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_parent_map([version_id])[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
current_components = set(pending_components)
949
pending_components = set()
950
for version_id, details in build_details.iteritems():
951
(index_memo, compression_parent, parents,
952
record_details) = details
953
method = record_details[0]
954
if compression_parent is not None:
955
pending_components.add(compression_parent)
956
component_data[version_id] = (record_details, index_memo,
958
missing = current_components.difference(build_details)
960
raise errors.RevisionNotPresent(missing.pop(), self.filename)
961
return component_data
963
def _get_content(self, version_id, parent_texts={}):
964
"""Returns a content object that makes up the specified
966
cached_version = parent_texts.get(version_id, None)
967
if cached_version is not None:
968
if not self.has_version(version_id):
969
raise RevisionNotPresent(version_id, self.filename)
970
return cached_version
972
text_map, contents_map = self._get_content_maps([version_id])
973
return contents_map[version_id]
975
def _check_versions_present(self, version_ids):
976
"""Check that all specified versions are present."""
977
self._index.check_versions_present(version_ids)
979
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
980
nostore_sha, random_id, check_content, left_matching_blocks):
981
"""See VersionedFile.add_lines_with_ghosts()."""
982
self._check_add(version_id, lines, random_id, check_content)
983
return self._add(version_id, lines, parents, self.delta,
984
parent_texts, left_matching_blocks, nostore_sha, random_id)
986
def _add_lines(self, version_id, parents, lines, parent_texts,
987
left_matching_blocks, nostore_sha, random_id, check_content):
988
"""See VersionedFile.add_lines."""
989
self._check_add(version_id, lines, random_id, check_content)
990
self._check_versions_present(parents)
991
return self._add(version_id, lines[:], parents, self.delta,
992
parent_texts, left_matching_blocks, nostore_sha, random_id)
994
def _check_add(self, version_id, lines, random_id, check_content):
995
"""check that version_id and lines are safe to add."""
996
if contains_whitespace(version_id):
997
raise InvalidRevisionId(version_id, self.filename)
998
self.check_not_reserved_id(version_id)
999
# Technically this could be avoided if we are happy to allow duplicate
1000
# id insertion when other things than bzr core insert texts, but it
1001
# seems useful for folk using the knit api directly to have some safety
1002
# blanket that we can disable.
1003
if not random_id and self.has_version(version_id):
1004
raise RevisionAlreadyPresent(version_id, self.filename)
1006
self._check_lines_not_unicode(lines)
1007
self._check_lines_are_lines(lines)
1009
def _add(self, version_id, lines, parents, delta, parent_texts,
1010
left_matching_blocks, nostore_sha, random_id):
1011
"""Add a set of lines on top of version specified by parents.
1013
If delta is true, compress the text as a line-delta against
1016
Any versions not present will be converted into ghosts.
1018
# first thing, if the content is something we don't need to store, find
1020
line_bytes = ''.join(lines)
1021
digest = sha_string(line_bytes)
1022
if nostore_sha == digest:
1023
raise errors.ExistingContent
1025
present_parents = []
1026
if parent_texts is None:
1028
for parent in parents:
1029
if self.has_version(parent):
1030
present_parents.append(parent)
1032
# can only compress against the left most present parent.
1034
(len(present_parents) == 0 or
1035
present_parents[0] != parents[0])):
1038
text_length = len(line_bytes)
1041
if lines[-1][-1] != '\n':
1042
# copy the contents of lines.
1044
options.append('no-eol')
1045
lines[-1] = lines[-1] + '\n'
1049
# To speed the extract of texts the delta chain is limited
1050
# to a fixed number of deltas. This should minimize both
1051
# I/O and the time spend applying deltas.
1052
delta = self._check_should_delta(present_parents)
1054
assert isinstance(version_id, str)
1055
content = self.factory.make(lines, version_id)
1056
if delta or (self.factory.annotated and len(present_parents) > 0):
1057
# Merge annotations from parent texts if needed.
1058
delta_hunks = self._merge_annotations(content, present_parents,
1059
parent_texts, delta, self.factory.annotated,
1060
left_matching_blocks)
1063
options.append('line-delta')
1064
store_lines = self.factory.lower_line_delta(delta_hunks)
1065
size, bytes = self._data._record_to_data(version_id, digest,
1068
options.append('fulltext')
1069
# isinstance is slower and we have no hierarchy.
1070
if self.factory.__class__ == KnitPlainFactory:
1071
# Use the already joined bytes saving iteration time in
1073
size, bytes = self._data._record_to_data(version_id, digest,
1074
lines, [line_bytes])
1076
# get mixed annotation + content and feed it into the
1078
store_lines = self.factory.lower_fulltext(content)
1079
size, bytes = self._data._record_to_data(version_id, digest,
1082
access_memo = self._data.add_raw_records([size], bytes)[0]
1083
self._index.add_versions(
1084
((version_id, options, access_memo, parents),),
1085
random_id=random_id)
1086
return digest, text_length, content
1088
def check(self, progress_bar=None):
1089
"""See VersionedFile.check()."""
1091
def _clone_text(self, new_version_id, old_version_id, parents):
1092
"""See VersionedFile.clone_text()."""
1093
# FIXME RBC 20060228 make fast by only inserting an index with null
1095
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
1097
def get_lines(self, version_id):
1098
"""See VersionedFile.get_lines()."""
1099
return self.get_line_list([version_id])[0]
1101
def _get_record_map(self, version_ids):
1102
"""Produce a dictionary of knit records.
1104
:return: {version_id:(record, record_details, digest, next)}
1106
data returned from read_records
1108
opaque information to pass to parse_record
1110
SHA1 digest of the full text after all steps are done
1112
build-parent of the version, i.e. the leftmost ancestor.
1113
Will be None if the record is not a delta.
1115
position_map = self._get_components_positions(version_ids)
1116
# c = component_id, r = record_details, i_m = index_memo, n = next
1117
records = [(c, i_m) for c, (r, i_m, n)
1118
in position_map.iteritems()]
1120
for component_id, record, digest in \
1121
self._data.read_records_iter(records):
1122
(record_details, index_memo, next) = position_map[component_id]
1123
record_map[component_id] = record, record_details, digest, next
1127
def get_text(self, version_id):
1128
"""See VersionedFile.get_text"""
1129
return self.get_texts([version_id])[0]
1131
def get_texts(self, version_ids):
1132
return [''.join(l) for l in self.get_line_list(version_ids)]
1134
def get_line_list(self, version_ids):
1135
"""Return the texts of listed versions as a list of strings."""
1136
for version_id in version_ids:
1137
self.check_not_reserved_id(version_id)
1138
text_map, content_map = self._get_content_maps(version_ids)
1139
return [text_map[v] for v in version_ids]
1141
_get_lf_split_line_list = get_line_list
1143
def _get_content_maps(self, version_ids):
1144
"""Produce maps of text and KnitContents
1146
:return: (text_map, content_map) where text_map contains the texts for
1147
the requested versions and content_map contains the KnitContents.
1148
Both dicts take version_ids as their keys.
1150
# FUTURE: This function could be improved for the 'extract many' case
1151
# by tracking each component and only doing the copy when the number of
1152
# children than need to apply delta's to it is > 1 or it is part of the
1154
version_ids = list(version_ids)
1155
multiple_versions = len(version_ids) != 1
1156
record_map = self._get_record_map(version_ids)
1161
for version_id in version_ids:
1164
while cursor is not None:
1165
record, record_details, digest, next = record_map[cursor]
1166
components.append((cursor, record, record_details, digest))
1167
if cursor in content_map:
1172
for (component_id, record, record_details,
1173
digest) in reversed(components):
1174
if component_id in content_map:
1175
content = content_map[component_id]
1177
content, delta = self.factory.parse_record(version_id,
1178
record, record_details, content,
1179
copy_base_content=multiple_versions)
1180
if multiple_versions:
1181
content_map[component_id] = content
1183
content.cleanup_eol(copy_on_mutate=multiple_versions)
1184
final_content[version_id] = content
1186
# digest here is the digest from the last applied component.
1187
text = content.text()
1188
actual_sha = sha_strings(text)
1189
if actual_sha != digest:
1190
raise KnitCorrupt(self.filename,
1192
'\n of reconstructed text does not match'
1194
'\n for version %s' %
1195
(actual_sha, digest, version_id))
1196
text_map[version_id] = text
1197
return text_map, final_content
1199
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1201
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1202
if version_ids is None:
1203
version_ids = self.versions()
1205
pb = progress.DummyProgress()
