1
# Copyright (C) 2005, 2006, 2007 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""Knit versionedfile implementation.
19
A knit is a versioned file implementation that supports efficient append only
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
end-marker; simply "end VERSION"
28
delta can be line or full contents.a
29
... the 8's there are the index number of the annotation.
30
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
8 e.set('executable', 'yes')
36
8 if elt.get('executable') == 'yes':
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
43
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
44
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
45
09:33 < lifeless> right
46
09:33 < jrydberg> lifeless: the position and size is the range in the data file
49
so the index sequence is the dictionary compressed sequence number used
50
in the deltas to provide line annotation
55
# 10:16 < lifeless> make partial index writes safe
56
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
57
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
59
# move sha1 out of the content so that join is faster at verifying parents
60
# record content length ?
64
from cStringIO import StringIO
65
from itertools import izip, chain
70
from zlib import Z_DEFAULT_COMPRESSION
73
from bzrlib.lazy_import import lazy_import
74
lazy_import(globals(), """
94
from bzrlib.errors import (
102
RevisionAlreadyPresent,
104
from bzrlib.graph import Graph
105
from bzrlib.osutils import (
111
from bzrlib.symbol_versioning import (
112
DEPRECATED_PARAMETER,
117
from bzrlib.tsort import topo_sort
118
from bzrlib.tuned_gzip import GzipFile, bytes_to_gzip
120
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
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 KnitContent(object):
142
"""Content of a knit version to which deltas can be applied."""
145
self._should_strip_eol = False
148
"""Return a list of (origin, text) tuples."""
149
return list(self.annotate_iter())
151
def apply_delta(self, delta, new_version_id):
152
"""Apply delta to this object to become new_version_id."""
153
raise NotImplementedError(self.apply_delta)
155
def cleanup_eol(self, copy_on_mutate=True):
156
if self._should_strip_eol:
158
self._lines = self._lines[:]
159
self.strip_last_line_newline()
161
def line_delta_iter(self, new_lines):
162
"""Generate line-based delta from this content to new_lines."""
163
new_texts = new_lines.text()
164
old_texts = self.text()
165
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
166
for tag, i1, i2, j1, j2 in s.get_opcodes():
169
# ofrom, oto, length, data
170
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
172
def line_delta(self, new_lines):
173
return list(self.line_delta_iter(new_lines))
176
def get_line_delta_blocks(knit_delta, source, target):
177
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
178
target_len = len(target)
181
for s_begin, s_end, t_len, new_text in knit_delta:
182
true_n = s_begin - s_pos
185
# knit deltas do not provide reliable info about whether the
186
# last line of a file matches, due to eol handling.
187
if source[s_pos + n -1] != target[t_pos + n -1]:
190
yield s_pos, t_pos, n
191
t_pos += t_len + true_n
193
n = target_len - t_pos
195
if source[s_pos + n -1] != target[t_pos + n -1]:
198
yield s_pos, t_pos, n
199
yield s_pos + (target_len - t_pos), target_len, 0
202
class AnnotatedKnitContent(KnitContent):
203
"""Annotated content."""
205
def __init__(self, lines):
206
KnitContent.__init__(self)
209
def annotate_iter(self):
210
"""Yield tuples of (origin, text) for each content line."""
211
return iter(self._lines)
213
def apply_delta(self, delta, new_version_id):
214
"""Apply delta to this object to become new_version_id."""
217
for start, end, count, delta_lines in delta:
218
lines[offset+start:offset+end] = delta_lines
219
offset = offset + (start - end) + count
221
def strip_last_line_newline(self):
222
line = self._lines[-1][1].rstrip('\n')
223
self._lines[-1] = (self._lines[-1][0], line)
224
self._should_strip_eol = False
228
lines = [text for origin, text in self._lines]
229
except ValueError, e:
230
# most commonly (only?) caused by the internal form of the knit
231
# missing annotation information because of a bug - see thread
233
raise KnitCorrupt(self,
234
"line in annotated knit missing annotation information: %s"
237
if self._should_strip_eol:
238
anno, line = lines[-1]
239
lines[-1] = (anno, line.rstrip('\n'))
243
return AnnotatedKnitContent(self._lines[:])
246
class PlainKnitContent(KnitContent):
247
"""Unannotated content.
249
When annotate[_iter] is called on this content, the same version is reported
250
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
254
def __init__(self, lines, version_id):
255
KnitContent.__init__(self)
257
self._version_id = version_id
259
def annotate_iter(self):
260
"""Yield tuples of (origin, text) for each content line."""
261
for line in self._lines:
262
yield self._version_id, line
264
def apply_delta(self, delta, new_version_id):
265
"""Apply delta to this object to become new_version_id."""
268
for start, end, count, delta_lines in delta:
269
lines[offset+start:offset+end] = delta_lines
270
offset = offset + (start - end) + count
271
self._version_id = new_version_id
274
return PlainKnitContent(self._lines[:], self._version_id)
276
def strip_last_line_newline(self):
277
self._lines[-1] = self._lines[-1].rstrip('\n')
278
self._should_strip_eol = False
282
if self._should_strip_eol:
284
lines[-1] = lines[-1].rstrip('\n')
288
class _KnitFactory(object):
289
"""Base class for common Factory functions."""
291
def parse_record(self, version_id, record, record_details,
292
base_content, copy_base_content=True):
293
"""Parse a record into a full content object.
295
:param version_id: The official version id for this content
296
:param record: The data returned by read_records_iter()
297
:param record_details: Details about the record returned by
299
:param base_content: If get_build_details returns a compression_parent,
300
you must return a base_content here, else use None
301
:param copy_base_content: When building from the base_content, decide
302
you can either copy it and return a new object, or modify it in
304
:return: (content, delta) A Content object and possibly a line-delta,
307
method, noeol = record_details
308
if method == 'line-delta':
309
assert base_content is not None
310
if copy_base_content:
311
content = base_content.copy()
313
content = base_content
314
delta = self.parse_line_delta(record, version_id)
315
content.apply_delta(delta, version_id)
317
content = self.parse_fulltext(record, version_id)
319
content._should_strip_eol = noeol
320
return (content, delta)
323
class KnitAnnotateFactory(_KnitFactory):
324
"""Factory for creating annotated Content objects."""
328
def make(self, lines, version_id):
329
num_lines = len(lines)
330
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
332
def parse_fulltext(self, content, version_id):
333
"""Convert fulltext to internal representation
335
fulltext content is of the format
336
revid(utf8) plaintext\n
337
internal representation is of the format:
340
# TODO: jam 20070209 The tests expect this to be returned as tuples,
341
# but the code itself doesn't really depend on that.
342
# Figure out a way to not require the overhead of turning the
343
# list back into tuples.
344
lines = [tuple(line.split(' ', 1)) for line in content]
345
return AnnotatedKnitContent(lines)
347
def parse_line_delta_iter(self, lines):
348
return iter(self.parse_line_delta(lines))
350
def parse_line_delta(self, lines, version_id, plain=False):
351
"""Convert a line based delta into internal representation.
353
line delta is in the form of:
354
intstart intend intcount
356
revid(utf8) newline\n
357
internal representation is
358
(start, end, count, [1..count tuples (revid, newline)])
360
:param plain: If True, the lines are returned as a plain
361
list without annotations, not as a list of (origin, content) tuples, i.e.
362
(start, end, count, [1..count newline])
369
def cache_and_return(line):
370
origin, text = line.split(' ', 1)
371
return cache.setdefault(origin, origin), text
373
# walk through the lines parsing.
374
# Note that the plain test is explicitly pulled out of the
375
# loop to minimise any performance impact
378
start, end, count = [int(n) for n in header.split(',')]
379
contents = [next().split(' ', 1)[1] for i in xrange(count)]
380
result.append((start, end, count, contents))
383
start, end, count = [int(n) for n in header.split(',')]
384
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
385
result.append((start, end, count, contents))
388
def get_fulltext_content(self, lines):
389
"""Extract just the content lines from a fulltext."""
390
return (line.split(' ', 1)[1] for line in lines)
392
def get_linedelta_content(self, lines):
393
"""Extract just the content from a line delta.
395
This doesn't return all of the extra information stored in a delta.
396
Only the actual content lines.
401
header = header.split(',')
402
count = int(header[2])
403
for i in xrange(count):
404
origin, text = next().split(' ', 1)
407
def lower_fulltext(self, content):
408
"""convert a fulltext content record into a serializable form.
410
see parse_fulltext which this inverts.
412
# TODO: jam 20070209 We only do the caching thing to make sure that
413
# the origin is a valid utf-8 line, eventually we could remove it
414
return ['%s %s' % (o, t) for o, t in content._lines]
416
def lower_line_delta(self, delta):
417
"""convert a delta into a serializable form.
419
See parse_line_delta which this inverts.
421
# TODO: jam 20070209 We only do the caching thing to make sure that
422
# the origin is a valid utf-8 line, eventually we could remove it
424
for start, end, c, lines in delta:
425
out.append('%d,%d,%d\n' % (start, end, c))
426
out.extend(origin + ' ' + text
427
for origin, text in lines)
430
def annotate_iter(self, knit, version_id):
431
content = knit._get_content(version_id)
432
return content.annotate_iter()
435
class KnitPlainFactory(_KnitFactory):
436
"""Factory for creating plain Content objects."""
440
def make(self, lines, version_id):
441
return PlainKnitContent(lines, version_id)
443
def parse_fulltext(self, content, version_id):
444
"""This parses an unannotated fulltext.
446
Note that this is not a noop - the internal representation
447
has (versionid, line) - its just a constant versionid.
449
return self.make(content, version_id)
451
def parse_line_delta_iter(self, lines, version_id):
453
num_lines = len(lines)
454
while cur < num_lines:
457
start, end, c = [int(n) for n in header.split(',')]
458
yield start, end, c, lines[cur:cur+c]
461
def parse_line_delta(self, lines, version_id):
462
return list(self.parse_line_delta_iter(lines, version_id))
464
def get_fulltext_content(self, lines):
465
"""Extract just the content lines from a fulltext."""
468
def get_linedelta_content(self, lines):
469
"""Extract just the content from a line delta.
471
This doesn't return all of the extra information stored in a delta.
472
Only the actual content lines.
477
header = header.split(',')
478
count = int(header[2])
479
for i in xrange(count):
482
def lower_fulltext(self, content):
483
return content.text()
485
def lower_line_delta(self, delta):
487
for start, end, c, lines in delta:
488
out.append('%d,%d,%d\n' % (start, end, c))
492
def annotate_iter(self, knit, version_id):
493
annotator = _KnitAnnotator(knit)
494
return iter(annotator.annotate(version_id))
497
def make_empty_knit(transport, relpath):
498
"""Construct a empty knit at the specified location."""
499
k = make_file_knit(transport, relpath, 'w', KnitPlainFactory)
502
def make_file_knit(name, transport, file_mode=None, access_mode='w',
503
factory=None, delta=True, create=False, create_parent_dir=False,
504
delay_create=False, dir_mode=None, get_scope=None):
505
"""Factory to create a KnitVersionedFile for a .knit/.kndx file pair."""
