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))
627
self._index.add_versions(index_entries)
629
def copy_to(self, name, transport):
630
"""See VersionedFile.copy_to()."""
631
# copy the current index to a temp index to avoid racing with local
633
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
634
self.transport.get(self._index._filename))
636
f = self._data._open_file()
638
transport.put_file(name + DATA_SUFFIX, f)
641
# move the copied index into place
642
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
644
def get_data_stream(self, required_versions):
645
"""Get a data stream for the specified versions.
647
Versions may be returned in any order, not necessarily the order
648
specified. They are returned in a partial order by compression
649
parent, so that the deltas can be applied as the data stream is
650
inserted; however note that compression parents will not be sent
651
unless they were specifically requested, as the client may already
654
:param required_versions: The exact set of versions to be extracted.
655
Unlike some other knit methods, this is not used to generate a
656
transitive closure, rather it is used precisely as given.
658
:returns: format_signature, list of (version, options, length, parents),
661
required_version_set = frozenset(required_versions)
663
# list of revisions that can just be sent without waiting for their
666
# map from revision to the children based on it
668
# first, read all relevant index data, enough to sort into the right
670
for version_id in required_versions:
671
options = self._index.get_options(version_id)
672
parents = self._index.get_parents_with_ghosts(version_id)
673
index_memo = self._index.get_position(version_id)
674
version_index[version_id] = (index_memo, options, parents)
675
if ('line-delta' in options
676
and parents[0] in required_version_set):
677
# must wait until the parent has been sent
678
deferred.setdefault(parents[0], []). \
681
# either a fulltext, or a delta whose parent the client did
682
# not ask for and presumably already has
683
ready_to_send.append(version_id)
684
# build a list of results to return, plus instructions for data to
686
copy_queue_records = []
687
temp_version_list = []
689
# XXX: pushing and popping lists may be a bit inefficient
690
version_id = ready_to_send.pop(0)
691
(index_memo, options, parents) = version_index[version_id]
692
copy_queue_records.append((version_id, index_memo))
693
none, data_pos, data_size = index_memo
694
temp_version_list.append((version_id, options, data_size,
696
if version_id in deferred:
697
# now we can send all the children of this revision - we could
698
# put them in anywhere, but we hope that sending them soon
699
# after the fulltext will give good locality in the receiver
700
ready_to_send[:0] = deferred.pop(version_id)
701
assert len(deferred) == 0, \
702
"Still have compressed child versions waiting to be sent"
703
# XXX: The stream format is such that we cannot stream it - we have to
704
# know the length of all the data a-priori.
706
result_version_list = []
707
for (version_id, raw_data), \
708
(version_id2, options, _, parents) in \
709
izip(self._data.read_records_iter_raw(copy_queue_records),
711
assert version_id == version_id2, \
712
'logic error, inconsistent results'
713
raw_datum.append(raw_data)
714
result_version_list.append(
715
(version_id, options, len(raw_data), parents))
716
# provide a callback to get data incrementally.
717
pseudo_file = StringIO(''.join(raw_datum))
720
return pseudo_file.read()
722
return pseudo_file.read(length)
723
return (self.get_format_signature(), result_version_list, read)
725
def _extract_blocks(self, version_id, source, target):
726
if self._index.get_method(version_id) != 'line-delta':
728
parent, sha1, noeol, delta = self.get_delta(version_id)
729
return KnitContent.get_line_delta_blocks(delta, source, target)
731
def get_delta(self, version_id):
732
"""Get a delta for constructing version from some other version."""
733
self.check_not_reserved_id(version_id)
734
parents = self.get_parent_map([version_id])[version_id]
739
index_memo = self._index.get_position(version_id)
740
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
741
noeol = 'no-eol' in self._index.get_options(version_id)
742
if 'fulltext' == self._index.get_method(version_id):
743
new_content = self.factory.parse_fulltext(data, version_id)
744
if parent is not None:
745
reference_content = self._get_content(parent)
746
old_texts = reference_content.text()
749
new_texts = new_content.text()
750
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
752
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
754
delta = self.factory.parse_line_delta(data, version_id)
755
return parent, sha1, noeol, delta
757
def get_format_signature(self):
758
"""See VersionedFile.get_format_signature()."""
759
if self.factory.annotated:
760
annotated_part = "annotated"
762
annotated_part = "plain"
763
return "knit-%s" % (annotated_part,)
765
@deprecated_method(one_four)
766
def get_graph_with_ghosts(self):
767
"""See VersionedFile.get_graph_with_ghosts()."""
768
return self.get_parent_map(self.versions())
770
def get_sha1s(self, version_ids):
771
"""See VersionedFile.get_sha1s()."""
772
record_map = self._get_record_map(version_ids)
773
# record entry 2 is the 'digest'.
774
return [record_map[v][2] for v in version_ids]
776
@deprecated_method(one_four)
777
def has_ghost(self, version_id):
778
"""True if there is a ghost reference in the file to version_id."""
780
if self.has_version(version_id):
782
# optimisable if needed by memoising the _ghosts set.
783
items = self.get_parent_map(self.versions())
784
for parents in items.itervalues():
785
for parent in parents:
786
if parent == version_id and parent not in items:
790
def insert_data_stream(self, (format, data_list, reader_callable)):
791
"""Insert knit records from a data stream into this knit.
793
If a version in the stream is already present in this knit, it will not
794
be inserted a second time. It will be checked for consistency with the
795
stored version however, and may cause a KnitCorrupt error to be raised
796
if the data in the stream disagrees with the already stored data.
798
:seealso: get_data_stream
800
if format != self.get_format_signature():
801
if 'knit' in debug.debug_flags:
803
'incompatible format signature inserting to %r', self)
804
source = self._knit_from_datastream(
805
(format, data_list, reader_callable))
809
for version_id, options, length, parents in data_list:
810
if self.has_version(version_id):
811
# First check: the list of parents.
812
my_parents = self.get_parents_with_ghosts(version_id)
813
if tuple(my_parents) != tuple(parents):
814
# XXX: KnitCorrupt is not quite the right exception here.
817
'parents list %r from data stream does not match '
818
'already recorded parents %r for %s'
819
% (parents, my_parents, version_id))
821
# Also check the SHA-1 of the fulltext this content will
823
raw_data = reader_callable(length)
824
my_fulltext_sha1 = self.get_sha1s([version_id])[0]
825
df, rec = self._data._parse_record_header(version_id, raw_data)
826
stream_fulltext_sha1 = rec[3]
827
if my_fulltext_sha1 != stream_fulltext_sha1:
828
# Actually, we don't know if it's this knit that's corrupt,
829
# or the data stream we're trying to insert.
831
self.filename, 'sha-1 does not match %s' % version_id)
833
if 'line-delta' in options:
834
# Make sure that this knit record is actually useful: a
835
# line-delta is no use unless we have its parent.
836
# Fetching from a broken repository with this problem
837
# shouldn't break the target repository.
839
# See https://bugs.launchpad.net/bzr/+bug/164443
840
if not self._index.has_version(parents[0]):
843
'line-delta from stream '
846
'missing parent %s\n'
847
'Try running "bzr check" '
848
'on the source repository, and "bzr reconcile" '
850
(version_id, parents[0]))
851
self._add_raw_records(
852
[(version_id, options, parents, length)],
853
reader_callable(length))
855
def _knit_from_datastream(self, (format, data_list, reader_callable)):
856
"""Create a knit object from a data stream.
858
This method exists to allow conversion of data streams that do not
859
match the signature of this knit. Generally it will be slower and use
860
more memory to use this method to insert data, but it will work.
862
:seealso: get_data_stream for details on datastreams.
863
:return: A knit versioned file which can be used to join the datastream
866
if format == "knit-plain":
867
factory = KnitPlainFactory()
868
elif format == "knit-annotated":
869
factory = KnitAnnotateFactory()
871
raise errors.KnitDataStreamUnknown(format)
872
index = _StreamIndex(data_list, self._index)
873
access = _StreamAccess(reader_callable, index, self, factory)
874
return KnitVersionedFile(self.filename, self.transport,
875
factory=factory, index=index, access_method=access)
878
"""See VersionedFile.versions."""
879
if 'evil' in debug.debug_flags:
880
trace.mutter_callsite(2, "versions scales with size of history")
881
return self._index.get_versions()
883
def has_version(self, version_id):
884
"""See VersionedFile.has_version."""
885
if 'evil' in debug.debug_flags:
886
trace.mutter_callsite(2, "has_version is a LBYL scenario")
887
return self._index.has_version(version_id)
889
__contains__ = has_version
891
def _merge_annotations(self, content, parents, parent_texts={},
892
delta=None, annotated=None,
893
left_matching_blocks=None):
894
"""Merge annotations for content. This is done by comparing
895
the annotations based on changed to the text.
897
if left_matching_blocks is not None:
898
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
902
for parent_id in parents:
903
merge_content = self._get_content(parent_id, parent_texts)
904
if (parent_id == parents[0] and delta_seq is not None):
907
seq = patiencediff.PatienceSequenceMatcher(
908
None, merge_content.text(), content.text())
909
for i, j, n in seq.get_matching_blocks():
912
# this appears to copy (origin, text) pairs across to the
913
# new content for any line that matches the last-checked
915
content._lines[j:j+n] = merge_content._lines[i:i+n]
917
if delta_seq is None:
918
reference_content = self._get_content(parents[0], parent_texts)
919
new_texts = content.text()
920
old_texts = reference_content.text()
921
delta_seq = patiencediff.PatienceSequenceMatcher(
922
None, old_texts, new_texts)
923
return self._make_line_delta(delta_seq, content)
925
def _make_line_delta(self, delta_seq, new_content):
926
"""Generate a line delta from delta_seq and new_content."""
