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 ValueError("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]))
847
# We received a line-delta record for a non-delta knit.
848
# Convert it to a fulltext.
849
gzip_bytes = reader_callable(length)
850
lines, sha1 = self._data._parse_record(
851
version_id, gzip_bytes)
852
delta = self.factory.parse_line_delta(lines,
854
content = self.factory.make(
855
self.get_lines(parents[0]), parents[0])
856
content.apply_delta(delta, version_id)
857
digest, len, content = self.add_lines(
858
version_id, parents, content.text())
860
raise errors.VersionedFileInvalidChecksum(version)
863
self._add_raw_records(
864
[(version_id, options, parents, length)],
865
reader_callable(length))
867
def _knit_from_datastream(self, (format, data_list, reader_callable)):
868
"""Create a knit object from a data stream.
870
This method exists to allow conversion of data streams that do not
871
match the signature of this knit. Generally it will be slower and use
872
more memory to use this method to insert data, but it will work.
874
:seealso: get_data_stream for details on datastreams.
875
:return: A knit versioned file which can be used to join the datastream
878
if format == "knit-plain":
879
factory = KnitPlainFactory()
880
elif format == "knit-annotated":
881
factory = KnitAnnotateFactory()
883
raise errors.KnitDataStreamUnknown(format)
884
index = _StreamIndex(data_list, self._index)
885
access = _StreamAccess(reader_callable, index, self, factory)
886
return KnitVersionedFile(self.filename, self.transport,
887
factory=factory, index=index, access_method=access)
890
"""See VersionedFile.versions."""
891
if 'evil' in debug.debug_flags:
892
trace.mutter_callsite(2, "versions scales with size of history")
893
return self._index.get_versions()
895
def has_version(self, version_id):
896
"""See VersionedFile.has_version."""
897
if 'evil' in debug.debug_flags:
898
trace.mutter_callsite(2, "has_version is a LBYL scenario")
899
return self._index.has_version(version_id)
901
__contains__ = has_version
903
def _merge_annotations(self, content, parents, parent_texts={},
904
delta=None, annotated=None,
905
left_matching_blocks=None):
906
"""Merge annotations for content. This is done by comparing
907
the annotations based on changed to the text.
909
if left_matching_blocks is not None:
910
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
914
for parent_id in parents:
915
merge_content = self._get_content(parent_id, parent_texts)
916
if (parent_id == parents[0] and delta_seq is not None):
919
seq = patiencediff.PatienceSequenceMatcher(
920
None, merge_content.text(), content.text())
921
for i, j, n in seq.get_matching_blocks():
924
# this appears to copy (origin, text) pairs across to the
925
# new content for any line that matches the last-checked
927
content._lines[j:j+n] = merge_content._lines[i:i+n]
929
if delta_seq is None:
930
reference_content = self._get_content(parents[0], parent_texts)
931
new_texts = content.text()
932
old_texts = reference_content.text()
933
delta_seq = patiencediff.PatienceSequenceMatcher(
934
None, old_texts, new_texts)
935
return self._make_line_delta(delta_seq, content)
937
def _make_line_delta(self, delta_seq, new_content):
938
"""Generate a line delta from delta_seq and new_content."""
940
for op in delta_seq.get_opcodes():
943
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
946
def _get_components_positions(self, version_ids):
947
"""Produce a map of position data for the components of versions.
949
This data is intended to be used for retrieving the knit records.
951
A dict of version_id to (record_details, index_memo, next, parents) is
953
method is the way referenced data should be applied.
954
index_memo is the handle to pass to the data access to actually get the
956
next is the build-parent of the version, or None for fulltexts.
957
parents is the version_ids of the parents of this version
960
pending_components = version_ids
961
while pending_components:
962
build_details = self._index.get_build_details(pending_components)
963
current_components = set(pending_components)
964
pending_components = set()
965
for version_id, details in build_details.iteritems():
966
(index_memo, compression_parent, parents,
967
record_details) = details
968
method = record_details[0]
969
if compression_parent is not None:
970
pending_components.add(compression_parent)
971
component_data[version_id] = (record_details, index_memo,
973
missing = current_components.difference(build_details)
975
raise errors.RevisionNotPresent(missing.pop(), self.filename)
976
return component_data
978
def _get_content(self, version_id, parent_texts={}):
979
"""Returns a content object that makes up the specified
981
cached_version = parent_texts.get(version_id, None)
982
if cached_version is not None:
983
if not self.has_version(version_id):
984
raise RevisionNotPresent(version_id, self.filename)
985
return cached_version
987
text_map, contents_map = self._get_content_maps([version_id])
988
return contents_map[version_id]
990
def _check_versions_present(self, version_ids):
991
"""Check that all specified versions are present."""
992
self._index.check_versions_present(version_ids)
994
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
995
nostore_sha, random_id, check_content, left_matching_blocks):
996
"""See VersionedFile.add_lines_with_ghosts()."""
997
self._check_add(version_id, lines, random_id, check_content)
998
return self._add(version_id, lines, parents, self.delta,
999
parent_texts, left_matching_blocks, nostore_sha, random_id)
1001
def _add_lines(self, version_id, parents, lines, parent_texts,
1002
left_matching_blocks, nostore_sha, random_id, check_content):
1003
"""See VersionedFile.add_lines."""
1004
self._check_add(version_id, lines, random_id, check_content)
1005
self._check_versions_present(parents)
1006
return self._add(version_id, lines[:], parents, self.delta,
1007
parent_texts, left_matching_blocks, nostore_sha, random_id)
1009
def _check_add(self, version_id, lines, random_id, check_content):
1010
"""check that version_id and lines are safe to add."""
1011
if contains_whitespace(version_id):
1012
raise InvalidRevisionId(version_id, self.filename)
1013
self.check_not_reserved_id(version_id)
1014
# Technically this could be avoided if we are happy to allow duplicate
1015
# id insertion when other things than bzr core insert texts, but it
1016
# seems useful for folk using the knit api directly to have some safety
1017
# blanket that we can disable.
1018
if not random_id and self.has_version(version_id):
1019
raise RevisionAlreadyPresent(version_id, self.filename)
1021
self._check_lines_not_unicode(lines)
1022
self._check_lines_are_lines(lines)
1024
def _add(self, version_id, lines, parents, delta, parent_texts,
1025
left_matching_blocks, nostore_sha, random_id):
1026
"""Add a set of lines on top of version specified by parents.
1028
If delta is true, compress the text as a line-delta against
1031
Any versions not present will be converted into ghosts.
1033
# first thing, if the content is something we don't need to store, find
1035
line_bytes = ''.join(lines)
1036
digest = sha_string(line_bytes)
1037
if nostore_sha == digest:
1038
raise errors.ExistingContent
1040
present_parents = []
1041
if parent_texts is None:
1043
for parent in parents:
1044
if self.has_version(parent):
1045
present_parents.append(parent)
1047
# can only compress against the left most present parent.
1049
(len(present_parents) == 0 or
1050
present_parents[0] != parents[0])):
1053
text_length = len(line_bytes)
1056
if lines[-1][-1] != '\n':
1057
# copy the contents of lines.
1059
options.append('no-eol')
1060
lines[-1] = lines[-1] + '\n'
1064
# To speed the extract of texts the delta chain is limited
1065
# to a fixed number of deltas. This should minimize both
1066
# I/O and the time spend applying deltas.
1067
delta = self._check_should_delta(present_parents)
1069
assert isinstance(version_id, str)
1070
content = self.factory.make(lines, version_id)
1071
if delta or (self.factory.annotated and len(present_parents) > 0):
1072
# Merge annotations from parent texts if needed.
1073
delta_hunks = self._merge_annotations(content, present_parents,
1074
parent_texts, delta, self.factory.annotated,
1075
left_matching_blocks)
1078
options.append('line-delta')
1079
store_lines = self.factory.lower_line_delta(delta_hunks)
1080
size, bytes = self._data._record_to_data(version_id, digest,
1083
options.append('fulltext')
1084
# isinstance is slower and we have no hierarchy.
1085
if self.factory.__class__ == KnitPlainFactory:
1086
# Use the already joined bytes saving iteration time in
1088
size, bytes = self._data._record_to_data(version_id, digest,
1089
lines, [line_bytes])
1091
# get mixed annotation + content and feed it into the
1093
store_lines = self.factory.lower_fulltext(content)
1094
size, bytes = self._data._record_to_data(version_id, digest,
1097
access_memo = self._data.add_raw_records([size], bytes)[0]
1098
self._index.add_versions(
1099
((version_id, options, access_memo, parents),),
1100
random_id=random_id)
1101
return digest, text_length, content
1103
def check(self, progress_bar=None):
1104
"""See VersionedFile.check()."""
