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 (
128
# TODO: Split out code specific to this format into an associated object.
130
# TODO: Can we put in some kind of value to check that the index and data
131
# files belong together?
133
# TODO: accommodate binaries, perhaps by storing a byte count
135
# TODO: function to check whole file
137
# TODO: atomically append data, then measure backwards from the cursor
138
# position after writing to work out where it was located. we may need to
139
# bypass python file buffering.
141
DATA_SUFFIX = '.knit'
142
INDEX_SUFFIX = '.kndx'
145
class KnitContentFactory(ContentFactory):
146
"""Content factory for streaming from knits.
148
:seealso ContentFactory:
151
def __init__(self, version, parents, build_details, sha1, raw_record,
152
annotated, knit=None):
153
"""Create a KnitContentFactory for version.
155
:param version: The version.
156
:param parents: The parents.
157
:param build_details: The build details as returned from
159
:param sha1: The sha1 expected from the full text of this object.
160
:param raw_record: The bytes of the knit data from disk.
161
:param annotated: True if the raw data is annotated.
163
ContentFactory.__init__(self)
165
self.key = (version,)
166
self.parents = tuple((parent,) for parent in parents)
167
if build_details[0] == 'line-delta':
172
annotated_kind = 'annotated-'
175
self.storage_kind = 'knit-%s%s-gz' % (annotated_kind, kind)
176
self._raw_record = raw_record
177
self._build_details = build_details
180
def get_bytes_as(self, storage_kind):
181
if storage_kind == self.storage_kind:
182
return self._raw_record
183
if storage_kind == 'fulltext' and self._knit is not None:
184
return self._knit.get_text(self.key[0])
186
raise errors.UnavailableRepresentation(self.key, storage_kind,
190
class KnitContent(object):
191
"""Content of a knit version to which deltas can be applied."""
194
self._should_strip_eol = False
196
def apply_delta(self, delta, new_version_id):
197
"""Apply delta to this object to become new_version_id."""
198
raise NotImplementedError(self.apply_delta)
200
def cleanup_eol(self, copy_on_mutate=True):
201
if self._should_strip_eol:
203
self._lines = self._lines[:]
204
self.strip_last_line_newline()
206
def line_delta_iter(self, new_lines):
207
"""Generate line-based delta from this content to new_lines."""
208
new_texts = new_lines.text()
209
old_texts = self.text()
210
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
211
for tag, i1, i2, j1, j2 in s.get_opcodes():
214
# ofrom, oto, length, data
215
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
217
def line_delta(self, new_lines):
218
return list(self.line_delta_iter(new_lines))
221
def get_line_delta_blocks(knit_delta, source, target):
222
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
223
target_len = len(target)
226
for s_begin, s_end, t_len, new_text in knit_delta:
227
true_n = s_begin - s_pos
230
# knit deltas do not provide reliable info about whether the
231
# last line of a file matches, due to eol handling.
232
if source[s_pos + n -1] != target[t_pos + n -1]:
235
yield s_pos, t_pos, n
236
t_pos += t_len + true_n
238
n = target_len - t_pos
240
if source[s_pos + n -1] != target[t_pos + n -1]:
243
yield s_pos, t_pos, n
244
yield s_pos + (target_len - t_pos), target_len, 0
247
class AnnotatedKnitContent(KnitContent):
248
"""Annotated content."""
250
def __init__(self, lines):
251
KnitContent.__init__(self)
255
"""Return a list of (origin, text) for each content line."""
256
return list(self._lines)
258
def apply_delta(self, delta, new_version_id):
259
"""Apply delta to this object to become new_version_id."""
262
for start, end, count, delta_lines in delta:
263
lines[offset+start:offset+end] = delta_lines
264
offset = offset + (start - end) + count
266
def strip_last_line_newline(self):
267
line = self._lines[-1][1].rstrip('\n')
268
self._lines[-1] = (self._lines[-1][0], line)
269
self._should_strip_eol = False
273
lines = [text for origin, text in self._lines]
274
except ValueError, e:
275
# most commonly (only?) caused by the internal form of the knit
276
# missing annotation information because of a bug - see thread
278
raise KnitCorrupt(self,
279
"line in annotated knit missing annotation information: %s"
282
if self._should_strip_eol:
283
anno, line = lines[-1]
284
lines[-1] = (anno, line.rstrip('\n'))
288
return AnnotatedKnitContent(self._lines[:])
291
class PlainKnitContent(KnitContent):
292
"""Unannotated content.
294
When annotate[_iter] is called on this content, the same version is reported
295
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
299
def __init__(self, lines, version_id):
300
KnitContent.__init__(self)
302
self._version_id = version_id
305
"""Return a list of (origin, text) for each content line."""
306
return [(self._version_id, line) for line in self._lines]
308
def apply_delta(self, delta, new_version_id):
309
"""Apply delta to this object to become new_version_id."""
312
for start, end, count, delta_lines in delta:
313
lines[offset+start:offset+end] = delta_lines
314
offset = offset + (start - end) + count
315
self._version_id = new_version_id
318
return PlainKnitContent(self._lines[:], self._version_id)
320
def strip_last_line_newline(self):
321
self._lines[-1] = self._lines[-1].rstrip('\n')
322
self._should_strip_eol = False
326
if self._should_strip_eol:
328
lines[-1] = lines[-1].rstrip('\n')
332
class _KnitFactory(object):
333
"""Base class for common Factory functions."""
335
def parse_record(self, version_id, record, record_details,
336
base_content, copy_base_content=True):
337
"""Parse a record into a full content object.
339
:param version_id: The official version id for this content
340
:param record: The data returned by read_records_iter()
341
:param record_details: Details about the record returned by
343
:param base_content: If get_build_details returns a compression_parent,
344
you must return a base_content here, else use None
345
:param copy_base_content: When building from the base_content, decide
346
you can either copy it and return a new object, or modify it in
348
:return: (content, delta) A Content object and possibly a line-delta,
351
method, noeol = record_details
352
if method == 'line-delta':
353
assert base_content is not None
354
if copy_base_content:
355
content = base_content.copy()
357
content = base_content
358
delta = self.parse_line_delta(record, version_id)
359
content.apply_delta(delta, version_id)
361
content = self.parse_fulltext(record, version_id)
363
content._should_strip_eol = noeol
364
return (content, delta)
367
class KnitAnnotateFactory(_KnitFactory):
368
"""Factory for creating annotated Content objects."""
372
def make(self, lines, version_id):
373
num_lines = len(lines)
374
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
376
def parse_fulltext(self, content, version_id):
377
"""Convert fulltext to internal representation
379
fulltext content is of the format
380
revid(utf8) plaintext\n
381
internal representation is of the format:
384
# TODO: jam 20070209 The tests expect this to be returned as tuples,
385
# but the code itself doesn't really depend on that.
386
# Figure out a way to not require the overhead of turning the
387
# list back into tuples.
388
lines = [tuple(line.split(' ', 1)) for line in content]
389
return AnnotatedKnitContent(lines)
391
def parse_line_delta_iter(self, lines):
392
return iter(self.parse_line_delta(lines))
394
def parse_line_delta(self, lines, version_id, plain=False):
395
"""Convert a line based delta into internal representation.
397
line delta is in the form of:
398
intstart intend intcount
400
revid(utf8) newline\n
401
internal representation is
402
(start, end, count, [1..count tuples (revid, newline)])
404
:param plain: If True, the lines are returned as a plain
405
list without annotations, not as a list of (origin, content) tuples, i.e.
406
(start, end, count, [1..count newline])
413
def cache_and_return(line):
414
origin, text = line.split(' ', 1)
415
return cache.setdefault(origin, origin), text
417
# walk through the lines parsing.
418
# Note that the plain test is explicitly pulled out of the
419
# loop to minimise any performance impact
422
start, end, count = [int(n) for n in header.split(',')]
423
contents = [next().split(' ', 1)[1] for i in xrange(count)]
424
result.append((start, end, count, contents))
427
start, end, count = [int(n) for n in header.split(',')]
428
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
429
result.append((start, end, count, contents))
432
def get_fulltext_content(self, lines):
433
"""Extract just the content lines from a fulltext."""
434
return (line.split(' ', 1)[1] for line in lines)
436
def get_linedelta_content(self, lines):
437
"""Extract just the content from a line delta.
439
This doesn't return all of the extra information stored in a delta.
440
Only the actual content lines.
445
header = header.split(',')
446
count = int(header[2])
447
for i in xrange(count):
448
origin, text = next().split(' ', 1)
451
def lower_fulltext(self, content):
452
"""convert a fulltext content record into a serializable form.
454
see parse_fulltext which this inverts.
456
# TODO: jam 20070209 We only do the caching thing to make sure that
457
# the origin is a valid utf-8 line, eventually we could remove it
458
return ['%s %s' % (o, t) for o, t in content._lines]
460
def lower_line_delta(self, delta):
461
"""convert a delta into a serializable form.
463
See parse_line_delta which this inverts.
465
# TODO: jam 20070209 We only do the caching thing to make sure that
466
# the origin is a valid utf-8 line, eventually we could remove it
468
for start, end, c, lines in delta:
469
out.append('%d,%d,%d\n' % (start, end, c))
470
out.extend(origin + ' ' + text
471
for origin, text in lines)
474
def annotate(self, knit, version_id):
475
content = knit._get_content(version_id)
476
return content.annotate()
479
class KnitPlainFactory(_KnitFactory):
480
"""Factory for creating plain Content objects."""
484
def make(self, lines, version_id):
485
return PlainKnitContent(lines, version_id)
487
def parse_fulltext(self, content, version_id):
488
"""This parses an unannotated fulltext.
490
Note that this is not a noop - the internal representation
491
has (versionid, line) - its just a constant versionid.
493
return self.make(content, version_id)
495
def parse_line_delta_iter(self, lines, version_id):
497
num_lines = len(lines)
498
while cur < num_lines:
501
start, end, c = [int(n) for n in header.split(',')]
502
yield start, end, c, lines[cur:cur+c]
505
def parse_line_delta(self, lines, version_id):
506
return list(self.parse_line_delta_iter(lines, version_id))
508
def get_fulltext_content(self, lines):
509
"""Extract just the content lines from a fulltext."""
512
def get_linedelta_content(self, lines):
513
"""Extract just the content from a line delta.
515
This doesn't return all of the extra information stored in a delta.
516
Only the actual content lines.
521
header = header.split(',')
522
count = int(header[2])
523
for i in xrange(count):
526
def lower_fulltext(self, content):
527
return content.text()
529
def lower_line_delta(self, delta):
531
for start, end, c, lines in delta:
532
out.append('%d,%d,%d\n' % (start, end, c))
536
def annotate(self, knit, version_id):
537
annotator = _KnitAnnotator(knit)
538
return annotator.annotate(version_id)
541
def make_empty_knit(transport, relpath):
542
"""Construct a empty knit at the specified location."""
543
k = make_file_knit(transport, relpath, 'w', KnitPlainFactory)
546
def make_file_knit(name, transport, file_mode=None, access_mode='w',
547
factory=None, delta=True, create=False, create_parent_dir=False,
548
delay_create=False, dir_mode=None, get_scope=None):
549
"""Factory to create a KnitVersionedFile for a .knit/.kndx file pair."""
