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 ?
63
from cStringIO import StringIO
64
from itertools import izip, chain
69
from zlib import Z_DEFAULT_COMPRESSION
72
from bzrlib.lazy_import import lazy_import
73
lazy_import(globals(), """
93
from bzrlib.errors import (
101
RevisionAlreadyPresent,
103
from bzrlib.graph import Graph
104
from bzrlib.osutils import (
111
from bzrlib.tsort import topo_sort
112
from bzrlib.tuned_gzip import GzipFile, bytes_to_gzip
114
from bzrlib.versionedfile import (
115
AbsentContentFactory,
124
# TODO: Split out code specific to this format into an associated object.
126
# TODO: Can we put in some kind of value to check that the index and data
127
# files belong together?
129
# TODO: accommodate binaries, perhaps by storing a byte count
131
# TODO: function to check whole file
133
# TODO: atomically append data, then measure backwards from the cursor
134
# position after writing to work out where it was located. we may need to
135
# bypass python file buffering.
137
DATA_SUFFIX = '.knit'
138
INDEX_SUFFIX = '.kndx'
141
class KnitAdapter(object):
142
"""Base class for knit record adaption."""
144
def __init__(self, basis_vf):
145
"""Create an adapter which accesses full texts from basis_vf.
147
:param basis_vf: A versioned file to access basis texts of deltas from.
148
May be None for adapters that do not need to access basis texts.
150
self._data = _KnitData(None)
151
self._annotate_factory = KnitAnnotateFactory()
152
self._plain_factory = KnitPlainFactory()
153
self._basis_vf = basis_vf
156
class FTAnnotatedToUnannotated(KnitAdapter):
157
"""An adapter from FT annotated knits to unannotated ones."""
159
def get_bytes(self, factory, annotated_compressed_bytes):
161
self._data._parse_record_unchecked(annotated_compressed_bytes)
162
content = self._annotate_factory.parse_fulltext(contents, rec[1])
163
size, bytes = self._data._record_to_data(rec[1], rec[3], content.text())
167
class DeltaAnnotatedToUnannotated(KnitAdapter):
168
"""An adapter for deltas from annotated to unannotated."""
170
def get_bytes(self, factory, annotated_compressed_bytes):
172
self._data._parse_record_unchecked(annotated_compressed_bytes)
173
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
175
contents = self._plain_factory.lower_line_delta(delta)
176
size, bytes = self._data._record_to_data(rec[1], rec[3], contents)
180
class FTAnnotatedToFullText(KnitAdapter):
181
"""An adapter from FT annotated knits to unannotated ones."""
183
def get_bytes(self, factory, annotated_compressed_bytes):
185
self._data._parse_record_unchecked(annotated_compressed_bytes)
186
content, delta = self._annotate_factory.parse_record(factory.key[0],
187
contents, factory._build_details, None)
188
return ''.join(content.text())
191
class DeltaAnnotatedToFullText(KnitAdapter):
192
"""An adapter for deltas from annotated to unannotated."""
194
def get_bytes(self, factory, annotated_compressed_bytes):
196
self._data._parse_record_unchecked(annotated_compressed_bytes)
197
delta = self._annotate_factory.parse_line_delta(contents, rec[1],
199
compression_parent = factory.parents[0][0]
200
basis_lines = self._basis_vf.get_lines(compression_parent)
201
# Manually apply the delta because we have one annotated content and
203
basis_content = PlainKnitContent(basis_lines, compression_parent)
204
basis_content.apply_delta(delta, rec[1])
205
basis_content._should_strip_eol = factory._build_details[1]
206
return ''.join(basis_content.text())
209
class FTPlainToFullText(KnitAdapter):
210
"""An adapter from FT plain knits to unannotated ones."""
212
def get_bytes(self, factory, compressed_bytes):
214
self._data._parse_record_unchecked(compressed_bytes)
215
content, delta = self._plain_factory.parse_record(factory.key[0],
216
contents, factory._build_details, None)
217
return ''.join(content.text())
220
class DeltaPlainToFullText(KnitAdapter):
221
"""An adapter for deltas from annotated to unannotated."""
223
def get_bytes(self, factory, compressed_bytes):
225
self._data._parse_record_unchecked(compressed_bytes)
226
delta = self._plain_factory.parse_line_delta(contents, rec[1])
227
compression_parent = factory.parents[0][0]
228
basis_lines = self._basis_vf.get_lines(compression_parent)
229
basis_content = PlainKnitContent(basis_lines, compression_parent)
230
# Manually apply the delta because we have one annotated content and
232
content, _ = self._plain_factory.parse_record(rec[1], contents,
233
factory._build_details, basis_content)
234
return ''.join(content.text())
237
class KnitContentFactory(ContentFactory):
238
"""Content factory for streaming from knits.
240
:seealso ContentFactory:
243
def __init__(self, version, parents, build_details, sha1, raw_record,
244
annotated, knit=None):
245
"""Create a KnitContentFactory for version.
247
:param version: The version.
248
:param parents: The parents.
249
:param build_details: The build details as returned from
251
:param sha1: The sha1 expected from the full text of this object.
252
:param raw_record: The bytes of the knit data from disk.
253
:param annotated: True if the raw data is annotated.
255
ContentFactory.__init__(self)
257
self.key = (version,)
258
self.parents = tuple((parent,) for parent in parents)
259
if build_details[0] == 'line-delta':
264
annotated_kind = 'annotated-'
267
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
268
self._raw_record = raw_record
269
self._build_details = build_details
272
def get_bytes_as(self, storage_kind):
273
if storage_kind == self.storage_kind:
274
return self._raw_record
275
if storage_kind == 'fulltext' and self._knit is not None:
276
return self._knit.get_text(self.key[0])
278
raise errors.UnavailableRepresentation(self.key, storage_kind,
282
class KnitContent(object):
283
"""Content of a knit version to which deltas can be applied.
285
This is always stored in memory as a list of lines with \n at the end,
286
plus a flag saying if the final ending is really there or not, because that
287
corresponds to the on-disk knit representation.
291
self._should_strip_eol = False
293
def apply_delta(self, delta, new_version_id):
294
"""Apply delta to this object to become new_version_id."""
295
raise NotImplementedError(self.apply_delta)
297
def line_delta_iter(self, new_lines):
298
"""Generate line-based delta from this content to new_lines."""
299
new_texts = new_lines.text()
300
old_texts = self.text()
301
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
302
for tag, i1, i2, j1, j2 in s.get_opcodes():
305
# ofrom, oto, length, data
306
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
308
def line_delta(self, new_lines):
309
return list(self.line_delta_iter(new_lines))
312
def get_line_delta_blocks(knit_delta, source, target):
313
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
314
target_len = len(target)
317
for s_begin, s_end, t_len, new_text in knit_delta:
318
true_n = s_begin - s_pos
321
# knit deltas do not provide reliable info about whether the
322
# last line of a file matches, due to eol handling.
323
if source[s_pos + n -1] != target[t_pos + n -1]:
326
yield s_pos, t_pos, n
327
t_pos += t_len + true_n
329
n = target_len - t_pos
331
if source[s_pos + n -1] != target[t_pos + n -1]:
334
yield s_pos, t_pos, n
335
yield s_pos + (target_len - t_pos), target_len, 0
338
class AnnotatedKnitContent(KnitContent):
339
"""Annotated content."""
341
def __init__(self, lines):
342
KnitContent.__init__(self)
346
"""Return a list of (origin, text) for each content line."""
347
lines = self._lines[:]
348
if self._should_strip_eol:
349
origin, last_line = lines[-1]
350
lines[-1] = (origin, last_line.rstrip('\n'))
353
def apply_delta(self, delta, new_version_id):
354
"""Apply delta to this object to become new_version_id."""
357
for start, end, count, delta_lines in delta:
358
lines[offset+start:offset+end] = delta_lines
359
offset = offset + (start - end) + count
363
lines = [text for origin, text in self._lines]
364
except ValueError, e:
365
# most commonly (only?) caused by the internal form of the knit
366
# missing annotation information because of a bug - see thread
368
raise KnitCorrupt(self,
369
"line in annotated knit missing annotation information: %s"
371
if self._should_strip_eol:
372
lines[-1] = lines[-1].rstrip('\n')
376
return AnnotatedKnitContent(self._lines[:])
379
class PlainKnitContent(KnitContent):
380
"""Unannotated content.
382
When annotate[_iter] is called on this content, the same version is reported
383
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
387
def __init__(self, lines, version_id):
388
KnitContent.__init__(self)
390
self._version_id = version_id
393
"""Return a list of (origin, text) for each content line."""
394
return [(self._version_id, line) for line in self._lines]
396
def apply_delta(self, delta, new_version_id):
397
"""Apply delta to this object to become new_version_id."""
400
for start, end, count, delta_lines in delta:
401
lines[offset+start:offset+end] = delta_lines
402
offset = offset + (start - end) + count
403
self._version_id = new_version_id
406
return PlainKnitContent(self._lines[:], self._version_id)
410
if self._should_strip_eol:
412
lines[-1] = lines[-1].rstrip('\n')
416
class _KnitFactory(object):
417
"""Base class for common Factory functions."""
419
def parse_record(self, version_id, record, record_details,
420
base_content, copy_base_content=True):
421
"""Parse a record into a full content object.
423
:param version_id: The official version id for this content
424
:param record: The data returned by read_records_iter()
425
:param record_details: Details about the record returned by
427
:param base_content: If get_build_details returns a compression_parent,
428
you must return a base_content here, else use None
429
:param copy_base_content: When building from the base_content, decide
430
you can either copy it and return a new object, or modify it in
432
:return: (content, delta) A Content object and possibly a line-delta,
435
method, noeol = record_details
436
if method == 'line-delta':
437
if copy_base_content:
438
content = base_content.copy()
440
content = base_content
441
delta = self.parse_line_delta(record, version_id)
442
content.apply_delta(delta, version_id)
444
content = self.parse_fulltext(record, version_id)
446
content._should_strip_eol = noeol
447
return (content, delta)
450
class KnitAnnotateFactory(_KnitFactory):
451
"""Factory for creating annotated Content objects."""
455
def make(self, lines, version_id):
456
num_lines = len(lines)
457
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
459
def parse_fulltext(self, content, version_id):
460
"""Convert fulltext to internal representation
462
fulltext content is of the format
463
revid(utf8) plaintext\n
464
internal representation is of the format:
467
# TODO: jam 20070209 The tests expect this to be returned as tuples,
468
# but the code itself doesn't really depend on that.
469
# Figure out a way to not require the overhead of turning the
470
# list back into tuples.
471
lines = [tuple(line.split(' ', 1)) for line in content]
472
return AnnotatedKnitContent(lines)
474
def parse_line_delta_iter(self, lines):
475
return iter(self.parse_line_delta(lines))
477
def parse_line_delta(self, lines, version_id, plain=False):
478
"""Convert a line based delta into internal representation.
480
line delta is in the form of:
481
intstart intend intcount
483
revid(utf8) newline\n
484
internal representation is
485
(start, end, count, [1..count tuples (revid, newline)])
487
:param plain: If True, the lines are returned as a plain
488
list without annotations, not as a list of (origin, content) tuples, i.e.
489
(start, end, count, [1..count newline])
496
def cache_and_return(line):
497
origin, text = line.split(' ', 1)
498
return cache.setdefault(origin, origin), text
500
# walk through the lines parsing.
501
# Note that the plain test is explicitly pulled out of the
502
# loop to minimise any performance impact
505
start, end, count = [int(n) for n in header.split(',')]
506
contents = [next().split(' ', 1)[1] for i in xrange(count)]
507
result.append((start, end, count, contents))
510
start, end, count = [int(n) for n in header.split(',')]
511
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
512
result.append((start, end, count, contents))
515
def get_fulltext_content(self, lines):
516
"""Extract just the content lines from a fulltext."""
517
return (line.split(' ', 1)[1] for line in lines)
519
def get_linedelta_content(self, lines):
520
"""Extract just the content from a line delta.
522
This doesn't return all of the extra information stored in a delta.
523
Only the actual content lines.
528
header = header.split(',')
529
count = int(header[2])
530
for i in xrange(count):
531
origin, text = next().split(' ', 1)
534
def lower_fulltext(self, content):
535
"""convert a fulltext content record into a serializable form.
537
see parse_fulltext which this inverts.
539
# TODO: jam 20070209 We only do the caching thing to make sure that
540
# the origin is a valid utf-8 line, eventually we could remove it
541
return ['%s %s' % (o, t) for o, t in content._lines]
543
def lower_line_delta(self, delta):
544
"""convert a delta into a serializable form.
