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
147
def apply_delta(self, delta, new_version_id):
148
"""Apply delta to this object to become new_version_id."""
149
raise NotImplementedError(self.apply_delta)
151
def cleanup_eol(self, copy_on_mutate=True):
152
if self._should_strip_eol:
154
self._lines = self._lines[:]
155
self.strip_last_line_newline()
157
def line_delta_iter(self, new_lines):
158
"""Generate line-based delta from this content to new_lines."""
159
new_texts = new_lines.text()
160
old_texts = self.text()
161
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
162
for tag, i1, i2, j1, j2 in s.get_opcodes():
165
# ofrom, oto, length, data
166
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
168
def line_delta(self, new_lines):
169
return list(self.line_delta_iter(new_lines))
172
def get_line_delta_blocks(knit_delta, source, target):
173
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
174
target_len = len(target)
177
for s_begin, s_end, t_len, new_text in knit_delta:
178
true_n = s_begin - s_pos
181
# knit deltas do not provide reliable info about whether the
182
# last line of a file matches, due to eol handling.
183
if source[s_pos + n -1] != target[t_pos + n -1]:
186
yield s_pos, t_pos, n
187
t_pos += t_len + true_n
189
n = target_len - t_pos
191
if source[s_pos + n -1] != target[t_pos + n -1]:
194
yield s_pos, t_pos, n
195
yield s_pos + (target_len - t_pos), target_len, 0
198
class AnnotatedKnitContent(KnitContent):
199
"""Annotated content."""
201
def __init__(self, lines):
202
KnitContent.__init__(self)
206
"""Return a list of (origin, text) for each content line."""
207
return list(self._lines)
209
def apply_delta(self, delta, new_version_id):
210
"""Apply delta to this object to become new_version_id."""
213
for start, end, count, delta_lines in delta:
214
lines[offset+start:offset+end] = delta_lines
215
offset = offset + (start - end) + count
217
def strip_last_line_newline(self):
218
line = self._lines[-1][1].rstrip('\n')
219
self._lines[-1] = (self._lines[-1][0], line)
220
self._should_strip_eol = False
224
lines = [text for origin, text in self._lines]
225
except ValueError, e:
226
# most commonly (only?) caused by the internal form of the knit
227
# missing annotation information because of a bug - see thread
229
raise KnitCorrupt(self,
230
"line in annotated knit missing annotation information: %s"
233
if self._should_strip_eol:
234
anno, line = lines[-1]
235
lines[-1] = (anno, line.rstrip('\n'))
239
return AnnotatedKnitContent(self._lines[:])
242
class PlainKnitContent(KnitContent):
243
"""Unannotated content.
245
When annotate[_iter] is called on this content, the same version is reported
246
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
250
def __init__(self, lines, version_id):
251
KnitContent.__init__(self)
253
self._version_id = version_id
256
"""Return a list of (origin, text) for each content line."""
257
return [(self._version_id, line) for line in self._lines]
259
def apply_delta(self, delta, new_version_id):
260
"""Apply delta to this object to become new_version_id."""
263
for start, end, count, delta_lines in delta:
264
lines[offset+start:offset+end] = delta_lines
265
offset = offset + (start - end) + count
266
self._version_id = new_version_id
269
return PlainKnitContent(self._lines[:], self._version_id)
271
def strip_last_line_newline(self):
272
self._lines[-1] = self._lines[-1].rstrip('\n')
273
self._should_strip_eol = False
277
if self._should_strip_eol:
279
lines[-1] = lines[-1].rstrip('\n')
283
class _KnitFactory(object):
284
"""Base class for common Factory functions."""
286
def parse_record(self, version_id, record, record_details,
287
base_content, copy_base_content=True):
288
"""Parse a record into a full content object.
290
:param version_id: The official version id for this content
291
:param record: The data returned by read_records_iter()
292
:param record_details: Details about the record returned by
294
:param base_content: If get_build_details returns a compression_parent,
295
you must return a base_content here, else use None
296
:param copy_base_content: When building from the base_content, decide
297
you can either copy it and return a new object, or modify it in
299
:return: (content, delta) A Content object and possibly a line-delta,
302
method, noeol = record_details
303
if method == 'line-delta':
304
assert base_content is not None
305
if copy_base_content:
306
content = base_content.copy()
308
content = base_content
309
delta = self.parse_line_delta(record, version_id)
310
content.apply_delta(delta, version_id)
312
content = self.parse_fulltext(record, version_id)
314
content._should_strip_eol = noeol
315
return (content, delta)
318
class KnitAnnotateFactory(_KnitFactory):
319
"""Factory for creating annotated Content objects."""
323
def make(self, lines, version_id):
324
num_lines = len(lines)
325
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
327
def parse_fulltext(self, content, version_id):
328
"""Convert fulltext to internal representation
330
fulltext content is of the format
331
revid(utf8) plaintext\n
332
internal representation is of the format:
335
# TODO: jam 20070209 The tests expect this to be returned as tuples,
336
# but the code itself doesn't really depend on that.
337
# Figure out a way to not require the overhead of turning the
338
# list back into tuples.
339
lines = [tuple(line.split(' ', 1)) for line in content]
340
return AnnotatedKnitContent(lines)
342
def parse_line_delta_iter(self, lines):
343
return iter(self.parse_line_delta(lines))
345
def parse_line_delta(self, lines, version_id, plain=False):
346
"""Convert a line based delta into internal representation.
348
line delta is in the form of:
349
intstart intend intcount
351
revid(utf8) newline\n
352
internal representation is
353
(start, end, count, [1..count tuples (revid, newline)])
355
:param plain: If True, the lines are returned as a plain
356
list without annotations, not as a list of (origin, content) tuples, i.e.
357
(start, end, count, [1..count newline])
364
def cache_and_return(line):
365
origin, text = line.split(' ', 1)
366
return cache.setdefault(origin, origin), text
368
# walk through the lines parsing.
369
# Note that the plain test is explicitly pulled out of the
370
# loop to minimise any performance impact
373
start, end, count = [int(n) for n in header.split(',')]
374
contents = [next().split(' ', 1)[1] for i in xrange(count)]
375
result.append((start, end, count, contents))
378
start, end, count = [int(n) for n in header.split(',')]
379
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
380
result.append((start, end, count, contents))
383
def get_fulltext_content(self, lines):
384
"""Extract just the content lines from a fulltext."""
385
return (line.split(' ', 1)[1] for line in lines)
387
def get_linedelta_content(self, lines):
388
"""Extract just the content from a line delta.
390
This doesn't return all of the extra information stored in a delta.
391
Only the actual content lines.
396
header = header.split(',')
397
count = int(header[2])
398
for i in xrange(count):
399
origin, text = next().split(' ', 1)
402
def lower_fulltext(self, content):
403
"""convert a fulltext content record into a serializable form.
405
see parse_fulltext which this inverts.
407
# TODO: jam 20070209 We only do the caching thing to make sure that
408
# the origin is a valid utf-8 line, eventually we could remove it
409
return ['%s %s' % (o, t) for o, t in content._lines]
411
def lower_line_delta(self, delta):
412
"""convert a delta into a serializable form.
414
See parse_line_delta which this inverts.
416
# TODO: jam 20070209 We only do the caching thing to make sure that
417
# the origin is a valid utf-8 line, eventually we could remove it
419
for start, end, c, lines in delta:
420
out.append('%d,%d,%d\n' % (start, end, c))
421
out.extend(origin + ' ' + text
422
for origin, text in lines)
425
def annotate(self, knit, version_id):
426
content = knit._get_content(version_id)
427
return content.annotate()
430
class KnitPlainFactory(_KnitFactory):
431
"""Factory for creating plain Content objects."""
435
def make(self, lines, version_id):
436
return PlainKnitContent(lines, version_id)
438
def parse_fulltext(self, content, version_id):
439
"""This parses an unannotated fulltext.
441
Note that this is not a noop - the internal representation
442
has (versionid, line) - its just a constant versionid.
444
return self.make(content, version_id)
446
def parse_line_delta_iter(self, lines, version_id):
448
num_lines = len(lines)
449
while cur < num_lines:
452
start, end, c = [int(n) for n in header.split(',')]
453
yield start, end, c, lines[cur:cur+c]
456
def parse_line_delta(self, lines, version_id):
457
return list(self.parse_line_delta_iter(lines, version_id))
459
def get_fulltext_content(self, lines):
460
"""Extract just the content lines from a fulltext."""
463
def get_linedelta_content(self, lines):
464
"""Extract just the content from a line delta.
466
This doesn't return all of the extra information stored in a delta.
467
Only the actual content lines.
472
header = header.split(',')
473
count = int(header[2])
474
for i in xrange(count):
477
def lower_fulltext(self, content):
478
return content.text()
480
def lower_line_delta(self, delta):
482
for start, end, c, lines in delta:
483
out.append('%d,%d,%d\n' % (start, end, c))
487
def annotate(self, knit, version_id):
488
annotator = _KnitAnnotator(knit)
489
return annotator.annotate(version_id)
492
def make_empty_knit(transport, relpath):
493
"""Construct a empty knit at the specified location."""
494
k = make_file_knit(transport, relpath, 'w', KnitPlainFactory)
497
def make_file_knit(name, transport, file_mode=None, access_mode='w',
498
factory=None, delta=True, create=False, create_parent_dir=False,
499
delay_create=False, dir_mode=None, get_scope=None):
500
"""Factory to create a KnitVersionedFile for a .knit/.kndx file pair."""
502
factory = KnitAnnotateFactory()
504
factory = KnitPlainFactory()
505
if get_scope is None:
506
get_scope = lambda:None
507
index = _KnitIndex(transport, name + INDEX_SUFFIX,
508
access_mode, create=create, file_mode=file_mode,
509
create_parent_dir=create_parent_dir, delay_create=delay_create,
510
dir_mode=dir_mode, get_scope=get_scope)
511
access = _KnitAccess(transport, name + DATA_SUFFIX, file_mode,
512
dir_mode, ((create and not len(index)) and delay_create),
514
return KnitVersionedFile(name, transport, factory=factory,
515
create=create, delay_create=delay_create, index=index,
516
access_method=access)
520
"""Return the suffixes used by file based knits."""
521
return [DATA_SUFFIX, INDEX_SUFFIX]
522
make_file_knit.get_suffixes = get_suffixes
525
class KnitVersionedFile(VersionedFile):
526
"""Weave-like structure with faster random access.
528
A knit stores a number of texts and a summary of the relationships
529
between them. Texts are identified by a string version-id. Texts
530
are normally stored and retrieved as a series of lines, but can
531
also be passed as single strings.
533
Lines are stored with the trailing newline (if any) included, to
534
avoid special cases for files with no final newline. Lines are
535
composed of 8-bit characters, not unicode. The combination of
536
these approaches should mean any 'binary' file can be safely
537
stored and retrieved.
