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."""
286
self._should_strip_eol = False
288
def apply_delta(self, delta, new_version_id):
289
"""Apply delta to this object to become new_version_id."""
290
raise NotImplementedError(self.apply_delta)
292
def cleanup_eol(self, copy_on_mutate=True):
293
if self._should_strip_eol:
295
self._lines = self._lines[:]
296
self.strip_last_line_newline()
298
def line_delta_iter(self, new_lines):
299
"""Generate line-based delta from this content to new_lines."""
300
new_texts = new_lines.text()
301
old_texts = self.text()
302
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
303
for tag, i1, i2, j1, j2 in s.get_opcodes():
306
# ofrom, oto, length, data
307
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
309
def line_delta(self, new_lines):
310
return list(self.line_delta_iter(new_lines))
313
def get_line_delta_blocks(knit_delta, source, target):
314
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
315
target_len = len(target)
318
for s_begin, s_end, t_len, new_text in knit_delta:
319
true_n = s_begin - s_pos
322
# knit deltas do not provide reliable info about whether the
323
# last line of a file matches, due to eol handling.
324
if source[s_pos + n -1] != target[t_pos + n -1]:
327
yield s_pos, t_pos, n
328
t_pos += t_len + true_n
330
n = target_len - t_pos
332
if source[s_pos + n -1] != target[t_pos + n -1]:
335
yield s_pos, t_pos, n
336
yield s_pos + (target_len - t_pos), target_len, 0
339
class AnnotatedKnitContent(KnitContent):
340
"""Annotated content."""
342
def __init__(self, lines):
343
KnitContent.__init__(self)
347
"""Return a list of (origin, text) for each content line."""
348
return list(self._lines)
350
def apply_delta(self, delta, new_version_id):
351
"""Apply delta to this object to become new_version_id."""
354
for start, end, count, delta_lines in delta:
355
lines[offset+start:offset+end] = delta_lines
356
offset = offset + (start - end) + count
358
def strip_last_line_newline(self):
359
line = self._lines[-1][1].rstrip('\n')
360
self._lines[-1] = (self._lines[-1][0], line)
361
self._should_strip_eol = False
365
lines = [text for origin, text in self._lines]
366
except ValueError, e:
367
# most commonly (only?) caused by the internal form of the knit
368
# missing annotation information because of a bug - see thread
370
raise KnitCorrupt(self,
371
"line in annotated knit missing annotation information: %s"
374
if self._should_strip_eol:
375
lines[-1] = lines[-1].rstrip('\n')
379
return AnnotatedKnitContent(self._lines[:])
382
class PlainKnitContent(KnitContent):
383
"""Unannotated content.
385
When annotate[_iter] is called on this content, the same version is reported
386
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
390
def __init__(self, lines, version_id):
391
KnitContent.__init__(self)
393
self._version_id = version_id
396
"""Return a list of (origin, text) for each content line."""
397
return [(self._version_id, line) for line in self._lines]
399
def apply_delta(self, delta, new_version_id):
400
"""Apply delta to this object to become new_version_id."""
403
for start, end, count, delta_lines in delta:
404
lines[offset+start:offset+end] = delta_lines
405
offset = offset + (start - end) + count
406
self._version_id = new_version_id
409
return PlainKnitContent(self._lines[:], self._version_id)
411
def strip_last_line_newline(self):
412
self._lines[-1] = self._lines[-1].rstrip('\n')
413
self._should_strip_eol = False
417
if self._should_strip_eol:
419
lines[-1] = lines[-1].rstrip('\n')
423
class _KnitFactory(object):
424
"""Base class for common Factory functions."""
426
def parse_record(self, version_id, record, record_details,
427
base_content, copy_base_content=True):
428
"""Parse a record into a full content object.
430
:param version_id: The official version id for this content
431
:param record: The data returned by read_records_iter()
432
:param record_details: Details about the record returned by
434
:param base_content: If get_build_details returns a compression_parent,
435
you must return a base_content here, else use None
436
:param copy_base_content: When building from the base_content, decide
437
you can either copy it and return a new object, or modify it in
439
:return: (content, delta) A Content object and possibly a line-delta,
442
method, noeol = record_details
443
if method == 'line-delta':
444
if copy_base_content:
445
content = base_content.copy()
447
content = base_content
448
delta = self.parse_line_delta(record, version_id)
449
content.apply_delta(delta, version_id)
451
content = self.parse_fulltext(record, version_id)
453
content._should_strip_eol = noeol
454
return (content, delta)
457
class KnitAnnotateFactory(_KnitFactory):
458
"""Factory for creating annotated Content objects."""
462
def make(self, lines, version_id):
463
num_lines = len(lines)
464
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
466
def parse_fulltext(self, content, version_id):
467
"""Convert fulltext to internal representation
469
fulltext content is of the format
470
revid(utf8) plaintext\n
471
internal representation is of the format:
474
# TODO: jam 20070209 The tests expect this to be returned as tuples,
475
# but the code itself doesn't really depend on that.
476
# Figure out a way to not require the overhead of turning the
477
# list back into tuples.
478
lines = [tuple(line.split(' ', 1)) for line in content]
479
return AnnotatedKnitContent(lines)
481
def parse_line_delta_iter(self, lines):
482
return iter(self.parse_line_delta(lines))
484
def parse_line_delta(self, lines, version_id, plain=False):
485
"""Convert a line based delta into internal representation.
487
line delta is in the form of:
488
intstart intend intcount
490
revid(utf8) newline\n
491
internal representation is
492
(start, end, count, [1..count tuples (revid, newline)])
494
:param plain: If True, the lines are returned as a plain
495
list without annotations, not as a list of (origin, content) tuples, i.e.
496
(start, end, count, [1..count newline])
503
def cache_and_return(line):
504
origin, text = line.split(' ', 1)
505
return cache.setdefault(origin, origin), text
507
# walk through the lines parsing.
508
# Note that the plain test is explicitly pulled out of the
509
# loop to minimise any performance impact
512
start, end, count = [int(n) for n in header.split(',')]
513
contents = [next().split(' ', 1)[1] for i in xrange(count)]
514
result.append((start, end, count, contents))
517
start, end, count = [int(n) for n in header.split(',')]
518
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
519
result.append((start, end, count, contents))
522
def get_fulltext_content(self, lines):
523
"""Extract just the content lines from a fulltext."""
524
return (line.split(' ', 1)[1] for line in lines)
526
def get_linedelta_content(self, lines):
527
"""Extract just the content from a line delta.
529
This doesn't return all of the extra information stored in a delta.
530
Only the actual content lines.
535
header = header.split(',')
536
count = int(header[2])
537
for i in xrange(count):
538
origin, text = next().split(' ', 1)
541
def lower_fulltext(self, content):
542
"""convert a fulltext content record into a serializable form.
544
see parse_fulltext which this inverts.
546
# TODO: jam 20070209 We only do the caching thing to make sure that
547
# the origin is a valid utf-8 line, eventually we could remove it
548
return ['%s %s' % (o, t) for o, t in content._lines]
550
def lower_line_delta(self, delta):
551
"""convert a delta into a serializable form.
553
See parse_line_delta which this inverts.
555
# TODO: jam 20070209 We only do the caching thing to make sure that
556
# the origin is a valid utf-8 line, eventually we could remove it
558
for start, end, c, lines in delta:
559
out.append('%d,%d,%d\n' % (start, end, c))
560
out.extend(origin + ' ' + text
561
for origin, text in lines)
564
def annotate(self, knit, version_id):
565
content = knit._get_content(version_id)
566
return content.annotate()
569
class KnitPlainFactory(_KnitFactory):
570
"""Factory for creating plain Content objects."""
574
def make(self, lines, version_id):
575
return PlainKnitContent(lines, version_id)
577
def parse_fulltext(self, content, version_id):
578
"""This parses an unannotated fulltext.
580
Note that this is not a noop - the internal representation
581
has (versionid, line) - its just a constant versionid.
583
return self.make(content, version_id)
585
def parse_line_delta_iter(self, lines, version_id):
587
num_lines = len(lines)
588
while cur < num_lines:
591
start, end, c = [int(n) for n in header.split(',')]
592
yield start, end, c, lines[cur:cur+c]
595
def parse_line_delta(self, lines, version_id):
596
return list(self.parse_line_delta_iter(lines, version_id))
598
def get_fulltext_content(self, lines):
599
"""Extract just the content lines from a fulltext."""
602
def get_linedelta_content(self, lines):
603
"""Extract just the content from a line delta.
605
This doesn't return all of the extra information stored in a delta.
606
Only the actual content lines.
611
header = header.split(',')
612
count = int(header[2])
613
for i in xrange(count):
616
def lower_fulltext(self, content):
617
return content.text()
619
def lower_line_delta(self, delta):
621
for start, end, c, lines in delta:
622
out.append('%d,%d,%d\n' % (start, end, c))
626
def annotate(self, knit, version_id):
627
annotator = _KnitAnnotator(knit)
628
return annotator.annotate(version_id)
631
def make_empty_knit(transport, relpath):
632
"""Construct a empty knit at the specified location."""
633
k = make_file_knit(transport, relpath, 'w', KnitPlainFactory)
636
def make_file_knit(name, transport, file_mode=None, access_mode='w',
637
factory=None, delta=True, create=False, create_parent_dir=False,
638
delay_create=False, dir_mode=None, get_scope=None):
639
"""Factory to create a KnitVersionedFile for a .knit/.kndx file pair."""
641
factory = KnitAnnotateFactory()
642
if get_scope is None:
643
get_scope = lambda:None
644
index = _KnitIndex(transport, name + INDEX_SUFFIX,
645
access_mode, create=create, file_mode=file_mode,
646
create_parent_dir=create_parent_dir, delay_create=delay_create,
647
dir_mode=dir_mode, get_scope=get_scope)
648
access = _KnitAccess(transport, name + DATA_SUFFIX, file_mode,
649
dir_mode, ((create and not len(index)) and delay_create),
651
return KnitVersionedFile(name, transport, factory=factory,
652
create=create, delay_create=delay_create, index=index,
653
access_method=access)
657
"""Return the suffixes used by file based knits."""
658
return [DATA_SUFFIX, INDEX_SUFFIX]
659
make_file_knit.get_suffixes = get_suffixes
662
class KnitVersionedFile(VersionedFile):
663
"""Weave-like structure with faster random access.
665
A knit stores a number of texts and a summary of the relationships
666
between them. Texts are identified by a string version-id. Texts
667
are normally stored and retrieved as a series of lines, but can
668
also be passed as single strings.
670
Lines are stored with the trailing newline (if any) included, to
671
avoid special cases for files with no final newline. Lines are
672
composed of 8-bit characters, not unicode. The combination of
673
these approaches should mean any 'binary' file can be safely
674
stored and retrieved.
677
def __init__(self, relpath, transport, file_mode=None,
678
factory=None, delta=True, create=False, create_parent_dir=False,
679
delay_create=False, dir_mode=None, index=None, access_method=None):
680
"""Construct a knit at location specified by relpath.
682
:param create: If not True, only open an existing knit.
683
:param create_parent_dir: If True, create the parent directory if
684
creating the file fails. (This is used for stores with
685
hash-prefixes that may not exist yet)
686
:param delay_create: The calling code is aware that the knit won't
687
actually be created until the first data is stored.
688
:param index: An index to use for the knit.
690
super(KnitVersionedFile, self).__init__()
691
self.transport = transport
692
self.filename = relpath
693
self.factory = factory or KnitAnnotateFactory()
696
self._max_delta_chain = 200
698
if None in (access_method, index):
699
raise ValueError("No default access_method or index any more")
701
_access = access_method
702
if create and not len(self) and not delay_create:
704
self._data = _KnitData(_access)
707
return '%s(%s)' % (self.__class__.__name__,
708
self.transport.abspath(self.filename))
710
def _check_should_delta(self, first_parents):
711
"""Iterate back through the parent listing, looking for a fulltext.