546
See parse_line_delta which this inverts.
548
# TODO: jam 20070209 We only do the caching thing to make sure that
549
# the origin is a valid utf-8 line, eventually we could remove it
551
for start, end, c, lines in delta:
552
out.append('%d,%d,%d\n' % (start, end, c))
553
out.extend(origin + ' ' + text
554
for origin, text in lines)
557
def annotate(self, knit, version_id):
558
content = knit._get_content(version_id)
559
return content.annotate()
562
class KnitPlainFactory(_KnitFactory):
563
"""Factory for creating plain Content objects."""
567
def make(self, lines, version_id):
568
return PlainKnitContent(lines, version_id)
570
def parse_fulltext(self, content, version_id):
571
"""This parses an unannotated fulltext.
573
Note that this is not a noop - the internal representation
574
has (versionid, line) - its just a constant versionid.
576
return self.make(content, version_id)
578
def parse_line_delta_iter(self, lines, version_id):
580
num_lines = len(lines)
581
while cur < num_lines:
584
start, end, c = [int(n) for n in header.split(',')]
585
yield start, end, c, lines[cur:cur+c]
588
def parse_line_delta(self, lines, version_id):
589
return list(self.parse_line_delta_iter(lines, version_id))
591
def get_fulltext_content(self, lines):
592
"""Extract just the content lines from a fulltext."""
595
def get_linedelta_content(self, lines):
596
"""Extract just the content from a line delta.
598
This doesn't return all of the extra information stored in a delta.
599
Only the actual content lines.
604
header = header.split(',')
605
count = int(header[2])
606
for i in xrange(count):
609
def lower_fulltext(self, content):
610
return content.text()
612
def lower_line_delta(self, delta):
614
for start, end, c, lines in delta:
615
out.append('%d,%d,%d\n' % (start, end, c))
619
def annotate(self, knit, version_id):
620
annotator = _KnitAnnotator(knit)
621
return annotator.annotate(version_id)
624
def make_empty_knit(transport, relpath):
625
"""Construct a empty knit at the specified location."""
626
k = make_file_knit(transport, relpath, 'w', KnitPlainFactory)
629
def make_file_knit(name, transport, file_mode=None, access_mode='w',
630
factory=None, delta=True, create=False, create_parent_dir=False,
631
delay_create=False, dir_mode=None, get_scope=None):
632
"""Factory to create a KnitVersionedFile for a .knit/.kndx file pair."""
634
factory = KnitAnnotateFactory()
635
if get_scope is None:
636
get_scope = lambda:None
637
index = _KnitIndex(transport, name + INDEX_SUFFIX,
638
access_mode, create=create, file_mode=file_mode,
639
create_parent_dir=create_parent_dir, delay_create=delay_create,
640
dir_mode=dir_mode, get_scope=get_scope)
641
access = _KnitAccess(transport, name + DATA_SUFFIX, file_mode,
642
dir_mode, ((create and not len(index)) and delay_create),
644
return KnitVersionedFile(name, transport, factory=factory,
645
create=create, delay_create=delay_create, index=index,
646
access_method=access)
650
"""Return the suffixes used by file based knits."""
651
return [DATA_SUFFIX, INDEX_SUFFIX]
652
make_file_knit.get_suffixes = get_suffixes
655
class KnitVersionedFile(VersionedFile):
656
"""Weave-like structure with faster random access.
658
A knit stores a number of texts and a summary of the relationships
659
between them. Texts are identified by a string version-id. Texts
660
are normally stored and retrieved as a series of lines, but can
661
also be passed as single strings.
663
Lines are stored with the trailing newline (if any) included, to
664
avoid special cases for files with no final newline. Lines are
665
composed of 8-bit characters, not unicode. The combination of
666
these approaches should mean any 'binary' file can be safely
667
stored and retrieved.
670
def __init__(self, relpath, transport, file_mode=None,
671
factory=None, delta=True, create=False, create_parent_dir=False,
672
delay_create=False, dir_mode=None, index=None, access_method=None):
673
"""Construct a knit at location specified by relpath.
675
:param create: If not True, only open an existing knit.
676
:param create_parent_dir: If True, create the parent directory if
677
creating the file fails. (This is used for stores with
678
hash-prefixes that may not exist yet)
679
:param delay_create: The calling code is aware that the knit won't
680
actually be created until the first data is stored.
681
:param index: An index to use for the knit.
683
super(KnitVersionedFile, self).__init__()
684
self.transport = transport
685
self.filename = relpath
686
self.factory = factory or KnitAnnotateFactory()
689
self._max_delta_chain = 200
691
if None in (access_method, index):
692
raise ValueError("No default access_method or index any more")
694
_access = access_method
695
if create and not len(self) and not delay_create:
697
self._data = _KnitData(_access)
700
return '%s(%s)' % (self.__class__.__name__,
701
self.transport.abspath(self.filename))
703
def _check_should_delta(self, first_parents):
704
"""Iterate back through the parent listing, looking for a fulltext.
706
This is used when we want to decide whether to add a delta or a new
707
fulltext. It searches for _max_delta_chain parents. When it finds a
708
fulltext parent, it sees if the total size of the deltas leading up to
709
it is large enough to indicate that we want a new full text anyway.
711
Return True if we should create a new delta, False if we should use a
716
delta_parents = first_parents
717
for count in xrange(self._max_delta_chain):
718
parent = delta_parents[0]
719
method = self._index.get_method(parent)
720
index, pos, size = self._index.get_position(parent)
721
if method == 'fulltext':
725
delta_parents = self._index.get_parent_map([parent])[parent]
727
# We couldn't find a fulltext, so we must create a new one
730
return fulltext_size > delta_size
732
def _check_write_ok(self):
733
return self._index._check_write_ok()
735
def _add_raw_records(self, records, data):
736
"""Add all the records 'records' with data pre-joined in 'data'.
738
:param records: A list of tuples(version_id, options, parents, size).
739
:param data: The data for the records. When it is written, the records
740
are adjusted to have pos pointing into data by the sum of
741
the preceding records sizes.
744
raw_record_sizes = [record[3] for record in records]
745
positions = self._data.add_raw_records(raw_record_sizes, data)
747
for (version_id, options, parents, _), access_memo in zip(
749
index_entries.append((version_id, options, access_memo, parents))
750
self._index.add_versions(index_entries)
752
def copy_to(self, name, transport):
753
"""See VersionedFile.copy_to()."""
754
# copy the current index to a temp index to avoid racing with local
756
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
757
self.transport.get(self._index._filename))
759
f = self._data._open_file()
761
transport.put_file(name + DATA_SUFFIX, f)
764
# move the copied index into place
765
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
767
def get_data_stream(self, required_versions):
768
"""Get a data stream for the specified versions.
770
Versions may be returned in any order, not necessarily the order
771
specified. They are returned in a partial order by compression
772
parent, so that the deltas can be applied as the data stream is
773
inserted; however note that compression parents will not be sent
774
unless they were specifically requested, as the client may already
777
:param required_versions: The exact set of versions to be extracted.
778
Unlike some other knit methods, this is not used to generate a
779
transitive closure, rather it is used precisely as given.
781
:returns: format_signature, list of (version, options, length, parents),
784
required_version_set = frozenset(required_versions)
786
# list of revisions that can just be sent without waiting for their
789
# map from revision to the children based on it
791
# first, read all relevant index data, enough to sort into the right
793
for version_id in required_versions:
794
options = self._index.get_options(version_id)
795
parents = self._index.get_parents_with_ghosts(version_id)
796
index_memo = self._index.get_position(version_id)
797
version_index[version_id] = (index_memo, options, parents)
798
if ('line-delta' in options
799
and parents[0] in required_version_set):
800
# must wait until the parent has been sent
801
deferred.setdefault(parents[0], []). \
804
# either a fulltext, or a delta whose parent the client did
805
# not ask for and presumably already has
806
ready_to_send.append(version_id)
807
# build a list of results to return, plus instructions for data to
809
copy_queue_records = []
810
temp_version_list = []
812
# XXX: pushing and popping lists may be a bit inefficient
813
version_id = ready_to_send.pop(0)
814
(index_memo, options, parents) = version_index[version_id]
815
copy_queue_records.append((version_id, index_memo))
816
none, data_pos, data_size = index_memo
817
temp_version_list.append((version_id, options, data_size,
819
if version_id in deferred:
820
# now we can send all the children of this revision - we could
821
# put them in anywhere, but we hope that sending them soon
822
# after the fulltext will give good locality in the receiver
823
ready_to_send[:0] = deferred.pop(version_id)
824
if not (len(deferred) == 0):
825
raise AssertionError("Still have compressed child versions waiting to be sent")
826
# XXX: The stream format is such that we cannot stream it - we have to
827
# know the length of all the data a-priori.
829
result_version_list = []
830
for (version_id, raw_data, _), \
831
(version_id2, options, _, parents) in \
832
izip(self._data.read_records_iter_raw(copy_queue_records),
834
if not (version_id == version_id2):
835
raise AssertionError('logic error, inconsistent results')
836
raw_datum.append(raw_data)
837
result_version_list.append(
838
(version_id, options, len(raw_data), parents))
839
# provide a callback to get data incrementally.
840
pseudo_file = StringIO(''.join(raw_datum))
843
return pseudo_file.read()
845
return pseudo_file.read(length)
846
return (self.get_format_signature(), result_version_list, read)
848
def get_record_stream(self, versions, ordering, include_delta_closure):
849
"""Get a stream of records for versions.
851
:param versions: The versions to include. Each version is a tuple
853
:param ordering: Either 'unordered' or 'topological'. A topologically
854
sorted stream has compression parents strictly before their
856
:param include_delta_closure: If True then the closure across any
857
compression parents will be included (in the opaque data).
858
:return: An iterator of ContentFactory objects, each of which is only
859
valid until the iterator is advanced.
861
if include_delta_closure:
862
# Nb: what we should do is plan the data to stream to allow
863
# reconstruction of all the texts without excessive buffering,
864
# including re-sending common bases as needed. This makes the most
865
# sense when we start serialising these streams though, so for now
866
# we just fallback to individual text construction behind the
867
# abstraction barrier.
871
# We end up doing multiple index lookups here for parents details and
872
# disk layout details - we need a unified api ?
873
parent_map = self.get_parent_map(versions)
874
absent_versions = set(versions) - set(parent_map)
875
if ordering == 'topological':
876
present_versions = topo_sort(parent_map)
878
# List comprehension to keep the requested order (as that seems
879
# marginally useful, at least until we start doing IO optimising
881
present_versions = [version for version in versions if version in
883
position_map = self._get_components_positions(present_versions)
884
records = [(version, position_map[version][1]) for version in
887
for version in absent_versions:
888
yield AbsentContentFactory((version,))
889
for version, raw_data, sha1 in \
890
self._data.read_records_iter_raw(records):
891
(record_details, index_memo, _) = position_map[version]
892
yield KnitContentFactory(version, parent_map[version],
893
record_details, sha1, raw_data, self.factory.annotated, knit)
895
def _extract_blocks(self, version_id, source, target):
896
if self._index.get_method(version_id) != 'line-delta':
898
parent, sha1, noeol, delta = self.get_delta(version_id)
899
return KnitContent.get_line_delta_blocks(delta, source, target)
901
def get_delta(self, version_id):
902
"""Get a delta for constructing version from some other version."""
903
self.check_not_reserved_id(version_id)
904
parents = self.get_parent_map([version_id])[version_id]
909
index_memo = self._index.get_position(version_id)
910
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
911
noeol = 'no-eol' in self._index.get_options(version_id)
912
if 'fulltext' == self._index.get_method(version_id):
913
new_content = self.factory.parse_fulltext(data, version_id)
914
if parent is not None:
915
reference_content = self._get_content(parent)
916
old_texts = reference_content.text()
919
new_texts = new_content.text()
920
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
922
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
924
delta = self.factory.parse_line_delta(data, version_id)
925
return parent, sha1, noeol, delta
927
def get_format_signature(self):
928
"""See VersionedFile.get_format_signature()."""
929
if self.factory.annotated:
930
annotated_part = "annotated"
932
annotated_part = "plain"
933
return "knit-%s" % (annotated_part,)
935
def get_sha1s(self, version_ids):
936
"""See VersionedFile.get_sha1s()."""
937
record_map = self._get_record_map(version_ids)
938
# record entry 2 is the 'digest'.
939
return [record_map[v][2] for v in version_ids]
941
def insert_data_stream(self, (format, data_list, reader_callable)):
942
"""Insert knit records from a data stream into this knit.
