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
147
def apply_delta(self, delta, new_version_id):
148
"""Apply delta to this object to become new_version_id."""
149
raise NotImplementedError(self.apply_delta)
151
def cleanup_eol(self, copy_on_mutate=True):
152
if self._should_strip_eol:
154
self._lines = self._lines[:]
155
self.strip_last_line_newline()
157
def line_delta_iter(self, new_lines):
158
"""Generate line-based delta from this content to new_lines."""
159
new_texts = new_lines.text()
160
old_texts = self.text()
161
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
162
for tag, i1, i2, j1, j2 in s.get_opcodes():
165
# ofrom, oto, length, data
166
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
168
def line_delta(self, new_lines):
169
return list(self.line_delta_iter(new_lines))
172
def get_line_delta_blocks(knit_delta, source, target):
173
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
174
target_len = len(target)
177
for s_begin, s_end, t_len, new_text in knit_delta:
178
true_n = s_begin - s_pos
181
# knit deltas do not provide reliable info about whether the
182
# last line of a file matches, due to eol handling.
183
if source[s_pos + n -1] != target[t_pos + n -1]:
186
yield s_pos, t_pos, n
187
t_pos += t_len + true_n
189
n = target_len - t_pos
191
if source[s_pos + n -1] != target[t_pos + n -1]:
194
yield s_pos, t_pos, n
195
yield s_pos + (target_len - t_pos), target_len, 0
198
class AnnotatedKnitContent(KnitContent):
199
"""Annotated content."""
201
def __init__(self, lines):
202
KnitContent.__init__(self)
206
"""Return a list of (origin, text) for each content line."""
207
return list(self._lines)
209
def apply_delta(self, delta, new_version_id):
210
"""Apply delta to this object to become new_version_id."""
213
for start, end, count, delta_lines in delta:
214
lines[offset+start:offset+end] = delta_lines
215
offset = offset + (start - end) + count
217
def strip_last_line_newline(self):
218
line = self._lines[-1][1].rstrip('\n')
219
self._lines[-1] = (self._lines[-1][0], line)
220
self._should_strip_eol = False
224
lines = [text for origin, text in self._lines]
225
except ValueError, e:
226
# most commonly (only?) caused by the internal form of the knit
227
# missing annotation information because of a bug - see thread
229
raise KnitCorrupt(self,
230
"line in annotated knit missing annotation information: %s"
233
if self._should_strip_eol:
234
anno, line = lines[-1]
235
lines[-1] = (anno, line.rstrip('\n'))
239
return AnnotatedKnitContent(self._lines[:])
242
class PlainKnitContent(KnitContent):
243
"""Unannotated content.
245
When annotate[_iter] is called on this content, the same version is reported
246
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
250
def __init__(self, lines, version_id):
251
KnitContent.__init__(self)
253
self._version_id = version_id
256
"""Return a list of (origin, text) for each content line."""
257
return [(self._version_id, line) for line in self._lines]
259
def apply_delta(self, delta, new_version_id):
260
"""Apply delta to this object to become new_version_id."""
263
for start, end, count, delta_lines in delta:
264
lines[offset+start:offset+end] = delta_lines
265
offset = offset + (start - end) + count
266
self._version_id = new_version_id
269
return PlainKnitContent(self._lines[:], self._version_id)
271
def strip_last_line_newline(self):
272
self._lines[-1] = self._lines[-1].rstrip('\n')
273
self._should_strip_eol = False
277
if self._should_strip_eol:
279
lines[-1] = lines[-1].rstrip('\n')
283
class _KnitFactory(object):
284
"""Base class for common Factory functions."""
286
def parse_record(self, version_id, record, record_details,
287
base_content, copy_base_content=True):
288
"""Parse a record into a full content object.
290
:param version_id: The official version id for this content
291
:param record: The data returned by read_records_iter()
292
:param record_details: Details about the record returned by
294
:param base_content: If get_build_details returns a compression_parent,
295
you must return a base_content here, else use None
296
:param copy_base_content: When building from the base_content, decide
297
you can either copy it and return a new object, or modify it in
299
:return: (content, delta) A Content object and possibly a line-delta,
302
method, noeol = record_details
303
if method == 'line-delta':
304
assert base_content is not None
305
if copy_base_content:
306
content = base_content.copy()
308
content = base_content
309
delta = self.parse_line_delta(record, version_id)
310
content.apply_delta(delta, version_id)
312
content = self.parse_fulltext(record, version_id)
314
content._should_strip_eol = noeol
315
return (content, delta)
318
class KnitAnnotateFactory(_KnitFactory):
319
"""Factory for creating annotated Content objects."""
323
def make(self, lines, version_id):
324
num_lines = len(lines)
325
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
327
def parse_fulltext(self, content, version_id):
328
"""Convert fulltext to internal representation
330
fulltext content is of the format
331
revid(utf8) plaintext\n
332
internal representation is of the format:
335
# TODO: jam 20070209 The tests expect this to be returned as tuples,
336
# but the code itself doesn't really depend on that.
337
# Figure out a way to not require the overhead of turning the
338
# list back into tuples.
339
lines = [tuple(line.split(' ', 1)) for line in content]
340
return AnnotatedKnitContent(lines)
342
def parse_line_delta_iter(self, lines):
343
return iter(self.parse_line_delta(lines))
345
def parse_line_delta(self, lines, version_id, plain=False):
346
"""Convert a line based delta into internal representation.
348
line delta is in the form of:
349
intstart intend intcount
351
revid(utf8) newline\n
352
internal representation is
353
(start, end, count, [1..count tuples (revid, newline)])
355
:param plain: If True, the lines are returned as a plain
356
list without annotations, not as a list of (origin, content) tuples, i.e.
357
(start, end, count, [1..count newline])
364
def cache_and_return(line):
365
origin, text = line.split(' ', 1)
366
return cache.setdefault(origin, origin), text
368
# walk through the lines parsing.
369
# Note that the plain test is explicitly pulled out of the
370
# loop to minimise any performance impact
373
start, end, count = [int(n) for n in header.split(',')]
374
contents = [next().split(' ', 1)[1] for i in xrange(count)]
375
result.append((start, end, count, contents))
378
start, end, count = [int(n) for n in header.split(',')]
379
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
380
result.append((start, end, count, contents))
383
def get_fulltext_content(self, lines):
384
"""Extract just the content lines from a fulltext."""
385
return (line.split(' ', 1)[1] for line in lines)
387
def get_linedelta_content(self, lines):
388
"""Extract just the content from a line delta.
390
This doesn't return all of the extra information stored in a delta.
391
Only the actual content lines.
396
header = header.split(',')
397
count = int(header[2])
398
for i in xrange(count):
399
origin, text = next().split(' ', 1)
402
def lower_fulltext(self, content):
403
"""convert a fulltext content record into a serializable form.
405
see parse_fulltext which this inverts.
407
# TODO: jam 20070209 We only do the caching thing to make sure that
408
# the origin is a valid utf-8 line, eventually we could remove it
409
return ['%s %s' % (o, t) for o, t in content._lines]
411
def lower_line_delta(self, delta):
412
"""convert a delta into a serializable form.
414
See parse_line_delta which this inverts.
416
# TODO: jam 20070209 We only do the caching thing to make sure that
417
# the origin is a valid utf-8 line, eventually we could remove it
419
for start, end, c, lines in delta:
420
out.append('%d,%d,%d\n' % (start, end, c))
421
out.extend(origin + ' ' + text
422
for origin, text in lines)
425
def annotate(self, knit, version_id):
426
content = knit._get_content(version_id)
427
return content.annotate()
430
class KnitPlainFactory(_KnitFactory):
431
"""Factory for creating plain Content objects."""
435
def make(self, lines, version_id):
436
return PlainKnitContent(lines, version_id)
438
def parse_fulltext(self, content, version_id):
439
"""This parses an unannotated fulltext.
441
Note that this is not a noop - the internal representation
442
has (versionid, line) - its just a constant versionid.
444
return self.make(content, version_id)
446
def parse_line_delta_iter(self, lines, version_id):
448
num_lines = len(lines)
449
while cur < num_lines:
452
start, end, c = [int(n) for n in header.split(',')]
453
yield start, end, c, lines[cur:cur+c]
456
def parse_line_delta(self, lines, version_id):
457
return list(self.parse_line_delta_iter(lines, version_id))
459
def get_fulltext_content(self, lines):
460
"""Extract just the content lines from a fulltext."""
463
def get_linedelta_content(self, lines):
464
"""Extract just the content from a line delta.
466
This doesn't return all of the extra information stored in a delta.
467
Only the actual content lines.
472
header = header.split(',')
473
count = int(header[2])
474
for i in xrange(count):
477
def lower_fulltext(self, content):
478
return content.text()
480
def lower_line_delta(self, delta):
482
for start, end, c, lines in delta:
483
out.append('%d,%d,%d\n' % (start, end, c))
487
def annotate(self, knit, version_id):
488
annotator = _KnitAnnotator(knit)
489
return annotator.annotate(version_id)
492
def make_empty_knit(transport, relpath):
493
"""Construct a empty knit at the specified location."""
494
k = make_file_knit(transport, relpath, 'w', KnitPlainFactory)
497
def make_file_knit(name, transport, file_mode=None, access_mode='w',
498
factory=None, delta=True, create=False, create_parent_dir=False,
499
delay_create=False, dir_mode=None, get_scope=None):
500
"""Factory to create a KnitVersionedFile for a .knit/.kndx file pair."""
502
factory = KnitAnnotateFactory()
504
factory = KnitPlainFactory()
505
if get_scope is None:
506
get_scope = lambda:None
507
index = _KnitIndex(transport, name + INDEX_SUFFIX,
508
access_mode, create=create, file_mode=file_mode,
509
create_parent_dir=create_parent_dir, delay_create=delay_create,
510
dir_mode=dir_mode, get_scope=get_scope)
511
access = _KnitAccess(transport, name + DATA_SUFFIX, file_mode,
512
dir_mode, ((create and not len(index)) and delay_create),
514
return KnitVersionedFile(name, transport, factory=factory,
515
create=create, delay_create=delay_create, index=index,
516
access_method=access)
520
"""Return the suffixes used by file based knits."""
521
return [DATA_SUFFIX, INDEX_SUFFIX]
522
make_file_knit.get_suffixes = get_suffixes
525
class KnitVersionedFile(VersionedFile):
526
"""Weave-like structure with faster random access.
528
A knit stores a number of texts and a summary of the relationships
529
between them. Texts are identified by a string version-id. Texts
530
are normally stored and retrieved as a series of lines, but can
531
also be passed as single strings.
533
Lines are stored with the trailing newline (if any) included, to
534
avoid special cases for files with no final newline. Lines are
535
composed of 8-bit characters, not unicode. The combination of
536
these approaches should mean any 'binary' file can be safely
537
stored and retrieved.
540
def __init__(self, relpath, transport, file_mode=None,
541
factory=None, delta=True, create=False, create_parent_dir=False,
542
delay_create=False, dir_mode=None, index=None, access_method=None):
543
"""Construct a knit at location specified by relpath.