713
This is used when we want to decide whether to add a delta or a new
714
fulltext. It searches for _max_delta_chain parents. When it finds a
715
fulltext parent, it sees if the total size of the deltas leading up to
716
it is large enough to indicate that we want a new full text anyway.
718
Return True if we should create a new delta, False if we should use a
723
delta_parents = first_parents
724
for count in xrange(self._max_delta_chain):
725
parent = delta_parents[0]
726
method = self._index.get_method(parent)
727
index, pos, size = self._index.get_position(parent)
728
if method == 'fulltext':
732
delta_parents = self._index.get_parent_map([parent])[parent]
734
# We couldn't find a fulltext, so we must create a new one
737
return fulltext_size > delta_size
739
def _check_write_ok(self):
740
return self._index._check_write_ok()
742
def _add_raw_records(self, records, data):
743
"""Add all the records 'records' with data pre-joined in 'data'.
745
:param records: A list of tuples(version_id, options, parents, size).
746
:param data: The data for the records. When it is written, the records
747
are adjusted to have pos pointing into data by the sum of
748
the preceding records sizes.
751
raw_record_sizes = [record[3] for record in records]
752
positions = self._data.add_raw_records(raw_record_sizes, data)
754
for (version_id, options, parents, _), access_memo in zip(
756
index_entries.append((version_id, options, access_memo, parents))
757
self._index.add_versions(index_entries)
759
def copy_to(self, name, transport):
760
"""See VersionedFile.copy_to()."""
761
# copy the current index to a temp index to avoid racing with local
763
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
764
self.transport.get(self._index._filename))
766
f = self._data._open_file()
768
transport.put_file(name + DATA_SUFFIX, f)
771
# move the copied index into place
772
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
774
def get_data_stream(self, required_versions):
775
"""Get a data stream for the specified versions.
777
Versions may be returned in any order, not necessarily the order
778
specified. They are returned in a partial order by compression
779
parent, so that the deltas can be applied as the data stream is
780
inserted; however note that compression parents will not be sent
781
unless they were specifically requested, as the client may already
784
:param required_versions: The exact set of versions to be extracted.
785
Unlike some other knit methods, this is not used to generate a
786
transitive closure, rather it is used precisely as given.
788
:returns: format_signature, list of (version, options, length, parents),
791
required_version_set = frozenset(required_versions)
793
# list of revisions that can just be sent without waiting for their
796
# map from revision to the children based on it
798
# first, read all relevant index data, enough to sort into the right
800
for version_id in required_versions:
801
options = self._index.get_options(version_id)
802
parents = self._index.get_parents_with_ghosts(version_id)
803
index_memo = self._index.get_position(version_id)
804
version_index[version_id] = (index_memo, options, parents)
805
if ('line-delta' in options
806
and parents[0] in required_version_set):
807
# must wait until the parent has been sent
808
deferred.setdefault(parents[0], []). \
811
# either a fulltext, or a delta whose parent the client did
812
# not ask for and presumably already has
813
ready_to_send.append(version_id)
814
# build a list of results to return, plus instructions for data to
816
copy_queue_records = []
817
temp_version_list = []
819
# XXX: pushing and popping lists may be a bit inefficient
820
version_id = ready_to_send.pop(0)
821
(index_memo, options, parents) = version_index[version_id]
822
copy_queue_records.append((version_id, index_memo))
823
none, data_pos, data_size = index_memo
824
temp_version_list.append((version_id, options, data_size,
826
if version_id in deferred:
827
# now we can send all the children of this revision - we could
828
# put them in anywhere, but we hope that sending them soon
829
# after the fulltext will give good locality in the receiver
830
ready_to_send[:0] = deferred.pop(version_id)
831
if not (len(deferred) == 0):
832
raise AssertionError("Still have compressed child versions waiting to be sent")
833
# XXX: The stream format is such that we cannot stream it - we have to
834
# know the length of all the data a-priori.
836
result_version_list = []
837
for (version_id, raw_data, _), \
838
(version_id2, options, _, parents) in \
839
izip(self._data.read_records_iter_raw(copy_queue_records),
841
if not (version_id == version_id2):
842
raise AssertionError('logic error, inconsistent results')
843
raw_datum.append(raw_data)
844
result_version_list.append(
845
(version_id, options, len(raw_data), parents))
846
# provide a callback to get data incrementally.
847
pseudo_file = StringIO(''.join(raw_datum))
850
return pseudo_file.read()
852
return pseudo_file.read(length)
853
return (self.get_format_signature(), result_version_list, read)
855
def get_record_stream(self, versions, ordering, include_delta_closure):
856
"""Get a stream of records for versions.
858
:param versions: The versions to include. Each version is a tuple
860
:param ordering: Either 'unordered' or 'topological'. A topologically
861
sorted stream has compression parents strictly before their
863
:param include_delta_closure: If True then the closure across any
864
compression parents will be included (in the opaque data).
865
:return: An iterator of ContentFactory objects, each of which is only
866
valid until the iterator is advanced.
868
if include_delta_closure:
869
# Nb: what we should do is plan the data to stream to allow
870
# reconstruction of all the texts without excessive buffering,
871
# including re-sending common bases as needed. This makes the most
872
# sense when we start serialising these streams though, so for now
873
# we just fallback to individual text construction behind the
874
# abstraction barrier.
878
# We end up doing multiple index lookups here for parents details and
879
# disk layout details - we need a unified api ?
880
parent_map = self.get_parent_map(versions)
881
absent_versions = set(versions) - set(parent_map)
882
if ordering == 'topological':
883
present_versions = topo_sort(parent_map)
885
# List comprehension to keep the requested order (as that seems
886
# marginally useful, at least until we start doing IO optimising
888
present_versions = [version for version in versions if version in
890
position_map = self._get_components_positions(present_versions)
891
records = [(version, position_map[version][1]) for version in
894
for version in absent_versions:
895
yield AbsentContentFactory((version,))
896
for version, raw_data, sha1 in \
897
self._data.read_records_iter_raw(records):
898
(record_details, index_memo, _) = position_map[version]
899
yield KnitContentFactory(version, parent_map[version],
900
record_details, sha1, raw_data, self.factory.annotated, knit)
902
def _extract_blocks(self, version_id, source, target):
903
if self._index.get_method(version_id) != 'line-delta':
905
parent, sha1, noeol, delta = self.get_delta(version_id)
906
return KnitContent.get_line_delta_blocks(delta, source, target)
908
def get_delta(self, version_id):
909
"""Get a delta for constructing version from some other version."""
910
self.check_not_reserved_id(version_id)
911
parents = self.get_parent_map([version_id])[version_id]
916
index_memo = self._index.get_position(version_id)
917
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
918
noeol = 'no-eol' in self._index.get_options(version_id)
919
if 'fulltext' == self._index.get_method(version_id):
920
new_content = self.factory.parse_fulltext(data, version_id)
921
if parent is not None:
922
reference_content = self._get_content(parent)
923
old_texts = reference_content.text()
926
new_texts = new_content.text()
927
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
929
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
931
delta = self.factory.parse_line_delta(data, version_id)
932
return parent, sha1, noeol, delta
934
def get_format_signature(self):
935
"""See VersionedFile.get_format_signature()."""
936
if self.factory.annotated:
937
annotated_part = "annotated"
939
annotated_part = "plain"
940
return "knit-%s" % (annotated_part,)
942
def get_sha1s(self, version_ids):
943
"""See VersionedFile.get_sha1s()."""
944
record_map = self._get_record_map(version_ids)
945
# record entry 2 is the 'digest'.
946
return [record_map[v][2] for v in version_ids]
948
def insert_data_stream(self, (format, data_list, reader_callable)):
949
"""Insert knit records from a data stream into this knit.
951
If a version in the stream is already present in this knit, it will not
952
be inserted a second time. It will be checked for consistency with the
953
stored version however, and may cause a KnitCorrupt error to be raised
954
if the data in the stream disagrees with the already stored data.
956
:seealso: get_data_stream
958
if format != self.get_format_signature():
959
if 'knit' in debug.debug_flags:
961
'incompatible format signature inserting to %r', self)
962
source = self._knit_from_datastream(
963
(format, data_list, reader_callable))
964
stream = source.get_record_stream(source.versions(), 'unordered', False)
965
self.insert_record_stream(stream)
968
for version_id, options, length, parents in data_list:
969
if self.has_version(version_id):
970
# First check: the list of parents.
971
my_parents = self.get_parents_with_ghosts(version_id)
972
if tuple(my_parents) != tuple(parents):
973
# XXX: KnitCorrupt is not quite the right exception here.
976
'parents list %r from data stream does not match '
977
'already recorded parents %r for %s'
978
% (parents, my_parents, version_id))
980
# Also check the SHA-1 of the fulltext this content will
982
raw_data = reader_callable(length)
983
my_fulltext_sha1 = self.get_sha1s([version_id])[0]
984
df, rec = self._data._parse_record_header(version_id, raw_data)
985
stream_fulltext_sha1 = rec[3]
986
if my_fulltext_sha1 != stream_fulltext_sha1:
987
# Actually, we don't know if it's this knit that's corrupt,
988
# or the data stream we're trying to insert.
990
self.filename, 'sha-1 does not match %s' % version_id)
992
if 'line-delta' in options:
993
# Make sure that this knit record is actually useful: a
994
# line-delta is no use unless we have its parent.
995
# Fetching from a broken repository with this problem
996
# shouldn't break the target repository.
998
# See https://bugs.launchpad.net/bzr/+bug/164443
999
if not self._index.has_version(parents[0]):
1002
'line-delta from stream '
1005
'missing parent %s\n'
1006
'Try running "bzr check" '
1007
'on the source repository, and "bzr reconcile" '
1009
(version_id, parents[0]))
1011
# We received a line-delta record for a non-delta knit.
1012
# Convert it to a fulltext.
1013
gzip_bytes = reader_callable(length)
1014
self._convert_line_delta_to_fulltext(
1015
gzip_bytes, version_id, parents)
1018
self._add_raw_records(
1019
[(version_id, options, parents, length)],
1020
reader_callable(length))
1022
def _convert_line_delta_to_fulltext(self, gzip_bytes, version_id, parents):
1023
lines, sha1 = self._data._parse_record(version_id, gzip_bytes)
1024
delta = self.factory.parse_line_delta(lines, version_id)
1025
content = self.factory.make(self.get_lines(parents[0]), parents[0])
1026
content.apply_delta(delta, version_id)
1027
digest, len, content = self.add_lines(
1028
version_id, parents, content.text())
1030
raise errors.VersionedFileInvalidChecksum(version_id)
1032
def _knit_from_datastream(self, (format, data_list, reader_callable)):
1033
"""Create a knit object from a data stream.
1035
This method exists to allow conversion of data streams that do not
1036
match the signature of this knit. Generally it will be slower and use
1037
more memory to use this method to insert data, but it will work.
1039
:seealso: get_data_stream for details on datastreams.
1040
:return: A knit versioned file which can be used to join the datastream
1043
if format == "knit-plain":
1044
factory = KnitPlainFactory()
1045
elif format == "knit-annotated":
1046
factory = KnitAnnotateFactory()
1048
raise errors.KnitDataStreamUnknown(format)
1049
index = _StreamIndex(data_list, self._index)
1050
access = _StreamAccess(reader_callable, index, self, factory)
1051
return KnitVersionedFile(self.filename, self.transport,
1052
factory=factory, index=index, access_method=access)
1054
def insert_record_stream(self, stream):
1055
"""Insert a record stream into this versioned file.
1057
:param stream: A stream of records to insert.
