1
# Copyright (C) 2005, 2006, 2007 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""Knit versionedfile implementation.
19
A knit is a versioned file implementation that supports efficient append only
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
end-marker; simply "end VERSION"
28
delta can be line or full contents.a
29
... the 8's there are the index number of the annotation.
30
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
8 e.set('executable', 'yes')
36
8 if elt.get('executable') == 'yes':
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
43
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
44
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
45
09:33 < lifeless> right
46
09:33 < jrydberg> lifeless: the position and size is the range in the data file
49
so the index sequence is the dictionary compressed sequence number used
50
in the deltas to provide line annotation
55
# 10:16 < lifeless> make partial index writes safe
56
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
57
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
59
# move sha1 out of the content so that join is faster at verifying parents
60
# record content length ?
64
from cStringIO import StringIO
65
from itertools import izip, chain
70
from zlib import Z_DEFAULT_COMPRESSION
73
from bzrlib.lazy_import import lazy_import
74
lazy_import(globals(), """
94
from bzrlib.errors import (
102
RevisionAlreadyPresent,
104
from bzrlib.graph import Graph
105
from bzrlib.osutils import (
111
from bzrlib.symbol_versioning import (
112
DEPRECATED_PARAMETER,
117
from bzrlib.tsort import topo_sort
118
from bzrlib.tuned_gzip import GzipFile, bytes_to_gzip
120
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
124
# TODO: Split out code specific to this format into an associated object.
126
# TODO: Can we put in some kind of value to check that the index and data
127
# files belong together?
129
# TODO: accommodate binaries, perhaps by storing a byte count
131
# TODO: function to check whole file
133
# TODO: atomically append data, then measure backwards from the cursor
134
# position after writing to work out where it was located. we may need to
135
# bypass python file buffering.
137
DATA_SUFFIX = '.knit'
138
INDEX_SUFFIX = '.kndx'
141
class KnitContent(object):
142
"""Content of a knit version to which deltas can be applied."""
145
self._should_strip_eol = False
148
"""Return a list of (origin, text) tuples."""
149
return list(self.annotate_iter())
151
def apply_delta(self, delta, new_version_id):
152
"""Apply delta to this object to become new_version_id."""
153
raise NotImplementedError(self.apply_delta)
155
def cleanup_eol(self, copy_on_mutate=True):
156
if self._should_strip_eol:
158
self._lines = self._lines[:]
159
self.strip_last_line_newline()
161
def line_delta_iter(self, new_lines):
162
"""Generate line-based delta from this content to new_lines."""
163
new_texts = new_lines.text()
164
old_texts = self.text()
165
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
166
for tag, i1, i2, j1, j2 in s.get_opcodes():
169
# ofrom, oto, length, data
170
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
172
def line_delta(self, new_lines):
173
return list(self.line_delta_iter(new_lines))
176
def get_line_delta_blocks(knit_delta, source, target):
177
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
178
target_len = len(target)
181
for s_begin, s_end, t_len, new_text in knit_delta:
182
true_n = s_begin - s_pos
185
# knit deltas do not provide reliable info about whether the
186
# last line of a file matches, due to eol handling.
187
if source[s_pos + n -1] != target[t_pos + n -1]:
190
yield s_pos, t_pos, n
191
t_pos += t_len + true_n
193
n = target_len - t_pos
195
if source[s_pos + n -1] != target[t_pos + n -1]:
198
yield s_pos, t_pos, n
199
yield s_pos + (target_len - t_pos), target_len, 0
202
class AnnotatedKnitContent(KnitContent):
203
"""Annotated content."""
205
def __init__(self, lines):
206
KnitContent.__init__(self)
209
def annotate_iter(self):
210
"""Yield tuples of (origin, text) for each content line."""
211
return iter(self._lines)
213
def apply_delta(self, delta, new_version_id):
214
"""Apply delta to this object to become new_version_id."""
217
for start, end, count, delta_lines in delta:
218
lines[offset+start:offset+end] = delta_lines
219
offset = offset + (start - end) + count
221
def strip_last_line_newline(self):
222
line = self._lines[-1][1].rstrip('\n')
223
self._lines[-1] = (self._lines[-1][0], line)
224
self._should_strip_eol = False
228
lines = [text for origin, text in self._lines]
229
except ValueError, e:
230
# most commonly (only?) caused by the internal form of the knit
231
# missing annotation information because of a bug - see thread
233
raise KnitCorrupt(self,
234
"line in annotated knit missing annotation information: %s"
237
if self._should_strip_eol:
238
anno, line = lines[-1]
239
lines[-1] = (anno, line.rstrip('\n'))
243
return AnnotatedKnitContent(self._lines[:])
246
class PlainKnitContent(KnitContent):
247
"""Unannotated content.
249
When annotate[_iter] is called on this content, the same version is reported
250
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
254
def __init__(self, lines, version_id):
255
KnitContent.__init__(self)
257
self._version_id = version_id
259
def annotate_iter(self):
260
"""Yield tuples of (origin, text) for each content line."""
261
for line in self._lines:
262
yield self._version_id, line
264
def apply_delta(self, delta, new_version_id):
265
"""Apply delta to this object to become new_version_id."""
268
for start, end, count, delta_lines in delta:
269
lines[offset+start:offset+end] = delta_lines
270
offset = offset + (start - end) + count
271
self._version_id = new_version_id
274
return PlainKnitContent(self._lines[:], self._version_id)
276
def strip_last_line_newline(self):
277
self._lines[-1] = self._lines[-1].rstrip('\n')
278
self._should_strip_eol = False
282
if self._should_strip_eol:
284
lines[-1] = lines[-1].rstrip('\n')
288
class _KnitFactory(object):
289
"""Base class for common Factory functions."""
291
def parse_record(self, version_id, record, record_details,
292
base_content, copy_base_content=True):
293
"""Parse a record into a full content object.
295
:param version_id: The official version id for this content
296
:param record: The data returned by read_records_iter()
297
:param record_details: Details about the record returned by
299
:param base_content: If get_build_details returns a compression_parent,
300
you must return a base_content here, else use None
301
:param copy_base_content: When building from the base_content, decide
302
you can either copy it and return a new object, or modify it in
304
:return: (content, delta) A Content object and possibly a line-delta,
307
method, noeol = record_details
308
if method == 'line-delta':
309
assert base_content is not None
310
if copy_base_content:
311
content = base_content.copy()
313
content = base_content
314
delta = self.parse_line_delta(record, version_id)
315
content.apply_delta(delta, version_id)
317
content = self.parse_fulltext(record, version_id)
319
content._should_strip_eol = noeol
320
return (content, delta)
323
class KnitAnnotateFactory(_KnitFactory):
324
"""Factory for creating annotated Content objects."""
328
def make(self, lines, version_id):
329
num_lines = len(lines)
330
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
332
def parse_fulltext(self, content, version_id):
333
"""Convert fulltext to internal representation
335
fulltext content is of the format
336
revid(utf8) plaintext\n
337
internal representation is of the format:
340
# TODO: jam 20070209 The tests expect this to be returned as tuples,
341
# but the code itself doesn't really depend on that.
342
# Figure out a way to not require the overhead of turning the
343
# list back into tuples.
344
lines = [tuple(line.split(' ', 1)) for line in content]
345
return AnnotatedKnitContent(lines)
347
def parse_line_delta_iter(self, lines):
348
return iter(self.parse_line_delta(lines))
350
def parse_line_delta(self, lines, version_id, plain=False):
351
"""Convert a line based delta into internal representation.
353
line delta is in the form of:
354
intstart intend intcount
356
revid(utf8) newline\n
357
internal representation is
358
(start, end, count, [1..count tuples (revid, newline)])
360
:param plain: If True, the lines are returned as a plain
361
list without annotations, not as a list of (origin, content) tuples, i.e.
362
(start, end, count, [1..count newline])
369
def cache_and_return(line):
370
origin, text = line.split(' ', 1)
371
return cache.setdefault(origin, origin), text
373
# walk through the lines parsing.
374
# Note that the plain test is explicitly pulled out of the
375
# loop to minimise any performance impact
378
start, end, count = [int(n) for n in header.split(',')]
379
contents = [next().split(' ', 1)[1] for i in xrange(count)]
380
result.append((start, end, count, contents))
383
start, end, count = [int(n) for n in header.split(',')]
384
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
385
result.append((start, end, count, contents))
388
def get_fulltext_content(self, lines):
389
"""Extract just the content lines from a fulltext."""
390
return (line.split(' ', 1)[1] for line in lines)
392
def get_linedelta_content(self, lines):
393
"""Extract just the content from a line delta.
395
This doesn't return all of the extra information stored in a delta.
396
Only the actual content lines.
401
header = header.split(',')
402
count = int(header[2])
403
for i in xrange(count):
404
origin, text = next().split(' ', 1)
407
def lower_fulltext(self, content):
408
"""convert a fulltext content record into a serializable form.
410
see parse_fulltext which this inverts.
412
# TODO: jam 20070209 We only do the caching thing to make sure that
413
# the origin is a valid utf-8 line, eventually we could remove it
414
return ['%s %s' % (o, t) for o, t in content._lines]
416
def lower_line_delta(self, delta):
417
"""convert a delta into a serializable form.
419
See parse_line_delta which this inverts.
421
# TODO: jam 20070209 We only do the caching thing to make sure that
422
# the origin is a valid utf-8 line, eventually we could remove it
424
for start, end, c, lines in delta:
425
out.append('%d,%d,%d\n' % (start, end, c))
426
out.extend(origin + ' ' + text
427
for origin, text in lines)
430
def annotate_iter(self, knit, version_id):
431
content = knit._get_content(version_id)
432
return content.annotate_iter()
435
class KnitPlainFactory(_KnitFactory):
436
"""Factory for creating plain Content objects."""
440
def make(self, lines, version_id):
441
return PlainKnitContent(lines, version_id)
443
def parse_fulltext(self, content, version_id):
444
"""This parses an unannotated fulltext.
446
Note that this is not a noop - the internal representation
447
has (versionid, line) - its just a constant versionid.
449
return self.make(content, version_id)
451
def parse_line_delta_iter(self, lines, version_id):
453
num_lines = len(lines)
454
while cur < num_lines:
457
start, end, c = [int(n) for n in header.split(',')]
458
yield start, end, c, lines[cur:cur+c]
461
def parse_line_delta(self, lines, version_id):
462
return list(self.parse_line_delta_iter(lines, version_id))
464
def get_fulltext_content(self, lines):
465
"""Extract just the content lines from a fulltext."""