540
def __init__(self, relpath, transport, file_mode=None,
541
factory=None, delta=True, create=False, create_parent_dir=False,
542
delay_create=False, dir_mode=None, index=None, access_method=None):
543
"""Construct a knit at location specified by relpath.
545
:param create: If not True, only open an existing knit.
546
:param create_parent_dir: If True, create the parent directory if
547
creating the file fails. (This is used for stores with
548
hash-prefixes that may not exist yet)
549
:param delay_create: The calling code is aware that the knit won't
550
actually be created until the first data is stored.
551
:param index: An index to use for the knit.
553
super(KnitVersionedFile, self).__init__()
554
self.transport = transport
555
self.filename = relpath
556
self.factory = factory or KnitAnnotateFactory()
559
self._max_delta_chain = 200
561
if None in (access_method, index):
562
raise NotImplementedError, "No default access_method or index any more"
564
_access = access_method
565
if create and not len(self) and not delay_create:
567
self._data = _KnitData(_access)
570
return '%s(%s)' % (self.__class__.__name__,
571
self.transport.abspath(self.filename))
573
def _check_should_delta(self, first_parents):
574
"""Iterate back through the parent listing, looking for a fulltext.
576
This is used when we want to decide whether to add a delta or a new
577
fulltext. It searches for _max_delta_chain parents. When it finds a
578
fulltext parent, it sees if the total size of the deltas leading up to
579
it is large enough to indicate that we want a new full text anyway.
581
Return True if we should create a new delta, False if we should use a
586
delta_parents = first_parents
587
for count in xrange(self._max_delta_chain):
588
parent = delta_parents[0]
589
method = self._index.get_method(parent)
590
index, pos, size = self._index.get_position(parent)
591
if method == 'fulltext':
595
delta_parents = self._index.get_parent_map([parent])[parent]
597
# We couldn't find a fulltext, so we must create a new one
600
return fulltext_size > delta_size
602
def _check_write_ok(self):
603
return self._index._check_write_ok()
605
def _add_raw_records(self, records, data):
606
"""Add all the records 'records' with data pre-joined in 'data'.
608
:param records: A list of tuples(version_id, options, parents, size).
609
:param data: The data for the records. When it is written, the records
610
are adjusted to have pos pointing into data by the sum of
611
the preceding records sizes.
614
raw_record_sizes = [record[3] for record in records]
615
positions = self._data.add_raw_records(raw_record_sizes, data)
618
for (version_id, options, parents, size), access_memo in zip(
620
index_entries.append((version_id, options, access_memo, parents))
622
self._index.add_versions(index_entries)
624
def copy_to(self, name, transport):
625
"""See VersionedFile.copy_to()."""
626
# copy the current index to a temp index to avoid racing with local
628
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
629
self.transport.get(self._index._filename))
631
f = self._data._open_file()
633
transport.put_file(name + DATA_SUFFIX, f)
636
# move the copied index into place
637
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
639
def get_data_stream(self, required_versions):
640
"""Get a data stream for the specified versions.
642
Versions may be returned in any order, not necessarily the order
643
specified. They are returned in a partial order by compression
644
parent, so that the deltas can be applied as the data stream is
645
inserted; however note that compression parents will not be sent
646
unless they were specifically requested, as the client may already
649
:param required_versions: The exact set of versions to be extracted.
650
Unlike some other knit methods, this is not used to generate a
651
transitive closure, rather it is used precisely as given.
653
:returns: format_signature, list of (version, options, length, parents),
656
required_version_set = frozenset(required_versions)
658
# list of revisions that can just be sent without waiting for their
661
# map from revision to the children based on it
663
# first, read all relevant index data, enough to sort into the right
665
for version_id in required_versions:
666
options = self._index.get_options(version_id)
667
parents = self._index.get_parents_with_ghosts(version_id)
668
index_memo = self._index.get_position(version_id)
669
version_index[version_id] = (index_memo, options, parents)
670
if ('line-delta' in options
671
and parents[0] in required_version_set):
672
# must wait until the parent has been sent
673
deferred.setdefault(parents[0], []). \
676
# either a fulltext, or a delta whose parent the client did
677
# not ask for and presumably already has
678
ready_to_send.append(version_id)
679
# build a list of results to return, plus instructions for data to
681
copy_queue_records = []
682
temp_version_list = []
684
# XXX: pushing and popping lists may be a bit inefficient
685
version_id = ready_to_send.pop(0)
686
(index_memo, options, parents) = version_index[version_id]
687
copy_queue_records.append((version_id, index_memo))
688
none, data_pos, data_size = index_memo
689
temp_version_list.append((version_id, options, data_size,
691
if version_id in deferred:
692
# now we can send all the children of this revision - we could
693
# put them in anywhere, but we hope that sending them soon
694
# after the fulltext will give good locality in the receiver
695
ready_to_send[:0] = deferred.pop(version_id)
696
assert len(deferred) == 0, \
697
"Still have compressed child versions waiting to be sent"
698
# XXX: The stream format is such that we cannot stream it - we have to
699
# know the length of all the data a-priori.
701
result_version_list = []
702
for (version_id, raw_data), \
703
(version_id2, options, _, parents) in \
704
izip(self._data.read_records_iter_raw(copy_queue_records),
706
assert version_id == version_id2, \
707
'logic error, inconsistent results'
708
raw_datum.append(raw_data)
709
result_version_list.append(
710
(version_id, options, len(raw_data), parents))
711
# provide a callback to get data incrementally.
712
pseudo_file = StringIO(''.join(raw_datum))
715
return pseudo_file.read()
717
return pseudo_file.read(length)
718
return (self.get_format_signature(), result_version_list, read)
720
def _extract_blocks(self, version_id, source, target):
721
if self._index.get_method(version_id) != 'line-delta':
723
parent, sha1, noeol, delta = self.get_delta(version_id)
724
return KnitContent.get_line_delta_blocks(delta, source, target)
726
def get_delta(self, version_id):
727
"""Get a delta for constructing version from some other version."""
728
self.check_not_reserved_id(version_id)
729
parents = self.get_parent_map([version_id])[version_id]
734
index_memo = self._index.get_position(version_id)
735
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
736
noeol = 'no-eol' in self._index.get_options(version_id)
737
if 'fulltext' == self._index.get_method(version_id):
738
new_content = self.factory.parse_fulltext(data, version_id)
739
if parent is not None:
740
reference_content = self._get_content(parent)
741
old_texts = reference_content.text()
744
new_texts = new_content.text()
745
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
747
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
749
delta = self.factory.parse_line_delta(data, version_id)
750
return parent, sha1, noeol, delta
752
def get_format_signature(self):
753
"""See VersionedFile.get_format_signature()."""
754
if self.factory.annotated:
755
annotated_part = "annotated"
757
annotated_part = "plain"
758
return "knit-%s" % (annotated_part,)
760
@deprecated_method(one_four)
761
def get_graph_with_ghosts(self):
762
"""See VersionedFile.get_graph_with_ghosts()."""
763
return self.get_parent_map(self.versions())
765
def get_sha1s(self, version_ids):
766
"""See VersionedFile.get_sha1s()."""
767
record_map = self._get_record_map(version_ids)
768
# record entry 2 is the 'digest'.
769
return [record_map[v][2] for v in version_ids]
771
@deprecated_method(one_four)
772
def has_ghost(self, version_id):
773
"""True if there is a ghost reference in the file to version_id."""
775
if self.has_version(version_id):
777
# optimisable if needed by memoising the _ghosts set.
778
items = self.get_parent_map(self.versions())
779
for parents in items.itervalues():
780
for parent in parents:
781
if parent == version_id and parent not in items:
785
def insert_data_stream(self, (format, data_list, reader_callable)):
786
"""Insert knit records from a data stream into this knit.
788
If a version in the stream is already present in this knit, it will not
789
be inserted a second time. It will be checked for consistency with the
790
stored version however, and may cause a KnitCorrupt error to be raised
791
if the data in the stream disagrees with the already stored data.
793
:seealso: get_data_stream
795
if format != self.get_format_signature():
796
if 'knit' in debug.debug_flags:
798
'incompatible format signature inserting to %r', self)
799
source = self._knit_from_datastream(
800
(format, data_list, reader_callable))
804
for version_id, options, length, parents in data_list:
805
if self.has_version(version_id):
806
# First check: the list of parents.
807
my_parents = self.get_parents_with_ghosts(version_id)
808
if tuple(my_parents) != tuple(parents):
809
# XXX: KnitCorrupt is not quite the right exception here.
812
'parents list %r from data stream does not match '
813
'already recorded parents %r for %s'
814
% (parents, my_parents, version_id))
816
# Also check the SHA-1 of the fulltext this content will
818
raw_data = reader_callable(length)
819
my_fulltext_sha1 = self.get_sha1s([version_id])[0]
820
df, rec = self._data._parse_record_header(version_id, raw_data)
821
stream_fulltext_sha1 = rec[3]
822
if my_fulltext_sha1 != stream_fulltext_sha1:
823
# Actually, we don't know if it's this knit that's corrupt,
824
# or the data stream we're trying to insert.
826
self.filename, 'sha-1 does not match %s' % version_id)
828
if 'line-delta' in options:
829
# Make sure that this knit record is actually useful: a
830
# line-delta is no use unless we have its parent.
831
# Fetching from a broken repository with this problem
832
# shouldn't break the target repository.
834
# See https://bugs.launchpad.net/bzr/+bug/164443
835
if not self._index.has_version(parents[0]):
838
'line-delta from stream '
841
'missing parent %s\n'
842
'Try running "bzr check" '
843
'on the source repository, and "bzr reconcile" '
845
(version_id, parents[0]))
846
self._add_raw_records(
847
[(version_id, options, parents, length)],
848
reader_callable(length))
850
def _knit_from_datastream(self, (format, data_list, reader_callable)):
851
"""Create a knit object from a data stream.
853
This method exists to allow conversion of data streams that do not
854
match the signature of this knit. Generally it will be slower and use
855
more memory to use this method to insert data, but it will work.
857
:seealso: get_data_stream for details on datastreams.
858
:return: A knit versioned file which can be used to join the datastream
861
if format == "knit-plain":
862
factory = KnitPlainFactory()
863
elif format == "knit-annotated":
864
factory = KnitAnnotateFactory()
866
raise errors.KnitDataStreamUnknown(format)
867
index = _StreamIndex(data_list, self._index)
868
access = _StreamAccess(reader_callable, index, self, factory)
869
return KnitVersionedFile(self.filename, self.transport,
870
factory=factory, index=index, access_method=access)
873
"""See VersionedFile.versions."""
874
if 'evil' in debug.debug_flags:
875
trace.mutter_callsite(2, "versions scales with size of history")
876
return self._index.get_versions()
878
def has_version(self, version_id):
879
"""See VersionedFile.has_version."""