1059
:seealso VersionedFile.get_record_stream:
1061
def get_adapter(adapter_key):
1063
return adapters[adapter_key]
1065
adapter_factory = adapter_registry.get(adapter_key)
1066
adapter = adapter_factory(self)
1067
adapters[adapter_key] = adapter
1069
if self.factory.annotated:
1070
# self is annotated, we need annotated knits to use directly.
1071
annotated = "annotated-"
1074
# self is not annotated, but we can strip annotations cheaply.
1076
convertibles = set(["knit-annotated-delta-gz",
1077
"knit-annotated-ft-gz"])
1078
# The set of types we can cheaply adapt without needing basis texts.
1079
native_types = set()
1080
native_types.add("knit-%sdelta-gz" % annotated)
1081
native_types.add("knit-%sft-gz" % annotated)
1082
knit_types = native_types.union(convertibles)
1084
# Buffer all index entries that we can't add immediately because their
1085
# basis parent is missing. We don't buffer all because generating
1086
# annotations may require access to some of the new records. However we
1087
# can't generate annotations from new deltas until their basis parent
1088
# is present anyway, so we get away with not needing an index that
1089
# includes the new keys.
1090
# key = basis_parent, value = index entry to add
1091
buffered_index_entries = {}
1092
for record in stream:
1093
# Raise an error when a record is missing.
1094
if record.storage_kind == 'absent':
1095
raise RevisionNotPresent([record.key[0]], self)
1096
# adapt to non-tuple interface
1097
parents = [parent[0] for parent in record.parents]
1098
if record.storage_kind in knit_types:
1099
if record.storage_kind not in native_types:
1101
adapter_key = (record.storage_kind, "knit-delta-gz")
1102
adapter = get_adapter(adapter_key)
1104
adapter_key = (record.storage_kind, "knit-ft-gz")
1105
adapter = get_adapter(adapter_key)
1106
bytes = adapter.get_bytes(
1107
record, record.get_bytes_as(record.storage_kind))
1109
bytes = record.get_bytes_as(record.storage_kind)
1110
options = [record._build_details[0]]
1111
if record._build_details[1]:
1112
options.append('no-eol')
1113
# Just blat it across.
1114
# Note: This does end up adding data on duplicate keys. As
1115
# modern repositories use atomic insertions this should not
1116
# lead to excessive growth in the event of interrupted fetches.
1117
# 'knit' repositories may suffer excessive growth, but as a
1118
# deprecated format this is tolerable. It can be fixed if
1119
# needed by in the kndx index support raising on a duplicate
1120
# add with identical parents and options.
1121
access_memo = self._data.add_raw_records([len(bytes)], bytes)[0]
1122
index_entry = (record.key[0], options, access_memo, parents)
1124
if 'fulltext' not in options:
1125
basis_parent = parents[0]
1126
if not self.has_version(basis_parent):
1127
pending = buffered_index_entries.setdefault(
1129
pending.append(index_entry)
1132
self._index.add_versions([index_entry])
1133
elif record.storage_kind == 'fulltext':
1134
self.add_lines(record.key[0], parents,
1135
split_lines(record.get_bytes_as('fulltext')))
1137
adapter_key = record.storage_kind, 'fulltext'
1138
adapter = get_adapter(adapter_key)
1139
lines = split_lines(adapter.get_bytes(
1140
record, record.get_bytes_as(record.storage_kind)))
1142
self.add_lines(record.key[0], parents, lines)
1143
except errors.RevisionAlreadyPresent:
1145
# Add any records whose basis parent is now available.
1146
added_keys = [record.key[0]]
1148
key = added_keys.pop(0)
1149
if key in buffered_index_entries:
1150
index_entries = buffered_index_entries[key]
1151
self._index.add_versions(index_entries)
1153
[index_entry[0] for index_entry in index_entries])
1154
del buffered_index_entries[key]
1155
# If there were any deltas which had a missing basis parent, error.
1156
if buffered_index_entries:
1157
raise errors.RevisionNotPresent(buffered_index_entries.keys()[0],
1161
"""See VersionedFile.versions."""
1162
if 'evil' in debug.debug_flags:
1163
trace.mutter_callsite(2, "versions scales with size of history")
1164
return self._index.get_versions()
1166
def has_version(self, version_id):
1167
"""See VersionedFile.has_version."""
1168
if 'evil' in debug.debug_flags:
1169
trace.mutter_callsite(2, "has_version is a LBYL scenario")
1170
return self._index.has_version(version_id)
1172
__contains__ = has_version
1174
def _merge_annotations(self, content, parents, parent_texts={},
1175
delta=None, annotated=None,
1176
left_matching_blocks=None):
1177
"""Merge annotations for content. This is done by comparing
1178
the annotations based on changed to the text.
1180
if left_matching_blocks is not None:
1181
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
1185
for parent_id in parents:
1186
merge_content = self._get_content(parent_id, parent_texts)
1187
if (parent_id == parents[0] and delta_seq is not None):
1190
seq = patiencediff.PatienceSequenceMatcher(
1191
None, merge_content.text(), content.text())
1192
for i, j, n in seq.get_matching_blocks():
1195
# this appears to copy (origin, text) pairs across to the
1196
# new content for any line that matches the last-checked
1198
content._lines[j:j+n] = merge_content._lines[i:i+n]
1200
if delta_seq is None:
1201
reference_content = self._get_content(parents[0], parent_texts)
1202
new_texts = content.text()
1203
old_texts = reference_content.text()
1204
delta_seq = patiencediff.PatienceSequenceMatcher(
1205
None, old_texts, new_texts)
1206
return self._make_line_delta(delta_seq, content)
1208
def _make_line_delta(self, delta_seq, new_content):
1209
"""Generate a line delta from delta_seq and new_content."""
1211
for op in delta_seq.get_opcodes():
1212
if op[0] == 'equal':
1214
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1217
def _get_components_positions(self, version_ids):
1218
"""Produce a map of position data for the components of versions.
1220
This data is intended to be used for retrieving the knit records.
1222
A dict of version_id to (record_details, index_memo, next, parents) is
1224
method is the way referenced data should be applied.
1225
index_memo is the handle to pass to the data access to actually get the
1227
next is the build-parent of the version, or None for fulltexts.
1228
parents is the version_ids of the parents of this version
1231
pending_components = version_ids
1232
while pending_components:
1233
build_details = self._index.get_build_details(pending_components)
1234
current_components = set(pending_components)
1235
pending_components = set()
1236
for version_id, details in build_details.iteritems():
1237
(index_memo, compression_parent, parents,
1238
record_details) = details
1239
method = record_details[0]
1240
if compression_parent is not None:
1241
pending_components.add(compression_parent)
1242
component_data[version_id] = (record_details, index_memo,
1244
missing = current_components.difference(build_details)
1246
raise errors.RevisionNotPresent(missing.pop(), self.filename)
1247
return component_data
1249
def _get_content(self, version_id, parent_texts={}):
1250
"""Returns a content object that makes up the specified
1252
cached_version = parent_texts.get(version_id, None)
1253
if cached_version is not None:
1254
if not self.has_version(version_id):
1255
raise RevisionNotPresent(version_id, self.filename)
1256
return cached_version
1258
text_map, contents_map = self._get_content_maps([version_id])
1259
return contents_map[version_id]
1261
def _check_versions_present(self, version_ids):
1262
"""Check that all specified versions are present."""
1263
self._index.check_versions_present(version_ids)
1265
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
1266
nostore_sha, random_id, check_content, left_matching_blocks):
1267
"""See VersionedFile.add_lines_with_ghosts()."""
1268
self._check_add(version_id, lines, random_id, check_content)
1269
return self._add(version_id, lines, parents, self.delta,
1270
parent_texts, left_matching_blocks, nostore_sha, random_id)
1272
def _add_lines(self, version_id, parents, lines, parent_texts,
1273
left_matching_blocks, nostore_sha, random_id, check_content):
1274
"""See VersionedFile.add_lines."""
1275
self._check_add(version_id, lines, random_id, check_content)
1276
self._check_versions_present(parents)
1277
return self._add(version_id, lines[:], parents, self.delta,
1278
parent_texts, left_matching_blocks, nostore_sha, random_id)
1280
def _check_add(self, version_id, lines, random_id, check_content):
1281
"""check that version_id and lines are safe to add."""
1282
if contains_whitespace(version_id):
1283
raise InvalidRevisionId(version_id, self.filename)
1284
self.check_not_reserved_id(version_id)
1285
# Technically this could be avoided if we are happy to allow duplicate
1286
# id insertion when other things than bzr core insert texts, but it
1287
# seems useful for folk using the knit api directly to have some safety
1288
# blanket that we can disable.
1289
if not random_id and self.has_version(version_id):
1290
raise RevisionAlreadyPresent(version_id, self.filename)
1292
self._check_lines_not_unicode(lines)
1293
self._check_lines_are_lines(lines)
1295
def _add(self, version_id, lines, parents, delta, parent_texts,
1296
left_matching_blocks, nostore_sha, random_id):
1297
"""Add a set of lines on top of version specified by parents.
1299
If delta is true, compress the text as a line-delta against
1302
Any versions not present will be converted into ghosts.
1304
# first thing, if the content is something we don't need to store, find
1306
line_bytes = ''.join(lines)
1307
digest = sha_string(line_bytes)
1308
if nostore_sha == digest:
1309
raise errors.ExistingContent
1311
present_parents = []
1312
if parent_texts is None:
1314
for parent in parents:
1315
if self.has_version(parent):
1316
present_parents.append(parent)
1318
# can only compress against the left most present parent.
1320
(len(present_parents) == 0 or
1321
present_parents[0] != parents[0])):
1324
text_length = len(line_bytes)
1327
if lines[-1][-1] != '\n':
1328
# copy the contents of lines.
1330
options.append('no-eol')
1331
lines[-1] = lines[-1] + '\n'
1335
# To speed the extract of texts the delta chain is limited
1336
# to a fixed number of deltas. This should minimize both
1337
# I/O and the time spend applying deltas.
1338
delta = self._check_should_delta(present_parents)
1340
content = self.factory.make(lines, version_id)
1341
if delta or (self.factory.annotated and len(present_parents) > 0):
1342
# Merge annotations from parent texts if needed.
1343
delta_hunks = self._merge_annotations(content, present_parents,
1344
parent_texts, delta, self.factory.annotated,
1345
left_matching_blocks)
1348
options.append('line-delta')
1349
store_lines = self.factory.lower_line_delta(delta_hunks)
1350
size, bytes = self._data._record_to_data(version_id, digest,
1353
options.append('fulltext')
1354
# isinstance is slower and we have no hierarchy.
1355
if self.factory.__class__ == KnitPlainFactory:
1356
# Use the already joined bytes saving iteration time in
1358
size, bytes = self._data._record_to_data(version_id, digest,
1359
lines, [line_bytes])
1361
# get mixed annotation + content and feed it into the
1363
store_lines = self.factory.lower_fulltext(content)
1364
size, bytes = self._data._record_to_data(version_id, digest,
1367
access_memo = self._data.add_raw_records([size], bytes)[0]
1368
self._index.add_versions(
1369
((version_id, options, access_memo, parents),),
1370
random_id=random_id)
1371
return digest, text_length, content
1373
def check(self, progress_bar=None):
1374
"""See VersionedFile.check()."""
1375
# This doesn't actually test extraction of everything, but that will
1376
# impact 'bzr check' substantially, and needs to be integrated with
1377
# care. However, it does check for the obvious problem of a delta with
1379
versions = self.versions()
1380
parent_map = self.get_parent_map(versions)
1381
for version in versions:
1382
if self._index.get_method(version) != 'fulltext':
1383
compression_parent = parent_map[version][0]
1384
if compression_parent not in parent_map:
1385
raise errors.KnitCorrupt(self,
1386
"Missing basis parent %s for %s" % (
1387
compression_parent, version))
1389
def get_lines(self, version_id):
1390
"""See VersionedFile.get_lines()."""
1391
return self.get_line_list([version_id])[0]
1393
def _get_record_map(self, version_ids):
1394
"""Produce a dictionary of knit records.