468
def get_linedelta_content(self, lines):
469
"""Extract just the content from a line delta.
471
This doesn't return all of the extra information stored in a delta.
472
Only the actual content lines.
477
header = header.split(',')
478
count = int(header[2])
479
for i in xrange(count):
482
def lower_fulltext(self, content):
483
return content.text()
485
def lower_line_delta(self, delta):
487
for start, end, c, lines in delta:
488
out.append('%d,%d,%d\n' % (start, end, c))
492
def annotate_iter(self, knit, version_id):
493
annotator = _KnitAnnotator(knit)
494
return iter(annotator.annotate(version_id))
497
def make_empty_knit(transport, relpath):
498
"""Construct a empty knit at the specified location."""
499
k = make_file_knit(transport, relpath, 'w', KnitPlainFactory)
502
def make_file_knit(name, transport, file_mode=None, access_mode='w',
503
factory=None, delta=True, create=False, create_parent_dir=False,
504
delay_create=False, dir_mode=None, get_scope=None):
505
"""Factory to create a KnitVersionedFile for a .knit/.kndx file pair."""
507
factory = KnitAnnotateFactory()
509
factory = KnitPlainFactory()
510
if get_scope is None:
511
get_scope = lambda:None
512
index = _KnitIndex(transport, name + INDEX_SUFFIX,
513
access_mode, create=create, file_mode=file_mode,
514
create_parent_dir=create_parent_dir, delay_create=delay_create,
515
dir_mode=dir_mode, get_scope=get_scope)
516
access = _KnitAccess(transport, name + DATA_SUFFIX, file_mode,
517
dir_mode, ((create and not len(index)) and delay_create),
519
return KnitVersionedFile(name, transport, factory=factory,
520
create=create, delay_create=delay_create, index=index,
521
access_method=access)
525
"""Return the suffixes used by file based knits."""
526
return [DATA_SUFFIX, INDEX_SUFFIX]
527
make_file_knit.get_suffixes = get_suffixes
530
class KnitVersionedFile(VersionedFile):
531
"""Weave-like structure with faster random access.
533
A knit stores a number of texts and a summary of the relationships
534
between them. Texts are identified by a string version-id. Texts
535
are normally stored and retrieved as a series of lines, but can
536
also be passed as single strings.
538
Lines are stored with the trailing newline (if any) included, to
539
avoid special cases for files with no final newline. Lines are
540
composed of 8-bit characters, not unicode. The combination of
541
these approaches should mean any 'binary' file can be safely
542
stored and retrieved.
545
def __init__(self, relpath, transport, file_mode=None,
546
factory=None, delta=True, create=False, create_parent_dir=False,
547
delay_create=False, dir_mode=None, index=None, access_method=None):
548
"""Construct a knit at location specified by relpath.
550
:param create: If not True, only open an existing knit.
551
:param create_parent_dir: If True, create the parent directory if
552
creating the file fails. (This is used for stores with
553
hash-prefixes that may not exist yet)
554
:param delay_create: The calling code is aware that the knit won't
555
actually be created until the first data is stored.
556
:param index: An index to use for the knit.
558
super(KnitVersionedFile, self).__init__()
559
self.transport = transport
560
self.filename = relpath
561
self.factory = factory or KnitAnnotateFactory()
564
self._max_delta_chain = 200
566
if None in (access_method, index):
567
raise NotImplementedError, "No default access_method or index any more"
569
_access = access_method
570
if create and not len(self) and not delay_create:
572
self._data = _KnitData(_access)
575
return '%s(%s)' % (self.__class__.__name__,
576
self.transport.abspath(self.filename))
578
def _check_should_delta(self, first_parents):
579
"""Iterate back through the parent listing, looking for a fulltext.
581
This is used when we want to decide whether to add a delta or a new
582
fulltext. It searches for _max_delta_chain parents. When it finds a
583
fulltext parent, it sees if the total size of the deltas leading up to
584
it is large enough to indicate that we want a new full text anyway.
586
Return True if we should create a new delta, False if we should use a
591
delta_parents = first_parents
592
for count in xrange(self._max_delta_chain):
593
parent = delta_parents[0]
594
method = self._index.get_method(parent)
595
index, pos, size = self._index.get_position(parent)
596
if method == 'fulltext':
600
delta_parents = self._index.get_parent_map([parent])[parent]
602
# We couldn't find a fulltext, so we must create a new one
605
return fulltext_size > delta_size
607
def _check_write_ok(self):
608
return self._index._check_write_ok()
610
def _add_raw_records(self, records, data):
611
"""Add all the records 'records' with data pre-joined in 'data'.
613
:param records: A list of tuples(version_id, options, parents, size).
614
:param data: The data for the records. When it is written, the records
615
are adjusted to have pos pointing into data by the sum of
616
the preceding records sizes.
619
raw_record_sizes = [record[3] for record in records]
620
positions = self._data.add_raw_records(raw_record_sizes, data)
623
for (version_id, options, parents, size), access_memo in zip(
625
index_entries.append((version_id, options, access_memo, parents))
626
if self._data._do_cache:
627
self._data._cache[version_id] = data[offset:offset+size]
629
self._index.add_versions(index_entries)
631
def enable_cache(self):
632
"""Start caching data for this knit"""
633
self._data.enable_cache()
635
def clear_cache(self):
636
"""Clear the data cache only."""
637
self._data.clear_cache()
639
def copy_to(self, name, transport):
640
"""See VersionedFile.copy_to()."""
641
# copy the current index to a temp index to avoid racing with local
643
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
644
self.transport.get(self._index._filename))
646
f = self._data._open_file()
648
transport.put_file(name + DATA_SUFFIX, f)
651
# move the copied index into place
652
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
654
def get_data_stream(self, required_versions):
655
"""Get a data stream for the specified versions.
657
Versions may be returned in any order, not necessarily the order
658
specified. They are returned in a partial order by compression
659
parent, so that the deltas can be applied as the data stream is
660
inserted; however note that compression parents will not be sent
661
unless they were specifically requested, as the client may already
664
:param required_versions: The exact set of versions to be extracted.
665
Unlike some other knit methods, this is not used to generate a
666
transitive closure, rather it is used precisely as given.
668
:returns: format_signature, list of (version, options, length, parents),
671
required_version_set = frozenset(required_versions)
673
# list of revisions that can just be sent without waiting for their
676
# map from revision to the children based on it
678
# first, read all relevant index data, enough to sort into the right
680
for version_id in required_versions:
681
options = self._index.get_options(version_id)
682
parents = self._index.get_parents_with_ghosts(version_id)
683
index_memo = self._index.get_position(version_id)
684
version_index[version_id] = (index_memo, options, parents)
685
if ('line-delta' in options
686
and parents[0] in required_version_set):
687
# must wait until the parent has been sent
688
deferred.setdefault(parents[0], []). \
691
# either a fulltext, or a delta whose parent the client did
692
# not ask for and presumably already has
693
ready_to_send.append(version_id)
694
# build a list of results to return, plus instructions for data to
696
copy_queue_records = []
697
temp_version_list = []
699
# XXX: pushing and popping lists may be a bit inefficient
700
version_id = ready_to_send.pop(0)
701
(index_memo, options, parents) = version_index[version_id]
702
copy_queue_records.append((version_id, index_memo))
703
none, data_pos, data_size = index_memo
704
temp_version_list.append((version_id, options, data_size,
706
if version_id in deferred:
707
# now we can send all the children of this revision - we could
708
# put them in anywhere, but we hope that sending them soon
709
# after the fulltext will give good locality in the receiver
710
ready_to_send[:0] = deferred.pop(version_id)
711
assert len(deferred) == 0, \
712
"Still have compressed child versions waiting to be sent"
713
# XXX: The stream format is such that we cannot stream it - we have to
714
# know the length of all the data a-priori.
716
result_version_list = []
717
for (version_id, raw_data), \
718
(version_id2, options, _, parents) in \
719
izip(self._data.read_records_iter_raw(copy_queue_records),
721
assert version_id == version_id2, \
722
'logic error, inconsistent results'
723
raw_datum.append(raw_data)
724
result_version_list.append(
725
(version_id, options, len(raw_data), parents))
726
# provide a callback to get data incrementally.
727
pseudo_file = StringIO(''.join(raw_datum))
730
return pseudo_file.read()
732
return pseudo_file.read(length)
733
return (self.get_format_signature(), result_version_list, read)
735
def _extract_blocks(self, version_id, source, target):
736
if self._index.get_method(version_id) != 'line-delta':
738
parent, sha1, noeol, delta = self.get_delta(version_id)
739
return KnitContent.get_line_delta_blocks(delta, source, target)
741
def get_delta(self, version_id):
742
"""Get a delta for constructing version from some other version."""
743
self.check_not_reserved_id(version_id)
744
parents = self.get_parent_map([version_id])[version_id]
749
index_memo = self._index.get_position(version_id)
750
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
751
noeol = 'no-eol' in self._index.get_options(version_id)
752
if 'fulltext' == self._index.get_method(version_id):
753
new_content = self.factory.parse_fulltext(data, version_id)
754
if parent is not None:
755
reference_content = self._get_content(parent)
756
old_texts = reference_content.text()
759
new_texts = new_content.text()
760
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
762
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
764
delta = self.factory.parse_line_delta(data, version_id)
765
return parent, sha1, noeol, delta
767
def get_format_signature(self):
768
"""See VersionedFile.get_format_signature()."""
769
if self.factory.annotated:
770
annotated_part = "annotated"
772
annotated_part = "plain"
773
return "knit-%s" % (annotated_part,)
775
@deprecated_method(one_four)
776
def get_graph_with_ghosts(self):
777
"""See VersionedFile.get_graph_with_ghosts()."""
778
return self.get_parent_map(self.versions())
780
def get_sha1(self, version_id):
781
return self.get_sha1s([version_id])[0]
783
def get_sha1s(self, version_ids):
784
"""See VersionedFile.get_sha1()."""
785
record_map = self._get_record_map(version_ids)
786
# record entry 2 is the 'digest'.
787
return [record_map[v][2] for v in version_ids]
789
@deprecated_method(one_four)
790
def has_ghost(self, version_id):
791
"""True if there is a ghost reference in the file to version_id."""
