used for stores with
462
hash-prefixes that may not exist yet)
463
:param delay_create: The calling code is aware that the knit won't
464
actually be created until the first data is stored.
465
:param index: An index to use for the knit.
467
if access_mode is None:
469
super(KnitVersionedFile, self).__init__(access_mode)
470
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
471
self.transport = transport
472
self.filename = relpath
473
self.factory = factory or KnitAnnotateFactory()
474
self.writable = (access_mode == 'w')
477
self._max_delta_chain = 200
480
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
481
access_mode, create=create, file_mode=file_mode,
482
create_parent_dir=create_parent_dir, delay_create=delay_create,
486
if access_method is None:
487
_access = _KnitAccess(transport, relpath + DATA_SUFFIX, file_mode, dir_mode,
488
((create and not len(self)) and delay_create), create_parent_dir)
490
_access = access_method
491
if create and not len(self) and not delay_create:
493
self._data = _KnitData(_access)
496
return '%s(%s)' % (self.__class__.__name__,
497
self.transport.abspath(self.filename))
499
def _check_should_delta(self, first_parents):
500
"""Iterate back through the parent listing, looking for a fulltext.
502
This is used when we want to decide whether to add a delta or a new
503
fulltext. It searches for _max_delta_chain parents. When it finds a
504
fulltext parent, it sees if the total size of the deltas leading up to
505
it is large enough to indicate that we want a new full text anyway.
507
Return True if we should create a new delta, False if we should use a
512
delta_parents = first_parents
513
for count in xrange(self._max_delta_chain):
514
parent = delta_parents[0]
515
method = self._index.get_method(parent)
516
index, pos, size = self._index.get_position(parent)
517
if method == 'fulltext':
521
delta_parents = self._index.get_parents(parent)
523
# We couldn't find a fulltext, so we must create a new one
526
return fulltext_size > delta_size
528
def _add_raw_records(self, records, data):
529
"""Add all the records 'records' with data pre-joined in 'data'.
531
:param records: A list of tuples(version_id, options, parents, size).
532
:param data: The data for the records. When it is written, the records
533
are adjusted to have pos pointing into data by the sum of
534
the preceding records sizes.
537
raw_record_sizes = [record[3] for record in records]
538
positions = self._data.add_raw_records(raw_record_sizes, data)
541
for (version_id, options, parents, size), access_memo in zip(
543
index_entries.append((version_id, options, access_memo, parents))
544
if self._data._do_cache:
545
self._data._cache[version_id] = data[offset:offset+size]
547
self._index.add_versions(index_entries)
549
def enable_cache(self):
550
"""Start caching data for this knit"""
551
self._data.enable_cache()
553
def clear_cache(self):
554
"""Clear the data cache only."""
555
self._data.clear_cache()
557
def copy_to(self, name, transport):
558
"""See VersionedFile.copy_to()."""
559
# copy the current index to a temp index to avoid racing with local
561
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
562
self.transport.get(self._index._filename))
564
f = self._data._open_file()
566
transport.put_file(name + DATA_SUFFIX, f)
569
# move the copied index into place
570
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
572
def create_empty(self, name, transport, mode=None):
573
return KnitVersionedFile(name, transport, factory=self.factory,
574
delta=self.delta, create=True)
576
def get_data_stream(self, required_versions):
577
"""Get a data stream for the specified versions.
579
Versions may be returned in any order, not necessarily the order
580
specified. They are returned in a partial order by compression
581
parent, so that the deltas can be applied as the data stream is
582
inserted; however note that compression parents will not be sent
583
unless they were specifically requested, as the client may already
586
:param required_versions: The exact set of versions to be extracted.
587
Unlike some other knit methods, this is not used to generate a
588
transitive closure, rather it is used precisely as given.
590
:returns: format_signature, list of (version, options, length, parents),
593
required_version_set = frozenset(required_versions)
595
# list of revisions that can just be sent without waiting for their
598
# map from revision to the children based on it
600
# first, read all relevant index data, enough to sort into the right
602
for version_id in required_versions:
603
options = self._index.get_options(version_id)
604
parents = self._index.get_parents_with_ghosts(version_id)
605
index_memo = self._index.get_position(version_id)
606
version_index[version_id] = (index_memo, options, parents)
607
if ('line-delta' in options
608
and parents[0] in required_version_set):
609
# must wait until the parent has been sent
610
deferred.setdefault(parents[0], []). \
613
# either a fulltext, or a delta whose parent the client did
614
# not ask for and presumably already has
615
ready_to_send.append(version_id)
616
# build a list of results to return, plus instructions for data to
618
copy_queue_records = []
619
temp_version_list = []
621
# XXX: pushing and popping lists may be a bit inefficient
622
version_id = ready_to_send.pop(0)
623
(index_memo, options, parents) = version_index[version_id]
624
copy_queue_records.append((version_id, index_memo))
625
none, data_pos, data_size = index_memo
626
temp_version_list.append((version_id, options, data_size,
628
if version_id in deferred:
629
# now we can send all the children of this revision - we could
630
# put them in anywhere, but we hope that sending them soon
631
# after the fulltext will give good locality in the receiver
632
ready_to_send[:0] = deferred.pop(version_id)
633
assert len(deferred) == 0, \
634
"Still have compressed child versions waiting to be sent"
635
# XXX: The stream format is such that we cannot stream it - we have to
636
# know the length of all the data a-priori.
638
result_version_list = []
639
for (version_id, raw_data), \
640
(version_id2, options, _, parents) in \
641
izip(self._data.read_records_iter_raw(copy_queue_records),
643
assert version_id == version_id2, \
644
'logic error, inconsistent results'
645
raw_datum.append(raw_data)
646
result_version_list.append(
647
(version_id, options, len(raw_data), parents))
648
# provide a callback to get data incrementally.
649
pseudo_file = StringIO(''.join(raw_datum))
652
return pseudo_file.read()
654
return pseudo_file.read(length)
655
return (self.get_format_signature(), result_version_list, read)
657
def _extract_blocks(self, version_id, source, target):
658
if self._index.get_method(version_id) != 'line-delta':
660
parent, sha1, noeol, delta = self.get_delta(version_id)
661
return KnitContent.get_line_delta_blocks(delta, source, target)
663
def get_delta(self, version_id):
664
"""Get a delta for constructing version from some other version."""
665
self.check_not_reserved_id(version_id)
666
parents = self.get_parents(version_id)
671
index_memo = self._index.get_position(version_id)
672
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
673
noeol = 'no-eol' in self._index.get_options(version_id)
674
if 'fulltext' == self._index.get_method(version_id):
675
new_content = self.factory.parse_fulltext(data, version_id)
676
if parent is not None:
677
reference_content = self._get_content(parent)
678
old_texts = reference_content.text()
681
new_texts = new_content.text()
682
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
684
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
686
delta = self.factory.parse_line_delta(data, version_id)
687
return parent, sha1, noeol, delta
689
def get_format_signature(self):
690
"""See VersionedFile.get_format_signature()."""
691
if self.factory.annotated:
692
annotated_part = "annotated"
694
annotated_part = "plain"
695
return "knit-%s" % (annotated_part,)
697
def get_graph_with_ghosts(self):
698
"""See VersionedFile.get_graph_with_ghosts()."""
699
graph_items = self._index.get_graph()
700
return dict(graph_items)
702
def get_sha1(self, version_id):
703
return self.get_sha1s([version_id])[0]
705
def get_sha1s(self, version_ids):
706
"""See VersionedFile.get_sha1()."""
707
record_map = self._get_record_map(version_ids)
708
# record entry 2 is the 'digest'.
709
return [record_map[v][2] for v in version_ids]
713
"""See VersionedFile.get_suffixes()."""
714
return [DATA_SUFFIX, INDEX_SUFFIX]
716
def has_ghost(self, version_id):
717
"""True if there is a ghost reference in the file to version_id."""
719
if self.has_version(version_id):
721
# optimisable if needed by memoising the _ghosts set.
722
items = self._index.get_graph()
723
for node, parents in items:
724
for parent in parents:
725
if parent not in self._index._cache:
726
if parent == version_id:
730
def insert_data_stream(self, (format, data_list, reader_callable)):
731
"""Insert knit records from a data stream into this knit.
733
If a version in the stream is already present in this knit, it will not
734
be inserted a second time. It will be checked for consistency with the
735
stored version however, and may cause a KnitCorrupt error to be raised
736
if the data in the stream disagrees with the already stored data.
738
:seealso: get_data_stream
740
if format != self.get_format_signature():
741
if 'knit' in debug.debug_flags:
743
'incompatible format signature inserting to %r', self)
744
source = self._knit_from_datastream(
745
(format, data_list, reader_callable))
749
for version_id, options, length, parents in data_list:
750
if self.has_version(version_id):
751
# First check: the list of parents.
752
my_parents = self.get_parents_with_ghosts(version_id)
753
if tuple(my_parents) != tuple(parents):
754
# XXX: KnitCorrupt is not quite the right exception here.
757
'parents list %r from data stream does not match '
758
'already recorded parents %r for %s'
759
% (parents, my_parents, version_id))
761
# Also check the SHA-1 of the fulltext this content will
763
raw_data = reader_callable(length)
764
my_fulltext_sha1 = self.get_sha1(version_id)
765
df, rec = self._data._parse_record_header(version_id, raw_data)
766
stream_fulltext_sha1 = rec[3]
767
if my_fulltext_sha1 != stream_fulltext_sha1:
768
# Actually, we don't know if it's this knit that's corrupt,
769
# or the data stream we're trying to insert.
771
self.filename, 'sha-1 does not match %s' % version_id)
773
if 'line-delta' in options:
774
# Make sure that this knit record is actually useful: a
775
# line-delta is no use unless we have its parent.
776
# Fetching from a broken repository with this problem
777
# shouldn't break the target repository.
779
# See https://bugs.launchpad.net/bzr/+bug/164443
780
if not self._index.has_version(parents[0]):
783
'line-delta from stream '
786
'missing parent %s\n'
787
'Try running "bzr check" '
788
'on the source repository, and "bzr reconcile" '
790
(version_id, parents[0]))
791
self._add_raw_records(
792
[(version_id, options, parents, length)],
793
reader_callable(length))
795
def _knit_from_datastream(self, (format, data_list, reader_callable)):
796
"""Create a knit object from a data stream.