1106
def get_lines(self, version_id):
1107
"""See VersionedFile.get_lines()."""
1108
return self.get_line_list([version_id])[0]
1110
def _get_record_map(self, version_ids):
1111
"""Produce a dictionary of knit records.
1113
:return: {version_id:(record, record_details, digest, next)}
1115
data returned from read_records
1117
opaque information to pass to parse_record
1119
SHA1 digest of the full text after all steps are done
1121
build-parent of the version, i.e. the leftmost ancestor.
1122
Will be None if the record is not a delta.
1124
position_map = self._get_components_positions(version_ids)
1125
# c = component_id, r = record_details, i_m = index_memo, n = next
1126
records = [(c, i_m) for c, (r, i_m, n)
1127
in position_map.iteritems()]
1129
for component_id, record, digest in \
1130
self._data.read_records_iter(records):
1131
(record_details, index_memo, next) = position_map[component_id]
1132
record_map[component_id] = record, record_details, digest, next
1136
def get_text(self, version_id):
1137
"""See VersionedFile.get_text"""
1138
return self.get_texts([version_id])[0]
1140
def get_texts(self, version_ids):
1141
return [''.join(l) for l in self.get_line_list(version_ids)]
1143
def get_line_list(self, version_ids):
1144
"""Return the texts of listed versions as a list of strings."""
1145
for version_id in version_ids:
1146
self.check_not_reserved_id(version_id)
1147
text_map, content_map = self._get_content_maps(version_ids)
1148
return [text_map[v] for v in version_ids]
1150
_get_lf_split_line_list = get_line_list
1152
def _get_content_maps(self, version_ids):
1153
"""Produce maps of text and KnitContents
1155
:return: (text_map, content_map) where text_map contains the texts for
1156
the requested versions and content_map contains the KnitContents.
1157
Both dicts take version_ids as their keys.
1159
# FUTURE: This function could be improved for the 'extract many' case
1160
# by tracking each component and only doing the copy when the number of
1161
# children than need to apply delta's to it is > 1 or it is part of the
1163
version_ids = list(version_ids)
1164
multiple_versions = len(version_ids) != 1
1165
record_map = self._get_record_map(version_ids)
1170
for version_id in version_ids:
1173
while cursor is not None:
1174
record, record_details, digest, next = record_map[cursor]
1175
components.append((cursor, record, record_details, digest))
1176
if cursor in content_map:
1181
for (component_id, record, record_details,
1182
digest) in reversed(components):
1183
if component_id in content_map:
1184
content = content_map[component_id]
1186
content, delta = self.factory.parse_record(version_id,
1187
record, record_details, content,
1188
copy_base_content=multiple_versions)
1189
if multiple_versions:
1190
content_map[component_id] = content
1192
content.cleanup_eol(copy_on_mutate=multiple_versions)
1193
final_content[version_id] = content
1195
# digest here is the digest from the last applied component.
1196
text = content.text()
1197
actual_sha = sha_strings(text)
1198
if actual_sha != digest:
1199
raise KnitCorrupt(self.filename,
1201
'\n of reconstructed text does not match'
1203
'\n for version %s' %
1204
(actual_sha, digest, version_id))
1205
text_map[version_id] = text
1206
return text_map, final_content
1208
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1210
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1211
if version_ids is None:
1212
version_ids = self.versions()
1214
pb = progress.DummyProgress()
1215
# we don't care about inclusions, the caller cares.
1216
# but we need to setup a list of records to visit.
1217
# we need version_id, position, length
1218
version_id_records = []
1219
requested_versions = set(version_ids)
1220
# filter for available versions
1221
for version_id in requested_versions:
1222
if not self.has_version(version_id):
1223
raise RevisionNotPresent(version_id, self.filename)
1224
# get a in-component-order queue:
1225
for version_id in self.versions():
1226
if version_id in requested_versions:
1227
index_memo = self._index.get_position(version_id)
1228
version_id_records.append((version_id, index_memo))
1230
total = len(version_id_records)
1231
for version_idx, (version_id, data, sha_value) in \
1232
enumerate(self._data.read_records_iter(version_id_records)):
1233
pb.update('Walking content.', version_idx, total)
1234
method = self._index.get_method(version_id)
1236
assert method in ('fulltext', 'line-delta')
1237
if method == 'fulltext':
1238
line_iterator = self.factory.get_fulltext_content(data)
1240
line_iterator = self.factory.get_linedelta_content(data)
1241
# XXX: It might be more efficient to yield (version_id,
1242
# line_iterator) in the future. However for now, this is a simpler
1243
# change to integrate into the rest of the codebase. RBC 20071110
1244
for line in line_iterator:
1245
yield line, version_id
1247
pb.update('Walking content.', total, total)
1249
def num_versions(self):
1250
"""See VersionedFile.num_versions()."""
1251
return self._index.num_versions()
1253
__len__ = num_versions
1255
def annotate(self, version_id):
1256
"""See VersionedFile.annotate."""
1257
return self.factory.annotate(self, version_id)
1259
def get_parent_map(self, version_ids):
1260
"""See VersionedFile.get_parent_map."""
1261
return self._index.get_parent_map(version_ids)
1263
def get_ancestry(self, versions, topo_sorted=True):
1264
"""See VersionedFile.get_ancestry."""
1265
if isinstance(versions, basestring):
1266
versions = [versions]
1269
return self._index.get_ancestry(versions, topo_sorted)
1271
def get_ancestry_with_ghosts(self, versions):
1272
"""See VersionedFile.get_ancestry_with_ghosts."""
1273
if isinstance(versions, basestring):
1274
versions = [versions]
1277
return self._index.get_ancestry_with_ghosts(versions)
1279
def plan_merge(self, ver_a, ver_b):
1280
"""See VersionedFile.plan_merge."""
1281
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1282
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1283
annotated_a = self.annotate(ver_a)
1284
annotated_b = self.annotate(ver_b)
1285
return merge._plan_annotate_merge(annotated_a, annotated_b,
1286
ancestors_a, ancestors_b)
1289
class _KnitComponentFile(object):
1290
"""One of the files used to implement a knit database"""
1292
def __init__(self, transport, filename, mode, file_mode=None,
1293
create_parent_dir=False, dir_mode=None):
1294
self._transport = transport
1295
self._filename = filename
1297
self._file_mode = file_mode
1298
self._dir_mode = dir_mode
1299
self._create_parent_dir = create_parent_dir
1300
self._need_to_create = False
1302
def _full_path(self):
1303
"""Return the full path to this file."""
1304
return self._transport.base + self._filename
1306
def check_header(self, fp):
1307
line = fp.readline()
1309
# An empty file can actually be treated as though the file doesn't
1311
raise errors.NoSuchFile(self._full_path())
1312
if line != self.HEADER:
1313
raise KnitHeaderError(badline=line,
1314
filename=self._transport.abspath(self._filename))
1317
return '%s(%s)' % (self.__class__.__name__, self._filename)
1320
class _KnitIndex(_KnitComponentFile):
1321
"""Manages knit index file.
1323
The index is already kept in memory and read on startup, to enable
1324
fast lookups of revision information. The cursor of the index
1325
file is always pointing to the end, making it easy to append
1328
_cache is a cache for fast mapping from version id to a Index
1331
_history is a cache for fast mapping from indexes to version ids.
1333
The index data format is dictionary compressed when it comes to
1334
parent references; a index entry may only have parents that with a
1335
lover index number. As a result, the index is topological sorted.
1337
Duplicate entries may be written to the index for a single version id
1338
if this is done then the latter one completely replaces the former:
1339
this allows updates to correct version and parent information.
1340
Note that the two entries may share the delta, and that successive
1341
annotations and references MUST point to the first entry.
1343
The index file on disc contains a header, followed by one line per knit
1344
record. The same revision can be present in an index file more than once.
1345
The first occurrence gets assigned a sequence number starting from 0.
1347
The format of a single line is
1348
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1349
REVISION_ID is a utf8-encoded revision id
1350
FLAGS is a comma separated list of flags about the record. Values include
1351
no-eol, line-delta, fulltext.
1352
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1353
that the the compressed data starts at.
1354
LENGTH is the ascii representation of the length of the data file.
1355
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1357
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1358
revision id already in the knit that is a parent of REVISION_ID.
1359
The ' :' marker is the end of record marker.
