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
return [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
def text_lines(self):
232
"""Return the official fulltext for this content.
234
This includes stripping the final newline if it should be done.
236
lines = [text for o, l in self._lines]
237
if self._should_strip_eol:
238
anno, line = lines[-1]
239
lines[-1] = (anno, line.rstrip('\n'))
243
return AnnotatedKnitContent(self._lines[:])
246
class PlainKnitContent(KnitContent):
247
"""Unannotated content.
249
When annotate[_iter] is called on this content, the same version is reported
250
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
254
def __init__(self, lines, version_id):
255
KnitContent.__init__(self)
257
self._version_id = version_id
259
def annotate_iter(self):
260
"""Yield tuples of (origin, text) for each content line."""
261
for line in self._lines:
262
yield self._version_id, line
264
def apply_delta(self, delta, new_version_id):
265
"""Apply delta to this object to become new_version_id."""
268
for start, end, count, delta_lines in delta:
269
lines[offset+start:offset+end] = delta_lines
270
offset = offset + (start - end) + count
271
self._version_id = new_version_id
274
return PlainKnitContent(self._lines[:], self._version_id)
276
def strip_last_line_newline(self):
277
self._lines[-1] = self._lines[-1].rstrip('\n')
278
self._should_strip_eol = False
283
def text_lines(self):
284
"""Return the official fulltext for this content.
286
This includes stripping the final newline if it should be done.
289
if self._should_strip_eol:
291
lines[-1] = lines[-1].rstrip('\n')
295
class _KnitFactory(object):
296
"""Base class for common Factory functions."""
298
def parse_record(self, version_id, record, record_details,
299
base_content, copy_base_content=True):
300
"""Parse a record into a full content object.
302
:param version_id: The official version id for this content
303
:param record: The data returned by read_records_iter()
304
:param record_details: Details about the record returned by
306
:param base_content: If get_build_details returns a compression_parent,
307
you must return a base_content here, else use None
308
:param copy_base_content: When building from the base_content, decide
309
you can either copy it and return a new object, or modify it in
311
:return: (content, delta) A Content object and possibly a line-delta,
314
method, noeol = record_details
315
if method == 'line-delta':
316
assert base_content is not None
317
if copy_base_content:
318
content = base_content.copy()
320
content = base_content
321
delta = self.parse_line_delta(record, version_id)
322
content.apply_delta(delta, version_id)
324
content = self.parse_fulltext(record, version_id)
326
content._should_strip_eol = noeol
327
return (content, delta)
330
class KnitAnnotateFactory(_KnitFactory):
331
"""Factory for creating annotated Content objects."""
335
def make(self, lines, version_id):
336
num_lines = len(lines)
337
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
339
def parse_fulltext(self, content, version_id):
340
"""Convert fulltext to internal representation
342
fulltext content is of the format
343
revid(utf8) plaintext\n
344
internal representation is of the format:
347
# TODO: jam 20070209 The tests expect this to be returned as tuples,
348
# but the code itself doesn't really depend on that.
349
# Figure out a way to not require the overhead of turning the
350
# list back into tuples.
351
lines = [tuple(line.split(' ', 1)) for line in content]
352
return AnnotatedKnitContent(lines)
354
def parse_line_delta_iter(self, lines):
355
return iter(self.parse_line_delta(lines))
357
def parse_line_delta(self, lines, version_id, plain=False):
358
"""Convert a line based delta into internal representation.
360
line delta is in the form of:
361
intstart intend intcount
363
revid(utf8) newline\n
364
internal representation is
365
(start, end, count, [1..count tuples (revid, newline)])
367
:param plain: If True, the lines are returned as a plain
368
list without annotations, not as a list of (origin, content) tuples, i.e.
369
(start, end, count, [1..count newline])
376
def cache_and_return(line):
377
origin, text = line.split(' ', 1)
378
return cache.setdefault(origin, origin), text
380
# walk through the lines parsing.
381
# Note that the plain test is explicitly pulled out of the
382
# loop to minimise any performance impact
385
start, end, count = [int(n) for n in header.split(',')]
386
contents = [next().split(' ', 1)[1] for i in xrange(count)]
387
result.append((start, end, count, contents))
390
start, end, count = [int(n) for n in header.split(',')]
391
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
392
result.append((start, end, count, contents))
395
def get_fulltext_content(self, lines):
396
"""Extract just the content lines from a fulltext."""
397
return (line.split(' ', 1)[1] for line in lines)
399
def get_linedelta_content(self, lines):
400
"""Extract just the content from a line delta.
402
This doesn't return all of the extra information stored in a delta.
403
Only the actual content lines.
408
header = header.split(',')
409
count = int(header[2])
410
for i in xrange(count):
411
origin, text = next().split(' ', 1)
414
def lower_fulltext(self, content):
415
"""convert a fulltext content record into a serializable form.
417
see parse_fulltext which this inverts.
419
# TODO: jam 20070209 We only do the caching thing to make sure that
420
# the origin is a valid utf-8 line, eventually we could remove it
421
return ['%s %s' % (o, t) for o, t in content._lines]
423
def lower_line_delta(self, delta):
424
"""convert a delta into a serializable form.
426
See parse_line_delta which this inverts.
428
# TODO: jam 20070209 We only do the caching thing to make sure that
429
# the origin is a valid utf-8 line, eventually we could remove it
431
for start, end, c, lines in delta:
432
out.append('%d,%d,%d\n' % (start, end, c))
433
out.extend(origin + ' ' + text
434
for origin, text in lines)
437
def annotate_iter(self, knit, version_id):
438
content = knit._get_content(version_id)
439
return content.annotate_iter()
442
class KnitPlainFactory(_KnitFactory):
443
"""Factory for creating plain Content objects."""
447
def make(self, lines, version_id):
448
return PlainKnitContent(lines, version_id)
450
def parse_fulltext(self, content, version_id):
451
"""This parses an unannotated fulltext.
453
Note that this is not a noop - the internal representation
454
has (versionid, line) - its just a constant versionid.
456
return self.make(content, version_id)
458
def parse_line_delta_iter(self, lines, version_id):
460
num_lines = len(lines)
461
while cur < num_lines:
464
start, end, c = [int(n) for n in header.split(',')]
465
yield start, end, c, lines[cur:cur+c]
468
def parse_line_delta(self, lines, version_id):
469
return list(self.parse_line_delta_iter(lines, version_id))
471
def get_fulltext_content(self, lines):
472
"""Extract just the content lines from a fulltext."""
475
def get_linedelta_content(self, lines):
476
"""Extract just the content from a line delta.
478
This doesn't return all of the extra information stored in a delta.
479
Only the actual content lines.
484
header = header.split(',')
485
count = int(header[2])
486
for i in xrange(count):
489
def lower_fulltext(self, content):
490
return content.text()
492
def lower_line_delta(self, delta):
494
for start, end, c, lines in delta:
495
out.append('%d,%d,%d\n' % (start, end, c))
499
def annotate_iter(self, knit, version_id):
500
annotator = _KnitAnnotator(knit)
501
return iter(annotator.get_annotated_lines(version_id))
504
def make_empty_knit(transport, relpath):
505
"""Construct a empty knit at the specified location."""
506
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
509
class KnitVersionedFile(VersionedFile):
510
"""Weave-like structure with faster random access.
512
A knit stores a number of texts and a summary of the relationships
513
between them. Texts are identified by a string version-id. Texts
514
are normally stored and retrieved as a series of lines, but can
515
also be passed as single strings.
517
Lines are stored with the trailing newline (if any) included, to
518
avoid special cases for files with no final newline. Lines are
519
composed of 8-bit characters, not unicode. The combination of
520
these approaches should mean any 'binary' file can be safely
521
stored and retrieved.
524
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
525
factory=None, delta=True, create=False, create_parent_dir=False,
526
delay_create=False, dir_mode=None, index=None, access_method=None):
527
"""Construct a knit at location specified by relpath.
529
:param create: If not True, only open an existing knit.
530
:param create_parent_dir: If True, create the parent directory if
531
creating the file fails. (This is used for stores with
532
hash-prefixes that may not exist yet)
533
:param delay_create: The calling code is aware that the knit won't
534
actually be created until the first data is stored.
535
:param index: An index to use for the knit.
537
if access_mode is None:
539
super(KnitVersionedFile, self).__init__(access_mode)
540
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
541
self.transport = transport
542
self.filename = relpath
543
self.factory = factory or KnitAnnotateFactory()
544
self.writable = (access_mode == 'w')
547
self._max_delta_chain = 200
550
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
551
access_mode, create=create, file_mode=file_mode,
552
create_parent_dir=create_parent_dir, delay_create=delay_create,
556
if access_method is None:
557
_access = _KnitAccess(transport, relpath + DATA_SUFFIX, file_mode, dir_mode,
558
((create and not len(self)) and delay_create), create_parent_dir)
560
_access = access_method
561
if create and not len(self) and not delay_create:
563
self._data = _KnitData(_access)
566
return '%s(%s)' % (self.__class__.__name__,
567
self.transport.abspath(self.filename))
569
def _check_should_delta(self, first_parents):
570
"""Iterate back through the parent listing, looking for a fulltext.
572
This is used when we want to decide whether to add a delta or a new
573
fulltext. It searches for _max_delta_chain parents. When it finds a
574
fulltext parent, it sees if the total size of the deltas leading up to
575
it is large enough to indicate that we want a new full text anyway.
577
Return True if we should create a new delta, False if we should use a
582
delta_parents = first_parents
583
for count in xrange(self._max_delta_chain):
584
parent = delta_parents[0]
585
method = self._index.get_method(parent)
586
index, pos, size = self._index.get_position(parent)
587
if method == 'fulltext':
591
delta_parents = self._index.get_parents(parent)
593
# We couldn't find a fulltext, so we must create a new one
596
return fulltext_size > delta_size
598
def _add_raw_records(self, records, data):
599
"""Add all the records 'records' with data pre-joined in 'data'.
601
:param records: A list of tuples(version_id, options, parents, size).
602
:param data: The data for the records. When it is written, the records
603
are adjusted to have pos pointing into data by the sum of
604
the preceding records sizes.
607
raw_record_sizes = [record[3] for record in records]
608
positions = self._data.add_raw_records(raw_record_sizes, data)
611
for (version_id, options, parents, size), access_memo in zip(
613
index_entries.append((version_id, options, access_memo, parents))
614
if self._data._do_cache:
615
self._data._cache[version_id] = data[offset:offset+size]
617
self._index.add_versions(index_entries)
619
def enable_cache(self):
620
"""Start caching data for this knit"""
621
self._data.enable_cache()
623
def clear_cache(self):
624
"""Clear the data cache only."""