798
This method exists to allow conversion of data streams that do not
799
match the signature of this knit. Generally it will be slower and use
800
more memory to use this method to insert data, but it will work.
802
:seealso: get_data_stream for details on datastreams.
803
:return: A knit versioned file which can be used to join the datastream
806
if format == "knit-plain":
807
factory = KnitPlainFactory()
808
elif format == "knit-annotated":
809
factory = KnitAnnotateFactory()
811
raise errors.KnitDataStreamUnknown(format)
812
index = _StreamIndex(data_list, self._index)
813
access = _StreamAccess(reader_callable, index, self, factory)
814
return KnitVersionedFile(self.filename, self.transport,
815
factory=factory, index=index, access_method=access)
818
"""See VersionedFile.versions."""
819
if 'evil' in debug.debug_flags:
820
trace.mutter_callsite(2, "versions scales with size of history")
821
return self._index.get_versions()
823
def has_version(self, version_id):
824
"""See VersionedFile.has_version."""
825
if 'evil' in debug.debug_flags:
826
trace.mutter_callsite(2, "has_version is a LBYL scenario")
827
return self._index.has_version(version_id)
829
__contains__ = has_version
831
def _merge_annotations(self, content, parents, parent_texts={},
832
delta=None, annotated=None,
833
left_matching_blocks=None):
834
"""Merge annotations for content. This is done by comparing
835
the annotations based on changed to the text.
837
if left_matching_blocks is not None:
838
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
842
for parent_id in parents:
843
merge_content = self._get_content(parent_id, parent_texts)
844
if (parent_id == parents[0] and delta_seq is not None):
847
seq = patiencediff.PatienceSequenceMatcher(
848
None, merge_content.text(), content.text())
849
for i, j, n in seq.get_matching_blocks():
852
# this appears to copy (origin, text) pairs across to the
853
# new content for any line that matches the last-checked
855
content._lines[j:j+n] = merge_content._lines[i:i+n]
857
if delta_seq is None:
858
reference_content = self._get_content(parents[0], parent_texts)
859
new_texts = content.text()
860
old_texts = reference_content.text()
861
delta_seq = patiencediff.PatienceSequenceMatcher(
862
None, old_texts, new_texts)
863
return self._make_line_delta(delta_seq, content)
865
def _make_line_delta(self, delta_seq, new_content):
866
"""Generate a line delta from delta_seq and new_content."""
868
for op in delta_seq.get_opcodes():
871
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
874
def _get_components_positions(self, version_ids):
875
"""Produce a map of position data for the components of versions.
877
This data is intended to be used for retrieving the knit records.
879
A dict of version_id to (method, index_memo, next, parents, noeol) is
881
method is the way referenced data should be applied.
882
index_memo is the handle to pass to the data access to actually get the
884
next is the build-parent of the version, or None for fulltexts.
885
parents is the version_ids of the parents of this version
886
noeol is a flag indicating if there is a final newline character
889
pending_components = version_ids
890
while pending_components:
891
build_details = self._index.get_build_details(pending_components)
892
pending_components = set()
893
for version_id, details in build_details.items():
894
(method, index_memo, compression_parent, parents,
896
if compression_parent is not None:
897
pending_components.add(compression_parent)
898
component_data[version_id] = (method, index_memo,
900
return component_data
902
def _get_content(self, version_id, parent_texts={}):
903
"""Returns a content object that makes up the specified
905
cached_version = parent_texts.get(version_id, None)
906
if cached_version is not None:
907
if not self.has_version(version_id):
908
raise RevisionNotPresent(version_id, self.filename)
909
return cached_version
911
text_map, contents_map = self._get_content_maps([version_id])
912
return contents_map[version_id]
914
def _check_versions_present(self, version_ids):
915
"""Check that all specified versions are present."""
916
self._index.check_versions_present(version_ids)
918
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
919
nostore_sha, random_id, check_content):
920
"""See VersionedFile.add_lines_with_ghosts()."""
921
self._check_add(version_id, lines, random_id, check_content)
922
return self._add(version_id, lines, parents, self.delta,
923
parent_texts, None, nostore_sha, random_id)
925
def _add_lines(self, version_id, parents, lines, parent_texts,
926
left_matching_blocks, nostore_sha, random_id, check_content):
927
"""See VersionedFile.add_lines."""
928
self._check_add(version_id, lines, random_id, check_content)
929
self._check_versions_present(parents)
930
return self._add(version_id, lines[:], parents, self.delta,
931
parent_texts, left_matching_blocks, nostore_sha, random_id)
933
def _check_add(self, version_id, lines, random_id, check_content):
934
"""check that version_id and lines are safe to add."""
935
if contains_whitespace(version_id):
936
raise InvalidRevisionId(version_id, self.filename)
937
self.check_not_reserved_id(version_id)
938
# Technically this could be avoided if we are happy to allow duplicate
939
# id insertion when other things than bzr core insert texts, but it
940
# seems useful for folk using the knit api directly to have some safety
941
# blanket that we can disable.
942
if not random_id and self.has_version(version_id):
943
raise RevisionAlreadyPresent(version_id, self.filename)
945
self._check_lines_not_unicode(lines)
946
self._check_lines_are_lines(lines)
948
def _add(self, version_id, lines, parents, delta, parent_texts,
949
left_matching_blocks, nostore_sha, random_id):
950
"""Add a set of lines on top of version specified by parents.
952
If delta is true, compress the text as a line-delta against
955
Any versions not present will be converted into ghosts.
957
# first thing, if the content is something we don't need to store, find
959
line_bytes = ''.join(lines)
960
digest = sha_string(line_bytes)
961
if nostore_sha == digest:
962
raise errors.ExistingContent
965
if parent_texts is None:
967
for parent in parents:
968
if self.has_version(parent):
969
present_parents.append(parent)
971
# can only compress against the left most present parent.
973
(len(present_parents) == 0 or
974
present_parents[0] != parents[0])):
977
text_length = len(line_bytes)
980
if lines[-1][-1] != '\n':
981
# copy the contents of lines.
983
options.append('no-eol')
984
lines[-1] = lines[-1] + '\n'
988
# To speed the extract of texts the delta chain is limited
989
# to a fixed number of deltas. This should minimize both
990
# I/O and the time spend applying deltas.
991
delta = self._check_should_delta(present_parents)
993
assert isinstance(version_id, str)
994
content = self.factory.make(lines, version_id)
995
if delta or (self.factory.annotated and len(present_parents) > 0):
996
# Merge annotations from parent texts if needed.
997
delta_hunks = self._merge_annotations(content, present_parents,
998
parent_texts, delta, self.factory.annotated,
999
left_matching_blocks)
1002
options.append('line-delta')
1003
store_lines = self.factory.lower_line_delta(delta_hunks)
1004
size, bytes = self._data._record_to_data(version_id, digest,
1007
options.append('fulltext')
1008
# isinstance is slower and we have no hierarchy.
1009
if self.factory.__class__ == KnitPlainFactory:
1010
# Use the already joined bytes saving iteration time in
1012
size, bytes = self._data._record_to_data(version_id, digest,
1013
lines, [line_bytes])
1015
# get mixed annotation + content and feed it into the
1017
store_lines = self.factory.lower_fulltext(content)
1018
size, bytes = self._data._record_to_data(version_id, digest,
1021
access_memo = self._data.add_raw_records([size], bytes)[0]
1022
self._index.add_versions(
1023
((version_id, options, access_memo, parents),),
1024
random_id=random_id)
1025
return digest, text_length, content
1027
def check(self, progress_bar=None):
1028
"""See VersionedFile.check()."""
1030
def _clone_text(self, new_version_id, old_version_id, parents):
1031
"""See VersionedFile.clone_text()."""
1032
# FIXME RBC 20060228 make fast by only inserting an index with null
1034
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
1036
def get_lines(self, version_id):
1037
"""See VersionedFile.get_lines()."""
1038
return self.get_line_list([version_id])[0]
1040
def _get_record_map(self, version_ids):
1041
"""Produce a dictionary of knit records.
1043
The keys are version_ids, the values are tuples of (method, content,
1045
method is the way the content should be applied.
1046
content is a KnitContent object.
1047
digest is the SHA1 digest of this version id after all steps are done
1048
next is the build-parent of the version, i.e. the leftmost ancestor.
1049
If the method is fulltext, next will be None.
1051
position_map = self._get_components_positions(version_ids)
1052
# c = component_id, m = method, i_m = index_memo, n = next
1053
# p = parent_ids, e = noeol
1054
records = [(c, i_m) for c, (m, i_m, n)
1055
in position_map.iteritems()]
1057
for component_id, content, digest in \
1058
self._data.read_records_iter(records):
1059
(method, index_memo, next) = position_map[component_id]
1060
record_map[component_id] = method, content, digest, next
1064
def get_text(self, version_id):
1065
"""See VersionedFile.get_text"""
1066
return self.get_texts([version_id])[0]
1068
def get_texts(self, version_ids):
1069
return [''.join(l) for l in self.get_line_list(version_ids)]
1071
def get_line_list(self, version_ids):
1072
"""Return the texts of listed versions as a list of strings."""