507
factory = KnitAnnotateFactory()
509
factory = KnitPlainFactory()
510
if get_scope is None:
511
get_scope = lambda:None
512
index = _KnitIndex(transport, name + INDEX_SUFFIX,
513
access_mode, create=create, file_mode=file_mode,
514
create_parent_dir=create_parent_dir, delay_create=delay_create,
515
dir_mode=dir_mode, get_scope=get_scope)
516
access = _KnitAccess(transport, name + DATA_SUFFIX, file_mode,
517
dir_mode, ((create and not len(index)) and delay_create),
519
return KnitVersionedFile(name, transport, factory=factory,
520
create=create, delay_create=delay_create, index=index,
521
access_method=access)
525
"""Return the suffixes used by file based knits."""
526
return [DATA_SUFFIX, INDEX_SUFFIX]
527
make_file_knit.get_suffixes = get_suffixes
530
class KnitVersionedFile(VersionedFile):
531
"""Weave-like structure with faster random access.
533
A knit stores a number of texts and a summary of the relationships
534
between them. Texts are identified by a string version-id. Texts
535
are normally stored and retrieved as a series of lines, but can
536
also be passed as single strings.
538
Lines are stored with the trailing newline (if any) included, to
539
avoid special cases for files with no final newline. Lines are
540
composed of 8-bit characters, not unicode. The combination of
541
these approaches should mean any 'binary' file can be safely
542
stored and retrieved.
545
def __init__(self, relpath, transport, file_mode=None,
546
factory=None, delta=True, create=False, create_parent_dir=False,
547
delay_create=False, dir_mode=None, index=None, access_method=None):
548
"""Construct a knit at location specified by relpath.
550
:param create: If not True, only open an existing knit.
551
:param create_parent_dir: If True, create the parent directory if
552
creating the file fails. (This is used for stores with
553
hash-prefixes that may not exist yet)
554
:param delay_create: The calling code is aware that the knit won't
555
actually be created until the first data is stored.
556
:param index: An index to use for the knit.
558
super(KnitVersionedFile, self).__init__()
559
self.transport = transport
560
self.filename = relpath
561
self.factory = factory or KnitAnnotateFactory()
564
self._max_delta_chain = 200
566
if None in (access_method, index):
567
raise NotImplementedError, "No default access_method or index any more"
569
_access = access_method
570
if create and not len(self) and not delay_create:
572
self._data = _KnitData(_access)
575
return '%s(%s)' % (self.__class__.__name__,
576
self.transport.abspath(self.filename))
578
def _check_should_delta(self, first_parents):
579
"""Iterate back through the parent listing, looking for a fulltext.
581
This is used when we want to decide whether to add a delta or a new
582
fulltext. It searches for _max_delta_chain parents. When it finds a
583
fulltext parent, it sees if the total size of the deltas leading up to
584
it is large enough to indicate that we want a new full text anyway.
586
Return True if we should create a new delta, False if we should use a
591
delta_parents = first_parents
592
for count in xrange(self._max_delta_chain):
593
parent = delta_parents[0]
594
method = self._index.get_method(parent)
595
index, pos, size = self._index.get_position(parent)
596
if method == 'fulltext':
600
delta_parents = self._index.get_parent_map([parent])[parent]
602
# We couldn't find a fulltext, so we must create a new one
605
return fulltext_size > delta_size
607
def _check_write_ok(self):
608
return self._index._check_write_ok()
610
def _add_raw_records(self, records, data):
611
"""Add all the records 'records' with data pre-joined in 'data'.
613
:param records: A list of tuples(version_id, options, parents, size).
614
:param data: The data for the records. When it is written, the records
615
are adjusted to have pos pointing into data by the sum of
616
the preceding records sizes.
619
raw_record_sizes = [record[3] for record in records]
620
positions = self._data.add_raw_records(raw_record_sizes, data)
623
for (version_id, options, parents, size), access_memo in zip(
625
index_entries.append((version_id, options, access_memo, parents))
626
if self._data._do_cache:
627
self._data._cache[version_id] = data[offset:offset+size]
629
self._index.add_versions(index_entries)
631
def enable_cache(self):
632
"""Start caching data for this knit"""
633
self._data.enable_cache()
635
def clear_cache(self):
636
"""Clear the data cache only."""
637
self._data.clear_cache()
639
def copy_to(self, name, transport):
640
"""See VersionedFile.copy_to()."""
641
# copy the current index to a temp index to avoid racing with local
643
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
644
self.transport.get(self._index._filename))
646
f = self._data._open_file()
648
transport.put_file(name + DATA_SUFFIX, f)
651
# move the copied index into place
652
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
654
def get_data_stream(self, required_versions):
655
"""Get a data stream for the specified versions.
657
Versions may be returned in any order, not necessarily the order
658
specified. They are returned in a partial order by compression
659
parent, so that the deltas can be applied as the data stream is
660
inserted; however note that compression parents will not be sent
661
unless they were specifically requested, as the client may already
664
:param required_versions: The exact set of versions to be extracted.
665
Unlike some other knit methods, this is not used to generate a
666
transitive closure, rather it is used precisely as given.
668
:returns: format_signature, list of (version, options, length, parents),
671
required_version_set = frozenset(required_versions)
673
# list of revisions that can just be sent without waiting for their
676
# map from revision to the children based on it
678
# first, read all relevant index data, enough to sort into the right
680
for version_id in required_versions:
681
options = self._index.get_options(version_id)
682
parents = self._index.get_parents_with_ghosts(version_id)
683
index_memo = self._index.get_position(version_id)
684
version_index[version_id] = (index_memo, options, parents)
685
if ('line-delta' in options
686
and parents[0] in required_version_set):
687
# must wait until the parent has been sent
688
deferred.setdefault(parents[0], []). \
691
# either a fulltext, or a delta whose parent the client did
692
# not ask for and presumably already has
693
ready_to_send.append(version_id)
694
# build a list of results to return, plus instructions for data to
696
copy_queue_records = []
697
temp_version_list = []
699
# XXX: pushing and popping lists may be a bit inefficient
700
version_id = ready_to_send.pop(0)
701
(index_memo, options, parents) = version_index[version_id]
702
copy_queue_records.append((version_id, index_memo))
703
none, data_pos, data_size = index_memo
704
temp_version_list.append((version_id, options, data_size,
706
if version_id in deferred:
707
# now we can send all the children of this revision - we could
708
# put them in anywhere, but we hope that sending them soon
709
# after the fulltext will give good locality in the receiver
710
ready_to_send[:0] = deferred.pop(version_id)
711
assert len(deferred) == 0, \
712
"Still have compressed child versions waiting to be sent"
713
# XXX: The stream format is such that we cannot stream it - we have to
714
# know the length of all the data a-priori.
716
result_version_list = []
717
for (version_id, raw_data), \
718
(version_id2, options, _, parents) in \
719
izip(self._data.read_records_iter_raw(copy_queue_records),
721
assert version_id == version_id2, \
722
'logic error, inconsistent results'
723
raw_datum.append(raw_data)
724
result_version_list.append(
725
(version_id, options, len(raw_data), parents))
726
# provide a callback to get data incrementally.
727
pseudo_file = StringIO(''.join(raw_datum))
730
return pseudo_file.read()
732
return pseudo_file.read(length)
733
return (self.get_format_signature(), result_version_list, read)
735
def _extract_blocks(self, version_id, source, target):
736
if self._index.get_method(version_id) != 'line-delta':
738
parent, sha1, noeol, delta = self.get_delta(version_id)
739
return KnitContent.get_line_delta_blocks(delta, source, target)
741
def get_delta(self, version_id):
742
"""Get a delta for constructing version from some other version."""
743
self.check_not_reserved_id(version_id)
744
parents = self.get_parent_map([version_id])[version_id]
749
index_memo = self._index.get_position(version_id)
750
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
751
noeol = 'no-eol' in self._index.get_options(version_id)
752
if 'fulltext' == self._index.get_method(version_id):
753
new_content = self.factory.parse_fulltext(data, version_id)
754
if parent is not None:
755
reference_content = self._get_content(parent)
756
old_texts = reference_content.text()
759
new_texts = new_content.text()
760
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
762
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
764
delta = self.factory.parse_line_delta(data, version_id)
765
return parent, sha1, noeol, delta
767
def get_format_signature(self):
768
"""See VersionedFile.get_format_signature()."""
769
if self.factory.annotated:
770
annotated_part = "annotated"
772
annotated_part = "plain"
773
return "knit-%s" % (annotated_part,)
775
@deprecated_method(one_four)
776
def get_graph_with_ghosts(self):
777
"""See VersionedFile.get_graph_with_ghosts()."""
778
return self.get_parent_map(self.versions())
780
def get_sha1s(self, version_ids):
781
"""See VersionedFile.get_sha1s()."""
782
record_map = self._get_record_map(version_ids)
783
# record entry 2 is the 'digest'.
784
return [record_map[v][2] for v in version_ids]
786
@deprecated_method(one_four)
787
def has_ghost(self, version_id):
788
"""True if there is a ghost reference in the file to version_id."""
790
if self.has_version(version_id):
792
# optimisable if needed by memoising the _ghosts set.
793
items = self.get_parent_map(self.versions())
794
for parents in items.itervalues():
795
for parent in parents:
796
if parent == version_id and parent not in items:
800
def insert_data_stream(self, (format, data_list, reader_callable)):
801
"""Insert knit records from a data stream into this knit.
803
If a version in the stream is already present in this knit, it will not
804
be inserted a second time. It will be checked for consistency with the
805
stored version however, and may cause a KnitCorrupt error to be raised
806
if the data in the stream disagrees with the already stored data.
808
:seealso: get_data_stream
810
if format != self.get_format_signature():
811
if 'knit' in debug.debug_flags:
813
'incompatible format signature inserting to %r', self)
814
source = self._knit_from_datastream(
815
(format, data_list, reader_callable))
819
for version_id, options, length, parents in data_list:
820
if self.has_version(version_id):
821
# First check: the list of parents.
822
my_parents = self.get_parents_with_ghosts(version_id)
823
if tuple(my_parents) != tuple(parents):
824
# XXX: KnitCorrupt is not quite the right exception here.
827
'parents list %r from data stream does not match '
828
'already recorded parents %r for %s'
829
% (parents, my_parents, version_id))
831
# Also check the SHA-1 of the fulltext this content will
833
raw_data = reader_callable(length)
834
my_fulltext_sha1 = self.get_sha1s([version_id])[0]
835
df, rec = self._data._parse_record_header(version_id, raw_data)
836
stream_fulltext_sha1 = rec[3]
837
if my_fulltext_sha1 != stream_fulltext_sha1:
838
# Actually, we don't know if it's this knit that's corrupt,
839
# or the data stream we're trying to insert.
841
self.filename, 'sha-1 does not match %s' % version_id)
843
if 'line-delta' in options:
844
# Make sure that this knit record is actually useful: a
845
# line-delta is no use unless we have its parent.
846
# Fetching from a broken repository with this problem
847
# shouldn't break the target repository.
849
# See https://bugs.launchpad.net/bzr/+bug/164443
850
if not self._index.has_version(parents[0]):
853
'line-delta from stream '
856
'missing parent %s\n'
857
'Try running "bzr check" '
858
'on the source repository, and "bzr reconcile" '
860
(version_id, parents[0]))
861
self._add_raw_records(
862
[(version_id, options, parents, length)],
863
reader_callable(length))
865
def _knit_from_datastream(self, (format, data_list, reader_callable)):
866
"""Create a knit object from a data stream.