1396
:return: {version_id:(record, record_details, digest, next)}
1398
data returned from read_records
1400
opaque information to pass to parse_record
1402
SHA1 digest of the full text after all steps are done
1404
build-parent of the version, i.e. the leftmost ancestor.
1405
Will be None if the record is not a delta.
1407
position_map = self._get_components_positions(version_ids)
1408
# c = component_id, r = record_details, i_m = index_memo, n = next
1409
records = [(c, i_m) for c, (r, i_m, n)
1410
in position_map.iteritems()]
1412
for component_id, record, digest in \
1413
self._data.read_records_iter(records):
1414
(record_details, index_memo, next) = position_map[component_id]
1415
record_map[component_id] = record, record_details, digest, next
1419
def get_text(self, version_id):
1420
"""See VersionedFile.get_text"""
1421
return self.get_texts([version_id])[0]
1423
def get_texts(self, version_ids):
1424
return [''.join(l) for l in self.get_line_list(version_ids)]
1426
def get_line_list(self, version_ids):
1427
"""Return the texts of listed versions as a list of strings."""
1428
for version_id in version_ids:
1429
self.check_not_reserved_id(version_id)
1430
text_map, content_map = self._get_content_maps(version_ids)
1431
return [text_map[v] for v in version_ids]
1433
_get_lf_split_line_list = get_line_list
1435
def _get_content_maps(self, version_ids):
1436
"""Produce maps of text and KnitContents
1438
:return: (text_map, content_map) where text_map contains the texts for
1439
the requested versions and content_map contains the KnitContents.
1440
Both dicts take version_ids as their keys.
1442
# FUTURE: This function could be improved for the 'extract many' case
1443
# by tracking each component and only doing the copy when the number of
1444
# children than need to apply delta's to it is > 1 or it is part of the
1446
version_ids = list(version_ids)
1447
multiple_versions = len(version_ids) != 1
1448
record_map = self._get_record_map(version_ids)
1453
for version_id in version_ids:
1456
while cursor is not None:
1457
record, record_details, digest, next = record_map[cursor]
1458
components.append((cursor, record, record_details, digest))
1459
if cursor in content_map:
1464
for (component_id, record, record_details,
1465
digest) in reversed(components):
1466
if component_id in content_map:
1467
content = content_map[component_id]
1469
content, delta = self.factory.parse_record(version_id,
1470
record, record_details, content,
1471
copy_base_content=multiple_versions)
1472
if multiple_versions:
1473
content_map[component_id] = content
1475
content.cleanup_eol(copy_on_mutate=multiple_versions)
1476
final_content[version_id] = content
1478
# digest here is the digest from the last applied component.
1479
text = content.text()
1480
actual_sha = sha_strings(text)
1481
if actual_sha != digest:
1482
raise KnitCorrupt(self.filename,
1484
'\n of reconstructed text does not match'
1486
'\n for version %s' %
1487
(actual_sha, digest, version_id))
1488
text_map[version_id] = text
1489
return text_map, final_content
1491
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1493
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1494
if version_ids is None:
1495
version_ids = self.versions()
1497
pb = progress.DummyProgress()
1498
# we don't care about inclusions, the caller cares.
1499
# but we need to setup a list of records to visit.
1500
# we need version_id, position, length
1501
version_id_records = []
1502
requested_versions = set(version_ids)
1503
# filter for available versions
1504
for version_id in requested_versions:
1505
if not self.has_version(version_id):
1506
raise RevisionNotPresent(version_id, self.filename)
1507
# get a in-component-order queue:
1508
for version_id in self.versions():
1509
if version_id in requested_versions:
1510
index_memo = self._index.get_position(version_id)
1511
version_id_records.append((version_id, index_memo))
1513
total = len(version_id_records)
1514
for version_idx, (version_id, data, sha_value) in \
1515
enumerate(self._data.read_records_iter(version_id_records)):
1516
pb.update('Walking content.', version_idx, total)
1517
method = self._index.get_method(version_id)
1518
if method == 'fulltext':
1519
line_iterator = self.factory.get_fulltext_content(data)
1520
elif method == 'line-delta':
1521
line_iterator = self.factory.get_linedelta_content(data)
1523
raise ValueError('invalid method %r' % (method,))
1524
# XXX: It might be more efficient to yield (version_id,
1525
# line_iterator) in the future. However for now, this is a simpler
1526
# change to integrate into the rest of the codebase. RBC 20071110
1527
for line in line_iterator:
1528
yield line, version_id
1530
pb.update('Walking content.', total, total)
1532
def num_versions(self):
1533
"""See VersionedFile.num_versions()."""
1534
return self._index.num_versions()
1536
__len__ = num_versions
1538
def annotate(self, version_id):
1539
"""See VersionedFile.annotate."""
1540
return self.factory.annotate(self, version_id)
1542
def get_parent_map(self, version_ids):
1543
"""See VersionedFile.get_parent_map."""
1544
return self._index.get_parent_map(version_ids)
1546
def get_ancestry(self, versions, topo_sorted=True):
1547
"""See VersionedFile.get_ancestry."""
1548
if isinstance(versions, basestring):
1549
versions = [versions]
1552
return self._index.get_ancestry(versions, topo_sorted)
1554
def get_ancestry_with_ghosts(self, versions):
1555
"""See VersionedFile.get_ancestry_with_ghosts."""
1556
if isinstance(versions, basestring):
1557
versions = [versions]
1560
return self._index.get_ancestry_with_ghosts(versions)
1562
def plan_merge(self, ver_a, ver_b):
1563
"""See VersionedFile.plan_merge."""
1564
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1565
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1566
annotated_a = self.annotate(ver_a)
1567
annotated_b = self.annotate(ver_b)
1568
return merge._plan_annotate_merge(annotated_a, annotated_b,
1569
ancestors_a, ancestors_b)
1572
class _KnitComponentFile(object):
1573
"""One of the files used to implement a knit database"""
1575
def __init__(self, transport, filename, mode, file_mode=None,
1576
create_parent_dir=False, dir_mode=None):
1577
self._transport = transport
1578
self._filename = filename
1580
self._file_mode = file_mode
1581
self._dir_mode = dir_mode
1582
self._create_parent_dir = create_parent_dir
1583
self._need_to_create = False
1585
def _full_path(self):
1586
"""Return the full path to this file."""
1587
return self._transport.base + self._filename
1589
def check_header(self, fp):
1590
line = fp.readline()
1592
# An empty file can actually be treated as though the file doesn't
1594
raise errors.NoSuchFile(self._full_path())
1595
if line != self.HEADER:
1596
raise KnitHeaderError(badline=line,
1597
filename=self._transport.abspath(self._filename))
1600
return '%s(%s)' % (self.__class__.__name__, self._filename)
1603
class _KnitIndex(_KnitComponentFile):
1604
"""Manages knit index file.
1606
The index is already kept in memory and read on startup, to enable
1607
fast lookups of revision information. The cursor of the index
1608
file is always pointing to the end, making it easy to append
1611
_cache is a cache for fast mapping from version id to a Index
1614
_history is a cache for fast mapping from indexes to version ids.
1616
The index data format is dictionary compressed when it comes to
1617
parent references; a index entry may only have parents that with a
1618
lover index number. As a result, the index is topological sorted.
1620
Duplicate entries may be written to the index for a single version id
1621
if this is done then the latter one completely replaces the former:
1622
this allows updates to correct version and parent information.
1623
Note that the two entries may share the delta, and that successive
1624
annotations and references MUST point to the first entry.
1626
The index file on disc contains a header, followed by one line per knit
1627
record. The same revision can be present in an index file more than once.
1628
The first occurrence gets assigned a sequence number starting from 0.
1630
The format of a single line is
1631
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1632
REVISION_ID is a utf8-encoded revision id
1633
FLAGS is a comma separated list of flags about the record. Values include
1634
no-eol, line-delta, fulltext.
1635
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1636
that the the compressed data starts at.
1637
LENGTH is the ascii representation of the length of the data file.
1638
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1640
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1641
revision id already in the knit that is a parent of REVISION_ID.
1642
The ' :' marker is the end of record marker.
1645
when a write is interrupted to the index file, it will result in a line
1646
that does not end in ' :'. If the ' :' is not present at the end of a line,
1647
or at the end of the file, then the record that is missing it will be
1648
ignored by the parser.
1650
When writing new records to the index file, the data is preceded by '\n'
1651
to ensure that records always start on new lines even if the last write was
1652
interrupted. As a result its normal for the last line in the index to be
1653
missing a trailing newline. One can be added with no harmful effects.
1656
HEADER = "# bzr knit index 8\n"
1658
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1659
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1661
def _cache_version(self, version_id, options, pos, size, parents):
1662
"""Cache a version record in the history array and index cache.
1664
This is inlined into _load_data for performance. KEEP IN SYNC.
1665
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1668
# only want the _history index to reference the 1st index entry
1670
if version_id not in self._cache:
1671
index = len(self._history)
1672
self._history.append(version_id)
1674
index = self._cache[version_id][5]
1675
self._cache[version_id] = (version_id,
1682
def _check_write_ok(self):
1683
if self._get_scope() != self._scope:
1684
raise errors.OutSideTransaction()
1685
if self._mode != 'w':
1686
raise errors.ReadOnlyObjectDirtiedError(self)
1688
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1689
create_parent_dir=False, delay_create=False, dir_mode=None,
1691
_KnitComponentFile.__init__(self, transport, filename, mode,
1692
file_mode=file_mode,
1693
create_parent_dir=create_parent_dir,
1696
# position in _history is the 'official' index for a revision
1697
# but the values may have come from a newer entry.
1698
# so - wc -l of a knit index is != the number of unique names
1702
fp = self._transport.get(self._filename)
1704
# _load_data may raise NoSuchFile if the target knit is
1706
_load_data(self, fp)
1710
if mode != 'w' or not create:
1713
self._need_to_create = True
1715
self._transport.put_bytes_non_atomic(
1716
self._filename, self.HEADER, mode=self._file_mode)
1717
self._scope = get_scope()
1718
self._get_scope = get_scope
1720
def get_ancestry(self, versions, topo_sorted=True):
1721
"""See VersionedFile.get_ancestry."""
1722
# get a graph of all the mentioned versions:
1724
pending = set(versions)
1727
version = pending.pop()
1730
parents = [p for p in cache[version][4] if p in cache]
1732
raise RevisionNotPresent(version, self._filename)
1733
# if not completed and not a ghost
1734
pending.update([p for p in parents if p not in graph])
1735
graph[version] = parents
1738
return topo_sort(graph.items())
1740
def get_ancestry_with_ghosts(self, versions):
1741
"""See VersionedFile.get_ancestry_with_ghosts."""
1742
# get a graph of all the mentioned versions:
1743
self.check_versions_present(versions)
1746
pending = set(versions)
1748
version = pending.pop()
1750
parents = cache[version][4]
1756
pending.update([p for p in parents if p not in graph])
1757
graph[version] = parents
1758
return topo_sort(graph.items())
1760
def get_build_details(self, version_ids):
1761
"""Get the method, index_memo and compression parent for version_ids.
1763
Ghosts are omitted from the result.
1765
:param version_ids: An iterable of version_ids.
1766
:return: A dict of version_id:(index_memo, compression_parent,
1767
parents, record_details).
1769
opaque structure to pass to read_records to extract the raw
1772
Content that this record is built upon, may be None
1774
Logical parents of this node
1776
extra information about the content which needs to be passed to
1777
Factory.parse_record
1780
for version_id in version_ids:
1781
if version_id not in self._cache:
1782
# ghosts are omitted
1784
method = self.get_method(version_id)
1785
parents = self.get_parents_with_ghosts(version_id)
1786
if method == 'fulltext':
1787
compression_parent = None
1789
compression_parent = parents[0]
1790
noeol = 'no-eol' in self.get_options(version_id)
1791
index_memo = self.get_position(version_id)
1792
result[version_id] = (index_memo, compression_parent,
1793
parents, (method, noeol))
1796
def num_versions(self):
1797
return len(self._history)
1799
__len__ = num_versions
1801
def get_versions(self):
1802
"""Get all the versions in the file. not topologically sorted."""