1362
when a write is interrupted to the index file, it will result in a line
1363
that does not end in ' :'. If the ' :' is not present at the end of a line,
1364
or at the end of the file, then the record that is missing it will be
1365
ignored by the parser.
1367
When writing new records to the index file, the data is preceded by '\n'
1368
to ensure that records always start on new lines even if the last write was
1369
interrupted. As a result its normal for the last line in the index to be
1370
missing a trailing newline. One can be added with no harmful effects.
1373
HEADER = "# bzr knit index 8\n"
1375
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1376
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1378
def _cache_version(self, version_id, options, pos, size, parents):
1379
"""Cache a version record in the history array and index cache.
1381
This is inlined into _load_data for performance. KEEP IN SYNC.
1382
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1385
# only want the _history index to reference the 1st index entry
1387
if version_id not in self._cache:
1388
index = len(self._history)
1389
self._history.append(version_id)
1391
index = self._cache[version_id][5]
1392
self._cache[version_id] = (version_id,
1399
def _check_write_ok(self):
1400
if self._get_scope() != self._scope:
1401
raise errors.OutSideTransaction()
1402
if self._mode != 'w':
1403
raise errors.ReadOnlyObjectDirtiedError(self)
1405
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1406
create_parent_dir=False, delay_create=False, dir_mode=None,
1408
_KnitComponentFile.__init__(self, transport, filename, mode,
1409
file_mode=file_mode,
1410
create_parent_dir=create_parent_dir,
1413
# position in _history is the 'official' index for a revision
1414
# but the values may have come from a newer entry.
1415
# so - wc -l of a knit index is != the number of unique names
1419
fp = self._transport.get(self._filename)
1421
# _load_data may raise NoSuchFile if the target knit is
1423
_load_data(self, fp)
1427
if mode != 'w' or not create:
1430
self._need_to_create = True
1432
self._transport.put_bytes_non_atomic(
1433
self._filename, self.HEADER, mode=self._file_mode)
1434
self._scope = get_scope()
1435
self._get_scope = get_scope
1437
def get_ancestry(self, versions, topo_sorted=True):
1438
"""See VersionedFile.get_ancestry."""
1439
# get a graph of all the mentioned versions:
1441
pending = set(versions)
1444
version = pending.pop()
1447
parents = [p for p in cache[version][4] if p in cache]
1449
raise RevisionNotPresent(version, self._filename)
1450
# if not completed and not a ghost
1451
pending.update([p for p in parents if p not in graph])
1452
graph[version] = parents
1455
return topo_sort(graph.items())
1457
def get_ancestry_with_ghosts(self, versions):
1458
"""See VersionedFile.get_ancestry_with_ghosts."""
1459
# get a graph of all the mentioned versions:
1460
self.check_versions_present(versions)
1463
pending = set(versions)
1465
version = pending.pop()
1467
parents = cache[version][4]
1473
pending.update([p for p in parents if p not in graph])
1474
graph[version] = parents
1475
return topo_sort(graph.items())
1477
def get_build_details(self, version_ids):
1478
"""Get the method, index_memo and compression parent for version_ids.
1480
Ghosts are omitted from the result.
1482
:param version_ids: An iterable of version_ids.
1483
:return: A dict of version_id:(index_memo, compression_parent,
1484
parents, record_details).
1486
opaque structure to pass to read_records to extract the raw
1489
Content that this record is built upon, may be None
1491
Logical parents of this node
1493
extra information about the content which needs to be passed to
1494
Factory.parse_record
1497
for version_id in version_ids:
1498
if version_id not in self._cache:
1499
# ghosts are omitted
1501
method = self.get_method(version_id)
1502
parents = self.get_parents_with_ghosts(version_id)
1503
if method == 'fulltext':
1504
compression_parent = None
1506
compression_parent = parents[0]
1507
noeol = 'no-eol' in self.get_options(version_id)
1508
index_memo = self.get_position(version_id)
1509
result[version_id] = (index_memo, compression_parent,
1510
parents, (method, noeol))
1513
def num_versions(self):
1514
return len(self._history)
1516
__len__ = num_versions
1518
def get_versions(self):
1519
"""Get all the versions in the file. not topologically sorted."""
1520
return self._history
1522
def _version_list_to_index(self, versions):
1525
for version in versions:
1526
if version in cache:
1527
# -- inlined lookup() --
1528
result_list.append(str(cache[version][5]))
1529
# -- end lookup () --
1531
result_list.append('.' + version)
1532
return ' '.join(result_list)
1534
def add_version(self, version_id, options, index_memo, parents):
1535
"""Add a version record to the index."""
1536
self.add_versions(((version_id, options, index_memo, parents),))
1538
def add_versions(self, versions, random_id=False):
1539
"""Add multiple versions to the index.
1541
:param versions: a list of tuples:
1542
(version_id, options, pos, size, parents).
1543
:param random_id: If True the ids being added were randomly generated
1544
and no check for existence will be performed.
1547
orig_history = self._history[:]
1548
orig_cache = self._cache.copy()
1551
for version_id, options, (index, pos, size), parents in versions:
1552
line = "\n%s %s %s %s %s :" % (version_id,
1556
self._version_list_to_index(parents))
1557
assert isinstance(line, str), \
1558
'content must be utf-8 encoded: %r' % (line,)
1560
self._cache_version(version_id, options, pos, size, tuple(parents))
1561
if not self._need_to_create:
1562
self._transport.append_bytes(self._filename, ''.join(lines))
1565
sio.write(self.HEADER)
1566
sio.writelines(lines)
1568
self._transport.put_file_non_atomic(self._filename, sio,
1569
create_parent_dir=self._create_parent_dir,
1570
mode=self._file_mode,
1571
dir_mode=self._dir_mode)
1572
self._need_to_create = False
1574
# If any problems happen, restore the original values and re-raise
1575
self._history = orig_history
1576
self._cache = orig_cache
1579
def has_version(self, version_id):
1580
"""True if the version is in the index."""
1581
return version_id in self._cache
1583
def get_position(self, version_id):
1584
"""Return details needed to access the version.
1586
.kndx indices do not support split-out data, so return None for the
1589
:return: a tuple (None, data position, size) to hand to the access
1590
logic to get the record.
1592
entry = self._cache[version_id]
1593
return None, entry[2], entry[3]
1595
def get_method(self, version_id):
1596
"""Return compression method of specified version."""
1598
options = self._cache[version_id][1]
1600
raise RevisionNotPresent(version_id, self._filename)
1601
if 'fulltext' in options:
1604
if 'line-delta' not in options:
1605
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1608
def get_options(self, version_id):
1609
"""Return a list representing options.
1613
return self._cache[version_id][1]
1615
def get_parent_map(self, version_ids):
1616
"""Passed through to by KnitVersionedFile.get_parent_map."""
1618
for version_id in version_ids:
1620
result[version_id] = tuple(self._cache[version_id][4])
1625
def get_parents_with_ghosts(self, version_id):
1626
"""Return parents of specified version with ghosts."""
1628
return self.get_parent_map([version_id])[version_id]
1630
raise RevisionNotPresent(version_id, self)
1632
def check_versions_present(self, version_ids):
1633
"""Check that all specified versions are present."""
1635
for version_id in version_ids:
1636
if version_id not in cache:
1637
raise RevisionNotPresent(version_id, self._filename)
1640
class KnitGraphIndex(object):
1641
"""A knit index that builds on GraphIndex."""
1643
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1644
"""Construct a KnitGraphIndex on a graph_index.
1646
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1647
:param deltas: Allow delta-compressed records.
1648
:param add_callback: If not None, allow additions to the index and call
1649
this callback with a list of added GraphIndex nodes:
1650
[(node, value, node_refs), ...]
1651
:param parents: If True, record knits parents, if not do not record
1654
self._graph_index = graph_index
1655
self._deltas = deltas
1656
self._add_callback = add_callback
1657
self._parents = parents
1658
if deltas and not parents:
1659
raise KnitCorrupt(self, "Cannot do delta compression without "
1662
def _check_write_ok(self):
1665
def _get_entries(self, keys, check_present=False):
1666
"""Get the entries for keys.
1668
:param keys: An iterable of index keys, - 1-tuples.
1673
for node in self._graph_index.iter_entries(keys):
1675
found_keys.add(node[1])
1677
# adapt parentless index to the rest of the code.
1678
for node in self._graph_index.iter_entries(keys):
1679
yield node[0], node[1], node[2], ()
1680
found_keys.add(node[1])
1682
missing_keys = keys.difference(found_keys)
1684
raise RevisionNotPresent(missing_keys.pop(), self)
1686
def _present_keys(self, version_ids):
1688
node[1] for node in self._get_entries(version_ids)])
1690
def _parentless_ancestry(self, versions):
1691
"""Honour the get_ancestry API for parentless knit indices."""
