1
# Copyright (C) 2005, 2006, 2007 Canonical Ltd
3
# This program is free software; you can redistribute it and/or modify
4
# it under the terms of the GNU General Public License as published by
5
# the Free Software Foundation; either version 2 of the License, or
6
# (at your option) any later version.
8
# This program is distributed in the hope that it will be useful,
9
# but WITHOUT ANY WARRANTY; without even the implied warranty of
10
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11
# GNU General Public License for more details.
13
# You should have received a copy of the GNU General Public License
14
# along with this program; if not, write to the Free Software
15
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17
"""Knit versionedfile implementation.
19
A knit is a versioned file implementation that supports efficient append only
23
lifeless: the data file is made up of "delta records". each delta record has a delta header
24
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
25
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
26
end-marker; simply "end VERSION"
28
delta can be line or full contents.a
29
... the 8's there are the index number of the annotation.
30
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
34
8 e.set('executable', 'yes')
36
8 if elt.get('executable') == 'yes':
37
8 ie.executable = True
38
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
42
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
43
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
44
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
45
09:33 < lifeless> right
46
09:33 < jrydberg> lifeless: the position and size is the range in the data file
49
so the index sequence is the dictionary compressed sequence number used
50
in the deltas to provide line annotation
55
# 10:16 < lifeless> make partial index writes safe
56
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
57
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
59
# move sha1 out of the content so that join is faster at verifying parents
60
# record content length ?
64
from cStringIO import StringIO
65
from itertools import izip, chain
70
from zlib import Z_DEFAULT_COMPRESSION
73
from bzrlib.lazy_import import lazy_import
74
lazy_import(globals(), """
94
from bzrlib.errors import (
102
RevisionAlreadyPresent,
104
from bzrlib.graph import Graph
105
from bzrlib.osutils import (
111
from bzrlib.symbol_versioning import (
112
DEPRECATED_PARAMETER,
117
from bzrlib.tsort import topo_sort
118
from bzrlib.tuned_gzip import GzipFile, bytes_to_gzip
120
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
124
# TODO: Split out code specific to this format into an associated object.
126
# TODO: Can we put in some kind of value to check that the index and data
127
# files belong together?
129
# TODO: accommodate binaries, perhaps by storing a byte count
131
# TODO: function to check whole file
133
# TODO: atomically append data, then measure backwards from the cursor
134
# position after writing to work out where it was located. we may need to
135
# bypass python file buffering.
137
DATA_SUFFIX = '.knit'
138
INDEX_SUFFIX = '.kndx'
141
class KnitContent(object):
142
"""Content of a knit version to which deltas can be applied."""
145
self._should_strip_eol = False
147
def apply_delta(self, delta, new_version_id):
148
"""Apply delta to this object to become new_version_id."""
149
raise NotImplementedError(self.apply_delta)
151
def cleanup_eol(self, copy_on_mutate=True):
152
if self._should_strip_eol:
154
self._lines = self._lines[:]
155
self.strip_last_line_newline()
157
def line_delta_iter(self, new_lines):
158
"""Generate line-based delta from this content to new_lines."""
159
new_texts = new_lines.text()
160
old_texts = self.text()
161
s = patiencediff.PatienceSequenceMatcher(None, old_texts, new_texts)
162
for tag, i1, i2, j1, j2 in s.get_opcodes():
165
# ofrom, oto, length, data
166
yield i1, i2, j2 - j1, new_lines._lines[j1:j2]
168
def line_delta(self, new_lines):
169
return list(self.line_delta_iter(new_lines))
172
def get_line_delta_blocks(knit_delta, source, target):
173
"""Extract SequenceMatcher.get_matching_blocks() from a knit delta"""
174
target_len = len(target)
177
for s_begin, s_end, t_len, new_text in knit_delta:
178
true_n = s_begin - s_pos
181
# knit deltas do not provide reliable info about whether the
182
# last line of a file matches, due to eol handling.
183
if source[s_pos + n -1] != target[t_pos + n -1]:
186
yield s_pos, t_pos, n
187
t_pos += t_len + true_n
189
n = target_len - t_pos
191
if source[s_pos + n -1] != target[t_pos + n -1]:
194
yield s_pos, t_pos, n
195
yield s_pos + (target_len - t_pos), target_len, 0
198
class AnnotatedKnitContent(KnitContent):
199
"""Annotated content."""
201
def __init__(self, lines):
202
KnitContent.__init__(self)
206
"""Return a list of (origin, text) for each content line."""
207
return list(self._lines)
209
def apply_delta(self, delta, new_version_id):
210
"""Apply delta to this object to become new_version_id."""
213
for start, end, count, delta_lines in delta:
214
lines[offset+start:offset+end] = delta_lines
215
offset = offset + (start - end) + count
217
def strip_last_line_newline(self):
218
line = self._lines[-1][1].rstrip('\n')
219
self._lines[-1] = (self._lines[-1][0], line)
220
self._should_strip_eol = False
224
lines = [text for origin, text in self._lines]
225
except ValueError, e:
226
# most commonly (only?) caused by the internal form of the knit
227
# missing annotation information because of a bug - see thread
229
raise KnitCorrupt(self,
230
"line in annotated knit missing annotation information: %s"
233
if self._should_strip_eol:
234
anno, line = lines[-1]
235
lines[-1] = (anno, line.rstrip('\n'))
239
return AnnotatedKnitContent(self._lines[:])
242
class PlainKnitContent(KnitContent):
243
"""Unannotated content.
245
When annotate[_iter] is called on this content, the same version is reported
246
for all lines. Generally, annotate[_iter] is not useful on PlainKnitContent
250
def __init__(self, lines, version_id):
251
KnitContent.__init__(self)
253
self._version_id = version_id
256
"""Return a list of (origin, text) for each content line."""
257
return [(self._version_id, line) for line in self._lines]
259
def apply_delta(self, delta, new_version_id):
260
"""Apply delta to this object to become new_version_id."""
263
for start, end, count, delta_lines in delta:
264
lines[offset+start:offset+end] = delta_lines
265
offset = offset + (start - end) + count
266
self._version_id = new_version_id
269
return PlainKnitContent(self._lines[:], self._version_id)
271
def strip_last_line_newline(self):
272
self._lines[-1] = self._lines[-1].rstrip('\n')
273
self._should_strip_eol = False
277
if self._should_strip_eol:
279
lines[-1] = lines[-1].rstrip('\n')
283
class _KnitFactory(object):
284
"""Base class for common Factory functions."""
286
def parse_record(self, version_id, record, record_details,
287
base_content, copy_base_content=True):
288
"""Parse a record into a full content object.
290
:param version_id: The official version id for this content
291
:param record: The data returned by read_records_iter()
292
:param record_details: Details about the record returned by
294
:param base_content: If get_build_details returns a compression_parent,
295
you must return a base_content here, else use None
296
:param copy_base_content: When building from the base_content, decide
297
you can either copy it and return a new object, or modify it in
299
:return: (content, delta) A Content object and possibly a line-delta,
302
method, noeol = record_details
303
if method == 'line-delta':
304
if copy_base_content:
305
content = base_content.copy()
307
content = base_content
308
delta = self.parse_line_delta(record, version_id)
309
content.apply_delta(delta, version_id)
311
content = self.parse_fulltext(record, version_id)
313
content._should_strip_eol = noeol
314
return (content, delta)
317
class KnitAnnotateFactory(_KnitFactory):
318
"""Factory for creating annotated Content objects."""
322
def make(self, lines, version_id):
323
num_lines = len(lines)
324
return AnnotatedKnitContent(zip([version_id] * num_lines, lines))
326
def parse_fulltext(self, content, version_id):
327
"""Convert fulltext to internal representation
329
fulltext content is of the format
330
revid(utf8) plaintext\n
331
internal representation is of the format:
334
# TODO: jam 20070209 The tests expect this to be returned as tuples,
335
# but the code itself doesn't really depend on that.
336
# Figure out a way to not require the overhead of turning the
337
# list back into tuples.
338
lines = [tuple(line.split(' ', 1)) for line in content]
339
return AnnotatedKnitContent(lines)
341
def parse_line_delta_iter(self, lines):
342
return iter(self.parse_line_delta(lines))
344
def parse_line_delta(self, lines, version_id, plain=False):
345
"""Convert a line based delta into internal representation.
347
line delta is in the form of:
348
intstart intend intcount
350
revid(utf8) newline\n
351
internal representation is
352
(start, end, count, [1..count tuples (revid, newline)])
354
:param plain: If True, the lines are returned as a plain
355
list without annotations, not as a list of (origin, content) tuples, i.e.
356
(start, end, count, [1..count newline])
363
def cache_and_return(line):
364
origin, text = line.split(' ', 1)
365
return cache.setdefault(origin, origin), text
367
# walk through the lines parsing.
368
# Note that the plain test is explicitly pulled out of the
369
# loop to minimise any performance impact
372
start, end, count = [int(n) for n in header.split(',')]
373
contents = [next().split(' ', 1)[1] for i in xrange(count)]
374
result.append((start, end, count, contents))
377
start, end, count = [int(n) for n in header.split(',')]
378
contents = [tuple(next().split(' ', 1)) for i in xrange(count)]
379
result.append((start, end, count, contents))
382
def get_fulltext_content(self, lines):
383
"""Extract just the content lines from a fulltext."""
384
return (line.split(' ', 1)[1] for line in lines)
386
def get_linedelta_content(self, lines):
387
"""Extract just the content from a line delta.
389
This doesn't return all of the extra information stored in a delta.
390
Only the actual content lines.
395
header = header.split(',')
396
count = int(header[2])
397
for i in xrange(count):
398
origin, text = next().split(' ', 1)
401
def lower_fulltext(self, content):
402
"""convert a fulltext content record into a serializable form.
404
see parse_fulltext which this inverts.
406
# TODO: jam 20070209 We only do the caching thing to make sure that
407
# the origin is a valid utf-8 line, eventually we could remove it
408
return ['%s %s' % (o, t) for o, t in content._lines]
410
def lower_line_delta(self, delta):
411
"""convert a delta into a serializable form.
413
See parse_line_delta which this inverts.
415
# TODO: jam 20070209 We only do the caching thing to make sure that
416
# the origin is a valid utf-8 line, eventually we could remove it
418
for start, end, c, lines in delta:
419
out.append('%d,%d,%d\n' % (start, end, c))
420
out.extend(origin + ' ' + text
421
for origin, text in lines)
424
def annotate(self, knit, version_id):
425
content = knit._get_content(version_id)
426
return content.annotate()
429
class KnitPlainFactory(_KnitFactory):
430
"""Factory for creating plain Content objects."""
434
def make(self, lines, version_id):
435
return PlainKnitContent(lines, version_id)
437
def parse_fulltext(self, content, version_id):
438
"""This parses an unannotated fulltext.
440
Note that this is not a noop - the internal representation
441
has (versionid, line) - its just a constant versionid.
443
return self.make(content, version_id)
445
def parse_line_delta_iter(self, lines, version_id):
447
num_lines = len(lines)
448
while cur < num_lines:
451
start, end, c = [int(n) for n in header.split(',')]
452
yield start, end, c, lines[cur:cur+c]
455
def parse_line_delta(self, lines, version_id):
456
return list(self.parse_line_delta_iter(lines, version_id))
458
def get_fulltext_content(self, lines):
459
"""Extract just the content lines from a fulltext."""
462
def get_linedelta_content(self, lines):
463
"""Extract just the content from a line delta.
465
This doesn't return all of the extra information stored in a delta.
466
Only the actual content lines.
