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."""
1264
lookup = self._index.get_parents_with_ghosts
1265
for version_id in version_ids:
1267
result[version_id] = tuple(lookup(version_id))
1268
except (KeyError, RevisionNotPresent):
1272
def get_ancestry(self, versions, topo_sorted=True):
1273
"""See VersionedFile.get_ancestry."""
1274
if isinstance(versions, basestring):
1275
versions = [versions]
1278
return self._index.get_ancestry(versions, topo_sorted)
1280
def get_ancestry_with_ghosts(self, versions):
1281
"""See VersionedFile.get_ancestry_with_ghosts."""
1282
if isinstance(versions, basestring):
1283
versions = [versions]
1286
return self._index.get_ancestry_with_ghosts(versions)
1288
def plan_merge(self, ver_a, ver_b):
1289
"""See VersionedFile.plan_merge."""
1290
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1291
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1292
annotated_a = self.annotate(ver_a)
1293
annotated_b = self.annotate(ver_b)
1294
return merge._plan_annotate_merge(annotated_a, annotated_b,
1295
ancestors_a, ancestors_b)
1298
class _KnitComponentFile(object):
1299
"""One of the files used to implement a knit database"""
1301
def __init__(self, transport, filename, mode, file_mode=None,
1302
create_parent_dir=False, dir_mode=None):
1303
self._transport = transport
1304
self._filename = filename
1306
self._file_mode = file_mode
1307
self._dir_mode = dir_mode
1308
self._create_parent_dir = create_parent_dir
1309
self._need_to_create = False
1311
def _full_path(self):
1312
"""Return the full path to this file."""
1313
return self._transport.base + self._filename
1315
def check_header(self, fp):
1316
line = fp.readline()
1318
# An empty file can actually be treated as though the file doesn't
1320
raise errors.NoSuchFile(self._full_path())
1321
if line != self.HEADER:
1322
raise KnitHeaderError(badline=line,
1323
filename=self._transport.abspath(self._filename))
1326
return '%s(%s)' % (self.__class__.__name__, self._filename)
1329
class _KnitIndex(_KnitComponentFile):
1330
"""Manages knit index file.
1332
The index is already kept in memory and read on startup, to enable
1333
fast lookups of revision information. The cursor of the index
1334
file is always pointing to the end, making it easy to append
1337
_cache is a cache for fast mapping from version id to a Index
1340
_history is a cache for fast mapping from indexes to version ids.
1342
The index data format is dictionary compressed when it comes to
1343
parent references; a index entry may only have parents that with a
1344
lover index number. As a result, the index is topological sorted.
1346
Duplicate entries may be written to the index for a single version id
1347
if this is done then the latter one completely replaces the former:
1348
this allows updates to correct version and parent information.
1349
Note that the two entries may share the delta, and that successive
1350
annotations and references MUST point to the first entry.
1352
The index file on disc contains a header, followed by one line per knit
1353
record. The same revision can be present in an index file more than once.
1354
The first occurrence gets assigned a sequence number starting from 0.
1356
The format of a single line is
1357
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1358
REVISION_ID is a utf8-encoded revision id
1359
FLAGS is a comma separated list of flags about the record. Values include
1360
no-eol, line-delta, fulltext.
1361
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1362
that the the compressed data starts at.
1363
LENGTH is the ascii representation of the length of the data file.
1364
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1366
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1367
revision id already in the knit that is a parent of REVISION_ID.
1368
The ' :' marker is the end of record marker.
1371
when a write is interrupted to the index file, it will result in a line
1372
that does not end in ' :'. If the ' :' is not present at the end of a line,
1373
or at the end of the file, then the record that is missing it will be
1374
ignored by the parser.
1376
When writing new records to the index file, the data is preceded by '\n'
1377
to ensure that records always start on new lines even if the last write was
1378
interrupted. As a result its normal for the last line in the index to be
1379
missing a trailing newline. One can be added with no harmful effects.
1382
HEADER = "# bzr knit index 8\n"
1384
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1385
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1387
def _cache_version(self, version_id, options, pos, size, parents):
1388
"""Cache a version record in the history array and index cache.
1390
This is inlined into _load_data for performance. KEEP IN SYNC.
1391
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1394
# only want the _history index to reference the 1st index entry
1396
if version_id not in self._cache:
1397
index = len(self._history)
1398
self._history.append(version_id)
1400
index = self._cache[version_id][5]
1401
self._cache[version_id] = (version_id,
1408
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1409
create_parent_dir=False, delay_create=False, dir_mode=None):
1410
_KnitComponentFile.__init__(self, transport, filename, mode,
1411
file_mode=file_mode,
1412
create_parent_dir=create_parent_dir,
1415
# position in _history is the 'official' index for a revision
1416
# but the values may have come from a newer entry.
1417
# so - wc -l of a knit index is != the number of unique names
1421
fp = self._transport.get(self._filename)
1423
# _load_data may raise NoSuchFile if the target knit is
1425
_load_data(self, fp)
1429
if mode != 'w' or not create:
1432
self._need_to_create = True
1434
self._transport.put_bytes_non_atomic(
1435
self._filename, self.HEADER, mode=self._file_mode)
1437
def get_graph(self):
1438
"""Return a list of the node:parents lists from this knit index."""
1439
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1441
def get_ancestry(self, versions, topo_sorted=True):
1442
"""See VersionedFile.get_ancestry."""
1443
# get a graph of all the mentioned versions:
1445
pending = set(versions)
1448
version = pending.pop()
1451
parents = [p for p in cache[version][4] if p in cache]
1453
raise RevisionNotPresent(version, self._filename)
1454
# if not completed and not a ghost
1455
pending.update([p for p in parents if p not in graph])
1456
graph[version] = parents
1459
return topo_sort(graph.items())
1461
def get_ancestry_with_ghosts(self, versions):
1462
"""See VersionedFile.get_ancestry_with_ghosts."""
1463
# get a graph of all the mentioned versions:
1464
self.check_versions_present(versions)
1467
pending = set(versions)
1469
version = pending.pop()
1471
parents = cache[version][4]
1477
pending.update([p for p in parents if p not in graph])
1478
graph[version] = parents
1479
return topo_sort(graph.items())
1481
def get_build_details(self, version_ids):
1482
"""Get the method, index_memo and compression parent for version_ids.
1484
Ghosts are omitted from the result.
1486
:param version_ids: An iterable of version_ids.
1487
:return: A dict of version_id:(index_memo, compression_parent,
1488
parents, record_details).
1490
opaque structure to pass to read_records to extract the raw
1493
Content that this record is built upon, may be None
1495
Logical parents of this node
1497
extra information about the content which needs to be passed to
1498
Factory.parse_record
1501
for version_id in version_ids:
1502
if version_id not in self._cache:
1503
# ghosts are omitted
1505
method = self.get_method(version_id)
1506
parents = self.get_parents_with_ghosts(version_id)
1507
if method == 'fulltext':
1508
compression_parent = None
1510
compression_parent = parents[0]
1511
noeol = 'no-eol' in self.get_options(version_id)
1512
index_memo = self.get_position(version_id)
1513
result[version_id] = (index_memo, compression_parent,
1514
parents, (method, noeol))
1517
def iter_parents(self, version_ids):
1518
"""Iterate through the parents for many version ids.
