1
# Copyright (C) 2005, 2006, 2007 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""Knit versionedfile implementation.
19
A knit is a versioned file implementation that supports efficient append only
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
end-marker; simply "end VERSION"
28
delta can be line or full contents.a
29
... the 8's there are the index number of the annotation.
30
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
8 e.set('executable', 'yes')
36
8 if elt.get('executable') == 'yes':
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
43
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
44
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
45
09:33 < lifeless> right
46
09:33 < jrydberg> lifeless: the position and size is the range in the data file
49
so the index sequence is the dictionary compressed sequence number used
50
in the deltas to provide line annotation
55
# 10:16 < lifeless> make partial index writes safe
56
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
57
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
59
# move sha1 out of the content so that join is faster at verifying parents
60
# record content length ?
64
from cStringIO import StringIO
65
from itertools import izip, chain
70
from zlib import Z_DEFAULT_COMPRESSION
73
from bzrlib.lazy_import import lazy_import
74
lazy_import(globals(), """
94
from bzrlib.errors import (
102
RevisionAlreadyPresent,
104
from bzrlib.graph import Graph
105
from bzrlib.osutils import (
111
from bzrlib.symbol_versioning import (
112
DEPRECATED_PARAMETER,
117
from bzrlib.tsort import topo_sort
118
from bzrlib.tuned_gzip import GzipFile, bytes_to_gzip
120
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
124
# TODO: Split out code specific to this format into an associated object.
126
# TODO: Can we put in some kind of value to check that the index and data
127
# files belong together?
129
# TODO: accommodate binaries, perhaps by storing a byte count
131
# TODO: function to check whole file
133
# TODO: atomically append data, then measure backwards from the cursor
134
# position after writing to work out where it was located. we may need to
135
# bypass python file buffering.
137
DATA_SUFFIX = '.knit'
138
INDEX_SUFFIX = '.kndx'
141
class KnitContent(object):
142
"""Content of a knit version to which deltas can be applied."""
145
self._should_strip_eol = False
148
"""Return a list of (origin, text) tuples."""
149
return list(self.annotate_iter())
151
def apply_delta(self, delta, new_version_id):
152
"""Apply delta to this object to become new_version_id."""
153
raise NotImplementedError(self.apply_delta)
155
def cleanup_eol(self, copy_on_mutate=True):
156
if self._should_strip_eol:
158
self._lines = self._lines[:]
159
self.strip_last_line_newline()
161
def line_delta_iter(self, new_lines):
162
"""Generate line-based delta from this content to new_lines."""
163
new_texts = new_lines.text()
164
old_texts = self.text()
165
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
166
for tag, i1, i2, j1, j2 in s.get_opcodes():
169
# ofrom, oto, length, data
170
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
172
def line_delta(self, new_lines):
173
return list(self.line_delta_iter(new_lines))
176
def get_line_delta_blocks(knit_delta, source, target):
177
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
178
target_len = len(target)
181
for s_begin, s_end, t_len, new_text in knit_delta:
182
true_n = s_begin - s_pos
185
# knit deltas do not provide reliable info about whether the
186
# last line of a file matches, due to eol handling.
187
if source[s_pos + n -1] != target[t_pos + n -1]:
190
yield s_pos, t_pos, n
191
t_pos += t_len + true_n
193
n = target_len - t_pos
195
if source[s_pos + n -1] != target[t_pos + n -1]:
198
yield s_pos, t_pos, n
199
yield s_pos + (target_len - t_pos), target_len, 0
202
class AnnotatedKnitContent(KnitContent):
203
"""Annotated content."""
205
def __init__(self, lines):
206
KnitContent.__init__(self)
209
def annotate_iter(self):
210
"""Yield tuples of (origin, text) for each content line."""
211
return iter(self._lines)
213
def apply_delta(self, delta, new_version_id):
214
"""Apply delta to this object to become new_version_id."""
217
for start, end, count, delta_lines in delta:
218
lines[offset+start:offset+end] = delta_lines
219
offset = offset + (start - end) + count
221
def strip_last_line_newline(self):
222
line = self._lines[-1][1].rstrip('\n')
223
self._lines[-1] = (self._lines[-1][0], line)
224
self._should_strip_eol = False
228
lines = [text for origin, text in self._lines]
229
except ValueError, e:
230
# most commonly (only?) caused by the internal form of the knit
231
# missing annotation information because of a bug - see thread
233
raise KnitCorrupt(self,
234
"line in annotated knit missing annotation information: %s"
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
282
if self._should_strip_eol:
284
lines[-1] = lines[-1].rstrip('\n')
288
class _KnitFactory(object):
289
"""Base class for common Factory functions."""
291
def parse_record(self, version_id, record, record_details,
292
base_content, copy_base_content=True):
293
"""Parse a record into a full content object.
295
:param version_id: The official version id for this content
296
:param record: The data returned by read_records_iter()
297
:param record_details: Details about the record returned by
299
:param base_content: If get_build_details returns a compression_parent,
300
you must return a base_content here, else use None
301
:param copy_base_content: When building from the base_content, decide
302
you can either copy it and return a new object, or modify it in
304
:return: (content, delta) A Content object and possibly a line-delta,
307
method, noeol = record_details
308
if method == 'line-delta':
309
assert base_content is not None
310
if copy_base_content:
311
content = base_content.copy()
313
content = base_content
314
delta = self.parse_line_delta(record, version_id)
315
content.apply_delta(delta, version_id)
317
content = self.parse_fulltext(record, version_id)
319
content._should_strip_eol = noeol
320
return (content, delta)
323
class KnitAnnotateFactory(_KnitFactory):
324
"""Factory for creating annotated Content objects."""
328
def make(self, lines, version_id):
329
num_lines = len(lines)
330
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
332
def parse_fulltext(self, content, version_id):
333
"""Convert fulltext to internal representation
335
fulltext content is of the format
336
revid(utf8) plaintext\n
337
internal representation is of the format:
340
# TODO: jam 20070209 The tests expect this to be returned as tuples,
341
# but the code itself doesn't really depend on that.
342
# Figure out a way to not require the overhead of turning the
343
# list back into tuples.
344
lines = [tuple(line.split(' ', 1)) for line in content]
345
return AnnotatedKnitContent(lines)
347
def parse_line_delta_iter(self, lines):
348
return iter(self.parse_line_delta(lines))
350
def parse_line_delta(self, lines, version_id, plain=False):
351
"""Convert a line based delta into internal representation.
353
line delta is in the form of:
354
intstart intend intcount
356
revid(utf8) newline\n
357
internal representation is
358
(start, end, count, [1..count tuples (revid, newline)])
360
:param plain: If True, the lines are returned as a plain
361
list without annotations, not as a list of (origin, content) tuples, i.e.
362
(start, end, count, [1..count newline])
369
def cache_and_return(line):
370
origin, text = line.split(' ', 1)
371
return cache.setdefault(origin, origin), text
373
# walk through the lines parsing.
374
# Note that the plain test is explicitly pulled out of the
375
# loop to minimise any performance impact
378
start, end, count = [int(n) for n in header.split(',')]
379
contents = [next().split(' ', 1)[1] for i in xrange(count)]
380
result.append((start, end, count, contents))
383
start, end, count = [int(n) for n in header.split(',')]
384
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
385
result.append((start, end, count, contents))
388
def get_fulltext_content(self, lines):
389
"""Extract just the content lines from a fulltext."""
390
return (line.split(' ', 1)[1] for line in lines)
392
def get_linedelta_content(self, lines):
393
"""Extract just the content from a line delta.
395
This doesn't return all of the extra information stored in a delta.
396
Only the actual content lines.
401
header = header.split(',')
402
count = int(header[2])
403
for i in xrange(count):
404
origin, text = next().split(' ', 1)
407
def lower_fulltext(self, content):
408
"""convert a fulltext content record into a serializable form.
410
see parse_fulltext which this inverts.
412
# TODO: jam 20070209 We only do the caching thing to make sure that
413
# the origin is a valid utf-8 line, eventually we could remove it
414
return ['%s %s' % (o, t) for o, t in content._lines]
416
def lower_line_delta(self, delta):
417
"""convert a delta into a serializable form.
419
See parse_line_delta which this inverts.
421
# TODO: jam 20070209 We only do the caching thing to make sure that
422
# the origin is a valid utf-8 line, eventually we could remove it
424
for start, end, c, lines in delta:
425
out.append('%d,%d,%d\n' % (start, end, c))
426
out.extend(origin + ' ' + text
427
for origin, text in lines)
430
def annotate_iter(self, knit, version_id):
431
content = knit._get_content(version_id)
432
return content.annotate_iter()
435
class KnitPlainFactory(_KnitFactory):
436
"""Factory for creating plain Content objects."""
440
def make(self, lines, version_id):
441
return PlainKnitContent(lines, version_id)
443
def parse_fulltext(self, content, version_id):
444
"""This parses an unannotated fulltext.
446
Note that this is not a noop - the internal representation
447
has (versionid, line) - its just a constant versionid.
449
return self.make(content, version_id)
451
def parse_line_delta_iter(self, lines, version_id):
453
num_lines = len(lines)
454
while cur < num_lines:
457
start, end, c = [int(n) for n in header.split(',')]
458
yield start, end, c, lines[cur:cur+c]
461
def parse_line_delta(self, lines, version_id):
462
return list(self.parse_line_delta_iter(lines, version_id))
464
def get_fulltext_content(self, lines):
465
"""Extract just the content lines from a fulltext."""
468
def get_linedelta_content(self, lines):
469
"""Extract just the content from a line delta.
471
This doesn't return all of the extra information stored in a delta.
472
Only the actual content lines.
477
header = header.split(',')
478
count = int(header[2])
479
for i in xrange(count):
482
def lower_fulltext(self, content):
483
return content.text()
485
def lower_line_delta(self, delta):
487
for start, end, c, lines in delta:
488
out.append('%d,%d,%d\n' % (start, end, c))
492
def annotate_iter(self, knit, version_id):
493
annotator = _KnitAnnotator(knit)
494
return iter(annotator.annotate(version_id))
497
def make_empty_knit(transport, relpath):
498
"""Construct a empty knit at the specified location."""
499
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
502
class KnitVersionedFile(VersionedFile):
503
"""Weave-like structure with faster random access.
505
A knit stores a number of texts and a summary of the relationships
506
between them. Texts are identified by a string version-id. Texts
507
are normally stored and retrieved as a series of lines, but can
508
also be passed as single strings.
510
Lines are stored with the trailing newline (if any) included, to
511
avoid special cases for files with no final newline. Lines are
512
composed of 8-bit characters, not unicode. The combination of
513
these approaches should mean any 'binary' file can be safely
514
stored and retrieved.
517
def __init__(self, relpath, transport, file_mode=None, access_mode=None,
518
factory=None, delta=True, create=False, create_parent_dir=False,
519
delay_create=False, dir_mode=None, index=None, access_method=None):
520
"""Construct a knit at location specified by relpath.
522
:param create: If not True, only open an existing knit.
523
:param create_parent_dir: If True, create the parent directory if
524
creating the file fails. (This is used for stores with
525
hash-prefixes that may not exist yet)
526
:param delay_create: The calling code is aware that the knit won't
527
actually be created until the first data is stored.
528
:param index: An index to use for the knit.