1073
for version_id in version_ids:
1074
self.check_not_reserved_id(version_id)
1075
text_map, content_map = self._get_content_maps(version_ids)
1076
return [text_map[v] for v in version_ids]
1078
_get_lf_split_line_list = get_line_list
1080
def _get_content_maps(self, version_ids):
1081
"""Produce maps of text and KnitContents
1083
:return: (text_map, content_map) where text_map contains the texts for
1084
the requested versions and content_map contains the KnitContents.
1085
Both dicts take version_ids as their keys.
1087
# FUTURE: This function could be improved for the 'extract many' case
1088
# by tracking each component and only doing the copy when the number of
1089
# children than need to apply delta's to it is > 1 or it is part of the
1091
version_ids = list(version_ids)
1092
multiple_versions = len(version_ids) != 1
1093
record_map = self._get_record_map(version_ids)
1098
for version_id in version_ids:
1101
while cursor is not None:
1102
method, data, digest, next = record_map[cursor]
1103
components.append((cursor, method, data, digest))
1104
if cursor in content_map:
1109
for component_id, method, data, digest in reversed(components):
1110
if component_id in content_map:
1111
content = content_map[component_id]
1113
if method == 'fulltext':
1114
assert content is None
1115
content = self.factory.parse_fulltext(data, version_id)
1116
elif method == 'line-delta':
1117
delta = self.factory.parse_line_delta(data, version_id)
1118
if multiple_versions:
1119
# only doing this when we want multiple versions
1120
# output avoids list copies - which reference and
1121
# dereference many strings.
1122
content = content.copy()
1123
content.apply_delta(delta, version_id)
1124
if multiple_versions:
1125
content_map[component_id] = content
1127
if 'no-eol' in self._index.get_options(version_id):
1128
if multiple_versions:
1129
content = content.copy()
1130
content.strip_last_line_newline()
1131
final_content[version_id] = content
1133
# digest here is the digest from the last applied component.
1134
text = content.text()
1135
actual_sha = sha_strings(text)
1136
if actual_sha != digest:
1137
raise KnitCorrupt(self.filename,
1139
'\n of reconstructed text does not match'
1141
'\n for version %s' %
1142
(actual_sha, digest, version_id))
1143
text_map[version_id] = text
1144
return text_map, final_content
1146
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1148
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1149
if version_ids is None:
1150
version_ids = self.versions()
1152
pb = progress.DummyProgress()
1153
# we don't care about inclusions, the caller cares.
1154
# but we need to setup a list of records to visit.
1155
# we need version_id, position, length
1156
version_id_records = []
1157
requested_versions = set(version_ids)
1158
# filter for available versions
1159
for version_id in requested_versions:
1160
if not self.has_version(version_id):
1161
raise RevisionNotPresent(version_id, self.filename)
1162
# get a in-component-order queue:
1163
for version_id in self.versions():
1164
if version_id in requested_versions:
1165
index_memo = self._index.get_position(version_id)
1166
version_id_records.append((version_id, index_memo))
1168
total = len(version_id_records)
1169
for version_idx, (version_id, data, sha_value) in \
1170
enumerate(self._data.read_records_iter(version_id_records)):
1171
pb.update('Walking content.', version_idx, total)
1172
method = self._index.get_method(version_id)
1174
assert method in ('fulltext', 'line-delta')
1175
if method == 'fulltext':
1176
line_iterator = self.factory.get_fulltext_content(data)
1178
line_iterator = self.factory.get_linedelta_content(data)
1179
# XXX: It might be more efficient to yield (version_id,
1180
# line_iterator) in the future. However for now, this is a simpler
1181
# change to integrate into the rest of the codebase. RBC 20071110
1182
for line in line_iterator:
1183
yield line, version_id
1185
pb.update('Walking content.', total, total)
1187
def iter_parents(self, version_ids):
1188
"""Iterate through the parents for many version ids.
1190
:param version_ids: An iterable yielding version_ids.
1191
:return: An iterator that yields (version_id, parents). Requested
1192
version_ids not present in the versioned file are simply skipped.
1193
The order is undefined, allowing for different optimisations in
1194
the underlying implementation.
1196
return self._index.iter_parents(version_ids)
1198
def num_versions(self):
1199
"""See VersionedFile.num_versions()."""
1200
return self._index.num_versions()
1202
__len__ = num_versions
1204
def annotate_iter(self, version_id):
1205
"""See VersionedFile.annotate_iter."""
1206
return self.factory.annotate_iter(self, version_id)
1208
def get_parents(self, version_id):
1209
"""See VersionedFile.get_parents."""
1212
# 52554 calls in 1264 872 internal down from 3674
1214
return self._index.get_parents(version_id)
1216
raise RevisionNotPresent(version_id, self.filename)
1218
def get_parents_with_ghosts(self, version_id):
1219
"""See VersionedFile.get_parents."""
1221
return self._index.get_parents_with_ghosts(version_id)
1223
raise RevisionNotPresent(version_id, self.filename)
1225
def get_ancestry(self, versions, topo_sorted=True):
1226
"""See VersionedFile.get_ancestry."""
1227
if isinstance(versions, basestring):
1228
versions = [versions]
1231
return self._index.get_ancestry(versions, topo_sorted)
1233
def get_ancestry_with_ghosts(self, versions):
1234
"""See VersionedFile.get_ancestry_with_ghosts."""
1235
if isinstance(versions, basestring):
1236
versions = [versions]
1239
return self._index.get_ancestry_with_ghosts(versions)
1241
def plan_merge(self, ver_a, ver_b):
1242
"""See VersionedFile.plan_merge."""
1243
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1244
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1245
annotated_a = self.annotate(ver_a)
1246
annotated_b = self.annotate(ver_b)
1247
return merge._plan_annotate_merge(annotated_a, annotated_b,
1248
ancestors_a, ancestors_b)
1251
class _KnitComponentFile(object):
1252
"""One of the files used to implement a knit database"""
1254
def __init__(self, transport, filename, mode, file_mode=None,
1255
create_parent_dir=False, dir_mode=None):
1256
self._transport = transport
1257
self._filename = filename
1259
self._file_mode = file_mode
1260
self._dir_mode = dir_mode
1261
self._create_parent_dir = create_parent_dir
1262
self._need_to_create = False
1264
def _full_path(self):
1265
"""Return the full path to this file."""
1266
return self._transport.base + self._filename
1268
def check_header(self, fp):
1269
line = fp.readline()
1271
# An empty file can actually be treated as though the file doesn't
1273
raise errors.NoSuchFile(self._full_path())
1274
if line != self.HEADER:
1275
raise KnitHeaderError(badline=line,
1276
filename=self._transport.abspath(self._filename))
1279
return '%s(%s)' % (self.__class__.__name__, self._filename)
1282
class _KnitIndex(_KnitComponentFile):
1283
"""Manages knit index file.
1285
The index is already kept in memory and read on startup, to enable
1286
fast lookups of revision information. The cursor of the index
1287
file is always pointing to the end, making it easy to append
1290
_cache is a cache for fast mapping from version id to a Index
1293
_history is a cache for fast mapping from indexes to version ids.
1295
The index data format is dictionary compressed when it comes to
1296
parent references; a index entry may only have parents that with a
1297
lover index number. As a result, the index is topological sorted.
1299
Duplicate entries may be written to the index for a single version id
1300
if this is done then the latter one completely replaces the former:
1301
this allows updates to correct version and parent information.
1302
Note that the two entries may share the delta, and that successive
1303
annotations and references MUST point to the first entry.
1305
The index file on disc contains a header, followed by one line per knit
1306
record. The same revision can be present in an index file more than once.
1307
The first occurrence gets assigned a sequence number starting from 0.
1309
The format of a single line is
1310
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1311
REVISION_ID is a utf8-encoded revision id
1312
FLAGS is a comma separated list of flags about the record. Values include
1313
no-eol, line-delta, fulltext.
1314
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1315
that the the compressed data starts at.
1316
LENGTH is the ascii representation of the length of the data file.
1317
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1319
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1320
revision id already in the knit that is a parent of REVISION_ID.
1321
The ' :' marker is the end of record marker.
1324
when a write is interrupted to the index file, it will result in a line
1325
that does not end in ' :'. If the ' :' is not present at the end of a line,
1326
or at the end of the file, then the record that is missing it will be
1327
ignored by the parser.
1329
When writing new records to the index file, the data is preceded by '\n'
1330
to ensure that records always start on new lines even if the last write was
1331
interrupted. As a result its normal for the last line in the index to be
1332
missing a trailing newline. One can be added with no harmful effects.
1335
HEADER = "# bzr knit index 8\n"
1337
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1338
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1340
def _cache_version(self, version_id, options, pos, size, parents):
1341
"""Cache a version record in the history array and index cache.
1343
This is inlined into _load_data for performance. KEEP IN SYNC.
1344
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1347
# only want the _history index to reference the 1st index entry
1349
if version_id not in self._cache:
1350
index = len(self._history)
1351
self._history.append(version_id)
1353
index = self._cache[version_id][5]
1354
self._cache[version_id] = (version_id,
1361
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1362
create_parent_dir=False, delay_create=False, dir_mode=None):
1363
_KnitComponentFile.__init__(self, transport, filename, mode,
1364
file_mode=file_mode,
1365
create_parent_dir=create_parent_dir,
1368
# position in _history is the 'official' index for a revision
1369
# but the values may have come from a newer entry.
1370
# so - wc -l of a knit index is != the number of unique names
1374
fp = self._transport.get(self._filename)
1376
# _load_data may raise NoSuchFile if the target knit is
1378
_load_data(self, fp)
1382
if mode != 'w' or not create:
1385
self._need_to_create = True
1387
self._transport.put_bytes_non_atomic(
1388
self._filename, self.HEADER, mode=self._file_mode)
1390
def get_graph(self):
1391
"""Return a list of the node:parents lists from this knit index."""