880
if 'evil' in debug.debug_flags:
881
trace.mutter_callsite(2, "has_version is a LBYL scenario")
882
return self._index.has_version(version_id)
884
__contains__ = has_version
886
def _merge_annotations(self, content, parents, parent_texts={},
887
delta=None, annotated=None,
888
left_matching_blocks=None):
889
"""Merge annotations for content. This is done by comparing
890
the annotations based on changed to the text.
892
if left_matching_blocks is not None:
893
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
897
for parent_id in parents:
898
merge_content = self._get_content(parent_id, parent_texts)
899
if (parent_id == parents[0] and delta_seq is not None):
902
seq = patiencediff.PatienceSequenceMatcher(
903
None, merge_content.text(), content.text())
904
for i, j, n in seq.get_matching_blocks():
907
# this appears to copy (origin, text) pairs across to the
908
# new content for any line that matches the last-checked
910
content._lines[j:j+n] = merge_content._lines[i:i+n]
912
if delta_seq is None:
913
reference_content = self._get_content(parents[0], parent_texts)
914
new_texts = content.text()
915
old_texts = reference_content.text()
916
delta_seq = patiencediff.PatienceSequenceMatcher(
917
None, old_texts, new_texts)
918
return self._make_line_delta(delta_seq, content)
920
def _make_line_delta(self, delta_seq, new_content):
921
"""Generate a line delta from delta_seq and new_content."""
923
for op in delta_seq.get_opcodes():
926
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
929
def _get_components_positions(self, version_ids):
930
"""Produce a map of position data for the components of versions.
932
This data is intended to be used for retrieving the knit records.
934
A dict of version_id to (record_details, index_memo, next, parents) is
936
method is the way referenced data should be applied.
937
index_memo is the handle to pass to the data access to actually get the
939
next is the build-parent of the version, or None for fulltexts.
940
parents is the version_ids of the parents of this version
943
pending_components = version_ids
944
while pending_components:
945
build_details = self._index.get_build_details(pending_components)
946
current_components = set(pending_components)
947
pending_components = set()
948
for version_id, details in build_details.iteritems():
949
(index_memo, compression_parent, parents,
950
record_details) = details
951
method = record_details[0]
952
if compression_parent is not None:
953
pending_components.add(compression_parent)
954
component_data[version_id] = (record_details, index_memo,
956
missing = current_components.difference(build_details)
958
raise errors.RevisionNotPresent(missing.pop(), self.filename)
959
return component_data
961
def _get_content(self, version_id, parent_texts={}):
962
"""Returns a content object that makes up the specified
964
cached_version = parent_texts.get(version_id, None)
965
if cached_version is not None:
966
if not self.has_version(version_id):
967
raise RevisionNotPresent(version_id, self.filename)
968
return cached_version
970
text_map, contents_map = self._get_content_maps([version_id])
971
return contents_map[version_id]
973
def _check_versions_present(self, version_ids):
974
"""Check that all specified versions are present."""
975
self._index.check_versions_present(version_ids)
977
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
978
nostore_sha, random_id, check_content, left_matching_blocks):
979
"""See VersionedFile.add_lines_with_ghosts()."""
980
self._check_add(version_id, lines, random_id, check_content)
981
return self._add(version_id, lines, parents, self.delta,
982
parent_texts, left_matching_blocks, nostore_sha, random_id)
984
def _add_lines(self, version_id, parents, lines, parent_texts,
985
left_matching_blocks, nostore_sha, random_id, check_content):
986
"""See VersionedFile.add_lines."""
987
self._check_add(version_id, lines, random_id, check_content)
988
self._check_versions_present(parents)
989
return self._add(version_id, lines[:], parents, self.delta,
990
parent_texts, left_matching_blocks, nostore_sha, random_id)
992
def _check_add(self, version_id, lines, random_id, check_content):
993
"""check that version_id and lines are safe to add."""
994
if contains_whitespace(version_id):
995
raise InvalidRevisionId(version_id, self.filename)
996
self.check_not_reserved_id(version_id)
997
# Technically this could be avoided if we are happy to allow duplicate
998
# id insertion when other things than bzr core insert texts, but it
999
# seems useful for folk using the knit api directly to have some safety
1000
# blanket that we can disable.
1001
if not random_id and self.has_version(version_id):
1002
raise RevisionAlreadyPresent(version_id, self.filename)
1004
self._check_lines_not_unicode(lines)
1005
self._check_lines_are_lines(lines)
1007
def _add(self, version_id, lines, parents, delta, parent_texts,
1008
left_matching_blocks, nostore_sha, random_id):
1009
"""Add a set of lines on top of version specified by parents.
1011
If delta is true, compress the text as a line-delta against
1014
Any versions not present will be converted into ghosts.
1016
# first thing, if the content is something we don't need to store, find
1018
line_bytes = ''.join(lines)
1019
digest = sha_string(line_bytes)
1020
if nostore_sha == digest:
1021
raise errors.ExistingContent
1023
present_parents = []
1024
if parent_texts is None:
1026
for parent in parents:
1027
if self.has_version(parent):
1028
present_parents.append(parent)
1030
# can only compress against the left most present parent.
1032
(len(present_parents) == 0 or
1033
present_parents[0] != parents[0])):
1036
text_length = len(line_bytes)
1039
if lines[-1][-1] != '\n':
1040
# copy the contents of lines.
1042
options.append('no-eol')
1043
lines[-1] = lines[-1] + '\n'
1047
# To speed the extract of texts the delta chain is limited
1048
# to a fixed number of deltas. This should minimize both
1049
# I/O and the time spend applying deltas.
1050
delta = self._check_should_delta(present_parents)
1052
assert isinstance(version_id, str)
1053
content = self.factory.make(lines, version_id)
1054
if delta or (self.factory.annotated and len(present_parents) > 0):
1055
# Merge annotations from parent texts if needed.
1056
delta_hunks = self._merge_annotations(content, present_parents,
1057
parent_texts, delta, self.factory.annotated,
1058
left_matching_blocks)
1061
options.append('line-delta')
1062
store_lines = self.factory.lower_line_delta(delta_hunks)
1063
size, bytes = self._data._record_to_data(version_id, digest,
1066
options.append('fulltext')
1067
# isinstance is slower and we have no hierarchy.
1068
if self.factory.__class__ == KnitPlainFactory:
1069
# Use the already joined bytes saving iteration time in
1071
size, bytes = self._data._record_to_data(version_id, digest,
1072
lines, [line_bytes])
1074
# get mixed annotation + content and feed it into the
1076
store_lines = self.factory.lower_fulltext(content)
1077
size, bytes = self._data._record_to_data(version_id, digest,
1080
access_memo = self._data.add_raw_records([size], bytes)[0]
1081
self._index.add_versions(
1082
((version_id, options, access_memo, parents),),
1083
random_id=random_id)
1084
return digest, text_length, content
1086
def check(self, progress_bar=None):
1087
"""See VersionedFile.check()."""
1089
def get_lines(self, version_id):
1090
"""See VersionedFile.get_lines()."""
1091
return self.get_line_list([version_id])[0]
1093
def _get_record_map(self, version_ids):
1094
"""Produce a dictionary of knit records.
1096
:return: {version_id:(record, record_details, digest, next)}
1098
data returned from read_records
1100
opaque information to pass to parse_record
1102
SHA1 digest of the full text after all steps are done
1104
build-parent of the version, i.e. the leftmost ancestor.
1105
Will be None if the record is not a delta.
1107
position_map = self._get_components_positions(version_ids)
1108
# c = component_id, r = record_details, i_m = index_memo, n = next
1109
records = [(c, i_m) for c, (r, i_m, n)
1110
in position_map.iteritems()]
1112
for component_id, record, digest in \
1113
self._data.read_records_iter(records):
1114
(record_details, index_memo, next) = position_map[component_id]
1115
record_map[component_id] = record, record_details, digest, next
1119
def get_text(self, version_id):
1120
"""See VersionedFile.get_text"""
1121
return self.get_texts([version_id])[0]
1123
def get_texts(self, version_ids):
1124
return [''.join(l) for l in self.get_line_list(version_ids)]
1126
def get_line_list(self, version_ids):
1127
"""Return the texts of listed versions as a list of strings."""
1128
for version_id in version_ids:
1129
self.check_not_reserved_id(version_id)
1130
text_map, content_map = self._get_content_maps(version_ids)
1131
return [text_map[v] for v in version_ids]
1133
_get_lf_split_line_list = get_line_list
1135
def _get_content_maps(self, version_ids):
1136
"""Produce maps of text and KnitContents
1138
:return: (text_map, content_map) where text_map contains the texts for
1139
the requested versions and content_map contains the KnitContents.
1140
Both dicts take version_ids as their keys.
1142
# FUTURE: This function could be improved for the 'extract many' case
1143
# by tracking each component and only doing the copy when the number of
1144
# children than need to apply delta's to it is > 1 or it is part of the
1146
version_ids = list(version_ids)
1147
multiple_versions = len(version_ids) != 1
1148
record_map = self._get_record_map(version_ids)
1153
for version_id in version_ids:
1156
while cursor is not None:
1157
record, record_details, digest, next = record_map[cursor]
1158
components.append((cursor, record, record_details, digest))
1159
if cursor in content_map:
1164
for (component_id, record, record_details,
1165
digest) in reversed(components):
1166
if component_id in content_map:
1167
content = content_map[component_id]
1169
content, delta = self.factory.parse_record(version_id,
1170
record, record_details, content,
1171
copy_base_content=multiple_versions)
1172
if multiple_versions:
1173
content_map[component_id] = content
1175
content.cleanup_eol(copy_on_mutate=multiple_versions)
1176
final_content[version_id] = content
1178
# digest here is the digest from the last applied component.
1179
text = content.text()
1180
actual_sha = sha_strings(text)
1181
if actual_sha != digest:
1182
raise KnitCorrupt(self.filename,
1184
'\n of reconstructed text does not match'
1186
'\n for version %s' %
1187
(actual_sha, digest, version_id))
1188
text_map[version_id] = text
1189
return text_map, final_content
1191
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1193
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1194
if version_ids is None:
1195
version_ids = self.versions()
1197
pb = progress.DummyProgress()
1198
# we don't care about inclusions, the caller cares.
1199
# but we need to setup a list of records to visit.
1200
# we need version_id, position, length
1201
version_id_records = []
1202
requested_versions = set(version_ids)
1203
# filter for available versions
1204
for version_id in requested_versions:
1205
if not self.has_version(version_id):
1206
raise RevisionNotPresent(version_id, self.filename)
1207
# get a in-component-order queue:
1208
for version_id in self.versions():
1209
if version_id in requested_versions:
1210
index_memo = self._index.get_position(version_id)
1211
version_id_records.append((version_id, index_memo))
1213
total = len(version_id_records)
1214
for version_idx, (version_id, data, sha_value) in \
1215
enumerate(self._data.read_records_iter(version_id_records)):
1216
pb.update('Walking content.', version_idx, total)
1217
method = self._index.get_method(version_id)
1219
assert method in ('fulltext', 'line-delta')
1220
if method == 'fulltext':
1221
line_iterator = self.factory.get_fulltext_content(data)
1223
line_iterator = self.factory.get_linedelta_content(data)
1224
# XXX: It might be more efficient to yield (version_id,
1225
# line_iterator) in the future. However for now, this is a simpler
1226
# change to integrate into the rest of the codebase. RBC 20071110
1227
for line in line_iterator:
1228
yield line, version_id
1230
pb.update('Walking content.', total, total)
1232
def num_versions(self):
1233
"""See VersionedFile.num_versions()."""