545
:param create: If not True, only open an existing knit.
546
:param create_parent_dir: If True, create the parent directory if
547
creating the file fails. (This is used for stores with
548
hash-prefixes that may not exist yet)
549
:param delay_create: The calling code is aware that the knit won't
550
actually be created until the first data is stored.
551
:param index: An index to use for the knit.
553
super(KnitVersionedFile, self).__init__()
554
self.transport = transport
555
self.filename = relpath
556
self.factory = factory or KnitAnnotateFactory()
559
self._max_delta_chain = 200
561
if None in (access_method, index):
562
raise ValueError("No default access_method or index any more")
564
_access = access_method
565
if create and not len(self) and not delay_create:
567
self._data = _KnitData(_access)
570
return '%s(%s)' % (self.__class__.__name__,
571
self.transport.abspath(self.filename))
573
def _check_should_delta(self, first_parents):
574
"""Iterate back through the parent listing, looking for a fulltext.
576
This is used when we want to decide whether to add a delta or a new
577
fulltext. It searches for _max_delta_chain parents. When it finds a
578
fulltext parent, it sees if the total size of the deltas leading up to
579
it is large enough to indicate that we want a new full text anyway.
581
Return True if we should create a new delta, False if we should use a
586
delta_parents = first_parents
587
for count in xrange(self._max_delta_chain):
588
parent = delta_parents[0]
589
method = self._index.get_method(parent)
590
index, pos, size = self._index.get_position(parent)
591
if method == 'fulltext':
595
delta_parents = self._index.get_parent_map([parent])[parent]
597
# We couldn't find a fulltext, so we must create a new one
600
return fulltext_size > delta_size
602
def _check_write_ok(self):
603
return self._index._check_write_ok()
605
def _add_raw_records(self, records, data):
606
"""Add all the records 'records' with data pre-joined in 'data'.
608
:param records: A list of tuples(version_id, options, parents, size).
609
:param data: The data for the records. When it is written, the records
610
are adjusted to have pos pointing into data by the sum of
611
the preceding records sizes.
614
raw_record_sizes = [record[3] for record in records]
615
positions = self._data.add_raw_records(raw_record_sizes, data)
618
for (version_id, options, parents, size), access_memo in zip(
620
index_entries.append((version_id, options, access_memo, parents))
622
self._index.add_versions(index_entries)
624
def copy_to(self, name, transport):
625
"""See VersionedFile.copy_to()."""
626
# copy the current index to a temp index to avoid racing with local
628
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
629
self.transport.get(self._index._filename))
631
f = self._data._open_file()
633
transport.put_file(name + DATA_SUFFIX, f)
636
# move the copied index into place
637
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
639
def get_data_stream(self, required_versions):
640
"""Get a data stream for the specified versions.
642
Versions may be returned in any order, not necessarily the order
643
specified. They are returned in a partial order by compression
644
parent, so that the deltas can be applied as the data stream is
645
inserted; however note that compression parents will not be sent
646
unless they were specifically requested, as the client may already
649
:param required_versions: The exact set of versions to be extracted.
650
Unlike some other knit methods, this is not used to generate a
651
transitive closure, rather it is used precisely as given.
653
:returns: format_signature, list of (version, options, length, parents),
656
required_version_set = frozenset(required_versions)
658
# list of revisions that can just be sent without waiting for their
661
# map from revision to the children based on it
663
# first, read all relevant index data, enough to sort into the right
665
for version_id in required_versions:
666
options = self._index.get_options(version_id)
667
parents = self._index.get_parents_with_ghosts(version_id)
668
index_memo = self._index.get_position(version_id)
669
version_index[version_id] = (index_memo, options, parents)
670
if ('line-delta' in options
671
and parents[0] in required_version_set):
672
# must wait until the parent has been sent
673
deferred.setdefault(parents[0], []). \
676
# either a fulltext, or a delta whose parent the client did
677
# not ask for and presumably already has
678
ready_to_send.append(version_id)
679
# build a list of results to return, plus instructions for data to
681
copy_queue_records = []
682
temp_version_list = []
684
# XXX: pushing and popping lists may be a bit inefficient
685
version_id = ready_to_send.pop(0)
686
(index_memo, options, parents) = version_index[version_id]
687
copy_queue_records.append((version_id, index_memo))
688
none, data_pos, data_size = index_memo
689
temp_version_list.append((version_id, options, data_size,
691
if version_id in deferred:
692
# now we can send all the children of this revision - we could
693
# put them in anywhere, but we hope that sending them soon
694
# after the fulltext will give good locality in the receiver
695
ready_to_send[:0] = deferred.pop(version_id)
696
assert len(deferred) == 0, \
697
"Still have compressed child versions waiting to be sent"
698
# XXX: The stream format is such that we cannot stream it - we have to
699
# know the length of all the data a-priori.
701
result_version_list = []
702
for (version_id, raw_data), \
703
(version_id2, options, _, parents) in \
704
izip(self._data.read_records_iter_raw(copy_queue_records),
706
assert version_id == version_id2, \
707
'logic error, inconsistent results'
708
raw_datum.append(raw_data)
709
result_version_list.append(
710
(version_id, options, len(raw_data), parents))
711
# provide a callback to get data incrementally.
712
pseudo_file = StringIO(''.join(raw_datum))
715
return pseudo_file.read()
717
return pseudo_file.read(length)
718
return (self.get_format_signature(), result_version_list, read)
720
def _extract_blocks(self, version_id, source, target):
721
if self._index.get_method(version_id) != 'line-delta':
723
parent, sha1, noeol, delta = self.get_delta(version_id)
724
return KnitContent.get_line_delta_blocks(delta, source, target)
726
def get_delta(self, version_id):
727
"""Get a delta for constructing version from some other version."""
728
self.check_not_reserved_id(version_id)
729
parents = self.get_parent_map([version_id])[version_id]
734
index_memo = self._index.get_position(version_id)
735
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
736
noeol = 'no-eol' in self._index.get_options(version_id)
737
if 'fulltext' == self._index.get_method(version_id):
738
new_content = self.factory.parse_fulltext(data, version_id)
739
if parent is not None:
740
reference_content = self._get_content(parent)
741
old_texts = reference_content.text()
744
new_texts = new_content.text()
745
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
747
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
749
delta = self.factory.parse_line_delta(data, version_id)
750
return parent, sha1, noeol, delta
752
def get_format_signature(self):
753
"""See VersionedFile.get_format_signature()."""
754
if self.factory.annotated:
755
annotated_part = "annotated"
757
annotated_part = "plain"
758
return "knit-%s" % (annotated_part,)
760
@deprecated_method(one_four)
761
def get_graph_with_ghosts(self):
762
"""See VersionedFile.get_graph_with_ghosts()."""
763
return self.get_parent_map(self.versions())
765
def get_sha1s(self, version_ids):
766
"""See VersionedFile.get_sha1s()."""
767
record_map = self._get_record_map(version_ids)
768
# record entry 2 is the 'digest'.
769
return [record_map[v][2] for v in version_ids]
771
@deprecated_method(one_four)
772
def has_ghost(self, version_id):
773
"""True if there is a ghost reference in the file to version_id."""
775
if self.has_version(version_id):
777
# optimisable if needed by memoising the _ghosts set.
778
items = self.get_parent_map(self.versions())
779
for parents in items.itervalues():
780
for parent in parents:
781
if parent == version_id and parent not in items:
785
def insert_data_stream(self, (format, data_list, reader_callable)):
786
"""Insert knit records from a data stream into this knit.
788
If a version in the stream is already present in this knit, it will not
789
be inserted a second time. It will be checked for consistency with the
790
stored version however, and may cause a KnitCorrupt error to be raised
791
if the data in the stream disagrees with the already stored data.
793
:seealso: get_data_stream
795
if format != self.get_format_signature():
796
if 'knit' in debug.debug_flags:
798
'incompatible format signature inserting to %r', self)
799
source = self._knit_from_datastream(
800
(format, data_list, reader_callable))
804
for version_id, options, length, parents in data_list:
805
if self.has_version(version_id):
806
# First check: the list of parents.
807
my_parents = self.get_parents_with_ghosts(version_id)
808
if tuple(my_parents) != tuple(parents):
809
# XXX: KnitCorrupt is not quite the right exception here.
812
'parents list %r from data stream does not match '
813
'already recorded parents %r for %s'
814
% (parents, my_parents, version_id))
816
# Also check the SHA-1 of the fulltext this content will
818
raw_data = reader_callable(length)
819
my_fulltext_sha1 = self.get_sha1s([version_id])[0]
820
df, rec = self._data._parse_record_header(version_id, raw_data)
821
stream_fulltext_sha1 = rec[3]
822
if my_fulltext_sha1 != stream_fulltext_sha1:
823
# Actually, we don't know if it's this knit that's corrupt,
824
# or the data stream we're trying to insert.
826
self.filename, 'sha-1 does not match %s' % version_id)
828
if 'line-delta' in options:
829
# Make sure that this knit record is actually useful: a
830
# line-delta is no use unless we have its parent.
831
# Fetching from a broken repository with this problem
832
# shouldn't break the target repository.
834
# See https://bugs.launchpad.net/bzr/+bug/164443
835
if not self._index.has_version(parents[0]):
838
'line-delta from stream '
841
'missing parent %s\n'
842
'Try running "bzr check" '
843
'on the source repository, and "bzr reconcile" '
845
(version_id, parents[0]))
847
# We received a line-delta record for a non-delta knit.
848
# Convert it to a fulltext.
849
gzip_bytes = reader_callable(length)
850
self._convert_line_delta_to_fulltext(
851
gzip_bytes, version_id, parents)
854
self._add_raw_records(
855
[(version_id, options, parents, length)],
856
reader_callable(length))
858
def _convert_line_delta_to_fulltext(self, gzip_bytes, version_id, parents):
859
lines, sha1 = self._data._parse_record(version_id, gzip_bytes)
860
delta = self.factory.parse_line_delta(lines, version_id)
861
content = self.factory.make(self.get_lines(parents[0]), parents[0])
862
content.apply_delta(delta, version_id)
863
digest, len, content = self.add_lines(
864
version_id, parents, content.text())
866
raise errors.VersionedFileInvalidChecksum(version_id)
868
def _knit_from_datastream(self, (format, data_list, reader_callable)):
869
"""Create a knit object from a data stream.
871
This method exists to allow conversion of data streams that do not
872
match the signature of this knit. Generally it will be slower and use
873
more memory to use this method to insert data, but it will work.
875
:seealso: get_data_stream for details on datastreams.
876
:return: A knit versioned file which can be used to join the datastream
879
if format == "knit-plain":
880
factory = KnitPlainFactory()
881
elif format == "knit-annotated":
882
factory = KnitAnnotateFactory()
884
raise errors.KnitDataStreamUnknown(format)
885
index = _StreamIndex(data_list, self._index)
886
access = _StreamAccess(reader_callable, index, self, factory)
887
return KnitVersionedFile(self.filename, self.transport,
888
factory=factory, index=index, access_method=access)
891
"""See VersionedFile.versions."""