944
If a version in the stream is already present in this knit, it will not
945
be inserted a second time. It will be checked for consistency with the
946
stored version however, and may cause a KnitCorrupt error to be raised
947
if the data in the stream disagrees with the already stored data.
949
:seealso: get_data_stream
951
if format != self.get_format_signature():
952
if 'knit' in debug.debug_flags:
954
'incompatible format signature inserting to %r', self)
955
source = self._knit_from_datastream(
956
(format, data_list, reader_callable))
957
stream = source.get_record_stream(source.versions(), 'unordered', False)
958
self.insert_record_stream(stream)
961
for version_id, options, length, parents in data_list:
962
if self.has_version(version_id):
963
# First check: the list of parents.
964
my_parents = self.get_parents_with_ghosts(version_id)
965
if tuple(my_parents) != tuple(parents):
966
# XXX: KnitCorrupt is not quite the right exception here.
969
'parents list %r from data stream does not match '
970
'already recorded parents %r for %s'
971
% (parents, my_parents, version_id))
973
# Also check the SHA-1 of the fulltext this content will
975
raw_data = reader_callable(length)
976
my_fulltext_sha1 = self.get_sha1s([version_id])[0]
977
df, rec = self._data._parse_record_header(version_id, raw_data)
978
stream_fulltext_sha1 = rec[3]
979
if my_fulltext_sha1 != stream_fulltext_sha1:
980
# Actually, we don't know if it's this knit that's corrupt,
981
# or the data stream we're trying to insert.
983
self.filename, 'sha-1 does not match %s' % version_id)
985
if 'line-delta' in options:
986
# Make sure that this knit record is actually useful: a
987
# line-delta is no use unless we have its parent.
988
# Fetching from a broken repository with this problem
989
# shouldn't break the target repository.
991
# See https://bugs.launchpad.net/bzr/+bug/164443
992
if not self._index.has_version(parents[0]):
995
'line-delta from stream '
998
'missing parent %s\n'
999
'Try running "bzr check" '
1000
'on the source repository, and "bzr reconcile" '
1002
(version_id, parents[0]))
1004
# We received a line-delta record for a non-delta knit.
1005
# Convert it to a fulltext.
1006
gzip_bytes = reader_callable(length)
1007
self._convert_line_delta_to_fulltext(
1008
gzip_bytes, version_id, parents)
1011
self._add_raw_records(
1012
[(version_id, options, parents, length)],
1013
reader_callable(length))
1015
def _convert_line_delta_to_fulltext(self, gzip_bytes, version_id, parents):
1016
lines, sha1 = self._data._parse_record(version_id, gzip_bytes)
1017
delta = self.factory.parse_line_delta(lines, version_id)
1018
content = self.factory.make(self.get_lines(parents[0]), parents[0])
1019
content.apply_delta(delta, version_id)
1020
digest, len, content = self.add_lines(
1021
version_id, parents, content.text())
1023
raise errors.VersionedFileInvalidChecksum(version_id)
1025
def _knit_from_datastream(self, (format, data_list, reader_callable)):
1026
"""Create a knit object from a data stream.
1028
This method exists to allow conversion of data streams that do not
1029
match the signature of this knit. Generally it will be slower and use
1030
more memory to use this method to insert data, but it will work.
1032
:seealso: get_data_stream for details on datastreams.
1033
:return: A knit versioned file which can be used to join the datastream
1036
if format == "knit-plain":
1037
factory = KnitPlainFactory()
1038
elif format == "knit-annotated":
1039
factory = KnitAnnotateFactory()
1041
raise errors.KnitDataStreamUnknown(format)
1042
index = _StreamIndex(data_list, self._index)
1043
access = _StreamAccess(reader_callable, index, self, factory)
1044
return KnitVersionedFile(self.filename, self.transport,
1045
factory=factory, index=index, access_method=access)
1047
def insert_record_stream(self, stream):
1048
"""Insert a record stream into this versioned file.
1050
:param stream: A stream of records to insert.
1052
:seealso VersionedFile.get_record_stream:
1054
def get_adapter(adapter_key):
1056
return adapters[adapter_key]
1058
adapter_factory = adapter_registry.get(adapter_key)
1059
adapter = adapter_factory(self)
1060
adapters[adapter_key] = adapter
1062
if self.factory.annotated:
1063
# self is annotated, we need annotated knits to use directly.
1064
annotated = "annotated-"
1067
# self is not annotated, but we can strip annotations cheaply.
1069
convertibles = set(["knit-annotated-delta-gz",
1070
"knit-annotated-ft-gz"])
1071
# The set of types we can cheaply adapt without needing basis texts.
1072
native_types = set()
1073
native_types.add("knit-%sdelta-gz" % annotated)
1074
native_types.add("knit-%sft-gz" % annotated)
1075
knit_types = native_types.union(convertibles)
1077
# Buffer all index entries that we can't add immediately because their
1078
# basis parent is missing. We don't buffer all because generating
1079
# annotations may require access to some of the new records. However we
1080
# can't generate annotations from new deltas until their basis parent
1081
# is present anyway, so we get away with not needing an index that
1082
# includes the new keys.
1083
# key = basis_parent, value = index entry to add
1084
buffered_index_entries = {}
1085
for record in stream:
1086
# Raise an error when a record is missing.
1087
if record.storage_kind == 'absent':
1088
raise RevisionNotPresent([record.key[0]], self)
1089
# adapt to non-tuple interface
1090
parents = [parent[0] for parent in record.parents]
1091
if record.storage_kind in knit_types:
1092
if record.storage_kind not in native_types:
1094
adapter_key = (record.storage_kind, "knit-delta-gz")
1095
adapter = get_adapter(adapter_key)
1097
adapter_key = (record.storage_kind, "knit-ft-gz")
1098
adapter = get_adapter(adapter_key)
1099
bytes = adapter.get_bytes(
1100
record, record.get_bytes_as(record.storage_kind))
1102
bytes = record.get_bytes_as(record.storage_kind)
1103
options = [record._build_details[0]]
1104
if record._build_details[1]:
1105
options.append('no-eol')
1106
# Just blat it across.
1107
# Note: This does end up adding data on duplicate keys. As
1108
# modern repositories use atomic insertions this should not
1109
# lead to excessive growth in the event of interrupted fetches.
1110
# 'knit' repositories may suffer excessive growth, but as a
1111
# deprecated format this is tolerable. It can be fixed if
1112
# needed by in the kndx index support raising on a duplicate
1113
# add with identical parents and options.
1114
access_memo = self._data.add_raw_records([len(bytes)], bytes)[0]
1115
index_entry = (record.key[0], options, access_memo, parents)
1117
if 'fulltext' not in options:
1118
basis_parent = parents[0]
1119
if not self.has_version(basis_parent):
1120
pending = buffered_index_entries.setdefault(
1122
pending.append(index_entry)
1125
self._index.add_versions([index_entry])
1126
elif record.storage_kind == 'fulltext':
1127
self.add_lines(record.key[0], parents,
1128
split_lines(record.get_bytes_as('fulltext')))
1130
adapter_key = record.storage_kind, 'fulltext'
1131
adapter = get_adapter(adapter_key)
1132
lines = split_lines(adapter.get_bytes(
1133
record, record.get_bytes_as(record.storage_kind)))
1135
self.add_lines(record.key[0], parents, lines)
1136
except errors.RevisionAlreadyPresent:
1138
# Add any records whose basis parent is now available.
1139
added_keys = [record.key[0]]
1141
key = added_keys.pop(0)
1142
if key in buffered_index_entries:
1143
index_entries = buffered_index_entries[key]
1144
self._index.add_versions(index_entries)
1146
[index_entry[0] for index_entry in index_entries])
1147
del buffered_index_entries[key]
1148
# If there were any deltas which had a missing basis parent, error.
1149
if buffered_index_entries:
1150
raise errors.RevisionNotPresent(buffered_index_entries.keys()[0],
1154
"""See VersionedFile.versions."""
1155
if 'evil' in debug.debug_flags:
1156
trace.mutter_callsite(2, "versions scales with size of history")
1157
return self._index.get_versions()
1159
def has_version(self, version_id):
1160
"""See VersionedFile.has_version."""
1161
if 'evil' in debug.debug_flags:
1162
trace.mutter_callsite(2, "has_version is a LBYL scenario")
1163
return self._index.has_version(version_id)
1165
__contains__ = has_version
1167
def _merge_annotations(self, content, parents, parent_texts={},
1168
delta=None, annotated=None,
1169
left_matching_blocks=None):
1170
"""Merge annotations for content. This is done by comparing
1171
the annotations based on changed to the text.
1173
if left_matching_blocks is not None:
1174
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1178
for parent_id in parents:
1179
merge_content = self._get_content(parent_id, parent_texts)
1180
if (parent_id == parents[0] and delta_seq is not None):
1183
seq = patiencediff.PatienceSequenceMatcher(
1184
None, merge_content.text(), content.text())
1185
for i, j, n in seq.get_matching_blocks():
1188
# this copies (origin, text) pairs across to the new
1189
# content for any line that matches the last-checked
1191
content._lines[j:j+n] = merge_content._lines[i:i+n]
1193
if delta_seq is None:
1194
reference_content = self._get_content(parents[0], parent_texts)
1195
new_texts = content.text()
1196
old_texts = reference_content.text()
1197
delta_seq = patiencediff.PatienceSequenceMatcher(
1198
None, old_texts, new_texts)
1199
return self._make_line_delta(delta_seq, content)
1201
def _make_line_delta(self, delta_seq, new_content):
1202
"""Generate a line delta from delta_seq and new_content."""
1204
for op in delta_seq.get_opcodes():
1205
if op[0] == 'equal':
1207
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1210
def _get_components_positions(self, version_ids):
1211
"""Produce a map of position data for the components of versions.
1213
This data is intended to be used for retrieving the knit records.
1215
A dict of version_id to (record_details, index_memo, next, parents) is
1217
method is the way referenced data should be applied.
1218
index_memo is the handle to pass to the data access to actually get the
1220
next is the build-parent of the version, or None for fulltexts.
1221
parents is the version_ids of the parents of this version
1224
pending_components = version_ids
1225
while pending_components:
1226
build_details = self._index.get_build_details(pending_components)
1227
current_components = set(pending_components)
1228
pending_components = set()
1229
for version_id, details in build_details.iteritems():
1230
(index_memo, compression_parent, parents,
1231
record_details) = details
1232
method = record_details[0]
1233
if compression_parent is not None:
1234
pending_components.add(compression_parent)
1235
component_data[version_id] = (record_details, index_memo,
1237
missing = current_components.difference(build_details)
1239
raise errors.RevisionNotPresent(missing.pop(), self.filename)
1240
return component_data
1242
def _get_content(self, version_id, parent_texts={}):
1243
"""Returns a content object that makes up the specified
1245
cached_version = parent_texts.get(version_id, None)
1246
if cached_version is not None:
1247
if not self.has_version(version_id):
1248
raise RevisionNotPresent(version_id, self.filename)
1249
return cached_version
1251
text_map, contents_map = self._get_content_maps([version_id])
1252
return contents_map[version_id]
1254
def _check_versions_present(self, version_ids):
1255
"""Check that all specified versions are present."""
1256
self._index.check_versions_present(version_ids)
1258
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
1259
nostore_sha, random_id, check_content, left_matching_blocks):
1260
"""See VersionedFile.add_lines_with_ghosts()."""
1261
self._check_add(version_id, lines, random_id, check_content)
1262
return self._add(version_id, lines, parents, self.delta,
1263
parent_texts, left_matching_blocks, nostore_sha, random_id)
1265
def _add_lines(self, version_id, parents, lines, parent_texts,
1266
left_matching_blocks, nostore_sha, random_id, check_content):
1267
"""See VersionedFile.add_lines."""
1268
self._check_add(version_id, lines, random_id, check_content)
1269
self._check_versions_present(parents)
1270
return self._add(version_id, lines[:], parents, self.delta,
1271
parent_texts, left_matching_blocks, nostore_sha, random_id)
1273
def _check_add(self, version_id, lines, random_id, check_content):
1274
"""check that version_id and lines are safe to add."""
1275
if contains_whitespace(version_id):
1276
raise InvalidRevisionId(version_id, self.filename)
1277
self.check_not_reserved_id(version_id)
1278
# Technically this could be avoided if we are happy to allow duplicate
1279
# id insertion when other things than bzr core insert texts, but it
1280
# seems useful for folk using the knit api directly to have some safety
1281
# blanket that we can disable.
1282
if not random_id and self.has_version(version_id):
1283
raise RevisionAlreadyPresent(version_id, self.filename)
1285
self._check_lines_not_unicode(lines)
1286
self._check_lines_are_lines(lines)
1288
def _add(self, version_id, lines, parents, delta, parent_texts,
1289
left_matching_blocks, nostore_sha, random_id):
1290
"""Add a set of lines on top of version specified by parents.