530
if access_mode is None:
532
super(KnitVersionedFile, self).__init__(access_mode)
533
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
534
self.transport = transport
535
self.filename = relpath
536
self.factory = factory or KnitAnnotateFactory()
537
self.writable = (access_mode == 'w')
540
self._max_delta_chain = 200
543
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
544
access_mode, create=create, file_mode=file_mode,
545
create_parent_dir=create_parent_dir, delay_create=delay_create,
549
if access_method is None:
550
_access = _KnitAccess(transport, relpath + DATA_SUFFIX, file_mode, dir_mode,
551
((create and not len(self)) and delay_create), create_parent_dir)
553
_access = access_method
554
if create and not len(self) and not delay_create:
556
self._data = _KnitData(_access)
559
return '%s(%s)' % (self.__class__.__name__,
560
self.transport.abspath(self.filename))
562
def _check_should_delta(self, first_parents):
563
"""Iterate back through the parent listing, looking for a fulltext.
565
This is used when we want to decide whether to add a delta or a new
566
fulltext. It searches for _max_delta_chain parents. When it finds a
567
fulltext parent, it sees if the total size of the deltas leading up to
568
it is large enough to indicate that we want a new full text anyway.
570
Return True if we should create a new delta, False if we should use a
575
delta_parents = first_parents
576
for count in xrange(self._max_delta_chain):
577
parent = delta_parents[0]
578
method = self._index.get_method(parent)
579
index, pos, size = self._index.get_position(parent)
580
if method == 'fulltext':
584
delta_parents = self._index.get_parent_map([parent])[parent]
586
# We couldn't find a fulltext, so we must create a new one
589
return fulltext_size > delta_size
591
def _add_raw_records(self, records, data):
592
"""Add all the records 'records' with data pre-joined in 'data'.
594
:param records: A list of tuples(version_id, options, parents, size).
595
:param data: The data for the records. When it is written, the records
596
are adjusted to have pos pointing into data by the sum of
597
the preceding records sizes.
600
raw_record_sizes = [record[3] for record in records]
601
positions = self._data.add_raw_records(raw_record_sizes, data)
604
for (version_id, options, parents, size), access_memo in zip(
606
index_entries.append((version_id, options, access_memo, parents))
607
if self._data._do_cache:
608
self._data._cache[version_id] = data[offset:offset+size]
610
self._index.add_versions(index_entries)
612
def enable_cache(self):
613
"""Start caching data for this knit"""
614
self._data.enable_cache()
616
def clear_cache(self):
617
"""Clear the data cache only."""
618
self._data.clear_cache()
620
def copy_to(self, name, transport):
621
"""See VersionedFile.copy_to()."""
622
# copy the current index to a temp index to avoid racing with local
624
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
625
self.transport.get(self._index._filename))
627
f = self._data._open_file()
629
transport.put_file(name + DATA_SUFFIX, f)
632
# move the copied index into place
633
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
635
def get_data_stream(self, required_versions):
636
"""Get a data stream for the specified versions.
638
Versions may be returned in any order, not necessarily the order
639
specified. They are returned in a partial order by compression
640
parent, so that the deltas can be applied as the data stream is
641
inserted; however note that compression parents will not be sent
642
unless they were specifically requested, as the client may already
645
:param required_versions: The exact set of versions to be extracted.
646
Unlike some other knit methods, this is not used to generate a
647
transitive closure, rather it is used precisely as given.
649
:returns: format_signature, list of (version, options, length, parents),
652
required_version_set = frozenset(required_versions)
654
# list of revisions that can just be sent without waiting for their
657
# map from revision to the children based on it
659
# first, read all relevant index data, enough to sort into the right
661
for version_id in required_versions:
662
options = self._index.get_options(version_id)
663
parents = self._index.get_parents_with_ghosts(version_id)
664
index_memo = self._index.get_position(version_id)
665
version_index[version_id] = (index_memo, options, parents)
666
if ('line-delta' in options
667
and parents[0] in required_version_set):
668
# must wait until the parent has been sent
669
deferred.setdefault(parents[0], []). \
672
# either a fulltext, or a delta whose parent the client did
673
# not ask for and presumably already has
674
ready_to_send.append(version_id)
675
# build a list of results to return, plus instructions for data to
677
copy_queue_records = []
678
temp_version_list = []
680
# XXX: pushing and popping lists may be a bit inefficient
681
version_id = ready_to_send.pop(0)
682
(index_memo, options, parents) = version_index[version_id]
683
copy_queue_records.append((version_id, index_memo))
684
none, data_pos, data_size = index_memo
685
temp_version_list.append((version_id, options, data_size,
687
if version_id in deferred:
688
# now we can send all the children of this revision - we could
689
# put them in anywhere, but we hope that sending them soon
690
# after the fulltext will give good locality in the receiver
691
ready_to_send[:0] = deferred.pop(version_id)
692
assert len(deferred) == 0, \
693
"Still have compressed child versions waiting to be sent"
694
# XXX: The stream format is such that we cannot stream it - we have to
695
# know the length of all the data a-priori.
697
result_version_list = []
698
for (version_id, raw_data), \
699
(version_id2, options, _, parents) in \
700
izip(self._data.read_records_iter_raw(copy_queue_records),
702
assert version_id == version_id2, \
703
'logic error, inconsistent results'
704
raw_datum.append(raw_data)
705
result_version_list.append(
706
(version_id, options, len(raw_data), parents))
707
# provide a callback to get data incrementally.
708
pseudo_file = StringIO(''.join(raw_datum))
711
return pseudo_file.read()
713
return pseudo_file.read(length)
714
return (self.get_format_signature(), result_version_list, read)
716
def _extract_blocks(self, version_id, source, target):
717
if self._index.get_method(version_id) != 'line-delta':
719
parent, sha1, noeol, delta = self.get_delta(version_id)
720
return KnitContent.get_line_delta_blocks(delta, source, target)
722
def get_delta(self, version_id):
723
"""Get a delta for constructing version from some other version."""
724
self.check_not_reserved_id(version_id)
725
parents = self.get_parent_map([version_id])[version_id]
730
index_memo = self._index.get_position(version_id)
731
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
732
noeol = 'no-eol' in self._index.get_options(version_id)
733
if 'fulltext' == self._index.get_method(version_id):
734
new_content = self.factory.parse_fulltext(data, version_id)
735
if parent is not None:
736
reference_content = self._get_content(parent)
737
old_texts = reference_content.text()
740
new_texts = new_content.text()
741
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
743
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
745
delta = self.factory.parse_line_delta(data, version_id)
746
return parent, sha1, noeol, delta
748
def get_format_signature(self):
749
"""See VersionedFile.get_format_signature()."""
750
if self.factory.annotated:
751
annotated_part = "annotated"
753
annotated_part = "plain"
754
return "knit-%s" % (annotated_part,)
756
@deprecated_method(one_four)
757
def get_graph_with_ghosts(self):
758
"""See VersionedFile.get_graph_with_ghosts()."""
759
return self.get_parent_map(self.versions())
761
def get_sha1(self, version_id):
762
return self.get_sha1s([version_id])[0]
764
def get_sha1s(self, version_ids):
765
"""See VersionedFile.get_sha1()."""
766
record_map = self._get_record_map(version_ids)
767
# record entry 2 is the 'digest'.
768
return [record_map[v][2] for v in version_ids]
772
"""See VersionedFile.get_suffixes()."""
773
return [DATA_SUFFIX, INDEX_SUFFIX]
775
@deprecated_method(one_four)
776
def has_ghost(self, version_id):
777
"""True if there is a ghost reference in the file to version_id."""
779
if self.has_version(version_id):
781
# optimisable if needed by memoising the _ghosts set.
782
items = self.get_parent_map(self.versions())
783
for parents in items.itervalues():
784
for parent in parents:
785
if parent == version_id and parent not in items:
789
def insert_data_stream(self, (format, data_list, reader_callable)):
790
"""Insert knit records from a data stream into this knit.
792
If a version in the stream is already present in this knit, it will not
793
be inserted a second time. It will be checked for consistency with the
794
stored version however, and may cause a KnitCorrupt error to be raised
795
if the data in the stream disagrees with the already stored data.
797
:seealso: get_data_stream
799
if format != self.get_format_signature():
800
if 'knit' in debug.debug_flags:
802
'incompatible format signature inserting to %r', self)
803
source = self._knit_from_datastream(
804
(format, data_list, reader_callable))
808
for version_id, options, length, parents in data_list:
809
if self.has_version(version_id):
810
# First check: the list of parents.
811
my_parents = self.get_parents_with_ghosts(version_id)
812
if tuple(my_parents) != tuple(parents):
813
# XXX: KnitCorrupt is not quite the right exception here.
816
'parents list %r from data stream does not match '
817
'already recorded parents %r for %s'
818
% (parents, my_parents, version_id))
820
# Also check the SHA-1 of the fulltext this content will
822
raw_data = reader_callable(length)
823
my_fulltext_sha1 = self.get_sha1(version_id)
824
df, rec = self._data._parse_record_header(version_id, raw_data)
825
stream_fulltext_sha1 = rec[3]
826
if my_fulltext_sha1 != stream_fulltext_sha1:
827
# Actually, we don't know if it's this knit that's corrupt,
828
# or the data stream we're trying to insert.
830
self.filename, 'sha-1 does not match %s' % version_id)
832
if 'line-delta' in options:
833
# Make sure that this knit record is actually useful: a
834
# line-delta is no use unless we have its parent.
835
# Fetching from a broken repository with this problem
836
# shouldn't break the target repository.
838
# See https://bugs.launchpad.net/bzr/+bug/164443
839
if not self._index.has_version(parents[0]):
842
'line-delta from stream '
845
'missing parent %s\n'
846
'Try running "bzr check" '
847
'on the source repository, and "bzr reconcile" '
849
(version_id, parents[0]))
850
self._add_raw_records(
851
[(version_id, options, parents, length)],
852
reader_callable(length))
854
def _knit_from_datastream(self, (format, data_list, reader_callable)):
855
"""Create a knit object from a data stream.
857
This method exists to allow conversion of data streams that do not
858
match the signature of this knit. Generally it will be slower and use
859
more memory to use this method to insert data, but it will work.
861
:seealso: get_data_stream for details on datastreams.
862
:return: A knit versioned file which can be used to join the datastream
865
if format == "knit-plain":
866
factory = KnitPlainFactory()
867
elif format == "knit-annotated":
868
factory = KnitAnnotateFactory()
870
raise errors.KnitDataStreamUnknown(format)
871
index = _StreamIndex(data_list, self._index)
872
access = _StreamAccess(reader_callable, index, self, factory)
873
return KnitVersionedFile(self.filename, self.transport,
874
factory=factory, index=index, access_method=access)
877
"""See VersionedFile.versions."""
878
if 'evil' in debug.debug_flags:
879
trace.mutter_callsite(2, "versions scales with size of history")
880
return self._index.get_versions()
882
def has_version(self, version_id):
883
"""See VersionedFile.has_version."""
884
if 'evil' in debug.debug_flags:
885
trace.mutter_callsite(2, "has_version is a LBYL scenario")
886
return self._index.has_version(version_id)
888
__contains__ = has_version
890
def _merge_annotations(self, content, parents, parent_texts={},
891
delta=None, annotated=None,
892
left_matching_blocks=None):
893
"""Merge annotations for content. This is done by comparing
894
the annotations based on changed to the text.