868
This method exists to allow conversion of data streams that do not
869
match the signature of this knit. Generally it will be slower and use
870
more memory to use this method to insert data, but it will work.
872
:seealso: get_data_stream for details on datastreams.
873
:return: A knit versioned file which can be used to join the datastream
876
if format == "knit-plain":
877
factory = KnitPlainFactory()
878
elif format == "knit-annotated":
879
factory = KnitAnnotateFactory()
881
raise errors.KnitDataStreamUnknown(format)
882
index = _StreamIndex(data_list, self._index)
883
access = _StreamAccess(reader_callable, index, self, factory)
884
return KnitVersionedFile(self.filename, self.transport,
885
factory=factory, index=index, access_method=access)
888
"""See VersionedFile.versions."""
889
if 'evil' in debug.debug_flags:
890
trace.mutter_callsite(2, "versions scales with size of history")
891
return self._index.get_versions()
893
def has_version(self, version_id):
894
"""See VersionedFile.has_version."""
895
if 'evil' in debug.debug_flags:
896
trace.mutter_callsite(2, "has_version is a LBYL scenario")
897
return self._index.has_version(version_id)
899
__contains__ = has_version
901
def _merge_annotations(self, content, parents, parent_texts={},
902
delta=None, annotated=None,
903
left_matching_blocks=None):
904
"""Merge annotations for content. This is done by comparing
905
the annotations based on changed to the text.
907
if left_matching_blocks is not None:
908
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
912
for parent_id in parents:
913
merge_content = self._get_content(parent_id, parent_texts)
914
if (parent_id == parents[0] and delta_seq is not None):
917
seq = patiencediff.PatienceSequenceMatcher(
918
None, merge_content.text(), content.text())
919
for i, j, n in seq.get_matching_blocks():
922
# this appears to copy (origin, text) pairs across to the
923
# new content for any line that matches the last-checked
925
content._lines[j:j+n] = merge_content._lines[i:i+n]
927
if delta_seq is None:
928
reference_content = self._get_content(parents[0], parent_texts)
929
new_texts = content.text()
930
old_texts = reference_content.text()
931
delta_seq = patiencediff.PatienceSequenceMatcher(
932
None, old_texts, new_texts)
933
return self._make_line_delta(delta_seq, content)
935
def _make_line_delta(self, delta_seq, new_content):
936
"""Generate a line delta from delta_seq and new_content."""
938
for op in delta_seq.get_opcodes():
941
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
944
def _get_components_positions(self, version_ids):
945
"""Produce a map of position data for the components of versions.
947
This data is intended to be used for retrieving the knit records.
949
A dict of version_id to (record_details, index_memo, next, parents) is
951
method is the way referenced data should be applied.
952
index_memo is the handle to pass to the data access to actually get the
954
next is the build-parent of the version, or None for fulltexts.
955
parents is the version_ids of the parents of this version
958
pending_components = version_ids
959
while pending_components:
960
build_details = self._index.get_build_details(pending_components)
961
current_components = set(pending_components)
962
pending_components = set()
963
for version_id, details in build_details.iteritems():
964
(index_memo, compression_parent, parents,
965
record_details) = details
966
method = record_details[0]
967
if compression_parent is not None:
968
pending_components.add(compression_parent)
969
component_data[version_id] = (record_details, index_memo,
971
missing = current_components.difference(build_details)
973
raise errors.RevisionNotPresent(missing.pop(), self.filename)
974
return component_data
976
def _get_content(self, version_id, parent_texts={}):
977
"""Returns a content object that makes up the specified
979
cached_version = parent_texts.get(version_id, None)
980
if cached_version is not None:
981
if not self.has_version(version_id):
982
raise RevisionNotPresent(version_id, self.filename)
983
return cached_version
985
text_map, contents_map = self._get_content_maps([version_id])
986
return contents_map[version_id]
988
def _check_versions_present(self, version_ids):
989
"""Check that all specified versions are present."""
990
self._index.check_versions_present(version_ids)
992
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
993
nostore_sha, random_id, check_content, left_matching_blocks):
994
"""See VersionedFile.add_lines_with_ghosts()."""
995
self._check_add(version_id, lines, random_id, check_content)
996
return self._add(version_id, lines, parents, self.delta,
997
parent_texts, left_matching_blocks, nostore_sha, random_id)
999
def _add_lines(self, version_id, parents, lines, parent_texts,
1000
left_matching_blocks, nostore_sha, random_id, check_content):
1001
"""See VersionedFile.add_lines."""
1002
self._check_add(version_id, lines, random_id, check_content)
1003
self._check_versions_present(parents)
1004
return self._add(version_id, lines[:], parents, self.delta,
1005
parent_texts, left_matching_blocks, nostore_sha, random_id)
1007
def _check_add(self, version_id, lines, random_id, check_content):
1008
"""check that version_id and lines are safe to add."""
1009
if contains_whitespace(version_id):
1010
raise InvalidRevisionId(version_id, self.filename)
1011
self.check_not_reserved_id(version_id)
1012
# Technically this could be avoided if we are happy to allow duplicate
1013
# id insertion when other things than bzr core insert texts, but it
1014
# seems useful for folk using the knit api directly to have some safety
1015
# blanket that we can disable.
1016
if not random_id and self.has_version(version_id):
1017
raise RevisionAlreadyPresent(version_id, self.filename)
1019
self._check_lines_not_unicode(lines)
1020
self._check_lines_are_lines(lines)
1022
def _add(self, version_id, lines, parents, delta, parent_texts,
1023
left_matching_blocks, nostore_sha, random_id):
1024
"""Add a set of lines on top of version specified by parents.
1026
If delta is true, compress the text as a line-delta against
1029
Any versions not present will be converted into ghosts.
1031
# first thing, if the content is something we don't need to store, find
1033
line_bytes = ''.join(lines)
1034
digest = sha_string(line_bytes)
1035
if nostore_sha == digest:
1036
raise errors.ExistingContent
1038
present_parents = []
1039
if parent_texts is None:
1041
for parent in parents:
1042
if self.has_version(parent):
1043
present_parents.append(parent)
1045
# can only compress against the left most present parent.
1047
(len(present_parents) == 0 or
1048
present_parents[0] != parents[0])):
1051
text_length = len(line_bytes)
1054
if lines[-1][-1] != '\n':
1055
# copy the contents of lines.
1057
options.append('no-eol')
1058
lines[-1] = lines[-1] + '\n'
1062
# To speed the extract of texts the delta chain is limited
1063
# to a fixed number of deltas. This should minimize both
1064
# I/O and the time spend applying deltas.
1065
delta = self._check_should_delta(present_parents)
1067
assert isinstance(version_id, str)
1068
content = self.factory.make(lines, version_id)
1069
if delta or (self.factory.annotated and len(present_parents) > 0):
1070
# Merge annotations from parent texts if needed.
1071
delta_hunks = self._merge_annotations(content, present_parents,
1072
parent_texts, delta, self.factory.annotated,
1073
left_matching_blocks)
1076
options.append('line-delta')
1077
store_lines = self.factory.lower_line_delta(delta_hunks)
1078
size, bytes = self._data._record_to_data(version_id, digest,
1081
options.append('fulltext')
1082
# isinstance is slower and we have no hierarchy.
1083
if self.factory.__class__ == KnitPlainFactory:
1084
# Use the already joined bytes saving iteration time in
1086
size, bytes = self._data._record_to_data(version_id, digest,
1087
lines, [line_bytes])
1089
# get mixed annotation + content and feed it into the
1091
store_lines = self.factory.lower_fulltext(content)
1092
size, bytes = self._data._record_to_data(version_id, digest,
1095
access_memo = self._data.add_raw_records([size], bytes)[0]
1096
self._index.add_versions(
1097
((version_id, options, access_memo, parents),),
1098
random_id=random_id)
1099
return digest, text_length, content
1101
def check(self, progress_bar=None):
1102
"""See VersionedFile.check()."""
1104
def _clone_text(self, new_version_id, old_version_id, parents):
1105
"""See VersionedFile.clone_text()."""
1106
# FIXME RBC 20060228 make fast by only inserting an index with null
1108
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
1110
def get_lines(self, version_id):
1111
"""See VersionedFile.get_lines()."""
1112
return self.get_line_list([version_id])[0]
1114
def _get_record_map(self, version_ids):
1115
"""Produce a dictionary of knit records.
1117
:return: {version_id:(record, record_details, digest, next)}
1119
data returned from read_records
1121
opaque information to pass to parse_record
1123
SHA1 digest of the full text after all steps are done
1125
build-parent of the version, i.e. the leftmost ancestor.
1126
Will be None if the record is not a delta.
1128
position_map = self._get_components_positions(version_ids)
1129
# c = component_id, r = record_details, i_m = index_memo, n = next
1130
records = [(c, i_m) for c, (r, i_m, n)
1131
in position_map.iteritems()]
1133
for component_id, record, digest in \
1134
self._data.read_records_iter(records):
1135
(record_details, index_memo, next) = position_map[component_id]
1136
record_map[component_id] = record, record_details, digest, next
1140
def get_text(self, version_id):
1141
"""See VersionedFile.get_text"""
1142
return self.get_texts([version_id])[0]
1144
def get_texts(self, version_ids):
1145
return [''.join(l) for l in self.get_line_list(version_ids)]
1147
def get_line_list(self, version_ids):
1148
"""Return the texts of listed versions as a list of strings."""
1149
for version_id in version_ids:
1150
self.check_not_reserved_id(version_id)
1151
text_map, content_map = self._get_content_maps(version_ids)
1152
return [text_map[v] for v in version_ids]
1154
_get_lf_split_line_list = get_line_list
1156
def _get_content_maps(self, version_ids):
1157
"""Produce maps of text and KnitContents
1159
:return: (text_map, content_map) where text_map contains the texts for
1160
the requested versions and content_map contains the KnitContents.
1161
Both dicts take version_ids as their keys.