1520
:param version_ids: An iterable yielding version_ids.
1521
:return: An iterator that yields (version_id, parents). Requested
1522
version_ids not present in the versioned file are simply skipped.
1523
The order is undefined, allowing for different optimisations in
1524
the underlying implementation.
1526
for version_id in version_ids:
1528
yield version_id, tuple(self.get_parents(version_id))
1532
def num_versions(self):
1533
return len(self._history)
1535
__len__ = num_versions
1537
def get_versions(self):
1538
"""Get all the versions in the file. not topologically sorted."""
1539
return self._history
1541
def _version_list_to_index(self, versions):
1544
for version in versions:
1545
if version in cache:
1546
# -- inlined lookup() --
1547
result_list.append(str(cache[version][5]))
1548
# -- end lookup () --
1550
result_list.append('.' + version)
1551
return ' '.join(result_list)
1553
def add_version(self, version_id, options, index_memo, parents):
1554
"""Add a version record to the index."""
1555
self.add_versions(((version_id, options, index_memo, parents),))
1557
def add_versions(self, versions, random_id=False):
1558
"""Add multiple versions to the index.
1560
:param versions: a list of tuples:
1561
(version_id, options, pos, size, parents).
1562
:param random_id: If True the ids being added were randomly generated
1563
and no check for existence will be performed.
1566
orig_history = self._history[:]
1567
orig_cache = self._cache.copy()
1570
for version_id, options, (index, pos, size), parents in versions:
1571
line = "\n%s %s %s %s %s :" % (version_id,
1575
self._version_list_to_index(parents))
1576
assert isinstance(line, str), \
1577
'content must be utf-8 encoded: %r' % (line,)
1579
self._cache_version(version_id, options, pos, size, parents)
1580
if not self._need_to_create:
1581
self._transport.append_bytes(self._filename, ''.join(lines))
1584
sio.write(self.HEADER)
1585
sio.writelines(lines)
1587
self._transport.put_file_non_atomic(self._filename, sio,
1588
create_parent_dir=self._create_parent_dir,
1589
mode=self._file_mode,
1590
dir_mode=self._dir_mode)
1591
self._need_to_create = False
1593
# If any problems happen, restore the original values and re-raise
1594
self._history = orig_history
1595
self._cache = orig_cache
1598
def has_version(self, version_id):
1599
"""True if the version is in the index."""
1600
return version_id in self._cache
1602
def get_position(self, version_id):
1603
"""Return details needed to access the version.
1605
.kndx indices do not support split-out data, so return None for the
1608
:return: a tuple (None, data position, size) to hand to the access
1609
logic to get the record.
1611
entry = self._cache[version_id]
1612
return None, entry[2], entry[3]
1614
def get_method(self, version_id):
1615
"""Return compression method of specified version."""
1617
options = self._cache[version_id][1]
1619
raise RevisionNotPresent(version_id, self._filename)
1620
if 'fulltext' in options:
1623
if 'line-delta' not in options:
1624
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1627
def get_options(self, version_id):
1628
"""Return a list representing options.
1632
return self._cache[version_id][1]
1634
def get_parents(self, version_id):
1635
"""Return parents of specified version ignoring ghosts."""
1636
return [parent for parent in self._cache[version_id][4]
1637
if parent in self._cache]
1639
def get_parents_with_ghosts(self, version_id):
1640
"""Return parents of specified version with ghosts."""
1641
return self._cache[version_id][4]
1643
def check_versions_present(self, version_ids):
1644
"""Check that all specified versions are present."""
1646
for version_id in version_ids:
1647
if version_id not in cache:
1648
raise RevisionNotPresent(version_id, self._filename)
1651
class KnitGraphIndex(object):
1652
"""A knit index that builds on GraphIndex."""
1654
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1655
"""Construct a KnitGraphIndex on a graph_index.
1657
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1658
:param deltas: Allow delta-compressed records.
1659
:param add_callback: If not None, allow additions to the index and call
1660
this callback with a list of added GraphIndex nodes:
1661
[(node, value, node_refs), ...]
1662
:param parents: If True, record knits parents, if not do not record
1665
self._graph_index = graph_index
1666
self._deltas = deltas
1667
self._add_callback = add_callback
1668
self._parents = parents
1669
if deltas and not parents:
1670
raise KnitCorrupt(self, "Cannot do delta compression without "
1673
def _get_entries(self, keys, check_present=False):
1674
"""Get the entries for keys.
1676
:param keys: An iterable of index keys, - 1-tuples.
1681
for node in self._graph_index.iter_entries(keys):
1683
found_keys.add(node[1])
1685
# adapt parentless index to the rest of the code.
1686
for node in self._graph_index.iter_entries(keys):
1687
yield node[0], node[1], node[2], ()
1688
found_keys.add(node[1])
1690
missing_keys = keys.difference(found_keys)
1692
raise RevisionNotPresent(missing_keys.pop(), self)
1694
def _present_keys(self, version_ids):
1696
node[1] for node in self._get_entries(version_ids)])
1698
def _parentless_ancestry(self, versions):
1699
"""Honour the get_ancestry API for parentless knit indices."""
1700
wanted_keys = self._version_ids_to_keys(versions)
1701
present_keys = self._present_keys(wanted_keys)
1702
missing = set(wanted_keys).difference(present_keys)
1704
raise RevisionNotPresent(missing.pop(), self)
1705
return list(self._keys_to_version_ids(present_keys))
1707
def get_ancestry(self, versions, topo_sorted=True):
1708
"""See VersionedFile.get_ancestry."""
1709
if not self._parents:
1710
return self._parentless_ancestry(versions)
1711
# XXX: This will do len(history) index calls - perhaps
1712
# it should be altered to be a index core feature?
1713
# get a graph of all the mentioned versions:
1716
versions = self._version_ids_to_keys(versions)
1717
pending = set(versions)
1719
# get all pending nodes
1720
this_iteration = pending
1721
new_nodes = self._get_entries(this_iteration)
1724
for (index, key, value, node_refs) in new_nodes:
1725
# dont ask for ghosties - otherwise
1726
# we we can end up looping with pending
1727
# being entirely ghosted.
1728
graph[key] = [parent for parent in node_refs[0]
1729
if parent not in ghosts]
1731
for parent in graph[key]:
1732
# dont examine known nodes again
1737
ghosts.update(this_iteration.difference(found))
1738
if versions.difference(graph):
1739
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1741
result_keys = topo_sort(graph.items())
1743
result_keys = graph.iterkeys()
1744
return [key[0] for key in result_keys]
1746
def get_ancestry_with_ghosts(self, versions):
1747
"""See VersionedFile.get_ancestry."""