793
if self.has_version(version_id):
795
# optimisable if needed by memoising the _ghosts set.
796
items = self.get_parent_map(self.versions())
797
for parents in items.itervalues():
798
for parent in parents:
799
if parent == version_id and parent not in items:
803
def insert_data_stream(self, (format, data_list, reader_callable)):
804
"""Insert knit records from a data stream into this knit.
806
If a version in the stream is already present in this knit, it will not
807
be inserted a second time. It will be checked for consistency with the
808
stored version however, and may cause a KnitCorrupt error to be raised
809
if the data in the stream disagrees with the already stored data.
811
:seealso: get_data_stream
813
if format != self.get_format_signature():
814
if 'knit' in debug.debug_flags:
816
'incompatible format signature inserting to %r', self)
817
source = self._knit_from_datastream(
818
(format, data_list, reader_callable))
822
for version_id, options, length, parents in data_list:
823
if self.has_version(version_id):
824
# First check: the list of parents.
825
my_parents = self.get_parents_with_ghosts(version_id)
826
if tuple(my_parents) != tuple(parents):
827
# XXX: KnitCorrupt is not quite the right exception here.
830
'parents list %r from data stream does not match '
831
'already recorded parents %r for %s'
832
% (parents, my_parents, version_id))
834
# Also check the SHA-1 of the fulltext this content will
836
raw_data = reader_callable(length)
837
my_fulltext_sha1 = self.get_sha1(version_id)
838
df, rec = self._data._parse_record_header(version_id, raw_data)
839
stream_fulltext_sha1 = rec[3]
840
if my_fulltext_sha1 != stream_fulltext_sha1:
841
# Actually, we don't know if it's this knit that's corrupt,
842
# or the data stream we're trying to insert.
844
self.filename, 'sha-1 does not match %s' % version_id)
846
if 'line-delta' in options:
847
# Make sure that this knit record is actually useful: a
848
# line-delta is no use unless we have its parent.
849
# Fetching from a broken repository with this problem
850
# shouldn't break the target repository.
852
# See https://bugs.launchpad.net/bzr/+bug/164443
853
if not self._index.has_version(parents[0]):
856
'line-delta from stream '
859
'missing parent %s\n'
860
'Try running "bzr check" '
861
'on the source repository, and "bzr reconcile" '
863
(version_id, parents[0]))
864
self._add_raw_records(
865
[(version_id, options, parents, length)],
866
reader_callable(length))
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 _clone_text(self, new_version_id, old_version_id, parents):
1108
"""See VersionedFile.clone_text()."""
1109
# FIXME RBC 20060228 make fast by only inserting an index with null
1111
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
1113
def get_lines(self, version_id):
1114
"""See VersionedFile.get_lines()."""
1115
return self.get_line_list([version_id])[0]
1117
def _get_record_map(self, version_ids):
1118
"""Produce a dictionary of knit records.
1120
:return: {version_id:(record, record_details, digest, next)}
1122
data returned from read_records
1124
opaque information to pass to parse_record
1126
SHA1 digest of the full text after all steps are done
1128
build-parent of the version, i.e. the leftmost ancestor.
1129
Will be None if the record is not a delta.
1131
position_map = self._get_components_positions(version_ids)
1132
# c = component_id, r = record_details, i_m = index_memo, n = next
1133
records = [(c, i_m) for c, (r, i_m, n)
1134
in position_map.iteritems()]
1136
for component_id, record, digest in \
1137
self._data.read_records_iter(records):
1138
(record_details, index_memo, next) = position_map[component_id]
1139
record_map[component_id] = record, record_details, digest, next
1143
def get_text(self, version_id):
1144
"""See VersionedFile.get_text"""
1145
return self.get_texts([version_id])[0]
1147
def get_texts(self, version_ids):
1148
return [''.join(l) for l in self.get_line_list(version_ids)]
1150
def get_line_list(self, version_ids):
1151
"""Return the texts of listed versions as a list of strings."""
1152
for version_id in version_ids:
1153
self.check_not_reserved_id(version_id)
1154
text_map, content_map = self._get_content_maps(version_ids)
1155
return [text_map[v] for v in version_ids]
1157
_get_lf_split_line_list = get_line_list
1159
def _get_content_maps(self, version_ids):
1160
"""Produce maps of text and KnitContents
1162
:return: (text_map, content_map) where text_map contains the texts for
1163
the requested versions and content_map contains the KnitContents.
1164
Both dicts take version_ids as their keys.
1166
# FUTURE: This function could be improved for the 'extract many' case
1167
# by tracking each component and only doing the copy when the number of
1168
# children than need to apply delta's to it is > 1 or it is part of the
1170
version_ids = list(version_ids)
1171
multiple_versions = len(version_ids) != 1
1172
record_map = self._get_record_map(version_ids)
1177
for version_id in version_ids:
1180
while cursor is not None:
1181
record, record_details, digest, next = record_map[cursor]
1182
components.append((cursor, record, record_details, digest))
1183
if cursor in content_map:
1188
for (component_id, record, record_details,
1189
digest) in reversed(components):
1190
if component_id in content_map:
1191
content = content_map[component_id]
1193
content, delta = self.factory.parse_record(version_id,
1194
record, record_details, content,
1195
copy_base_content=multiple_versions)
1196
if multiple_versions:
1197
content_map[component_id] = content
1199
content.cleanup_eol(copy_on_mutate=multiple_versions)
1200
final_content[version_id] = content
1202
# digest here is the digest from the last applied component.
1203
text = content.text()
1204
actual_sha = sha_strings(text)
1205
if actual_sha != digest:
1206
raise KnitCorrupt(self.filename,
1208
'\n of reconstructed text does not match'
1210
'\n for version %s' %
1211
(actual_sha, digest, version_id))
1212
text_map[version_id] = text
1213
return text_map, final_content
1215
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1217
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1218
if version_ids is None:
1219
version_ids = self.versions()
1221
pb = progress.DummyProgress()
1222
# we don't care about inclusions, the caller cares.
1223
# but we need to setup a list of records to visit.
1224
# we need version_id, position, length
1225
version_id_records = []
1226
requested_versions = set(version_ids)
1227
# filter for available versions
1228
for version_id in requested_versions:
1229
if not self.has_version(version_id):
1230
raise RevisionNotPresent(version_id, self.filename)
1231
# get a in-component-order queue:
1232
for version_id in self.versions():
1233
if version_id in requested_versions:
1234
index_memo = self._index.get_position(version_id)
1235
version_id_records.append((version_id, index_memo))
1237
total = len(version_id_records)
1238
for version_idx, (version_id, data, sha_value) in \
1239
enumerate(self._data.read_records_iter(version_id_records)):
1240
pb.update('Walking content.', version_idx, total)
1241
method = self._index.get_method(version_id)
1243
assert method in ('fulltext', 'line-delta')
1244
if method == 'fulltext':
1245
line_iterator = self.factory.get_fulltext_content(data)
1247
line_iterator = self.factory.get_linedelta_content(data)
1248
# XXX: It might be more efficient to yield (version_id,
1249
# line_iterator) in the future. However for now, this is a simpler
1250
# change to integrate into the rest of the codebase. RBC 20071110
1251
for line in line_iterator:
1252
yield line, version_id
1254
pb.update('Walking content.', total, total)
1256
def num_versions(self):
1257
"""See VersionedFile.num_versions()."""
1258
return self._index.num_versions()
1260
__len__ = num_versions
1262
def annotate_iter(self, version_id):
1263
"""See VersionedFile.annotate_iter."""
1264
return self.factory.annotate_iter(self, version_id)
1266
def get_parent_map(self, version_ids):
1267
"""See VersionedFile.get_parent_map."""
1268
return self._index.get_parent_map(version_ids)
1270
def get_ancestry(self, versions, topo_sorted=True):
1271
"""See VersionedFile.get_ancestry."""
1272
if isinstance(versions, basestring):
1273
versions = [versions]
1276
return self._index.get_ancestry(versions, topo_sorted)
1278
def get_ancestry_with_ghosts(self, versions):
1279
"""See VersionedFile.get_ancestry_with_ghosts."""
1280
if isinstance(versions, basestring):
1281
versions = [versions]
1284
return self._index.get_ancestry_with_ghosts(versions)
1286
def plan_merge(self, ver_a, ver_b):
1287
"""See VersionedFile.plan_merge."""
1288
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1289
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1290
annotated_a = self.annotate(ver_a)
1291
annotated_b = self.annotate(ver_b)
1292
return merge._plan_annotate_merge(annotated_a, annotated_b,
1293
ancestors_a, ancestors_b)
1296
class _KnitComponentFile(object):
1297
"""One of the files used to implement a knit database"""
1299
def __init__(self, transport, filename, mode, file_mode=None,
1300
create_parent_dir=False, dir_mode=None):
1301
self._transport = transport
1302
self._filename = filename
1304
self._file_mode = file_mode
1305
self._dir_mode = dir_mode
1306
self._create_parent_dir = create_parent_dir
1307
self._need_to_create = False
1309
def _full_path(self):
1310
"""Return the full path to this file."""
1311
return self._transport.base + self._filename
1313
def check_header(self, fp):
1314
line = fp.readline()
1316
# An empty file can actually be treated as though the file doesn't
1318
raise errors.NoSuchFile(self._full_path())
1319
if line != self.HEADER:
1320
raise KnitHeaderError(badline=line,
1321
filename=self._transport.abspath(self._filename))
1324
return '%s(%s)' % (self.__class__.__name__, self._filename)
1327
class _KnitIndex(_KnitComponentFile):
1328
"""Manages knit index file.
1330
The index is already kept in memory and read on startup, to enable
1331
fast lookups of revision information. The cursor of the index
1332
file is always pointing to the end, making it easy to append
1335
_cache is a cache for fast mapping from version id to a Index
1338
_history is a cache for fast mapping from indexes to version ids.
1340
The index data format is dictionary compressed when it comes to
1341
parent references; a index entry may only have parents that with a
1342
lover index number. As a result, the index is topological sorted.
1344
Duplicate entries may be written to the index for a single version id
1345
if this is done then the latter one completely replaces the former:
1346
this allows updates to correct version and parent information.
1347
Note that the two entries may share the delta, and that successive
1348
annotations and references MUST point to the first entry.
1350
The index file on disc contains a header, followed by one line per knit
1351
record. The same revision can be present in an index file more than once.
1352
The first occurrence gets assigned a sequence number starting from 0.