1392
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1394
def get_ancestry(self, versions, topo_sorted=True):
1395
"""See VersionedFile.get_ancestry."""
1396
# get a graph of all the mentioned versions:
1398
pending = set(versions)
1401
version = pending.pop()
1404
parents = [p for p in cache[version][4] if p in cache]
1406
raise RevisionNotPresent(version, self._filename)
1407
# if not completed and not a ghost
1408
pending.update([p for p in parents if p not in graph])
1409
graph[version] = parents
1412
return topo_sort(graph.items())
1414
def get_ancestry_with_ghosts(self, versions):
1415
"""See VersionedFile.get_ancestry_with_ghosts."""
1416
# get a graph of all the mentioned versions:
1417
self.check_versions_present(versions)
1420
pending = set(versions)
1422
version = pending.pop()
1424
parents = cache[version][4]
1430
pending.update([p for p in parents if p not in graph])
1431
graph[version] = parents
1432
return topo_sort(graph.items())
1434
def get_build_details(self, version_ids):
1435
"""Get the method, index_memo and compression parent for version_ids.
1437
:param version_ids: An iterable of version_ids.
1438
:return: A dict of version_id:(method, index_memo, compression_parent,
1442
for version_id in version_ids:
1443
method = self.get_method(version_id)
1444
parents = self.get_parents_with_ghosts(version_id)
1445
if method == 'fulltext':
1446
compression_parent = None
1448
compression_parent = parents[0]
1449
noeol = 'no-eol' in self.get_options(version_id)
1450
index_memo = self.get_position(version_id)
1451
result[version_id] = (method, index_memo, compression_parent,
1455
def iter_parents(self, version_ids):
1456
"""Iterate through the parents for many version ids.
1458
:param version_ids: An iterable yielding version_ids.
1459
:return: An iterator that yields (version_id, parents). Requested
1460
version_ids not present in the versioned file are simply skipped.
1461
The order is undefined, allowing for different optimisations in
1462
the underlying implementation.
1464
for version_id in version_ids:
1466
yield version_id, tuple(self.get_parents(version_id))
1470
def num_versions(self):
1471
return len(self._history)
1473
__len__ = num_versions
1475
def get_versions(self):
1476
"""Get all the versions in the file. not topologically sorted."""
1477
return self._history
1479
def _version_list_to_index(self, versions):
1482
for version in versions:
1483
if version in cache:
1484
# -- inlined lookup() --
1485
result_list.append(str(cache[version][5]))
1486
# -- end lookup () --
1488
result_list.append('.' + version)
1489
return ' '.join(result_list)
1491
def add_version(self, version_id, options, index_memo, parents):
1492
"""Add a version record to the index."""
1493
self.add_versions(((version_id, options, index_memo, parents),))
1495
def add_versions(self, versions, random_id=False):
1496
"""Add multiple versions to the index.
1498
:param versions: a list of tuples:
1499
(version_id, options, pos, size, parents).
1500
:param random_id: If True the ids being added were randomly generated
1501
and no check for existence will be performed.
1504
orig_history = self._history[:]
1505
orig_cache = self._cache.copy()
1508
for version_id, options, (index, pos, size), parents in versions:
1509
line = "\n%s %s %s %s %s :" % (version_id,
1513
self._version_list_to_index(parents))
1514
assert isinstance(line, str), \
1515
'content must be utf-8 encoded: %r' % (line,)
1517
self._cache_version(version_id, options, pos, size, parents)
1518
if not self._need_to_create:
1519
self._transport.append_bytes(self._filename, ''.join(lines))
1522
sio.write(self.HEADER)
1523
sio.writelines(lines)
1525
self._transport.put_file_non_atomic(self._filename, sio,
1526
create_parent_dir=self._create_parent_dir,
1527
mode=self._file_mode,
1528
dir_mode=self._dir_mode)
1529
self._need_to_create = False
1531
# If any problems happen, restore the original values and re-raise
1532
self._history = orig_history
1533
self._cache = orig_cache
1536
def has_version(self, version_id):
1537
"""True if the version is in the index."""
1538
return version_id in self._cache
1540
def get_position(self, version_id):
1541
"""Return details needed to access the version.
1543
.kndx indices do not support split-out data, so return None for the
1546
:return: a tuple (None, data position, size) to hand to the access
1547
logic to get the record.
1549
entry = self._cache[version_id]
1550
return None, entry[2], entry[3]
1552
def get_method(self, version_id):
1553
"""Return compression method of specified version."""
1555
options = self._cache[version_id][1]
1557
raise RevisionNotPresent(version_id, self._filename)
1558
if 'fulltext' in options:
1561
if 'line-delta' not in options:
1562
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1565
def get_options(self, version_id):
1566
"""Return a list representing options.
1570
return self._cache[version_id][1]
1572
def get_parents(self, version_id):
1573
"""Return parents of specified version ignoring ghosts."""
1574
return [parent for parent in self._cache[version_id][4]
1575
if parent in self._cache]
1577
def get_parents_with_ghosts(self, version_id):
1578
"""Return parents of specified version with ghosts."""
1579
return self._cache[version_id][4]
1581
def check_versions_present(self, version_ids):
1582
"""Check that all specified versions are present."""
1584
for version_id in version_ids:
1585
if version_id not in cache:
1586
raise RevisionNotPresent(version_id, self._filename)
1589
class KnitGraphIndex(object):
1590
"""A knit index that builds on GraphIndex."""
1592
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1593
"""Construct a KnitGraphIndex on a graph_index.
1595
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1596
:param deltas: Allow delta-compressed records.
1597
:param add_callback: If not None, allow additions to the index and call
1598
this callback with a list of added GraphIndex nodes:
1599
[(node, value, node_refs), ...]
1600
:param parents: If True, record knits parents, if not do not record
1603
self._graph_index = graph_index
1604
self._deltas = deltas
1605
self._add_callback = add_callback
1606
self._parents = parents
1607
if deltas and not parents:
1608
raise KnitCorrupt(self, "Cannot do delta compression without "
1611
def _get_entries(self, keys, check_present=False):
1612
"""Get the entries for keys.
1614
:param keys: An iterable of index keys, - 1-tuples.
1619
for node in self._graph_index.iter_entries(keys):
1621
found_keys.add(node[1])
1623
# adapt parentless index to the rest of the code.
1624
for node in self._graph_index.iter_entries(keys):
1625
yield node[0], node[1], node[2], ()
1626
found_keys.add(node[1])
1628
missing_keys = keys.difference(found_keys)
1630
raise RevisionNotPresent(missing_keys.pop(), self)
1632
def _present_keys(self, version_ids):
1634
node[1] for node in self._get_entries(version_ids)])
1636
def _parentless_ancestry(self, versions):
1637
"""Honour the get_ancestry API for parentless knit indices."""
1638
wanted_keys = self._version_ids_to_keys(versions)
1639
present_keys = self._present_keys(wanted_keys)
1640
missing = set(wanted_keys).difference(present_keys)
1642
raise RevisionNotPresent(missing.pop(), self)
1643
return list(self._keys_to_version_ids(present_keys))
1645
def get_ancestry(self, versions, topo_sorted=True):
1646
"""See VersionedFile.get_ancestry."""
1647
if not self._parents:
1648
return self._parentless_ancestry(versions)
1649
# XXX: This will do len(history) index calls - perhaps
1650
# it should be altered to be a index core feature?
1651
# get a graph of all the mentioned versions:
1654
versions = self._version_ids_to_keys(versions)
1655
pending = set(versions)
1657
# get all pending nodes
1658
this_iteration = pending
1659
new_nodes = self._get_entries(this_iteration)
1662
for (index, key, value, node_refs) in new_nodes:
1663
# dont ask for ghosties - otherwise
1664
# we we can end up looping with pending
1665
# being entirely ghosted.
1666
graph[key] = [parent for parent in node_refs[0]
1667
if parent not in ghosts]
1669
for parent in graph[key]:
1670
# dont examine known nodes again
1675
ghosts.update(this_iteration.difference(found))
1676
if versions.difference(graph):
1677
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1679
result_keys = topo_sort(graph.items())
1681
result_keys = graph.iterkeys()
1682
return [key[0] for key in result_keys]
1684
def get_ancestry_with_ghosts(self, versions):
1685
"""See VersionedFile.get_ancestry."""
1686
if not self._parents:
1687
return self._parentless_ancestry(versions)
1688
# XXX: This will do len(history) index calls - perhaps
1689
# it should be altered to be a index core feature?
1690
# get a graph of all the mentioned versions:
1692
versions = self._version_ids_to_keys(versions)
1693
pending = set(versions)
1695
# get all pending nodes
1696
this_iteration = pending
1697
new_nodes = self._get_entries(this_iteration)
1699
for (index, key, value, node_refs) in new_nodes:
1700
graph[key] = node_refs[0]
1702
for parent in graph[key]:
1703
# dont examine known nodes again
1707
missing_versions = this_iteration.difference(graph)
1708
missing_needed = versions.intersection(missing_versions)
1710
raise RevisionNotPresent(missing_needed.pop(), self)
1711
for missing_version in missing_versions:
1712
# add a key, no parents
1713
graph[missing_version] = []
1714
pending.discard(missing_version) # don't look for it
1715
result_keys = topo_sort(graph.items())
1716
return [key[0] for key in result_keys]
1718
def get_build_details(self, version_ids):
1719
"""Get the method, index_memo and compression parent for version_ids.