625
self._data.clear_cache()
627
def copy_to(self, name, transport):
628
"""See VersionedFile.copy_to()."""
629
# copy the current index to a temp index to avoid racing with local
631
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
632
self.transport.get(self._index._filename))
634
f = self._data._open_file()
636
transport.put_file(name + DATA_SUFFIX, f)
639
# move the copied index into place
640
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
642
def create_empty(self, name, transport, mode=None):
643
return KnitVersionedFile(name, transport, factory=self.factory,
644
delta=self.delta, create=True)
646
def get_data_stream(self, required_versions):
647
"""Get a data stream for the specified versions.
649
Versions may be returned in any order, not necessarily the order
650
specified. They are returned in a partial order by compression
651
parent, so that the deltas can be applied as the data stream is
652
inserted; however note that compression parents will not be sent
653
unless they were specifically requested, as the client may already
656
:param required_versions: The exact set of versions to be extracted.
657
Unlike some other knit methods, this is not used to generate a
658
transitive closure, rather it is used precisely as given.
660
:returns: format_signature, list of (version, options, length, parents),
663
required_version_set = frozenset(required_versions)
665
# list of revisions that can just be sent without waiting for their
668
# map from revision to the children based on it
670
# first, read all relevant index data, enough to sort into the right
672
for version_id in required_versions:
673
options = self._index.get_options(version_id)
674
parents = self._index.get_parents_with_ghosts(version_id)
675
index_memo = self._index.get_position(version_id)
676
version_index[version_id] = (index_memo, options, parents)
677
if ('line-delta' in options
678
and parents[0] in required_version_set):
679
# must wait until the parent has been sent
680
deferred.setdefault(parents[0], []). \
683
# either a fulltext, or a delta whose parent the client did
684
# not ask for and presumably already has
685
ready_to_send.append(version_id)
686
# build a list of results to return, plus instructions for data to
688
copy_queue_records = []
689
temp_version_list = []
691
# XXX: pushing and popping lists may be a bit inefficient
692
version_id = ready_to_send.pop(0)
693
(index_memo, options, parents) = version_index[version_id]
694
copy_queue_records.append((version_id, index_memo))
695
none, data_pos, data_size = index_memo
696
temp_version_list.append((version_id, options, data_size,
698
if version_id in deferred:
699
# now we can send all the children of this revision - we could
700
# put them in anywhere, but we hope that sending them soon
701
# after the fulltext will give good locality in the receiver
702
ready_to_send[:0] = deferred.pop(version_id)
703
assert len(deferred) == 0, \
704
"Still have compressed child versions waiting to be sent"
705
# XXX: The stream format is such that we cannot stream it - we have to
706
# know the length of all the data a-priori.
708
result_version_list = []
709
for (version_id, raw_data), \
710
(version_id2, options, _, parents) in \
711
izip(self._data.read_records_iter_raw(copy_queue_records),
713
assert version_id == version_id2, \
714
'logic error, inconsistent results'
715
raw_datum.append(raw_data)
716
result_version_list.append(
717
(version_id, options, len(raw_data), parents))
718
# provide a callback to get data incrementally.
719
pseudo_file = StringIO(''.join(raw_datum))
722
return pseudo_file.read()
724
return pseudo_file.read(length)
725
return (self.get_format_signature(), result_version_list, read)
727
def _extract_blocks(self, version_id, source, target):
728
if self._index.get_method(version_id) != 'line-delta':
730
parent, sha1, noeol, delta = self.get_delta(version_id)
731
return KnitContent.get_line_delta_blocks(delta, source, target)
733
def get_delta(self, version_id):
734
"""Get a delta for constructing version from some other version."""
735
self.check_not_reserved_id(version_id)
736
parents = self.get_parents(version_id)
741
index_memo = self._index.get_position(version_id)
742
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
743
noeol = 'no-eol' in self._index.get_options(version_id)
744
if 'fulltext' == self._index.get_method(version_id):
745
new_content = self.factory.parse_fulltext(data, version_id)
746
if parent is not None:
747
reference_content = self._get_content(parent)
748
old_texts = reference_content.text()
751
new_texts = new_content.text()
752
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
754
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
756
delta = self.factory.parse_line_delta(data, version_id)
757
return parent, sha1, noeol, delta
759
def get_format_signature(self):
760
"""See VersionedFile.get_format_signature()."""
761
if self.factory.annotated:
762
annotated_part = "annotated"
764
annotated_part = "plain"
765
return "knit-%s" % (annotated_part,)
767
def get_graph_with_ghosts(self):
768
"""See VersionedFile.get_graph_with_ghosts()."""
769
graph_items = self._index.get_graph()
770
return dict(graph_items)
772
def get_sha1(self, version_id):
773
return self.get_sha1s([version_id])[0]
775
def get_sha1s(self, version_ids):
776
"""See VersionedFile.get_sha1()."""
777
record_map = self._get_record_map(version_ids)
778
# record entry 2 is the 'digest'.
779
return [record_map[v][2] for v in version_ids]
783
"""See VersionedFile.get_suffixes()."""
784
return [DATA_SUFFIX, INDEX_SUFFIX]
786
def has_ghost(self, version_id):
787
"""True if there is a ghost reference in the file to version_id."""
789
if self.has_version(version_id):
791
# optimisable if needed by memoising the _ghosts set.
792
items = self._index.get_graph()
793
for node, parents in items:
794
for parent in parents:
795
if parent not in self._index._cache:
796
if parent == version_id:
800
def insert_data_stream(self, (format, data_list, reader_callable)):
801
"""Insert knit records from a data stream into this knit.
803
If a version in the stream is already present in this knit, it will not
804
be inserted a second time. It will be checked for consistency with the
805
stored version however, and may cause a KnitCorrupt error to be raised
806
if the data in the stream disagrees with the already stored data.
808
:seealso: get_data_stream
810
if format != self.get_format_signature():
811
if 'knit' in debug.debug_flags:
813
'incompatible format signature inserting to %r', self)
814
source = self._knit_from_datastream(
815
(format, data_list, reader_callable))
819
for version_id, options, length, parents in data_list:
820
if self.has_version(version_id):
821
# First check: the list of parents.
822
my_parents = self.get_parents_with_ghosts(version_id)
823
if tuple(my_parents) != tuple(parents):
824
# XXX: KnitCorrupt is not quite the right exception here.
827
'parents list %r from data stream does not match '
828
'already recorded parents %r for %s'
829
% (parents, my_parents, version_id))
831
# Also check the SHA-1 of the fulltext this content will
833
raw_data = reader_callable(length)
834
my_fulltext_sha1 = self.get_sha1(version_id)
835
df, rec = self._data._parse_record_header(version_id, raw_data)
836
stream_fulltext_sha1 = rec[3]
837
if my_fulltext_sha1 != stream_fulltext_sha1:
838
# Actually, we don't know if it's this knit that's corrupt,
839
# or the data stream we're trying to insert.
841
self.filename, 'sha-1 does not match %s' % version_id)
843
if 'line-delta' in options:
844
# Make sure that this knit record is actually useful: a
845
# line-delta is no use unless we have its parent.
846
# Fetching from a broken repository with this problem
847
# shouldn't break the target repository.
849
# See https://bugs.launchpad.net/bzr/+bug/164443
850
if not self._index.has_version(parents[0]):
853
'line-delta from stream '
856
'missing parent %s\n'
857
'Try running "bzr check" '
858
'on the source repository, and "bzr reconcile" '
860
(version_id, parents[0]))
861
self._add_raw_records(
862
[(version_id, options, parents, length)],
863
reader_callable(length))
865
def _knit_from_datastream(self, (format, data_list, reader_callable)):
866
"""Create a knit object from a data stream.
868
This method exists to allow conversion of data streams that do not
869
match the signature of this knit. Generally it will be slower and use
870
more memory to use this method to insert data, but it will work.
872
:seealso: get_data_stream for details on datastreams.
873
:return: A knit versioned file which can be used to join the datastream
876
if format == "knit-plain":
877
factory = KnitPlainFactory()
878
elif format == "knit-annotated":
879
factory = KnitAnnotateFactory()
881
raise errors.KnitDataStreamUnknown(format)
882
index = _StreamIndex(data_list, self._index)
883
access = _StreamAccess(reader_callable, index, self, factory)
884
return KnitVersionedFile(self.filename, self.transport,
885
factory=factory, index=index, access_method=access)
888
"""See VersionedFile.versions."""
889
if 'evil' in debug.debug_flags:
890
trace.mutter_callsite(2, "versions scales with size of history")
891
return self._index.get_versions()
893
def has_version(self, version_id):
894
"""See VersionedFile.has_version."""
895
if 'evil' in debug.debug_flags:
896
trace.mutter_callsite(2, "has_version is a LBYL scenario")
897
return self._index.has_version(version_id)
899
__contains__ = has_version
901
def _merge_annotations(self, content, parents, parent_texts={},
902
delta=None, annotated=None,
903
left_matching_blocks=None):
904
"""Merge annotations for content. This is done by comparing
905
the annotations based on changed to the text.
907
if left_matching_blocks is not None:
908
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
912
for parent_id in parents:
913
merge_content = self._get_content(parent_id, parent_texts)
914
if (parent_id == parents[0] and delta_seq is not None):
917
seq = patiencediff.PatienceSequenceMatcher(
918
None, merge_content.text(), content.text())
919
for i, j, n in seq.get_matching_blocks():
922
# this appears to copy (origin, text) pairs across to the
923
# new content for any line that matches the last-checked
925
content._lines[j:j+n] = merge_content._lines[i:i+n]
927
if delta_seq is None:
928
reference_content = self._get_content(parents[0], parent_texts)
929
new_texts = content.text()
930
old_texts = reference_content.text()
931
delta_seq = patiencediff.PatienceSequenceMatcher(
932
None, old_texts, new_texts)
933
return self._make_line_delta(delta_seq, content)
935
def _make_line_delta(self, delta_seq, new_content):
936
"""Generate a line delta from delta_seq and new_content."""
938
for op in delta_seq.get_opcodes():
941
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
944
def _get_components_positions(self, version_ids):
945
"""Produce a map of position data for the components of versions.
947
This data is intended to be used for retrieving the knit records.
949
A dict of version_id to (record_details, index_memo, next, parents) is
951
method is the way referenced data should be applied.
952
index_memo is the handle to pass to the data access to actually get the
954
next is the build-parent of the version, or None for fulltexts.