1354
The format of a single line is
1355
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1356
REVISION_ID is a utf8-encoded revision id
1357
FLAGS is a comma separated list of flags about the record. Values include
1358
no-eol, line-delta, fulltext.
1359
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1360
that the the compressed data starts at.
1361
LENGTH is the ascii representation of the length of the data file.
1362
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1364
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1365
revision id already in the knit that is a parent of REVISION_ID.
1366
The ' :' marker is the end of record marker.
1369
when a write is interrupted to the index file, it will result in a line
1370
that does not end in ' :'. If the ' :' is not present at the end of a line,
1371
or at the end of the file, then the record that is missing it will be
1372
ignored by the parser.
1374
When writing new records to the index file, the data is preceded by '\n'
1375
to ensure that records always start on new lines even if the last write was
1376
interrupted. As a result its normal for the last line in the index to be
1377
missing a trailing newline. One can be added with no harmful effects.
1380
HEADER = "# bzr knit index 8\n"
1382
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1383
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1385
def _cache_version(self, version_id, options, pos, size, parents):
1386
"""Cache a version record in the history array and index cache.
1388
This is inlined into _load_data for performance. KEEP IN SYNC.
1389
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1392
# only want the _history index to reference the 1st index entry
1394
if version_id not in self._cache:
1395
index = len(self._history)
1396
self._history.append(version_id)
1398
index = self._cache[version_id][5]
1399
self._cache[version_id] = (version_id,
1406
def _check_write_ok(self):
1407
if self._get_scope() != self._scope:
1408
raise errors.OutSideTransaction()
1409
if self._mode != 'w':
1410
raise errors.ReadOnlyObjectDirtiedError(self)
1412
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1413
create_parent_dir=False, delay_create=False, dir_mode=None,
1415
_KnitComponentFile.__init__(self, transport, filename, mode,
1416
file_mode=file_mode,
1417
create_parent_dir=create_parent_dir,
1420
# position in _history is the 'official' index for a revision
1421
# but the values may have come from a newer entry.
1422
# so - wc -l of a knit index is != the number of unique names
1426
fp = self._transport.get(self._filename)
1428
# _load_data may raise NoSuchFile if the target knit is
1430
_load_data(self, fp)
1434
if mode != 'w' or not create:
1437
self._need_to_create = True
1439
self._transport.put_bytes_non_atomic(
1440
self._filename, self.HEADER, mode=self._file_mode)
1441
self._scope = get_scope()
1442
self._get_scope = get_scope
1444
def get_ancestry(self, versions, topo_sorted=True):
1445
"""See VersionedFile.get_ancestry."""
1446
# get a graph of all the mentioned versions:
1448
pending = set(versions)
1451
version = pending.pop()
1454
parents = [p for p in cache[version][4] if p in cache]
1456
raise RevisionNotPresent(version, self._filename)
1457
# if not completed and not a ghost
1458
pending.update([p for p in parents if p not in graph])
1459
graph[version] = parents
1462
return topo_sort(graph.items())
1464
def get_ancestry_with_ghosts(self, versions):
1465
"""See VersionedFile.get_ancestry_with_ghosts."""
1466
# get a graph of all the mentioned versions:
1467
self.check_versions_present(versions)
1470
pending = set(versions)
1472
version = pending.pop()
1474
parents = cache[version][4]
1480
pending.update([p for p in parents if p not in graph])
1481
graph[version] = parents
1482
return topo_sort(graph.items())
1484
def get_build_details(self, version_ids):
1485
"""Get the method, index_memo and compression parent for version_ids.
1487
Ghosts are omitted from the result.
1489
:param version_ids: An iterable of version_ids.
1490
:return: A dict of version_id:(index_memo, compression_parent,
1491
parents, record_details).
1493
opaque structure to pass to read_records to extract the raw
1496
Content that this record is built upon, may be None
1498
Logical parents of this node
1500
extra information about the content which needs to be passed to
1501
Factory.parse_record
1504
for version_id in version_ids:
1505
if version_id not in self._cache:
1506
# ghosts are omitted
1508
method = self.get_method(version_id)
1509
parents = self.get_parents_with_ghosts(version_id)
1510
if method == 'fulltext':
1511
compression_parent = None
1513
compression_parent = parents[0]
1514
noeol = 'no-eol' in self.get_options(version_id)
1515
index_memo = self.get_position(version_id)
1516
result[version_id] = (index_memo, compression_parent,
1517
parents, (method, noeol))
1520
def num_versions(self):
1521
return len(self._history)
1523
__len__ = num_versions
1525
def get_versions(self):
1526
"""Get all the versions in the file. not topologically sorted."""
1527
return self._history
1529
def _version_list_to_index(self, versions):
1532
for version in versions:
1533
if version in cache:
1534
# -- inlined lookup() --
1535
result_list.append(str(cache[version][5]))
1536
# -- end lookup () --
1538
result_list.append('.' + version)
1539
return ' '.join(result_list)
1541
def add_version(self, version_id, options, index_memo, parents):
1542
"""Add a version record to the index."""
1543
self.add_versions(((version_id, options, index_memo, parents),))
1545
def add_versions(self, versions, random_id=False):
1546
"""Add multiple versions to the index.
1548
:param versions: a list of tuples:
1549
(version_id, options, pos, size, parents).
1550
:param random_id: If True the ids being added were randomly generated
1551
and no check for existence will be performed.
1554
orig_history = self._history[:]
1555
orig_cache = self._cache.copy()
1558
for version_id, options, (index, pos, size), parents in versions:
1559
line = "\n%s %s %s %s %s :" % (version_id,
1563
self._version_list_to_index(parents))
1564
assert isinstance(line, str), \
1565
'content must be utf-8 encoded: %r' % (line,)
1567
self._cache_version(version_id, options, pos, size, tuple(parents))
1568
if not self._need_to_create:
1569
self._transport.append_bytes(self._filename, ''.join(lines))
1572
sio.write(self.HEADER)
1573
sio.writelines(lines)
1575
self._transport.put_file_non_atomic(self._filename, sio,
1576
create_parent_dir=self._create_parent_dir,
1577
mode=self._file_mode,
1578
dir_mode=self._dir_mode)
1579
self._need_to_create = False
1581
# If any problems happen, restore the original values and re-raise
1582
self._history = orig_history
1583
self._cache = orig_cache
1586
def has_version(self, version_id):
1587
"""True if the version is in the index."""
1588
return version_id in self._cache
1590
def get_position(self, version_id):
1591
"""Return details needed to access the version.
1593
.kndx indices do not support split-out data, so return None for the
1596
:return: a tuple (None, data position, size) to hand to the access
1597
logic to get the record.
1599
entry = self._cache[version_id]
1600
return None, entry[2], entry[3]
1602
def get_method(self, version_id):
1603
"""Return compression method of specified version."""
1605
options = self._cache[version_id][1]
1607
raise RevisionNotPresent(version_id, self._filename)
1608
if 'fulltext' in options:
1611
if 'line-delta' not in options:
1612
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1615
def get_options(self, version_id):
1616
"""Return a list representing options.
1620
return self._cache[version_id][1]
1622
def get_parent_map(self, version_ids):
1623
"""Passed through to by KnitVersionedFile.get_parent_map."""
1625
for version_id in version_ids:
1627
result[version_id] = tuple(self._cache[version_id][4])
1632
def get_parents_with_ghosts(self, version_id):
1633
"""Return parents of specified version with ghosts."""
1635
return self.get_parent_map([version_id])[version_id]
1637
raise RevisionNotPresent(version_id, self)
1639
def check_versions_present(self, version_ids):
1640
"""Check that all specified versions are present."""
1642
for version_id in version_ids:
1643
if version_id not in cache:
1644
raise RevisionNotPresent(version_id, self._filename)
1647
class KnitGraphIndex(object):
1648
"""A knit index that builds on GraphIndex."""
1650
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1651
"""Construct a KnitGraphIndex on a graph_index.
1653
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1654
:param deltas: Allow delta-compressed records.
1655
:param add_callback: If not None, allow additions to the index and call
1656
this callback with a list of added GraphIndex nodes:
1657
[(node, value, node_refs), ...]
1658
:param parents: If True, record knits parents, if not do not record
1661
self._graph_index = graph_index
1662
self._deltas = deltas
1663
self._add_callback = add_callback
1664
self._parents = parents
1665
if deltas and not parents:
1666
raise KnitCorrupt(self, "Cannot do delta compression without "
1669
def _check_write_ok(self):
1672
def _get_entries(self, keys, check_present=False):
1673
"""Get the entries for keys.
1675
:param keys: An iterable of index keys, - 1-tuples.
1680
for node in self._graph_index.iter_entries(keys):
1682
found_keys.add(node[1])
1684
# adapt parentless index to the rest of the code.
1685
for node in self._graph_index.iter_entries(keys):
1686
yield node[0], node[1], node[2], ()
1687
found_keys.add(node[1])
1689
missing_keys = keys.difference(found_keys)
1691
raise RevisionNotPresent(missing_keys.pop(), self)
1693
def _present_keys(self, version_ids):
1695
node[1] for node in self._get_entries(version_ids)])
1697
def _parentless_ancestry(self, versions):
1698
"""Honour the get_ancestry API for parentless knit indices."""
1699
wanted_keys = self._version_ids_to_keys(versions)
1700
present_keys = self._present_keys(wanted_keys)
1701
missing = set(wanted_keys).difference(present_keys)
1703
raise RevisionNotPresent(missing.pop(), self)
1704
return list(self._keys_to_version_ids(present_keys))
1706
def get_ancestry(self, versions, topo_sorted=True):
1707
"""See VersionedFile.get_ancestry."""
1708
if not self._parents:
1709
return self._parentless_ancestry(versions)
1710
# XXX: This will do len(history) index calls - perhaps
1711
# it should be altered to be a index core feature?
1712
# get a graph of all the mentioned versions:
1715
versions = self._version_ids_to_keys(versions)
1716
pending = set(versions)
1718
# get all pending nodes
1719
this_iteration = pending
1720
new_nodes = self._get_entries(this_iteration)
1723
for (index, key, value, node_refs) in new_nodes:
1724
# dont ask for ghosties - otherwise
1725
# we we can end up looping with pending
1726
# being entirely ghosted.
1727
graph[key] = [parent for parent in node_refs[0]
1728
if parent not in ghosts]
1730
for parent in graph[key]:
1731
# dont examine known nodes again
1736
ghosts.update(this_iteration.difference(found))
1737
if versions.difference(graph):
1738
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1740
result_keys = topo_sort(graph.items())
1742
result_keys = graph.iterkeys()
1743
return [key[0] for key in result_keys]
1745
def get_ancestry_with_ghosts(self, versions):
1746
"""See VersionedFile.get_ancestry."""