1692
wanted_keys = self._version_ids_to_keys(versions)
1693
present_keys = self._present_keys(wanted_keys)
1694
missing = set(wanted_keys).difference(present_keys)
1696
raise RevisionNotPresent(missing.pop(), self)
1697
return list(self._keys_to_version_ids(present_keys))
1699
def get_ancestry(self, versions, topo_sorted=True):
1700
"""See VersionedFile.get_ancestry."""
1701
if not self._parents:
1702
return self._parentless_ancestry(versions)
1703
# XXX: This will do len(history) index calls - perhaps
1704
# it should be altered to be a index core feature?
1705
# get a graph of all the mentioned versions:
1708
versions = self._version_ids_to_keys(versions)
1709
pending = set(versions)
1711
# get all pending nodes
1712
this_iteration = pending
1713
new_nodes = self._get_entries(this_iteration)
1716
for (index, key, value, node_refs) in new_nodes:
1717
# dont ask for ghosties - otherwise
1718
# we we can end up looping with pending
1719
# being entirely ghosted.
1720
graph[key] = [parent for parent in node_refs[0]
1721
if parent not in ghosts]
1723
for parent in graph[key]:
1724
# dont examine known nodes again
1729
ghosts.update(this_iteration.difference(found))
1730
if versions.difference(graph):
1731
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1733
result_keys = topo_sort(graph.items())
1735
result_keys = graph.iterkeys()
1736
return [key[0] for key in result_keys]
1738
def get_ancestry_with_ghosts(self, versions):
1739
"""See VersionedFile.get_ancestry."""
1740
if not self._parents:
1741
return self._parentless_ancestry(versions)
1742
# XXX: This will do len(history) index calls - perhaps
1743
# it should be altered to be a index core feature?
1744
# get a graph of all the mentioned versions:
1746
versions = self._version_ids_to_keys(versions)
1747
pending = set(versions)
1749
# get all pending nodes
1750
this_iteration = pending
1751
new_nodes = self._get_entries(this_iteration)
1753
for (index, key, value, node_refs) in new_nodes:
1754
graph[key] = node_refs[0]
1756
for parent in graph[key]:
1757
# dont examine known nodes again
1761
missing_versions = this_iteration.difference(graph)
1762
missing_needed = versions.intersection(missing_versions)
1764
raise RevisionNotPresent(missing_needed.pop(), self)
1765
for missing_version in missing_versions:
1766
# add a key, no parents
1767
graph[missing_version] = []
1768
pending.discard(missing_version) # don't look for it
1769
result_keys = topo_sort(graph.items())
1770
return [key[0] for key in result_keys]
1772
def get_build_details(self, version_ids):
1773
"""Get the method, index_memo and compression parent for version_ids.
1775
Ghosts are omitted from the result.
1777
:param version_ids: An iterable of version_ids.
1778
:return: A dict of version_id:(index_memo, compression_parent,
1779
parents, record_details).
1781
opaque structure to pass to read_records to extract the raw
1784
Content that this record is built upon, may be None
1786
Logical parents of this node
1788
extra information about the content which needs to be passed to
1789
Factory.parse_record
1792
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
1793
for entry in entries:
1794
version_id = self._keys_to_version_ids((entry[1],))[0]
1795
if not self._parents:
1798
parents = self._keys_to_version_ids(entry[3][0])
1799
if not self._deltas:
1800
compression_parent = None
1802
compression_parent_key = self._compression_parent(entry)
1803
if compression_parent_key:
1804
compression_parent = self._keys_to_version_ids(
1805
(compression_parent_key,))[0]
1807
compression_parent = None
1808
noeol = (entry[2][0] == 'N')
1809
if compression_parent:
1810
method = 'line-delta'
1813
result[version_id] = (self._node_to_position(entry),
1814
compression_parent, parents,
1818
def _compression_parent(self, an_entry):
1819
# return the key that an_entry is compressed against, or None
1820
# Grab the second parent list (as deltas implies parents currently)
1821
compression_parents = an_entry[3][1]
1822
if not compression_parents:
1824
assert len(compression_parents) == 1
1825
return compression_parents[0]
1827
def _get_method(self, node):
1828
if not self._deltas:
1830
if self._compression_parent(node):
1835
def num_versions(self):
1836
return len(list(self._graph_index.iter_all_entries()))
1838
__len__ = num_versions
1840
def get_versions(self):
1841
"""Get all the versions in the file. not topologically sorted."""
1842
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1844
def has_version(self, version_id):
1845
"""True if the version is in the index."""
1846
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1848
def _keys_to_version_ids(self, keys):
1849
return tuple(key[0] for key in keys)
1851
def get_position(self, version_id):
1852
"""Return details needed to access the version.
1854
:return: a tuple (index, data position, size) to hand to the access
1855
logic to get the record.
1857
node = self._get_node(version_id)
1858
return self._node_to_position(node)
1860
def _node_to_position(self, node):
1861
"""Convert an index value to position details."""
1862
bits = node[2][1:].split(' ')
1863
return node[0], int(bits[0]), int(bits[1])
1865
def get_method(self, version_id):
1866
"""Return compression method of specified version."""
1867
return self._get_method(self._get_node(version_id))
1869
def _get_node(self, version_id):
1871
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1873
raise RevisionNotPresent(version_id, self)
1875
def get_options(self, version_id):
1876
"""Return a list representing options.
1880
node = self._get_node(version_id)
1881
options = [self._get_method(node)]
1882
if node[2][0] == 'N':
1883
options.append('no-eol')
1886
def get_parent_map(self, version_ids):
1887
"""Passed through to by KnitVersionedFile.get_parent_map."""
1888
nodes = self._get_entries(self._version_ids_to_keys(version_ids))
1892
result[node[1][0]] = self._keys_to_version_ids(node[3][0])
1895
result[node[1][0]] = ()
1898
def get_parents_with_ghosts(self, version_id):
1899
"""Return parents of specified version with ghosts."""
1901
return self.get_parent_map([version_id])[version_id]
1903
raise RevisionNotPresent(version_id, self)
1905
def check_versions_present(self, version_ids):
1906
"""Check that all specified versions are present."""
1907
keys = self._version_ids_to_keys(version_ids)
1908
present = self._present_keys(keys)
1909
missing = keys.difference(present)
1911
raise RevisionNotPresent(missing.pop(), self)
1913
def add_version(self, version_id, options, access_memo, parents):
1914
"""Add a version record to the index."""
1915
return self.add_versions(((version_id, options, access_memo, parents),))
1917
def add_versions(self, versions, random_id=False):
1918
"""Add multiple versions to the index.
1920
This function does not insert data into the Immutable GraphIndex
1921
backing the KnitGraphIndex, instead it prepares data for insertion by
1922
the caller and checks that it is safe to insert then calls
1923
self._add_callback with the prepared GraphIndex nodes.
1925
:param versions: a list of tuples:
1926
(version_id, options, pos, size, parents).
1927
:param random_id: If True the ids being added were randomly generated
1928
and no check for existence will be performed.
1930
if not self._add_callback:
1931
raise errors.ReadOnlyError(self)
1932
# we hope there are no repositories with inconsistent parentage
1937
for (version_id, options, access_memo, parents) in versions:
1938
index, pos, size = access_memo
1939
key = (version_id, )
1940
parents = tuple((parent, ) for parent in parents)
1941
if 'no-eol' in options:
1945
value += "%d %d" % (pos, size)
1946
if not self._deltas:
1947
if 'line-delta' in options:
1948
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1951
if 'line-delta' in options:
1952
node_refs = (parents, (parents[0],))
1954
node_refs = (parents, ())
1956
node_refs = (parents, )
1959
raise KnitCorrupt(self, "attempt to add node with parents "
1960
"in parentless index.")
1962
keys[key] = (value, node_refs)
1964
present_nodes = self._get_entries(keys)
1965
for (index, key, value, node_refs) in present_nodes:
1966
if (value, node_refs) != keys[key]:
1967
raise KnitCorrupt(self, "inconsistent details in add_versions"
1968
": %s %s" % ((value, node_refs), keys[key]))
1972
for key, (value, node_refs) in keys.iteritems():
1973
result.append((key, value, node_refs))
1975
for key, (value, node_refs) in keys.iteritems():
1976
result.append((key, value))
1977
self._add_callback(result)
1979
def _version_ids_to_keys(self, version_ids):
1980
return set((version_id, ) for version_id in version_ids)
1983
class _KnitAccess(object):
1984
"""Access to knit records in a .knit file."""