1163
# FUTURE: This function could be improved for the 'extract many' case
1164
# by tracking each component and only doing the copy when the number of
1165
# children than need to apply delta's to it is > 1 or it is part of the
1167
version_ids = list(version_ids)
1168
multiple_versions = len(version_ids) != 1
1169
record_map = self._get_record_map(version_ids)
1174
for version_id in version_ids:
1177
while cursor is not None:
1178
record, record_details, digest, next = record_map[cursor]
1179
components.append((cursor, record, record_details, digest))
1180
if cursor in content_map:
1185
for (component_id, record, record_details,
1186
digest) in reversed(components):
1187
if component_id in content_map:
1188
content = content_map[component_id]
1190
content, delta = self.factory.parse_record(version_id,
1191
record, record_details, content,
1192
copy_base_content=multiple_versions)
1193
if multiple_versions:
1194
content_map[component_id] = content
1196
content.cleanup_eol(copy_on_mutate=multiple_versions)
1197
final_content[version_id] = content
1199
# digest here is the digest from the last applied component.
1200
text = content.text()
1201
actual_sha = sha_strings(text)
1202
if actual_sha != digest:
1203
raise KnitCorrupt(self.filename,
1205
'\n of reconstructed text does not match'
1207
'\n for version %s' %
1208
(actual_sha, digest, version_id))
1209
text_map[version_id] = text
1210
return text_map, final_content
1212
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1214
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1215
if version_ids is None:
1216
version_ids = self.versions()
1218
pb = progress.DummyProgress()
1219
# we don't care about inclusions, the caller cares.
1220
# but we need to setup a list of records to visit.
1221
# we need version_id, position, length
1222
version_id_records = []
1223
requested_versions = set(version_ids)
1224
# filter for available versions
1225
for version_id in requested_versions:
1226
if not self.has_version(version_id):
1227
raise RevisionNotPresent(version_id, self.filename)
1228
# get a in-component-order queue:
1229
for version_id in self.versions():
1230
if version_id in requested_versions:
1231
index_memo = self._index.get_position(version_id)
1232
version_id_records.append((version_id, index_memo))
1234
total = len(version_id_records)
1235
for version_idx, (version_id, data, sha_value) in \
1236
enumerate(self._data.read_records_iter(version_id_records)):
1237
pb.update('Walking content.', version_idx, total)
1238
method = self._index.get_method(version_id)
1240
assert method in ('fulltext', 'line-delta')
1241
if method == 'fulltext':
1242
line_iterator = self.factory.get_fulltext_content(data)
1244
line_iterator = self.factory.get_linedelta_content(data)
1245
# XXX: It might be more efficient to yield (version_id,
1246
# line_iterator) in the future. However for now, this is a simpler
1247
# change to integrate into the rest of the codebase. RBC 20071110
1248
for line in line_iterator:
1249
yield line, version_id
1251
pb.update('Walking content.', total, total)
1253
def num_versions(self):
1254
"""See VersionedFile.num_versions()."""
1255
return self._index.num_versions()
1257
__len__ = num_versions
1259
def annotate_iter(self, version_id):
1260
"""See VersionedFile.annotate_iter."""
1261
return self.factory.annotate_iter(self, version_id)
1263
def get_parent_map(self, version_ids):
1264
"""See VersionedFile.get_parent_map."""
1265
return self._index.get_parent_map(version_ids)
1267
def get_ancestry(self, versions, topo_sorted=True):
1268
"""See VersionedFile.get_ancestry."""
1269
if isinstance(versions, basestring):
1270
versions = [versions]
1273
return self._index.get_ancestry(versions, topo_sorted)
1275
def get_ancestry_with_ghosts(self, versions):
1276
"""See VersionedFile.get_ancestry_with_ghosts."""
1277
if isinstance(versions, basestring):
1278
versions = [versions]
1281
return self._index.get_ancestry_with_ghosts(versions)
1283
def plan_merge(self, ver_a, ver_b):
1284
"""See VersionedFile.plan_merge."""
1285
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1286
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1287
annotated_a = self.annotate(ver_a)
1288
annotated_b = self.annotate(ver_b)
1289
return merge._plan_annotate_merge(annotated_a, annotated_b,
1290
ancestors_a, ancestors_b)
1293
class _KnitComponentFile(object):
1294
"""One of the files used to implement a knit database"""
1296
def __init__(self, transport, filename, mode, file_mode=None,
1297
create_parent_dir=False, dir_mode=None):
1298
self._transport = transport
1299
self._filename = filename
1301
self._file_mode = file_mode
1302
self._dir_mode = dir_mode
1303
self._create_parent_dir = create_parent_dir
1304
self._need_to_create = False
1306
def _full_path(self):
1307
"""Return the full path to this file."""
1308
return self._transport.base + self._filename
1310
def check_header(self, fp):
1311
line = fp.readline()
1313
# An empty file can actually be treated as though the file doesn't
1315
raise errors.NoSuchFile(self._full_path())
1316
if line != self.HEADER:
1317
raise KnitHeaderError(badline=line,
1318
filename=self._transport.abspath(self._filename))
1321
return '%s(%s)' % (self.__class__.__name__, self._filename)
1324
class _KnitIndex(_KnitComponentFile):
1325
"""Manages knit index file.
1327
The index is already kept in memory and read on startup, to enable
1328
fast lookups of revision information. The cursor of the index
1329
file is always pointing to the end, making it easy to append
1332
_cache is a cache for fast mapping from version id to a Index
1335
_history is a cache for fast mapping from indexes to version ids.
1337
The index data format is dictionary compressed when it comes to
1338
parent references; a index entry may only have parents that with a
1339
lover index number. As a result, the index is topological sorted.
1341
Duplicate entries may be written to the index for a single version id
1342
if this is done then the latter one completely replaces the former:
1343
this allows updates to correct version and parent information.
1344
Note that the two entries may share the delta, and that successive
1345
annotations and references MUST point to the first entry.
1347
The index file on disc contains a header, followed by one line per knit
1348
record. The same revision can be present in an index file more than once.
1349
The first occurrence gets assigned a sequence number starting from 0.
1351
The format of a single line is
1352
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1353
REVISION_ID is a utf8-encoded revision id
1354
FLAGS is a comma separated list of flags about the record. Values include
1355
no-eol, line-delta, fulltext.
1356
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1357
that the the compressed data starts at.
1358
LENGTH is the ascii representation of the length of the data file.
1359
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1361
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1362
revision id already in the knit that is a parent of REVISION_ID.
1363
The ' :' marker is the end of record marker.
1366
when a write is interrupted to the index file, it will result in a line
1367
that does not end in ' :'. If the ' :' is not present at the end of a line,
1368
or at the end of the file, then the record that is missing it will be
1369
ignored by the parser.
1371
When writing new records to the index file, the data is preceded by '\n'
1372
to ensure that records always start on new lines even if the last write was
1373
interrupted. As a result its normal for the last line in the index to be
1374
missing a trailing newline. One can be added with no harmful effects.
1377
HEADER = "# bzr knit index 8\n"
1379
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1380
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1382
def _cache_version(self, version_id, options, pos, size, parents):
1383
"""Cache a version record in the history array and index cache.
1385
This is inlined into _load_data for performance. KEEP IN SYNC.
1386
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1389
# only want the _history index to reference the 1st index entry
1391
if version_id not in self._cache:
1392
index = len(self._history)
1393
self._history.append(version_id)
1395
index = self._cache[version_id][5]
1396
self._cache[version_id] = (version_id,
1403
def _check_write_ok(self):
1404
if self._get_scope() != self._scope:
1405
raise errors.OutSideTransaction()
1406
if self._mode != 'w':
1407
raise errors.ReadOnlyObjectDirtiedError(self)
1409
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1410
create_parent_dir=False, delay_create=False, dir_mode=None,
1412
_KnitComponentFile.__init__(self, transport, filename, mode,
1413
file_mode=file_mode,
1414
create_parent_dir=create_parent_dir,
1417
# position in _history is the 'official' index for a revision
1418
# but the values may have come from a newer entry.
1419
# so - wc -l of a knit index is != the number of unique names
1423
fp = self._transport.get(self._filename)
1425
# _load_data may raise NoSuchFile if the target knit is
1427
_load_data(self, fp)
1431
if mode != 'w' or not create:
1434
self._need_to_create = True
1436
self._transport.put_bytes_non_atomic(
1437
self._filename, self.HEADER, mode=self._file_mode)
1438
self._scope = get_scope()
1439
self._get_scope = get_scope
1441
def get_ancestry(self, versions, topo_sorted=True):
1442
"""See VersionedFile.get_ancestry."""
1443
# get a graph of all the mentioned versions:
1445
pending = set(versions)
1448
version = pending.pop()
1451
parents = [p for p in cache[version][4] if p in cache]
1453
raise RevisionNotPresent(version, self._filename)
1454
# if not completed and not a ghost
1455
pending.update([p for p in parents if p not in graph])
1456
graph[version] = parents
1459
return topo_sort(graph.items())
1461
def get_ancestry_with_ghosts(self, versions):
1462
"""See VersionedFile.get_ancestry_with_ghosts."""
1463
# get a graph of all the mentioned versions:
1464
self.check_versions_present(versions)
1467
pending = set(versions)
1469
version = pending.pop()
1471
parents = cache[version][4]
1477
pending.update([p for p in parents if p not in graph])
1478
graph[version] = parents
1479
return topo_sort(graph.items())
1481
def get_build_details(self, version_ids):
1482
"""Get the method, index_memo and compression parent for version_ids.
1484
Ghosts are omitted from the result.
1486
:param version_ids: An iterable of version_ids.
1487
:return: A dict of version_id:(index_memo, compression_parent,
1488
parents, record_details).
1490
opaque structure to pass to read_records to extract the raw
1493
Content that this record is built upon, may be None
1495
Logical parents of this node
1497
extra information about the content which needs to be passed to
1498
Factory.parse_record
1501
for version_id in version_ids:
1502
if version_id not in self._cache:
1503
# ghosts are omitted
1505
method = self.get_method(version_id)
1506
parents = self.get_parents_with_ghosts(version_id)
1507
if method == 'fulltext':
1508
compression_parent = None
1510
compression_parent = parents[0]
1511
noeol = 'no-eol' in self.get_options(version_id)
1512
index_memo = self.get_position(version_id)
1513
result[version_id] = (index_memo, compression_parent,
1514
parents, (method, noeol))
1517
def num_versions(self):
1518
return len(self._history)
1520
__len__ = num_versions
1522
def get_versions(self):
1523
"""Get all the versions in the file. not topologically sorted."""
1524
return self._history
1526
def _version_list_to_index(self, versions):
1529
for version in versions:
1530
if version in cache:
1531
# -- inlined lookup() --
1532
result_list.append(str(cache[version][5]))
1533
# -- end lookup () --
1535
result_list.append('.' + version)
1536
return ' '.join(result_list)
1538
def add_version(self, version_id, options, index_memo, parents):
1539
"""Add a version record to the index."""
1540
self.add_versions(((version_id, options, index_memo, parents),))
1542
def add_versions(self, versions, random_id=False):
1543
"""Add multiple versions to the index.
1545
:param versions: a list of tuples:
1546
(version_id, options, pos, size, parents).
1547
:param random_id: If True the ids being added were randomly generated
1548
and no check for existence will be performed.
1551
orig_history = self._history[:]
1552
orig_cache = self._cache.copy()
1555
for version_id, options, (index, pos, size), parents in versions:
1556
line = "\n%s %s %s %s %s :" % (version_id,
1560
self._version_list_to_index(parents))
1561
assert isinstance(line, str), \
1562
'content must be utf-8 encoded: %r' % (line,)
1564
self._cache_version(version_id, options, pos, size, tuple(parents))
1565
if not self._need_to_create:
1566
self._transport.append_bytes(self._filename, ''.join(lines))
1569
sio.write(self.HEADER)
1570
sio.writelines(lines)
1572
self._transport.put_file_non_atomic(self._filename, sio,
1573
create_parent_dir=self._create_parent_dir,
1574
mode=self._file_mode,
1575
dir_mode=self._dir_mode)
1576
self._need_to_create = False
1578
# If any problems happen, restore the original values and re-raise
1579
self._history = orig_history
1580
self._cache = orig_cache
1583
def has_version(self, version_id):
1584
"""True if the version is in the index."""
1585
return version_id in self._cache
1587
def get_position(self, version_id):
1588
"""Return details needed to access the version.
1590
.kndx indices do not support split-out data, so return None for the
1593
:return: a tuple (None, data position, size) to hand to the access
1594
logic to get the record.