1747
if not self._parents:
1748
return self._parentless_ancestry(versions)
1749
# XXX: This will do len(history) index calls - perhaps
1750
# it should be altered to be a index core feature?
1751
# get a graph of all the mentioned versions:
1753
versions = self._version_ids_to_keys(versions)
1754
pending = set(versions)
1756
# get all pending nodes
1757
this_iteration = pending
1758
new_nodes = self._get_entries(this_iteration)
1760
for (index, key, value, node_refs) in new_nodes:
1761
graph[key] = node_refs[0]
1763
for parent in graph[key]:
1764
# dont examine known nodes again
1768
missing_versions = this_iteration.difference(graph)
1769
missing_needed = versions.intersection(missing_versions)
1771
raise RevisionNotPresent(missing_needed.pop(), self)
1772
for missing_version in missing_versions:
1773
# add a key, no parents
1774
graph[missing_version] = []
1775
pending.discard(missing_version) # don't look for it
1776
result_keys = topo_sort(graph.items())
1777
return [key[0] for key in result_keys]
1779
def get_build_details(self, version_ids):
1780
"""Get the method, index_memo and compression parent for version_ids.
1782
Ghosts are omitted from the result.
1784
:param version_ids: An iterable of version_ids.
1785
:return: A dict of version_id:(index_memo, compression_parent,
1786
parents, record_details).
1788
opaque structure to pass to read_records to extract the raw
1791
Content that this record is built upon, may be None
1793
Logical parents of this node
1795
extra information about the content which needs to be passed to
1796
Factory.parse_record
1799
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
1800
for entry in entries:
1801
version_id = self._keys_to_version_ids((entry[1],))[0]
1802
if not self._parents:
1805
parents = self._keys_to_version_ids(entry[3][0])
1806
if not self._deltas:
1807
compression_parent = None
1809
compression_parent_key = self._compression_parent(entry)
1810
if compression_parent_key:
1811
compression_parent = self._keys_to_version_ids(
1812
(compression_parent_key,))[0]
1814
compression_parent = None
1815
noeol = (entry[2][0] == 'N')
1816
if compression_parent:
1817
method = 'line-delta'
1820
result[version_id] = (self._node_to_position(entry),
1821
compression_parent, parents,
1825
def _compression_parent(self, an_entry):
1826
# return the key that an_entry is compressed against, or None
1827
# Grab the second parent list (as deltas implies parents currently)
1828
compression_parents = an_entry[3][1]
1829
if not compression_parents:
1831
assert len(compression_parents) == 1
1832
return compression_parents[0]
1834
def _get_method(self, node):
1835
if not self._deltas:
1837
if self._compression_parent(node):
1842
def num_versions(self):
1843
return len(list(self._graph_index.iter_all_entries()))
1845
__len__ = num_versions
1847
def get_versions(self):
1848
"""Get all the versions in the file. not topologically sorted."""
1849
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1851
def has_version(self, version_id):
1852
"""True if the version is in the index."""
1853
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1855
def _keys_to_version_ids(self, keys):
1856
return tuple(key[0] for key in keys)
1858
def get_position(self, version_id):
1859
"""Return details needed to access the version.
1861
:return: a tuple (index, data position, size) to hand to the access
1862
logic to get the record.
1864
node = self._get_node(version_id)
1865
return self._node_to_position(node)
1867
def _node_to_position(self, node):
1868
"""Convert an index value to position details."""
1869
bits = node[2][1:].split(' ')
1870
return node[0], int(bits[0]), int(bits[1])
1872
def get_method(self, version_id):
1873
"""Return compression method of specified version."""
1874
return self._get_method(self._get_node(version_id))
1876
def _get_node(self, version_id):
1878
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1880
raise RevisionNotPresent(version_id, self)
1882
def get_options(self, version_id):
1883
"""Return a list representing options.
1887
node = self._get_node(version_id)
1888
options = [self._get_method(node)]
1889
if node[2][0] == 'N':
1890
options.append('no-eol')
1893
def get_parent_map(self, version_ids):
1894
"""Passed through to by KnitVersionedFile.get_parent_map."""
1895
nodes = self._get_entries(self._version_ids_to_keys(version_ids))
1899
result[node[1][0]] = self._keys_to_version_ids(node[3][0])
1902
result[node[1][0]] = ()
1905
def get_parents_with_ghosts(self, version_id):
1906
"""Return parents of specified version with ghosts."""
1908
return self.get_parent_map([version_id])[version_id]
1910
raise RevisionNotPresent(version_id, self)
1912
def check_versions_present(self, version_ids):
1913
"""Check that all specified versions are present."""
1914
keys = self._version_ids_to_keys(version_ids)
1915
present = self._present_keys(keys)
1916
missing = keys.difference(present)
1918
raise RevisionNotPresent(missing.pop(), self)
1920
def add_version(self, version_id, options, access_memo, parents):
1921
"""Add a version record to the index."""
1922
return self.add_versions(((version_id, options, access_memo, parents),))
1924
def add_versions(self, versions, random_id=False):
1925
"""Add multiple versions to the index.
1927
This function does not insert data into the Immutable GraphIndex
1928
backing the KnitGraphIndex, instead it prepares data for insertion by
1929
the caller and checks that it is safe to insert then calls
1930
self._add_callback with the prepared GraphIndex nodes.
1932
:param versions: a list of tuples:
1933
(version_id, options, pos, size, parents).
1934
:param random_id: If True the ids being added were randomly generated
1935
and no check for existence will be performed.
1937
if not self._add_callback:
1938
raise errors.ReadOnlyError(self)
1939
# we hope there are no repositories with inconsistent parentage
1944
for (version_id, options, access_memo, parents) in versions:
1945
index, pos, size = access_memo
1946
key = (version_id, )
1947
parents = tuple((parent, ) for parent in parents)
1948
if 'no-eol' in options:
1952
value += "%d %d" % (pos, size)
1953
if not self._deltas:
1954
if 'line-delta' in options:
1955
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1958
if 'line-delta' in options:
1959
node_refs = (parents, (parents[0],))
1961
node_refs = (parents, ())
1963
node_refs = (parents, )
1966
raise KnitCorrupt(self, "attempt to add node with parents "
1967
"in parentless index.")
1969
keys[key] = (value, node_refs)
1971
present_nodes = self._get_entries(keys)
1972
for (index, key, value, node_refs) in present_nodes:
1973
if (value, node_refs) != keys[key]:
1974
raise KnitCorrupt(self, "inconsistent details in add_versions"
1975
": %s %s" % ((value, node_refs), keys[key]))
1979
for key, (value, node_refs) in keys.iteritems():
1980
result.append((key, value, node_refs))
1982
for key, (value, node_refs) in keys.iteritems():
1983
result.append((key, value))
1984
self._add_callback(result)
1986
def _version_ids_to_keys(self, version_ids):
1987
return set((version_id, ) for version_id in version_ids)
1990
class _KnitAccess(object):
1991
"""Access to knit records in a .knit file."""
1993
def __init__(self, transport, filename, _file_mode, _dir_mode,
1994
_need_to_create, _create_parent_dir):
1995
"""Create a _KnitAccess for accessing and inserting data.
1997
:param transport: The transport the .knit is located on.
1998
:param filename: The filename of the .knit.
2000
self._transport = transport
2001
self._filename = filename
2002
self._file_mode = _file_mode
2003
self._dir_mode = _dir_mode
2004
self._need_to_create = _need_to_create
2005
self._create_parent_dir = _create_parent_dir
2007
def add_raw_records(self, sizes, raw_data):
2008
"""Add raw knit bytes to a storage area.
2010
The data is spooled to whereever the access method is storing data.
2012
:param sizes: An iterable containing the size of each raw data segment.
2013
:param raw_data: A bytestring containing the data.
2014
:return: A list of memos to retrieve the record later. Each memo is a
2015
tuple - (index, pos, length), where the index field is always None
2016
for the .knit access method.
2018
assert type(raw_data) == str, \
2019
'data must be plain bytes was %s' % type(raw_data)
2020
if not self._need_to_create:
2021
base = self._transport.append_bytes(self._filename, raw_data)
2023
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2024
create_parent_dir=self._create_parent_dir,
2025
mode=self._file_mode,
2026
dir_mode=self._dir_mode)
2027
self._need_to_create = False
2031
result.append((None, base, size))
2036
"""IFF this data access has its own storage area, initialise it.
2040
self._transport.put_bytes_non_atomic(self._filename, '',
2041
mode=self._file_mode)
2043
def open_file(self):
2044
"""IFF this data access can be represented as a single file, open it.
2046
For knits that are not mapped to a single file on disk this will
2049
:return: None or a file handle.
2052
return self._transport.get(self._filename)
2057
def get_raw_records(self, memos_for_retrieval):
2058
"""Get the raw bytes for a records.
2060
:param memos_for_retrieval: An iterable containing the (index, pos,
2061
length) memo for retrieving the bytes. The .knit method ignores
2062
the index as there is always only a single file.
2063
:return: An iterator over the bytes of the records.
2065
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2066
for pos, data in self._transport.readv(self._filename, read_vector):
2070
class _PackAccess(object):
2071
"""Access to knit records via a collection of packs."""
2073
def __init__(self, index_to_packs, writer=None):
2074
"""Create a _PackAccess object.
2076
:param index_to_packs: A dict mapping index objects to the transport
2077
and file names for obtaining data.
2078
:param writer: A tuple (pack.ContainerWriter, write_index) which
2079
contains the pack to write, and the index that reads from it will
2083
self.container_writer = writer[0]
2084
self.write_index = writer[1]
2086
self.container_writer = None
2087
self.write_index = None
2088
self.indices = index_to_packs
2090
def add_raw_records(self, sizes, raw_data):
2091
"""Add raw knit bytes to a storage area.
2093
The data is spooled to the container writer in one bytes-record per
2096
:param sizes: An iterable containing the size of each raw data segment.
2097
:param raw_data: A bytestring containing the data.
2098
:return: A list of memos to retrieve the record later. Each memo is a
2099
tuple - (index, pos, length), where the index field is the
2100
write_index object supplied to the PackAccess object.