471
header = header.split(',')
472
count = int(header[2])
473
for i in xrange(count):
476
def lower_fulltext(self, content):
477
return content.text()
479
def lower_line_delta(self, delta):
481
for start, end, c, lines in delta:
482
out.append('%d,%d,%d\n' % (start, end, c))
486
def annotate(self, knit, version_id):
487
annotator = _KnitAnnotator(knit)
488
return annotator.annotate(version_id)
491
def make_empty_knit(transport, relpath):
492
"""Construct a empty knit at the specified location."""
493
k = make_file_knit(transport, relpath, 'w', KnitPlainFactory)
496
def make_file_knit(name, transport, file_mode=None, access_mode='w',
497
factory=None, delta=True, create=False, create_parent_dir=False,
498
delay_create=False, dir_mode=None, get_scope=None):
499
"""Factory to create a KnitVersionedFile for a .knit/.kndx file pair."""
501
factory = KnitAnnotateFactory()
503
factory = KnitPlainFactory()
504
if get_scope is None:
505
get_scope = lambda:None
506
index = _KnitIndex(transport, name + INDEX_SUFFIX,
507
access_mode, create=create, file_mode=file_mode,
508
create_parent_dir=create_parent_dir, delay_create=delay_create,
509
dir_mode=dir_mode, get_scope=get_scope)
510
access = _KnitAccess(transport, name + DATA_SUFFIX, file_mode,
511
dir_mode, ((create and not len(index)) and delay_create),
513
return KnitVersionedFile(name, transport, factory=factory,
514
create=create, delay_create=delay_create, index=index,
515
access_method=access)
519
"""Return the suffixes used by file based knits."""
520
return [DATA_SUFFIX, INDEX_SUFFIX]
521
make_file_knit.get_suffixes = get_suffixes
524
class KnitVersionedFile(VersionedFile):
525
"""Weave-like structure with faster random access.
527
A knit stores a number of texts and a summary of the relationships
528
between them. Texts are identified by a string version-id. Texts
529
are normally stored and retrieved as a series of lines, but can
530
also be passed as single strings.
532
Lines are stored with the trailing newline (if any) included, to
533
avoid special cases for files with no final newline. Lines are
534
composed of 8-bit characters, not unicode. The combination of
535
these approaches should mean any 'binary' file can be safely
536
stored and retrieved.
539
def __init__(self, relpath, transport, file_mode=None,
540
factory=None, delta=True, create=False, create_parent_dir=False,
541
delay_create=False, dir_mode=None, index=None, access_method=None):
542
"""Construct a knit at location specified by relpath.
544
:param create: If not True, only open an existing knit.
545
:param create_parent_dir: If True, create the parent directory if
546
creating the file fails. (This is used for stores with
547
hash-prefixes that may not exist yet)
548
:param delay_create: The calling code is aware that the knit won't
549
actually be created until the first data is stored.
550
:param index: An index to use for the knit.
552
super(KnitVersionedFile, self).__init__()
553
self.transport = transport
554
self.filename = relpath
555
self.factory = factory or KnitAnnotateFactory()
558
self._max_delta_chain = 200
560
if None in (access_method, index):
561
raise ValueError("No default access_method or index any more")
563
_access = access_method
564
if create and not len(self) and not delay_create:
566
self._data = _KnitData(_access)
569
return '%s(%s)' % (self.__class__.__name__,
570
self.transport.abspath(self.filename))
572
def _check_should_delta(self, first_parents):
573
"""Iterate back through the parent listing, looking for a fulltext.
575
This is used when we want to decide whether to add a delta or a new
576
fulltext. It searches for _max_delta_chain parents. When it finds a
577
fulltext parent, it sees if the total size of the deltas leading up to
578
it is large enough to indicate that we want a new full text anyway.
580
Return True if we should create a new delta, False if we should use a
585
delta_parents = first_parents
586
for count in xrange(self._max_delta_chain):
587
parent = delta_parents[0]
588
method = self._index.get_method(parent)
589
index, pos, size = self._index.get_position(parent)
590
if method == 'fulltext':
594
delta_parents = self._index.get_parent_map([parent])[parent]
596
# We couldn't find a fulltext, so we must create a new one
599
return fulltext_size > delta_size
601
def _check_write_ok(self):
602
return self._index._check_write_ok()
604
def _add_raw_records(self, records, data):
605
"""Add all the records 'records' with data pre-joined in 'data'.
607
:param records: A list of tuples(version_id, options, parents, size).
608
:param data: The data for the records. When it is written, the records
609
are adjusted to have pos pointing into data by the sum of
610
the preceding records sizes.
613
raw_record_sizes = [record[3] for record in records]
614
positions = self._data.add_raw_records(raw_record_sizes, data)
617
for (version_id, options, parents, size), access_memo in zip(
619
index_entries.append((version_id, options, access_memo, parents))
621
self._index.add_versions(index_entries)
623
def copy_to(self, name, transport):
624
"""See VersionedFile.copy_to()."""
625
# copy the current index to a temp index to avoid racing with local
627
transport.put_file_non_atomic(name + INDEX_SUFFIX + '.tmp',
628
self.transport.get(self._index._filename))
630
f = self._data._open_file()
632
transport.put_file(name + DATA_SUFFIX, f)
635
# move the copied index into place
636
transport.move(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
638
def get_data_stream(self, required_versions):
639
"""Get a data stream for the specified versions.
641
Versions may be returned in any order, not necessarily the order
642
specified. They are returned in a partial order by compression
643
parent, so that the deltas can be applied as the data stream is
644
inserted; however note that compression parents will not be sent
645
unless they were specifically requested, as the client may already
648
:param required_versions: The exact set of versions to be extracted.
649
Unlike some other knit methods, this is not used to generate a
650
transitive closure, rather it is used precisely as given.
652
:returns: format_signature, list of (version, options, length, parents),
655
required_version_set = frozenset(required_versions)
657
# list of revisions that can just be sent without waiting for their
660
# map from revision to the children based on it
662
# first, read all relevant index data, enough to sort into the right
664
for version_id in required_versions:
665
options = self._index.get_options(version_id)
666
parents = self._index.get_parents_with_ghosts(version_id)
667
index_memo = self._index.get_position(version_id)
668
version_index[version_id] = (index_memo, options, parents)
669
if ('line-delta' in options
670
and parents[0] in required_version_set):
671
# must wait until the parent has been sent
672
deferred.setdefault(parents[0], []). \
675
# either a fulltext, or a delta whose parent the client did
676
# not ask for and presumably already has
677
ready_to_send.append(version_id)
678
# build a list of results to return, plus instructions for data to
680
copy_queue_records = []
681
temp_version_list = []
683
# XXX: pushing and popping lists may be a bit inefficient
684
version_id = ready_to_send.pop(0)
685
(index_memo, options, parents) = version_index[version_id]
686
copy_queue_records.append((version_id, index_memo))
687
none, data_pos, data_size = index_memo
688
temp_version_list.append((version_id, options, data_size,
690
if version_id in deferred:
691
# now we can send all the children of this revision - we could
692
# put them in anywhere, but we hope that sending them soon
693
# after the fulltext will give good locality in the receiver
694
ready_to_send[:0] = deferred.pop(version_id)
695
if not (len(deferred) == 0):
696
raise AssertionError("Still have compressed child versions waiting to be sent")
697
# XXX: The stream format is such that we cannot stream it - we have to
698
# know the length of all the data a-priori.
700
result_version_list = []
701
for (version_id, raw_data), \
702
(version_id2, options, _, parents) in \
703
izip(self._data.read_records_iter_raw(copy_queue_records),
705
if not (version_id == version_id2):
706
raise AssertionError('logic error, inconsistent results')
707
raw_datum.append(raw_data)
708
result_version_list.append(
709
(version_id, options, len(raw_data), parents))
710
# provide a callback to get data incrementally.
711
pseudo_file = StringIO(''.join(raw_datum))
714
return pseudo_file.read()
716
return pseudo_file.read(length)
717
return (self.get_format_signature(), result_version_list, read)
719
def _extract_blocks(self, version_id, source, target):
720
if self._index.get_method(version_id) != 'line-delta':
722
parent, sha1, noeol, delta = self.get_delta(version_id)
723
return KnitContent.get_line_delta_blocks(delta, source, target)
725
def get_delta(self, version_id):
726
"""Get a delta for constructing version from some other version."""
727
self.check_not_reserved_id(version_id)
728
parents = self.get_parent_map([version_id])[version_id]
733
index_memo = self._index.get_position(version_id)
734
data, sha1 = self._data.read_records(((version_id, index_memo),))[version_id]
735
noeol = 'no-eol' in self._index.get_options(version_id)
736
if 'fulltext' == self._index.get_method(version_id):
737
new_content = self.factory.parse_fulltext(data, version_id)
738
if parent is not None:
739
reference_content = self._get_content(parent)
740
old_texts = reference_content.text()
743
new_texts = new_content.text()
744
delta_seq = patiencediff.PatienceSequenceMatcher(None, old_texts,
746
return parent, sha1, noeol, self._make_line_delta(delta_seq, new_content)
748
delta = self.factory.parse_line_delta(data, version_id)
749
return parent, sha1, noeol, delta
751
def get_format_signature(self):
752
"""See VersionedFile.get_format_signature()."""
753
if self.factory.annotated:
754
annotated_part = "annotated"
756
annotated_part = "plain"
757
return "knit-%s" % (annotated_part,)
759
@deprecated_method(one_four)
760
def get_graph_with_ghosts(self):
761
"""See VersionedFile.get_graph_with_ghosts()."""
762
return self.get_parent_map(self.versions())
764
def get_sha1s(self, version_ids):
765
"""See VersionedFile.get_sha1s()."""
766
record_map = self._get_record_map(version_ids)
767
# record entry 2 is the 'digest'.
768
return [record_map[v][2] for v in version_ids]
770
@deprecated_method(one_four)
771
def has_ghost(self, version_id):
772
"""True if there is a ghost reference in the file to version_id."""
774
if self.has_version(version_id):
776
# optimisable if needed by memoising the _ghosts set.
777
items = self.get_parent_map(self.versions())
778
for parents in items.itervalues():
779
for parent in parents:
780
if parent == version_id and parent not in items:
784
def insert_data_stream(self, (format, data_list, reader_callable)):
785
"""Insert knit records from a data stream into this knit.
787
If a version in the stream is already present in this knit, it will not
788
be inserted a second time. It will be checked for consistency with the
789
stored version however, and may cause a KnitCorrupt error to be raised
790
if the data in the stream disagrees with the already stored data.
792
:seealso: get_data_stream
794
if format != self.get_format_signature():
795
if 'knit' in debug.debug_flags:
797
'incompatible format signature inserting to %r', self)
798
source = self._knit_from_datastream(
799
(format, data_list, reader_callable))
803
for version_id, options, length, parents in data_list:
804
if self.has_version(version_id):
805
# First check: the list of parents.
806
my_parents = self.get_parents_with_ghosts(version_id)
807
if tuple(my_parents) != tuple(parents):
808
# XXX: KnitCorrupt is not quite the right exception here.
811
'parents list %r from data stream does not match '
812
'already recorded parents %r for %s'
813
% (parents, my_parents, version_id))
815
# Also check the SHA-1 of the fulltext this content will
817
raw_data = reader_callable(length)
818
my_fulltext_sha1 = self.get_sha1s([version_id])[0]
819
df, rec = self._data._parse_record_header(version_id, raw_data)
820
stream_fulltext_sha1 = rec[3]
821
if my_fulltext_sha1 != stream_fulltext_sha1:
822
# Actually, we don't know if it's this knit that's corrupt,
823
# or the data stream we're trying to insert.
825
self.filename, 'sha-1 does not match %s' % version_id)
827
if 'line-delta' in options:
828
# Make sure that this knit record is actually useful: a
829
# line-delta is no use unless we have its parent.
830
# Fetching from a broken repository with this problem
831
# shouldn't break the target repository.
833
# See https://bugs.launchpad.net/bzr/+bug/164443
834
if not self._index.has_version(parents[0]):
837
'line-delta from stream '
840
'missing parent %s\n'
841
'Try running "bzr check" '
842
'on the source repository, and "bzr reconcile" '
844
(version_id, parents[0]))
845
self._add_raw_records(
846
[(version_id, options, parents, length)],
847
reader_callable(length))
849
def _knit_from_datastream(self, (format, data_list, reader_callable)):
850
"""Create a knit object from a data stream.
