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.tuned_gzip import GzipFile, bytes_to_gzip
105
from bzrlib.osutils import (
111
from bzrlib.symbol_versioning import DEPRECATED_PARAMETER, deprecated_passed
112
from bzrlib.tsort import topo_sort
115
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
118
# TODO: Split out code specific to this format into an associated object.
120
# TODO: Can we put in some kind of value to check that the index and data
121
# files belong together?
123
# TODO: accommodate binaries, perhaps by storing a byte count
125
# TODO: function to check whole file
127
# TODO: atomically append data, then measure backwards from the cursor
128
# position after writing to work out where it was located. we may need to
129
# bypass python file buffering.
131
DATA_SUFFIX = '.knit'
132
INDEX_SUFFIX = '.kndx'
135
class KnitContent(object):
136
"""Content of a knit version to which deltas can be applied."""
139
self._should_strip_eol = False
142
"""Return a list of (origin, text) tuples."""
143
return list(self.annotate_iter())
145
def apply_delta(self, delta, new_version_id):
146
"""Apply delta to this object to become new_version_id."""
147
raise NotImplementedError(self.apply_delta)
149
def cleanup_eol(self, copy_on_mutate=True):
150
if self._should_strip_eol:
152
self._lines = self._lines[:]
153
self.strip_last_line_newline()
155
def line_delta_iter(self, new_lines):
156
"""Generate line-based delta from this content to new_lines."""
157
new_texts = new_lines.text()
158
old_texts = self.text()
159
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
160
for tag, i1, i2, j1, j2 in s.get_opcodes():
163
# ofrom, oto, length, data
164
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
166
def line_delta(self, new_lines):
167
return list(self.line_delta_iter(new_lines))
170
def get_line_delta_blocks(knit_delta, source, target):
171
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
172
target_len = len(target)
175
for s_begin, s_end, t_len, new_text in knit_delta:
176
true_n = s_begin - s_pos
179
# knit deltas do not provide reliable info about whether the
180
# last line of a file matches, due to eol handling.
181
if source[s_pos + n -1] != target[t_pos + n -1]:
184
yield s_pos, t_pos, n
185
t_pos += t_len + true_n
187
n = target_len - t_pos
189
if source[s_pos + n -1] != target[t_pos + n -1]:
192
yield s_pos, t_pos, n
193
yield s_pos + (target_len - t_pos), target_len, 0
196
class AnnotatedKnitContent(KnitContent):
197
"""Annotated content."""
199
def __init__(self, lines):
200
KnitContent.__init__(self)
203
def annotate_iter(self):
204
"""Yield tuples of (origin, text) for each content line."""
205
return iter(self._lines)
207
def apply_delta(self, delta, new_version_id):
208
"""Apply delta to this object to become new_version_id."""
211
for start, end, count, delta_lines in delta:
212
lines[offset+start:offset+end] = delta_lines
213
offset = offset + (start - end) + count
215
def strip_last_line_newline(self):
216
line = self._lines[-1][1].rstrip('\n')
217
self._lines[-1] = (self._lines[-1][0], line)
218
self._should_strip_eol = False
222
lines = [text for origin, text in self._lines]
223
except ValueError, e:
224
# most commonly (only?) caused by the internal form of the knit
225
# missing annotation information because of a bug - see thread
227
raise KnitCorrupt(self,
228
"line in annotated knit missing annotation information: %s"
231
if self._should_strip_eol:
232
anno, line = lines[-1]
233
lines[-1] = (anno, line.rstrip('\n'))
237
return AnnotatedKnitContent(self._lines[:])
240
class PlainKnitContent(KnitContent):
241
"""Unannotated content.
243
When annotate[_iter] is called on this content, the same version is reported
244
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
248
def __init__(self, lines, version_id):
249
KnitContent.__init__(self)
251
self._version_id = version_id
253
def annotate_iter(self):
254
"""Yield tuples of (origin, text) for each content line."""
255
for line in self._lines:
256
yield self._version_id, line
258
def apply_delta(self, delta, new_version_id):
259
"""Apply delta to this object to become new_version_id."""
262
for start, end, count, delta_lines in delta:
263
lines[offset+start:offset+end] = delta_lines
264
offset = offset + (start - end) + count
265
self._version_id = new_version_id
268
return PlainKnitContent(self._lines[:], self._version_id)
270
def strip_last_line_newline(self):
271
self._lines[-1] = self._lines[-1].rstrip('\n')
272
self._should_strip_eol = False
276
if self._should_strip_eol:
278
lines[-1] = lines[-1].rstrip('\n')
282
class _KnitFactory(object):
283
"""Base class for common Factory functions."""
285
def parse_record(self, version_id, record, record_details,
286
base_content, copy_base_content=True):
287
"""Parse a record into a full content object.
289
:param version_id: The official version id for this content
290
:param record: The data returned by read_records_iter()
291
:param record_details: Details about the record returned by
293
:param base_content: If get_build_details returns a compression_parent,
294
you must return a base_content here, else use None
295
:param copy_base_content: When building from the base_content, decide
296
you can either copy it and return a new object, or modify it in
298
:return: (content, delta) A Content object and possibly a line-delta,
301
method, noeol = record_details
302
if method == 'line-delta':
303
assert base_content is not None
304
if copy_base_content:
305
content = base_content.copy()
307
content = base_content
308
delta = self.parse_line_delta(record, version_id)
309
content.apply_delta(delta, version_id)
311
content = self.parse_fulltext(record, version_id)
313
content._should_strip_eol = noeol
314
return (content, delta)
317
class KnitAnnotateFactory(_KnitFactory):
318
"""Factory for creating annotated Content objects."""
322
def make(self, lines, version_id):
323
num_lines = len(lines)
324
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
326
def parse_fulltext(self, content, version_id):
327
"""Convert fulltext to internal representation
329
fulltext content is of the format
330
revid(utf8) plaintext\n
331
internal representation is of the format:
334
# TODO: jam 20070209 The tests expect this to be returned as tuples,
335
# but the code itself doesn't really depend on that.
336
# Figure out a way to not require the overhead of turning the
337
# list back into tuples.
338
lines = [tuple(line.split(' ', 1)) for line in content]
339
return AnnotatedKnitContent(lines)
341
def parse_line_delta_iter(self, lines):
342
return iter(self.parse_line_delta(lines))
344
def parse_line_delta(self, lines, version_id, plain=False):
345
"""Convert a line based delta into internal representation.
347
line delta is in the form of:
348
intstart intend intcount
350
revid(utf8) newline\n
351
internal representation is
352
(start, end, count, [1..count tuples (revid, newline)])
354
:param plain: If True, the lines are returned as a plain
355
list without annotations, not as a list of (origin, content) tuples, i.e.
356
(start, end, count, [1..count newline])
363
def cache_and_return(line):
364
origin, text = line.split(' ', 1)
365
return cache.setdefault(origin, origin), text
367
# walk through the lines parsing.
368
# Note that the plain test is explicitly pulled out of the
369
# loop to minimise any performance impact
372
start, end, count = [int(n) for n in header.split(',')]
373
contents = [next().split(' ', 1)[1] for i in xrange(count)]
374
result.append((start, end, count, contents))
377
start, end, count = [int(n) for n in header.split(',')]
378
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
379
result.append((start, end, count, contents))
382
def get_fulltext_content(self, lines):
383
"""Extract just the content lines from a fulltext."""
384
return (line.split(' ', 1)[1] for line in lines)
386
def get_linedelta_content(self, lines):
387
"""Extract just the content from a line delta.
389
This doesn't return all of the extra information stored in a delta.
390
Only the actual content lines.
395
header = header.split(',')
396
count = int(header[2])
397
for i in xrange(count):
398
origin, text = next().split(' ', 1)
401
def lower_fulltext(self, content):
402
"""convert a fulltext content record into a serializable form.
404
see parse_fulltext which this inverts.
406
# TODO: jam 20070209 We only do the caching thing to make sure that
407
# the origin is a valid utf-8 line, eventually we could remove it
408
return ['%s %s' % (o, t) for o, t in content._lines]
410
def lower_line_delta(self, delta):
411
"""convert a delta into a serializable form.
413
See parse_line_delta which this inverts.
415
# TODO: jam 20070209 We only do the caching thing to make sure that
416
# the origin is a valid utf-8 line, eventually we could remove it
418
for start, end, c, lines in delta:
419
out.append('%d,%d,%d\n' % (start, end, c))
420
out.extend(origin + ' ' + text
421
for origin, text in lines)
424
def annotate_iter(self, knit, version_id):
425
content = knit._get_content(version_id)
426
return content.annotate_iter()
429
class KnitPlainFactory(_KnitFactory):
430
"""Factory for creating plain Content objects."""
434
def make(self, lines, version_id):
435
return PlainKnitContent(lines, version_id)
437
def parse_fulltext(self, content, version_id):
438
"""This parses an unannotated fulltext.
440
Note that this is not a noop - the internal representation
441
has (versionid, line) - its just a constant versionid.
443
return self.make(content, version_id)
445
def parse_line_delta_iter(self, lines, version_id):
447
num_lines = len(lines)
448
while cur < num_lines:
451
start, end, c = [int(n) for n in header.split(',')]
452
yield start, end, c, lines[cur:cur+c]
455
def parse_line_delta(self, lines, version_id):
456
return list(self.parse_line_delta_iter(lines, version_id))
458
def get_fulltext_content(self, lines):
459
"""Extract just the content lines from a fulltext."""
462
def get_linedelta_content(self, lines):
463
"""Extract just the content from a line delta.
465
This doesn't return all of the extra information stored in a delta.
466
Only the actual content lines.
471
header = header.split(',')
472
count = int(header[2])
473
for i in xrange(count):
476
def lower_fulltext(self, content):
477
return content.text()
479
def lower_line_delta(self, delta):
481
for start, end, c, lines in delta:
482
out.append('%d,%d,%d\n' % (start, end, c))
486
def annotate_iter(self, knit, version_id):
487
annotator = _KnitAnnotator(knit)
488
return iter(annotator.annotate(version_id))
491
def make_empty_knit(transport, relpath):
492
"""Construct a empty knit at the specified location."""
493
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
496
class KnitVersionedFile(VersionedFile):
497
"""Weave-like structure with faster random access.
499
A knit stores a number of texts and a summary of the relationships
500
between them. Texts are identified by a string version-id. Texts
501
are normally stored and retrieved as a series of lines, but can
502
also be passed as single strings.
504
Lines are stored with the trailing newline (if any) included, to
505
avoid special cases for files with no final newline. Lines are
506
composed of 8-bit characters, not unicode. The combination of
507
these approaches should mean any 'binary' file can be safely
508
stored and retrieved.
511
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
512
factory=None, delta=True, create=False, create_parent_dir=False,
513
delay_create=False, dir_mode=None, index=None, access_method=None):
514
"""Construct a knit at location specified by relpath.
516
:param create: If not True, only open an existing knit.
517
:param create_parent_dir: If True, create the parent directory if
518
creating the file fails. (This is used for stores with
519
hash-prefixes that may not exist yet)
520
:param delay_create: The calling code is aware that the knit won't
521
actually be created until the first data is stored.