1596
entry = self._cache[version_id]
1597
return None, entry[2], entry[3]
1599
def get_method(self, version_id):
1600
"""Return compression method of specified version."""
1602
options = self._cache[version_id][1]
1604
raise RevisionNotPresent(version_id, self._filename)
1605
if 'fulltext' in options:
1608
if 'line-delta' not in options:
1609
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1612
def get_options(self, version_id):
1613
"""Return a list representing options.
1617
return self._cache[version_id][1]
1619
def get_parent_map(self, version_ids):
1620
"""Passed through to by KnitVersionedFile.get_parent_map."""
1622
for version_id in version_ids:
1624
result[version_id] = tuple(self._cache[version_id][4])
1629
def get_parents_with_ghosts(self, version_id):
1630
"""Return parents of specified version with ghosts."""
1632
return self.get_parent_map([version_id])[version_id]
1634
raise RevisionNotPresent(version_id, self)
1636
def check_versions_present(self, version_ids):
1637
"""Check that all specified versions are present."""
1639
for version_id in version_ids:
1640
if version_id not in cache:
1641
raise RevisionNotPresent(version_id, self._filename)
1644
class KnitGraphIndex(object):
1645
"""A knit index that builds on GraphIndex."""
1647
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1648
"""Construct a KnitGraphIndex on a graph_index.
1650
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1651
:param deltas: Allow delta-compressed records.
1652
:param add_callback: If not None, allow additions to the index and call
1653
this callback with a list of added GraphIndex nodes:
1654
[(node, value, node_refs), ...]
1655
:param parents: If True, record knits parents, if not do not record
1658
self._graph_index = graph_index
1659
self._deltas = deltas
1660
self._add_callback = add_callback
1661
self._parents = parents
1662
if deltas and not parents:
1663
raise KnitCorrupt(self, "Cannot do delta compression without "
1666
def _check_write_ok(self):
1669
def _get_entries(self, keys, check_present=False):
1670
"""Get the entries for keys.
1672
:param keys: An iterable of index keys, - 1-tuples.
1677
for node in self._graph_index.iter_entries(keys):
1679
found_keys.add(node[1])
1681
# adapt parentless index to the rest of the code.
1682
for node in self._graph_index.iter_entries(keys):
1683
yield node[0], node[1], node[2], ()
1684
found_keys.add(node[1])
1686
missing_keys = keys.difference(found_keys)
1688
raise RevisionNotPresent(missing_keys.pop(), self)
1690
def _present_keys(self, version_ids):
1692
node[1] for node in self._get_entries(version_ids)])
1694
def _parentless_ancestry(self, versions):
1695
"""Honour the get_ancestry API for parentless knit indices."""
1696
wanted_keys = self._version_ids_to_keys(versions)
1697
present_keys = self._present_keys(wanted_keys)
1698
missing = set(wanted_keys).difference(present_keys)
1700
raise RevisionNotPresent(missing.pop(), self)
1701
return list(self._keys_to_version_ids(present_keys))
1703
def get_ancestry(self, versions, topo_sorted=True):
1704
"""See VersionedFile.get_ancestry."""
1705
if not self._parents:
1706
return self._parentless_ancestry(versions)
1707
# XXX: This will do len(history) index calls - perhaps
1708
# it should be altered to be a index core feature?
1709
# get a graph of all the mentioned versions:
1712
versions = self._version_ids_to_keys(versions)
1713
pending = set(versions)
1715
# get all pending nodes
1716
this_iteration = pending
1717
new_nodes = self._get_entries(this_iteration)
1720
for (index, key, value, node_refs) in new_nodes:
1721
# dont ask for ghosties - otherwise
1722
# we we can end up looping with pending
1723
# being entirely ghosted.
1724
graph[key] = [parent for parent in node_refs[0]
1725
if parent not in ghosts]
1727
for parent in graph[key]:
1728
# dont examine known nodes again
1733
ghosts.update(this_iteration.difference(found))
1734
if versions.difference(graph):
1735
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1737
result_keys = topo_sort(graph.items())
1739
result_keys = graph.iterkeys()
1740
return [key[0] for key in result_keys]
1742
def get_ancestry_with_ghosts(self, versions):
1743
"""See VersionedFile.get_ancestry."""
1744
if not self._parents:
1745
return self._parentless_ancestry(versions)
1746
# XXX: This will do len(history) index calls - perhaps
1747
# it should be altered to be a index core feature?
1748
# get a graph of all the mentioned versions:
1750
versions = self._version_ids_to_keys(versions)
1751
pending = set(versions)
1753
# get all pending nodes
1754
this_iteration = pending
1755
new_nodes = self._get_entries(this_iteration)
1757
for (index, key, value, node_refs) in new_nodes:
1758
graph[key] = node_refs[0]
1760
for parent in graph[key]:
1761
# dont examine known nodes again
1765
missing_versions = this_iteration.difference(graph)
1766
missing_needed = versions.intersection(missing_versions)
1768
raise RevisionNotPresent(missing_needed.pop(), self)
1769
for missing_version in missing_versions:
1770
# add a key, no parents
1771
graph[missing_version] = []
1772
pending.discard(missing_version) # don't look for it
1773
result_keys = topo_sort(graph.items())
1774
return [key[0] for key in result_keys]
1776
def get_build_details(self, version_ids):
1777
"""Get the method, index_memo and compression parent for version_ids.
1779
Ghosts are omitted from the result.
1781
:param version_ids: An iterable of version_ids.
1782
:return: A dict of version_id:(index_memo, compression_parent,
1783
parents, record_details).
1785
opaque structure to pass to read_records to extract the raw
1788
Content that this record is built upon, may be None
1790
Logical parents of this node
1792
extra information about the content which needs to be passed to
1793
Factory.parse_record
1796
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
1797
for entry in entries:
1798
version_id = self._keys_to_version_ids((entry[1],))[0]
1799
if not self._parents:
1802
parents = self._keys_to_version_ids(entry[3][0])
1803
if not self._deltas:
1804
compression_parent = None
1806
compression_parent_key = self._compression_parent(entry)
1807
if compression_parent_key:
1808
compression_parent = self._keys_to_version_ids(
1809
(compression_parent_key,))[0]
1811
compression_parent = None
1812
noeol = (entry[2][0] == 'N')
1813
if compression_parent:
1814
method = 'line-delta'
1817
result[version_id] = (self._node_to_position(entry),
1818
compression_parent, parents,
1822
def _compression_parent(self, an_entry):
1823
# return the key that an_entry is compressed against, or None
1824
# Grab the second parent list (as deltas implies parents currently)
1825
compression_parents = an_entry[3][1]
1826
if not compression_parents:
1828
assert len(compression_parents) == 1
1829
return compression_parents[0]
1831
def _get_method(self, node):
1832
if not self._deltas:
1834
if self._compression_parent(node):
1839
def num_versions(self):
1840
return len(list(self._graph_index.iter_all_entries()))
1842
__len__ = num_versions
1844
def get_versions(self):
1845
"""Get all the versions in the file. not topologically sorted."""
1846
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1848
def has_version(self, version_id):
1849
"""True if the version is in the index."""
1850
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1852
def _keys_to_version_ids(self, keys):
1853
return tuple(key[0] for key in keys)
1855
def get_position(self, version_id):
1856
"""Return details needed to access the version.
1858
:return: a tuple (index, data position, size) to hand to the access
1859
logic to get the record.
1861
node = self._get_node(version_id)
1862
return self._node_to_position(node)
1864
def _node_to_position(self, node):
1865
"""Convert an index value to position details."""
1866
bits = node[2][1:].split(' ')
1867
return node[0], int(bits[0]), int(bits[1])
1869
def get_method(self, version_id):
1870
"""Return compression method of specified version."""
1871
return self._get_method(self._get_node(version_id))
1873
def _get_node(self, version_id):
1875
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1877
raise RevisionNotPresent(version_id, self)
1879
def get_options(self, version_id):
1880
"""Return a list representing options.
1884
node = self._get_node(version_id)
1885
options = [self._get_method(node)]
1886
if node[2][0] == 'N':
1887
options.append('no-eol')
1890
def get_parent_map(self, version_ids):
1891
"""Passed through to by KnitVersionedFile.get_parent_map."""
1892
nodes = self._get_entries(self._version_ids_to_keys(version_ids))
1896
result[node[1][0]] = self._keys_to_version_ids(node[3][0])
1899
result[node[1][0]] = ()
1902
def get_parents_with_ghosts(self, version_id):
1903
"""Return parents of specified version with ghosts."""
1905
return self.get_parent_map([version_id])[version_id]
1907
raise RevisionNotPresent(version_id, self)
1909
def check_versions_present(self, version_ids):
1910
"""Check that all specified versions are present."""
1911
keys = self._version_ids_to_keys(version_ids)
1912
present = self._present_keys(keys)
1913
missing = keys.difference(present)
1915
raise RevisionNotPresent(missing.pop(), self)
1917
def add_version(self, version_id, options, access_memo, parents):
1918
"""Add a version record to the index."""
1919
return self.add_versions(((version_id, options, access_memo, parents),))
1921
def add_versions(self, versions, random_id=False):
1922
"""Add multiple versions to the index.
1924
This function does not insert data into the Immutable GraphIndex
1925
backing the KnitGraphIndex, instead it prepares data for insertion by
1926
the caller and checks that it is safe to insert then calls
1927
self._add_callback with the prepared GraphIndex nodes.
1929
:param versions: a list of tuples:
1930
(version_id, options, pos, size, parents).
1931
:param random_id: If True the ids being added were randomly generated
1932
and no check for existence will be performed.
1934
if not self._add_callback:
1935
raise errors.ReadOnlyError(self)
1936
# we hope there are no repositories with inconsistent parentage
1941
for (version_id, options, access_memo, parents) in versions:
1942
index, pos, size = access_memo
1943
key = (version_id, )
1944
parents = tuple((parent, ) for parent in parents)
1945
if 'no-eol' in options:
1949
value += "%d %d" % (pos, size)
1950
if not self._deltas:
1951
if 'line-delta' in options:
1952
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1955
if 'line-delta' in options:
1956
node_refs = (parents, (parents[0],))
1958
node_refs = (parents, ())
1960
node_refs = (parents, )
1963
raise KnitCorrupt(self, "attempt to add node with parents "
1964
"in parentless index.")
1966
keys[key] = (value, node_refs)
1968
present_nodes = self._get_entries(keys)
1969
for (index, key, value, node_refs) in present_nodes:
1970
if (value, node_refs) != keys[key]:
1971
raise KnitCorrupt(self, "inconsistent details in add_versions"
1972
": %s %s" % ((value, node_refs), keys[key]))
1976
for key, (value, node_refs) in keys.iteritems():
1977
result.append((key, value, node_refs))
1979
for key, (value, node_refs) in keys.iteritems():
1980
result.append((key, value))
1981
self._add_callback(result)
1983
def _version_ids_to_keys(self, version_ids):
1984
return set((version_id, ) for version_id in version_ids)
1987
class _KnitAccess(object):
1988
"""Access to knit records in a .knit file."""
1990
def __init__(self, transport, filename, _file_mode, _dir_mode,
1991
_need_to_create, _create_parent_dir):
1992
"""Create a _KnitAccess for accessing and inserting data.
1994
:param transport: The transport the .knit is located on.
1995
:param filename: The filename of the .knit.
