1
# Copyright (C) 2005, 2006 by Canonical Ltd
2
# Written by Martin Pool.
3
# Modified by Johan Rydberg <jrydberg@gnu.org>
4
# Modified by Robert Collins <robert.collins@canonical.com>
6
# This program is free software; you can redistribute it and/or modify
7
# it under the terms of the GNU General Public License as published by
8
# the Free Software Foundation; either version 2 of the License, or
9
# (at your option) any later version.
11
# This program is distributed in the hope that it will be useful,
12
# but WITHOUT ANY WARRANTY; without even the implied warranty of
13
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14
# GNU General Public License for more details.
16
# You should have received a copy of the GNU General Public License
17
# along with this program; if not, write to the Free Software
18
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20
"""Knit versionedfile implementation.
22
A knit is a versioned file implementation that supports efficient append only
26
lifeless: the data file is made up of "delta records". each delta record has a delta header
27
that contains; (1) a version id, (2) the size of the delta (in lines), and (3) the digest of
28
the -expanded data- (ie, the delta applied to the parent). the delta also ends with a
29
end-marker; simply "end VERSION"
31
delta can be line or full contents.a
32
... the 8's there are the index number of the annotation.
33
version robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad 7 c7d23b2a5bd6ca00e8e266cec0ec228158ee9f9e
37
8 e.set('executable', 'yes')
39
8 if elt.get('executable') == 'yes':
40
8 ie.executable = True
41
end robertc@robertcollins.net-20051003014215-ee2990904cc4c7ad
45
09:33 < jrydberg> lifeless: each index is made up of a tuple of; version id, options, position, size, parents
46
09:33 < jrydberg> lifeless: the parents are currently dictionary compressed
47
09:33 < jrydberg> lifeless: (meaning it currently does not support ghosts)
48
09:33 < lifeless> right
49
09:33 < jrydberg> lifeless: the position and size is the range in the data file
52
so the index sequence is the dictionary compressed sequence number used
53
in the deltas to provide line annotation
58
# 10:16 < lifeless> make partial index writes safe
59
# 10:16 < lifeless> implement 'knit.check()' like weave.check()
60
# 10:17 < lifeless> record known ghosts so we can detect when they are filled in rather than the current 'reweave
62
# move sha1 out of the content so that join is faster at verifying parents
63
# record content length ?
67
from cStringIO import StringIO
69
from difflib import SequenceMatcher
70
from gzip import GzipFile
71
from itertools import izip
76
import bzrlib.errors as errors
77
from bzrlib.errors import FileExists, NoSuchFile, KnitError, \
78
InvalidRevisionId, KnitCorrupt, KnitHeaderError, \
79
RevisionNotPresent, RevisionAlreadyPresent
80
from bzrlib.trace import mutter
81
from bzrlib.osutils import contains_whitespace, contains_linebreaks, \
83
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
84
from bzrlib.tsort import topo_sort
87
# TODO: Split out code specific to this format into an associated object.
89
# TODO: Can we put in some kind of value to check that the index and data
90
# files belong together?
92
# TODO: accomodate binaries, perhaps by storing a byte count
94
# TODO: function to check whole file
96
# TODO: atomically append data, then measure backwards from the cursor
97
# position after writing to work out where it was located. we may need to
98
# bypass python file buffering.
100
DATA_SUFFIX = '.knit'
101
INDEX_SUFFIX = '.kndx'
104
class KnitContent(object):
105
"""Content of a knit version to which deltas can be applied."""
107
def __init__(self, lines):
110
def annotate_iter(self):
111
"""Yield tuples of (origin, text) for each content line."""
112
for origin, text in self._lines:
116
"""Return a list of (origin, text) tuples."""
117
return list(self.annotate_iter())
119
def apply_delta(self, delta):
120
"""Apply delta to this content."""
122
for start, end, count, lines in delta:
123
self._lines[offset+start:offset+end] = lines
124
offset = offset + (start - end) + count
126
def line_delta_iter(self, new_lines):
127
"""Generate line-based delta from new_lines to this content."""
128
new_texts = [text for origin, text in new_lines._lines]
129
old_texts = [text for origin, text in self._lines]
130
s = difflib.SequenceMatcher(None, old_texts, new_texts)
131
for op in s.get_opcodes():
134
yield (op[1], op[2], op[4]-op[3], new_lines._lines[op[3]:op[4]])
136
def line_delta(self, new_lines):
137
return list(self.line_delta_iter(new_lines))
140
return [text for origin, text in self._lines]
143
class _KnitFactory(object):
144
"""Base factory for creating content objects."""