522
:param index: An index to use for the knit.
524
if access_mode is None:
526
super(KnitVersionedFile, self).__init__(access_mode)
527
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
528
self.transport = transport
529
self.filename = relpath
530
self.factory = factory or KnitAnnotateFactory()
531
self.writable = (access_mode == 'w')
534
self._max_delta_chain = 200
537
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
538
access_mode, create=create, file_mode=file_mode,
539
create_parent_dir=create_parent_dir, delay_create=delay_create,
543
if access_method is None:
544
_access = _KnitAccess(transport, relpath + DATA_SUFFIX, file_mode, dir_mode,
545
((create and not len(self)) and delay_create), create_parent_dir)
547
_access = access_method
548
if create and not len(self) and not delay_create:
550
self._data = _KnitData(_access)
553
return '%s(%s)' % (self.__class__.__name__,
554
self.transport.abspath(self.filename))
556
def _check_should_delta(self, first_parents):
557
"""Iterate back through the parent listing, looking for a fulltext.
559
This is used when we want to decide whether to add a delta or a new
560
fulltext. It searches for _max_delta_chain parents. When it finds a
561
fulltext parent, it sees if the total size of the deltas leading up to
562
it is large enough to indicate that we want a new full text anyway.
564
Return True if we should create a new delta, False if we should use a
569
delta_parents = first_parents
570
for count in xrange(self._max_delta_chain):
571
parent = delta_parents[0]
572
method = self._index.get_method(parent)
573
index, pos, size = self._index.get_position(parent)
574
if method == 'fulltext':
578
delta_parents = self._index.get_parents(parent)
580
# We couldn't find a fulltext, so we must create a new one
583
return fulltext_size > delta_size
585
def _add_raw_records(self, records, data):
586
"""Add all the records 'records' with data pre-joined in 'data'.
588
:param records: A list of tuples(version_id, options, parents, size).
589
:param data: The data for the records. When it is written, the records
590
are adjusted to have pos pointing into data by the sum of
591
the preceding records sizes.
594
raw_record_sizes = [record[3] for record in records]
595
positions = self._data.add_raw_records(raw_record_sizes, data)
598
for (version_id, options, parents, size), access_memo in zip(
600
index_entries.append((version_id, options, access_memo, parents))
601
if self._data._do_cache:
602
self._data._cache[version_id] = data[offset:offset+size]
604
self._index.add_versions(index_entries)
606
def enable_cache(self):
607
"""Start caching data for this knit"""
608
self._data.enable_cache()
610
def clear_cache(self):
611
"""Clear the data cache only."""
612
self._data.clear_cache()
614
def copy_to(self, name, transport):
615
"""See VersionedFile.copy_to()."""
616
# copy the current index to a temp index to avoid racing with local
618
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
619
self.transport.get(self._index._filename))
621
f = self._data._open_file()
623
transport.put_file(name + DATA_SUFFIX, f)
626
# move the copied index into place
627
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
629
def create_empty(self, name, transport, mode=None):
630
return KnitVersionedFile(name, transport, factory=self.factory,
631
delta=self.delta, create=True)
633
def get_data_stream(self, required_versions):
634
"""Get a data stream for the specified versions.
636
Versions may be returned in any order, not necessarily the order
637
specified. They are returned in a partial order by compression
638
parent, so that the deltas can be applied as the data stream is
639
inserted; however note that compression parents will not be sent
640
unless they were specifically requested, as the client may already
643
:param required_versions: The exact set of versions to be extracted.
644
Unlike some other knit methods, this is not used to generate a
645
transitive closure, rather it is used precisely as given.
647
:returns: format_signature, list of (version, options, length, parents),
650
required_version_set = frozenset(required_versions)
652
# list of revisions that can just be sent without waiting for their
655
# map from revision to the children based on it
657
# first, read all relevant index data, enough to sort into the right
659
for version_id in required_versions:
660
options = self._index.get_options(version_id)
661
parents = self._index.get_parents_with_ghosts(version_id)
662
index_memo = self._index.get_position(version_id)
663
version_index[version_id] = (index_memo, options, parents)
664
if ('line-delta' in options
665
and parents[0] in required_version_set):
666
# must wait until the parent has been sent
667
deferred.setdefault(parents[0], []). \
670
# either a fulltext, or a delta whose parent the client did
671
# not ask for and presumably already has
672
ready_to_send.append(version_id)
673
# build a list of results to return, plus instructions for data to
675
copy_queue_records = []
676
temp_version_list = []
678
# XXX: pushing and popping lists may be a bit inefficient
679
version_id = ready_to_send.pop(0)
680
(index_memo, options, parents) = version_index[version_id]
681
copy_queue_records.append((version_id, index_memo))
682
none, data_pos, data_size = index_memo
683
temp_version_list.append((version_id, options, data_size,
685
if version_id in deferred:
686
# now we can send all the children of this revision - we could
687
# put them in anywhere, but we hope that sending them soon
688
# after the fulltext will give good locality in the receiver
689
ready_to_send[:0] = deferred.pop(version_id)
690
assert len(deferred) == 0, \
691
"Still have compressed child versions waiting to be sent"
692
# XXX: The stream format is such that we cannot stream it - we have to
693
# know the length of all the data a-priori.
695
result_version_list = []
696
for (version_id, raw_data), \
697
(version_id2, options, _, parents) in \
698
izip(self._data.read_records_iter_raw(copy_queue_records),
700
assert version_id == version_id2, \
701
'logic error, inconsistent results'
702
raw_datum.append(raw_data)
703
result_version_list.append(
704
(version_id, options, len(raw_data), parents))
705
# provide a callback to get data incrementally.
706
pseudo_file = StringIO(''.join(raw_datum))
709
return pseudo_file.read()
711
return pseudo_file.read(length)
712
return (self.get_format_signature(), result_version_list, read)
714
def _extract_blocks(self, version_id, source, target):
715
if self._index.get_method(version_id) != 'line-delta':
717
parent, sha1, noeol, delta = self.get_delta(version_id)
718
return KnitContent.get_line_delta_blocks(delta, source, target)
720
def get_delta(self, version_id):
721
"""Get a delta for constructing version from some other version."""
722
self.check_not_reserved_id(version_id)
723
parents = self.get_parents(version_id)
728
index_memo = self._index.get_position(version_id)
729
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
730
noeol = 'no-eol' in self._index.get_options(version_id)
731
if 'fulltext' == self._index.get_method(version_id):
732
new_content = self.factory.parse_fulltext(data, version_id)
733
if parent is not None:
734
reference_content = self._get_content(parent)
735
old_texts = reference_content.text()
738
new_texts = new_content.text()
739
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
741
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
743
delta = self.factory.parse_line_delta(data, version_id)
744
return parent, sha1, noeol, delta
746
def get_format_signature(self):
747
"""See VersionedFile.get_format_signature()."""
748
if self.factory.annotated:
749
annotated_part = "annotated"
751
annotated_part = "plain"
752
return "knit-%s" % (annotated_part,)
754
def get_graph_with_ghosts(self):
755
"""See VersionedFile.get_graph_with_ghosts()."""
756
graph_items = self._index.get_graph()
757
return dict(graph_items)
759
def get_sha1(self, version_id):
760
return self.get_sha1s([version_id])[0]
762
def get_sha1s(self, version_ids):
763
"""See VersionedFile.get_sha1()."""
764
record_map = self._get_record_map(version_ids)
765
# record entry 2 is the 'digest'.
766
return [record_map[v][2] for v in version_ids]
770
"""See VersionedFile.get_suffixes()."""
771
return [DATA_SUFFIX, INDEX_SUFFIX]
773
def has_ghost(self, version_id):
774
"""True if there is a ghost reference in the file to version_id."""
776
if self.has_version(version_id):
778
# optimisable if needed by memoising the _ghosts set.
779
items = self._index.get_graph()
780
for node, parents in items:
781
for parent in parents:
782
if parent not in self._index._cache:
783
if parent == version_id:
787
def insert_data_stream(self, (format, data_list, reader_callable)):
788
"""Insert knit records from a data stream into this knit.
790
If a version in the stream is already present in this knit, it will not
791
be inserted a second time. It will be checked for consistency with the
792
stored version however, and may cause a KnitCorrupt error to be raised
793
if the data in the stream disagrees with the already stored data.
795
:seealso: get_data_stream
797
if format != self.get_format_signature():
798
if 'knit' in debug.debug_flags:
800
'incompatible format signature inserting to %r', self)
801
source = self._knit_from_datastream(
802
(format, data_list, reader_callable))
806
for version_id, options, length, parents in data_list:
807
if self.has_version(version_id):
808
# First check: the list of parents.
809
my_parents = self.get_parents_with_ghosts(version_id)
810
if tuple(my_parents) != tuple(parents):
811
# XXX: KnitCorrupt is not quite the right exception here.
814
'parents list %r from data stream does not match '
815
'already recorded parents %r for %s'
816
% (parents, my_parents, version_id))
818
# Also check the SHA-1 of the fulltext this content will
820
raw_data = reader_callable(length)
821
my_fulltext_sha1 = self.get_sha1(version_id)
822
df, rec = self._data._parse_record_header(version_id, raw_data)
823
stream_fulltext_sha1 = rec[3]
824
if my_fulltext_sha1 != stream_fulltext_sha1:
825
# Actually, we don't know if it's this knit that's corrupt,
826
# or the data stream we're trying to insert.
828
self.filename, 'sha-1 does not match %s' % version_id)
830
if 'line-delta' in options:
831
# Make sure that this knit record is actually useful: a
832
# line-delta is no use unless we have its parent.
833
# Fetching from a broken repository with this problem
834
# shouldn't break the target repository.
836
# See https://bugs.launchpad.net/bzr/+bug/164443
837
if not self._index.has_version(parents[0]):
840
'line-delta from stream '
843
'missing parent %s\n'
844
'Try running "bzr check" '
845
'on the source repository, and "bzr reconcile" '
847
(version_id, parents[0]))
848
self._add_raw_records(
849
[(version_id, options, parents, length)],
850
reader_callable(length))
852
def _knit_from_datastream(self, (format, data_list, reader_callable)):
853
"""Create a knit object from a data stream.
855
This method exists to allow conversion of data streams that do not
856
match the signature of this knit. Generally it will be slower and use
857
more memory to use this method to insert data, but it will work.
859
:seealso: get_data_stream for details on datastreams.
860
:return: A knit versioned file which can be used to join the datastream
863
if format == "knit-plain":
864
factory = KnitPlainFactory()
865
elif format == "knit-annotated":
866
factory = KnitAnnotateFactory()
868
raise errors.KnitDataStreamUnknown(format)
869
index = _StreamIndex(data_list, self._index)
870
access = _StreamAccess(reader_callable, index, self, factory)
871
return KnitVersionedFile(self.filename, self.transport,
872
factory=factory, index=index, access_method=access)
875
"""See VersionedFile.versions."""
876
if 'evil' in debug.debug_flags:
877
trace.mutter_callsite(2, "versions scales with size of history")
878
return self._index.get_versions()
880
def has_version(self, version_id):
881
"""See VersionedFile.has_version."""