928
for op in delta_seq.get_opcodes():
931
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
934
def _get_components_positions(self, version_ids):
935
"""Produce a map of position data for the components of versions.
937
This data is intended to be used for retrieving the knit records.
939
A dict of version_id to (record_details, index_memo, next, parents) is
941
method is the way referenced data should be applied.
942
index_memo is the handle to pass to the data access to actually get the
944
next is the build-parent of the version, or None for fulltexts.
945
parents is the version_ids of the parents of this version
948
pending_components = version_ids
949
while pending_components:
950
build_details = self._index.get_build_details(pending_components)
951
current_components = set(pending_components)
952
pending_components = set()
953
for version_id, details in build_details.iteritems():
954
(index_memo, compression_parent, parents,
955
record_details) = details
956
method = record_details[0]
957
if compression_parent is not None:
958
pending_components.add(compression_parent)
959
component_data[version_id] = (record_details, index_memo,
961
missing = current_components.difference(build_details)
963
raise errors.RevisionNotPresent(missing.pop(), self.filename)
964
return component_data
966
def _get_content(self, version_id, parent_texts={}):
967
"""Returns a content object that makes up the specified
969
cached_version = parent_texts.get(version_id, None)
970
if cached_version is not None:
971
if not self.has_version(version_id):
972
raise RevisionNotPresent(version_id, self.filename)
973
return cached_version
975
text_map, contents_map = self._get_content_maps([version_id])
976
return contents_map[version_id]
978
def _check_versions_present(self, version_ids):
979
"""Check that all specified versions are present."""
980
self._index.check_versions_present(version_ids)
982
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
983
nostore_sha, random_id, check_content, left_matching_blocks):
984
"""See VersionedFile.add_lines_with_ghosts()."""
985
self._check_add(version_id, lines, random_id, check_content)
986
return self._add(version_id, lines, parents, self.delta,
987
parent_texts, left_matching_blocks, nostore_sha, random_id)
989
def _add_lines(self, version_id, parents, lines, parent_texts,
990
left_matching_blocks, nostore_sha, random_id, check_content):
991
"""See VersionedFile.add_lines."""
992
self._check_add(version_id, lines, random_id, check_content)
993
self._check_versions_present(parents)
994
return self._add(version_id, lines[:], parents, self.delta,
995
parent_texts, left_matching_blocks, nostore_sha, random_id)
997
def _check_add(self, version_id, lines, random_id, check_content):
998
"""check that version_id and lines are safe to add."""
999
if contains_whitespace(version_id):
1000
raise InvalidRevisionId(version_id, self.filename)
1001
self.check_not_reserved_id(version_id)
1002
# Technically this could be avoided if we are happy to allow duplicate
1003
# id insertion when other things than bzr core insert texts, but it
1004
# seems useful for folk using the knit api directly to have some safety
1005
# blanket that we can disable.
1006
if not random_id and self.has_version(version_id):
1007
raise RevisionAlreadyPresent(version_id, self.filename)
1009
self._check_lines_not_unicode(lines)
1010
self._check_lines_are_lines(lines)
1012
def _add(self, version_id, lines, parents, delta, parent_texts,
1013
left_matching_blocks, nostore_sha, random_id):
1014
"""Add a set of lines on top of version specified by parents.
1016
If delta is true, compress the text as a line-delta against
1019
Any versions not present will be converted into ghosts.
1021
# first thing, if the content is something we don't need to store, find
1023
line_bytes = ''.join(lines)
1024
digest = sha_string(line_bytes)
1025
if nostore_sha == digest:
1026
raise errors.ExistingContent
1028
present_parents = []
1029
if parent_texts is None:
1031
for parent in parents:
1032
if self.has_version(parent):
1033
present_parents.append(parent)
1035
# can only compress against the left most present parent.
1037
(len(present_parents) == 0 or
1038
present_parents[0] != parents[0])):
1041
text_length = len(line_bytes)
1044
if lines[-1][-1] != '\n':
1045
# copy the contents of lines.
1047
options.append('no-eol')
1048
lines[-1] = lines[-1] + '\n'
1052
# To speed the extract of texts the delta chain is limited
1053
# to a fixed number of deltas. This should minimize both
1054
# I/O and the time spend applying deltas.
1055
delta = self._check_should_delta(present_parents)
1057
assert isinstance(version_id, str)
1058
content = self.factory.make(lines, version_id)
1059
if delta or (self.factory.annotated and len(present_parents) > 0):
1060
# Merge annotations from parent texts if needed.
1061
delta_hunks = self._merge_annotations(content, present_parents,
1062
parent_texts, delta, self.factory.annotated,
1063
left_matching_blocks)
1066
options.append('line-delta')
1067
store_lines = self.factory.lower_line_delta(delta_hunks)
1068
size, bytes = self._data._record_to_data(version_id, digest,
1071
options.append('fulltext')
1072
# isinstance is slower and we have no hierarchy.
1073
if self.factory.__class__ == KnitPlainFactory:
1074
# Use the already joined bytes saving iteration time in
1076
size, bytes = self._data._record_to_data(version_id, digest,
1077
lines, [line_bytes])
1079
# get mixed annotation + content and feed it into the
1081
store_lines = self.factory.lower_fulltext(content)
1082
size, bytes = self._data._record_to_data(version_id, digest,
1085
access_memo = self._data.add_raw_records([size], bytes)[0]
1086
self._index.add_versions(
1087
((version_id, options, access_memo, parents),),
1088
random_id=random_id)
1089
return digest, text_length, content
1091
def check(self, progress_bar=None):
1092
"""See VersionedFile.check()."""
1094
def get_lines(self, version_id):
1095
"""See VersionedFile.get_lines()."""
1096
return self.get_line_list([version_id])[0]
1098
def _get_record_map(self, version_ids):
1099
"""Produce a dictionary of knit records.
1101
:return: {version_id:(record, record_details, digest, next)}
1103
data returned from read_records
1105
opaque information to pass to parse_record
1107
SHA1 digest of the full text after all steps are done
1109
build-parent of the version, i.e. the leftmost ancestor.
1110
Will be None if the record is not a delta.
1112
position_map = self._get_components_positions(version_ids)
1113
# c = component_id, r = record_details, i_m = index_memo, n = next
1114
records = [(c, i_m) for c, (r, i_m, n)
1115
in position_map.iteritems()]
1117
for component_id, record, digest in \
1118
self._data.read_records_iter(records):
1119
(record_details, index_memo, next) = position_map[component_id]
1120
record_map[component_id] = record, record_details, digest, next
1124
def get_text(self, version_id):
1125
"""See VersionedFile.get_text"""
1126
return self.get_texts([version_id])[0]
1128
def get_texts(self, version_ids):
1129
return [''.join(l) for l in self.get_line_list(version_ids)]
1131
def get_line_list(self, version_ids):
1132
"""Return the texts of listed versions as a list of strings."""
1133
for version_id in version_ids:
1134
self.check_not_reserved_id(version_id)
1135
text_map, content_map = self._get_content_maps(version_ids)
1136
return [text_map[v] for v in version_ids]
1138
_get_lf_split_line_list = get_line_list
1140
def _get_content_maps(self, version_ids):
1141
"""Produce maps of text and KnitContents
1143
:return: (text_map, content_map) where text_map contains the texts for
1144
the requested versions and content_map contains the KnitContents.
1145
Both dicts take version_ids as their keys.
1147
# FUTURE: This function could be improved for the 'extract many' case
1148
# by tracking each component and only doing the copy when the number of
1149
# children than need to apply delta's to it is > 1 or it is part of the
1151
version_ids = list(version_ids)
1152
multiple_versions = len(version_ids) != 1
1153
record_map = self._get_record_map(version_ids)
1158
for version_id in version_ids:
1161
while cursor is not None:
1162
record, record_details, digest, next = record_map[cursor]
1163
components.append((cursor, record, record_details, digest))
1164
if cursor in content_map:
1169
for (component_id, record, record_details,
1170
digest) in reversed(components):
1171
if component_id in content_map:
1172
content = content_map[component_id]
1174
content, delta = self.factory.parse_record(version_id,
1175
record, record_details, content,
1176
copy_base_content=multiple_versions)
1177
if multiple_versions:
1178
content_map[component_id] = content
1180
content.cleanup_eol(copy_on_mutate=multiple_versions)
1181
final_content[version_id] = content
1183
# digest here is the digest from the last applied component.
1184
text = content.text()
1185
actual_sha = sha_strings(text)
1186
if actual_sha != digest:
1187
raise KnitCorrupt(self.filename,
1189
'\n of reconstructed text does not match'
1191
'\n for version %s' %
1192
(actual_sha, digest, version_id))
1193
text_map[version_id] = text
1194
return text_map, final_content
1196
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1198
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1199
if version_ids is None:
1200
version_ids = self.versions()
1202
pb = progress.DummyProgress()