146
def make(self, lines, version):
147
num_lines = len(lines)
148
return KnitContent(zip([version] * num_lines, lines))
151
class KnitAnnotateFactory(_KnitFactory):
152
"""Factory for creating annotated Content objects."""
156
def parse_fulltext(self, content, version):
157
"""Convert fulltext to internal representation
159
fulltext content is of the format
160
revid(utf8) plaintext\n
161
internal representation is of the format:
166
origin, text = line.split(' ', 1)
167
lines.append((origin.decode('utf-8'), text))
168
return KnitContent(lines)
170
def parse_line_delta_iter(self, lines):
171
"""Convert a line based delta into internal representation.
173
line delta is in the form of:
174
intstart intend intcount
176
revid(utf8) newline\n
177
internal represnetation is
178
(start, end, count, [1..count tuples (revid, newline)])
181
header = lines.pop(0)
182
start, end, c = [int(n) for n in header.split(',')]
185
origin, text = lines.pop(0).split(' ', 1)
186
contents.append((origin.decode('utf-8'), text))
187
yield start, end, c, contents
189
def parse_line_delta(self, lines, version):
190
return list(self.parse_line_delta_iter(lines))
192
def lower_fulltext(self, content):
193
"""convert a fulltext content record into a serializable form.
195
see parse_fulltext which this inverts.
197
return ['%s %s' % (o.encode('utf-8'), t) for o, t in content._lines]
199
def lower_line_delta(self, delta):
200
"""convert a delta into a serializable form.
202
See parse_line_delta_iter which this inverts.
205
for start, end, c, lines in delta:
206
out.append('%d,%d,%d\n' % (start, end, c))
207
for origin, text in lines:
208
out.append('%s %s' % (origin.encode('utf-8'), text))
212
class KnitPlainFactory(_KnitFactory):
213
"""Factory for creating plain Content objects."""
217
def parse_fulltext(self, content, version):
218
"""This parses an unannotated fulltext.
220
Note that this is not a noop - the internal representation
221
has (versionid, line) - its just a constant versionid.
223
return self.make(content, version)
225
def parse_line_delta_iter(self, lines, version):
227
header = lines.pop(0)
228
start, end, c = [int(n) for n in header.split(',')]
229
yield start, end, c, zip([version] * c, lines[:c])
232
def parse_line_delta(self, lines, version):
233
return list(self.parse_line_delta_iter(lines, version))
235
def lower_fulltext(self, content):
236
return content.text()
238
def lower_line_delta(self, delta):
240
for start, end, c, lines in delta:
241
out.append('%d,%d,%d\n' % (start, end, c))
242
out.extend([text for origin, text in lines])
246
def make_empty_knit(transport, relpath):
247
"""Construct a empty knit at the specified location."""
248
k = KnitVersionedFile(transport, relpath, 'w', KnitPlainFactory)
252
class KnitVersionedFile(VersionedFile):
253
"""Weave-like structure with faster random access.
255
A knit stores a number of texts and a summary of the relationships
256
between them. Texts are identified by a string version-id. Texts
257
are normally stored and retrieved as a series of lines, but can
258
also be passed as single strings.
260
Lines are stored with the trailing newline (if any) included, to
261
avoid special cases for files with no final newline. Lines are
262
composed of 8-bit characters, not unicode. The combination of
263
these approaches should mean any 'binary' file can be safely
264
stored and retrieved.
267
def __init__(self, relpath, transport, file_mode=None, access_mode=None, factory=None,
268
basis_knit=None, delta=True, create=False):
269
"""Construct a knit at location specified by relpath.
271
:param create: If not True, only open an existing knit.
273
if access_mode is None:
275
super(KnitVersionedFile, self).__init__(access_mode)
276
assert access_mode in ('r', 'w'), "invalid mode specified %r" % access_mode
277
assert not basis_knit or isinstance(basis_knit, KnitVersionedFile), \
280
self.transport = transport
281
self.filename = relpath
282
self.basis_knit = basis_knit
283
self.factory = factory or KnitAnnotateFactory()
284
self.writable = (access_mode == 'w')
287
self._index = _KnitIndex(transport, relpath + INDEX_SUFFIX,
288
access_mode, create=create)
289
self._data = _KnitData(transport, relpath + DATA_SUFFIX,
290
access_mode, create=not len(self.versions()))
292
def clear_cache(self):
293
"""Clear the data cache only."""