882
if 'evil' in debug.debug_flags:
883
trace.mutter_callsite(2, "has_version is a LBYL scenario")
884
return self._index.has_version(version_id)
886
__contains__ = has_version
888
def _merge_annotations(self, content, parents, parent_texts={},
889
delta=None, annotated=None,
890
left_matching_blocks=None):
891
"""Merge annotations for content. This is done by comparing
892
the annotations based on changed to the text.
894
if left_matching_blocks is not None:
895
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
899
for parent_id in parents:
900
merge_content = self._get_content(parent_id, parent_texts)
901
if (parent_id == parents[0] and delta_seq is not None):
904
seq = patiencediff.PatienceSequenceMatcher(
905
None, merge_content.text(), content.text())
906
for i, j, n in seq.get_matching_blocks():
909
# this appears to copy (origin, text) pairs across to the
910
# new content for any line that matches the last-checked
912
content._lines[j:j+n] = merge_content._lines[i:i+n]
914
if delta_seq is None:
915
reference_content = self._get_content(parents[0], parent_texts)
916
new_texts = content.text()
917
old_texts = reference_content.text()
918
delta_seq = patiencediff.PatienceSequenceMatcher(
919
None, old_texts, new_texts)
920
return self._make_line_delta(delta_seq, content)
922
def _make_line_delta(self, delta_seq, new_content):
923
"""Generate a line delta from delta_seq and new_content."""
925
for op in delta_seq.get_opcodes():
928
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
931
def _get_components_positions(self, version_ids):
932
"""Produce a map of position data for the components of versions.
934
This data is intended to be used for retrieving the knit records.
936
A dict of version_id to (record_details, index_memo, next, parents) is
938
method is the way referenced data should be applied.
939
index_memo is the handle to pass to the data access to actually get the
941
next is the build-parent of the version, or None for fulltexts.
942
parents is the version_ids of the parents of this version
945
pending_components = version_ids
946
while pending_components:
947
build_details = self._index.get_build_details(pending_components)
948
pending_components = set()
949
for version_id, details in build_details.items():
950
(index_memo, compression_parent, parents, record_details) = details
951
method = record_details[0]
952
if compression_parent is not None:
953
pending_components.add(compression_parent)
954
component_data[version_id] = (record_details, index_memo,
956
return component_data
958
def _get_content(self, version_id, parent_texts={}):
959
"""Returns a content object that makes up the specified
961
cached_version = parent_texts.get(version_id, None)
962
if cached_version is not None:
963
if not self.has_version(version_id):
964
raise RevisionNotPresent(version_id, self.filename)
965
return cached_version
967
text_map, contents_map = self._get_content_maps([version_id])
968
return contents_map[version_id]
970
def _check_versions_present(self, version_ids):
971
"""Check that all specified versions are present."""
972
self._index.check_versions_present(version_ids)
974
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
975
nostore_sha, random_id, check_content):
976
"""See VersionedFile.add_lines_with_ghosts()."""
977
self._check_add(version_id, lines, random_id, check_content)
978
return self._add(version_id, lines, parents, self.delta,
979
parent_texts, None, nostore_sha, random_id)
981
def _add_lines(self, version_id, parents, lines, parent_texts,
982
left_matching_blocks, nostore_sha, random_id, check_content):
983
"""See VersionedFile.add_lines."""
984
self._check_add(version_id, lines, random_id, check_content)
985
self._check_versions_present(parents)
986
return self._add(version_id, lines[:], parents, self.delta,
987
parent_texts, left_matching_blocks, nostore_sha, random_id)
989
def _check_add(self, version_id, lines, random_id, check_content):
990
"""check that version_id and lines are safe to add."""
991
if contains_whitespace(version_id):
992
raise InvalidRevisionId(version_id, self.filename)
993
self.check_not_reserved_id(version_id)
994
# Technically this could be avoided if we are happy to allow duplicate
995
# id insertion when other things than bzr core insert texts, but it
996
# seems useful for folk using the knit api directly to have some safety
997
# blanket that we can disable.
998
if not random_id and self.has_version(version_id):
999
raise RevisionAlreadyPresent(version_id, self.filename)
1001
self._check_lines_not_unicode(lines)
1002
self._check_lines_are_lines(lines)
1004
def _add(self, version_id, lines, parents, delta, parent_texts,
1005
left_matching_blocks, nostore_sha, random_id):
1006
"""Add a set of lines on top of version specified by parents.
1008
If delta is true, compress the text as a line-delta against
1011
Any versions not present will be converted into ghosts.
1013
# first thing, if the content is something we don't need to store, find
1015
line_bytes = ''.join(lines)
1016
digest = sha_string(line_bytes)
1017
if nostore_sha == digest:
1018
raise errors.ExistingContent
1020
present_parents = []
1021
if parent_texts is None:
1023
for parent in parents:
1024
if self.has_version(parent):
1025
present_parents.append(parent)
1027
# can only compress against the left most present parent.
1029
(len(present_parents) == 0 or
1030
present_parents[0] != parents[0])):
1033
text_length = len(line_bytes)
1036
if lines[-1][-1] != '\n':
1037
# copy the contents of lines.
1039
options.append('no-eol')
1040
lines[-1] = lines[-1] + '\n'
1044
# To speed the extract of texts the delta chain is limited
1045
# to a fixed number of deltas. This should minimize both
1046
# I/O and the time spend applying deltas.
1047
delta = self._check_should_delta(present_parents)
1049
assert isinstance(version_id, str)
1050
content = self.factory.make(lines, version_id)
1051
if delta or (self.factory.annotated and len(present_parents) > 0):
1052
# Merge annotations from parent texts if needed.
1053
delta_hunks = self._merge_annotations(content, present_parents,
1054
parent_texts, delta, self.factory.annotated,
1055
left_matching_blocks)
1058
options.append('line-delta')
1059
store_lines = self.factory.lower_line_delta(delta_hunks)
1060
size, bytes = self._data._record_to_data(version_id, digest,
1063
options.append('fulltext')
1064
# isinstance is slower and we have no hierarchy.
1065
if self.factory.__class__ == KnitPlainFactory:
1066
# Use the already joined bytes saving iteration time in
1068
size, bytes = self._data._record_to_data(version_id, digest,
1069
lines, [line_bytes])
1071
# get mixed annotation + content and feed it into the
1073
store_lines = self.factory.lower_fulltext(content)
1074
size, bytes = self._data._record_to_data(version_id, digest,
1077
access_memo = self._data.add_raw_records([size], bytes)[0]
1078
self._index.add_versions(
1079
((version_id, options, access_memo, parents),),
1080
random_id=random_id)
1081
return digest, text_length, content
1083
def check(self, progress_bar=None):
1084
"""See VersionedFile.check()."""
1086
def _clone_text(self, new_version_id, old_version_id, parents):
1087
"""See VersionedFile.clone_text()."""
1088
# FIXME RBC 20060228 make fast by only inserting an index with null
1090
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
1092
def get_lines(self, version_id):
1093
"""See VersionedFile.get_lines()."""
1094
return self.get_line_list([version_id])[0]
1096
def _get_record_map(self, version_ids):
1097
"""Produce a dictionary of knit records.
1099
:return: {version_id:(record, record_details, digest, next)}
1101
data returned from read_records
1103
opaque information to pass to parse_record
1105
SHA1 digest of the full text after all steps are done
1107
build-parent of the version, i.e. the leftmost ancestor.
1108
Will be None if the record is not a delta.
1110
position_map = self._get_components_positions(version_ids)
1111
# c = component_id, r = record_details, i_m = index_memo, n = next
1112
records = [(c, i_m) for c, (r, i_m, n)
1113
in position_map.iteritems()]
1115
for component_id, record, digest in \
1116
self._data.read_records_iter(records):
1117
(record_details, index_memo, next) = position_map[component_id]
1118
record_map[component_id] = record, record_details, digest, next
1122
def get_text(self, version_id):
1123
"""See VersionedFile.get_text"""
1124
return self.get_texts([version_id])[0]
1126
def get_texts(self, version_ids):
1127
return [''.join(l) for l in self.get_line_list(version_ids)]
1129
def get_line_list(self, version_ids):
1130
"""Return the texts of listed versions as a list of strings."""
1131
for version_id in version_ids:
1132
self.check_not_reserved_id(version_id)
1133
text_map, content_map = self._get_content_maps(version_ids)
1134
return [text_map[v] for v in version_ids]
1136
_get_lf_split_line_list = get_line_list
1138
def _get_content_maps(self, version_ids):
1139
"""Produce maps of text and KnitContents
1141
:return: (text_map, content_map) where text_map contains the texts for
1142
the requested versions and content_map contains the KnitContents.
1143
Both dicts take version_ids as their keys.
1145
# FUTURE: This function could be improved for the 'extract many' case
1146
# by tracking each component and only doing the copy when the number of
1147
# children than need to apply delta's to it is > 1 or it is part of the
1149
version_ids = list(version_ids)
1150
multiple_versions = len(version_ids) != 1
1151
record_map = self._get_record_map(version_ids)
1156
for version_id in version_ids:
1159
while cursor is not None:
1160
record, record_details, digest, next = record_map[cursor]
1161
components.append((cursor, record, record_details, digest))
1162
if cursor in content_map:
1167
for (component_id, record, record_details,
1168
digest) in reversed(components):
1169
if component_id in content_map:
1170
content = content_map[component_id]
1172
content, delta = self.factory.parse_record(version_id,
1173
record, record_details, content,
1174
copy_base_content=multiple_versions)
1175
if multiple_versions:
1176
content_map[component_id] = content
1178
content.cleanup_eol(copy_on_mutate=multiple_versions)
1179
final_content[version_id] = content
1181
# digest here is the digest from the last applied component.
1182
text = content.text()
1183
actual_sha = sha_strings(text)
1184
if actual_sha != digest:
1185
raise KnitCorrupt(self.filename,
1187
'\n of reconstructed text does not match'
1189
'\n for version %s' %
1190
(actual_sha, digest, version_id))
1191
text_map[version_id] = text
1192
return text_map, final_content
1194
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1196
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1197
if version_ids is None:
1198
version_ids = self.versions()
1200
pb = progress.DummyProgress()
1201
# we don't care about inclusions, the caller cares.
1202
# but we need to setup a list of records to visit.
1203
# we need version_id, position, length
1204
version_id_records = []
1205
requested_versions = set(version_ids)
1206
# filter for available versions
1207
for version_id in requested_versions:
1208
if not self.has_version(version_id):
1209
raise RevisionNotPresent(version_id, self.filename)
1210
# get a in-component-order queue:
1211
for version_id in self.versions():
1212
if version_id in requested_versions:
1213
index_memo = self._index.get_position(version_id)
1214
version_id_records.append((version_id, index_memo))
1216
total = len(version_id_records)
1217
for version_idx, (version_id, data, sha_value) in \
1218
enumerate(self._data.read_records_iter(version_id_records)):
1219
pb.update('Walking content.', version_idx, total)
1220
method = self._index.get_method(version_id)
1222
assert method in ('fulltext', 'line-delta')
1223
if method == 'fulltext':
1224
line_iterator = self.factory.get_fulltext_content(data)
1226
line_iterator = self.factory.get_linedelta_content(data)
1227
# XXX: It might be more efficient to yield (version_id,
1228
# line_iterator) in the future. However for now, this is a simpler
1229
# change to integrate into the rest of the codebase. RBC 20071110
1230
for line in line_iterator:
1231
yield line, version_id
1233
pb.update('Walking content.', total, total)
1235
def iter_parents(self, version_ids):
1236
"""Iterate through the parents for many version ids.