1748
if not self._parents:
1749
return self._parentless_ancestry(versions)
1750
# XXX: This will do len(history) index calls - perhaps
1751
# it should be altered to be a index core feature?
1752
# get a graph of all the mentioned versions:
1754
versions = self._version_ids_to_keys(versions)
1755
pending = set(versions)
1757
# get all pending nodes
1758
this_iteration = pending
1759
new_nodes = self._get_entries(this_iteration)
1761
for (index, key, value, node_refs) in new_nodes:
1762
graph[key] = node_refs[0]
1764
for parent in graph[key]:
1765
# dont examine known nodes again
1769
missing_versions = this_iteration.difference(graph)
1770
missing_needed = versions.intersection(missing_versions)
1772
raise RevisionNotPresent(missing_needed.pop(), self)
1773
for missing_version in missing_versions:
1774
# add a key, no parents
1775
graph[missing_version] = []
1776
pending.discard(missing_version) # don't look for it
1777
result_keys = topo_sort(graph.items())
1778
return [key[0] for key in result_keys]
1780
def get_build_details(self, version_ids):
1781
"""Get the method, index_memo and compression parent for version_ids.
1783
Ghosts are omitted from the result.
1785
:param version_ids: An iterable of version_ids.
1786
:return: A dict of version_id:(index_memo, compression_parent,
1787
parents, record_details).
1789
opaque structure to pass to read_records to extract the raw
1792
Content that this record is built upon, may be None
1794
Logical parents of this node
1796
extra information about the content which needs to be passed to
1797
Factory.parse_record
1800
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
1801
for entry in entries:
1802
version_id = self._keys_to_version_ids((entry[1],))[0]
1803
if not self._parents:
1806
parents = self._keys_to_version_ids(entry[3][0])
1807
if not self._deltas:
1808
compression_parent = None
1810
compression_parent_key = self._compression_parent(entry)
1811
if compression_parent_key:
1812
compression_parent = self._keys_to_version_ids(
1813
(compression_parent_key,))[0]
1815
compression_parent = None
1816
noeol = (entry[2][0] == 'N')
1817
if compression_parent:
1818
method = 'line-delta'
1821
result[version_id] = (self._node_to_position(entry),
1822
compression_parent, parents,
1826
def _compression_parent(self, an_entry):
1827
# return the key that an_entry is compressed against, or None
1828
# Grab the second parent list (as deltas implies parents currently)
1829
compression_parents = an_entry[3][1]
1830
if not compression_parents:
1832
assert len(compression_parents) == 1
1833
return compression_parents[0]
1835
def _get_method(self, node):
1836
if not self._deltas:
1838
if self._compression_parent(node):
1843
def get_graph(self):
1844
"""Return a list of the node:parents lists from this knit index."""
1845
if not self._parents:
1846
return [(key, ()) for key in self.get_versions()]
1848
for index, key, value, refs in self._graph_index.iter_all_entries():
1849
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1852
def iter_parents(self, version_ids):
1853
"""Iterate through the parents for many version ids.
1855
:param version_ids: An iterable yielding version_ids.
1856
:return: An iterator that yields (version_id, parents). Requested
1857
version_ids not present in the versioned file are simply skipped.
1858
The order is undefined, allowing for different optimisations in
1859
the underlying implementation.
1862
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1864
present_parents = set()
1865
for node in all_nodes:
1866
all_parents.update(node[3][0])
1867
# any node we are querying must be present
1868
present_parents.add(node[1])
1869
unknown_parents = all_parents.difference(present_parents)
1870
present_parents.update(self._present_keys(unknown_parents))
1871
for node in all_nodes:
1873
for parent in node[3][0]:
1874
if parent in present_parents:
1875
parents.append(parent[0])
1876
yield node[1][0], tuple(parents)
1878
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1879
yield node[1][0], ()
1881
def num_versions(self):
1882
return len(list(self._graph_index.iter_all_entries()))
1884
__len__ = num_versions
1886
def get_versions(self):
1887
"""Get all the versions in the file. not topologically sorted."""
1888
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1890
def has_version(self, version_id):
1891
"""True if the version is in the index."""
1892
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1894
def _keys_to_version_ids(self, keys):
1895
return tuple(key[0] for key in keys)
1897
def get_position(self, version_id):
1898
"""Return details needed to access the version.
1900
:return: a tuple (index, data position, size) to hand to the access
1901
logic to get the record.
1903
node = self._get_node(version_id)
1904
return self._node_to_position(node)
1906
def _node_to_position(self, node):
1907
"""Convert an index value to position details."""
1908
bits = node[2][1:].split(' ')
1909
return node[0], int(bits[0]), int(bits[1])
1911
def get_method(self, version_id):
1912
"""Return compression method of specified version."""
1913
return self._get_method(self._get_node(version_id))
1915
def _get_node(self, version_id):
1917
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1919
raise RevisionNotPresent(version_id, self)
1921
def get_options(self, version_id):
1922
"""Return a list representing options.
1926
node = self._get_node(version_id)
1927
options = [self._get_method(node)]
1928
if node[2][0] == 'N':
1929
options.append('no-eol')
1932
def get_parents(self, version_id):
1933
"""Return parents of specified version ignoring ghosts."""
1934
parents = list(self.iter_parents([version_id]))
1937
raise errors.RevisionNotPresent(version_id, self)
1938
return parents[0][1]
1940
def get_parents_with_ghosts(self, version_id):
1941
"""Return parents of specified version with ghosts."""
1942
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1943
check_present=True))
1944
if not self._parents:
1946
return self._keys_to_version_ids(nodes[0][3][0])
1948
def check_versions_present(self, version_ids):
1949
"""Check that all specified versions are present."""
1950
keys = self._version_ids_to_keys(version_ids)
1951
present = self._present_keys(keys)
1952
missing = keys.difference(present)
1954
raise RevisionNotPresent(missing.pop(), self)
1956
def add_version(self, version_id, options, access_memo, parents):
1957
"""Add a version record to the index."""
1958
return self.add_versions(((version_id, options, access_memo, parents),))
1960
def add_versions(self, versions, random_id=False):
1961
"""Add multiple versions to the index.
1963
This function does not insert data into the Immutable GraphIndex
1964
backing the KnitGraphIndex, instead it prepares data for insertion by
1965
the caller and checks that it is safe to insert then calls
1966
self._add_callback with the prepared GraphIndex nodes.
1968
:param versions: a list of tuples:
1969
(version_id, options, pos, size, parents).
1970
:param random_id: If True the ids being added were randomly generated
1971
and no check for existence will be performed.
1973
if not self._add_callback:
1974
raise errors.ReadOnlyError(self)
1975
# we hope there are no repositories with inconsistent parentage
1980
for (version_id, options, access_memo, parents) in versions:
1981
index, pos, size = access_memo
1982
key = (version_id, )
1983
parents = tuple((parent, ) for parent in parents)
1984
if 'no-eol' in options:
1988
value += "%d %d" % (pos, size)
1989
if not self._deltas:
1990
if 'line-delta' in options:
1991
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1994
if 'line-delta' in options:
1995
node_refs = (parents, (parents[0],))
1997
node_refs = (parents, ())
1999
node_refs = (parents, )
2002
raise KnitCorrupt(self, "attempt to add node with parents "
2003
"in parentless index.")