1721
:param version_ids: An iterable of version_ids.
1722
:return: A dict of version_id:(method, index_memo, compression_parent,
1726
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
1727
for entry in entries:
1728
version_id = self._keys_to_version_ids((entry[1],))[0]
1729
parents = self._keys_to_version_ids(entry[3][0])
1730
if not self._deltas:
1731
compression_parent = None
1733
compression_parent_key = self._compression_parent(entry)
1734
if compression_parent_key:
1735
compression_parent = self._keys_to_version_ids(
1736
(compression_parent_key,))[0]
1738
compression_parent = None
1739
noeol = (entry[2][0] == 'N')
1740
if compression_parent:
1741
method = 'line-delta'
1744
result[version_id] = (method, self._node_to_position(entry),
1745
compression_parent, parents, noeol)
1748
def _compression_parent(self, an_entry):
1749
# return the key that an_entry is compressed against, or None
1750
# Grab the second parent list (as deltas implies parents currently)
1751
compression_parents = an_entry[3][1]
1752
if not compression_parents:
1754
assert len(compression_parents) == 1
1755
return compression_parents[0]
1757
def _get_method(self, node):
1758
if not self._deltas:
1760
if self._compression_parent(node):
1765
def get_graph(self):
1766
"""Return a list of the node:parents lists from this knit index."""
1767
if not self._parents:
1768
return [(key, ()) for key in self.get_versions()]
1770
for index, key, value, refs in self._graph_index.iter_all_entries():
1771
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1774
def iter_parents(self, version_ids):
1775
"""Iterate through the parents for many version ids.
1777
:param version_ids: An iterable yielding version_ids.
1778
:return: An iterator that yields (version_id, parents). Requested
1779
version_ids not present in the versioned file are simply skipped.
1780
The order is undefined, allowing for different optimisations in
1781
the underlying implementation.
1784
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1786
present_parents = set()
1787
for node in all_nodes:
1788
all_parents.update(node[3][0])
1789
# any node we are querying must be present
1790
present_parents.add(node[1])
1791
unknown_parents = all_parents.difference(present_parents)
1792
present_parents.update(self._present_keys(unknown_parents))
1793
for node in all_nodes:
1795
for parent in node[3][0]:
1796
if parent in present_parents:
1797
parents.append(parent[0])
1798
yield node[1][0], tuple(parents)
1800
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1801
yield node[1][0], ()
1803
def num_versions(self):
1804
return len(list(self._graph_index.iter_all_entries()))
1806
__len__ = num_versions
1808
def get_versions(self):
1809
"""Get all the versions in the file. not topologically sorted."""
1810
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1812
def has_version(self, version_id):
1813
"""True if the version is in the index."""
1814
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1816
def _keys_to_version_ids(self, keys):
1817
return tuple(key[0] for key in keys)
1819
def get_position(self, version_id):
1820
"""Return details needed to access the version.
1822
:return: a tuple (index, data position, size) to hand to the access
1823
logic to get the record.
1825
node = self._get_node(version_id)
1826
return self._node_to_position(node)
1828
def _node_to_position(self, node):
1829
"""Convert an index value to position details."""
1830
bits = node[2][1:].split(' ')
1831
return node[0], int(bits[0]), int(bits[1])
1833
def get_method(self, version_id):
1834
"""Return compression method of specified version."""
1835
return self._get_method(self._get_node(version_id))
1837
def _get_node(self, version_id):
1839
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1841
raise RevisionNotPresent(version_id, self)
1843
def get_options(self, version_id):
1844
"""Return a list representing options.
1848
node = self._get_node(version_id)
1849
options = [self._get_method(node)]
1850
if node[2][0] == 'N':
1851
options.append('no-eol')
1854
def get_parents(self, version_id):
1855
"""Return parents of specified version ignoring ghosts."""
1856
parents = list(self.iter_parents([version_id]))
1859
raise errors.RevisionNotPresent(version_id, self)
1860
return parents[0][1]
1862
def get_parents_with_ghosts(self, version_id):
1863
"""Return parents of specified version with ghosts."""
1864
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1865
check_present=True))
1866
if not self._parents:
1868
return self._keys_to_version_ids(nodes[0][3][0])
1870
def check_versions_present(self, version_ids):
1871
"""Check that all specified versions are present."""
1872
keys = self._version_ids_to_keys(version_ids)
1873
present = self._present_keys(keys)
1874
missing = keys.difference(present)
1876
raise RevisionNotPresent(missing.pop(), self)
1878
def add_version(self, version_id, options, access_memo, parents):
1879
"""Add a version record to the index."""
1880
return self.add_versions(((version_id, options, access_memo, parents),))
1882
def add_versions(self, versions, random_id=False):
1883
"""Add multiple versions to the index.
1885
This function does not insert data into the Immutable GraphIndex
1886
backing the KnitGraphIndex, instead it prepares data for insertion by
1887
the caller and checks that it is safe to insert then calls
1888
self._add_callback with the prepared GraphIndex nodes.
1890
:param versions: a list of tuples:
1891
(version_id, options, pos, size, parents).
1892
:param random_id: If True the ids being added were randomly generated
1893
and no check for existence will be performed.
1895
if not self._add_callback:
1896
raise errors.ReadOnlyError(self)
1897
# we hope there are no repositories with inconsistent parentage
1902
for (version_id, options, access_memo, parents) in versions:
1903
index, pos, size = access_memo
1904
key = (version_id, )
1905
parents = tuple((parent, ) for parent in parents)
1906
if 'no-eol' in options:
1910
value += "%d %d" % (pos, size)
1911
if not self._deltas:
1912
if 'line-delta' in options:
1913
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1916
if 'line-delta' in options:
1917
node_refs = (parents, (parents[0],))
1919
node_refs = (parents, ())
1921
node_refs = (parents, )
1924
raise KnitCorrupt(self, "attempt to add node with parents "
1925
"in parentless index.")
1927
keys[key] = (value, node_refs)
1929
present_nodes = self._get_entries(keys)
1930
for (index, key, value, node_refs) in present_nodes:
1931
if (value, node_refs) != keys[key]:
1932
raise KnitCorrupt(self, "inconsistent details in add_versions"
1933
": %s %s" % ((value, node_refs), keys[key]))
1937
for key, (value, node_refs) in keys.iteritems():
1938
result.append((key, value, node_refs))
1940
for key, (value, node_refs) in keys.iteritems():
1941
result.append((key, value))
1942
self._add_callback(result)
1944
def _version_ids_to_keys(self, version_ids):
1945
return set((version_id, ) for version_id in version_ids)
1948
class _KnitAccess(object):
1949
"""Access to knit records in a .knit file."""
1951
def __init__(self, transport, filename, _file_mode, _dir_mode,
1952
_need_to_create, _create_parent_dir):
1953
"""Create a _KnitAccess for accessing and inserting data.
1955
:param transport: The transport the .knit is located on.
1956
:param filename: The filename of the .knit.
1958
self._transport = transport
1959
self._filename = filename
1960
self._file_mode = _file_mode
1961
self._dir_mode = _dir_mode
1962
self._need_to_create = _need_to_create
1963
self._create_parent_dir = _create_parent_dir
1965
def add_raw_records(self, sizes, raw_data):
1966
"""Add raw knit bytes to a storage area.
1968
The data is spooled to whereever the access method is storing data.
1970
:param sizes: An iterable containing the size of each raw data segment.
1971
:param raw_data: A bytestring containing the data.
1972
:return: A list of memos to retrieve the record later. Each memo is a
1973
tuple - (index, pos, length), where the index field is always None
1974
for the .knit access method.
1976
assert type(raw_data) == str, \
1977
'data must be plain bytes was %s' % type(raw_data)
1978
if not self._need_to_create:
1979
base = self._transport.append_bytes(self._filename, raw_data)
1981
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1982
create_parent_dir=self._create_parent_dir,
1983
mode=self._file_mode,
1984
dir_mode=self._dir_mode)
1985
self._need_to_create = False
1989
result.append((None, base, size))
1994
"""IFF this data access has its own storage area, initialise it.
1998
self._transport.put_bytes_non_atomic(self._filename, '',
1999
mode=self._file_mode)
2001
def open_file(self):
2002
"""IFF this data access can be represented as a single file, open it.
2004
For knits that are not mapped to a single file on disk this will
2007
:return: None or a file handle.
2010
return self._transport.get(self._filename)
2015
def get_raw_records(self, memos_for_retrieval):
2016
"""Get the raw bytes for a records.
2018
:param memos_for_retrieval: An iterable containing the (index, pos,
2019
length) memo for retrieving the bytes. The .knit method ignores
2020
the index as there is always only a single file.
2021
:return: An iterator over the bytes of the records.
2023
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2024
for pos, data in self._transport.readv(self._filename, read_vector):
2028
class _PackAccess(object):
2029
"""Access to knit records via a collection of packs."""
2031
def __init__(self, index_to_packs, writer=None):
2032
"""Create a _PackAccess object.
2034
:param index_to_packs: A dict mapping index objects to the transport
2035
and file names for obtaining data.
2036
:param writer: A tuple (pack.ContainerWriter, write_index) which
2037
contains the pack to write, and the index that reads from it will
2041
self.container_writer = writer[0]
2042
self.write_index = writer[1]
2044
self.container_writer = None
2045
self.write_index = None
2046
self.indices = index_to_packs
2048
def add_raw_records(self, sizes, raw_data):
2049
"""Add raw knit bytes to a storage area.
2051
The data is spooled to the container writer in one bytes-record per
2054
:param sizes: An iterable containing the size of each raw data segment.