294
self._data.clear_cache()
296
def copy_to(self, name, transport):
297
"""See VersionedFile.copy_to()."""
298
# copy the current index to a temp index to avoid racing with local
300
transport.put(name + INDEX_SUFFIX + '.tmp', self.transport.get(self._index._filename))
302
transport.put(name + DATA_SUFFIX, self._data._open_file())
303
# rename the copied index into place
304
transport.rename(name + INDEX_SUFFIX + '.tmp', name + INDEX_SUFFIX)
306
def create_empty(self, name, transport, mode=None):
307
return KnitVersionedFile(name, transport, factory=self.factory, delta=self.delta, create=True)
309
def _fix_parents(self, version, new_parents):
310
"""Fix the parents list for version.
312
This is done by appending a new version to the index
313
with identical data except for the parents list.
314
the parents list must be a superset of the current
317
current_values = self._index._cache[version]
318
assert set(current_values[4]).difference(set(new_parents)) == set()
319
self._index.add_version(version,
325
def get_graph_with_ghosts(self):
326
"""See VersionedFile.get_graph_with_ghosts()."""
327
graph_items = self._index.get_graph()
328
return dict(graph_items)
332
"""See VersionedFile.get_suffixes()."""
333
return [DATA_SUFFIX, INDEX_SUFFIX]
335
def has_ghost(self, version_id):
336
"""True if there is a ghost reference in the file to version_id."""
338
if self.has_version(version_id):
340
# optimisable if needed by memoising the _ghosts set.
341
items = self._index.get_graph()
342
for node, parents in items:
343
for parent in parents:
344
if parent not in self._index._cache:
345
if parent == version_id:
350
"""See VersionedFile.versions."""
351
return self._index.get_versions()
353
def has_version(self, version_id):
354
"""See VersionedFile.has_version."""
355
return self._index.has_version(version_id)
357
__contains__ = has_version
359
def _merge_annotations(self, content, parents):
360
"""Merge annotations for content. This is done by comparing
361
the annotations based on changed to the text."""
362
for parent_id in parents:
363
merge_content = self._get_content(parent_id)
364
seq = SequenceMatcher(None, merge_content.text(), content.text())
365
for i, j, n in seq.get_matching_blocks():
368
content._lines[j:j+n] = merge_content._lines[i:i+n]
370
def _get_components(self, version_id):
371
"""Return a list of (version_id, method, data) tuples that
372
makes up version specified by version_id of the knit.
374
The components should be applied in the order of the returned
377
The basis knit will be used to the largest extent possible
378
since it is assumed that accesses to it is faster.
380
# needed_revisions holds a list of (method, version_id) of
381
# versions that is needed to be fetched to construct the final
382
# version of the file.
384
# basis_revisions is a list of versions that needs to be
385
# fetched but exists in the basis knit.
387
basis = self.basis_knit
394
if basis and basis._index.has_version(cursor):
396
basis_versions.append(cursor)
397
method = picked_knit._index.get_method(cursor)
398
needed_versions.append((method, cursor))
399
if method == 'fulltext':
401
cursor = picked_knit.get_parents(cursor)[0]
406
for comp_id in basis_versions:
407
data_pos, data_size = basis._index.get_data_position(comp_id)
408
records.append((piece_id, data_pos, data_size))
409
components.update(basis._data.read_records(records))
412
for comp_id in [vid for method, vid in needed_versions
413
if vid not in basis_versions]:
414
data_pos, data_size = self._index.get_position(comp_id)
415
records.append((comp_id, data_pos, data_size))
416
components.update(self._data.read_records(records))
418
# get_data_records returns a mapping with the version id as
419
# index and the value as data. The order the components need
420
# to be applied is held by needed_versions (reversed).
422
for method, comp_id in reversed(needed_versions):
423
out.append((comp_id, method, components[comp_id]))
427
def _get_content(self, version_id):
428
"""Returns a content object that makes up the specified
430
if not self.has_version(version_id):
431
raise RevisionNotPresent(version_id, self.filename)
433
if self.basis_knit and version_id in self.basis_knit:
434
return self.basis_knit._get_content(version_id)
437
components = self._get_components(version_id)
438
for component_id, method, (data, digest) in components:
439
version_idx = self._index.lookup(component_id)
440
if method == 'fulltext':
441
assert content is None
442
content = self.factory.parse_fulltext(data, version_idx)
443
elif method == 'line-delta':
444
delta = self.factory.parse_line_delta(data, version_idx)
445
content.apply_delta(delta)
447
if 'no-eol' in self._index.get_options(version_id):
448
line = content._lines[-1][1].rstrip('\n')
449
content._lines[-1] = (content._lines[-1][0], line)
451
if sha_strings(content.text()) != digest:
452
raise KnitCorrupt(self.filename, 'sha-1 does not match')
456
def _check_versions_present(self, version_ids):
457
"""Check that all specified versions are present."""