1292
If delta is true, compress the text as a line-delta against
1295
Any versions not present will be converted into ghosts.
1297
# first thing, if the content is something we don't need to store, find
1299
line_bytes = ''.join(lines)
1300
digest = sha_string(line_bytes)
1301
if nostore_sha == digest:
1302
raise errors.ExistingContent
1304
present_parents = []
1305
if parent_texts is None:
1307
for parent in parents:
1308
if self.has_version(parent):
1309
present_parents.append(parent)
1311
# can only compress against the left most present parent.
1313
(len(present_parents) == 0 or
1314
present_parents[0] != parents[0])):
1317
text_length = len(line_bytes)
1320
if lines[-1][-1] != '\n':
1321
# copy the contents of lines.
1323
options.append('no-eol')
1324
lines[-1] = lines[-1] + '\n'
1328
# To speed the extract of texts the delta chain is limited
1329
# to a fixed number of deltas. This should minimize both
1330
# I/O and the time spend applying deltas.
1331
delta = self._check_should_delta(present_parents)
1333
content = self.factory.make(lines, version_id)
1334
if 'no-eol' in options:
1335
# Hint to the content object that its text() call should strip the
1337
content._should_strip_eol = True
1338
if delta or (self.factory.annotated and len(present_parents) > 0):
1339
# Merge annotations from parent texts if needed.
1340
delta_hunks = self._merge_annotations(content, present_parents,
1341
parent_texts, delta, self.factory.annotated,
1342
left_matching_blocks)
1345
options.append('line-delta')
1346
store_lines = self.factory.lower_line_delta(delta_hunks)
1347
size, bytes = self._data._record_to_data(version_id, digest,
1350
options.append('fulltext')
1351
# isinstance is slower and we have no hierarchy.
1352
if self.factory.__class__ == KnitPlainFactory:
1353
# Use the already joined bytes saving iteration time in
1355
size, bytes = self._data._record_to_data(version_id, digest,
1356
lines, [line_bytes])
1358
# get mixed annotation + content and feed it into the
1360
store_lines = self.factory.lower_fulltext(content)
1361
size, bytes = self._data._record_to_data(version_id, digest,
1364
access_memo = self._data.add_raw_records([size], bytes)[0]
1365
self._index.add_versions(
1366
((version_id, options, access_memo, parents),),
1367
random_id=random_id)
1368
return digest, text_length, content
1370
def check(self, progress_bar=None):
1371
"""See VersionedFile.check()."""
1372
# This doesn't actually test extraction of everything, but that will
1373
# impact 'bzr check' substantially, and needs to be integrated with
1374
# care. However, it does check for the obvious problem of a delta with
1376
versions = self.versions()
1377
parent_map = self.get_parent_map(versions)
1378
for version in versions:
1379
if self._index.get_method(version) != 'fulltext':
1380
compression_parent = parent_map[version][0]
1381
if compression_parent not in parent_map:
1382
raise errors.KnitCorrupt(self,
1383
"Missing basis parent %s for %s" % (
1384
compression_parent, version))
1386
def get_lines(self, version_id):
1387
"""See VersionedFile.get_lines()."""
1388
return self.get_line_list([version_id])[0]
1390
def _get_record_map(self, version_ids):
1391
"""Produce a dictionary of knit records.
1393
:return: {version_id:(record, record_details, digest, next)}
1395
data returned from read_records
1397
opaque information to pass to parse_record
1399
SHA1 digest of the full text after all steps are done
1401
build-parent of the version, i.e. the leftmost ancestor.
1402
Will be None if the record is not a delta.
1404
position_map = self._get_components_positions(version_ids)
1405
# c = component_id, r = record_details, i_m = index_memo, n = next
1406
records = [(c, i_m) for c, (r, i_m, n)
1407
in position_map.iteritems()]
1409
for component_id, record, digest in \
1410
self._data.read_records_iter(records):
1411
(record_details, index_memo, next) = position_map[component_id]
1412
record_map[component_id] = record, record_details, digest, next
1416
def get_text(self, version_id):
1417
"""See VersionedFile.get_text"""
1418
return self.get_texts([version_id])[0]
1420
def get_texts(self, version_ids):
1421
return [''.join(l) for l in self.get_line_list(version_ids)]
1423
def get_line_list(self, version_ids):
1424
"""Return the texts of listed versions as a list of strings."""
1425
for version_id in version_ids:
1426
self.check_not_reserved_id(version_id)
1427
text_map, content_map = self._get_content_maps(version_ids)
1428
return [text_map[v] for v in version_ids]
1430
_get_lf_split_line_list = get_line_list
1432
def _get_content_maps(self, version_ids):
1433
"""Produce maps of text and KnitContents
1435
:return: (text_map, content_map) where text_map contains the texts for
1436
the requested versions and content_map contains the KnitContents.
1437
Both dicts take version_ids as their keys.
1439
# FUTURE: This function could be improved for the 'extract many' case
1440
# by tracking each component and only doing the copy when the number of
1441
# children than need to apply delta's to it is > 1 or it is part of the
1443
version_ids = list(version_ids)
1444
multiple_versions = len(version_ids) != 1
1445
record_map = self._get_record_map(version_ids)
1450
for version_id in version_ids:
1453
while cursor is not None:
1454
record, record_details, digest, next = record_map[cursor]
1455
components.append((cursor, record, record_details, digest))
1456
if cursor in content_map:
1461
for (component_id, record, record_details,
1462
digest) in reversed(components):
1463
if component_id in content_map:
1464
content = content_map[component_id]
1466
content, delta = self.factory.parse_record(version_id,
1467
record, record_details, content,
1468
copy_base_content=multiple_versions)
1469
if multiple_versions:
1470
content_map[component_id] = content
1472
final_content[version_id] = content
1474
# digest here is the digest from the last applied component.
1475
text = content.text()
1476
actual_sha = sha_strings(text)
1477
if actual_sha != digest:
1478
raise KnitCorrupt(self.filename,
1480
'\n of reconstructed text does not match'
1482
'\n for version %s' %
1483
(actual_sha, digest, version_id))
1484
text_map[version_id] = text
1485
return text_map, final_content
1487
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1489
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1490
if version_ids is None:
1491
version_ids = self.versions()
1493
pb = progress.DummyProgress()
1494
# we don't care about inclusions, the caller cares.
1495
# but we need to setup a list of records to visit.
1496
# we need version_id, position, length
1497
version_id_records = []
1498
requested_versions = set(version_ids)
1499
# filter for available versions
1500
for version_id in requested_versions:
1501
if not self.has_version(version_id):
1502
raise RevisionNotPresent(version_id, self.filename)
1503
# get a in-component-order queue:
1504
for version_id in self.versions():
1505
if version_id in requested_versions:
1506
index_memo = self._index.get_position(version_id)
1507
version_id_records.append((version_id, index_memo))
1509
total = len(version_id_records)
1510
for version_idx, (version_id, data, sha_value) in \
1511
enumerate(self._data.read_records_iter(version_id_records)):
1512
pb.update('Walking content.', version_idx, total)
1513
method = self._index.get_method(version_id)
1514
if method == 'fulltext':
1515
line_iterator = self.factory.get_fulltext_content(data)
1516
elif method == 'line-delta':
1517
line_iterator = self.factory.get_linedelta_content(data)
1519
raise ValueError('invalid method %r' % (method,))
1520
# XXX: It might be more efficient to yield (version_id,
1521
# line_iterator) in the future. However for now, this is a simpler
1522
# change to integrate into the rest of the codebase. RBC 20071110
1523
for line in line_iterator:
1524
yield line, version_id
1526
pb.update('Walking content.', total, total)
1528
def num_versions(self):
1529
"""See VersionedFile.num_versions()."""
1530
return self._index.num_versions()
1532
__len__ = num_versions
1534
def annotate(self, version_id):
1535
"""See VersionedFile.annotate."""
1536
return self.factory.annotate(self, version_id)
1538
def get_parent_map(self, version_ids):
1539
"""See VersionedFile.get_parent_map."""
1540
return self._index.get_parent_map(version_ids)
1542
def get_ancestry(self, versions, topo_sorted=True):
1543
"""See VersionedFile.get_ancestry."""
1544
if isinstance(versions, basestring):
1545
versions = [versions]
1548
return self._index.get_ancestry(versions, topo_sorted)
1550
def get_ancestry_with_ghosts(self, versions):
1551
"""See VersionedFile.get_ancestry_with_ghosts."""
1552
if isinstance(versions, basestring):
1553
versions = [versions]
1556
return self._index.get_ancestry_with_ghosts(versions)
1558
def plan_merge(self, ver_a, ver_b):
1559
"""See VersionedFile.plan_merge."""
1560
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1561
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1562
annotated_a = self.annotate(ver_a)
1563
annotated_b = self.annotate(ver_b)
1564
return merge._plan_annotate_merge(annotated_a, annotated_b,
1565
ancestors_a, ancestors_b)
1568
class _KnitComponentFile(object):
1569
"""One of the files used to implement a knit database"""
1571
def __init__(self, transport, filename, mode, file_mode=None,
1572
create_parent_dir=False, dir_mode=None):
1573
self._transport = transport
1574
self._filename = filename
1576
self._file_mode = file_mode
1577
self._dir_mode = dir_mode
1578
self._create_parent_dir = create_parent_dir
1579
self._need_to_create = False
1581
def _full_path(self):
1582
"""Return the full path to this file."""
1583
return self._transport.base + self._filename
1585
def check_header(self, fp):
1586
line = fp.readline()
1588
# An empty file can actually be treated as though the file doesn't
1590
raise errors.NoSuchFile(self._full_path())
1591
if line != self.HEADER:
1592
raise KnitHeaderError(badline=line,
1593
filename=self._transport.abspath(self._filename))
1596
return '%s(%s)' % (self.__class__.__name__, self._filename)
1599
class _KnitIndex(_KnitComponentFile):
1600
"""Manages knit index file.
1602
The index is already kept in memory and read on startup, to enable
1603
fast lookups of revision information. The cursor of the index
1604
file is always pointing to the end, making it easy to append
1607
_cache is a cache for fast mapping from version id to a Index
1610
_history is a cache for fast mapping from indexes to version ids.
1612
The index data format is dictionary compressed when it comes to
1613
parent references; a index entry may only have parents that with a
1614
lover index number. As a result, the index is topological sorted.
1616
Duplicate entries may be written to the index for a single version id
1617
if this is done then the latter one completely replaces the former:
1618
this allows updates to correct version and parent information.
1619
Note that the two entries may share the delta, and that successive
1620
annotations and references MUST point to the first entry.
1622
The index file on disc contains a header, followed by one line per knit
1623
record. The same revision can be present in an index file more than once.
1624
The first occurrence gets assigned a sequence number starting from 0.
1626
The format of a single line is
1627
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1628
REVISION_ID is a utf8-encoded revision id
1629
FLAGS is a comma separated list of flags about the record. Values include
1630
no-eol, line-delta, fulltext.
1631
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1632
that the the compressed data starts at.
1633
LENGTH is the ascii representation of the length of the data file.
1634
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1636
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1637
revision id already in the knit that is a parent of REVISION_ID.
1638
The ' :' marker is the end of record marker.
1641
when a write is interrupted to the index file, it will result in a line
1642
that does not end in ' :'. If the ' :' is not present at the end of a line,
1643
or at the end of the file, then the record that is missing it will be
1644
ignored by the parser.
1646
When writing new records to the index file, the data is preceded by '\n'
1647
to ensure that records always start on new lines even if the last write was
1648
interrupted. As a result its normal for the last line in the index to be
1649
missing a trailing newline. One can be added with no harmful effects.
1652
HEADER = "# bzr knit index 8\n"
1654
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1655
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1657
def _cache_version(self, version_id, options, pos, size, parents):
1658
"""Cache a version record in the history array and index cache.
1660
This is inlined into _load_data for performance. KEEP IN SYNC.