1206
# we don't care about inclusions, the caller cares.
1207
# but we need to setup a list of records to visit.
1208
# we need version_id, position, length
1209
version_id_records = []
1210
requested_versions = set(version_ids)
1211
# filter for available versions
1212
for version_id in requested_versions:
1213
if not self.has_version(version_id):
1214
raise RevisionNotPresent(version_id, self.filename)
1215
# get a in-component-order queue:
1216
for version_id in self.versions():
1217
if version_id in requested_versions:
1218
index_memo = self._index.get_position(version_id)
1219
version_id_records.append((version_id, index_memo))
1221
total = len(version_id_records)
1222
for version_idx, (version_id, data, sha_value) in \
1223
enumerate(self._data.read_records_iter(version_id_records)):
1224
pb.update('Walking content.', version_idx, total)
1225
method = self._index.get_method(version_id)
1227
assert method in ('fulltext', 'line-delta')
1228
if method == 'fulltext':
1229
line_iterator = self.factory.get_fulltext_content(data)
1231
line_iterator = self.factory.get_linedelta_content(data)
1232
# XXX: It might be more efficient to yield (version_id,
1233
# line_iterator) in the future. However for now, this is a simpler
1234
# change to integrate into the rest of the codebase. RBC 20071110
1235
for line in line_iterator:
1236
yield line, version_id
1238
pb.update('Walking content.', total, total)
1240
def iter_parents(self, version_ids):
1241
"""Iterate through the parents for many version ids.
1243
:param version_ids: An iterable yielding version_ids.
1244
:return: An iterator that yields (version_id, parents). Requested
1245
version_ids not present in the versioned file are simply skipped.
1246
The order is undefined, allowing for different optimisations in
1247
the underlying implementation.
1249
return self._index.iter_parents(version_ids)
1251
def num_versions(self):
1252
"""See VersionedFile.num_versions()."""
1253
return self._index.num_versions()
1255
__len__ = num_versions
1257
def annotate_iter(self, version_id):
1258
"""See VersionedFile.annotate_iter."""
1259
return self.factory.annotate_iter(self, version_id)
1261
def get_parent_map(self, version_ids):
1262
"""See VersionedFile.get_parent_map."""
1263
return self._index.get_parent_map(version_ids)
1265
def get_ancestry(self, versions, topo_sorted=True):
1266
"""See VersionedFile.get_ancestry."""
1267
if isinstance(versions, basestring):
1268
versions = [versions]
1271
return self._index.get_ancestry(versions, topo_sorted)
1273
def get_ancestry_with_ghosts(self, versions):
1274
"""See VersionedFile.get_ancestry_with_ghosts."""
1275
if isinstance(versions, basestring):
1276
versions = [versions]
1279
return self._index.get_ancestry_with_ghosts(versions)
1281
def plan_merge(self, ver_a, ver_b):
1282
"""See VersionedFile.plan_merge."""
1283
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1284
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1285
annotated_a = self.annotate(ver_a)
1286
annotated_b = self.annotate(ver_b)
1287
return merge._plan_annotate_merge(annotated_a, annotated_b,
1288
ancestors_a, ancestors_b)
1291
class _KnitComponentFile(object):
1292
"""One of the files used to implement a knit database"""
1294
def __init__(self, transport, filename, mode, file_mode=None,
1295
create_parent_dir=False, dir_mode=None):
1296
self._transport = transport
1297
self._filename = filename
1299
self._file_mode = file_mode
1300
self._dir_mode = dir_mode
1301
self._create_parent_dir = create_parent_dir
1302
self._need_to_create = False
1304
def _full_path(self):
1305
"""Return the full path to this file."""
1306
return self._transport.base + self._filename
1308
def check_header(self, fp):
1309
line = fp.readline()
1311
# An empty file can actually be treated as though the file doesn't
1313
raise errors.NoSuchFile(self._full_path())
1314
if line != self.HEADER:
1315
raise KnitHeaderError(badline=line,
1316
filename=self._transport.abspath(self._filename))
1319
return '%s(%s)' % (self.__class__.__name__, self._filename)
1322
class _KnitIndex(_KnitComponentFile):
1323
"""Manages knit index file.
1325
The index is already kept in memory and read on startup, to enable
1326
fast lookups of revision information. The cursor of the index
1327
file is always pointing to the end, making it easy to append
1330
_cache is a cache for fast mapping from version id to a Index
1333
_history is a cache for fast mapping from indexes to version ids.
1335
The index data format is dictionary compressed when it comes to
1336
parent references; a index entry may only have parents that with a
1337
lover index number. As a result, the index is topological sorted.
1339
Duplicate entries may be written to the index for a single version id
1340
if this is done then the latter one completely replaces the former:
1341
this allows updates to correct version and parent information.
1342
Note that the two entries may share the delta, and that successive
1343
annotations and references MUST point to the first entry.
1345
The index file on disc contains a header, followed by one line per knit
1346
record. The same revision can be present in an index file more than once.
1347
The first occurrence gets assigned a sequence number starting from 0.
1349
The format of a single line is
1350
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1351
REVISION_ID is a utf8-encoded revision id
1352
FLAGS is a comma separated list of flags about the record. Values include
1353
no-eol, line-delta, fulltext.
1354
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1355
that the the compressed data starts at.
1356
LENGTH is the ascii representation of the length of the data file.
1357
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1359
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1360
revision id already in the knit that is a parent of REVISION_ID.
1361
The ' :' marker is the end of record marker.
1364
when a write is interrupted to the index file, it will result in a line
1365
that does not end in ' :'. If the ' :' is not present at the end of a line,
1366
or at the end of the file, then the record that is missing it will be
1367
ignored by the parser.
1369
When writing new records to the index file, the data is preceded by '\n'
1370
to ensure that records always start on new lines even if the last write was
1371
interrupted. As a result its normal for the last line in the index to be
1372
missing a trailing newline. One can be added with no harmful effects.
1375
HEADER = "# bzr knit index 8\n"
1377
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1378
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1380
def _cache_version(self, version_id, options, pos, size, parents):
1381
"""Cache a version record in the history array and index cache.
1383
This is inlined into _load_data for performance. KEEP IN SYNC.
1384
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1387
# only want the _history index to reference the 1st index entry
1389
if version_id not in self._cache:
1390
index = len(self._history)
1391
self._history.append(version_id)
1393
index = self._cache[version_id][5]
1394
self._cache[version_id] = (version_id,
1401
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1402
create_parent_dir=False, delay_create=False, dir_mode=None):
1403
_KnitComponentFile.__init__(self, transport, filename, mode,
1404
file_mode=file_mode,
1405
create_parent_dir=create_parent_dir,
1408
# position in _history is the 'official' index for a revision
1409
# but the values may have come from a newer entry.
1410
# so - wc -l of a knit index is != the number of unique names
1414
fp = self._transport.get(self._filename)
1416
# _load_data may raise NoSuchFile if the target knit is
1418
_load_data(self, fp)
1422
if mode != 'w' or not create:
1425
self._need_to_create = True
1427
self._transport.put_bytes_non_atomic(
1428
self._filename, self.HEADER, mode=self._file_mode)
1430
def get_graph(self):
1431
"""Return a list of the node:parents lists from this knit index."""
1432
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1434
def get_ancestry(self, versions, topo_sorted=True):
1435
"""See VersionedFile.get_ancestry."""
1436
# get a graph of all the mentioned versions:
1438
pending = set(versions)
1441
version = pending.pop()
1444
parents = [p for p in cache[version][4] if p in cache]
1446
raise RevisionNotPresent(version, self._filename)
1447
# if not completed and not a ghost
1448
pending.update([p for p in parents if p not in graph])
1449
graph[version] = parents
1452
return topo_sort(graph.items())
1454
def get_ancestry_with_ghosts(self, versions):
1455
"""See VersionedFile.get_ancestry_with_ghosts."""
1456
# get a graph of all the mentioned versions:
1457
self.check_versions_present(versions)
1460
pending = set(versions)
1462
version = pending.pop()
1464
parents = cache[version][4]
1470
pending.update([p for p in parents if p not in graph])
1471
graph[version] = parents
1472
return topo_sort(graph.items())
1474
def get_build_details(self, version_ids):
1475
"""Get the method, index_memo and compression parent for version_ids.
1477
Ghosts are omitted from the result.
1479
:param version_ids: An iterable of version_ids.