458
version_ids = set(version_ids)
459
for r in list(version_ids):
460
if self._index.has_version(r):
461
version_ids.remove(r)
463
raise RevisionNotPresent(list(version_ids)[0], self.filename)
465
def _add_lines_with_ghosts(self, version_id, parents, lines):
466
"""See VersionedFile.add_lines_with_ghosts()."""
467
self._check_add(version_id, lines)
468
return self._add(version_id, lines[:], parents, self.delta)
470
def _add_lines(self, version_id, parents, lines):
471
"""See VersionedFile.add_lines."""
472
self._check_add(version_id, lines)
473
self._check_versions_present(parents)
474
return self._add(version_id, lines[:], parents, self.delta)
476
def _check_add(self, version_id, lines):
477
"""check that version_id and lines are safe to add."""
478
assert self.writable, "knit is not opened for write"
479
### FIXME escape. RBC 20060228
480
if contains_whitespace(version_id):
481
raise InvalidRevisionId(version_id)
482
if self.has_version(version_id):
483
raise RevisionAlreadyPresent(version_id, self.filename)
485
if False or __debug__:
487
assert '\n' not in l[:-1]
489
def _add(self, version_id, lines, parents, delta):
490
"""Add a set of lines on top of version specified by parents.
492
If delta is true, compress the text as a line-delta against
495
Any versions not present will be converted into ghosts.
497
ghostless_parents = []
499
for parent in parents:
500
if not self.has_version(parent):
501
ghosts.append(parent)
503
ghostless_parents.append(parent)
505
if delta and not len(ghostless_parents):
508
digest = sha_strings(lines)
511
if lines[-1][-1] != '\n':
512
options.append('no-eol')
513
lines[-1] = lines[-1] + '\n'
515
lines = self.factory.make(lines, version_id) #len(self._index))
516
if self.factory.annotated and len(ghostless_parents) > 0:
517
# Merge annotations from parent texts if so is needed.
518
self._merge_annotations(lines, ghostless_parents)
520
if len(ghostless_parents) and delta:
521
# To speed the extract of texts the delta chain is limited
522
# to a fixed number of deltas. This should minimize both
523
# I/O and the time spend applying deltas.
525
delta_parents = ghostless_parents
527
parent = delta_parents[0]
528
method = self._index.get_method(parent)
529
if method == 'fulltext':
531
delta_parents = self._index.get_parents(parent)
533
if method == 'line-delta':
537
options.append('line-delta')
538
content = self._get_content(ghostless_parents[0])
539
delta_hunks = content.line_delta(lines)
540
store_lines = self.factory.lower_line_delta(delta_hunks)
542
options.append('fulltext')
543
store_lines = self.factory.lower_fulltext(lines)
545
where, size = self._data.add_record(version_id, digest, store_lines)
546
self._index.add_version(version_id, options, where, size, parents)
548
def check(self, progress_bar=None):
549
"""See VersionedFile.check()."""
551
def _clone_text(self, new_version_id, old_version_id, parents):
552
"""See VersionedFile.clone_text()."""
553
# FIXME RBC 20060228 make fast by only inserting an index with null delta.
554
self.add_lines(new_version_id, parents, self.get_lines(old_version_id))
556
def get_lines(self, version_id):
557
"""See VersionedFile.get_lines()."""
558
return self._get_content(version_id).text()
560
def iter_lines_added_or_present_in_versions(self, version_ids=None):
561
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
562
if version_ids is None:
563
version_ids = self.versions()