896
if left_matching_blocks is not None:
897
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
901
for parent_id in parents:
902
merge_content = self._get_content(parent_id, parent_texts)
903
if (parent_id == parents[0] and delta_seq is not None):
906
seq = patiencediff.PatienceSequenceMatcher(
907
None, merge_content.text(), content.text())
908
for i, j, n in seq.get_matching_blocks():
911
# this appears to copy (origin, text) pairs across to the
912
# new content for any line that matches the last-checked
914
content._lines[j:j+n] = merge_content._lines[i:i+n]
916
if delta_seq is None:
917
reference_content = self._get_content(parents[0], parent_texts)
918
new_texts = content.text()
919
old_texts = reference_content.text()
920
delta_seq = patiencediff.PatienceSequenceMatcher(
921
None, old_texts, new_texts)
922
return self._make_line_delta(delta_seq, content)
924
def _make_line_delta(self, delta_seq, new_content):
925
"""Generate a line delta from delta_seq and new_content."""
927
for op in delta_seq.get_opcodes():
930
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
933
def _get_components_positions(self, version_ids):
934
"""Produce a map of position data for the components of versions.
936
This data is intended to be used for retrieving the knit records.
938
A dict of version_id to (record_details, index_memo, next, parents) is
940
method is the way referenced data should be applied.
941
index_memo is the handle to pass to the data access to actually get the
943
next is the build-parent of the version, or None for fulltexts.
944
parents is the version_ids of the parents of this version
947
pending_components = version_ids
948
while pending_components:
949
build_details = self._index.get_build_details(pending_components)
950
current_components = set(pending_components)
951
pending_components = set()
952
for version_id, details in build_details.iteritems():
953
(index_memo, compression_parent, parents,
954
record_details) = details
955
method = record_details[0]
956
if compression_parent is not None:
957
pending_components.add(compression_parent)
958
component_data[version_id] = (record_details, index_memo,
960
missing = current_components.difference(build_details)
962
raise errors.RevisionNotPresent(missing.pop(), self.filename)
963
return component_data
965
def _get_content(self, version_id, parent_texts={}):
966
"""Returns a content object that makes up the specified
968
cached_version = parent_texts.get(version_id, None)
969
if cached_version is not None:
970
if not self.has_version(version_id):
971
raise RevisionNotPresent(version_id, self.filename)
972
return cached_version
974
text_map, contents_map = self._get_content_maps([version_id])
975
return contents_map[version_id]
977
def _check_versions_present(self, version_ids):
978
"""Check that all specified versions are present."""
979
self._index.check_versions_present(version_ids)
981
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
982
nostore_sha, random_id, check_content, left_matching_blocks):
983
"""See VersionedFile.add_lines_with_ghosts()."""
984
self._check_add(version_id, lines, random_id, check_content)
985
return self._add(version_id, lines, parents, self.delta,
986
parent_texts, left_matching_blocks, nostore_sha, random_id)
988
def _add_lines(self, version_id, parents, lines, parent_texts,
989
left_matching_blocks, nostore_sha, random_id, check_content):
990
"""See VersionedFile.add_lines."""
991
self._check_add(version_id, lines, random_id, check_content)
992
self._check_versions_present(parents)
993
return self._add(version_id, lines[:], parents, self.delta,
994
parent_texts, left_matching_blocks, nostore_sha, random_id)
996
def _check_add(self, version_id, lines, random_id, check_content):
997
"""check that version_id and lines are safe to add."""
998
if contains_whitespace(version_id):
999
raise InvalidRevisionId(version_id, self.filename)
1000
self.check_not_reserved_id(version_id)
1001
# Technically this could be avoided if we are happy to allow duplicate
1002
# id insertion when other things than bzr core insert texts, but it
1003
# seems useful for folk using the knit api directly to have some safety
1004
# blanket that we can disable.
1005
if not random_id and self.has_version(version_id):
1006
raise RevisionAlreadyPresent(version_id, self.filename)
1008
self._check_lines_not_unicode(lines)
1009
self._check_lines_are_lines(lines)
1011
def _add(self, version_id, lines, parents, delta, parent_texts,
1012
left_matching_blocks, nostore_sha, random_id):
1013
"""Add a set of lines on top of version specified by parents.
1015
If delta is true, compress the text as a line-delta against
1018
Any versions not present will be converted into ghosts.
1020
# first thing, if the content is something we don't need to store, find
1022
line_bytes = ''.join(lines)
1023
digest = sha_string(line_bytes)
1024
if nostore_sha == digest:
1025
raise errors.ExistingContent
1027
present_parents = []
1028
if parent_texts is None:
1030
for parent in parents:
1031
if self.has_version(parent):
1032
present_parents.append(parent)
1034
# can only compress against the left most present parent.
1036
(len(present_parents) == 0 or
1037
present_parents[0] != parents[0])):
1040
text_length = len(line_bytes)
1043
if lines[-1][-1] != '\n':
1044
# copy the contents of lines.
1046
options.append('no-eol')
1047
lines[-1] = lines[-1] + '\n'
1051
# To speed the extract of texts the delta chain is limited
1052
# to a fixed number of deltas. This should minimize both
1053
# I/O and the time spend applying deltas.
1054
delta = self._check_should_delta(present_parents)
1056
assert isinstance(version_id, str)
1057
content = self.factory.make(lines, version_id)
1058
if delta or (self.factory.annotated and len(present_parents) > 0):
1059
# Merge annotations from parent texts if needed.
1060
delta_hunks = self._merge_annotations(content, present_parents,
1061
parent_texts, delta, self.factory.annotated,
1062
left_matching_blocks)
1065
options.append('line-delta')
1066
store_lines = self.factory.lower_line_delta(delta_hunks)
1067
size, bytes = self._data._record_to_data(version_id, digest,
1070
options.append('fulltext')
1071
# isinstance is slower and we have no hierarchy.
1072
if self.factory.__class__ == KnitPlainFactory:
1073
# Use the already joined bytes saving iteration time in
1075
size, bytes = self._data._record_to_data(version_id, digest,
1076
lines, [line_bytes])
1078
# get mixed annotation + content and feed it into the
1080
store_lines = self.factory.lower_fulltext(content)
1081
size, bytes = self._data._record_to_data(version_id, digest,
1084
access_memo = self._data.add_raw_records([size], bytes)[0]
1085
self._index.add_versions(
1086
((version_id, options, access_memo, parents),),
1087
random_id=random_id)
1088
return digest, text_length, content
1090
def check(self, progress_bar=None):
1091
"""See VersionedFile.check()."""
1093
def _clone_text(self, new_version_id, old_version_id, parents):
1094
"""See VersionedFile.clone_text()."""
1095
# FIXME RBC 20060228 make fast by only inserting an index with null
1097
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
1099
def get_lines(self, version_id):
1100
"""See VersionedFile.get_lines()."""
1101
return self.get_line_list([version_id])[0]
1103
def _get_record_map(self, version_ids):
1104
"""Produce a dictionary of knit records.
1106
:return: {version_id:(record, record_details, digest, next)}
1108
data returned from read_records
1110
opaque information to pass to parse_record
1112
SHA1 digest of the full text after all steps are done
1114
build-parent of the version, i.e. the leftmost ancestor.
1115
Will be None if the record is not a delta.
1117
position_map = self._get_components_positions(version_ids)
1118
# c = component_id, r = record_details, i_m = index_memo, n = next
1119
records = [(c, i_m) for c, (r, i_m, n)
1120
in position_map.iteritems()]
1122
for component_id, record, digest in \
1123
self._data.read_records_iter(records):
1124
(record_details, index_memo, next) = position_map[component_id]
1125
record_map[component_id] = record, record_details, digest, next
1129
def get_text(self, version_id):
1130
"""See VersionedFile.get_text"""
1131
return self.get_texts([version_id])[0]
1133
def get_texts(self, version_ids):
1134
return [''.join(l) for l in self.get_line_list(version_ids)]
1136
def get_line_list(self, version_ids):
1137
"""Return the texts of listed versions as a list of strings."""
1138
for version_id in version_ids:
1139
self.check_not_reserved_id(version_id)
1140
text_map, content_map = self._get_content_maps(version_ids)
1141
return [text_map[v] for v in version_ids]
1143
_get_lf_split_line_list = get_line_list
1145
def _get_content_maps(self, version_ids):
1146
"""Produce maps of text and KnitContents
1148
:return: (text_map, content_map) where text_map contains the texts for
1149
the requested versions and content_map contains the KnitContents.
1150
Both dicts take version_ids as their keys.
1152
# FUTURE: This function could be improved for the 'extract many' case
1153
# by tracking each component and only doing the copy when the number of
1154
# children than need to apply delta's to it is > 1 or it is part of the
1156
version_ids = list(version_ids)
1157
multiple_versions = len(version_ids) != 1
1158
record_map = self._get_record_map(version_ids)
1163
for version_id in version_ids:
1166
while cursor is not None:
1167
record, record_details, digest, next = record_map[cursor]
1168
components.append((cursor, record, record_details, digest))
1169
if cursor in content_map:
1174
for (component_id, record, record_details,
1175
digest) in reversed(components):
1176
if component_id in content_map:
1177
content = content_map[component_id]
1179
content, delta = self.factory.parse_record(version_id,
1180
record, record_details, content,
1181
copy_base_content=multiple_versions)
1182
if multiple_versions:
1183
content_map[component_id] = content
1185
content.cleanup_eol(copy_on_mutate=multiple_versions)
1186
final_content[version_id] = content
1188
# digest here is the digest from the last applied component.
1189
text = content.text()
1190
actual_sha = sha_strings(text)
1191
if actual_sha != digest:
1192
raise KnitCorrupt(self.filename,
1194
'\n of reconstructed text does not match'
1196
'\n for version %s' %
1197
(actual_sha, digest, version_id))
1198
text_map[version_id] = text
1199
return text_map, final_content
1201
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1203
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1204
if version_ids is None:
1205
version_ids = self.versions()
1207
pb = progress.DummyProgress()
1208
# we don't care about inclusions, the caller cares.
1209
# but we need to setup a list of records to visit.
1210
# we need version_id, position, length
1211
version_id_records = []
1212
requested_versions = set(version_ids)
1213
# filter for available versions
1214
for version_id in requested_versions:
1215
if not self.has_version(version_id):
1216
raise RevisionNotPresent(version_id, self.filename)
1217
# get a in-component-order queue:
1218
for version_id in self.versions():
1219
if version_id in requested_versions:
1220
index_memo = self._index.get_position(version_id)
1221
version_id_records.append((version_id, index_memo))
1223
total = len(version_id_records)
1224
for version_idx, (version_id, data, sha_value) in \
1225
enumerate(self._data.read_records_iter(version_id_records)):
1226
pb.update('Walking content.', version_idx, total)
1227
method = self._index.get_method(version_id)
1229
assert method in ('fulltext', 'line-delta')
1230
if method == 'fulltext':
1231
line_iterator = self.factory.get_fulltext_content(data)
1233
line_iterator = self.factory.get_linedelta_content(data)
1234
# XXX: It might be more efficient to yield (version_id,
1235
# line_iterator) in the future. However for now, this is a simpler
1236
# change to integrate into the rest of the codebase. RBC 20071110
1237
for line in line_iterator:
1238
yield line, version_id
1240
pb.update('Walking content.', total, total)
1242
def iter_parents(self, version_ids):
1243
"""Iterate through the parents for many version ids.
1245
:param version_ids: An iterable yielding version_ids.
1246
:return: An iterator that yields (version_id, parents). Requested
1247
version_ids not present in the versioned file are simply skipped.