1234
return self._index.num_versions()
1236
__len__ = num_versions
1238
def annotate(self, version_id):
1239
"""See VersionedFile.annotate."""
1240
return self.factory.annotate(self, version_id)
1242
def get_parent_map(self, version_ids):
1243
"""See VersionedFile.get_parent_map."""
1244
return self._index.get_parent_map(version_ids)
1246
def get_ancestry(self, versions, topo_sorted=True):
1247
"""See VersionedFile.get_ancestry."""
1248
if isinstance(versions, basestring):
1249
versions = [versions]
1252
return self._index.get_ancestry(versions, topo_sorted)
1254
def get_ancestry_with_ghosts(self, versions):
1255
"""See VersionedFile.get_ancestry_with_ghosts."""
1256
if isinstance(versions, basestring):
1257
versions = [versions]
1260
return self._index.get_ancestry_with_ghosts(versions)
1262
def plan_merge(self, ver_a, ver_b):
1263
"""See VersionedFile.plan_merge."""
1264
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1265
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1266
annotated_a = self.annotate(ver_a)
1267
annotated_b = self.annotate(ver_b)
1268
return merge._plan_annotate_merge(annotated_a, annotated_b,
1269
ancestors_a, ancestors_b)
1272
class _KnitComponentFile(object):
1273
"""One of the files used to implement a knit database"""
1275
def __init__(self, transport, filename, mode, file_mode=None,
1276
create_parent_dir=False, dir_mode=None):
1277
self._transport = transport
1278
self._filename = filename
1280
self._file_mode = file_mode
1281
self._dir_mode = dir_mode
1282
self._create_parent_dir = create_parent_dir
1283
self._need_to_create = False
1285
def _full_path(self):
1286
"""Return the full path to this file."""
1287
return self._transport.base + self._filename
1289
def check_header(self, fp):
1290
line = fp.readline()
1292
# An empty file can actually be treated as though the file doesn't
1294
raise errors.NoSuchFile(self._full_path())
1295
if line != self.HEADER:
1296
raise KnitHeaderError(badline=line,
1297
filename=self._transport.abspath(self._filename))
1300
return '%s(%s)' % (self.__class__.__name__, self._filename)
1303
class _KnitIndex(_KnitComponentFile):
1304
"""Manages knit index file.
1306
The index is already kept in memory and read on startup, to enable
1307
fast lookups of revision information. The cursor of the index
1308
file is always pointing to the end, making it easy to append
1311
_cache is a cache for fast mapping from version id to a Index
1314
_history is a cache for fast mapping from indexes to version ids.
1316
The index data format is dictionary compressed when it comes to
1317
parent references; a index entry may only have parents that with a
1318
lover index number. As a result, the index is topological sorted.
1320
Duplicate entries may be written to the index for a single version id
1321
if this is done then the latter one completely replaces the former:
1322
this allows updates to correct version and parent information.
1323
Note that the two entries may share the delta, and that successive
1324
annotations and references MUST point to the first entry.
1326
The index file on disc contains a header, followed by one line per knit
1327
record. The same revision can be present in an index file more than once.
1328
The first occurrence gets assigned a sequence number starting from 0.
1330
The format of a single line is
1331
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1332
REVISION_ID is a utf8-encoded revision id
1333
FLAGS is a comma separated list of flags about the record. Values include
1334
no-eol, line-delta, fulltext.
1335
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1336
that the the compressed data starts at.
1337
LENGTH is the ascii representation of the length of the data file.
1338
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1340
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1341
revision id already in the knit that is a parent of REVISION_ID.
1342
The ' :' marker is the end of record marker.
1345
when a write is interrupted to the index file, it will result in a line
1346
that does not end in ' :'. If the ' :' is not present at the end of a line,
1347
or at the end of the file, then the record that is missing it will be
1348
ignored by the parser.
1350
When writing new records to the index file, the data is preceded by '\n'
1351
to ensure that records always start on new lines even if the last write was
1352
interrupted. As a result its normal for the last line in the index to be
1353
missing a trailing newline. One can be added with no harmful effects.
1356
HEADER = "# bzr knit index 8\n"
1358
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1359
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1361
def _cache_version(self, version_id, options, pos, size, parents):
1362
"""Cache a version record in the history array and index cache.
1364
This is inlined into _load_data for performance. KEEP IN SYNC.
1365
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1368
# only want the _history index to reference the 1st index entry
1370
if version_id not in self._cache:
1371
index = len(self._history)
1372
self._history.append(version_id)
1374
index = self._cache[version_id][5]
1375
self._cache[version_id] = (version_id,
1382
def _check_write_ok(self):
1383
if self._get_scope() != self._scope:
1384
raise errors.OutSideTransaction()
1385
if self._mode != 'w':
1386
raise errors.ReadOnlyObjectDirtiedError(self)
1388
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1389
create_parent_dir=False, delay_create=False, dir_mode=None,
1391
_KnitComponentFile.__init__(self, transport, filename, mode,
1392
file_mode=file_mode,
1393
create_parent_dir=create_parent_dir,
1396
# position in _history is the 'official' index for a revision
1397
# but the values may have come from a newer entry.
1398
# so - wc -l of a knit index is != the number of unique names
1402
fp = self._transport.get(self._filename)
1404
# _load_data may raise NoSuchFile if the target knit is
1406
_load_data(self, fp)
1410
if mode != 'w' or not create:
1413
self._need_to_create = True
1415
self._transport.put_bytes_non_atomic(
1416
self._filename, self.HEADER, mode=self._file_mode)
1417
self._scope = get_scope()
1418
self._get_scope = get_scope
1420
def get_ancestry(self, versions, topo_sorted=True):
1421
"""See VersionedFile.get_ancestry."""
1422
# get a graph of all the mentioned versions:
1424
pending = set(versions)
1427
version = pending.pop()
1430
parents = [p for p in cache[version][4] if p in cache]
1432
raise RevisionNotPresent(version, self._filename)
1433
# if not completed and not a ghost
1434
pending.update([p for p in parents if p not in graph])
1435
graph[version] = parents
1438
return topo_sort(graph.items())
1440
def get_ancestry_with_ghosts(self, versions):
1441
"""See VersionedFile.get_ancestry_with_ghosts."""
1442
# get a graph of all the mentioned versions:
1443
self.check_versions_present(versions)
1446
pending = set(versions)
1448
version = pending.pop()
1450
parents = cache[version][4]
1456
pending.update([p for p in parents if p not in graph])
1457
graph[version] = parents
1458
return topo_sort(graph.items())
1460
def get_build_details(self, version_ids):
1461
"""Get the method, index_memo and compression parent for version_ids.
1463
Ghosts are omitted from the result.
1465
:param version_ids: An iterable of version_ids.
1466
:return: A dict of version_id:(index_memo, compression_parent,
1467
parents, record_details).
1469
opaque structure to pass to read_records to extract the raw
1472
Content that this record is built upon, may be None
1474
Logical parents of this node
1476
extra information about the content which needs to be passed to
1477
Factory.parse_record
1480
for version_id in version_ids:
1481
if version_id not in self._cache:
1482
# ghosts are omitted
1484
method = self.get_method(version_id)
1485
parents = self.get_parents_with_ghosts(version_id)
1486
if method == 'fulltext':
1487
compression_parent = None
1489
compression_parent = parents[0]
1490
noeol = 'no-eol' in self.get_options(version_id)
1491
index_memo = self.get_position(version_id)
1492
result[version_id] = (index_memo, compression_parent,
1493
parents, (method, noeol))
1496
def num_versions(self):
1497
return len(self._history)
1499
__len__ = num_versions
1501
def get_versions(self):
1502
"""Get all the versions in the file. not topologically sorted."""
1503
return self._history
1505
def _version_list_to_index(self, versions):
1508
for version in versions:
1509
if version in cache:
1510
# -- inlined lookup() --
1511
result_list.append(str(cache[version][5]))
1512
# -- end lookup () --
1514
result_list.append('.' + version)
1515
return ' '.join(result_list)
1517
def add_version(self, version_id, options, index_memo, parents):
1518
"""Add a version record to the index."""
1519
self.add_versions(((version_id, options, index_memo, parents),))
1521
def add_versions(self, versions, random_id=False):
1522
"""Add multiple versions to the index.
1524
:param versions: a list of tuples:
1525
(version_id, options, pos, size, parents).
1526
:param random_id: If True the ids being added were randomly generated
1527
and no check for existence will be performed.
1530
orig_history = self._history[:]
1531
orig_cache = self._cache.copy()
1534
for version_id, options, (index, pos, size), parents in versions:
1535
line = "\n%s %s %s %s %s :" % (version_id,
1539
self._version_list_to_index(parents))
1540
assert isinstance(line, str), \
1541
'content must be utf-8 encoded: %r' % (line,)
1543
self._cache_version(version_id, options, pos, size, tuple(parents))
1544
if not self._need_to_create:
1545
self._transport.append_bytes(self._filename, ''.join(lines))
1548
sio.write(self.HEADER)
1549
sio.writelines(lines)
1551
self._transport.put_file_non_atomic(self._filename, sio,
1552
create_parent_dir=self._create_parent_dir,
1553
mode=self._file_mode,
1554
dir_mode=self._dir_mode)
1555
self._need_to_create = False
1557
# If any problems happen, restore the original values and re-raise
1558
self._history = orig_history
1559
self._cache = orig_cache
1562
def has_version(self, version_id):
1563
"""True if the version is in the index."""
1564
return version_id in self._cache
1566
def get_position(self, version_id):
1567
"""Return details needed to access the version.
1569
.kndx indices do not support split-out data, so return None for the
1572
:return: a tuple (None, data position, size) to hand to the access
1573
logic to get the record.
1575
entry = self._cache[version_id]
1576
return None, entry[2], entry[3]
1578
def get_method(self, version_id):
1579
"""Return compression method of specified version."""
1581
options = self._cache[version_id][1]
1583
raise RevisionNotPresent(version_id, self._filename)
1584
if 'fulltext' in options:
1587
if 'line-delta' not in options:
1588
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1591
def get_options(self, version_id):
1592
"""Return a list representing options.
1596
return self._cache[version_id][1]
1598
def get_parent_map(self, version_ids):
1599
"""Passed through to by KnitVersionedFile.get_parent_map."""