955
parents is the version_ids of the parents of this version
958
pending_components = version_ids
959
while pending_components:
960
build_details = self._index.get_build_details(pending_components)
961
pending_components = set()
962
for version_id, details in build_details.items():
963
(index_memo, compression_parent, parents, record_details) = details
964
method = record_details[0]
965
if compression_parent is not None:
966
pending_components.add(compression_parent)
967
component_data[version_id] = (record_details, index_memo,
969
return component_data
971
def _get_content(self, version_id, parent_texts={}):
972
"""Returns a content object that makes up the specified
974
cached_version = parent_texts.get(version_id, None)
975
if cached_version is not None:
976
if not self.has_version(version_id):
977
raise RevisionNotPresent(version_id, self.filename)
978
return cached_version
980
text_map, contents_map = self._get_content_maps([version_id])
981
return contents_map[version_id]
983
def _check_versions_present(self, version_ids):
984
"""Check that all specified versions are present."""
985
self._index.check_versions_present(version_ids)
987
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
988
nostore_sha, random_id, check_content):
989
"""See VersionedFile.add_lines_with_ghosts()."""
990
self._check_add(version_id, lines, random_id, check_content)
991
return self._add(version_id, lines, parents, self.delta,
992
parent_texts, None, nostore_sha, random_id)
994
def _add_lines(self, version_id, parents, lines, parent_texts,
995
left_matching_blocks, nostore_sha, random_id, check_content):
996
"""See VersionedFile.add_lines."""
997
self._check_add(version_id, lines, random_id, check_content)
998
self._check_versions_present(parents)
999
return self._add(version_id, lines[:], parents, self.delta,
1000
parent_texts, left_matching_blocks, nostore_sha, random_id)
1002
def _check_add(self, version_id, lines, random_id, check_content):
1003
"""check that version_id and lines are safe to add."""
1004
if contains_whitespace(version_id):
1005
raise InvalidRevisionId(version_id, self.filename)
1006
self.check_not_reserved_id(version_id)
1007
# Technically this could be avoided if we are happy to allow duplicate
1008
# id insertion when other things than bzr core insert texts, but it
1009
# seems useful for folk using the knit api directly to have some safety
1010
# blanket that we can disable.
1011
if not random_id and self.has_version(version_id):
1012
raise RevisionAlreadyPresent(version_id, self.filename)
1014
self._check_lines_not_unicode(lines)
1015
self._check_lines_are_lines(lines)
1017
def _add(self, version_id, lines, parents, delta, parent_texts,
1018
left_matching_blocks, nostore_sha, random_id):
1019
"""Add a set of lines on top of version specified by parents.
1021
If delta is true, compress the text as a line-delta against
1024
Any versions not present will be converted into ghosts.
1026
# first thing, if the content is something we don't need to store, find
1028
line_bytes = ''.join(lines)
1029
digest = sha_string(line_bytes)
1030
if nostore_sha == digest:
1031
raise errors.ExistingContent
1033
present_parents = []
1034
if parent_texts is None:
1036
for parent in parents:
1037
if self.has_version(parent):
1038
present_parents.append(parent)
1040
# can only compress against the left most present parent.
1042
(len(present_parents) == 0 or
1043
present_parents[0] != parents[0])):
1046
text_length = len(line_bytes)
1049
if lines[-1][-1] != '\n':
1050
# copy the contents of lines.
1052
options.append('no-eol')
1053
lines[-1] = lines[-1] + '\n'
1057
# To speed the extract of texts the delta chain is limited
1058
# to a fixed number of deltas. This should minimize both
1059
# I/O and the time spend applying deltas.
1060
delta = self._check_should_delta(present_parents)
1062
assert isinstance(version_id, str)
1063
content = self.factory.make(lines, version_id)
1064
if delta or (self.factory.annotated and len(present_parents) > 0):
1065
# Merge annotations from parent texts if needed.
1066
delta_hunks = self._merge_annotations(content, present_parents,
1067
parent_texts, delta, self.factory.annotated,
1068
left_matching_blocks)
1071
options.append('line-delta')
1072
store_lines = self.factory.lower_line_delta(delta_hunks)
1073
size, bytes = self._data._record_to_data(version_id, digest,
1076
options.append('fulltext')
1077
# isinstance is slower and we have no hierarchy.
1078
if self.factory.__class__ == KnitPlainFactory:
1079
# Use the already joined bytes saving iteration time in
1081
size, bytes = self._data._record_to_data(version_id, digest,
1082
lines, [line_bytes])
1084
# get mixed annotation + content and feed it into the
1086
store_lines = self.factory.lower_fulltext(content)
1087
size, bytes = self._data._record_to_data(version_id, digest,
1090
access_memo = self._data.add_raw_records([size], bytes)[0]
1091
self._index.add_versions(
1092
((version_id, options, access_memo, parents),),
1093
random_id=random_id)
1094
return digest, text_length, content
1096
def check(self, progress_bar=None):
1097
"""See VersionedFile.check()."""
1099
def _clone_text(self, new_version_id, old_version_id, parents):
1100
"""See VersionedFile.clone_text()."""
1101
# FIXME RBC 20060228 make fast by only inserting an index with null
1103
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
1105
def get_lines(self, version_id):
1106
"""See VersionedFile.get_lines()."""
1107
return self.get_line_list([version_id])[0]
1109
def _get_record_map(self, version_ids):
1110
"""Produce a dictionary of knit records.
1112
The keys are version_ids, the values are tuples of (record_details,
1113
content, digest, next).
1114
method is the way the content should be applied.
1115
content is a KnitContent object.
1116
digest is the SHA1 digest of this version id after all steps are done
1117
next is the build-parent of the version, i.e. the leftmost ancestor.
1118
If the method is fulltext, next will be None.
1120
position_map = self._get_components_positions(version_ids)
1121
# c = component_id, r = record_details, i_m = index_memo, n = next
1122
records = [(c, i_m) for c, (r, i_m, n)
1123
in position_map.iteritems()]
1125
for component_id, content, digest in \
1126
self._data.read_records_iter(records):
1127
(record_details, index_memo, next) = position_map[component_id]
1128
record_map[component_id] = record_details, content, digest, next
1132
def get_text(self, version_id):
1133
"""See VersionedFile.get_text"""
1134
return self.get_texts([version_id])[0]
1136
def get_texts(self, version_ids):
1137
return [''.join(l) for l in self.get_line_list(version_ids)]
1139
def get_line_list(self, version_ids):
1140
"""Return the texts of listed versions as a list of strings."""
1141
for version_id in version_ids:
1142
self.check_not_reserved_id(version_id)
1143
text_map, content_map = self._get_content_maps(version_ids)
1144
return [text_map[v] for v in version_ids]
1146
_get_lf_split_line_list = get_line_list
1148
def _get_content_maps(self, version_ids):
1149
"""Produce maps of text and KnitContents
1151
:return: (text_map, content_map) where text_map contains the texts for
1152
the requested versions and content_map contains the KnitContents.
1153
Both dicts take version_ids as their keys.
1155
# FUTURE: This function could be improved for the 'extract many' case
1156
# by tracking each component and only doing the copy when the number of
1157
# children than need to apply delta's to it is > 1 or it is part of the
1159
version_ids = list(version_ids)
1160
multiple_versions = len(version_ids) != 1
1161
record_map = self._get_record_map(version_ids)
1166
for version_id in version_ids:
1169
while cursor is not None:
1170
record_details, data, digest, next = record_map[cursor]
1171
components.append((cursor, record_details, data, digest))
1172
if cursor in content_map:
1177
for (component_id, record_details, data,
1178
digest) in reversed(components):
1179
if component_id in content_map:
1180
content = content_map[component_id]
1182
content, delta = self.factory.parse_record(version_id,
1183
data, record_details, content,
1184
copy_base_content=multiple_versions)
1185
if multiple_versions:
1186
content_map[component_id] = content
1188
content.cleanup_eol(copy_on_mutate=multiple_versions)
1189
final_content[version_id] = content
1191
# digest here is the digest from the last applied component.
1192
text = content.text()
1193
actual_sha = sha_strings(text)
1194
if actual_sha != digest:
1195
raise KnitCorrupt(self.filename,
1197
'\n of reconstructed text does not match'
1199
'\n for version %s' %
1200
(actual_sha, digest, version_id))
1201
text_map[version_id] = text
1202
return text_map, final_content
1204
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1206
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1207
if version_ids is None:
1208
version_ids = self.versions()
1210
pb = progress.DummyProgress()
1211
# we don't care about inclusions, the caller cares.
1212
# but we need to setup a list of records to visit.
1213
# we need version_id, position, length
1214
version_id_records = []
1215
requested_versions = set(version_ids)
1216
# filter for available versions
1217
for version_id in requested_versions:
1218
if not self.has_version(version_id):
1219
raise RevisionNotPresent(version_id, self.filename)
1220
# get a in-component-order queue:
1221
for version_id in self.versions():
1222
if version_id in requested_versions:
1223
index_memo = self._index.get_position(version_id)
1224
version_id_records.append((version_id, index_memo))
1226
total = len(version_id_records)
1227
for version_idx, (version_id, data, sha_value) in \
1228
enumerate(self._data.read_records_iter(version_id_records)):
1229
pb.update('Walking content.', version_idx, total)
1230
method = self._index.get_method(version_id)
1232
assert method in ('fulltext', 'line-delta')
1233
if method == 'fulltext':
1234
line_iterator = self.factory.get_fulltext_content(data)
1236
line_iterator = self.factory.get_linedelta_content(data)
1237
# XXX: It might be more efficient to yield (version_id,
1238
# line_iterator) in the future. However for now, this is a simpler
1239
# change to integrate into the rest of the codebase. RBC 20071110
1240
for line in line_iterator:
1241
yield line, version_id
1243
pb.update('Walking content.', total, total)
1245
def iter_parents(self, version_ids):
1246
"""Iterate through the parents for many version ids.
1248
:param version_ids: An iterable yielding version_ids.
1249
:return: An iterator that yields (version_id, parents). Requested
1250
version_ids not present in the versioned file are simply skipped.
1251
The order is undefined, allowing for different optimisations in
1252
the underlying implementation.
1254
return self._index.iter_parents(version_ids)
1256
def num_versions(self):
1257
"""See VersionedFile.num_versions()."""
1258
return self._index.num_versions()
1260
__len__ = num_versions
1262
def annotate_iter(self, version_id):
1263
"""See VersionedFile.annotate_iter."""
1264
return self.factory.annotate_iter(self, version_id)
1266
def get_parents(self, version_id):
1267
"""See VersionedFile.get_parents."""