892
if 'evil' in debug.debug_flags:
893
trace.mutter_callsite(2, "versions scales with size of history")
894
return self._index.get_versions()
896
def has_version(self, version_id):
897
"""See VersionedFile.has_version."""
898
if 'evil' in debug.debug_flags:
899
trace.mutter_callsite(2, "has_version is a LBYL scenario")
900
return self._index.has_version(version_id)
902
__contains__ = has_version
904
def _merge_annotations(self, content, parents, parent_texts={},
905
delta=None, annotated=None,
906
left_matching_blocks=None):
907
"""Merge annotations for content. This is done by comparing
908
the annotations based on changed to the text.
910
if left_matching_blocks is not None:
911
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
915
for parent_id in parents:
916
merge_content = self._get_content(parent_id, parent_texts)
917
if (parent_id == parents[0] and delta_seq is not None):
920
seq = patiencediff.PatienceSequenceMatcher(
921
None, merge_content.text(), content.text())
922
for i, j, n in seq.get_matching_blocks():
925
# this appears to copy (origin, text) pairs across to the
926
# new content for any line that matches the last-checked
928
content._lines[j:j+n] = merge_content._lines[i:i+n]
930
if delta_seq is None:
931
reference_content = self._get_content(parents[0], parent_texts)
932
new_texts = content.text()
933
old_texts = reference_content.text()
934
delta_seq = patiencediff.PatienceSequenceMatcher(
935
None, old_texts, new_texts)
936
return self._make_line_delta(delta_seq, content)
938
def _make_line_delta(self, delta_seq, new_content):
939
"""Generate a line delta from delta_seq and new_content."""
941
for op in delta_seq.get_opcodes():
944
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
947
def _get_components_positions(self, version_ids):
948
"""Produce a map of position data for the components of versions.
950
This data is intended to be used for retrieving the knit records.
952
A dict of version_id to (record_details, index_memo, next, parents) is
954
method is the way referenced data should be applied.
955
index_memo is the handle to pass to the data access to actually get the
957
next is the build-parent of the version, or None for fulltexts.
958
parents is the version_ids of the parents of this version
961
pending_components = version_ids
962
while pending_components:
963
build_details = self._index.get_build_details(pending_components)
964
current_components = set(pending_components)
965
pending_components = set()
966
for version_id, details in build_details.iteritems():
967
(index_memo, compression_parent, parents,
968
record_details) = details
969
method = record_details[0]
970
if compression_parent is not None:
971
pending_components.add(compression_parent)
972
component_data[version_id] = (record_details, index_memo,
974
missing = current_components.difference(build_details)
976
raise errors.RevisionNotPresent(missing.pop(), self.filename)
977
return component_data
979
def _get_content(self, version_id, parent_texts={}):
980
"""Returns a content object that makes up the specified
982
cached_version = parent_texts.get(version_id, None)
983
if cached_version is not None:
984
if not self.has_version(version_id):
985
raise RevisionNotPresent(version_id, self.filename)
986
return cached_version
988
text_map, contents_map = self._get_content_maps([version_id])
989
return contents_map[version_id]
991
def _check_versions_present(self, version_ids):
992
"""Check that all specified versions are present."""
993
self._index.check_versions_present(version_ids)
995
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
996
nostore_sha, random_id, check_content, left_matching_blocks):
997
"""See VersionedFile.add_lines_with_ghosts()."""
998
self._check_add(version_id, lines, random_id, check_content)
999
return self._add(version_id, lines, parents, self.delta,
1000
parent_texts, left_matching_blocks, nostore_sha, random_id)
1002
def _add_lines(self, version_id, parents, lines, parent_texts,
1003
left_matching_blocks, nostore_sha, random_id, check_content):
1004
"""See VersionedFile.add_lines."""
1005
self._check_add(version_id, lines, random_id, check_content)
1006
self._check_versions_present(parents)
1007
return self._add(version_id, lines[:], parents, self.delta,
1008
parent_texts, left_matching_blocks, nostore_sha, random_id)
1010
def _check_add(self, version_id, lines, random_id, check_content):
1011
"""check that version_id and lines are safe to add."""
1012
if contains_whitespace(version_id):
1013
raise InvalidRevisionId(version_id, self.filename)
1014
self.check_not_reserved_id(version_id)
1015
# Technically this could be avoided if we are happy to allow duplicate
1016
# id insertion when other things than bzr core insert texts, but it
1017
# seems useful for folk using the knit api directly to have some safety
1018
# blanket that we can disable.
1019
if not random_id and self.has_version(version_id):
1020
raise RevisionAlreadyPresent(version_id, self.filename)
1022
self._check_lines_not_unicode(lines)
1023
self._check_lines_are_lines(lines)
1025
def _add(self, version_id, lines, parents, delta, parent_texts,
1026
left_matching_blocks, nostore_sha, random_id):
1027
"""Add a set of lines on top of version specified by parents.
1029
If delta is true, compress the text as a line-delta against
1032
Any versions not present will be converted into ghosts.
1034
# first thing, if the content is something we don't need to store, find
1036
line_bytes = ''.join(lines)
1037
digest = sha_string(line_bytes)
1038
if nostore_sha == digest:
1039
raise errors.ExistingContent
1041
present_parents = []
1042
if parent_texts is None:
1044
for parent in parents:
1045
if self.has_version(parent):
1046
present_parents.append(parent)
1048
# can only compress against the left most present parent.
1050
(len(present_parents) == 0 or
1051
present_parents[0] != parents[0])):
1054
text_length = len(line_bytes)
1057
if lines[-1][-1] != '\n':
1058
# copy the contents of lines.
1060
options.append('no-eol')
1061
lines[-1] = lines[-1] + '\n'
1065
# To speed the extract of texts the delta chain is limited
1066
# to a fixed number of deltas. This should minimize both
1067
# I/O and the time spend applying deltas.
1068
delta = self._check_should_delta(present_parents)
1070
assert isinstance(version_id, str)
1071
content = self.factory.make(lines, version_id)
1072
if delta or (self.factory.annotated and len(present_parents) > 0):
1073
# Merge annotations from parent texts if needed.
1074
delta_hunks = self._merge_annotations(content, present_parents,
1075
parent_texts, delta, self.factory.annotated,
1076
left_matching_blocks)
1079
options.append('line-delta')
1080
store_lines = self.factory.lower_line_delta(delta_hunks)
1081
size, bytes = self._data._record_to_data(version_id, digest,
1084
options.append('fulltext')
1085
# isinstance is slower and we have no hierarchy.
1086
if self.factory.__class__ == KnitPlainFactory:
1087
# Use the already joined bytes saving iteration time in
1089
size, bytes = self._data._record_to_data(version_id, digest,
1090
lines, [line_bytes])
1092
# get mixed annotation + content and feed it into the
1094
store_lines = self.factory.lower_fulltext(content)
1095
size, bytes = self._data._record_to_data(version_id, digest,
1098
access_memo = self._data.add_raw_records([size], bytes)[0]
1099
self._index.add_versions(
1100
((version_id, options, access_memo, parents),),
1101
random_id=random_id)
1102
return digest, text_length, content
1104
def check(self, progress_bar=None):
1105
"""See VersionedFile.check()."""
1107
def get_lines(self, version_id):
1108
"""See VersionedFile.get_lines()."""
1109
return self.get_line_list([version_id])[0]
1111
def _get_record_map(self, version_ids):
1112
"""Produce a dictionary of knit records.
1114
:return: {version_id:(record, record_details, digest, next)}
1116
data returned from read_records
1118
opaque information to pass to parse_record
1120
SHA1 digest of the full text after all steps are done
1122
build-parent of the version, i.e. the leftmost ancestor.
1123
Will be None if the record is not a delta.
1125
position_map = self._get_components_positions(version_ids)
1126
# c = component_id, r = record_details, i_m = index_memo, n = next
1127
records = [(c, i_m) for c, (r, i_m, n)
1128
in position_map.iteritems()]
1130
for component_id, record, digest in \
1131
self._data.read_records_iter(records):
1132
(record_details, index_memo, next) = position_map[component_id]
1133
record_map[component_id] = record, record_details, digest, next
1137
def get_text(self, version_id):
1138
"""See VersionedFile.get_text"""
1139
return self.get_texts([version_id])[0]
1141
def get_texts(self, version_ids):
1142
return [''.join(l) for l in self.get_line_list(version_ids)]
1144
def get_line_list(self, version_ids):
1145
"""Return the texts of listed versions as a list of strings."""
1146
for version_id in version_ids:
1147
self.check_not_reserved_id(version_id)
1148
text_map, content_map = self._get_content_maps(version_ids)
1149
return [text_map[v] for v in version_ids]
1151
_get_lf_split_line_list = get_line_list
1153
def _get_content_maps(self, version_ids):
1154
"""Produce maps of text and KnitContents
1156
:return: (text_map, content_map) where text_map contains the texts for
1157
the requested versions and content_map contains the KnitContents.
1158
Both dicts take version_ids as their keys.
1160
# FUTURE: This function could be improved for the 'extract many' case
1161
# by tracking each component and only doing the copy when the number of
1162
# children than need to apply delta's to it is > 1 or it is part of the
1164
version_ids = list(version_ids)
1165
multiple_versions = len(version_ids) != 1
1166
record_map = self._get_record_map(version_ids)
1171
for version_id in version_ids:
1174
while cursor is not None:
1175
record, record_details, digest, next = record_map[cursor]
1176
components.append((cursor, record, record_details, digest))
1177
if cursor in content_map:
1182
for (component_id, record, record_details,
1183
digest) in reversed(components):
1184
if component_id in content_map:
1185
content = content_map[component_id]
1187
content, delta = self.factory.parse_record(version_id,
1188
record, record_details, content,
1189
copy_base_content=multiple_versions)
1190
if multiple_versions:
1191
content_map[component_id] = content
1193
content.cleanup_eol(copy_on_mutate=multiple_versions)
1194
final_content[version_id] = content
1196
# digest here is the digest from the last applied component.
1197
text = content.text()
1198
actual_sha = sha_strings(text)
1199
if actual_sha != digest:
1200
raise KnitCorrupt(self.filename,
1202
'\n of reconstructed text does not match'
1204
'\n for version %s' %
1205
(actual_sha, digest, version_id))
1206
text_map[version_id] = text
1207
return text_map, final_content
1209
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1211
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1212
if version_ids is None:
1213
version_ids = self.versions()
1215
pb = progress.DummyProgress()
1216
# we don't care about inclusions, the caller cares.
1217
# but we need to setup a list of records to visit.
1218
# we need version_id, position, length
1219
version_id_records = []
1220
requested_versions = set(version_ids)
1221
# filter for available versions
1222
for version_id in requested_versions:
1223
if not self.has_version(version_id):
1224
raise RevisionNotPresent(version_id, self.filename)
1225
# get a in-component-order queue:
1226
for version_id in self.versions():
1227
if version_id in requested_versions:
1228
index_memo = self._index.get_position(version_id)
1229
version_id_records.append((version_id, index_memo))
1231
total = len(version_id_records)
1232
for version_idx, (version_id, data, sha_value) in \
1233
enumerate(self._data.read_records_iter(version_id_records)):
1234
pb.update('Walking content.', version_idx, total)
1235
method = self._index.get_method(version_id)
1237
assert method in ('fulltext', 'line-delta')
1238
if method == 'fulltext':
1239
line_iterator = self.factory.get_fulltext_content(data)
1241
line_iterator = self.factory.get_linedelta_content(data)
1242
# XXX: It might be more efficient to yield (version_id,
1243
# line_iterator) in the future. However for now, this is a simpler
1244
# change to integrate into the rest of the codebase. RBC 20071110
1245
for line in line_iterator:
1246
yield line, version_id
1248
pb.update('Walking content.', total, total)
1250
def num_versions(self):
1251
"""See VersionedFile.num_versions()."""