1997
self._transport = transport
1998
self._filename = filename
1999
self._file_mode = _file_mode
2000
self._dir_mode = _dir_mode
2001
self._need_to_create = _need_to_create
2002
self._create_parent_dir = _create_parent_dir
2004
def add_raw_records(self, sizes, raw_data):
2005
"""Add raw knit bytes to a storage area.
2007
The data is spooled to whereever the access method is storing data.
2009
:param sizes: An iterable containing the size of each raw data segment.
2010
:param raw_data: A bytestring containing the data.
2011
:return: A list of memos to retrieve the record later. Each memo is a
2012
tuple - (index, pos, length), where the index field is always None
2013
for the .knit access method.
2015
assert type(raw_data) == str, \
2016
'data must be plain bytes was %s' % type(raw_data)
2017
if not self._need_to_create:
2018
base = self._transport.append_bytes(self._filename, raw_data)
2020
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2021
create_parent_dir=self._create_parent_dir,
2022
mode=self._file_mode,
2023
dir_mode=self._dir_mode)
2024
self._need_to_create = False
2028
result.append((None, base, size))
2033
"""IFF this data access has its own storage area, initialise it.
2037
self._transport.put_bytes_non_atomic(self._filename, '',
2038
mode=self._file_mode)
2040
def open_file(self):
2041
"""IFF this data access can be represented as a single file, open it.
2043
For knits that are not mapped to a single file on disk this will
2046
:return: None or a file handle.
2049
return self._transport.get(self._filename)
2054
def get_raw_records(self, memos_for_retrieval):
2055
"""Get the raw bytes for a records.
2057
:param memos_for_retrieval: An iterable containing the (index, pos,
2058
length) memo for retrieving the bytes. The .knit method ignores
2059
the index as there is always only a single file.
2060
:return: An iterator over the bytes of the records.
2062
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2063
for pos, data in self._transport.readv(self._filename, read_vector):
2067
class _PackAccess(object):
2068
"""Access to knit records via a collection of packs."""
2070
def __init__(self, index_to_packs, writer=None):
2071
"""Create a _PackAccess object.
2073
:param index_to_packs: A dict mapping index objects to the transport
2074
and file names for obtaining data.
2075
:param writer: A tuple (pack.ContainerWriter, write_index) which
2076
contains the pack to write, and the index that reads from it will
2080
self.container_writer = writer[0]
2081
self.write_index = writer[1]
2083
self.container_writer = None
2084
self.write_index = None
2085
self.indices = index_to_packs
2087
def add_raw_records(self, sizes, raw_data):
2088
"""Add raw knit bytes to a storage area.
2090
The data is spooled to the container writer in one bytes-record per
2093
:param sizes: An iterable containing the size of each raw data segment.
2094
:param raw_data: A bytestring containing the data.
2095
:return: A list of memos to retrieve the record later. Each memo is a
2096
tuple - (index, pos, length), where the index field is the
2097
write_index object supplied to the PackAccess object.
2099
assert type(raw_data) == str, \
2100
'data must be plain bytes was %s' % type(raw_data)
2104
p_offset, p_length = self.container_writer.add_bytes_record(
2105
raw_data[offset:offset+size], [])
2107
result.append((self.write_index, p_offset, p_length))
2111
"""Pack based knits do not get individually created."""
2113
def get_raw_records(self, memos_for_retrieval):
2114
"""Get the raw bytes for a records.
2116
:param memos_for_retrieval: An iterable containing the (index, pos,
2117
length) memo for retrieving the bytes. The Pack access method
2118
looks up the pack to use for a given record in its index_to_pack
2120
:return: An iterator over the bytes of the records.
2122
# first pass, group into same-index requests
2124
current_index = None
2125
for (index, offset, length) in memos_for_retrieval:
2126
if current_index == index:
2127
current_list.append((offset, length))
2129
if current_index is not None:
2130
request_lists.append((current_index, current_list))
2131
current_index = index
2132
current_list = [(offset, length)]
2133
# handle the last entry
2134
if current_index is not None:
2135
request_lists.append((current_index, current_list))
2136
for index, offsets in request_lists:
2137
transport, path = self.indices[index]
2138
reader = pack.make_readv_reader(transport, path, offsets)
2139
for names, read_func in reader.iter_records():
2140
yield read_func(None)
2142
def open_file(self):
2143
"""Pack based knits have no single file."""
2146
def set_writer(self, writer, index, (transport, packname)):
2147
"""Set a writer to use for adding data."""
2148
if index is not None:
2149
self.indices[index] = (transport, packname)
2150
self.container_writer = writer
2151
self.write_index = index
2154
class _StreamAccess(object):
2155
"""A Knit Access object that provides data from a datastream.
2157
It also provides a fallback to present as unannotated data, annotated data
2158
from a *backing* access object.
2160
This is triggered by a index_memo which is pointing to a different index
2161
than this was constructed with, and is used to allow extracting full
2162
unannotated texts for insertion into annotated knits.
2165
def __init__(self, reader_callable, stream_index, backing_knit,
2167
"""Create a _StreamAccess object.
2169
:param reader_callable: The reader_callable from the datastream.
2170
This is called to buffer all the data immediately, for
2172
:param stream_index: The index the data stream this provides access to
2173
which will be present in native index_memo's.
2174
:param backing_knit: The knit object that will provide access to
2175
annotated texts which are not available in the stream, so as to
2176
create unannotated texts.
2177
:param orig_factory: The original content factory used to generate the
2178
stream. This is used for checking whether the thunk code for
2179
supporting _copy_texts will generate the correct form of data.
2181
self.data = reader_callable(None)
2182
self.stream_index = stream_index
2183
self.backing_knit = backing_knit
2184
self.orig_factory = orig_factory
2186
def get_raw_records(self, memos_for_retrieval):
2187
"""Get the raw bytes for a records.
2189
:param memos_for_retrieval: An iterable containing the (thunk_flag,
2190
index, start, end) memo for retrieving the bytes.
2191
:return: An iterator over the bytes of the records.
2193
# use a generator for memory friendliness
2194
for thunk_flag, version_id, start, end in memos_for_retrieval:
2195
if version_id is self.stream_index:
2196
yield self.data[start:end]
2198
# we have been asked to thunk. This thunking only occurs when
2199
# we are obtaining plain texts from an annotated backing knit
2200
# so that _copy_texts will work.
2201
# We could improve performance here by scanning for where we need
2202
# to do this and using get_line_list, then interleaving the output
2203
# as desired. However, for now, this is sufficient.
2204
if self.orig_factory.__class__ != KnitPlainFactory:
2205
raise errors.KnitCorrupt(
2206
self, 'Bad thunk request %r' % version_id)
2207
lines = self.backing_knit.get_lines(version_id)
2208
line_bytes = ''.join(lines)
2209
digest = sha_string(line_bytes)
2211
if lines[-1][-1] != '\n':
2212
lines[-1] = lines[-1] + '\n'
2214
orig_options = list(self.backing_knit._index.get_options(version_id))
2215
if 'fulltext' not in orig_options:
2216
if 'line-delta' not in orig_options:
2217
raise errors.KnitCorrupt(self,
2218
'Unknown compression method %r' % orig_options)
2219
orig_options.remove('line-delta')
2220
orig_options.append('fulltext')
2221
# We want plain data, because we expect to thunk only to allow text
2223
size, bytes = self.backing_knit._data._record_to_data(version_id,
2224
digest, lines, line_bytes)
2228
class _StreamIndex(object):
2229
"""A Knit Index object that uses the data map from a datastream."""
2231
def __init__(self, data_list, backing_index):
2232
"""Create a _StreamIndex object.
2234
:param data_list: The data_list from the datastream.
2235
:param backing_index: The index which will supply values for nodes
2236
referenced outside of this stream.
2238
self.data_list = data_list
2239
self.backing_index = backing_index
2240
self._by_version = {}
2242
for key, options, length, parents in data_list:
2243
self._by_version[key] = options, (pos, pos + length), parents
2246
def get_ancestry(self, versions, topo_sorted):
2247
"""Get an ancestry list for versions."""
2249
# Not needed for basic joins
2250
raise NotImplementedError(self.get_ancestry)
2251
# get a graph of all the mentioned versions:
2252
# Little ugly - basically copied from KnitIndex, but don't want to
2253
# accidentally incorporate too much of that index's code.
2255
pending = set(versions)
2256
cache = self._by_version
2258
version = pending.pop()
2261
parents = [p for p in cache[version][2] if p in cache]
2263
raise RevisionNotPresent(version, self)
2264
# if not completed and not a ghost
2265
pending.update([p for p in parents if p not in ancestry])
2266
ancestry.add(version)
2267
return list(ancestry)
2269
def get_build_details(self, version_ids):
2270
"""Get the method, index_memo and compression parent for version_ids.
2272
Ghosts are omitted from the result.
2274
:param version_ids: An iterable of version_ids.
2275
:return: A dict of version_id:(index_memo, compression_parent,
2276
parents, record_details).
2278
opaque structure to pass to read_records to extract the raw
2281
Content that this record is built upon, may be None
2283
Logical parents of this node
2285
extra information about the content which needs to be passed to
2286
Factory.parse_record
2289
for version_id in version_ids:
2291
method = self.get_method(version_id)
2292
except errors.RevisionNotPresent:
2293
# ghosts are omitted
2295
parent_ids = self.get_parents_with_ghosts(version_id)
2296
noeol = ('no-eol' in self.get_options(version_id))
2297
if method == 'fulltext':
2298
compression_parent = None
2300
compression_parent = parent_ids[0]
2301
index_memo = self.get_position(version_id)
2302
result[version_id] = (index_memo, compression_parent,
2303
parent_ids, (method, noeol))
2306
def get_method(self, version_id):
2307
"""Return compression method of specified version."""
2309
options = self._by_version[version_id][0]
2311
# Strictly speaking this should check in the backing knit, but
2312
# until we have a test to discriminate, this will do.
2313
return self.backing_index.get_method(version_id)
2314
if 'fulltext' in options:
2316
elif 'line-delta' in options:
2319
raise errors.KnitIndexUnknownMethod(self, options)
2321
def get_options(self, version_id):
2322
"""Return a list representing options.
2327
return self._by_version[version_id][0]
2329
return self.backing_index.get_options(version_id)
2331
def get_parent_map(self, version_ids):
2332
"""Passed through to by KnitVersionedFile.get_parent_map."""
2335
for version_id in version_ids:
2337
result[version_id] = self._by_version[version_id][2]
2339
pending_ids.add(version_id)
2340
result.update(self.backing_index.get_parent_map(pending_ids))
2343
def get_parents_with_ghosts(self, version_id):
2344
"""Return parents of specified version with ghosts."""
2346
return self.get_parent_map([version_id])[version_id]
2348
raise RevisionNotPresent(version_id, self)
2350
def get_position(self, version_id):
2351
"""Return details needed to access the version.
2353
_StreamAccess has the data as a big array, so we return slice
2354
coordinates into that (as index_memo's are opaque outside the
2355
index and matching access class).
2357
:return: a tuple (thunk_flag, index, start, end). If thunk_flag is
2358
False, index will be self, otherwise it will be a version id.
2361
start, end = self._by_version[version_id][1]
2362
return False, self, start, end
2364
# Signal to the access object to handle this from the backing knit.
2365
return (True, version_id, None, None)
2367
def get_versions(self):
2368
"""Get all the versions in the stream."""