2055
:param raw_data: A bytestring containing the data.
2056
:return: A list of memos to retrieve the record later. Each memo is a
2057
tuple - (index, pos, length), where the index field is the
2058
write_index object supplied to the PackAccess object.
2060
assert type(raw_data) == str, \
2061
'data must be plain bytes was %s' % type(raw_data)
2065
p_offset, p_length = self.container_writer.add_bytes_record(
2066
raw_data[offset:offset+size], [])
2068
result.append((self.write_index, p_offset, p_length))
2072
"""Pack based knits do not get individually created."""
2074
def get_raw_records(self, memos_for_retrieval):
2075
"""Get the raw bytes for a records.
2077
:param memos_for_retrieval: An iterable containing the (index, pos,
2078
length) memo for retrieving the bytes. The Pack access method
2079
looks up the pack to use for a given record in its index_to_pack
2081
:return: An iterator over the bytes of the records.
2083
# first pass, group into same-index requests
2085
current_index = None
2086
for (index, offset, length) in memos_for_retrieval:
2087
if current_index == index:
2088
current_list.append((offset, length))
2090
if current_index is not None:
2091
request_lists.append((current_index, current_list))
2092
current_index = index
2093
current_list = [(offset, length)]
2094
# handle the last entry
2095
if current_index is not None:
2096
request_lists.append((current_index, current_list))
2097
for index, offsets in request_lists:
2098
transport, path = self.indices[index]
2099
reader = pack.make_readv_reader(transport, path, offsets)
2100
for names, read_func in reader.iter_records():
2101
yield read_func(None)
2103
def open_file(self):
2104
"""Pack based knits have no single file."""
2107
def set_writer(self, writer, index, (transport, packname)):
2108
"""Set a writer to use for adding data."""
2109
if index is not None:
2110
self.indices[index] = (transport, packname)
2111
self.container_writer = writer
2112
self.write_index = index
2115
class _StreamAccess(object):
2116
"""A Knit Access object that provides data from a datastream.
2118
It also provides a fallback to present as unannotated data, annotated data
2119
from a *backing* access object.
2121
This is triggered by a index_memo which is pointing to a different index
2122
than this was constructed with, and is used to allow extracting full
2123
unannotated texts for insertion into annotated knits.
2126
def __init__(self, reader_callable, stream_index, backing_knit,
2128
"""Create a _StreamAccess object.
2130
:param reader_callable: The reader_callable from the datastream.
2131
This is called to buffer all the data immediately, for
2133
:param stream_index: The index the data stream this provides access to
2134
which will be present in native index_memo's.
2135
:param backing_knit: The knit object that will provide access to
2136
annotated texts which are not available in the stream, so as to
2137
create unannotated texts.
2138
:param orig_factory: The original content factory used to generate the
2139
stream. This is used for checking whether the thunk code for
2140
supporting _copy_texts will generate the correct form of data.
2142
self.data = reader_callable(None)
2143
self.stream_index = stream_index
2144
self.backing_knit = backing_knit
2145
self.orig_factory = orig_factory
2147
def get_raw_records(self, memos_for_retrieval):
2148
"""Get the raw bytes for a records.
2150
:param memos_for_retrieval: An iterable containing the (thunk_flag,
2151
index, start, end) memo for retrieving the bytes.
2152
:return: An iterator over the bytes of the records.
2154
# use a generator for memory friendliness
2155
for thunk_flag, version_id, start, end in memos_for_retrieval:
2156
if version_id is self.stream_index:
2157
yield self.data[start:end]
2159
# we have been asked to thunk. This thunking only occurs when
2160
# we are obtaining plain texts from an annotated backing knit
2161
# so that _copy_texts will work.
2162
# We could improve performance here by scanning for where we need
2163
# to do this and using get_line_list, then interleaving the output
2164
# as desired. However, for now, this is sufficient.
2165
if self.orig_factory.__class__ != KnitPlainFactory:
2166
raise errors.KnitCorrupt(
2167
self, 'Bad thunk request %r' % version_id)
2168
lines = self.backing_knit.get_lines(version_id)
2169
line_bytes = ''.join(lines)
2170
digest = sha_string(line_bytes)
2172
if lines[-1][-1] != '\n':
2173
lines[-1] = lines[-1] + '\n'
2175
orig_options = list(self.backing_knit._index.get_options(version_id))
2176
if 'fulltext' not in orig_options:
2177
if 'line-delta' not in orig_options:
2178
raise errors.KnitCorrupt(self,
2179
'Unknown compression method %r' % orig_options)
2180
orig_options.remove('line-delta')
2181
orig_options.append('fulltext')
2182
# We want plain data, because we expect to thunk only to allow text
2184
size, bytes = self.backing_knit._data._record_to_data(version_id,
2185
digest, lines, line_bytes)
2189
class _StreamIndex(object):
2190
"""A Knit Index object that uses the data map from a datastream."""
2192
def __init__(self, data_list, backing_index):
2193
"""Create a _StreamIndex object.
2195
:param data_list: The data_list from the datastream.
2196
:param backing_index: The index which will supply values for nodes
2197
referenced outside of this stream.
2199
self.data_list = data_list
2200
self.backing_index = backing_index
2201
self._by_version = {}
2203
for key, options, length, parents in data_list:
2204
self._by_version[key] = options, (pos, pos + length), parents
2207
def get_ancestry(self, versions, topo_sorted):
2208
"""Get an ancestry list for versions."""
2210
# Not needed for basic joins
2211
raise NotImplementedError(self.get_ancestry)
2212
# get a graph of all the mentioned versions:
2213
# Little ugly - basically copied from KnitIndex, but don't want to
2214
# accidentally incorporate too much of that index's code.
2216
pending = set(versions)
2217
cache = self._by_version
2219
version = pending.pop()
2222
parents = [p for p in cache[version][2] if p in cache]
2224
raise RevisionNotPresent(version, self)
2225
# if not completed and not a ghost
2226
pending.update([p for p in parents if p not in ancestry])
2227
ancestry.add(version)
2228
return list(ancestry)
2230
def get_build_details(self, version_ids):
2231
"""Get the method, index_memo and compression parent for version_ids.
2233
:param version_ids: An iterable of version_ids.
2234
:return: A dict of version_id:(method, index_memo, compression_parent,
2238
for version_id in version_ids:
2239
method = self.get_method(version_id)
2240
parent_ids = self.get_parents_with_ghosts(version_id)
2241
noeol = ('no-eol' in self.get_options(version_id))
2242
if method == 'fulltext':
2243
compression_parent = None
2245
compression_parent = parent_ids[0]
2246
index_memo = self.get_position(version_id)
2247
result[version_id] = (method, index_memo, compression_parent,
2251
def get_method(self, version_id):
2252
"""Return compression method of specified version."""
2254
options = self._by_version[version_id][0]
2256
# Strictly speaking this should check in the backing knit, but
2257
# until we have a test to discriminate, this will do.
2258
return self.backing_index.get_method(version_id)
2259
if 'fulltext' in options:
2261
elif 'line-delta' in options:
2264
raise errors.KnitIndexUnknownMethod(self, options)
2266
def get_options(self, version_id):
2267
"""Return a list representing options.
2272
return self._by_version[version_id][0]
2274
return self.backing_index.get_options(version_id)
2276
def get_parents_with_ghosts(self, version_id):
2277
"""Return parents of specified version with ghosts."""
2279
return self._by_version[version_id][2]
2281
return self.backing_index.get_parents_with_ghosts(version_id)
2283
def get_position(self, version_id):
2284
"""Return details needed to access the version.
2286
_StreamAccess has the data as a big array, so we return slice
2287
coordinates into that (as index_memo's are opaque outside the
2288
index and matching access class).
2290
:return: a tuple (thunk_flag, index, start, end). If thunk_flag is
2291
False, index will be self, otherwise it will be a version id.
2294
start, end = self._by_version[version_id][1]
2295
return False, self, start, end
2297
# Signal to the access object to handle this from the backing knit.
2298
return (True, version_id, None, None)
2300
def get_versions(self):
2301
"""Get all the versions in the stream."""
2302
return self._by_version.keys()
2304
def iter_parents(self, version_ids):
2305
"""Iterate through the parents for many version ids.
2307
:param version_ids: An iterable yielding version_ids.
2308
:return: An iterator that yields (version_id, parents). Requested
2309
version_ids not present in the versioned file are simply skipped.
2310
The order is undefined, allowing for different optimisations in
2311
the underlying implementation.
2314
for version in version_ids:
2316
result.append((version, self._by_version[version][2]))
2322
class _KnitData(object):
2323
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2325
The KnitData class provides the logic for parsing and using knit records,
2326
making use of an access method for the low level read and write operations.
2329
def __init__(self, access):
2330
"""Create a KnitData object.
2332
:param access: The access method to use. Access methods such as
2333
_KnitAccess manage the insertion of raw records and the subsequent
2334
retrieval of the same.
2336
self._access = access
2337
self._checked = False
2338
# TODO: jam 20060713 conceptually, this could spill to disk
2339
# if the cached size gets larger than a certain amount
2340
# but it complicates the model a bit, so for now just use
2341
# a simple dictionary
2343
self._do_cache = False
2345
def enable_cache(self):
2346
"""Enable caching of reads."""
2347
self._do_cache = True
2349
def clear_cache(self):
2350
"""Clear the record cache."""
2351
self._do_cache = False
2354
def _open_file(self):
2355
return self._access.open_file()
2357
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2358
"""Convert version_id, digest, lines into a raw data block.