1238
:param version_ids: An iterable yielding version_ids.
1239
:return: An iterator that yields (version_id, parents). Requested
1240
version_ids not present in the versioned file are simply skipped.
1241
The order is undefined, allowing for different optimisations in
1242
the underlying implementation.
1244
return self._index.iter_parents(version_ids)
1246
def num_versions(self):
1247
"""See VersionedFile.num_versions()."""
1248
return self._index.num_versions()
1250
__len__ = num_versions
1252
def annotate_iter(self, version_id):
1253
"""See VersionedFile.annotate_iter."""
1254
return self.factory.annotate_iter(self, version_id)
1256
def get_parents(self, version_id):
1257
"""See VersionedFile.get_parents."""
1260
# 52554 calls in 1264 872 internal down from 3674
1262
return self._index.get_parents(version_id)
1264
raise RevisionNotPresent(version_id, self.filename)
1266
def get_parents_with_ghosts(self, version_id):
1267
"""See VersionedFile.get_parents."""
1269
return self._index.get_parents_with_ghosts(version_id)
1271
raise RevisionNotPresent(version_id, self.filename)
1273
def get_ancestry(self, versions, topo_sorted=True):
1274
"""See VersionedFile.get_ancestry."""
1275
if isinstance(versions, basestring):
1276
versions = [versions]
1279
return self._index.get_ancestry(versions, topo_sorted)
1281
def get_ancestry_with_ghosts(self, versions):
1282
"""See VersionedFile.get_ancestry_with_ghosts."""
1283
if isinstance(versions, basestring):
1284
versions = [versions]
1287
return self._index.get_ancestry_with_ghosts(versions)
1289
def plan_merge(self, ver_a, ver_b):
1290
"""See VersionedFile.plan_merge."""
1291
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1292
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1293
annotated_a = self.annotate(ver_a)
1294
annotated_b = self.annotate(ver_b)
1295
return merge._plan_annotate_merge(annotated_a, annotated_b,
1296
ancestors_a, ancestors_b)
1299
class _KnitComponentFile(object):
1300
"""One of the files used to implement a knit database"""
1302
def __init__(self, transport, filename, mode, file_mode=None,
1303
create_parent_dir=False, dir_mode=None):
1304
self._transport = transport
1305
self._filename = filename
1307
self._file_mode = file_mode
1308
self._dir_mode = dir_mode
1309
self._create_parent_dir = create_parent_dir
1310
self._need_to_create = False
1312
def _full_path(self):
1313
"""Return the full path to this file."""
1314
return self._transport.base + self._filename
1316
def check_header(self, fp):
1317
line = fp.readline()
1319
# An empty file can actually be treated as though the file doesn't
1321
raise errors.NoSuchFile(self._full_path())
1322
if line != self.HEADER:
1323
raise KnitHeaderError(badline=line,
1324
filename=self._transport.abspath(self._filename))
1327
return '%s(%s)' % (self.__class__.__name__, self._filename)
1330
class _KnitIndex(_KnitComponentFile):
1331
"""Manages knit index file.
1333
The index is already kept in memory and read on startup, to enable
1334
fast lookups of revision information. The cursor of the index
1335
file is always pointing to the end, making it easy to append
1338
_cache is a cache for fast mapping from version id to a Index
1341
_history is a cache for fast mapping from indexes to version ids.
1343
The index data format is dictionary compressed when it comes to
1344
parent references; a index entry may only have parents that with a
1345
lover index number. As a result, the index is topological sorted.
1347
Duplicate entries may be written to the index for a single version id
1348
if this is done then the latter one completely replaces the former:
1349
this allows updates to correct version and parent information.
1350
Note that the two entries may share the delta, and that successive
1351
annotations and references MUST point to the first entry.
1353
The index file on disc contains a header, followed by one line per knit
1354
record. The same revision can be present in an index file more than once.
1355
The first occurrence gets assigned a sequence number starting from 0.
1357
The format of a single line is
1358
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1359
REVISION_ID is a utf8-encoded revision id
1360
FLAGS is a comma separated list of flags about the record. Values include
1361
no-eol, line-delta, fulltext.
1362
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1363
that the the compressed data starts at.
1364
LENGTH is the ascii representation of the length of the data file.
1365
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1367
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1368
revision id already in the knit that is a parent of REVISION_ID.
1369
The ' :' marker is the end of record marker.
1372
when a write is interrupted to the index file, it will result in a line
1373
that does not end in ' :'. If the ' :' is not present at the end of a line,
1374
or at the end of the file, then the record that is missing it will be
1375
ignored by the parser.
1377
When writing new records to the index file, the data is preceded by '\n'
1378
to ensure that records always start on new lines even if the last write was
1379
interrupted. As a result its normal for the last line in the index to be
1380
missing a trailing newline. One can be added with no harmful effects.
1383
HEADER = "# bzr knit index 8\n"
1385
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1386
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1388
def _cache_version(self, version_id, options, pos, size, parents):
1389
"""Cache a version record in the history array and index cache.
1391
This is inlined into _load_data for performance. KEEP IN SYNC.
1392
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1395
# only want the _history index to reference the 1st index entry
1397
if version_id not in self._cache:
1398
index = len(self._history)
1399
self._history.append(version_id)
1401
index = self._cache[version_id][5]
1402
self._cache[version_id] = (version_id,
1409
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1410
create_parent_dir=False, delay_create=False, dir_mode=None):
1411
_KnitComponentFile.__init__(self, transport, filename, mode,
1412
file_mode=file_mode,
1413
create_parent_dir=create_parent_dir,
1416
# position in _history is the 'official' index for a revision
1417
# but the values may have come from a newer entry.
1418
# so - wc -l of a knit index is != the number of unique names
1422
fp = self._transport.get(self._filename)
1424
# _load_data may raise NoSuchFile if the target knit is
1426
_load_data(self, fp)
1430
if mode != 'w' or not create:
1433
self._need_to_create = True
1435
self._transport.put_bytes_non_atomic(
1436
self._filename, self.HEADER, mode=self._file_mode)
1438
def get_graph(self):
1439
"""Return a list of the node:parents lists from this knit index."""
1440
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1442
def get_ancestry(self, versions, topo_sorted=True):
1443
"""See VersionedFile.get_ancestry."""
1444
# get a graph of all the mentioned versions:
1446
pending = set(versions)
1449
version = pending.pop()
1452
parents = [p for p in cache[version][4] if p in cache]
1454
raise RevisionNotPresent(version, self._filename)
1455
# if not completed and not a ghost
1456
pending.update([p for p in parents if p not in graph])
1457
graph[version] = parents
1460
return topo_sort(graph.items())
1462
def get_ancestry_with_ghosts(self, versions):
1463
"""See VersionedFile.get_ancestry_with_ghosts."""
1464
# get a graph of all the mentioned versions:
1465
self.check_versions_present(versions)
1468
pending = set(versions)
1470
version = pending.pop()
1472
parents = cache[version][4]
1478
pending.update([p for p in parents if p not in graph])
1479
graph[version] = parents
1480
return topo_sort(graph.items())
1482
def get_build_details(self, version_ids):
1483
"""Get the method, index_memo and compression parent for version_ids.
1485
:param version_ids: An iterable of version_ids.
1486
:return: A dict of version_id:(index_memo, compression_parent,
1487
parents, record_details).
1489
opaque structure to pass to read_records to extract the raw
1492
Content that this record is built upon, may be None
1494
Logical parents of this node
1496
extra information about the content which needs to be passed to
1497
Factory.parse_record
1500
for version_id in version_ids:
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 iter_parents(self, version_ids):
1514
"""Iterate through the parents for many version ids.
1516
:param version_ids: An iterable yielding version_ids.
1517
:return: An iterator that yields (version_id, parents). Requested
1518
version_ids not present in the versioned file are simply skipped.
1519
The order is undefined, allowing for different optimisations in
1520
the underlying implementation.
1522
for version_id in version_ids:
1524
yield version_id, tuple(self.get_parents(version_id))
1528
def num_versions(self):
1529
return len(self._history)
1531
__len__ = num_versions
1533
def get_versions(self):
1534
"""Get all the versions in the file. not topologically sorted."""
1535
return self._history
1537
def _version_list_to_index(self, versions):
1540
for version in versions:
1541
if version in cache:
1542
# -- inlined lookup() --
1543
result_list.append(str(cache[version][5]))
1544
# -- end lookup () --
1546
result_list.append('.' + version)
1547
return ' '.join(result_list)
1549
def add_version(self, version_id, options, index_memo, parents):
1550
"""Add a version record to the index."""
1551
self.add_versions(((version_id, options, index_memo, parents),))
1553
def add_versions(self, versions, random_id=False):
1554
"""Add multiple versions to the index.
1556
:param versions: a list of tuples:
1557
(version_id, options, pos, size, parents).
1558
:param random_id: If True the ids being added were randomly generated
1559
and no check for existence will be performed.
1562
orig_history = self._history[:]
1563
orig_cache = self._cache.copy()
1566
for version_id, options, (index, pos, size), parents in versions:
1567
line = "\n%s %s %s %s %s :" % (version_id,
1571
self._version_list_to_index(parents))
1572
assert isinstance(line, str), \
1573
'content must be utf-8 encoded: %r' % (line,)
1575
self._cache_version(version_id, options, pos, size, parents)
1576
if not self._need_to_create:
1577
self._transport.append_bytes(self._filename, ''.join(lines))
1580
sio.write(self.HEADER)
1581
sio.writelines(lines)
1583
self._transport.put_file_non_atomic(self._filename, sio,
1584
create_parent_dir=self._create_parent_dir,
1585
mode=self._file_mode,
1586
dir_mode=self._dir_mode)
1587
self._need_to_create = False
1589
# If any problems happen, restore the original values and re-raise
1590
self._history = orig_history
1591
self._cache = orig_cache
1594
def has_version(self, version_id):
1595
"""True if the version is in the index."""
1596
return version_id in self._cache
1598
def get_position(self, version_id):
1599
"""Return details needed to access the version.
1601
.kndx indices do not support split-out data, so return None for the
1604
:return: a tuple (None, data position, size) to hand to the access
1605
logic to get the record.
1607
entry = self._cache[version_id]
1608
return None, entry[2], entry[3]
1610
def get_method(self, version_id):
1611
"""Return compression method of specified version."""
1613
options = self._cache[version_id][1]
1615
raise RevisionNotPresent(version_id, self._filename)
1616
if 'fulltext' in options:
1619
if 'line-delta' not in options:
1620
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1623
def get_options(self, version_id):
1624
"""Return a list representing options.
1628
return self._cache[version_id][1]
1630
def get_parents(self, version_id):
1631
"""Return parents of specified version ignoring ghosts."""