852
This method exists to allow conversion of data streams that do not
853
match the signature of this knit. Generally it will be slower and use
854
more memory to use this method to insert data, but it will work.
856
:seealso: get_data_stream for details on datastreams.
857
:return: A knit versioned file which can be used to join the datastream
860
if format == "knit-plain":
861
factory = KnitPlainFactory()
862
elif format == "knit-annotated":
863
factory = KnitAnnotateFactory()
865
raise errors.KnitDataStreamUnknown(format)
866
index = _StreamIndex(data_list, self._index)
867
access = _StreamAccess(reader_callable, index, self, factory)
868
return KnitVersionedFile(self.filename, self.transport,
869
factory=factory, index=index, access_method=access)
872
"""See VersionedFile.versions."""
873
if 'evil' in debug.debug_flags:
874
trace.mutter_callsite(2, "versions scales with size of history")
875
return self._index.get_versions()
877
def has_version(self, version_id):
878
"""See VersionedFile.has_version."""
879
if 'evil' in debug.debug_flags:
880
trace.mutter_callsite(2, "has_version is a LBYL scenario")
881
return self._index.has_version(version_id)
883
__contains__ = has_version
885
def _merge_annotations(self, content, parents, parent_texts={},
886
delta=None, annotated=None,
887
left_matching_blocks=None):
888
"""Merge annotations for content. This is done by comparing
889
the annotations based on changed to the text.
891
if left_matching_blocks is not None:
892
delta_seq = diff._PrematchedMatcher(left_matching_blocks)
896
for parent_id in parents:
897
merge_content = self._get_content(parent_id, parent_texts)
898
if (parent_id == parents[0] and delta_seq is not None):
901
seq = patiencediff.PatienceSequenceMatcher(
902
None, merge_content.text(), content.text())
903
for i, j, n in seq.get_matching_blocks():
906
# this appears to copy (origin, text) pairs across to the
907
# new content for any line that matches the last-checked
909
content._lines[j:j+n] = merge_content._lines[i:i+n]
911
if delta_seq is None:
912
reference_content = self._get_content(parents[0], parent_texts)
913
new_texts = content.text()
914
old_texts = reference_content.text()
915
delta_seq = patiencediff.PatienceSequenceMatcher(
916
None, old_texts, new_texts)
917
return self._make_line_delta(delta_seq, content)
919
def _make_line_delta(self, delta_seq, new_content):
920
"""Generate a line delta from delta_seq and new_content."""
922
for op in delta_seq.get_opcodes():
925
diff_hunks.append((op[1], op[2], op[4]-op[3], new_content._lines[op[3]:op[4]]))
928
def _get_components_positions(self, version_ids):
929
"""Produce a map of position data for the components of versions.
931
This data is intended to be used for retrieving the knit records.
933
A dict of version_id to (record_details, index_memo, next, parents) is
935
method is the way referenced data should be applied.
936
index_memo is the handle to pass to the data access to actually get the
938
next is the build-parent of the version, or None for fulltexts.
939
parents is the version_ids of the parents of this version
942
pending_components = version_ids
943
while pending_components:
944
build_details = self._index.get_build_details(pending_components)
945
current_components = set(pending_components)
946
pending_components = set()
947
for version_id, details in build_details.iteritems():
948
(index_memo, compression_parent, parents,
949
record_details) = details
950
method = record_details[0]
951
if compression_parent is not None:
952
pending_components.add(compression_parent)
953
component_data[version_id] = (record_details, index_memo,
955
missing = current_components.difference(build_details)
957
raise errors.RevisionNotPresent(missing.pop(), self.filename)
958
return component_data
960
def _get_content(self, version_id, parent_texts={}):
961
"""Returns a content object that makes up the specified
963
cached_version = parent_texts.get(version_id, None)
964
if cached_version is not None:
965
if not self.has_version(version_id):
966
raise RevisionNotPresent(version_id, self.filename)
967
return cached_version
969
text_map, contents_map = self._get_content_maps([version_id])
970
return contents_map[version_id]
972
def _check_versions_present(self, version_ids):
973
"""Check that all specified versions are present."""
974
self._index.check_versions_present(version_ids)
976
def _add_lines_with_ghosts(self, version_id, parents, lines, parent_texts,
977
nostore_sha, random_id, check_content, left_matching_blocks):
978
"""See VersionedFile.add_lines_with_ghosts()."""
979
self._check_add(version_id, lines, random_id, check_content)
980
return self._add(version_id, lines, parents, self.delta,
981
parent_texts, left_matching_blocks, nostore_sha, random_id)
983
def _add_lines(self, version_id, parents, lines, parent_texts,
984
left_matching_blocks, nostore_sha, random_id, check_content):
985
"""See VersionedFile.add_lines."""
986
self._check_add(version_id, lines, random_id, check_content)
987
self._check_versions_present(parents)
988
return self._add(version_id, lines[:], parents, self.delta,
989
parent_texts, left_matching_blocks, nostore_sha, random_id)
991
def _check_add(self, version_id, lines, random_id, check_content):
992
"""check that version_id and lines are safe to add."""
993
if contains_whitespace(version_id):
994
raise InvalidRevisionId(version_id, self.filename)
995
self.check_not_reserved_id(version_id)
996
# Technically this could be avoided if we are happy to allow duplicate
997
# id insertion when other things than bzr core insert texts, but it
998
# seems useful for folk using the knit api directly to have some safety
999
# blanket that we can disable.
1000
if not random_id and self.has_version(version_id):
1001
raise RevisionAlreadyPresent(version_id, self.filename)
1003
self._check_lines_not_unicode(lines)
1004
self._check_lines_are_lines(lines)
1006
def _add(self, version_id, lines, parents, delta, parent_texts,
1007
left_matching_blocks, nostore_sha, random_id):
1008
"""Add a set of lines on top of version specified by parents.
1010
If delta is true, compress the text as a line-delta against
1013
Any versions not present will be converted into ghosts.
1015
# first thing, if the content is something we don't need to store, find
1017
line_bytes = ''.join(lines)
1018
digest = sha_string(line_bytes)
1019
if nostore_sha == digest:
1020
raise errors.ExistingContent
1022
present_parents = []
1023
if parent_texts is None:
1025
for parent in parents:
1026
if self.has_version(parent):
1027
present_parents.append(parent)
1029
# can only compress against the left most present parent.
1031
(len(present_parents) == 0 or
1032
present_parents[0] != parents[0])):
1035
text_length = len(line_bytes)
1038
if lines[-1][-1] != '\n':
1039
# copy the contents of lines.
1041
options.append('no-eol')
1042
lines[-1] = lines[-1] + '\n'
1046
# To speed the extract of texts the delta chain is limited
1047
# to a fixed number of deltas. This should minimize both
1048
# I/O and the time spend applying deltas.
1049
delta = self._check_should_delta(present_parents)
1051
content = self.factory.make(lines, version_id)
1052
if delta or (self.factory.annotated and len(present_parents) > 0):
1053
# Merge annotations from parent texts if needed.
1054
delta_hunks = self._merge_annotations(content, present_parents,
1055
parent_texts, delta, self.factory.annotated,
1056
left_matching_blocks)
1059
options.append('line-delta')
1060
store_lines = self.factory.lower_line_delta(delta_hunks)
1061
size, bytes = self._data._record_to_data(version_id, digest,
1064
options.append('fulltext')
1065
# isinstance is slower and we have no hierarchy.
1066
if self.factory.__class__ == KnitPlainFactory:
1067
# Use the already joined bytes saving iteration time in
1069
size, bytes = self._data._record_to_data(version_id, digest,
1070
lines, [line_bytes])
1072
# get mixed annotation + content and feed it into the
1074
store_lines = self.factory.lower_fulltext(content)
1075
size, bytes = self._data._record_to_data(version_id, digest,
1078
access_memo = self._data.add_raw_records([size], bytes)[0]
1079
self._index.add_versions(
1080
((version_id, options, access_memo, parents),),
1081
random_id=random_id)
1082
return digest, text_length, content
1084
def check(self, progress_bar=None):
1085
"""See VersionedFile.check()."""
1087
def get_lines(self, version_id):
1088
"""See VersionedFile.get_lines()."""
1089
return self.get_line_list([version_id])[0]
1091
def _get_record_map(self, version_ids):
1092
"""Produce a dictionary of knit records.
1094
:return: {version_id:(record, record_details, digest, next)}
1096
data returned from read_records
1098
opaque information to pass to parse_record
1100
SHA1 digest of the full text after all steps are done
1102
build-parent of the version, i.e. the leftmost ancestor.
1103
Will be None if the record is not a delta.
1105
position_map = self._get_components_positions(version_ids)
1106
# c = component_id, r = record_details, i_m = index_memo, n = next
1107
records = [(c, i_m) for c, (r, i_m, n)
1108
in position_map.iteritems()]
1110
for component_id, record, digest in \
1111
self._data.read_records_iter(records):
1112
(record_details, index_memo, next) = position_map[component_id]
1113
record_map[component_id] = record, record_details, digest, next
1117
def get_text(self, version_id):
1118
"""See VersionedFile.get_text"""
1119
return self.get_texts([version_id])[0]
1121
def get_texts(self, version_ids):
1122
return [''.join(l) for l in self.get_line_list(version_ids)]
1124
def get_line_list(self, version_ids):
1125
"""Return the texts of listed versions as a list of strings."""
1126
for version_id in version_ids:
1127
self.check_not_reserved_id(version_id)
1128
text_map, content_map = self._get_content_maps(version_ids)
1129
return [text_map[v] for v in version_ids]
1131
_get_lf_split_line_list = get_line_list
1133
def _get_content_maps(self, version_ids):
1134
"""Produce maps of text and KnitContents
1136
:return: (text_map, content_map) where text_map contains the texts for
1137
the requested versions and content_map contains the KnitContents.
1138
Both dicts take version_ids as their keys.