1661
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1664
# only want the _history index to reference the 1st index entry
1666
if version_id not in self._cache:
1667
index = len(self._history)
1668
self._history.append(version_id)
1670
index = self._cache[version_id][5]
1671
self._cache[version_id] = (version_id,
1678
def _check_write_ok(self):
1679
if self._get_scope() != self._scope:
1680
raise errors.OutSideTransaction()
1681
if self._mode != 'w':
1682
raise errors.ReadOnlyObjectDirtiedError(self)
1684
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1685
create_parent_dir=False, delay_create=False, dir_mode=None,
1687
_KnitComponentFile.__init__(self, transport, filename, mode,
1688
file_mode=file_mode,
1689
create_parent_dir=create_parent_dir,
1692
# position in _history is the 'official' index for a revision
1693
# but the values may have come from a newer entry.
1694
# so - wc -l of a knit index is != the number of unique names
1698
fp = self._transport.get(self._filename)
1700
# _load_data may raise NoSuchFile if the target knit is
1702
_load_data(self, fp)
1706
if mode != 'w' or not create:
1709
self._need_to_create = True
1711
self._transport.put_bytes_non_atomic(
1712
self._filename, self.HEADER, mode=self._file_mode)
1713
self._scope = get_scope()
1714
self._get_scope = get_scope
1716
def get_ancestry(self, versions, topo_sorted=True):
1717
"""See VersionedFile.get_ancestry."""
1718
# get a graph of all the mentioned versions:
1720
pending = set(versions)
1723
version = pending.pop()
1726
parents = [p for p in cache[version][4] if p in cache]
1728
raise RevisionNotPresent(version, self._filename)
1729
# if not completed and not a ghost
1730
pending.update([p for p in parents if p not in graph])
1731
graph[version] = parents
1734
return topo_sort(graph.items())
1736
def get_ancestry_with_ghosts(self, versions):
1737
"""See VersionedFile.get_ancestry_with_ghosts."""
1738
# get a graph of all the mentioned versions:
1739
self.check_versions_present(versions)
1742
pending = set(versions)
1744
version = pending.pop()
1746
parents = cache[version][4]
1752
pending.update([p for p in parents if p not in graph])
1753
graph[version] = parents
1754
return topo_sort(graph.items())
1756
def get_build_details(self, version_ids):
1757
"""Get the method, index_memo and compression parent for version_ids.
1759
Ghosts are omitted from the result.
1761
:param version_ids: An iterable of version_ids.
1762
:return: A dict of version_id:(index_memo, compression_parent,
1763
parents, record_details).
1765
opaque structure to pass to read_records to extract the raw
1768
Content that this record is built upon, may be None
1770
Logical parents of this node
1772
extra information about the content which needs to be passed to
1773
Factory.parse_record
1776
for version_id in version_ids:
1777
if version_id not in self._cache:
1778
# ghosts are omitted
1780
method = self.get_method(version_id)
1781
parents = self.get_parents_with_ghosts(version_id)
1782
if method == 'fulltext':
1783
compression_parent = None
1785
compression_parent = parents[0]
1786
noeol = 'no-eol' in self.get_options(version_id)
1787
index_memo = self.get_position(version_id)
1788
result[version_id] = (index_memo, compression_parent,
1789
parents, (method, noeol))
1792
def num_versions(self):
1793
return len(self._history)
1795
__len__ = num_versions
1797
def get_versions(self):
1798
"""Get all the versions in the file. not topologically sorted."""
1799
return self._history
1801
def _version_list_to_index(self, versions):
1804
for version in versions:
1805
if version in cache:
1806
# -- inlined lookup() --
1807
result_list.append(str(cache[version][5]))
1808
# -- end lookup () --
1810
result_list.append('.' + version)
1811
return ' '.join(result_list)
1813
def add_version(self, version_id, options, index_memo, parents):
1814
"""Add a version record to the index."""
1815
self.add_versions(((version_id, options, index_memo, parents),))
1817
def add_versions(self, versions, random_id=False):
1818
"""Add multiple versions to the index.
1820
:param versions: a list of tuples:
1821
(version_id, options, pos, size, parents).
1822
:param random_id: If True the ids being added were randomly generated
1823
and no check for existence will be performed.
1826
orig_history = self._history[:]
1827
orig_cache = self._cache.copy()
1830
for version_id, options, (index, pos, size), parents in versions:
1831
line = "\n%s %s %s %s %s :" % (version_id,
1835
self._version_list_to_index(parents))
1837
self._cache_version(version_id, options, pos, size, tuple(parents))
1838
if not self._need_to_create:
1839
self._transport.append_bytes(self._filename, ''.join(lines))
1842
sio.write(self.HEADER)
1843
sio.writelines(lines)
1845
self._transport.put_file_non_atomic(self._filename, sio,
1846
create_parent_dir=self._create_parent_dir,
1847
mode=self._file_mode,
1848
dir_mode=self._dir_mode)
1849
self._need_to_create = False
1851
# If any problems happen, restore the original values and re-raise
1852
self._history = orig_history
1853
self._cache = orig_cache
1856
def has_version(self, version_id):
1857
"""True if the version is in the index."""
1858
return version_id in self._cache
1860
def get_position(self, version_id):
1861
"""Return details needed to access the version.
1863
.kndx indices do not support split-out data, so return None for the
1866
:return: a tuple (None, data position, size) to hand to the access
1867
logic to get the record.
1869
entry = self._cache[version_id]
1870
return None, entry[2], entry[3]
1872
def get_method(self, version_id):
1873
"""Return compression method of specified version."""
1875
options = self._cache[version_id][1]
1877
raise RevisionNotPresent(version_id, self._filename)
1878
if 'fulltext' in options:
1881
if 'line-delta' not in options:
1882
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1885
def get_options(self, version_id):
1886
"""Return a list representing options.
1890
return self._cache[version_id][1]
1892
def get_parent_map(self, version_ids):
1893
"""Passed through to by KnitVersionedFile.get_parent_map."""
1895
for version_id in version_ids:
1897
result[version_id] = tuple(self._cache[version_id][4])
1902
def get_parents_with_ghosts(self, version_id):
1903
"""Return parents of specified version with ghosts."""
1905
return self.get_parent_map([version_id])[version_id]
1907
raise RevisionNotPresent(version_id, self)
1909
def check_versions_present(self, version_ids):
1910
"""Check that all specified versions are present."""
1912
for version_id in version_ids:
1913
if version_id not in cache:
1914
raise RevisionNotPresent(version_id, self._filename)
1917
class KnitGraphIndex(object):
1918
"""A knit index that builds on GraphIndex."""
1920
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1921
"""Construct a KnitGraphIndex on a graph_index.
1923
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1924
:param deltas: Allow delta-compressed records.
1925
:param add_callback: If not None, allow additions to the index and call
1926
this callback with a list of added GraphIndex nodes:
1927
[(node, value, node_refs), ...]
1928
:param parents: If True, record knits parents, if not do not record
1931
self._graph_index = graph_index
1932
self._deltas = deltas
1933
self._add_callback = add_callback
1934
self._parents = parents
1935
if deltas and not parents:
1936
raise KnitCorrupt(self, "Cannot do delta compression without "
1939
def _check_write_ok(self):
1942
def _get_entries(self, keys, check_present=False):
1943
"""Get the entries for keys.
1945
:param keys: An iterable of index keys, - 1-tuples.
1950
for node in self._graph_index.iter_entries(keys):
1952
found_keys.add(node[1])
1954
# adapt parentless index to the rest of the code.
1955
for node in self._graph_index.iter_entries(keys):
1956
yield node[0], node[1], node[2], ()
1957
found_keys.add(node[1])
1959
missing_keys = keys.difference(found_keys)
1961
raise RevisionNotPresent(missing_keys.pop(), self)
1963
def _present_keys(self, version_ids):
1965
node[1] for node in self._get_entries(version_ids)])
1967
def _parentless_ancestry(self, versions):
1968
"""Honour the get_ancestry API for parentless knit indices."""
1969
wanted_keys = self._version_ids_to_keys(versions)
1970
present_keys = self._present_keys(wanted_keys)
1971
missing = set(wanted_keys).difference(present_keys)
1973
raise RevisionNotPresent(missing.pop(), self)
1974
return list(self._keys_to_version_ids(present_keys))
1976
def get_ancestry(self, versions, topo_sorted=True):
1977
"""See VersionedFile.get_ancestry."""
1978
if not self._parents:
1979
return self._parentless_ancestry(versions)
1980
# XXX: This will do len(history) index calls - perhaps
1981
# it should be altered to be a index core feature?
1982
# get a graph of all the mentioned versions:
1985
versions = self._version_ids_to_keys(versions)
1986
pending = set(versions)
1988
# get all pending nodes
1989
this_iteration = pending
1990
new_nodes = self._get_entries(this_iteration)
1993
for (index, key, value, node_refs) in new_nodes:
1994
# dont ask for ghosties - otherwise
1995
# we we can end up looping with pending
1996
# being entirely ghosted.
1997
graph[key] = [parent for parent in node_refs[0]
1998
if parent not in ghosts]
2000
for parent in graph[key]:
2001
# dont examine known nodes again
2006
ghosts.update(this_iteration.difference(found))
2007
if versions.difference(graph):
2008
raise RevisionNotPresent(versions.difference(graph).pop(), self)
2010
result_keys = topo_sort(graph.items())
2012
result_keys = graph.iterkeys()
2013
return [key[0] for key in result_keys]
2015
def get_ancestry_with_ghosts(self, versions):
2016
"""See VersionedFile.get_ancestry."""
2017
if not self._parents:
2018
return self._parentless_ancestry(versions)
2019
# XXX: This will do len(history) index calls - perhaps
2020
# it should be altered to be a index core feature?
2021
# get a graph of all the mentioned versions:
2023
versions = self._version_ids_to_keys(versions)
2024
pending = set(versions)
2026
# get all pending nodes
2027
this_iteration = pending
2028
new_nodes = self._get_entries(this_iteration)
2030
for (index, key, value, node_refs) in new_nodes:
2031
graph[key] = node_refs[0]
2033
for parent in graph[key]:
2034
# dont examine known nodes again
2038
missing_versions = this_iteration.difference(graph)
2039
missing_needed = versions.intersection(missing_versions)
2041
raise RevisionNotPresent(missing_needed.pop(), self)
2042
for missing_version in missing_versions:
2043
# add a key, no parents
2044
graph[missing_version] = []
2045
pending.discard(missing_version) # don't look for it
2046
result_keys = topo_sort(graph.items())
2047
return [key[0] for key in result_keys]
2049
def get_build_details(self, version_ids):
2050
"""Get the method, index_memo and compression parent for version_ids.
2052
Ghosts are omitted from the result.
2054
:param version_ids: An iterable of version_ids.
2055
:return: A dict of version_id:(index_memo, compression_parent,
2056
parents, record_details).
2058
opaque structure to pass to read_records to extract the raw
2061
Content that this record is built upon, may be None
2063
Logical parents of this node
2065
extra information about the content which needs to be passed to
2066
Factory.parse_record
2069
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
2070
for entry in entries:
2071
version_id = self._keys_to_version_ids((entry[1],))[0]
2072
if not self._parents:
2075
parents = self._keys_to_version_ids(entry[3][0])
2076
if not self._deltas:
2077
compression_parent = None
2079
compression_parent_key = self._compression_parent(entry)
2080
if compression_parent_key:
2081
compression_parent = self._keys_to_version_ids(
2082
(compression_parent_key,))[0]
2084
compression_parent = None
2085
noeol = (entry[2][0] == 'N')
2086
if compression_parent:
2087
method = 'line-delta'
2090
result[version_id] = (self._node_to_position(entry),
2091
compression_parent, parents,
2095
def _compression_parent(self, an_entry):
2096
# return the key that an_entry is compressed against, or None
2097
# Grab the second parent list (as deltas implies parents currently)
2098
compression_parents = an_entry[3][1]
2099
if not compression_parents:
2101
return compression_parents[0]
2103
def _get_method(self, node):
2104
if not self._deltas:
2106
if self._compression_parent(node):
2111
def num_versions(self):
2112
return len(list(self._graph_index.iter_all_entries()))
2114
__len__ = num_versions
2116
def get_versions(self):
2117
"""Get all the versions in the file. not topologically sorted."""
2118
return [node[1][0] for node in self._graph_index.iter_all_entries()]
2120
def has_version(self, version_id):
2121
"""True if the version is in the index."""
2122
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
2124
def _keys_to_version_ids(self, keys):
2125
return tuple(key[0] for key in keys)
2127
def get_position(self, version_id):
2128
"""Return details needed to access the version.
2130
:return: a tuple (index, data position, size) to hand to the access
2131
logic to get the record.
2133
node = self._get_node(version_id)
2134
return self._node_to_position(node)
2136
def _node_to_position(self, node):
2137
"""Convert an index value to position details."""