2005
keys[key] = (value, node_refs)
2007
present_nodes = self._get_entries(keys)
2008
for (index, key, value, node_refs) in present_nodes:
2009
if (value, node_refs) != keys[key]:
2010
raise KnitCorrupt(self, "inconsistent details in add_versions"
2011
": %s %s" % ((value, node_refs), keys[key]))
2015
for key, (value, node_refs) in keys.iteritems():
2016
result.append((key, value, node_refs))
2018
for key, (value, node_refs) in keys.iteritems():
2019
result.append((key, value))
2020
self._add_callback(result)
2022
def _version_ids_to_keys(self, version_ids):
2023
return set((version_id, ) for version_id in version_ids)
2026
class _KnitAccess(object):
2027
"""Access to knit records in a .knit file."""
2029
def __init__(self, transport, filename, _file_mode, _dir_mode,
2030
_need_to_create, _create_parent_dir):
2031
"""Create a _KnitAccess for accessing and inserting data.
2033
:param transport: The transport the .knit is located on.
2034
:param filename: The filename of the .knit.
2036
self._transport = transport
2037
self._filename = filename
2038
self._file_mode = _file_mode
2039
self._dir_mode = _dir_mode
2040
self._need_to_create = _need_to_create
2041
self._create_parent_dir = _create_parent_dir
2043
def add_raw_records(self, sizes, raw_data):
2044
"""Add raw knit bytes to a storage area.
2046
The data is spooled to whereever the access method is storing data.
2048
:param sizes: An iterable containing the size of each raw data segment.
2049
:param raw_data: A bytestring containing the data.
2050
:return: A list of memos to retrieve the record later. Each memo is a
2051
tuple - (index, pos, length), where the index field is always None
2052
for the .knit access method.
2054
assert type(raw_data) == str, \
2055
'data must be plain bytes was %s' % type(raw_data)
2056
if not self._need_to_create:
2057
base = self._transport.append_bytes(self._filename, raw_data)
2059
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2060
create_parent_dir=self._create_parent_dir,
2061
mode=self._file_mode,
2062
dir_mode=self._dir_mode)
2063
self._need_to_create = False
2067
result.append((None, base, size))
2072
"""IFF this data access has its own storage area, initialise it.
2076
self._transport.put_bytes_non_atomic(self._filename, '',
2077
mode=self._file_mode)
2079
def open_file(self):
2080
"""IFF this data access can be represented as a single file, open it.
2082
For knits that are not mapped to a single file on disk this will
2085
:return: None or a file handle.
2088
return self._transport.get(self._filename)
2093
def get_raw_records(self, memos_for_retrieval):
2094
"""Get the raw bytes for a records.
2096
:param memos_for_retrieval: An iterable containing the (index, pos,
2097
length) memo for retrieving the bytes. The .knit method ignores
2098
the index as there is always only a single file.
2099
:return: An iterator over the bytes of the records.
2101
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2102
for pos, data in self._transport.readv(self._filename, read_vector):
2106
class _PackAccess(object):
2107
"""Access to knit records via a collection of packs."""
2109
def __init__(self, index_to_packs, writer=None):
2110
"""Create a _PackAccess object.
2112
:param index_to_packs: A dict mapping index objects to the transport
2113
and file names for obtaining data.
2114
:param writer: A tuple (pack.ContainerWriter, write_index) which
2115
contains the pack to write, and the index that reads from it will
2119
self.container_writer = writer[0]
2120
self.write_index = writer[1]
2122
self.container_writer = None
2123
self.write_index = None
2124
self.indices = index_to_packs
2126
def add_raw_records(self, sizes, raw_data):
2127
"""Add raw knit bytes to a storage area.
2129
The data is spooled to the container writer in one bytes-record per
2132
:param sizes: An iterable containing the size of each raw data segment.
2133
:param raw_data: A bytestring containing the data.
2134
:return: A list of memos to retrieve the record later. Each memo is a
2135
tuple - (index, pos, length), where the index field is the
2136
write_index object supplied to the PackAccess object.
2138
assert type(raw_data) == str, \
2139
'data must be plain bytes was %s' % type(raw_data)
2143
p_offset, p_length = self.container_writer.add_bytes_record(
2144
raw_data[offset:offset+size], [])
2146
result.append((self.write_index, p_offset, p_length))
2150
"""Pack based knits do not get individually created."""
2152
def get_raw_records(self, memos_for_retrieval):
2153
"""Get the raw bytes for a records.
2155
:param memos_for_retrieval: An iterable containing the (index, pos,
2156
length) memo for retrieving the bytes. The Pack access method
2157
looks up the pack to use for a given record in its index_to_pack
2159
:return: An iterator over the bytes of the records.
2161
# first pass, group into same-index requests
2163
current_index = None
2164
for (index, offset, length) in memos_for_retrieval:
2165
if current_index == index:
2166
current_list.append((offset, length))
2168
if current_index is not None:
2169
request_lists.append((current_index, current_list))
2170
current_index = index
2171
current_list = [(offset, length)]
2172
# handle the last entry
2173
if current_index is not None:
2174
request_lists.append((current_index, current_list))
2175
for index, offsets in request_lists:
2176
transport, path = self.indices[index]
2177
reader = pack.make_readv_reader(transport, path, offsets)
2178
for names, read_func in reader.iter_records():
2179
yield read_func(None)
2181
def open_file(self):
2182
"""Pack based knits have no single file."""
2185
def set_writer(self, writer, index, (transport, packname)):
2186
"""Set a writer to use for adding data."""
2187
if index is not None:
2188
self.indices[index] = (transport, packname)
2189
self.container_writer = writer
2190
self.write_index = index
2193
class _StreamAccess(object):
2194
"""A Knit Access object that provides data from a datastream.
2196
It also provides a fallback to present as unannotated data, annotated data
2197
from a *backing* access object.
2199
This is triggered by a index_memo which is pointing to a different index
2200
than this was constructed with, and is used to allow extracting full
2201
unannotated texts for insertion into annotated knits.
2204
def __init__(self, reader_callable, stream_index, backing_knit,
2206
"""Create a _StreamAccess object.
2208
:param reader_callable: The reader_callable from the datastream.
2209
This is called to buffer all the data immediately, for
2211
:param stream_index: The index the data stream this provides access to
2212
which will be present in native index_memo's.
2213
:param backing_knit: The knit object that will provide access to
2214
annotated texts which are not available in the stream, so as to
2215
create unannotated texts.
2216
:param orig_factory: The original content factory used to generate the
2217
stream. This is used for checking whether the thunk code for
2218
supporting _copy_texts will generate the correct form of data.
2220
self.data = reader_callable(None)
2221
self.stream_index = stream_index
2222
self.backing_knit = backing_knit
2223
self.orig_factory = orig_factory
2225
def get_raw_records(self, memos_for_retrieval):
2226
"""Get the raw bytes for a records.