1480
:return: A dict of version_id:(index_memo, compression_parent,
1481
parents, record_details).
1483
opaque structure to pass to read_records to extract the raw
1486
Content that this record is built upon, may be None
1488
Logical parents of this node
1490
extra information about the content which needs to be passed to
1491
Factory.parse_record
1494
for version_id in version_ids:
1495
if version_id not in self._cache:
1496
# ghosts are omitted
1498
method = self.get_method(version_id)
1499
parents = self.get_parents_with_ghosts(version_id)
1500
if method == 'fulltext':
1501
compression_parent = None
1503
compression_parent = parents[0]
1504
noeol = 'no-eol' in self.get_options(version_id)
1505
index_memo = self.get_position(version_id)
1506
result[version_id] = (index_memo, compression_parent,
1507
parents, (method, noeol))
1510
def iter_parents(self, version_ids):
1511
"""Iterate through the parents for many version ids.
1513
:param version_ids: An iterable yielding version_ids.
1514
:return: An iterator that yields (version_id, parents). Requested
1515
version_ids not present in the versioned file are simply skipped.
1516
The order is undefined, allowing for different optimisations in
1517
the underlying implementation.
1519
for version_id in version_ids:
1521
yield version_id, tuple(self.get_parents(version_id))
1525
def num_versions(self):
1526
return len(self._history)
1528
__len__ = num_versions
1530
def get_versions(self):
1531
"""Get all the versions in the file. not topologically sorted."""
1532
return self._history
1534
def _version_list_to_index(self, versions):
1537
for version in versions:
1538
if version in cache:
1539
# -- inlined lookup() --
1540
result_list.append(str(cache[version][5]))
1541
# -- end lookup () --
1543
result_list.append('.' + version)
1544
return ' '.join(result_list)
1546
def add_version(self, version_id, options, index_memo, parents):
1547
"""Add a version record to the index."""
1548
self.add_versions(((version_id, options, index_memo, parents),))
1550
def add_versions(self, versions, random_id=False):
1551
"""Add multiple versions to the index.
1553
:param versions: a list of tuples:
1554
(version_id, options, pos, size, parents).
1555
:param random_id: If True the ids being added were randomly generated
1556
and no check for existence will be performed.
1559
orig_history = self._history[:]
1560
orig_cache = self._cache.copy()
1563
for version_id, options, (index, pos, size), parents in versions:
1564
line = "\n%s %s %s %s %s :" % (version_id,
1568
self._version_list_to_index(parents))
1569
assert isinstance(line, str), \
1570
'content must be utf-8 encoded: %r' % (line,)
1572
self._cache_version(version_id, options, pos, size, tuple(parents))
1573
if not self._need_to_create:
1574
self._transport.append_bytes(self._filename, ''.join(lines))
1577
sio.write(self.HEADER)
1578
sio.writelines(lines)
1580
self._transport.put_file_non_atomic(self._filename, sio,
1581
create_parent_dir=self._create_parent_dir,
1582
mode=self._file_mode,
1583
dir_mode=self._dir_mode)
1584
self._need_to_create = False
1586
# If any problems happen, restore the original values and re-raise
1587
self._history = orig_history
1588
self._cache = orig_cache
1591
def has_version(self, version_id):
1592
"""True if the version is in the index."""
1593
return version_id in self._cache
1595
def get_position(self, version_id):
1596
"""Return details needed to access the version.
1598
.kndx indices do not support split-out data, so return None for the
1601
:return: a tuple (None, data position, size) to hand to the access
1602
logic to get the record.
1604
entry = self._cache[version_id]
1605
return None, entry[2], entry[3]
1607
def get_method(self, version_id):
1608
"""Return compression method of specified version."""
1610
options = self._cache[version_id][1]
1612
raise RevisionNotPresent(version_id, self._filename)
1613
if 'fulltext' in options:
1616
if 'line-delta' not in options:
1617
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1620
def get_options(self, version_id):
1621
"""Return a list representing options.
1625
return self._cache[version_id][1]
1627
def get_parent_map(self, version_ids):
1628
"""Passed through to by KnitVersionedFile.get_parent_map."""
1630
for version_id in version_ids:
1632
result[version_id] = tuple(self._cache[version_id][4])
1637
def get_parents(self, version_id):
1638
"""Return parents of specified version ignoring ghosts."""
1639
return [parent for parent in self._cache[version_id][4]
1640
if parent in self._cache]
1642
def get_parents_with_ghosts(self, version_id):
1643
"""Return parents of specified version with ghosts."""
1645
return self.get_parent_map([version_id])[version_id]
1647
raise RevisionNotPresent(version_id, self)
1649
def check_versions_present(self, version_ids):
1650
"""Check that all specified versions are present."""
1652
for version_id in version_ids:
1653
if version_id not in cache:
1654
raise RevisionNotPresent(version_id, self._filename)
1657
class KnitGraphIndex(object):
1658
"""A knit index that builds on GraphIndex."""
1660
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1661
"""Construct a KnitGraphIndex on a graph_index.
1663
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1664
:param deltas: Allow delta-compressed records.
1665
:param add_callback: If not None, allow additions to the index and call
1666
this callback with a list of added GraphIndex nodes:
1667
[(node, value, node_refs), ...]
1668
:param parents: If True, record knits parents, if not do not record
1671
self._graph_index = graph_index
1672
self._deltas = deltas
1673
self._add_callback = add_callback
1674
self._parents = parents
1675
if deltas and not parents:
1676
raise KnitCorrupt(self, "Cannot do delta compression without "
1679
def _get_entries(self, keys, check_present=False):
1680
"""Get the entries for keys.
1682
:param keys: An iterable of index keys, - 1-tuples.
1687
for node in self._graph_index.iter_entries(keys):
1689
found_keys.add(node[1])
1691
# adapt parentless index to the rest of the code.
1692
for node in self._graph_index.iter_entries(keys):
1693
yield node[0], node[1], node[2], ()
1694
found_keys.add(node[1])
1696
missing_keys = keys.difference(found_keys)
1698
raise RevisionNotPresent(missing_keys.pop(), self)
1700
def _present_keys(self, version_ids):
1702
node[1] for node in self._get_entries(version_ids)])
1704
def _parentless_ancestry(self, versions):
1705
"""Honour the get_ancestry API for parentless knit indices."""
1706
wanted_keys = self._version_ids_to_keys(versions)
1707
present_keys = self._present_keys(wanted_keys)
1708
missing = set(wanted_keys).difference(present_keys)
1710
raise RevisionNotPresent(missing.pop(), self)
1711
return list(self._keys_to_version_ids(present_keys))
1713
def get_ancestry(self, versions, topo_sorted=True):
1714
"""See VersionedFile.get_ancestry."""
1715
if not self._parents:
1716
return self._parentless_ancestry(versions)
1717
# XXX: This will do len(history) index calls - perhaps
1718
# it should be altered to be a index core feature?
1719
# get a graph of all the mentioned versions:
1722
versions = self._version_ids_to_keys(versions)
1723
pending = set(versions)
1725
# get all pending nodes
1726
this_iteration = pending
1727
new_nodes = self._get_entries(this_iteration)
1730
for (index, key, value, node_refs) in new_nodes:
1731
# dont ask for ghosties - otherwise
1732
# we we can end up looping with pending
1733
# being entirely ghosted.
1734
graph[key] = [parent for parent in node_refs[0]
1735
if parent not in ghosts]
1737
for parent in graph[key]:
1738
# dont examine known nodes again
1743
ghosts.update(this_iteration.difference(found))
1744
if versions.difference(graph):
1745
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1747
result_keys = topo_sort(graph.items())
1749
result_keys = graph.iterkeys()
1750
return [key[0] for key in result_keys]
1752
def get_ancestry_with_ghosts(self, versions):
1753
"""See VersionedFile.get_ancestry."""
1754
if not self._parents:
1755
return self._parentless_ancestry(versions)
1756
# XXX: This will do len(history) index calls - perhaps
1757
# it should be altered to be a index core feature?
1758
# get a graph of all the mentioned versions:
1760
versions = self._version_ids_to_keys(versions)
1761
pending = set(versions)
1763
# get all pending nodes
1764
this_iteration = pending
1765
new_nodes = self._get_entries(this_iteration)
1767
for (index, key, value, node_refs) in new_nodes:
1768
graph[key] = node_refs[0]
1770
for parent in graph[key]:
1771
# dont examine known nodes again
1775
missing_versions = this_iteration.difference(graph)
1776
missing_needed = versions.intersection(missing_versions)
1778
raise RevisionNotPresent(missing_needed.pop(), self)
1779
for missing_version in missing_versions:
1780
# add a key, no parents
1781
graph[missing_version] = []
1782
pending.discard(missing_version) # don't look for it
1783
result_keys = topo_sort(graph.items())
1784
return [key[0] for key in result_keys]
1786
def get_build_details(self, version_ids):
1787
"""Get the method, index_memo and compression parent for version_ids.