2138
bits = node[2][1:].split(' ')
2139
return node[0], int(bits[0]), int(bits[1])
2141
def get_method(self, version_id):
2142
"""Return compression method of specified version."""
2143
return self._get_method(self._get_node(version_id))
2145
def _get_node(self, version_id):
2147
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
2149
raise RevisionNotPresent(version_id, self)
2151
def get_options(self, version_id):
2152
"""Return a list representing options.
2156
node = self._get_node(version_id)
2157
options = [self._get_method(node)]
2158
if node[2][0] == 'N':
2159
options.append('no-eol')
2162
def get_parent_map(self, version_ids):
2163
"""Passed through to by KnitVersionedFile.get_parent_map."""
2164
nodes = self._get_entries(self._version_ids_to_keys(version_ids))
2168
result[node[1][0]] = self._keys_to_version_ids(node[3][0])
2171
result[node[1][0]] = ()
2174
def get_parents_with_ghosts(self, version_id):
2175
"""Return parents of specified version with ghosts."""
2177
return self.get_parent_map([version_id])[version_id]
2179
raise RevisionNotPresent(version_id, self)
2181
def check_versions_present(self, version_ids):
2182
"""Check that all specified versions are present."""
2183
keys = self._version_ids_to_keys(version_ids)
2184
present = self._present_keys(keys)
2185
missing = keys.difference(present)
2187
raise RevisionNotPresent(missing.pop(), self)
2189
def add_version(self, version_id, options, access_memo, parents):
2190
"""Add a version record to the index."""
2191
return self.add_versions(((version_id, options, access_memo, parents),))
2193
def add_versions(self, versions, random_id=False):
2194
"""Add multiple versions to the index.
2196
This function does not insert data into the Immutable GraphIndex
2197
backing the KnitGraphIndex, instead it prepares data for insertion by
2198
the caller and checks that it is safe to insert then calls
2199
self._add_callback with the prepared GraphIndex nodes.
2201
:param versions: a list of tuples:
2202
(version_id, options, pos, size, parents).
2203
:param random_id: If True the ids being added were randomly generated
2204
and no check for existence will be performed.
2206
if not self._add_callback:
2207
raise errors.ReadOnlyError(self)
2208
# we hope there are no repositories with inconsistent parentage
2213
for (version_id, options, access_memo, parents) in versions:
2214
index, pos, size = access_memo
2215
key = (version_id, )
2216
parents = tuple((parent, ) for parent in parents)
2217
if 'no-eol' in options:
2221
value += "%d %d" % (pos, size)
2222
if not self._deltas:
2223
if 'line-delta' in options:
2224
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2227
if 'line-delta' in options:
2228
node_refs = (parents, (parents[0],))
2230
node_refs = (parents, ())
2232
node_refs = (parents, )
2235
raise KnitCorrupt(self, "attempt to add node with parents "
2236
"in parentless index.")
2238
keys[key] = (value, node_refs)
2240
present_nodes = self._get_entries(keys)
2241
for (index, key, value, node_refs) in present_nodes:
2242
if (value, node_refs) != keys[key]:
2243
raise KnitCorrupt(self, "inconsistent details in add_versions"
2244
": %s %s" % ((value, node_refs), keys[key]))
2248
for key, (value, node_refs) in keys.iteritems():
2249
result.append((key, value, node_refs))
2251
for key, (value, node_refs) in keys.iteritems():
2252
result.append((key, value))
2253
self._add_callback(result)
2255
def _version_ids_to_keys(self, version_ids):
2256
return set((version_id, ) for version_id in version_ids)
2259
class _KnitAccess(object):
2260
"""Access to knit records in a .knit file."""
2262
def __init__(self, transport, filename, _file_mode, _dir_mode,
2263
_need_to_create, _create_parent_dir):
2264
"""Create a _KnitAccess for accessing and inserting data.
2266
:param transport: The transport the .knit is located on.
2267
:param filename: The filename of the .knit.
2269
self._transport = transport
2270
self._filename = filename
2271
self._file_mode = _file_mode
2272
self._dir_mode = _dir_mode
2273
self._need_to_create = _need_to_create
2274
self._create_parent_dir = _create_parent_dir
2277
return "%s(%r)" % (self.__class__.__name__,
2278
self._transport.abspath(self._filename))
2280
def add_raw_records(self, sizes, raw_data):
2281
"""Add raw knit bytes to a storage area.
2283
The data is spooled to whereever the access method is storing data.
2285
:param sizes: An iterable containing the size of each raw data segment.
2286
:param raw_data: A bytestring containing the data.
2287
:return: A list of memos to retrieve the record later. Each memo is a
2288
tuple - (index, pos, length), where the index field is always None
2289
for the .knit access method.
2291
if not self._need_to_create:
2292
base = self._transport.append_bytes(self._filename, raw_data)
2294
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2295
create_parent_dir=self._create_parent_dir,
2296
mode=self._file_mode,
2297
dir_mode=self._dir_mode)
2298
self._need_to_create = False
2302
result.append((None, base, size))
2307
"""IFF this data access has its own storage area, initialise it.
2311
self._transport.put_bytes_non_atomic(self._filename, '',
2312
mode=self._file_mode)
2314
def open_file(self):
2315
"""IFF this data access can be represented as a single file, open it.
2317
For knits that are not mapped to a single file on disk this will
2320
:return: None or a file handle.
2323
return self._transport.get(self._filename)
2328
def get_raw_records(self, memos_for_retrieval):
2329
"""Get the raw bytes for a records.
2331
:param memos_for_retrieval: An iterable containing the (index, pos,
2332
length) memo for retrieving the bytes. The .knit method ignores
2333
the index as there is always only a single file.
2334
:return: An iterator over the bytes of the records.
2336
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2337
for pos, data in self._transport.readv(self._filename, read_vector):
2341
class _PackAccess(object):
2342
"""Access to knit records via a collection of packs."""
2344
def __init__(self, index_to_packs, writer=None):
2345
"""Create a _PackAccess object.
2347
:param index_to_packs: A dict mapping index objects to the transport
2348
and file names for obtaining data.
2349
:param writer: A tuple (pack.ContainerWriter, write_index) which
2350
contains the pack to write, and the index that reads from it will
2354
self.container_writer = writer[0]
2355
self.write_index = writer[1]
2357
self.container_writer = None
2358
self.write_index = None
2359
self.indices = index_to_packs
2361
def add_raw_records(self, sizes, raw_data):
2362
"""Add raw knit bytes to a storage area.
2364
The data is spooled to the container writer in one bytes-record per
2367
:param sizes: An iterable containing the size of each raw data segment.
2368
:param raw_data: A bytestring containing the data.
2369
:return: A list of memos to retrieve the record later. Each memo is a
2370
tuple - (index, pos, length), where the index field is the
2371
write_index object supplied to the PackAccess object.
2376
p_offset, p_length = self.container_writer.add_bytes_record(
2377
raw_data[offset:offset+size], [])
2379
result.append((self.write_index, p_offset, p_length))
2383
"""Pack based knits do not get individually created."""
2385
def get_raw_records(self, memos_for_retrieval):
2386
"""Get the raw bytes for a records.
2388
:param memos_for_retrieval: An iterable containing the (index, pos,
2389
length) memo for retrieving the bytes. The Pack access method
2390
looks up the pack to use for a given record in its index_to_pack
2392
:return: An iterator over the bytes of the records.
2394
# first pass, group into same-index requests
2396
current_index = None
2397
for (index, offset, length) in memos_for_retrieval:
2398
if current_index == index:
2399
current_list.append((offset, length))
2401
if current_index is not None:
2402
request_lists.append((current_index, current_list))
2403
current_index = index
2404
current_list = [(offset, length)]
2405
# handle the last entry
2406
if current_index is not None:
2407
request_lists.append((current_index, current_list))
2408
for index, offsets in request_lists:
2409
transport, path = self.indices[index]
2410
reader = pack.make_readv_reader(transport, path, offsets)
2411
for names, read_func in reader.iter_records():
2412
yield read_func(None)
2414
def open_file(self):
2415
"""Pack based knits have no single file."""
2418
def set_writer(self, writer, index, (transport, packname)):
2419
"""Set a writer to use for adding data."""
2420
if index is not None:
2421
self.indices[index] = (transport, packname)
2422
self.container_writer = writer
2423
self.write_index = index
2426
class _StreamAccess(object):
2427
"""A Knit Access object that provides data from a datastream.
2429
It also provides a fallback to present as unannotated data, annotated data
2430
from a *backing* access object.
2432
This is triggered by a index_memo which is pointing to a different index
2433
than this was constructed with, and is used to allow extracting full
2434
unannotated texts for insertion into annotated knits.
2437
def __init__(self, reader_callable, stream_index, backing_knit,
2439
"""Create a _StreamAccess object.
2441
:param reader_callable: The reader_callable from the datastream.
2442
This is called to buffer all the data immediately, for
2444
:param stream_index: The index the data stream this provides access to
2445
which will be present in native index_memo's.
2446
:param backing_knit: The knit object that will provide access to
2447
annotated texts which are not available in the stream, so as to
2448
create unannotated texts.
2449
:param orig_factory: The original content factory used to generate the
2450
stream. This is used for checking whether the thunk code for
2451
supporting _copy_texts will generate the correct form of data.
2453
self.data = reader_callable(None)
2454
self.stream_index = stream_index
2455
self.backing_knit = backing_knit
2456
self.orig_factory = orig_factory
2458
def get_raw_records(self, memos_for_retrieval):
2459
"""Get the raw bytes for a records.
2461
:param memos_for_retrieval: An iterable of memos from the
2462
_StreamIndex object identifying bytes to read; for these classes
2463
they are (from_backing_knit, index, start, end) and can point to
2464
either the backing knit or streamed data.
2465
:return: An iterator yielding a byte string for each record in
2466
memos_for_retrieval.
2468
# use a generator for memory friendliness
2469
for from_backing_knit, version_id, start, end in memos_for_retrieval:
2470
if not from_backing_knit:
2471
if version_id is not self.stream_index:
2472
raise AssertionError()
2473
yield self.data[start:end]
2475
# we have been asked to thunk. This thunking only occurs when
2476
# we are obtaining plain texts from an annotated backing knit
2477
# so that _copy_texts will work.
2478
# We could improve performance here by scanning for where we need
2479
# to do this and using get_line_list, then interleaving the output
2480
# as desired. However, for now, this is sufficient.
2481
if self.orig_factory.__class__ != KnitPlainFactory:
2482
raise errors.KnitCorrupt(
2483
self, 'Bad thunk request %r cannot be backed by %r' %
2484
(version_id, self.orig_factory))
2485
lines = self.backing_knit.get_lines(version_id)
2486
line_bytes = ''.join(lines)
2487
digest = sha_string(line_bytes)
2488
# the packed form of the fulltext always has a trailing newline,
2489
# even if the actual text does not, unless the file is empty. the
2490
# record options including the noeol flag are passed through by
2491
# _StreamIndex, so this is safe.
2493
if lines[-1][-1] != '\n':
2494
lines[-1] = lines[-1] + '\n'
2496
# We want plain data, because we expect to thunk only to allow text
2498
size, bytes = self.backing_knit._data._record_to_data(version_id,
2499
digest, lines, line_bytes)
2503
class _StreamIndex(object):
2504
"""A Knit Index object that uses the data map from a datastream."""
2506
def __init__(self, data_list, backing_index):
2507
"""Create a _StreamIndex object.
2509
:param data_list: The data_list from the datastream.
2510
:param backing_index: The index which will supply values for nodes
2511
referenced outside of this stream.
2513
self.data_list = data_list
2514
self.backing_index = backing_index
2515
self._by_version = {}
2517
for key, options, length, parents in data_list:
2518
self._by_version[key] = options, (pos, pos + length), parents
2521
def get_ancestry(self, versions, topo_sorted):
2522
"""Get an ancestry list for versions."""
2524
# Not needed for basic joins
2525
raise NotImplementedError(self.get_ancestry)
2526
# get a graph of all the mentioned versions:
2527
# Little ugly - basically copied from KnitIndex, but don't want to
2528
# accidentally incorporate too much of that index's code.
2530
pending = set(versions)
2531
cache = self._by_version
2533
version = pending.pop()
2536
parents = [p for p in cache[version][2] if p in cache]
2538
raise RevisionNotPresent(version, self)
2539
# if not completed and not a ghost
2540
pending.update([p for p in parents if p not in ancestry])
2541
ancestry.add(version)
2542
return list(ancestry)
2544
def get_build_details(self, version_ids):
2545
"""Get the method, index_memo and compression parent for version_ids.