551
factory = KnitAnnotateFactory()
553
factory = KnitPlainFactory()
554
if get_scope is None:
555
get_scope = lambda:None
556
index = _KnitIndex(transport, name + INDEX_SUFFIX,
557
access_mode, create=create, file_mode=file_mode,
558
create_parent_dir=create_parent_dir, delay_create=delay_create,
559
dir_mode=dir_mode, get_scope=get_scope)
560
access = _KnitAccess(transport, name + DATA_SUFFIX, file_mode,
561
dir_mode, ((create and not len(index)) and delay_create),
563
return KnitVersionedFile(name, transport, factory=factory,
564
create=create, delay_create=delay_create, index=index,
565
access_method=access)
569
"""Return the suffixes used by file based knits."""
570
return [DATA_SUFFIX, INDEX_SUFFIX]
571
make_file_knit.get_suffixes = get_suffixes
574
class KnitVersionedFile(VersionedFile):
575
"""Weave-like structure with faster random access.
577
A knit stores a number of texts and a summary of the relationships
578
between them. Texts are identified by a string version-id. Texts
579
are normally stored and retrieved as a series of lines, but can
580
also be passed as single strings.
582
Lines are stored with the trailing newline (if any) included, to
583
avoid special cases for files with no final newline. Lines are
584
composed of 8-bit characters, not unicode. The combination of
585
these approaches should mean any 'binary' file can be safely
586
stored and retrieved.
589
def __init__(self, relpath, transport, file_mode=None,
590
factory=None, delta=True, create=False, create_parent_dir=False,
591
delay_create=False, dir_mode=None, index=None, access_method=None):
592
"""Construct a knit at location specified by relpath.
594
:param create: If not True, only open an existing knit.
595
:param create_parent_dir: If True, create the parent directory if
596
creating the file fails. (This is used for stores with
597
hash-prefixes that may not exist yet)
598
:param delay_create: The calling code is aware that the knit won't
599
actually be created until the first data is stored.
600
:param index: An index to use for the knit.
602
super(KnitVersionedFile, self).__init__()
603
self.transport = transport
604
self.filename = relpath
605
self.factory = factory or KnitAnnotateFactory()
608
self._max_delta_chain = 200
610
if None in (access_method, index):
611
raise ValueError("No default access_method or index any more")
613
_access = access_method
614
if create and not len(self) and not delay_create:
616
self._data = _KnitData(_access)
619
return '%s(%s)' % (self.__class__.__name__,
620
self.transport.abspath(self.filename))
622
def _check_should_delta(self, first_parents):
623
"""Iterate back through the parent listing, looking for a fulltext.
625
This is used when we want to decide whether to add a delta or a new
626
fulltext. It searches for _max_delta_chain parents. When it finds a
627
fulltext parent, it sees if the total size of the deltas leading up to
628
it is large enough to indicate that we want a new full text anyway.
630
Return True if we should create a new delta, False if we should use a
635
delta_parents = first_parents
636
for count in xrange(self._max_delta_chain):
637
parent = delta_parents[0]
638
method = self._index.get_method(parent)
639
index, pos, size = self._index.get_position(parent)
640
if method == 'fulltext':
644
delta_parents = self._index.get_parent_map([parent])[parent]
646
# We couldn't find a fulltext, so we must create a new one
649
return fulltext_size > delta_size
651
def _check_write_ok(self):
652
return self._index._check_write_ok()
654
def _add_raw_records(self, records, data):
655
"""Add all the records 'records' with data pre-joined in 'data'.
657
:param records: A list of tuples(version_id, options, parents, size).
658
:param data: The data for the records. When it is written, the records
659
are adjusted to have pos pointing into data by the sum of
660
the preceding records sizes.
663
raw_record_sizes = [record[3] for record in records]
664
positions = self._data.add_raw_records(raw_record_sizes, data)
667
for (version_id, options, parents, size), access_memo in zip(
669
index_entries.append((version_id, options, access_memo, parents))
671
self._index.add_versions(index_entries)
673
def copy_to(self, name, transport):
674
"""See VersionedFile.copy_to()."""
675
# copy the current index to a temp index to avoid racing with local
677
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
678
self.transport.get(self._index._filename))
680
f = self._data._open_file()
682
transport.put_file(name + DATA_SUFFIX, f)
685
# move the copied index into place
686
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
688
def get_data_stream(self, required_versions):
689
"""Get a data stream for the specified versions.
691
Versions may be returned in any order, not necessarily the order
692
specified. They are returned in a partial order by compression
693
parent, so that the deltas can be applied as the data stream is
694
inserted; however note that compression parents will not be sent
695
unless they were specifically requested, as the client may already
698
:param required_versions: The exact set of versions to be extracted.
699
Unlike some other knit methods, this is not used to generate a
700
transitive closure, rather it is used precisely as given.
702
:returns: format_signature, list of (version, options, length, parents),
705
required_version_set = frozenset(required_versions)
707
# list of revisions that can just be sent without waiting for their
710
# map from revision to the children based on it
712
# first, read all relevant index data, enough to sort into the right
714
for version_id in required_versions:
715
options = self._index.get_options(version_id)
716
parents = self._index.get_parents_with_ghosts(version_id)
717
index_memo = self._index.get_position(version_id)
718
version_index[version_id] = (index_memo, options, parents)
719
if ('line-delta' in options
720
and parents[0] in required_version_set):
721
# must wait until the parent has been sent
722
deferred.setdefault(parents[0], []). \
725
# either a fulltext, or a delta whose parent the client did
726
# not ask for and presumably already has
727
ready_to_send.append(version_id)
728
# build a list of results to return, plus instructions for data to
730
copy_queue_records = []
731
temp_version_list = []
733
# XXX: pushing and popping lists may be a bit inefficient
734
version_id = ready_to_send.pop(0)
735
(index_memo, options, parents) = version_index[version_id]
736
copy_queue_records.append((version_id, index_memo))
737
none, data_pos, data_size = index_memo
738
temp_version_list.append((version_id, options, data_size,
740
if version_id in deferred:
741
# now we can send all the children of this revision - we could
742
# put them in anywhere, but we hope that sending them soon
743
# after the fulltext will give good locality in the receiver
744
ready_to_send[:0] = deferred.pop(version_id)
745
assert len(deferred) == 0, \
746
"Still have compressed child versions waiting to be sent"
747
# XXX: The stream format is such that we cannot stream it - we have to
748
# know the length of all the data a-priori.
750
result_version_list = []
751
for (version_id, raw_data, _), \
752
(version_id2, options, _, parents) in \
753
izip(self._data.read_records_iter_raw(copy_queue_records),
755
assert version_id == version_id2, \
756
'logic error, inconsistent results'
757
raw_datum.append(raw_data)
758
result_version_list.append(
759
(version_id, options, len(raw_data), parents))
760
# provide a callback to get data incrementally.
761
pseudo_file = StringIO(''.join(raw_datum))
764
return pseudo_file.read()
766
return pseudo_file.read(length)
767
return (self.get_format_signature(), result_version_list, read)
769
def get_record_stream(self, versions, ordering, include_delta_closure):
770
"""Get a stream of records for versions.
772
:param versions: The versions to include. Each version is a tuple
774
:param ordering: Either 'unordered' or 'topological'. A topologically
775
sorted stream has compression parents strictly before their
777
:param include_delta_closure: If True then the closure across any
778
compression parents will be included (in the opaque data).
779
:return: An iterator of ContentFactory objects, each of which is only
780
valid until the iterator is advanced.
782
if include_delta_closure:
783
# Nb: what we should do is plan the data to stream to allow
784
# reconstruction of all the texts without excessive buffering,
785
# including re-sending common bases as needed. This makes the most
786
# sense when we start serialising these streams though, so for now
787
# we just fallback to individual text construction behind the
788
# abstraction barrier.
792
# Double index lookups here : need a unified api ?
793
parent_map = self.get_parent_map(versions)
794
position_map = self._get_components_positions(versions)
795
if ordering == 'topological':
796
versions = topo_sort(parent_map)
797
# c = component_id, r = record_details, i_m = index_memo, n = next
798
records = [(version, position_map[version][1]) for version in versions]
800
for version, raw_data, sha1 in \
801
self._data.read_records_iter_raw(records):
802
(record_details, index_memo, _) = position_map[version]
803
yield KnitContentFactory(version, parent_map[version],
804
record_details, sha1, raw_data, self.factory.annotated, knit)
806
def _extract_blocks(self, version_id, source, target):
807
if self._index.get_method(version_id) != 'line-delta':
809
parent, sha1, noeol, delta = self.get_delta(version_id)
810
return KnitContent.get_line_delta_blocks(delta, source, target)
812
def get_delta(self, version_id):
813
"""Get a delta for constructing version from some other version."""
814
self.check_not_reserved_id(version_id)
815
parents = self.get_parent_map([version_id])[version_id]
820
index_memo = self._index.get_position(version_id)
821
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
822
noeol = 'no-eol' in self._index.get_options(version_id)
823
if 'fulltext' == self._index.get_method(version_id):
824
new_content = self.factory.parse_fulltext(data, version_id)
825
if parent is not None:
826
reference_content = self._get_content(parent)
827
old_texts = reference_content.text()
830
new_texts = new_content.text()
831
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
833
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
835
delta = self.factory.parse_line_delta(data, version_id)
836
return parent, sha1, noeol, delta
838
def get_format_signature(self):
839
"""See VersionedFile.get_format_signature()."""
840
if self.factory.annotated:
841
annotated_part = "annotated"
843
annotated_part = "plain"
844
return "knit-%s" % (annotated_part,)
846
@deprecated_method(one_four)
847
def get_graph_with_ghosts(self):
848
"""See VersionedFile.get_graph_with_ghosts()."""
849
return self.get_parent_map(self.versions())
851
def get_sha1s(self, version_ids):
852
"""See VersionedFile.get_sha1s()."""
853
record_map = self._get_record_map(version_ids)
854
# record entry 2 is the 'digest'.
855
return [record_map[v][2] for v in version_ids]
857
@deprecated_method(one_four)
858
def has_ghost(self, version_id):
859
"""True if there is a ghost reference in the file to version_id."""
861
if self.has_version(version_id):
863
# optimisable if needed by memoising the _ghosts set.
864
items = self.get_parent_map(self.versions())
865
for parents in items.itervalues():
866
for parent in parents:
867
if parent == version_id and parent not in items:
871
def insert_data_stream(self, (format, data_list, reader_callable)):
872
"""Insert knit records from a data stream into this knit.
874
If a version in the stream is already present in this knit, it will not
875
be inserted a second time. It will be checked for consistency with the
876
stored version however, and may cause a KnitCorrupt error to be raised
877
if the data in the stream disagrees with the already stored data.