1203
# we don't care about inclusions, the caller cares.
1204
# but we need to setup a list of records to visit.
1205
# we need version_id, position, length
1206
version_id_records = []
1207
requested_versions = set(version_ids)
1208
# filter for available versions
1209
for version_id in requested_versions:
1210
if not self.has_version(version_id):
1211
raise RevisionNotPresent(version_id, self.filename)
1212
# get a in-component-order queue:
1213
for version_id in self.versions():
1214
if version_id in requested_versions:
1215
index_memo = self._index.get_position(version_id)
1216
version_id_records.append((version_id, index_memo))
1218
total = len(version_id_records)
1219
for version_idx, (version_id, data, sha_value) in \
1220
enumerate(self._data.read_records_iter(version_id_records)):
1221
pb.update('Walking content.', version_idx, total)
1222
method = self._index.get_method(version_id)
1224
assert method in ('fulltext', 'line-delta')
1225
if method == 'fulltext':
1226
line_iterator = self.factory.get_fulltext_content(data)
1228
line_iterator = self.factory.get_linedelta_content(data)
1229
# XXX: It might be more efficient to yield (version_id,
1230
# line_iterator) in the future. However for now, this is a simpler
1231
# change to integrate into the rest of the codebase. RBC 20071110
1232
for line in line_iterator:
1233
yield line, version_id
1235
pb.update('Walking content.', total, total)
1237
def num_versions(self):
1238
"""See VersionedFile.num_versions()."""
1239
return self._index.num_versions()
1241
__len__ = num_versions
1243
def annotate_iter(self, version_id):
1244
"""See VersionedFile.annotate_iter."""
1245
return self.factory.annotate_iter(self, version_id)
1247
def get_parent_map(self, version_ids):
1248
"""See VersionedFile.get_parent_map."""
1249
return self._index.get_parent_map(version_ids)
1251
def get_ancestry(self, versions, topo_sorted=True):
1252
"""See VersionedFile.get_ancestry."""
1253
if isinstance(versions, basestring):
1254
versions = [versions]
1257
return self._index.get_ancestry(versions, topo_sorted)
1259
def get_ancestry_with_ghosts(self, versions):
1260
"""See VersionedFile.get_ancestry_with_ghosts."""
1261
if isinstance(versions, basestring):
1262
versions = [versions]
1265
return self._index.get_ancestry_with_ghosts(versions)
1267
def plan_merge(self, ver_a, ver_b):
1268
"""See VersionedFile.plan_merge."""
1269
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1270
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1271
annotated_a = self.annotate(ver_a)
1272
annotated_b = self.annotate(ver_b)
1273
return merge._plan_annotate_merge(annotated_a, annotated_b,
1274
ancestors_a, ancestors_b)
1277
class _KnitComponentFile(object):
1278
"""One of the files used to implement a knit database"""
1280
def __init__(self, transport, filename, mode, file_mode=None,
1281
create_parent_dir=False, dir_mode=None):
1282
self._transport = transport
1283
self._filename = filename
1285
self._file_mode = file_mode
1286
self._dir_mode = dir_mode
1287
self._create_parent_dir = create_parent_dir
1288
self._need_to_create = False
1290
def _full_path(self):
1291
"""Return the full path to this file."""
1292
return self._transport.base + self._filename
1294
def check_header(self, fp):
1295
line = fp.readline()
1297
# An empty file can actually be treated as though the file doesn't
1299
raise errors.NoSuchFile(self._full_path())
1300
if line != self.HEADER:
1301
raise KnitHeaderError(badline=line,
1302
filename=self._transport.abspath(self._filename))
1305
return '%s(%s)' % (self.__class__.__name__, self._filename)
1308
class _KnitIndex(_KnitComponentFile):
1309
"""Manages knit index file.
1311
The index is already kept in memory and read on startup, to enable
1312
fast lookups of revision information. The cursor of the index
1313
file is always pointing to the end, making it easy to append
1316
_cache is a cache for fast mapping from version id to a Index
1319
_history is a cache for fast mapping from indexes to version ids.
1321
The index data format is dictionary compressed when it comes to
1322
parent references; a index entry may only have parents that with a
1323
lover index number. As a result, the index is topological sorted.
1325
Duplicate entries may be written to the index for a single version id
1326
if this is done then the latter one completely replaces the former:
1327
this allows updates to correct version and parent information.
1328
Note that the two entries may share the delta, and that successive
1329
annotations and references MUST point to the first entry.
1331
The index file on disc contains a header, followed by one line per knit
1332
record. The same revision can be present in an index file more than once.
1333
The first occurrence gets assigned a sequence number starting from 0.
1335
The format of a single line is
1336
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1337
REVISION_ID is a utf8-encoded revision id
1338
FLAGS is a comma separated list of flags about the record. Values include
1339
no-eol, line-delta, fulltext.
1340
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1341
that the the compressed data starts at.
1342
LENGTH is the ascii representation of the length of the data file.
1343
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1345
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1346
revision id already in the knit that is a parent of REVISION_ID.
1347
The ' :' marker is the end of record marker.
1350
when a write is interrupted to the index file, it will result in a line
1351
that does not end in ' :'. If the ' :' is not present at the end of a line,
1352
or at the end of the file, then the record that is missing it will be
1353
ignored by the parser.
1355
When writing new records to the index file, the data is preceded by '\n'
1356
to ensure that records always start on new lines even if the last write was
1357
interrupted. As a result its normal for the last line in the index to be
1358
missing a trailing newline. One can be added with no harmful effects.
1361
HEADER = "# bzr knit index 8\n"
1363
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1364
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1366
def _cache_version(self, version_id, options, pos, size, parents):
1367
"""Cache a version record in the history array and index cache.
1369
This is inlined into _load_data for performance. KEEP IN SYNC.
1370
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1373
# only want the _history index to reference the 1st index entry
1375
if version_id not in self._cache:
1376
index = len(self._history)
1377
self._history.append(version_id)
1379
index = self._cache[version_id][5]
1380
self._cache[version_id] = (version_id,
1387
def _check_write_ok(self):
1388
if self._get_scope() != self._scope:
1389
raise errors.OutSideTransaction()
1390
if self._mode != 'w':
1391
raise errors.ReadOnlyObjectDirtiedError(self)
1393
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1394
create_parent_dir=False, delay_create=False, dir_mode=None,
1396
_KnitComponentFile.__init__(self, transport, filename, mode,
1397
file_mode=file_mode,
1398
create_parent_dir=create_parent_dir,
1401
# position in _history is the 'official' index for a revision
1402
# but the values may have come from a newer entry.
1403
# so - wc -l of a knit index is != the number of unique names
1407
fp = self._transport.get(self._filename)
1409
# _load_data may raise NoSuchFile if the target knit is
1411
_load_data(self, fp)
1415
if mode != 'w' or not create:
1418
self._need_to_create = True
1420
self._transport.put_bytes_non_atomic(
1421
self._filename, self.HEADER, mode=self._file_mode)
1422
self._scope = get_scope()
1423
self._get_scope = get_scope
1425
def get_ancestry(self, versions, topo_sorted=True):
1426
"""See VersionedFile.get_ancestry."""
1427
# get a graph of all the mentioned versions:
1429
pending = set(versions)
1432
version = pending.pop()
1435
parents = [p for p in cache[version][4] if p in cache]
1437
raise RevisionNotPresent(version, self._filename)
1438
# if not completed and not a ghost
1439
pending.update([p for p in parents if p not in graph])
1440
graph[version] = parents
1443
return topo_sort(graph.items())
1445
def get_ancestry_with_ghosts(self, versions):
1446
"""See VersionedFile.get_ancestry_with_ghosts."""
1447
# get a graph of all the mentioned versions:
1448
self.check_versions_present(versions)
1451
pending = set(versions)
1453
version = pending.pop()
1455
parents = cache[version][4]
1461
pending.update([p for p in parents if p not in graph])
1462
graph[version] = parents
1463
return topo_sort(graph.items())
1465
def get_build_details(self, version_ids):
1466
"""Get the method, index_memo and compression parent for version_ids.
1468
Ghosts are omitted from the result.
1470
:param version_ids: An iterable of version_ids.
1471
:return: A dict of version_id:(index_memo, compression_parent,
1472
parents, record_details).
1474
opaque structure to pass to read_records to extract the raw
1477
Content that this record is built upon, may be None
1479
Logical parents of this node
1481
extra information about the content which needs to be passed to
1482
Factory.parse_record
1485
for version_id in version_ids:
1486
if version_id not in self._cache:
1487
# ghosts are omitted
1489
method = self.get_method(version_id)
1490
parents = self.get_parents_with_ghosts(version_id)
1491
if method == 'fulltext':
1492
compression_parent = None
1494
compression_parent = parents[0]
1495
noeol = 'no-eol' in self.get_options(version_id)
1496
index_memo = self.get_position(version_id)
1497
result[version_id] = (index_memo, compression_parent,
1498
parents, (method, noeol))
1501
def num_versions(self):
1502
return len(self._history)
1504
__len__ = num_versions
1506
def get_versions(self):
1507
"""Get all the versions in the file. not topologically sorted."""