1789
Ghosts are omitted from the result.
1791
:param version_ids: An iterable of version_ids.
1792
:return: A dict of version_id:(index_memo, compression_parent,
1793
parents, record_details).
1795
opaque structure to pass to read_records to extract the raw
1798
Content that this record is built upon, may be None
1800
Logical parents of this node
1802
extra information about the content which needs to be passed to
1803
Factory.parse_record
1806
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
1807
for entry in entries:
1808
version_id = self._keys_to_version_ids((entry[1],))[0]
1809
if not self._parents:
1812
parents = self._keys_to_version_ids(entry[3][0])
1813
if not self._deltas:
1814
compression_parent = None
1816
compression_parent_key = self._compression_parent(entry)
1817
if compression_parent_key:
1818
compression_parent = self._keys_to_version_ids(
1819
(compression_parent_key,))[0]
1821
compression_parent = None
1822
noeol = (entry[2][0] == 'N')
1823
if compression_parent:
1824
method = 'line-delta'
1827
result[version_id] = (self._node_to_position(entry),
1828
compression_parent, parents,
1832
def _compression_parent(self, an_entry):
1833
# return the key that an_entry is compressed against, or None
1834
# Grab the second parent list (as deltas implies parents currently)
1835
compression_parents = an_entry[3][1]
1836
if not compression_parents:
1838
assert len(compression_parents) == 1
1839
return compression_parents[0]
1841
def _get_method(self, node):
1842
if not self._deltas:
1844
if self._compression_parent(node):
1849
def get_graph(self):
1850
"""Return a list of the node:parents lists from this knit index."""
1851
if not self._parents:
1852
return [(key, ()) for key in self.get_versions()]
1854
for index, key, value, refs in self._graph_index.iter_all_entries():
1855
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1858
def iter_parents(self, version_ids):
1859
"""Iterate through the parents for many version ids.
1861
:param version_ids: An iterable yielding version_ids.
1862
:return: An iterator that yields (version_id, parents). Requested
1863
version_ids not present in the versioned file are simply skipped.
1864
The order is undefined, allowing for different optimisations in
1865
the underlying implementation.
1868
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1870
present_parents = set()
1871
for node in all_nodes:
1872
all_parents.update(node[3][0])
1873
# any node we are querying must be present
1874
present_parents.add(node[1])
1875
unknown_parents = all_parents.difference(present_parents)
1876
present_parents.update(self._present_keys(unknown_parents))
1877
for node in all_nodes:
1879
for parent in node[3][0]:
1880
if parent in present_parents:
1881
parents.append(parent[0])
1882
yield node[1][0], tuple(parents)
1884
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1885
yield node[1][0], ()
1887
def num_versions(self):
1888
return len(list(self._graph_index.iter_all_entries()))
1890
__len__ = num_versions
1892
def get_versions(self):
1893
"""Get all the versions in the file. not topologically sorted."""
1894
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1896
def has_version(self, version_id):
1897
"""True if the version is in the index."""
1898
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1900
def _keys_to_version_ids(self, keys):
1901
return tuple(key[0] for key in keys)
1903
def get_position(self, version_id):
1904
"""Return details needed to access the version.
1906
:return: a tuple (index, data position, size) to hand to the access
1907
logic to get the record.
1909
node = self._get_node(version_id)
1910
return self._node_to_position(node)
1912
def _node_to_position(self, node):
1913
"""Convert an index value to position details."""
1914
bits = node[2][1:].split(' ')
1915
return node[0], int(bits[0]), int(bits[1])
1917
def get_method(self, version_id):
1918
"""Return compression method of specified version."""
1919
return self._get_method(self._get_node(version_id))
1921
def _get_node(self, version_id):
1923
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1925
raise RevisionNotPresent(version_id, self)
1927
def get_options(self, version_id):
1928
"""Return a list representing options.
1932
node = self._get_node(version_id)
1933
options = [self._get_method(node)]
1934
if node[2][0] == 'N':
1935
options.append('no-eol')
1938
def get_parent_map(self, version_ids):
1939
"""Passed through to by KnitVersionedFile.get_parent_map."""
1940
nodes = self._get_entries(self._version_ids_to_keys(version_ids))
1944
result[node[1][0]] = self._keys_to_version_ids(node[3][0])
1947
result[node[1][0]] = ()
1950
def get_parents(self, version_id):
1951
"""Return parents of specified version ignoring ghosts."""
1952
parents = list(self.iter_parents([version_id]))
1955
raise errors.RevisionNotPresent(version_id, self)
1956
return parents[0][1]
1958
def get_parents_with_ghosts(self, version_id):
1959
"""Return parents of specified version with ghosts."""
1961
return self.get_parent_map([version_id])[version_id]
1963
raise RevisionNotPresent(version_id, self)
1965
def check_versions_present(self, version_ids):
1966
"""Check that all specified versions are present."""
1967
keys = self._version_ids_to_keys(version_ids)
1968
present = self._present_keys(keys)
1969
missing = keys.difference(present)
1971
raise RevisionNotPresent(missing.pop(), self)
1973
def add_version(self, version_id, options, access_memo, parents):
1974
"""Add a version record to the index."""
1975
return self.add_versions(((version_id, options, access_memo, parents),))
1977
def add_versions(self, versions, random_id=False):
1978
"""Add multiple versions to the index.
1980
This function does not insert data into the Immutable GraphIndex
1981
backing the KnitGraphIndex, instead it prepares data for insertion by
1982
the caller and checks that it is safe to insert then calls
1983
self._add_callback with the prepared GraphIndex nodes.
1985
:param versions: a list of tuples:
1986
(version_id, options, pos, size, parents).
1987
:param random_id: If True the ids being added were randomly generated
1988
and no check for existence will be performed.
1990
if not self._add_callback:
1991
raise errors.ReadOnlyError(self)
1992
# we hope there are no repositories with inconsistent parentage
1997
for (version_id, options, access_memo, parents) in versions:
1998
index, pos, size = access_memo
1999
key = (version_id, )
2000
parents = tuple((parent, ) for parent in parents)
2001
if 'no-eol' in options:
2005
value += "%d %d" % (pos, size)
2006
if not self._deltas:
2007
if 'line-delta' in options:
2008
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2011
if 'line-delta' in options:
2012
node_refs = (parents, (parents[0],))
2014
node_refs = (parents, ())
2016
node_refs = (parents, )
2019
raise KnitCorrupt(self, "attempt to add node with parents "
2020
"in parentless index.")
2022
keys[key] = (value, node_refs)
2024
present_nodes = self._get_entries(keys)
2025
for (index, key, value, node_refs) in present_nodes:
2026
if (value, node_refs) != keys[key]:
2027
raise KnitCorrupt(self, "inconsistent details in add_versions"
2028
": %s %s" % ((value, node_refs), keys[key]))
2032
for key, (value, node_refs) in keys.iteritems():
2033
result.append((key, value, node_refs))
2035
for key, (value, node_refs) in keys.iteritems():
2036
result.append((key, value))
2037
self._add_callback(result)
2039
def _version_ids_to_keys(self, version_ids):
2040
return set((version_id, ) for version_id in version_ids)
2043
class _KnitAccess(object):
2044
"""Access to knit records in a .knit file."""
2046
def __init__(self, transport, filename, _file_mode, _dir_mode,
2047
_need_to_create, _create_parent_dir):
2048
"""Create a _KnitAccess for accessing and inserting data.
2050
:param transport: The transport the .knit is located on.
2051
:param filename: The filename of the .knit.
2053
self._transport = transport
2054
self._filename = filename
2055
self._file_mode = _file_mode
2056
self._dir_mode = _dir_mode
2057
self._need_to_create = _need_to_create
2058
self._create_parent_dir = _create_parent_dir
2060
def add_raw_records(self, sizes, raw_data):
2061
"""Add raw knit bytes to a storage area.
2063
The data is spooled to whereever the access method is storing data.