1632
return [parent for parent in self._cache[version_id][4]
1633
if parent in self._cache]
1635
def get_parents_with_ghosts(self, version_id):
1636
"""Return parents of specified version with ghosts."""
1637
return self._cache[version_id][4]
1639
def check_versions_present(self, version_ids):
1640
"""Check that all specified versions are present."""
1642
for version_id in version_ids:
1643
if version_id not in cache:
1644
raise RevisionNotPresent(version_id, self._filename)
1647
class KnitGraphIndex(object):
1648
"""A knit index that builds on GraphIndex."""
1650
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1651
"""Construct a KnitGraphIndex on a graph_index.
1653
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1654
:param deltas: Allow delta-compressed records.
1655
:param add_callback: If not None, allow additions to the index and call
1656
this callback with a list of added GraphIndex nodes:
1657
[(node, value, node_refs), ...]
1658
:param parents: If True, record knits parents, if not do not record
1661
self._graph_index = graph_index
1662
self._deltas = deltas
1663
self._add_callback = add_callback
1664
self._parents = parents
1665
if deltas and not parents:
1666
raise KnitCorrupt(self, "Cannot do delta compression without "
1669
def _get_entries(self, keys, check_present=False):
1670
"""Get the entries for keys.
1672
:param keys: An iterable of index keys, - 1-tuples.
1677
for node in self._graph_index.iter_entries(keys):
1679
found_keys.add(node[1])
1681
# adapt parentless index to the rest of the code.
1682
for node in self._graph_index.iter_entries(keys):
1683
yield node[0], node[1], node[2], ()
1684
found_keys.add(node[1])
1686
missing_keys = keys.difference(found_keys)
1688
raise RevisionNotPresent(missing_keys.pop(), self)
1690
def _present_keys(self, version_ids):
1692
node[1] for node in self._get_entries(version_ids)])
1694
def _parentless_ancestry(self, versions):
1695
"""Honour the get_ancestry API for parentless knit indices."""
1696
wanted_keys = self._version_ids_to_keys(versions)
1697
present_keys = self._present_keys(wanted_keys)
1698
missing = set(wanted_keys).difference(present_keys)
1700
raise RevisionNotPresent(missing.pop(), self)
1701
return list(self._keys_to_version_ids(present_keys))
1703
def get_ancestry(self, versions, topo_sorted=True):
1704
"""See VersionedFile.get_ancestry."""
1705
if not self._parents:
1706
return self._parentless_ancestry(versions)
1707
# XXX: This will do len(history) index calls - perhaps
1708
# it should be altered to be a index core feature?
1709
# get a graph of all the mentioned versions:
1712
versions = self._version_ids_to_keys(versions)
1713
pending = set(versions)
1715
# get all pending nodes
1716
this_iteration = pending
1717
new_nodes = self._get_entries(this_iteration)
1720
for (index, key, value, node_refs) in new_nodes:
1721
# dont ask for ghosties - otherwise
1722
# we we can end up looping with pending
1723
# being entirely ghosted.
1724
graph[key] = [parent for parent in node_refs[0]
1725
if parent not in ghosts]
1727
for parent in graph[key]:
1728
# dont examine known nodes again
1733
ghosts.update(this_iteration.difference(found))
1734
if versions.difference(graph):
1735
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1737
result_keys = topo_sort(graph.items())
1739
result_keys = graph.iterkeys()
1740
return [key[0] for key in result_keys]
1742
def get_ancestry_with_ghosts(self, versions):
1743
"""See VersionedFile.get_ancestry."""
1744
if not self._parents:
1745
return self._parentless_ancestry(versions)
1746
# XXX: This will do len(history) index calls - perhaps
1747
# it should be altered to be a index core feature?
1748
# get a graph of all the mentioned versions:
1750
versions = self._version_ids_to_keys(versions)
1751
pending = set(versions)
1753
# get all pending nodes
1754
this_iteration = pending
1755
new_nodes = self._get_entries(this_iteration)
1757
for (index, key, value, node_refs) in new_nodes:
1758
graph[key] = node_refs[0]
1760
for parent in graph[key]:
1761
# dont examine known nodes again
1765
missing_versions = this_iteration.difference(graph)
1766
missing_needed = versions.intersection(missing_versions)
1768
raise RevisionNotPresent(missing_needed.pop(), self)
1769
for missing_version in missing_versions:
1770
# add a key, no parents
1771
graph[missing_version] = []
1772
pending.discard(missing_version) # don't look for it
1773
result_keys = topo_sort(graph.items())
1774
return [key[0] for key in result_keys]
1776
def get_build_details(self, version_ids):
1777
"""Get the method, index_memo and compression parent for version_ids.
1779
:param version_ids: An iterable of version_ids.
1780
:return: A dict of version_id:(index_memo, compression_parent,
1781
parents, record_details).
1783
opaque structure to pass to read_records to extract the raw
1786
Content that this record is built upon, may be None
1788
Logical parents of this node
1790
extra information about the content which needs to be passed to
1791
Factory.parse_record
1794
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
1795
for entry in entries:
1796
version_id = self._keys_to_version_ids((entry[1],))[0]
1797
parents = self._keys_to_version_ids(entry[3][0])
1798
if not self._deltas:
1799
compression_parent = None
1801
compression_parent_key = self._compression_parent(entry)
1802
if compression_parent_key:
1803
compression_parent = self._keys_to_version_ids(
1804
(compression_parent_key,))[0]
1806
compression_parent = None
1807
noeol = (entry[2][0] == 'N')
1808
if compression_parent:
1809
method = 'line-delta'
1812
result[version_id] = (self._node_to_position(entry),
1813
compression_parent, parents,
1817
def _compression_parent(self, an_entry):
1818
# return the key that an_entry is compressed against, or None
1819
# Grab the second parent list (as deltas implies parents currently)
1820
compression_parents = an_entry[3][1]
1821
if not compression_parents:
1823
assert len(compression_parents) == 1
1824
return compression_parents[0]
1826
def _get_method(self, node):
1827
if not self._deltas:
1829
if self._compression_parent(node):
1834
def get_graph(self):
1835
"""Return a list of the node:parents lists from this knit index."""
1836
if not self._parents:
1837
return [(key, ()) for key in self.get_versions()]
1839
for index, key, value, refs in self._graph_index.iter_all_entries():
1840
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1843
def iter_parents(self, version_ids):
1844
"""Iterate through the parents for many version ids.
1846
:param version_ids: An iterable yielding version_ids.
1847
:return: An iterator that yields (version_id, parents). Requested
1848
version_ids not present in the versioned file are simply skipped.
1849
The order is undefined, allowing for different optimisations in
1850
the underlying implementation.
1853
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1855
present_parents = set()
1856
for node in all_nodes:
1857
all_parents.update(node[3][0])
1858
# any node we are querying must be present
1859
present_parents.add(node[1])
1860
unknown_parents = all_parents.difference(present_parents)
1861
present_parents.update(self._present_keys(unknown_parents))
1862
for node in all_nodes:
1864
for parent in node[3][0]:
1865
if parent in present_parents:
1866
parents.append(parent[0])
1867
yield node[1][0], tuple(parents)
1869
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1870
yield node[1][0], ()
1872
def num_versions(self):
1873
return len(list(self._graph_index.iter_all_entries()))
1875
__len__ = num_versions
1877
def get_versions(self):
1878
"""Get all the versions in the file. not topologically sorted."""
1879
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1881
def has_version(self, version_id):
1882
"""True if the version is in the index."""
1883
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1885
def _keys_to_version_ids(self, keys):
1886
return tuple(key[0] for key in keys)
1888
def get_position(self, version_id):
1889
"""Return details needed to access the version.
1891
:return: a tuple (index, data position, size) to hand to the access
1892
logic to get the record.
1894
node = self._get_node(version_id)
1895
return self._node_to_position(node)
1897
def _node_to_position(self, node):
1898
"""Convert an index value to position details."""
1899
bits = node[2][1:].split(' ')
1900
return node[0], int(bits[0]), int(bits[1])
1902
def get_method(self, version_id):
1903
"""Return compression method of specified version."""
1904
return self._get_method(self._get_node(version_id))
1906
def _get_node(self, version_id):
1908
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1910
raise RevisionNotPresent(version_id, self)
1912
def get_options(self, version_id):
1913
"""Return a list representing options.
1917
node = self._get_node(version_id)
1918
options = [self._get_method(node)]
1919
if node[2][0] == 'N':
1920
options.append('no-eol')
1923
def get_parents(self, version_id):
1924
"""Return parents of specified version ignoring ghosts."""
1925
parents = list(self.iter_parents([version_id]))
1928
raise errors.RevisionNotPresent(version_id, self)
1929
return parents[0][1]
1931
def get_parents_with_ghosts(self, version_id):
1932
"""Return parents of specified version with ghosts."""
1933
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1934
check_present=True))
1935
if not self._parents:
1937
return self._keys_to_version_ids(nodes[0][3][0])
1939
def check_versions_present(self, version_ids):
1940
"""Check that all specified versions are present."""
1941
keys = self._version_ids_to_keys(version_ids)
1942
present = self._present_keys(keys)
1943
missing = keys.difference(present)
1945
raise RevisionNotPresent(missing.pop(), self)
1947
def add_version(self, version_id, options, access_memo, parents):
1948
"""Add a version record to the index."""
1949
return self.add_versions(((version_id, options, access_memo, parents),))
1951
def add_versions(self, versions, random_id=False):
1952
"""Add multiple versions to the index.
1954
This function does not insert data into the Immutable GraphIndex
1955
backing the KnitGraphIndex, instead it prepares data for insertion by
1956
the caller and checks that it is safe to insert then calls
1957
self._add_callback with the prepared GraphIndex nodes.
1959
:param versions: a list of tuples:
1960
(version_id, options, pos, size, parents).
1961
:param random_id: If True the ids being added were randomly generated
1962
and no check for existence will be performed.
1964
if not self._add_callback:
1965
raise errors.ReadOnlyError(self)
1966
# we hope there are no repositories with inconsistent parentage
1971
for (version_id, options, access_memo, parents) in versions:
1972
index, pos, size = access_memo
1973
key = (version_id, )
1974
parents = tuple((parent, ) for parent in parents)
1975
if 'no-eol' in options:
1979
value += "%d %d" % (pos, size)
1980
if not self._deltas:
1981
if 'line-delta' in options:
1982
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1985
if 'line-delta' in options:
1986
node_refs = (parents, (parents[0],))
1988
node_refs = (parents, ())
1990
node_refs = (parents, )
1993
raise KnitCorrupt(self, "attempt to add node with parents "
1994
"in parentless index.")
1996
keys[key] = (value, node_refs)
1998
present_nodes = self._get_entries(keys)
1999
for (index, key, value, node_refs) in present_nodes:
2000
if (value, node_refs) != keys[key]:
2001
raise KnitCorrupt(self, "inconsistent details in add_versions"
2002
": %s %s" % ((value, node_refs), keys[key]))
2006
for key, (value, node_refs) in keys.iteritems():
2007
result.append((key, value, node_refs))
2009
for key, (value, node_refs) in keys.iteritems():
2010
result.append((key, value))
2011
self._add_callback(result)
2013
def _version_ids_to_keys(self, version_ids):
2014
return set((version_id, ) for version_id in version_ids)
2017
class _KnitAccess(object):
2018
"""Access to knit records in a .knit file."""
