1
# Copyright (C) 2005, 2006, 2007 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""Knit versionedfile implementation.
19
A knit is a versioned file implementation that supports efficient append only
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
end-marker; simply "end VERSION"
28
delta can be line or full contents.a
29
... the 8's there are the index number of the annotation.
30
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
8 e.set('executable', 'yes')
36
8 if elt.get('executable') == 'yes':
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
43
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
44
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
45
09:33 < lifeless> right
46
09:33 < jrydberg> lifeless: the position and size is the range in the data file
49
so the index sequence is the dictionary compressed sequence number used
50
in the deltas to provide line annotation
55
# 10:16 < lifeless> make partial index writes safe
56
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
57
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
59
# move sha1 out of the content so that join is faster at verifying parents
60
# record content length ?
64
from cStringIO import StringIO
65
from itertools import izip, chain
70
from zlib import Z_DEFAULT_COMPRESSION
73
from bzrlib.lazy_import import lazy_import
74
lazy_import(globals(), """
94
from bzrlib.errors import (
102
RevisionAlreadyPresent,
104
from bzrlib.graph import Graph
105
from bzrlib.osutils import (
111
from bzrlib.symbol_versioning import (
112
DEPRECATED_PARAMETER,
117
from bzrlib.tsort import topo_sort
118
from bzrlib.tuned_gzip import GzipFile, bytes_to_gzip
120
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
124
# TODO: Split out code specific to this format into an associated object.
126
# TODO: Can we put in some kind of value to check that the index and data
127
# files belong together?
129
# TODO: accommodate binaries, perhaps by storing a byte count
131
# TODO: function to check whole file
133
# TODO: atomically append data, then measure backwards from the cursor
134
# position after writing to work out where it was located. we may need to
135
# bypass python file buffering.
137
DATA_SUFFIX = '.knit'
138
INDEX_SUFFIX = '.kndx'
141
class KnitContent(object):
142
"""Content of a knit version to which deltas can be applied."""
145
self._should_strip_eol = False
147
def apply_delta(self, delta, new_version_id):
148
"""Apply delta to this object to become new_version_id."""
149
raise NotImplementedError(self.apply_delta)
151
def cleanup_eol(self, copy_on_mutate=True):
152
if self._should_strip_eol:
154
self._lines = self._lines[:]
155
self.strip_last_line_newline()
157
def line_delta_iter(self, new_lines):
158
"""Generate line-based delta from this content to new_lines."""
159
new_texts = new_lines.text()
160
old_texts = self.text()
161
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
162
for tag, i1, i2, j1, j2 in s.get_opcodes():
165
# ofrom, oto, length, data
166
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
168
def line_delta(self, new_lines):
169
return list(self.line_delta_iter(new_lines))
172
def get_line_delta_blocks(knit_delta, source, target):
173
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
174
target_len = len(target)
177
for s_begin, s_end, t_len, new_text in knit_delta:
178
true_n = s_begin - s_pos
181
# knit deltas do not provide reliable info about whether the
182
# last line of a file matches, due to eol handling.
183
if source[s_pos + n -1] != target[t_pos + n -1]:
186
yield s_pos, t_pos, n
187
t_pos += t_len + true_n
189
n = target_len - t_pos
191
if source[s_pos + n -1] != target[t_pos + n -1]:
194
yield s_pos, t_pos, n
195
yield s_pos + (target_len - t_pos), target_len, 0
198
class AnnotatedKnitContent(KnitContent):
199
"""Annotated content."""
201
def __init__(self, lines):
202
KnitContent.__init__(self)
206
"""Return a list of (origin, text) for each content line."""
207
return list(self._lines)
209
def apply_delta(self, delta, new_version_id):
210
"""Apply delta to this object to become new_version_id."""
213
for start, end, count, delta_lines in delta:
214
lines[offset+start:offset+end] = delta_lines
215
offset = offset + (start - end) + count
217
def strip_last_line_newline(self):
218
line = self._lines[-1][1].rstrip('\n')
219
self._lines[-1] = (self._lines[-1][0], line)
220
self._should_strip_eol = False
224
lines = [text for origin, text in self._lines]
225
except ValueError, e:
226
# most commonly (only?) caused by the internal form of the knit
227
# missing annotation information because of a bug - see thread
229
raise KnitCorrupt(self,
230
"line in annotated knit missing annotation information: %s"
233
if self._should_strip_eol:
234
anno, line = lines[-1]
235
lines[-1] = (anno, line.rstrip('\n'))
239
return AnnotatedKnitContent(self._lines[:])
242
class PlainKnitContent(KnitContent):
243
"""Unannotated content.
245
When annotate[_iter] is called on this content, the same version is reported
246
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
250
def __init__(self, lines, version_id):
251
KnitContent.__init__(self)
253
self._version_id = version_id
256
"""Return a list of (origin, text) for each content line."""
257
return [(self._version_id, line) for line in self._lines]
259
def apply_delta(self, delta, new_version_id):
260
"""Apply delta to this object to become new_version_id."""
263
for start, end, count, delta_lines in delta:
264
lines[offset+start:offset+end] = delta_lines
265
offset = offset + (start - end) + count
266
self._version_id = new_version_id
269
return PlainKnitContent(self._lines[:], self._version_id)
271
def strip_last_line_newline(self):
272
self._lines[-1] = self._lines[-1].rstrip('\n')
273
self._should_strip_eol = False
277
if self._should_strip_eol:
279
lines[-1] = lines[-1].rstrip('\n')
283
class _KnitFactory(object):
284
"""Base class for common Factory functions."""
286
def parse_record(self, version_id, record, record_details,
287
base_content, copy_base_content=True):
288
"""Parse a record into a full content object.
290
:param version_id: The official version id for this content
291
:param record: The data returned by read_records_iter()
292
:param record_details: Details about the record returned by
294
:param base_content: If get_build_details returns a compression_parent,
295
you must return a base_content here, else use None
296
:param copy_base_content: When building from the base_content, decide
297
you can either copy it and return a new object, or modify it in
299
:return: (content, delta) A Content object and possibly a line-delta,
302
method, noeol = record_details
303
if method == 'line-delta':
304
if copy_base_content:
305
content = base_content.copy()
307
content = base_content
308
delta = self.parse_line_delta(record, version_id)
309
content.apply_delta(delta, version_id)
311
content = self.parse_fulltext(record, version_id)
313
content._should_strip_eol = noeol
314
return (content, delta)
317
class KnitAnnotateFactory(_KnitFactory):
318
"""Factory for creating annotated Content objects."""
322
def make(self, lines, version_id):
323
num_lines = len(lines)
324
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
326
def parse_fulltext(self, content, version_id):
327
"""Convert fulltext to internal representation
329
fulltext content is of the format
330
revid(utf8) plaintext\n
331
internal representation is of the format:
334
# TODO: jam 20070209 The tests expect this to be returned as tuples,
335
# but the code itself doesn't really depend on that.
336
# Figure out a way to not require the overhead of turning the
337
# list back into tuples.
338
lines = [tuple(line.split(' ', 1)) for line in content]
339
return AnnotatedKnitContent(lines)
341
def parse_line_delta_iter(self, lines):
342
return iter(self.parse_line_delta(lines))
344
def parse_line_delta(self, lines, version_id, plain=False):
345
"""Convert a line based delta into internal representation.
347
line delta is in the form of:
348
intstart intend intcount
350
revid(utf8) newline\n
351
internal representation is
352
(start, end, count, [1..count tuples (revid, newline)])
354
:param plain: If True, the lines are returned as a plain
355
list without annotations, not as a list of (origin, content) tuples, i.e.
356
(start, end, count, [1..count newline])
363
def cache_and_return(line):
364
origin, text = line.split(' ', 1)
365
return cache.setdefault(origin, origin), text
367
# walk through the lines parsing.
368
# Note that the plain test is explicitly pulled out of the
369
# loop to minimise any performance impact
372
start, end, count = [int(n) for n in header.split(',')]
373
contents = [next().split(' ', 1)[1] for i in xrange(count)]
374
result.append((start, end, count, contents))
377
start, end, count = [int(n) for n in header.split(',')]
378
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
379
result.append((start, end, count, contents))
382
def get_fulltext_content(self, lines):
383
"""Extract just the content lines from a fulltext."""
384
return (line.split(' ', 1)[1] for line in lines)
386
def get_linedelta_content(self, lines):
387
"""Extract just the content from a line delta.
389
This doesn't return all of the extra information stored in a delta.
390
Only the actual content lines.
395
header = header.split(',')
396
count = int(header[2])
397
for i in xrange(count):
398
origin, text = next().split(' ', 1)
401
def lower_fulltext(self, content):
402
"""convert a fulltext content record into a serializable form.
404
see parse_fulltext which this inverts.
406
# TODO: jam 20070209 We only do the caching thing to make sure that
407
# the origin is a valid utf-8 line, eventually we could remove it
408
return ['%s %s' % (o, t) for o, t in content._lines]
410
def lower_line_delta(self, delta):
411
"""convert a delta into a serializable form.
413
See parse_line_delta which this inverts.
415
# TODO: jam 20070209 We only do the caching thing to make sure that
416
# the origin is a valid utf-8 line, eventually we could remove it
418
for start, end, c, lines in delta:
419
out.append('%d,%d,%d\n' % (start, end, c))
420
out.extend(origin + ' ' + text
421
for origin, text in lines)
424
def annotate(self, knit, version_id):
425
content = knit._get_content(version_id)
426
return content.annotate()
429
class KnitPlainFactory(_KnitFactory):
430
"""Factory for creating plain Content objects."""
434
def make(self, lines, version_id):
435
return PlainKnitContent(lines, version_id)
437
def parse_fulltext(self, content, version_id):
438
"""This parses an unannotated fulltext.
440
Note that this is not a noop - the internal representation
441
has (versionid, line) - its just a constant versionid.
443
return self.make(content, version_id)
445
def parse_line_delta_iter(self, lines, version_id):
447
num_lines = len(lines)
448
while cur < num_lines:
451
start, end, c = [int(n) for n in header.split(',')]
452
yield start, end, c, lines[cur:cur+c]
455
def parse_line_delta(self, lines, version_id):
456
return list(self.parse_line_delta_iter(lines, version_id))
458
def get_fulltext_content(self, lines):
459
"""Extract just the content lines from a fulltext."""
462
def get_linedelta_content(self, lines):
463
"""Extract just the content from a line delta.
465
This doesn't return all of the extra information stored in a delta.
466
Only the actual content lines.
471
header = header.split(',')
472
count = int(header[2])
473
for i in xrange(count):
476
def lower_fulltext(self, content):
477
return content.text()
479
def lower_line_delta(self, delta):
481
for start, end, c, lines in delta:
482
out.append('%d,%d,%d\n' % (start, end, c))
486
def annotate(self, knit, version_id):
487
annotator = _KnitAnnotator(knit)
488
return annotator.annotate(version_id)
491
def make_empty_knit(transport, relpath):
492
"""Construct a empty knit at the specified location."""