1252
return self._index.num_versions()
1254
__len__ = num_versions
1256
def annotate(self, version_id):
1257
"""See VersionedFile.annotate."""
1258
return self.factory.annotate(self, version_id)
1260
def get_parent_map(self, version_ids):
1261
"""See VersionedFile.get_parent_map."""
1262
return self._index.get_parent_map(version_ids)
1264
def get_ancestry(self, versions, topo_sorted=True):
1265
"""See VersionedFile.get_ancestry."""
1266
if isinstance(versions, basestring):
1267
versions = [versions]
1270
return self._index.get_ancestry(versions, topo_sorted)
1272
def get_ancestry_with_ghosts(self, versions):
1273
"""See VersionedFile.get_ancestry_with_ghosts."""
1274
if isinstance(versions, basestring):
1275
versions = [versions]
1278
return self._index.get_ancestry_with_ghosts(versions)
1280
def plan_merge(self, ver_a, ver_b):
1281
"""See VersionedFile.plan_merge."""
1282
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1283
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1284
annotated_a = self.annotate(ver_a)
1285
annotated_b = self.annotate(ver_b)
1286
return merge._plan_annotate_merge(annotated_a, annotated_b,
1287
ancestors_a, ancestors_b)
1290
class _KnitComponentFile(object):
1291
"""One of the files used to implement a knit database"""
1293
def __init__(self, transport, filename, mode, file_mode=None,
1294
create_parent_dir=False, dir_mode=None):
1295
self._transport = transport
1296
self._filename = filename
1298
self._file_mode = file_mode
1299
self._dir_mode = dir_mode
1300
self._create_parent_dir = create_parent_dir
1301
self._need_to_create = False
1303
def _full_path(self):
1304
"""Return the full path to this file."""
1305
return self._transport.base + self._filename
1307
def check_header(self, fp):
1308
line = fp.readline()
1310
# An empty file can actually be treated as though the file doesn't
1312
raise errors.NoSuchFile(self._full_path())
1313
if line != self.HEADER:
1314
raise KnitHeaderError(badline=line,
1315
filename=self._transport.abspath(self._filename))
1318
return '%s(%s)' % (self.__class__.__name__, self._filename)
1321
class _KnitIndex(_KnitComponentFile):
1322
"""Manages knit index file.
1324
The index is already kept in memory and read on startup, to enable
1325
fast lookups of revision information. The cursor of the index
1326
file is always pointing to the end, making it easy to append
1329
_cache is a cache for fast mapping from version id to a Index
1332
_history is a cache for fast mapping from indexes to version ids.
1334
The index data format is dictionary compressed when it comes to
1335
parent references; a index entry may only have parents that with a
1336
lover index number. As a result, the index is topological sorted.
1338
Duplicate entries may be written to the index for a single version id
1339
if this is done then the latter one completely replaces the former:
1340
this allows updates to correct version and parent information.
1341
Note that the two entries may share the delta, and that successive
1342
annotations and references MUST point to the first entry.
1344
The index file on disc contains a header, followed by one line per knit
1345
record. The same revision can be present in an index file more than once.
1346
The first occurrence gets assigned a sequence number starting from 0.
1348
The format of a single line is
1349
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1350
REVISION_ID is a utf8-encoded revision id
1351
FLAGS is a comma separated list of flags about the record. Values include
1352
no-eol, line-delta, fulltext.
1353
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1354
that the the compressed data starts at.
1355
LENGTH is the ascii representation of the length of the data file.
1356
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1358
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1359
revision id already in the knit that is a parent of REVISION_ID.
1360
The ' :' marker is the end of record marker.
1363
when a write is interrupted to the index file, it will result in a line
1364
that does not end in ' :'. If the ' :' is not present at the end of a line,
1365
or at the end of the file, then the record that is missing it will be
1366
ignored by the parser.
1368
When writing new records to the index file, the data is preceded by '\n'
1369
to ensure that records always start on new lines even if the last write was
1370
interrupted. As a result its normal for the last line in the index to be
1371
missing a trailing newline. One can be added with no harmful effects.
1374
HEADER = "# bzr knit index 8\n"
1376
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1377
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1379
def _cache_version(self, version_id, options, pos, size, parents):
1380
"""Cache a version record in the history array and index cache.
1382
This is inlined into _load_data for performance. KEEP IN SYNC.
1383
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1386
# only want the _history index to reference the 1st index entry
1388
if version_id not in self._cache:
1389
index = len(self._history)
1390
self._history.append(version_id)
1392
index = self._cache[version_id][5]
1393
self._cache[version_id] = (version_id,
1400
def _check_write_ok(self):
1401
if self._get_scope() != self._scope:
1402
raise errors.OutSideTransaction()
1403
if self._mode != 'w':
1404
raise errors.ReadOnlyObjectDirtiedError(self)
1406
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1407
create_parent_dir=False, delay_create=False, dir_mode=None,
1409
_KnitComponentFile.__init__(self, transport, filename, mode,
1410
file_mode=file_mode,
1411
create_parent_dir=create_parent_dir,
1414
# position in _history is the 'official' index for a revision
1415
# but the values may have come from a newer entry.
1416
# so - wc -l of a knit index is != the number of unique names
1420
fp = self._transport.get(self._filename)
1422
# _load_data may raise NoSuchFile if the target knit is
1424
_load_data(self, fp)
1428
if mode != 'w' or not create:
1431
self._need_to_create = True
1433
self._transport.put_bytes_non_atomic(
1434
self._filename, self.HEADER, mode=self._file_mode)
1435
self._scope = get_scope()
1436
self._get_scope = get_scope
1438
def get_ancestry(self, versions, topo_sorted=True):
1439
"""See VersionedFile.get_ancestry."""
1440
# get a graph of all the mentioned versions:
1442
pending = set(versions)
1445
version = pending.pop()
1448
parents = [p for p in cache[version][4] if p in cache]
1450
raise RevisionNotPresent(version, self._filename)
1451
# if not completed and not a ghost
1452
pending.update([p for p in parents if p not in graph])
1453
graph[version] = parents
1456
return topo_sort(graph.items())
1458
def get_ancestry_with_ghosts(self, versions):
1459
"""See VersionedFile.get_ancestry_with_ghosts."""
1460
# get a graph of all the mentioned versions:
1461
self.check_versions_present(versions)
1464
pending = set(versions)
1466
version = pending.pop()
1468
parents = cache[version][4]
1474
pending.update([p for p in parents if p not in graph])
1475
graph[version] = parents
1476
return topo_sort(graph.items())
1478
def get_build_details(self, version_ids):
1479
"""Get the method, index_memo and compression parent for version_ids.
1481
Ghosts are omitted from the result.
1483
:param version_ids: An iterable of version_ids.
1484
:return: A dict of version_id:(index_memo, compression_parent,
1485
parents, record_details).
1487
opaque structure to pass to read_records to extract the raw
1490
Content that this record is built upon, may be None
1492
Logical parents of this node
1494
extra information about the content which needs to be passed to
1495
Factory.parse_record
1498
for version_id in version_ids:
1499
if version_id not in self._cache:
1500
# ghosts are omitted
1502
method = self.get_method(version_id)
1503
parents = self.get_parents_with_ghosts(version_id)
1504
if method == 'fulltext':
1505
compression_parent = None
1507
compression_parent = parents[0]
1508
noeol = 'no-eol' in self.get_options(version_id)
1509
index_memo = self.get_position(version_id)
1510
result[version_id] = (index_memo, compression_parent,
1511
parents, (method, noeol))
1514
def num_versions(self):
1515
return len(self._history)
1517
__len__ = num_versions
1519
def get_versions(self):
1520
"""Get all the versions in the file. not topologically sorted."""
1521
return self._history
1523
def _version_list_to_index(self, versions):
1526
for version in versions:
1527
if version in cache:
1528
# -- inlined lookup() --
1529
result_list.append(str(cache[version][5]))
1530
# -- end lookup () --
1532
result_list.append('.' + version)
1533
return ' '.join(result_list)
1535
def add_version(self, version_id, options, index_memo, parents):
1536
"""Add a version record to the index."""
1537
self.add_versions(((version_id, options, index_memo, parents),))
1539
def add_versions(self, versions, random_id=False):
1540
"""Add multiple versions to the index.
1542
:param versions: a list of tuples:
1543
(version_id, options, pos, size, parents).
1544
:param random_id: If True the ids being added were randomly generated
1545
and no check for existence will be performed.
1548
orig_history = self._history[:]
1549
orig_cache = self._cache.copy()
1552
for version_id, options, (index, pos, size), parents in versions:
1553
line = "\n%s %s %s %s %s :" % (version_id,
1557
self._version_list_to_index(parents))
1558
assert isinstance(line, str), \
1559
'content must be utf-8 encoded: %r' % (line,)
1561
self._cache_version(version_id, options, pos, size, tuple(parents))
1562
if not self._need_to_create:
1563
self._transport.append_bytes(self._filename, ''.join(lines))
1566
sio.write(self.HEADER)
1567
sio.writelines(lines)
1569
self._transport.put_file_non_atomic(self._filename, sio,
1570
create_parent_dir=self._create_parent_dir,
1571
mode=self._file_mode,
1572
dir_mode=self._dir_mode)
1573
self._need_to_create = False
1575
# If any problems happen, restore the original values and re-raise
1576
self._history = orig_history
1577
self._cache = orig_cache
1580
def has_version(self, version_id):
1581
"""True if the version is in the index."""
1582
return version_id in self._cache
1584
def get_position(self, version_id):
1585
"""Return details needed to access the version.
1587
.kndx indices do not support split-out data, so return None for the
1590
:return: a tuple (None, data position, size) to hand to the access
1591
logic to get the record.
1593
entry = self._cache[version_id]
1594
return None, entry[2], entry[3]
1596
def get_method(self, version_id):
1597
"""Return compression method of specified version."""
1599
options = self._cache[version_id][1]
1601
raise RevisionNotPresent(version_id, self._filename)
1602
if 'fulltext' in options:
1605
if 'line-delta' not in options:
1606
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1609
def get_options(self, version_id):
1610
"""Return a list representing options.
1614
return self._cache[version_id][1]
1616
def get_parent_map(self, version_ids):
1617
"""Passed through to by KnitVersionedFile.get_parent_map."""