2020
def __init__(self, transport, filename, _file_mode, _dir_mode,
2021
_need_to_create, _create_parent_dir):
2022
"""Create a _KnitAccess for accessing and inserting data.
2024
:param transport: The transport the .knit is located on.
2025
:param filename: The filename of the .knit.
2027
self._transport = transport
2028
self._filename = filename
2029
self._file_mode = _file_mode
2030
self._dir_mode = _dir_mode
2031
self._need_to_create = _need_to_create
2032
self._create_parent_dir = _create_parent_dir
2034
def add_raw_records(self, sizes, raw_data):
2035
"""Add raw knit bytes to a storage area.
2037
The data is spooled to whereever the access method is storing data.
2039
:param sizes: An iterable containing the size of each raw data segment.
2040
:param raw_data: A bytestring containing the data.
2041
:return: A list of memos to retrieve the record later. Each memo is a
2042
tuple - (index, pos, length), where the index field is always None
2043
for the .knit access method.
2045
assert type(raw_data) == str, \
2046
'data must be plain bytes was %s' % type(raw_data)
2047
if not self._need_to_create:
2048
base = self._transport.append_bytes(self._filename, raw_data)
2050
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2051
create_parent_dir=self._create_parent_dir,
2052
mode=self._file_mode,
2053
dir_mode=self._dir_mode)
2054
self._need_to_create = False
2058
result.append((None, base, size))
2063
"""IFF this data access has its own storage area, initialise it.
2067
self._transport.put_bytes_non_atomic(self._filename, '',
2068
mode=self._file_mode)
2070
def open_file(self):
2071
"""IFF this data access can be represented as a single file, open it.
2073
For knits that are not mapped to a single file on disk this will
2076
:return: None or a file handle.
2079
return self._transport.get(self._filename)
2084
def get_raw_records(self, memos_for_retrieval):
2085
"""Get the raw bytes for a records.
2087
:param memos_for_retrieval: An iterable containing the (index, pos,
2088
length) memo for retrieving the bytes. The .knit method ignores
2089
the index as there is always only a single file.
2090
:return: An iterator over the bytes of the records.
2092
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2093
for pos, data in self._transport.readv(self._filename, read_vector):
2097
class _PackAccess(object):
2098
"""Access to knit records via a collection of packs."""
2100
def __init__(self, index_to_packs, writer=None):
2101
"""Create a _PackAccess object.
2103
:param index_to_packs: A dict mapping index objects to the transport
2104
and file names for obtaining data.
2105
:param writer: A tuple (pack.ContainerWriter, write_index) which
2106
contains the pack to write, and the index that reads from it will
2110
self.container_writer = writer[0]
2111
self.write_index = writer[1]
2113
self.container_writer = None
2114
self.write_index = None
2115
self.indices = index_to_packs
2117
def add_raw_records(self, sizes, raw_data):
2118
"""Add raw knit bytes to a storage area.
2120
The data is spooled to the container writer in one bytes-record per
2123
:param sizes: An iterable containing the size of each raw data segment.
2124
:param raw_data: A bytestring containing the data.
2125
:return: A list of memos to retrieve the record later. Each memo is a
2126
tuple - (index, pos, length), where the index field is the
2127
write_index object supplied to the PackAccess object.
2129
assert type(raw_data) == str, \
2130
'data must be plain bytes was %s' % type(raw_data)
2134
p_offset, p_length = self.container_writer.add_bytes_record(
2135
raw_data[offset:offset+size], [])
2137
result.append((self.write_index, p_offset, p_length))
2141
"""Pack based knits do not get individually created."""
2143
def get_raw_records(self, memos_for_retrieval):
2144
"""Get the raw bytes for a records.
2146
:param memos_for_retrieval: An iterable containing the (index, pos,
2147
length) memo for retrieving the bytes. The Pack access method
2148
looks up the pack to use for a given record in its index_to_pack
2150
:return: An iterator over the bytes of the records.
2152
# first pass, group into same-index requests
2154
current_index = None
2155
for (index, offset, length) in memos_for_retrieval:
2156
if current_index == index:
2157
current_list.append((offset, length))
2159
if current_index is not None:
2160
request_lists.append((current_index, current_list))
2161
current_index = index
2162
current_list = [(offset, length)]
2163
# handle the last entry
2164
if current_index is not None:
2165
request_lists.append((current_index, current_list))
2166
for index, offsets in request_lists:
2167
transport, path = self.indices[index]
2168
reader = pack.make_readv_reader(transport, path, offsets)
2169
for names, read_func in reader.iter_records():
2170
yield read_func(None)
2172
def open_file(self):
2173
"""Pack based knits have no single file."""
2176
def set_writer(self, writer, index, (transport, packname)):
2177
"""Set a writer to use for adding data."""
2178
if index is not None:
2179
self.indices[index] = (transport, packname)
2180
self.container_writer = writer
2181
self.write_index = index
2184
class _StreamAccess(object):
2185
"""A Knit Access object that provides data from a datastream.
2187
It also provides a fallback to present as unannotated data, annotated data
2188
from a *backing* access object.
2190
This is triggered by a index_memo which is pointing to a different index
2191
than this was constructed with, and is used to allow extracting full
2192
unannotated texts for insertion into annotated knits.
2195
def __init__(self, reader_callable, stream_index, backing_knit,
2197
"""Create a _StreamAccess object.
2199
:param reader_callable: The reader_callable from the datastream.
2200
This is called to buffer all the data immediately, for
2202
:param stream_index: The index the data stream this provides access to
2203
which will be present in native index_memo's.
2204
:param backing_knit: The knit object that will provide access to
2205
annotated texts which are not available in the stream, so as to
2206
create unannotated texts.
2207
:param orig_factory: The original content factory used to generate the
2208
stream. This is used for checking whether the thunk code for
2209
supporting _copy_texts will generate the correct form of data.
2211
self.data = reader_callable(None)
2212
self.stream_index = stream_index
2213
self.backing_knit = backing_knit
2214
self.orig_factory = orig_factory
2216
def get_raw_records(self, memos_for_retrieval):
2217
"""Get the raw bytes for a records.
2219
:param memos_for_retrieval: An iterable containing the (thunk_flag,
2220
index, start, end) memo for retrieving the bytes.
2221
:return: An iterator over the bytes of the records.
2223
# use a generator for memory friendliness
2224
for thunk_flag, version_id, start, end in memos_for_retrieval:
2225
if version_id is self.stream_index:
2226
yield self.data[start:end]
2228
# we have been asked to thunk. This thunking only occurs when
2229
# we are obtaining plain texts from an annotated backing knit
2230
# so that _copy_texts will work.
2231
# We could improve performance here by scanning for where we need
2232
# to do this and using get_line_list, then interleaving the output
2233
# as desired. However, for now, this is sufficient.
2234
if self.orig_factory.__class__ != KnitPlainFactory:
2235
raise errors.KnitCorrupt(
2236
self, 'Bad thunk request %r' % version_id)
2237
lines = self.backing_knit.get_lines(version_id)
2238
line_bytes = ''.join(lines)
2239
digest = sha_string(line_bytes)
2241
if lines[-1][-1] != '\n':
2242
lines[-1] = lines[-1] + '\n'
2244
orig_options = list(self.backing_knit._index.get_options(version_id))
2245
if 'fulltext' not in orig_options:
2246
if 'line-delta' not in orig_options:
2247
raise errors.KnitCorrupt(self,
2248
'Unknown compression method %r' % orig_options)
2249
orig_options.remove('line-delta')
2250
orig_options.append('fulltext')
2251
# We want plain data, because we expect to thunk only to allow text
2253
size, bytes = self.backing_knit._data._record_to_data(version_id,
2254
digest, lines, line_bytes)
2258
class _StreamIndex(object):
2259
"""A Knit Index object that uses the data map from a datastream."""
2261
def __init__(self, data_list, backing_index):
2262
"""Create a _StreamIndex object.
2264
:param data_list: The data_list from the datastream.
2265
:param backing_index: The index which will supply values for nodes
2266
referenced outside of this stream.
2268
self.data_list = data_list
2269
self.backing_index = backing_index
2270
self._by_version = {}
2272
for key, options, length, parents in data_list:
2273
self._by_version[key] = options, (pos, pos + length), parents
2276
def get_ancestry(self, versions, topo_sorted):
2277
"""Get an ancestry list for versions."""
2279
# Not needed for basic joins
2280
raise NotImplementedError(self.get_ancestry)
2281
# get a graph of all the mentioned versions:
2282
# Little ugly - basically copied from KnitIndex, but don't want to
2283
# accidentally incorporate too much of that index's code.
2285
pending = set(versions)
2286
cache = self._by_version
2288
version = pending.pop()
2291
parents = [p for p in cache[version][2] if p in cache]
2293
raise RevisionNotPresent(version, self)
2294
# if not completed and not a ghost
2295
pending.update([p for p in parents if p not in ancestry])
2296
ancestry.add(version)
2297
return list(ancestry)
2299
def get_build_details(self, version_ids):
2300
"""Get the method, index_memo and compression parent for version_ids.
2302
:param version_ids: An iterable of version_ids.
2303
:return: A dict of version_id:(index_memo, compression_parent,
2304
parents, record_details).
2306
opaque structure to pass to read_records to extract the raw
2309
Content that this record is built upon, may be None
2311
Logical parents of this node
2313
extra information about the content which needs to be passed to
2314
Factory.parse_record
2317
for version_id in version_ids:
2318
method = self.get_method(version_id)
2319
parent_ids = self.get_parents_with_ghosts(version_id)
2320
noeol = ('no-eol' in self.get_options(version_id))
2321
if method == 'fulltext':
2322
compression_parent = None
2324
compression_parent = parent_ids[0]
2325
index_memo = self.get_position(version_id)
2326
result[version_id] = (index_memo, compression_parent,
2327
parent_ids, (method, noeol))
2330
def get_method(self, version_id):
2331
"""Return compression method of specified version."""
2333
options = self._by_version[version_id][0]
2335
# Strictly speaking this should check in the backing knit, but
2336
# until we have a test to discriminate, this will do.
2337
return self.backing_index.get_method(version_id)
2338
if 'fulltext' in options:
2340
elif 'line-delta' in options:
2343
raise errors.KnitIndexUnknownMethod(self, options)
2345
def get_options(self, version_id):
2346
"""Return a list representing options.
2351
return self._by_version[version_id][0]
2353
return self.backing_index.get_options(version_id)
2355
def get_parents_with_ghosts(self, version_id):
2356
"""Return parents of specified version with ghosts."""
2358
return self._by_version[version_id][2]
2360
return self.backing_index.get_parents_with_ghosts(version_id)
2362
def get_position(self, version_id):
2363
"""Return details needed to access the version.
2365
_StreamAccess has the data as a big array, so we return slice
2366
coordinates into that (as index_memo's are opaque outside the
2367
index and matching access class).