493
k = make_file_knit(transport, relpath, 'w', KnitPlainFactory)
496
def make_file_knit(name, transport, file_mode=None, access_mode='w',
497
factory=None, delta=True, create=False, create_parent_dir=False,
498
delay_create=False, dir_mode=None, get_scope=None):
499
"""Factory to create a KnitVersionedFile for a .knit/.kndx file pair."""
501
factory = KnitAnnotateFactory()
503
factory = KnitPlainFactory()
504
if get_scope is None:
505
get_scope = lambda:None
506
index = _KnitIndex(transport, name + INDEX_SUFFIX,
507
access_mode, create=create, file_mode=file_mode,
508
create_parent_dir=create_parent_dir, delay_create=delay_create,
509
dir_mode=dir_mode, get_scope=get_scope)
510
access = _KnitAccess(transport, name + DATA_SUFFIX, file_mode,
511
dir_mode, ((create and not len(index)) and delay_create),
513
return KnitVersionedFile(name, transport, factory=factory,
514
create=create, delay_create=delay_create, index=index,
515
access_method=access)
519
"""Return the suffixes used by file based knits."""
520
return [DATA_SUFFIX, INDEX_SUFFIX]
521
make_file_knit.get_suffixes = get_suffixes
524
class KnitVersionedFile(VersionedFile):
525
"""Weave-like structure with faster random access.
527
A knit stores a number of texts and a summary of the relationships
528
between them. Texts are identified by a string version-id. Texts
529
are normally stored and retrieved as a series of lines, but can
530
also be passed as single strings.
532
Lines are stored with the trailing newline (if any) included, to
533
avoid special cases for files with no final newline. Lines are
534
composed of 8-bit characters, not unicode. The combination of
535
these approaches should mean any 'binary' file can be safely
536
stored and retrieved.
539
def __init__(self, relpath, transport, file_mode=None,
540
factory=None, delta=True, create=False, create_parent_dir=False,
541
delay_create=False, dir_mode=None, index=None, access_method=None):
542
"""Construct a knit at location specified by relpath.
544
:param create: If not True, only open an existing knit.
545
:param create_parent_dir: If True, create the parent directory if
546
creating the file fails. (This is used for stores with
547
hash-prefixes that may not exist yet)
548
:param delay_create: The calling code is aware that the knit won't
549
actually be created until the first data is stored.
550
:param index: An index to use for the knit.
552
super(KnitVersionedFile, self).__init__()
553
self.transport = transport
554
self.filename = relpath
555
self.factory = factory or KnitAnnotateFactory()
558
self._max_delta_chain = 200
560
if None in (access_method, index):
561
raise ValueError("No default access_method or index any more")
563
_access = access_method
564
if create and not len(self) and not delay_create:
566
self._data = _KnitData(_access)
569
return '%s(%s)' % (self.__class__.__name__,
570
self.transport.abspath(self.filename))
572
def _check_should_delta(self, first_parents):
573
"""Iterate back through the parent listing, looking for a fulltext.
575
This is used when we want to decide whether to add a delta or a new
576
fulltext. It searches for _max_delta_chain parents. When it finds a
577
fulltext parent, it sees if the total size of the deltas leading up to
578
it is large enough to indicate that we want a new full text anyway.
580
Return True if we should create a new delta, False if we should use a
585
delta_parents = first_parents
586
for count in xrange(self._max_delta_chain):
587
parent = delta_parents[0]
588
method = self._index.get_method(parent)
589
index, pos, size = self._index.get_position(parent)
590
if method == 'fulltext':
594
delta_parents = self._index.get_parent_map([parent])[parent]
596
# We couldn't find a fulltext, so we must create a new one
599
return fulltext_size > delta_size
601
def _check_write_ok(self):
602
return self._index._check_write_ok()
604
def _add_raw_records(self, records, data):
605
"""Add all the records 'records' with data pre-joined in 'data'.
607
:param records: A list of tuples(version_id, options, parents, size).
608
:param data: The data for the records. When it is written, the records
609
are adjusted to have pos pointing into data by the sum of
610
the preceding records sizes.
613
raw_record_sizes = [record[3] for record in records]
614
positions = self._data.add_raw_records(raw_record_sizes, data)
617
for (version_id, options, parents, size), access_memo in zip(
619
index_entries.append((version_id, options, access_memo, parents))
621
self._index.add_versions(index_entries)
623
def copy_to(self, name, transport):
624
"""See VersionedFile.copy_to()."""
625
# copy the current index to a temp index to avoid racing with local
627
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
628
self.transport.get(self._index._filename))
630
f = self._data._open_file()
632
transport.put_file(name + DATA_SUFFIX, f)
635
# move the copied index into place
636
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
638
def get_data_stream(self, required_versions):
639
"""Get a data stream for the specified versions.
641
Versions may be returned in any order, not necessarily the order
642
specified. They are returned in a partial order by compression
643
parent, so that the deltas can be applied as the data stream is
644
inserted; however note that compression parents will not be sent
645
unless they were specifically requested, as the client may already
648
:param required_versions: The exact set of versions to be extracted.
649
Unlike some other knit methods, this is not used to generate a
650
transitive closure, rather it is used precisely as given.
652
:returns: format_signature, list of (version, options, length, parents),
655
required_version_set = frozenset(required_versions)
657
# list of revisions that can just be sent without waiting for their
660
# map from revision to the children based on it
662
# first, read all relevant index data, enough to sort into the right
664
for version_id in required_versions:
665
options = self._index.get_options(version_id)
666
parents = self._index.get_parents_with_ghosts(version_id)
667
index_memo = self._index.get_position(version_id)
668
version_index[version_id] = (index_memo, options, parents)
669
if ('line-delta' in options
670
and parents[0] in required_version_set):
671
# must wait until the parent has been sent
672
deferred.setdefault(parents[0], []). \
675
# either a fulltext, or a delta whose parent the client did
676
# not ask for and presumably already has
677
ready_to_send.append(version_id)
678
# build a list of results to return, plus instructions for data to
680
copy_queue_records = []
681
temp_version_list = []
683
# XXX: pushing and popping lists may be a bit inefficient
684
version_id = ready_to_send.pop(0)
685
(index_memo, options, parents) = version_index[version_id]
686
copy_queue_records.append((version_id, index_memo))
687
none, data_pos, data_size = index_memo
688
temp_version_list.append((version_id, options, data_size,
690
if version_id in deferred:
691
# now we can send all the children of this revision - we could
692
# put them in anywhere, but we hope that sending them soon
693
# after the fulltext will give good locality in the receiver
694
ready_to_send[:0] = deferred.pop(version_id)
695
if not (len(deferred) == 0):
696
raise AssertionError("Still have compressed child versions waiting to be sent")
697
# XXX: The stream format is such that we cannot stream it - we have to
698
# know the length of all the data a-priori.
700
result_version_list = []
701
for (version_id, raw_data), \
702
(version_id2, options, _, parents) in \
703
izip(self._data.read_records_iter_raw(copy_queue_records),
705
if not (version_id == version_id2):
706
raise AssertionError('logic error, inconsistent results')
707
raw_datum.append(raw_data)
708
result_version_list.append(
709
(version_id, options, len(raw_data), parents))
710
# provide a callback to get data incrementally.
711
pseudo_file = StringIO(''.join(raw_datum))
714
return pseudo_file.read()
716
return pseudo_file.read(length)
717
return (self.get_format_signature(), result_version_list, read)
719
def _extract_blocks(self, version_id, source, target):
720
if self._index.get_method(version_id) != 'line-delta':
722
parent, sha1, noeol, delta = self.get_delta(version_id)
723
return KnitContent.get_line_delta_blocks(delta, source, target)
725
def get_delta(self, version_id):
726
"""Get a delta for constructing version from some other version."""
727
self.check_not_reserved_id(version_id)
728
parents = self.get_parent_map([version_id])[version_id]
733
index_memo = self._index.get_position(version_id)
734
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
735
noeol = 'no-eol' in self._index.get_options(version_id)
736
if 'fulltext' == self._index.get_method(version_id):
737
new_content = self.factory.parse_fulltext(data, version_id)
738
if parent is not None:
739
reference_content = self._get_content(parent)
740
old_texts = reference_content.text()
743
new_texts = new_content.text()
744
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
746
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
748
delta = self.factory.parse_line_delta(data, version_id)
749
return parent, sha1, noeol, delta
751
def get_format_signature(self):
752
"""See VersionedFile.get_format_signature()."""
753
if self.factory.annotated:
754
annotated_part = "annotated"
756
annotated_part = "plain"
757
return "knit-%s" % (annotated_part,)
759
@deprecated_method(one_four)
760
def get_graph_with_ghosts(self):
761
"""See VersionedFile.get_graph_with_ghosts()."""
762
return self.get_parent_map(self.versions())
764
def get_sha1s(self, version_ids):
765
"""See VersionedFile.get_sha1s()."""
766
record_map = self._get_record_map(version_ids)
767
# record entry 2 is the 'digest'.
768
return [record_map[v][2] for v in version_ids]
770
@deprecated_method(one_four)
771
def has_ghost(self, version_id):
772
"""True if there is a ghost reference in the file to version_id."""
774
if self.has_version(version_id):
776
# optimisable if needed by memoising the _ghosts set.
777
items = self.get_parent_map(self.versions())
778
for parents in items.itervalues():
779
for parent in parents:
780
if parent == version_id and parent not in items:
784
def insert_data_stream(self, (format, data_list, reader_callable)):
785
"""Insert knit records from a data stream into this knit.
787
If a version in the stream is already present in this knit, it will not
788
be inserted a second time. It will be checked for consistency with the
789
stored version however, and may cause a KnitCorrupt error to be raised
790
if the data in the stream disagrees with the already stored data.
792
:seealso: get_data_stream
794
if format != self.get_format_signature():
795
if 'knit' in debug.debug_flags:
797
'incompatible format signature inserting to %r', self)
798
source = self._knit_from_datastream(
799
(format, data_list, reader_callable))
803
for version_id, options, length, parents in data_list:
804
if self.has_version(version_id):
805
# First check: the list of parents.
806
my_parents = self.get_parents_with_ghosts(version_id)
807
if tuple(my_parents) != tuple(parents):
808
# XXX: KnitCorrupt is not quite the right exception here.
811
'parents list %r from data stream does not match '
812
'already recorded parents %r for %s'
813
% (parents, my_parents, version_id))
815
# Also check the SHA-1 of the fulltext this content will
817
raw_data = reader_callable(length)
818
my_fulltext_sha1 = self.get_sha1s([version_id])[0]
819
df, rec = self._data._parse_record_header(version_id, raw_data)
820
stream_fulltext_sha1 = rec[3]
821
if my_fulltext_sha1 != stream_fulltext_sha1:
822
# Actually, we don't know if it's this knit that's corrupt,
823
# or the data stream we're trying to insert.
825
self.filename, 'sha-1 does not match %s' % version_id)
827
if 'line-delta' in options:
828
# Make sure that this knit record is actually useful: a
829
# line-delta is no use unless we have its parent.
830
# Fetching from a broken repository with this problem
831
# shouldn't break the target repository.
833
# See https://bugs.launchpad.net/bzr/+bug/164443
834
if not self._index.has_version(parents[0]):
837
'line-delta from stream '
840
'missing parent %s\n'
841
'Try running "bzr check" '
842
'on the source repository, and "bzr reconcile" '
844
(version_id, parents[0]))
846
# We received a line-delta record for a non-delta knit.
847
# Convert it to a fulltext.
848
gzip_bytes = reader_callable(length)
849
self._convert_line_delta_to_fulltext(
850
gzip_bytes, version_id, parents)
853
self._add_raw_records(
854
[(version_id, options, parents, length)],
855
reader_callable(length))
857
def _convert_line_delta_to_fulltext(self, gzip_bytes, version_id, parents):
858
lines, sha1 = self._data._parse_record(version_id, gzip_bytes)
859
delta = self.factory.parse_line_delta(lines, version_id)
860
content = self.factory.make(self.get_lines(parents[0]), parents[0])
861
content.apply_delta(delta, version_id)
862
digest, len, content = self.add_lines(
863
version_id, parents, content.text())
865
raise errors.VersionedFileInvalidChecksum(version_id)
867
def _knit_from_datastream(self, (format, data_list, reader_callable)):
868
"""Create a knit object from a data stream.