2065
:param sizes: An iterable containing the size of each raw data segment.
2066
:param raw_data: A bytestring containing the data.
2067
:return: A list of memos to retrieve the record later. Each memo is a
2068
tuple - (index, pos, length), where the index field is always None
2069
for the .knit access method.
2071
assert type(raw_data) == str, \
2072
'data must be plain bytes was %s' % type(raw_data)
2073
if not self._need_to_create:
2074
base = self._transport.append_bytes(self._filename, raw_data)
2076
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2077
create_parent_dir=self._create_parent_dir,
2078
mode=self._file_mode,
2079
dir_mode=self._dir_mode)
2080
self._need_to_create = False
2084
result.append((None, base, size))
2089
"""IFF this data access has its own storage area, initialise it.
2093
self._transport.put_bytes_non_atomic(self._filename, '',
2094
mode=self._file_mode)
2096
def open_file(self):
2097
"""IFF this data access can be represented as a single file, open it.
2099
For knits that are not mapped to a single file on disk this will
2102
:return: None or a file handle.
2105
return self._transport.get(self._filename)
2110
def get_raw_records(self, memos_for_retrieval):
2111
"""Get the raw bytes for a records.
2113
:param memos_for_retrieval: An iterable containing the (index, pos,
2114
length) memo for retrieving the bytes. The .knit method ignores
2115
the index as there is always only a single file.
2116
:return: An iterator over the bytes of the records.
2118
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2119
for pos, data in self._transport.readv(self._filename, read_vector):
2123
class _PackAccess(object):
2124
"""Access to knit records via a collection of packs."""
2126
def __init__(self, index_to_packs, writer=None):
2127
"""Create a _PackAccess object.
2129
:param index_to_packs: A dict mapping index objects to the transport
2130
and file names for obtaining data.
2131
:param writer: A tuple (pack.ContainerWriter, write_index) which
2132
contains the pack to write, and the index that reads from it will
2136
self.container_writer = writer[0]
2137
self.write_index = writer[1]
2139
self.container_writer = None
2140
self.write_index = None
2141
self.indices = index_to_packs
2143
def add_raw_records(self, sizes, raw_data):
2144
"""Add raw knit bytes to a storage area.
2146
The data is spooled to the container writer in one bytes-record per
2149
:param sizes: An iterable containing the size of each raw data segment.
2150
:param raw_data: A bytestring containing the data.
2151
:return: A list of memos to retrieve the record later. Each memo is a
2152
tuple - (index, pos, length), where the index field is the
2153
write_index object supplied to the PackAccess object.
2155
assert type(raw_data) == str, \
2156
'data must be plain bytes was %s' % type(raw_data)
2160
p_offset, p_length = self.container_writer.add_bytes_record(
2161
raw_data[offset:offset+size], [])
2163
result.append((self.write_index, p_offset, p_length))
2167
"""Pack based knits do not get individually created."""
2169
def get_raw_records(self, memos_for_retrieval):
2170
"""Get the raw bytes for a records.
2172
:param memos_for_retrieval: An iterable containing the (index, pos,
2173
length) memo for retrieving the bytes. The Pack access method
2174
looks up the pack to use for a given record in its index_to_pack
2176
:return: An iterator over the bytes of the records.
2178
# first pass, group into same-index requests
2180
current_index = None
2181
for (index, offset, length) in memos_for_retrieval:
2182
if current_index == index:
2183
current_list.append((offset, length))
2185
if current_index is not None:
2186
request_lists.append((current_index, current_list))
2187
current_index = index
2188
current_list = [(offset, length)]
2189
# handle the last entry
2190
if current_index is not None:
2191
request_lists.append((current_index, current_list))
2192
for index, offsets in request_lists:
2193
transport, path = self.indices[index]
2194
reader = pack.make_readv_reader(transport, path, offsets)
2195
for names, read_func in reader.iter_records():
2196
yield read_func(None)
2198
def open_file(self):
2199
"""Pack based knits have no single file."""
2202
def set_writer(self, writer, index, (transport, packname)):
2203
"""Set a writer to use for adding data."""
2204
if index is not None:
2205
self.indices[index] = (transport, packname)
2206
self.container_writer = writer
2207
self.write_index = index
2210
class _StreamAccess(object):
2211
"""A Knit Access object that provides data from a datastream.
2213
It also provides a fallback to present as unannotated data, annotated data
2214
from a *backing* access object.
2216
This is triggered by a index_memo which is pointing to a different index
2217
than this was constructed with, and is used to allow extracting full
2218
unannotated texts for insertion into annotated knits.
2221
def __init__(self, reader_callable, stream_index, backing_knit,
2223
"""Create a _StreamAccess object.
2225
:param reader_callable: The reader_callable from the datastream.
2226
This is called to buffer all the data immediately, for
2228
:param stream_index: The index the data stream this provides access to
2229
which will be present in native index_memo's.
2230
:param backing_knit: The knit object that will provide access to
2231
annotated texts which are not available in the stream, so as to
2232
create unannotated texts.
2233
:param orig_factory: The original content factory used to generate the
2234
stream. This is used for checking whether the thunk code for
2235
supporting _copy_texts will generate the correct form of data.
2237
self.data = reader_callable(None)
2238
self.stream_index = stream_index
2239
self.backing_knit = backing_knit
2240
self.orig_factory = orig_factory
2242
def get_raw_records(self, memos_for_retrieval):
2243
"""Get the raw bytes for a records.
2245
:param memos_for_retrieval: An iterable containing the (thunk_flag,
2246
index, start, end) memo for retrieving the bytes.
2247
:return: An iterator over the bytes of the records.
2249
# use a generator for memory friendliness
2250
for thunk_flag, version_id, start, end in memos_for_retrieval:
2251
if version_id is self.stream_index:
2252
yield self.data[start:end]
2254
# we have been asked to thunk. This thunking only occurs when
2255
# we are obtaining plain texts from an annotated backing knit
2256
# so that _copy_texts will work.
2257
# We could improve performance here by scanning for where we need
2258
# to do this and using get_line_list, then interleaving the output
2259
# as desired. However, for now, this is sufficient.
2260
if self.orig_factory.__class__ != KnitPlainFactory:
2261
raise errors.KnitCorrupt(
2262
self, 'Bad thunk request %r' % version_id)
2263
lines = self.backing_knit.get_lines(version_id)
2264
line_bytes = ''.join(lines)
2265
digest = sha_string(line_bytes)
2267
if lines[-1][-1] != '\n':
2268
lines[-1] = lines[-1] + '\n'
2270
orig_options = list(self.backing_knit._index.get_options(version_id))
2271
if 'fulltext' not in orig_options:
2272
if 'line-delta' not in orig_options:
2273
raise errors.KnitCorrupt(self,
2274
'Unknown compression method %r' % orig_options)
2275
orig_options.remove('line-delta')
2276
orig_options.append('fulltext')
2277
# We want plain data, because we expect to thunk only to allow text
2279
size, bytes = self.backing_knit._data._record_to_data(version_id,
2280
digest, lines, line_bytes)
2284
class _StreamIndex(object):
2285
"""A Knit Index object that uses the data map from a datastream."""
2287
def __init__(self, data_list, backing_index):
2288
"""Create a _StreamIndex object.
2290
:param data_list: The data_list from the datastream.
2291
:param backing_index: The index which will supply values for nodes
2292
referenced outside of this stream.
2294
self.data_list = data_list
2295
self.backing_index = backing_index
2296
self._by_version = {}
2298
for key, options, length, parents in data_list:
2299
self._by_version[key] = options, (pos, pos + length), parents
2302
def get_ancestry(self, versions, topo_sorted):
2303
"""Get an ancestry list for versions."""
2305
# Not needed for basic joins
2306
raise NotImplementedError(self.get_ancestry)
2307
# get a graph of all the mentioned versions:
2308
# Little ugly - basically copied from KnitIndex, but don't want to
2309
# accidentally incorporate too much of that index's code.
2311
pending = set(versions)
2312
cache = self._by_version
2314
version = pending.pop()
2317
parents = [p for p in cache[version][2] if p in cache]
2319
raise RevisionNotPresent(version, self)
2320
# if not completed and not a ghost
2321
pending.update([p for p in parents if p not in ancestry])
2322
ancestry.add(version)
2323
return list(ancestry)
2325
def get_build_details(self, version_ids):
2326
"""Get the method, index_memo and compression parent for version_ids.
2328
Ghosts are omitted from the result.