1248
The order is undefined, allowing for different optimisations in
1249
the underlying implementation.
1251
return self._index.iter_parents(version_ids)
1253
def num_versions(self):
1254
"""See VersionedFile.num_versions()."""
1255
return self._index.num_versions()
1257
__len__ = num_versions
1259
def annotate_iter(self, version_id):
1260
"""See VersionedFile.annotate_iter."""
1261
return self.factory.annotate_iter(self, version_id)
1263
def get_parent_map(self, version_ids):
1264
"""See VersionedFile.get_parent_map."""
1265
return self._index.get_parent_map(version_ids)
1267
def get_ancestry(self, versions, topo_sorted=True):
1268
"""See VersionedFile.get_ancestry."""
1269
if isinstance(versions, basestring):
1270
versions = [versions]
1273
return self._index.get_ancestry(versions, topo_sorted)
1275
def get_ancestry_with_ghosts(self, versions):
1276
"""See VersionedFile.get_ancestry_with_ghosts."""
1277
if isinstance(versions, basestring):
1278
versions = [versions]
1281
return self._index.get_ancestry_with_ghosts(versions)
1283
def plan_merge(self, ver_a, ver_b):
1284
"""See VersionedFile.plan_merge."""
1285
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1286
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1287
annotated_a = self.annotate(ver_a)
1288
annotated_b = self.annotate(ver_b)
1289
return merge._plan_annotate_merge(annotated_a, annotated_b,
1290
ancestors_a, ancestors_b)
1293
class _KnitComponentFile(object):
1294
"""One of the files used to implement a knit database"""
1296
def __init__(self, transport, filename, mode, file_mode=None,
1297
create_parent_dir=False, dir_mode=None):
1298
self._transport = transport
1299
self._filename = filename
1301
self._file_mode = file_mode
1302
self._dir_mode = dir_mode
1303
self._create_parent_dir = create_parent_dir
1304
self._need_to_create = False
1306
def _full_path(self):
1307
"""Return the full path to this file."""
1308
return self._transport.base + self._filename
1310
def check_header(self, fp):
1311
line = fp.readline()
1313
# An empty file can actually be treated as though the file doesn't
1315
raise errors.NoSuchFile(self._full_path())
1316
if line != self.HEADER:
1317
raise KnitHeaderError(badline=line,
1318
filename=self._transport.abspath(self._filename))
1321
return '%s(%s)' % (self.__class__.__name__, self._filename)
1324
class _KnitIndex(_KnitComponentFile):
1325
"""Manages knit index file.
1327
The index is already kept in memory and read on startup, to enable
1328
fast lookups of revision information. The cursor of the index
1329
file is always pointing to the end, making it easy to append
1332
_cache is a cache for fast mapping from version id to a Index
1335
_history is a cache for fast mapping from indexes to version ids.
1337
The index data format is dictionary compressed when it comes to
1338
parent references; a index entry may only have parents that with a
1339
lover index number. As a result, the index is topological sorted.
1341
Duplicate entries may be written to the index for a single version id
1342
if this is done then the latter one completely replaces the former:
1343
this allows updates to correct version and parent information.
1344
Note that the two entries may share the delta, and that successive
1345
annotations and references MUST point to the first entry.
1347
The index file on disc contains a header, followed by one line per knit
1348
record. The same revision can be present in an index file more than once.
1349
The first occurrence gets assigned a sequence number starting from 0.
1351
The format of a single line is
1352
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1353
REVISION_ID is a utf8-encoded revision id
1354
FLAGS is a comma separated list of flags about the record. Values include
1355
no-eol, line-delta, fulltext.
1356
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1357
that the the compressed data starts at.
1358
LENGTH is the ascii representation of the length of the data file.
1359
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1361
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1362
revision id already in the knit that is a parent of REVISION_ID.
1363
The ' :' marker is the end of record marker.
1366
when a write is interrupted to the index file, it will result in a line
1367
that does not end in ' :'. If the ' :' is not present at the end of a line,
1368
or at the end of the file, then the record that is missing it will be
1369
ignored by the parser.
1371
When writing new records to the index file, the data is preceded by '\n'
1372
to ensure that records always start on new lines even if the last write was
1373
interrupted. As a result its normal for the last line in the index to be
1374
missing a trailing newline. One can be added with no harmful effects.
1377
HEADER = "# bzr knit index 8\n"
1379
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1380
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1382
def _cache_version(self, version_id, options, pos, size, parents):
1383
"""Cache a version record in the history array and index cache.
1385
This is inlined into _load_data for performance. KEEP IN SYNC.
1386
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1389
# only want the _history index to reference the 1st index entry
1391
if version_id not in self._cache:
1392
index = len(self._history)
1393
self._history.append(version_id)
1395
index = self._cache[version_id][5]
1396
self._cache[version_id] = (version_id,
1403
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1404
create_parent_dir=False, delay_create=False, dir_mode=None):
1405
_KnitComponentFile.__init__(self, transport, filename, mode,
1406
file_mode=file_mode,
1407
create_parent_dir=create_parent_dir,
1410
# position in _history is the 'official' index for a revision
1411
# but the values may have come from a newer entry.
1412
# so - wc -l of a knit index is != the number of unique names
1416
fp = self._transport.get(self._filename)
1418
# _load_data may raise NoSuchFile if the target knit is
1420
_load_data(self, fp)
1424
if mode != 'w' or not create:
1427
self._need_to_create = True
1429
self._transport.put_bytes_non_atomic(
1430
self._filename, self.HEADER, mode=self._file_mode)
1432
def get_ancestry(self, versions, topo_sorted=True):
1433
"""See VersionedFile.get_ancestry."""
1434
# get a graph of all the mentioned versions:
1436
pending = set(versions)
1439
version = pending.pop()
1442
parents = [p for p in cache[version][4] if p in cache]
1444
raise RevisionNotPresent(version, self._filename)
1445
# if not completed and not a ghost
1446
pending.update([p for p in parents if p not in graph])
1447
graph[version] = parents
1450
return topo_sort(graph.items())
1452
def get_ancestry_with_ghosts(self, versions):
1453
"""See VersionedFile.get_ancestry_with_ghosts."""
1454
# get a graph of all the mentioned versions:
1455
self.check_versions_present(versions)
1458
pending = set(versions)
1460
version = pending.pop()
1462
parents = cache[version][4]
1468
pending.update([p for p in parents if p not in graph])
1469
graph[version] = parents
1470
return topo_sort(graph.items())
1472
def get_build_details(self, version_ids):
1473
"""Get the method, index_memo and compression parent for version_ids.
1475
Ghosts are omitted from the result.
1477
:param version_ids: An iterable of version_ids.
1478
:return: A dict of version_id:(index_memo, compression_parent,
1479
parents, record_details).
1481
opaque structure to pass to read_records to extract the raw
1484
Content that this record is built upon, may be None
1486
Logical parents of this node
1488
extra information about the content which needs to be passed to
1489
Factory.parse_record
1492
for version_id in version_ids:
1493
if version_id not in self._cache:
1494
# ghosts are omitted
1496
method = self.get_method(version_id)
1497
parents = self.get_parents_with_ghosts(version_id)
1498
if method == 'fulltext':
1499
compression_parent = None
1501
compression_parent = parents[0]
1502
noeol = 'no-eol' in self.get_options(version_id)
1503
index_memo = self.get_position(version_id)
1504
result[version_id] = (index_memo, compression_parent,
1505
parents, (method, noeol))
1508
def iter_parents(self, version_ids):
1509
"""Iterate through the parents for many version ids.
1511
:param version_ids: An iterable yielding version_ids.
1512
:return: An iterator that yields (version_id, parents). Requested
1513
version_ids not present in the versioned file are simply skipped.
1514
The order is undefined, allowing for different optimisations in
1515
the underlying implementation.
1517
parent_map = self.get_parent_map(version_ids)
1518
parent_map_set = set(parent_map)
1519
unknown_existence = set()
1520
for parents in parent_map.itervalues():
1521
unknown_existence.update(parents)
1522
unknown_existence.difference_update(parent_map_set)
1523
present_parents = set(self.get_parent_map(unknown_existence))
1524
present_parents.update(parent_map_set)
1525
for version_id, parents in parent_map.iteritems():
1526
parents = tuple(parent for parent in parents
1527
if parent in present_parents)
1528
yield version_id, parents
1530
def num_versions(self):
1531
return len(self._history)
1533
__len__ = num_versions
1535
def get_versions(self):
1536
"""Get all the versions in the file. not topologically sorted."""
1537
return self._history
1539
def _version_list_to_index(self, versions):
1542
for version in versions:
1543
if version in cache:
1544
# -- inlined lookup() --
1545
result_list.append(str(cache[version][5]))
1546
# -- end lookup () --
1548
result_list.append('.' + version)
1549
return ' '.join(result_list)
1551
def add_version(self, version_id, options, index_memo, parents):
1552
"""Add a version record to the index."""
1553
self.add_versions(((version_id, options, index_memo, parents),))
1555
def add_versions(self, versions, random_id=False):
1556
"""Add multiple versions to the index.
1558
:param versions: a list of tuples:
1559
(version_id, options, pos, size, parents).
1560
:param random_id: If True the ids being added were randomly generated
1561
and no check for existence will be performed.
1564
orig_history = self._history[:]
1565
orig_cache = self._cache.copy()
1568
for version_id, options, (index, pos, size), parents in versions:
1569
line = "\n%s %s %s %s %s :" % (version_id,
1573
self._version_list_to_index(parents))
1574
assert isinstance(line, str), \
1575
'content must be utf-8 encoded: %r' % (line,)
1577
self._cache_version(version_id, options, pos, size, tuple(parents))
1578
if not self._need_to_create:
1579
self._transport.append_bytes(self._filename, ''.join(lines))
1582
sio.write(self.HEADER)
1583
sio.writelines(lines)
1585
self._transport.put_file_non_atomic(self._filename, sio,
1586
create_parent_dir=self._create_parent_dir,
1587
mode=self._file_mode,
1588
dir_mode=self._dir_mode)
1589
self._need_to_create = False
1591
# If any problems happen, restore the original values and re-raise
1592
self._history = orig_history
1593
self._cache = orig_cache
1596
def has_version(self, version_id):
1597
"""True if the version is in the index."""
1598
return version_id in self._cache
1600
def get_position(self, version_id):
1601
"""Return details needed to access the version.
1603
.kndx indices do not support split-out data, so return None for the
1606
:return: a tuple (None, data position, size) to hand to the access
1607
logic to get the record.
1609
entry = self._cache[version_id]
1610
return None, entry[2], entry[3]
1612
def get_method(self, version_id):
1613
"""Return compression method of specified version."""
1615
options = self._cache[version_id][1]
1617
raise RevisionNotPresent(version_id, self._filename)
1618
if 'fulltext' in options:
1621
if 'line-delta' not in options:
1622
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1625
def get_options(self, version_id):
1626
"""Return a list representing options.
1630
return self._cache[version_id][1]
1632
def get_parent_map(self, version_ids):
1633
"""Passed through to by KnitVersionedFile.get_parent_map."""
1635
for version_id in version_ids:
1637
result[version_id] = tuple(self._cache[version_id][4])
1642
def get_parents_with_ghosts(self, version_id):
1643
"""Return parents of specified version with ghosts."""
1645
return self.get_parent_map([version_id])[version_id]
1647
raise RevisionNotPresent(version_id, self)
1649
def check_versions_present(self, version_ids):
1650
"""Check that all specified versions are present."""