870
This method exists to allow conversion of data streams that do not
871
match the signature of this knit. Generally it will be slower and use
872
more memory to use this method to insert data, but it will work.
874
:seealso: get_data_stream for details on datastreams.
875
:return: A knit versioned file which can be used to join the datastream
878
if format == "knit-plain":
879
factory = KnitPlainFactory()
880
elif format == "knit-annotated":
881
factory = KnitAnnotateFactory()
883
raise errors.KnitDataStreamUnknown(format)
884
index = _StreamIndex(data_list, self._index)
885
access = _StreamAccess(reader_callable, index, self, factory)
886
return KnitVersionedFile(self.filename, self.transport,
887
factory=factory, index=index, access_method=access)
890
"""See VersionedFile.versions."""
891
if 'evil' in debug.debug_flags:
892
trace.mutter_callsite(2, "versions scales with size of history")
893
return self._index.get_versions()
895
def has_version(self, version_id):
896
"""See VersionedFile.has_version."""
897
if 'evil' in debug.debug_flags:
898
trace.mutter_callsite(2, "has_version is a LBYL scenario")
899
return self._index.has_version(version_id)
901
__contains__ = has_version
903
def _merge_annotations(self, content, parents, parent_texts={},
904
delta=None, annotated=None,
905
left_matching_blocks=None):
906
"""Merge annotations for content. This is done by comparing
907
the annotations based on changed to the text.
909
if left_matching_blocks is not None:
910
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
914
for parent_id in parents:
915
merge_content = self._get_content(parent_id, parent_texts)
916
if (parent_id == parents[0] and delta_seq is not None):
919
seq = patiencediff.PatienceSequenceMatcher(
920
None, merge_content.text(), content.text())
921
for i, j, n in seq.get_matching_blocks():
924
# this appears to copy (origin, text) pairs across to the
925
# new content for any line that matches the last-checked
927
content._lines[j:j+n] = merge_content._lines[i:i+n]
929
if delta_seq is None:
930
reference_content = self._get_content(parents[0], parent_texts)
931
new_texts = content.text()
932
old_texts = reference_content.text()
933
delta_seq = patiencediff.PatienceSequenceMatcher(
934
None, old_texts, new_texts)
935
return self._make_line_delta(delta_seq, content)
937
def _make_line_delta(self, delta_seq, new_content):
938
"""Generate a line delta from delta_seq and new_content."""
940
for op in delta_seq.get_opcodes():
943
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
946
def _get_components_positions(self, version_ids):
947
"""Produce a map of position data for the components of versions.
949
This data is intended to be used for retrieving the knit records.
951
A dict of version_id to (record_details, index_memo, next, parents) is
953
method is the way referenced data should be applied.
954
index_memo is the handle to pass to the data access to actually get the
956
next is the build-parent of the version, or None for fulltexts.
957
parents is the version_ids of the parents of this version
960
pending_components = version_ids
961
while pending_components:
962
build_details = self._index.get_build_details(pending_components)
963
current_components = set(pending_components)
964
pending_components = set()
965
for version_id, details in build_details.iteritems():
966
(index_memo, compression_parent, parents,
967
record_details) = details
968
method = record_details[0]
969
if compression_parent is not None:
970
pending_components.add(compression_parent)
971
component_data[version_id] = (record_details, index_memo,
973
missing = current_components.difference(build_details)
975
raise errors.RevisionNotPresent(missing.pop(), self.filename)
976
return component_data
978
def _get_content(self, version_id, parent_texts={}):
979
"""Returns a content object that makes up the specified
981
cached_version = parent_texts.get(version_id, None)
982
if cached_version is not None:
983
if not self.has_version(version_id):
984
raise RevisionNotPresent(version_id, self.filename)
985
return cached_version
987
text_map, contents_map = self._get_content_maps([version_id])
988
return contents_map[version_id]
990
def _check_versions_present(self, version_ids):
991
"""Check that all specified versions are present."""
992
self._index.check_versions_present(version_ids)
994
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
995
nostore_sha, random_id, check_content, left_matching_blocks):
996
"""See VersionedFile.add_lines_with_ghosts()."""
997
self._check_add(version_id, lines, random_id, check_content)
998
return self._add(version_id, lines, parents, self.delta,
999
parent_texts, left_matching_blocks, nostore_sha, random_id)
1001
def _add_lines(self, version_id, parents, lines, parent_texts,
1002
left_matching_blocks, nostore_sha, random_id, check_content):
1003
"""See VersionedFile.add_lines."""
1004
self._check_add(version_id, lines, random_id, check_content)
1005
self._check_versions_present(parents)
1006
return self._add(version_id, lines[:], parents, self.delta,
1007
parent_texts, left_matching_blocks, nostore_sha, random_id)
1009
def _check_add(self, version_id, lines, random_id, check_content):
1010
"""check that version_id and lines are safe to add."""
1011
if contains_whitespace(version_id):
1012
raise InvalidRevisionId(version_id, self.filename)
1013
self.check_not_reserved_id(version_id)
1014
# Technically this could be avoided if we are happy to allow duplicate
1015
# id insertion when other things than bzr core insert texts, but it
1016
# seems useful for folk using the knit api directly to have some safety
1017
# blanket that we can disable.
1018
if not random_id and self.has_version(version_id):
1019
raise RevisionAlreadyPresent(version_id, self.filename)
1021
self._check_lines_not_unicode(lines)
1022
self._check_lines_are_lines(lines)
1024
def _add(self, version_id, lines, parents, delta, parent_texts,
1025
left_matching_blocks, nostore_sha, random_id):
1026
"""Add a set of lines on top of version specified by parents.
1028
If delta is true, compress the text as a line-delta against
1031
Any versions not present will be converted into ghosts.
1033
# first thing, if the content is something we don't need to store, find
1035
line_bytes = ''.join(lines)
1036
digest = sha_string(line_bytes)
1037
if nostore_sha == digest:
1038
raise errors.ExistingContent
1040
present_parents = []
1041
if parent_texts is None:
1043
for parent in parents:
1044
if self.has_version(parent):
1045
present_parents.append(parent)
1047
# can only compress against the left most present parent.
1049
(len(present_parents) == 0 or
1050
present_parents[0] != parents[0])):
1053
text_length = len(line_bytes)
1056
if lines[-1][-1] != '\n':
1057
# copy the contents of lines.
1059
options.append('no-eol')
1060
lines[-1] = lines[-1] + '\n'
1064
# To speed the extract of texts the delta chain is limited
1065
# to a fixed number of deltas. This should minimize both
1066
# I/O and the time spend applying deltas.
1067
delta = self._check_should_delta(present_parents)
1069
content = self.factory.make(lines, version_id)
1070
if delta or (self.factory.annotated and len(present_parents) > 0):
1071
# Merge annotations from parent texts if needed.
1072
delta_hunks = self._merge_annotations(content, present_parents,
1073
parent_texts, delta, self.factory.annotated,
1074
left_matching_blocks)
1077
options.append('line-delta')
1078
store_lines = self.factory.lower_line_delta(delta_hunks)
1079
size, bytes = self._data._record_to_data(version_id, digest,
1082
options.append('fulltext')
1083
# isinstance is slower and we have no hierarchy.
1084
if self.factory.__class__ == KnitPlainFactory:
1085
# Use the already joined bytes saving iteration time in
1087
size, bytes = self._data._record_to_data(version_id, digest,
1088
lines, [line_bytes])
1090
# get mixed annotation + content and feed it into the
1092
store_lines = self.factory.lower_fulltext(content)
1093
size, bytes = self._data._record_to_data(version_id, digest,
1096
access_memo = self._data.add_raw_records([size], bytes)[0]
1097
self._index.add_versions(
1098
((version_id, options, access_memo, parents),),
1099
random_id=random_id)
1100
return digest, text_length, content
1102
def check(self, progress_bar=None):
1103
"""See VersionedFile.check()."""
1105
def get_lines(self, version_id):
1106
"""See VersionedFile.get_lines()."""
1107
return self.get_line_list([version_id])[0]
1109
def _get_record_map(self, version_ids):
1110
"""Produce a dictionary of knit records.
1112
:return: {version_id:(record, record_details, digest, next)}
1114
data returned from read_records
1116
opaque information to pass to parse_record
1118
SHA1 digest of the full text after all steps are done
1120
build-parent of the version, i.e. the leftmost ancestor.
1121
Will be None if the record is not a delta.
1123
position_map = self._get_components_positions(version_ids)
1124
# c = component_id, r = record_details, i_m = index_memo, n = next
1125
records = [(c, i_m) for c, (r, i_m, n)
1126
in position_map.iteritems()]
1128
for component_id, record, digest in \
1129
self._data.read_records_iter(records):
1130
(record_details, index_memo, next) = position_map[component_id]
1131
record_map[component_id] = record, record_details, digest, next
1135
def get_text(self, version_id):
1136
"""See VersionedFile.get_text"""
1137
return self.get_texts([version_id])[0]
1139
def get_texts(self, version_ids):
1140
return [''.join(l) for l in self.get_line_list(version_ids)]
1142
def get_line_list(self, version_ids):
1143
"""Return the texts of listed versions as a list of strings."""
1144
for version_id in version_ids:
1145
self.check_not_reserved_id(version_id)
1146
text_map, content_map = self._get_content_maps(version_ids)
1147
return [text_map[v] for v in version_ids]
1149
_get_lf_split_line_list = get_line_list
1151
def _get_content_maps(self, version_ids):
1152
"""Produce maps of text and KnitContents
1154
:return: (text_map, content_map) where text_map contains the texts for
1155
the requested versions and content_map contains the KnitContents.
1156
Both dicts take version_ids as their keys.