1140
# FUTURE: This function could be improved for the 'extract many' case
1141
# by tracking each component and only doing the copy when the number of
1142
# children than need to apply delta's to it is > 1 or it is part of the
1144
version_ids = list(version_ids)
1145
multiple_versions = len(version_ids) != 1
1146
record_map = self._get_record_map(version_ids)
1151
for version_id in version_ids:
1154
while cursor is not None:
1155
record, record_details, digest, next = record_map[cursor]
1156
components.append((cursor, record, record_details, digest))
1157
if cursor in content_map:
1162
for (component_id, record, record_details,
1163
digest) in reversed(components):
1164
if component_id in content_map:
1165
content = content_map[component_id]
1167
content, delta = self.factory.parse_record(version_id,
1168
record, record_details, content,
1169
copy_base_content=multiple_versions)
1170
if multiple_versions:
1171
content_map[component_id] = content
1173
content.cleanup_eol(copy_on_mutate=multiple_versions)
1174
final_content[version_id] = content
1176
# digest here is the digest from the last applied component.
1177
text = content.text()
1178
actual_sha = sha_strings(text)
1179
if actual_sha != digest:
1180
raise KnitCorrupt(self.filename,
1182
'\n of reconstructed text does not match'
1184
'\n for version %s' %
1185
(actual_sha, digest, version_id))
1186
text_map[version_id] = text
1187
return text_map, final_content
1189
def iter_lines_added_or_present_in_versions(self, version_ids=None,
1191
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
1192
if version_ids is None:
1193
version_ids = self.versions()
1195
pb = progress.DummyProgress()
1196
# we don't care about inclusions, the caller cares.
1197
# but we need to setup a list of records to visit.
1198
# we need version_id, position, length
1199
version_id_records = []
1200
requested_versions = set(version_ids)
1201
# filter for available versions
1202
for version_id in requested_versions:
1203
if not self.has_version(version_id):
1204
raise RevisionNotPresent(version_id, self.filename)
1205
# get a in-component-order queue:
1206
for version_id in self.versions():
1207
if version_id in requested_versions:
1208
index_memo = self._index.get_position(version_id)
1209
version_id_records.append((version_id, index_memo))
1211
total = len(version_id_records)
1212
for version_idx, (version_id, data, sha_value) in \
1213
enumerate(self._data.read_records_iter(version_id_records)):
1214
pb.update('Walking content.', version_idx, total)
1215
method = self._index.get_method(version_id)
1216
if method == 'fulltext':
1217
line_iterator = self.factory.get_fulltext_content(data)
1218
elif method == 'line-delta':
1219
line_iterator = self.factory.get_linedelta_content(data)
1221
raise ValueError('invalid method %r' % (method,))
1222
# XXX: It might be more efficient to yield (version_id,
1223
# line_iterator) in the future. However for now, this is a simpler
1224
# change to integrate into the rest of the codebase. RBC 20071110
1225
for line in line_iterator:
1226
yield line, version_id
1228
pb.update('Walking content.', total, total)
1230
def num_versions(self):
1231
"""See VersionedFile.num_versions()."""
1232
return self._index.num_versions()
1234
__len__ = num_versions
1236
def annotate(self, version_id):
1237
"""See VersionedFile.annotate."""
1238
return self.factory.annotate(self, version_id)
1240
def get_parent_map(self, version_ids):
1241
"""See VersionedFile.get_parent_map."""
1242
return self._index.get_parent_map(version_ids)
1244
def get_ancestry(self, versions, topo_sorted=True):
1245
"""See VersionedFile.get_ancestry."""
1246
if isinstance(versions, basestring):
1247
versions = [versions]
1250
return self._index.get_ancestry(versions, topo_sorted)
1252
def get_ancestry_with_ghosts(self, versions):
1253
"""See VersionedFile.get_ancestry_with_ghosts."""
1254
if isinstance(versions, basestring):
1255
versions = [versions]
1258
return self._index.get_ancestry_with_ghosts(versions)
1260
def plan_merge(self, ver_a, ver_b):
1261
"""See VersionedFile.plan_merge."""
1262
ancestors_b = set(self.get_ancestry(ver_b, topo_sorted=False))
1263
ancestors_a = set(self.get_ancestry(ver_a, topo_sorted=False))
1264
annotated_a = self.annotate(ver_a)
1265
annotated_b = self.annotate(ver_b)
1266
return merge._plan_annotate_merge(annotated_a, annotated_b,
1267
ancestors_a, ancestors_b)
1270
class _KnitComponentFile(object):
1271
"""One of the files used to implement a knit database"""
1273
def __init__(self, transport, filename, mode, file_mode=None,
1274
create_parent_dir=False, dir_mode=None):
1275
self._transport = transport
1276
self._filename = filename
1278
self._file_mode = file_mode
1279
self._dir_mode = dir_mode
1280
self._create_parent_dir = create_parent_dir
1281
self._need_to_create = False
1283
def _full_path(self):
1284
"""Return the full path to this file."""
1285
return self._transport.base + self._filename
1287
def check_header(self, fp):
1288
line = fp.readline()
1290
# An empty file can actually be treated as though the file doesn't
1292
raise errors.NoSuchFile(self._full_path())
1293
if line != self.HEADER:
1294
raise KnitHeaderError(badline=line,
1295
filename=self._transport.abspath(self._filename))
1298
return '%s(%s)' % (self.__class__.__name__, self._filename)
1301
class _KnitIndex(_KnitComponentFile):
1302
"""Manages knit index file.
1304
The index is already kept in memory and read on startup, to enable
1305
fast lookups of revision information. The cursor of the index
1306
file is always pointing to the end, making it easy to append
1309
_cache is a cache for fast mapping from version id to a Index
1312
_history is a cache for fast mapping from indexes to version ids.
1314
The index data format is dictionary compressed when it comes to
1315
parent references; a index entry may only have parents that with a
1316
lover index number. As a result, the index is topological sorted.
1318
Duplicate entries may be written to the index for a single version id
1319
if this is done then the latter one completely replaces the former:
1320
this allows updates to correct version and parent information.
1321
Note that the two entries may share the delta, and that successive
1322
annotations and references MUST point to the first entry.
1324
The index file on disc contains a header, followed by one line per knit
1325
record. The same revision can be present in an index file more than once.
1326
The first occurrence gets assigned a sequence number starting from 0.
1328
The format of a single line is
1329
REVISION_ID FLAGS BYTE_OFFSET LENGTH( PARENT_ID|PARENT_SEQUENCE_ID)* :\n
1330
REVISION_ID is a utf8-encoded revision id
1331
FLAGS is a comma separated list of flags about the record. Values include
1332
no-eol, line-delta, fulltext.
1333
BYTE_OFFSET is the ascii representation of the byte offset in the data file
1334
that the the compressed data starts at.
1335
LENGTH is the ascii representation of the length of the data file.
1336
PARENT_ID a utf-8 revision id prefixed by a '.' that is a parent of
1338
PARENT_SEQUENCE_ID the ascii representation of the sequence number of a
1339
revision id already in the knit that is a parent of REVISION_ID.
1340
The ' :' marker is the end of record marker.
1343
when a write is interrupted to the index file, it will result in a line
1344
that does not end in ' :'. If the ' :' is not present at the end of a line,
1345
or at the end of the file, then the record that is missing it will be
1346
ignored by the parser.
1348
When writing new records to the index file, the data is preceded by '\n'
1349
to ensure that records always start on new lines even if the last write was
1350
interrupted. As a result its normal for the last line in the index to be
1351
missing a trailing newline. One can be added with no harmful effects.
1354
HEADER = "# bzr knit index 8\n"
1356
# speed of knit parsing went from 280 ms to 280 ms with slots addition.
1357
# __slots__ = ['_cache', '_history', '_transport', '_filename']
1359
def _cache_version(self, version_id, options, pos, size, parents):
1360
"""Cache a version record in the history array and index cache.
1362
This is inlined into _load_data for performance. KEEP IN SYNC.
1363
(It saves 60ms, 25% of the __init__ overhead on local 4000 record
1366
# only want the _history index to reference the 1st index entry
1368
if version_id not in self._cache:
1369
index = len(self._history)
1370
self._history.append(version_id)
1372
index = self._cache[version_id][5]
1373
self._cache[version_id] = (version_id,
1380
def _check_write_ok(self):
1381
if self._get_scope() != self._scope:
1382
raise errors.OutSideTransaction()
1383
if self._mode != 'w':
1384
raise errors.ReadOnlyObjectDirtiedError(self)
1386
def __init__(self, transport, filename, mode, create=False, file_mode=None,
1387
create_parent_dir=False, delay_create=False, dir_mode=None,
1389
_KnitComponentFile.__init__(self, transport, filename, mode,
1390
file_mode=file_mode,
1391
create_parent_dir=create_parent_dir,
1394
# position in _history is the 'official' index for a revision
1395
# but the values may have come from a newer entry.
1396
# so - wc -l of a knit index is != the number of unique names
1400
fp = self._transport.get(self._filename)
1402
# _load_data may raise NoSuchFile if the target knit is
1404
_load_data(self, fp)
1408
if mode != 'w' or not create:
1411
self._need_to_create = True
1413
self._transport.put_bytes_non_atomic(
1414
self._filename, self.HEADER, mode=self._file_mode)
1415
self._scope = get_scope()
1416
self._get_scope = get_scope
1418
def get_ancestry(self, versions, topo_sorted=True):
1419
"""See VersionedFile.get_ancestry."""
1420
# get a graph of all the mentioned versions:
1422
pending = set(versions)
1425
version = pending.pop()
1428
parents = [p for p in cache[version][4] if p in cache]
1430
raise RevisionNotPresent(version, self._filename)
1431
# if not completed and not a ghost
1432
pending.update([p for p in parents if p not in graph])
1433
graph[version] = parents
1436
return topo_sort(graph.items())
1438
def get_ancestry_with_ghosts(self, versions):
1439
"""See VersionedFile.get_ancestry_with_ghosts."""
1440
# get a graph of all the mentioned versions:
1441
self.check_versions_present(versions)
1444
pending = set(versions)
1446
version = pending.pop()
1448
parents = cache[version][4]
1454
pending.update([p for p in parents if p not in graph])
1455
graph[version] = parents
1456
return topo_sort(graph.items())
1458
def get_build_details(self, version_ids):
1459
"""Get the method, index_memo and compression parent for version_ids.
1461
Ghosts are omitted from the result.
1463
:param version_ids: An iterable of version_ids.
1464
:return: A dict of version_id:(index_memo, compression_parent,
1465
parents, record_details).
1467
opaque structure to pass to read_records to extract the raw
1470
Content that this record is built upon, may be None
1472
Logical parents of this node
1474
extra information about the content which needs to be passed to
1475
Factory.parse_record
1478
for version_id in version_ids:
1479
if version_id not in self._cache:
1480
# ghosts are omitted
1482
method = self.get_method(version_id)
1483
parents = self.get_parents_with_ghosts(version_id)
1484
if method == 'fulltext':
1485
compression_parent = None
1487
compression_parent = parents[0]
1488
noeol = 'no-eol' in self.get_options(version_id)
1489
index_memo = self.get_position(version_id)
1490
result[version_id] = (index_memo, compression_parent,
1491
parents, (method, noeol))
1494
def num_versions(self):
1495
return len(self._history)
1497
__len__ = num_versions
1499
def get_versions(self):
1500
"""Get all the versions in the file. not topologically sorted."""
1501
return self._history
1503
def _version_list_to_index(self, versions):
1506
for version in versions:
1507
if version in cache:
1508
# -- inlined lookup() --
1509
result_list.append(str(cache[version][5]))
1510
# -- end lookup () --
1512
result_list.append('.' + version)
1513
return ' '.join(result_list)
1515
def add_version(self, version_id, options, index_memo, parents):
1516
"""Add a version record to the index."""
1517
self.add_versions(((version_id, options, index_memo, parents),))
1519
def add_versions(self, versions, random_id=False):
1520
"""Add multiple versions to the index.
1522
:param versions: a list of tuples:
1523
(version_id, options, pos, size, parents).
1524
:param random_id: If True the ids being added were randomly generated
1525
and no check for existence will be performed.
1528
orig_history = self._history[:]
1529
orig_cache = self._cache.copy()
1532
for version_id, options, (index, pos, size), parents in versions:
1533
line = "\n%s %s %s %s %s :" % (version_id,
1537
self._version_list_to_index(parents))
1539
self._cache_version(version_id, options, pos, size, tuple(parents))
1540
if not self._need_to_create:
1541
self._transport.append_bytes(self._filename, ''.join(lines))
1544
sio.write(self.HEADER)
1545
sio.writelines(lines)
1547
self._transport.put_file_non_atomic(self._filename, sio,
1548
create_parent_dir=self._create_parent_dir,
1549
mode=self._file_mode,
1550
dir_mode=self._dir_mode)
1551
self._need_to_create = False
1553
# If any problems happen, restore the original values and re-raise
1554
self._history = orig_history
1555
self._cache = orig_cache
1558
def has_version(self, version_id):
1559
"""True if the version is in the index."""