2330
:param version_ids: An iterable of version_ids.
2331
:return: A dict of version_id:(index_memo, compression_parent,
2332
parents, record_details).
2334
opaque structure to pass to read_records to extract the raw
2337
Content that this record is built upon, may be None
2339
Logical parents of this node
2341
extra information about the content which needs to be passed to
2342
Factory.parse_record
2345
for version_id in version_ids:
2347
method = self.get_method(version_id)
2348
except errors.RevisionNotPresent:
2349
# ghosts are omitted
2351
parent_ids = self.get_parents_with_ghosts(version_id)
2352
noeol = ('no-eol' in self.get_options(version_id))
2353
if method == 'fulltext':
2354
compression_parent = None
2356
compression_parent = parent_ids[0]
2357
index_memo = self.get_position(version_id)
2358
result[version_id] = (index_memo, compression_parent,
2359
parent_ids, (method, noeol))
2362
def get_method(self, version_id):
2363
"""Return compression method of specified version."""
2365
options = self._by_version[version_id][0]
2367
# Strictly speaking this should check in the backing knit, but
2368
# until we have a test to discriminate, this will do.
2369
return self.backing_index.get_method(version_id)
2370
if 'fulltext' in options:
2372
elif 'line-delta' in options:
2375
raise errors.KnitIndexUnknownMethod(self, options)
2377
def get_options(self, version_id):
2378
"""Return a list representing options.
2383
return self._by_version[version_id][0]
2385
return self.backing_index.get_options(version_id)
2387
def get_parent_map(self, version_ids):
2388
"""Passed through to by KnitVersionedFile.get_parent_map."""
2391
for version_id in version_ids:
2393
result[version_id] = self._by_version[version_id][2]
2395
pending_ids.add(version_id)
2396
result.update(self.backing_index.get_parent_map(pending_ids))
2399
def get_parents_with_ghosts(self, version_id):
2400
"""Return parents of specified version with ghosts."""
2402
return self.get_parent_map([version_id])[version_id]
2404
raise RevisionNotPresent(version_id, self)
2406
def get_position(self, version_id):
2407
"""Return details needed to access the version.
2409
_StreamAccess has the data as a big array, so we return slice
2410
coordinates into that (as index_memo's are opaque outside the
2411
index and matching access class).
2413
:return: a tuple (thunk_flag, index, start, end). If thunk_flag is
2414
False, index will be self, otherwise it will be a version id.
2417
start, end = self._by_version[version_id][1]
2418
return False, self, start, end
2420
# Signal to the access object to handle this from the backing knit.
2421
return (True, version_id, None, None)
2423
def get_versions(self):
2424
"""Get all the versions in the stream."""
2425
return self._by_version.keys()
2427
def iter_parents(self, version_ids):
2428
"""Iterate through the parents for many version ids.
2430
:param version_ids: An iterable yielding version_ids.
2431
:return: An iterator that yields (version_id, parents). Requested
2432
version_ids not present in the versioned file are simply skipped.
2433
The order is undefined, allowing for different optimisations in
2434
the underlying implementation.
2437
for version in version_ids:
2439
result.append((version, self._by_version[version][2]))
2445
class _KnitData(object):
2446
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2448
The KnitData class provides the logic for parsing and using knit records,
2449
making use of an access method for the low level read and write operations.
2452
def __init__(self, access):
2453
"""Create a KnitData object.
2455
:param access: The access method to use. Access methods such as
2456
_KnitAccess manage the insertion of raw records and the subsequent
2457
retrieval of the same.
2459
self._access = access
2460
self._checked = False
2461
# TODO: jam 20060713 conceptually, this could spill to disk
2462
# if the cached size gets larger than a certain amount
2463
# but it complicates the model a bit, so for now just use
2464
# a simple dictionary
2466
self._do_cache = False
2468
def enable_cache(self):
2469
"""Enable caching of reads."""
2470
self._do_cache = True
2472
def clear_cache(self):
2473
"""Clear the record cache."""
2474
self._do_cache = False
2477
def _open_file(self):
2478
return self._access.open_file()
2480
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2481
"""Convert version_id, digest, lines into a raw data block.
2483
:param dense_lines: The bytes of lines but in a denser form. For
2484
instance, if lines is a list of 1000 bytestrings each ending in \n,
2485
dense_lines may be a list with one line in it, containing all the
2486
1000's lines and their \n's. Using dense_lines if it is already
2487
known is a win because the string join to create bytes in this
2488
function spends less time resizing the final string.
2489
:return: (len, a StringIO instance with the raw data ready to read.)
2491
# Note: using a string copy here increases memory pressure with e.g.
2492
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2493
# when doing the initial commit of a mozilla tree. RBC 20070921
2494
bytes = ''.join(chain(
2495
["version %s %d %s\n" % (version_id,
2498
dense_lines or lines,
2499
["end %s\n" % version_id]))
2500
assert bytes.__class__ == str
2501
compressed_bytes = bytes_to_gzip(bytes)
2502
return len(compressed_bytes), compressed_bytes
2504
def add_raw_records(self, sizes, raw_data):
2505
"""Append a prepared record to the data file.
2507
:param sizes: An iterable containing the size of each raw data segment.
2508
:param raw_data: A bytestring containing the data.
2509
:return: a list of index data for the way the data was stored.
2510
See the access method add_raw_records documentation for more
2513
return self._access.add_raw_records(sizes, raw_data)
2515
def _parse_record_header(self, version_id, raw_data):
2516
"""Parse a record header for consistency.
2518
:return: the header and the decompressor stream.
2519
as (stream, header_record)
2521
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2523
rec = self._check_header(version_id, df.readline())
2524
except Exception, e:
2525
raise KnitCorrupt(self._access,
2526
"While reading {%s} got %s(%s)"
2527
% (version_id, e.__class__.__name__, str(e)))
2530
def _check_header(self, version_id, line):
2533
raise KnitCorrupt(self._access,
2534
'unexpected number of elements in record header')
2535
if rec[1] != version_id:
2536
raise KnitCorrupt(self._access,
2537
'unexpected version, wanted %r, got %r'
2538
% (version_id, rec[1]))
2541
def _parse_record(self, version_id, data):
2543
# 4168 calls in 2880 217 internal
2544
# 4168 calls to _parse_record_header in 2121
2545
# 4168 calls to readlines in 330
2546
df = GzipFile(mode='rb', fileobj=StringIO(data))
2549
record_contents = df.readlines()
2550
except Exception, e:
2551
raise KnitCorrupt(self._access,
2552
"While reading {%s} got %s(%s)"
2553
% (version_id, e.__class__.__name__, str(e)))
2554
header = record_contents.pop(0)
2555
rec = self._check_header(version_id, header)
2557
last_line = record_contents.pop()
2558
if len(record_contents) != int(rec[2]):
2559
raise KnitCorrupt(self._access,
2560
'incorrect number of lines %s != %s'
2562
% (len(record_contents), int(rec[2]),
2564
if last_line != 'end %s\n' % rec[1]:
2565
raise KnitCorrupt(self._access,
2566
'unexpected version end line %r, wanted %r'
2567
% (last_line, version_id))
2569
return record_contents, rec[3]
2571
def read_records_iter_raw(self, records):
2572
"""Read text records from data file and yield raw data.
2574
This unpacks enough of the text record to validate the id is
2575
as expected but thats all.
2577
# setup an iterator of the external records:
2578
# uses readv so nice and fast we hope.
2580
# grab the disk data needed.
2582
# Don't check _cache if it is empty
2583
needed_offsets = [index_memo for version_id, index_memo
2585
if version_id not in self._cache]
2587
needed_offsets = [index_memo for version_id, index_memo
2590
raw_records = self._access.get_raw_records(needed_offsets)
2592
for version_id, index_memo in records:
2593
if version_id in self._cache:
2594
# This data has already been validated
2595
data = self._cache[version_id]
2597
data = raw_records.next()
2599
self._cache[version_id] = data
2601
# validate the header
2602
df, rec = self._parse_record_header(version_id, data)
2604
yield version_id, data
2606
def read_records_iter(self, records):
2607
"""Read text records from data file and yield result.
2609
The result will be returned in whatever is the fastest to read.
2610
Not by the order requested. Also, multiple requests for the same
2611
record will only yield 1 response.