2369
:return: a tuple (thunk_flag, index, start, end). If thunk_flag is
2370
False, index will be self, otherwise it will be a version id.
2373
start, end = self._by_version[version_id][1]
2374
return False, self, start, end
2376
# Signal to the access object to handle this from the backing knit.
2377
return (True, version_id, None, None)
2379
def get_versions(self):
2380
"""Get all the versions in the stream."""
2381
return self._by_version.keys()
2383
def iter_parents(self, version_ids):
2384
"""Iterate through the parents for many version ids.
2386
:param version_ids: An iterable yielding version_ids.
2387
:return: An iterator that yields (version_id, parents). Requested
2388
version_ids not present in the versioned file are simply skipped.
2389
The order is undefined, allowing for different optimisations in
2390
the underlying implementation.
2393
for version in version_ids:
2395
result.append((version, self._by_version[version][2]))
2401
class _KnitData(object):
2402
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2404
The KnitData class provides the logic for parsing and using knit records,
2405
making use of an access method for the low level read and write operations.
2408
def __init__(self, access):
2409
"""Create a KnitData object.
2411
:param access: The access method to use. Access methods such as
2412
_KnitAccess manage the insertion of raw records and the subsequent
2413
retrieval of the same.
2415
self._access = access
2416
self._checked = False
2417
# TODO: jam 20060713 conceptually, this could spill to disk
2418
# if the cached size gets larger than a certain amount
2419
# but it complicates the model a bit, so for now just use
2420
# a simple dictionary
2422
self._do_cache = False
2424
def enable_cache(self):
2425
"""Enable caching of reads."""
2426
self._do_cache = True
2428
def clear_cache(self):
2429
"""Clear the record cache."""
2430
self._do_cache = False
2433
def _open_file(self):
2434
return self._access.open_file()
2436
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2437
"""Convert version_id, digest, lines into a raw data block.
2439
:param dense_lines: The bytes of lines but in a denser form. For
2440
instance, if lines is a list of 1000 bytestrings each ending in \n,
2441
dense_lines may be a list with one line in it, containing all the
2442
1000's lines and their \n's. Using dense_lines if it is already
2443
known is a win because the string join to create bytes in this
2444
function spends less time resizing the final string.
2445
:return: (len, a StringIO instance with the raw data ready to read.)
2447
# Note: using a string copy here increases memory pressure with e.g.
2448
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2449
# when doing the initial commit of a mozilla tree. RBC 20070921
2450
bytes = ''.join(chain(
2451
["version %s %d %s\n" % (version_id,
2454
dense_lines or lines,
2455
["end %s\n" % version_id]))
2456
assert bytes.__class__ == str
2457
compressed_bytes = bytes_to_gzip(bytes)
2458
return len(compressed_bytes), compressed_bytes
2460
def add_raw_records(self, sizes, raw_data):
2461
"""Append a prepared record to the data file.
2463
:param sizes: An iterable containing the size of each raw data segment.
2464
:param raw_data: A bytestring containing the data.
2465
:return: a list of index data for the way the data was stored.
2466
See the access method add_raw_records documentation for more
2469
return self._access.add_raw_records(sizes, raw_data)
2471
def _parse_record_header(self, version_id, raw_data):
2472
"""Parse a record header for consistency.
2474
:return: the header and the decompressor stream.
2475
as (stream, header_record)
2477
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2479
rec = self._check_header(version_id, df.readline())
2480
except Exception, e:
2481
raise KnitCorrupt(self._access,
2482
"While reading {%s} got %s(%s)"
2483
% (version_id, e.__class__.__name__, str(e)))
2486
def _check_header(self, version_id, line):
2489
raise KnitCorrupt(self._access,
2490
'unexpected number of elements in record header')
2491
if rec[1] != version_id:
2492
raise KnitCorrupt(self._access,
2493
'unexpected version, wanted %r, got %r'
2494
% (version_id, rec[1]))
2497
def _parse_record(self, version_id, data):
2499
# 4168 calls in 2880 217 internal
2500
# 4168 calls to _parse_record_header in 2121
2501
# 4168 calls to readlines in 330
2502
df = GzipFile(mode='rb', fileobj=StringIO(data))
2505
record_contents = df.readlines()
2506
except Exception, e:
2507
raise KnitCorrupt(self._access,
2508
"While reading {%s} got %s(%s)"
2509
% (version_id, e.__class__.__name__, str(e)))
2510
header = record_contents.pop(0)
2511
rec = self._check_header(version_id, header)
2513
last_line = record_contents.pop()
2514
if len(record_contents) != int(rec[2]):
2515
raise KnitCorrupt(self._access,
2516
'incorrect number of lines %s != %s'
2518
% (len(record_contents), int(rec[2]),
2520
if last_line != 'end %s\n' % rec[1]:
2521
raise KnitCorrupt(self._access,
2522
'unexpected version end line %r, wanted %r'
2523
% (last_line, version_id))
2525
return record_contents, rec[3]
2527
def read_records_iter_raw(self, records):
2528
"""Read text records from data file and yield raw data.
2530
This unpacks enough of the text record to validate the id is
2531
as expected but thats all.
2533
# setup an iterator of the external records:
2534
# uses readv so nice and fast we hope.
2536
# grab the disk data needed.
2538
# Don't check _cache if it is empty
2539
needed_offsets = [index_memo for version_id, index_memo
2541
if version_id not in self._cache]
2543
needed_offsets = [index_memo for version_id, index_memo
2546
raw_records = self._access.get_raw_records(needed_offsets)
2548
for version_id, index_memo in records:
2549
if version_id in self._cache:
2550
# This data has already been validated
2551
data = self._cache[version_id]
2553
data = raw_records.next()
2555
self._cache[version_id] = data
2557
# validate the header
2558
df, rec = self._parse_record_header(version_id, data)
2560
yield version_id, data
2562
def read_records_iter(self, records):
2563
"""Read text records from data file and yield result.
2565
The result will be returned in whatever is the fastest to read.
2566
Not by the order requested. Also, multiple requests for the same
2567
record will only yield 1 response.
2568
:param records: A list of (version_id, pos, len) entries
2569
:return: Yields (version_id, contents, digest) in the order
2570
read, not the order requested
2576
# Skip records we have alread seen
2577
yielded_records = set()
2578
needed_records = set()
2579
for record in records:
2580
if record[0] in self._cache:
2581
if record[0] in yielded_records:
2583
yielded_records.add(record[0])
2584
data = self._cache[record[0]]
2585
content, digest = self._parse_record(record[0], data)
2586
yield (record[0], content, digest)
2588
needed_records.add(record)
2589
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2591
needed_records = sorted(set(records), key=operator.itemgetter(1))
2593
if not needed_records:
2596
# The transport optimizes the fetching as well
2597
# (ie, reads continuous ranges.)
2598
raw_data = self._access.get_raw_records(
2599
[index_memo for version_id, index_memo in needed_records])
2601
for (version_id, index_memo), data in \
2602
izip(iter(needed_records), raw_data):
2603
content, digest = self._parse_record(version_id, data)
2605
self._cache[version_id] = data
2606
yield version_id, content, digest
2608
def read_records(self, records):
2609
"""Read records into a dictionary."""
2611
for record_id, content, digest in \
2612
self.read_records_iter(records):
2613
components[record_id] = (content, digest)
2617
class InterKnit(InterVersionedFile):
2618
"""Optimised code paths for knit to knit operations."""
2620
_matching_file_from_factory = KnitVersionedFile
2621
_matching_file_to_factory = KnitVersionedFile
2624
def is_compatible(source, target):
2625
"""Be compatible with knits. """
2627
return (isinstance(source, KnitVersionedFile) and
2628
isinstance(target, KnitVersionedFile))
2629
except AttributeError:
2632
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2633
"""Copy texts to the target by extracting and adding them one by one.
2635
see join() for the parameter definitions.
2637
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2638
graph = self.source.get_graph(version_ids)
2639
order = topo_sort(graph.items())
2641
def size_of_content(content):
2642
return sum(len(line) for line in content.text())
2643
# Cache at most 10MB of parent texts
2644
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2645
compute_size=size_of_content)
2646
# TODO: jam 20071116 It would be nice to have a streaming interface to
2647
# get multiple texts from a source. The source could be smarter
2648
# about how it handled intermediate stages.
2649
# get_line_list() or make_mpdiffs() seem like a possibility, but
2650
# at the moment they extract all full texts into memory, which
2651
# causes us to store more than our 3x fulltext goal.
2652
# Repository.iter_files_bytes() may be another possibility
2653
to_process = [version for version in order
2654
if version not in self.target]
2655
total = len(to_process)
2656
pb = ui.ui_factory.nested_progress_bar()
2658
for index, version in enumerate(to_process):
2659
pb.update('Converting versioned data', index, total)
2660
sha1, num_bytes, parent_text = self.target.add_lines(version,
2661
self.source.get_parents_with_ghosts(version),
2662
self.source.get_lines(version),
2663
parent_texts=parent_cache)
2664
parent_cache[version] = parent_text
2669
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2670
"""See InterVersionedFile.join."""
2671
assert isinstance(self.source, KnitVersionedFile)
2672
assert isinstance(self.target, KnitVersionedFile)
2674
# If the source and target are mismatched w.r.t. annotations vs
2675
# plain, the data needs to be converted accordingly
2676
if self.source.factory.annotated == self.target.factory.annotated:
2678
elif self.source.factory.annotated:
2679
converter = self._anno_to_plain_converter
2681
# We're converting from a plain to an annotated knit. Copy them
2682
# across by full texts.
2683
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2685
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2689
pb = ui.ui_factory.nested_progress_bar()
2691
version_ids = list(version_ids)
2692
if None in version_ids:
2693
version_ids.remove(None)
2695
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2697
this_versions = set(self.target._index.get_versions())
2698
# XXX: For efficiency we should not look at the whole index,
2699
# we only need to consider the referenced revisions - they
2700
# must all be present, or the method must be full-text.
2701
# TODO, RBC 20070919
2702
needed_versions = self.source_ancestry - this_versions
2704
if not needed_versions:
2706
full_list = topo_sort(self.source.get_graph())
2708
version_list = [i for i in full_list if (not self.target.has_version(i)
2709
and i in needed_versions)]
2713
copy_queue_records = []
2715
for version_id in version_list:
2716
options = self.source._index.get_options(version_id)
2717
parents = self.source._index.get_parents_with_ghosts(version_id)
2718
# check that its will be a consistent copy:
2719
for parent in parents:
2720
# if source has the parent, we must :
2721
# * already have it or
2722
# * have it scheduled already
2723
# otherwise we don't care
2724
assert (self.target.has_version(parent) or
2725
parent in copy_set or
2726
not self.source.has_version(parent))
2727
index_memo = self.source._index.get_position(version_id)
2728
copy_queue_records.append((version_id, index_memo))
2729
copy_queue.append((version_id, options, parents))
2730
copy_set.add(version_id)
2732
# data suck the join:
2734
total = len(version_list)
2737
for (version_id, raw_data), \
2738
(version_id2, options, parents) in \
2739
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2741
assert version_id == version_id2, 'logic error, inconsistent results'
2743
pb.update("Joining knit", count, total)
2745
size, raw_data = converter(raw_data, version_id, options,
2748
size = len(raw_data)
2749
raw_records.append((version_id, options, parents, size))
2750
raw_datum.append(raw_data)
2751
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2756
def _anno_to_plain_converter(self, raw_data, version_id, options,
2758
"""Convert annotated content to plain content."""