1560
return version_id in self._cache
1562
def get_position(self, version_id):
1563
"""Return details needed to access the version.
1565
.kndx indices do not support split-out data, so return None for the
1568
:return: a tuple (None, data position, size) to hand to the access
1569
logic to get the record.
1571
entry = self._cache[version_id]
1572
return None, entry[2], entry[3]
1574
def get_method(self, version_id):
1575
"""Return compression method of specified version."""
1577
options = self._cache[version_id][1]
1579
raise RevisionNotPresent(version_id, self._filename)
1580
if 'fulltext' in options:
1583
if 'line-delta' not in options:
1584
raise errors.KnitIndexUnknownMethod(self._full_path(), options)
1587
def get_options(self, version_id):
1588
"""Return a list representing options.
1592
return self._cache[version_id][1]
1594
def get_parent_map(self, version_ids):
1595
"""Passed through to by KnitVersionedFile.get_parent_map."""
1597
for version_id in version_ids:
1599
result[version_id] = tuple(self._cache[version_id][4])
1604
def get_parents_with_ghosts(self, version_id):
1605
"""Return parents of specified version with ghosts."""
1607
return self.get_parent_map([version_id])[version_id]
1609
raise RevisionNotPresent(version_id, self)
1611
def check_versions_present(self, version_ids):
1612
"""Check that all specified versions are present."""
1614
for version_id in version_ids:
1615
if version_id not in cache:
1616
raise RevisionNotPresent(version_id, self._filename)
1619
class KnitGraphIndex(object):
1620
"""A knit index that builds on GraphIndex."""
1622
def __init__(self, graph_index, deltas=False, parents=True, add_callback=None):
1623
"""Construct a KnitGraphIndex on a graph_index.
1625
:param graph_index: An implementation of bzrlib.index.GraphIndex.
1626
:param deltas: Allow delta-compressed records.
1627
:param add_callback: If not None, allow additions to the index and call
1628
this callback with a list of added GraphIndex nodes:
1629
[(node, value, node_refs), ...]
1630
:param parents: If True, record knits parents, if not do not record
1633
self._graph_index = graph_index
1634
self._deltas = deltas
1635
self._add_callback = add_callback
1636
self._parents = parents
1637
if deltas and not parents:
1638
raise KnitCorrupt(self, "Cannot do delta compression without "
1641
def _check_write_ok(self):
1644
def _get_entries(self, keys, check_present=False):
1645
"""Get the entries for keys.
1647
:param keys: An iterable of index keys, - 1-tuples.
1652
for node in self._graph_index.iter_entries(keys):
1654
found_keys.add(node[1])
1656
# adapt parentless index to the rest of the code.
1657
for node in self._graph_index.iter_entries(keys):
1658
yield node[0], node[1], node[2], ()
1659
found_keys.add(node[1])
1661
missing_keys = keys.difference(found_keys)
1663
raise RevisionNotPresent(missing_keys.pop(), self)
1665
def _present_keys(self, version_ids):
1667
node[1] for node in self._get_entries(version_ids)])
1669
def _parentless_ancestry(self, versions):
1670
"""Honour the get_ancestry API for parentless knit indices."""
1671
wanted_keys = self._version_ids_to_keys(versions)
1672
present_keys = self._present_keys(wanted_keys)
1673
missing = set(wanted_keys).difference(present_keys)
1675
raise RevisionNotPresent(missing.pop(), self)
1676
return list(self._keys_to_version_ids(present_keys))
1678
def get_ancestry(self, versions, topo_sorted=True):
1679
"""See VersionedFile.get_ancestry."""
1680
if not self._parents:
1681
return self._parentless_ancestry(versions)
1682
# XXX: This will do len(history) index calls - perhaps
1683
# it should be altered to be a index core feature?
1684
# get a graph of all the mentioned versions:
1687
versions = self._version_ids_to_keys(versions)
1688
pending = set(versions)
1690
# get all pending nodes
1691
this_iteration = pending
1692
new_nodes = self._get_entries(this_iteration)
1695
for (index, key, value, node_refs) in new_nodes:
1696
# dont ask for ghosties - otherwise
1697
# we we can end up looping with pending
1698
# being entirely ghosted.
1699
graph[key] = [parent for parent in node_refs[0]
1700
if parent not in ghosts]
1702
for parent in graph[key]:
1703
# dont examine known nodes again
1708
ghosts.update(this_iteration.difference(found))
1709
if versions.difference(graph):
1710
raise RevisionNotPresent(versions.difference(graph).pop(), self)
1712
result_keys = topo_sort(graph.items())
1714
result_keys = graph.iterkeys()
1715
return [key[0] for key in result_keys]
1717
def get_ancestry_with_ghosts(self, versions):
1718
"""See VersionedFile.get_ancestry."""
1719
if not self._parents:
1720
return self._parentless_ancestry(versions)
1721
# XXX: This will do len(history) index calls - perhaps
1722
# it should be altered to be a index core feature?
1723
# get a graph of all the mentioned versions:
1725
versions = self._version_ids_to_keys(versions)
1726
pending = set(versions)
1728
# get all pending nodes
1729
this_iteration = pending
1730
new_nodes = self._get_entries(this_iteration)
1732
for (index, key, value, node_refs) in new_nodes:
1733
graph[key] = node_refs[0]
1735
for parent in graph[key]:
1736
# dont examine known nodes again
1740
missing_versions = this_iteration.difference(graph)
1741
missing_needed = versions.intersection(missing_versions)
1743
raise RevisionNotPresent(missing_needed.pop(), self)
1744
for missing_version in missing_versions:
1745
# add a key, no parents
1746
graph[missing_version] = []
1747
pending.discard(missing_version) # don't look for it
1748
result_keys = topo_sort(graph.items())
1749
return [key[0] for key in result_keys]
1751
def get_build_details(self, version_ids):
1752
"""Get the method, index_memo and compression parent for version_ids.
1754
Ghosts are omitted from the result.
1756
:param version_ids: An iterable of version_ids.
1757
:return: A dict of version_id:(index_memo, compression_parent,
1758
parents, record_details).
1760
opaque structure to pass to read_records to extract the raw
1763
Content that this record is built upon, may be None
1765
Logical parents of this node
1767
extra information about the content which needs to be passed to
1768
Factory.parse_record
1771
entries = self._get_entries(self._version_ids_to_keys(version_ids), True)
1772
for entry in entries:
1773
version_id = self._keys_to_version_ids((entry[1],))[0]
1774
if not self._parents:
1777
parents = self._keys_to_version_ids(entry[3][0])
1778
if not self._deltas:
1779
compression_parent = None
1781
compression_parent_key = self._compression_parent(entry)
1782
if compression_parent_key:
1783
compression_parent = self._keys_to_version_ids(
1784
(compression_parent_key,))[0]
1786
compression_parent = None
1787
noeol = (entry[2][0] == 'N')
1788
if compression_parent:
1789
method = 'line-delta'
1792
result[version_id] = (self._node_to_position(entry),
1793
compression_parent, parents,
1797
def _compression_parent(self, an_entry):
1798
# return the key that an_entry is compressed against, or None
1799
# Grab the second parent list (as deltas implies parents currently)
1800
compression_parents = an_entry[3][1]
1801
if not compression_parents:
1803
return compression_parents[0]
1805
def _get_method(self, node):
1806
if not self._deltas:
1808
if self._compression_parent(node):
1813
def num_versions(self):
1814
return len(list(self._graph_index.iter_all_entries()))
1816
__len__ = num_versions
1818
def get_versions(self):
1819
"""Get all the versions in the file. not topologically sorted."""
1820
return [node[1][0] for node in self._graph_index.iter_all_entries()]
1822
def has_version(self, version_id):
1823
"""True if the version is in the index."""
1824
return len(self._present_keys(self._version_ids_to_keys([version_id]))) == 1
1826
def _keys_to_version_ids(self, keys):
1827
return tuple(key[0] for key in keys)
1829
def get_position(self, version_id):
1830
"""Return details needed to access the version.
1832
:return: a tuple (index, data position, size) to hand to the access
1833
logic to get the record.
1835
node = self._get_node(version_id)
1836
return self._node_to_position(node)
1838
def _node_to_position(self, node):
1839
"""Convert an index value to position details."""
1840
bits = node[2][1:].split(' ')
1841
return node[0], int(bits[0]), int(bits[1])
1843
def get_method(self, version_id):
1844
"""Return compression method of specified version."""
1845
return self._get_method(self._get_node(version_id))
1847
def _get_node(self, version_id):
1849
return list(self._get_entries(self._version_ids_to_keys([version_id])))[0]
1851
raise RevisionNotPresent(version_id, self)
1853
def get_options(self, version_id):
1854
"""Return a list representing options.
1858
node = self._get_node(version_id)
1859
options = [self._get_method(node)]
1860
if node[2][0] == 'N':
1861
options.append('no-eol')
1864
def get_parent_map(self, version_ids):
1865
"""Passed through to by KnitVersionedFile.get_parent_map."""
1866
nodes = self._get_entries(self._version_ids_to_keys(version_ids))
1870
result[node[1][0]] = self._keys_to_version_ids(node[3][0])
1873
result[node[1][0]] = ()
1876
def get_parents_with_ghosts(self, version_id):
1877
"""Return parents of specified version with ghosts."""
1879
return self.get_parent_map([version_id])[version_id]
1881
raise RevisionNotPresent(version_id, self)
1883
def check_versions_present(self, version_ids):
1884
"""Check that all specified versions are present."""
1885
keys = self._version_ids_to_keys(version_ids)
1886
present = self._present_keys(keys)
1887
missing = keys.difference(present)
1889
raise RevisionNotPresent(missing.pop(), self)
1891
def add_version(self, version_id, options, access_memo, parents):
1892
"""Add a version record to the index."""
1893
return self.add_versions(((version_id, options, access_memo, parents),))
1895
def add_versions(self, versions, random_id=False):
1896
"""Add multiple versions to the index.
1898
This function does not insert data into the Immutable GraphIndex
1899
backing the KnitGraphIndex, instead it prepares data for insertion by
1900
the caller and checks that it is safe to insert then calls
1901
self._add_callback with the prepared GraphIndex nodes.
1903
:param versions: a list of tuples:
1904
(version_id, options, pos, size, parents).
1905
:param random_id: If True the ids being added were randomly generated
1906
and no check for existence will be performed.
1908
if not self._add_callback:
1909
raise errors.ReadOnlyError(self)
1910
# we hope there are no repositories with inconsistent parentage
1915
for (version_id, options, access_memo, parents) in versions:
1916
index, pos, size = access_memo
1917
key = (version_id, )
1918
parents = tuple((parent, ) for parent in parents)
1919
if 'no-eol' in options:
1923
value += "%d %d" % (pos, size)
1924
if not self._deltas:
1925
if 'line-delta' in options:
1926
raise KnitCorrupt(self, "attempt to add line-delta in non-delta knit")
1929
if 'line-delta' in options:
1930
node_refs = (parents, (parents[0],))
1932
node_refs = (parents, ())
1934
node_refs = (parents, )
1937
raise KnitCorrupt(self, "attempt to add node with parents "
1938
"in parentless index.")
1940
keys[key] = (value, node_refs)
1942
present_nodes = self._get_entries(keys)
1943
for (index, key, value, node_refs) in present_nodes:
1944
if (value, node_refs) != keys[key]:
1945
raise KnitCorrupt(self, "inconsistent details in add_versions"
1946
": %s %s" % ((value, node_refs), keys[key]))
1950
for key, (value, node_refs) in keys.iteritems():
1951
result.append((key, value, node_refs))
1953
for key, (value, node_refs) in keys.iteritems():
1954
result.append((key, value))
1955
self._add_callback(result)
1957
def _version_ids_to_keys(self, version_ids):
1958
return set((version_id, ) for version_id in version_ids)
1961
class _KnitAccess(object):
1962
"""Access to knit records in a .knit file."""