1986
def __init__(self, transport, filename, _file_mode, _dir_mode,
1987
_need_to_create, _create_parent_dir):
1988
"""Create a _KnitAccess for accessing and inserting data.
1990
:param transport: The transport the .knit is located on.
1991
:param filename: The filename of the .knit.
1993
self._transport = transport
1994
self._filename = filename
1995
self._file_mode = _file_mode
1996
self._dir_mode = _dir_mode
1997
self._need_to_create = _need_to_create
1998
self._create_parent_dir = _create_parent_dir
2000
def add_raw_records(self, sizes, raw_data):
2001
"""Add raw knit bytes to a storage area.
2003
The data is spooled to whereever the access method is storing data.
2005
:param sizes: An iterable containing the size of each raw data segment.
2006
:param raw_data: A bytestring containing the data.
2007
:return: A list of memos to retrieve the record later. Each memo is a
2008
tuple - (index, pos, length), where the index field is always None
2009
for the .knit access method.
2011
assert type(raw_data) == str, \
2012
'data must be plain bytes was %s' % type(raw_data)
2013
if not self._need_to_create:
2014
base = self._transport.append_bytes(self._filename, raw_data)
2016
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2017
create_parent_dir=self._create_parent_dir,
2018
mode=self._file_mode,
2019
dir_mode=self._dir_mode)
2020
self._need_to_create = False
2024
result.append((None, base, size))
2029
"""IFF this data access has its own storage area, initialise it.
2033
self._transport.put_bytes_non_atomic(self._filename, '',
2034
mode=self._file_mode)
2036
def open_file(self):
2037
"""IFF this data access can be represented as a single file, open it.
2039
For knits that are not mapped to a single file on disk this will
2042
:return: None or a file handle.
2045
return self._transport.get(self._filename)
2050
def get_raw_records(self, memos_for_retrieval):
2051
"""Get the raw bytes for a records.
2053
:param memos_for_retrieval: An iterable containing the (index, pos,
2054
length) memo for retrieving the bytes. The .knit method ignores
2055
the index as there is always only a single file.
2056
:return: An iterator over the bytes of the records.
2058
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2059
for pos, data in self._transport.readv(self._filename, read_vector):
2063
class _PackAccess(object):
2064
"""Access to knit records via a collection of packs."""
2066
def __init__(self, index_to_packs, writer=None):
2067
"""Create a _PackAccess object.
2069
:param index_to_packs: A dict mapping index objects to the transport
2070
and file names for obtaining data.
2071
:param writer: A tuple (pack.ContainerWriter, write_index) which
2072
contains the pack to write, and the index that reads from it will
2076
self.container_writer = writer[0]
2077
self.write_index = writer[1]
2079
self.container_writer = None
2080
self.write_index = None
2081
self.indices = index_to_packs
2083
def add_raw_records(self, sizes, raw_data):
2084
"""Add raw knit bytes to a storage area.
2086
The data is spooled to the container writer in one bytes-record per
2089
:param sizes: An iterable containing the size of each raw data segment.
2090
:param raw_data: A bytestring containing the data.
2091
:return: A list of memos to retrieve the record later. Each memo is a
2092
tuple - (index, pos, length), where the index field is the
2093
write_index object supplied to the PackAccess object.
2095
assert type(raw_data) == str, \
2096
'data must be plain bytes was %s' % type(raw_data)
2100
p_offset, p_length = self.container_writer.add_bytes_record(
2101
raw_data[offset:offset+size], [])
2103
result.append((self.write_index, p_offset, p_length))
2107
"""Pack based knits do not get individually created."""
2109
def get_raw_records(self, memos_for_retrieval):
2110
"""Get the raw bytes for a records.
2112
:param memos_for_retrieval: An iterable containing the (index, pos,
2113
length) memo for retrieving the bytes. The Pack access method
2114
looks up the pack to use for a given record in its index_to_pack
2116
:return: An iterator over the bytes of the records.
2118
# first pass, group into same-index requests
2120
current_index = None
2121
for (index, offset, length) in memos_for_retrieval:
2122
if current_index == index:
2123
current_list.append((offset, length))
2125
if current_index is not None:
2126
request_lists.append((current_index, current_list))
2127
current_index = index
2128
current_list = [(offset, length)]
2129
# handle the last entry
2130
if current_index is not None:
2131
request_lists.append((current_index, current_list))
2132
for index, offsets in request_lists:
2133
transport, path = self.indices[index]
2134
reader = pack.make_readv_reader(transport, path, offsets)
2135
for names, read_func in reader.iter_records():
2136
yield read_func(None)
2138
def open_file(self):
2139
"""Pack based knits have no single file."""
2142
def set_writer(self, writer, index, (transport, packname)):
2143
"""Set a writer to use for adding data."""
2144
if index is not None:
2145
self.indices[index] = (transport, packname)
2146
self.container_writer = writer
2147
self.write_index = index
2150
class _StreamAccess(object):
2151
"""A Knit Access object that provides data from a datastream.
2153
It also provides a fallback to present as unannotated data, annotated data
2154
from a *backing* access object.
2156
This is triggered by a index_memo which is pointing to a different index
2157
than this was constructed with, and is used to allow extracting full
2158
unannotated texts for insertion into annotated knits.
2161
def __init__(self, reader_callable, stream_index, backing_knit,
2163
"""Create a _StreamAccess object.
2165
:param reader_callable: The reader_callable from the datastream.
2166
This is called to buffer all the data immediately, for
2168
:param stream_index: The index the data stream this provides access to
2169
which will be present in native index_memo's.
2170
:param backing_knit: The knit object that will provide access to
2171
annotated texts which are not available in the stream, so as to
2172
create unannotated texts.
2173
:param orig_factory: The original content factory used to generate the
2174
stream. This is used for checking whether the thunk code for
2175
supporting _copy_texts will generate the correct form of data.
2177
self.data = reader_callable(None)
2178
self.stream_index = stream_index
2179
self.backing_knit = backing_knit
2180
self.orig_factory = orig_factory
2182
def get_raw_records(self, memos_for_retrieval):
2183
"""Get the raw bytes for a records.
2185
:param memos_for_retrieval: An iterable of memos from the
2186
_StreamIndex object identifying bytes to read; for these classes
2187
they are (from_backing_knit, index, start, end) and can point to
2188
either the backing knit or streamed data.
2189
:return: An iterator yielding a byte string for each record in
2190
memos_for_retrieval.
2192
# use a generator for memory friendliness
2193
for from_backing_knit, version_id, start, end in memos_for_retrieval:
2194
if not from_backing_knit:
2195
assert version_id is self.stream_index
2196
yield self.data[start:end]
2198
# we have been asked to thunk. This thunking only occurs when
2199
# we are obtaining plain texts from an annotated backing knit
2200
# so that _copy_texts will work.
2201
# We could improve performance here by scanning for where we need
2202
# to do this and using get_line_list, then interleaving the output
2203
# as desired. However, for now, this is sufficient.
2204
if self.orig_factory.__class__ != KnitPlainFactory:
2205
raise errors.KnitCorrupt(
2206
self, 'Bad thunk request %r cannot be backed by %r' %
2207
(version_id, self.orig_factory))
2208
lines = self.backing_knit.get_lines(version_id)
2209
line_bytes = ''.join(lines)
2210
digest = sha_string(line_bytes)
2211
# the packed form of the fulltext always has a trailing newline,
2212
# even if the actual text does not, unless the file is empty. the
2213
# record options including the noeol flag are passed through by
2214
# _StreamIndex, so this is safe.
2216
if lines[-1][-1] != '\n':
2217
lines[-1] = lines[-1] + '\n'
2219
# We want plain data, because we expect to thunk only to allow text
2221
size, bytes = self.backing_knit._data._record_to_data(version_id,
2222
digest, lines, line_bytes)
2226
class _StreamIndex(object):
2227
"""A Knit Index object that uses the data map from a datastream."""
2229
def __init__(self, data_list, backing_index):
2230
"""Create a _StreamIndex object.
2232
:param data_list: The data_list from the datastream.
2233
:param backing_index: The index which will supply values for nodes
2234
referenced outside of this stream.