1803
return self._history
1805
def _version_list_to_index(self, versions):
1808
for version in versions:
1809
if version in cache:
1810
# -- inlined lookup() --
1811
result_list.append(str(cache[version][5]))
1812
# -- end lookup () --
1814
result_list.append('.' + version)
1815
return ' '.join(result_list)
1817
def add_version(self, version_id, options, index_memo, parents):
1818
"""Add a version record to the index."""
1819
self.add_versions(((version_id, options, index_memo, parents),))
1821
def add_versions(self, versions, random_id=False):
1822
"""Add multiple versions to the index.
1824
:param versions: a list of tuples:
1825
(version_id, options, pos, size, parents).
1826
:param random_id: If True the ids being added were randomly generated
1827
and no check for existence will be performed.
1830
orig_history = self._history[:]
1831
orig_cache = self._cache.copy()
1834
for version_id, options, (index, pos, size), parents in versions:
1835
line = "\n%s %s %s %s %s :" % (version_id,
1839
self._version_list_to_index(parents))
1841
self._cache_version(version_id, options, pos, size, tuple(parents))
1842
if not self._need_to_create:
1843
self._transport.append_bytes(self._filename, ''.join(lines))
1846
sio.write(self.HEADER)
1847
sio.writelines(lines)
1849
self._transport.put_file_non_atomic(self._filename, sio,
1850
create_parent_dir=self._create_parent_dir,
1851
mode=self._file_mode,
1852
dir_mode=self._dir_mode)
1853
self._need_to_create = False
1855
# If any problems happen, restore the original values and re-raise
1856
self._history = orig_history
1857
self._cache = orig_cache
1860
def has_version(self, version_id):
1861
"""True if the version is in the index."""
1862
return version_id in self._cache
1864
def get_position(self, version_id):
1865
"""Return details needed to access the version.
1867
.kndx indices do not support split-out data, so return None for the
1870
:return: a tuple (None, data position, size) to hand to the access
1871
logic to get the record.
1873
entry = self._cache[version_id]
1874
return None, entry[2], entry[3]
1876
def get_method(self, version_id):
1877
"""Return compression method of specified version."""
1879
options = self._cache[version_id][1]
1881
raise RevisionNotPresent(version_id, self._filename)
1882
if 'fulltext' in options:
1885
if 'line-delta' not in options:
1886
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1889
def get_options(self, version_id):
1890
"""Return a list representing options.
1894
return self._cache[version_id][1]
1896
def get_parent_map(self, version_ids):
1897
"""Passed through to by KnitVersionedFile.get_parent_map."""
1899
for version_id in version_ids:
1901
result[version_id] = tuple(self._cache[version_id][4])
1906
def get_parents_with_ghosts(self, version_id):
1907
"""Return parents of specified version with ghosts."""
1909
return self.get_parent_map([version_id])[version_id]
1911
raise RevisionNotPresent(version_id, self)
1913
def check_versions_present(self, version_ids):
1914
"""Check that all specified versions are present."""
1916
for version_id in version_ids:
1917
if version_id not in cache:
1918
raise RevisionNotPresent(version_id, self._filename)
1921
class KnitGraphIndex(object):
1922
"""A knit index that builds on GraphIndex."""
1924
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1925
"""Construct a KnitGraphIndex on a graph_index.
1927
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1928
:param deltas: Allow delta-compressed records.
1929
:param add_callback: If not None, allow additions to the index and call
1930
this callback with a list of added GraphIndex nodes:
1931
[(node, value, node_refs), ...]
1932
:param parents: If True, record knits parents, if not do not record
1935
self._graph_index = graph_index
1936
self._deltas = deltas
1937
self._add_callback = add_callback
1938
self._parents = parents
1939
if deltas and not parents:
1940
raise KnitCorrupt(self, "Cannot do delta compression without "
1943
def _check_write_ok(self):
1946
def _get_entries(self, keys, check_present=False):
1947
"""Get the entries for keys.
1949
:param keys: An iterable of index keys, - 1-tuples.
1954
for node in self._graph_index.iter_entries(keys):
1956
found_keys.add(node[1])
1958
# adapt parentless index to the rest of the code.
1959
for node in self._graph_index.iter_entries(keys):
1960
yield node[0], node[1], node[2], ()
1961
found_keys.add(node[1])
1963
missing_keys = keys.difference(found_keys)
1965
raise RevisionNotPresent(missing_keys.pop(), self)
1967
def _present_keys(self, version_ids):
1969
node[1] for node in self._get_entries(version_ids)])
1971
def _parentless_ancestry(self, versions):
1972
"""Honour the get_ancestry API for parentless knit indices."""
1973
wanted_keys = self._version_ids_to_keys(versions)
1974
present_keys = self._present_keys(wanted_keys)
1975
missing = set(wanted_keys).difference(present_keys)
1977
raise RevisionNotPresent(missing.pop(), self)
1978
return list(self._keys_to_version_ids(present_keys))
1980
def get_ancestry(self, versions, topo_sorted=True):
1981
"""See VersionedFile.get_ancestry."""
1982
if not self._parents:
1983
return self._parentless_ancestry(versions)
1984
# XXX: This will do len(history) index calls - perhaps
1985
# it should be altered to be a index core feature?
1986
# get a graph of all the mentioned versions:
1989
versions = self._version_ids_to_keys(versions)
1990
pending = set(versions)
1992
# get all pending nodes
1993
this_iteration = pending
1994
new_nodes = self._get_entries(this_iteration)
1997
for (index, key, value, node_refs) in new_nodes:
1998
# dont ask for ghosties - otherwise
1999
# we we can end up looping with pending
2000
# being entirely ghosted.
2001
graph[key] = [parent for parent in node_refs[0]
2002
if parent not in ghosts]
2004
for parent in graph[key]:
2005
# dont examine known nodes again
2010
ghosts.update(this_iteration.difference(found))
2011
if versions.difference(graph):
2012
raise RevisionNotPresent(versions.difference(graph).pop(), self)
2014
result_keys = topo_sort(graph.items())
2016
result_keys = graph.iterkeys()
2017
return [key[0] for key in result_keys]
2019
def get_ancestry_with_ghosts(self, versions):
2020
"""See VersionedFile.get_ancestry."""
2021
if not self._parents:
2022
return self._parentless_ancestry(versions)
2023
# XXX: This will do len(history) index calls - perhaps
2024
# it should be altered to be a index core feature?
2025
# get a graph of all the mentioned versions:
2027
versions = self._version_ids_to_keys(versions)
2028
pending = set(versions)
2030
# get all pending nodes
2031
this_iteration = pending
2032
new_nodes = self._get_entries(this_iteration)
2034
for (index, key, value, node_refs) in new_nodes:
2035
graph[key] = node_refs[0]
2037
for parent in graph[key]:
2038
# dont examine known nodes again
2042
missing_versions = this_iteration.difference(graph)
2043
missing_needed = versions.intersection(missing_versions)
2045
raise RevisionNotPresent(missing_needed.pop(), self)
2046
for missing_version in missing_versions:
2047
# add a key, no parents
2048
graph[missing_version] = []
2049
pending.discard(missing_version) # don't look for it
2050
result_keys = topo_sort(graph.items())
2051
return [key[0] for key in result_keys]
2053
def get_build_details(self, version_ids):
2054
"""Get the method, index_memo and compression parent for version_ids.
2056
Ghosts are omitted from the result.
2058
:param version_ids: An iterable of version_ids.
2059
:return: A dict of version_id:(index_memo, compression_parent,
2060
parents, record_details).
2062
opaque structure to pass to read_records to extract the raw
2065
Content that this record is built upon, may be None
2067
Logical parents of this node
2069
extra information about the content which needs to be passed to
2070
Factory.parse_record
2073
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
2074
for entry in entries:
2075
version_id = self._keys_to_version_ids((entry[1],))[0]
2076
if not self._parents:
2079
parents = self._keys_to_version_ids(entry[3][0])
2080
if not self._deltas:
2081
compression_parent = None
2083
compression_parent_key = self._compression_parent(entry)
2084
if compression_parent_key:
2085
compression_parent = self._keys_to_version_ids(
2086
(compression_parent_key,))[0]
2088
compression_parent = None
2089
noeol = (entry[2][0] == 'N')
2090
if compression_parent:
2091
method = 'line-delta'
2094
result[version_id] = (self._node_to_position(entry),
2095
compression_parent, parents,
2099
def _compression_parent(self, an_entry):
2100
# return the key that an_entry is compressed against, or None
2101
# Grab the second parent list (as deltas implies parents currently)
2102
compression_parents = an_entry[3][1]
2103
if not compression_parents:
2105
return compression_parents[0]
2107
def _get_method(self, node):
2108
if not self._deltas:
2110
if self._compression_parent(node):
2115
def num_versions(self):
2116
return len(list(self._graph_index.iter_all_entries()))
2118
__len__ = num_versions
2120
def get_versions(self):
2121
"""Get all the versions in the file. not topologically sorted."""
2122
return [node[1][0] for node in self._graph_index.iter_all_entries()]
2124
def has_version(self, version_id):
2125
"""True if the version is in the index."""
2126
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
2128
def _keys_to_version_ids(self, keys):
2129
return tuple(key[0] for key in keys)
2131
def get_position(self, version_id):
2132
"""Return details needed to access the version.
2134
:return: a tuple (index, data position, size) to hand to the access
2135
logic to get the record.
2137
node = self._get_node(version_id)
2138
return self._node_to_position(node)
2140
def _node_to_position(self, node):
2141
"""Convert an index value to position details."""
2142
bits = node[2][1:].split(' ')
2143
return node[0], int(bits[0]), int(bits[1])
2145
def get_method(self, version_id):
2146
"""Return compression method of specified version."""
2147
return self._get_method(self._get_node(version_id))
2149
def _get_node(self, version_id):
2151
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
2153
raise RevisionNotPresent(version_id, self)
2155
def get_options(self, version_id):
2156
"""Return a list representing options.
2160
node = self._get_node(version_id)
2161
options = [self._get_method(node)]
2162
if node[2][0] == 'N':
2163
options.append('no-eol')
2166
def get_parent_map(self, version_ids):
2167
"""Passed through to by KnitVersionedFile.get_parent_map."""
2168
nodes = self._get_entries(self._version_ids_to_keys(version_ids))
2172
result[node[1][0]] = self._keys_to_version_ids(node[3][0])
2175
result[node[1][0]] = ()
2178
def get_parents_with_ghosts(self, version_id):
2179
"""Return parents of specified version with ghosts."""
2181
return self.get_parent_map([version_id])[version_id]
2183
raise RevisionNotPresent(version_id, self)
2185
def check_versions_present(self, version_ids):
2186
"""Check that all specified versions are present."""
2187
keys = self._version_ids_to_keys(version_ids)
2188
present = self._present_keys(keys)
2189
missing = keys.difference(present)
2191
raise RevisionNotPresent(missing.pop(), self)
2193
def add_version(self, version_id, options, access_memo, parents):
2194
"""Add a version record to the index."""
2195
return self.add_versions(((version_id, options, access_memo, parents),))
2197
def add_versions(self, versions, random_id=False):
2198
"""Add multiple versions to the index.
2200
This function does not insert data into the Immutable GraphIndex
2201
backing the KnitGraphIndex, instead it prepares data for insertion by
2202
the caller and checks that it is safe to insert then calls
2203
self._add_callback with the prepared GraphIndex nodes.
2205
:param versions: a list of tuples:
2206
(version_id, options, pos, size, parents).