2612
:param records: A list of (version_id, pos, len) entries
2613
:return: Yields (version_id, contents, digest) in the order
2614
read, not the order requested
2620
# Skip records we have alread seen
2621
yielded_records = set()
2622
needed_records = set()
2623
for record in records:
2624
if record[0] in self._cache:
2625
if record[0] in yielded_records:
2627
yielded_records.add(record[0])
2628
data = self._cache[record[0]]
2629
content, digest = self._parse_record(record[0], data)
2630
yield (record[0], content, digest)
2632
needed_records.add(record)
2633
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2635
needed_records = sorted(set(records), key=operator.itemgetter(1))
2637
if not needed_records:
2640
# The transport optimizes the fetching as well
2641
# (ie, reads continuous ranges.)
2642
raw_data = self._access.get_raw_records(
2643
[index_memo for version_id, index_memo in needed_records])
2645
for (version_id, index_memo), data in \
2646
izip(iter(needed_records), raw_data):
2647
content, digest = self._parse_record(version_id, data)
2649
self._cache[version_id] = data
2650
yield version_id, content, digest
2652
def read_records(self, records):
2653
"""Read records into a dictionary."""
2655
for record_id, content, digest in \
2656
self.read_records_iter(records):
2657
components[record_id] = (content, digest)
2661
class InterKnit(InterVersionedFile):
2662
"""Optimised code paths for knit to knit operations."""
2664
_matching_file_from_factory = KnitVersionedFile
2665
_matching_file_to_factory = KnitVersionedFile
2668
def is_compatible(source, target):
2669
"""Be compatible with knits. """
2671
return (isinstance(source, KnitVersionedFile) and
2672
isinstance(target, KnitVersionedFile))
2673
except AttributeError:
2676
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2677
"""Copy texts to the target by extracting and adding them one by one.
2679
see join() for the parameter definitions.
2681
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2682
graph = self.source.get_graph(version_ids)
2683
order = topo_sort(graph.items())
2685
def size_of_content(content):
2686
return sum(len(line) for line in content.text())
2687
# Cache at most 10MB of parent texts
2688
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2689
compute_size=size_of_content)
2690
# TODO: jam 20071116 It would be nice to have a streaming interface to
2691
# get multiple texts from a source. The source could be smarter
2692
# about how it handled intermediate stages.
2693
# get_line_list() or make_mpdiffs() seem like a possibility, but
2694
# at the moment they extract all full texts into memory, which
2695
# causes us to store more than our 3x fulltext goal.
2696
# Repository.iter_files_bytes() may be another possibility
2697
to_process = [version for version in order
2698
if version not in self.target]
2699
total = len(to_process)
2700
pb = ui.ui_factory.nested_progress_bar()
2702
for index, version in enumerate(to_process):
2703
pb.update('Converting versioned data', index, total)
2704
sha1, num_bytes, parent_text = self.target.add_lines(version,
2705
self.source.get_parents_with_ghosts(version),
2706
self.source.get_lines(version),
2707
parent_texts=parent_cache)
2708
parent_cache[version] = parent_text
2713
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2714
"""See InterVersionedFile.join."""
2715
assert isinstance(self.source, KnitVersionedFile)
2716
assert isinstance(self.target, KnitVersionedFile)
2718
# If the source and target are mismatched w.r.t. annotations vs
2719
# plain, the data needs to be converted accordingly
2720
if self.source.factory.annotated == self.target.factory.annotated:
2722
elif self.source.factory.annotated:
2723
converter = self._anno_to_plain_converter
2725
# We're converting from a plain to an annotated knit. Copy them
2726
# across by full texts.
2727
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2729
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2733
pb = ui.ui_factory.nested_progress_bar()
2735
version_ids = list(version_ids)
2736
if None in version_ids:
2737
version_ids.remove(None)
2739
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2741
this_versions = set(self.target._index.get_versions())
2742
# XXX: For efficiency we should not look at the whole index,
2743
# we only need to consider the referenced revisions - they
2744
# must all be present, or the method must be full-text.
2745
# TODO, RBC 20070919
2746
needed_versions = self.source_ancestry - this_versions
2748
if not needed_versions:
2750
full_list = topo_sort(self.source.get_graph())
2752
version_list = [i for i in full_list if (not self.target.has_version(i)
2753
and i in needed_versions)]
2757
copy_queue_records = []
2759
for version_id in version_list:
2760
options = self.source._index.get_options(version_id)
2761
parents = self.source._index.get_parents_with_ghosts(version_id)
2762
# check that its will be a consistent copy:
2763
for parent in parents:
2764
# if source has the parent, we must :
2765
# * already have it or
2766
# * have it scheduled already
2767
# otherwise we don't care
2768
assert (self.target.has_version(parent) or
2769
parent in copy_set or
2770
not self.source.has_version(parent))
2771
index_memo = self.source._index.get_position(version_id)
2772
copy_queue_records.append((version_id, index_memo))
2773
copy_queue.append((version_id, options, parents))
2774
copy_set.add(version_id)
2776
# data suck the join:
2778
total = len(version_list)
2781
for (version_id, raw_data), \
2782
(version_id2, options, parents) in \
2783
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2785
assert version_id == version_id2, 'logic error, inconsistent results'
2787
pb.update("Joining knit", count, total)
2789
size, raw_data = converter(raw_data, version_id, options,
2792
size = len(raw_data)
2793
raw_records.append((version_id, options, parents, size))
2794
raw_datum.append(raw_data)
2795
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2800
def _anno_to_plain_converter(self, raw_data, version_id, options,
2802
"""Convert annotated content to plain content."""
2803
data, digest = self.source._data._parse_record(version_id, raw_data)
2804
if 'fulltext' in options:
2805
content = self.source.factory.parse_fulltext(data, version_id)
2806
lines = self.target.factory.lower_fulltext(content)
2808
delta = self.source.factory.parse_line_delta(data, version_id,
2810
lines = self.target.factory.lower_line_delta(delta)
2811
return self.target._data._record_to_data(version_id, digest, lines)
2814
InterVersionedFile.register_optimiser(InterKnit)
2817
class WeaveToKnit(InterVersionedFile):
2818
"""Optimised code paths for weave to knit operations."""
2820
_matching_file_from_factory = bzrlib.weave.WeaveFile
2821
_matching_file_to_factory = KnitVersionedFile
2824
def is_compatible(source, target):
2825
"""Be compatible with weaves to knits."""
2827
return (isinstance(source, bzrlib.weave.Weave) and
2828
isinstance(target, KnitVersionedFile))
2829
except AttributeError:
2832
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2833
"""See InterVersionedFile.join."""
2834
assert isinstance(self.source, bzrlib.weave.Weave)
2835
assert isinstance(self.target, KnitVersionedFile)
2837
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2842
pb = ui.ui_factory.nested_progress_bar()
2844
version_ids = list(version_ids)
2846
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2847
this_versions = set(self.target._index.get_versions())
2848
needed_versions = self.source_ancestry - this_versions
2850
if not needed_versions:
2852
full_list = topo_sort(self.source.get_graph())
2854
version_list = [i for i in full_list if (not self.target.has_version(i)
2855
and i in needed_versions)]
2859
total = len(version_list)
2860
parent_map = self.source.get_parent_map(version_list)
2861
for version_id in version_list:
2862
pb.update("Converting to knit", count, total)
2863
parents = parent_map[version_id]
2864
# check that its will be a consistent copy:
2865
for parent in parents:
2866
# if source has the parent, we must already have it
2867
assert (self.target.has_version(parent))
2868
self.target.add_lines(
2869
version_id, parents, self.source.get_lines(version_id))
2876
InterVersionedFile.register_optimiser(WeaveToKnit)
2879
# Deprecated, use PatienceSequenceMatcher instead
2880
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2883
def annotate_knit(knit, revision_id):
2884
"""Annotate a knit with no cached annotations.
2886
This implementation is for knits with no cached annotations.
2887
It will work for knits with cached annotations, but this is not
2890
annotator = _KnitAnnotator(knit)
2891
return iter(annotator.annotate(revision_id))
2894
class _KnitAnnotator(object):
2895
"""Build up the annotations for a text."""