1619
for version_id in version_ids:
1621
result[version_id] = tuple(self._cache[version_id][4])
1626
def get_parents_with_ghosts(self, version_id):
1627
"""Return parents of specified version with ghosts."""
1629
return self.get_parent_map([version_id])[version_id]
1631
raise RevisionNotPresent(version_id, self)
1633
def check_versions_present(self, version_ids):
1634
"""Check that all specified versions are present."""
1636
for version_id in version_ids:
1637
if version_id not in cache:
1638
raise RevisionNotPresent(version_id, self._filename)
1641
class KnitGraphIndex(object):
1642
"""A knit index that builds on GraphIndex."""
1644
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1645
"""Construct a KnitGraphIndex on a graph_index.
1647
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1648
:param deltas: Allow delta-compressed records.
1649
:param add_callback: If not None, allow additions to the index and call
1650
this callback with a list of added GraphIndex nodes:
1651
[(node, value, node_refs), ...]
1652
:param parents: If True, record knits parents, if not do not record
1655
self._graph_index = graph_index
1656
self._deltas = deltas
1657
self._add_callback = add_callback
1658
self._parents = parents
1659
if deltas and not parents:
1660
raise KnitCorrupt(self, "Cannot do delta compression without "
1663
def _check_write_ok(self):
1666
def _get_entries(self, keys, check_present=False):
1667
"""Get the entries for keys.
1669
:param keys: An iterable of index keys, - 1-tuples.
1674
for node in self._graph_index.iter_entries(keys):
1676
found_keys.add(node[1])
1678
# adapt parentless index to the rest of the code.
1679
for node in self._graph_index.iter_entries(keys):
1680
yield node[0], node[1], node[2], ()
1681
found_keys.add(node[1])
1683
missing_keys = keys.difference(found_keys)
1685
raise RevisionNotPresent(missing_keys.pop(), self)
1687
def _present_keys(self, version_ids):
1689
node[1] for node in self._get_entries(version_ids)])
1691
def _parentless_ancestry(self, versions):
1692
"""Honour the get_ancestry API for parentless knit indices."""
1693
wanted_keys = self._version_ids_to_keys(versions)
1694
present_keys = self._present_keys(wanted_keys)
1695
missing = set(wanted_keys).difference(present_keys)
1697
raise RevisionNotPresent(missing.pop(), self)
1698
return list(self._keys_to_version_ids(present_keys))
1700
def get_ancestry(self, versions, topo_sorted=True):
1701
"""See VersionedFile.get_ancestry."""
1702
if not self._parents:
1703
return self._parentless_ancestry(versions)
1704
# XXX: This will do len(history) index calls - perhaps
1705
# it should be altered to be a index core feature?
1706
# get a graph of all the mentioned versions:
1709
versions = self._version_ids_to_keys(versions)
1710
pending = set(versions)
1712
# get all pending nodes
1713
this_iteration = pending
1714
new_nodes = self._get_entries(this_iteration)
1717
for (index, key, value, node_refs) in new_nodes:
1718
# dont ask for ghosties - otherwise
1719
# we we can end up looping with pending
1720
# being entirely ghosted.
1721
graph[key] = [parent for parent in node_refs[0]
1722
if parent not in ghosts]
1724
for parent in graph[key]:
1725
# dont examine known nodes again
1730
ghosts.update(this_iteration.difference(found))
1731
if versions.difference(graph):
1732
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1734
result_keys = topo_sort(graph.items())
1736
result_keys = graph.iterkeys()
1737
return [key[0] for key in result_keys]
1739
def get_ancestry_with_ghosts(self, versions):
1740
"""See VersionedFile.get_ancestry."""
1741
if not self._parents:
1742
return self._parentless_ancestry(versions)
1743
# XXX: This will do len(history) index calls - perhaps
1744
# it should be altered to be a index core feature?
1745
# get a graph of all the mentioned versions:
1747
versions = self._version_ids_to_keys(versions)
1748
pending = set(versions)
1750
# get all pending nodes
1751
this_iteration = pending
1752
new_nodes = self._get_entries(this_iteration)
1754
for (index, key, value, node_refs) in new_nodes:
1755
graph[key] = node_refs[0]
1757
for parent in graph[key]:
1758
# dont examine known nodes again
1762
missing_versions = this_iteration.difference(graph)
1763
missing_needed = versions.intersection(missing_versions)
1765
raise RevisionNotPresent(missing_needed.pop(), self)
1766
for missing_version in missing_versions:
1767
# add a key, no parents
1768
graph[missing_version] = []
1769
pending.discard(missing_version) # don't look for it
1770
result_keys = topo_sort(graph.items())
1771
return [key[0] for key in result_keys]
1773
def get_build_details(self, version_ids):
1774
"""Get the method, index_memo and compression parent for version_ids.
1776
Ghosts are omitted from the result.
1778
:param version_ids: An iterable of version_ids.
1779
:return: A dict of version_id:(index_memo, compression_parent,
1780
parents, record_details).
1782
opaque structure to pass to read_records to extract the raw
1785
Content that this record is built upon, may be None
1787
Logical parents of this node
1789
extra information about the content which needs to be passed to
1790
Factory.parse_record
1793
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
1794
for entry in entries:
1795
version_id = self._keys_to_version_ids((entry[1],))[0]
1796
if not self._parents:
1799
parents = self._keys_to_version_ids(entry[3][0])
1800
if not self._deltas:
1801
compression_parent = None
1803
compression_parent_key = self._compression_parent(entry)
1804
if compression_parent_key:
1805
compression_parent = self._keys_to_version_ids(
1806
(compression_parent_key,))[0]
1808
compression_parent = None
1809
noeol = (entry[2][0] == 'N')
1810
if compression_parent:
1811
method = 'line-delta'
1814
result[version_id] = (self._node_to_position(entry),
1815
compression_parent, parents,
1819
def _compression_parent(self, an_entry):
1820
# return the key that an_entry is compressed against, or None
1821
# Grab the second parent list (as deltas implies parents currently)
1822
compression_parents = an_entry[3][1]
1823
if not compression_parents:
1825
assert len(compression_parents) == 1
1826
return compression_parents[0]
1828
def _get_method(self, node):
1829
if not self._deltas:
1831
if self._compression_parent(node):
1836
def num_versions(self):
1837
return len(list(self._graph_index.iter_all_entries()))
1839
__len__ = num_versions
1841
def get_versions(self):
1842
"""Get all the versions in the file. not topologically sorted."""
1843
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1845
def has_version(self, version_id):
1846
"""True if the version is in the index."""
1847
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1849
def _keys_to_version_ids(self, keys):
1850
return tuple(key[0] for key in keys)
1852
def get_position(self, version_id):
1853
"""Return details needed to access the version.
1855
:return: a tuple (index, data position, size) to hand to the access
1856
logic to get the record.
1858
node = self._get_node(version_id)
1859
return self._node_to_position(node)
1861
def _node_to_position(self, node):
1862
"""Convert an index value to position details."""
1863
bits = node[2][1:].split(' ')
1864
return node[0], int(bits[0]), int(bits[1])
1866
def get_method(self, version_id):
1867
"""Return compression method of specified version."""
1868
return self._get_method(self._get_node(version_id))
1870
def _get_node(self, version_id):
1872
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1874
raise RevisionNotPresent(version_id, self)
1876
def get_options(self, version_id):
1877
"""Return a list representing options.
1881
node = self._get_node(version_id)
1882
options = [self._get_method(node)]
1883
if node[2][0] == 'N':
1884
options.append('no-eol')
1887
def get_parent_map(self, version_ids):
1888
"""Passed through to by KnitVersionedFile.get_parent_map."""
1889
nodes = self._get_entries(self._version_ids_to_keys(version_ids))
1893
result[node[1][0]] = self._keys_to_version_ids(node[3][0])
1896
result[node[1][0]] = ()
1899
def get_parents_with_ghosts(self, version_id):
1900
"""Return parents of specified version with ghosts."""
1902
return self.get_parent_map([version_id])[version_id]
1904
raise RevisionNotPresent(version_id, self)
1906
def check_versions_present(self, version_ids):
1907
"""Check that all specified versions are present."""
1908
keys = self._version_ids_to_keys(version_ids)
1909
present = self._present_keys(keys)
1910
missing = keys.difference(present)
1912
raise RevisionNotPresent(missing.pop(), self)
1914
def add_version(self, version_id, options, access_memo, parents):
1915
"""Add a version record to the index."""
1916
return self.add_versions(((version_id, options, access_memo, parents),))
1918
def add_versions(self, versions, random_id=False):
1919
"""Add multiple versions to the index.
1921
This function does not insert data into the Immutable GraphIndex
1922
backing the KnitGraphIndex, instead it prepares data for insertion by
1923
the caller and checks that it is safe to insert then calls
1924
self._add_callback with the prepared GraphIndex nodes.
1926
:param versions: a list of tuples:
1927
(version_id, options, pos, size, parents).
1928
:param random_id: If True the ids being added were randomly generated
1929
and no check for existence will be performed.
1931
if not self._add_callback:
1932
raise errors.ReadOnlyError(self)
1933
# we hope there are no repositories with inconsistent parentage
1938
for (version_id, options, access_memo, parents) in versions:
1939
index, pos, size = access_memo
1940
key = (version_id, )
1941
parents = tuple((parent, ) for parent in parents)
1942
if 'no-eol' in options:
1946
value += "%d %d" % (pos, size)
1947
if not self._deltas:
1948
if 'line-delta' in options:
1949
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1952
if 'line-delta' in options:
1953
node_refs = (parents, (parents[0],))
1955
node_refs = (parents, ())
1957
node_refs = (parents, )
1960
raise KnitCorrupt(self, "attempt to add node with parents "
1961
"in parentless index.")
1963
keys[key] = (value, node_refs)
1965
present_nodes = self._get_entries(keys)
1966
for (index, key, value, node_refs) in present_nodes:
1967
if (value, node_refs) != keys[key]:
1968
raise KnitCorrupt(self, "inconsistent details in add_versions"
1969
": %s %s" % ((value, node_refs), keys[key]))
1973
for key, (value, node_refs) in keys.iteritems():
1974
result.append((key, value, node_refs))
1976
for key, (value, node_refs) in keys.iteritems():
1977
result.append((key, value))
1978
self._add_callback(result)
1980
def _version_ids_to_keys(self, version_ids):
1981
return set((version_id, ) for version_id in version_ids)
1984
class _KnitAccess(object):
1985
"""Access to knit records in a .knit file."""
1987
def __init__(self, transport, filename, _file_mode, _dir_mode,
1988
_need_to_create, _create_parent_dir):
1989
"""Create a _KnitAccess for accessing and inserting data.
1991
:param transport: The transport the .knit is located on.
1992
:param filename: The filename of the .knit.
1994
self._transport = transport
1995
self._filename = filename
1996
self._file_mode = _file_mode
1997
self._dir_mode = _dir_mode
1998
self._need_to_create = _need_to_create
1999
self._create_parent_dir = _create_parent_dir
2001
def add_raw_records(self, sizes, raw_data):
2002
"""Add raw knit bytes to a storage area.