879
:seealso: get_data_stream
881
if format != self.get_format_signature():
882
if 'knit' in debug.debug_flags:
884
'incompatible format signature inserting to %r', self)
885
source = self._knit_from_datastream(
886
(format, data_list, reader_callable))
890
for version_id, options, length, parents in data_list:
891
if self.has_version(version_id):
892
# First check: the list of parents.
893
my_parents = self.get_parents_with_ghosts(version_id)
894
if tuple(my_parents) != tuple(parents):
895
# XXX: KnitCorrupt is not quite the right exception here.
898
'parents list %r from data stream does not match '
899
'already recorded parents %r for %s'
900
% (parents, my_parents, version_id))
902
# Also check the SHA-1 of the fulltext this content will
904
raw_data = reader_callable(length)
905
my_fulltext_sha1 = self.get_sha1s([version_id])[0]
906
df, rec = self._data._parse_record_header(version_id, raw_data)
907
stream_fulltext_sha1 = rec[3]
908
if my_fulltext_sha1 != stream_fulltext_sha1:
909
# Actually, we don't know if it's this knit that's corrupt,
910
# or the data stream we're trying to insert.
912
self.filename, 'sha-1 does not match %s' % version_id)
914
if 'line-delta' in options:
915
# Make sure that this knit record is actually useful: a
916
# line-delta is no use unless we have its parent.
917
# Fetching from a broken repository with this problem
918
# shouldn't break the target repository.
920
# See https://bugs.launchpad.net/bzr/+bug/164443
921
if not self._index.has_version(parents[0]):
924
'line-delta from stream '
927
'missing parent %s\n'
928
'Try running "bzr check" '
929
'on the source repository, and "bzr reconcile" '
931
(version_id, parents[0]))
932
self._add_raw_records(
933
[(version_id, options, parents, length)],
934
reader_callable(length))
936
def _knit_from_datastream(self, (format, data_list, reader_callable)):
937
"""Create a knit object from a data stream.
939
This method exists to allow conversion of data streams that do not
940
match the signature of this knit. Generally it will be slower and use
941
more memory to use this method to insert data, but it will work.
943
:seealso: get_data_stream for details on datastreams.
944
:return: A knit versioned file which can be used to join the datastream
947
if format == "knit-plain":
948
factory = KnitPlainFactory()
949
elif format == "knit-annotated":
950
factory = KnitAnnotateFactory()
952
raise errors.KnitDataStreamUnknown(format)
953
index = _StreamIndex(data_list, self._index)
954
access = _StreamAccess(reader_callable, index, self, factory)
955
return KnitVersionedFile(self.filename, self.transport,
956
factory=factory, index=index, access_method=access)
959
"""See VersionedFile.versions."""
960
if 'evil' in debug.debug_flags:
961
trace.mutter_callsite(2, "versions scales with size of history")
962
return self._index.get_versions()
964
def has_version(self, version_id):
965
"""See VersionedFile.has_version."""
966
if 'evil' in debug.debug_flags:
967
trace.mutter_callsite(2, "has_version is a LBYL scenario")
968
return self._index.has_version(version_id)
970
__contains__ = has_version
972
def _merge_annotations(self, content, parents, parent_texts={},
973
delta=None, annotated=None,
974
left_matching_blocks=None):
975
"""Merge annotations for content. This is done by comparing
976
the annotations based on changed to the text.
978
if left_matching_blocks is not None:
979
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
983
for parent_id in parents:
984
merge_content = self._get_content(parent_id, parent_texts)
985
if (parent_id == parents[0] and delta_seq is not None):
988
seq = patiencediff.PatienceSequenceMatcher(
989
None, merge_content.text(), content.text())
990
for i, j, n in seq.get_matching_blocks():
993
# this appears to copy (origin, text) pairs across to the
994
# new content for any line that matches the last-checked
996
content._lines[j:j+n] = merge_content._lines[i:i+n]
998
if delta_seq is None:
999
reference_content = self._get_content(parents[0], parent_texts)
1000
new_texts = content.text()
1001
old_texts = reference_content.text()
1002
delta_seq = patiencediff.PatienceSequenceMatcher(
1003
None, old_texts, new_texts)
1004
return self._make_line_delta(delta_seq, content)
1006
def _make_line_delta(self, delta_seq, new_content):
1007
"""Generate a line delta from delta_seq and new_content."""
1009
for op in delta_seq.get_opcodes():
1010
if op[0] == 'equal':
1012
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
1015
def _get_components_positions(self, version_ids):
1016
"""Produce a map of position data for the components of versions.
1018
This data is intended to be used for retrieving the knit records.
1020
A dict of version_id to (record_details, index_memo, next, parents) is
1022
method is the way referenced data should be applied.
1023
index_memo is the handle to pass to the data access to actually get the
1025
next is the build-parent of the version, or None for fulltexts.
1026
parents is the version_ids of the parents of this version
1029
pending_components = version_ids
1030
while pending_components:
1031
build_details = self._index.get_build_details(pending_components)
1032
current_components = set(pending_components)
1033
pending_components = set()
1034
for version_id, details in build_details.iteritems():
1035
(index_memo, compression_parent, parents,
1036
record_details) = details
1037
method = record_details[0]
1038
if compression_parent is not None:
1039
pending_components.add(compression_parent)
1040
component_data[version_id] = (record_details, index_memo,
1042
missing = current_components.difference(build_details)
1044
raise errors.RevisionNotPresent(missing.pop(), self.filename)
1045
return component_data
1047
def _get_content(self, version_id, parent_texts={}):
1048
"""Returns a content object that makes up the specified
1050
cached_version = parent_texts.get(version_id, None)
1051
if cached_version is not None:
1052
if not self.has_version(version_id):
1053
raise RevisionNotPresent(version_id, self.filename)
1054
return cached_version
1056
text_map, contents_map = self._get_content_maps([version_id])
1057
return contents_map[version_id]
1059
def _check_versions_present(self, version_ids):
1060
"""Check that all specified versions are present."""
1061
self._index.check_versions_present(version_ids)
1063
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
1064
nostore_sha, random_id, check_content, left_matching_blocks):
1065
"""See VersionedFile.add_lines_with_ghosts()."""
1066
self._check_add(version_id, lines, random_id, check_content)
1067
return self._add(version_id, lines, parents, self.delta,
1068
parent_texts, left_matching_blocks, nostore_sha, random_id)
1070
def _add_lines(self, version_id, parents, lines, parent_texts,
1071
left_matching_blocks, nostore_sha, random_id, check_content):
1072
"""See VersionedFile.add_lines."""
1073
self._check_add(version_id, lines, random_id, check_content)
1074
self._check_versions_present(parents)
1075
return self._add(version_id, lines[:], parents, self.delta,
1076
parent_texts, left_matching_blocks, nostore_sha, random_id)
1078
def _check_add(self, version_id, lines, random_id, check_content):
1079
"""check that version_id and lines are safe to add."""
1080
if contains_whitespace(version_id):
1081
raise InvalidRevisionId(version_id, self.filename)
1082
self.check_not_reserved_id(version_id)
1083
# Technically this could be avoided if we are happy to allow duplicate
1084
# id insertion when other things than bzr core insert texts, but it
1085
# seems useful for folk using the knit api directly to have some safety
1086
# blanket that we can disable.
1087
if not random_id and self.has_version(version_id):
1088
raise RevisionAlreadyPresent(version_id, self.filename)
1090
self._check_lines_not_unicode(lines)
1091
self._check_lines_are_lines(lines)
1093
def _add(self, version_id, lines, parents, delta, parent_texts,
1094
left_matching_blocks, nostore_sha, random_id):
1095
"""Add a set of lines on top of version specified by parents.
1097
If delta is true, compress the text as a line-delta against
1100
Any versions not present will be converted into ghosts.
1102
# first thing, if the content is something we don't need to store, find
1104
line_bytes = ''.join(lines)
1105
digest = sha_string(line_bytes)
1106
if nostore_sha == digest:
1107
raise errors.ExistingContent
1109
present_parents = []
1110
if parent_texts is None:
1112
for parent in parents:
1113
if self.has_version(parent):
1114
present_parents.append(parent)
1116
# can only compress against the left most present parent.
1118
(len(present_parents) == 0 or
1119
present_parents[0] != parents[0])):
1122
text_length = len(line_bytes)
1125
if lines[-1][-1] != '\n':
1126
# copy the contents of lines.
1128
options.append('no-eol')
1129
lines[-1] = lines[-1] + '\n'
1133
# To speed the extract of texts the delta chain is limited
1134
# to a fixed number of deltas. This should minimize both
1135
# I/O and the time spend applying deltas.
1136
delta = self._check_should_delta(present_parents)
1138
assert isinstance(version_id, str)
1139
content = self.factory.make(lines, version_id)
1140
if delta or (self.factory.annotated and len(present_parents) > 0):
1141
# Merge annotations from parent texts if needed.
1142
delta_hunks = self._merge_annotations(content, present_parents,
1143
parent_texts, delta, self.factory.annotated,
1144
left_matching_blocks)
1147
options.append('line-delta')
1148
store_lines = self.factory.lower_line_delta(delta_hunks)
1149
size, bytes = self._data._record_to_data(version_id, digest,
1152
options.append('fulltext')
1153
# isinstance is slower and we have no hierarchy.
1154
if self.factory.__class__ == KnitPlainFactory:
1155
# Use the already joined bytes saving iteration time in
1157
size, bytes = self._data._record_to_data(version_id, digest,
1158
lines, [line_bytes])
1160
# get mixed annotation + content and feed it into the
1162
store_lines = self.factory.lower_fulltext(content)
1163
size, bytes = self._data._record_to_data(version_id, digest,
1166
access_memo = self._data.add_raw_records([size], bytes)[0]
1167
self._index.add_versions(
1168
((version_id, options, access_memo, parents),),
1169
random_id=random_id)
1170
return digest, text_length, content
1172
def check(self, progress_bar=None):
1173
"""See VersionedFile.check()."""
1175
def get_lines(self, version_id):
1176
"""See VersionedFile.get_lines()."""
1177
return self.get_line_list([version_id])[0]
1179
def _get_record_map(self, version_ids):
1180
"""Produce a dictionary of knit records.
1182
:return: {version_id:(record, record_details, digest, next)}
1184
data returned from read_records
1186
opaque information to pass to parse_record
1188
SHA1 digest of the full text after all steps are done
1190
build-parent of the version, i.e. the leftmost ancestor.
1191
Will be None if the record is not a delta.
1193
position_map = self._get_components_positions(version_ids)
1194
# c = component_id, r = record_details, i_m = index_memo, n = next
1195
records = [(c, i_m) for c, (r, i_m, n)
1196
in position_map.iteritems()]
1198
for component_id, record, digest in \
1199
self._data.read_records_iter(records):
1200
(record_details, index_memo, next) = position_map[component_id]
1201
record_map[component_id] = record, record_details, digest, next
1205
def get_text(self, version_id):
1206
"""See VersionedFile.get_text"""
1207
return self.get_texts([version_id])[0]
1209
def get_texts(self, version_ids):
1210
return [''.join(l) for l in self.get_line_list(version_ids)]
1212
def get_line_list(self, version_ids):
1213
"""Return the texts of listed versions as a list of strings."""
1214
for version_id in version_ids:
1215
self.check_not_reserved_id(version_id)
1216
text_map, content_map = self._get_content_maps(version_ids)
1217
return [text_map[v] for v in version_ids]
1219
_get_lf_split_line_list = get_line_list
1221
def _get_content_maps(self, version_ids):
1222
"""Produce maps of text and KnitContents
1224
:return: (text_map, content_map) where text_map contains the texts for
1225
the requested versions and content_map contains the KnitContents.
1226
Both dicts take version_ids as their keys.