2897
def __init__(self, knit):
2900
# Content objects, differs from fulltexts because of how final newlines
2901
# are treated by knits. the content objects here will always have a
2903
self._fulltext_contents = {}
2905
# Annotated lines of specific revisions
2906
self._annotated_lines = {}
2908
# Track the raw data for nodes that we could not process yet.
2909
# This maps the revision_id of the base to a list of children that will
2910
# annotated from it.
2911
self._pending_children = {}
2913
# Nodes which cannot be extracted
2914
self._ghosts = set()
2916
# Track how many children this node has, so we know if we need to keep
2918
self._annotate_children = {}
2919
self._compression_children = {}
2921
self._all_build_details = {}
2922
# The children => parent revision_id graph
2923
self._revision_id_graph = {}
2925
self._heads_provider = None
2927
self._nodes_to_keep_annotations = set()
2928
self._generations_until_keep = 100
2930
def set_generations_until_keep(self, value):
2931
"""Set the number of generations before caching a node.
2933
Setting this to -1 will cache every merge node, setting this higher
2934
will cache fewer nodes.
2936
self._generations_until_keep = value
2938
def _add_fulltext_content(self, revision_id, content_obj):
2939
self._fulltext_contents[revision_id] = content_obj
2940
# TODO: jam 20080305 It might be good to check the sha1digest here
2941
return content_obj.text()
2943
def _check_parents(self, child, nodes_to_annotate):
2944
"""Check if all parents have been processed.
2946
:param child: A tuple of (rev_id, parents, raw_content)
2947
:param nodes_to_annotate: If child is ready, add it to
2948
nodes_to_annotate, otherwise put it back in self._pending_children
2950
for parent_id in child[1]:
2951
if (parent_id not in self._annotated_lines):
2952
# This parent is present, but another parent is missing
2953
self._pending_children.setdefault(parent_id,
2957
# This one is ready to be processed
2958
nodes_to_annotate.append(child)
2960
def _add_annotation(self, revision_id, fulltext, parent_ids,
2961
left_matching_blocks=None):
2962
"""Add an annotation entry.
2964
All parents should already have been annotated.
2965
:return: A list of children that now have their parents satisfied.
2967
a = self._annotated_lines
2968
annotated_parent_lines = [a[p] for p in parent_ids]
2969
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2970
fulltext, revision_id, left_matching_blocks,
2971
heads_provider=self._get_heads_provider()))
2972
self._annotated_lines[revision_id] = annotated_lines
2973
for p in parent_ids:
2974
ann_children = self._annotate_children[p]
2975
ann_children.remove(revision_id)
2976
if (not ann_children
2977
and p not in self._nodes_to_keep_annotations):
2978
del self._annotated_lines[p]
2979
del self._all_build_details[p]
2980
if p in self._fulltext_contents:
2981
del self._fulltext_contents[p]
2982
# Now that we've added this one, see if there are any pending
2983
# deltas to be done, certainly this parent is finished
2984
nodes_to_annotate = []
2985
for child in self._pending_children.pop(revision_id, []):
2986
self._check_parents(child, nodes_to_annotate)
2987
return nodes_to_annotate
2989
def _get_build_graph(self, revision_id):
2990
"""Get the graphs for building texts and annotations.
2992
The data you need for creating a full text may be different than the
2993
data you need to annotate that text. (At a minimum, you need both
2994
parents to create an annotation, but only need 1 parent to generate the
2997
:return: A list of (revision_id, index_memo) records, suitable for
2998
passing to read_records_iter to start reading in the raw data fro/
3001
if revision_id in self._annotated_lines:
3004
pending = set([revision_id])
3009
# get all pending nodes
3011
this_iteration = pending
3012
build_details = self._knit._index.get_build_details(this_iteration)
3013
self._all_build_details.update(build_details)
3014
# new_nodes = self._knit._index._get_entries(this_iteration)
3016
for rev_id, details in build_details.iteritems():
3017
(index_memo, compression_parent, parents,
3018
record_details) = details
3019
self._revision_id_graph[rev_id] = parents
3020
records.append((rev_id, index_memo))
3021
# Do we actually need to check _annotated_lines?
3022
pending.update(p for p in parents
3023
if p not in self._all_build_details)
3024
if compression_parent:
3025
self._compression_children.setdefault(compression_parent,
3028
for parent in parents:
3029
self._annotate_children.setdefault(parent,
3031
num_gens = generation - kept_generation
3032
if ((num_gens >= self._generations_until_keep)
3033
and len(parents) > 1):
3034
kept_generation = generation
3035
self._nodes_to_keep_annotations.add(rev_id)
3037
missing_versions = this_iteration.difference(build_details.keys())
3038
self._ghosts.update(missing_versions)
3039
for missing_version in missing_versions:
3040
# add a key, no parents
3041
self._revision_id_graph[missing_version] = ()
3042
pending.discard(missing_version) # don't look for it
3043
# XXX: This should probably be a real exception, as it is a data
3045
assert not self._ghosts.intersection(self._compression_children), \
3046
"We cannot have nodes which have a compression parent of a ghost."
3047
# Cleanout anything that depends on a ghost so that we don't wait for
3048
# the ghost to show up
3049
for node in self._ghosts:
3050
if node in self._annotate_children:
3051
# We won't be building this node
3052
del self._annotate_children[node]
3053
# Generally we will want to read the records in reverse order, because
3054
# we find the parent nodes after the children
3058
def _annotate_records(self, records):
3059
"""Build the annotations for the listed records."""
3060
# We iterate in the order read, rather than a strict order requested
3061
# However, process what we can, and put off to the side things that
3062
# still need parents, cleaning them up when those parents are
3064
for (rev_id, record,
3065
digest) in self._knit._data.read_records_iter(records):
3066
if rev_id in self._annotated_lines:
3068
parent_ids = self._revision_id_graph[rev_id]
3069
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3070
details = self._all_build_details[rev_id]
3071
(index_memo, compression_parent, parents,
3072
record_details) = details
3073
nodes_to_annotate = []
3074
# TODO: Remove the punning between compression parents, and
3075
# parent_ids, we should be able to do this without assuming
3077
if len(parent_ids) == 0:
3078
# There are no parents for this node, so just add it
3079
# TODO: This probably needs to be decoupled
3080
assert compression_parent is None
3081
fulltext_content, delta = self._knit.factory.parse_record(
3082
rev_id, record, record_details, None)
3083
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3084
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3085
parent_ids, left_matching_blocks=None))
3087
child = (rev_id, parent_ids, record)
3088
# Check if all the parents are present
3089
self._check_parents(child, nodes_to_annotate)
3090
while nodes_to_annotate:
3091
# Should we use a queue here instead of a stack?
3092
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3093
(index_memo, compression_parent, parents,
3094
record_details) = self._all_build_details[rev_id]
3095
if compression_parent is not None:
3096
comp_children = self._compression_children[compression_parent]
3097
assert rev_id in comp_children
3098
# If there is only 1 child, it is safe to reuse this
3100
reuse_content = (len(comp_children) == 1
3101
and compression_parent not in
3102
self._nodes_to_keep_annotations)
3104
# Remove it from the cache since it will be changing
3105
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3106
# Make sure to copy the fulltext since it might be
3108
parent_fulltext = list(parent_fulltext_content.text())
3110
parent_fulltext_content = self._fulltext_contents[compression_parent]
3111
parent_fulltext = parent_fulltext_content.text()
3112
comp_children.remove(rev_id)
3113
fulltext_content, delta = self._knit.factory.parse_record(
3114
rev_id, record, record_details,
3115
parent_fulltext_content,
3116
copy_base_content=(not reuse_content))
3117
fulltext = self._add_fulltext_content(rev_id,
3119
blocks = KnitContent.get_line_delta_blocks(delta,
3120
parent_fulltext, fulltext)
3122
fulltext_content = self._knit.factory.parse_fulltext(
3124
fulltext = self._add_fulltext_content(rev_id,
3127
nodes_to_annotate.extend(
3128
self._add_annotation(rev_id, fulltext, parent_ids,
3129
left_matching_blocks=blocks))
3131
def _get_heads_provider(self):
3132
"""Create a heads provider for resolving ancestry issues."""
3133
if self._heads_provider is not None:
3134
return self._heads_provider
3135
parent_provider = _mod_graph.DictParentsProvider(
3136
self._revision_id_graph)
3137
graph_obj = _mod_graph.Graph(parent_provider)
3138
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3139
self._heads_provider = head_cache
3142
def annotate(self, revision_id):
3143
"""Return the annotated fulltext at the given revision.
3145
:param revision_id: The revision id for this file
3147
records = self._get_build_graph(revision_id)
3148
if revision_id in self._ghosts:
3149
raise errors.RevisionNotPresent(revision_id, self._knit)
3150
self._annotate_records(records)
3151
return self._annotated_lines[revision_id]
3155
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3157
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