2102
assert type(raw_data) == str, \
2103
'data must be plain bytes was %s' % type(raw_data)
2107
p_offset, p_length = self.container_writer.add_bytes_record(
2108
raw_data[offset:offset+size], [])
2110
result.append((self.write_index, p_offset, p_length))
2114
"""Pack based knits do not get individually created."""
2116
def get_raw_records(self, memos_for_retrieval):
2117
"""Get the raw bytes for a records.
2119
:param memos_for_retrieval: An iterable containing the (index, pos,
2120
length) memo for retrieving the bytes. The Pack access method
2121
looks up the pack to use for a given record in its index_to_pack
2123
:return: An iterator over the bytes of the records.
2125
# first pass, group into same-index requests
2127
current_index = None
2128
for (index, offset, length) in memos_for_retrieval:
2129
if current_index == index:
2130
current_list.append((offset, length))
2132
if current_index is not None:
2133
request_lists.append((current_index, current_list))
2134
current_index = index
2135
current_list = [(offset, length)]
2136
# handle the last entry
2137
if current_index is not None:
2138
request_lists.append((current_index, current_list))
2139
for index, offsets in request_lists:
2140
transport, path = self.indices[index]
2141
reader = pack.make_readv_reader(transport, path, offsets)
2142
for names, read_func in reader.iter_records():
2143
yield read_func(None)
2145
def open_file(self):
2146
"""Pack based knits have no single file."""
2149
def set_writer(self, writer, index, (transport, packname)):
2150
"""Set a writer to use for adding data."""
2151
if index is not None:
2152
self.indices[index] = (transport, packname)
2153
self.container_writer = writer
2154
self.write_index = index
2157
class _StreamAccess(object):
2158
"""A Knit Access object that provides data from a datastream.
2160
It also provides a fallback to present as unannotated data, annotated data
2161
from a *backing* access object.
2163
This is triggered by a index_memo which is pointing to a different index
2164
than this was constructed with, and is used to allow extracting full
2165
unannotated texts for insertion into annotated knits.
2168
def __init__(self, reader_callable, stream_index, backing_knit,
2170
"""Create a _StreamAccess object.
2172
:param reader_callable: The reader_callable from the datastream.
2173
This is called to buffer all the data immediately, for
2175
:param stream_index: The index the data stream this provides access to
2176
which will be present in native index_memo's.
2177
:param backing_knit: The knit object that will provide access to
2178
annotated texts which are not available in the stream, so as to
2179
create unannotated texts.
2180
:param orig_factory: The original content factory used to generate the
2181
stream. This is used for checking whether the thunk code for
2182
supporting _copy_texts will generate the correct form of data.
2184
self.data = reader_callable(None)
2185
self.stream_index = stream_index
2186
self.backing_knit = backing_knit
2187
self.orig_factory = orig_factory
2189
def get_raw_records(self, memos_for_retrieval):
2190
"""Get the raw bytes for a records.
2192
:param memos_for_retrieval: An iterable containing the (thunk_flag,
2193
index, start, end) memo for retrieving the bytes.
2194
:return: An iterator over the bytes of the records.
2196
# use a generator for memory friendliness
2197
for thunk_flag, version_id, start, end in memos_for_retrieval:
2198
if version_id is self.stream_index:
2199
yield self.data[start:end]
2201
# we have been asked to thunk. This thunking only occurs when
2202
# we are obtaining plain texts from an annotated backing knit
2203
# so that _copy_texts will work.
2204
# We could improve performance here by scanning for where we need
2205
# to do this and using get_line_list, then interleaving the output
2206
# as desired. However, for now, this is sufficient.
2207
if self.orig_factory.__class__ != KnitPlainFactory:
2208
raise errors.KnitCorrupt(
2209
self, 'Bad thunk request %r' % version_id)
2210
lines = self.backing_knit.get_lines(version_id)
2211
line_bytes = ''.join(lines)
2212
digest = sha_string(line_bytes)
2214
if lines[-1][-1] != '\n':
2215
lines[-1] = lines[-1] + '\n'
2217
orig_options = list(self.backing_knit._index.get_options(version_id))
2218
if 'fulltext' not in orig_options:
2219
if 'line-delta' not in orig_options:
2220
raise errors.KnitCorrupt(self,
2221
'Unknown compression method %r' % orig_options)
2222
orig_options.remove('line-delta')
2223
orig_options.append('fulltext')
2224
# We want plain data, because we expect to thunk only to allow text
2226
size, bytes = self.backing_knit._data._record_to_data(version_id,
2227
digest, lines, line_bytes)
2231
class _StreamIndex(object):
2232
"""A Knit Index object that uses the data map from a datastream."""
2234
def __init__(self, data_list, backing_index):
2235
"""Create a _StreamIndex object.
2237
:param data_list: The data_list from the datastream.
2238
:param backing_index: The index which will supply values for nodes
2239
referenced outside of this stream.
2241
self.data_list = data_list
2242
self.backing_index = backing_index
2243
self._by_version = {}
2245
for key, options, length, parents in data_list:
2246
self._by_version[key] = options, (pos, pos + length), parents
2249
def get_ancestry(self, versions, topo_sorted):
2250
"""Get an ancestry list for versions."""
2252
# Not needed for basic joins
2253
raise NotImplementedError(self.get_ancestry)
2254
# get a graph of all the mentioned versions:
2255
# Little ugly - basically copied from KnitIndex, but don't want to
2256
# accidentally incorporate too much of that index's code.
2258
pending = set(versions)
2259
cache = self._by_version
2261
version = pending.pop()
2264
parents = [p for p in cache[version][2] if p in cache]
2266
raise RevisionNotPresent(version, self)
2267
# if not completed and not a ghost
2268
pending.update([p for p in parents if p not in ancestry])
2269
ancestry.add(version)
2270
return list(ancestry)
2272
def get_build_details(self, version_ids):
2273
"""Get the method, index_memo and compression parent for version_ids.
2275
Ghosts are omitted from the result.
2277
:param version_ids: An iterable of version_ids.
2278
:return: A dict of version_id:(index_memo, compression_parent,
2279
parents, record_details).
2281
opaque structure to pass to read_records to extract the raw
2284
Content that this record is built upon, may be None
2286
Logical parents of this node
2288
extra information about the content which needs to be passed to
2289
Factory.parse_record
2292
for version_id in version_ids:
2294
method = self.get_method(version_id)
2295
except errors.RevisionNotPresent:
2296
# ghosts are omitted
2298
parent_ids = self.get_parents_with_ghosts(version_id)
2299
noeol = ('no-eol' in self.get_options(version_id))
2300
if method == 'fulltext':
2301
compression_parent = None
2303
compression_parent = parent_ids[0]
2304
index_memo = self.get_position(version_id)
2305
result[version_id] = (index_memo, compression_parent,
2306
parent_ids, (method, noeol))
2309
def get_method(self, version_id):
2310
"""Return compression method of specified version."""
2312
options = self._by_version[version_id][0]
2314
# Strictly speaking this should check in the backing knit, but
2315
# until we have a test to discriminate, this will do.
2316
return self.backing_index.get_method(version_id)
2317
if 'fulltext' in options:
2319
elif 'line-delta' in options:
2322
raise errors.KnitIndexUnknownMethod(self, options)
2324
def get_options(self, version_id):
2325
"""Return a list representing options.
2330
return self._by_version[version_id][0]
2332
return self.backing_index.get_options(version_id)
2334
def get_parent_map(self, version_ids):
2335
"""Passed through to by KnitVersionedFile.get_parent_map."""
2338
for version_id in version_ids:
2340
result[version_id] = self._by_version[version_id][2]
2342
pending_ids.add(version_id)
2343
result.update(self.backing_index.get_parent_map(pending_ids))
2346
def get_parents_with_ghosts(self, version_id):
2347
"""Return parents of specified version with ghosts."""
2349
return self.get_parent_map([version_id])[version_id]
2351
raise RevisionNotPresent(version_id, self)
2353
def get_position(self, version_id):
2354
"""Return details needed to access the version.
2356
_StreamAccess has the data as a big array, so we return slice
2357
coordinates into that (as index_memo's are opaque outside the
2358
index and matching access class).
2360
:return: a tuple (thunk_flag, index, start, end). If thunk_flag is
2361
False, index will be self, otherwise it will be a version id.
2364
start, end = self._by_version[version_id][1]
2365
return False, self, start, end
2367
# Signal to the access object to handle this from the backing knit.
2368
return (True, version_id, None, None)
2370
def get_versions(self):
2371
"""Get all the versions in the stream."""
2372
return self._by_version.keys()
2375
class _KnitData(object):
2376
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2378
The KnitData class provides the logic for parsing and using knit records,
2379
making use of an access method for the low level read and write operations.
2382
def __init__(self, access):
2383
"""Create a KnitData object.
2385
:param access: The access method to use. Access methods such as
2386
_KnitAccess manage the insertion of raw records and the subsequent
2387
retrieval of the same.
2389
self._access = access
2390
self._checked = False
2391
# TODO: jam 20060713 conceptually, this could spill to disk
2392
# if the cached size gets larger than a certain amount
2393
# but it complicates the model a bit, so for now just use
2394
# a simple dictionary
2396
self._do_cache = False
2398
def enable_cache(self):
2399
"""Enable caching of reads."""
2400
self._do_cache = True
2402
def clear_cache(self):
2403
"""Clear the record cache."""
2404
self._do_cache = False
2407
def _open_file(self):
2408
return self._access.open_file()
2410
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2411
"""Convert version_id, digest, lines into a raw data block.
2413
:param dense_lines: The bytes of lines but in a denser form. For
2414
instance, if lines is a list of 1000 bytestrings each ending in \n,
2415
dense_lines may be a list with one line in it, containing all the
2416
1000's lines and their \n's. Using dense_lines if it is already
2417
known is a win because the string join to create bytes in this
2418
function spends less time resizing the final string.
2419
:return: (len, a StringIO instance with the raw data ready to read.)
2421
# Note: using a string copy here increases memory pressure with e.g.
2422
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2423
# when doing the initial commit of a mozilla tree. RBC 20070921
2424
bytes = ''.join(chain(
2425
["version %s %d %s\n" % (version_id,
2428
dense_lines or lines,
2429
["end %s\n" % version_id]))
2430
assert bytes.__class__ == str
2431
compressed_bytes = bytes_to_gzip(bytes)
2432
return len(compressed_bytes), compressed_bytes
2434
def add_raw_records(self, sizes, raw_data):
2435
"""Append a prepared record to the data file.