1964
def __init__(self, transport, filename, _file_mode, _dir_mode,
1965
_need_to_create, _create_parent_dir):
1966
"""Create a _KnitAccess for accessing and inserting data.
1968
:param transport: The transport the .knit is located on.
1969
:param filename: The filename of the .knit.
1971
self._transport = transport
1972
self._filename = filename
1973
self._file_mode = _file_mode
1974
self._dir_mode = _dir_mode
1975
self._need_to_create = _need_to_create
1976
self._create_parent_dir = _create_parent_dir
1978
def add_raw_records(self, sizes, raw_data):
1979
"""Add raw knit bytes to a storage area.
1981
The data is spooled to whereever the access method is storing data.
1983
:param sizes: An iterable containing the size of each raw data segment.
1984
:param raw_data: A bytestring containing the data.
1985
:return: A list of memos to retrieve the record later. Each memo is a
1986
tuple - (index, pos, length), where the index field is always None
1987
for the .knit access method.
1989
if not self._need_to_create:
1990
base = self._transport.append_bytes(self._filename, raw_data)
1992
self._transport.put_bytes_non_atomic(self._filename, raw_data,
1993
create_parent_dir=self._create_parent_dir,
1994
mode=self._file_mode,
1995
dir_mode=self._dir_mode)
1996
self._need_to_create = False
2000
result.append((None, base, size))
2005
"""IFF this data access has its own storage area, initialise it.
2009
self._transport.put_bytes_non_atomic(self._filename, '',
2010
mode=self._file_mode)
2012
def open_file(self):
2013
"""IFF this data access can be represented as a single file, open it.
2015
For knits that are not mapped to a single file on disk this will
2018
:return: None or a file handle.
2021
return self._transport.get(self._filename)
2026
def get_raw_records(self, memos_for_retrieval):
2027
"""Get the raw bytes for a records.
2029
:param memos_for_retrieval: An iterable containing the (index, pos,
2030
length) memo for retrieving the bytes. The .knit method ignores
2031
the index as there is always only a single file.
2032
:return: An iterator over the bytes of the records.
2034
read_vector = [(pos, size) for (index, pos, size) in memos_for_retrieval]
2035
for pos, data in self._transport.readv(self._filename, read_vector):
2039
class _PackAccess(object):
2040
"""Access to knit records via a collection of packs."""
2042
def __init__(self, index_to_packs, writer=None):
2043
"""Create a _PackAccess object.
2045
:param index_to_packs: A dict mapping index objects to the transport
2046
and file names for obtaining data.
2047
:param writer: A tuple (pack.ContainerWriter, write_index) which
2048
contains the pack to write, and the index that reads from it will
2052
self.container_writer = writer[0]
2053
self.write_index = writer[1]
2055
self.container_writer = None
2056
self.write_index = None
2057
self.indices = index_to_packs
2059
def add_raw_records(self, sizes, raw_data):
2060
"""Add raw knit bytes to a storage area.
2062
The data is spooled to the container writer in one bytes-record per
2065
:param sizes: An iterable containing the size of each raw data segment.
2066
:param raw_data: A bytestring containing the data.
2067
:return: A list of memos to retrieve the record later. Each memo is a
2068
tuple - (index, pos, length), where the index field is the
2069
write_index object supplied to the PackAccess object.
2074
p_offset, p_length = self.container_writer.add_bytes_record(
2075
raw_data[offset:offset+size], [])
2077
result.append((self.write_index, p_offset, p_length))
2081
"""Pack based knits do not get individually created."""
2083
def get_raw_records(self, memos_for_retrieval):
2084
"""Get the raw bytes for a records.
2086
:param memos_for_retrieval: An iterable containing the (index, pos,
2087
length) memo for retrieving the bytes. The Pack access method
2088
looks up the pack to use for a given record in its index_to_pack
2090
:return: An iterator over the bytes of the records.
2092
# first pass, group into same-index requests
2094
current_index = None
2095
for (index, offset, length) in memos_for_retrieval:
2096
if current_index == index:
2097
current_list.append((offset, length))
2099
if current_index is not None:
2100
request_lists.append((current_index, current_list))
2101
current_index = index
2102
current_list = [(offset, length)]
2103
# handle the last entry
2104
if current_index is not None:
2105
request_lists.append((current_index, current_list))
2106
for index, offsets in request_lists:
2107
transport, path = self.indices[index]
2108
reader = pack.make_readv_reader(transport, path, offsets)
2109
for names, read_func in reader.iter_records():
2110
yield read_func(None)
2112
def open_file(self):
2113
"""Pack based knits have no single file."""
2116
def set_writer(self, writer, index, (transport, packname)):
2117
"""Set a writer to use for adding data."""
2118
if index is not None:
2119
self.indices[index] = (transport, packname)
2120
self.container_writer = writer
2121
self.write_index = index
2124
class _StreamAccess(object):
2125
"""A Knit Access object that provides data from a datastream.
2127
It also provides a fallback to present as unannotated data, annotated data
2128
from a *backing* access object.
2130
This is triggered by a index_memo which is pointing to a different index
2131
than this was constructed with, and is used to allow extracting full
2132
unannotated texts for insertion into annotated knits.
2135
def __init__(self, reader_callable, stream_index, backing_knit,
2137
"""Create a _StreamAccess object.
2139
:param reader_callable: The reader_callable from the datastream.
2140
This is called to buffer all the data immediately, for
2142
:param stream_index: The index the data stream this provides access to
2143
which will be present in native index_memo's.
2144
:param backing_knit: The knit object that will provide access to
2145
annotated texts which are not available in the stream, so as to
2146
create unannotated texts.
2147
:param orig_factory: The original content factory used to generate the
2148
stream. This is used for checking whether the thunk code for
2149
supporting _copy_texts will generate the correct form of data.
2151
self.data = reader_callable(None)
2152
self.stream_index = stream_index
2153
self.backing_knit = backing_knit
2154
self.orig_factory = orig_factory
2156
def get_raw_records(self, memos_for_retrieval):
2157
"""Get the raw bytes for a records.
2159
:param memos_for_retrieval: An iterable of memos from the
2160
_StreamIndex object identifying bytes to read; for these classes
2161
they are (from_backing_knit, index, start, end) and can point to
2162
either the backing knit or streamed data.
2163
:return: An iterator yielding a byte string for each record in
2164
memos_for_retrieval.
2166
# use a generator for memory friendliness
2167
for from_backing_knit, version_id, start, end in memos_for_retrieval:
2168
if not from_backing_knit:
2169
if version_id is not self.stream_index:
2170
raise AssertionError()
2171
yield self.data[start:end]
2173
# we have been asked to thunk. This thunking only occurs when
2174
# we are obtaining plain texts from an annotated backing knit
2175
# so that _copy_texts will work.
2176
# We could improve performance here by scanning for where we need
2177
# to do this and using get_line_list, then interleaving the output
2178
# as desired. However, for now, this is sufficient.
2179
if self.orig_factory.__class__ != KnitPlainFactory:
2180
raise errors.KnitCorrupt(
2181
self, 'Bad thunk request %r cannot be backed by %r' %
2182
(version_id, self.orig_factory))
2183
lines = self.backing_knit.get_lines(version_id)
2184
line_bytes = ''.join(lines)
2185
digest = sha_string(line_bytes)
2186
# the packed form of the fulltext always has a trailing newline,
2187
# even if the actual text does not, unless the file is empty. the
2188
# record options including the noeol flag are passed through by
2189
# _StreamIndex, so this is safe.
2191
if lines[-1][-1] != '\n':
2192
lines[-1] = lines[-1] + '\n'
2194
# We want plain data, because we expect to thunk only to allow text
2196
size, bytes = self.backing_knit._data._record_to_data(version_id,
2197
digest, lines, line_bytes)
2201
class _StreamIndex(object):
2202
"""A Knit Index object that uses the data map from a datastream."""
2204
def __init__(self, data_list, backing_index):
2205
"""Create a _StreamIndex object.
2207
:param data_list: The data_list from the datastream.
2208
:param backing_index: The index which will supply values for nodes
2209
referenced outside of this stream.
2211
self.data_list = data_list
2212
self.backing_index = backing_index
2213
self._by_version = {}
2215
for key, options, length, parents in data_list:
2216
self._by_version[key] = options, (pos, pos + length), parents
2219
def get_ancestry(self, versions, topo_sorted):
2220
"""Get an ancestry list for versions."""
2222
# Not needed for basic joins
2223
raise NotImplementedError(self.get_ancestry)
2224
# get a graph of all the mentioned versions:
2225
# Little ugly - basically copied from KnitIndex, but don't want to
2226
# accidentally incorporate too much of that index's code.
2228
pending = set(versions)
2229
cache = self._by_version
2231
version = pending.pop()
2234
parents = [p for p in cache[version][2] if p in cache]
2236
raise RevisionNotPresent(version, self)
2237
# if not completed and not a ghost
2238
pending.update([p for p in parents if p not in ancestry])
2239
ancestry.add(version)
2240
return list(ancestry)
2242
def get_build_details(self, version_ids):
2243
"""Get the method, index_memo and compression parent for version_ids.
2245
Ghosts are omitted from the result.
2247
:param version_ids: An iterable of version_ids.
2248
:return: A dict of version_id:(index_memo, compression_parent,
2249
parents, record_details).
2251
opaque memo that can be passed to _StreamAccess.read_records
2252
to extract the raw data; for these classes it is
2253
(from_backing_knit, index, start, end)
2255
Content that this record is built upon, may be None
2257
Logical parents of this node
2259
extra information about the content which needs to be passed to
2260
Factory.parse_record
2263
for version_id in version_ids:
2265
method = self.get_method(version_id)
2266
except errors.RevisionNotPresent:
2267
# ghosts are omitted
2269
parent_ids = self.get_parents_with_ghosts(version_id)
2270
noeol = ('no-eol' in self.get_options(version_id))
2271
index_memo = self.get_position(version_id)
2272
from_backing_knit = index_memo[0]
2273
if from_backing_knit:
2274
# texts retrieved from the backing knit are always full texts
2276
if method == 'fulltext':
2277
compression_parent = None
2279
compression_parent = parent_ids[0]
2280
result[version_id] = (index_memo, compression_parent,
2281
parent_ids, (method, noeol))
2284
def get_method(self, version_id):
2285
"""Return compression method of specified version."""
2286
options = self.get_options(version_id)
2287
if 'fulltext' in options:
2289
elif 'line-delta' in options:
2292
raise errors.KnitIndexUnknownMethod(self, options)
2294
def get_options(self, version_id):
2295
"""Return a list representing options.
2300
return self._by_version[version_id][0]
2302
options = list(self.backing_index.get_options(version_id))
2303
if 'fulltext' in options:
2305
elif 'line-delta' in options:
2306
# Texts from the backing knit are always returned from the stream
2308
options.remove('line-delta')
2309
options.append('fulltext')
2311
raise errors.KnitIndexUnknownMethod(self, options)
2312
return tuple(options)
2314
def get_parent_map(self, version_ids):
2315
"""Passed through to by KnitVersionedFile.get_parent_map."""
2318
for version_id in version_ids:
2320
result[version_id] = self._by_version[version_id][2]
2322
pending_ids.add(version_id)
2323
result.update(self.backing_index.get_parent_map(pending_ids))
2326
def get_parents_with_ghosts(self, version_id):
2327
"""Return parents of specified version with ghosts."""
2329
return self.get_parent_map([version_id])[version_id]
2331
raise RevisionNotPresent(version_id, self)
2333
def get_position(self, version_id):
2334
"""Return details needed to access the version.