2228
:param memos_for_retrieval: An iterable containing the (thunk_flag,
2229
index, start, end) memo for retrieving the bytes.
2230
:return: An iterator over the bytes of the records.
2232
# use a generator for memory friendliness
2233
for thunk_flag, version_id, start, end in memos_for_retrieval:
2234
if version_id is self.stream_index:
2235
yield self.data[start:end]
2237
# we have been asked to thunk. This thunking only occurs when
2238
# we are obtaining plain texts from an annotated backing knit
2239
# so that _copy_texts will work.
2240
# We could improve performance here by scanning for where we need
2241
# to do this and using get_line_list, then interleaving the output
2242
# as desired. However, for now, this is sufficient.
2243
if self.orig_factory.__class__ != KnitPlainFactory:
2244
raise errors.KnitCorrupt(
2245
self, 'Bad thunk request %r' % version_id)
2246
lines = self.backing_knit.get_lines(version_id)
2247
line_bytes = ''.join(lines)
2248
digest = sha_string(line_bytes)
2250
if lines[-1][-1] != '\n':
2251
lines[-1] = lines[-1] + '\n'
2253
orig_options = list(self.backing_knit._index.get_options(version_id))
2254
if 'fulltext' not in orig_options:
2255
if 'line-delta' not in orig_options:
2256
raise errors.KnitCorrupt(self,
2257
'Unknown compression method %r' % orig_options)
2258
orig_options.remove('line-delta')
2259
orig_options.append('fulltext')
2260
# We want plain data, because we expect to thunk only to allow text
2262
size, bytes = self.backing_knit._data._record_to_data(version_id,
2263
digest, lines, line_bytes)
2267
class _StreamIndex(object):
2268
"""A Knit Index object that uses the data map from a datastream."""
2270
def __init__(self, data_list, backing_index):
2271
"""Create a _StreamIndex object.
2273
:param data_list: The data_list from the datastream.
2274
:param backing_index: The index which will supply values for nodes
2275
referenced outside of this stream.
2277
self.data_list = data_list
2278
self.backing_index = backing_index
2279
self._by_version = {}
2281
for key, options, length, parents in data_list:
2282
self._by_version[key] = options, (pos, pos + length), parents
2285
def get_ancestry(self, versions, topo_sorted):
2286
"""Get an ancestry list for versions."""
2288
# Not needed for basic joins
2289
raise NotImplementedError(self.get_ancestry)
2290
# get a graph of all the mentioned versions:
2291
# Little ugly - basically copied from KnitIndex, but don't want to
2292
# accidentally incorporate too much of that index's code.
2294
pending = set(versions)
2295
cache = self._by_version
2297
version = pending.pop()
2300
parents = [p for p in cache[version][2] if p in cache]
2302
raise RevisionNotPresent(version, self)
2303
# if not completed and not a ghost
2304
pending.update([p for p in parents if p not in ancestry])
2305
ancestry.add(version)
2306
return list(ancestry)
2308
def get_build_details(self, version_ids):
2309
"""Get the method, index_memo and compression parent for version_ids.
2311
Ghosts are omitted from the result.
2313
:param version_ids: An iterable of version_ids.
2314
:return: A dict of version_id:(index_memo, compression_parent,
2315
parents, record_details).
2317
opaque structure to pass to read_records to extract the raw
2320
Content that this record is built upon, may be None
2322
Logical parents of this node
2324
extra information about the content which needs to be passed to
2325
Factory.parse_record
2328
for version_id in version_ids:
2330
method = self.get_method(version_id)
2331
except errors.RevisionNotPresent:
2332
# ghosts are omitted
2334
parent_ids = self.get_parents_with_ghosts(version_id)
2335
noeol = ('no-eol' in self.get_options(version_id))
2336
if method == 'fulltext':
2337
compression_parent = None
2339
compression_parent = parent_ids[0]
2340
index_memo = self.get_position(version_id)
2341
result[version_id] = (index_memo, compression_parent,
2342
parent_ids, (method, noeol))
2345
def get_method(self, version_id):
2346
"""Return compression method of specified version."""
2348
options = self._by_version[version_id][0]
2350
# Strictly speaking this should check in the backing knit, but
2351
# until we have a test to discriminate, this will do.
2352
return self.backing_index.get_method(version_id)
2353
if 'fulltext' in options:
2355
elif 'line-delta' in options:
2358
raise errors.KnitIndexUnknownMethod(self, options)
2360
def get_options(self, version_id):
2361
"""Return a list representing options.
2366
return self._by_version[version_id][0]
2368
return self.backing_index.get_options(version_id)
2370
def get_parents_with_ghosts(self, version_id):
2371
"""Return parents of specified version with ghosts."""
2373
return self._by_version[version_id][2]
2375
return self.backing_index.get_parents_with_ghosts(version_id)
2377
def get_position(self, version_id):
2378
"""Return details needed to access the version.
2380
_StreamAccess has the data as a big array, so we return slice
2381
coordinates into that (as index_memo's are opaque outside the
2382
index and matching access class).
2384
:return: a tuple (thunk_flag, index, start, end). If thunk_flag is
2385
False, index will be self, otherwise it will be a version id.
2388
start, end = self._by_version[version_id][1]
2389
return False, self, start, end
2391
# Signal to the access object to handle this from the backing knit.
2392
return (True, version_id, None, None)
2394
def get_versions(self):
2395
"""Get all the versions in the stream."""
2396
return self._by_version.keys()
2398
def iter_parents(self, version_ids):
2399
"""Iterate through the parents for many version ids.
2401
:param version_ids: An iterable yielding version_ids.
2402
:return: An iterator that yields (version_id, parents). Requested
2403
version_ids not present in the versioned file are simply skipped.
2404
The order is undefined, allowing for different optimisations in
2405
the underlying implementation.
2408
for version in version_ids:
2410
result.append((version, self._by_version[version][2]))
2416
class _KnitData(object):
2417
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2419
The KnitData class provides the logic for parsing and using knit records,
2420
making use of an access method for the low level read and write operations.
2423
def __init__(self, access):
2424
"""Create a KnitData object.
2426
:param access: The access method to use. Access methods such as
2427
_KnitAccess manage the insertion of raw records and the subsequent
2428
retrieval of the same.
2430
self._access = access
2431
self._checked = False
2432
# TODO: jam 20060713 conceptually, this could spill to disk
2433
# if the cached size gets larger than a certain amount
2434
# but it complicates the model a bit, so for now just use
2435
# a simple dictionary
2437
self._do_cache = False
2439
def enable_cache(self):
2440
"""Enable caching of reads."""
2441
self._do_cache = True
2443
def clear_cache(self):
2444
"""Clear the record cache."""
2445
self._do_cache = False
2448
def _open_file(self):
2449
return self._access.open_file()
2451
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2452
"""Convert version_id, digest, lines into a raw data block.
2454
:param dense_lines: The bytes of lines but in a denser form. For
2455
instance, if lines is a list of 1000 bytestrings each ending in \n,
2456
dense_lines may be a list with one line in it, containing all the
2457
1000's lines and their \n's. Using dense_lines if it is already
2458
known is a win because the string join to create bytes in this
2459
function spends less time resizing the final string.