1158
# FUTURE: This function could be improved for the 'extract many' case
1159
# by tracking each component and only doing the copy when the number of
1160
# children than need to apply delta's to it is > 1 or it is part of the
1162
version_ids = list(version_ids)
1163
multiple_versions = len(version_ids) != 1
1164
record_map = self._get_record_map(version_ids)
1169
for version_id in version_ids:
1172
while cursor is not None:
1173
record, record_details, digest, next = record_map[cursor]
1174
components.append((cursor, record, record_details, digest))
1175
if cursor in content_map:
1180
for (component_id, record, record_details,
1181
digest) in reversed(components):
1182
if component_id in content_map:
1183
content = content_map[component_id]
1185
content, delta = self.factory.parse_record(version_id,
1186
record, record_details, content,
1187
copy_base_content=multiple_versions)
1188
if multiple_versions:
1189
content_map[component_id] = content
1191
content.cleanup_eol(copy_on_mutate=multiple_versions)
1192
final_content[version_id] = content
1194
# digest here is the digest from the last applied component.
1195
text = content.text()
1196
actual_sha = sha_strings(text)
1197
if actual_sha != digest:
1198
raise KnitCorrupt(self.filename,
1200
'\n of reconstructed text does not match'
1202
'\n for version %s' %
1203
(actual_sha, digest, version_id))
1204
text_map[version_id] = text
1205
return text_map, final_content
1207
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1209
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1210
if version_ids is None:
1211
version_ids = self.versions()
1213
pb = progress.DummyProgress()
1214
# we don't care about inclusions, the caller cares.
1215
# but we need to setup a list of records to visit.
1216
# we need version_id, position, length
1217
version_id_records = []
1218
requested_versions = set(version_ids)
1219
# filter for available versions
1220
for version_id in requested_versions:
1221
if not self.has_version(version_id):
1222
raise RevisionNotPresent(version_id, self.filename)
1223
# get a in-component-order queue:
1224
for version_id in self.versions():
1225
if version_id in requested_versions:
1226
index_memo = self._index.get_position(version_id)
1227
version_id_records.append((version_id, index_memo))
1229
total = len(version_id_records)
1230
for version_idx, (version_id, data, sha_value) in \
1231
enumerate(self._data.read_records_iter(version_id_records)):
1232
pb.update('Walking content.', version_idx, total)
1233
method = self._index.get_method(version_id)
1234
if method == 'fulltext':
1235
line_iterator = self.factory.get_fulltext_content(data)
1236
elif method == 'line-delta':
1237
line_iterator = self.factory.get_linedelta_content(data)
1239
raise ValueError('invalid method %r' % (method,))
1240
# XXX: It might be more efficient to yield (version_id,
1241
# line_iterator) in the future. However for now, this is a simpler
1242
# change to integrate into the rest of the codebase. RBC 20071110
1243
for line in line_iterator:
1244
yield line, version_id
1246
pb.update('Walking content.', total, total)
1248
def num_versions(self):
1249
"""See VersionedFile.num_versions()."""
1250
return self._index.num_versions()
1252
__len__ = num_versions
1254
def annotate(self, version_id):
1255
"""See VersionedFile.annotate."""
1256
return self.factory.annotate(self, version_id)
1258
def get_parent_map(self, version_ids):
1259
"""See VersionedFile.get_parent_map."""
1260
return self._index.get_parent_map(version_ids)
1262
def get_ancestry(self, versions, topo_sorted=True):
1263
"""See VersionedFile.get_ancestry."""
1264
if isinstance(versions, basestring):
1265
versions = [versions]
1268
return self._index.get_ancestry(versions, topo_sorted)
1270
def get_ancestry_with_ghosts(self, versions):
1271
"""See VersionedFile.get_ancestry_with_ghosts."""
1272
if isinstance(versions, basestring):
1273
versions = [versions]
1276
return self._index.get_ancestry_with_ghosts(versions)
1278
def plan_merge(self, ver_a, ver_b):
1279
"""See VersionedFile.plan_merge."""
1280
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1281
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1282
annotated_a = self.annotate(ver_a)
1283
annotated_b = self.annotate(ver_b)
1284
return merge._plan_annotate_merge(annotated_a, annotated_b,
1285
ancestors_a, ancestors_b)
1288
class _KnitComponentFile(object):
1289
"""One of the files used to implement a knit database"""
1291
def __init__(self, transport, filename, mode, file_mode=None,
1292
create_parent_dir=False, dir_mode=None):
1293
self._transport = transport
1294
self._filename = filename
1296
self._file_mode = file_mode
1297
self._dir_mode = dir_mode
1298
self._create_parent_dir = create_parent_dir
1299
self._need_to_create = False
1301
def _full_path(self):
1302
"""Return the full path to this file."""
1303
return self._transport.base + self._filename
1305
def check_header(self, fp):
1306
line = fp.readline()
1308
# An empty file can actually be treated as though the file doesn't
1310
raise errors.NoSuchFile(self._full_path())
1311
if line != self.HEADER:
1312
raise KnitHeaderError(badline=line,
1313
filename=self._transport.abspath(self._filename))
1316
return '%s(%s)' % (self.__class__.__name__, self._filename)
1319
class _KnitIndex(_KnitComponentFile):
1320
"""Manages knit index file.
1322
The index is already kept in memory and read on startup, to enable
1323
fast lookups of revision information. The cursor of the index
1324
file is always pointing to the end, making it easy to append
1327
_cache is a cache for fast mapping from version id to a Index
1330
_history is a cache for fast mapping from indexes to version ids.
1332
The index data format is dictionary compressed when it comes to
1333
parent references; a index entry may only have parents that with a
1334
lover index number. As a result, the index is topological sorted.
1336
Duplicate entries may be written to the index for a single version id
1337
if this is done then the latter one completely replaces the former:
1338
this allows updates to correct version and parent information.
1339
Note that the two entries may share the delta, and that successive
1340
annotations and references MUST point to the first entry.
1342
The index file on disc contains a header, followed by one line per knit
1343
record. The same revision can be present in an index file more than once.
1344
The first occurrence gets assigned a sequence number starting from 0.
1346
The format of a single line is
1347
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1348
REVISION_ID is a utf8-encoded revision id
1349
FLAGS is a comma separated list of flags about the record. Values include
1350
no-eol, line-delta, fulltext.
1351
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1352
that the the compressed data starts at.
1353
LENGTH is the ascii representation of the length of the data file.
1354
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1356
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1357
revision id already in the knit that is a parent of REVISION_ID.
1358
The ' :' marker is the end of record marker.
1361
when a write is interrupted to the index file, it will result in a line
1362
that does not end in ' :'. If the ' :' is not present at the end of a line,
1363
or at the end of the file, then the record that is missing it will be
1364
ignored by the parser.
1366
When writing new records to the index file, the data is preceded by '\n'
1367
to ensure that records always start on new lines even if the last write was
1368
interrupted. As a result its normal for the last line in the index to be
1369
missing a trailing newline. One can be added with no harmful effects.
1372
HEADER = "# bzr knit index 8\n"
1374
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1375
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1377
def _cache_version(self, version_id, options, pos, size, parents):
1378
"""Cache a version record in the history array and index cache.
1380
This is inlined into _load_data for performance. KEEP IN SYNC.
1381
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1384
# only want the _history index to reference the 1st index entry
1386
if version_id not in self._cache:
1387
index = len(self._history)
1388
self._history.append(version_id)
1390
index = self._cache[version_id][5]
1391
self._cache[version_id] = (version_id,
1398
def _check_write_ok(self):
1399
if self._get_scope() != self._scope:
1400
raise errors.OutSideTransaction()
1401
if self._mode != 'w':
1402
raise errors.ReadOnlyObjectDirtiedError(self)
1404
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1405
create_parent_dir=False, delay_create=False, dir_mode=None,
1407
_KnitComponentFile.__init__(self, transport, filename, mode,
1408
file_mode=file_mode,
1409
create_parent_dir=create_parent_dir,
1412
# position in _history is the 'official' index for a revision
1413
# but the values may have come from a newer entry.
1414
# so - wc -l of a knit index is != the number of unique names
1418
fp = self._transport.get(self._filename)
1420
# _load_data may raise NoSuchFile if the target knit is
1422
_load_data(self, fp)
1426
if mode != 'w' or not create:
1429
self._need_to_create = True
1431
self._transport.put_bytes_non_atomic(
1432
self._filename, self.HEADER, mode=self._file_mode)
1433
self._scope = get_scope()
1434
self._get_scope = get_scope
1436
def get_ancestry(self, versions, topo_sorted=True):
1437
"""See VersionedFile.get_ancestry."""
1438
# get a graph of all the mentioned versions:
1440
pending = set(versions)
1443
version = pending.pop()
1446
parents = [p for p in cache[version][4] if p in cache]
1448
raise RevisionNotPresent(version, self._filename)
1449
# if not completed and not a ghost
1450
pending.update([p for p in parents if p not in graph])
1451
graph[version] = parents
1454
return topo_sort(graph.items())
1456
def get_ancestry_with_ghosts(self, versions):
1457
"""See VersionedFile.get_ancestry_with_ghosts."""
1458
# get a graph of all the mentioned versions:
1459
self.check_versions_present(versions)
1462
pending = set(versions)
1464
version = pending.pop()
1466
parents = cache[version][4]
1472
pending.update([p for p in parents if p not in graph])
1473
graph[version] = parents
1474
return topo_sort(graph.items())
1476
def get_build_details(self, version_ids):
1477
"""Get the method, index_memo and compression parent for version_ids.
1479
Ghosts are omitted from the result.
1481
:param version_ids: An iterable of version_ids.
1482
:return: A dict of version_id:(index_memo, compression_parent,
1483
parents, record_details).
1485
opaque structure to pass to read_records to extract the raw
1488
Content that this record is built upon, may be None
1490
Logical parents of this node
1492
extra information about the content which needs to be passed to
1493
Factory.parse_record
1496
for version_id in version_ids:
1497
if version_id not in self._cache:
1498
# ghosts are omitted
1500
method = self.get_method(version_id)
1501
parents = self.get_parents_with_ghosts(version_id)
1502
if method == 'fulltext':
1503
compression_parent = None
1505
compression_parent = parents[0]
1506
noeol = 'no-eol' in self.get_options(version_id)
1507
index_memo = self.get_position(version_id)
1508
result[version_id] = (index_memo, compression_parent,
1509
parents, (method, noeol))
1512
def num_versions(self):
1513
return len(self._history)
1515
__len__ = num_versions
1517
def get_versions(self):
1518
"""Get all the versions in the file. not topologically sorted."""
1519
return self._history
1521
def _version_list_to_index(self, versions):
1524
for version in versions:
1525
if version in cache:
1526
# -- inlined lookup() --
1527
result_list.append(str(cache[version][5]))
1528
# -- end lookup () --
1530
result_list.append('.' + version)
1531
return ' '.join(result_list)
1533
def add_version(self, version_id, options, index_memo, parents):
1534
"""Add a version record to the index."""
1535
self.add_versions(((version_id, options, index_memo, parents),))
1537
def add_versions(self, versions, random_id=False):
1538
"""Add multiple versions to the index.
1540
:param versions: a list of tuples:
1541
(version_id, options, pos, size, parents).
1542
:param random_id: If True the ids being added were randomly generated
1543
and no check for existence will be performed.