2207
:param random_id: If True the ids being added were randomly generated
2208
and no check for existence will be performed.
2210
if not self._add_callback:
2211
raise errors.ReadOnlyError(self)
2212
# we hope there are no repositories with inconsistent parentage
2217
for (version_id, options, access_memo, parents) in versions:
2218
index, pos, size = access_memo
2219
key = (version_id, )
2220
parents = tuple((parent, ) for parent in parents)
2221
if 'no-eol' in options:
2225
value += "%d %d" % (pos, size)
2226
if not self._deltas:
2227
if 'line-delta' in options:
2228
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2231
if 'line-delta' in options:
2232
node_refs = (parents, (parents[0],))
2234
node_refs = (parents, ())
2236
node_refs = (parents, )
2239
raise KnitCorrupt(self, "attempt to add node with parents "
2240
"in parentless index.")
2242
keys[key] = (value, node_refs)
2244
present_nodes = self._get_entries(keys)
2245
for (index, key, value, node_refs) in present_nodes:
2246
if (value, node_refs) != keys[key]:
2247
raise KnitCorrupt(self, "inconsistent details in add_versions"
2248
": %s %s" % ((value, node_refs), keys[key]))
2252
for key, (value, node_refs) in keys.iteritems():
2253
result.append((key, value, node_refs))
2255
for key, (value, node_refs) in keys.iteritems():
2256
result.append((key, value))
2257
self._add_callback(result)
2259
def _version_ids_to_keys(self, version_ids):
2260
return set((version_id, ) for version_id in version_ids)
2263
class _KnitAccess(object):
2264
"""Access to knit records in a .knit file."""
2266
def __init__(self, transport, filename, _file_mode, _dir_mode,
2267
_need_to_create, _create_parent_dir):
2268
"""Create a _KnitAccess for accessing and inserting data.
2270
:param transport: The transport the .knit is located on.
2271
:param filename: The filename of the .knit.
2273
self._transport = transport
2274
self._filename = filename
2275
self._file_mode = _file_mode
2276
self._dir_mode = _dir_mode
2277
self._need_to_create = _need_to_create
2278
self._create_parent_dir = _create_parent_dir
2282
return "%s(%r)" % (self.__class__.__name__,
2283
self._transport.abspath(self._filename))
2285
return "_KnitAccess(**unprintable**)"
2287
def add_raw_records(self, sizes, raw_data):
2288
"""Add raw knit bytes to a storage area.
2290
The data is spooled to whereever the access method is storing data.
2292
:param sizes: An iterable containing the size of each raw data segment.
2293
:param raw_data: A bytestring containing the data.
2294
:return: A list of memos to retrieve the record later. Each memo is a
2295
tuple - (index, pos, length), where the index field is always None
2296
for the .knit access method.
2298
if not self._need_to_create:
2299
base = self._transport.append_bytes(self._filename, raw_data)
2301
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2302
create_parent_dir=self._create_parent_dir,
2303
mode=self._file_mode,
2304
dir_mode=self._dir_mode)
2305
self._need_to_create = False
2309
result.append((None, base, size))
2314
"""IFF this data access has its own storage area, initialise it.
2318
self._transport.put_bytes_non_atomic(self._filename, '',
2319
mode=self._file_mode)
2321
def open_file(self):
2322
"""IFF this data access can be represented as a single file, open it.
2324
For knits that are not mapped to a single file on disk this will
2327
:return: None or a file handle.
2330
return self._transport.get(self._filename)
2335
def get_raw_records(self, memos_for_retrieval):
2336
"""Get the raw bytes for a records.
2338
:param memos_for_retrieval: An iterable containing the (index, pos,
2339
length) memo for retrieving the bytes. The .knit method ignores
2340
the index as there is always only a single file.
2341
:return: An iterator over the bytes of the records.
2343
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2344
for pos, data in self._transport.readv(self._filename, read_vector):
2348
class _PackAccess(object):
2349
"""Access to knit records via a collection of packs."""
2351
def __init__(self, index_to_packs, writer=None):
2352
"""Create a _PackAccess object.
2354
:param index_to_packs: A dict mapping index objects to the transport
2355
and file names for obtaining data.
2356
:param writer: A tuple (pack.ContainerWriter, write_index) which
2357
contains the pack to write, and the index that reads from it will
2361
self.container_writer = writer[0]
2362
self.write_index = writer[1]
2364
self.container_writer = None
2365
self.write_index = None
2366
self.indices = index_to_packs
2368
def add_raw_records(self, sizes, raw_data):
2369
"""Add raw knit bytes to a storage area.
2371
The data is spooled to the container writer in one bytes-record per
2374
:param sizes: An iterable containing the size of each raw data segment.
2375
:param raw_data: A bytestring containing the data.
2376
:return: A list of memos to retrieve the record later. Each memo is a
2377
tuple - (index, pos, length), where the index field is the
2378
write_index object supplied to the PackAccess object.
2383
p_offset, p_length = self.container_writer.add_bytes_record(
2384
raw_data[offset:offset+size], [])
2386
result.append((self.write_index, p_offset, p_length))
2390
"""Pack based knits do not get individually created."""
2392
def get_raw_records(self, memos_for_retrieval):
2393
"""Get the raw bytes for a records.
2395
:param memos_for_retrieval: An iterable containing the (index, pos,
2396
length) memo for retrieving the bytes. The Pack access method
2397
looks up the pack to use for a given record in its index_to_pack
2399
:return: An iterator over the bytes of the records.
2401
# first pass, group into same-index requests
2403
current_index = None
2404
for (index, offset, length) in memos_for_retrieval:
2405
if current_index == index:
2406
current_list.append((offset, length))
2408
if current_index is not None:
2409
request_lists.append((current_index, current_list))
2410
current_index = index
2411
current_list = [(offset, length)]
2412
# handle the last entry
2413
if current_index is not None:
2414
request_lists.append((current_index, current_list))
2415
for index, offsets in request_lists:
2416
transport, path = self.indices[index]
2417
reader = pack.make_readv_reader(transport, path, offsets)
2418
for names, read_func in reader.iter_records():
2419
yield read_func(None)
2421
def open_file(self):
2422
"""Pack based knits have no single file."""
2425
def set_writer(self, writer, index, (transport, packname)):
2426
"""Set a writer to use for adding data."""
2427
if index is not None:
2428
self.indices[index] = (transport, packname)
2429
self.container_writer = writer
2430
self.write_index = index
2433
class _StreamAccess(object):
2434
"""A Knit Access object that provides data from a datastream.
2436
It also provides a fallback to present as unannotated data, annotated data
2437
from a *backing* access object.
2439
This is triggered by a index_memo which is pointing to a different index
2440
than this was constructed with, and is used to allow extracting full
2441
unannotated texts for insertion into annotated knits.
2444
def __init__(self, reader_callable, stream_index, backing_knit,
2446
"""Create a _StreamAccess object.
2448
:param reader_callable: The reader_callable from the datastream.
2449
This is called to buffer all the data immediately, for
2451
:param stream_index: The index the data stream this provides access to
2452
which will be present in native index_memo's.
2453
:param backing_knit: The knit object that will provide access to
2454
annotated texts which are not available in the stream, so as to
2455
create unannotated texts.
2456
:param orig_factory: The original content factory used to generate the
2457
stream. This is used for checking whether the thunk code for
2458
supporting _copy_texts will generate the correct form of data.
2460
self.data = reader_callable(None)
2461
self.stream_index = stream_index
2462
self.backing_knit = backing_knit
2463
self.orig_factory = orig_factory
2465
def get_raw_records(self, memos_for_retrieval):
2466
"""Get the raw bytes for a records.
2468
:param memos_for_retrieval: An iterable of memos from the
2469
_StreamIndex object identifying bytes to read; for these classes
2470
they are (from_backing_knit, index, start, end) and can point to
2471
either the backing knit or streamed data.
2472
:return: An iterator yielding a byte string for each record in
2473
memos_for_retrieval.
2475
# use a generator for memory friendliness
2476
for from_backing_knit, version_id, start, end in memos_for_retrieval:
2477
if not from_backing_knit:
2478
if version_id is not self.stream_index:
2479
raise AssertionError()
2480
yield self.data[start:end]
2482
# we have been asked to thunk. This thunking only occurs when
2483
# we are obtaining plain texts from an annotated backing knit
2484
# so that _copy_texts will work.
2485
# We could improve performance here by scanning for where we need
2486
# to do this and using get_line_list, then interleaving the output
2487
# as desired. However, for now, this is sufficient.
2488
if self.orig_factory.__class__ != KnitPlainFactory:
2489
raise errors.KnitCorrupt(
2490
self, 'Bad thunk request %r cannot be backed by %r' %
2491
(version_id, self.orig_factory))
2492
lines = self.backing_knit.get_lines(version_id)
2493
line_bytes = ''.join(lines)
2494
digest = sha_string(line_bytes)
2495
# the packed form of the fulltext always has a trailing newline,
2496
# even if the actual text does not, unless the file is empty. the
2497
# record options including the noeol flag are passed through by
2498
# _StreamIndex, so this is safe.
2500
if lines[-1][-1] != '\n':
2501
lines[-1] = lines[-1] + '\n'
2503
# We want plain data, because we expect to thunk only to allow text
2505
size, bytes = self.backing_knit._data._record_to_data(version_id,
2506
digest, lines, line_bytes)
2510
class _StreamIndex(object):
2511
"""A Knit Index object that uses the data map from a datastream."""
2513
def __init__(self, data_list, backing_index):
2514
"""Create a _StreamIndex object.
2516
:param data_list: The data_list from the datastream.
2517
:param backing_index: The index which will supply values for nodes
2518
referenced outside of this stream.
2520
self.data_list = data_list
2521
self.backing_index = backing_index
2522
self._by_version = {}
2524
for key, options, length, parents in data_list:
2525
self._by_version[key] = options, (pos, pos + length), parents
2528
def get_ancestry(self, versions, topo_sorted):
2529
"""Get an ancestry list for versions."""
2531
# Not needed for basic joins
2532
raise NotImplementedError(self.get_ancestry)
2533
# get a graph of all the mentioned versions:
2534
# Little ugly - basically copied from KnitIndex, but don't want to
2535
# accidentally incorporate too much of that index's code.
2537
pending = set(versions)
2538
cache = self._by_version
2540
version = pending.pop()
2543
parents = [p for p in cache[version][2] if p in cache]
2545
raise RevisionNotPresent(version, self)
2546
# if not completed and not a ghost
2547
pending.update([p for p in parents if p not in ancestry])
2548
ancestry.add(version)
2549
return list(ancestry)
2551
def get_build_details(self, version_ids):
2552
"""Get the method, index_memo and compression parent for version_ids.
2554
Ghosts are omitted from the result.
2556
:param version_ids: An iterable of version_ids.
2557
:return: A dict of version_id:(index_memo, compression_parent,
2558
parents, record_details).
2560
opaque memo that can be passed to _StreamAccess.read_records
2561
to extract the raw data; for these classes it is
2562
(from_backing_knit, index, start, end)
2564
Content that this record is built upon, may be None
2566
Logical parents of this node
2568
extra information about the content which needs to be passed to
2569
Factory.parse_record
2572
for version_id in version_ids:
2574
method = self.get_method(version_id)
2575
except errors.RevisionNotPresent:
2576
# ghosts are omitted
2578
parent_ids = self.get_parents_with_ghosts(version_id)
2579
noeol = ('no-eol' in self.get_options(version_id))
2580
index_memo = self.get_position(version_id)
2581
from_backing_knit = index_memo[0]
2582
if from_backing_knit:
2583
# texts retrieved from the backing knit are always full texts
2585
if method == 'fulltext':
2586
compression_parent = None
2588
compression_parent = parent_ids[0]
2589
result[version_id] = (index_memo, compression_parent,
2590
parent_ids, (method, noeol))
2593
def get_method(self, version_id):
2594
"""Return compression method of specified version."""