2460
:return: (len, a StringIO instance with the raw data ready to read.)
2462
# Note: using a string copy here increases memory pressure with e.g.
2463
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2464
# when doing the initial commit of a mozilla tree. RBC 20070921
2465
bytes = ''.join(chain(
2466
["version %s %d %s\n" % (version_id,
2469
dense_lines or lines,
2470
["end %s\n" % version_id]))
2471
assert bytes.__class__ == str
2472
compressed_bytes = bytes_to_gzip(bytes)
2473
return len(compressed_bytes), compressed_bytes
2475
def add_raw_records(self, sizes, raw_data):
2476
"""Append a prepared record to the data file.
2478
:param sizes: An iterable containing the size of each raw data segment.
2479
:param raw_data: A bytestring containing the data.
2480
:return: a list of index data for the way the data was stored.
2481
See the access method add_raw_records documentation for more
2484
return self._access.add_raw_records(sizes, raw_data)
2486
def _parse_record_header(self, version_id, raw_data):
2487
"""Parse a record header for consistency.
2489
:return: the header and the decompressor stream.
2490
as (stream, header_record)
2492
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2494
rec = self._check_header(version_id, df.readline())
2495
except Exception, e:
2496
raise KnitCorrupt(self._access,
2497
"While reading {%s} got %s(%s)"
2498
% (version_id, e.__class__.__name__, str(e)))
2501
def _check_header(self, version_id, line):
2504
raise KnitCorrupt(self._access,
2505
'unexpected number of elements in record header')
2506
if rec[1] != version_id:
2507
raise KnitCorrupt(self._access,
2508
'unexpected version, wanted %r, got %r'
2509
% (version_id, rec[1]))
2512
def _parse_record(self, version_id, data):
2514
# 4168 calls in 2880 217 internal
2515
# 4168 calls to _parse_record_header in 2121
2516
# 4168 calls to readlines in 330
2517
df = GzipFile(mode='rb', fileobj=StringIO(data))
2520
record_contents = df.readlines()
2521
except Exception, e:
2522
raise KnitCorrupt(self._access,
2523
"While reading {%s} got %s(%s)"
2524
% (version_id, e.__class__.__name__, str(e)))
2525
header = record_contents.pop(0)
2526
rec = self._check_header(version_id, header)
2528
last_line = record_contents.pop()
2529
if len(record_contents) != int(rec[2]):
2530
raise KnitCorrupt(self._access,
2531
'incorrect number of lines %s != %s'
2533
% (len(record_contents), int(rec[2]),
2535
if last_line != 'end %s\n' % rec[1]:
2536
raise KnitCorrupt(self._access,
2537
'unexpected version end line %r, wanted %r'
2538
% (last_line, version_id))
2540
return record_contents, rec[3]
2542
def read_records_iter_raw(self, records):
2543
"""Read text records from data file and yield raw data.
2545
This unpacks enough of the text record to validate the id is
2546
as expected but thats all.
2548
# setup an iterator of the external records:
2549
# uses readv so nice and fast we hope.
2551
# grab the disk data needed.
2553
# Don't check _cache if it is empty
2554
needed_offsets = [index_memo for version_id, index_memo
2556
if version_id not in self._cache]
2558
needed_offsets = [index_memo for version_id, index_memo
2561
raw_records = self._access.get_raw_records(needed_offsets)
2563
for version_id, index_memo in records:
2564
if version_id in self._cache:
2565
# This data has already been validated
2566
data = self._cache[version_id]
2568
data = raw_records.next()
2570
self._cache[version_id] = data
2572
# validate the header
2573
df, rec = self._parse_record_header(version_id, data)
2575
yield version_id, data
2577
def read_records_iter(self, records):
2578
"""Read text records from data file and yield result.
2580
The result will be returned in whatever is the fastest to read.
2581
Not by the order requested. Also, multiple requests for the same
2582
record will only yield 1 response.
2583
:param records: A list of (version_id, pos, len) entries
2584
:return: Yields (version_id, contents, digest) in the order
2585
read, not the order requested
2591
# Skip records we have alread seen
2592
yielded_records = set()
2593
needed_records = set()
2594
for record in records:
2595
if record[0] in self._cache:
2596
if record[0] in yielded_records:
2598
yielded_records.add(record[0])
2599
data = self._cache[record[0]]
2600
content, digest = self._parse_record(record[0], data)
2601
yield (record[0], content, digest)
2603
needed_records.add(record)
2604
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2606
needed_records = sorted(set(records), key=operator.itemgetter(1))
2608
if not needed_records:
2611
# The transport optimizes the fetching as well
2612
# (ie, reads continuous ranges.)
2613
raw_data = self._access.get_raw_records(
2614
[index_memo for version_id, index_memo in needed_records])
2616
for (version_id, index_memo), data in \
2617
izip(iter(needed_records), raw_data):
2618
content, digest = self._parse_record(version_id, data)
2620
self._cache[version_id] = data
2621
yield version_id, content, digest
2623
def read_records(self, records):
2624
"""Read records into a dictionary."""
2626
for record_id, content, digest in \
2627
self.read_records_iter(records):
2628
components[record_id] = (content, digest)
2632
class InterKnit(InterVersionedFile):
2633
"""Optimised code paths for knit to knit operations."""
2635
_matching_file_from_factory = KnitVersionedFile
2636
_matching_file_to_factory = KnitVersionedFile
2639
def is_compatible(source, target):
2640
"""Be compatible with knits. """
2642
return (isinstance(source, KnitVersionedFile) and
2643
isinstance(target, KnitVersionedFile))
2644
except AttributeError:
2647
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2648
"""Copy texts to the target by extracting and adding them one by one.
2650
see join() for the parameter definitions.
2652
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2653
graph = self.source.get_graph(version_ids)
2654
order = topo_sort(graph.items())
2656
def size_of_content(content):
2657
return sum(len(line) for line in content.text())
2658
# Cache at most 10MB of parent texts
2659
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2660
compute_size=size_of_content)
2661
# TODO: jam 20071116 It would be nice to have a streaming interface to
2662
# get multiple texts from a source. The source could be smarter
2663
# about how it handled intermediate stages.
2664
# get_line_list() or make_mpdiffs() seem like a possibility, but
2665
# at the moment they extract all full texts into memory, which
2666
# causes us to store more than our 3x fulltext goal.
2667
# Repository.iter_files_bytes() may be another possibility
2668
to_process = [version for version in order
2669
if version not in self.target]
2670
total = len(to_process)
2671
pb = ui.ui_factory.nested_progress_bar()
2673
for index, version in enumerate(to_process):
2674
pb.update('Converting versioned data', index, total)
2675
sha1, num_bytes, parent_text = self.target.add_lines(version,
2676
self.source.get_parents_with_ghosts(version),
2677
self.source.get_lines(version),
2678
parent_texts=parent_cache)
2679
parent_cache[version] = parent_text
2684
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2685
"""See InterVersionedFile.join."""