2547
Ghosts are omitted from the result.
2549
:param version_ids: An iterable of version_ids.
2550
:return: A dict of version_id:(index_memo, compression_parent,
2551
parents, record_details).
2553
opaque memo that can be passed to _StreamAccess.read_records
2554
to extract the raw data; for these classes it is
2555
(from_backing_knit, index, start, end)
2557
Content that this record is built upon, may be None
2559
Logical parents of this node
2561
extra information about the content which needs to be passed to
2562
Factory.parse_record
2565
for version_id in version_ids:
2567
method = self.get_method(version_id)
2568
except errors.RevisionNotPresent:
2569
# ghosts are omitted
2571
parent_ids = self.get_parents_with_ghosts(version_id)
2572
noeol = ('no-eol' in self.get_options(version_id))
2573
index_memo = self.get_position(version_id)
2574
from_backing_knit = index_memo[0]
2575
if from_backing_knit:
2576
# texts retrieved from the backing knit are always full texts
2578
if method == 'fulltext':
2579
compression_parent = None
2581
compression_parent = parent_ids[0]
2582
result[version_id] = (index_memo, compression_parent,
2583
parent_ids, (method, noeol))
2586
def get_method(self, version_id):
2587
"""Return compression method of specified version."""
2588
options = self.get_options(version_id)
2589
if 'fulltext' in options:
2591
elif 'line-delta' in options:
2594
raise errors.KnitIndexUnknownMethod(self, options)
2596
def get_options(self, version_id):
2597
"""Return a list representing options.
2602
return self._by_version[version_id][0]
2604
options = list(self.backing_index.get_options(version_id))
2605
if 'fulltext' in options:
2607
elif 'line-delta' in options:
2608
# Texts from the backing knit are always returned from the stream
2610
options.remove('line-delta')
2611
options.append('fulltext')
2613
raise errors.KnitIndexUnknownMethod(self, options)
2614
return tuple(options)
2616
def get_parent_map(self, version_ids):
2617
"""Passed through to by KnitVersionedFile.get_parent_map."""
2620
for version_id in version_ids:
2622
result[version_id] = self._by_version[version_id][2]
2624
pending_ids.add(version_id)
2625
result.update(self.backing_index.get_parent_map(pending_ids))
2628
def get_parents_with_ghosts(self, version_id):
2629
"""Return parents of specified version with ghosts."""
2631
return self.get_parent_map([version_id])[version_id]
2633
raise RevisionNotPresent(version_id, self)
2635
def get_position(self, version_id):
2636
"""Return details needed to access the version.
2638
_StreamAccess has the data as a big array, so we return slice
2639
coordinates into that (as index_memo's are opaque outside the
2640
index and matching access class).
2642
:return: a tuple (from_backing_knit, index, start, end) that can
2643
be passed e.g. to get_raw_records.
2644
If from_backing_knit is False, index will be self, otherwise it
2645
will be a version id.
2648
start, end = self._by_version[version_id][1]
2649
return False, self, start, end
2651
# Signal to the access object to handle this from the backing knit.
2652
return (True, version_id, None, None)
2654
def get_versions(self):
2655
"""Get all the versions in the stream."""
2656
return self._by_version.keys()
2659
class _KnitData(object):
2660
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2662
The KnitData class provides the logic for parsing and using knit records,
2663
making use of an access method for the low level read and write operations.
2666
def __init__(self, access):
2667
"""Create a KnitData object.
2669
:param access: The access method to use. Access methods such as
2670
_KnitAccess manage the insertion of raw records and the subsequent
2671
retrieval of the same.
2673
self._access = access
2674
self._checked = False
2676
def _open_file(self):
2677
return self._access.open_file()
2679
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2680
"""Convert version_id, digest, lines into a raw data block.
2682
:param dense_lines: The bytes of lines but in a denser form. For
2683
instance, if lines is a list of 1000 bytestrings each ending in \n,
2684
dense_lines may be a list with one line in it, containing all the
2685
1000's lines and their \n's. Using dense_lines if it is already
2686
known is a win because the string join to create bytes in this
2687
function spends less time resizing the final string.
2688
:return: (len, a StringIO instance with the raw data ready to read.)
2690
# Note: using a string copy here increases memory pressure with e.g.
2691
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2692
# when doing the initial commit of a mozilla tree. RBC 20070921
2693
bytes = ''.join(chain(
2694
["version %s %d %s\n" % (version_id,
2697
dense_lines or lines,
2698
["end %s\n" % version_id]))
2699
if lines and lines[-1][-1] != '\n':
2700
raise ValueError('corrupt lines value %r' % lines)
2701
compressed_bytes = bytes_to_gzip(bytes)
2702
return len(compressed_bytes), compressed_bytes
2704
def add_raw_records(self, sizes, raw_data):
2705
"""Append a prepared record to the data file.
2707
:param sizes: An iterable containing the size of each raw data segment.
2708
:param raw_data: A bytestring containing the data.
2709
:return: a list of index data for the way the data was stored.
2710
See the access method add_raw_records documentation for more
2713
return self._access.add_raw_records(sizes, raw_data)
2715
def _parse_record_header(self, version_id, raw_data):
2716
"""Parse a record header for consistency.
2718
:return: the header and the decompressor stream.
2719
as (stream, header_record)
2721
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2723
rec = self._check_header(version_id, df.readline())
2724
except Exception, e:
2725
raise KnitCorrupt(self._access,
2726
"While reading {%s} got %s(%s)"
2727
% (version_id, e.__class__.__name__, str(e)))
2730
def _split_header(self, line):
2733
raise KnitCorrupt(self._access,
2734
'unexpected number of elements in record header')
2737
def _check_header_version(self, rec, version_id):
2738
if rec[1] != version_id:
2739
raise KnitCorrupt(self._access,
2740
'unexpected version, wanted %r, got %r'
2741
% (version_id, rec[1]))
2743
def _check_header(self, version_id, line):
2744
rec = self._split_header(line)
2745
self._check_header_version(rec, version_id)
2748
def _parse_record_unchecked(self, data):
2750
# 4168 calls in 2880 217 internal
2751
# 4168 calls to _parse_record_header in 2121
2752
# 4168 calls to readlines in 330
2753
df = GzipFile(mode='rb', fileobj=StringIO(data))
2755
record_contents = df.readlines()
2756
except Exception, e:
2757
raise KnitCorrupt(self._access, "Corrupt compressed record %r, got %s(%s)" %
2758
(data, e.__class__.__name__, str(e)))
2759
header = record_contents.pop(0)
2760
rec = self._split_header(header)
2761
last_line = record_contents.pop()
2762
if len(record_contents) != int(rec[2]):
2763
raise KnitCorrupt(self._access,
2764
'incorrect number of lines %s != %s'
2766
% (len(record_contents), int(rec[2]),
2768
if last_line != 'end %s\n' % rec[1]:
2769
raise KnitCorrupt(self._access,
2770
'unexpected version end line %r, wanted %r'
2771
% (last_line, rec[1]))
2773
return rec, record_contents
2775
def _parse_record(self, version_id, data):
2776
rec, record_contents = self._parse_record_unchecked(data)
2777
self._check_header_version(rec, version_id)
2778
return record_contents, rec[3]
2780
def read_records_iter_raw(self, records):
2781
"""Read text records from data file and yield raw data.
2783
This unpacks enough of the text record to validate the id is
2784
as expected but thats all.
2786
Each item the iterator yields is (version_id, bytes,
2789
# setup an iterator of the external records:
2790
# uses readv so nice and fast we hope.
2792
# grab the disk data needed.
2793
needed_offsets = [index_memo for version_id, index_memo
2795
raw_records = self._access.get_raw_records(needed_offsets)
2797
for version_id, index_memo in records:
2798
data = raw_records.next()
2799
# validate the header
2800
df, rec = self._parse_record_header(version_id, data)
2802
yield version_id, data, rec[3]
2804
def read_records_iter(self, records):
2805
"""Read text records from data file and yield result.
2807
The result will be returned in whatever is the fastest to read.
2808
Not by the order requested. Also, multiple requests for the same
2809
record will only yield 1 response.
2810
:param records: A list of (version_id, pos, len) entries
2811
:return: Yields (version_id, contents, digest) in the order
2812
read, not the order requested
2817
needed_records = sorted(set(records), key=operator.itemgetter(1))
2818
if not needed_records:
2821
# The transport optimizes the fetching as well
2822
# (ie, reads continuous ranges.)
2823
raw_data = self._access.get_raw_records(
2824
[index_memo for version_id, index_memo in needed_records])
2826
for (version_id, index_memo), data in \
2827
izip(iter(needed_records), raw_data):
2828
content, digest = self._parse_record(version_id, data)
2829
yield version_id, content, digest
2831
def read_records(self, records):
2832
"""Read records into a dictionary."""
2834
for record_id, content, digest in \
2835
self.read_records_iter(records):
2836
components[record_id] = (content, digest)
2840
class InterKnit(InterVersionedFile):
2841
"""Optimised code paths for knit to knit operations."""
2843
_matching_file_from_factory = staticmethod(make_file_knit)
2844
_matching_file_to_factory = staticmethod(make_file_knit)
2847
def is_compatible(source, target):
2848
"""Be compatible with knits. """
2850
return (isinstance(source, KnitVersionedFile) and
2851
isinstance(target, KnitVersionedFile))
2852
except AttributeError:
2855
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2856
"""Copy texts to the target by extracting and adding them one by one.
2858
see join() for the parameter definitions.
2860
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2861
# --- the below is factorable out with VersionedFile.join, but wait for
2862
# VersionedFiles, it may all be simpler then.
2863
graph = Graph(self.source)
2864
search = graph._make_breadth_first_searcher(version_ids)
2865
transitive_ids = set()
2866
map(transitive_ids.update, list(search))
2867
parent_map = self.source.get_parent_map(transitive_ids)
2868
order = topo_sort(parent_map.items())
2870
def size_of_content(content):
2871
return sum(len(line) for line in content.text())
2872
# Cache at most 10MB of parent texts
2873
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2874
compute_size=size_of_content)
2875
# TODO: jam 20071116 It would be nice to have a streaming interface to
2876
# get multiple texts from a source. The source could be smarter
2877
# about how it handled intermediate stages.
2878
# get_line_list() or make_mpdiffs() seem like a possibility, but
2879
# at the moment they extract all full texts into memory, which
2880
# causes us to store more than our 3x fulltext goal.
2881
# Repository.iter_files_bytes() may be another possibility
2882
to_process = [version for version in order
2883
if version not in self.target]
2884
total = len(to_process)
2885
pb = ui.ui_factory.nested_progress_bar()
2887
for index, version in enumerate(to_process):
2888
pb.update('Converting versioned data', index, total)
2889
sha1, num_bytes, parent_text = self.target.add_lines(version,
2890
self.source.get_parents_with_ghosts(version),
2891
self.source.get_lines(version),
2892
parent_texts=parent_cache)
2893
parent_cache[version] = parent_text
2898
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2899
"""See InterVersionedFile.join."""
2900
# If the source and target are mismatched w.r.t. annotations vs
2901
# plain, the data needs to be converted accordingly
2902
if self.source.factory.annotated == self.target.factory.annotated:
2904
elif self.source.factory.annotated:
2905
converter = self._anno_to_plain_converter
2907
# We're converting from a plain to an annotated knit. Copy them
2908
# across by full texts.
2909
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2911
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2915
pb = ui.ui_factory.nested_progress_bar()
2917
version_ids = list(version_ids)
2918
if None in version_ids:
2919
version_ids.remove(None)
2921
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2923
this_versions = set(self.target._index.get_versions())
2924
# XXX: For efficiency we should not look at the whole index,
2925
# we only need to consider the referenced revisions - they
2926
# must all be present, or the method must be full-text.
2927
# TODO, RBC 20070919
2928
needed_versions = self.source_ancestry - this_versions
2930
if not needed_versions:
2932
full_list = topo_sort(
2933
self.source.get_parent_map(self.source.versions()))
2935
version_list = [i for i in full_list if (not self.target.has_version(i)
2936
and i in needed_versions)]
2940
copy_queue_records = []
2942
for version_id in version_list:
2943
options = self.source._index.get_options(version_id)
2944
parents = self.source._index.get_parents_with_ghosts(version_id)
2945
# check that its will be a consistent copy:
2946
for parent in parents:
2947
# if source has the parent, we must :
2948
# * already have it or
2949
# * have it scheduled already
2950
# otherwise we don't care
2951
if not (self.target.has_version(parent) or
2952
parent in copy_set or
2953
not self.source.has_version(parent)):
2954
raise AssertionError("problem joining parent %r "
2956
% (parent, self.source, self.target))
2957
index_memo = self.source._index.get_position(version_id)
2958
copy_queue_records.append((version_id, index_memo))
2959
copy_queue.append((version_id, options, parents))
2960
copy_set.add(version_id)
2962
# data suck the join:
2964
total = len(version_list)
2967
for (version_id, raw_data, _), \
2968
(version_id2, options, parents) in \
2969
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2971
if not (version_id == version_id2):
2972
raise AssertionError('logic error, inconsistent results')
2974
pb.update("Joining knit", count, total)
2976
size, raw_data = converter(raw_data, version_id, options,
2979
size = len(raw_data)
2980
raw_records.append((version_id, options, parents, size))
2981
raw_datum.append(raw_data)
2982
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2987
def _anno_to_plain_converter(self, raw_data, version_id, options,
2989
"""Convert annotated content to plain content."""