2004
The data is spooled to whereever the access method is storing data.
2006
:param sizes: An iterable containing the size of each raw data segment.
2007
:param raw_data: A bytestring containing the data.
2008
:return: A list of memos to retrieve the record later. Each memo is a
2009
tuple - (index, pos, length), where the index field is always None
2010
for the .knit access method.
2012
assert type(raw_data) == str, \
2013
'data must be plain bytes was %s' % type(raw_data)
2014
if not self._need_to_create:
2015
base = self._transport.append_bytes(self._filename, raw_data)
2017
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2018
create_parent_dir=self._create_parent_dir,
2019
mode=self._file_mode,
2020
dir_mode=self._dir_mode)
2021
self._need_to_create = False
2025
result.append((None, base, size))
2030
"""IFF this data access has its own storage area, initialise it.
2034
self._transport.put_bytes_non_atomic(self._filename, '',
2035
mode=self._file_mode)
2037
def open_file(self):
2038
"""IFF this data access can be represented as a single file, open it.
2040
For knits that are not mapped to a single file on disk this will
2043
:return: None or a file handle.
2046
return self._transport.get(self._filename)
2051
def get_raw_records(self, memos_for_retrieval):
2052
"""Get the raw bytes for a records.
2054
:param memos_for_retrieval: An iterable containing the (index, pos,
2055
length) memo for retrieving the bytes. The .knit method ignores
2056
the index as there is always only a single file.
2057
:return: An iterator over the bytes of the records.
2059
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2060
for pos, data in self._transport.readv(self._filename, read_vector):
2064
class _PackAccess(object):
2065
"""Access to knit records via a collection of packs."""
2067
def __init__(self, index_to_packs, writer=None):
2068
"""Create a _PackAccess object.
2070
:param index_to_packs: A dict mapping index objects to the transport
2071
and file names for obtaining data.
2072
:param writer: A tuple (pack.ContainerWriter, write_index) which
2073
contains the pack to write, and the index that reads from it will
2077
self.container_writer = writer[0]
2078
self.write_index = writer[1]
2080
self.container_writer = None
2081
self.write_index = None
2082
self.indices = index_to_packs
2084
def add_raw_records(self, sizes, raw_data):
2085
"""Add raw knit bytes to a storage area.
2087
The data is spooled to the container writer in one bytes-record per
2090
:param sizes: An iterable containing the size of each raw data segment.
2091
:param raw_data: A bytestring containing the data.
2092
:return: A list of memos to retrieve the record later. Each memo is a
2093
tuple - (index, pos, length), where the index field is the
2094
write_index object supplied to the PackAccess object.
2096
assert type(raw_data) == str, \
2097
'data must be plain bytes was %s' % type(raw_data)
2101
p_offset, p_length = self.container_writer.add_bytes_record(
2102
raw_data[offset:offset+size], [])
2104
result.append((self.write_index, p_offset, p_length))
2108
"""Pack based knits do not get individually created."""
2110
def get_raw_records(self, memos_for_retrieval):
2111
"""Get the raw bytes for a records.
2113
:param memos_for_retrieval: An iterable containing the (index, pos,
2114
length) memo for retrieving the bytes. The Pack access method
2115
looks up the pack to use for a given record in its index_to_pack
2117
:return: An iterator over the bytes of the records.
2119
# first pass, group into same-index requests
2121
current_index = None
2122
for (index, offset, length) in memos_for_retrieval:
2123
if current_index == index:
2124
current_list.append((offset, length))
2126
if current_index is not None:
2127
request_lists.append((current_index, current_list))
2128
current_index = index
2129
current_list = [(offset, length)]
2130
# handle the last entry
2131
if current_index is not None:
2132
request_lists.append((current_index, current_list))
2133
for index, offsets in request_lists:
2134
transport, path = self.indices[index]
2135
reader = pack.make_readv_reader(transport, path, offsets)
2136
for names, read_func in reader.iter_records():
2137
yield read_func(None)
2139
def open_file(self):
2140
"""Pack based knits have no single file."""
2143
def set_writer(self, writer, index, (transport, packname)):
2144
"""Set a writer to use for adding data."""
2145
if index is not None:
2146
self.indices[index] = (transport, packname)
2147
self.container_writer = writer
2148
self.write_index = index
2151
class _StreamAccess(object):
2152
"""A Knit Access object that provides data from a datastream.
2154
It also provides a fallback to present as unannotated data, annotated data
2155
from a *backing* access object.
2157
This is triggered by a index_memo which is pointing to a different index
2158
than this was constructed with, and is used to allow extracting full
2159
unannotated texts for insertion into annotated knits.
2162
def __init__(self, reader_callable, stream_index, backing_knit,
2164
"""Create a _StreamAccess object.
2166
:param reader_callable: The reader_callable from the datastream.
2167
This is called to buffer all the data immediately, for
2169
:param stream_index: The index the data stream this provides access to
2170
which will be present in native index_memo's.
2171
:param backing_knit: The knit object that will provide access to
2172
annotated texts which are not available in the stream, so as to
2173
create unannotated texts.
2174
:param orig_factory: The original content factory used to generate the
2175
stream. This is used for checking whether the thunk code for
2176
supporting _copy_texts will generate the correct form of data.
2178
self.data = reader_callable(None)
2179
self.stream_index = stream_index
2180
self.backing_knit = backing_knit
2181
self.orig_factory = orig_factory
2183
def get_raw_records(self, memos_for_retrieval):
2184
"""Get the raw bytes for a records.
2186
:param memos_for_retrieval: An iterable of memos from the
2187
_StreamIndex object identifying bytes to read; for these classes
2188
they are (from_backing_knit, index, start, end) and can point to
2189
either the backing knit or streamed data.
2190
:return: An iterator yielding a byte string for each record in
2191
memos_for_retrieval.
2193
# use a generator for memory friendliness
2194
for from_backing_knit, version_id, start, end in memos_for_retrieval:
2195
if not from_backing_knit:
2196
assert version_id is self.stream_index
2197
yield self.data[start:end]
2199
# we have been asked to thunk. This thunking only occurs when
2200
# we are obtaining plain texts from an annotated backing knit
2201
# so that _copy_texts will work.
2202
# We could improve performance here by scanning for where we need
2203
# to do this and using get_line_list, then interleaving the output
2204
# as desired. However, for now, this is sufficient.
2205
if self.orig_factory.__class__ != KnitPlainFactory:
2206
raise errors.KnitCorrupt(
2207
self, 'Bad thunk request %r cannot be backed by %r' %
2208
(version_id, self.orig_factory))
2209
lines = self.backing_knit.get_lines(version_id)
2210
line_bytes = ''.join(lines)
2211
digest = sha_string(line_bytes)
2212
# the packed form of the fulltext always has a trailing newline,
2213
# even if the actual text does not, unless the file is empty. the
2214
# record options including the noeol flag are passed through by
2215
# _StreamIndex, so this is safe.
2217
if lines[-1][-1] != '\n':
2218
lines[-1] = lines[-1] + '\n'
2220
# We want plain data, because we expect to thunk only to allow text
2222
size, bytes = self.backing_knit._data._record_to_data(version_id,
2223
digest, lines, line_bytes)
2227
class _StreamIndex(object):
2228
"""A Knit Index object that uses the data map from a datastream."""
2230
def __init__(self, data_list, backing_index):
2231
"""Create a _StreamIndex object.
2233
:param data_list: The data_list from the datastream.
2234
:param backing_index: The index which will supply values for nodes
2235
referenced outside of this stream.
2237
self.data_list = data_list
2238
self.backing_index = backing_index
2239
self._by_version = {}
2241
for key, options, length, parents in data_list:
2242
self._by_version[key] = options, (pos, pos + length), parents
2245
def get_ancestry(self, versions, topo_sorted):
2246
"""Get an ancestry list for versions."""
2248
# Not needed for basic joins
2249
raise NotImplementedError(self.get_ancestry)
2250
# get a graph of all the mentioned versions:
2251
# Little ugly - basically copied from KnitIndex, but don't want to
2252
# accidentally incorporate too much of that index's code.
2254
pending = set(versions)
2255
cache = self._by_version
2257
version = pending.pop()
2260
parents = [p for p in cache[version][2] if p in cache]
2262
raise RevisionNotPresent(version, self)
2263
# if not completed and not a ghost
2264
pending.update([p for p in parents if p not in ancestry])
2265
ancestry.add(version)
2266
return list(ancestry)
2268
def get_build_details(self, version_ids):
2269
"""Get the method, index_memo and compression parent for version_ids.
2271
Ghosts are omitted from the result.
2273
:param version_ids: An iterable of version_ids.
2274
:return: A dict of version_id:(index_memo, compression_parent,
2275
parents, record_details).
2277
opaque memo that can be passed to _StreamAccess.read_records
2278
to extract the raw data; for these classes it is
2279
(from_backing_knit, index, start, end)
2281
Content that this record is built upon, may be None
2283
Logical parents of this node
2285
extra information about the content which needs to be passed to
2286
Factory.parse_record
2289
for version_id in version_ids:
2291
method = self.get_method(version_id)
2292
except errors.RevisionNotPresent:
2293
# ghosts are omitted
2295
parent_ids = self.get_parents_with_ghosts(version_id)
2296
noeol = ('no-eol' in self.get_options(version_id))
2297
index_memo = self.get_position(version_id)
2298
from_backing_knit = index_memo[0]
2299
if from_backing_knit:
2300
# texts retrieved from the backing knit are always full texts
2302
if method == 'fulltext':
2303
compression_parent = None
2305
compression_parent = parent_ids[0]
2306
result[version_id] = (index_memo, compression_parent,
2307
parent_ids, (method, noeol))
2310
def get_method(self, version_id):
2311
"""Return compression method of specified version."""
2312
options = self.get_options(version_id)
2313
if 'fulltext' in options:
2315
elif 'line-delta' in options:
2318
raise errors.KnitIndexUnknownMethod(self, options)
2320
def get_options(self, version_id):
2321
"""Return a list representing options.
2326
return self._by_version[version_id][0]
2328
options = list(self.backing_index.get_options(version_id))
2329
if 'fulltext' in options:
2331
elif 'line-delta' in options:
2332
# Texts from the backing knit are always returned from the stream
2334
options.remove('line-delta')
2335
options.append('fulltext')
2337
raise errors.KnitIndexUnknownMethod(self, options)
2338
return tuple(options)
2340
def get_parent_map(self, version_ids):
2341
"""Passed through to by KnitVersionedFile.get_parent_map."""
2344
for version_id in version_ids:
2346
result[version_id] = self._by_version[version_id][2]
2348
pending_ids.add(version_id)
2349
result.update(self.backing_index.get_parent_map(pending_ids))
2352
def get_parents_with_ghosts(self, version_id):
2353
"""Return parents of specified version with ghosts."""
2355
return self.get_parent_map([version_id])[version_id]
2357
raise RevisionNotPresent(version_id, self)
2359
def get_position(self, version_id):
2360
"""Return details needed to access the version.