2369
return self._by_version.keys()
2372
class _KnitData(object):
2373
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2375
The KnitData class provides the logic for parsing and using knit records,
2376
making use of an access method for the low level read and write operations.
2379
def __init__(self, access):
2380
"""Create a KnitData object.
2382
:param access: The access method to use. Access methods such as
2383
_KnitAccess manage the insertion of raw records and the subsequent
2384
retrieval of the same.
2386
self._access = access
2387
self._checked = False
2388
# TODO: jam 20060713 conceptually, this could spill to disk
2389
# if the cached size gets larger than a certain amount
2390
# but it complicates the model a bit, so for now just use
2391
# a simple dictionary
2393
self._do_cache = False
2395
def enable_cache(self):
2396
"""Enable caching of reads."""
2397
self._do_cache = True
2399
def clear_cache(self):
2400
"""Clear the record cache."""
2401
self._do_cache = False
2404
def _open_file(self):
2405
return self._access.open_file()
2407
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2408
"""Convert version_id, digest, lines into a raw data block.
2410
:param dense_lines: The bytes of lines but in a denser form. For
2411
instance, if lines is a list of 1000 bytestrings each ending in \n,
2412
dense_lines may be a list with one line in it, containing all the
2413
1000's lines and their \n's. Using dense_lines if it is already
2414
known is a win because the string join to create bytes in this
2415
function spends less time resizing the final string.
2416
:return: (len, a StringIO instance with the raw data ready to read.)
2418
# Note: using a string copy here increases memory pressure with e.g.
2419
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2420
# when doing the initial commit of a mozilla tree. RBC 20070921
2421
bytes = ''.join(chain(
2422
["version %s %d %s\n" % (version_id,
2425
dense_lines or lines,
2426
["end %s\n" % version_id]))
2427
assert bytes.__class__ == str
2428
compressed_bytes = bytes_to_gzip(bytes)
2429
return len(compressed_bytes), compressed_bytes
2431
def add_raw_records(self, sizes, raw_data):
2432
"""Append a prepared record to the data file.
2434
:param sizes: An iterable containing the size of each raw data segment.
2435
:param raw_data: A bytestring containing the data.
2436
:return: a list of index data for the way the data was stored.
2437
See the access method add_raw_records documentation for more
2440
return self._access.add_raw_records(sizes, raw_data)
2442
def _parse_record_header(self, version_id, raw_data):
2443
"""Parse a record header for consistency.
2445
:return: the header and the decompressor stream.
2446
as (stream, header_record)
2448
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2450
rec = self._check_header(version_id, df.readline())
2451
except Exception, e:
2452
raise KnitCorrupt(self._access,
2453
"While reading {%s} got %s(%s)"
2454
% (version_id, e.__class__.__name__, str(e)))
2457
def _check_header(self, version_id, line):
2460
raise KnitCorrupt(self._access,
2461
'unexpected number of elements in record header')
2462
if rec[1] != version_id:
2463
raise KnitCorrupt(self._access,
2464
'unexpected version, wanted %r, got %r'
2465
% (version_id, rec[1]))
2468
def _parse_record(self, version_id, data):
2470
# 4168 calls in 2880 217 internal
2471
# 4168 calls to _parse_record_header in 2121
2472
# 4168 calls to readlines in 330
2473
df = GzipFile(mode='rb', fileobj=StringIO(data))
2476
record_contents = df.readlines()
2477
except Exception, e:
2478
raise KnitCorrupt(self._access,
2479
"While reading {%s} got %s(%s)"
2480
% (version_id, e.__class__.__name__, str(e)))
2481
header = record_contents.pop(0)
2482
rec = self._check_header(version_id, header)
2484
last_line = record_contents.pop()
2485
if len(record_contents) != int(rec[2]):
2486
raise KnitCorrupt(self._access,
2487
'incorrect number of lines %s != %s'
2489
% (len(record_contents), int(rec[2]),
2491
if last_line != 'end %s\n' % rec[1]:
2492
raise KnitCorrupt(self._access,
2493
'unexpected version end line %r, wanted %r'
2494
% (last_line, version_id))
2496
return record_contents, rec[3]
2498
def read_records_iter_raw(self, records):
2499
"""Read text records from data file and yield raw data.
2501
This unpacks enough of the text record to validate the id is
2502
as expected but thats all.
2504
# setup an iterator of the external records:
2505
# uses readv so nice and fast we hope.
2507
# grab the disk data needed.
2509
# Don't check _cache if it is empty
2510
needed_offsets = [index_memo for version_id, index_memo
2512
if version_id not in self._cache]
2514
needed_offsets = [index_memo for version_id, index_memo
2517
raw_records = self._access.get_raw_records(needed_offsets)
2519
for version_id, index_memo in records:
2520
if version_id in self._cache:
2521
# This data has already been validated
2522
data = self._cache[version_id]
2524
data = raw_records.next()
2526
self._cache[version_id] = data
2528
# validate the header
2529
df, rec = self._parse_record_header(version_id, data)
2531
yield version_id, data
2533
def read_records_iter(self, records):
2534
"""Read text records from data file and yield result.
2536
The result will be returned in whatever is the fastest to read.
2537
Not by the order requested. Also, multiple requests for the same
2538
record will only yield 1 response.
2539
:param records: A list of (version_id, pos, len) entries
2540
:return: Yields (version_id, contents, digest) in the order
2541
read, not the order requested
2547
# Skip records we have alread seen
2548
yielded_records = set()
2549
needed_records = set()
2550
for record in records:
2551
if record[0] in self._cache:
2552
if record[0] in yielded_records:
2554
yielded_records.add(record[0])
2555
data = self._cache[record[0]]
2556
content, digest = self._parse_record(record[0], data)
2557
yield (record[0], content, digest)
2559
needed_records.add(record)
2560
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2562
needed_records = sorted(set(records), key=operator.itemgetter(1))
2564
if not needed_records:
2567
# The transport optimizes the fetching as well
2568
# (ie, reads continuous ranges.)
2569
raw_data = self._access.get_raw_records(
2570
[index_memo for version_id, index_memo in needed_records])
2572
for (version_id, index_memo), data in \
2573
izip(iter(needed_records), raw_data):
2574
content, digest = self._parse_record(version_id, data)
2576
self._cache[version_id] = data
2577
yield version_id, content, digest
2579
def read_records(self, records):
2580
"""Read records into a dictionary."""
2582
for record_id, content, digest in \
2583
self.read_records_iter(records):
2584
components[record_id] = (content, digest)
2588
class InterKnit(InterVersionedFile):
2589
"""Optimised code paths for knit to knit operations."""
2591
_matching_file_from_factory = staticmethod(make_file_knit)
2592
_matching_file_to_factory = staticmethod(make_file_knit)
2595
def is_compatible(source, target):
2596
"""Be compatible with knits. """
2598
return (isinstance(source, KnitVersionedFile) and
2599
isinstance(target, KnitVersionedFile))
2600
except AttributeError:
2603
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2604
"""Copy texts to the target by extracting and adding them one by one.
2606
see join() for the parameter definitions.
2608
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2609
# --- the below is factorable out with VersionedFile.join, but wait for
2610
# VersionedFiles, it may all be simpler then.
2611
graph = Graph(self.source)
2612
search = graph._make_breadth_first_searcher(version_ids)
2613
transitive_ids = set()
2614
map(transitive_ids.update, list(search))
2615
parent_map = self.source.get_parent_map(transitive_ids)
2616
order = topo_sort(parent_map.items())
2618
def size_of_content(content):
2619
return sum(len(line) for line in content.text())
2620
# Cache at most 10MB of parent texts
2621
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2622
compute_size=size_of_content)
2623
# TODO: jam 20071116 It would be nice to have a streaming interface to
2624
# get multiple texts from a source. The source could be smarter
2625
# about how it handled intermediate stages.
2626
# get_line_list() or make_mpdiffs() seem like a possibility, but
2627
# at the moment they extract all full texts into memory, which
2628
# causes us to store more than our 3x fulltext goal.
2629
# Repository.iter_files_bytes() may be another possibility
2630
to_process = [version for version in order
2631
if version not in self.target]
2632
total = len(to_process)
2633
pb = ui.ui_factory.nested_progress_bar()
2635
for index, version in enumerate(to_process):
2636
pb.update('Converting versioned data', index, total)
2637
sha1, num_bytes, parent_text = self.target.add_lines(version,
2638
self.source.get_parents_with_ghosts(version),
2639
self.source.get_lines(version),
2640
parent_texts=parent_cache)
2641
parent_cache[version] = parent_text
2646
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2647
"""See InterVersionedFile.join."""
2648
assert isinstance(self.source, KnitVersionedFile)
2649
assert isinstance(self.target, KnitVersionedFile)
2651
# If the source and target are mismatched w.r.t. annotations vs
2652
# plain, the data needs to be converted accordingly
2653
if self.source.factory.annotated == self.target.factory.annotated:
2655
elif self.source.factory.annotated:
2656
converter = self._anno_to_plain_converter
2658
# We're converting from a plain to an annotated knit. Copy them
2659
# across by full texts.
2660
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2662
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2666
pb = ui.ui_factory.nested_progress_bar()
2668
version_ids = list(version_ids)
2669
if None in version_ids:
2670
version_ids.remove(None)
2672
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2674
this_versions = set(self.target._index.get_versions())
2675
# XXX: For efficiency we should not look at the whole index,
2676
# we only need to consider the referenced revisions - they
2677
# must all be present, or the method must be full-text.
2678
# TODO, RBC 20070919
2679
needed_versions = self.source_ancestry - this_versions
2681
if not needed_versions:
2683
full_list = topo_sort(
2684
self.source.get_parent_map(self.source.versions()))
2686
version_list = [i for i in full_list if (not self.target.has_version(i)
2687
and i in needed_versions)]
2691
copy_queue_records = []
2693
for version_id in version_list:
2694
options = self.source._index.get_options(version_id)
2695
parents = self.source._index.get_parents_with_ghosts(version_id)
2696
# check that its will be a consistent copy:
2697
for parent in parents:
2698
# if source has the parent, we must :
2699
# * already have it or
2700
# * have it scheduled already
2701
# otherwise we don't care
2702
assert (self.target.has_version(parent) or
2703
parent in copy_set or
2704
not self.source.has_version(parent))
2705
index_memo = self.source._index.get_position(version_id)
2706
copy_queue_records.append((version_id, index_memo))
2707
copy_queue.append((version_id, options, parents))
2708
copy_set.add(version_id)
2710
# data suck the join:
2712
total = len(version_list)
2715
for (version_id, raw_data), \
2716
(version_id2, options, parents) in \
2717
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2719
assert version_id == version_id2, 'logic error, inconsistent results'
2721
pb.update("Joining knit", count, total)
2723
size, raw_data = converter(raw_data, version_id, options,
2726
size = len(raw_data)
2727
raw_records.append((version_id, options, parents, size))
2728
raw_datum.append(raw_data)
2729
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2734
def _anno_to_plain_converter(self, raw_data, version_id, options,
2736
"""Convert annotated content to plain content."""