1601
for version_id in version_ids:
1603
result[version_id] = tuple(self._cache[version_id][4])
1608
def get_parents_with_ghosts(self, version_id):
1609
"""Return parents of specified version with ghosts."""
1611
return self.get_parent_map([version_id])[version_id]
1613
raise RevisionNotPresent(version_id, self)
1615
def check_versions_present(self, version_ids):
1616
"""Check that all specified versions are present."""
1618
for version_id in version_ids:
1619
if version_id not in cache:
1620
raise RevisionNotPresent(version_id, self._filename)
1623
class KnitGraphIndex(object):
1624
"""A knit index that builds on GraphIndex."""
1626
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1627
"""Construct a KnitGraphIndex on a graph_index.
1629
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1630
:param deltas: Allow delta-compressed records.
1631
:param add_callback: If not None, allow additions to the index and call
1632
this callback with a list of added GraphIndex nodes:
1633
[(node, value, node_refs), ...]
1634
:param parents: If True, record knits parents, if not do not record
1637
self._graph_index = graph_index
1638
self._deltas = deltas
1639
self._add_callback = add_callback
1640
self._parents = parents
1641
if deltas and not parents:
1642
raise KnitCorrupt(self, "Cannot do delta compression without "
1645
def _check_write_ok(self):
1648
def _get_entries(self, keys, check_present=False):
1649
"""Get the entries for keys.
1651
:param keys: An iterable of index keys, - 1-tuples.
1656
for node in self._graph_index.iter_entries(keys):
1658
found_keys.add(node[1])
1660
# adapt parentless index to the rest of the code.
1661
for node in self._graph_index.iter_entries(keys):
1662
yield node[0], node[1], node[2], ()
1663
found_keys.add(node[1])
1665
missing_keys = keys.difference(found_keys)
1667
raise RevisionNotPresent(missing_keys.pop(), self)
1669
def _present_keys(self, version_ids):
1671
node[1] for node in self._get_entries(version_ids)])
1673
def _parentless_ancestry(self, versions):
1674
"""Honour the get_ancestry API for parentless knit indices."""
1675
wanted_keys = self._version_ids_to_keys(versions)
1676
present_keys = self._present_keys(wanted_keys)
1677
missing = set(wanted_keys).difference(present_keys)
1679
raise RevisionNotPresent(missing.pop(), self)
1680
return list(self._keys_to_version_ids(present_keys))
1682
def get_ancestry(self, versions, topo_sorted=True):
1683
"""See VersionedFile.get_ancestry."""
1684
if not self._parents:
1685
return self._parentless_ancestry(versions)
1686
# XXX: This will do len(history) index calls - perhaps
1687
# it should be altered to be a index core feature?
1688
# get a graph of all the mentioned versions:
1691
versions = self._version_ids_to_keys(versions)
1692
pending = set(versions)
1694
# get all pending nodes
1695
this_iteration = pending
1696
new_nodes = self._get_entries(this_iteration)
1699
for (index, key, value, node_refs) in new_nodes:
1700
# dont ask for ghosties - otherwise
1701
# we we can end up looping with pending
1702
# being entirely ghosted.
1703
graph[key] = [parent for parent in node_refs[0]
1704
if parent not in ghosts]
1706
for parent in graph[key]:
1707
# dont examine known nodes again
1712
ghosts.update(this_iteration.difference(found))
1713
if versions.difference(graph):
1714
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1716
result_keys = topo_sort(graph.items())
1718
result_keys = graph.iterkeys()
1719
return [key[0] for key in result_keys]
1721
def get_ancestry_with_ghosts(self, versions):
1722
"""See VersionedFile.get_ancestry."""
1723
if not self._parents:
1724
return self._parentless_ancestry(versions)
1725
# XXX: This will do len(history) index calls - perhaps
1726
# it should be altered to be a index core feature?
1727
# get a graph of all the mentioned versions:
1729
versions = self._version_ids_to_keys(versions)
1730
pending = set(versions)
1732
# get all pending nodes
1733
this_iteration = pending
1734
new_nodes = self._get_entries(this_iteration)
1736
for (index, key, value, node_refs) in new_nodes:
1737
graph[key] = node_refs[0]
1739
for parent in graph[key]:
1740
# dont examine known nodes again
1744
missing_versions = this_iteration.difference(graph)
1745
missing_needed = versions.intersection(missing_versions)
1747
raise RevisionNotPresent(missing_needed.pop(), self)
1748
for missing_version in missing_versions:
1749
# add a key, no parents
1750
graph[missing_version] = []
1751
pending.discard(missing_version) # don't look for it
1752
result_keys = topo_sort(graph.items())
1753
return [key[0] for key in result_keys]
1755
def get_build_details(self, version_ids):
1756
"""Get the method, index_memo and compression parent for version_ids.
1758
Ghosts are omitted from the result.
1760
:param version_ids: An iterable of version_ids.
1761
:return: A dict of version_id:(index_memo, compression_parent,
1762
parents, record_details).
1764
opaque structure to pass to read_records to extract the raw
1767
Content that this record is built upon, may be None
1769
Logical parents of this node
1771
extra information about the content which needs to be passed to
1772
Factory.parse_record
1775
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
1776
for entry in entries:
1777
version_id = self._keys_to_version_ids((entry[1],))[0]
1778
if not self._parents:
1781
parents = self._keys_to_version_ids(entry[3][0])
1782
if not self._deltas:
1783
compression_parent = None
1785
compression_parent_key = self._compression_parent(entry)
1786
if compression_parent_key:
1787
compression_parent = self._keys_to_version_ids(
1788
(compression_parent_key,))[0]
1790
compression_parent = None
1791
noeol = (entry[2][0] == 'N')
1792
if compression_parent:
1793
method = 'line-delta'
1796
result[version_id] = (self._node_to_position(entry),
1797
compression_parent, parents,
1801
def _compression_parent(self, an_entry):
1802
# return the key that an_entry is compressed against, or None
1803
# Grab the second parent list (as deltas implies parents currently)
1804
compression_parents = an_entry[3][1]
1805
if not compression_parents:
1807
assert len(compression_parents) == 1
1808
return compression_parents[0]
1810
def _get_method(self, node):
1811
if not self._deltas:
1813
if self._compression_parent(node):
1818
def num_versions(self):
1819
return len(list(self._graph_index.iter_all_entries()))
1821
__len__ = num_versions
1823
def get_versions(self):
1824
"""Get all the versions in the file. not topologically sorted."""
1825
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1827
def has_version(self, version_id):
1828
"""True if the version is in the index."""
1829
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1831
def _keys_to_version_ids(self, keys):
1832
return tuple(key[0] for key in keys)
1834
def get_position(self, version_id):
1835
"""Return details needed to access the version.
1837
:return: a tuple (index, data position, size) to hand to the access
1838
logic to get the record.
1840
node = self._get_node(version_id)
1841
return self._node_to_position(node)
1843
def _node_to_position(self, node):
1844
"""Convert an index value to position details."""
1845
bits = node[2][1:].split(' ')
1846
return node[0], int(bits[0]), int(bits[1])
1848
def get_method(self, version_id):
1849
"""Return compression method of specified version."""
1850
return self._get_method(self._get_node(version_id))
1852
def _get_node(self, version_id):
1854
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1856
raise RevisionNotPresent(version_id, self)
1858
def get_options(self, version_id):
1859
"""Return a list representing options.
1863
node = self._get_node(version_id)
1864
options = [self._get_method(node)]
1865
if node[2][0] == 'N':
1866
options.append('no-eol')
1869
def get_parent_map(self, version_ids):
1870
"""Passed through to by KnitVersionedFile.get_parent_map."""
1871
nodes = self._get_entries(self._version_ids_to_keys(version_ids))
1875
result[node[1][0]] = self._keys_to_version_ids(node[3][0])
1878
result[node[1][0]] = ()
1881
def get_parents_with_ghosts(self, version_id):
1882
"""Return parents of specified version with ghosts."""
1884
return self.get_parent_map([version_id])[version_id]
1886
raise RevisionNotPresent(version_id, self)
1888
def check_versions_present(self, version_ids):
1889
"""Check that all specified versions are present."""
1890
keys = self._version_ids_to_keys(version_ids)
1891
present = self._present_keys(keys)
1892
missing = keys.difference(present)
1894
raise RevisionNotPresent(missing.pop(), self)
1896
def add_version(self, version_id, options, access_memo, parents):
1897
"""Add a version record to the index."""
1898
return self.add_versions(((version_id, options, access_memo, parents),))
1900
def add_versions(self, versions, random_id=False):
1901
"""Add multiple versions to the index.
1903
This function does not insert data into the Immutable GraphIndex
1904
backing the KnitGraphIndex, instead it prepares data for insertion by
1905
the caller and checks that it is safe to insert then calls
1906
self._add_callback with the prepared GraphIndex nodes.
1908
:param versions: a list of tuples:
1909
(version_id, options, pos, size, parents).
1910
:param random_id: If True the ids being added were randomly generated
1911
and no check for existence will be performed.
1913
if not self._add_callback:
1914
raise errors.ReadOnlyError(self)
1915
# we hope there are no repositories with inconsistent parentage
1920
for (version_id, options, access_memo, parents) in versions:
1921
index, pos, size = access_memo
1922
key = (version_id, )
1923
parents = tuple((parent, ) for parent in parents)
1924
if 'no-eol' in options:
1928
value += "%d %d" % (pos, size)
1929
if not self._deltas:
1930
if 'line-delta' in options:
1931
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1934
if 'line-delta' in options:
1935
node_refs = (parents, (parents[0],))
1937
node_refs = (parents, ())
1939
node_refs = (parents, )
1942
raise KnitCorrupt(self, "attempt to add node with parents "
1943
"in parentless index.")
1945
keys[key] = (value, node_refs)
1947
present_nodes = self._get_entries(keys)
1948
for (index, key, value, node_refs) in present_nodes:
1949
if (value, node_refs) != keys[key]:
1950
raise KnitCorrupt(self, "inconsistent details in add_versions"
1951
": %s %s" % ((value, node_refs), keys[key]))
1955
for key, (value, node_refs) in keys.iteritems():
1956
result.append((key, value, node_refs))
1958
for key, (value, node_refs) in keys.iteritems():
1959
result.append((key, value))
1960
self._add_callback(result)
1962
def _version_ids_to_keys(self, version_ids):
1963
return set((version_id, ) for version_id in version_ids)
1966
class _KnitAccess(object):
1967
"""Access to knit records in a .knit file."""
1969
def __init__(self, transport, filename, _file_mode, _dir_mode,
1970
_need_to_create, _create_parent_dir):
1971
"""Create a _KnitAccess for accessing and inserting data.
1973
:param transport: The transport the .knit is located on.
1974
:param filename: The filename of the .knit.
1976
self._transport = transport
1977
self._filename = filename
1978
self._file_mode = _file_mode
1979
self._dir_mode = _dir_mode
1980
self._need_to_create = _need_to_create
1981
self._create_parent_dir = _create_parent_dir
1983
def add_raw_records(self, sizes, raw_data):
1984
"""Add raw knit bytes to a storage area.