1546
orig_history = self._history[:]
1547
orig_cache = self._cache.copy()
1550
for version_id, options, (index, pos, size), parents in versions:
1551
line = "\n%s %s %s %s %s :" % (version_id,
1555
self._version_list_to_index(parents))
1557
self._cache_version(version_id, options, pos, size, tuple(parents))
1558
if not self._need_to_create:
1559
self._transport.append_bytes(self._filename, ''.join(lines))
1562
sio.write(self.HEADER)
1563
sio.writelines(lines)
1565
self._transport.put_file_non_atomic(self._filename, sio,
1566
create_parent_dir=self._create_parent_dir,
1567
mode=self._file_mode,
1568
dir_mode=self._dir_mode)
1569
self._need_to_create = False
1571
# If any problems happen, restore the original values and re-raise
1572
self._history = orig_history
1573
self._cache = orig_cache
1576
def has_version(self, version_id):
1577
"""True if the version is in the index."""
1578
return version_id in self._cache
1580
def get_position(self, version_id):
1581
"""Return details needed to access the version.
1583
.kndx indices do not support split-out data, so return None for the
1586
:return: a tuple (None, data position, size) to hand to the access
1587
logic to get the record.
1589
entry = self._cache[version_id]
1590
return None, entry[2], entry[3]
1592
def get_method(self, version_id):
1593
"""Return compression method of specified version."""
1595
options = self._cache[version_id][1]
1597
raise RevisionNotPresent(version_id, self._filename)
1598
if 'fulltext' in options:
1601
if 'line-delta' not in options:
1602
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1605
def get_options(self, version_id):
1606
"""Return a list representing options.
1610
return self._cache[version_id][1]
1612
def get_parent_map(self, version_ids):
1613
"""Passed through to by KnitVersionedFile.get_parent_map."""
1615
for version_id in version_ids:
1617
result[version_id] = tuple(self._cache[version_id][4])
1622
def get_parents_with_ghosts(self, version_id):
1623
"""Return parents of specified version with ghosts."""
1625
return self.get_parent_map([version_id])[version_id]
1627
raise RevisionNotPresent(version_id, self)
1629
def check_versions_present(self, version_ids):
1630
"""Check that all specified versions are present."""
1632
for version_id in version_ids:
1633
if version_id not in cache:
1634
raise RevisionNotPresent(version_id, self._filename)
1637
class KnitGraphIndex(object):
1638
"""A knit index that builds on GraphIndex."""
1640
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1641
"""Construct a KnitGraphIndex on a graph_index.
1643
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1644
:param deltas: Allow delta-compressed records.
1645
:param add_callback: If not None, allow additions to the index and call
1646
this callback with a list of added GraphIndex nodes:
1647
[(node, value, node_refs), ...]
1648
:param parents: If True, record knits parents, if not do not record
1651
self._graph_index = graph_index
1652
self._deltas = deltas
1653
self._add_callback = add_callback
1654
self._parents = parents
1655
if deltas and not parents:
1656
raise KnitCorrupt(self, "Cannot do delta compression without "
1659
def _check_write_ok(self):
1662
def _get_entries(self, keys, check_present=False):
1663
"""Get the entries for keys.
1665
:param keys: An iterable of index keys, - 1-tuples.
1670
for node in self._graph_index.iter_entries(keys):
1672
found_keys.add(node[1])
1674
# adapt parentless index to the rest of the code.
1675
for node in self._graph_index.iter_entries(keys):
1676
yield node[0], node[1], node[2], ()
1677
found_keys.add(node[1])
1679
missing_keys = keys.difference(found_keys)
1681
raise RevisionNotPresent(missing_keys.pop(), self)
1683
def _present_keys(self, version_ids):
1685
node[1] for node in self._get_entries(version_ids)])
1687
def _parentless_ancestry(self, versions):
1688
"""Honour the get_ancestry API for parentless knit indices."""
1689
wanted_keys = self._version_ids_to_keys(versions)
1690
present_keys = self._present_keys(wanted_keys)
1691
missing = set(wanted_keys).difference(present_keys)
1693
raise RevisionNotPresent(missing.pop(), self)
1694
return list(self._keys_to_version_ids(present_keys))
1696
def get_ancestry(self, versions, topo_sorted=True):
1697
"""See VersionedFile.get_ancestry."""
1698
if not self._parents:
1699
return self._parentless_ancestry(versions)
1700
# XXX: This will do len(history) index calls - perhaps
1701
# it should be altered to be a index core feature?
1702
# get a graph of all the mentioned versions:
1705
versions = self._version_ids_to_keys(versions)
1706
pending = set(versions)
1708
# get all pending nodes
1709
this_iteration = pending
1710
new_nodes = self._get_entries(this_iteration)
1713
for (index, key, value, node_refs) in new_nodes:
1714
# dont ask for ghosties - otherwise
1715
# we we can end up looping with pending
1716
# being entirely ghosted.
1717
graph[key] = [parent for parent in node_refs[0]
1718
if parent not in ghosts]
1720
for parent in graph[key]:
1721
# dont examine known nodes again
1726
ghosts.update(this_iteration.difference(found))
1727
if versions.difference(graph):
1728
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1730
result_keys = topo_sort(graph.items())
1732
result_keys = graph.iterkeys()
1733
return [key[0] for key in result_keys]
1735
def get_ancestry_with_ghosts(self, versions):
1736
"""See VersionedFile.get_ancestry."""
1737
if not self._parents:
1738
return self._parentless_ancestry(versions)
1739
# XXX: This will do len(history) index calls - perhaps
1740
# it should be altered to be a index core feature?
1741
# get a graph of all the mentioned versions:
1743
versions = self._version_ids_to_keys(versions)
1744
pending = set(versions)
1746
# get all pending nodes
1747
this_iteration = pending
1748
new_nodes = self._get_entries(this_iteration)
1750
for (index, key, value, node_refs) in new_nodes:
1751
graph[key] = node_refs[0]
1753
for parent in graph[key]:
1754
# dont examine known nodes again
1758
missing_versions = this_iteration.difference(graph)
1759
missing_needed = versions.intersection(missing_versions)
1761
raise RevisionNotPresent(missing_needed.pop(), self)
1762
for missing_version in missing_versions:
1763
# add a key, no parents
1764
graph[missing_version] = []
1765
pending.discard(missing_version) # don't look for it
1766
result_keys = topo_sort(graph.items())
1767
return [key[0] for key in result_keys]
1769
def get_build_details(self, version_ids):
1770
"""Get the method, index_memo and compression parent for version_ids.
1772
Ghosts are omitted from the result.
1774
:param version_ids: An iterable of version_ids.
1775
:return: A dict of version_id:(index_memo, compression_parent,
1776
parents, record_details).
1778
opaque structure to pass to read_records to extract the raw
1781
Content that this record is built upon, may be None
1783
Logical parents of this node
1785
extra information about the content which needs to be passed to
1786
Factory.parse_record
1789
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
1790
for entry in entries:
1791
version_id = self._keys_to_version_ids((entry[1],))[0]
1792
if not self._parents:
1795
parents = self._keys_to_version_ids(entry[3][0])
1796
if not self._deltas:
1797
compression_parent = None
1799
compression_parent_key = self._compression_parent(entry)
1800
if compression_parent_key:
1801
compression_parent = self._keys_to_version_ids(
1802
(compression_parent_key,))[0]
1804
compression_parent = None
1805
noeol = (entry[2][0] == 'N')
1806
if compression_parent:
1807
method = 'line-delta'
1810
result[version_id] = (self._node_to_position(entry),
1811
compression_parent, parents,
1815
def _compression_parent(self, an_entry):
1816
# return the key that an_entry is compressed against, or None
1817
# Grab the second parent list (as deltas implies parents currently)
1818
compression_parents = an_entry[3][1]
1819
if not compression_parents:
1821
return compression_parents[0]
1823
def _get_method(self, node):
1824
if not self._deltas:
1826
if self._compression_parent(node):
1831
def num_versions(self):
1832
return len(list(self._graph_index.iter_all_entries()))
1834
__len__ = num_versions
1836
def get_versions(self):
1837
"""Get all the versions in the file. not topologically sorted."""
1838
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1840
def has_version(self, version_id):
1841
"""True if the version is in the index."""
1842
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1844
def _keys_to_version_ids(self, keys):
1845
return tuple(key[0] for key in keys)
1847
def get_position(self, version_id):
1848
"""Return details needed to access the version.
1850
:return: a tuple (index, data position, size) to hand to the access
1851
logic to get the record.
1853
node = self._get_node(version_id)
1854
return self._node_to_position(node)
1856
def _node_to_position(self, node):
1857
"""Convert an index value to position details."""
1858
bits = node[2][1:].split(' ')
1859
return node[0], int(bits[0]), int(bits[1])
1861
def get_method(self, version_id):
1862
"""Return compression method of specified version."""
1863
return self._get_method(self._get_node(version_id))
1865
def _get_node(self, version_id):
1867
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1869
raise RevisionNotPresent(version_id, self)
1871
def get_options(self, version_id):
1872
"""Return a list representing options.
1876
node = self._get_node(version_id)
1877
options = [self._get_method(node)]
1878
if node[2][0] == 'N':
1879
options.append('no-eol')
1882
def get_parent_map(self, version_ids):
1883
"""Passed through to by KnitVersionedFile.get_parent_map."""
1884
nodes = self._get_entries(self._version_ids_to_keys(version_ids))
1888
result[node[1][0]] = self._keys_to_version_ids(node[3][0])
1891
result[node[1][0]] = ()
1894
def get_parents_with_ghosts(self, version_id):
1895
"""Return parents of specified version with ghosts."""
1897
return self.get_parent_map([version_id])[version_id]
1899
raise RevisionNotPresent(version_id, self)
1901
def check_versions_present(self, version_ids):
1902
"""Check that all specified versions are present."""
1903
keys = self._version_ids_to_keys(version_ids)
1904
present = self._present_keys(keys)
1905
missing = keys.difference(present)
1907
raise RevisionNotPresent(missing.pop(), self)
1909
def add_version(self, version_id, options, access_memo, parents):
1910
"""Add a version record to the index."""
1911
return self.add_versions(((version_id, options, access_memo, parents),))
1913
def add_versions(self, versions, random_id=False):
1914
"""Add multiple versions to the index.
1916
This function does not insert data into the Immutable GraphIndex
1917
backing the KnitGraphIndex, instead it prepares data for insertion by
1918
the caller and checks that it is safe to insert then calls
1919
self._add_callback with the prepared GraphIndex nodes.
1921
:param versions: a list of tuples:
1922
(version_id, options, pos, size, parents).
1923
:param random_id: If True the ids being added were randomly generated
1924
and no check for existence will be performed.
1926
if not self._add_callback:
1927
raise errors.ReadOnlyError(self)
1928
# we hope there are no repositories with inconsistent parentage
1933
for (version_id, options, access_memo, parents) in versions:
1934
index, pos, size = access_memo
1935
key = (version_id, )
1936
parents = tuple((parent, ) for parent in parents)
1937
if 'no-eol' in options:
1941
value += "%d %d" % (pos, size)
1942
if not self._deltas:
1943
if 'line-delta' in options:
1944
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1947
if 'line-delta' in options:
1948
node_refs = (parents, (parents[0],))
1950
node_refs = (parents, ())
1952
node_refs = (parents, )
1955
raise KnitCorrupt(self, "attempt to add node with parents "
1956
"in parentless index.")