1270
# 52554 calls in 1264 872 internal down from 3674
1272
return self._index.get_parents(version_id)
1274
raise RevisionNotPresent(version_id, self.filename)
1276
def get_parents_with_ghosts(self, version_id):
1277
"""See VersionedFile.get_parents."""
1279
return self._index.get_parents_with_ghosts(version_id)
1281
raise RevisionNotPresent(version_id, self.filename)
1283
def get_ancestry(self, versions, topo_sorted=True):
1284
"""See VersionedFile.get_ancestry."""
1285
if isinstance(versions, basestring):
1286
versions = [versions]
1289
return self._index.get_ancestry(versions, topo_sorted)
1291
def get_ancestry_with_ghosts(self, versions):
1292
"""See VersionedFile.get_ancestry_with_ghosts."""
1293
if isinstance(versions, basestring):
1294
versions = [versions]
1297
return self._index.get_ancestry_with_ghosts(versions)
1299
def plan_merge(self, ver_a, ver_b):
1300
"""See VersionedFile.plan_merge."""
1301
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1302
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1303
annotated_a = self.annotate(ver_a)
1304
annotated_b = self.annotate(ver_b)
1305
return merge._plan_annotate_merge(annotated_a, annotated_b,
1306
ancestors_a, ancestors_b)
1309
class _KnitComponentFile(object):
1310
"""One of the files used to implement a knit database"""
1312
def __init__(self, transport, filename, mode, file_mode=None,
1313
create_parent_dir=False, dir_mode=None):
1314
self._transport = transport
1315
self._filename = filename
1317
self._file_mode = file_mode
1318
self._dir_mode = dir_mode
1319
self._create_parent_dir = create_parent_dir
1320
self._need_to_create = False
1322
def _full_path(self):
1323
"""Return the full path to this file."""
1324
return self._transport.base + self._filename
1326
def check_header(self, fp):
1327
line = fp.readline()
1329
# An empty file can actually be treated as though the file doesn't
1331
raise errors.NoSuchFile(self._full_path())
1332
if line != self.HEADER:
1333
raise KnitHeaderError(badline=line,
1334
filename=self._transport.abspath(self._filename))
1337
return '%s(%s)' % (self.__class__.__name__, self._filename)
1340
class _KnitIndex(_KnitComponentFile):
1341
"""Manages knit index file.
1343
The index is already kept in memory and read on startup, to enable
1344
fast lookups of revision information. The cursor of the index
1345
file is always pointing to the end, making it easy to append
1348
_cache is a cache for fast mapping from version id to a Index
1351
_history is a cache for fast mapping from indexes to version ids.
1353
The index data format is dictionary compressed when it comes to
1354
parent references; a index entry may only have parents that with a
1355
lover index number. As a result, the index is topological sorted.
1357
Duplicate entries may be written to the index for a single version id
1358
if this is done then the latter one completely replaces the former:
1359
this allows updates to correct version and parent information.
1360
Note that the two entries may share the delta, and that successive
1361
annotations and references MUST point to the first entry.
1363
The index file on disc contains a header, followed by one line per knit
1364
record. The same revision can be present in an index file more than once.
1365
The first occurrence gets assigned a sequence number starting from 0.
1367
The format of a single line is
1368
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1369
REVISION_ID is a utf8-encoded revision id
1370
FLAGS is a comma separated list of flags about the record. Values include
1371
no-eol, line-delta, fulltext.
1372
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1373
that the the compressed data starts at.
1374
LENGTH is the ascii representation of the length of the data file.
1375
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1377
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1378
revision id already in the knit that is a parent of REVISION_ID.
1379
The ' :' marker is the end of record marker.
1382
when a write is interrupted to the index file, it will result in a line
1383
that does not end in ' :'. If the ' :' is not present at the end of a line,
1384
or at the end of the file, then the record that is missing it will be
1385
ignored by the parser.
1387
When writing new records to the index file, the data is preceded by '\n'
1388
to ensure that records always start on new lines even if the last write was
1389
interrupted. As a result its normal for the last line in the index to be
1390
missing a trailing newline. One can be added with no harmful effects.
1393
HEADER = "# bzr knit index 8\n"
1395
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1396
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1398
def _cache_version(self, version_id, options, pos, size, parents):
1399
"""Cache a version record in the history array and index cache.
1401
This is inlined into _load_data for performance. KEEP IN SYNC.
1402
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1405
# only want the _history index to reference the 1st index entry
1407
if version_id not in self._cache:
1408
index = len(self._history)
1409
self._history.append(version_id)
1411
index = self._cache[version_id][5]
1412
self._cache[version_id] = (version_id,
1419
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1420
create_parent_dir=False, delay_create=False, dir_mode=None):
1421
_KnitComponentFile.__init__(self, transport, filename, mode,
1422
file_mode=file_mode,
1423
create_parent_dir=create_parent_dir,
1426
# position in _history is the 'official' index for a revision
1427
# but the values may have come from a newer entry.
1428
# so - wc -l of a knit index is != the number of unique names
1432
fp = self._transport.get(self._filename)
1434
# _load_data may raise NoSuchFile if the target knit is
1436
_load_data(self, fp)
1440
if mode != 'w' or not create:
1443
self._need_to_create = True
1445
self._transport.put_bytes_non_atomic(
1446
self._filename, self.HEADER, mode=self._file_mode)
1448
def get_graph(self):
1449
"""Return a list of the node:parents lists from this knit index."""
1450
return [(vid, idx[4]) for vid, idx in self._cache.iteritems()]
1452
def get_ancestry(self, versions, topo_sorted=True):
1453
"""See VersionedFile.get_ancestry."""
1454
# get a graph of all the mentioned versions:
1456
pending = set(versions)
1459
version = pending.pop()
1462
parents = [p for p in cache[version][4] if p in cache]
1464
raise RevisionNotPresent(version, self._filename)
1465
# if not completed and not a ghost
1466
pending.update([p for p in parents if p not in graph])
1467
graph[version] = parents
1470
return topo_sort(graph.items())
1472
def get_ancestry_with_ghosts(self, versions):
1473
"""See VersionedFile.get_ancestry_with_ghosts."""
1474
# get a graph of all the mentioned versions:
1475
self.check_versions_present(versions)
1478
pending = set(versions)
1480
version = pending.pop()
1482
parents = cache[version][4]
1488
pending.update([p for p in parents if p not in graph])
1489
graph[version] = parents
1490
return topo_sort(graph.items())
1492
def get_build_details(self, version_ids):
1493
"""Get the method, index_memo and compression parent for version_ids.
1495
:param version_ids: An iterable of version_ids.
1496
:return: A dict of version_id:(index_memo, compression_parent,
1497
parents, record_details).
1499
opaque structure to pass to read_records to extract the raw
1502
Content that this record is built upon, may be None
1504
Logical parents of this node
1506
extra information about the content which needs to be passed to
1507
Factory.parse_record
1510
for version_id in version_ids:
1511
method = self.get_method(version_id)
1512
parents = self.get_parents_with_ghosts(version_id)
1513
if method == 'fulltext':
1514
compression_parent = None
1516
compression_parent = parents[0]
1517
noeol = 'no-eol' in self.get_options(version_id)
1518
index_memo = self.get_position(version_id)
1519
result[version_id] = (index_memo, compression_parent,
1520
parents, (method, noeol))
1523
def iter_parents(self, version_ids):
1524
"""Iterate through the parents for many version ids.
1526
:param version_ids: An iterable yielding version_ids.
1527
:return: An iterator that yields (version_id, parents). Requested
1528
version_ids not present in the versioned file are simply skipped.
1529
The order is undefined, allowing for different optimisations in
1530
the underlying implementation.
1532
for version_id in version_ids:
1534
yield version_id, tuple(self.get_parents(version_id))
1538
def num_versions(self):
1539
return len(self._history)
1541
__len__ = num_versions
1543
def get_versions(self):
1544
"""Get all the versions in the file. not topologically sorted."""
1545
return self._history
1547
def _version_list_to_index(self, versions):
1550
for version in versions:
1551
if version in cache:
1552
# -- inlined lookup() --
1553
result_list.append(str(cache[version][5]))
1554
# -- end lookup () --
1556
result_list.append('.' + version)
1557
return ' '.join(result_list)
1559
def add_version(self, version_id, options, index_memo, parents):
1560
"""Add a version record to the index."""
1561
self.add_versions(((version_id, options, index_memo, parents),))
1563
def add_versions(self, versions, random_id=False):
1564
"""Add multiple versions to the index.
1566
:param versions: a list of tuples:
1567
(version_id, options, pos, size, parents).
1568
:param random_id: If True the ids being added were randomly generated
1569
and no check for existence will be performed.
1572
orig_history = self._history[:]
1573
orig_cache = self._cache.copy()
1576
for version_id, options, (index, pos, size), parents in versions:
1577
line = "\n%s %s %s %s %s :" % (version_id,
1581
self._version_list_to_index(parents))
1582
assert isinstance(line, str), \
1583
'content must be utf-8 encoded: %r' % (line,)
1585
self._cache_version(version_id, options, pos, size, parents)
1586
if not self._need_to_create:
1587
self._transport.append_bytes(self._filename, ''.join(lines))
1590
sio.write(self.HEADER)
1591
sio.writelines(lines)
1593
self._transport.put_file_non_atomic(self._filename, sio,
1594
create_parent_dir=self._create_parent_dir,
1595
mode=self._file_mode,
1596
dir_mode=self._dir_mode)
1597
self._need_to_create = False
1599
# If any problems happen, restore the original values and re-raise
1600
self._history = orig_history
1601
self._cache = orig_cache
1604
def has_version(self, version_id):
1605
"""True if the version is in the index."""
1606
return version_id in self._cache
1608
def get_position(self, version_id):
1609
"""Return details needed to access the version.
1611
.kndx indices do not support split-out data, so return None for the
1614
:return: a tuple (None, data position, size) to hand to the access
1615
logic to get the record.
1617
entry = self._cache[version_id]
1618
return None, entry[2], entry[3]
1620
def get_method(self, version_id):
1621
"""Return compression method of specified version."""
1623
options = self._cache[version_id][1]
1625
raise RevisionNotPresent(version_id, self._filename)
1626
if 'fulltext' in options:
1629
if 'line-delta' not in options:
1630
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1633
def get_options(self, version_id):
1634
"""Return a list representing options.
1638
return self._cache[version_id][1]
1640
def get_parents(self, version_id):
1641
"""Return parents of specified version ignoring ghosts."""
1642
return [parent for parent in self._cache[version_id][4]
1643
if parent in self._cache]
1645
def get_parents_with_ghosts(self, version_id):
1646
"""Return parents of specified version with ghosts."""