1986
The data is spooled to whereever the access method is storing data.
1988
:param sizes: An iterable containing the size of each raw data segment.
1989
:param raw_data: A bytestring containing the data.
1990
:return: A list of memos to retrieve the record later. Each memo is a
1991
tuple - (index, pos, length), where the index field is always None
1992
for the .knit access method.
1994
assert type(raw_data) == str, \
1995
'data must be plain bytes was %s' % type(raw_data)
1996
if not self._need_to_create:
1997
base = self._transport.append_bytes(self._filename, raw_data)
1999
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2000
create_parent_dir=self._create_parent_dir,
2001
mode=self._file_mode,
2002
dir_mode=self._dir_mode)
2003
self._need_to_create = False
2007
result.append((None, base, size))
2012
"""IFF this data access has its own storage area, initialise it.
2016
self._transport.put_bytes_non_atomic(self._filename, '',
2017
mode=self._file_mode)
2019
def open_file(self):
2020
"""IFF this data access can be represented as a single file, open it.
2022
For knits that are not mapped to a single file on disk this will
2025
:return: None or a file handle.
2028
return self._transport.get(self._filename)
2033
def get_raw_records(self, memos_for_retrieval):
2034
"""Get the raw bytes for a records.
2036
:param memos_for_retrieval: An iterable containing the (index, pos,
2037
length) memo for retrieving the bytes. The .knit method ignores
2038
the index as there is always only a single file.
2039
:return: An iterator over the bytes of the records.
2041
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2042
for pos, data in self._transport.readv(self._filename, read_vector):
2046
class _PackAccess(object):
2047
"""Access to knit records via a collection of packs."""
2049
def __init__(self, index_to_packs, writer=None):
2050
"""Create a _PackAccess object.
2052
:param index_to_packs: A dict mapping index objects to the transport
2053
and file names for obtaining data.
2054
:param writer: A tuple (pack.ContainerWriter, write_index) which
2055
contains the pack to write, and the index that reads from it will
2059
self.container_writer = writer[0]
2060
self.write_index = writer[1]
2062
self.container_writer = None
2063
self.write_index = None
2064
self.indices = index_to_packs
2066
def add_raw_records(self, sizes, raw_data):
2067
"""Add raw knit bytes to a storage area.
2069
The data is spooled to the container writer in one bytes-record per
2072
:param sizes: An iterable containing the size of each raw data segment.
2073
:param raw_data: A bytestring containing the data.
2074
:return: A list of memos to retrieve the record later. Each memo is a
2075
tuple - (index, pos, length), where the index field is the
2076
write_index object supplied to the PackAccess object.
2078
assert type(raw_data) == str, \
2079
'data must be plain bytes was %s' % type(raw_data)
2083
p_offset, p_length = self.container_writer.add_bytes_record(
2084
raw_data[offset:offset+size], [])
2086
result.append((self.write_index, p_offset, p_length))
2090
"""Pack based knits do not get individually created."""
2092
def get_raw_records(self, memos_for_retrieval):
2093
"""Get the raw bytes for a records.
2095
:param memos_for_retrieval: An iterable containing the (index, pos,
2096
length) memo for retrieving the bytes. The Pack access method
2097
looks up the pack to use for a given record in its index_to_pack
2099
:return: An iterator over the bytes of the records.
2101
# first pass, group into same-index requests
2103
current_index = None
2104
for (index, offset, length) in memos_for_retrieval:
2105
if current_index == index:
2106
current_list.append((offset, length))
2108
if current_index is not None:
2109
request_lists.append((current_index, current_list))
2110
current_index = index
2111
current_list = [(offset, length)]
2112
# handle the last entry
2113
if current_index is not None:
2114
request_lists.append((current_index, current_list))
2115
for index, offsets in request_lists:
2116
transport, path = self.indices[index]
2117
reader = pack.make_readv_reader(transport, path, offsets)
2118
for names, read_func in reader.iter_records():
2119
yield read_func(None)
2121
def open_file(self):
2122
"""Pack based knits have no single file."""
2125
def set_writer(self, writer, index, (transport, packname)):
2126
"""Set a writer to use for adding data."""
2127
if index is not None:
2128
self.indices[index] = (transport, packname)
2129
self.container_writer = writer
2130
self.write_index = index
2133
class _StreamAccess(object):
2134
"""A Knit Access object that provides data from a datastream.
2136
It also provides a fallback to present as unannotated data, annotated data
2137
from a *backing* access object.
2139
This is triggered by a index_memo which is pointing to a different index
2140
than this was constructed with, and is used to allow extracting full
2141
unannotated texts for insertion into annotated knits.
2144
def __init__(self, reader_callable, stream_index, backing_knit,
2146
"""Create a _StreamAccess object.
2148
:param reader_callable: The reader_callable from the datastream.
2149
This is called to buffer all the data immediately, for
2151
:param stream_index: The index the data stream this provides access to
2152
which will be present in native index_memo's.
2153
:param backing_knit: The knit object that will provide access to
2154
annotated texts which are not available in the stream, so as to
2155
create unannotated texts.
2156
:param orig_factory: The original content factory used to generate the
2157
stream. This is used for checking whether the thunk code for
2158
supporting _copy_texts will generate the correct form of data.
2160
self.data = reader_callable(None)
2161
self.stream_index = stream_index
2162
self.backing_knit = backing_knit
2163
self.orig_factory = orig_factory
2165
def get_raw_records(self, memos_for_retrieval):
2166
"""Get the raw bytes for a records.
2168
:param memos_for_retrieval: An iterable containing the (thunk_flag,
2169
index, start, end) memo for retrieving the bytes.
2170
:return: An iterator over the bytes of the records.
2172
# use a generator for memory friendliness
2173
for thunk_flag, version_id, start, end in memos_for_retrieval:
2174
if version_id is self.stream_index:
2175
yield self.data[start:end]
2177
# we have been asked to thunk. This thunking only occurs when
2178
# we are obtaining plain texts from an annotated backing knit
2179
# so that _copy_texts will work.
2180
# We could improve performance here by scanning for where we need
2181
# to do this and using get_line_list, then interleaving the output
2182
# as desired. However, for now, this is sufficient.
2183
if self.orig_factory.__class__ != KnitPlainFactory:
2184
raise errors.KnitCorrupt(
2185
self, 'Bad thunk request %r' % version_id)
2186
lines = self.backing_knit.get_lines(version_id)
2187
line_bytes = ''.join(lines)
2188
digest = sha_string(line_bytes)
2190
if lines[-1][-1] != '\n':
2191
lines[-1] = lines[-1] + '\n'
2193
orig_options = list(self.backing_knit._index.get_options(version_id))
2194
if 'fulltext' not in orig_options:
2195
if 'line-delta' not in orig_options:
2196
raise errors.KnitCorrupt(self,
2197
'Unknown compression method %r' % orig_options)
2198
orig_options.remove('line-delta')
2199
orig_options.append('fulltext')
2200
# We want plain data, because we expect to thunk only to allow text
2202
size, bytes = self.backing_knit._data._record_to_data(version_id,
2203
digest, lines, line_bytes)
2207
class _StreamIndex(object):
2208
"""A Knit Index object that uses the data map from a datastream."""
2210
def __init__(self, data_list, backing_index):
2211
"""Create a _StreamIndex object.
2213
:param data_list: The data_list from the datastream.
2214
:param backing_index: The index which will supply values for nodes
2215
referenced outside of this stream.
2217
self.data_list = data_list
2218
self.backing_index = backing_index
2219
self._by_version = {}
2221
for key, options, length, parents in data_list:
2222
self._by_version[key] = options, (pos, pos + length), parents
2225
def get_ancestry(self, versions, topo_sorted):
2226
"""Get an ancestry list for versions."""
2228
# Not needed for basic joins
2229
raise NotImplementedError(self.get_ancestry)
2230
# get a graph of all the mentioned versions:
2231
# Little ugly - basically copied from KnitIndex, but don't want to
2232
# accidentally incorporate too much of that index's code.
2234
pending = set(versions)
2235
cache = self._by_version
2237
version = pending.pop()
2240
parents = [p for p in cache[version][2] if p in cache]
2242
raise RevisionNotPresent(version, self)
2243
# if not completed and not a ghost
2244
pending.update([p for p in parents if p not in ancestry])
2245
ancestry.add(version)
2246
return list(ancestry)
2248
def get_build_details(self, version_ids):
2249
"""Get the method, index_memo and compression parent for version_ids.
2251
Ghosts are omitted from the result.
2253
:param version_ids: An iterable of version_ids.
2254
:return: A dict of version_id:(index_memo, compression_parent,
2255
parents, record_details).
2257
opaque structure to pass to read_records to extract the raw
2260
Content that this record is built upon, may be None
2262
Logical parents of this node
2264
extra information about the content which needs to be passed to
2265
Factory.parse_record
2268
for version_id in version_ids:
2270
method = self.get_method(version_id)
2271
except errors.RevisionNotPresent:
2272
# ghosts are omitted
2274
parent_ids = self.get_parents_with_ghosts(version_id)
2275
noeol = ('no-eol' in self.get_options(version_id))
2276
if method == 'fulltext':
2277
compression_parent = None
2279
compression_parent = parent_ids[0]
2280
index_memo = self.get_position(version_id)
2281
result[version_id] = (index_memo, compression_parent,
2282
parent_ids, (method, noeol))
2285
def get_method(self, version_id):
2286
"""Return compression method of specified version."""
2288
options = self._by_version[version_id][0]
2290
# Strictly speaking this should check in the backing knit, but
2291
# until we have a test to discriminate, this will do.
2292
return self.backing_index.get_method(version_id)
2293
if 'fulltext' in options:
2295
elif 'line-delta' in options:
2298
raise errors.KnitIndexUnknownMethod(self, options)
2300
def get_options(self, version_id):
2301
"""Return a list representing options.
2306
return self._by_version[version_id][0]
2308
return self.backing_index.get_options(version_id)
2310
def get_parent_map(self, version_ids):
2311
"""Passed through to by KnitVersionedFile.get_parent_map."""
2314
for version_id in version_ids:
2316
result[version_id] = self._by_version[version_id][2]
2318
pending_ids.add(version_id)
2319
result.update(self.backing_index.get_parent_map(pending_ids))
2322
def get_parents_with_ghosts(self, version_id):
2323
"""Return parents of specified version with ghosts."""
2325
return self.get_parent_map([version_id])[version_id]
2327
raise RevisionNotPresent(version_id, self)
2329
def get_position(self, version_id):
2330
"""Return details needed to access the version.
2332
_StreamAccess has the data as a big array, so we return slice
2333
coordinates into that (as index_memo's are opaque outside the
2334
index and matching access class).