2990
data, digest = self.source._data._parse_record(version_id, raw_data)
2991
if 'fulltext' in options:
2992
content = self.source.factory.parse_fulltext(data, version_id)
2993
lines = self.target.factory.lower_fulltext(content)
2995
delta = self.source.factory.parse_line_delta(data, version_id,
2997
lines = self.target.factory.lower_line_delta(delta)
2998
return self.target._data._record_to_data(version_id, digest, lines)
3001
InterVersionedFile.register_optimiser(InterKnit)
3004
class WeaveToKnit(InterVersionedFile):
3005
"""Optimised code paths for weave to knit operations."""
3007
_matching_file_from_factory = bzrlib.weave.WeaveFile
3008
_matching_file_to_factory = staticmethod(make_file_knit)
3011
def is_compatible(source, target):
3012
"""Be compatible with weaves to knits."""
3014
return (isinstance(source, bzrlib.weave.Weave) and
3015
isinstance(target, KnitVersionedFile))
3016
except AttributeError:
3019
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
3020
"""See InterVersionedFile.join."""
3021
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
3026
pb = ui.ui_factory.nested_progress_bar()
3028
version_ids = list(version_ids)
3030
self.source_ancestry = set(self.source.get_ancestry(version_ids))
3031
this_versions = set(self.target._index.get_versions())
3032
needed_versions = self.source_ancestry - this_versions
3034
if not needed_versions:
3036
full_list = topo_sort(
3037
self.source.get_parent_map(self.source.versions()))
3039
version_list = [i for i in full_list if (not self.target.has_version(i)
3040
and i in needed_versions)]
3044
total = len(version_list)
3045
parent_map = self.source.get_parent_map(version_list)
3046
for version_id in version_list:
3047
pb.update("Converting to knit", count, total)
3048
parents = parent_map[version_id]
3049
# check that its will be a consistent copy:
3050
for parent in parents:
3051
# if source has the parent, we must already have it
3052
if not self.target.has_version(parent):
3053
raise AssertionError("%r does not have parent %r"
3054
% (self.target, parent))
3055
self.target.add_lines(
3056
version_id, parents, self.source.get_lines(version_id))
3063
InterVersionedFile.register_optimiser(WeaveToKnit)
3066
# Deprecated, use PatienceSequenceMatcher instead
3067
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
3070
def annotate_knit(knit, revision_id):
3071
"""Annotate a knit with no cached annotations.
3073
This implementation is for knits with no cached annotations.
3074
It will work for knits with cached annotations, but this is not
3077
annotator = _KnitAnnotator(knit)
3078
return iter(annotator.annotate(revision_id))
3081
class _KnitAnnotator(object):
3082
"""Build up the annotations for a text."""
3084
def __init__(self, knit):
3087
# Content objects, differs from fulltexts because of how final newlines
3088
# are treated by knits. the content objects here will always have a
3090
self._fulltext_contents = {}
3092
# Annotated lines of specific revisions
3093
self._annotated_lines = {}
3095
# Track the raw data for nodes that we could not process yet.
3096
# This maps the revision_id of the base to a list of children that will
3097
# annotated from it.
3098
self._pending_children = {}
3100
# Nodes which cannot be extracted
3101
self._ghosts = set()
3103
# Track how many children this node has, so we know if we need to keep
3105
self._annotate_children = {}
3106
self._compression_children = {}
3108
self._all_build_details = {}
3109
# The children => parent revision_id graph
3110
self._revision_id_graph = {}
3112
self._heads_provider = None
3114
self._nodes_to_keep_annotations = set()
3115
self._generations_until_keep = 100
3117
def set_generations_until_keep(self, value):
3118
"""Set the number of generations before caching a node.
3120
Setting this to -1 will cache every merge node, setting this higher
3121
will cache fewer nodes.
3123
self._generations_until_keep = value
3125
def _add_fulltext_content(self, revision_id, content_obj):
3126
self._fulltext_contents[revision_id] = content_obj
3127
# TODO: jam 20080305 It might be good to check the sha1digest here
3128
return content_obj.text()
3130
def _check_parents(self, child, nodes_to_annotate):
3131
"""Check if all parents have been processed.
3133
:param child: A tuple of (rev_id, parents, raw_content)
3134
:param nodes_to_annotate: If child is ready, add it to
3135
nodes_to_annotate, otherwise put it back in self._pending_children
3137
for parent_id in child[1]:
3138
if (parent_id not in self._annotated_lines):
3139
# This parent is present, but another parent is missing
3140
self._pending_children.setdefault(parent_id,
3144
# This one is ready to be processed
3145
nodes_to_annotate.append(child)
3147
def _add_annotation(self, revision_id, fulltext, parent_ids,
3148
left_matching_blocks=None):
3149
"""Add an annotation entry.
3151
All parents should already have been annotated.
3152
:return: A list of children that now have their parents satisfied.
3154
a = self._annotated_lines
3155
annotated_parent_lines = [a[p] for p in parent_ids]
3156
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
3157
fulltext, revision_id, left_matching_blocks,
3158
heads_provider=self._get_heads_provider()))
3159
self._annotated_lines[revision_id] = annotated_lines
3160
for p in parent_ids:
3161
ann_children = self._annotate_children[p]
3162
ann_children.remove(revision_id)
3163
if (not ann_children
3164
and p not in self._nodes_to_keep_annotations):
3165
del self._annotated_lines[p]
3166
del self._all_build_details[p]
3167
if p in self._fulltext_contents:
3168
del self._fulltext_contents[p]
3169
# Now that we've added this one, see if there are any pending
3170
# deltas to be done, certainly this parent is finished
3171
nodes_to_annotate = []
3172
for child in self._pending_children.pop(revision_id, []):
3173
self._check_parents(child, nodes_to_annotate)
3174
return nodes_to_annotate
3176
def _get_build_graph(self, revision_id):
3177
"""Get the graphs for building texts and annotations.
3179
The data you need for creating a full text may be different than the
3180
data you need to annotate that text. (At a minimum, you need both
3181
parents to create an annotation, but only need 1 parent to generate the
3184
:return: A list of (revision_id, index_memo) records, suitable for
3185
passing to read_records_iter to start reading in the raw data fro/
3188
if revision_id in self._annotated_lines:
3191
pending = set([revision_id])
3196
# get all pending nodes
3198
this_iteration = pending
3199
build_details = self._knit._index.get_build_details(this_iteration)
3200
self._all_build_details.update(build_details)
3201
# new_nodes = self._knit._index._get_entries(this_iteration)
3203
for rev_id, details in build_details.iteritems():
3204
(index_memo, compression_parent, parents,
3205
record_details) = details
3206
self._revision_id_graph[rev_id] = parents
3207
records.append((rev_id, index_memo))
3208
# Do we actually need to check _annotated_lines?
3209
pending.update(p for p in parents
3210
if p not in self._all_build_details)
3211
if compression_parent:
3212
self._compression_children.setdefault(compression_parent,
3215
for parent in parents:
3216
self._annotate_children.setdefault(parent,
3218
num_gens = generation - kept_generation
3219
if ((num_gens >= self._generations_until_keep)
3220
and len(parents) > 1):
3221
kept_generation = generation
3222
self._nodes_to_keep_annotations.add(rev_id)
3224
missing_versions = this_iteration.difference(build_details.keys())
3225
self._ghosts.update(missing_versions)
3226
for missing_version in missing_versions:
3227
# add a key, no parents
3228
self._revision_id_graph[missing_version] = ()
3229
pending.discard(missing_version) # don't look for it
3230
if self._ghosts.intersection(self._compression_children):
3232
"We cannot have nodes which have a ghost compression parent:\n"
3234
"compression children: %r"
3235
% (self._ghosts, self._compression_children))
3236
# Cleanout anything that depends on a ghost so that we don't wait for
3237
# the ghost to show up
3238
for node in self._ghosts:
3239
if node in self._annotate_children:
3240
# We won't be building this node
3241
del self._annotate_children[node]
3242
# Generally we will want to read the records in reverse order, because
3243
# we find the parent nodes after the children
3247
def _annotate_records(self, records):
3248
"""Build the annotations for the listed records."""
3249
# We iterate in the order read, rather than a strict order requested
3250
# However, process what we can, and put off to the side things that
3251
# still need parents, cleaning them up when those parents are
3253
for (rev_id, record,
3254
digest) in self._knit._data.read_records_iter(records):
3255
if rev_id in self._annotated_lines:
3257
parent_ids = self._revision_id_graph[rev_id]
3258
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3259
details = self._all_build_details[rev_id]
3260
(index_memo, compression_parent, parents,
3261
record_details) = details
3262
nodes_to_annotate = []
3263
# TODO: Remove the punning between compression parents, and
3264
# parent_ids, we should be able to do this without assuming
3266
if len(parent_ids) == 0:
3267
# There are no parents for this node, so just add it
3268
# TODO: This probably needs to be decoupled
3269
fulltext_content, delta = self._knit.factory.parse_record(
3270
rev_id, record, record_details, None)
3271
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3272
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3273
parent_ids, left_matching_blocks=None))
3275
child = (rev_id, parent_ids, record)
3276
# Check if all the parents are present
3277
self._check_parents(child, nodes_to_annotate)
3278
while nodes_to_annotate:
3279
# Should we use a queue here instead of a stack?
3280
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3281
(index_memo, compression_parent, parents,
3282
record_details) = self._all_build_details[rev_id]
3283
if compression_parent is not None:
3284
comp_children = self._compression_children[compression_parent]
3285
if rev_id not in comp_children:
3286
raise AssertionError("%r not in compression children %r"
3287
% (rev_id, comp_children))
3288
# If there is only 1 child, it is safe to reuse this
3290
reuse_content = (len(comp_children) == 1
3291
and compression_parent not in
3292
self._nodes_to_keep_annotations)
3294
# Remove it from the cache since it will be changing
3295
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3296
# Make sure to copy the fulltext since it might be
3298
parent_fulltext = list(parent_fulltext_content.text())
3300
parent_fulltext_content = self._fulltext_contents[compression_parent]
3301
parent_fulltext = parent_fulltext_content.text()
3302
comp_children.remove(rev_id)
3303
fulltext_content, delta = self._knit.factory.parse_record(
3304
rev_id, record, record_details,
3305
parent_fulltext_content,
3306
copy_base_content=(not reuse_content))
3307
fulltext = self._add_fulltext_content(rev_id,
3309
blocks = KnitContent.get_line_delta_blocks(delta,
3310
parent_fulltext, fulltext)
3312
fulltext_content = self._knit.factory.parse_fulltext(
3314
fulltext = self._add_fulltext_content(rev_id,
3317
nodes_to_annotate.extend(
3318
self._add_annotation(rev_id, fulltext, parent_ids,
3319
left_matching_blocks=blocks))
3321
def _get_heads_provider(self):
3322
"""Create a heads provider for resolving ancestry issues."""
3323
if self._heads_provider is not None:
3324
return self._heads_provider
3325
parent_provider = _mod_graph.DictParentsProvider(
3326
self._revision_id_graph)
3327
graph_obj = _mod_graph.Graph(parent_provider)
3328
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3329
self._heads_provider = head_cache
3332
def annotate(self, revision_id):
3333
"""Return the annotated fulltext at the given revision.
3335
:param revision_id: The revision id for this file
3337
records = self._get_build_graph(revision_id)
3338
if revision_id in self._ghosts:
3339
raise errors.RevisionNotPresent(revision_id, self._knit)
3340
self._annotate_records(records)
3341
return self._annotated_lines[revision_id]
3345
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3347
from bzrlib._knit_load_data_py import _load_data_py as _load_data