1647
return self._cache[version_id][4]
1649
def check_versions_present(self, version_ids):
1650
"""Check that all specified versions are present."""
1652
for version_id in version_ids:
1653
if version_id not in cache:
1654
raise RevisionNotPresent(version_id, self._filename)
1657
class KnitGraphIndex(object):
1658
"""A knit index that builds on GraphIndex."""
1660
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1661
"""Construct a KnitGraphIndex on a graph_index.
1663
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1664
:param deltas: Allow delta-compressed records.
1665
:param add_callback: If not None, allow additions to the index and call
1666
this callback with a list of added GraphIndex nodes:
1667
[(node, value, node_refs), ...]
1668
:param parents: If True, record knits parents, if not do not record
1671
self._graph_index = graph_index
1672
self._deltas = deltas
1673
self._add_callback = add_callback
1674
self._parents = parents
1675
if deltas and not parents:
1676
raise KnitCorrupt(self, "Cannot do delta compression without "
1679
def _get_entries(self, keys, check_present=False):
1680
"""Get the entries for keys.
1682
:param keys: An iterable of index keys, - 1-tuples.
1687
for node in self._graph_index.iter_entries(keys):
1689
found_keys.add(node[1])
1691
# adapt parentless index to the rest of the code.
1692
for node in self._graph_index.iter_entries(keys):
1693
yield node[0], node[1], node[2], ()
1694
found_keys.add(node[1])
1696
missing_keys = keys.difference(found_keys)
1698
raise RevisionNotPresent(missing_keys.pop(), self)
1700
def _present_keys(self, version_ids):
1702
node[1] for node in self._get_entries(version_ids)])
1704
def _parentless_ancestry(self, versions):
1705
"""Honour the get_ancestry API for parentless knit indices."""
1706
wanted_keys = self._version_ids_to_keys(versions)
1707
present_keys = self._present_keys(wanted_keys)
1708
missing = set(wanted_keys).difference(present_keys)
1710
raise RevisionNotPresent(missing.pop(), self)
1711
return list(self._keys_to_version_ids(present_keys))
1713
def get_ancestry(self, versions, topo_sorted=True):
1714
"""See VersionedFile.get_ancestry."""
1715
if not self._parents:
1716
return self._parentless_ancestry(versions)
1717
# XXX: This will do len(history) index calls - perhaps
1718
# it should be altered to be a index core feature?
1719
# get a graph of all the mentioned versions:
1722
versions = self._version_ids_to_keys(versions)
1723
pending = set(versions)
1725
# get all pending nodes
1726
this_iteration = pending
1727
new_nodes = self._get_entries(this_iteration)
1730
for (index, key, value, node_refs) in new_nodes:
1731
# dont ask for ghosties - otherwise
1732
# we we can end up looping with pending
1733
# being entirely ghosted.
1734
graph[key] = [parent for parent in node_refs[0]
1735
if parent not in ghosts]
1737
for parent in graph[key]:
1738
# dont examine known nodes again
1743
ghosts.update(this_iteration.difference(found))
1744
if versions.difference(graph):
1745
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1747
result_keys = topo_sort(graph.items())
1749
result_keys = graph.iterkeys()
1750
return [key[0] for key in result_keys]
1752
def get_ancestry_with_ghosts(self, versions):
1753
"""See VersionedFile.get_ancestry."""
1754
if not self._parents:
1755
return self._parentless_ancestry(versions)
1756
# XXX: This will do len(history) index calls - perhaps
1757
# it should be altered to be a index core feature?
1758
# get a graph of all the mentioned versions:
1760
versions = self._version_ids_to_keys(versions)
1761
pending = set(versions)
1763
# get all pending nodes
1764
this_iteration = pending
1765
new_nodes = self._get_entries(this_iteration)
1767
for (index, key, value, node_refs) in new_nodes:
1768
graph[key] = node_refs[0]
1770
for parent in graph[key]:
1771
# dont examine known nodes again
1775
missing_versions = this_iteration.difference(graph)
1776
missing_needed = versions.intersection(missing_versions)
1778
raise RevisionNotPresent(missing_needed.pop(), self)
1779
for missing_version in missing_versions:
1780
# add a key, no parents
1781
graph[missing_version] = []
1782
pending.discard(missing_version) # don't look for it
1783
result_keys = topo_sort(graph.items())
1784
return [key[0] for key in result_keys]
1786
def get_build_details(self, version_ids):
1787
"""Get the method, index_memo and compression parent for version_ids.
1789
:param version_ids: An iterable of version_ids.
1790
:return: A dict of version_id:(method, index_memo, compression_parent,
1794
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
1795
for entry in entries:
1796
version_id = self._keys_to_version_ids((entry[1],))[0]
1797
parents = self._keys_to_version_ids(entry[3][0])
1798
if not self._deltas:
1799
compression_parent = None
1801
compression_parent_key = self._compression_parent(entry)
1802
if compression_parent_key:
1803
compression_parent = self._keys_to_version_ids(
1804
(compression_parent_key,))[0]
1806
compression_parent = None
1807
noeol = (entry[2][0] == 'N')
1808
if compression_parent:
1809
method = 'line-delta'
1812
result[version_id] = (self._node_to_position(entry),
1813
compression_parent, parents,
1817
def _compression_parent(self, an_entry):
1818
# return the key that an_entry is compressed against, or None
1819
# Grab the second parent list (as deltas implies parents currently)
1820
compression_parents = an_entry[3][1]
1821
if not compression_parents:
1823
assert len(compression_parents) == 1
1824
return compression_parents[0]
1826
def _get_method(self, node):
1827
if not self._deltas:
1829
if self._compression_parent(node):
1834
def get_graph(self):
1835
"""Return a list of the node:parents lists from this knit index."""
1836
if not self._parents:
1837
return [(key, ()) for key in self.get_versions()]
1839
for index, key, value, refs in self._graph_index.iter_all_entries():
1840
result.append((key[0], tuple([ref[0] for ref in refs[0]])))
1843
def iter_parents(self, version_ids):
1844
"""Iterate through the parents for many version ids.
1846
:param version_ids: An iterable yielding version_ids.
1847
:return: An iterator that yields (version_id, parents). Requested
1848
version_ids not present in the versioned file are simply skipped.
1849
The order is undefined, allowing for different optimisations in
1850
the underlying implementation.
1853
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1855
present_parents = set()
1856
for node in all_nodes:
1857
all_parents.update(node[3][0])
1858
# any node we are querying must be present
1859
present_parents.add(node[1])
1860
unknown_parents = all_parents.difference(present_parents)
1861
present_parents.update(self._present_keys(unknown_parents))
1862
for node in all_nodes:
1864
for parent in node[3][0]:
1865
if parent in present_parents:
1866
parents.append(parent[0])
1867
yield node[1][0], tuple(parents)
1869
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1870
yield node[1][0], ()
1872
def num_versions(self):
1873
return len(list(self._graph_index.iter_all_entries()))
1875
__len__ = num_versions
1877
def get_versions(self):
1878
"""Get all the versions in the file. not topologically sorted."""
1879
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1881
def has_version(self, version_id):
1882
"""True if the version is in the index."""
1883
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1885
def _keys_to_version_ids(self, keys):
1886
return tuple(key[0] for key in keys)
1888
def get_position(self, version_id):
1889
"""Return details needed to access the version.
1891
:return: a tuple (index, data position, size) to hand to the access
1892
logic to get the record.
1894
node = self._get_node(version_id)
1895
return self._node_to_position(node)
1897
def _node_to_position(self, node):
1898
"""Convert an index value to position details."""
1899
bits = node[2][1:].split(' ')
1900
return node[0], int(bits[0]), int(bits[1])
1902
def get_method(self, version_id):
1903
"""Return compression method of specified version."""
1904
return self._get_method(self._get_node(version_id))
1906
def _get_node(self, version_id):
1908
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1910
raise RevisionNotPresent(version_id, self)
1912
def get_options(self, version_id):
1913
"""Return a list representing options.
1917
node = self._get_node(version_id)
1918
options = [self._get_method(node)]
1919
if node[2][0] == 'N':
1920
options.append('no-eol')
1923
def get_parents(self, version_id):
1924
"""Return parents of specified version ignoring ghosts."""
1925
parents = list(self.iter_parents([version_id]))
1928
raise errors.RevisionNotPresent(version_id, self)
1929
return parents[0][1]
1931
def get_parents_with_ghosts(self, version_id):
1932
"""Return parents of specified version with ghosts."""
1933
nodes = list(self._get_entries(self._version_ids_to_keys([version_id]),
1934
check_present=True))
1935
if not self._parents:
1937
return self._keys_to_version_ids(nodes[0][3][0])
1939
def check_versions_present(self, version_ids):
1940
"""Check that all specified versions are present."""
1941
keys = self._version_ids_to_keys(version_ids)
1942
present = self._present_keys(keys)
1943
missing = keys.difference(present)
1945
raise RevisionNotPresent(missing.pop(), self)
1947
def add_version(self, version_id, options, access_memo, parents):
1948
"""Add a version record to the index."""
1949
return self.add_versions(((version_id, options, access_memo, parents),))
1951
def add_versions(self, versions, random_id=False):
1952
"""Add multiple versions to the index.
1954
This function does not insert data into the Immutable GraphIndex
1955
backing the KnitGraphIndex, instead it prepares data for insertion by
1956
the caller and checks that it is safe to insert then calls
1957
self._add_callback with the prepared GraphIndex nodes.
1959
:param versions: a list of tuples:
1960
(version_id, options, pos, size, parents).
1961
:param random_id: If True the ids being added were randomly generated
1962
and no check for existence will be performed.
1964
if not self._add_callback:
1965
raise errors.ReadOnlyError(self)
1966
# we hope there are no repositories with inconsistent parentage
1971
for (version_id, options, access_memo, parents) in versions:
1972
index, pos, size = access_memo
1973
key = (version_id, )
1974
parents = tuple((parent, ) for parent in parents)
1975
if 'no-eol' in options:
1979
value += "%d %d" % (pos, size)
1980
if not self._deltas:
1981
if 'line-delta' in options:
1982
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1985
if 'line-delta' in options:
1986
node_refs = (parents, (parents[0],))
1988
node_refs = (parents, ())
1990
node_refs = (parents, )
1993
raise KnitCorrupt(self, "attempt to add node with parents "
1994
"in parentless index.")