1228
# FUTURE: This function could be improved for the 'extract many' case
1229
# by tracking each component and only doing the copy when the number of
1230
# children than need to apply delta's to it is > 1 or it is part of the
1232
version_ids = list(version_ids)
1233
multiple_versions = len(version_ids) != 1
1234
record_map = self._get_record_map(version_ids)
1239
for version_id in version_ids:
1242
while cursor is not None:
1243
record, record_details, digest, next = record_map[cursor]
1244
components.append((cursor, record, record_details, digest))
1245
if cursor in content_map:
1250
for (component_id, record, record_details,
1251
digest) in reversed(components):
1252
if component_id in content_map:
1253
content = content_map[component_id]
1255
content, delta = self.factory.parse_record(version_id,
1256
record, record_details, content,
1257
copy_base_content=multiple_versions)
1258
if multiple_versions:
1259
content_map[component_id] = content
1261
content.cleanup_eol(copy_on_mutate=multiple_versions)
1262
final_content[version_id] = content
1264
# digest here is the digest from the last applied component.
1265
text = content.text()
1266
actual_sha = sha_strings(text)
1267
if actual_sha != digest:
1268
raise KnitCorrupt(self.filename,
1270
'\n of reconstructed text does not match'
1272
'\n for version %s' %
1273
(actual_sha, digest, version_id))
1274
text_map[version_id] = text
1275
return text_map, final_content
1277
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1279
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1280
if version_ids is None:
1281
version_ids = self.versions()
1283
pb = progress.DummyProgress()
1284
# we don't care about inclusions, the caller cares.
1285
# but we need to setup a list of records to visit.
1286
# we need version_id, position, length
1287
version_id_records = []
1288
requested_versions = set(version_ids)
1289
# filter for available versions
1290
for version_id in requested_versions:
1291
if not self.has_version(version_id):
1292
raise RevisionNotPresent(version_id, self.filename)
1293
# get a in-component-order queue:
1294
for version_id in self.versions():
1295
if version_id in requested_versions:
1296
index_memo = self._index.get_position(version_id)
1297
version_id_records.append((version_id, index_memo))
1299
total = len(version_id_records)
1300
for version_idx, (version_id, data, sha_value) in \
1301
enumerate(self._data.read_records_iter(version_id_records)):
1302
pb.update('Walking content.', version_idx, total)
1303
method = self._index.get_method(version_id)
1305
assert method in ('fulltext', 'line-delta')
1306
if method == 'fulltext':
1307
line_iterator = self.factory.get_fulltext_content(data)
1309
line_iterator = self.factory.get_linedelta_content(data)
1310
# XXX: It might be more efficient to yield (version_id,
1311
# line_iterator) in the future. However for now, this is a simpler
1312
# change to integrate into the rest of the codebase. RBC 20071110
1313
for line in line_iterator:
1314
yield line, version_id
1316
pb.update('Walking content.', total, total)
1318
def num_versions(self):
1319
"""See VersionedFile.num_versions()."""
1320
return self._index.num_versions()
1322
__len__ = num_versions
1324
def annotate(self, version_id):
1325
"""See VersionedFile.annotate."""
1326
return self.factory.annotate(self, version_id)
1328
def get_parent_map(self, version_ids):
1329
"""See VersionedFile.get_parent_map."""
1330
return self._index.get_parent_map(version_ids)
1332
def get_ancestry(self, versions, topo_sorted=True):
1333
"""See VersionedFile.get_ancestry."""
1334
if isinstance(versions, basestring):
1335
versions = [versions]
1338
return self._index.get_ancestry(versions, topo_sorted)
1340
def get_ancestry_with_ghosts(self, versions):
1341
"""See VersionedFile.get_ancestry_with_ghosts."""
1342
if isinstance(versions, basestring):
1343
versions = [versions]
1346
return self._index.get_ancestry_with_ghosts(versions)
1348
def plan_merge(self, ver_a, ver_b):
1349
"""See VersionedFile.plan_merge."""
1350
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1351
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1352
annotated_a = self.annotate(ver_a)
1353
annotated_b = self.annotate(ver_b)
1354
return merge._plan_annotate_merge(annotated_a, annotated_b,
1355
ancestors_a, ancestors_b)
1358
class _KnitComponentFile(object):
1359
"""One of the files used to implement a knit database"""
1361
def __init__(self, transport, filename, mode, file_mode=None,
1362
create_parent_dir=False, dir_mode=None):
1363
self._transport = transport
1364
self._filename = filename
1366
self._file_mode = file_mode
1367
self._dir_mode = dir_mode
1368
self._create_parent_dir = create_parent_dir
1369
self._need_to_create = False
1371
def _full_path(self):
1372
"""Return the full path to this file."""
1373
return self._transport.base + self._filename
1375
def check_header(self, fp):
1376
line = fp.readline()
1378
# An empty file can actually be treated as though the file doesn't
1380
raise errors.NoSuchFile(self._full_path())
1381
if line != self.HEADER:
1382
raise KnitHeaderError(badline=line,
1383
filename=self._transport.abspath(self._filename))
1386
return '%s(%s)' % (self.__class__.__name__, self._filename)
1389
class _KnitIndex(_KnitComponentFile):
1390
"""Manages knit index file.
1392
The index is already kept in memory and read on startup, to enable
1393
fast lookups of revision information. The cursor of the index
1394
file is always pointing to the end, making it easy to append
1397
_cache is a cache for fast mapping from version id to a Index
1400
_history is a cache for fast mapping from indexes to version ids.
1402
The index data format is dictionary compressed when it comes to
1403
parent references; a index entry may only have parents that with a
1404
lover index number. As a result, the index is topological sorted.
1406
Duplicate entries may be written to the index for a single version id
1407
if this is done then the latter one completely replaces the former:
1408
this allows updates to correct version and parent information.
1409
Note that the two entries may share the delta, and that successive
1410
annotations and references MUST point to the first entry.
1412
The index file on disc contains a header, followed by one line per knit
1413
record. The same revision can be present in an index file more than once.
1414
The first occurrence gets assigned a sequence number starting from 0.
1416
The format of a single line is
1417
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1418
REVISION_ID is a utf8-encoded revision id
1419
FLAGS is a comma separated list of flags about the record. Values include
1420
no-eol, line-delta, fulltext.
1421
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1422
that the the compressed data starts at.
1423
LENGTH is the ascii representation of the length of the data file.
1424
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1426
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1427
revision id already in the knit that is a parent of REVISION_ID.
1428
The ' :' marker is the end of record marker.
1431
when a write is interrupted to the index file, it will result in a line
1432
that does not end in ' :'. If the ' :' is not present at the end of a line,
1433
or at the end of the file, then the record that is missing it will be
1434
ignored by the parser.
1436
When writing new records to the index file, the data is preceded by '\n'
1437
to ensure that records always start on new lines even if the last write was
1438
interrupted. As a result its normal for the last line in the index to be
1439
missing a trailing newline. One can be added with no harmful effects.
1442
HEADER = "# bzr knit index 8\n"
1444
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1445
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1447
def _cache_version(self, version_id, options, pos, size, parents):
1448
"""Cache a version record in the history array and index cache.
1450
This is inlined into _load_data for performance. KEEP IN SYNC.
1451
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1454
# only want the _history index to reference the 1st index entry
1456
if version_id not in self._cache:
1457
index = len(self._history)
1458
self._history.append(version_id)
1460
index = self._cache[version_id][5]
1461
self._cache[version_id] = (version_id,
1468
def _check_write_ok(self):
1469
if self._get_scope() != self._scope:
1470
raise errors.OutSideTransaction()
1471
if self._mode != 'w':
1472
raise errors.ReadOnlyObjectDirtiedError(self)
1474
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1475
create_parent_dir=False, delay_create=False, dir_mode=None,
1477
_KnitComponentFile.__init__(self, transport, filename, mode,
1478
file_mode=file_mode,
1479
create_parent_dir=create_parent_dir,
1482
# position in _history is the 'official' index for a revision
1483
# but the values may have come from a newer entry.
1484
# so - wc -l of a knit index is != the number of unique names
1488
fp = self._transport.get(self._filename)
1490
# _load_data may raise NoSuchFile if the target knit is
1492
_load_data(self, fp)
1496
if mode != 'w' or not create:
1499
self._need_to_create = True
1501
self._transport.put_bytes_non_atomic(
1502
self._filename, self.HEADER, mode=self._file_mode)
1503
self._scope = get_scope()
1504
self._get_scope = get_scope
1506
def get_ancestry(self, versions, topo_sorted=True):
1507
"""See VersionedFile.get_ancestry."""
1508
# get a graph of all the mentioned versions:
1510
pending = set(versions)
1513
version = pending.pop()
1516
parents = [p for p in cache[version][4] if p in cache]
1518
raise RevisionNotPresent(version, self._filename)
1519
# if not completed and not a ghost
1520
pending.update([p for p in parents if p not in graph])
1521
graph[version] = parents
1524
return topo_sort(graph.items())
1526
def get_ancestry_with_ghosts(self, versions):
1527
"""See VersionedFile.get_ancestry_with_ghosts."""
1528
# get a graph of all the mentioned versions:
1529
self.check_versions_present(versions)
1532
pending = set(versions)
1534
version = pending.pop()
1536
parents = cache[version][4]
1542
pending.update([p for p in parents if p not in graph])
1543
graph[version] = parents
1544
return topo_sort(graph.items())
1546
def get_build_details(self, version_ids):
1547
"""Get the method, index_memo and compression parent for version_ids.
1549
Ghosts are omitted from the result.
1551
:param version_ids: An iterable of version_ids.
1552
:return: A dict of version_id:(index_memo, compression_parent,
1553
parents, record_details).
1555
opaque structure to pass to read_records to extract the raw
1558
Content that this record is built upon, may be None
1560
Logical parents of this node
1562
extra information about the content which needs to be passed to
1563
Factory.parse_record
1566
for version_id in version_ids:
1567
if version_id not in self._cache:
1568
# ghosts are omitted
1570
method = self.get_method(version_id)
1571
parents = self.get_parents_with_ghosts(version_id)
1572
if method == 'fulltext':
1573
compression_parent = None
1575
compression_parent = parents[0]
1576
noeol = 'no-eol' in self.get_options(version_id)
1577
index_memo = self.get_position(version_id)
1578
result[version_id] = (index_memo, compression_parent,
1579
parents, (method, noeol))
1582
def num_versions(self):
1583
return len(self._history)
1585
__len__ = num_versions
1587
def get_versions(self):
1588
"""Get all the versions in the file. not topologically sorted."""
1589
return self._history
1591
def _version_list_to_index(self, versions):
1594
for version in versions:
1595
if version in cache:
1596
# -- inlined lookup() --
1597
result_list.append(str(cache[version][5]))
1598
# -- end lookup () --
1600
result_list.append('.' + version)
1601
return ' '.join(result_list)
1603
def add_version(self, version_id, options, index_memo, parents):
1604
"""Add a version record to the index."""
1605
self.add_versions(((version_id, options, index_memo, parents),))
1607
def add_versions(self, versions, random_id=False):
1608
"""Add multiple versions to the index.
1610
:param versions: a list of tuples:
1611
(version_id, options, pos, size, parents).
1612
:param random_id: If True the ids being added were randomly generated
1613
and no check for existence will be performed.
1616
orig_history = self._history[:]
1617
orig_cache = self._cache.copy()
1620
for version_id, options, (index, pos, size), parents in versions:
1621
line = "\n%s %s %s %s %s :" % (version_id,
1625
self._version_list_to_index(parents))
1626
assert isinstance(line, str), \
1627
'content must be utf-8 encoded: %r' % (line,)
1629
self._cache_version(version_id, options, pos, size, tuple(parents))
1630
if not self._need_to_create:
1631
self._transport.append_bytes(self._filename, ''.join(lines))
1634
sio.write(self.HEADER)
1635
sio.writelines(lines)
1637
self._transport.put_file_non_atomic(self._filename, sio,
1638
create_parent_dir=self._create_parent_dir,
1639
mode=self._file_mode,
1640
dir_mode=self._dir_mode)
1641
self._need_to_create = False
1643
# If any problems happen, restore the original values and re-raise
1644
self._history = orig_history
1645
self._cache = orig_cache
1648
def has_version(self, version_id):
1649
"""True if the version is in the index."""