564
# we dont care about inclusions, the caller cares.
565
# but we need to setup a list of records to visit.
566
# we need version_id, position, length
567
version_id_records = []
568
requested_versions = list(version_ids)
569
# filter for available versions
570
for version_id in requested_versions:
571
if not self.has_version(version_id):
572
raise RevisionNotPresent(version_id, self.filename)
573
# get a in-component-order queue:
575
for version_id in self.versions():
576
if version_id in requested_versions:
577
version_ids.append(version_id)
578
data_pos, length = self._index.get_position(version_id)
579
version_id_records.append((version_id, data_pos, length))
581
pb = bzrlib.ui.ui_factory.nested_progress_bar()
583
total = len(version_id_records)
585
pb.update('Walking content.', count, total)
586
for version_id, data, sha_value in \
587
self._data.read_records_iter(version_id_records):
588
pb.update('Walking content.', count, total)
589
method = self._index.get_method(version_id)
590
version_idx = self._index.lookup(version_id)
591
assert method in ('fulltext', 'line-delta')
592
if method == 'fulltext':
593
content = self.factory.parse_fulltext(data, version_idx)
594
for line in content.text():
597
delta = self.factory.parse_line_delta(data, version_idx)
598
for start, end, count, lines in delta:
599
for origin, line in lines:
602
pb.update('Walking content.', total, total)
605
pb.update('Walking content.', total, total)
609
def num_versions(self):
610
"""See VersionedFile.num_versions()."""
611
return self._index.num_versions()
613
__len__ = num_versions
615
def annotate_iter(self, version_id):
616
"""See VersionedFile.annotate_iter."""
617
content = self._get_content(version_id)
618
for origin, text in content.annotate_iter():
621
def get_parents(self, version_id):
622
"""See VersionedFile.get_parents."""
623
self._check_versions_present([version_id])
624
return list(self._index.get_parents(version_id))
626
def get_parents_with_ghosts(self, version_id):
627
"""See VersionedFile.get_parents."""
628
self._check_versions_present([version_id])
629
return list(self._index.get_parents_with_ghosts(version_id))
631
def get_ancestry(self, versions):
632
"""See VersionedFile.get_ancestry."""
633
if isinstance(versions, basestring):
634
versions = [versions]
637
self._check_versions_present(versions)
638
return self._index.get_ancestry(versions)
640
def get_ancestry_with_ghosts(self, versions):
641
"""See VersionedFile.get_ancestry_with_ghosts."""
642
if isinstance(versions, basestring):
643
versions = [versions]
646
self._check_versions_present(versions)
647
return self._index.get_ancestry_with_ghosts(versions)
649
#@deprecated_method(zero_eight)
650
def walk(self, version_ids):
651
"""See VersionedFile.walk."""
652
# We take the short path here, and extract all relevant texts
653
# and put them in a weave and let that do all the work. Far
654
# from optimal, but is much simpler.
655
# FIXME RB 20060228 this really is inefficient!
656
from bzrlib.weave import Weave
658
w = Weave(self.filename)
659
ancestry = self.get_ancestry(version_ids)
660
sorted_graph = topo_sort(self._index.get_graph())
661
version_list = [vid for vid in sorted_graph if vid in ancestry]
663
for version_id in version_list:
664
lines = self.get_lines(version_id)
665
w.add_lines(version_id, self.get_parents(version_id), lines)
667
for lineno, insert_id, dset, line in w.walk(version_ids):
668
yield lineno, insert_id, dset, line
671
class _KnitComponentFile(object):
672
"""One of the files used to implement a knit database"""
674
def __init__(self, transport, filename, mode):
675
self._transport = transport
676
self._filename = filename
679
def write_header(self):
680
old_len = self._transport.append(self._filename, StringIO(self.HEADER))
682
raise KnitCorrupt(self._filename, 'misaligned after writing header')
684
def check_header(self, fp):
685
line = fp.read(len(self.HEADER))
686
if line != self.HEADER:
687
raise KnitHeaderError(badline=line)
690
"""Commit is a nop."""
693
return '%s(%s)' % (self.__class__.__name__, self._filename)
696
class _KnitIndex(_KnitComponentFile):
697
"""Manages knit index file.
699
The index is already kept in memory and read on startup, to enable
700
fast lookups of revision information. The cursor of the index
701
file is always pointing to the end, making it easy to append
704
_cache is a cache for fast mapping from version id to a Index
707
_history is a cache for fast mapping from indexes to version ids.
709
The index data format is dictionary compressed when it comes to
710
parent references; a index entry may only have parents that with a
711
lover index number. As a result, the index is topological sorted.
713
Duplicate entries may be written to the index for a single version id
714
if this is done then the latter one completely replaces the former:
715
this allows updates to correct version and parent information.
716
Note that the two entries may share the delta, and that successive
717
annotations and references MUST point to the first entry.