2236
self.data_list = data_list
2237
self.backing_index = backing_index
2238
self._by_version = {}
2240
for key, options, length, parents in data_list:
2241
self._by_version[key] = options, (pos, pos + length), parents
2244
def get_ancestry(self, versions, topo_sorted):
2245
"""Get an ancestry list for versions."""
2247
# Not needed for basic joins
2248
raise NotImplementedError(self.get_ancestry)
2249
# get a graph of all the mentioned versions:
2250
# Little ugly - basically copied from KnitIndex, but don't want to
2251
# accidentally incorporate too much of that index's code.
2253
pending = set(versions)
2254
cache = self._by_version
2256
version = pending.pop()
2259
parents = [p for p in cache[version][2] if p in cache]
2261
raise RevisionNotPresent(version, self)
2262
# if not completed and not a ghost
2263
pending.update([p for p in parents if p not in ancestry])
2264
ancestry.add(version)
2265
return list(ancestry)
2267
def get_build_details(self, version_ids):
2268
"""Get the method, index_memo and compression parent for version_ids.
2270
Ghosts are omitted from the result.
2272
:param version_ids: An iterable of version_ids.
2273
:return: A dict of version_id:(index_memo, compression_parent,
2274
parents, record_details).
2276
opaque memo that can be passed to _StreamAccess.read_records
2277
to extract the raw data; for these classes it is
2278
(from_backing_knit, index, start, end)
2280
Content that this record is built upon, may be None
2282
Logical parents of this node
2284
extra information about the content which needs to be passed to
2285
Factory.parse_record
2288
for version_id in version_ids:
2290
method = self.get_method(version_id)
2291
except errors.RevisionNotPresent:
2292
# ghosts are omitted
2294
parent_ids = self.get_parents_with_ghosts(version_id)
2295
noeol = ('no-eol' in self.get_options(version_id))
2296
index_memo = self.get_position(version_id)
2297
from_backing_knit = index_memo[0]
2298
if from_backing_knit:
2299
# texts retrieved from the backing knit are always full texts
2301
if method == 'fulltext':
2302
compression_parent = None
2304
compression_parent = parent_ids[0]
2305
result[version_id] = (index_memo, compression_parent,
2306
parent_ids, (method, noeol))
2309
def get_method(self, version_id):
2310
"""Return compression method of specified version."""
2311
options = self.get_options(version_id)
2312
if 'fulltext' in options:
2314
elif 'line-delta' in options:
2317
raise errors.KnitIndexUnknownMethod(self, options)
2319
def get_options(self, version_id):
2320
"""Return a list representing options.
2325
return self._by_version[version_id][0]
2327
options = list(self.backing_index.get_options(version_id))
2328
if 'fulltext' in options:
2330
elif 'line-delta' in options:
2331
# Texts from the backing knit are always returned from the stream
2333
options.remove('line-delta')
2334
options.append('fulltext')
2336
raise errors.KnitIndexUnknownMethod(self, options)
2337
return tuple(options)
2339
def get_parent_map(self, version_ids):
2340
"""Passed through to by KnitVersionedFile.get_parent_map."""
2343
for version_id in version_ids:
2345
result[version_id] = self._by_version[version_id][2]
2347
pending_ids.add(version_id)
2348
result.update(self.backing_index.get_parent_map(pending_ids))
2351
def get_parents_with_ghosts(self, version_id):
2352
"""Return parents of specified version with ghosts."""
2354
return self.get_parent_map([version_id])[version_id]
2356
raise RevisionNotPresent(version_id, self)
2358
def get_position(self, version_id):
2359
"""Return details needed to access the version.
2361
_StreamAccess has the data as a big array, so we return slice
2362
coordinates into that (as index_memo's are opaque outside the
2363
index and matching access class).
2365
:return: a tuple (from_backing_knit, index, start, end) that can
2366
be passed e.g. to get_raw_records.
2367
If from_backing_knit is False, index will be self, otherwise it
2368
will be a version id.
2371
start, end = self._by_version[version_id][1]
2372
return False, self, start, end
2374
# Signal to the access object to handle this from the backing knit.
2375
return (True, version_id, None, None)
2377
def get_versions(self):
2378
"""Get all the versions in the stream."""
2379
return self._by_version.keys()
2382
class _KnitData(object):
2383
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2385
The KnitData class provides the logic for parsing and using knit records,
2386
making use of an access method for the low level read and write operations.
2389
def __init__(self, access):
2390
"""Create a KnitData object.
2392
:param access: The access method to use. Access methods such as
2393
_KnitAccess manage the insertion of raw records and the subsequent
2394
retrieval of the same.
2396
self._access = access
2397
self._checked = False
2399
def _open_file(self):
2400
return self._access.open_file()
2402
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2403
"""Convert version_id, digest, lines into a raw data block.
2405
:param dense_lines: The bytes of lines but in a denser form. For
2406
instance, if lines is a list of 1000 bytestrings each ending in \n,
2407
dense_lines may be a list with one line in it, containing all the
2408
1000's lines and their \n's. Using dense_lines if it is already
2409
known is a win because the string join to create bytes in this
2410
function spends less time resizing the final string.
2411
:return: (len, a StringIO instance with the raw data ready to read.)
2413
# Note: using a string copy here increases memory pressure with e.g.
2414
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2415
# when doing the initial commit of a mozilla tree. RBC 20070921
2416
bytes = ''.join(chain(
2417
["version %s %d %s\n" % (version_id,
2420
dense_lines or lines,
2421
["end %s\n" % version_id]))
2422
assert bytes.__class__ == str
2423
compressed_bytes = bytes_to_gzip(bytes)
2424
return len(compressed_bytes), compressed_bytes
2426
def add_raw_records(self, sizes, raw_data):
2427
"""Append a prepared record to the data file.
2429
:param sizes: An iterable containing the size of each raw data segment.
2430
:param raw_data: A bytestring containing the data.
2431
:return: a list of index data for the way the data was stored.
2432
See the access method add_raw_records documentation for more
2435
return self._access.add_raw_records(sizes, raw_data)
2437
def _parse_record_header(self, version_id, raw_data):
2438
"""Parse a record header for consistency.
2440
:return: the header and the decompressor stream.
2441
as (stream, header_record)
2443
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2445
rec = self._check_header(version_id, df.readline())
2446
except Exception, e:
2447
raise KnitCorrupt(self._access,
2448
"While reading {%s} got %s(%s)"
2449
% (version_id, e.__class__.__name__, str(e)))
2452
def _check_header(self, version_id, line):
2455
raise KnitCorrupt(self._access,
2456
'unexpected number of elements in record header')
2457
if rec[1] != version_id:
2458
raise KnitCorrupt(self._access,
2459
'unexpected version, wanted %r, got %r'
2460
% (version_id, rec[1]))
2463
def _parse_record(self, version_id, data):
2465
# 4168 calls in 2880 217 internal
2466
# 4168 calls to _parse_record_header in 2121
2467
# 4168 calls to readlines in 330
2468
df = GzipFile(mode='rb', fileobj=StringIO(data))
2471
record_contents = df.readlines()
2472
except Exception, e:
2473
raise KnitCorrupt(self._access,
2474
"While reading {%s} got %s(%s)"
2475
% (version_id, e.__class__.__name__, str(e)))
2476
header = record_contents.pop(0)
2477
rec = self._check_header(version_id, header)
2479
last_line = record_contents.pop()
2480
if len(record_contents) != int(rec[2]):
2481
raise KnitCorrupt(self._access,
2482
'incorrect number of lines %s != %s'
2484
% (len(record_contents), int(rec[2]),
2486
if last_line != 'end %s\n' % rec[1]:
2487
raise KnitCorrupt(self._access,
2488
'unexpected version end line %r, wanted %r'
2489
% (last_line, version_id))
2491
return record_contents, rec[3]
2493
def read_records_iter_raw(self, records):
2494
"""Read text records from data file and yield raw data.
2496
This unpacks enough of the text record to validate the id is
2497
as expected but thats all.
2499
# setup an iterator of the external records:
2500
# uses readv so nice and fast we hope.
2502
# grab the disk data needed.
2503
needed_offsets = [index_memo for version_id, index_memo
2505
raw_records = self._access.get_raw_records(needed_offsets)
2507
for version_id, index_memo in records:
2508
data = raw_records.next()
2509
# validate the header
2510
df, rec = self._parse_record_header(version_id, data)
2512
yield version_id, data
2514
def read_records_iter(self, records):
2515
"""Read text records from data file and yield result.
2517
The result will be returned in whatever is the fastest to read.
2518
Not by the order requested. Also, multiple requests for the same
2519
record will only yield 1 response.