1652
for version_id in version_ids:
1653
if version_id not in cache:
1654
raise RevisionNotPresent(version_id, self._filename)
1657
class KnitGraphIndex(object):
1658
"""A knit index that builds on GraphIndex."""
1660
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1661
"""Construct a KnitGraphIndex on a graph_index.
1663
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1664
:param deltas: Allow delta-compressed records.
1665
:param add_callback: If not None, allow additions to the index and call
1666
this callback with a list of added GraphIndex nodes:
1667
[(node, value, node_refs), ...]
1668
:param parents: If True, record knits parents, if not do not record
1671
self._graph_index = graph_index
1672
self._deltas = deltas
1673
self._add_callback = add_callback
1674
self._parents = parents
1675
if deltas and not parents:
1676
raise KnitCorrupt(self, "Cannot do delta compression without "
1679
def _get_entries(self, keys, check_present=False):
1680
"""Get the entries for keys.
1682
:param keys: An iterable of index keys, - 1-tuples.
1687
for node in self._graph_index.iter_entries(keys):
1689
found_keys.add(node[1])
1691
# adapt parentless index to the rest of the code.
1692
for node in self._graph_index.iter_entries(keys):
1693
yield node[0], node[1], node[2], ()
1694
found_keys.add(node[1])
1696
missing_keys = keys.difference(found_keys)
1698
raise RevisionNotPresent(missing_keys.pop(), self)
1700
def _present_keys(self, version_ids):
1702
node[1] for node in self._get_entries(version_ids)])
1704
def _parentless_ancestry(self, versions):
1705
"""Honour the get_ancestry API for parentless knit indices."""
1706
wanted_keys = self._version_ids_to_keys(versions)
1707
present_keys = self._present_keys(wanted_keys)
1708
missing = set(wanted_keys).difference(present_keys)
1710
raise RevisionNotPresent(missing.pop(), self)
1711
return list(self._keys_to_version_ids(present_keys))
1713
def get_ancestry(self, versions, topo_sorted=True):
1714
"""See VersionedFile.get_ancestry."""
1715
if not self._parents:
1716
return self._parentless_ancestry(versions)
1717
# XXX: This will do len(history) index calls - perhaps
1718
# it should be altered to be a index core feature?
1719
# get a graph of all the mentioned versions:
1722
versions = self._version_ids_to_keys(versions)
1723
pending = set(versions)
1725
# get all pending nodes
1726
this_iteration = pending
1727
new_nodes = self._get_entries(this_iteration)
1730
for (index, key, value, node_refs) in new_nodes:
1731
# dont ask for ghosties - otherwise
1732
# we we can end up looping with pending
1733
# being entirely ghosted.
1734
graph[key] = [parent for parent in node_refs[0]
1735
if parent not in ghosts]
1737
for parent in graph[key]:
1738
# dont examine known nodes again
1743
ghosts.update(this_iteration.difference(found))
1744
if versions.difference(graph):
1745
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1747
result_keys = topo_sort(graph.items())
1749
result_keys = graph.iterkeys()
1750
return [key[0] for key in result_keys]
1752
def get_ancestry_with_ghosts(self, versions):
1753
"""See VersionedFile.get_ancestry."""
1754
if not self._parents:
1755
return self._parentless_ancestry(versions)
1756
# XXX: This will do len(history) index calls - perhaps
1757
# it should be altered to be a index core feature?
1758
# get a graph of all the mentioned versions:
1760
versions = self._version_ids_to_keys(versions)
1761
pending = set(versions)
1763
# get all pending nodes
1764
this_iteration = pending
1765
new_nodes = self._get_entries(this_iteration)
1767
for (index, key, value, node_refs) in new_nodes:
1768
graph[key] = node_refs[0]
1770
for parent in graph[key]:
1771
# dont examine known nodes again
1775
missing_versions = this_iteration.difference(graph)
1776
missing_needed = versions.intersection(missing_versions)
1778
raise RevisionNotPresent(missing_needed.pop(), self)
1779
for missing_version in missing_versions:
1780
# add a key, no parents
1781
graph[missing_version] = []
1782
pending.discard(missing_version) # don't look for it
1783
result_keys = topo_sort(graph.items())
1784
return [key[0] for key in result_keys]
1786
def get_build_details(self, version_ids):
1787
"""Get the method, index_memo and compression parent for version_ids.
1789
Ghosts are omitted from the result.
1791
:param version_ids: An iterable of version_ids.
1792
:return: A dict of version_id:(index_memo, compression_parent,
1793
parents, record_details).
1795
opaque structure to pass to read_records to extract the raw
1798
Content that this record is built upon, may be None
1800
Logical parents of this node
1802
extra information about the content which needs to be passed to
1803
Factory.parse_record
1806
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
1807
for entry in entries:
1808
version_id = self._keys_to_version_ids((entry[1],))[0]
1809
if not self._parents:
1812
parents = self._keys_to_version_ids(entry[3][0])
1813
if not self._deltas:
1814
compression_parent = None
1816
compression_parent_key = self._compression_parent(entry)
1817
if compression_parent_key:
1818
compression_parent = self._keys_to_version_ids(
1819
(compression_parent_key,))[0]
1821
compression_parent = None
1822
noeol = (entry[2][0] == 'N')
1823
if compression_parent:
1824
method = 'line-delta'
1827
result[version_id] = (self._node_to_position(entry),
1828
compression_parent, parents,
1832
def _compression_parent(self, an_entry):
1833
# return the key that an_entry is compressed against, or None
1834
# Grab the second parent list (as deltas implies parents currently)
1835
compression_parents = an_entry[3][1]
1836
if not compression_parents:
1838
assert len(compression_parents) == 1
1839
return compression_parents[0]
1841
def _get_method(self, node):
1842
if not self._deltas:
1844
if self._compression_parent(node):
1849
def iter_parents(self, version_ids):
1850
"""Iterate through the parents for many version ids.
1852
:param version_ids: An iterable yielding version_ids.
1853
:return: An iterator that yields (version_id, parents). Requested
1854
version_ids not present in the versioned file are simply skipped.
1855
The order is undefined, allowing for different optimisations in
1856
the underlying implementation.
1859
all_nodes = set(self._get_entries(self._version_ids_to_keys(version_ids)))
1861
present_parents = set()
1862
for node in all_nodes:
1863
all_parents.update(node[3][0])
1864
# any node we are querying must be present
1865
present_parents.add(node[1])
1866
unknown_parents = all_parents.difference(present_parents)
1867
present_parents.update(self._present_keys(unknown_parents))
1868
for node in all_nodes:
1870
for parent in node[3][0]:
1871
if parent in present_parents:
1872
parents.append(parent[0])
1873
yield node[1][0], tuple(parents)
1875
for node in self._get_entries(self._version_ids_to_keys(version_ids)):
1876
yield node[1][0], ()
1878
def num_versions(self):
1879
return len(list(self._graph_index.iter_all_entries()))
1881
__len__ = num_versions
1883
def get_versions(self):
1884
"""Get all the versions in the file. not topologically sorted."""
1885
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1887
def has_version(self, version_id):
1888
"""True if the version is in the index."""
1889
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1891
def _keys_to_version_ids(self, keys):
1892
return tuple(key[0] for key in keys)
1894
def get_position(self, version_id):
1895
"""Return details needed to access the version.
1897
:return: a tuple (index, data position, size) to hand to the access
1898
logic to get the record.
1900
node = self._get_node(version_id)
1901
return self._node_to_position(node)
1903
def _node_to_position(self, node):
1904
"""Convert an index value to position details."""
1905
bits = node[2][1:].split(' ')
1906
return node[0], int(bits[0]), int(bits[1])
1908
def get_method(self, version_id):
1909
"""Return compression method of specified version."""
1910
return self._get_method(self._get_node(version_id))
1912
def _get_node(self, version_id):
1914
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1916
raise RevisionNotPresent(version_id, self)
1918
def get_options(self, version_id):
1919
"""Return a list representing options.
1923
node = self._get_node(version_id)
1924
options = [self._get_method(node)]
1925
if node[2][0] == 'N':
1926
options.append('no-eol')
1929
def get_parent_map(self, version_ids):
1930
"""Passed through to by KnitVersionedFile.get_parent_map."""
1931
nodes = self._get_entries(self._version_ids_to_keys(version_ids))
1935
result[node[1][0]] = self._keys_to_version_ids(node[3][0])
1938
result[node[1][0]] = ()
1941
def get_parents_with_ghosts(self, version_id):
1942
"""Return parents of specified version with ghosts."""
1944
return self.get_parent_map([version_id])[version_id]
1946
raise RevisionNotPresent(version_id, self)
1948
def check_versions_present(self, version_ids):
1949
"""Check that all specified versions are present."""
1950
keys = self._version_ids_to_keys(version_ids)
1951
present = self._present_keys(keys)
1952
missing = keys.difference(present)
1954
raise RevisionNotPresent(missing.pop(), self)
1956
def add_version(self, version_id, options, access_memo, parents):
1957
"""Add a version record to the index."""
1958
return self.add_versions(((version_id, options, access_memo, parents),))
1960
def add_versions(self, versions, random_id=False):
1961
"""Add multiple versions to the index.
1963
This function does not insert data into the Immutable GraphIndex
1964
backing the KnitGraphIndex, instead it prepares data for insertion by
1965
the caller and checks that it is safe to insert then calls
1966
self._add_callback with the prepared GraphIndex nodes.
1968
:param versions: a list of tuples:
1969
(version_id, options, pos, size, parents).
1970
:param random_id: If True the ids being added were randomly generated
1971
and no check for existence will be performed.
1973
if not self._add_callback:
1974
raise errors.ReadOnlyError(self)
1975
# we hope there are no repositories with inconsistent parentage
1980
for (version_id, options, access_memo, parents) in versions:
1981
index, pos, size = access_memo
1982
key = (version_id, )
1983
parents = tuple((parent, ) for parent in parents)
1984
if 'no-eol' in options:
1988
value += "%d %d" % (pos, size)
1989
if not self._deltas:
1990
if 'line-delta' in options:
1991
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1994
if 'line-delta' in options:
1995
node_refs = (parents, (parents[0],))
1997
node_refs = (parents, ())
1999
node_refs = (parents, )
2002
raise KnitCorrupt(self, "attempt to add node with parents "
2003
"in parentless index.")
2005
keys[key] = (value, node_refs)
2007
present_nodes = self._get_entries(keys)
2008
for (index, key, value, node_refs) in present_nodes:
2009
if (value, node_refs) != keys[key]:
2010
raise KnitCorrupt(self, "inconsistent details in add_versions"
2011
": %s %s" % ((value, node_refs), keys[key]))
2015
for key, (value, node_refs) in keys.iteritems():
2016
result.append((key, value, node_refs))
2018
for key, (value, node_refs) in keys.iteritems():
2019
result.append((key, value))
2020
self._add_callback(result)
2022
def _version_ids_to_keys(self, version_ids):
2023
return set((version_id, ) for version_id in version_ids)
2026
class _KnitAccess(object):
2027
"""Access to knit records in a .knit file."""
2029
def __init__(self, transport, filename, _file_mode, _dir_mode,
2030
_need_to_create, _create_parent_dir):
2031
"""Create a _KnitAccess for accessing and inserting data.
2033
:param transport: The transport the .knit is located on.
2034
:param filename: The filename of the .knit.