2362
_StreamAccess has the data as a big array, so we return slice
2363
coordinates into that (as index_memo's are opaque outside the
2364
index and matching access class).
2366
:return: a tuple (from_backing_knit, index, start, end) that can
2367
be passed e.g. to get_raw_records.
2368
If from_backing_knit is False, index will be self, otherwise it
2369
will be a version id.
2372
start, end = self._by_version[version_id][1]
2373
return False, self, start, end
2375
# Signal to the access object to handle this from the backing knit.
2376
return (True, version_id, None, None)
2378
def get_versions(self):
2379
"""Get all the versions in the stream."""
2380
return self._by_version.keys()
2383
class _KnitData(object):
2384
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2386
The KnitData class provides the logic for parsing and using knit records,
2387
making use of an access method for the low level read and write operations.
2390
def __init__(self, access):
2391
"""Create a KnitData object.
2393
:param access: The access method to use. Access methods such as
2394
_KnitAccess manage the insertion of raw records and the subsequent
2395
retrieval of the same.
2397
self._access = access
2398
self._checked = False
2400
def _open_file(self):
2401
return self._access.open_file()
2403
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2404
"""Convert version_id, digest, lines into a raw data block.
2406
:param dense_lines: The bytes of lines but in a denser form. For
2407
instance, if lines is a list of 1000 bytestrings each ending in \n,
2408
dense_lines may be a list with one line in it, containing all the
2409
1000's lines and their \n's. Using dense_lines if it is already
2410
known is a win because the string join to create bytes in this
2411
function spends less time resizing the final string.
2412
:return: (len, a StringIO instance with the raw data ready to read.)
2414
# Note: using a string copy here increases memory pressure with e.g.
2415
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2416
# when doing the initial commit of a mozilla tree. RBC 20070921
2417
bytes = ''.join(chain(
2418
["version %s %d %s\n" % (version_id,
2421
dense_lines or lines,
2422
["end %s\n" % version_id]))
2423
assert bytes.__class__ == str
2424
compressed_bytes = bytes_to_gzip(bytes)
2425
return len(compressed_bytes), compressed_bytes
2427
def add_raw_records(self, sizes, raw_data):
2428
"""Append a prepared record to the data file.
2430
:param sizes: An iterable containing the size of each raw data segment.
2431
:param raw_data: A bytestring containing the data.
2432
:return: a list of index data for the way the data was stored.
2433
See the access method add_raw_records documentation for more
2436
return self._access.add_raw_records(sizes, raw_data)
2438
def _parse_record_header(self, version_id, raw_data):
2439
"""Parse a record header for consistency.
2441
:return: the header and the decompressor stream.
2442
as (stream, header_record)
2444
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2446
rec = self._check_header(version_id, df.readline())
2447
except Exception, e:
2448
raise KnitCorrupt(self._access,
2449
"While reading {%s} got %s(%s)"
2450
% (version_id, e.__class__.__name__, str(e)))
2453
def _check_header(self, version_id, line):
2456
raise KnitCorrupt(self._access,
2457
'unexpected number of elements in record header')
2458
if rec[1] != version_id:
2459
raise KnitCorrupt(self._access,
2460
'unexpected version, wanted %r, got %r'
2461
% (version_id, rec[1]))
2464
def _parse_record(self, version_id, data):
2466
# 4168 calls in 2880 217 internal
2467
# 4168 calls to _parse_record_header in 2121
2468
# 4168 calls to readlines in 330
2469
df = GzipFile(mode='rb', fileobj=StringIO(data))
2472
record_contents = df.readlines()
2473
except Exception, e:
2474
raise KnitCorrupt(self._access,
2475
"While reading {%s} got %s(%s)"
2476
% (version_id, e.__class__.__name__, str(e)))
2477
header = record_contents.pop(0)
2478
rec = self._check_header(version_id, header)
2480
last_line = record_contents.pop()
2481
if len(record_contents) != int(rec[2]):
2482
raise KnitCorrupt(self._access,
2483
'incorrect number of lines %s != %s'
2485
% (len(record_contents), int(rec[2]),
2487
if last_line != 'end %s\n' % rec[1]:
2488
raise KnitCorrupt(self._access,
2489
'unexpected version end line %r, wanted %r'
2490
% (last_line, version_id))
2492
return record_contents, rec[3]
2494
def read_records_iter_raw(self, records):
2495
"""Read text records from data file and yield raw data.
2497
This unpacks enough of the text record to validate the id is
2498
as expected but thats all.
2500
# setup an iterator of the external records:
2501
# uses readv so nice and fast we hope.
2503
# grab the disk data needed.
2504
needed_offsets = [index_memo for version_id, index_memo
2506
raw_records = self._access.get_raw_records(needed_offsets)
2508
for version_id, index_memo in records:
2509
data = raw_records.next()
2510
# validate the header
2511
df, rec = self._parse_record_header(version_id, data)
2513
yield version_id, data
2515
def read_records_iter(self, records):
2516
"""Read text records from data file and yield result.
2518
The result will be returned in whatever is the fastest to read.
2519
Not by the order requested. Also, multiple requests for the same
2520
record will only yield 1 response.
2521
:param records: A list of (version_id, pos, len) entries
2522
:return: Yields (version_id, contents, digest) in the order
2523
read, not the order requested
2528
needed_records = sorted(set(records), key=operator.itemgetter(1))
2529
if not needed_records:
2532
# The transport optimizes the fetching as well
2533
# (ie, reads continuous ranges.)
2534
raw_data = self._access.get_raw_records(
2535
[index_memo for version_id, index_memo in needed_records])
2537
for (version_id, index_memo), data in \
2538
izip(iter(needed_records), raw_data):
2539
content, digest = self._parse_record(version_id, data)
2540
yield version_id, content, digest
2542
def read_records(self, records):
2543
"""Read records into a dictionary."""
2545
for record_id, content, digest in \
2546
self.read_records_iter(records):
2547
components[record_id] = (content, digest)
2551
class InterKnit(InterVersionedFile):
2552
"""Optimised code paths for knit to knit operations."""
2554
_matching_file_from_factory = staticmethod(make_file_knit)
2555
_matching_file_to_factory = staticmethod(make_file_knit)
2558
def is_compatible(source, target):
2559
"""Be compatible with knits. """
2561
return (isinstance(source, KnitVersionedFile) and
2562
isinstance(target, KnitVersionedFile))
2563
except AttributeError:
2566
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2567
"""Copy texts to the target by extracting and adding them one by one.
2569
see join() for the parameter definitions.
2571
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2572
# --- the below is factorable out with VersionedFile.join, but wait for
2573
# VersionedFiles, it may all be simpler then.
2574
graph = Graph(self.source)
2575
search = graph._make_breadth_first_searcher(version_ids)
2576
transitive_ids = set()
2577
map(transitive_ids.update, list(search))
2578
parent_map = self.source.get_parent_map(transitive_ids)
2579
order = topo_sort(parent_map.items())
2581
def size_of_content(content):
2582
return sum(len(line) for line in content.text())
2583
# Cache at most 10MB of parent texts
2584
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2585
compute_size=size_of_content)
2586
# TODO: jam 20071116 It would be nice to have a streaming interface to
2587
# get multiple texts from a source. The source could be smarter
2588
# about how it handled intermediate stages.
2589
# get_line_list() or make_mpdiffs() seem like a possibility, but
2590
# at the moment they extract all full texts into memory, which
2591
# causes us to store more than our 3x fulltext goal.
2592
# Repository.iter_files_bytes() may be another possibility
2593
to_process = [version for version in order
2594
if version not in self.target]
2595
total = len(to_process)
2596
pb = ui.ui_factory.nested_progress_bar()
2598
for index, version in enumerate(to_process):
2599
pb.update('Converting versioned data', index, total)
2600
sha1, num_bytes, parent_text = self.target.add_lines(version,
2601
self.source.get_parents_with_ghosts(version),
2602
self.source.get_lines(version),
2603
parent_texts=parent_cache)
2604
parent_cache[version] = parent_text
2609
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2610
"""See InterVersionedFile.join."""
2611
assert isinstance(self.source, KnitVersionedFile)
2612
assert isinstance(self.target, KnitVersionedFile)
2614
# If the source and target are mismatched w.r.t. annotations vs
2615
# plain, the data needs to be converted accordingly
2616
if self.source.factory.annotated == self.target.factory.annotated:
2618
elif self.source.factory.annotated:
2619
converter = self._anno_to_plain_converter
2621
# We're converting from a plain to an annotated knit. Copy them
2622
# across by full texts.
2623
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2625
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2629
pb = ui.ui_factory.nested_progress_bar()
2631
version_ids = list(version_ids)
2632
if None in version_ids:
2633
version_ids.remove(None)
2635
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2637
this_versions = set(self.target._index.get_versions())
2638
# XXX: For efficiency we should not look at the whole index,
2639
# we only need to consider the referenced revisions - they
2640
# must all be present, or the method must be full-text.
2641
# TODO, RBC 20070919
2642
needed_versions = self.source_ancestry - this_versions
2644
if not needed_versions:
2646
full_list = topo_sort(
2647
self.source.get_parent_map(self.source.versions()))
2649
version_list = [i for i in full_list if (not self.target.has_version(i)
2650
and i in needed_versions)]
2654
copy_queue_records = []
2656
for version_id in version_list:
2657
options = self.source._index.get_options(version_id)
2658
parents = self.source._index.get_parents_with_ghosts(version_id)
2659
# check that its will be a consistent copy:
2660
for parent in parents:
2661
# if source has the parent, we must :
2662
# * already have it or
2663
# * have it scheduled already
2664
# otherwise we don't care
2665
assert (self.target.has_version(parent) or
2666
parent in copy_set or
2667
not self.source.has_version(parent))
2668
index_memo = self.source._index.get_position(version_id)
2669
copy_queue_records.append((version_id, index_memo))
2670
copy_queue.append((version_id, options, parents))
2671
copy_set.add(version_id)
2673
# data suck the join:
2675
total = len(version_list)
2678
for (version_id, raw_data), \
2679
(version_id2, options, parents) in \
2680
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2682
assert version_id == version_id2, 'logic error, inconsistent results'
2684
pb.update("Joining knit", count, total)
2686
size, raw_data = converter(raw_data, version_id, options,
2689
size = len(raw_data)
2690
raw_records.append((version_id, options, parents, size))
2691
raw_datum.append(raw_data)
2692
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2697
def _anno_to_plain_converter(self, raw_data, version_id, options,
2699
"""Convert annotated content to plain content."""
2700
data, digest = self.source._data._parse_record(version_id, raw_data)
2701
if 'fulltext' in options:
2702
content = self.source.factory.parse_fulltext(data, version_id)
2703
lines = self.target.factory.lower_fulltext(content)
2705
delta = self.source.factory.parse_line_delta(data, version_id,
2707
lines = self.target.factory.lower_line_delta(delta)
2708
return self.target._data._record_to_data(version_id, digest, lines)
2711
InterVersionedFile.register_optimiser(InterKnit)
2714
class WeaveToKnit(InterVersionedFile):
2715
"""Optimised code paths for weave to knit operations."""