1996
keys[key] = (value, node_refs)
1998
present_nodes = self._get_entries(keys)
1999
for (index, key, value, node_refs) in present_nodes:
2000
if (value, node_refs) != keys[key]:
2001
raise KnitCorrupt(self, "inconsistent details in add_versions"
2002
": %s %s" % ((value, node_refs), keys[key]))
2006
for key, (value, node_refs) in keys.iteritems():
2007
result.append((key, value, node_refs))
2009
for key, (value, node_refs) in keys.iteritems():
2010
result.append((key, value))
2011
self._add_callback(result)
2013
def _version_ids_to_keys(self, version_ids):
2014
return set((version_id, ) for version_id in version_ids)
2017
class _KnitAccess(object):
2018
"""Access to knit records in a .knit file."""
2020
def __init__(self, transport, filename, _file_mode, _dir_mode,
2021
_need_to_create, _create_parent_dir):
2022
"""Create a _KnitAccess for accessing and inserting data.
2024
:param transport: The transport the .knit is located on.
2025
:param filename: The filename of the .knit.
2027
self._transport = transport
2028
self._filename = filename
2029
self._file_mode = _file_mode
2030
self._dir_mode = _dir_mode
2031
self._need_to_create = _need_to_create
2032
self._create_parent_dir = _create_parent_dir
2034
def add_raw_records(self, sizes, raw_data):
2035
"""Add raw knit bytes to a storage area.
2037
The data is spooled to whereever the access method is storing data.
2039
:param sizes: An iterable containing the size of each raw data segment.
2040
:param raw_data: A bytestring containing the data.
2041
:return: A list of memos to retrieve the record later. Each memo is a
2042
tuple - (index, pos, length), where the index field is always None
2043
for the .knit access method.
2045
assert type(raw_data) == str, \
2046
'data must be plain bytes was %s' % type(raw_data)
2047
if not self._need_to_create:
2048
base = self._transport.append_bytes(self._filename, raw_data)
2050
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2051
create_parent_dir=self._create_parent_dir,
2052
mode=self._file_mode,
2053
dir_mode=self._dir_mode)
2054
self._need_to_create = False
2058
result.append((None, base, size))
2063
"""IFF this data access has its own storage area, initialise it.
2067
self._transport.put_bytes_non_atomic(self._filename, '',
2068
mode=self._file_mode)
2070
def open_file(self):
2071
"""IFF this data access can be represented as a single file, open it.
2073
For knits that are not mapped to a single file on disk this will
2076
:return: None or a file handle.
2079
return self._transport.get(self._filename)
2084
def get_raw_records(self, memos_for_retrieval):
2085
"""Get the raw bytes for a records.
2087
:param memos_for_retrieval: An iterable containing the (index, pos,
2088
length) memo for retrieving the bytes. The .knit method ignores
2089
the index as there is always only a single file.
2090
:return: An iterator over the bytes of the records.
2092
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2093
for pos, data in self._transport.readv(self._filename, read_vector):
2097
class _PackAccess(object):
2098
"""Access to knit records via a collection of packs."""
2100
def __init__(self, index_to_packs, writer=None):
2101
"""Create a _PackAccess object.
2103
:param index_to_packs: A dict mapping index objects to the transport
2104
and file names for obtaining data.
2105
:param writer: A tuple (pack.ContainerWriter, write_index) which
2106
contains the pack to write, and the index that reads from it will
2110
self.container_writer = writer[0]
2111
self.write_index = writer[1]
2113
self.container_writer = None
2114
self.write_index = None
2115
self.indices = index_to_packs
2117
def add_raw_records(self, sizes, raw_data):
2118
"""Add raw knit bytes to a storage area.
2120
The data is spooled to the container writer in one bytes-record per
2123
:param sizes: An iterable containing the size of each raw data segment.
2124
:param raw_data: A bytestring containing the data.
2125
:return: A list of memos to retrieve the record later. Each memo is a
2126
tuple - (index, pos, length), where the index field is the
2127
write_index object supplied to the PackAccess object.
2129
assert type(raw_data) == str, \
2130
'data must be plain bytes was %s' % type(raw_data)
2134
p_offset, p_length = self.container_writer.add_bytes_record(
2135
raw_data[offset:offset+size], [])
2137
result.append((self.write_index, p_offset, p_length))
2141
"""Pack based knits do not get individually created."""
2143
def get_raw_records(self, memos_for_retrieval):
2144
"""Get the raw bytes for a records.
2146
:param memos_for_retrieval: An iterable containing the (index, pos,
2147
length) memo for retrieving the bytes. The Pack access method
2148
looks up the pack to use for a given record in its index_to_pack
2150
:return: An iterator over the bytes of the records.
2152
# first pass, group into same-index requests
2154
current_index = None
2155
for (index, offset, length) in memos_for_retrieval:
2156
if current_index == index:
2157
current_list.append((offset, length))
2159
if current_index is not None:
2160
request_lists.append((current_index, current_list))
2161
current_index = index
2162
current_list = [(offset, length)]
2163
# handle the last entry
2164
if current_index is not None:
2165
request_lists.append((current_index, current_list))
2166
for index, offsets in request_lists:
2167
transport, path = self.indices[index]
2168
reader = pack.make_readv_reader(transport, path, offsets)
2169
for names, read_func in reader.iter_records():
2170
yield read_func(None)
2172
def open_file(self):
2173
"""Pack based knits have no single file."""
2176
def set_writer(self, writer, index, (transport, packname)):
2177
"""Set a writer to use for adding data."""
2178
if index is not None:
2179
self.indices[index] = (transport, packname)
2180
self.container_writer = writer
2181
self.write_index = index
2184
class _StreamAccess(object):
2185
"""A Knit Access object that provides data from a datastream.
2187
It also provides a fallback to present as unannotated data, annotated data
2188
from a *backing* access object.
2190
This is triggered by a index_memo which is pointing to a different index
2191
than this was constructed with, and is used to allow extracting full
2192
unannotated texts for insertion into annotated knits.
2195
def __init__(self, reader_callable, stream_index, backing_knit,
2197
"""Create a _StreamAccess object.
2199
:param reader_callable: The reader_callable from the datastream.
2200
This is called to buffer all the data immediately, for
2202
:param stream_index: The index the data stream this provides access to
2203
which will be present in native index_memo's.
2204
:param backing_knit: The knit object that will provide access to
2205
annotated texts which are not available in the stream, so as to
2206
create unannotated texts.
2207
:param orig_factory: The original content factory used to generate the
2208
stream. This is used for checking whether the thunk code for
2209
supporting _copy_texts will generate the correct form of data.
2211
self.data = reader_callable(None)
2212
self.stream_index = stream_index
2213
self.backing_knit = backing_knit
2214
self.orig_factory = orig_factory
2216
def get_raw_records(self, memos_for_retrieval):
2217
"""Get the raw bytes for a records.
2219
:param memos_for_retrieval: An iterable containing the (thunk_flag,
2220
index, start, end) memo for retrieving the bytes.
2221
:return: An iterator over the bytes of the records.
2223
# use a generator for memory friendliness
2224
for thunk_flag, version_id, start, end in memos_for_retrieval:
2225
if version_id is self.stream_index:
2226
yield self.data[start:end]
2228
# we have been asked to thunk. This thunking only occurs when
2229
# we are obtaining plain texts from an annotated backing knit
2230
# so that _copy_texts will work.
2231
# We could improve performance here by scanning for where we need
2232
# to do this and using get_line_list, then interleaving the output
2233
# as desired. However, for now, this is sufficient.
2234
if self.orig_factory.__class__ != KnitPlainFactory:
2235
raise errors.KnitCorrupt(
2236
self, 'Bad thunk request %r' % version_id)
2237
lines = self.backing_knit.get_lines(version_id)
2238
line_bytes = ''.join(lines)
2239
digest = sha_string(line_bytes)
2241
if lines[-1][-1] != '\n':
2242
lines[-1] = lines[-1] + '\n'
2244
orig_options = list(self.backing_knit._index.get_options(version_id))
2245
if 'fulltext' not in orig_options:
2246
if 'line-delta' not in orig_options:
2247
raise errors.KnitCorrupt(self,
2248
'Unknown compression method %r' % orig_options)
2249
orig_options.remove('line-delta')
2250
orig_options.append('fulltext')
2251
# We want plain data, because we expect to thunk only to allow text
2253
size, bytes = self.backing_knit._data._record_to_data(version_id,
2254
digest, lines, line_bytes)
2258
class _StreamIndex(object):
2259
"""A Knit Index object that uses the data map from a datastream."""
2261
def __init__(self, data_list, backing_index):
2262
"""Create a _StreamIndex object.
2264
:param data_list: The data_list from the datastream.
2265
:param backing_index: The index which will supply values for nodes
2266
referenced outside of this stream.
2268
self.data_list = data_list
2269
self.backing_index = backing_index
2270
self._by_version = {}
2272
for key, options, length, parents in data_list:
2273
self._by_version[key] = options, (pos, pos + length), parents
2276
def get_ancestry(self, versions, topo_sorted):
2277
"""Get an ancestry list for versions."""
2279
# Not needed for basic joins
2280
raise NotImplementedError(self.get_ancestry)
2281
# get a graph of all the mentioned versions:
2282
# Little ugly - basically copied from KnitIndex, but don't want to
2283
# accidentally incorporate too much of that index's code.
2285
pending = set(versions)
2286
cache = self._by_version
2288
version = pending.pop()
2291
parents = [p for p in cache[version][2] if p in cache]
2293
raise RevisionNotPresent(version, self)
2294
# if not completed and not a ghost
2295
pending.update([p for p in parents if p not in ancestry])
2296
ancestry.add(version)
2297
return list(ancestry)
2299
def get_build_details(self, version_ids):
2300
"""Get the method, index_memo and compression parent for version_ids.
2302
:param version_ids: An iterable of version_ids.
2303
:return: A dict of version_id:(method, index_memo, compression_parent,
2307
for version_id in version_ids:
2308
method = self.get_method(version_id)
2309
parent_ids = self.get_parents_with_ghosts(version_id)
2310
noeol = ('no-eol' in self.get_options(version_id))
2311
if method == 'fulltext':
2312
compression_parent = None
2314
compression_parent = parent_ids[0]
2315
index_memo = self.get_position(version_id)
2316
result[version_id] = (index_memo, compression_parent,
2317
parent_ids, (method, noeol))
2320
def get_method(self, version_id):
2321
"""Return compression method of specified version."""
2323
options = self._by_version[version_id][0]
2325
# Strictly speaking this should check in the backing knit, but
2326
# until we have a test to discriminate, this will do.
2327
return self.backing_index.get_method(version_id)
2328
if 'fulltext' in options:
2330
elif 'line-delta' in options:
2333
raise errors.KnitIndexUnknownMethod(self, options)
2335
def get_options(self, version_id):
2336
"""Return a list representing options.