2360
:param dense_lines: The bytes of lines but in a denser form. For
2361
instance, if lines is a list of 1000 bytestrings each ending in \n,
2362
dense_lines may be a list with one line in it, containing all the
2363
1000's lines and their \n's. Using dense_lines if it is already
2364
known is a win because the string join to create bytes in this
2365
function spends less time resizing the final string.
2366
:return: (len, a StringIO instance with the raw data ready to read.)
2368
# Note: using a string copy here increases memory pressure with e.g.
2369
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2370
# when doing the initial commit of a mozilla tree. RBC 20070921
2371
bytes = ''.join(chain(
2372
["version %s %d %s\n" % (version_id,
2375
dense_lines or lines,
2376
["end %s\n" % version_id]))
2377
assert bytes.__class__ == str
2378
compressed_bytes = bytes_to_gzip(bytes)
2379
return len(compressed_bytes), compressed_bytes
2381
def add_raw_records(self, sizes, raw_data):
2382
"""Append a prepared record to the data file.
2384
:param sizes: An iterable containing the size of each raw data segment.
2385
:param raw_data: A bytestring containing the data.
2386
:return: a list of index data for the way the data was stored.
2387
See the access method add_raw_records documentation for more
2390
return self._access.add_raw_records(sizes, raw_data)
2392
def _parse_record_header(self, version_id, raw_data):
2393
"""Parse a record header for consistency.
2395
:return: the header and the decompressor stream.
2396
as (stream, header_record)
2398
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2400
rec = self._check_header(version_id, df.readline())
2401
except Exception, e:
2402
raise KnitCorrupt(self._access,
2403
"While reading {%s} got %s(%s)"
2404
% (version_id, e.__class__.__name__, str(e)))
2407
def _check_header(self, version_id, line):
2410
raise KnitCorrupt(self._access,
2411
'unexpected number of elements in record header')
2412
if rec[1] != version_id:
2413
raise KnitCorrupt(self._access,
2414
'unexpected version, wanted %r, got %r'
2415
% (version_id, rec[1]))
2418
def _parse_record(self, version_id, data):
2420
# 4168 calls in 2880 217 internal
2421
# 4168 calls to _parse_record_header in 2121
2422
# 4168 calls to readlines in 330
2423
df = GzipFile(mode='rb', fileobj=StringIO(data))
2426
record_contents = df.readlines()
2427
except Exception, e:
2428
raise KnitCorrupt(self._access,
2429
"While reading {%s} got %s(%s)"
2430
% (version_id, e.__class__.__name__, str(e)))
2431
header = record_contents.pop(0)
2432
rec = self._check_header(version_id, header)
2434
last_line = record_contents.pop()
2435
if len(record_contents) != int(rec[2]):
2436
raise KnitCorrupt(self._access,
2437
'incorrect number of lines %s != %s'
2439
% (len(record_contents), int(rec[2]),
2441
if last_line != 'end %s\n' % rec[1]:
2442
raise KnitCorrupt(self._access,
2443
'unexpected version end line %r, wanted %r'
2444
% (last_line, version_id))
2446
return record_contents, rec[3]
2448
def read_records_iter_raw(self, records):
2449
"""Read text records from data file and yield raw data.
2451
This unpacks enough of the text record to validate the id is
2452
as expected but thats all.
2454
# setup an iterator of the external records:
2455
# uses readv so nice and fast we hope.
2457
# grab the disk data needed.
2459
# Don't check _cache if it is empty
2460
needed_offsets = [index_memo for version_id, index_memo
2462
if version_id not in self._cache]
2464
needed_offsets = [index_memo for version_id, index_memo
2467
raw_records = self._access.get_raw_records(needed_offsets)
2469
for version_id, index_memo in records:
2470
if version_id in self._cache:
2471
# This data has already been validated
2472
data = self._cache[version_id]
2474
data = raw_records.next()
2476
self._cache[version_id] = data
2478
# validate the header
2479
df, rec = self._parse_record_header(version_id, data)
2481
yield version_id, data
2483
def read_records_iter(self, records):
2484
"""Read text records from data file and yield result.
2486
The result will be returned in whatever is the fastest to read.
2487
Not by the order requested. Also, multiple requests for the same
2488
record will only yield 1 response.
2489
:param records: A list of (version_id, pos, len) entries
2490
:return: Yields (version_id, contents, digest) in the order
2491
read, not the order requested
2497
# Skip records we have alread seen
2498
yielded_records = set()
2499
needed_records = set()
2500
for record in records:
2501
if record[0] in self._cache:
2502
if record[0] in yielded_records:
2504
yielded_records.add(record[0])
2505
data = self._cache[record[0]]
2506
content, digest = self._parse_record(record[0], data)
2507
yield (record[0], content, digest)
2509
needed_records.add(record)
2510
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2512
needed_records = sorted(set(records), key=operator.itemgetter(1))
2514
if not needed_records:
2517
# The transport optimizes the fetching as well
2518
# (ie, reads continuous ranges.)
2519
raw_data = self._access.get_raw_records(
2520
[index_memo for version_id, index_memo in needed_records])
2522
for (version_id, index_memo), data in \
2523
izip(iter(needed_records), raw_data):
2524
content, digest = self._parse_record(version_id, data)
2526
self._cache[version_id] = data
2527
yield version_id, content, digest
2529
def read_records(self, records):
2530
"""Read records into a dictionary."""
2532
for record_id, content, digest in \
2533
self.read_records_iter(records):
2534
components[record_id] = (content, digest)
2538
class InterKnit(InterVersionedFile):
2539
"""Optimised code paths for knit to knit operations."""
2541
_matching_file_from_factory = KnitVersionedFile
2542
_matching_file_to_factory = KnitVersionedFile
2545
def is_compatible(source, target):
2546
"""Be compatible with knits. """
2548
return (isinstance(source, KnitVersionedFile) and
2549
isinstance(target, KnitVersionedFile))
2550
except AttributeError:
2553
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2554
"""Copy texts to the target by extracting and adding them one by one.
2556
see join() for the parameter definitions.
2558
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2559
graph = self.source.get_graph(version_ids)
2560
order = topo_sort(graph.items())
2562
def size_of_content(content):
2563
return sum(len(line) for line in content.text())
2564
# Cache at most 10MB of parent texts
2565
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2566
compute_size=size_of_content)
2567
# TODO: jam 20071116 It would be nice to have a streaming interface to
2568
# get multiple texts from a source. The source could be smarter
2569
# about how it handled intermediate stages.
2570
# get_line_list() or make_mpdiffs() seem like a possibility, but
2571
# at the moment they extract all full texts into memory, which
2572
# causes us to store more than our 3x fulltext goal.
2573
# Repository.iter_files_bytes() may be another possibility
2574
to_process = [version for version in order
2575
if version not in self.target]
2576
total = len(to_process)
2577
pb = ui.ui_factory.nested_progress_bar()
2579
for index, version in enumerate(to_process):
2580
pb.update('Converting versioned data', index, total)
2581
sha1, num_bytes, parent_text = self.target.add_lines(version,
2582
self.source.get_parents_with_ghosts(version),
2583
self.source.get_lines(version),
2584
parent_texts=parent_cache)
2585
parent_cache[version] = parent_text
2590
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2591
"""See InterVersionedFile.join."""
2592
assert isinstance(self.source, KnitVersionedFile)
2593
assert isinstance(self.target, KnitVersionedFile)
2595
# If the source and target are mismatched w.r.t. annotations vs
2596
# plain, the data needs to be converted accordingly
2597
if self.source.factory.annotated == self.target.factory.annotated:
2599
elif self.source.factory.annotated:
2600
converter = self._anno_to_plain_converter
2602
# We're converting from a plain to an annotated knit. Copy them
2603
# across by full texts.
2604
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2606
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2610
pb = ui.ui_factory.nested_progress_bar()
2612
version_ids = list(version_ids)
2613
if None in version_ids:
2614
version_ids.remove(None)
2616
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2618
this_versions = set(self.target._index.get_versions())
2619
# XXX: For efficiency we should not look at the whole index,
2620
# we only need to consider the referenced revisions - they
2621
# must all be present, or the method must be full-text.
2622
# TODO, RBC 20070919
2623
needed_versions = self.source_ancestry - this_versions
2625
if not needed_versions:
2627
full_list = topo_sort(self.source.get_graph())
2629
version_list = [i for i in full_list if (not self.target.has_version(i)
2630
and i in needed_versions)]
2634
copy_queue_records = []
2636
for version_id in version_list:
2637
options = self.source._index.get_options(version_id)
2638
parents = self.source._index.get_parents_with_ghosts(version_id)
2639
# check that its will be a consistent copy:
2640
for parent in parents:
2641
# if source has the parent, we must :
2642
# * already have it or
2643
# * have it scheduled already
2644
# otherwise we don't care
2645
assert (self.target.has_version(parent) or
2646
parent in copy_set or
2647
not self.source.has_version(parent))
2648
index_memo = self.source._index.get_position(version_id)
2649
copy_queue_records.append((version_id, index_memo))
2650
copy_queue.append((version_id, options, parents))
2651
copy_set.add(version_id)
2653
# data suck the join:
2655
total = len(version_list)
2658
for (version_id, raw_data), \
2659
(version_id2, options, parents) in \
2660
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2662
assert version_id == version_id2, 'logic error, inconsistent results'
2664
pb.update("Joining knit", count, total)
2666
size, raw_data = converter(raw_data, version_id, options,
2669
size = len(raw_data)
2670
raw_records.append((version_id, options, parents, size))
2671
raw_datum.append(raw_data)
2672
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2677
def _anno_to_plain_converter(self, raw_data, version_id, options,
2679
"""Convert annotated content to plain content."""