1958
keys[key] = (value, node_refs)
1960
present_nodes = self._get_entries(keys)
1961
for (index, key, value, node_refs) in present_nodes:
1962
if (value, node_refs) != keys[key]:
1963
raise KnitCorrupt(self, "inconsistent details in add_versions"
1964
": %s %s" % ((value, node_refs), keys[key]))
1968
for key, (value, node_refs) in keys.iteritems():
1969
result.append((key, value, node_refs))
1971
for key, (value, node_refs) in keys.iteritems():
1972
result.append((key, value))
1973
self._add_callback(result)
1975
def _version_ids_to_keys(self, version_ids):
1976
return set((version_id, ) for version_id in version_ids)
1979
class _KnitAccess(object):
1980
"""Access to knit records in a .knit file."""
1982
def __init__(self, transport, filename, _file_mode, _dir_mode,
1983
_need_to_create, _create_parent_dir):
1984
"""Create a _KnitAccess for accessing and inserting data.
1986
:param transport: The transport the .knit is located on.
1987
:param filename: The filename of the .knit.
1989
self._transport = transport
1990
self._filename = filename
1991
self._file_mode = _file_mode
1992
self._dir_mode = _dir_mode
1993
self._need_to_create = _need_to_create
1994
self._create_parent_dir = _create_parent_dir
1996
def add_raw_records(self, sizes, raw_data):
1997
"""Add raw knit bytes to a storage area.
1999
The data is spooled to whereever the access method is storing data.
2001
:param sizes: An iterable containing the size of each raw data segment.
2002
:param raw_data: A bytestring containing the data.
2003
:return: A list of memos to retrieve the record later. Each memo is a
2004
tuple - (index, pos, length), where the index field is always None
2005
for the .knit access method.
2007
if not self._need_to_create:
2008
base = self._transport.append_bytes(self._filename, raw_data)
2010
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2011
create_parent_dir=self._create_parent_dir,
2012
mode=self._file_mode,
2013
dir_mode=self._dir_mode)
2014
self._need_to_create = False
2018
result.append((None, base, size))
2023
"""IFF this data access has its own storage area, initialise it.
2027
self._transport.put_bytes_non_atomic(self._filename, '',
2028
mode=self._file_mode)
2030
def open_file(self):
2031
"""IFF this data access can be represented as a single file, open it.
2033
For knits that are not mapped to a single file on disk this will
2036
:return: None or a file handle.
2039
return self._transport.get(self._filename)
2044
def get_raw_records(self, memos_for_retrieval):
2045
"""Get the raw bytes for a records.
2047
:param memos_for_retrieval: An iterable containing the (index, pos,
2048
length) memo for retrieving the bytes. The .knit method ignores
2049
the index as there is always only a single file.
2050
:return: An iterator over the bytes of the records.
2052
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2053
for pos, data in self._transport.readv(self._filename, read_vector):
2057
class _PackAccess(object):
2058
"""Access to knit records via a collection of packs."""
2060
def __init__(self, index_to_packs, writer=None):
2061
"""Create a _PackAccess object.
2063
:param index_to_packs: A dict mapping index objects to the transport
2064
and file names for obtaining data.
2065
:param writer: A tuple (pack.ContainerWriter, write_index) which
2066
contains the pack to write, and the index that reads from it will
2070
self.container_writer = writer[0]
2071
self.write_index = writer[1]
2073
self.container_writer = None
2074
self.write_index = None
2075
self.indices = index_to_packs
2077
def add_raw_records(self, sizes, raw_data):
2078
"""Add raw knit bytes to a storage area.
2080
The data is spooled to the container writer in one bytes-record per
2083
:param sizes: An iterable containing the size of each raw data segment.
2084
:param raw_data: A bytestring containing the data.
2085
:return: A list of memos to retrieve the record later. Each memo is a
2086
tuple - (index, pos, length), where the index field is the
2087
write_index object supplied to the PackAccess object.
2092
p_offset, p_length = self.container_writer.add_bytes_record(
2093
raw_data[offset:offset+size], [])
2095
result.append((self.write_index, p_offset, p_length))
2099
"""Pack based knits do not get individually created."""
2101
def get_raw_records(self, memos_for_retrieval):
2102
"""Get the raw bytes for a records.
2104
:param memos_for_retrieval: An iterable containing the (index, pos,
2105
length) memo for retrieving the bytes. The Pack access method
2106
looks up the pack to use for a given record in its index_to_pack
2108
:return: An iterator over the bytes of the records.
2110
# first pass, group into same-index requests
2112
current_index = None
2113
for (index, offset, length) in memos_for_retrieval:
2114
if current_index == index:
2115
current_list.append((offset, length))
2117
if current_index is not None:
2118
request_lists.append((current_index, current_list))
2119
current_index = index
2120
current_list = [(offset, length)]
2121
# handle the last entry
2122
if current_index is not None:
2123
request_lists.append((current_index, current_list))
2124
for index, offsets in request_lists:
2125
transport, path = self.indices[index]
2126
reader = pack.make_readv_reader(transport, path, offsets)
2127
for names, read_func in reader.iter_records():
2128
yield read_func(None)
2130
def open_file(self):
2131
"""Pack based knits have no single file."""
2134
def set_writer(self, writer, index, (transport, packname)):
2135
"""Set a writer to use for adding data."""
2136
if index is not None:
2137
self.indices[index] = (transport, packname)
2138
self.container_writer = writer
2139
self.write_index = index
2142
class _StreamAccess(object):
2143
"""A Knit Access object that provides data from a datastream.
2145
It also provides a fallback to present as unannotated data, annotated data
2146
from a *backing* access object.
2148
This is triggered by a index_memo which is pointing to a different index
2149
than this was constructed with, and is used to allow extracting full
2150
unannotated texts for insertion into annotated knits.
2153
def __init__(self, reader_callable, stream_index, backing_knit,
2155
"""Create a _StreamAccess object.
2157
:param reader_callable: The reader_callable from the datastream.
2158
This is called to buffer all the data immediately, for
2160
:param stream_index: The index the data stream this provides access to
2161
which will be present in native index_memo's.
2162
:param backing_knit: The knit object that will provide access to
2163
annotated texts which are not available in the stream, so as to
2164
create unannotated texts.
2165
:param orig_factory: The original content factory used to generate the
2166
stream. This is used for checking whether the thunk code for
2167
supporting _copy_texts will generate the correct form of data.
2169
self.data = reader_callable(None)
2170
self.stream_index = stream_index
2171
self.backing_knit = backing_knit
2172
self.orig_factory = orig_factory
2174
def get_raw_records(self, memos_for_retrieval):
2175
"""Get the raw bytes for a records.
2177
:param memos_for_retrieval: An iterable of memos from the
2178
_StreamIndex object identifying bytes to read; for these classes
2179
they are (from_backing_knit, index, start, end) and can point to
2180
either the backing knit or streamed data.
2181
:return: An iterator yielding a byte string for each record in
2182
memos_for_retrieval.
2184
# use a generator for memory friendliness
2185
for from_backing_knit, version_id, start, end in memos_for_retrieval:
2186
if not from_backing_knit:
2187
if version_id is not self.stream_index:
2188
raise AssertionError()
2189
yield self.data[start:end]
2191
# we have been asked to thunk. This thunking only occurs when
2192
# we are obtaining plain texts from an annotated backing knit
2193
# so that _copy_texts will work.
2194
# We could improve performance here by scanning for where we need
2195
# to do this and using get_line_list, then interleaving the output
2196
# as desired. However, for now, this is sufficient.
2197
if self.orig_factory.__class__ != KnitPlainFactory:
2198
raise errors.KnitCorrupt(
2199
self, 'Bad thunk request %r cannot be backed by %r' %
2200
(version_id, self.orig_factory))
2201
lines = self.backing_knit.get_lines(version_id)
2202
line_bytes = ''.join(lines)
2203
digest = sha_string(line_bytes)
2204
# the packed form of the fulltext always has a trailing newline,
2205
# even if the actual text does not, unless the file is empty. the
2206
# record options including the noeol flag are passed through by
2207
# _StreamIndex, so this is safe.
2209
if lines[-1][-1] != '\n':
2210
lines[-1] = lines[-1] + '\n'
2212
# We want plain data, because we expect to thunk only to allow text
2214
size, bytes = self.backing_knit._data._record_to_data(version_id,
2215
digest, lines, line_bytes)
2219
class _StreamIndex(object):
2220
"""A Knit Index object that uses the data map from a datastream."""
2222
def __init__(self, data_list, backing_index):
2223
"""Create a _StreamIndex object.
2225
:param data_list: The data_list from the datastream.
2226
:param backing_index: The index which will supply values for nodes
2227
referenced outside of this stream.
2229
self.data_list = data_list
2230
self.backing_index = backing_index
2231
self._by_version = {}
2233
for key, options, length, parents in data_list:
2234
self._by_version[key] = options, (pos, pos + length), parents
2237
def get_ancestry(self, versions, topo_sorted):
2238
"""Get an ancestry list for versions."""
2240
# Not needed for basic joins
2241
raise NotImplementedError(self.get_ancestry)
2242
# get a graph of all the mentioned versions:
2243
# Little ugly - basically copied from KnitIndex, but don't want to
2244
# accidentally incorporate too much of that index's code.
2246
pending = set(versions)
2247
cache = self._by_version
2249
version = pending.pop()
2252
parents = [p for p in cache[version][2] if p in cache]
2254
raise RevisionNotPresent(version, self)
2255
# if not completed and not a ghost
2256
pending.update([p for p in parents if p not in ancestry])
2257
ancestry.add(version)
2258
return list(ancestry)
2260
def get_build_details(self, version_ids):
2261
"""Get the method, index_memo and compression parent for version_ids.
2263
Ghosts are omitted from the result.
2265
:param version_ids: An iterable of version_ids.
2266
:return: A dict of version_id:(index_memo, compression_parent,
2267
parents, record_details).
2269
opaque memo that can be passed to _StreamAccess.read_records
2270
to extract the raw data; for these classes it is
2271
(from_backing_knit, index, start, end)
2273
Content that this record is built upon, may be None
2275
Logical parents of this node
2277
extra information about the content which needs to be passed to
2278
Factory.parse_record
2281
for version_id in version_ids:
2283
method = self.get_method(version_id)
2284
except errors.RevisionNotPresent:
2285
# ghosts are omitted
2287
parent_ids = self.get_parents_with_ghosts(version_id)
2288
noeol = ('no-eol' in self.get_options(version_id))
2289
index_memo = self.get_position(version_id)
2290
from_backing_knit = index_memo[0]
2291
if from_backing_knit:
2292
# texts retrieved from the backing knit are always full texts
2294
if method == 'fulltext':
2295
compression_parent = None
2297
compression_parent = parent_ids[0]
2298
result[version_id] = (index_memo, compression_parent,
2299
parent_ids, (method, noeol))
2302
def get_method(self, version_id):
2303
"""Return compression method of specified version."""
2304
options = self.get_options(version_id)
2305
if 'fulltext' in options:
2307
elif 'line-delta' in options:
2310
raise errors.KnitIndexUnknownMethod(self, options)
2312
def get_options(self, version_id):
2313
"""Return a list representing options.