2341
return self._by_version[version_id][0]
2343
return self.backing_index.get_options(version_id)
2345
def get_parents_with_ghosts(self, version_id):
2346
"""Return parents of specified version with ghosts."""
2348
return self._by_version[version_id][2]
2350
return self.backing_index.get_parents_with_ghosts(version_id)
2352
def get_position(self, version_id):
2353
"""Return details needed to access the version.
2355
_StreamAccess has the data as a big array, so we return slice
2356
coordinates into that (as index_memo's are opaque outside the
2357
index and matching access class).
2359
:return: a tuple (thunk_flag, index, start, end). If thunk_flag is
2360
False, index will be self, otherwise it will be a version id.
2363
start, end = self._by_version[version_id][1]
2364
return False, self, start, end
2366
# Signal to the access object to handle this from the backing knit.
2367
return (True, version_id, None, None)
2369
def get_versions(self):
2370
"""Get all the versions in the stream."""
2371
return self._by_version.keys()
2373
def iter_parents(self, version_ids):
2374
"""Iterate through the parents for many version ids.
2376
:param version_ids: An iterable yielding version_ids.
2377
:return: An iterator that yields (version_id, parents). Requested
2378
version_ids not present in the versioned file are simply skipped.
2379
The order is undefined, allowing for different optimisations in
2380
the underlying implementation.
2383
for version in version_ids:
2385
result.append((version, self._by_version[version][2]))
2391
class _KnitData(object):
2392
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2394
The KnitData class provides the logic for parsing and using knit records,
2395
making use of an access method for the low level read and write operations.
2398
def __init__(self, access):
2399
"""Create a KnitData object.
2401
:param access: The access method to use. Access methods such as
2402
_KnitAccess manage the insertion of raw records and the subsequent
2403
retrieval of the same.
2405
self._access = access
2406
self._checked = False
2407
# TODO: jam 20060713 conceptually, this could spill to disk
2408
# if the cached size gets larger than a certain amount
2409
# but it complicates the model a bit, so for now just use
2410
# a simple dictionary
2412
self._do_cache = False
2414
def enable_cache(self):
2415
"""Enable caching of reads."""
2416
self._do_cache = True
2418
def clear_cache(self):
2419
"""Clear the record cache."""
2420
self._do_cache = False
2423
def _open_file(self):
2424
return self._access.open_file()
2426
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2427
"""Convert version_id, digest, lines into a raw data block.
2429
:param dense_lines: The bytes of lines but in a denser form. For
2430
instance, if lines is a list of 1000 bytestrings each ending in \n,
2431
dense_lines may be a list with one line in it, containing all the
2432
1000's lines and their \n's. Using dense_lines if it is already
2433
known is a win because the string join to create bytes in this
2434
function spends less time resizing the final string.
2435
:return: (len, a StringIO instance with the raw data ready to read.)
2437
# Note: using a string copy here increases memory pressure with e.g.
2438
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2439
# when doing the initial commit of a mozilla tree. RBC 20070921
2440
bytes = ''.join(chain(
2441
["version %s %d %s\n" % (version_id,
2444
dense_lines or lines,
2445
["end %s\n" % version_id]))
2446
assert bytes.__class__ == str
2447
compressed_bytes = bytes_to_gzip(bytes)
2448
return len(compressed_bytes), compressed_bytes
2450
def add_raw_records(self, sizes, raw_data):
2451
"""Append a prepared record to the data file.
2453
:param sizes: An iterable containing the size of each raw data segment.
2454
:param raw_data: A bytestring containing the data.
2455
:return: a list of index data for the way the data was stored.
2456
See the access method add_raw_records documentation for more
2459
return self._access.add_raw_records(sizes, raw_data)
2461
def _parse_record_header(self, version_id, raw_data):
2462
"""Parse a record header for consistency.
2464
:return: the header and the decompressor stream.
2465
as (stream, header_record)
2467
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2469
rec = self._check_header(version_id, df.readline())
2470
except Exception, e:
2471
raise KnitCorrupt(self._access,
2472
"While reading {%s} got %s(%s)"
2473
% (version_id, e.__class__.__name__, str(e)))
2476
def _check_header(self, version_id, line):
2479
raise KnitCorrupt(self._access,
2480
'unexpected number of elements in record header')
2481
if rec[1] != version_id:
2482
raise KnitCorrupt(self._access,
2483
'unexpected version, wanted %r, got %r'
2484
% (version_id, rec[1]))
2487
def _parse_record(self, version_id, data):
2489
# 4168 calls in 2880 217 internal
2490
# 4168 calls to _parse_record_header in 2121
2491
# 4168 calls to readlines in 330
2492
df = GzipFile(mode='rb', fileobj=StringIO(data))
2495
record_contents = df.readlines()
2496
except Exception, e:
2497
raise KnitCorrupt(self._access,
2498
"While reading {%s} got %s(%s)"
2499
% (version_id, e.__class__.__name__, str(e)))
2500
header = record_contents.pop(0)
2501
rec = self._check_header(version_id, header)
2503
last_line = record_contents.pop()
2504
if len(record_contents) != int(rec[2]):
2505
raise KnitCorrupt(self._access,
2506
'incorrect number of lines %s != %s'
2508
% (len(record_contents), int(rec[2]),
2510
if last_line != 'end %s\n' % rec[1]:
2511
raise KnitCorrupt(self._access,
2512
'unexpected version end line %r, wanted %r'
2513
% (last_line, version_id))
2515
return record_contents, rec[3]
2517
def read_records_iter_raw(self, records):
2518
"""Read text records from data file and yield raw data.
2520
This unpacks enough of the text record to validate the id is
2521
as expected but thats all.
2523
# setup an iterator of the external records:
2524
# uses readv so nice and fast we hope.
2526
# grab the disk data needed.
2528
# Don't check _cache if it is empty
2529
needed_offsets = [index_memo for version_id, index_memo
2531
if version_id not in self._cache]
2533
needed_offsets = [index_memo for version_id, index_memo
2536
raw_records = self._access.get_raw_records(needed_offsets)
2538
for version_id, index_memo in records:
2539
if version_id in self._cache:
2540
# This data has already been validated
2541
data = self._cache[version_id]
2543
data = raw_records.next()
2545
self._cache[version_id] = data
2547
# validate the header
2548
df, rec = self._parse_record_header(version_id, data)
2550
yield version_id, data
2552
def read_records_iter(self, records):
2553
"""Read text records from data file and yield result.
2555
The result will be returned in whatever is the fastest to read.
2556
Not by the order requested. Also, multiple requests for the same
2557
record will only yield 1 response.
2558
:param records: A list of (version_id, pos, len) entries
2559
:return: Yields (version_id, contents, digest) in the order
2560
read, not the order requested
2566
# Skip records we have alread seen
2567
yielded_records = set()
2568
needed_records = set()
2569
for record in records:
2570
if record[0] in self._cache:
2571
if record[0] in yielded_records:
2573
yielded_records.add(record[0])
2574
data = self._cache[record[0]]
2575
content, digest = self._parse_record(record[0], data)
2576
yield (record[0], content, digest)
2578
needed_records.add(record)
2579
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2581
needed_records = sorted(set(records), key=operator.itemgetter(1))
2583
if not needed_records:
2586
# The transport optimizes the fetching as well
2587
# (ie, reads continuous ranges.)
2588
raw_data = self._access.get_raw_records(
2589
[index_memo for version_id, index_memo in needed_records])
2591
for (version_id, index_memo), data in \
2592
izip(iter(needed_records), raw_data):
2593
content, digest = self._parse_record(version_id, data)
2595
self._cache[version_id] = data
2596
yield version_id, content, digest
2598
def read_records(self, records):
2599
"""Read records into a dictionary."""
2601
for record_id, content, digest in \
2602
self.read_records_iter(records):
2603
components[record_id] = (content, digest)
2607
class InterKnit(InterVersionedFile):
2608
"""Optimised code paths for knit to knit operations."""
2610
_matching_file_from_factory = KnitVersionedFile
2611
_matching_file_to_factory = KnitVersionedFile
2614
def is_compatible(source, target):
2615
"""Be compatible with knits. """
2617
return (isinstance(source, KnitVersionedFile) and
2618
isinstance(target, KnitVersionedFile))
2619
except AttributeError:
2622
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2623
"""Copy texts to the target by extracting and adding them one by one.
2625
see join() for the parameter definitions.
2627
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2628
graph = self.source.get_graph(version_ids)
2629
order = topo_sort(graph.items())
2631
def size_of_content(content):
2632
return sum(len(line) for line in content.text())
2633
# Cache at most 10MB of parent texts
2634
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2635
compute_size=size_of_content)
2636
# TODO: jam 20071116 It would be nice to have a streaming interface to
2637
# get multiple texts from a source. The source could be smarter
2638
# about how it handled intermediate stages.
2639
# get_line_list() or make_mpdiffs() seem like a possibility, but
2640
# at the moment they extract all full texts into memory, which
2641
# causes us to store more than our 3x fulltext goal.
2642
# Repository.iter_files_bytes() may be another possibility
2643
to_process = [version for version in order
2644
if version not in self.target]
2645
total = len(to_process)
2646
pb = ui.ui_factory.nested_progress_bar()
2648
for index, version in enumerate(to_process):
2649
pb.update('Converting versioned data', index, total)
2650
sha1, num_bytes, parent_text = self.target.add_lines(version,
2651
self.source.get_parents_with_ghosts(version),
2652
self.source.get_lines(version),
2653
parent_texts=parent_cache)
2654
parent_cache[version] = parent_text
2659
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2660
"""See InterVersionedFile.join."""