2036
self._transport = transport
2037
self._filename = filename
2038
self._file_mode = _file_mode
2039
self._dir_mode = _dir_mode
2040
self._need_to_create = _need_to_create
2041
self._create_parent_dir = _create_parent_dir
2043
def add_raw_records(self, sizes, raw_data):
2044
"""Add raw knit bytes to a storage area.
2046
The data is spooled to whereever the access method is storing data.
2048
:param sizes: An iterable containing the size of each raw data segment.
2049
:param raw_data: A bytestring containing the data.
2050
:return: A list of memos to retrieve the record later. Each memo is a
2051
tuple - (index, pos, length), where the index field is always None
2052
for the .knit access method.
2054
assert type(raw_data) == str, \
2055
'data must be plain bytes was %s' % type(raw_data)
2056
if not self._need_to_create:
2057
base = self._transport.append_bytes(self._filename, raw_data)
2059
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2060
create_parent_dir=self._create_parent_dir,
2061
mode=self._file_mode,
2062
dir_mode=self._dir_mode)
2063
self._need_to_create = False
2067
result.append((None, base, size))
2072
"""IFF this data access has its own storage area, initialise it.
2076
self._transport.put_bytes_non_atomic(self._filename, '',
2077
mode=self._file_mode)
2079
def open_file(self):
2080
"""IFF this data access can be represented as a single file, open it.
2082
For knits that are not mapped to a single file on disk this will
2085
:return: None or a file handle.
2088
return self._transport.get(self._filename)
2093
def get_raw_records(self, memos_for_retrieval):
2094
"""Get the raw bytes for a records.
2096
:param memos_for_retrieval: An iterable containing the (index, pos,
2097
length) memo for retrieving the bytes. The .knit method ignores
2098
the index as there is always only a single file.
2099
:return: An iterator over the bytes of the records.
2101
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2102
for pos, data in self._transport.readv(self._filename, read_vector):
2106
class _PackAccess(object):
2107
"""Access to knit records via a collection of packs."""
2109
def __init__(self, index_to_packs, writer=None):
2110
"""Create a _PackAccess object.
2112
:param index_to_packs: A dict mapping index objects to the transport
2113
and file names for obtaining data.
2114
:param writer: A tuple (pack.ContainerWriter, write_index) which
2115
contains the pack to write, and the index that reads from it will
2119
self.container_writer = writer[0]
2120
self.write_index = writer[1]
2122
self.container_writer = None
2123
self.write_index = None
2124
self.indices = index_to_packs
2126
def add_raw_records(self, sizes, raw_data):
2127
"""Add raw knit bytes to a storage area.
2129
The data is spooled to the container writer in one bytes-record per
2132
:param sizes: An iterable containing the size of each raw data segment.
2133
:param raw_data: A bytestring containing the data.
2134
:return: A list of memos to retrieve the record later. Each memo is a
2135
tuple - (index, pos, length), where the index field is the
2136
write_index object supplied to the PackAccess object.
2138
assert type(raw_data) == str, \
2139
'data must be plain bytes was %s' % type(raw_data)
2143
p_offset, p_length = self.container_writer.add_bytes_record(
2144
raw_data[offset:offset+size], [])
2146
result.append((self.write_index, p_offset, p_length))
2150
"""Pack based knits do not get individually created."""
2152
def get_raw_records(self, memos_for_retrieval):
2153
"""Get the raw bytes for a records.
2155
:param memos_for_retrieval: An iterable containing the (index, pos,
2156
length) memo for retrieving the bytes. The Pack access method
2157
looks up the pack to use for a given record in its index_to_pack
2159
:return: An iterator over the bytes of the records.
2161
# first pass, group into same-index requests
2163
current_index = None
2164
for (index, offset, length) in memos_for_retrieval:
2165
if current_index == index:
2166
current_list.append((offset, length))
2168
if current_index is not None:
2169
request_lists.append((current_index, current_list))
2170
current_index = index
2171
current_list = [(offset, length)]
2172
# handle the last entry
2173
if current_index is not None:
2174
request_lists.append((current_index, current_list))
2175
for index, offsets in request_lists:
2176
transport, path = self.indices[index]
2177
reader = pack.make_readv_reader(transport, path, offsets)
2178
for names, read_func in reader.iter_records():
2179
yield read_func(None)
2181
def open_file(self):
2182
"""Pack based knits have no single file."""
2185
def set_writer(self, writer, index, (transport, packname)):
2186
"""Set a writer to use for adding data."""
2187
if index is not None:
2188
self.indices[index] = (transport, packname)
2189
self.container_writer = writer
2190
self.write_index = index
2193
class _StreamAccess(object):
2194
"""A Knit Access object that provides data from a datastream.
2196
It also provides a fallback to present as unannotated data, annotated data
2197
from a *backing* access object.
2199
This is triggered by a index_memo which is pointing to a different index
2200
than this was constructed with, and is used to allow extracting full
2201
unannotated texts for insertion into annotated knits.
2204
def __init__(self, reader_callable, stream_index, backing_knit,
2206
"""Create a _StreamAccess object.
2208
:param reader_callable: The reader_callable from the datastream.
2209
This is called to buffer all the data immediately, for
2211
:param stream_index: The index the data stream this provides access to
2212
which will be present in native index_memo's.
2213
:param backing_knit: The knit object that will provide access to
2214
annotated texts which are not available in the stream, so as to
2215
create unannotated texts.
2216
:param orig_factory: The original content factory used to generate the
2217
stream. This is used for checking whether the thunk code for
2218
supporting _copy_texts will generate the correct form of data.
2220
self.data = reader_callable(None)
2221
self.stream_index = stream_index
2222
self.backing_knit = backing_knit
2223
self.orig_factory = orig_factory
2225
def get_raw_records(self, memos_for_retrieval):
2226
"""Get the raw bytes for a records.
2228
:param memos_for_retrieval: An iterable containing the (thunk_flag,
2229
index, start, end) memo for retrieving the bytes.
2230
:return: An iterator over the bytes of the records.
2232
# use a generator for memory friendliness
2233
for thunk_flag, version_id, start, end in memos_for_retrieval:
2234
if version_id is self.stream_index:
2235
yield self.data[start:end]
2237
# we have been asked to thunk. This thunking only occurs when
2238
# we are obtaining plain texts from an annotated backing knit
2239
# so that _copy_texts will work.
2240
# We could improve performance here by scanning for where we need
2241
# to do this and using get_line_list, then interleaving the output
2242
# as desired. However, for now, this is sufficient.
2243
if self.orig_factory.__class__ != KnitPlainFactory:
2244
raise errors.KnitCorrupt(
2245
self, 'Bad thunk request %r' % version_id)
2246
lines = self.backing_knit.get_lines(version_id)
2247
line_bytes = ''.join(lines)
2248
digest = sha_string(line_bytes)
2250
if lines[-1][-1] != '\n':
2251
lines[-1] = lines[-1] + '\n'
2253
orig_options = list(self.backing_knit._index.get_options(version_id))
2254
if 'fulltext' not in orig_options:
2255
if 'line-delta' not in orig_options:
2256
raise errors.KnitCorrupt(self,
2257
'Unknown compression method %r' % orig_options)
2258
orig_options.remove('line-delta')
2259
orig_options.append('fulltext')
2260
# We want plain data, because we expect to thunk only to allow text
2262
size, bytes = self.backing_knit._data._record_to_data(version_id,
2263
digest, lines, line_bytes)
2267
class _StreamIndex(object):
2268
"""A Knit Index object that uses the data map from a datastream."""
2270
def __init__(self, data_list, backing_index):
2271
"""Create a _StreamIndex object.
2273
:param data_list: The data_list from the datastream.
2274
:param backing_index: The index which will supply values for nodes
2275
referenced outside of this stream.
2277
self.data_list = data_list
2278
self.backing_index = backing_index
2279
self._by_version = {}
2281
for key, options, length, parents in data_list:
2282
self._by_version[key] = options, (pos, pos + length), parents
2285
def get_ancestry(self, versions, topo_sorted):
2286
"""Get an ancestry list for versions."""
2288
# Not needed for basic joins
2289
raise NotImplementedError(self.get_ancestry)
2290
# get a graph of all the mentioned versions:
2291
# Little ugly - basically copied from KnitIndex, but don't want to
2292
# accidentally incorporate too much of that index's code.
2294
pending = set(versions)
2295
cache = self._by_version
2297
version = pending.pop()
2300
parents = [p for p in cache[version][2] if p in cache]
2302
raise RevisionNotPresent(version, self)
2303
# if not completed and not a ghost
2304
pending.update([p for p in parents if p not in ancestry])
2305
ancestry.add(version)
2306
return list(ancestry)
2308
def get_build_details(self, version_ids):
2309
"""Get the method, index_memo and compression parent for version_ids.
2311
Ghosts are omitted from the result.
2313
:param version_ids: An iterable of version_ids.
2314
:return: A dict of version_id:(index_memo, compression_parent,
2315
parents, record_details).
2317
opaque structure to pass to read_records to extract the raw
2320
Content that this record is built upon, may be None
2322
Logical parents of this node
2324
extra information about the content which needs to be passed to
2325
Factory.parse_record
2328
for version_id in version_ids:
2330
method = self.get_method(version_id)
2331
except errors.RevisionNotPresent:
2332
# ghosts are omitted
2334
parent_ids = self.get_parents_with_ghosts(version_id)
2335
noeol = ('no-eol' in self.get_options(version_id))
2336
if method == 'fulltext':
2337
compression_parent = None
2339
compression_parent = parent_ids[0]
2340
index_memo = self.get_position(version_id)
2341
result[version_id] = (index_memo, compression_parent,
2342
parent_ids, (method, noeol))
2345
def get_method(self, version_id):
2346
"""Return compression method of specified version."""
2348
options = self._by_version[version_id][0]
2350
# Strictly speaking this should check in the backing knit, but
2351
# until we have a test to discriminate, this will do.
2352
return self.backing_index.get_method(version_id)
2353
if 'fulltext' in options:
2355
elif 'line-delta' in options:
2358
raise errors.KnitIndexUnknownMethod(self, options)
2360
def get_options(self, version_id):
2361
"""Return a list representing options.
2366
return self._by_version[version_id][0]
2368
return self.backing_index.get_options(version_id)
2370
def get_parent_map(self, version_ids):
2371
"""Passed through to by KnitVersionedFile.get_parent_map."""
2374
for version_id in version_ids:
2376
result[version_id] = self._by_version[version_id][2]
2378
pending_ids.add(version_id)
2379
result.update(self.backing_index.get_parent_map(pending_ids))
2382
def get_parents_with_ghosts(self, version_id):
2383
"""Return parents of specified version with ghosts."""
2385
return self.get_parent_map([version_id])[version_id]
2387
raise RevisionNotPresent(version_id, self)
2389
def get_position(self, version_id):
2390
"""Return details needed to access the version.
2392
_StreamAccess has the data as a big array, so we return slice
2393
coordinates into that (as index_memo's are opaque outside the
2394
index and matching access class).
2396
:return: a tuple (thunk_flag, index, start, end). If thunk_flag is
2397
False, index will be self, otherwise it will be a version id.
2400
start, end = self._by_version[version_id][1]
2401
return False, self, start, end
2403
# Signal to the access object to handle this from the backing knit.
2404
return (True, version_id, None, None)
2406
def get_versions(self):
2407
"""Get all the versions in the stream."""