2318
return self._by_version[version_id][0]
2320
options = list(self.backing_index.get_options(version_id))
2321
if 'fulltext' in options:
2323
elif 'line-delta' in options:
2324
# Texts from the backing knit are always returned from the stream
2326
options.remove('line-delta')
2327
options.append('fulltext')
2329
raise errors.KnitIndexUnknownMethod(self, options)
2330
return tuple(options)
2332
def get_parent_map(self, version_ids):
2333
"""Passed through to by KnitVersionedFile.get_parent_map."""
2336
for version_id in version_ids:
2338
result[version_id] = self._by_version[version_id][2]
2340
pending_ids.add(version_id)
2341
result.update(self.backing_index.get_parent_map(pending_ids))
2344
def get_parents_with_ghosts(self, version_id):
2345
"""Return parents of specified version with ghosts."""
2347
return self.get_parent_map([version_id])[version_id]
2349
raise RevisionNotPresent(version_id, self)
2351
def get_position(self, version_id):
2352
"""Return details needed to access the version.
2354
_StreamAccess has the data as a big array, so we return slice
2355
coordinates into that (as index_memo's are opaque outside the
2356
index and matching access class).
2358
:return: a tuple (from_backing_knit, index, start, end) that can
2359
be passed e.g. to get_raw_records.
2360
If from_backing_knit is False, index will be self, otherwise it
2361
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
2392
def _open_file(self):
2393
return self._access.open_file()
2395
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2396
"""Convert version_id, digest, lines into a raw data block.
2398
:param dense_lines: The bytes of lines but in a denser form. For
2399
instance, if lines is a list of 1000 bytestrings each ending in \n,
2400
dense_lines may be a list with one line in it, containing all the
2401
1000's lines and their \n's. Using dense_lines if it is already
2402
known is a win because the string join to create bytes in this
2403
function spends less time resizing the final string.
2404
:return: (len, a StringIO instance with the raw data ready to read.)
2406
# Note: using a string copy here increases memory pressure with e.g.
2407
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2408
# when doing the initial commit of a mozilla tree. RBC 20070921
2409
bytes = ''.join(chain(
2410
["version %s %d %s\n" % (version_id,
2413
dense_lines or lines,
2414
["end %s\n" % version_id]))
2415
compressed_bytes = bytes_to_gzip(bytes)
2416
return len(compressed_bytes), compressed_bytes
2418
def add_raw_records(self, sizes, raw_data):
2419
"""Append a prepared record to the data file.
2421
:param sizes: An iterable containing the size of each raw data segment.
2422
:param raw_data: A bytestring containing the data.
2423
:return: a list of index data for the way the data was stored.
2424
See the access method add_raw_records documentation for more
2427
return self._access.add_raw_records(sizes, raw_data)
2429
def _parse_record_header(self, version_id, raw_data):
2430
"""Parse a record header for consistency.
2432
:return: the header and the decompressor stream.
2433
as (stream, header_record)
2435
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2437
rec = self._check_header(version_id, df.readline())
2438
except Exception, e:
2439
raise KnitCorrupt(self._access,
2440
"While reading {%s} got %s(%s)"
2441
% (version_id, e.__class__.__name__, str(e)))
2444
def _check_header(self, version_id, line):
2447
raise KnitCorrupt(self._access,
2448
'unexpected number of elements in record header')
2449
if rec[1] != version_id:
2450
raise KnitCorrupt(self._access,
2451
'unexpected version, wanted %r, got %r'
2452
% (version_id, rec[1]))
2455
def _parse_record(self, version_id, data):
2457
# 4168 calls in 2880 217 internal
2458
# 4168 calls to _parse_record_header in 2121
2459
# 4168 calls to readlines in 330
2460
df = GzipFile(mode='rb', fileobj=StringIO(data))
2463
record_contents = df.readlines()
2464
except Exception, e:
2465
raise KnitCorrupt(self._access,
2466
"While reading {%s} got %s(%s)"
2467
% (version_id, e.__class__.__name__, str(e)))
2468
header = record_contents.pop(0)
2469
rec = self._check_header(version_id, header)
2471
last_line = record_contents.pop()
2472
if len(record_contents) != int(rec[2]):
2473
raise KnitCorrupt(self._access,
2474
'incorrect number of lines %s != %s'
2476
% (len(record_contents), int(rec[2]),
2478
if last_line != 'end %s\n' % rec[1]:
2479
raise KnitCorrupt(self._access,
2480
'unexpected version end line %r, wanted %r'
2481
% (last_line, version_id))
2483
return record_contents, rec[3]
2485
def read_records_iter_raw(self, records):
2486
"""Read text records from data file and yield raw data.
2488
This unpacks enough of the text record to validate the id is
2489
as expected but thats all.
2491
# setup an iterator of the external records:
2492
# uses readv so nice and fast we hope.
2494
# grab the disk data needed.
2495
needed_offsets = [index_memo for version_id, index_memo
2497
raw_records = self._access.get_raw_records(needed_offsets)
2499
for version_id, index_memo in records:
2500
data = raw_records.next()
2501
# validate the header
2502
df, rec = self._parse_record_header(version_id, data)
2504
yield version_id, data
2506
def read_records_iter(self, records):
2507
"""Read text records from data file and yield result.
2509
The result will be returned in whatever is the fastest to read.
2510
Not by the order requested. Also, multiple requests for the same
2511
record will only yield 1 response.
2512
:param records: A list of (version_id, pos, len) entries
2513
:return: Yields (version_id, contents, digest) in the order
2514
read, not the order requested
2519
needed_records = sorted(set(records), key=operator.itemgetter(1))
2520
if not needed_records:
2523
# The transport optimizes the fetching as well
2524
# (ie, reads continuous ranges.)
2525
raw_data = self._access.get_raw_records(
2526
[index_memo for version_id, index_memo in needed_records])
2528
for (version_id, index_memo), data in \
2529
izip(iter(needed_records), raw_data):
2530
content, digest = self._parse_record(version_id, data)
2531
yield version_id, content, digest
2533
def read_records(self, records):
2534
"""Read records into a dictionary."""
2536
for record_id, content, digest in \
2537
self.read_records_iter(records):
2538
components[record_id] = (content, digest)
2542
class InterKnit(InterVersionedFile):
2543
"""Optimised code paths for knit to knit operations."""
2545
_matching_file_from_factory = staticmethod(make_file_knit)
2546
_matching_file_to_factory = staticmethod(make_file_knit)
2549
def is_compatible(source, target):
2550
"""Be compatible with knits. """
2552
return (isinstance(source, KnitVersionedFile) and
2553
isinstance(target, KnitVersionedFile))
2554
except AttributeError:
2557
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2558
"""Copy texts to the target by extracting and adding them one by one.
2560
see join() for the parameter definitions.
2562
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2563
# --- the below is factorable out with VersionedFile.join, but wait for
2564
# VersionedFiles, it may all be simpler then.
2565
graph = Graph(self.source)
2566
search = graph._make_breadth_first_searcher(version_ids)
2567
transitive_ids = set()
2568
map(transitive_ids.update, list(search))
2569
parent_map = self.source.get_parent_map(transitive_ids)
2570
order = topo_sort(parent_map.items())
2572
def size_of_content(content):
2573
return sum(len(line) for line in content.text())
2574
# Cache at most 10MB of parent texts
2575
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2576
compute_size=size_of_content)
2577
# TODO: jam 20071116 It would be nice to have a streaming interface to
2578
# get multiple texts from a source. The source could be smarter
2579
# about how it handled intermediate stages.
2580
# get_line_list() or make_mpdiffs() seem like a possibility, but
2581
# at the moment they extract all full texts into memory, which
2582
# causes us to store more than our 3x fulltext goal.
2583
# Repository.iter_files_bytes() may be another possibility
2584
to_process = [version for version in order
2585
if version not in self.target]
2586
total = len(to_process)
2587
pb = ui.ui_factory.nested_progress_bar()
2589
for index, version in enumerate(to_process):
2590
pb.update('Converting versioned data', index, total)
2591
sha1, num_bytes, parent_text = self.target.add_lines(version,
2592
self.source.get_parents_with_ghosts(version),
2593
self.source.get_lines(version),
2594
parent_texts=parent_cache)
2595
parent_cache[version] = parent_text
2600
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2601
"""See InterVersionedFile.join."""
2602
# If the source and target are mismatched w.r.t. annotations vs
2603
# plain, the data needs to be converted accordingly
2604
if self.source.factory.annotated == self.target.factory.annotated:
2606
elif self.source.factory.annotated:
2607
converter = self._anno_to_plain_converter
2609
# We're converting from a plain to an annotated knit. Copy them
2610
# across by full texts.
2611
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2613
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2617
pb = ui.ui_factory.nested_progress_bar()
2619
version_ids = list(version_ids)
2620
if None in version_ids:
2621
version_ids.remove(None)
2623
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2625
this_versions = set(self.target._index.get_versions())
2626
# XXX: For efficiency we should not look at the whole index,
2627
# we only need to consider the referenced revisions - they
2628
# must all be present, or the method must be full-text.
2629
# TODO, RBC 20070919
2630
needed_versions = self.source_ancestry - this_versions
2632
if not needed_versions:
2634
full_list = topo_sort(
2635
self.source.get_parent_map(self.source.versions()))
2637
version_list = [i for i in full_list if (not self.target.has_version(i)
2638
and i in needed_versions)]
2642
copy_queue_records = []
2644
for version_id in version_list:
2645
options = self.source._index.get_options(version_id)
2646
parents = self.source._index.get_parents_with_ghosts(version_id)
2647
# check that its will be a consistent copy:
2648
for parent in parents:
2649
# if source has the parent, we must :
2650
# * already have it or
2651
# * have it scheduled already
2652
# otherwise we don't care
2653
if not (self.target.has_version(parent) or
2654
parent in copy_set or
2655
not self.source.has_version(parent)):
2656
raise AssertionError("problem joining parent %r "
2658
% (parent, self.source, self.target))
2659
index_memo = self.source._index.get_position(version_id)
2660
copy_queue_records.append((version_id, index_memo))
2661
copy_queue.append((version_id, options, parents))
2662
copy_set.add(version_id)
2664
# data suck the join:
2666
total = len(version_list)
2669
for (version_id, raw_data), \
2670
(version_id2, options, parents) in \
2671
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2673
if not (version_id == version_id2):
2674
raise AssertionError('logic error, inconsistent results')
2676
pb.update("Joining knit", count, total)
2678
size, raw_data = converter(raw_data, version_id, options,
2681
size = len(raw_data)
2682
raw_records.append((version_id, options, parents, size))
2683
raw_datum.append(raw_data)
2684
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2689
def _anno_to_plain_converter(self, raw_data, version_id, options,
2691
"""Convert annotated content to plain content."""
2692
data, digest = self.source._data._parse_record(version_id, raw_data)
2693
if 'fulltext' in options:
2694
content = self.source.factory.parse_fulltext(data, version_id)
2695
lines = self.target.factory.lower_fulltext(content)
2697
delta = self.source.factory.parse_line_delta(data, version_id,
2699
lines = self.target.factory.lower_line_delta(delta)
2700
return self.target._data._record_to_data(version_id, digest, lines)
2703
InterVersionedFile.register_optimiser(InterKnit)
2706
class WeaveToKnit(InterVersionedFile):
2707
"""Optimised code paths for weave to knit operations."""