2717
_matching_file_from_factory = bzrlib.weave.WeaveFile
2718
_matching_file_to_factory = staticmethod(make_file_knit)
2721
def is_compatible(source, target):
2722
"""Be compatible with weaves to knits."""
2724
return (isinstance(source, bzrlib.weave.Weave) and
2725
isinstance(target, KnitVersionedFile))
2726
except AttributeError:
2729
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2730
"""See InterVersionedFile.join."""
2731
assert isinstance(self.source, bzrlib.weave.Weave)
2732
assert isinstance(self.target, KnitVersionedFile)
2734
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2739
pb = ui.ui_factory.nested_progress_bar()
2741
version_ids = list(version_ids)
2743
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2744
this_versions = set(self.target._index.get_versions())
2745
needed_versions = self.source_ancestry - this_versions
2747
if not needed_versions:
2749
full_list = topo_sort(
2750
self.source.get_parent_map(self.source.versions()))
2752
version_list = [i for i in full_list if (not self.target.has_version(i)
2753
and i in needed_versions)]
2757
total = len(version_list)
2758
parent_map = self.source.get_parent_map(version_list)
2759
for version_id in version_list:
2760
pb.update("Converting to knit", count, total)
2761
parents = parent_map[version_id]
2762
# check that its will be a consistent copy:
2763
for parent in parents:
2764
# if source has the parent, we must already have it
2765
assert (self.target.has_version(parent))
2766
self.target.add_lines(
2767
version_id, parents, self.source.get_lines(version_id))
2774
InterVersionedFile.register_optimiser(WeaveToKnit)
2777
# Deprecated, use PatienceSequenceMatcher instead
2778
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2781
def annotate_knit(knit, revision_id):
2782
"""Annotate a knit with no cached annotations.
2784
This implementation is for knits with no cached annotations.
2785
It will work for knits with cached annotations, but this is not
2788
annotator = _KnitAnnotator(knit)
2789
return iter(annotator.annotate(revision_id))
2792
class _KnitAnnotator(object):
2793
"""Build up the annotations for a text."""
2795
def __init__(self, knit):
2798
# Content objects, differs from fulltexts because of how final newlines
2799
# are treated by knits. the content objects here will always have a
2801
self._fulltext_contents = {}
2803
# Annotated lines of specific revisions
2804
self._annotated_lines = {}
2806
# Track the raw data for nodes that we could not process yet.
2807
# This maps the revision_id of the base to a list of children that will
2808
# annotated from it.
2809
self._pending_children = {}
2811
# Nodes which cannot be extracted
2812
self._ghosts = set()
2814
# Track how many children this node has, so we know if we need to keep
2816
self._annotate_children = {}
2817
self._compression_children = {}
2819
self._all_build_details = {}
2820
# The children => parent revision_id graph
2821
self._revision_id_graph = {}
2823
self._heads_provider = None
2825
self._nodes_to_keep_annotations = set()
2826
self._generations_until_keep = 100
2828
def set_generations_until_keep(self, value):
2829
"""Set the number of generations before caching a node.
2831
Setting this to -1 will cache every merge node, setting this higher
2832
will cache fewer nodes.
2834
self._generations_until_keep = value
2836
def _add_fulltext_content(self, revision_id, content_obj):
2837
self._fulltext_contents[revision_id] = content_obj
2838
# TODO: jam 20080305 It might be good to check the sha1digest here
2839
return content_obj.text()
2841
def _check_parents(self, child, nodes_to_annotate):
2842
"""Check if all parents have been processed.
2844
:param child: A tuple of (rev_id, parents, raw_content)
2845
:param nodes_to_annotate: If child is ready, add it to
2846
nodes_to_annotate, otherwise put it back in self._pending_children
2848
for parent_id in child[1]:
2849
if (parent_id not in self._annotated_lines):
2850
# This parent is present, but another parent is missing
2851
self._pending_children.setdefault(parent_id,
2855
# This one is ready to be processed
2856
nodes_to_annotate.append(child)
2858
def _add_annotation(self, revision_id, fulltext, parent_ids,
2859
left_matching_blocks=None):
2860
"""Add an annotation entry.
2862
All parents should already have been annotated.
2863
:return: A list of children that now have their parents satisfied.
2865
a = self._annotated_lines
2866
annotated_parent_lines = [a[p] for p in parent_ids]
2867
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2868
fulltext, revision_id, left_matching_blocks,
2869
heads_provider=self._get_heads_provider()))
2870
self._annotated_lines[revision_id] = annotated_lines
2871
for p in parent_ids:
2872
ann_children = self._annotate_children[p]
2873
ann_children.remove(revision_id)
2874
if (not ann_children
2875
and p not in self._nodes_to_keep_annotations):
2876
del self._annotated_lines[p]
2877
del self._all_build_details[p]
2878
if p in self._fulltext_contents:
2879
del self._fulltext_contents[p]
2880
# Now that we've added this one, see if there are any pending
2881
# deltas to be done, certainly this parent is finished
2882
nodes_to_annotate = []
2883
for child in self._pending_children.pop(revision_id, []):
2884
self._check_parents(child, nodes_to_annotate)
2885
return nodes_to_annotate
2887
def _get_build_graph(self, revision_id):
2888
"""Get the graphs for building texts and annotations.
2890
The data you need for creating a full text may be different than the
2891
data you need to annotate that text. (At a minimum, you need both
2892
parents to create an annotation, but only need 1 parent to generate the
2895
:return: A list of (revision_id, index_memo) records, suitable for
2896
passing to read_records_iter to start reading in the raw data fro/
2899
if revision_id in self._annotated_lines:
2902
pending = set([revision_id])
2907
# get all pending nodes
2909
this_iteration = pending
2910
build_details = self._knit._index.get_build_details(this_iteration)
2911
self._all_build_details.update(build_details)
2912
# new_nodes = self._knit._index._get_entries(this_iteration)
2914
for rev_id, details in build_details.iteritems():
2915
(index_memo, compression_parent, parents,
2916
record_details) = details
2917
self._revision_id_graph[rev_id] = parents
2918
records.append((rev_id, index_memo))
2919
# Do we actually need to check _annotated_lines?
2920
pending.update(p for p in parents
2921
if p not in self._all_build_details)
2922
if compression_parent:
2923
self._compression_children.setdefault(compression_parent,
2926
for parent in parents:
2927
self._annotate_children.setdefault(parent,
2929
num_gens = generation - kept_generation
2930
if ((num_gens >= self._generations_until_keep)
2931
and len(parents) > 1):
2932
kept_generation = generation
2933
self._nodes_to_keep_annotations.add(rev_id)
2935
missing_versions = this_iteration.difference(build_details.keys())
2936
self._ghosts.update(missing_versions)
2937
for missing_version in missing_versions:
2938
# add a key, no parents
2939
self._revision_id_graph[missing_version] = ()
2940
pending.discard(missing_version) # don't look for it
2941
# XXX: This should probably be a real exception, as it is a data
2943
assert not self._ghosts.intersection(self._compression_children), \
2944
"We cannot have nodes which have a compression parent of a ghost."
2945
# Cleanout anything that depends on a ghost so that we don't wait for
2946
# the ghost to show up
2947
for node in self._ghosts:
2948
if node in self._annotate_children:
2949
# We won't be building this node
2950
del self._annotate_children[node]
2951
# Generally we will want to read the records in reverse order, because
2952
# we find the parent nodes after the children
2956
def _annotate_records(self, records):
2957
"""Build the annotations for the listed records."""
2958
# We iterate in the order read, rather than a strict order requested
2959
# However, process what we can, and put off to the side things that
2960
# still need parents, cleaning them up when those parents are
2962
for (rev_id, record,
2963
digest) in self._knit._data.read_records_iter(records):
2964
if rev_id in self._annotated_lines:
2966
parent_ids = self._revision_id_graph[rev_id]
2967
parent_ids = [p for p in parent_ids if p not in self._ghosts]
2968
details = self._all_build_details[rev_id]
2969
(index_memo, compression_parent, parents,
2970
record_details) = details
2971
nodes_to_annotate = []
2972
# TODO: Remove the punning between compression parents, and
2973
# parent_ids, we should be able to do this without assuming
2975
if len(parent_ids) == 0:
2976
# There are no parents for this node, so just add it
2977
# TODO: This probably needs to be decoupled
2978
assert compression_parent is None
2979
fulltext_content, delta = self._knit.factory.parse_record(
2980
rev_id, record, record_details, None)
2981
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
2982
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
2983
parent_ids, left_matching_blocks=None))
2985
child = (rev_id, parent_ids, record)
2986
# Check if all the parents are present
2987
self._check_parents(child, nodes_to_annotate)
2988
while nodes_to_annotate:
2989
# Should we use a queue here instead of a stack?
2990
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
2991
(index_memo, compression_parent, parents,
2992
record_details) = self._all_build_details[rev_id]
2993
if compression_parent is not None:
2994
comp_children = self._compression_children[compression_parent]
2995
assert rev_id in comp_children
2996
# If there is only 1 child, it is safe to reuse this
2998
reuse_content = (len(comp_children) == 1
2999
and compression_parent not in
3000
self._nodes_to_keep_annotations)
3002
# Remove it from the cache since it will be changing
3003
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3004
# Make sure to copy the fulltext since it might be
3006
parent_fulltext = list(parent_fulltext_content.text())
3008
parent_fulltext_content = self._fulltext_contents[compression_parent]
3009
parent_fulltext = parent_fulltext_content.text()
3010
comp_children.remove(rev_id)
3011
fulltext_content, delta = self._knit.factory.parse_record(
3012
rev_id, record, record_details,
3013
parent_fulltext_content,
3014
copy_base_content=(not reuse_content))
3015
fulltext = self._add_fulltext_content(rev_id,
3017
blocks = KnitContent.get_line_delta_blocks(delta,
3018
parent_fulltext, fulltext)
3020
fulltext_content = self._knit.factory.parse_fulltext(
3022
fulltext = self._add_fulltext_content(rev_id,
3025
nodes_to_annotate.extend(
3026
self._add_annotation(rev_id, fulltext, parent_ids,
3027
left_matching_blocks=blocks))
3029
def _get_heads_provider(self):
3030
"""Create a heads provider for resolving ancestry issues."""
3031
if self._heads_provider is not None:
3032
return self._heads_provider
3033
parent_provider = _mod_graph.DictParentsProvider(
3034
self._revision_id_graph)
3035
graph_obj = _mod_graph.Graph(parent_provider)
3036
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3037
self._heads_provider = head_cache
3040
def annotate(self, revision_id):
3041
"""Return the annotated fulltext at the given revision.
3043
:param revision_id: The revision id for this file
3045
records = self._get_build_graph(revision_id)
3046
if revision_id in self._ghosts:
3047
raise errors.RevisionNotPresent(revision_id, self._knit)
3048
self._annotate_records(records)
3049
return self._annotated_lines[revision_id]
3053
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3055
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