2595
options = self.get_options(version_id)
2596
if 'fulltext' in options:
2598
elif 'line-delta' in options:
2601
raise errors.KnitIndexUnknownMethod(self, options)
2603
def get_options(self, version_id):
2604
"""Return a list representing options.
2609
return self._by_version[version_id][0]
2611
options = list(self.backing_index.get_options(version_id))
2612
if 'fulltext' in options:
2614
elif 'line-delta' in options:
2615
# Texts from the backing knit are always returned from the stream
2617
options.remove('line-delta')
2618
options.append('fulltext')
2620
raise errors.KnitIndexUnknownMethod(self, options)
2621
return tuple(options)
2623
def get_parent_map(self, version_ids):
2624
"""Passed through to by KnitVersionedFile.get_parent_map."""
2627
for version_id in version_ids:
2629
result[version_id] = self._by_version[version_id][2]
2631
pending_ids.add(version_id)
2632
result.update(self.backing_index.get_parent_map(pending_ids))
2635
def get_parents_with_ghosts(self, version_id):
2636
"""Return parents of specified version with ghosts."""
2638
return self.get_parent_map([version_id])[version_id]
2640
raise RevisionNotPresent(version_id, self)
2642
def get_position(self, version_id):
2643
"""Return details needed to access the version.
2645
_StreamAccess has the data as a big array, so we return slice
2646
coordinates into that (as index_memo's are opaque outside the
2647
index and matching access class).
2649
:return: a tuple (from_backing_knit, index, start, end) that can
2650
be passed e.g. to get_raw_records.
2651
If from_backing_knit is False, index will be self, otherwise it
2652
will be a version id.
2655
start, end = self._by_version[version_id][1]
2656
return False, self, start, end
2658
# Signal to the access object to handle this from the backing knit.
2659
return (True, version_id, None, None)
2661
def get_versions(self):
2662
"""Get all the versions in the stream."""
2663
return self._by_version.keys()
2666
class _KnitData(object):
2667
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2669
The KnitData class provides the logic for parsing and using knit records,
2670
making use of an access method for the low level read and write operations.
2673
def __init__(self, access):
2674
"""Create a KnitData object.
2676
:param access: The access method to use. Access methods such as
2677
_KnitAccess manage the insertion of raw records and the subsequent
2678
retrieval of the same.
2680
self._access = access
2681
self._checked = False
2683
def _open_file(self):
2684
return self._access.open_file()
2686
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2687
"""Convert version_id, digest, lines into a raw data block.
2689
:param dense_lines: The bytes of lines but in a denser form. For
2690
instance, if lines is a list of 1000 bytestrings each ending in \n,
2691
dense_lines may be a list with one line in it, containing all the
2692
1000's lines and their \n's. Using dense_lines if it is already
2693
known is a win because the string join to create bytes in this
2694
function spends less time resizing the final string.
2695
:return: (len, a StringIO instance with the raw data ready to read.)
2697
# Note: using a string copy here increases memory pressure with e.g.
2698
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2699
# when doing the initial commit of a mozilla tree. RBC 20070921
2700
bytes = ''.join(chain(
2701
["version %s %d %s\n" % (version_id,
2704
dense_lines or lines,
2705
["end %s\n" % version_id]))
2706
compressed_bytes = bytes_to_gzip(bytes)
2707
return len(compressed_bytes), compressed_bytes
2709
def add_raw_records(self, sizes, raw_data):
2710
"""Append a prepared record to the data file.
2712
:param sizes: An iterable containing the size of each raw data segment.
2713
:param raw_data: A bytestring containing the data.
2714
:return: a list of index data for the way the data was stored.
2715
See the access method add_raw_records documentation for more
2718
return self._access.add_raw_records(sizes, raw_data)
2720
def _parse_record_header(self, version_id, raw_data):
2721
"""Parse a record header for consistency.
2723
:return: the header and the decompressor stream.
2724
as (stream, header_record)
2726
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2728
rec = self._check_header(version_id, df.readline())
2729
except Exception, e:
2730
raise KnitCorrupt(self._access,
2731
"While reading {%s} got %s(%s)"
2732
% (version_id, e.__class__.__name__, str(e)))
2735
def _split_header(self, line):
2738
raise KnitCorrupt(self._access,
2739
'unexpected number of elements in record header')
2742
def _check_header_version(self, rec, version_id):
2743
if rec[1] != version_id:
2744
raise KnitCorrupt(self._access,
2745
'unexpected version, wanted %r, got %r'
2746
% (version_id, rec[1]))
2748
def _check_header(self, version_id, line):
2749
rec = self._split_header(line)
2750
self._check_header_version(rec, version_id)
2753
def _parse_record_unchecked(self, data):
2755
# 4168 calls in 2880 217 internal
2756
# 4168 calls to _parse_record_header in 2121
2757
# 4168 calls to readlines in 330
2758
df = GzipFile(mode='rb', fileobj=StringIO(data))
2760
record_contents = df.readlines()
2761
except Exception, e:
2762
raise KnitCorrupt(self._access, "Corrupt compressed record %r, got %s(%s)" %
2763
(data, e.__class__.__name__, str(e)))
2764
header = record_contents.pop(0)
2765
rec = self._split_header(header)
2766
last_line = record_contents.pop()
2767
if len(record_contents) != int(rec[2]):
2768
raise KnitCorrupt(self._access,
2769
'incorrect number of lines %s != %s'
2771
% (len(record_contents), int(rec[2]),
2773
if last_line != 'end %s\n' % rec[1]:
2774
raise KnitCorrupt(self._access,
2775
'unexpected version end line %r, wanted %r'
2776
% (last_line, rec[1]))
2778
return rec, record_contents
2780
def _parse_record(self, version_id, data):
2781
rec, record_contents = self._parse_record_unchecked(data)
2782
self._check_header_version(rec, version_id)
2783
return record_contents, rec[3]
2785
def read_records_iter_raw(self, records):
2786
"""Read text records from data file and yield raw data.
2788
This unpacks enough of the text record to validate the id is
2789
as expected but thats all.
2791
Each item the iterator yields is (version_id, bytes,
2794
# setup an iterator of the external records:
2795
# uses readv so nice and fast we hope.
2797
# grab the disk data needed.
2798
needed_offsets = [index_memo for version_id, index_memo
2800
raw_records = self._access.get_raw_records(needed_offsets)
2802
for version_id, index_memo in records:
2803
data = raw_records.next()
2804
# validate the header
2805
df, rec = self._parse_record_header(version_id, data)
2807
yield version_id, data, rec[3]
2809
def read_records_iter(self, records):
2810
"""Read text records from data file and yield result.
2812
The result will be returned in whatever is the fastest to read.
2813
Not by the order requested. Also, multiple requests for the same
2814
record will only yield 1 response.
2815
:param records: A list of (version_id, pos, len) entries
2816
:return: Yields (version_id, contents, digest) in the order
2817
read, not the order requested
2822
needed_records = sorted(set(records), key=operator.itemgetter(1))
2823
if not needed_records:
2826
# The transport optimizes the fetching as well
2827
# (ie, reads continuous ranges.)
2828
raw_data = self._access.get_raw_records(
2829
[index_memo for version_id, index_memo in needed_records])
2831
for (version_id, index_memo), data in \
2832
izip(iter(needed_records), raw_data):
2833
content, digest = self._parse_record(version_id, data)
2834
yield version_id, content, digest
2836
def read_records(self, records):
2837
"""Read records into a dictionary."""
2839
for record_id, content, digest in \
2840
self.read_records_iter(records):
2841
components[record_id] = (content, digest)
2845
class InterKnit(InterVersionedFile):
2846
"""Optimised code paths for knit to knit operations."""
2848
_matching_file_from_factory = staticmethod(make_file_knit)
2849
_matching_file_to_factory = staticmethod(make_file_knit)
2852
def is_compatible(source, target):
2853
"""Be compatible with knits. """
2855
return (isinstance(source, KnitVersionedFile) and
2856
isinstance(target, KnitVersionedFile))
2857
except AttributeError:
2860
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2861
"""Copy texts to the target by extracting and adding them one by one.
2863
see join() for the parameter definitions.
2865
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2866
# --- the below is factorable out with VersionedFile.join, but wait for
2867
# VersionedFiles, it may all be simpler then.
2868
graph = Graph(self.source)
2869
search = graph._make_breadth_first_searcher(version_ids)
2870
transitive_ids = set()
2871
map(transitive_ids.update, list(search))
2872
parent_map = self.source.get_parent_map(transitive_ids)
2873
order = topo_sort(parent_map.items())
2875
def size_of_content(content):
2876
return sum(len(line) for line in content.text())
2877
# Cache at most 10MB of parent texts
2878
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2879
compute_size=size_of_content)
2880
# TODO: jam 20071116 It would be nice to have a streaming interface to
2881
# get multiple texts from a source. The source could be smarter
2882
# about how it handled intermediate stages.
2883
# get_line_list() or make_mpdiffs() seem like a possibility, but
2884
# at the moment they extract all full texts into memory, which
2885
# causes us to store more than our 3x fulltext goal.
2886
# Repository.iter_files_bytes() may be another possibility
2887
to_process = [version for version in order
2888
if version not in self.target]
2889
total = len(to_process)
2890
pb = ui.ui_factory.nested_progress_bar()
2892
for index, version in enumerate(to_process):
2893
pb.update('Converting versioned data', index, total)
2894
sha1, num_bytes, parent_text = self.target.add_lines(version,
2895
self.source.get_parents_with_ghosts(version),
2896
self.source.get_lines(version),
2897
parent_texts=parent_cache)
2898
parent_cache[version] = parent_text
2903
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2904
"""See InterVersionedFile.join."""
2905
# If the source and target are mismatched w.r.t. annotations vs
2906
# plain, the data needs to be converted accordingly
2907
if self.source.factory.annotated == self.target.factory.annotated:
2909
elif self.source.factory.annotated:
2910
converter = self._anno_to_plain_converter
2912
# We're converting from a plain to an annotated knit. Copy them
2913
# across by full texts.
2914
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2916
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2920
pb = ui.ui_factory.nested_progress_bar()
2922
version_ids = list(version_ids)
2923
if None in version_ids:
2924
version_ids.remove(None)
2926
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2928
this_versions = set(self.target._index.get_versions())
2929
# XXX: For efficiency we should not look at the whole index,
2930
# we only need to consider the referenced revisions - they
2931
# must all be present, or the method must be full-text.
2932
# TODO, RBC 20070919
2933
needed_versions = self.source_ancestry - this_versions
2935
if not needed_versions:
2937
full_list = topo_sort(
2938
self.source.get_parent_map(self.source.versions()))
2940
version_list = [i for i in full_list if (not self.target.has_version(i)
2941
and i in needed_versions)]
2945
copy_queue_records = []
2947
for version_id in version_list:
2948
options = self.source._index.get_options(version_id)
2949
parents = self.source._index.get_parents_with_ghosts(version_id)
2950
# check that its will be a consistent copy:
2951
for parent in parents:
2952
# if source has the parent, we must :
2953
# * already have it or
2954
# * have it scheduled already
2955
# otherwise we don't care
2956
if not (self.target.has_version(parent) or
2957
parent in copy_set or
2958
not self.source.has_version(parent)):
2959
raise AssertionError("problem joining parent %r "
2961
% (parent, self.source, self.target))
2962
index_memo = self.source._index.get_position(version_id)
2963
copy_queue_records.append((version_id, index_memo))
2964
copy_queue.append((version_id, options, parents))
2965
copy_set.add(version_id)
2967
# data suck the join:
2969
total = len(version_list)
2972
for (version_id, raw_data, _), \
2973
(version_id2, options, parents) in \
2974
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2976
if not (version_id == version_id2):
2977
raise AssertionError('logic error, inconsistent results')
2979
pb.update("Joining knit", count, total)
2981
size, raw_data = converter(raw_data, version_id, options,
2984
size = len(raw_data)
2985
raw_records.append((version_id, options, parents, size))
2986
raw_datum.append(raw_data)
2987
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2992
def _anno_to_plain_converter(self, raw_data, version_id, options,
2994
"""Convert annotated content to plain content."""