1508
return self._history
1510
def _version_list_to_index(self, versions):
1513
for version in versions:
1514
if version in cache:
1515
# -- inlined lookup() --
1516
result_list.append(str(cache[version][5]))
1517
# -- end lookup () --
1519
result_list.append('.' + version)
1520
return ' '.join(result_list)
1522
def add_version(self, version_id, options, index_memo, parents):
1523
"""Add a version record to the index."""
1524
self.add_versions(((version_id, options, index_memo, parents),))
1526
def add_versions(self, versions, random_id=False):
1527
"""Add multiple versions to the index.
1529
:param versions: a list of tuples:
1530
(version_id, options, pos, size, parents).
1531
:param random_id: If True the ids being added were randomly generated
1532
and no check for existence will be performed.
1535
orig_history = self._history[:]
1536
orig_cache = self._cache.copy()
1539
for version_id, options, (index, pos, size), parents in versions:
1540
line = "\n%s %s %s %s %s :" % (version_id,
1544
self._version_list_to_index(parents))
1545
assert isinstance(line, str), \
1546
'content must be utf-8 encoded: %r' % (line,)
1548
self._cache_version(version_id, options, pos, size, tuple(parents))
1549
if not self._need_to_create:
1550
self._transport.append_bytes(self._filename, ''.join(lines))
1553
sio.write(self.HEADER)
1554
sio.writelines(lines)
1556
self._transport.put_file_non_atomic(self._filename, sio,
1557
create_parent_dir=self._create_parent_dir,
1558
mode=self._file_mode,
1559
dir_mode=self._dir_mode)
1560
self._need_to_create = False
1562
# If any problems happen, restore the original values and re-raise
1563
self._history = orig_history
1564
self._cache = orig_cache
1567
def has_version(self, version_id):
1568
"""True if the version is in the index."""
1569
return version_id in self._cache
1571
def get_position(self, version_id):
1572
"""Return details needed to access the version.
1574
.kndx indices do not support split-out data, so return None for the
1577
:return: a tuple (None, data position, size) to hand to the access
1578
logic to get the record.
1580
entry = self._cache[version_id]
1581
return None, entry[2], entry[3]
1583
def get_method(self, version_id):
1584
"""Return compression method of specified version."""
1586
options = self._cache[version_id][1]
1588
raise RevisionNotPresent(version_id, self._filename)
1589
if 'fulltext' in options:
1592
if 'line-delta' not in options:
1593
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1596
def get_options(self, version_id):
1597
"""Return a list representing options.
1601
return self._cache[version_id][1]
1603
def get_parent_map(self, version_ids):
1604
"""Passed through to by KnitVersionedFile.get_parent_map."""
1606
for version_id in version_ids:
1608
result[version_id] = tuple(self._cache[version_id][4])
1613
def get_parents_with_ghosts(self, version_id):
1614
"""Return parents of specified version with ghosts."""
1616
return self.get_parent_map([version_id])[version_id]
1618
raise RevisionNotPresent(version_id, self)
1620
def check_versions_present(self, version_ids):
1621
"""Check that all specified versions are present."""
1623
for version_id in version_ids:
1624
if version_id not in cache:
1625
raise RevisionNotPresent(version_id, self._filename)
1628
class KnitGraphIndex(object):
1629
"""A knit index that builds on GraphIndex."""
1631
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1632
"""Construct a KnitGraphIndex on a graph_index.
1634
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1635
:param deltas: Allow delta-compressed records.
1636
:param add_callback: If not None, allow additions to the index and call
1637
this callback with a list of added GraphIndex nodes:
1638
[(node, value, node_refs), ...]
1639
:param parents: If True, record knits parents, if not do not record
1642
self._graph_index = graph_index
1643
self._deltas = deltas
1644
self._add_callback = add_callback
1645
self._parents = parents
1646
if deltas and not parents:
1647
raise KnitCorrupt(self, "Cannot do delta compression without "
1650
def _check_write_ok(self):
1653
def _get_entries(self, keys, check_present=False):
1654
"""Get the entries for keys.
1656
:param keys: An iterable of index keys, - 1-tuples.
1661
for node in self._graph_index.iter_entries(keys):
1663
found_keys.add(node[1])
1665
# adapt parentless index to the rest of the code.
1666
for node in self._graph_index.iter_entries(keys):
1667
yield node[0], node[1], node[2], ()
1668
found_keys.add(node[1])
1670
missing_keys = keys.difference(found_keys)
1672
raise RevisionNotPresent(missing_keys.pop(), self)
1674
def _present_keys(self, version_ids):
1676
node[1] for node in self._get_entries(version_ids)])
1678
def _parentless_ancestry(self, versions):
1679
"""Honour the get_ancestry API for parentless knit indices."""
1680
wanted_keys = self._version_ids_to_keys(versions)
1681
present_keys = self._present_keys(wanted_keys)
1682
missing = set(wanted_keys).difference(present_keys)
1684
raise RevisionNotPresent(missing.pop(), self)
1685
return list(self._keys_to_version_ids(present_keys))
1687
def get_ancestry(self, versions, topo_sorted=True):
1688
"""See VersionedFile.get_ancestry."""
1689
if not self._parents:
1690
return self._parentless_ancestry(versions)
1691
# XXX: This will do len(history) index calls - perhaps
1692
# it should be altered to be a index core feature?
1693
# get a graph of all the mentioned versions:
1696
versions = self._version_ids_to_keys(versions)
1697
pending = set(versions)
1699
# get all pending nodes
1700
this_iteration = pending
1701
new_nodes = self._get_entries(this_iteration)
1704
for (index, key, value, node_refs) in new_nodes:
1705
# dont ask for ghosties - otherwise
1706
# we we can end up looping with pending
1707
# being entirely ghosted.
1708
graph[key] = [parent for parent in node_refs[0]
1709
if parent not in ghosts]
1711
for parent in graph[key]:
1712
# dont examine known nodes again
1717
ghosts.update(this_iteration.difference(found))
1718
if versions.difference(graph):
1719
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1721
result_keys = topo_sort(graph.items())
1723
result_keys = graph.iterkeys()
1724
return [key[0] for key in result_keys]
1726
def get_ancestry_with_ghosts(self, versions):
1727
"""See VersionedFile.get_ancestry."""
1728
if not self._parents:
1729
return self._parentless_ancestry(versions)
1730
# XXX: This will do len(history) index calls - perhaps
1731
# it should be altered to be a index core feature?
1732
# get a graph of all the mentioned versions:
1734
versions = self._version_ids_to_keys(versions)
1735
pending = set(versions)
1737
# get all pending nodes
1738
this_iteration = pending
1739
new_nodes = self._get_entries(this_iteration)
1741
for (index, key, value, node_refs) in new_nodes:
1742
graph[key] = node_refs[0]
1744
for parent in graph[key]:
1745
# dont examine known nodes again
1749
missing_versions = this_iteration.difference(graph)
1750
missing_needed = versions.intersection(missing_versions)
1752
raise RevisionNotPresent(missing_needed.pop(), self)
1753
for missing_version in missing_versions:
1754
# add a key, no parents
1755
graph[missing_version] = []
1756
pending.discard(missing_version) # don't look for it
1757
result_keys = topo_sort(graph.items())
1758
return [key[0] for key in result_keys]
1760
def get_build_details(self, version_ids):
1761
"""Get the method, index_memo and compression parent for version_ids.
1763
Ghosts are omitted from the result.
1765
:param version_ids: An iterable of version_ids.
1766
:return: A dict of version_id:(index_memo, compression_parent,
1767
parents, record_details).
1769
opaque structure to pass to read_records to extract the raw
1772
Content that this record is built upon, may be None
1774
Logical parents of this node
1776
extra information about the content which needs to be passed to
1777
Factory.parse_record
1780
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
1781
for entry in entries:
1782
version_id = self._keys_to_version_ids((entry[1],))[0]
1783
if not self._parents:
1786
parents = self._keys_to_version_ids(entry[3][0])
1787
if not self._deltas:
1788
compression_parent = None
1790
compression_parent_key = self._compression_parent(entry)
1791
if compression_parent_key:
1792
compression_parent = self._keys_to_version_ids(
1793
(compression_parent_key,))[0]
1795
compression_parent = None
1796
noeol = (entry[2][0] == 'N')
1797
if compression_parent:
1798
method = 'line-delta'
1801
result[version_id] = (self._node_to_position(entry),
1802
compression_parent, parents,
1806
def _compression_parent(self, an_entry):
1807
# return the key that an_entry is compressed against, or None
1808
# Grab the second parent list (as deltas implies parents currently)
1809
compression_parents = an_entry[3][1]
1810
if not compression_parents:
1812
assert len(compression_parents) == 1
1813
return compression_parents[0]
1815
def _get_method(self, node):
1816
if not self._deltas:
1818
if self._compression_parent(node):
1823
def num_versions(self):
1824
return len(list(self._graph_index.iter_all_entries()))
1826
__len__ = num_versions
1828
def get_versions(self):
1829
"""Get all the versions in the file. not topologically sorted."""
1830
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1832
def has_version(self, version_id):
1833
"""True if the version is in the index."""
1834
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1836
def _keys_to_version_ids(self, keys):
1837
return tuple(key[0] for key in keys)
1839
def get_position(self, version_id):
1840
"""Return details needed to access the version.