2408
return self._by_version.keys()
2410
def iter_parents(self, version_ids):
2411
"""Iterate through the parents for many version ids.
2413
:param version_ids: An iterable yielding version_ids.
2414
:return: An iterator that yields (version_id, parents). Requested
2415
version_ids not present in the versioned file are simply skipped.
2416
The order is undefined, allowing for different optimisations in
2417
the underlying implementation.
2420
for version in version_ids:
2422
result.append((version, self._by_version[version][2]))
2428
class _KnitData(object):
2429
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2431
The KnitData class provides the logic for parsing and using knit records,
2432
making use of an access method for the low level read and write operations.
2435
def __init__(self, access):
2436
"""Create a KnitData object.
2438
:param access: The access method to use. Access methods such as
2439
_KnitAccess manage the insertion of raw records and the subsequent
2440
retrieval of the same.
2442
self._access = access
2443
self._checked = False
2444
# TODO: jam 20060713 conceptually, this could spill to disk
2445
# if the cached size gets larger than a certain amount
2446
# but it complicates the model a bit, so for now just use
2447
# a simple dictionary
2449
self._do_cache = False
2451
def enable_cache(self):
2452
"""Enable caching of reads."""
2453
self._do_cache = True
2455
def clear_cache(self):
2456
"""Clear the record cache."""
2457
self._do_cache = False
2460
def _open_file(self):
2461
return self._access.open_file()
2463
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2464
"""Convert version_id, digest, lines into a raw data block.
2466
:param dense_lines: The bytes of lines but in a denser form. For
2467
instance, if lines is a list of 1000 bytestrings each ending in \n,
2468
dense_lines may be a list with one line in it, containing all the
2469
1000's lines and their \n's. Using dense_lines if it is already
2470
known is a win because the string join to create bytes in this
2471
function spends less time resizing the final string.
2472
:return: (len, a StringIO instance with the raw data ready to read.)
2474
# Note: using a string copy here increases memory pressure with e.g.
2475
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2476
# when doing the initial commit of a mozilla tree. RBC 20070921
2477
bytes = ''.join(chain(
2478
["version %s %d %s\n" % (version_id,
2481
dense_lines or lines,
2482
["end %s\n" % version_id]))
2483
assert bytes.__class__ == str
2484
compressed_bytes = bytes_to_gzip(bytes)
2485
return len(compressed_bytes), compressed_bytes
2487
def add_raw_records(self, sizes, raw_data):
2488
"""Append a prepared record to the data file.
2490
:param sizes: An iterable containing the size of each raw data segment.
2491
:param raw_data: A bytestring containing the data.
2492
:return: a list of index data for the way the data was stored.
2493
See the access method add_raw_records documentation for more
2496
return self._access.add_raw_records(sizes, raw_data)
2498
def _parse_record_header(self, version_id, raw_data):
2499
"""Parse a record header for consistency.
2501
:return: the header and the decompressor stream.
2502
as (stream, header_record)
2504
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2506
rec = self._check_header(version_id, df.readline())
2507
except Exception, e:
2508
raise KnitCorrupt(self._access,
2509
"While reading {%s} got %s(%s)"
2510
% (version_id, e.__class__.__name__, str(e)))
2513
def _check_header(self, version_id, line):
2516
raise KnitCorrupt(self._access,
2517
'unexpected number of elements in record header')
2518
if rec[1] != version_id:
2519
raise KnitCorrupt(self._access,
2520
'unexpected version, wanted %r, got %r'
2521
% (version_id, rec[1]))
2524
def _parse_record(self, version_id, data):
2526
# 4168 calls in 2880 217 internal
2527
# 4168 calls to _parse_record_header in 2121
2528
# 4168 calls to readlines in 330
2529
df = GzipFile(mode='rb', fileobj=StringIO(data))
2532
record_contents = df.readlines()
2533
except Exception, e:
2534
raise KnitCorrupt(self._access,
2535
"While reading {%s} got %s(%s)"
2536
% (version_id, e.__class__.__name__, str(e)))
2537
header = record_contents.pop(0)
2538
rec = self._check_header(version_id, header)
2540
last_line = record_contents.pop()
2541
if len(record_contents) != int(rec[2]):
2542
raise KnitCorrupt(self._access,
2543
'incorrect number of lines %s != %s'
2545
% (len(record_contents), int(rec[2]),
2547
if last_line != 'end %s\n' % rec[1]:
2548
raise KnitCorrupt(self._access,
2549
'unexpected version end line %r, wanted %r'
2550
% (last_line, version_id))
2552
return record_contents, rec[3]
2554
def read_records_iter_raw(self, records):
2555
"""Read text records from data file and yield raw data.
2557
This unpacks enough of the text record to validate the id is
2558
as expected but thats all.
2560
# setup an iterator of the external records:
2561
# uses readv so nice and fast we hope.
2563
# grab the disk data needed.
2565
# Don't check _cache if it is empty
2566
needed_offsets = [index_memo for version_id, index_memo
2568
if version_id not in self._cache]
2570
needed_offsets = [index_memo for version_id, index_memo
2573
raw_records = self._access.get_raw_records(needed_offsets)
2575
for version_id, index_memo in records:
2576
if version_id in self._cache:
2577
# This data has already been validated
2578
data = self._cache[version_id]
2580
data = raw_records.next()
2582
self._cache[version_id] = data
2584
# validate the header
2585
df, rec = self._parse_record_header(version_id, data)
2587
yield version_id, data
2589
def read_records_iter(self, records):
2590
"""Read text records from data file and yield result.
2592
The result will be returned in whatever is the fastest to read.
2593
Not by the order requested. Also, multiple requests for the same
2594
record will only yield 1 response.
2595
:param records: A list of (version_id, pos, len) entries
2596
:return: Yields (version_id, contents, digest) in the order
2597
read, not the order requested
2603
# Skip records we have alread seen
2604
yielded_records = set()
2605
needed_records = set()
2606
for record in records:
2607
if record[0] in self._cache:
2608
if record[0] in yielded_records:
2610
yielded_records.add(record[0])
2611
data = self._cache[record[0]]
2612
content, digest = self._parse_record(record[0], data)
2613
yield (record[0], content, digest)
2615
needed_records.add(record)
2616
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2618
needed_records = sorted(set(records), key=operator.itemgetter(1))
2620
if not needed_records:
2623
# The transport optimizes the fetching as well
2624
# (ie, reads continuous ranges.)
2625
raw_data = self._access.get_raw_records(
2626
[index_memo for version_id, index_memo in needed_records])
2628
for (version_id, index_memo), data in \
2629
izip(iter(needed_records), raw_data):
2630
content, digest = self._parse_record(version_id, data)
2632
self._cache[version_id] = data
2633
yield version_id, content, digest
2635
def read_records(self, records):
2636
"""Read records into a dictionary."""
2638
for record_id, content, digest in \
2639
self.read_records_iter(records):
2640
components[record_id] = (content, digest)
2644
class InterKnit(InterVersionedFile):
2645
"""Optimised code paths for knit to knit operations."""
2647
_matching_file_from_factory = KnitVersionedFile
2648
_matching_file_to_factory = KnitVersionedFile
2651
def is_compatible(source, target):
2652
"""Be compatible with knits. """
2654
return (isinstance(source, KnitVersionedFile) and
2655
isinstance(target, KnitVersionedFile))
2656
except AttributeError:
2659
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2660
"""Copy texts to the target by extracting and adding them one by one.
2662
see join() for the parameter definitions.
2664
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2665
# --- the below is factorable out with VersionedFile.join, but wait for
2666
# VersionedFiles, it may all be simpler then.
2667
graph = Graph(self.source)
2668
search = graph._make_breadth_first_searcher(version_ids)
2669
transitive_ids = set()
2670
map(transitive_ids.update, list(search))
2671
parent_map = self.source.get_parent_map(transitive_ids)
2672
order = topo_sort(parent_map.items())
2674
def size_of_content(content):
2675
return sum(len(line) for line in content.text())
2676
# Cache at most 10MB of parent texts
2677
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2678
compute_size=size_of_content)
2679
# TODO: jam 20071116 It would be nice to have a streaming interface to
2680
# get multiple texts from a source. The source could be smarter
2681
# about how it handled intermediate stages.
2682
# get_line_list() or make_mpdiffs() seem like a possibility, but
2683
# at the moment they extract all full texts into memory, which
2684
# causes us to store more than our 3x fulltext goal.
2685
# Repository.iter_files_bytes() may be another possibility
2686
to_process = [version for version in order
2687
if version not in self.target]
2688
total = len(to_process)
2689
pb = ui.ui_factory.nested_progress_bar()
2691
for index, version in enumerate(to_process):
2692
pb.update('Converting versioned data', index, total)
2693
sha1, num_bytes, parent_text = self.target.add_lines(version,
2694
self.source.get_parents_with_ghosts(version),
2695
self.source.get_lines(version),
2696
parent_texts=parent_cache)
2697
parent_cache[version] = parent_text
2702
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2703
"""See InterVersionedFile.join."""
2704
assert isinstance(self.source, KnitVersionedFile)
2705
assert isinstance(self.target, KnitVersionedFile)
2707
# If the source and target are mismatched w.r.t. annotations vs
2708
# plain, the data needs to be converted accordingly
2709
if self.source.factory.annotated == self.target.factory.annotated:
2711
elif self.source.factory.annotated:
2712
converter = self._anno_to_plain_converter
2714
# We're converting from a plain to an annotated knit. Copy them
2715
# across by full texts.
2716
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2718
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2722
pb = ui.ui_factory.nested_progress_bar()
2724
version_ids = list(version_ids)
2725
if None in version_ids:
2726
version_ids.remove(None)
2728
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2730
this_versions = set(self.target._index.get_versions())
2731
# XXX: For efficiency we should not look at the whole index,
2732
# we only need to consider the referenced revisions - they
2733
# must all be present, or the method must be full-text.
2734
# TODO, RBC 20070919
2735
needed_versions = self.source_ancestry - this_versions
2737
if not needed_versions:
2739
full_list = topo_sort(
2740
self.source.get_parent_map(self.source.versions()))
2742
version_list = [i for i in full_list if (not self.target.has_version(i)
2743
and i in needed_versions)]
2747
copy_queue_records = []
2749
for version_id in version_list:
2750
options = self.source._index.get_options(version_id)
2751
parents = self.source._index.get_parents_with_ghosts(version_id)
2752
# check that its will be a consistent copy:
2753
for parent in parents:
2754
# if source has the parent, we must :
2755
# * already have it or
2756
# * have it scheduled already
2757
# otherwise we don't care
2758
assert (self.target.has_version(parent) or
2759
parent in copy_set or
2760
not self.source.has_version(parent))
2761
index_memo = self.source._index.get_position(version_id)
2762
copy_queue_records.append((version_id, index_memo))
2763
copy_queue.append((version_id, options, parents))
2764
copy_set.add(version_id)
2766
# data suck the join:
2768
total = len(version_list)
2771
for (version_id, raw_data), \
2772
(version_id2, options, parents) in \
2773
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2775
assert version_id == version_id2, 'logic error, inconsistent results'
2777
pb.update("Joining knit", count, total)
2779
size, raw_data = converter(raw_data, version_id, options,
2782
size = len(raw_data)
2783
raw_records.append((version_id, options, parents, size))
2784
raw_datum.append(raw_data)
2785
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2790
def _anno_to_plain_converter(self, raw_data, version_id, options,
2792
"""Convert annotated content to plain content."""