2661
assert isinstance(self.source, KnitVersionedFile)
2662
assert isinstance(self.target, KnitVersionedFile)
2664
# If the source and target are mismatched w.r.t. annotations vs
2665
# plain, the data needs to be converted accordingly
2666
if self.source.factory.annotated == self.target.factory.annotated:
2668
elif self.source.factory.annotated:
2669
converter = self._anno_to_plain_converter
2671
# We're converting from a plain to an annotated knit. Copy them
2672
# across by full texts.
2673
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2675
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2679
pb = ui.ui_factory.nested_progress_bar()
2681
version_ids = list(version_ids)
2682
if None in version_ids:
2683
version_ids.remove(None)
2685
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2687
this_versions = set(self.target._index.get_versions())
2688
# XXX: For efficiency we should not look at the whole index,
2689
# we only need to consider the referenced revisions - they
2690
# must all be present, or the method must be full-text.
2691
# TODO, RBC 20070919
2692
needed_versions = self.source_ancestry - this_versions
2694
if not needed_versions:
2696
full_list = topo_sort(self.source.get_graph())
2698
version_list = [i for i in full_list if (not self.target.has_version(i)
2699
and i in needed_versions)]
2703
copy_queue_records = []
2705
for version_id in version_list:
2706
options = self.source._index.get_options(version_id)
2707
parents = self.source._index.get_parents_with_ghosts(version_id)
2708
# check that its will be a consistent copy:
2709
for parent in parents:
2710
# if source has the parent, we must :
2711
# * already have it or
2712
# * have it scheduled already
2713
# otherwise we don't care
2714
assert (self.target.has_version(parent) or
2715
parent in copy_set or
2716
not self.source.has_version(parent))
2717
index_memo = self.source._index.get_position(version_id)
2718
copy_queue_records.append((version_id, index_memo))
2719
copy_queue.append((version_id, options, parents))
2720
copy_set.add(version_id)
2722
# data suck the join:
2724
total = len(version_list)
2727
for (version_id, raw_data), \
2728
(version_id2, options, parents) in \
2729
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2731
assert version_id == version_id2, 'logic error, inconsistent results'
2733
pb.update("Joining knit", count, total)
2735
size, raw_data = converter(raw_data, version_id, options,
2738
size = len(raw_data)
2739
raw_records.append((version_id, options, parents, size))
2740
raw_datum.append(raw_data)
2741
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2746
def _anno_to_plain_converter(self, raw_data, version_id, options,
2748
"""Convert annotated content to plain content."""
2749
data, digest = self.source._data._parse_record(version_id, raw_data)
2750
if 'fulltext' in options:
2751
content = self.source.factory.parse_fulltext(data, version_id)
2752
lines = self.target.factory.lower_fulltext(content)
2754
delta = self.source.factory.parse_line_delta(data, version_id,
2756
lines = self.target.factory.lower_line_delta(delta)
2757
return self.target._data._record_to_data(version_id, digest, lines)
2760
InterVersionedFile.register_optimiser(InterKnit)
2763
class WeaveToKnit(InterVersionedFile):
2764
"""Optimised code paths for weave to knit operations."""
2766
_matching_file_from_factory = bzrlib.weave.WeaveFile
2767
_matching_file_to_factory = KnitVersionedFile
2770
def is_compatible(source, target):
2771
"""Be compatible with weaves to knits."""
2773
return (isinstance(source, bzrlib.weave.Weave) and
2774
isinstance(target, KnitVersionedFile))
2775
except AttributeError:
2778
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2779
"""See InterVersionedFile.join."""
2780
assert isinstance(self.source, bzrlib.weave.Weave)
2781
assert isinstance(self.target, KnitVersionedFile)
2783
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2788
pb = ui.ui_factory.nested_progress_bar()
2790
version_ids = list(version_ids)
2792
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2793
this_versions = set(self.target._index.get_versions())
2794
needed_versions = self.source_ancestry - this_versions
2796
if not needed_versions:
2798
full_list = topo_sort(self.source.get_graph())
2800
version_list = [i for i in full_list if (not self.target.has_version(i)
2801
and i in needed_versions)]
2805
total = len(version_list)
2806
for version_id in version_list:
2807
pb.update("Converting to knit", count, total)
2808
parents = self.source.get_parents(version_id)
2809
# check that its will be a consistent copy:
2810
for parent in parents:
2811
# if source has the parent, we must already have it
2812
assert (self.target.has_version(parent))
2813
self.target.add_lines(
2814
version_id, parents, self.source.get_lines(version_id))
2821
InterVersionedFile.register_optimiser(WeaveToKnit)
2824
# Deprecated, use PatienceSequenceMatcher instead
2825
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2828
def annotate_knit(knit, revision_id):
2829
"""Annotate a knit with no cached annotations.
2831
This implementation is for knits with no cached annotations.
2832
It will work for knits with cached annotations, but this is not
2835
annotator = _KnitAnnotator(knit)
2836
return iter(annotator.get_annotated_lines(revision_id))
2839
class _KnitAnnotator(object):
2840
"""Build up the annotations for a text."""
2842
def __init__(self, knit):
2845
# unannotated lines of various revisions, this will have the final
2847
self._fulltexts = {}
2848
# Content objects, differs from fulltexts because of how final newlines
2849
# are treated by knits. the content objects here will always have a
2851
self._fulltext_contents = {}
2853
# Annotated lines of specific revisions
2854
self._annotated_lines = {}
2856
# Track the raw data for nodes that we could not process yet.
2857
# This maps the revision_id of the base to a list of children that will
2858
# annotated from it.
2859
self._pending_children = {}
2861
self._all_build_details = {}
2862
# The children => parent revision_id graph
2863
self._revision_id_graph = {}
2865
self._heads_provider = None
2867
def _add_fulltext_content(self, revision_id, content_obj):
2868
self._fulltext_contents[revision_id] = content_obj
2869
fulltext = content_obj.text_lines()
2870
self._fulltexts[revision_id] = fulltext
2871
# XXX: It would probably be good to check the sha1digest here
2874
def _check_parents(self, child, nodes_to_annotate):
2875
"""Check if all parents have been processed.
2877
:param child: A tuple of (rev_id, parents, raw_content)
2878
:param nodes_to_annotate: If child is ready, add it to
2879
nodes_to_annotate, otherwise put it back in self._pending_children
2881
for parent_id in child[1]:
2882
if parent_id not in self._annotated_lines:
2883
# This parent is present, but another parent is missing
2884
self._pending_children.setdefault(parent_id,
2888
# This one is ready to be processed
2889
nodes_to_annotate.append(child)
2891
def _add_annotation(self, revision_id, fulltext, parent_ids,
2892
left_matching_blocks=None):
2893
"""Add an annotation entry.
2895
All parents should already have been annotated.
2896
:return: A list of children that now have their parents satisfied.
2898
a = self._annotated_lines
2899
annotated_parent_lines = [a[p] for p in parent_ids]
2900
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2901
fulltext, revision_id, left_matching_blocks,
2902
heads_provider=self._get_heads_provider()))
2903
self._annotated_lines[revision_id] = annotated_lines
2904
# Now that we've added this one, see if there are any pending
2905
# deltas to be done, certainly this parent is finished
2906
nodes_to_annotate = []
2907
for child in self._pending_children.pop(revision_id, []):
2908
self._check_parents(child, nodes_to_annotate)
2909
return nodes_to_annotate
2911
def _get_build_graph(self, revision_id):
2912
"""Get the graphs for building texts and annotations.
2914
The data you need for creating a full text may be different than the
2915
data you need to annotate that text. (At a minimum, you need both
2916
parents to create an annotation, but only need 1 parent to generate the
2919
:return: A list of (revision_id, index_memo) records, suitable for
2920
passing to read_records_iter to start reading in the raw data from
2923
if revision_id in self._annotated_lines:
2926
pending = set([revision_id])
2929
# get all pending nodes
2930
this_iteration = pending
2931
build_details = self._knit._index.get_build_details(this_iteration)
2932
self._all_build_details.update(build_details)
2933
# new_nodes = self._knit._index._get_entries(this_iteration)
2935
for rev_id, details in build_details.iteritems():
2936
(index_memo, compression_parent, parents,
2937
record_details) = details
2938
self._revision_id_graph[rev_id] = parents
2939
records.append((rev_id, index_memo))
2940
pending.update(p for p in parents
2941
if p not in self._all_build_details)
2943
missing_versions = this_iteration.difference(build_details.keys())
2944
for missing_version in missing_versions:
2945
# add a key, no parents
2946
self._revision_id_graph[missing_versions] = ()
2947
pending.discard(missing_version) # don't look for it
2948
# Generally we will want to read the records in reverse order, because
2949
# we find the parent nodes after the children
2953
def _annotate_records(self, records):
2954
"""Build the annotations for the listed records."""
2955
# We iterate in the order read, rather than a strict order requested
2956
# However, process what we can, and put off to the side things that still
2957
# need parents, cleaning them up when those parents are processed.
2958
for (rev_id, record,
2959
digest) in self._knit._data.read_records_iter(records):
2960
if rev_id in self._annotated_lines:
2962
parent_ids = self._revision_id_graph[rev_id]
2963
details = self._all_build_details[rev_id]
2964
(index_memo, compression_parent, parents,
2965
record_details) = details
2966
nodes_to_annotate = []
2967
# TODO: Remove the punning between compression parents, and
2968
# parent_ids, we should be able to do this without assuming
2970
if len(parent_ids) == 0:
2971
# There are no parents for this node, so just add it
2972
# TODO: This probably needs to be decoupled
2973
assert compression_parent is None
2974
fulltext_content, delta = self._knit.factory.parse_record(
2975
rev_id, record, record_details, None)
2976
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
2977
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
2978
parent_ids, left_matching_blocks=None))
2980
child = (rev_id, parent_ids, record)
2981
# Check if all the parents are present
2982
self._check_parents(child, nodes_to_annotate)
2983
while nodes_to_annotate:
2984
# Should we use a queue here instead of a stack?
2985
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
2986
(index_memo, compression_parent, parents,
2987
record_details) = self._all_build_details[rev_id]
2988
if compression_parent is not None:
2989
parent_fulltext_content = self._fulltext_contents[compression_parent]
2990
fulltext_content, delta = self._knit.factory.parse_record(
2991
rev_id, record, record_details, parent_fulltext_content,
2992
copy_base_content=True)
2993
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
2994
parent_fulltext = self._fulltexts[parent_ids[0]]
2995
blocks = KnitContent.get_line_delta_blocks(delta,
2996
parent_fulltext, fulltext)
2998
fulltext_content = self._knit.factory.parse_fulltext(
3000
fulltext = self._add_fulltext_content(rev_id,
3003
nodes_to_annotate.extend(
3004
self._add_annotation(rev_id, fulltext, parent_ids,
3005
left_matching_blocks=blocks))
3007
def _get_heads_provider(self):
3008
"""Create a heads provider for resolving ancestry issues."""
3009
if self._heads_provider is not None:
3010
return self._heads_provider
3011
parent_provider = _mod_graph.DictParentsProvider(
3012
self._revision_id_graph)
3013
graph_obj = _mod_graph.Graph(parent_provider)
3014
head_cache = _mod_graph.HeadsCache(graph_obj)
3015
self._heads_provider = head_cache
3018
def get_annotated_lines(self, revision_id):
3019
"""Return the annotated fulltext at the given revision.
3021
:param revision_id: The revision id for this file
3023
records = self._get_build_graph(revision_id)
3024
self._annotate_records(records)
3025
return self._annotated_lines[revision_id]
3029
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3031
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