1842
:return: a tuple (index, data position, size) to hand to the access
1843
logic to get the record.
1845
node = self._get_node(version_id)
1846
return self._node_to_position(node)
1848
def _node_to_position(self, node):
1849
"""Convert an index value to position details."""
1850
bits = node[2][1:].split(' ')
1851
return node[0], int(bits[0]), int(bits[1])
1853
def get_method(self, version_id):
1854
"""Return compression method of specified version."""
1855
return self._get_method(self._get_node(version_id))
1857
def _get_node(self, version_id):
1859
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1861
raise RevisionNotPresent(version_id, self)
1863
def get_options(self, version_id):
1864
"""Return a list representing options.
1868
node = self._get_node(version_id)
1869
options = [self._get_method(node)]
1870
if node[2][0] == 'N':
1871
options.append('no-eol')
1874
def get_parent_map(self, version_ids):
1875
"""Passed through to by KnitVersionedFile.get_parent_map."""
1876
nodes = self._get_entries(self._version_ids_to_keys(version_ids))
1880
result[node[1][0]] = self._keys_to_version_ids(node[3][0])
1883
result[node[1][0]] = ()
1886
def get_parents_with_ghosts(self, version_id):
1887
"""Return parents of specified version with ghosts."""
1889
return self.get_parent_map([version_id])[version_id]
1891
raise RevisionNotPresent(version_id, self)
1893
def check_versions_present(self, version_ids):
1894
"""Check that all specified versions are present."""
1895
keys = self._version_ids_to_keys(version_ids)
1896
present = self._present_keys(keys)
1897
missing = keys.difference(present)
1899
raise RevisionNotPresent(missing.pop(), self)
1901
def add_version(self, version_id, options, access_memo, parents):
1902
"""Add a version record to the index."""
1903
return self.add_versions(((version_id, options, access_memo, parents),))
1905
def add_versions(self, versions, random_id=False):
1906
"""Add multiple versions to the index.
1908
This function does not insert data into the Immutable GraphIndex
1909
backing the KnitGraphIndex, instead it prepares data for insertion by
1910
the caller and checks that it is safe to insert then calls
1911
self._add_callback with the prepared GraphIndex nodes.
1913
:param versions: a list of tuples:
1914
(version_id, options, pos, size, parents).
1915
:param random_id: If True the ids being added were randomly generated
1916
and no check for existence will be performed.
1918
if not self._add_callback:
1919
raise errors.ReadOnlyError(self)
1920
# we hope there are no repositories with inconsistent parentage
1925
for (version_id, options, access_memo, parents) in versions:
1926
index, pos, size = access_memo
1927
key = (version_id, )
1928
parents = tuple((parent, ) for parent in parents)
1929
if 'no-eol' in options:
1933
value += "%d %d" % (pos, size)
1934
if not self._deltas:
1935
if 'line-delta' in options:
1936
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1939
if 'line-delta' in options:
1940
node_refs = (parents, (parents[0],))
1942
node_refs = (parents, ())
1944
node_refs = (parents, )
1947
raise KnitCorrupt(self, "attempt to add node with parents "
1948
"in parentless index.")
1950
keys[key] = (value, node_refs)
1952
present_nodes = self._get_entries(keys)
1953
for (index, key, value, node_refs) in present_nodes:
1954
if (value, node_refs) != keys[key]:
1955
raise KnitCorrupt(self, "inconsistent details in add_versions"
1956
": %s %s" % ((value, node_refs), keys[key]))
1960
for key, (value, node_refs) in keys.iteritems():
1961
result.append((key, value, node_refs))
1963
for key, (value, node_refs) in keys.iteritems():
1964
result.append((key, value))
1965
self._add_callback(result)
1967
def _version_ids_to_keys(self, version_ids):
1968
return set((version_id, ) for version_id in version_ids)
1971
class _KnitAccess(object):
1972
"""Access to knit records in a .knit file."""
1974
def __init__(self, transport, filename, _file_mode, _dir_mode,
1975
_need_to_create, _create_parent_dir):
1976
"""Create a _KnitAccess for accessing and inserting data.
1978
:param transport: The transport the .knit is located on.
1979
:param filename: The filename of the .knit.
1981
self._transport = transport
1982
self._filename = filename
1983
self._file_mode = _file_mode
1984
self._dir_mode = _dir_mode
1985
self._need_to_create = _need_to_create
1986
self._create_parent_dir = _create_parent_dir
1988
def add_raw_records(self, sizes, raw_data):
1989
"""Add raw knit bytes to a storage area.
1991
The data is spooled to whereever the access method is storing data.
1993
:param sizes: An iterable containing the size of each raw data segment.
1994
:param raw_data: A bytestring containing the data.
1995
:return: A list of memos to retrieve the record later. Each memo is a
1996
tuple - (index, pos, length), where the index field is always None
1997
for the .knit access method.
1999
assert type(raw_data) == str, \
2000
'data must be plain bytes was %s' % type(raw_data)
2001
if not self._need_to_create:
2002
base = self._transport.append_bytes(self._filename, raw_data)
2004
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2005
create_parent_dir=self._create_parent_dir,
2006
mode=self._file_mode,
2007
dir_mode=self._dir_mode)
2008
self._need_to_create = False
2012
result.append((None, base, size))
2017
"""IFF this data access has its own storage area, initialise it.
2021
self._transport.put_bytes_non_atomic(self._filename, '',
2022
mode=self._file_mode)
2024
def open_file(self):
2025
"""IFF this data access can be represented as a single file, open it.
2027
For knits that are not mapped to a single file on disk this will
2030
:return: None or a file handle.
2033
return self._transport.get(self._filename)
2038
def get_raw_records(self, memos_for_retrieval):
2039
"""Get the raw bytes for a records.
2041
:param memos_for_retrieval: An iterable containing the (index, pos,
2042
length) memo for retrieving the bytes. The .knit method ignores
2043
the index as there is always only a single file.
2044
:return: An iterator over the bytes of the records.
2046
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2047
for pos, data in self._transport.readv(self._filename, read_vector):
2051
class _PackAccess(object):
2052
"""Access to knit records via a collection of packs."""
2054
def __init__(self, index_to_packs, writer=None):
2055
"""Create a _PackAccess object.
2057
:param index_to_packs: A dict mapping index objects to the transport
2058
and file names for obtaining data.
2059
:param writer: A tuple (pack.ContainerWriter, write_index) which
2060
contains the pack to write, and the index that reads from it will
2064
self.container_writer = writer[0]
2065
self.write_index = writer[1]
2067
self.container_writer = None
2068
self.write_index = None
2069
self.indices = index_to_packs
2071
def add_raw_records(self, sizes, raw_data):
2072
"""Add raw knit bytes to a storage area.
2074
The data is spooled to the container writer in one bytes-record per
2077
:param sizes: An iterable containing the size of each raw data segment.
2078
:param raw_data: A bytestring containing the data.
2079
:return: A list of memos to retrieve the record later. Each memo is a
2080
tuple - (index, pos, length), where the index field is the
2081
write_index object supplied to the PackAccess object.
2083
assert type(raw_data) == str, \
2084
'data must be plain bytes was %s' % type(raw_data)
2088
p_offset, p_length = self.container_writer.add_bytes_record(
2089
raw_data[offset:offset+size], [])
2091
result.append((self.write_index, p_offset, p_length))
2095
"""Pack based knits do not get individually created."""
2097
def get_raw_records(self, memos_for_retrieval):
2098
"""Get the raw bytes for a records.
2100
:param memos_for_retrieval: An iterable containing the (index, pos,
2101
length) memo for retrieving the bytes. The Pack access method
2102
looks up the pack to use for a given record in its index_to_pack
2104
:return: An iterator over the bytes of the records.
2106
# first pass, group into same-index requests
2108
current_index = None
2109
for (index, offset, length) in memos_for_retrieval:
2110
if current_index == index:
2111
current_list.append((offset, length))
2113
if current_index is not None:
2114
request_lists.append((current_index, current_list))
2115
current_index = index
2116
current_list = [(offset, length)]
2117
# handle the last entry
2118
if current_index is not None:
2119
request_lists.append((current_index, current_list))
2120
for index, offsets in request_lists:
2121
transport, path = self.indices[index]
2122
reader = pack.make_readv_reader(transport, path, offsets)
2123
for names, read_func in reader.iter_records():
2124
yield read_func(None)
2126
def open_file(self):
2127
"""Pack based knits have no single file."""
2130
def set_writer(self, writer, index, (transport, packname)):
2131
"""Set a writer to use for adding data."""
2132
if index is not None:
2133
self.indices[index] = (transport, packname)
2134
self.container_writer = writer
2135
self.write_index = index
2138
class _StreamAccess(object):
2139
"""A Knit Access object that provides data from a datastream.
2141
It also provides a fallback to present as unannotated data, annotated data
2142
from a *backing* access object.
2144
This is triggered by a index_memo which is pointing to a different index
2145
than this was constructed with, and is used to allow extracting full
2146
unannotated texts for insertion into annotated knits.
2149
def __init__(self, reader_callable, stream_index, backing_knit,
2151
"""Create a _StreamAccess object.
2153
:param reader_callable: The reader_callable from the datastream.
2154
This is called to buffer all the data immediately, for
2156
:param stream_index: The index the data stream this provides access to
2157
which will be present in native index_memo's.