1650
return version_id in self._cache
1652
def get_position(self, version_id):
1653
"""Return details needed to access the version.
1655
.kndx indices do not support split-out data, so return None for the
1658
:return: a tuple (None, data position, size) to hand to the access
1659
logic to get the record.
1661
entry = self._cache[version_id]
1662
return None, entry[2], entry[3]
1664
def get_method(self, version_id):
1665
"""Return compression method of specified version."""
1667
options = self._cache[version_id][1]
1669
raise RevisionNotPresent(version_id, self._filename)
1670
if 'fulltext' in options:
1673
if 'line-delta' not in options:
1674
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1677
def get_options(self, version_id):
1678
"""Return a list representing options.
1682
return self._cache[version_id][1]
1684
def get_parent_map(self, version_ids):
1685
"""Passed through to by KnitVersionedFile.get_parent_map."""
1687
for version_id in version_ids:
1689
result[version_id] = tuple(self._cache[version_id][4])
1694
def get_parents_with_ghosts(self, version_id):
1695
"""Return parents of specified version with ghosts."""
1697
return self.get_parent_map([version_id])[version_id]
1699
raise RevisionNotPresent(version_id, self)
1701
def check_versions_present(self, version_ids):
1702
"""Check that all specified versions are present."""
1704
for version_id in version_ids:
1705
if version_id not in cache:
1706
raise RevisionNotPresent(version_id, self._filename)
1709
class KnitGraphIndex(object):
1710
"""A knit index that builds on GraphIndex."""
1712
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1713
"""Construct a KnitGraphIndex on a graph_index.
1715
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1716
:param deltas: Allow delta-compressed records.
1717
:param add_callback: If not None, allow additions to the index and call
1718
this callback with a list of added GraphIndex nodes:
1719
[(node, value, node_refs), ...]
1720
:param parents: If True, record knits parents, if not do not record
1723
self._graph_index = graph_index
1724
self._deltas = deltas
1725
self._add_callback = add_callback
1726
self._parents = parents
1727
if deltas and not parents:
1728
raise KnitCorrupt(self, "Cannot do delta compression without "
1731
def _check_write_ok(self):
1734
def _get_entries(self, keys, check_present=False):
1735
"""Get the entries for keys.
1737
:param keys: An iterable of index keys, - 1-tuples.
1742
for node in self._graph_index.iter_entries(keys):
1744
found_keys.add(node[1])
1746
# adapt parentless index to the rest of the code.
1747
for node in self._graph_index.iter_entries(keys):
1748
yield node[0], node[1], node[2], ()
1749
found_keys.add(node[1])
1751
missing_keys = keys.difference(found_keys)
1753
raise RevisionNotPresent(missing_keys.pop(), self)
1755
def _present_keys(self, version_ids):
1757
node[1] for node in self._get_entries(version_ids)])
1759
def _parentless_ancestry(self, versions):
1760
"""Honour the get_ancestry API for parentless knit indices."""
1761
wanted_keys = self._version_ids_to_keys(versions)
1762
present_keys = self._present_keys(wanted_keys)
1763
missing = set(wanted_keys).difference(present_keys)
1765
raise RevisionNotPresent(missing.pop(), self)
1766
return list(self._keys_to_version_ids(present_keys))
1768
def get_ancestry(self, versions, topo_sorted=True):
1769
"""See VersionedFile.get_ancestry."""
1770
if not self._parents:
1771
return self._parentless_ancestry(versions)
1772
# XXX: This will do len(history) index calls - perhaps
1773
# it should be altered to be a index core feature?
1774
# get a graph of all the mentioned versions:
1777
versions = self._version_ids_to_keys(versions)
1778
pending = set(versions)
1780
# get all pending nodes
1781
this_iteration = pending
1782
new_nodes = self._get_entries(this_iteration)
1785
for (index, key, value, node_refs) in new_nodes:
1786
# dont ask for ghosties - otherwise
1787
# we we can end up looping with pending
1788
# being entirely ghosted.
1789
graph[key] = [parent for parent in node_refs[0]
1790
if parent not in ghosts]
1792
for parent in graph[key]:
1793
# dont examine known nodes again
1798
ghosts.update(this_iteration.difference(found))
1799
if versions.difference(graph):
1800
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1802
result_keys = topo_sort(graph.items())
1804
result_keys = graph.iterkeys()
1805
return [key[0] for key in result_keys]
1807
def get_ancestry_with_ghosts(self, versions):
1808
"""See VersionedFile.get_ancestry."""
1809
if not self._parents:
1810
return self._parentless_ancestry(versions)
1811
# XXX: This will do len(history) index calls - perhaps
1812
# it should be altered to be a index core feature?
1813
# get a graph of all the mentioned versions:
1815
versions = self._version_ids_to_keys(versions)
1816
pending = set(versions)
1818
# get all pending nodes
1819
this_iteration = pending
1820
new_nodes = self._get_entries(this_iteration)
1822
for (index, key, value, node_refs) in new_nodes:
1823
graph[key] = node_refs[0]
1825
for parent in graph[key]:
1826
# dont examine known nodes again
1830
missing_versions = this_iteration.difference(graph)
1831
missing_needed = versions.intersection(missing_versions)
1833
raise RevisionNotPresent(missing_needed.pop(), self)
1834
for missing_version in missing_versions:
1835
# add a key, no parents
1836
graph[missing_version] = []
1837
pending.discard(missing_version) # don't look for it
1838
result_keys = topo_sort(graph.items())
1839
return [key[0] for key in result_keys]
1841
def get_build_details(self, version_ids):
1842
"""Get the method, index_memo and compression parent for version_ids.
1844
Ghosts are omitted from the result.
1846
:param version_ids: An iterable of version_ids.
1847
:return: A dict of version_id:(index_memo, compression_parent,
1848
parents, record_details).
1850
opaque structure to pass to read_records to extract the raw
1853
Content that this record is built upon, may be None
1855
Logical parents of this node
1857
extra information about the content which needs to be passed to
1858
Factory.parse_record
1861
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
1862
for entry in entries:
1863
version_id = self._keys_to_version_ids((entry[1],))[0]
1864
if not self._parents:
1867
parents = self._keys_to_version_ids(entry[3][0])
1868
if not self._deltas:
1869
compression_parent = None
1871
compression_parent_key = self._compression_parent(entry)
1872
if compression_parent_key:
1873
compression_parent = self._keys_to_version_ids(
1874
(compression_parent_key,))[0]
1876
compression_parent = None
1877
noeol = (entry[2][0] == 'N')
1878
if compression_parent:
1879
method = 'line-delta'
1882
result[version_id] = (self._node_to_position(entry),
1883
compression_parent, parents,
1887
def _compression_parent(self, an_entry):
1888
# return the key that an_entry is compressed against, or None
1889
# Grab the second parent list (as deltas implies parents currently)
1890
compression_parents = an_entry[3][1]
1891
if not compression_parents:
1893
assert len(compression_parents) == 1
1894
return compression_parents[0]
1896
def _get_method(self, node):
1897
if not self._deltas:
1899
if self._compression_parent(node):
1904
def num_versions(self):
1905
return len(list(self._graph_index.iter_all_entries()))
1907
__len__ = num_versions
1909
def get_versions(self):
1910
"""Get all the versions in the file. not topologically sorted."""
1911
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1913
def has_version(self, version_id):
1914
"""True if the version is in the index."""
1915
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1917
def _keys_to_version_ids(self, keys):
1918
return tuple(key[0] for key in keys)
1920
def get_position(self, version_id):
1921
"""Return details needed to access the version.
1923
:return: a tuple (index, data position, size) to hand to the access
1924
logic to get the record.
1926
node = self._get_node(version_id)
1927
return self._node_to_position(node)
1929
def _node_to_position(self, node):
1930
"""Convert an index value to position details."""
1931
bits = node[2][1:].split(' ')
1932
return node[0], int(bits[0]), int(bits[1])
1934
def get_method(self, version_id):
1935
"""Return compression method of specified version."""
1936
return self._get_method(self._get_node(version_id))
1938
def _get_node(self, version_id):
1940
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1942
raise RevisionNotPresent(version_id, self)
1944
def get_options(self, version_id):
1945
"""Return a list representing options.
1949
node = self._get_node(version_id)
1950
options = [self._get_method(node)]
1951
if node[2][0] == 'N':
1952
options.append('no-eol')
1955
def get_parent_map(self, version_ids):
1956
"""Passed through to by KnitVersionedFile.get_parent_map."""
1957
nodes = self._get_entries(self._version_ids_to_keys(version_ids))
1961
result[node[1][0]] = self._keys_to_version_ids(node[3][0])
1964
result[node[1][0]] = ()
1967
def get_parents_with_ghosts(self, version_id):
1968
"""Return parents of specified version with ghosts."""
1970
return self.get_parent_map([version_id])[version_id]
1972
raise RevisionNotPresent(version_id, self)
1974
def check_versions_present(self, version_ids):
1975
"""Check that all specified versions are present."""
1976
keys = self._version_ids_to_keys(version_ids)
1977
present = self._present_keys(keys)
1978
missing = keys.difference(present)
1980
raise RevisionNotPresent(missing.pop(), self)
1982
def add_version(self, version_id, options, access_memo, parents):
1983
"""Add a version record to the index."""
1984
return self.add_versions(((version_id, options, access_memo, parents),))
1986
def add_versions(self, versions, random_id=False):
1987
"""Add multiple versions to the index.
1989
This function does not insert data into the Immutable GraphIndex
1990
backing the KnitGraphIndex, instead it prepares data for insertion by
1991
the caller and checks that it is safe to insert then calls
1992
self._add_callback with the prepared GraphIndex nodes.
1994
:param versions: a list of tuples:
1995
(version_id, options, pos, size, parents).
1996
:param random_id: If True the ids being added were randomly generated
1997
and no check for existence will be performed.
1999
if not self._add_callback:
2000
raise errors.ReadOnlyError(self)
2001
# we hope there are no repositories with inconsistent parentage
2006
for (version_id, options, access_memo, parents) in versions:
2007
index, pos, size = access_memo
2008
key = (version_id, )
2009
parents = tuple((parent, ) for parent in parents)
2010
if 'no-eol' in options:
2014
value += "%d %d" % (pos, size)
2015
if not self._deltas:
2016
if 'line-delta' in options:
2017
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
2020
if 'line-delta' in options:
2021
node_refs = (parents, (parents[0],))
2023
node_refs = (parents, ())
2025
node_refs = (parents, )
2028
raise KnitCorrupt(self, "attempt to add node with parents "
2029
"in parentless index.")