2336
:return: a tuple (thunk_flag, index, start, end). If thunk_flag is
2337
False, index will be self, otherwise it will be a version id.
2340
start, end = self._by_version[version_id][1]
2341
return False, self, start, end
2343
# Signal to the access object to handle this from the backing knit.
2344
return (True, version_id, None, None)
2346
def get_versions(self):
2347
"""Get all the versions in the stream."""
2348
return self._by_version.keys()
2351
class _KnitData(object):
2352
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2354
The KnitData class provides the logic for parsing and using knit records,
2355
making use of an access method for the low level read and write operations.
2358
def __init__(self, access):
2359
"""Create a KnitData object.
2361
:param access: The access method to use. Access methods such as
2362
_KnitAccess manage the insertion of raw records and the subsequent
2363
retrieval of the same.
2365
self._access = access
2366
self._checked = False
2368
def _open_file(self):
2369
return self._access.open_file()
2371
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2372
"""Convert version_id, digest, lines into a raw data block.
2374
:param dense_lines: The bytes of lines but in a denser form. For
2375
instance, if lines is a list of 1000 bytestrings each ending in \n,
2376
dense_lines may be a list with one line in it, containing all the
2377
1000's lines and their \n's. Using dense_lines if it is already
2378
known is a win because the string join to create bytes in this
2379
function spends less time resizing the final string.
2380
:return: (len, a StringIO instance with the raw data ready to read.)
2382
# Note: using a string copy here increases memory pressure with e.g.
2383
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2384
# when doing the initial commit of a mozilla tree. RBC 20070921
2385
bytes = ''.join(chain(
2386
["version %s %d %s\n" % (version_id,
2389
dense_lines or lines,
2390
["end %s\n" % version_id]))
2391
assert bytes.__class__ == str
2392
compressed_bytes = bytes_to_gzip(bytes)
2393
return len(compressed_bytes), compressed_bytes
2395
def add_raw_records(self, sizes, raw_data):
2396
"""Append a prepared record to the data file.
2398
:param sizes: An iterable containing the size of each raw data segment.
2399
:param raw_data: A bytestring containing the data.
2400
:return: a list of index data for the way the data was stored.
2401
See the access method add_raw_records documentation for more
2404
return self._access.add_raw_records(sizes, raw_data)
2406
def _parse_record_header(self, version_id, raw_data):
2407
"""Parse a record header for consistency.
2409
:return: the header and the decompressor stream.
2410
as (stream, header_record)
2412
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2414
rec = self._check_header(version_id, df.readline())
2415
except Exception, e:
2416
raise KnitCorrupt(self._access,
2417
"While reading {%s} got %s(%s)"
2418
% (version_id, e.__class__.__name__, str(e)))
2421
def _check_header(self, version_id, line):
2424
raise KnitCorrupt(self._access,
2425
'unexpected number of elements in record header')
2426
if rec[1] != version_id:
2427
raise KnitCorrupt(self._access,
2428
'unexpected version, wanted %r, got %r'
2429
% (version_id, rec[1]))
2432
def _parse_record(self, version_id, data):
2434
# 4168 calls in 2880 217 internal
2435
# 4168 calls to _parse_record_header in 2121
2436
# 4168 calls to readlines in 330
2437
df = GzipFile(mode='rb', fileobj=StringIO(data))
2440
record_contents = df.readlines()
2441
except Exception, e:
2442
raise KnitCorrupt(self._access,
2443
"While reading {%s} got %s(%s)"
2444
% (version_id, e.__class__.__name__, str(e)))
2445
header = record_contents.pop(0)
2446
rec = self._check_header(version_id, header)
2448
last_line = record_contents.pop()
2449
if len(record_contents) != int(rec[2]):
2450
raise KnitCorrupt(self._access,
2451
'incorrect number of lines %s != %s'
2453
% (len(record_contents), int(rec[2]),
2455
if last_line != 'end %s\n' % rec[1]:
2456
raise KnitCorrupt(self._access,
2457
'unexpected version end line %r, wanted %r'
2458
% (last_line, version_id))
2460
return record_contents, rec[3]
2462
def read_records_iter_raw(self, records):
2463
"""Read text records from data file and yield raw data.
2465
This unpacks enough of the text record to validate the id is
2466
as expected but thats all.
2468
# setup an iterator of the external records:
2469
# uses readv so nice and fast we hope.
2471
# grab the disk data needed.
2472
needed_offsets = [index_memo for version_id, index_memo
2474
raw_records = self._access.get_raw_records(needed_offsets)
2476
for version_id, index_memo in records:
2477
data = raw_records.next()
2478
# validate the header
2479
df, rec = self._parse_record_header(version_id, data)
2481
yield version_id, data
2483
def read_records_iter(self, records):
2484
"""Read text records from data file and yield result.
2486
The result will be returned in whatever is the fastest to read.
2487
Not by the order requested. Also, multiple requests for the same
2488
record will only yield 1 response.
2489
:param records: A list of (version_id, pos, len) entries
2490
:return: Yields (version_id, contents, digest) in the order
2491
read, not the order requested
2496
needed_records = sorted(set(records), key=operator.itemgetter(1))
2497
if not needed_records:
2500
# The transport optimizes the fetching as well
2501
# (ie, reads continuous ranges.)
2502
raw_data = self._access.get_raw_records(
2503
[index_memo for version_id, index_memo in needed_records])
2505
for (version_id, index_memo), data in \
2506
izip(iter(needed_records), raw_data):
2507
content, digest = self._parse_record(version_id, data)
2508
yield version_id, content, digest
2510
def read_records(self, records):
2511
"""Read records into a dictionary."""
2513
for record_id, content, digest in \
2514
self.read_records_iter(records):
2515
components[record_id] = (content, digest)
2519
class InterKnit(InterVersionedFile):
2520
"""Optimised code paths for knit to knit operations."""
2522
_matching_file_from_factory = staticmethod(make_file_knit)
2523
_matching_file_to_factory = staticmethod(make_file_knit)
2526
def is_compatible(source, target):
2527
"""Be compatible with knits. """
2529
return (isinstance(source, KnitVersionedFile) and
2530
isinstance(target, KnitVersionedFile))
2531
except AttributeError:
2534
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2535
"""Copy texts to the target by extracting and adding them one by one.
2537
see join() for the parameter definitions.
2539
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2540
# --- the below is factorable out with VersionedFile.join, but wait for
2541
# VersionedFiles, it may all be simpler then.
2542
graph = Graph(self.source)
2543
search = graph._make_breadth_first_searcher(version_ids)
2544
transitive_ids = set()
2545
map(transitive_ids.update, list(search))
2546
parent_map = self.source.get_parent_map(transitive_ids)
2547
order = topo_sort(parent_map.items())
2549
def size_of_content(content):
2550
return sum(len(line) for line in content.text())
2551
# Cache at most 10MB of parent texts
2552
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2553
compute_size=size_of_content)
2554
# TODO: jam 20071116 It would be nice to have a streaming interface to
2555
# get multiple texts from a source. The source could be smarter
2556
# about how it handled intermediate stages.
2557
# get_line_list() or make_mpdiffs() seem like a possibility, but
2558
# at the moment they extract all full texts into memory, which
2559
# causes us to store more than our 3x fulltext goal.
2560
# Repository.iter_files_bytes() may be another possibility
2561
to_process = [version for version in order
2562
if version not in self.target]
2563
total = len(to_process)
2564
pb = ui.ui_factory.nested_progress_bar()
2566
for index, version in enumerate(to_process):
2567
pb.update('Converting versioned data', index, total)
2568
sha1, num_bytes, parent_text = self.target.add_lines(version,
2569
self.source.get_parents_with_ghosts(version),
2570
self.source.get_lines(version),
2571
parent_texts=parent_cache)
2572
parent_cache[version] = parent_text
2577
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2578
"""See InterVersionedFile.join."""
2579
assert isinstance(self.source, KnitVersionedFile)
2580
assert isinstance(self.target, KnitVersionedFile)
2582
# If the source and target are mismatched w.r.t. annotations vs
2583
# plain, the data needs to be converted accordingly
2584
if self.source.factory.annotated == self.target.factory.annotated:
2586
elif self.source.factory.annotated:
2587
converter = self._anno_to_plain_converter
2589
# We're converting from a plain to an annotated knit. Copy them
2590
# across by full texts.
2591
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2593
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2597
pb = ui.ui_factory.nested_progress_bar()
2599
version_ids = list(version_ids)
2600
if None in version_ids:
2601
version_ids.remove(None)
2603
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2605
this_versions = set(self.target._index.get_versions())
2606
# XXX: For efficiency we should not look at the whole index,
2607
# we only need to consider the referenced revisions - they
2608
# must all be present, or the method must be full-text.
2609
# TODO, RBC 20070919
2610
needed_versions = self.source_ancestry - this_versions
2612
if not needed_versions:
2614
full_list = topo_sort(
2615
self.source.get_parent_map(self.source.versions()))
2617
version_list = [i for i in full_list if (not self.target.has_version(i)
2618
and i in needed_versions)]
2622
copy_queue_records = []
2624
for version_id in version_list:
2625
options = self.source._index.get_options(version_id)
2626
parents = self.source._index.get_parents_with_ghosts(version_id)
2627
# check that its will be a consistent copy:
2628
for parent in parents:
2629
# if source has the parent, we must :
2630
# * already have it or
2631
# * have it scheduled already
2632
# otherwise we don't care
2633
assert (self.target.has_version(parent) or
2634
parent in copy_set or
2635
not self.source.has_version(parent))
2636
index_memo = self.source._index.get_position(version_id)
2637
copy_queue_records.append((version_id, index_memo))
2638
copy_queue.append((version_id, options, parents))
2639
copy_set.add(version_id)
2641
# data suck the join:
2643
total = len(version_list)
2646
for (version_id, raw_data), \
2647
(version_id2, options, parents) in \
2648
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2650
assert version_id == version_id2, 'logic error, inconsistent results'
2652
pb.update("Joining knit", count, total)
2654
size, raw_data = converter(raw_data, version_id, options,
2657
size = len(raw_data)
2658
raw_records.append((version_id, options, parents, size))
2659
raw_datum.append(raw_data)
2660
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2665
def _anno_to_plain_converter(self, raw_data, version_id, options,
2667
"""Convert annotated content to plain content."""
2668
data, digest = self.source._data._parse_record(version_id, raw_data)
2669
if 'fulltext' in options:
2670
content = self.source.factory.parse_fulltext(data, version_id)
2671
lines = self.target.factory.lower_fulltext(content)
2673
delta = self.source.factory.parse_line_delta(data, version_id,
2675
lines = self.target.factory.lower_line_delta(delta)
2676
return self.target._data._record_to_data(version_id, digest, lines)
2679
InterVersionedFile.register_optimiser(InterKnit)
2682
class WeaveToKnit(InterVersionedFile):
2683
"""Optimised code paths for weave to knit operations."""