2686
assert isinstance(self.source, KnitVersionedFile)
2687
assert isinstance(self.target, KnitVersionedFile)
2689
# If the source and target are mismatched w.r.t. annotations vs
2690
# plain, the data needs to be converted accordingly
2691
if self.source.factory.annotated == self.target.factory.annotated:
2693
elif self.source.factory.annotated:
2694
converter = self._anno_to_plain_converter
2696
# We're converting from a plain to an annotated knit. Copy them
2697
# across by full texts.
2698
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2700
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2704
pb = ui.ui_factory.nested_progress_bar()
2706
version_ids = list(version_ids)
2707
if None in version_ids:
2708
version_ids.remove(None)
2710
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2712
this_versions = set(self.target._index.get_versions())
2713
# XXX: For efficiency we should not look at the whole index,
2714
# we only need to consider the referenced revisions - they
2715
# must all be present, or the method must be full-text.
2716
# TODO, RBC 20070919
2717
needed_versions = self.source_ancestry - this_versions
2719
if not needed_versions:
2721
full_list = topo_sort(self.source.get_graph())
2723
version_list = [i for i in full_list if (not self.target.has_version(i)
2724
and i in needed_versions)]
2728
copy_queue_records = []
2730
for version_id in version_list:
2731
options = self.source._index.get_options(version_id)
2732
parents = self.source._index.get_parents_with_ghosts(version_id)
2733
# check that its will be a consistent copy:
2734
for parent in parents:
2735
# if source has the parent, we must :
2736
# * already have it or
2737
# * have it scheduled already
2738
# otherwise we don't care
2739
assert (self.target.has_version(parent) or
2740
parent in copy_set or
2741
not self.source.has_version(parent))
2742
index_memo = self.source._index.get_position(version_id)
2743
copy_queue_records.append((version_id, index_memo))
2744
copy_queue.append((version_id, options, parents))
2745
copy_set.add(version_id)
2747
# data suck the join:
2749
total = len(version_list)
2752
for (version_id, raw_data), \
2753
(version_id2, options, parents) in \
2754
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2756
assert version_id == version_id2, 'logic error, inconsistent results'
2758
pb.update("Joining knit", count, total)
2760
size, raw_data = converter(raw_data, version_id, options,
2763
size = len(raw_data)
2764
raw_records.append((version_id, options, parents, size))
2765
raw_datum.append(raw_data)
2766
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2771
def _anno_to_plain_converter(self, raw_data, version_id, options,
2773
"""Convert annotated content to plain content."""
2774
data, digest = self.source._data._parse_record(version_id, raw_data)
2775
if 'fulltext' in options:
2776
content = self.source.factory.parse_fulltext(data, version_id)
2777
lines = self.target.factory.lower_fulltext(content)
2779
delta = self.source.factory.parse_line_delta(data, version_id,
2781
lines = self.target.factory.lower_line_delta(delta)
2782
return self.target._data._record_to_data(version_id, digest, lines)
2785
InterVersionedFile.register_optimiser(InterKnit)
2788
class WeaveToKnit(InterVersionedFile):
2789
"""Optimised code paths for weave to knit operations."""
2791
_matching_file_from_factory = bzrlib.weave.WeaveFile
2792
_matching_file_to_factory = KnitVersionedFile
2795
def is_compatible(source, target):
2796
"""Be compatible with weaves to knits."""
2798
return (isinstance(source, bzrlib.weave.Weave) and
2799
isinstance(target, KnitVersionedFile))
2800
except AttributeError:
2803
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2804
"""See InterVersionedFile.join."""
2805
assert isinstance(self.source, bzrlib.weave.Weave)
2806
assert isinstance(self.target, KnitVersionedFile)
2808
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2813
pb = ui.ui_factory.nested_progress_bar()
2815
version_ids = list(version_ids)
2817
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2818
this_versions = set(self.target._index.get_versions())
2819
needed_versions = self.source_ancestry - this_versions
2821
if not needed_versions:
2823
full_list = topo_sort(self.source.get_graph())
2825
version_list = [i for i in full_list if (not self.target.has_version(i)
2826
and i in needed_versions)]
2830
total = len(version_list)
2831
parent_map = self.source.get_parent_map(version_list)
2832
for version_id in version_list:
2833
pb.update("Converting to knit", count, total)
2834
parents = parent_map[version_id]
2835
# check that its will be a consistent copy:
2836
for parent in parents:
2837
# if source has the parent, we must already have it
2838
assert (self.target.has_version(parent))
2839
self.target.add_lines(
2840
version_id, parents, self.source.get_lines(version_id))
2847
InterVersionedFile.register_optimiser(WeaveToKnit)
2850
# Deprecated, use PatienceSequenceMatcher instead
2851
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2854
def annotate_knit(knit, revision_id):
2855
"""Annotate a knit with no cached annotations.
2857
This implementation is for knits with no cached annotations.
2858
It will work for knits with cached annotations, but this is not
2861
annotator = _KnitAnnotator(knit)
2862
return iter(annotator.annotate(revision_id))
2865
class _KnitAnnotator(object):
2866
"""Build up the annotations for a text."""
2868
def __init__(self, knit):
2871
# Content objects, differs from fulltexts because of how final newlines
2872
# are treated by knits. the content objects here will always have a
2874
self._fulltext_contents = {}
2876
# Annotated lines of specific revisions
2877
self._annotated_lines = {}
2879
# Track the raw data for nodes that we could not process yet.
2880
# This maps the revision_id of the base to a list of children that will
2881
# annotated from it.
2882
self._pending_children = {}
2884
# Nodes which cannot be extracted
2885
self._ghosts = set()
2887
# Track how many children this node has, so we know if we need to keep
2889
self._annotate_children = {}
2890
self._compression_children = {}
2892
self._all_build_details = {}
2893
# The children => parent revision_id graph
2894
self._revision_id_graph = {}
2896
self._heads_provider = None
2898
self._nodes_to_keep_annotations = set()
2899
self._generations_until_keep = 100
2901
def set_generations_until_keep(self, value):
2902
"""Set the number of generations before caching a node.
2904
Setting this to -1 will cache every merge node, setting this higher
2905
will cache fewer nodes.
2907
self._generations_until_keep = value
2909
def _add_fulltext_content(self, revision_id, content_obj):
2910
self._fulltext_contents[revision_id] = content_obj
2911
# TODO: jam 20080305 It might be good to check the sha1digest here
2912
return content_obj.text()
2914
def _check_parents(self, child, nodes_to_annotate):
2915
"""Check if all parents have been processed.
2917
:param child: A tuple of (rev_id, parents, raw_content)
2918
:param nodes_to_annotate: If child is ready, add it to
2919
nodes_to_annotate, otherwise put it back in self._pending_children
2921
for parent_id in child[1]:
2922
if (parent_id not in self._annotated_lines):
2923
# This parent is present, but another parent is missing
2924
self._pending_children.setdefault(parent_id,
2928
# This one is ready to be processed
2929
nodes_to_annotate.append(child)
2931
def _add_annotation(self, revision_id, fulltext, parent_ids,
2932
left_matching_blocks=None):
2933
"""Add an annotation entry.