2031
keys[key] = (value, node_refs)
2033
present_nodes = self._get_entries(keys)
2034
for (index, key, value, node_refs) in present_nodes:
2035
if (value, node_refs) != keys[key]:
2036
raise KnitCorrupt(self, "inconsistent details in add_versions"
2037
": %s %s" % ((value, node_refs), keys[key]))
2041
for key, (value, node_refs) in keys.iteritems():
2042
result.append((key, value, node_refs))
2044
for key, (value, node_refs) in keys.iteritems():
2045
result.append((key, value))
2046
self._add_callback(result)
2048
def _version_ids_to_keys(self, version_ids):
2049
return set((version_id, ) for version_id in version_ids)
2052
class _KnitAccess(object):
2053
"""Access to knit records in a .knit file."""
2055
def __init__(self, transport, filename, _file_mode, _dir_mode,
2056
_need_to_create, _create_parent_dir):
2057
"""Create a _KnitAccess for accessing and inserting data.
2059
:param transport: The transport the .knit is located on.
2060
:param filename: The filename of the .knit.
2062
self._transport = transport
2063
self._filename = filename
2064
self._file_mode = _file_mode
2065
self._dir_mode = _dir_mode
2066
self._need_to_create = _need_to_create
2067
self._create_parent_dir = _create_parent_dir
2069
def add_raw_records(self, sizes, raw_data):
2070
"""Add raw knit bytes to a storage area.
2072
The data is spooled to whereever the access method is storing data.
2074
:param sizes: An iterable containing the size of each raw data segment.
2075
:param raw_data: A bytestring containing the data.
2076
:return: A list of memos to retrieve the record later. Each memo is a
2077
tuple - (index, pos, length), where the index field is always None
2078
for the .knit access method.
2080
assert type(raw_data) == str, \
2081
'data must be plain bytes was %s' % type(raw_data)
2082
if not self._need_to_create:
2083
base = self._transport.append_bytes(self._filename, raw_data)
2085
self._transport.put_bytes_non_atomic(self._filename, raw_data,
2086
create_parent_dir=self._create_parent_dir,
2087
mode=self._file_mode,
2088
dir_mode=self._dir_mode)
2089
self._need_to_create = False
2093
result.append((None, base, size))
2098
"""IFF this data access has its own storage area, initialise it.
2102
self._transport.put_bytes_non_atomic(self._filename, '',
2103
mode=self._file_mode)
2105
def open_file(self):
2106
"""IFF this data access can be represented as a single file, open it.
2108
For knits that are not mapped to a single file on disk this will
2111
:return: None or a file handle.
2114
return self._transport.get(self._filename)
2119
def get_raw_records(self, memos_for_retrieval):
2120
"""Get the raw bytes for a records.
2122
:param memos_for_retrieval: An iterable containing the (index, pos,
2123
length) memo for retrieving the bytes. The .knit method ignores
2124
the index as there is always only a single file.
2125
:return: An iterator over the bytes of the records.
2127
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2128
for pos, data in self._transport.readv(self._filename, read_vector):
2132
class _PackAccess(object):
2133
"""Access to knit records via a collection of packs."""
2135
def __init__(self, index_to_packs, writer=None):
2136
"""Create a _PackAccess object.
2138
:param index_to_packs: A dict mapping index objects to the transport
2139
and file names for obtaining data.
2140
:param writer: A tuple (pack.ContainerWriter, write_index) which
2141
contains the pack to write, and the index that reads from it will
2145
self.container_writer = writer[0]
2146
self.write_index = writer[1]
2148
self.container_writer = None
2149
self.write_index = None
2150
self.indices = index_to_packs
2152
def add_raw_records(self, sizes, raw_data):
2153
"""Add raw knit bytes to a storage area.
2155
The data is spooled to the container writer in one bytes-record per
2158
:param sizes: An iterable containing the size of each raw data segment.
2159
:param raw_data: A bytestring containing the data.
2160
:return: A list of memos to retrieve the record later. Each memo is a
2161
tuple - (index, pos, length), where the index field is the
2162
write_index object supplied to the PackAccess object.
2164
assert type(raw_data) == str, \
2165
'data must be plain bytes was %s' % type(raw_data)
2169
p_offset, p_length = self.container_writer.add_bytes_record(
2170
raw_data[offset:offset+size], [])
2172
result.append((self.write_index, p_offset, p_length))
2176
"""Pack based knits do not get individually created."""
2178
def get_raw_records(self, memos_for_retrieval):
2179
"""Get the raw bytes for a records.
2181
:param memos_for_retrieval: An iterable containing the (index, pos,
2182
length) memo for retrieving the bytes. The Pack access method
2183
looks up the pack to use for a given record in its index_to_pack
2185
:return: An iterator over the bytes of the records.
2187
# first pass, group into same-index requests
2189
current_index = None
2190
for (index, offset, length) in memos_for_retrieval:
2191
if current_index == index:
2192
current_list.append((offset, length))
2194
if current_index is not None:
2195
request_lists.append((current_index, current_list))
2196
current_index = index
2197
current_list = [(offset, length)]
2198
# handle the last entry
2199
if current_index is not None:
2200
request_lists.append((current_index, current_list))
2201
for index, offsets in request_lists:
2202
transport, path = self.indices[index]
2203
reader = pack.make_readv_reader(transport, path, offsets)
2204
for names, read_func in reader.iter_records():
2205
yield read_func(None)
2207
def open_file(self):
2208
"""Pack based knits have no single file."""
2211
def set_writer(self, writer, index, (transport, packname)):
2212
"""Set a writer to use for adding data."""
2213
if index is not None:
2214
self.indices[index] = (transport, packname)
2215
self.container_writer = writer
2216
self.write_index = index
2219
class _StreamAccess(object):
2220
"""A Knit Access object that provides data from a datastream.
2222
It also provides a fallback to present as unannotated data, annotated data
2223
from a *backing* access object.
2225
This is triggered by a index_memo which is pointing to a different index
2226
than this was constructed with, and is used to allow extracting full
2227
unannotated texts for insertion into annotated knits.
2230
def __init__(self, reader_callable, stream_index, backing_knit,
2232
"""Create a _StreamAccess object.
2234
:param reader_callable: The reader_callable from the datastream.
2235
This is called to buffer all the data immediately, for
2237
:param stream_index: The index the data stream this provides access to
2238
which will be present in native index_memo's.
2239
:param backing_knit: The knit object that will provide access to
2240
annotated texts which are not available in the stream, so as to
2241
create unannotated texts.
2242
:param orig_factory: The original content factory used to generate the
2243
stream. This is used for checking whether the thunk code for
2244
supporting _copy_texts will generate the correct form of data.
2246
self.data = reader_callable(None)
2247
self.stream_index = stream_index
2248
self.backing_knit = backing_knit
2249
self.orig_factory = orig_factory
2251
def get_raw_records(self, memos_for_retrieval):
2252
"""Get the raw bytes for a records.
2254
:param memos_for_retrieval: An iterable of memos from the
2255
_StreamIndex object identifying bytes to read; for these classes
2256
they are (from_backing_knit, index, start, end) and can point to
2257
either the backing knit or streamed data.
2258
:return: An iterator yielding a byte string for each record in
2259
memos_for_retrieval.
2261
# use a generator for memory friendliness
2262
for from_backing_knit, version_id, start, end in memos_for_retrieval:
2263
if not from_backing_knit:
2264
assert version_id is self.stream_index
2265
yield self.data[start:end]
2267
# we have been asked to thunk. This thunking only occurs when
2268
# we are obtaining plain texts from an annotated backing knit
2269
# so that _copy_texts will work.
2270
# We could improve performance here by scanning for where we need
2271
# to do this and using get_line_list, then interleaving the output
2272
# as desired. However, for now, this is sufficient.
2273
if self.orig_factory.__class__ != KnitPlainFactory:
2274
raise errors.KnitCorrupt(
2275
self, 'Bad thunk request %r cannot be backed by %r' %
2276
(version_id, self.orig_factory))
2277
lines = self.backing_knit.get_lines(version_id)
2278
line_bytes = ''.join(lines)
2279
digest = sha_string(line_bytes)
2280
# the packed form of the fulltext always has a trailing newline,
2281
# even if the actual text does not, unless the file is empty. the
2282
# record options including the noeol flag are passed through by
2283
# _StreamIndex, so this is safe.
2285
if lines[-1][-1] != '\n':
2286
lines[-1] = lines[-1] + '\n'
2288
# We want plain data, because we expect to thunk only to allow text
2290
size, bytes = self.backing_knit._data._record_to_data(version_id,
2291
digest, lines, line_bytes)
2295
class _StreamIndex(object):
2296
"""A Knit Index object that uses the data map from a datastream."""
2298
def __init__(self, data_list, backing_index):
2299
"""Create a _StreamIndex object.
2301
:param data_list: The data_list from the datastream.
2302
:param backing_index: The index which will supply values for nodes
2303
referenced outside of this stream.
2305
self.data_list = data_list
2306
self.backing_index = backing_index
2307
self._by_version = {}
2309
for key, options, length, parents in data_list:
2310
self._by_version[key] = options, (pos, pos + length), parents
2313
def get_ancestry(self, versions, topo_sorted):
2314
"""Get an ancestry list for versions."""
2316
# Not needed for basic joins
2317
raise NotImplementedError(self.get_ancestry)
2318
# get a graph of all the mentioned versions:
2319
# Little ugly - basically copied from KnitIndex, but don't want to
2320
# accidentally incorporate too much of that index's code.
2322
pending = set(versions)
2323
cache = self._by_version
2325
version = pending.pop()
2328
parents = [p for p in cache[version][2] if p in cache]
2330
raise RevisionNotPresent(version, self)
2331
# if not completed and not a ghost
2332
pending.update([p for p in parents if p not in ancestry])
2333
ancestry.add(version)
2334
return list(ancestry)
2336
def get_build_details(self, version_ids):
2337
"""Get the method, index_memo and compression parent for version_ids.
2339
Ghosts are omitted from the result.
2341
:param version_ids: An iterable of version_ids.
2342
:return: A dict of version_id:(index_memo, compression_parent,
2343
parents, record_details).
2345
opaque memo that can be passed to _StreamAccess.read_records
2346
to extract the raw data; for these classes it is
2347
(from_backing_knit, index, start, end)
2349
Content that this record is built upon, may be None
2351
Logical parents of this node
2353
extra information about the content which needs to be passed to
2354
Factory.parse_record
2357
for version_id in version_ids:
2359
method = self.get_method(version_id)
2360
except errors.RevisionNotPresent:
2361
# ghosts are omitted
2363
parent_ids = self.get_parents_with_ghosts(version_id)
2364
noeol = ('no-eol' in self.get_options(version_id))
2365
index_memo = self.get_position(version_id)
2366
from_backing_knit = index_memo[0]
2367
if from_backing_knit:
2368
# texts retrieved from the backing knit are always full texts
2370
if method == 'fulltext':
2371
compression_parent = None
2373
compression_parent = parent_ids[0]
2374
result[version_id] = (index_memo, compression_parent,
2375
parent_ids, (method, noeol))
2378
def get_method(self, version_id):
2379
"""Return compression method of specified version."""
2380
options = self.get_options(version_id)
2381
if 'fulltext' in options:
2383
elif 'line-delta' in options:
2386
raise errors.KnitIndexUnknownMethod(self, options)
2388
def get_options(self, version_id):
2389
"""Return a list representing options.
2394
return self._by_version[version_id][0]
2396
options = list(self.backing_index.get_options(version_id))
2397
if 'fulltext' in options:
2399
elif 'line-delta' in options:
2400
# Texts from the backing knit are always returned from the stream
2402
options.remove('line-delta')
2403
options.append('fulltext')
2405
raise errors.KnitIndexUnknownMethod(self, options)
2406
return tuple(options)
2408
def get_parent_map(self, version_ids):
2409
"""Passed through to by KnitVersionedFile.get_parent_map."""