2520
:param records: A list of (version_id, pos, len) entries
2521
:return: Yields (version_id, contents, digest) in the order
2522
read, not the order requested
2527
needed_records = sorted(set(records), key=operator.itemgetter(1))
2528
if not needed_records:
2531
# The transport optimizes the fetching as well
2532
# (ie, reads continuous ranges.)
2533
raw_data = self._access.get_raw_records(
2534
[index_memo for version_id, index_memo in needed_records])
2536
for (version_id, index_memo), data in \
2537
izip(iter(needed_records), raw_data):
2538
content, digest = self._parse_record(version_id, data)
2539
yield version_id, content, digest
2541
def read_records(self, records):
2542
"""Read records into a dictionary."""
2544
for record_id, content, digest in \
2545
self.read_records_iter(records):
2546
components[record_id] = (content, digest)
2550
class InterKnit(InterVersionedFile):
2551
"""Optimised code paths for knit to knit operations."""
2553
_matching_file_from_factory = staticmethod(make_file_knit)
2554
_matching_file_to_factory = staticmethod(make_file_knit)
2557
def is_compatible(source, target):
2558
"""Be compatible with knits. """
2560
return (isinstance(source, KnitVersionedFile) and
2561
isinstance(target, KnitVersionedFile))
2562
except AttributeError:
2565
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2566
"""Copy texts to the target by extracting and adding them one by one.
2568
see join() for the parameter definitions.
2570
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2571
# --- the below is factorable out with VersionedFile.join, but wait for
2572
# VersionedFiles, it may all be simpler then.
2573
graph = Graph(self.source)
2574
search = graph._make_breadth_first_searcher(version_ids)
2575
transitive_ids = set()
2576
map(transitive_ids.update, list(search))
2577
parent_map = self.source.get_parent_map(transitive_ids)
2578
order = topo_sort(parent_map.items())
2580
def size_of_content(content):
2581
return sum(len(line) for line in content.text())
2582
# Cache at most 10MB of parent texts
2583
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2584
compute_size=size_of_content)
2585
# TODO: jam 20071116 It would be nice to have a streaming interface to
2586
# get multiple texts from a source. The source could be smarter
2587
# about how it handled intermediate stages.
2588
# get_line_list() or make_mpdiffs() seem like a possibility, but
2589
# at the moment they extract all full texts into memory, which
2590
# causes us to store more than our 3x fulltext goal.
2591
# Repository.iter_files_bytes() may be another possibility
2592
to_process = [version for version in order
2593
if version not in self.target]
2594
total = len(to_process)
2595
pb = ui.ui_factory.nested_progress_bar()
2597
for index, version in enumerate(to_process):
2598
pb.update('Converting versioned data', index, total)
2599
sha1, num_bytes, parent_text = self.target.add_lines(version,
2600
self.source.get_parents_with_ghosts(version),
2601
self.source.get_lines(version),
2602
parent_texts=parent_cache)
2603
parent_cache[version] = parent_text
2608
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2609
"""See InterVersionedFile.join."""
2610
assert isinstance(self.source, KnitVersionedFile)
2611
assert isinstance(self.target, KnitVersionedFile)
2613
# If the source and target are mismatched w.r.t. annotations vs
2614
# plain, the data needs to be converted accordingly
2615
if self.source.factory.annotated == self.target.factory.annotated:
2617
elif self.source.factory.annotated:
2618
converter = self._anno_to_plain_converter
2620
# We're converting from a plain to an annotated knit. Copy them
2621
# across by full texts.
2622
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2624
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2628
pb = ui.ui_factory.nested_progress_bar()
2630
version_ids = list(version_ids)
2631
if None in version_ids:
2632
version_ids.remove(None)
2634
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2636
this_versions = set(self.target._index.get_versions())
2637
# XXX: For efficiency we should not look at the whole index,
2638
# we only need to consider the referenced revisions - they
2639
# must all be present, or the method must be full-text.
2640
# TODO, RBC 20070919
2641
needed_versions = self.source_ancestry - this_versions
2643
if not needed_versions:
2645
full_list = topo_sort(
2646
self.source.get_parent_map(self.source.versions()))
2648
version_list = [i for i in full_list if (not self.target.has_version(i)
2649
and i in needed_versions)]
2653
copy_queue_records = []
2655
for version_id in version_list:
2656
options = self.source._index.get_options(version_id)
2657
parents = self.source._index.get_parents_with_ghosts(version_id)
2658
# check that its will be a consistent copy:
2659
for parent in parents:
2660
# if source has the parent, we must :
2661
# * already have it or
2662
# * have it scheduled already
2663
# otherwise we don't care
2664
assert (self.target.has_version(parent) or
2665
parent in copy_set or
2666
not self.source.has_version(parent))
2667
index_memo = self.source._index.get_position(version_id)
2668
copy_queue_records.append((version_id, index_memo))
2669
copy_queue.append((version_id, options, parents))
2670
copy_set.add(version_id)
2672
# data suck the join:
2674
total = len(version_list)
2677
for (version_id, raw_data), \
2678
(version_id2, options, parents) in \
2679
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2681
assert version_id == version_id2, 'logic error, inconsistent results'
2683
pb.update("Joining knit", count, total)
2685
size, raw_data = converter(raw_data, version_id, options,
2688
size = len(raw_data)
2689
raw_records.append((version_id, options, parents, size))
2690
raw_datum.append(raw_data)
2691
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2696
def _anno_to_plain_converter(self, raw_data, version_id, options,
2698
"""Convert annotated content to plain content."""
2699
data, digest = self.source._data._parse_record(version_id, raw_data)
2700
if 'fulltext' in options:
2701
content = self.source.factory.parse_fulltext(data, version_id)
2702
lines = self.target.factory.lower_fulltext(content)
2704
delta = self.source.factory.parse_line_delta(data, version_id,
2706
lines = self.target.factory.lower_line_delta(delta)
2707
return self.target._data._record_to_data(version_id, digest, lines)
2710
InterVersionedFile.register_optimiser(InterKnit)
2713
class WeaveToKnit(InterVersionedFile):
2714
"""Optimised code paths for weave to knit operations."""
2716
_matching_file_from_factory = bzrlib.weave.WeaveFile
2717
_matching_file_to_factory = staticmethod(make_file_knit)
2720
def is_compatible(source, target):
2721
"""Be compatible with weaves to knits."""
2723
return (isinstance(source, bzrlib.weave.Weave) and
2724
isinstance(target, KnitVersionedFile))
2725
except AttributeError:
2728
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2729
"""See InterVersionedFile.join."""
2730
assert isinstance(self.source, bzrlib.weave.Weave)
2731
assert isinstance(self.target, KnitVersionedFile)
2733
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2738
pb = ui.ui_factory.nested_progress_bar()
2740
version_ids = list(version_ids)
2742
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2743
this_versions = set(self.target._index.get_versions())
2744
needed_versions = self.source_ancestry - this_versions
2746
if not needed_versions:
2748
full_list = topo_sort(
2749
self.source.get_parent_map(self.source.versions()))
2751
version_list = [i for i in full_list if (not self.target.has_version(i)
2752
and i in needed_versions)]
2756
total = len(version_list)
2757
parent_map = self.source.get_parent_map(version_list)
2758
for version_id in version_list:
2759
pb.update("Converting to knit", count, total)
2760
parents = parent_map[version_id]
2761
# check that its will be a consistent copy:
2762
for parent in parents:
2763
# if source has the parent, we must already have it
2764
assert (self.target.has_version(parent))
2765
self.target.add_lines(
2766
version_id, parents, self.source.get_lines(version_id))
2773
InterVersionedFile.register_optimiser(WeaveToKnit)
2776
# Deprecated, use PatienceSequenceMatcher instead
2777
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2780
def annotate_knit(knit, revision_id):
2781
"""Annotate a knit with no cached annotations.
2783
This implementation is for knits with no cached annotations.
2784
It will work for knits with cached annotations, but this is not
2787
annotator = _KnitAnnotator(knit)
2788
return iter(annotator.annotate(revision_id))
2791
class _KnitAnnotator(object):
2792
"""Build up the annotations for a text."""