720
HEADER = "# bzr knit index 7\n"
722
def _cache_version(self, version_id, options, pos, size, parents):
723
val = (version_id, options, pos, size, parents)
724
self._cache[version_id] = val
725
if not version_id in self._history:
726
self._history.append(version_id)
728
def _iter_index(self, fp):
734
#for l in lines.splitlines(False):
737
def __init__(self, transport, filename, mode, create=False):
738
_KnitComponentFile.__init__(self, transport, filename, mode)
740
# position in _history is the 'official' index for a revision
741
# but the values may have come from a newer entry.
742
# so - wc -l of a knit index is != the number of uniqe names
745
pb = bzrlib.ui.ui_factory.nested_progress_bar()
750
pb.update('read knit index', count, total)
751
fp = self._transport.get(self._filename)
752
self.check_header(fp)
753
for rec in self._iter_index(fp):
756
pb.update('read knit index', count, total)
757
parents = self._parse_parents(rec[4:])
758
self._cache_version(rec[0], rec[1].split(','), int(rec[2]), int(rec[3]),
760
except NoSuchFile, e:
761
if mode != 'w' or not create:
765
pb.update('read knit index', total, total)
768
def _parse_parents(self, compressed_parents):
769
"""convert a list of string parent values into version ids.
771
ints are looked up in the index.
772
.FOO values are ghosts and converted in to FOO.
775
for value in compressed_parents:
776
if value.startswith('.'):
777
result.append(value[1:])
779
assert isinstance(value, str)
781
result.append(self._history[int(value)])
783
import pdb;pdb.set_trace()
788
for version_id, index in self._cache.iteritems():
789
graph.append((version_id, index[4]))
792
def get_ancestry(self, versions):
793
"""See VersionedFile.get_ancestry."""
794
# get a graph of all the mentioned versions:
796
pending = set(versions)
798
version = pending.pop()
799
parents = self._cache[version][4]
802
parents = [parent for parent in parents if parent in self._cache]
803
for parent in parents:
804
# if not completed and not a ghost
805
if parent not in graph:
807
graph[version] = parents
808
return topo_sort(graph.items())
810
def get_ancestry_with_ghosts(self, versions):
811
"""See VersionedFile.get_ancestry_with_ghosts."""
812
# get a graph of all the mentioned versions:
814
pending = set(versions)
816
version = pending.pop()
818
parents = self._cache[version][4]
825
for parent in parents:
826
if parent not in graph:
828
graph[version] = parents
829
return topo_sort(graph.items())
831
def num_versions(self):
832
return len(self._history)
834
__len__ = num_versions
836
def get_versions(self):
839
def idx_to_name(self, idx):
840
return self._history[idx]
842
def lookup(self, version_id):
843
assert version_id in self._cache
844
return self._history.index(version_id)
846
def _version_list_to_index(self, versions):
848
for version in versions:
849
if version in self._cache:
850
result_list.append(str(self._history.index(version)))
852
result_list.append('.' + version.encode('utf-8'))
853
return ' '.join(result_list)
855
def add_version(self, version_id, options, pos, size, parents):
856
"""Add a version record to the index."""
857
self._cache_version(version_id, options, pos, size, parents)
859
content = "%s %s %s %s %s\n" % (version_id.encode('utf-8'),
863
self._version_list_to_index(parents))
864
assert isinstance(content, str), 'content must be utf-8 encoded'
865
self._transport.append(self._filename, StringIO(content))
867
def has_version(self, version_id):
868
"""True if the version is in the index."""
869
return self._cache.has_key(version_id)
871
def get_position(self, version_id):
872
"""Return data position and size of specified version."""
873
return (self._cache[version_id][2], \
874
self._cache[version_id][3])
876
def get_method(self, version_id):
877
"""Return compression method of specified version."""
878
options = self._cache[version_id][1]
879
if 'fulltext' in options:
882
assert 'line-delta' in options
885
def get_options(self, version_id):
886
return self._cache[version_id][1]
888
def get_parents(self, version_id):
889
"""Return parents of specified version ignoring ghosts."""
890
return [parent for parent in self._cache[version_id][4]
891
if parent in self._cache]
893
def get_parents_with_ghosts(self, version_id):
894
"""Return parents of specified version wth ghosts."""
895
return self._cache[version_id][4]
897
def check_versions_present(self, version_ids):
898
"""Check that all specified versions are present."""