2793
data, digest = self.source._data._parse_record(version_id, raw_data)
2794
if 'fulltext' in options:
2795
content = self.source.factory.parse_fulltext(data, version_id)
2796
lines = self.target.factory.lower_fulltext(content)
2798
delta = self.source.factory.parse_line_delta(data, version_id,
2800
lines = self.target.factory.lower_line_delta(delta)
2801
return self.target._data._record_to_data(version_id, digest, lines)
2804
InterVersionedFile.register_optimiser(InterKnit)
2807
class WeaveToKnit(InterVersionedFile):
2808
"""Optimised code paths for weave to knit operations."""
2810
_matching_file_from_factory = bzrlib.weave.WeaveFile
2811
_matching_file_to_factory = KnitVersionedFile
2814
def is_compatible(source, target):
2815
"""Be compatible with weaves to knits."""
2817
return (isinstance(source, bzrlib.weave.Weave) and
2818
isinstance(target, KnitVersionedFile))
2819
except AttributeError:
2822
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2823
"""See InterVersionedFile.join."""
2824
assert isinstance(self.source, bzrlib.weave.Weave)
2825
assert isinstance(self.target, KnitVersionedFile)
2827
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2832
pb = ui.ui_factory.nested_progress_bar()
2834
version_ids = list(version_ids)
2836
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2837
this_versions = set(self.target._index.get_versions())
2838
needed_versions = self.source_ancestry - this_versions
2840
if not needed_versions:
2842
full_list = topo_sort(
2843
self.source.get_parent_map(self.source.versions()))
2845
version_list = [i for i in full_list if (not self.target.has_version(i)
2846
and i in needed_versions)]
2850
total = len(version_list)
2851
parent_map = self.source.get_parent_map(version_list)
2852
for version_id in version_list:
2853
pb.update("Converting to knit", count, total)
2854
parents = parent_map[version_id]
2855
# check that its will be a consistent copy:
2856
for parent in parents:
2857
# if source has the parent, we must already have it
2858
assert (self.target.has_version(parent))
2859
self.target.add_lines(
2860
version_id, parents, self.source.get_lines(version_id))
2867
InterVersionedFile.register_optimiser(WeaveToKnit)
2870
# Deprecated, use PatienceSequenceMatcher instead
2871
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2874
def annotate_knit(knit, revision_id):
2875
"""Annotate a knit with no cached annotations.
2877
This implementation is for knits with no cached annotations.
2878
It will work for knits with cached annotations, but this is not
2881
annotator = _KnitAnnotator(knit)
2882
return iter(annotator.annotate(revision_id))
2885
class _KnitAnnotator(object):
2886
"""Build up the annotations for a text."""
2888
def __init__(self, knit):
2891
# Content objects, differs from fulltexts because of how final newlines
2892
# are treated by knits. the content objects here will always have a
2894
self._fulltext_contents = {}
2896
# Annotated lines of specific revisions
2897
self._annotated_lines = {}
2899
# Track the raw data for nodes that we could not process yet.
2900
# This maps the revision_id of the base to a list of children that will
2901
# annotated from it.
2902
self._pending_children = {}
2904
# Nodes which cannot be extracted
2905
self._ghosts = set()
2907
# Track how many children this node has, so we know if we need to keep
2909
self._annotate_children = {}
2910
self._compression_children = {}
2912
self._all_build_details = {}
2913
# The children => parent revision_id graph
2914
self._revision_id_graph = {}
2916
self._heads_provider = None
2918
self._nodes_to_keep_annotations = set()
2919
self._generations_until_keep = 100
2921
def set_generations_until_keep(self, value):
2922
"""Set the number of generations before caching a node.
2924
Setting this to -1 will cache every merge node, setting this higher
2925
will cache fewer nodes.
2927
self._generations_until_keep = value
2929
def _add_fulltext_content(self, revision_id, content_obj):
2930
self._fulltext_contents[revision_id] = content_obj
2931
# TODO: jam 20080305 It might be good to check the sha1digest here
2932
return content_obj.text()
2934
def _check_parents(self, child, nodes_to_annotate):
2935
"""Check if all parents have been processed.
2937
:param child: A tuple of (rev_id, parents, raw_content)
2938
:param nodes_to_annotate: If child is ready, add it to
2939
nodes_to_annotate, otherwise put it back in self._pending_children
2941
for parent_id in child[1]:
2942
if (parent_id not in self._annotated_lines):
2943
# This parent is present, but another parent is missing
2944
self._pending_children.setdefault(parent_id,
2948
# This one is ready to be processed
2949
nodes_to_annotate.append(child)
2951
def _add_annotation(self, revision_id, fulltext, parent_ids,
2952
left_matching_blocks=None):
2953
"""Add an annotation entry.
2955
All parents should already have been annotated.
2956
:return: A list of children that now have their parents satisfied.
2958
a = self._annotated_lines
2959
annotated_parent_lines = [a[p] for p in parent_ids]
2960
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2961
fulltext, revision_id, left_matching_blocks,
2962
heads_provider=self._get_heads_provider()))
2963
self._annotated_lines[revision_id] = annotated_lines
2964
for p in parent_ids:
2965
ann_children = self._annotate_children[p]
2966
ann_children.remove(revision_id)
2967
if (not ann_children
2968
and p not in self._nodes_to_keep_annotations):
2969
del self._annotated_lines[p]
2970
del self._all_build_details[p]
2971
if p in self._fulltext_contents:
2972
del self._fulltext_contents[p]
2973
# Now that we've added this one, see if there are any pending
2974
# deltas to be done, certainly this parent is finished
2975
nodes_to_annotate = []
2976
for child in self._pending_children.pop(revision_id, []):
2977
self._check_parents(child, nodes_to_annotate)
2978
return nodes_to_annotate
2980
def _get_build_graph(self, revision_id):
2981
"""Get the graphs for building texts and annotations.
2983
The data you need for creating a full text may be different than the
2984
data you need to annotate that text. (At a minimum, you need both
2985
parents to create an annotation, but only need 1 parent to generate the
2988
:return: A list of (revision_id, index_memo) records, suitable for
2989
passing to read_records_iter to start reading in the raw data fro/
2992
if revision_id in self._annotated_lines:
2995
pending = set([revision_id])
3000
# get all pending nodes
3002
this_iteration = pending
3003
build_details = self._knit._index.get_build_details(this_iteration)
3004
self._all_build_details.update(build_details)
3005
# new_nodes = self._knit._index._get_entries(this_iteration)
3007
for rev_id, details in build_details.iteritems():
3008
(index_memo, compression_parent, parents,
3009
record_details) = details
3010
self._revision_id_graph[rev_id] = parents
3011
records.append((rev_id, index_memo))
3012
# Do we actually need to check _annotated_lines?
3013
pending.update(p for p in parents
3014
if p not in self._all_build_details)
3015
if compression_parent:
3016
self._compression_children.setdefault(compression_parent,
3019
for parent in parents:
3020
self._annotate_children.setdefault(parent,
3022
num_gens = generation - kept_generation
3023
if ((num_gens >= self._generations_until_keep)
3024
and len(parents) > 1):
3025
kept_generation = generation
3026
self._nodes_to_keep_annotations.add(rev_id)
3028
missing_versions = this_iteration.difference(build_details.keys())
3029
self._ghosts.update(missing_versions)
3030
for missing_version in missing_versions:
3031
# add a key, no parents
3032
self._revision_id_graph[missing_version] = ()
3033
pending.discard(missing_version) # don't look for it
3034
# XXX: This should probably be a real exception, as it is a data
3036
assert not self._ghosts.intersection(self._compression_children), \
3037
"We cannot have nodes which have a compression parent of a ghost."
3038
# Cleanout anything that depends on a ghost so that we don't wait for
3039
# the ghost to show up
3040
for node in self._ghosts:
3041
if node in self._annotate_children:
3042
# We won't be building this node
3043
del self._annotate_children[node]
3044
# Generally we will want to read the records in reverse order, because
3045
# we find the parent nodes after the children
3049
def _annotate_records(self, records):
3050
"""Build the annotations for the listed records."""
3051
# We iterate in the order read, rather than a strict order requested
3052
# However, process what we can, and put off to the side things that
3053
# still need parents, cleaning them up when those parents are
3055
for (rev_id, record,
3056
digest) in self._knit._data.read_records_iter(records):
3057
if rev_id in self._annotated_lines:
3059
parent_ids = self._revision_id_graph[rev_id]
3060
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3061
details = self._all_build_details[rev_id]
3062
(index_memo, compression_parent, parents,
3063
record_details) = details
3064
nodes_to_annotate = []
3065
# TODO: Remove the punning between compression parents, and
3066
# parent_ids, we should be able to do this without assuming
3068
if len(parent_ids) == 0:
3069
# There are no parents for this node, so just add it
3070
# TODO: This probably needs to be decoupled
3071
assert compression_parent is None
3072
fulltext_content, delta = self._knit.factory.parse_record(
3073
rev_id, record, record_details, None)
3074
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3075
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3076
parent_ids, left_matching_blocks=None))
3078
child = (rev_id, parent_ids, record)
3079
# Check if all the parents are present
3080
self._check_parents(child, nodes_to_annotate)
3081
while nodes_to_annotate:
3082
# Should we use a queue here instead of a stack?
3083
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3084
(index_memo, compression_parent, parents,
3085
record_details) = self._all_build_details[rev_id]
3086
if compression_parent is not None:
3087
comp_children = self._compression_children[compression_parent]
3088
assert rev_id in comp_children
3089
# If there is only 1 child, it is safe to reuse this
3091
reuse_content = (len(comp_children) == 1
3092
and compression_parent not in
3093
self._nodes_to_keep_annotations)
3095
# Remove it from the cache since it will be changing
3096
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3097
# Make sure to copy the fulltext since it might be
3099
parent_fulltext = list(parent_fulltext_content.text())
3101
parent_fulltext_content = self._fulltext_contents[compression_parent]
3102
parent_fulltext = parent_fulltext_content.text()
3103
comp_children.remove(rev_id)
3104
fulltext_content, delta = self._knit.factory.parse_record(
3105
rev_id, record, record_details,
3106
parent_fulltext_content,
3107
copy_base_content=(not reuse_content))
3108
fulltext = self._add_fulltext_content(rev_id,
3110
blocks = KnitContent.get_line_delta_blocks(delta,
3111
parent_fulltext, fulltext)
3113
fulltext_content = self._knit.factory.parse_fulltext(
3115
fulltext = self._add_fulltext_content(rev_id,
3118
nodes_to_annotate.extend(
3119
self._add_annotation(rev_id, fulltext, parent_ids,
3120
left_matching_blocks=blocks))
3122
def _get_heads_provider(self):
3123
"""Create a heads provider for resolving ancestry issues."""
3124
if self._heads_provider is not None:
3125
return self._heads_provider
3126
parent_provider = _mod_graph.DictParentsProvider(
3127
self._revision_id_graph)
3128
graph_obj = _mod_graph.Graph(parent_provider)
3129
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3130
self._heads_provider = head_cache
3133
def annotate(self, revision_id):
3134
"""Return the annotated fulltext at the given revision.
3136
:param revision_id: The revision id for this file
3138
records = self._get_build_graph(revision_id)
3139
if revision_id in self._ghosts:
3140
raise errors.RevisionNotPresent(revision_id, self._knit)
3141
self._annotate_records(records)
3142
return self._annotated_lines[revision_id]
3146
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3148
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