2759
data, digest = self.source._data._parse_record(version_id, raw_data)
2760
if 'fulltext' in options:
2761
content = self.source.factory.parse_fulltext(data, version_id)
2762
lines = self.target.factory.lower_fulltext(content)
2764
delta = self.source.factory.parse_line_delta(data, version_id,
2766
lines = self.target.factory.lower_line_delta(delta)
2767
return self.target._data._record_to_data(version_id, digest, lines)
2770
InterVersionedFile.register_optimiser(InterKnit)
2773
class WeaveToKnit(InterVersionedFile):
2774
"""Optimised code paths for weave to knit operations."""
2776
_matching_file_from_factory = bzrlib.weave.WeaveFile
2777
_matching_file_to_factory = KnitVersionedFile
2780
def is_compatible(source, target):
2781
"""Be compatible with weaves to knits."""
2783
return (isinstance(source, bzrlib.weave.Weave) and
2784
isinstance(target, KnitVersionedFile))
2785
except AttributeError:
2788
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2789
"""See InterVersionedFile.join."""
2790
assert isinstance(self.source, bzrlib.weave.Weave)
2791
assert isinstance(self.target, KnitVersionedFile)
2793
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2798
pb = ui.ui_factory.nested_progress_bar()
2800
version_ids = list(version_ids)
2802
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2803
this_versions = set(self.target._index.get_versions())
2804
needed_versions = self.source_ancestry - this_versions
2806
if not needed_versions:
2808
full_list = topo_sort(self.source.get_graph())
2810
version_list = [i for i in full_list if (not self.target.has_version(i)
2811
and i in needed_versions)]
2815
total = len(version_list)
2816
for version_id in version_list:
2817
pb.update("Converting to knit", count, total)
2818
parents = self.source.get_parents(version_id)
2819
# check that its will be a consistent copy:
2820
for parent in parents:
2821
# if source has the parent, we must already have it
2822
assert (self.target.has_version(parent))
2823
self.target.add_lines(
2824
version_id, parents, self.source.get_lines(version_id))
2831
InterVersionedFile.register_optimiser(WeaveToKnit)
2834
# Deprecated, use PatienceSequenceMatcher instead
2835
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2838
def annotate_knit(knit, revision_id):
2839
"""Annotate a knit with no cached annotations.
2841
This implementation is for knits with no cached annotations.
2842
It will work for knits with cached annotations, but this is not
2845
annotator = _KnitAnnotator(knit)
2846
return iter(annotator.annotate(revision_id))
2849
class _KnitAnnotator(object):
2850
"""Build up the annotations for a text."""
2852
def __init__(self, knit):
2855
# Content objects, differs from fulltexts because of how final newlines
2856
# are treated by knits. the content objects here will always have a
2858
self._fulltext_contents = {}
2860
# Annotated lines of specific revisions
2861
self._annotated_lines = {}
2863
# Track the raw data for nodes that we could not process yet.
2864
# This maps the revision_id of the base to a list of children that will
2865
# annotated from it.
2866
self._pending_children = {}
2868
# Track how many children this node has, so we know if we need to keep
2870
self._annotate_children = {}
2871
self._compression_children = {}
2873
self._all_build_details = {}
2874
# The children => parent revision_id graph
2875
self._revision_id_graph = {}
2877
self._heads_provider = None
2879
self._nodes_to_keep_annotations = set()
2880
self._generations_until_keep = 100
2882
def set_generations_until_keep(self, value):
2883
"""Set the number of generations before caching a node.
2885
Setting this to -1 will cache every merge node, setting this higher
2886
will cache fewer nodes.
2888
self._generations_until_keep = value
2890
def _add_fulltext_content(self, revision_id, content_obj):
2891
self._fulltext_contents[revision_id] = content_obj
2892
# TODO: jam 20080305 It might be good to check the sha1digest here
2893
return content_obj.text()
2895
def _check_parents(self, child, nodes_to_annotate):
2896
"""Check if all parents have been processed.
2898
:param child: A tuple of (rev_id, parents, raw_content)
2899
:param nodes_to_annotate: If child is ready, add it to
2900
nodes_to_annotate, otherwise put it back in self._pending_children
2902
for parent_id in child[1]:
2903
if parent_id not in self._annotated_lines:
2904
# This parent is present, but another parent is missing
2905
self._pending_children.setdefault(parent_id,
2909
# This one is ready to be processed
2910
nodes_to_annotate.append(child)
2912
def _add_annotation(self, revision_id, fulltext, parent_ids,
2913
left_matching_blocks=None):
2914
"""Add an annotation entry.
2916
All parents should already have been annotated.
2917
:return: A list of children that now have their parents satisfied.
2919
a = self._annotated_lines
2920
annotated_parent_lines = [a[p] for p in parent_ids]
2921
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2922
fulltext, revision_id, left_matching_blocks,
2923
heads_provider=self._get_heads_provider()))
2924
self._annotated_lines[revision_id] = annotated_lines
2925
for p in parent_ids:
2926
ann_children = self._annotate_children[p]
2927
ann_children.remove(revision_id)
2928
if (not ann_children
2929
and p not in self._nodes_to_keep_annotations):
2930
del self._annotated_lines[p]
2931
del self._all_build_details[p]
2932
if p in self._fulltext_contents:
2933
del self._fulltext_contents[p]
2934
# Now that we've added this one, see if there are any pending
2935
# deltas to be done, certainly this parent is finished
2936
nodes_to_annotate = []
2937
for child in self._pending_children.pop(revision_id, []):
2938
self._check_parents(child, nodes_to_annotate)
2939
return nodes_to_annotate
2941
def _get_build_graph(self, revision_id):
2942
"""Get the graphs for building texts and annotations.
2944
The data you need for creating a full text may be different than the
2945
data you need to annotate that text. (At a minimum, you need both
2946
parents to create an annotation, but only need 1 parent to generate the
2949
:return: A list of (revision_id, index_memo) records, suitable for
2950
passing to read_records_iter to start reading in the raw data fro/
2953
if revision_id in self._annotated_lines:
2956
pending = set([revision_id])
2961
# get all pending nodes
2963
this_iteration = pending
2964
build_details = self._knit._index.get_build_details(this_iteration)
2965
self._all_build_details.update(build_details)
2966
# new_nodes = self._knit._index._get_entries(this_iteration)
2968
for rev_id, details in build_details.iteritems():
2969
(index_memo, compression_parent, parents,
2970
record_details) = details
2971
self._revision_id_graph[rev_id] = parents
2972
records.append((rev_id, index_memo))
2973
# Do we actually need to check _annotated_lines?
2974
pending.update(p for p in parents
2975
if p not in self._all_build_details)
2976
if compression_parent:
2977
self._compression_children.setdefault(compression_parent,
2980
for parent in parents:
2981
self._annotate_children.setdefault(parent,
2983
num_gens = generation - kept_generation
2984
if ((num_gens >= self._generations_until_keep)
2985
and len(parents) > 1):
2986
kept_generation = generation
2987
self._nodes_to_keep_annotations.add(rev_id)
2989
missing_versions = this_iteration.difference(build_details.keys())
2990
for missing_version in missing_versions:
2991
# add a key, no parents
2992
self._revision_id_graph[missing_versions] = ()
2993
pending.discard(missing_version) # don't look for it
2994
# Generally we will want to read the records in reverse order, because
2995
# we find the parent nodes after the children
2999
def _annotate_records(self, records):
3000
"""Build the annotations for the listed records."""
3001
# We iterate in the order read, rather than a strict order requested
3002
# However, process what we can, and put off to the side things that
3003
# still need parents, cleaning them up when those parents are
3005
for (rev_id, record,
3006
digest) in self._knit._data.read_records_iter(records):
3007
if rev_id in self._annotated_lines:
3009
parent_ids = self._revision_id_graph[rev_id]
3010
details = self._all_build_details[rev_id]
3011
(index_memo, compression_parent, parents,
3012
record_details) = details
3013
nodes_to_annotate = []
3014
# TODO: Remove the punning between compression parents, and
3015
# parent_ids, we should be able to do this without assuming
3017
if len(parent_ids) == 0:
3018
# There are no parents for this node, so just add it
3019
# TODO: This probably needs to be decoupled
3020
assert compression_parent is None
3021
fulltext_content, delta = self._knit.factory.parse_record(
3022
rev_id, record, record_details, None)
3023
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3024
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3025
parent_ids, left_matching_blocks=None))
3027
child = (rev_id, parent_ids, record)
3028
# Check if all the parents are present
3029
self._check_parents(child, nodes_to_annotate)
3030
while nodes_to_annotate:
3031
# Should we use a queue here instead of a stack?
3032
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3033
(index_memo, compression_parent, parents,
3034
record_details) = self._all_build_details[rev_id]
3035
if compression_parent is not None:
3036
comp_children = self._compression_children[compression_parent]
3037
assert rev_id in comp_children
3038
# If there is only 1 child, it is safe to reuse this
3040
reuse_content = (len(comp_children) == 1
3041
and compression_parent not in
3042
self._nodes_to_keep_annotations)
3044
# Remove it from the cache since it will be changing
3045
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3046
# Make sure to copy the fulltext since it might be
3048
parent_fulltext = list(parent_fulltext_content.text())
3050
parent_fulltext_content = self._fulltext_contents[compression_parent]
3051
parent_fulltext = parent_fulltext_content.text()
3052
comp_children.remove(rev_id)
3053
fulltext_content, delta = self._knit.factory.parse_record(
3054
rev_id, record, record_details,
3055
parent_fulltext_content,
3056
copy_base_content=(not reuse_content))
3057
fulltext = self._add_fulltext_content(rev_id,
3059
blocks = KnitContent.get_line_delta_blocks(delta,
3060
parent_fulltext, fulltext)
3062
fulltext_content = self._knit.factory.parse_fulltext(
3064
fulltext = self._add_fulltext_content(rev_id,
3067
nodes_to_annotate.extend(
3068
self._add_annotation(rev_id, fulltext, parent_ids,
3069
left_matching_blocks=blocks))
3071
def _get_heads_provider(self):
3072
"""Create a heads provider for resolving ancestry issues."""
3073
if self._heads_provider is not None:
3074
return self._heads_provider
3075
parent_provider = _mod_graph.DictParentsProvider(
3076
self._revision_id_graph)
3077
graph_obj = _mod_graph.Graph(parent_provider)
3078
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3079
self._heads_provider = head_cache
3082
def annotate(self, revision_id):
3083
"""Return the annotated fulltext at the given revision.
3085
:param revision_id: The revision id for this file
3087
records = self._get_build_graph(revision_id)
3088
self._annotate_records(records)
3089
return self._annotated_lines[revision_id]
3093
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3095
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