899
version_ids = set(version_ids)
900
for version_id in list(version_ids):
901
if version_id in self._cache:
902
version_ids.remove(version_id)
904
raise RevisionNotPresent(list(version_ids)[0], self.filename)
907
class _KnitData(_KnitComponentFile):
908
"""Contents of the knit data file"""
910
HEADER = "# bzr knit data 7\n"
912
def __init__(self, transport, filename, mode, create=False):
913
_KnitComponentFile.__init__(self, transport, filename, mode)
915
self._checked = False
917
self._transport.put(self._filename, StringIO(''))
920
def clear_cache(self):
921
"""Clear the record cache."""
924
def _open_file(self):
925
if self._file is None:
927
self._file = self._transport.get(self._filename)
932
def _record_to_data(self, version_id, digest, lines):
933
"""Convert version_id, digest, lines into a raw data block.
935
:return: (len, a StringIO instance with the raw data ready to read.)
938
data_file = GzipFile(None, mode='wb', fileobj=sio)
939
print >>data_file, "version %s %d %s" % (version_id.encode('utf-8'), len(lines), digest)
940
data_file.writelines(lines)
941
print >>data_file, "end %s\n" % version_id.encode('utf-8')
947
def add_raw_record(self, raw_data):
948
"""Append a prepared record to the data file."""
949
assert isinstance(raw_data, str), 'data must be plain bytes'
950
start_pos = self._transport.append(self._filename, StringIO(raw_data))
951
return start_pos, len(raw_data)
953
def add_record(self, version_id, digest, lines):
954
"""Write new text record to disk. Returns the position in the
955
file where it was written."""
956
size, sio = self._record_to_data(version_id, digest, lines)
958
self._records[version_id] = (digest, lines)
960
start_pos = self._transport.append(self._filename, sio)
961
return start_pos, size
963
def _parse_record_header(self, version_id, raw_data):
964
"""Parse a record header for consistency.
966
:return: the header and the decompressor stream.
967
as (stream, header_record)
969
df = GzipFile(mode='rb', fileobj=StringIO(raw_data))
970
rec = df.readline().split()
972
raise KnitCorrupt(self._filename, 'unexpected number of elements in record header')
973
if rec[1].decode('utf-8')!= version_id:
974
raise KnitCorrupt(self._filename,
975
'unexpected version, wanted %r, got %r' % (
979
def _parse_record(self, version_id, data):
980
df, rec = self._parse_record_header(version_id, data)
982
record_contents = self._read_record_contents(df, lines)
984
if l.decode('utf-8') != 'end %s\n' % version_id:
985
raise KnitCorrupt(self._filename, 'unexpected version end line %r, wanted %r'
988
return record_contents, rec[3]
990
def _read_record_contents(self, df, record_lines):
991
"""Read and return n lines from datafile."""
993
for i in range(record_lines):
994
r.append(df.readline())
997
def read_records_iter_raw(self, records):
998
"""Read text records from data file and yield raw data.
1000
This unpacks enough of the text record to validate the id is
1001
as expected but thats all.
1003
It will actively recompress currently cached records on the
1004
basis that that is cheaper than I/O activity.
1007
for version_id, pos, size in records:
1008
if version_id not in self._records:
1009
needed_records.append((version_id, pos, size))
1011
# setup an iterator of the external records:
1012
# uses readv so nice and fast we hope.
1013
if len(needed_records):
1014
# grab the disk data needed.
1015
raw_records = self._transport.readv(self._filename,
1016
[(pos, size) for version_id, pos, size in needed_records])
1018
for version_id, pos, size in records:
1019
if version_id in self._records:
1020
# compress a new version
1021
size, sio = self._record_to_data(version_id,
1022
self._records[version_id][0],
1023
self._records[version_id][1])
1024
yield version_id, sio.getvalue()
1026
pos, data = raw_records.next()
1027
# validate the header
1028
df, rec = self._parse_record_header(version_id, data)
1030
yield version_id, data
1033
def read_records_iter(self, records):
1034
"""Read text records from data file and yield result.
1036
Each passed record is a tuple of (version_id, pos, len) and
1037
will be read in the given order. Yields (version_id,
1042
for version_id, pos, size in records:
1043
if version_id not in self._records:
1044
needed_records.append((version_id, pos, size))
1046
if len(needed_records):
1047
# We take it that the transport optimizes the fetching as good
1048
# as possible (ie, reads continous ranges.)
1049
response = self._transport.readv(self._filename,
1050
[(pos, size) for version_id, pos, size in needed_records])
1052
for (record_id, pos, size), (pos, data) in izip(iter(needed_records), response):
1053
content, digest = self._parse_record(record_id, data)
1054
self._records[record_id] = (digest, content)
1056
for version_id, pos, size in records:
1057
yield version_id, copy(self._records[version_id][1]), copy(self._records[version_id][0])
1059
def read_records(self, records):
1060
"""Read records into a dictionary."""