2412
for version_id in version_ids:
2414
result[version_id] = self._by_version[version_id][2]
2416
pending_ids.add(version_id)
2417
result.update(self.backing_index.get_parent_map(pending_ids))
2420
def get_parents_with_ghosts(self, version_id):
2421
"""Return parents of specified version with ghosts."""
2423
return self.get_parent_map([version_id])[version_id]
2425
raise RevisionNotPresent(version_id, self)
2427
def get_position(self, version_id):
2428
"""Return details needed to access the version.
2430
_StreamAccess has the data as a big array, so we return slice
2431
coordinates into that (as index_memo's are opaque outside the
2432
index and matching access class).
2434
:return: a tuple (from_backing_knit, index, start, end) that can
2435
be passed e.g. to get_raw_records.
2436
If from_backing_knit is False, index will be self, otherwise it
2437
will be a version id.
2440
start, end = self._by_version[version_id][1]
2441
return False, self, start, end
2443
# Signal to the access object to handle this from the backing knit.
2444
return (True, version_id, None, None)
2446
def get_versions(self):
2447
"""Get all the versions in the stream."""
2448
return self._by_version.keys()
2451
class _KnitData(object):
2452
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2454
The KnitData class provides the logic for parsing and using knit records,
2455
making use of an access method for the low level read and write operations.
2458
def __init__(self, access):
2459
"""Create a KnitData object.
2461
:param access: The access method to use. Access methods such as
2462
_KnitAccess manage the insertion of raw records and the subsequent
2463
retrieval of the same.
2465
self._access = access
2466
self._checked = False
2468
def _open_file(self):
2469
return self._access.open_file()
2471
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2472
"""Convert version_id, digest, lines into a raw data block.
2474
:param dense_lines: The bytes of lines but in a denser form. For
2475
instance, if lines is a list of 1000 bytestrings each ending in \n,
2476
dense_lines may be a list with one line in it, containing all the
2477
1000's lines and their \n's. Using dense_lines if it is already
2478
known is a win because the string join to create bytes in this
2479
function spends less time resizing the final string.
2480
:return: (len, a StringIO instance with the raw data ready to read.)
2482
# Note: using a string copy here increases memory pressure with e.g.
2483
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2484
# when doing the initial commit of a mozilla tree. RBC 20070921
2485
bytes = ''.join(chain(
2486
["version %s %d %s\n" % (version_id,
2489
dense_lines or lines,
2490
["end %s\n" % version_id]))
2491
assert bytes.__class__ == str
2492
compressed_bytes = bytes_to_gzip(bytes)
2493
return len(compressed_bytes), compressed_bytes
2495
def add_raw_records(self, sizes, raw_data):
2496
"""Append a prepared record to the data file.
2498
:param sizes: An iterable containing the size of each raw data segment.
2499
:param raw_data: A bytestring containing the data.
2500
:return: a list of index data for the way the data was stored.
2501
See the access method add_raw_records documentation for more
2504
return self._access.add_raw_records(sizes, raw_data)
2506
def _parse_record_header(self, version_id, raw_data):
2507
"""Parse a record header for consistency.
2509
:return: the header and the decompressor stream.
2510
as (stream, header_record)
2512
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2514
rec = self._check_header(version_id, df.readline())
2515
except Exception, e:
2516
raise KnitCorrupt(self._access,
2517
"While reading {%s} got %s(%s)"
2518
% (version_id, e.__class__.__name__, str(e)))
2521
def _check_header(self, version_id, line):
2524
raise KnitCorrupt(self._access,
2525
'unexpected number of elements in record header')
2526
if rec[1] != version_id:
2527
raise KnitCorrupt(self._access,
2528
'unexpected version, wanted %r, got %r'
2529
% (version_id, rec[1]))
2532
def _parse_record(self, version_id, data):
2534
# 4168 calls in 2880 217 internal
2535
# 4168 calls to _parse_record_header in 2121
2536
# 4168 calls to readlines in 330
2537
df = GzipFile(mode='rb', fileobj=StringIO(data))
2540
record_contents = df.readlines()
2541
except Exception, e:
2542
raise KnitCorrupt(self._access,
2543
"While reading {%s} got %s(%s)"
2544
% (version_id, e.__class__.__name__, str(e)))
2545
header = record_contents.pop(0)
2546
rec = self._check_header(version_id, header)
2548
last_line = record_contents.pop()
2549
if len(record_contents) != int(rec[2]):
2550
raise KnitCorrupt(self._access,
2551
'incorrect number of lines %s != %s'
2553
% (len(record_contents), int(rec[2]),
2555
if last_line != 'end %s\n' % rec[1]:
2556
raise KnitCorrupt(self._access,
2557
'unexpected version end line %r, wanted %r'
2558
% (last_line, version_id))
2560
return record_contents, rec[3]
2562
def read_records_iter_raw(self, records):
2563
"""Read text records from data file and yield raw data.
2565
This unpacks enough of the text record to validate the id is
2566
as expected but thats all.
2568
Each item the iterator yields is (version_id, bytes,
2571
# setup an iterator of the external records:
2572
# uses readv so nice and fast we hope.
2574
# grab the disk data needed.
2575
needed_offsets = [index_memo for version_id, index_memo
2577
raw_records = self._access.get_raw_records(needed_offsets)
2579
for version_id, index_memo in records:
2580
data = raw_records.next()
2581
# validate the header
2582
df, rec = self._parse_record_header(version_id, data)
2584
yield version_id, data, rec[3]
2586
def read_records_iter(self, records):
2587
"""Read text records from data file and yield result.
2589
The result will be returned in whatever is the fastest to read.
2590
Not by the order requested. Also, multiple requests for the same
2591
record will only yield 1 response.
2592
:param records: A list of (version_id, pos, len) entries
2593
:return: Yields (version_id, contents, digest) in the order
2594
read, not the order requested
2599
needed_records = sorted(set(records), key=operator.itemgetter(1))
2600
if not needed_records:
2603
# The transport optimizes the fetching as well
2604
# (ie, reads continuous ranges.)
2605
raw_data = self._access.get_raw_records(
2606
[index_memo for version_id, index_memo in needed_records])
2608
for (version_id, index_memo), data in \
2609
izip(iter(needed_records), raw_data):
2610
content, digest = self._parse_record(version_id, data)
2611
yield version_id, content, digest
2613
def read_records(self, records):
2614
"""Read records into a dictionary."""
2616
for record_id, content, digest in \
2617
self.read_records_iter(records):
2618
components[record_id] = (content, digest)
2622
class InterKnit(InterVersionedFile):
2623
"""Optimised code paths for knit to knit operations."""
2625
_matching_file_from_factory = staticmethod(make_file_knit)
2626
_matching_file_to_factory = staticmethod(make_file_knit)
2629
def is_compatible(source, target):
2630
"""Be compatible with knits. """
2632
return (isinstance(source, KnitVersionedFile) and
2633
isinstance(target, KnitVersionedFile))
2634
except AttributeError:
2637
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2638
"""Copy texts to the target by extracting and adding them one by one.
2640
see join() for the parameter definitions.
2642
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2643
# --- the below is factorable out with VersionedFile.join, but wait for
2644
# VersionedFiles, it may all be simpler then.
2645
graph = Graph(self.source)
2646
search = graph._make_breadth_first_searcher(version_ids)
2647
transitive_ids = set()
2648
map(transitive_ids.update, list(search))
2649
parent_map = self.source.get_parent_map(transitive_ids)
2650
order = topo_sort(parent_map.items())
2652
def size_of_content(content):
2653
return sum(len(line) for line in content.text())
2654
# Cache at most 10MB of parent texts
2655
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2656
compute_size=size_of_content)
2657
# TODO: jam 20071116 It would be nice to have a streaming interface to
2658
# get multiple texts from a source. The source could be smarter
2659
# about how it handled intermediate stages.
2660
# get_line_list() or make_mpdiffs() seem like a possibility, but
2661
# at the moment they extract all full texts into memory, which
2662
# causes us to store more than our 3x fulltext goal.
2663
# Repository.iter_files_bytes() may be another possibility
2664
to_process = [version for version in order
2665
if version not in self.target]
2666
total = len(to_process)
2667
pb = ui.ui_factory.nested_progress_bar()
2669
for index, version in enumerate(to_process):
2670
pb.update('Converting versioned data', index, total)
2671
sha1, num_bytes, parent_text = self.target.add_lines(version,
2672
self.source.get_parents_with_ghosts(version),
2673
self.source.get_lines(version),
2674
parent_texts=parent_cache)
2675
parent_cache[version] = parent_text
2680
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2681
"""See InterVersionedFile.join."""
2682
assert isinstance(self.source, KnitVersionedFile)
2683
assert isinstance(self.target, KnitVersionedFile)
2685
# If the source and target are mismatched w.r.t. annotations vs
2686
# plain, the data needs to be converted accordingly
2687
if self.source.factory.annotated == self.target.factory.annotated:
2689
elif self.source.factory.annotated:
2690
converter = self._anno_to_plain_converter
2692
# We're converting from a plain to an annotated knit. Copy them
2693
# across by full texts.
2694
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2696
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2700
pb = ui.ui_factory.nested_progress_bar()
2702
version_ids = list(version_ids)
2703
if None in version_ids:
2704
version_ids.remove(None)
2706
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2708
this_versions = set(self.target._index.get_versions())
2709
# XXX: For efficiency we should not look at the whole index,
2710
# we only need to consider the referenced revisions - they
2711
# must all be present, or the method must be full-text.
2712
# TODO, RBC 20070919
2713
needed_versions = self.source_ancestry - this_versions
2715
if not needed_versions:
2717
full_list = topo_sort(
2718
self.source.get_parent_map(self.source.versions()))
2720
version_list = [i for i in full_list if (not self.target.has_version(i)
2721
and i in needed_versions)]
2725
copy_queue_records = []
2727
for version_id in version_list:
2728
options = self.source._index.get_options(version_id)
2729
parents = self.source._index.get_parents_with_ghosts(version_id)
2730
# check that its will be a consistent copy:
2731
for parent in parents:
2732
# if source has the parent, we must :
2733
# * already have it or
2734
# * have it scheduled already
2735
# otherwise we don't care
2736
assert (self.target.has_version(parent) or
2737
parent in copy_set or
2738
not self.source.has_version(parent))
2739
index_memo = self.source._index.get_position(version_id)
2740
copy_queue_records.append((version_id, index_memo))
2741
copy_queue.append((version_id, options, parents))
2742
copy_set.add(version_id)
2744
# data suck the join:
2746
total = len(version_list)
2749
for (version_id, raw_data, _), \
2750
(version_id2, options, parents) in \
2751
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2753
assert version_id == version_id2, 'logic error, inconsistent results'
2755
pb.update("Joining knit", count, total)
2757
size, raw_data = converter(raw_data, version_id, options,
2760
size = len(raw_data)
2761
raw_records.append((version_id, options, parents, size))
2762
raw_datum.append(raw_data)
2763
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2768
def _anno_to_plain_converter(self, raw_data, version_id, options,
2770
"""Convert annotated content to plain content."""