2437
:param sizes: An iterable containing the size of each raw data segment.
2438
:param raw_data: A bytestring containing the data.
2439
:return: a list of index data for the way the data was stored.
2440
See the access method add_raw_records documentation for more
2443
return self._access.add_raw_records(sizes, raw_data)
2445
def _parse_record_header(self, version_id, raw_data):
2446
"""Parse a record header for consistency.
2448
:return: the header and the decompressor stream.
2449
as (stream, header_record)
2451
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2453
rec = self._check_header(version_id, df.readline())
2454
except Exception, e:
2455
raise KnitCorrupt(self._access,
2456
"While reading {%s} got %s(%s)"
2457
% (version_id, e.__class__.__name__, str(e)))
2460
def _check_header(self, version_id, line):
2463
raise KnitCorrupt(self._access,
2464
'unexpected number of elements in record header')
2465
if rec[1] != version_id:
2466
raise KnitCorrupt(self._access,
2467
'unexpected version, wanted %r, got %r'
2468
% (version_id, rec[1]))
2471
def _parse_record(self, version_id, data):
2473
# 4168 calls in 2880 217 internal
2474
# 4168 calls to _parse_record_header in 2121
2475
# 4168 calls to readlines in 330
2476
df = GzipFile(mode='rb', fileobj=StringIO(data))
2479
record_contents = df.readlines()
2480
except Exception, e:
2481
raise KnitCorrupt(self._access,
2482
"While reading {%s} got %s(%s)"
2483
% (version_id, e.__class__.__name__, str(e)))
2484
header = record_contents.pop(0)
2485
rec = self._check_header(version_id, header)
2487
last_line = record_contents.pop()
2488
if len(record_contents) != int(rec[2]):
2489
raise KnitCorrupt(self._access,
2490
'incorrect number of lines %s != %s'
2492
% (len(record_contents), int(rec[2]),
2494
if last_line != 'end %s\n' % rec[1]:
2495
raise KnitCorrupt(self._access,
2496
'unexpected version end line %r, wanted %r'
2497
% (last_line, version_id))
2499
return record_contents, rec[3]
2501
def read_records_iter_raw(self, records):
2502
"""Read text records from data file and yield raw data.
2504
This unpacks enough of the text record to validate the id is
2505
as expected but thats all.
2507
# setup an iterator of the external records:
2508
# uses readv so nice and fast we hope.
2510
# grab the disk data needed.
2512
# Don't check _cache if it is empty
2513
needed_offsets = [index_memo for version_id, index_memo
2515
if version_id not in self._cache]
2517
needed_offsets = [index_memo for version_id, index_memo
2520
raw_records = self._access.get_raw_records(needed_offsets)
2522
for version_id, index_memo in records:
2523
if version_id in self._cache:
2524
# This data has already been validated
2525
data = self._cache[version_id]
2527
data = raw_records.next()
2529
self._cache[version_id] = data
2531
# validate the header
2532
df, rec = self._parse_record_header(version_id, data)
2534
yield version_id, data
2536
def read_records_iter(self, records):
2537
"""Read text records from data file and yield result.
2539
The result will be returned in whatever is the fastest to read.
2540
Not by the order requested. Also, multiple requests for the same
2541
record will only yield 1 response.
2542
:param records: A list of (version_id, pos, len) entries
2543
:return: Yields (version_id, contents, digest) in the order
2544
read, not the order requested
2550
# Skip records we have alread seen
2551
yielded_records = set()
2552
needed_records = set()
2553
for record in records:
2554
if record[0] in self._cache:
2555
if record[0] in yielded_records:
2557
yielded_records.add(record[0])
2558
data = self._cache[record[0]]
2559
content, digest = self._parse_record(record[0], data)
2560
yield (record[0], content, digest)
2562
needed_records.add(record)
2563
needed_records = sorted(needed_records, key=operator.itemgetter(1))
2565
needed_records = sorted(set(records), key=operator.itemgetter(1))
2567
if not needed_records:
2570
# The transport optimizes the fetching as well
2571
# (ie, reads continuous ranges.)
2572
raw_data = self._access.get_raw_records(
2573
[index_memo for version_id, index_memo in needed_records])
2575
for (version_id, index_memo), data in \
2576
izip(iter(needed_records), raw_data):
2577
content, digest = self._parse_record(version_id, data)
2579
self._cache[version_id] = data
2580
yield version_id, content, digest
2582
def read_records(self, records):
2583
"""Read records into a dictionary."""
2585
for record_id, content, digest in \
2586
self.read_records_iter(records):
2587
components[record_id] = (content, digest)
2591
class InterKnit(InterVersionedFile):
2592
"""Optimised code paths for knit to knit operations."""
2594
_matching_file_from_factory = staticmethod(make_file_knit)
2595
_matching_file_to_factory = staticmethod(make_file_knit)
2598
def is_compatible(source, target):
2599
"""Be compatible with knits. """
2601
return (isinstance(source, KnitVersionedFile) and
2602
isinstance(target, KnitVersionedFile))
2603
except AttributeError:
2606
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2607
"""Copy texts to the target by extracting and adding them one by one.
2609
see join() for the parameter definitions.
2611
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2612
# --- the below is factorable out with VersionedFile.join, but wait for
2613
# VersionedFiles, it may all be simpler then.
2614
graph = Graph(self.source)
2615
search = graph._make_breadth_first_searcher(version_ids)
2616
transitive_ids = set()
2617
map(transitive_ids.update, list(search))
2618
parent_map = self.source.get_parent_map(transitive_ids)
2619
order = topo_sort(parent_map.items())
2621
def size_of_content(content):
2622
return sum(len(line) for line in content.text())
2623
# Cache at most 10MB of parent texts
2624
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2625
compute_size=size_of_content)
2626
# TODO: jam 20071116 It would be nice to have a streaming interface to
2627
# get multiple texts from a source. The source could be smarter
2628
# about how it handled intermediate stages.
2629
# get_line_list() or make_mpdiffs() seem like a possibility, but
2630
# at the moment they extract all full texts into memory, which
2631
# causes us to store more than our 3x fulltext goal.
2632
# Repository.iter_files_bytes() may be another possibility
2633
to_process = [version for version in order
2634
if version not in self.target]
2635
total = len(to_process)
2636
pb = ui.ui_factory.nested_progress_bar()
2638
for index, version in enumerate(to_process):
2639
pb.update('Converting versioned data', index, total)
2640
sha1, num_bytes, parent_text = self.target.add_lines(version,
2641
self.source.get_parents_with_ghosts(version),
2642
self.source.get_lines(version),
2643
parent_texts=parent_cache)
2644
parent_cache[version] = parent_text
2649
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2650
"""See InterVersionedFile.join."""
2651
assert isinstance(self.source, KnitVersionedFile)
2652
assert isinstance(self.target, KnitVersionedFile)
2654
# If the source and target are mismatched w.r.t. annotations vs
2655
# plain, the data needs to be converted accordingly
2656
if self.source.factory.annotated == self.target.factory.annotated:
2658
elif self.source.factory.annotated:
2659
converter = self._anno_to_plain_converter
2661
# We're converting from a plain to an annotated knit. Copy them
2662
# across by full texts.
2663
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2665
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2669
pb = ui.ui_factory.nested_progress_bar()
2671
version_ids = list(version_ids)
2672
if None in version_ids:
2673
version_ids.remove(None)
2675
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2677
this_versions = set(self.target._index.get_versions())
2678
# XXX: For efficiency we should not look at the whole index,
2679
# we only need to consider the referenced revisions - they
2680
# must all be present, or the method must be full-text.
2681
# TODO, RBC 20070919
2682
needed_versions = self.source_ancestry - this_versions
2684
if not needed_versions:
2686
full_list = topo_sort(
2687
self.source.get_parent_map(self.source.versions()))
2689
version_list = [i for i in full_list if (not self.target.has_version(i)
2690
and i in needed_versions)]
2694
copy_queue_records = []
2696
for version_id in version_list:
2697
options = self.source._index.get_options(version_id)
2698
parents = self.source._index.get_parents_with_ghosts(version_id)
2699
# check that its will be a consistent copy:
2700
for parent in parents:
2701
# if source has the parent, we must :
2702
# * already have it or
2703
# * have it scheduled already
2704
# otherwise we don't care
2705
assert (self.target.has_version(parent) or
2706
parent in copy_set or
2707
not self.source.has_version(parent))
2708
index_memo = self.source._index.get_position(version_id)
2709
copy_queue_records.append((version_id, index_memo))
2710
copy_queue.append((version_id, options, parents))
2711
copy_set.add(version_id)
2713
# data suck the join:
2715
total = len(version_list)
2718
for (version_id, raw_data), \
2719
(version_id2, options, parents) in \
2720
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2722
assert version_id == version_id2, 'logic error, inconsistent results'
2724
pb.update("Joining knit", count, total)
2726
size, raw_data = converter(raw_data, version_id, options,
2729
size = len(raw_data)
2730
raw_records.append((version_id, options, parents, size))
2731
raw_datum.append(raw_data)
2732
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2737
def _anno_to_plain_converter(self, raw_data, version_id, options,
2739
"""Convert annotated content to plain content."""
2740
data, digest = self.source._data._parse_record(version_id, raw_data)
2741
if 'fulltext' in options:
2742
content = self.source.factory.parse_fulltext(data, version_id)
2743
lines = self.target.factory.lower_fulltext(content)
2745
delta = self.source.factory.parse_line_delta(data, version_id,
2747
lines = self.target.factory.lower_line_delta(delta)
2748
return self.target._data._record_to_data(version_id, digest, lines)
2751
InterVersionedFile.register_optimiser(InterKnit)
2754
class WeaveToKnit(InterVersionedFile):
2755
"""Optimised code paths for weave to knit operations."""
2757
_matching_file_from_factory = bzrlib.weave.WeaveFile
2758
_matching_file_to_factory = staticmethod(make_file_knit)
2761
def is_compatible(source, target):
2762
"""Be compatible with weaves to knits."""
2764
return (isinstance(source, bzrlib.weave.Weave) and
2765
isinstance(target, KnitVersionedFile))
2766
except AttributeError:
2769
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2770
"""See InterVersionedFile.join."""