1062
for record_id, content, digest in self.read_records_iter(records):
1063
components[record_id] = (content, digest)
1067
class InterKnit(InterVersionedFile):
1068
"""Optimised code paths for knit to knit operations."""
1070
_matching_file_factory = KnitVersionedFile
1073
def is_compatible(source, target):
1074
"""Be compatible with knits. """
1076
return (isinstance(source, KnitVersionedFile) and
1077
isinstance(target, KnitVersionedFile))
1078
except AttributeError:
1081
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1082
"""See InterVersionedFile.join."""
1083
assert isinstance(self.source, KnitVersionedFile)
1084
assert isinstance(self.target, KnitVersionedFile)
1086
if version_ids is None:
1087
version_ids = self.source.versions()
1089
if not ignore_missing:
1090
self.source._check_versions_present(version_ids)
1092
version_ids = set(self.source.versions()).intersection(
1098
pb = bzrlib.ui.ui_factory.nested_progress_bar()
1100
version_ids = list(version_ids)
1101
if None in version_ids:
1102
version_ids.remove(None)
1104
self.source_ancestry = set(self.source.get_ancestry(version_ids))
1105
this_versions = set(self.target._index.get_versions())
1106
needed_versions = self.source_ancestry - this_versions
1107
cross_check_versions = self.source_ancestry.intersection(this_versions)
1108
mismatched_versions = set()
1109
for version in cross_check_versions:
1110
# scan to include needed parents.
1111
n1 = set(self.target.get_parents_with_ghosts(version))
1112
n2 = set(self.source.get_parents_with_ghosts(version))
1114
# FIXME TEST this check for cycles being introduced works
1115
# the logic is we have a cycle if in our graph we are an
1116
# ancestor of any of the n2 revisions.
1122
parent_ancestors = self.source.get_ancestry(parent)
1123
if version in parent_ancestors:
1124
raise errors.GraphCycleError([parent, version])
1125
# ensure this parent will be available later.
1126
new_parents = n2.difference(n1)
1127
needed_versions.update(new_parents.difference(this_versions))
1128
mismatched_versions.add(version)
1130
if not needed_versions and not cross_check_versions:
1132
full_list = topo_sort(self.source.get_graph())
1134
version_list = [i for i in full_list if (not self.target.has_version(i)
1135
and i in needed_versions)]
1139
copy_queue_records = []
1141
for version_id in version_list:
1142
options = self.source._index.get_options(version_id)
1143
parents = self.source._index.get_parents_with_ghosts(version_id)
1144
# check that its will be a consistent copy:
1145
for parent in parents:
1146
# if source has the parent, we must :
1147
# * already have it or
1148
# * have it scheduled already
1149
# otherwise we dont care
1150
assert (self.target.has_version(parent) or
1151
parent in copy_set or
1152
not self.source.has_version(parent))
1153
data_pos, data_size = self.source._index.get_position(version_id)
1154
copy_queue_records.append((version_id, data_pos, data_size))
1155
copy_queue.append((version_id, options, parents))
1156
copy_set.add(version_id)
1158
# data suck the join:
1160
total = len(version_list)
1161
# we want the raw gzip for bulk copying, but the record validated
1162
# just enough to be sure its the right one.
1163
# TODO: consider writev or write combining to reduce
1164
# death of a thousand cuts feeling.
1165
for (version_id, raw_data), \
1166
(version_id2, options, parents) in \
1167
izip(self.source._data.read_records_iter_raw(copy_queue_records),
1169
assert version_id == version_id2, 'logic error, inconsistent results'
1171
pb.update("Joining knit", count, total)
1172
pos, size = self.target._data.add_raw_record(raw_data)
1173
self.target._index.add_version(version_id, options, pos, size, parents)
1175
for version in mismatched_versions:
1176
# FIXME RBC 20060309 is this needed?
1177
n1 = set(self.target.get_parents_with_ghosts(version))
1178
n2 = set(self.source.get_parents_with_ghosts(version))
1179
# write a combined record to our history preserving the current
1180
# parents as first in the list
1181
new_parents = self.target.get_parents_with_ghosts(version) + list(n2.difference(n1))
1182
self.target.fix_parents(version, new_parents)
1189
InterVersionedFile.register_optimiser(InterKnit)
added
removed
bzrlib/knit.py