2680
data, digest = self.source._data._parse_record(version_id, raw_data)
2681
if 'fulltext' in options:
2682
content = self.source.factory.parse_fulltext(data, version_id)
2683
lines = self.target.factory.lower_fulltext(content)
2685
delta = self.source.factory.parse_line_delta(data, version_id,
2687
lines = self.target.factory.lower_line_delta(delta)
2688
return self.target._data._record_to_data(version_id, digest, lines)
2691
InterVersionedFile.register_optimiser(InterKnit)
2694
class WeaveToKnit(InterVersionedFile):
2695
"""Optimised code paths for weave to knit operations."""
2697
_matching_file_from_factory = bzrlib.weave.WeaveFile
2698
_matching_file_to_factory = KnitVersionedFile
2701
def is_compatible(source, target):
2702
"""Be compatible with weaves to knits."""
2704
return (isinstance(source, bzrlib.weave.Weave) and
2705
isinstance(target, KnitVersionedFile))
2706
except AttributeError:
2709
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2710
"""See InterVersionedFile.join."""
2711
assert isinstance(self.source, bzrlib.weave.Weave)
2712
assert isinstance(self.target, KnitVersionedFile)
2714
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2719
pb = ui.ui_factory.nested_progress_bar()
2721
version_ids = list(version_ids)
2723
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2724
this_versions = set(self.target._index.get_versions())
2725
needed_versions = self.source_ancestry - this_versions
2727
if not needed_versions:
2729
full_list = topo_sort(self.source.get_graph())
2731
version_list = [i for i in full_list if (not self.target.has_version(i)
2732
and i in needed_versions)]
2736
total = len(version_list)
2737
for version_id in version_list:
2738
pb.update("Converting to knit", count, total)
2739
parents = self.source.get_parents(version_id)
2740
# check that its will be a consistent copy:
2741
for parent in parents:
2742
# if source has the parent, we must already have it
2743
assert (self.target.has_version(parent))
2744
self.target.add_lines(
2745
version_id, parents, self.source.get_lines(version_id))
2752
InterVersionedFile.register_optimiser(WeaveToKnit)
2755
# Deprecated, use PatienceSequenceMatcher instead
2756
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2759
def annotate_knit(knit, revision_id):
2760
"""Annotate a knit with no cached annotations.
2762
This implementation is for knits with no cached annotations.
2763
It will work for knits with cached annotations, but this is not
2766
annotator = _KnitAnnotator(knit)
2767
return iter(annotator.get_annotated_lines(revision_id))
2770
class _KnitAnnotator(object):
2771
"""Build up the annotations for a text."""
2773
def __init__(self, knit):
2776
# unannotated lines of various revisions, this will have the final
2778
self._fulltexts = {}
2779
# Content objects, differs from fulltexts because of how final newlines
2780
# are treated by knits. the content objects here will always have a
2782
self._fulltext_contents = {}
2784
# Annotated lines of specific revisions
2785
self._annotated_lines = {}
2787
# Track the raw data for nodes that we could not process yet.
2788
# This maps the revision_id of the base to a list of children that will
2789
# annotated from it.
2790
self._pending_children = {}
2792
self._all_build_details = {}
2793
# The children => parent revision_id graph
2794
self._revision_id_graph = {}
2796
self._heads_provider = None
2798
def _add_fulltext_content(self, revision_id, content_obj, noeol_flag):
2799
self._fulltext_contents[revision_id] = content_obj
2801
content_obj = content_obj.copy()
2802
content_obj.strip_last_line_newline()
2803
fulltext = content_obj.text()
2804
self._fulltexts[revision_id] = fulltext
2805
# XXX: It would probably be good to check the sha1digest here
2808
def _check_parents(self, child, nodes_to_annotate):
2809
"""Check if all parents have been processed.
2811
:param child: A tuple of (rev_id, parents, raw_content)
2812
:param nodes_to_annotate: If child is ready, add it to
2813
nodes_to_annotate, otherwise put it back in self._pending_children
2815
for parent_id in child[1]:
2816
if parent_id not in self._annotated_lines:
2817
# This parent is present, but another parent is missing
2818
self._pending_children.setdefault(parent_id,
2822
# This one is ready to be processed
2823
nodes_to_annotate.append(child)
2825
def _add_annotation(self, revision_id, fulltext, parent_ids,
2826
left_matching_blocks=None):
2827
"""Add an annotation entry.
2829
All parents should already have been annotated.
2830
:return: A list of children that now have their parents satisfied.
2832
a = self._annotated_lines
2833
annotated_parent_lines = [a[p] for p in parent_ids]
2834
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2835
fulltext, revision_id, left_matching_blocks,
2836
heads_provider=self._get_heads_provider()))
2837
self._annotated_lines[revision_id] = annotated_lines
2838
# Now that we've added this one, see if there are any pending
2839
# deltas to be done, certainly this parent is finished
2840
nodes_to_annotate = []
2841
for child in self._pending_children.pop(revision_id, []):
2842
self._check_parents(child, nodes_to_annotate)
2843
return nodes_to_annotate
2845
def _get_build_graph(self, revision_id):
2846
"""Get the graphs for building texts and annotations.
2848
The data you need for creating a full text may be different than the
2849
data you need to annotate that text. (At a minimum, you need both
2850
parents to create an annotation, but only need 1 parent to generate the
2853
:return: A list of (revision_id, index_memo) records, suitable for
2854
passing to read_records_iter to start reading in the raw data from
2857
if revision_id in self._annotated_lines:
2860
pending = set([revision_id])
2863
# get all pending nodes
2864
this_iteration = pending
2865
build_details = self._knit._index.get_build_details(this_iteration)
2866
self._all_build_details.update(build_details)
2867
# new_nodes = self._knit._index._get_entries(this_iteration)
2869
for rev_id, details in build_details.iteritems():
2870
(method, index_memo, compression_parent, parents,
2872
self._revision_id_graph[rev_id] = parents
2873
records.append((rev_id, index_memo))
2874
pending.update(p for p in parents
2875
if p not in self._all_build_details)
2877
missing_versions = this_iteration.difference(build_details.keys())
2878
for missing_version in missing_versions:
2879
# add a key, no parents
2880
self._revision_id_graph[missing_versions] = ()
2881
pending.discard(missing_version) # don't look for it
2882
# Generally we will want to read the records in reverse order, because
2883
# we find the parent nodes after the children
2887
def _annotate_records(self, records):
2888
"""Build the annotations for the listed records."""
2889
# We iterate in the order read, rather than a strict order requested
2890
# However, process what we can, and put off to the side things that still
2891
# need parents, cleaning them up when those parents are processed.
2892
for (rev_id, raw_content,
2893
digest) in self._knit._data.read_records_iter(records):
2894
if rev_id in self._annotated_lines:
2896
parent_ids = self._revision_id_graph[rev_id]
2897
details = self._all_build_details[rev_id]
2898
(method, index_memo, compression_parent, parent_ids,
2900
nodes_to_annotate = []
2901
# TODO: Remove the punning between compression parents, and
2902
# parent_ids, we should be able to do this without assuming
2904
if len(parent_ids) == 0:
2905
# There are no parents for this node, so just add it
2906
# TODO: This probably needs to be decoupled
2907
assert compression_parent is None and method == 'fulltext'
2908
fulltext_content = self._knit.factory.parse_fulltext(
2909
raw_content, rev_id)
2910
fulltext = self._add_fulltext_content(rev_id, fulltext_content,
2912
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
2913
parent_ids, left_matching_blocks=None))
2915
child = (rev_id, parent_ids, raw_content)
2916
# Check if all the parents are present
2917
self._check_parents(child, nodes_to_annotate)
2918
while nodes_to_annotate:
2919
# Should we use a queue here instead of a stack?
2920
(rev_id, parent_ids, raw_content) = nodes_to_annotate.pop()
2921
(method, index_memo, compression_parent, parent_ids,
2922
noeol) = self._all_build_details[rev_id]
2923
if method == 'line-delta':
2924
parent_fulltext_content = self._fulltext_contents[compression_parent]
2925
delta = self._knit.factory.parse_line_delta(raw_content,
2927
# TODO: only copy when the parent is still needed elsewhere
2928
fulltext_content = parent_fulltext_content.copy()
2929
fulltext_content.apply_delta(delta, rev_id)
2930
fulltext = self._add_fulltext_content(rev_id,
2931
fulltext_content, noeol)
2932
parent_fulltext = self._fulltexts[parent_ids[0]]
2933
blocks = KnitContent.get_line_delta_blocks(delta,
2934
parent_fulltext, fulltext)
2936
assert method == 'fulltext'
2937
fulltext_content = self._knit.factory.parse_fulltext(
2938
raw_content, rev_id)
2939
fulltext = self._add_fulltext_content(rev_id,
2940
fulltext_content, noeol)
2942
nodes_to_annotate.extend(
2943
self._add_annotation(rev_id, fulltext, parent_ids,
2944
left_matching_blocks=blocks))
2946
def _get_heads_provider(self):
2947
"""Create a heads provider for resolving ancestry issues."""
2948
if self._heads_provider is not None:
2949
return self._heads_provider
2950
parent_provider = _mod_graph.DictParentsProvider(
2951
self._revision_id_graph)
2952
graph_obj = _mod_graph.Graph(parent_provider)
2953
head_cache = _mod_graph.HeadsCache(graph_obj)
2954
self._heads_provider = head_cache
2957
def get_annotated_lines(self, revision_id):
2958
"""Return the annotated fulltext at the given revision.
2960
:param revision_id: The revision id for this file
2962
records = self._get_build_graph(revision_id)
2963
self._annotate_records(records)
2964
return self._annotated_lines[revision_id]
2968
from bzrlib._knit_load_data_c import _load_data_c as _load_data
2970
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