2336
_StreamAccess has the data as a big array, so we return slice
2337
coordinates into that (as index_memo's are opaque outside the
2338
index and matching access class).
2340
:return: a tuple (from_backing_knit, index, start, end) that can
2341
be passed e.g. to get_raw_records.
2342
If from_backing_knit is False, index will be self, otherwise it
2343
will be a version id.
2346
start, end = self._by_version[version_id][1]
2347
return False, self, start, end
2349
# Signal to the access object to handle this from the backing knit.
2350
return (True, version_id, None, None)
2352
def get_versions(self):
2353
"""Get all the versions in the stream."""
2354
return self._by_version.keys()
2357
class _KnitData(object):
2358
"""Manage extraction of data from a KnitAccess, caching and decompressing.
2360
The KnitData class provides the logic for parsing and using knit records,
2361
making use of an access method for the low level read and write operations.
2364
def __init__(self, access):
2365
"""Create a KnitData object.
2367
:param access: The access method to use. Access methods such as
2368
_KnitAccess manage the insertion of raw records and the subsequent
2369
retrieval of the same.
2371
self._access = access
2372
self._checked = False
2374
def _open_file(self):
2375
return self._access.open_file()
2377
def _record_to_data(self, version_id, digest, lines, dense_lines=None):
2378
"""Convert version_id, digest, lines into a raw data block.
2380
:param dense_lines: The bytes of lines but in a denser form. For
2381
instance, if lines is a list of 1000 bytestrings each ending in \n,
2382
dense_lines may be a list with one line in it, containing all the
2383
1000's lines and their \n's. Using dense_lines if it is already
2384
known is a win because the string join to create bytes in this
2385
function spends less time resizing the final string.
2386
:return: (len, a StringIO instance with the raw data ready to read.)
2388
# Note: using a string copy here increases memory pressure with e.g.
2389
# ISO's, but it is about 3 seconds faster on a 1.2Ghz intel machine
2390
# when doing the initial commit of a mozilla tree. RBC 20070921
2391
bytes = ''.join(chain(
2392
["version %s %d %s\n" % (version_id,
2395
dense_lines or lines,
2396
["end %s\n" % version_id]))
2397
compressed_bytes = bytes_to_gzip(bytes)
2398
return len(compressed_bytes), compressed_bytes
2400
def add_raw_records(self, sizes, raw_data):
2401
"""Append a prepared record to the data file.
2403
:param sizes: An iterable containing the size of each raw data segment.
2404
:param raw_data: A bytestring containing the data.
2405
:return: a list of index data for the way the data was stored.
2406
See the access method add_raw_records documentation for more
2409
return self._access.add_raw_records(sizes, raw_data)
2411
def _parse_record_header(self, version_id, raw_data):
2412
"""Parse a record header for consistency.
2414
:return: the header and the decompressor stream.
2415
as (stream, header_record)
2417
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
2419
rec = self._check_header(version_id, df.readline())
2420
except Exception, e:
2421
raise KnitCorrupt(self._access,
2422
"While reading {%s} got %s(%s)"
2423
% (version_id, e.__class__.__name__, str(e)))
2426
def _check_header(self, version_id, line):
2429
raise KnitCorrupt(self._access,
2430
'unexpected number of elements in record header')
2431
if rec[1] != version_id:
2432
raise KnitCorrupt(self._access,
2433
'unexpected version, wanted %r, got %r'
2434
% (version_id, rec[1]))
2437
def _parse_record(self, version_id, data):
2439
# 4168 calls in 2880 217 internal
2440
# 4168 calls to _parse_record_header in 2121
2441
# 4168 calls to readlines in 330
2442
df = GzipFile(mode='rb', fileobj=StringIO(data))
2445
record_contents = df.readlines()
2446
except Exception, e:
2447
raise KnitCorrupt(self._access,
2448
"While reading {%s} got %s(%s)"
2449
% (version_id, e.__class__.__name__, str(e)))
2450
header = record_contents.pop(0)
2451
rec = self._check_header(version_id, header)
2453
last_line = record_contents.pop()
2454
if len(record_contents) != int(rec[2]):
2455
raise KnitCorrupt(self._access,
2456
'incorrect number of lines %s != %s'
2458
% (len(record_contents), int(rec[2]),
2460
if last_line != 'end %s\n' % rec[1]:
2461
raise KnitCorrupt(self._access,
2462
'unexpected version end line %r, wanted %r'
2463
% (last_line, version_id))
2465
return record_contents, rec[3]
2467
def read_records_iter_raw(self, records):
2468
"""Read text records from data file and yield raw data.
2470
This unpacks enough of the text record to validate the id is
2471
as expected but thats all.
2473
# setup an iterator of the external records:
2474
# uses readv so nice and fast we hope.
2476
# grab the disk data needed.
2477
needed_offsets = [index_memo for version_id, index_memo
2479
raw_records = self._access.get_raw_records(needed_offsets)
2481
for version_id, index_memo in records:
2482
data = raw_records.next()
2483
# validate the header
2484
df, rec = self._parse_record_header(version_id, data)
2486
yield version_id, data
2488
def read_records_iter(self, records):
2489
"""Read text records from data file and yield result.
2491
The result will be returned in whatever is the fastest to read.
2492
Not by the order requested. Also, multiple requests for the same
2493
record will only yield 1 response.
2494
:param records: A list of (version_id, pos, len) entries
2495
:return: Yields (version_id, contents, digest) in the order
2496
read, not the order requested
2501
needed_records = sorted(set(records), key=operator.itemgetter(1))
2502
if not needed_records:
2505
# The transport optimizes the fetching as well
2506
# (ie, reads continuous ranges.)
2507
raw_data = self._access.get_raw_records(
2508
[index_memo for version_id, index_memo in needed_records])
2510
for (version_id, index_memo), data in \
2511
izip(iter(needed_records), raw_data):
2512
content, digest = self._parse_record(version_id, data)
2513
yield version_id, content, digest
2515
def read_records(self, records):
2516
"""Read records into a dictionary."""
2518
for record_id, content, digest in \
2519
self.read_records_iter(records):
2520
components[record_id] = (content, digest)
2524
class InterKnit(InterVersionedFile):
2525
"""Optimised code paths for knit to knit operations."""
2527
_matching_file_from_factory = staticmethod(make_file_knit)
2528
_matching_file_to_factory = staticmethod(make_file_knit)
2531
def is_compatible(source, target):
2532
"""Be compatible with knits. """
2534
return (isinstance(source, KnitVersionedFile) and
2535
isinstance(target, KnitVersionedFile))
2536
except AttributeError:
2539
def _copy_texts(self, pb, msg, version_ids, ignore_missing=False):
2540
"""Copy texts to the target by extracting and adding them one by one.
2542
see join() for the parameter definitions.
2544
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2545
# --- the below is factorable out with VersionedFile.join, but wait for
2546
# VersionedFiles, it may all be simpler then.
2547
graph = Graph(self.source)
2548
search = graph._make_breadth_first_searcher(version_ids)
2549
transitive_ids = set()
2550
map(transitive_ids.update, list(search))
2551
parent_map = self.source.get_parent_map(transitive_ids)
2552
order = topo_sort(parent_map.items())
2554
def size_of_content(content):
2555
return sum(len(line) for line in content.text())
2556
# Cache at most 10MB of parent texts
2557
parent_cache = lru_cache.LRUSizeCache(max_size=10*1024*1024,
2558
compute_size=size_of_content)
2559
# TODO: jam 20071116 It would be nice to have a streaming interface to
2560
# get multiple texts from a source. The source could be smarter
2561
# about how it handled intermediate stages.
2562
# get_line_list() or make_mpdiffs() seem like a possibility, but
2563
# at the moment they extract all full texts into memory, which
2564
# causes us to store more than our 3x fulltext goal.
2565
# Repository.iter_files_bytes() may be another possibility
2566
to_process = [version for version in order
2567
if version not in self.target]
2568
total = len(to_process)
2569
pb = ui.ui_factory.nested_progress_bar()
2571
for index, version in enumerate(to_process):
2572
pb.update('Converting versioned data', index, total)
2573
sha1, num_bytes, parent_text = self.target.add_lines(version,
2574
self.source.get_parents_with_ghosts(version),
2575
self.source.get_lines(version),
2576
parent_texts=parent_cache)
2577
parent_cache[version] = parent_text
2582
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2583
"""See InterVersionedFile.join."""
2584
# If the source and target are mismatched w.r.t. annotations vs
2585
# plain, the data needs to be converted accordingly
2586
if self.source.factory.annotated == self.target.factory.annotated:
2588
elif self.source.factory.annotated:
2589
converter = self._anno_to_plain_converter
2591
# We're converting from a plain to an annotated knit. Copy them
2592
# across by full texts.
2593
return self._copy_texts(pb, msg, version_ids, ignore_missing)
2595
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2599
pb = ui.ui_factory.nested_progress_bar()
2601
version_ids = list(version_ids)
2602
if None in version_ids:
2603
version_ids.remove(None)
2605
self.source_ancestry = set(self.source.get_ancestry(version_ids,
2607
this_versions = set(self.target._index.get_versions())
2608
# XXX: For efficiency we should not look at the whole index,
2609
# we only need to consider the referenced revisions - they
2610
# must all be present, or the method must be full-text.
2611
# TODO, RBC 20070919
2612
needed_versions = self.source_ancestry - this_versions
2614
if not needed_versions:
2616
full_list = topo_sort(
2617
self.source.get_parent_map(self.source.versions()))
2619
version_list = [i for i in full_list if (not self.target.has_version(i)
2620
and i in needed_versions)]
2624
copy_queue_records = []
2626
for version_id in version_list:
2627
options = self.source._index.get_options(version_id)
2628
parents = self.source._index.get_parents_with_ghosts(version_id)
2629
# check that its will be a consistent copy:
2630
for parent in parents:
2631
# if source has the parent, we must :
2632
# * already have it or
2633
# * have it scheduled already
2634
# otherwise we don't care
2635
if not (self.target.has_version(parent) or
2636
parent in copy_set or
2637
not self.source.has_version(parent)):
2638
raise AssertionError("problem joining parent %r "
2640
% (parent, self.source, self.target))
2641
index_memo = self.source._index.get_position(version_id)
2642
copy_queue_records.append((version_id, index_memo))
2643
copy_queue.append((version_id, options, parents))
2644
copy_set.add(version_id)
2646
# data suck the join:
2648
total = len(version_list)
2651
for (version_id, raw_data), \
2652
(version_id2, options, parents) in \
2653
izip(self.source._data.read_records_iter_raw(copy_queue_records),
2655
if not (version_id == version_id2):
2656
raise AssertionError('logic error, inconsistent results')
2658
pb.update("Joining knit", count, total)
2660
size, raw_data = converter(raw_data, version_id, options,
2663
size = len(raw_data)
2664
raw_records.append((version_id, options, parents, size))
2665
raw_datum.append(raw_data)
2666
self.target._add_raw_records(raw_records, ''.join(raw_datum))
2671
def _anno_to_plain_converter(self, raw_data, version_id, options,
2673
"""Convert annotated content to plain content."""
2674
data, digest = self.source._data._parse_record(version_id, raw_data)
2675
if 'fulltext' in options:
2676
content = self.source.factory.parse_fulltext(data, version_id)
2677
lines = self.target.factory.lower_fulltext(content)
2679
delta = self.source.factory.parse_line_delta(data, version_id,
2681
lines = self.target.factory.lower_line_delta(delta)
2682
return self.target._data._record_to_data(version_id, digest, lines)
2685
InterVersionedFile.register_optimiser(InterKnit)
2688
class WeaveToKnit(InterVersionedFile):
2689
"""Optimised code paths for weave to knit operations."""