2158
:param backing_knit: The knit object that will provide access to
2159
annotated texts which are not available in the stream, so as to
2160
create unannotated texts.
2161
:param orig_factory: The original content factory used to generate the
2162
stream. This is used for checking whether the thunk code for
2163
supporting _copy_texts will generate the correct form of data.
2165
self.data = reader_callable(None)
2166
self.stream_index = stream_index
2167
self.backing_knit = backing_knit
2168
self.orig_factory = orig_factory
2170
def get_raw_records(self, memos_for_retrieval):
2171
"""Get the raw bytes for a records.
2173
:param memos_for_retrieval: An iterable containing the (thunk_flag,
2174
index, start, end) memo for retrieving the bytes.
2175
:return: An iterator over the bytes of the records.
2177
# use a generator for memory friendliness
2178
for thunk_flag, version_id, start, end in memos_for_retrieval:
2179
if version_id is self.stream_index:
2180
yield self.data[start:end]
2182
# we have been asked to thunk. This thunking only occurs when
2183
# we are obtaining plain texts from an annotated backing knit
2184
# so that _copy_texts will work.
2185
# We could improve performance here by scanning for where we need
2186
# to do this and using get_line_list, then interleaving the output
2187
# as desired. However, for now, this is sufficient.
2188
if self.orig_factory.__class__ != KnitPlainFactory:
2189
raise errors.KnitCorrupt(
2190
self, 'Bad thunk request %r' % version_id)
2191
lines = self.backing_knit.get_lines(version_id)
2192
line_bytes = ''.join(lines)
2193
digest = sha_string(line_bytes)
2195
if lines[-1][-1] != '\n':
2196
lines[-1] = lines[-1] + '\n'
2198
orig_options = list(self.backing_knit._index.get_options(version_id))
2199
if 'fulltext' not in orig_options:
2200
if 'line-delta' not in orig_options:
2201
raise errors.KnitCorrupt(self,
2202
'Unknown compression method %r' % orig_options)
2203
orig_options.remove('line-delta')
2204
orig_options.append('fulltext')
2205
# We want plain data, because we expect to thunk only to allow text
2207
size, bytes = self.backing_knit._data._record_to_data(version_id,
2208
digest, lines, line_bytes)
2212
class _StreamIndex(object):
2213
"""A Knit Index object that uses the data map from a datastream."""
2215
def __init__(self, data_list, backing_index):
2216
"""Create a _StreamIndex object.
2218
:param data_list: The data_list from the datastream.
2219
:param backing_index: The index which will supply values for nodes
2220
referenced outside of this stream.
2222
self.data_list = data_list
2223
self.backing_index = backing_index
2224
self._by_version = {}
2226
for key, options, length, parents in data_list:
2227
self._by_version[key] = options, (pos, pos + length), parents
2230
def get_ancestry(self, versions, topo_sorted):
2231
"""Get an ancestry list for versions."""
2233
# Not needed for basic joins
2234
raise NotImplementedError(self.get_ancestry)
2235
# get a graph of all the mentioned versions:
2236
# Little ugly - basically copied from KnitIndex, but don't want to
2237
# accidentally incorporate too much of that index's code.
2239
pending = set(versions)
2240
cache = self._by_version
2242
version = pending.pop()
2245
parents = [p for p in cache[version][2] if p in cache]
2247
raise RevisionNotPresent(version, self)
2248
# if not completed and not a ghost
2249
pending.update([p for p in parents if p not in ancestry])
2250
ancestry.add(version)
2251
return list(ancestry)
2253
def get_build_details(self, version_ids):
2254
"""Get the method, index_memo and compression parent for version_ids.
2256
Ghosts are omitted from the result.
2258
:param version_ids: An iterable of version_ids.
2259
:return: A dict of version_id:(index_memo, compression_parent,
2260
parents, record_details).
2262
opaque structure to pass to read_records to extract the raw
2265
Content that this record is built upon, may be None
2267
Logical parents of this node
2269
extra information about the content which needs to be passed to
2270
Factory.parse_record
2273
for version_id in version_ids:
2275
method = self.get_method(version_id)
2276
except errors.RevisionNotPresent:
2277
# ghosts are omitted
2279
parent_ids = self.get_parents_with_ghosts(version_id)
2280
noeol = ('no-eol' in self.get_options(version_id))
2281
if method == 'fulltext':
2282
compression_parent = None
2284
compression_parent = parent_ids[0]
2285
index_memo = self.get_position(version_id)
2286
result[version_id] = (index_memo, compression_parent,
2287
parent_ids, (method, noeol))
2290
def get_method(self, version_id):
2291
"""Return compression method of specified version."""
2293
options = self._by_version[version_id][0]
2295
# Strictly speaking this should check in the backing knit, but
2296
# until we have a test to discriminate, this will do.
2297
return self.backing_index.get_method(version_id)
2298
if 'fulltext' in options:
2300
elif 'line-delta' in options:
2303
raise errors.KnitIndexUnknownMethod(self, options)
2305
def get_options(self, version_id):
2306
"""Return a list representing options.
2311
return self._by_version[version_id][0]
2313
return self.backing_index.get_options(version_id)
2315
def get_parent_map(self, version_ids):
2316
"""Passed through to by KnitVersionedFile.get_parent_map."""
2319
for version_id in version_ids:
2321
result[version_id] = self._by_version[version_id][2]
2323
pending_ids.add(version_id)
2324
result.update(self.backing_index.get_parent_map(pending_ids))
2327
def get_parents_with_ghosts(self, version_id):
2328
"""Return parents of specified version with ghosts."""
2330
return self.get_parent_map([version_id])[version_id]
2332
raise RevisionNotPresent(version_id, self)
2334
def get_position(self, version_id):
2335
"""Return details needed to access the version.
2337
_StreamAccess has the data as a big array, so we return slice
2338
coordinates into that (as index_memo's are opaque outside the
2339
index and matching access class).
2341
:return: a tuple (thunk_flag, index, start, end). If thunk_flag is
2342
False, index will be self, otherwise it will be a version id.
2345
start, end = self._by_version[version_id][1]
2346
return False, self, start, end
2348
# Signal to the access object to handle this from the backing knit.
2349
return (True, version_id, None, None)
2351
def get_versions(self):
2352
"""Get all the versions in the stream."""
2353
return self._by_version.keys()
2356
class _KnitData(object):
2357
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2359
The KnitData class provides the logic for parsing and using knit records,
2360
making use of an access method for the low level read and write operations.
2363
def __init__(self, access):
2364
"""Create a KnitData object.
2366
:param access: The access method to use. Access methods such as
2367
_KnitAccess manage the insertion of raw records and the subsequent
2368
retrieval of the same.
2370
self._access = access
2371
self._checked = False
2373
def _open_file(self):
2374
return self._access.open_file()
2376
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2377
"""Convert version_id, digest, lines into a raw data block.
2379
:param dense_lines: The bytes of lines but in a denser form. For
2380
instance, if lines is a list of 1000 bytestrings each ending in \n,
2381
dense_lines may be a list with one line in it, containing all the
2382
1000's lines and their \n's. Using dense_lines if it is already
2383
known is a win because the string join to create bytes in this
2384
function spends less time resizing the final string.
2385
:return: (len, a StringIO instance with the raw data ready to read.)
2387
# Note: using a string copy here increases memory pressure with e.g.
2388
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2389
# when doing the initial commit of a mozilla tree. RBC 20070921
2390
bytes = ''.join(chain(
2391
["version %s %d %s\n" % (version_id,
2394
dense_lines or lines,
2395
["end %s\n" % version_id]))
2396
assert bytes.__class__ == str
2397
compressed_bytes = bytes_to_gzip(bytes)
2398
return len(compressed_bytes), compressed_bytes
2400
def add_raw_records(self, sizes, raw_data):
2401
"""Append a prepared record to the data file.
2403
:param sizes: An iterable containing the size of each raw data segment.
2404
:param raw_data: A bytestring containing the data.
2405
:return: a list of index data for the way the data was stored.
2406
See the access method add_raw_records documentation for more
2409
return self._access.add_raw_records(sizes, raw_data)
2411
def _parse_record_header(self, version_id, raw_data):
2412
"""Parse a record header for consistency.
2414
:return: the header and the decompressor stream.
2415
as (stream, header_record)
2417
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2419
rec = self._check_header(version_id, df.readline())
2420
except Exception, e:
2421
raise KnitCorrupt(self._access,
2422
"While reading {%s} got %s(%s)"
2423
% (version_id, e.__class__.__name__, str(e)))
2426
def _check_header(self, version_id, line):
2429
raise KnitCorrupt(self._access,
2430
'unexpected number of elements in record header')
2431
if rec[1] != version_id:
2432
raise KnitCorrupt(self._access,
2433
'unexpected version, wanted %r, got %r'
2434
% (version_id, rec[1]))
2437
def _parse_record(self, version_id, data):
2439
# 4168 calls in 2880 217 internal
2440
# 4168 calls to _parse_record_header in 2121
2441
# 4168 calls to readlines in 330
2442
df = GzipFile(mode='rb', fileobj=StringIO(data))
2445
record_contents = df.readlines()
2446
except Exception, e:
2447
raise KnitCorrupt(self._access,
2448
"While reading {%s} got %s(%s)"
2449
% (version_id, e.__class__.__name__, str(e)))
2450
header = record_contents.pop(0)
2451
rec = self._check_header(version_id, header)
2453
last_line = record_contents.pop()
2454
if len(record_contents) != int(rec[2]):
2455
raise KnitCorrupt(self._access,
2456
'incorrect number of lines %s != %s'
2458
% (len(record_contents), int(rec[2]),
2460
if last_line != 'end %s\n' % rec[1]:
2461
raise KnitCorrupt(self._access,
2462
'unexpected version end line %r, wanted %r'
2463
% (last_line, version_id))
2465
return record_contents, rec[3]
2467
def read_records_iter_raw(self, records):
2468
"""Read text records from data file and yield raw data.