2709
_matching_file_from_factory = bzrlib.weave.WeaveFile
2710
_matching_file_to_factory = staticmethod(make_file_knit)
2713
def is_compatible(source, target):
2714
"""Be compatible with weaves to knits."""
2716
return (isinstance(source, bzrlib.weave.Weave) and
2717
isinstance(target, KnitVersionedFile))
2718
except AttributeError:
2721
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2722
"""See InterVersionedFile.join."""
2723
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2728
pb = ui.ui_factory.nested_progress_bar()
2730
version_ids = list(version_ids)
2732
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2733
this_versions = set(self.target._index.get_versions())
2734
needed_versions = self.source_ancestry - this_versions
2736
if not needed_versions:
2738
full_list = topo_sort(
2739
self.source.get_parent_map(self.source.versions()))
2741
version_list = [i for i in full_list if (not self.target.has_version(i)
2742
and i in needed_versions)]
2746
total = len(version_list)
2747
parent_map = self.source.get_parent_map(version_list)
2748
for version_id in version_list:
2749
pb.update("Converting to knit", count, total)
2750
parents = parent_map[version_id]
2751
# check that its will be a consistent copy:
2752
for parent in parents:
2753
# if source has the parent, we must already have it
2754
if not self.target.has_version(parent):
2755
raise AssertionError("%r does not have parent %r"
2756
% (self.target, parent))
2757
self.target.add_lines(
2758
version_id, parents, self.source.get_lines(version_id))
2765
InterVersionedFile.register_optimiser(WeaveToKnit)
2768
# Deprecated, use PatienceSequenceMatcher instead
2769
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2772
def annotate_knit(knit, revision_id):
2773
"""Annotate a knit with no cached annotations.
2775
This implementation is for knits with no cached annotations.
2776
It will work for knits with cached annotations, but this is not
2779
annotator = _KnitAnnotator(knit)
2780
return iter(annotator.annotate(revision_id))
2783
class _KnitAnnotator(object):
2784
"""Build up the annotations for a text."""
2786
def __init__(self, knit):
2789
# Content objects, differs from fulltexts because of how final newlines
2790
# are treated by knits. the content objects here will always have a
2792
self._fulltext_contents = {}
2794
# Annotated lines of specific revisions
2795
self._annotated_lines = {}
2797
# Track the raw data for nodes that we could not process yet.
2798
# This maps the revision_id of the base to a list of children that will
2799
# annotated from it.
2800
self._pending_children = {}
2802
# Nodes which cannot be extracted
2803
self._ghosts = set()
2805
# Track how many children this node has, so we know if we need to keep
2807
self._annotate_children = {}
2808
self._compression_children = {}
2810
self._all_build_details = {}
2811
# The children => parent revision_id graph
2812
self._revision_id_graph = {}
2814
self._heads_provider = None
2816
self._nodes_to_keep_annotations = set()
2817
self._generations_until_keep = 100
2819
def set_generations_until_keep(self, value):
2820
"""Set the number of generations before caching a node.
2822
Setting this to -1 will cache every merge node, setting this higher
2823
will cache fewer nodes.
2825
self._generations_until_keep = value
2827
def _add_fulltext_content(self, revision_id, content_obj):
2828
self._fulltext_contents[revision_id] = content_obj
2829
# TODO: jam 20080305 It might be good to check the sha1digest here
2830
return content_obj.text()
2832
def _check_parents(self, child, nodes_to_annotate):
2833
"""Check if all parents have been processed.
2835
:param child: A tuple of (rev_id, parents, raw_content)
2836
:param nodes_to_annotate: If child is ready, add it to
2837
nodes_to_annotate, otherwise put it back in self._pending_children
2839
for parent_id in child[1]:
2840
if (parent_id not in self._annotated_lines):
2841
# This parent is present, but another parent is missing
2842
self._pending_children.setdefault(parent_id,
2846
# This one is ready to be processed
2847
nodes_to_annotate.append(child)
2849
def _add_annotation(self, revision_id, fulltext, parent_ids,
2850
left_matching_blocks=None):
2851
"""Add an annotation entry.
2853
All parents should already have been annotated.
2854
:return: A list of children that now have their parents satisfied.
2856
a = self._annotated_lines
2857
annotated_parent_lines = [a[p] for p in parent_ids]
2858
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2859
fulltext, revision_id, left_matching_blocks,
2860
heads_provider=self._get_heads_provider()))
2861
self._annotated_lines[revision_id] = annotated_lines
2862
for p in parent_ids:
2863
ann_children = self._annotate_children[p]
2864
ann_children.remove(revision_id)
2865
if (not ann_children
2866
and p not in self._nodes_to_keep_annotations):
2867
del self._annotated_lines[p]
2868
del self._all_build_details[p]
2869
if p in self._fulltext_contents:
2870
del self._fulltext_contents[p]
2871
# Now that we've added this one, see if there are any pending
2872
# deltas to be done, certainly this parent is finished
2873
nodes_to_annotate = []
2874
for child in self._pending_children.pop(revision_id, []):
2875
self._check_parents(child, nodes_to_annotate)
2876
return nodes_to_annotate
2878
def _get_build_graph(self, revision_id):
2879
"""Get the graphs for building texts and annotations.
2881
The data you need for creating a full text may be different than the
2882
data you need to annotate that text. (At a minimum, you need both
2883
parents to create an annotation, but only need 1 parent to generate the
2886
:return: A list of (revision_id, index_memo) records, suitable for
2887
passing to read_records_iter to start reading in the raw data fro/
2890
if revision_id in self._annotated_lines:
2893
pending = set([revision_id])
2898
# get all pending nodes
2900
this_iteration = pending
2901
build_details = self._knit._index.get_build_details(this_iteration)
2902
self._all_build_details.update(build_details)
2903
# new_nodes = self._knit._index._get_entries(this_iteration)
2905
for rev_id, details in build_details.iteritems():
2906
(index_memo, compression_parent, parents,
2907
record_details) = details
2908
self._revision_id_graph[rev_id] = parents
2909
records.append((rev_id, index_memo))
2910
# Do we actually need to check _annotated_lines?
2911
pending.update(p for p in parents
2912
if p not in self._all_build_details)
2913
if compression_parent:
2914
self._compression_children.setdefault(compression_parent,
2917
for parent in parents:
2918
self._annotate_children.setdefault(parent,
2920
num_gens = generation - kept_generation
2921
if ((num_gens >= self._generations_until_keep)
2922
and len(parents) > 1):
2923
kept_generation = generation
2924
self._nodes_to_keep_annotations.add(rev_id)
2926
missing_versions = this_iteration.difference(build_details.keys())
2927
self._ghosts.update(missing_versions)
2928
for missing_version in missing_versions:
2929
# add a key, no parents
2930
self._revision_id_graph[missing_version] = ()
2931
pending.discard(missing_version) # don't look for it
2932
if self._ghosts.intersection(self._compression_children):
2934
"We cannot have nodes which have a ghost compression parent:\n"
2936
"compression children: %r"
2937
% (self._ghosts, self._compression_children))
2938
# Cleanout anything that depends on a ghost so that we don't wait for
2939
# the ghost to show up
2940
for node in self._ghosts:
2941
if node in self._annotate_children:
2942
# We won't be building this node
2943
del self._annotate_children[node]
2944
# Generally we will want to read the records in reverse order, because
2945
# we find the parent nodes after the children
2949
def _annotate_records(self, records):
2950
"""Build the annotations for the listed records."""
2951
# We iterate in the order read, rather than a strict order requested
2952
# However, process what we can, and put off to the side things that
2953
# still need parents, cleaning them up when those parents are
2955
for (rev_id, record,
2956
digest) in self._knit._data.read_records_iter(records):
2957
if rev_id in self._annotated_lines:
2959
parent_ids = self._revision_id_graph[rev_id]
2960
parent_ids = [p for p in parent_ids if p not in self._ghosts]
2961
details = self._all_build_details[rev_id]
2962
(index_memo, compression_parent, parents,
2963
record_details) = details
2964
nodes_to_annotate = []
2965
# TODO: Remove the punning between compression parents, and
2966
# parent_ids, we should be able to do this without assuming
2968
if len(parent_ids) == 0:
2969
# There are no parents for this node, so just add it
2970
# TODO: This probably needs to be decoupled
2971
fulltext_content, delta = self._knit.factory.parse_record(
2972
rev_id, record, record_details, None)
2973
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
2974
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
2975
parent_ids, left_matching_blocks=None))
2977
child = (rev_id, parent_ids, record)
2978
# Check if all the parents are present
2979
self._check_parents(child, nodes_to_annotate)
2980
while nodes_to_annotate:
2981
# Should we use a queue here instead of a stack?
2982
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
2983
(index_memo, compression_parent, parents,
2984
record_details) = self._all_build_details[rev_id]
2985
if compression_parent is not None:
2986
comp_children = self._compression_children[compression_parent]
2987
if rev_id not in comp_children:
2988
raise AssertionError("%r not in compression children %r"
2989
% (rev_id, comp_children))
2990
# If there is only 1 child, it is safe to reuse this
2992
reuse_content = (len(comp_children) == 1
2993
and compression_parent not in
2994
self._nodes_to_keep_annotations)
2996
# Remove it from the cache since it will be changing
2997
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
2998
# Make sure to copy the fulltext since it might be
3000
parent_fulltext = list(parent_fulltext_content.text())
3002
parent_fulltext_content = self._fulltext_contents[compression_parent]
3003
parent_fulltext = parent_fulltext_content.text()
3004
comp_children.remove(rev_id)
3005
fulltext_content, delta = self._knit.factory.parse_record(
3006
rev_id, record, record_details,
3007
parent_fulltext_content,
3008
copy_base_content=(not reuse_content))
3009
fulltext = self._add_fulltext_content(rev_id,
3011
blocks = KnitContent.get_line_delta_blocks(delta,
3012
parent_fulltext, fulltext)
3014
fulltext_content = self._knit.factory.parse_fulltext(
3016
fulltext = self._add_fulltext_content(rev_id,
3019
nodes_to_annotate.extend(
3020
self._add_annotation(rev_id, fulltext, parent_ids,
3021
left_matching_blocks=blocks))
3023
def _get_heads_provider(self):
3024
"""Create a heads provider for resolving ancestry issues."""
3025
if self._heads_provider is not None:
3026
return self._heads_provider
3027
parent_provider = _mod_graph.DictParentsProvider(
3028
self._revision_id_graph)
3029
graph_obj = _mod_graph.Graph(parent_provider)
3030
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3031
self._heads_provider = head_cache
3034
def annotate(self, revision_id):
3035
"""Return the annotated fulltext at the given revision.
3037
:param revision_id: The revision id for this file
3039
records = self._get_build_graph(revision_id)
3040
if revision_id in self._ghosts:
3041
raise errors.RevisionNotPresent(revision_id, self._knit)
3042
self._annotate_records(records)
3043
return self._annotated_lines[revision_id]
3047
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3049
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