2691
_matching_file_from_factory = bzrlib.weave.WeaveFile
2692
_matching_file_to_factory = staticmethod(make_file_knit)
2695
def is_compatible(source, target):
2696
"""Be compatible with weaves to knits."""
2698
return (isinstance(source, bzrlib.weave.Weave) and
2699
isinstance(target, KnitVersionedFile))
2700
except AttributeError:
2703
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
2704
"""See InterVersionedFile.join."""
2705
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
2710
pb = ui.ui_factory.nested_progress_bar()
2712
version_ids = list(version_ids)
2714
self.source_ancestry = set(self.source.get_ancestry(version_ids))
2715
this_versions = set(self.target._index.get_versions())
2716
needed_versions = self.source_ancestry - this_versions
2718
if not needed_versions:
2720
full_list = topo_sort(
2721
self.source.get_parent_map(self.source.versions()))
2723
version_list = [i for i in full_list if (not self.target.has_version(i)
2724
and i in needed_versions)]
2728
total = len(version_list)
2729
parent_map = self.source.get_parent_map(version_list)
2730
for version_id in version_list:
2731
pb.update("Converting to knit", count, total)
2732
parents = parent_map[version_id]
2733
# check that its will be a consistent copy:
2734
for parent in parents:
2735
# if source has the parent, we must already have it
2736
if not self.target.has_version(parent):
2737
raise AssertionError("%r does not have parent %r"
2738
% (self.target, parent))
2739
self.target.add_lines(
2740
version_id, parents, self.source.get_lines(version_id))
2747
InterVersionedFile.register_optimiser(WeaveToKnit)
2750
# Deprecated, use PatienceSequenceMatcher instead
2751
KnitSequenceMatcher = patiencediff.PatienceSequenceMatcher
2754
def annotate_knit(knit, revision_id):
2755
"""Annotate a knit with no cached annotations.
2757
This implementation is for knits with no cached annotations.
2758
It will work for knits with cached annotations, but this is not
2761
annotator = _KnitAnnotator(knit)
2762
return iter(annotator.annotate(revision_id))
2765
class _KnitAnnotator(object):
2766
"""Build up the annotations for a text."""
2768
def __init__(self, knit):
2771
# Content objects, differs from fulltexts because of how final newlines
2772
# are treated by knits. the content objects here will always have a
2774
self._fulltext_contents = {}
2776
# Annotated lines of specific revisions
2777
self._annotated_lines = {}
2779
# Track the raw data for nodes that we could not process yet.
2780
# This maps the revision_id of the base to a list of children that will
2781
# annotated from it.
2782
self._pending_children = {}
2784
# Nodes which cannot be extracted
2785
self._ghosts = set()
2787
# Track how many children this node has, so we know if we need to keep
2789
self._annotate_children = {}
2790
self._compression_children = {}
2792
self._all_build_details = {}
2793
# The children => parent revision_id graph
2794
self._revision_id_graph = {}
2796
self._heads_provider = None
2798
self._nodes_to_keep_annotations = set()
2799
self._generations_until_keep = 100
2801
def set_generations_until_keep(self, value):
2802
"""Set the number of generations before caching a node.
2804
Setting this to -1 will cache every merge node, setting this higher
2805
will cache fewer nodes.
2807
self._generations_until_keep = value
2809
def _add_fulltext_content(self, revision_id, content_obj):
2810
self._fulltext_contents[revision_id] = content_obj
2811
# TODO: jam 20080305 It might be good to check the sha1digest here
2812
return content_obj.text()
2814
def _check_parents(self, child, nodes_to_annotate):
2815
"""Check if all parents have been processed.
2817
:param child: A tuple of (rev_id, parents, raw_content)
2818
:param nodes_to_annotate: If child is ready, add it to
2819
nodes_to_annotate, otherwise put it back in self._pending_children
2821
for parent_id in child[1]:
2822
if (parent_id not in self._annotated_lines):
2823
# This parent is present, but another parent is missing
2824
self._pending_children.setdefault(parent_id,
2828
# This one is ready to be processed
2829
nodes_to_annotate.append(child)
2831
def _add_annotation(self, revision_id, fulltext, parent_ids,
2832
left_matching_blocks=None):
2833
"""Add an annotation entry.
2835
All parents should already have been annotated.
2836
:return: A list of children that now have their parents satisfied.
2838
a = self._annotated_lines
2839
annotated_parent_lines = [a[p] for p in parent_ids]
2840
annotated_lines = list(annotate.reannotate(annotated_parent_lines,
2841
fulltext, revision_id, left_matching_blocks,
2842
heads_provider=self._get_heads_provider()))
2843
self._annotated_lines[revision_id] = annotated_lines
2844
for p in parent_ids:
2845
ann_children = self._annotate_children[p]
2846
ann_children.remove(revision_id)
2847
if (not ann_children
2848
and p not in self._nodes_to_keep_annotations):
2849
del self._annotated_lines[p]
2850
del self._all_build_details[p]
2851
if p in self._fulltext_contents:
2852
del self._fulltext_contents[p]
2853
# Now that we've added this one, see if there are any pending
2854
# deltas to be done, certainly this parent is finished
2855
nodes_to_annotate = []
2856
for child in self._pending_children.pop(revision_id, []):
2857
self._check_parents(child, nodes_to_annotate)
2858
return nodes_to_annotate
2860
def _get_build_graph(self, revision_id):
2861
"""Get the graphs for building texts and annotations.
2863
The data you need for creating a full text may be different than the
2864
data you need to annotate that text. (At a minimum, you need both
2865
parents to create an annotation, but only need 1 parent to generate the
2868
:return: A list of (revision_id, index_memo) records, suitable for
2869
passing to read_records_iter to start reading in the raw data fro/
2872
if revision_id in self._annotated_lines:
2875
pending = set([revision_id])
2880
# get all pending nodes
2882
this_iteration = pending
2883
build_details = self._knit._index.get_build_details(this_iteration)
2884
self._all_build_details.update(build_details)
2885
# new_nodes = self._knit._index._get_entries(this_iteration)
2887
for rev_id, details in build_details.iteritems():
2888
(index_memo, compression_parent, parents,
2889
record_details) = details
2890
self._revision_id_graph[rev_id] = parents
2891
records.append((rev_id, index_memo))
2892
# Do we actually need to check _annotated_lines?
2893
pending.update(p for p in parents
2894
if p not in self._all_build_details)
2895
if compression_parent:
2896
self._compression_children.setdefault(compression_parent,
2899
for parent in parents:
2900
self._annotate_children.setdefault(parent,
2902
num_gens = generation - kept_generation
2903
if ((num_gens >= self._generations_until_keep)
2904
and len(parents) > 1):
2905
kept_generation = generation
2906
self._nodes_to_keep_annotations.add(rev_id)
2908
missing_versions = this_iteration.difference(build_details.keys())
2909
self._ghosts.update(missing_versions)
2910
for missing_version in missing_versions:
2911
# add a key, no parents
2912
self._revision_id_graph[missing_version] = ()
2913
pending.discard(missing_version) # don't look for it
2914
if self._ghosts.intersection(self._compression_children):
2916
"We cannot have nodes which have a ghost compression parent:\n"
2918
"compression children: %r"
2919
% (self._ghosts, self._compression_children))
2920
# Cleanout anything that depends on a ghost so that we don't wait for
2921
# the ghost to show up
2922
for node in self._ghosts:
2923
if node in self._annotate_children:
2924
# We won't be building this node
2925
del self._annotate_children[node]
2926
# Generally we will want to read the records in reverse order, because
2927
# we find the parent nodes after the children
2931
def _annotate_records(self, records):
2932
"""Build the annotations for the listed records."""
2933
# We iterate in the order read, rather than a strict order requested
2934
# However, process what we can, and put off to the side things that
2935
# still need parents, cleaning them up when those parents are
2937
for (rev_id, record,
2938
digest) in self._knit._data.read_records_iter(records):
2939
if rev_id in self._annotated_lines:
2941
parent_ids = self._revision_id_graph[rev_id]
2942
parent_ids = [p for p in parent_ids if p not in self._ghosts]
2943
details = self._all_build_details[rev_id]
2944
(index_memo, compression_parent, parents,
2945
record_details) = details
2946
nodes_to_annotate = []
2947
# TODO: Remove the punning between compression parents, and
2948
# parent_ids, we should be able to do this without assuming
2950
if len(parent_ids) == 0:
2951
# There are no parents for this node, so just add it
2952
# TODO: This probably needs to be decoupled
2953
fulltext_content, delta = self._knit.factory.parse_record(
2954
rev_id, record, record_details, None)
2955
fulltext = self._add_fulltext_content(rev_id, fulltext_content)
2956
nodes_to_annotate.extend(self._add_annotation(rev_id, fulltext,
2957
parent_ids, left_matching_blocks=None))
2959
child = (rev_id, parent_ids, record)
2960
# Check if all the parents are present
2961
self._check_parents(child, nodes_to_annotate)
2962
while nodes_to_annotate:
2963
# Should we use a queue here instead of a stack?
2964
(rev_id, parent_ids, record) = nodes_to_annotate.pop()
2965
(index_memo, compression_parent, parents,
2966
record_details) = self._all_build_details[rev_id]
2967
if compression_parent is not None:
2968
comp_children = self._compression_children[compression_parent]
2969
if rev_id not in comp_children:
2970
raise AssertionError("%r not in compression children %r"
2971
% (rev_id, comp_children))
2972
# If there is only 1 child, it is safe to reuse this
2974
reuse_content = (len(comp_children) == 1
2975
and compression_parent not in
2976
self._nodes_to_keep_annotations)
2978
# Remove it from the cache since it will be changing
2979
parent_fulltext_content = self._fulltext_contents.pop(compression_parent)
2980
# Make sure to copy the fulltext since it might be
2982
parent_fulltext = list(parent_fulltext_content.text())
2984
parent_fulltext_content = self._fulltext_contents[compression_parent]
2985
parent_fulltext = parent_fulltext_content.text()
2986
comp_children.remove(rev_id)
2987
fulltext_content, delta = self._knit.factory.parse_record(
2988
rev_id, record, record_details,
2989
parent_fulltext_content,
2990
copy_base_content=(not reuse_content))
2991
fulltext = self._add_fulltext_content(rev_id,
2993
blocks = KnitContent.get_line_delta_blocks(delta,
2994
parent_fulltext, fulltext)
2996
fulltext_content = self._knit.factory.parse_fulltext(
2998
fulltext = self._add_fulltext_content(rev_id,
3001
nodes_to_annotate.extend(
3002
self._add_annotation(rev_id, fulltext, parent_ids,
3003
left_matching_blocks=blocks))
3005
def _get_heads_provider(self):
3006
"""Create a heads provider for resolving ancestry issues."""
3007
if self._heads_provider is not None:
3008
return self._heads_provider
3009
parent_provider = _mod_graph.DictParentsProvider(
3010
self._revision_id_graph)
3011
graph_obj = _mod_graph.Graph(parent_provider)
3012
head_cache = _mod_graph.FrozenHeadsCache(graph_obj)
3013
self._heads_provider = head_cache
3016
def annotate(self, revision_id):
3017
"""Return the annotated fulltext at the given revision.
3019
:param revision_id: The revision id for this file
3021
records = self._get_build_graph(revision_id)
3022
if revision_id in self._ghosts:
3023
raise errors.RevisionNotPresent(revision_id, self._knit)
3024
self._annotate_records(records)
3025
return self._annotated_lines[revision_id]
3029
from bzrlib._knit_load_data_c import _load_data_c as _load_data
3031
from bzrlib._knit_load_data_py import _load_data_py as _load_data
added
removed
bzrlib/knit.py