2935
All parents should already have been annotated.
2936
:return: A list of children that now have their parents satisfied.
2938
a = self._annotated_lines
2939
annotated_parent_lines = [a[p] for p in parent_ids]
2940
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2941
fulltext, revision_id, left_matching_blocks,
2942
heads_provider=self._get_heads_provider()))
2943
self._annotated_lines[revision_id] = annotated_lines
2944
for p in parent_ids:
2945
ann_children = self._annotate_children[p]
2946
ann_children.remove(revision_id)
2947
if (not ann_children
2948
and p not in self._nodes_to_keep_annotations):
2949
del self._annotated_lines[p]
2950
del self._all_build_details[p]
2951
if p in self._fulltext_contents:
2952
del self._fulltext_contents[p]
2953
# Now that we've added this one, see if there are any pending
2954
# deltas to be done, certainly this parent is finished
2955
nodes_to_annotate = []
2956
for child in self._pending_children.pop(revision_id, []):
2957
self._check_parents(child, nodes_to_annotate)
2958
return nodes_to_annotate
2960
def _get_build_graph(self, revision_id):
2961
"""Get the graphs for building texts and annotations.
2963
The data you need for creating a full text may be different than the
2964
data you need to annotate that text. (At a minimum, you need both
2965
parents to create an annotation, but only need 1 parent to generate the
2968
:return: A list of (revision_id, index_memo) records, suitable for
2969
passing to read_records_iter to start reading in the raw data fro/
2972
if revision_id in self._annotated_lines:
2975
pending = set([revision_id])
2980
# get all pending nodes
2982
this_iteration = pending
2983
build_details = self._knit._index.get_build_details(this_iteration)
2984
self._all_build_details.update(build_details)
2985
# new_nodes = self._knit._index._get_entries(this_iteration)
2987
for rev_id, details in build_details.iteritems():
2988
(index_memo, compression_parent, parents,
2989
record_details) = details
2990
self._revision_id_graph[rev_id] = parents
2991
records.append((rev_id, index_memo))
2992
# Do we actually need to check _annotated_lines?
2993
pending.update(p for p in parents
2994
if p not in self._all_build_details)
2995
if compression_parent:
2996
self._compression_children.setdefault(compression_parent,
2999
for parent in parents:
3000
self._annotate_children.setdefault(parent,
3002
num_gens = generation - kept_generation
3003
if ((num_gens >= self._generations_until_keep)
3004
and len(parents) > 1):
3005
kept_generation = generation
3006
self._nodes_to_keep_annotations.add(rev_id)
3008
missing_versions = this_iteration.difference(build_details.keys())
3009
self._ghosts.update(missing_versions)
3010
for missing_version in missing_versions:
3011
# add a key, no parents
3012
self._revision_id_graph[missing_version] = ()
3013
pending.discard(missing_version) # don't look for it
3014
# XXX: This should probably be a real exception, as it is a data
3016
assert not self._ghosts.intersection(self._compression_children), \
3017
"We cannot have nodes which have a compression parent of a ghost."
3018
# Cleanout anything that depends on a ghost so that we don't wait for
3019
# the ghost to show up
3020
for node in self._ghosts:
3021
if node in self._annotate_children:
3022
# We won't be building this node
3023
del self._annotate_children[node]
3024
# Generally we will want to read the records in reverse order, because
3025
# we find the parent nodes after the children
3029
def _annotate_records(self, records):
3030
"""Build the annotations for the listed records."""
3031
# We iterate in the order read, rather than a strict order requested
3032
# However, process what we can, and put off to the side things that
3033
# still need parents, cleaning them up when those parents are
3035
for (rev_id, record,
3036
digest) in self._knit._data.read_records_iter(records):
3037
if rev_id in self._annotated_lines:
3039
parent_ids = self._revision_id_graph[rev_id]
3040
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3041
details = self._all_build_details[rev_id]
3042
(index_memo, compression_parent, parents,
3043
record_details) = details
3044
nodes_to_annotate = []
3045
# TODO: Remove the punning between compression parents, and
3046
# parent_ids, we should be able to do this without assuming
3048
if len(parent_ids) == 0:
3049
# There are no parents for this node, so just add it
3050
# TODO: This probably needs to be decoupled
3051
assert compression_parent is None
3052
fulltext_content, delta = self._knit.factory.parse_record(
3053
rev_id, record, record_details, None)
3054
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3055
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3056
parent_ids, left_matching_blocks=None))
3058
child = (rev_id, parent_ids, record)
3059
# Check if all the parents are present
3060
self._check_parents(child, nodes_to_annotate)
3061
while nodes_to_annotate:
3062
# Should we use a queue here instead of a stack?
3063
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3064
(index_memo, compression_parent, parents,
3065
record_details) = self._all_build_details[rev_id]
3066
if compression_parent is not None:
3067
comp_children = self._compression_children[compression_parent]
3068
assert rev_id in comp_children
3069
# If there is only 1 child, it is safe to reuse this
3071
reuse_content = (len(comp_children) == 1
3072
and compression_parent not in
3073
self._nodes_to_keep_annotations)
3075
# Remove it from the cache since it will be changing
3076
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3077
# Make sure to copy the fulltext since it might be
3079
parent_fulltext = list(parent_fulltext_content.text())
3081
parent_fulltext_content = self._fulltext_contents[compression_parent]
3082
parent_fulltext = parent_fulltext_content.text()
3083
comp_children.remove(rev_id)
3084
fulltext_content, delta = self._knit.factory.parse_record(
3085
rev_id, record, record_details,
3086
parent_fulltext_content,
3087
copy_base_content=(not reuse_content))
3088
fulltext = self._add_fulltext_content(rev_id,
3090
blocks = KnitContent.get_line_delta_blocks(delta,
3091
parent_fulltext, fulltext)
3093
fulltext_content = self._knit.factory.parse_fulltext(
3095
fulltext = self._add_fulltext_content(rev_id,
3098
nodes_to_annotate.extend(
3099
self._add_annotation(rev_id, fulltext, parent_ids,
3100
left_matching_blocks=blocks))
3102
def _get_heads_provider(self):
3103
"""Create a heads provider for resolving ancestry issues."""
3104
if self._heads_provider is not None:
3105
return self._heads_provider
3106
parent_provider = _mod_graph.DictParentsProvider(
3107
self._revision_id_graph)
3108
graph_obj = _mod_graph.Graph(parent_provider)
3109
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3110
self._heads_provider = head_cache
3113
def annotate(self, revision_id):
3114
"""Return the annotated fulltext at the given revision.
3116
:param revision_id: The revision id for this file
3118
records = self._get_build_graph(revision_id)
3119
if revision_id in self._ghosts:
3120
raise errors.RevisionNotPresent(revision_id, self._knit)
3121
self._annotate_records(records)
3122
return self._annotated_lines[revision_id]
3126
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3128
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