2685
_matching_file_from_factory = bzrlib.weave.WeaveFile
2686
_matching_file_to_factory = staticmethod(make_file_knit)
2689
def is_compatible(source, target):
2690
"""Be compatible with weaves to knits."""
2692
return (isinstance(source, bzrlib.weave.Weave) and
2693
isinstance(target, KnitVersionedFile))
2694
except AttributeError:
2697
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2698
"""See InterVersionedFile.join."""
2699
assert isinstance(self.source, bzrlib.weave.Weave)
2700
assert isinstance(self.target, KnitVersionedFile)
2702
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2707
pb = ui.ui_factory.nested_progress_bar()
2709
version_ids = list(version_ids)
2711
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2712
this_versions = set(self.target._index.get_versions())
2713
needed_versions = self.source_ancestry - this_versions
2715
if not needed_versions:
2717
full_list = topo_sort(
2718
self.source.get_parent_map(self.source.versions()))
2720
version_list = [i for i in full_list if (not self.target.has_version(i)
2721
and i in needed_versions)]
2725
total = len(version_list)
2726
parent_map = self.source.get_parent_map(version_list)
2727
for version_id in version_list:
2728
pb.update("Converting to knit", count, total)
2729
parents = parent_map[version_id]
2730
# check that its will be a consistent copy:
2731
for parent in parents:
2732
# if source has the parent, we must already have it
2733
assert (self.target.has_version(parent))
2734
self.target.add_lines(
2735
version_id, parents, self.source.get_lines(version_id))
2742
InterVersionedFile.register_optimiser(WeaveToKnit)
2745
# Deprecated, use PatienceSequenceMatcher instead
2746
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2749
def annotate_knit(knit, revision_id):
2750
"""Annotate a knit with no cached annotations.
2752
This implementation is for knits with no cached annotations.
2753
It will work for knits with cached annotations, but this is not
2756
annotator = _KnitAnnotator(knit)
2757
return iter(annotator.annotate(revision_id))
2760
class _KnitAnnotator(object):
2761
"""Build up the annotations for a text."""
2763
def __init__(self, knit):
2766
# Content objects, differs from fulltexts because of how final newlines
2767
# are treated by knits. the content objects here will always have a
2769
self._fulltext_contents = {}
2771
# Annotated lines of specific revisions
2772
self._annotated_lines = {}
2774
# Track the raw data for nodes that we could not process yet.
2775
# This maps the revision_id of the base to a list of children that will
2776
# annotated from it.
2777
self._pending_children = {}
2779
# Nodes which cannot be extracted
2780
self._ghosts = set()
2782
# Track how many children this node has, so we know if we need to keep
2784
self._annotate_children = {}
2785
self._compression_children = {}
2787
self._all_build_details = {}
2788
# The children => parent revision_id graph
2789
self._revision_id_graph = {}
2791
self._heads_provider = None
2793
self._nodes_to_keep_annotations = set()
2794
self._generations_until_keep = 100
2796
def set_generations_until_keep(self, value):
2797
"""Set the number of generations before caching a node.
2799
Setting this to -1 will cache every merge node, setting this higher
2800
will cache fewer nodes.
2802
self._generations_until_keep = value
2804
def _add_fulltext_content(self, revision_id, content_obj):
2805
self._fulltext_contents[revision_id] = content_obj
2806
# TODO: jam 20080305 It might be good to check the sha1digest here
2807
return content_obj.text()
2809
def _check_parents(self, child, nodes_to_annotate):
2810
"""Check if all parents have been processed.
2812
:param child: A tuple of (rev_id, parents, raw_content)
2813
:param nodes_to_annotate: If child is ready, add it to
2814
nodes_to_annotate, otherwise put it back in self._pending_children
2816
for parent_id in child[1]:
2817
if (parent_id not in self._annotated_lines):
2818
# This parent is present, but another parent is missing
2819
self._pending_children.setdefault(parent_id,
2823
# This one is ready to be processed
2824
nodes_to_annotate.append(child)
2826
def _add_annotation(self, revision_id, fulltext, parent_ids,
2827
left_matching_blocks=None):
2828
"""Add an annotation entry.
2830
All parents should already have been annotated.
2831
:return: A list of children that now have their parents satisfied.
2833
a = self._annotated_lines
2834
annotated_parent_lines = [a[p] for p in parent_ids]
2835
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2836
fulltext, revision_id, left_matching_blocks,
2837
heads_provider=self._get_heads_provider()))
2838
self._annotated_lines[revision_id] = annotated_lines
2839
for p in parent_ids:
2840
ann_children = self._annotate_children[p]
2841
ann_children.remove(revision_id)
2842
if (not ann_children
2843
and p not in self._nodes_to_keep_annotations):
2844
del self._annotated_lines[p]
2845
del self._all_build_details[p]
2846
if p in self._fulltext_contents:
2847
del self._fulltext_contents[p]
2848
# Now that we've added this one, see if there are any pending
2849
# deltas to be done, certainly this parent is finished
2850
nodes_to_annotate = []
2851
for child in self._pending_children.pop(revision_id, []):
2852
self._check_parents(child, nodes_to_annotate)
2853
return nodes_to_annotate
2855
def _get_build_graph(self, revision_id):
2856
"""Get the graphs for building texts and annotations.
2858
The data you need for creating a full text may be different than the
2859
data you need to annotate that text. (At a minimum, you need both
2860
parents to create an annotation, but only need 1 parent to generate the
2863
:return: A list of (revision_id, index_memo) records, suitable for
2864
passing to read_records_iter to start reading in the raw data fro/
2867
if revision_id in self._annotated_lines:
2870
pending = set([revision_id])
2875
# get all pending nodes
2877
this_iteration = pending
2878
build_details = self._knit._index.get_build_details(this_iteration)
2879
self._all_build_details.update(build_details)
2880
# new_nodes = self._knit._index._get_entries(this_iteration)
2882
for rev_id, details in build_details.iteritems():
2883
(index_memo, compression_parent, parents,
2884
record_details) = details
2885
self._revision_id_graph[rev_id] = parents
2886
records.append((rev_id, index_memo))
2887
# Do we actually need to check _annotated_lines?
2888
pending.update(p for p in parents
2889
if p not in self._all_build_details)
2890
if compression_parent:
2891
self._compression_children.setdefault(compression_parent,
2894
for parent in parents:
2895
self._annotate_children.setdefault(parent,
2897
num_gens = generation - kept_generation
2898
if ((num_gens >= self._generations_until_keep)
2899
and len(parents) > 1):
2900
kept_generation = generation
2901
self._nodes_to_keep_annotations.add(rev_id)
2903
missing_versions = this_iteration.difference(build_details.keys())
2904
self._ghosts.update(missing_versions)
2905
for missing_version in missing_versions:
2906
# add a key, no parents
2907
self._revision_id_graph[missing_version] = ()
2908
pending.discard(missing_version) # don't look for it
2909
# XXX: This should probably be a real exception, as it is a data
2911
assert not self._ghosts.intersection(self._compression_children), \
2912
"We cannot have nodes which have a compression parent of a ghost."
2913
# Cleanout anything that depends on a ghost so that we don't wait for
2914
# the ghost to show up
2915
for node in self._ghosts:
2916
if node in self._annotate_children:
2917
# We won't be building this node
2918
del self._annotate_children[node]
2919
# Generally we will want to read the records in reverse order, because
2920
# we find the parent nodes after the children
2924
def _annotate_records(self, records):
2925
"""Build the annotations for the listed records."""
2926
# We iterate in the order read, rather than a strict order requested
2927
# However, process what we can, and put off to the side things that
2928
# still need parents, cleaning them up when those parents are
2930
for (rev_id, record,
2931
digest) in self._knit._data.read_records_iter(records):
2932
if rev_id in self._annotated_lines:
2934
parent_ids = self._revision_id_graph[rev_id]
2935
parent_ids = [p for p in parent_ids if p not in self._ghosts]
2936
details = self._all_build_details[rev_id]
2937
(index_memo, compression_parent, parents,
2938
record_details) = details
2939
nodes_to_annotate = []
2940
# TODO: Remove the punning between compression parents, and
2941
# parent_ids, we should be able to do this without assuming
2943
if len(parent_ids) == 0:
2944
# There are no parents for this node, so just add it
2945
# TODO: This probably needs to be decoupled
2946
assert compression_parent is None
2947
fulltext_content, delta = self._knit.factory.parse_record(
2948
rev_id, record, record_details, None)
2949
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
2950
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
2951
parent_ids, left_matching_blocks=None))
2953
child = (rev_id, parent_ids, record)
2954
# Check if all the parents are present
2955
self._check_parents(child, nodes_to_annotate)
2956
while nodes_to_annotate:
2957
# Should we use a queue here instead of a stack?
2958
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
2959
(index_memo, compression_parent, parents,
2960
record_details) = self._all_build_details[rev_id]
2961
if compression_parent is not None:
2962
comp_children = self._compression_children[compression_parent]
2963
assert rev_id in comp_children
2964
# If there is only 1 child, it is safe to reuse this
2966
reuse_content = (len(comp_children) == 1
2967
and compression_parent not in
2968
self._nodes_to_keep_annotations)
2970
# Remove it from the cache since it will be changing
2971
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
2972
# Make sure to copy the fulltext since it might be
2974
parent_fulltext = list(parent_fulltext_content.text())
2976
parent_fulltext_content = self._fulltext_contents[compression_parent]
2977
parent_fulltext = parent_fulltext_content.text()
2978
comp_children.remove(rev_id)
2979
fulltext_content, delta = self._knit.factory.parse_record(
2980
rev_id, record, record_details,
2981
parent_fulltext_content,
2982
copy_base_content=(not reuse_content))
2983
fulltext = self._add_fulltext_content(rev_id,
2985
blocks = KnitContent.get_line_delta_blocks(delta,
2986
parent_fulltext, fulltext)
2988
fulltext_content = self._knit.factory.parse_fulltext(
2990
fulltext = self._add_fulltext_content(rev_id,
2993
nodes_to_annotate.extend(
2994
self._add_annotation(rev_id, fulltext, parent_ids,
2995
left_matching_blocks=blocks))
2997
def _get_heads_provider(self):
2998
"""Create a heads provider for resolving ancestry issues."""
2999
if self._heads_provider is not None:
3000
return self._heads_provider
3001
parent_provider = _mod_graph.DictParentsProvider(
3002
self._revision_id_graph)
3003
graph_obj = _mod_graph.Graph(parent_provider)
3004
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3005
self._heads_provider = head_cache
3008
def annotate(self, revision_id):
3009
"""Return the annotated fulltext at the given revision.
3011
:param revision_id: The revision id for this file
3013
records = self._get_build_graph(revision_id)
3014
if revision_id in self._ghosts:
3015
raise errors.RevisionNotPresent(revision_id, self._knit)
3016
self._annotate_records(records)
3017
return self._annotated_lines[revision_id]
3021
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3023
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