2470
This unpacks enough of the text record to validate the id is
2471
as expected but thats all.
2473
# setup an iterator of the external records:
2474
# uses readv so nice and fast we hope.
2476
# grab the disk data needed.
2477
needed_offsets = [index_memo for version_id, index_memo
2479
raw_records = self._access.get_raw_records(needed_offsets)
2481
for version_id, index_memo in records:
2482
data = raw_records.next()
2483
# validate the header
2484
df, rec = self._parse_record_header(version_id, data)
2486
yield version_id, data
2488
def read_records_iter(self, records):
2489
"""Read text records from data file and yield result.
2491
The result will be returned in whatever is the fastest to read.
2492
Not by the order requested. Also, multiple requests for the same
2493
record will only yield 1 response.
2494
:param records: A list of (version_id, pos, len) entries
2495
:return: Yields (version_id, contents, digest) in the order
2496
read, not the order requested
2501
needed_records = sorted(set(records), key=operator.itemgetter(1))
2502
if not needed_records:
2505
# The transport optimizes the fetching as well
2506
# (ie, reads continuous ranges.)
2507
raw_data = self._access.get_raw_records(
2508
[index_memo for version_id, index_memo in needed_records])
2510
for (version_id, index_memo), data in \
2511
izip(iter(needed_records), raw_data):
2512
content, digest = self._parse_record(version_id, data)
2513
yield version_id, content, digest
2515
def read_records(self, records):
2516
"""Read records into a dictionary."""
2518
for record_id, content, digest in \
2519
self.read_records_iter(records):
2520
components[record_id] = (content, digest)
2524
class InterKnit(InterVersionedFile):
2525
"""Optimised code paths for knit to knit operations."""
2527
_matching_file_from_factory = staticmethod(make_file_knit)
2528
_matching_file_to_factory = staticmethod(make_file_knit)
2531
def is_compatible(source, target):
2532
"""Be compatible with knits. """
2534
return (isinstance(source, KnitVersionedFile) and
2535
isinstance(target, KnitVersionedFile))
2536
except AttributeError:
2539
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2540
"""Copy texts to the target by extracting and adding them one by one.
2542
see join() for the parameter definitions.
2544
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2545
# --- the below is factorable out with VersionedFile.join, but wait for
2546
# VersionedFiles, it may all be simpler then.
2547
graph = Graph(self.source)
2548
search = graph._make_breadth_first_searcher(version_ids)
2549
transitive_ids = set()
2550
map(transitive_ids.update, list(search))
2551
parent_map = self.source.get_parent_map(transitive_ids)
2552
order = topo_sort(parent_map.items())
2554
def size_of_content(content):
2555
return sum(len(line) for line in content.text())
2556
# Cache at most 10MB of parent texts
2557
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2558
compute_size=size_of_content)
2559
# TODO: jam 20071116 It would be nice to have a streaming interface to
2560
# get multiple texts from a source. The source could be smarter
2561
# about how it handled intermediate stages.
2562
# get_line_list() or make_mpdiffs() seem like a possibility, but
2563
# at the moment they extract all full texts into memory, which
2564
# causes us to store more than our 3x fulltext goal.
2565
# Repository.iter_files_bytes() may be another possibility
2566
to_process = [version for version in order
2567
if version not in self.target]
2568
total = len(to_process)
2569
pb = ui.ui_factory.nested_progress_bar()
2571
for index, version in enumerate(to_process):
2572
pb.update('Converting versioned data', index, total)
2573
sha1, num_bytes, parent_text = self.target.add_lines(version,
2574
self.source.get_parents_with_ghosts(version),
2575
self.source.get_lines(version),
2576
parent_texts=parent_cache)
2577
parent_cache[version] = parent_text
2582
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2583
"""See InterVersionedFile.join."""
2584
assert isinstance(self.source, KnitVersionedFile)
2585
assert isinstance(self.target, KnitVersionedFile)
2587
# If the source and target are mismatched w.r.t. annotations vs
2588
# plain, the data needs to be converted accordingly
2589
if self.source.factory.annotated == self.target.factory.annotated:
2591
elif self.source.factory.annotated:
2592
converter = self._anno_to_plain_converter
2594
# We're converting from a plain to an annotated knit. Copy them
2595
# across by full texts.
2596
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2598
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2602
pb = ui.ui_factory.nested_progress_bar()
2604
version_ids = list(version_ids)
2605
if None in version_ids:
2606
version_ids.remove(None)
2608
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2610
this_versions = set(self.target._index.get_versions())
2611
# XXX: For efficiency we should not look at the whole index,
2612
# we only need to consider the referenced revisions - they
2613
# must all be present, or the method must be full-text.
2614
# TODO, RBC 20070919
2615
needed_versions = self.source_ancestry - this_versions
2617
if not needed_versions:
2619
full_list = topo_sort(
2620
self.source.get_parent_map(self.source.versions()))
2622
version_list = [i for i in full_list if (not self.target.has_version(i)
2623
and i in needed_versions)]
2627
copy_queue_records = []
2629
for version_id in version_list:
2630
options = self.source._index.get_options(version_id)
2631
parents = self.source._index.get_parents_with_ghosts(version_id)
2632
# check that its will be a consistent copy:
2633
for parent in parents:
2634
# if source has the parent, we must :
2635
# * already have it or
2636
# * have it scheduled already
2637
# otherwise we don't care
2638
assert (self.target.has_version(parent) or
2639
parent in copy_set or
2640
not self.source.has_version(parent))
2641
index_memo = self.source._index.get_position(version_id)
2642
copy_queue_records.append((version_id, index_memo))
2643
copy_queue.append((version_id, options, parents))
2644
copy_set.add(version_id)
2646
# data suck the join:
2648
total = len(version_list)
2651
for (version_id, raw_data), \
2652
(version_id2, options, parents) in \
2653
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2655
assert version_id == version_id2, 'logic error, inconsistent results'
2657
pb.update("Joining knit", count, total)
2659
size, raw_data = converter(raw_data, version_id, options,
2662
size = len(raw_data)
2663
raw_records.append((version_id, options, parents, size))
2664
raw_datum.append(raw_data)
2665
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2670
def _anno_to_plain_converter(self, raw_data, version_id, options,
2672
"""Convert annotated content to plain content."""
2673
data, digest = self.source._data._parse_record(version_id, raw_data)
2674
if 'fulltext' in options:
2675
content = self.source.factory.parse_fulltext(data, version_id)
2676
lines = self.target.factory.lower_fulltext(content)
2678
delta = self.source.factory.parse_line_delta(data, version_id,
2680
lines = self.target.factory.lower_line_delta(delta)
2681
return self.target._data._record_to_data(version_id, digest, lines)
2684
InterVersionedFile.register_optimiser(InterKnit)
2687
class WeaveToKnit(InterVersionedFile):
2688
"""Optimised code paths for weave to knit operations."""
2690
_matching_file_from_factory = bzrlib.weave.WeaveFile
2691
_matching_file_to_factory = staticmethod(make_file_knit)
2694
def is_compatible(source, target):
2695
"""Be compatible with weaves to knits."""
2697
return (isinstance(source, bzrlib.weave.Weave) and
2698
isinstance(target, KnitVersionedFile))
2699
except AttributeError:
2702
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2703
"""See InterVersionedFile.join."""
2704
assert isinstance(self.source, bzrlib.weave.Weave)
2705
assert isinstance(self.target, KnitVersionedFile)
2707
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2712
pb = ui.ui_factory.nested_progress_bar()
2714
version_ids = list(version_ids)
2716
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2717
this_versions = set(self.target._index.get_versions())
2718
needed_versions = self.source_ancestry - this_versions
2720
if not needed_versions:
2722
full_list = topo_sort(
2723
self.source.get_parent_map(self.source.versions()))
2725
version_list = [i for i in full_list if (not self.target.has_version(i)
2726
and i in needed_versions)]
2730
total = len(version_list)
2731
parent_map = self.source.get_parent_map(version_list)
2732
for version_id in version_list:
2733
pb.update("Converting to knit", count, total)
2734
parents = parent_map[version_id]
2735
# check that its will be a consistent copy:
2736
for parent in parents:
2737
# if source has the parent, we must already have it
2738
assert (self.target.has_version(parent))
2739
self.target.add_lines(
2740
version_id, parents, self.source.get_lines(version_id))
2747
InterVersionedFile.register_optimiser(WeaveToKnit)
2750
# Deprecated, use PatienceSequenceMatcher instead
2751
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2754
def annotate_knit(knit, revision_id):
2755
"""Annotate a knit with no cached annotations.