2737
data, digest = self.source._data._parse_record(version_id, raw_data)
2738
if 'fulltext' in options:
2739
content = self.source.factory.parse_fulltext(data, version_id)
2740
lines = self.target.factory.lower_fulltext(content)
2742
delta = self.source.factory.parse_line_delta(data, version_id,
2744
lines = self.target.factory.lower_line_delta(delta)
2745
return self.target._data._record_to_data(version_id, digest, lines)
2748
InterVersionedFile.register_optimiser(InterKnit)
2751
class WeaveToKnit(InterVersionedFile):
2752
"""Optimised code paths for weave to knit operations."""
2754
_matching_file_from_factory = bzrlib.weave.WeaveFile
2755
_matching_file_to_factory = staticmethod(make_file_knit)
2758
def is_compatible(source, target):
2759
"""Be compatible with weaves to knits."""
2761
return (isinstance(source, bzrlib.weave.Weave) and
2762
isinstance(target, KnitVersionedFile))
2763
except AttributeError:
2766
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2767
"""See InterVersionedFile.join."""
2768
assert isinstance(self.source, bzrlib.weave.Weave)
2769
assert isinstance(self.target, KnitVersionedFile)
2771
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2776
pb = ui.ui_factory.nested_progress_bar()
2778
version_ids = list(version_ids)
2780
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2781
this_versions = set(self.target._index.get_versions())
2782
needed_versions = self.source_ancestry - this_versions
2784
if not needed_versions:
2786
full_list = topo_sort(
2787
self.source.get_parent_map(self.source.versions()))
2789
version_list = [i for i in full_list if (not self.target.has_version(i)
2790
and i in needed_versions)]
2794
total = len(version_list)
2795
parent_map = self.source.get_parent_map(version_list)
2796
for version_id in version_list:
2797
pb.update("Converting to knit", count, total)
2798
parents = parent_map[version_id]
2799
# check that its will be a consistent copy:
2800
for parent in parents:
2801
# if source has the parent, we must already have it
2802
assert (self.target.has_version(parent))
2803
self.target.add_lines(
2804
version_id, parents, self.source.get_lines(version_id))
2811
InterVersionedFile.register_optimiser(WeaveToKnit)
2814
# Deprecated, use PatienceSequenceMatcher instead
2815
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2818
def annotate_knit(knit, revision_id):
2819
"""Annotate a knit with no cached annotations.
2821
This implementation is for knits with no cached annotations.
2822
It will work for knits with cached annotations, but this is not
2825
annotator = _KnitAnnotator(knit)
2826
return iter(annotator.annotate(revision_id))
2829
class _KnitAnnotator(object):
2830
"""Build up the annotations for a text."""
2832
def __init__(self, knit):
2835
# Content objects, differs from fulltexts because of how final newlines
2836
# are treated by knits. the content objects here will always have a
2838
self._fulltext_contents = {}
2840
# Annotated lines of specific revisions
2841
self._annotated_lines = {}
2843
# Track the raw data for nodes that we could not process yet.
2844
# This maps the revision_id of the base to a list of children that will
2845
# annotated from it.
2846
self._pending_children = {}
2848
# Nodes which cannot be extracted
2849
self._ghosts = set()
2851
# Track how many children this node has, so we know if we need to keep
2853
self._annotate_children = {}
2854
self._compression_children = {}
2856
self._all_build_details = {}
2857
# The children => parent revision_id graph
2858
self._revision_id_graph = {}
2860
self._heads_provider = None
2862
self._nodes_to_keep_annotations = set()
2863
self._generations_until_keep = 100
2865
def set_generations_until_keep(self, value):
2866
"""Set the number of generations before caching a node.
2868
Setting this to -1 will cache every merge node, setting this higher
2869
will cache fewer nodes.
2871
self._generations_until_keep = value
2873
def _add_fulltext_content(self, revision_id, content_obj):
2874
self._fulltext_contents[revision_id] = content_obj
2875
# TODO: jam 20080305 It might be good to check the sha1digest here
2876
return content_obj.text()
2878
def _check_parents(self, child, nodes_to_annotate):
2879
"""Check if all parents have been processed.
2881
:param child: A tuple of (rev_id, parents, raw_content)
2882
:param nodes_to_annotate: If child is ready, add it to
2883
nodes_to_annotate, otherwise put it back in self._pending_children
2885
for parent_id in child[1]:
2886
if (parent_id not in self._annotated_lines):
2887
# This parent is present, but another parent is missing
2888
self._pending_children.setdefault(parent_id,
2892
# This one is ready to be processed
2893
nodes_to_annotate.append(child)
2895
def _add_annotation(self, revision_id, fulltext, parent_ids,
2896
left_matching_blocks=None):
2897
"""Add an annotation entry.
2899
All parents should already have been annotated.
2900
:return: A list of children that now have their parents satisfied.
2902
a = self._annotated_lines
2903
annotated_parent_lines = [a[p] for p in parent_ids]
2904
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2905
fulltext, revision_id, left_matching_blocks,
2906
heads_provider=self._get_heads_provider()))
2907
self._annotated_lines[revision_id] = annotated_lines
2908
for p in parent_ids:
2909
ann_children = self._annotate_children[p]
2910
ann_children.remove(revision_id)
2911
if (not ann_children
2912
and p not in self._nodes_to_keep_annotations):
2913
del self._annotated_lines[p]
2914
del self._all_build_details[p]
2915
if p in self._fulltext_contents:
2916
del self._fulltext_contents[p]
2917
# Now that we've added this one, see if there are any pending
2918
# deltas to be done, certainly this parent is finished
2919
nodes_to_annotate = []
2920
for child in self._pending_children.pop(revision_id, []):
2921
self._check_parents(child, nodes_to_annotate)
2922
return nodes_to_annotate
2924
def _get_build_graph(self, revision_id):
2925
"""Get the graphs for building texts and annotations.
2927
The data you need for creating a full text may be different than the
2928
data you need to annotate that text. (At a minimum, you need both
2929
parents to create an annotation, but only need 1 parent to generate the
2932
:return: A list of (revision_id, index_memo) records, suitable for
2933
passing to read_records_iter to start reading in the raw data fro/
2936
if revision_id in self._annotated_lines:
2939
pending = set([revision_id])
2944
# get all pending nodes
2946
this_iteration = pending
2947
build_details = self._knit._index.get_build_details(this_iteration)
2948
self._all_build_details.update(build_details)
2949
# new_nodes = self._knit._index._get_entries(this_iteration)
2951
for rev_id, details in build_details.iteritems():
2952
(index_memo, compression_parent, parents,
2953
record_details) = details
2954
self._revision_id_graph[rev_id] = parents
2955
records.append((rev_id, index_memo))
2956
# Do we actually need to check _annotated_lines?
2957
pending.update(p for p in parents
2958
if p not in self._all_build_details)
2959
if compression_parent:
2960
self._compression_children.setdefault(compression_parent,
2963
for parent in parents:
2964
self._annotate_children.setdefault(parent,
2966
num_gens = generation - kept_generation
2967
if ((num_gens >= self._generations_until_keep)
2968
and len(parents) > 1):
2969
kept_generation = generation
2970
self._nodes_to_keep_annotations.add(rev_id)
2972
missing_versions = this_iteration.difference(build_details.keys())
2973
self._ghosts.update(missing_versions)
2974
for missing_version in missing_versions:
2975
# add a key, no parents
2976
self._revision_id_graph[missing_version] = ()
2977
pending.discard(missing_version) # don't look for it
2978
# XXX: This should probably be a real exception, as it is a data
2980
assert not self._ghosts.intersection(self._compression_children), \
2981
"We cannot have nodes which have a compression parent of a ghost."
2982
# Cleanout anything that depends on a ghost so that we don't wait for
2983
# the ghost to show up
2984
for node in self._ghosts:
2985
if node in self._annotate_children:
2986
# We won't be building this node
2987
del self._annotate_children[node]
2988
# Generally we will want to read the records in reverse order, because
2989
# we find the parent nodes after the children
2993
def _annotate_records(self, records):
2994
"""Build the annotations for the listed records."""
2995
# We iterate in the order read, rather than a strict order requested
2996
# However, process what we can, and put off to the side things that
2997
# still need parents, cleaning them up when those parents are
2999
for (rev_id, record,
3000
digest) in self._knit._data.read_records_iter(records):
3001
if rev_id in self._annotated_lines:
3003
parent_ids = self._revision_id_graph[rev_id]
3004
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3005
details = self._all_build_details[rev_id]
3006
(index_memo, compression_parent, parents,
3007
record_details) = details
3008
nodes_to_annotate = []
3009
# TODO: Remove the punning between compression parents, and
3010
# parent_ids, we should be able to do this without assuming
3012
if len(parent_ids) == 0:
3013
# There are no parents for this node, so just add it
3014
# TODO: This probably needs to be decoupled
3015
assert compression_parent is None
3016
fulltext_content, delta = self._knit.factory.parse_record(
3017
rev_id, record, record_details, None)
3018
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3019
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3020
parent_ids, left_matching_blocks=None))
3022
child = (rev_id, parent_ids, record)
3023
# Check if all the parents are present
3024
self._check_parents(child, nodes_to_annotate)
3025
while nodes_to_annotate:
3026
# Should we use a queue here instead of a stack?
3027
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3028
(index_memo, compression_parent, parents,
3029
record_details) = self._all_build_details[rev_id]
3030
if compression_parent is not None:
3031
comp_children = self._compression_children[compression_parent]
3032
assert rev_id in comp_children
3033
# If there is only 1 child, it is safe to reuse this
3035
reuse_content = (len(comp_children) == 1
3036
and compression_parent not in
3037
self._nodes_to_keep_annotations)
3039
# Remove it from the cache since it will be changing
3040
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3041
# Make sure to copy the fulltext since it might be
3043
parent_fulltext = list(parent_fulltext_content.text())
3045
parent_fulltext_content = self._fulltext_contents[compression_parent]
3046
parent_fulltext = parent_fulltext_content.text()
3047
comp_children.remove(rev_id)
3048
fulltext_content, delta = self._knit.factory.parse_record(
3049
rev_id, record, record_details,
3050
parent_fulltext_content,
3051
copy_base_content=(not reuse_content))
3052
fulltext = self._add_fulltext_content(rev_id,
3054
blocks = KnitContent.get_line_delta_blocks(delta,
3055
parent_fulltext, fulltext)
3057
fulltext_content = self._knit.factory.parse_fulltext(
3059
fulltext = self._add_fulltext_content(rev_id,
3062
nodes_to_annotate.extend(
3063
self._add_annotation(rev_id, fulltext, parent_ids,
3064
left_matching_blocks=blocks))
3066
def _get_heads_provider(self):
3067
"""Create a heads provider for resolving ancestry issues."""
3068
if self._heads_provider is not None:
3069
return self._heads_provider
3070
parent_provider = _mod_graph.DictParentsProvider(
3071
self._revision_id_graph)
3072
graph_obj = _mod_graph.Graph(parent_provider)
3073
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3074
self._heads_provider = head_cache
3077
def annotate(self, revision_id):
3078
"""Return the annotated fulltext at the given revision.
3080
:param revision_id: The revision id for this file
3082
records = self._get_build_graph(revision_id)
3083
if revision_id in self._ghosts:
3084
raise errors.RevisionNotPresent(revision_id, self._knit)
3085
self._annotate_records(records)
3086
return self._annotated_lines[revision_id]
3090
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3092
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