2771
data, digest = self.source._data._parse_record(version_id, raw_data)
2772
if 'fulltext' in options:
2773
content = self.source.factory.parse_fulltext(data, version_id)
2774
lines = self.target.factory.lower_fulltext(content)
2776
delta = self.source.factory.parse_line_delta(data, version_id,
2778
lines = self.target.factory.lower_line_delta(delta)
2779
return self.target._data._record_to_data(version_id, digest, lines)
2782
InterVersionedFile.register_optimiser(InterKnit)
2785
class WeaveToKnit(InterVersionedFile):
2786
"""Optimised code paths for weave to knit operations."""
2788
_matching_file_from_factory = bzrlib.weave.WeaveFile
2789
_matching_file_to_factory = staticmethod(make_file_knit)
2792
def is_compatible(source, target):
2793
"""Be compatible with weaves to knits."""
2795
return (isinstance(source, bzrlib.weave.Weave) and
2796
isinstance(target, KnitVersionedFile))
2797
except AttributeError:
2800
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2801
"""See InterVersionedFile.join."""
2802
assert isinstance(self.source, bzrlib.weave.Weave)
2803
assert isinstance(self.target, KnitVersionedFile)
2805
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2810
pb = ui.ui_factory.nested_progress_bar()
2812
version_ids = list(version_ids)
2814
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2815
this_versions = set(self.target._index.get_versions())
2816
needed_versions = self.source_ancestry - this_versions
2818
if not needed_versions:
2820
full_list = topo_sort(
2821
self.source.get_parent_map(self.source.versions()))
2823
version_list = [i for i in full_list if (not self.target.has_version(i)
2824
and i in needed_versions)]
2828
total = len(version_list)
2829
parent_map = self.source.get_parent_map(version_list)
2830
for version_id in version_list:
2831
pb.update("Converting to knit", count, total)
2832
parents = parent_map[version_id]
2833
# check that its will be a consistent copy:
2834
for parent in parents:
2835
# if source has the parent, we must already have it
2836
assert (self.target.has_version(parent))
2837
self.target.add_lines(
2838
version_id, parents, self.source.get_lines(version_id))
2845
InterVersionedFile.register_optimiser(WeaveToKnit)
2848
# Deprecated, use PatienceSequenceMatcher instead
2849
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2852
def annotate_knit(knit, revision_id):
2853
"""Annotate a knit with no cached annotations.
2855
This implementation is for knits with no cached annotations.
2856
It will work for knits with cached annotations, but this is not
2859
annotator = _KnitAnnotator(knit)
2860
return iter(annotator.annotate(revision_id))
2863
class _KnitAnnotator(object):
2864
"""Build up the annotations for a text."""
2866
def __init__(self, knit):
2869
# Content objects, differs from fulltexts because of how final newlines
2870
# are treated by knits. the content objects here will always have a
2872
self._fulltext_contents = {}
2874
# Annotated lines of specific revisions
2875
self._annotated_lines = {}
2877
# Track the raw data for nodes that we could not process yet.
2878
# This maps the revision_id of the base to a list of children that will
2879
# annotated from it.
2880
self._pending_children = {}
2882
# Nodes which cannot be extracted
2883
self._ghosts = set()
2885
# Track how many children this node has, so we know if we need to keep
2887
self._annotate_children = {}
2888
self._compression_children = {}
2890
self._all_build_details = {}
2891
# The children => parent revision_id graph
2892
self._revision_id_graph = {}
2894
self._heads_provider = None
2896
self._nodes_to_keep_annotations = set()
2897
self._generations_until_keep = 100
2899
def set_generations_until_keep(self, value):
2900
"""Set the number of generations before caching a node.
2902
Setting this to -1 will cache every merge node, setting this higher
2903
will cache fewer nodes.
2905
self._generations_until_keep = value
2907
def _add_fulltext_content(self, revision_id, content_obj):
2908
self._fulltext_contents[revision_id] = content_obj
2909
# TODO: jam 20080305 It might be good to check the sha1digest here
2910
return content_obj.text()
2912
def _check_parents(self, child, nodes_to_annotate):
2913
"""Check if all parents have been processed.
2915
:param child: A tuple of (rev_id, parents, raw_content)
2916
:param nodes_to_annotate: If child is ready, add it to
2917
nodes_to_annotate, otherwise put it back in self._pending_children
2919
for parent_id in child[1]:
2920
if (parent_id not in self._annotated_lines):
2921
# This parent is present, but another parent is missing
2922
self._pending_children.setdefault(parent_id,
2926
# This one is ready to be processed
2927
nodes_to_annotate.append(child)
2929
def _add_annotation(self, revision_id, fulltext, parent_ids,
2930
left_matching_blocks=None):
2931
"""Add an annotation entry.
2933
All parents should already have been annotated.
2934
:return: A list of children that now have their parents satisfied.
2936
a = self._annotated_lines
2937
annotated_parent_lines = [a[p] for p in parent_ids]
2938
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2939
fulltext, revision_id, left_matching_blocks,
2940
heads_provider=self._get_heads_provider()))
2941
self._annotated_lines[revision_id] = annotated_lines
2942
for p in parent_ids:
2943
ann_children = self._annotate_children[p]
2944
ann_children.remove(revision_id)
2945
if (not ann_children
2946
and p not in self._nodes_to_keep_annotations):
2947
del self._annotated_lines[p]
2948
del self._all_build_details[p]
2949
if p in self._fulltext_contents:
2950
del self._fulltext_contents[p]
2951
# Now that we've added this one, see if there are any pending
2952
# deltas to be done, certainly this parent is finished
2953
nodes_to_annotate = []
2954
for child in self._pending_children.pop(revision_id, []):
2955
self._check_parents(child, nodes_to_annotate)
2956
return nodes_to_annotate
2958
def _get_build_graph(self, revision_id):
2959
"""Get the graphs for building texts and annotations.
2961
The data you need for creating a full text may be different than the
2962
data you need to annotate that text. (At a minimum, you need both
2963
parents to create an annotation, but only need 1 parent to generate the
2966
:return: A list of (revision_id, index_memo) records, suitable for
2967
passing to read_records_iter to start reading in the raw data fro/
2970
if revision_id in self._annotated_lines:
2973
pending = set([revision_id])
2978
# get all pending nodes
2980
this_iteration = pending
2981
build_details = self._knit._index.get_build_details(this_iteration)
2982
self._all_build_details.update(build_details)
2983
# new_nodes = self._knit._index._get_entries(this_iteration)
2985
for rev_id, details in build_details.iteritems():
2986
(index_memo, compression_parent, parents,
2987
record_details) = details
2988
self._revision_id_graph[rev_id] = parents
2989
records.append((rev_id, index_memo))
2990
# Do we actually need to check _annotated_lines?
2991
pending.update(p for p in parents
2992
if p not in self._all_build_details)
2993
if compression_parent:
2994
self._compression_children.setdefault(compression_parent,
2997
for parent in parents:
2998
self._annotate_children.setdefault(parent,
3000
num_gens = generation - kept_generation
3001
if ((num_gens >= self._generations_until_keep)
3002
and len(parents) > 1):
3003
kept_generation = generation
3004
self._nodes_to_keep_annotations.add(rev_id)
3006
missing_versions = this_iteration.difference(build_details.keys())
3007
self._ghosts.update(missing_versions)
3008
for missing_version in missing_versions:
3009
# add a key, no parents
3010
self._revision_id_graph[missing_version] = ()
3011
pending.discard(missing_version) # don't look for it
3012
# XXX: This should probably be a real exception, as it is a data
3014
assert not self._ghosts.intersection(self._compression_children), \
3015
"We cannot have nodes which have a compression parent of a ghost."
3016
# Cleanout anything that depends on a ghost so that we don't wait for
3017
# the ghost to show up
3018
for node in self._ghosts:
3019
if node in self._annotate_children:
3020
# We won't be building this node
3021
del self._annotate_children[node]
3022
# Generally we will want to read the records in reverse order, because
3023
# we find the parent nodes after the children
3027
def _annotate_records(self, records):
3028
"""Build the annotations for the listed records."""
3029
# We iterate in the order read, rather than a strict order requested
3030
# However, process what we can, and put off to the side things that
3031
# still need parents, cleaning them up when those parents are
3033
for (rev_id, record,
3034
digest) in self._knit._data.read_records_iter(records):
3035
if rev_id in self._annotated_lines:
3037
parent_ids = self._revision_id_graph[rev_id]
3038
parent_ids = [p for p in parent_ids if p not in self._ghosts]
3039
details = self._all_build_details[rev_id]
3040
(index_memo, compression_parent, parents,
3041
record_details) = details
3042
nodes_to_annotate = []
3043
# TODO: Remove the punning between compression parents, and
3044
# parent_ids, we should be able to do this without assuming
3046
if len(parent_ids) == 0:
3047
# There are no parents for this node, so just add it
3048
# TODO: This probably needs to be decoupled
3049
assert compression_parent is None
3050
fulltext_content, delta = self._knit.factory.parse_record(
3051
rev_id, record, record_details, None)
3052
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
3053
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
3054
parent_ids, left_matching_blocks=None))
3056
child = (rev_id, parent_ids, record)
3057
# Check if all the parents are present
3058
self._check_parents(child, nodes_to_annotate)
3059
while nodes_to_annotate:
3060
# Should we use a queue here instead of a stack?
3061
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
3062
(index_memo, compression_parent, parents,
3063
record_details) = self._all_build_details[rev_id]
3064
if compression_parent is not None:
3065
comp_children = self._compression_children[compression_parent]
3066
assert rev_id in comp_children
3067
# If there is only 1 child, it is safe to reuse this
3069
reuse_content = (len(comp_children) == 1
3070
and compression_parent not in
3071
self._nodes_to_keep_annotations)
3073
# Remove it from the cache since it will be changing
3074
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
3075
# Make sure to copy the fulltext since it might be
3077
parent_fulltext = list(parent_fulltext_content.text())
3079
parent_fulltext_content = self._fulltext_contents[compression_parent]
3080
parent_fulltext = parent_fulltext_content.text()
3081
comp_children.remove(rev_id)
3082
fulltext_content, delta = self._knit.factory.parse_record(
3083
rev_id, record, record_details,
3084
parent_fulltext_content,
3085
copy_base_content=(not reuse_content))
3086
fulltext = self._add_fulltext_content(rev_id,
3088
blocks = KnitContent.get_line_delta_blocks(delta,
3089
parent_fulltext, fulltext)
3091
fulltext_content = self._knit.factory.parse_fulltext(
3093
fulltext = self._add_fulltext_content(rev_id,
3096
nodes_to_annotate.extend(
3097
self._add_annotation(rev_id, fulltext, parent_ids,
3098
left_matching_blocks=blocks))
3100
def _get_heads_provider(self):
3101
"""Create a heads provider for resolving ancestry issues."""
3102
if self._heads_provider is not None:
3103
return self._heads_provider
3104
parent_provider = _mod_graph.DictParentsProvider(
3105
self._revision_id_graph)
3106
graph_obj = _mod_graph.Graph(parent_provider)
3107
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3108
self._heads_provider = head_cache
3111
def annotate(self, revision_id):
3112
"""Return the annotated fulltext at the given revision.
3114
:param revision_id: The revision id for this file
3116
records = self._get_build_graph(revision_id)
3117
if revision_id in self._ghosts:
3118
raise errors.RevisionNotPresent(revision_id, self._knit)
3119
self._annotate_records(records)
3120
return self._annotated_lines[revision_id]
3124
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3126
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