2995
data, digest = self.source._data._parse_record(version_id, raw_data)
2996
if 'fulltext' in options:
2997
content = self.source.factory.parse_fulltext(data, version_id)
2998
lines = self.target.factory.lower_fulltext(content)
3000
delta = self.source.factory.parse_line_delta(data, version_id,
3002
lines = self.target.factory.lower_line_delta(delta)
3003
return self.target._data._record_to_data(version_id, digest, lines)
3006
InterVersionedFile.register_optimiser(InterKnit)
3009
class WeaveToKnit(InterVersionedFile):
3010
"""Optimised code paths for weave to knit operations."""
3012
_matching_file_from_factory = bzrlib.weave.WeaveFile
3013
_matching_file_to_factory = staticmethod(make_file_knit)
3016
def is_compatible(source, target):
3017
"""Be compatible with weaves to knits."""
3019
return (isinstance(source, bzrlib.weave.Weave) and
3020
isinstance(target, KnitVersionedFile))
3021
except AttributeError:
3024
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
3025
"""See InterVersionedFile.join."""
3026
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
3031
pb = ui.ui_factory.nested_progress_bar()
3033
version_ids = list(version_ids)
3035
self.source_ancestry = set(self.source.get_ancestry(version_ids))
3036
this_versions = set(self.target._index.get_versions())
3037
needed_versions = self.source_ancestry - this_versions
3039
if not needed_versions:
3041
full_list = topo_sort(
3042
self.source.get_parent_map(self.source.versions()))
3044
version_list = [i for i in full_list if (not self.target.has_version(i)
3045
and i in needed_versions)]
3049
total = len(version_list)
3050
parent_map = self.source.get_parent_map(version_list)
3051
for version_id in version_list:
3052
pb.update("Converting to knit", count, total)
3053
parents = parent_map[version_id]
3054
# check that its will be a consistent copy:
3055
for parent in parents:
3056
# if source has the parent, we must already have it
3057
if not self.target.has_version(parent):
3058
raise AssertionError("%r does not have parent %r"
3059
% (self.target, parent))
3060
self.target.add_lines(
3061
version_id, parents, self.source.get_lines(version_id))
3068
InterVersionedFile.register_optimiser(WeaveToKnit)
3071
# Deprecated, use PatienceSequenceMatcher instead
3072
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
3075
def annotate_knit(knit, revision_id):
3076
"""Annotate a knit with no cached annotations.
3078
This implementation is for knits with no cached annotations.
3079
It will work for knits with cached annotations, but this is not
3082
annotator = _KnitAnnotator(knit)
3083
return iter(annotator.annotate(revision_id))
3086
class _KnitAnnotator(object):
3087
"""Build up the annotations for a text."""
3089
def __init__(self, knit):
3092
# Content objects, differs from fulltexts because of how final newlines
3093
# are treated by knits. the content objects here will always have a
3095
self._fulltext_contents = {}
3097
# Annotated lines of specific revisions
3098
self._annotated_lines = {}
3100
# Track the raw data for nodes that we could not process yet.
3101
# This maps the revision_id of the base to a list of children that will
3102
# annotated from it.
3103
self._pending_children = {}
3105
# Nodes which cannot be extracted
3106
self._ghosts = set()
3108
# Track how many children this node has, so we know if we need to keep
3110
self._annotate_children = {}
3111
self._compression_children = {}
3113
self._all_build_details = {}
3114
# The children => parent revision_id graph
3115
self._revision_id_graph = {}
3117
self._heads_provider = None
3119
self._nodes_to_keep_annotations = set()
3120
self._generations_until_keep = 100
3122
def set_generations_until_keep(self, value):
3123
"""Set the number of generations before caching a node.
3125
Setting this to -1 will cache every merge node, setting this higher
3126
will cache fewer nodes.
3128
self._generations_until_keep = value
3130
def _add_fulltext_content(self, revision_id, content_obj):
3131
self._fulltext_contents[revision_id] = content_obj
3132
# TODO: jam 20080305 It might be good to check the sha1digest here
3133
return content_obj.text()
3135
def _check_parents(self, child, nodes_to_annotate):
3136
"""Check if all parents have been processed.
3138
:param child: A tuple of (rev_id, parents, raw_content)
3139
:param nodes_to_annotate: If child is ready, add it to
3140
nodes_to_annotate, otherwise put it back in self._pending_children
3142
for parent_id in child[1]:
3143
if (parent_id not in self._annotated_lines):
3144
# This parent is present, but another parent is missing
3145
self._pending_children.setdefault(parent_id,
3149
# This one is ready to be processed
3150
nodes_to_annotate.append(child)
3152
def _add_annotation(self, revision_id, fulltext, parent_ids,
3153
left_matching_blocks=None):
3154
"""Add an annotation entry.
3156
All parents should already have been annotated.
3157
:return: A list of children that now have their parents satisfied.
3159
a = self._annotated_lines
3160
annotated_parent_lines = [a[p] for p in parent_ids]
3161
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
3162
fulltext, revision_id, left_matching_blocks,
3163
heads_provider=self._get_heads_provider()))
3164
self._annotated_lines[revision_id] = annotated_lines
3165
for p in parent_ids:
3166
ann_children = self._annotate_children[p]
3167
ann_children.remove(revision_id)
3168
if (not ann_children
3169
and p not in self._nodes_to_keep_annotations):
3170
del self._annotated_lines[p]
3171
del self._all_build_details[p]
3172
if p in self._fulltext_contents:
3173
del self._fulltext_contents[p]
3174
# Now that we've added this one, see if there are any pending
3175
# deltas to be done, certainly this parent is finished
3176
nodes_to_annotate = []
3177
for child in self._pending_children.pop(revision_id, []):
3178
self._check_parents(child, nodes_to_annotate)
3179
return nodes_to_annotate
3181
def _get_build_graph(self, revision_id):
3182
"""Get the graphs for building texts and annotations.
3184
The data you need for creating a full text may be different than the
3185
data you need to annotate that text. (At a minimum, you need both
3186
parents to create an annotation, but only need 1 parent to generate the
3189
:return: A list of (revision_id, index_memo) records, suitable for
3190
passing to read_records_iter to start reading in the raw data fro/
3193
if revision_id in self._annotated_lines:
3196
pending = set([revision_id])
3201
# get all pending nodes
3203
this_iteration = pending
3204
build_details = self._knit._index.get_build_details(this_iteration)
3205
self._all_build_details.update(build_details)
3206
# new_nodes = self._knit._index._get_entries(this_iteration)
3208
for rev_id, details in build_details.iteritems():
3209
(index_memo, compression_parent, parents,
3210
record_details) = details
3211
self._revision_id_graph[rev_id] = parents
3212
records.append((rev_id, index_memo))
3213
# Do we actually need to check _annotated_lines?
3214
pending.update(p for p in parents
3215
if p not in self._all_build_details)
3216
if compression_parent:
3217
self._compression_children.setdefault(compression_parent,
3220
for parent in parents:
3221
self._annotate_children.setdefault(parent,
3223
num_gens = generation - kept_generation
3224
if ((num_gens >= self._generations_until_keep)
3225
and len(parents) > 1):
3226
kept_generation = generation
3227
self._nodes_to_keep_annotations.add(rev_id)
3229
missing_versions = this_iteration.difference(build_details.keys())
3230
self._ghosts.update(missing_versions)
3231
for missing_version in missing_versions:
3232
# add a key, no parents
3233
self._revision_id_graph[missing_version] = ()
3234
pending.discard(missing_version) # don't look for it
3235
if self._ghosts.intersection(self._compression_children):
3237
"We cannot have nodes which have a ghost compression parent:\n"
3239
"compression children: %r"
3240
% (self._ghosts, self._compression_children))
3241
# Cleanout anything that depends on a ghost so that we don't wait for
3242
# the ghost to show up
3243
for node in self._ghosts:
3244
if node in self._annotate_children:
3245
# We won't be building this node
3246
del self._annotate_children[node]
3247
# Generally we will want to read the records in reverse order, because
3248
# we find the parent nodes after the children
3252
def _annotate_records(self, records):
3253
"""Build the annotations for the listed records."""
3254
# We iterate in the order read, rather than a strict order requested
3255
# However, process what we can, and put off to the side things that
3256
# still need parents, cleaning them up when those parents are
3258
for (rev_id, record,
3259
digest) in self._knit._data.read_records_iter(records):
3260
if rev_id in self._annotated_lines:
3262
parent_ids = self._revision_id_graph[rev_id]
3263
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3264
details = self._all_build_details[rev_id]
3265
(index_memo, compression_parent, parents,
3266
record_details) = details
3267
nodes_to_annotate = []
3268
# TODO: Remove the punning between compression parents, and
3269
# parent_ids, we should be able to do this without assuming
3271
if len(parent_ids) == 0:
3272
# There are no parents for this node, so just add it
3273
# TODO: This probably needs to be decoupled
3274
fulltext_content, delta = self._knit.factory.parse_record(
3275
rev_id, record, record_details, None)
3276
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3277
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3278
parent_ids, left_matching_blocks=None))
3280
child = (rev_id, parent_ids, record)
3281
# Check if all the parents are present
3282
self._check_parents(child, nodes_to_annotate)
3283
while nodes_to_annotate:
3284
# Should we use a queue here instead of a stack?
3285
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3286
(index_memo, compression_parent, parents,
3287
record_details) = self._all_build_details[rev_id]
3288
if compression_parent is not None:
3289
comp_children = self._compression_children[compression_parent]
3290
if rev_id not in comp_children:
3291
raise AssertionError("%r not in compression children %r"
3292
% (rev_id, comp_children))
3293
# If there is only 1 child, it is safe to reuse this
3295
reuse_content = (len(comp_children) == 1
3296
and compression_parent not in
3297
self._nodes_to_keep_annotations)
3299
# Remove it from the cache since it will be changing
3300
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3301
# Make sure to copy the fulltext since it might be
3303
parent_fulltext = list(parent_fulltext_content.text())
3305
parent_fulltext_content = self._fulltext_contents[compression_parent]
3306
parent_fulltext = parent_fulltext_content.text()
3307
comp_children.remove(rev_id)
3308
fulltext_content, delta = self._knit.factory.parse_record(
3309
rev_id, record, record_details,
3310
parent_fulltext_content,
3311
copy_base_content=(not reuse_content))
3312
fulltext = self._add_fulltext_content(rev_id,
3314
blocks = KnitContent.get_line_delta_blocks(delta,
3315
parent_fulltext, fulltext)
3317
fulltext_content = self._knit.factory.parse_fulltext(
3319
fulltext = self._add_fulltext_content(rev_id,
3322
nodes_to_annotate.extend(
3323
self._add_annotation(rev_id, fulltext, parent_ids,
3324
left_matching_blocks=blocks))
3326
def _get_heads_provider(self):
3327
"""Create a heads provider for resolving ancestry issues."""
3328
if self._heads_provider is not None:
3329
return self._heads_provider
3330
parent_provider = _mod_graph.DictParentsProvider(
3331
self._revision_id_graph)
3332
graph_obj = _mod_graph.Graph(parent_provider)
3333
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3334
self._heads_provider = head_cache
3337
def annotate(self, revision_id):
3338
"""Return the annotated fulltext at the given revision.
3340
:param revision_id: The revision id for this file
3342
records = self._get_build_graph(revision_id)
3343
if revision_id in self._ghosts:
3344
raise errors.RevisionNotPresent(revision_id, self._knit)
3345
self._annotate_records(records)
3346
return self._annotated_lines[revision_id]
3350
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3352
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