2771
assert isinstance(self.source, bzrlib.weave.Weave)
2772
assert isinstance(self.target, KnitVersionedFile)
2774
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2779
pb = ui.ui_factory.nested_progress_bar()
2781
version_ids = list(version_ids)
2783
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2784
this_versions = set(self.target._index.get_versions())
2785
needed_versions = self.source_ancestry - this_versions
2787
if not needed_versions:
2789
full_list = topo_sort(
2790
self.source.get_parent_map(self.source.versions()))
2792
version_list = [i for i in full_list if (not self.target.has_version(i)
2793
and i in needed_versions)]
2797
total = len(version_list)
2798
parent_map = self.source.get_parent_map(version_list)
2799
for version_id in version_list:
2800
pb.update("Converting to knit", count, total)
2801
parents = parent_map[version_id]
2802
# check that its will be a consistent copy:
2803
for parent in parents:
2804
# if source has the parent, we must already have it
2805
assert (self.target.has_version(parent))
2806
self.target.add_lines(
2807
version_id, parents, self.source.get_lines(version_id))
2814
InterVersionedFile.register_optimiser(WeaveToKnit)
2817
# Deprecated, use PatienceSequenceMatcher instead
2818
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2821
def annotate_knit(knit, revision_id):
2822
"""Annotate a knit with no cached annotations.
2824
This implementation is for knits with no cached annotations.
2825
It will work for knits with cached annotations, but this is not
2828
annotator = _KnitAnnotator(knit)
2829
return iter(annotator.annotate(revision_id))
2832
class _KnitAnnotator(object):
2833
"""Build up the annotations for a text."""
2835
def __init__(self, knit):
2838
# Content objects, differs from fulltexts because of how final newlines
2839
# are treated by knits. the content objects here will always have a
2841
self._fulltext_contents = {}
2843
# Annotated lines of specific revisions
2844
self._annotated_lines = {}
2846
# Track the raw data for nodes that we could not process yet.
2847
# This maps the revision_id of the base to a list of children that will
2848
# annotated from it.
2849
self._pending_children = {}
2851
# Nodes which cannot be extracted
2852
self._ghosts = set()
2854
# Track how many children this node has, so we know if we need to keep
2856
self._annotate_children = {}
2857
self._compression_children = {}
2859
self._all_build_details = {}
2860
# The children => parent revision_id graph
2861
self._revision_id_graph = {}
2863
self._heads_provider = None
2865
self._nodes_to_keep_annotations = set()
2866
self._generations_until_keep = 100
2868
def set_generations_until_keep(self, value):
2869
"""Set the number of generations before caching a node.
2871
Setting this to -1 will cache every merge node, setting this higher
2872
will cache fewer nodes.
2874
self._generations_until_keep = value
2876
def _add_fulltext_content(self, revision_id, content_obj):
2877
self._fulltext_contents[revision_id] = content_obj
2878
# TODO: jam 20080305 It might be good to check the sha1digest here
2879
return content_obj.text()
2881
def _check_parents(self, child, nodes_to_annotate):
2882
"""Check if all parents have been processed.
2884
:param child: A tuple of (rev_id, parents, raw_content)
2885
:param nodes_to_annotate: If child is ready, add it to
2886
nodes_to_annotate, otherwise put it back in self._pending_children
2888
for parent_id in child[1]:
2889
if (parent_id not in self._annotated_lines):
2890
# This parent is present, but another parent is missing
2891
self._pending_children.setdefault(parent_id,
2895
# This one is ready to be processed
2896
nodes_to_annotate.append(child)
2898
def _add_annotation(self, revision_id, fulltext, parent_ids,
2899
left_matching_blocks=None):
2900
"""Add an annotation entry.
2902
All parents should already have been annotated.
2903
:return: A list of children that now have their parents satisfied.
2905
a = self._annotated_lines
2906
annotated_parent_lines = [a[p] for p in parent_ids]
2907
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2908
fulltext, revision_id, left_matching_blocks,
2909
heads_provider=self._get_heads_provider()))
2910
self._annotated_lines[revision_id] = annotated_lines
2911
for p in parent_ids:
2912
ann_children = self._annotate_children[p]
2913
ann_children.remove(revision_id)
2914
if (not ann_children
2915
and p not in self._nodes_to_keep_annotations):
2916
del self._annotated_lines[p]
2917
del self._all_build_details[p]
2918
if p in self._fulltext_contents:
2919
del self._fulltext_contents[p]
2920
# Now that we've added this one, see if there are any pending
2921
# deltas to be done, certainly this parent is finished
2922
nodes_to_annotate = []
2923
for child in self._pending_children.pop(revision_id, []):
2924
self._check_parents(child, nodes_to_annotate)
2925
return nodes_to_annotate
2927
def _get_build_graph(self, revision_id):
2928
"""Get the graphs for building texts and annotations.
2930
The data you need for creating a full text may be different than the
2931
data you need to annotate that text. (At a minimum, you need both
2932
parents to create an annotation, but only need 1 parent to generate the
2935
:return: A list of (revision_id, index_memo) records, suitable for
2936
passing to read_records_iter to start reading in the raw data fro/
2939
if revision_id in self._annotated_lines:
2942
pending = set([revision_id])
2947
# get all pending nodes
2949
this_iteration = pending
2950
build_details = self._knit._index.get_build_details(this_iteration)
2951
self._all_build_details.update(build_details)
2952
# new_nodes = self._knit._index._get_entries(this_iteration)
2954
for rev_id, details in build_details.iteritems():
2955
(index_memo, compression_parent, parents,
2956
record_details) = details
2957
self._revision_id_graph[rev_id] = parents
2958
records.append((rev_id, index_memo))
2959
# Do we actually need to check _annotated_lines?
2960
pending.update(p for p in parents
2961
if p not in self._all_build_details)
2962
if compression_parent:
2963
self._compression_children.setdefault(compression_parent,
2966
for parent in parents:
2967
self._annotate_children.setdefault(parent,
2969
num_gens = generation - kept_generation
2970
if ((num_gens >= self._generations_until_keep)
2971
and len(parents) > 1):
2972
kept_generation = generation
2973
self._nodes_to_keep_annotations.add(rev_id)
2975
missing_versions = this_iteration.difference(build_details.keys())
2976
self._ghosts.update(missing_versions)
2977
for missing_version in missing_versions:
2978
# add a key, no parents
2979
self._revision_id_graph[missing_version] = ()
2980
pending.discard(missing_version) # don't look for it
2981
# XXX: This should probably be a real exception, as it is a data
2983
assert not self._ghosts.intersection(self._compression_children), \
2984
"We cannot have nodes which have a compression parent of a ghost."
2985
# Cleanout anything that depends on a ghost so that we don't wait for
2986
# the ghost to show up
2987
for node in self._ghosts:
2988
if node in self._annotate_children:
2989
# We won't be building this node
2990
del self._annotate_children[node]
2991
# Generally we will want to read the records in reverse order, because
2992
# we find the parent nodes after the children
2996
def _annotate_records(self, records):
2997
"""Build the annotations for the listed records."""
2998
# We iterate in the order read, rather than a strict order requested
2999
# However, process what we can, and put off to the side things that
3000
# still need parents, cleaning them up when those parents are
3002
for (rev_id, record,
3003
digest) in self._knit._data.read_records_iter(records):
3004
if rev_id in self._annotated_lines:
3006
parent_ids = self._revision_id_graph[rev_id]
3007
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3008
details = self._all_build_details[rev_id]
3009
(index_memo, compression_parent, parents,
3010
record_details) = details
3011
nodes_to_annotate = []
3012
# TODO: Remove the punning between compression parents, and
3013
# parent_ids, we should be able to do this without assuming
3015
if len(parent_ids) == 0:
3016
# There are no parents for this node, so just add it
3017
# TODO: This probably needs to be decoupled
3018
assert compression_parent is None
3019
fulltext_content, delta = self._knit.factory.parse_record(
3020
rev_id, record, record_details, None)
3021
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3022
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3023
parent_ids, left_matching_blocks=None))
3025
child = (rev_id, parent_ids, record)
3026
# Check if all the parents are present
3027
self._check_parents(child, nodes_to_annotate)
3028
while nodes_to_annotate:
3029
# Should we use a queue here instead of a stack?
3030
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3031
(index_memo, compression_parent, parents,
3032
record_details) = self._all_build_details[rev_id]
3033
if compression_parent is not None:
3034
comp_children = self._compression_children[compression_parent]
3035
assert rev_id in comp_children
3036
# If there is only 1 child, it is safe to reuse this
3038
reuse_content = (len(comp_children) == 1
3039
and compression_parent not in
3040
self._nodes_to_keep_annotations)
3042
# Remove it from the cache since it will be changing
3043
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3044
# Make sure to copy the fulltext since it might be
3046
parent_fulltext = list(parent_fulltext_content.text())
3048
parent_fulltext_content = self._fulltext_contents[compression_parent]
3049
parent_fulltext = parent_fulltext_content.text()
3050
comp_children.remove(rev_id)
3051
fulltext_content, delta = self._knit.factory.parse_record(
3052
rev_id, record, record_details,
3053
parent_fulltext_content,
3054
copy_base_content=(not reuse_content))
3055
fulltext = self._add_fulltext_content(rev_id,
3057
blocks = KnitContent.get_line_delta_blocks(delta,
3058
parent_fulltext, fulltext)
3060
fulltext_content = self._knit.factory.parse_fulltext(
3062
fulltext = self._add_fulltext_content(rev_id,
3065
nodes_to_annotate.extend(
3066
self._add_annotation(rev_id, fulltext, parent_ids,
3067
left_matching_blocks=blocks))
3069
def _get_heads_provider(self):
3070
"""Create a heads provider for resolving ancestry issues."""
3071
if self._heads_provider is not None:
3072
return self._heads_provider
3073
parent_provider = _mod_graph.DictParentsProvider(
3074
self._revision_id_graph)
3075
graph_obj = _mod_graph.Graph(parent_provider)
3076
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3077
self._heads_provider = head_cache
3080
def annotate(self, revision_id):
3081
"""Return the annotated fulltext at the given revision.
3083
:param revision_id: The revision id for this file
3085
records = self._get_build_graph(revision_id)
3086
if revision_id in self._ghosts:
3087
raise errors.RevisionNotPresent(revision_id, self._knit)
3088
self._annotate_records(records)
3089
return self._annotated_lines[revision_id]
3093
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3095
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