2794
def __init__(self, knit):
2797
# Content objects, differs from fulltexts because of how final newlines
2798
# are treated by knits. the content objects here will always have a
2800
self._fulltext_contents = {}
2802
# Annotated lines of specific revisions
2803
self._annotated_lines = {}
2805
# Track the raw data for nodes that we could not process yet.
2806
# This maps the revision_id of the base to a list of children that will
2807
# annotated from it.
2808
self._pending_children = {}
2810
# Nodes which cannot be extracted
2811
self._ghosts = set()
2813
# Track how many children this node has, so we know if we need to keep
2815
self._annotate_children = {}
2816
self._compression_children = {}
2818
self._all_build_details = {}
2819
# The children => parent revision_id graph
2820
self._revision_id_graph = {}
2822
self._heads_provider = None
2824
self._nodes_to_keep_annotations = set()
2825
self._generations_until_keep = 100
2827
def set_generations_until_keep(self, value):
2828
"""Set the number of generations before caching a node.
2830
Setting this to -1 will cache every merge node, setting this higher
2831
will cache fewer nodes.
2833
self._generations_until_keep = value
2835
def _add_fulltext_content(self, revision_id, content_obj):
2836
self._fulltext_contents[revision_id] = content_obj
2837
# TODO: jam 20080305 It might be good to check the sha1digest here
2838
return content_obj.text()
2840
def _check_parents(self, child, nodes_to_annotate):
2841
"""Check if all parents have been processed.
2843
:param child: A tuple of (rev_id, parents, raw_content)
2844
:param nodes_to_annotate: If child is ready, add it to
2845
nodes_to_annotate, otherwise put it back in self._pending_children
2847
for parent_id in child[1]:
2848
if (parent_id not in self._annotated_lines):
2849
# This parent is present, but another parent is missing
2850
self._pending_children.setdefault(parent_id,
2854
# This one is ready to be processed
2855
nodes_to_annotate.append(child)
2857
def _add_annotation(self, revision_id, fulltext, parent_ids,
2858
left_matching_blocks=None):
2859
"""Add an annotation entry.
2861
All parents should already have been annotated.
2862
:return: A list of children that now have their parents satisfied.
2864
a = self._annotated_lines
2865
annotated_parent_lines = [a[p] for p in parent_ids]
2866
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2867
fulltext, revision_id, left_matching_blocks,
2868
heads_provider=self._get_heads_provider()))
2869
self._annotated_lines[revision_id] = annotated_lines
2870
for p in parent_ids:
2871
ann_children = self._annotate_children[p]
2872
ann_children.remove(revision_id)
2873
if (not ann_children
2874
and p not in self._nodes_to_keep_annotations):
2875
del self._annotated_lines[p]
2876
del self._all_build_details[p]
2877
if p in self._fulltext_contents:
2878
del self._fulltext_contents[p]
2879
# Now that we've added this one, see if there are any pending
2880
# deltas to be done, certainly this parent is finished
2881
nodes_to_annotate = []
2882
for child in self._pending_children.pop(revision_id, []):
2883
self._check_parents(child, nodes_to_annotate)
2884
return nodes_to_annotate
2886
def _get_build_graph(self, revision_id):
2887
"""Get the graphs for building texts and annotations.
2889
The data you need for creating a full text may be different than the
2890
data you need to annotate that text. (At a minimum, you need both
2891
parents to create an annotation, but only need 1 parent to generate the
2894
:return: A list of (revision_id, index_memo) records, suitable for
2895
passing to read_records_iter to start reading in the raw data fro/
2898
if revision_id in self._annotated_lines:
2901
pending = set([revision_id])
2906
# get all pending nodes
2908
this_iteration = pending
2909
build_details = self._knit._index.get_build_details(this_iteration)
2910
self._all_build_details.update(build_details)
2911
# new_nodes = self._knit._index._get_entries(this_iteration)
2913
for rev_id, details in build_details.iteritems():
2914
(index_memo, compression_parent, parents,
2915
record_details) = details
2916
self._revision_id_graph[rev_id] = parents
2917
records.append((rev_id, index_memo))
2918
# Do we actually need to check _annotated_lines?
2919
pending.update(p for p in parents
2920
if p not in self._all_build_details)
2921
if compression_parent:
2922
self._compression_children.setdefault(compression_parent,
2925
for parent in parents:
2926
self._annotate_children.setdefault(parent,
2928
num_gens = generation - kept_generation
2929
if ((num_gens >= self._generations_until_keep)
2930
and len(parents) > 1):
2931
kept_generation = generation
2932
self._nodes_to_keep_annotations.add(rev_id)
2934
missing_versions = this_iteration.difference(build_details.keys())
2935
self._ghosts.update(missing_versions)
2936
for missing_version in missing_versions:
2937
# add a key, no parents
2938
self._revision_id_graph[missing_version] = ()
2939
pending.discard(missing_version) # don't look for it
2940
# XXX: This should probably be a real exception, as it is a data
2942
assert not self._ghosts.intersection(self._compression_children), \
2943
"We cannot have nodes which have a compression parent of a ghost."
2944
# Cleanout anything that depends on a ghost so that we don't wait for
2945
# the ghost to show up
2946
for node in self._ghosts:
2947
if node in self._annotate_children:
2948
# We won't be building this node
2949
del self._annotate_children[node]
2950
# Generally we will want to read the records in reverse order, because
2951
# we find the parent nodes after the children
2955
def _annotate_records(self, records):
2956
"""Build the annotations for the listed records."""
2957
# We iterate in the order read, rather than a strict order requested
2958
# However, process what we can, and put off to the side things that
2959
# still need parents, cleaning them up when those parents are
2961
for (rev_id, record,
2962
digest) in self._knit._data.read_records_iter(records):
2963
if rev_id in self._annotated_lines:
2965
parent_ids = self._revision_id_graph[rev_id]
2966
parent_ids = [p for p in parent_ids if p not in self._ghosts]
2967
details = self._all_build_details[rev_id]
2968
(index_memo, compression_parent, parents,
2969
record_details) = details
2970
nodes_to_annotate = []
2971
# TODO: Remove the punning between compression parents, and
2972
# parent_ids, we should be able to do this without assuming
2974
if len(parent_ids) == 0:
2975
# There are no parents for this node, so just add it
2976
# TODO: This probably needs to be decoupled
2977
assert compression_parent is None
2978
fulltext_content, delta = self._knit.factory.parse_record(
2979
rev_id, record, record_details, None)
2980
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
2981
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
2982
parent_ids, left_matching_blocks=None))
2984
child = (rev_id, parent_ids, record)
2985
# Check if all the parents are present
2986
self._check_parents(child, nodes_to_annotate)
2987
while nodes_to_annotate:
2988
# Should we use a queue here instead of a stack?
2989
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
2990
(index_memo, compression_parent, parents,
2991
record_details) = self._all_build_details[rev_id]
2992
if compression_parent is not None:
2993
comp_children = self._compression_children[compression_parent]
2994
assert rev_id in comp_children
2995
# If there is only 1 child, it is safe to reuse this
2997
reuse_content = (len(comp_children) == 1
2998
and compression_parent not in
2999
self._nodes_to_keep_annotations)
3001
# Remove it from the cache since it will be changing
3002
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3003
# Make sure to copy the fulltext since it might be
3005
parent_fulltext = list(parent_fulltext_content.text())
3007
parent_fulltext_content = self._fulltext_contents[compression_parent]
3008
parent_fulltext = parent_fulltext_content.text()
3009
comp_children.remove(rev_id)
3010
fulltext_content, delta = self._knit.factory.parse_record(
3011
rev_id, record, record_details,
3012
parent_fulltext_content,
3013
copy_base_content=(not reuse_content))
3014
fulltext = self._add_fulltext_content(rev_id,
3016
blocks = KnitContent.get_line_delta_blocks(delta,
3017
parent_fulltext, fulltext)
3019
fulltext_content = self._knit.factory.parse_fulltext(
3021
fulltext = self._add_fulltext_content(rev_id,
3024
nodes_to_annotate.extend(
3025
self._add_annotation(rev_id, fulltext, parent_ids,
3026
left_matching_blocks=blocks))
3028
def _get_heads_provider(self):
3029
"""Create a heads provider for resolving ancestry issues."""
3030
if self._heads_provider is not None:
3031
return self._heads_provider
3032
parent_provider = _mod_graph.DictParentsProvider(
3033
self._revision_id_graph)
3034
graph_obj = _mod_graph.Graph(parent_provider)
3035
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3036
self._heads_provider = head_cache
3039
def annotate(self, revision_id):
3040
"""Return the annotated fulltext at the given revision.
3042
:param revision_id: The revision id for this file
3044
records = self._get_build_graph(revision_id)
3045
if revision_id in self._ghosts:
3046
raise errors.RevisionNotPresent(revision_id, self._knit)
3047
self._annotate_records(records)
3048
return self._annotated_lines[revision_id]
3052
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3054
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