2757
This implementation is for knits with no cached annotations.
2758
It will work for knits with cached annotations, but this is not
2761
annotator = _KnitAnnotator(knit)
2762
return iter(annotator.annotate(revision_id))
2765
class _KnitAnnotator(object):
2766
"""Build up the annotations for a text."""
2768
def __init__(self, knit):
2771
# Content objects, differs from fulltexts because of how final newlines
2772
# are treated by knits. the content objects here will always have a
2774
self._fulltext_contents = {}
2776
# Annotated lines of specific revisions
2777
self._annotated_lines = {}
2779
# Track the raw data for nodes that we could not process yet.
2780
# This maps the revision_id of the base to a list of children that will
2781
# annotated from it.
2782
self._pending_children = {}
2784
# Nodes which cannot be extracted
2785
self._ghosts = set()
2787
# Track how many children this node has, so we know if we need to keep
2789
self._annotate_children = {}
2790
self._compression_children = {}
2792
self._all_build_details = {}
2793
# The children => parent revision_id graph
2794
self._revision_id_graph = {}
2796
self._heads_provider = None
2798
self._nodes_to_keep_annotations = set()
2799
self._generations_until_keep = 100
2801
def set_generations_until_keep(self, value):
2802
"""Set the number of generations before caching a node.
2804
Setting this to -1 will cache every merge node, setting this higher
2805
will cache fewer nodes.
2807
self._generations_until_keep = value
2809
def _add_fulltext_content(self, revision_id, content_obj):
2810
self._fulltext_contents[revision_id] = content_obj
2811
# TODO: jam 20080305 It might be good to check the sha1digest here
2812
return content_obj.text()
2814
def _check_parents(self, child, nodes_to_annotate):
2815
"""Check if all parents have been processed.
2817
:param child: A tuple of (rev_id, parents, raw_content)
2818
:param nodes_to_annotate: If child is ready, add it to
2819
nodes_to_annotate, otherwise put it back in self._pending_children
2821
for parent_id in child[1]:
2822
if (parent_id not in self._annotated_lines):
2823
# This parent is present, but another parent is missing
2824
self._pending_children.setdefault(parent_id,
2828
# This one is ready to be processed
2829
nodes_to_annotate.append(child)
2831
def _add_annotation(self, revision_id, fulltext, parent_ids,
2832
left_matching_blocks=None):
2833
"""Add an annotation entry.
2835
All parents should already have been annotated.
2836
:return: A list of children that now have their parents satisfied.
2838
a = self._annotated_lines
2839
annotated_parent_lines = [a[p] for p in parent_ids]
2840
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2841
fulltext, revision_id, left_matching_blocks,
2842
heads_provider=self._get_heads_provider()))
2843
self._annotated_lines[revision_id] = annotated_lines
2844
for p in parent_ids:
2845
ann_children = self._annotate_children[p]
2846
ann_children.remove(revision_id)
2847
if (not ann_children
2848
and p not in self._nodes_to_keep_annotations):
2849
del self._annotated_lines[p]
2850
del self._all_build_details[p]
2851
if p in self._fulltext_contents:
2852
del self._fulltext_contents[p]
2853
# Now that we've added this one, see if there are any pending
2854
# deltas to be done, certainly this parent is finished
2855
nodes_to_annotate = []
2856
for child in self._pending_children.pop(revision_id, []):
2857
self._check_parents(child, nodes_to_annotate)
2858
return nodes_to_annotate
2860
def _get_build_graph(self, revision_id):
2861
"""Get the graphs for building texts and annotations.
2863
The data you need for creating a full text may be different than the
2864
data you need to annotate that text. (At a minimum, you need both
2865
parents to create an annotation, but only need 1 parent to generate the
2868
:return: A list of (revision_id, index_memo) records, suitable for
2869
passing to read_records_iter to start reading in the raw data fro/
2872
if revision_id in self._annotated_lines:
2875
pending = set([revision_id])
2880
# get all pending nodes
2882
this_iteration = pending
2883
build_details = self._knit._index.get_build_details(this_iteration)
2884
self._all_build_details.update(build_details)
2885
# new_nodes = self._knit._index._get_entries(this_iteration)
2887
for rev_id, details in build_details.iteritems():
2888
(index_memo, compression_parent, parents,
2889
record_details) = details
2890
self._revision_id_graph[rev_id] = parents
2891
records.append((rev_id, index_memo))
2892
# Do we actually need to check _annotated_lines?
2893
pending.update(p for p in parents
2894
if p not in self._all_build_details)
2895
if compression_parent:
2896
self._compression_children.setdefault(compression_parent,
2899
for parent in parents:
2900
self._annotate_children.setdefault(parent,
2902
num_gens = generation - kept_generation
2903
if ((num_gens >= self._generations_until_keep)
2904
and len(parents) > 1):
2905
kept_generation = generation
2906
self._nodes_to_keep_annotations.add(rev_id)
2908
missing_versions = this_iteration.difference(build_details.keys())
2909
self._ghosts.update(missing_versions)
2910
for missing_version in missing_versions:
2911
# add a key, no parents
2912
self._revision_id_graph[missing_version] = ()
2913
pending.discard(missing_version) # don't look for it
2914
# XXX: This should probably be a real exception, as it is a data
2916
assert not self._ghosts.intersection(self._compression_children), \
2917
"We cannot have nodes which have a compression parent of a ghost."
2918
# Cleanout anything that depends on a ghost so that we don't wait for
2919
# the ghost to show up
2920
for node in self._ghosts:
2921
if node in self._annotate_children:
2922
# We won't be building this node
2923
del self._annotate_children[node]
2924
# Generally we will want to read the records in reverse order, because
2925
# we find the parent nodes after the children
2929
def _annotate_records(self, records):
2930
"""Build the annotations for the listed records."""
2931
# We iterate in the order read, rather than a strict order requested
2932
# However, process what we can, and put off to the side things that
2933
# still need parents, cleaning them up when those parents are
2935
for (rev_id, record,
2936
digest) in self._knit._data.read_records_iter(records):
2937
if rev_id in self._annotated_lines:
2939
parent_ids = self._revision_id_graph[rev_id]
2940
parent_ids = [p for p in parent_ids if p not in self._ghosts]
2941
details = self._all_build_details[rev_id]
2942
(index_memo, compression_parent, parents,
2943
record_details) = details
2944
nodes_to_annotate = []
2945
# TODO: Remove the punning between compression parents, and
2946
# parent_ids, we should be able to do this without assuming
2948
if len(parent_ids) == 0:
2949
# There are no parents for this node, so just add it
2950
# TODO: This probably needs to be decoupled
2951
assert compression_parent is None
2952
fulltext_content, delta = self._knit.factory.parse_record(
2953
rev_id, record, record_details, None)
2954
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
2955
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
2956
parent_ids, left_matching_blocks=None))
2958
child = (rev_id, parent_ids, record)
2959
# Check if all the parents are present
2960
self._check_parents(child, nodes_to_annotate)
2961
while nodes_to_annotate:
2962
# Should we use a queue here instead of a stack?
2963
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
2964
(index_memo, compression_parent, parents,
2965
record_details) = self._all_build_details[rev_id]
2966
if compression_parent is not None:
2967
comp_children = self._compression_children[compression_parent]
2968
assert rev_id in comp_children
2969
# If there is only 1 child, it is safe to reuse this
2971
reuse_content = (len(comp_children) == 1
2972
and compression_parent not in
2973
self._nodes_to_keep_annotations)
2975
# Remove it from the cache since it will be changing
2976
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
2977
# Make sure to copy the fulltext since it might be
2979
parent_fulltext = list(parent_fulltext_content.text())
2981
parent_fulltext_content = self._fulltext_contents[compression_parent]
2982
parent_fulltext = parent_fulltext_content.text()
2983
comp_children.remove(rev_id)
2984
fulltext_content, delta = self._knit.factory.parse_record(
2985
rev_id, record, record_details,
2986
parent_fulltext_content,
2987
copy_base_content=(not reuse_content))
2988
fulltext = self._add_fulltext_content(rev_id,
2990
blocks = KnitContent.get_line_delta_blocks(delta,
2991
parent_fulltext, fulltext)
2993
fulltext_content = self._knit.factory.parse_fulltext(
2995
fulltext = self._add_fulltext_content(rev_id,
2998
nodes_to_annotate.extend(
2999
self._add_annotation(rev_id, fulltext, parent_ids,
3000
left_matching_blocks=blocks))
3002
def _get_heads_provider(self):
3003
"""Create a heads provider for resolving ancestry issues."""
3004
if self._heads_provider is not None:
3005
return self._heads_provider
3006
parent_provider = _mod_graph.DictParentsProvider(
3007
self._revision_id_graph)
3008
graph_obj = _mod_graph.Graph(parent_provider)
3009
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3010
self._heads_provider = head_cache
3013
def annotate(self, revision_id):
3014
"""Return the annotated fulltext at the given revision.
3016
:param revision_id: The revision id for this file
3018
records = self._get_build_graph(revision_id)
3019
if revision_id in self._ghosts:
3020
raise errors.RevisionNotPresent(revision_id, self._knit)
3021
self._annotate_records(records)
3022
return self._annotated_lines[revision_id]
3026
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3028
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
