3
# Copyright (C) 2005 Canonical Ltd
5
# This program is free software; you can redistribute it and/or modify
6
# it under the terms of the GNU General Public License as published by
7
# the Free Software Foundation; either version 2 of the License, or
8
# (at your option) any later version.
10
# This program is distributed in the hope that it will be useful,
11
# but WITHOUT ANY WARRANTY; without even the implied warranty of
12
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13
# GNU General Public License for more details.
15
# You should have received a copy of the GNU General Public License
16
# along with this program; if not, write to the Free Software
17
# Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19
# Author: Martin Pool <mbp@canonical.com>
22
"""Weave - storage of related text file versions"""
25
# XXX: If we do weaves this way, will a merge still behave the same
26
# way if it's done in a different order? That's a pretty desirable
29
# TODO: Nothing here so far assumes the lines are really \n newlines,
30
# rather than being split up in some other way. We could accommodate
31
# binaries, perhaps by naively splitting on \n or perhaps using
32
# something like a rolling checksum.
34
# TODO: End marker for each version so we can stop reading?
36
# TODO: Check that no insertion occurs inside a deletion that was
37
# active in the version of the insertion.
39
# TODO: In addition to the SHA-1 check, perhaps have some code that
40
# checks structural constraints of the weave: ie that insertions are
41
# properly nested, that there is no text outside of an insertion, that
42
# insertions or deletions are not repeated, etc.
44
# TODO: Parallel-extract that passes back each line along with a
45
# description of which revisions include it. Nice for checking all
46
# shas or calculating stats in parallel.
48
# TODO: Using a single _extract routine and then processing the output
49
# is probably inefficient. It's simple enough that we can afford to
50
# have slight specializations for different ways its used: annotate,
51
# basis for add, get, etc.
53
# TODO: Probably the API should work only in names to hide the integer
54
# indexes from the user.
56
# TODO: Is there any potential performance win by having an add()
57
# variant that is passed a pre-cooked version of the single basis
60
# TODO: Reweave can possibly be made faster by remembering diffs
61
# where the basis and destination are unchanged.
63
# FIXME: Sometimes we will be given a parents list for a revision
64
# that includes some redundant parents (i.e. already a parent of
65
# something in the list.) We should eliminate them. This can
66
# be done fairly efficiently because the sequence numbers constrain
67
# the possible relationships.
69
# FIXME: the conflict markers should be *7* characters
72
from cStringIO import StringIO
81
from bzrlib.trace import mutter
82
from bzrlib.errors import (WeaveError, WeaveFormatError, WeaveParentMismatch,
83
RevisionAlreadyPresent,
85
WeaveRevisionAlreadyPresent,
86
WeaveRevisionNotPresent,
88
import bzrlib.errors as errors
89
from bzrlib.osutils import sha_strings
90
import bzrlib.patiencediff
91
from bzrlib.tsort import topo_sort
92
from bzrlib.versionedfile import VersionedFile, InterVersionedFile
93
from bzrlib.weavefile import _read_weave_v5, write_weave_v5
96
class Weave(VersionedFile):
97
"""weave - versioned text file storage.
99
A Weave manages versions of line-based text files, keeping track
100
of the originating version for each line.
102
To clients the "lines" of the file are represented as a list of strings.
103
These strings will typically have terminal newline characters, but
104
this is not required. In particular files commonly do not have a newline
105
at the end of the file.
107
Texts can be identified in either of two ways:
109
* a nonnegative index number.
111
* a version-id string.
113
Typically the index number will be valid only inside this weave and
114
the version-id is used to reference it in the larger world.
116
The weave is represented as a list mixing edit instructions and
117
literal text. Each entry in _weave can be either a string (or
118
unicode), or a tuple. If a string, it means that the given line
119
should be output in the currently active revisions.
121
If a tuple, it gives a processing instruction saying in which
122
revisions the enclosed lines are active. The tuple has the form
123
(instruction, version).
125
The instruction can be '{' or '}' for an insertion block, and '['
126
and ']' for a deletion block respectively. The version is the
127
integer version index. There is no replace operator, only deletes
128
and inserts. For '}', the end of an insertion, there is no
129
version parameter because it always closes the most recently
134
* A later version can delete lines that were introduced by any
135
number of ancestor versions; this implies that deletion
136
instructions can span insertion blocks without regard to the
137
insertion block's nesting.
139
* Similarly, deletions need not be properly nested with regard to
140
each other, because they might have been generated by
141
independent revisions.
143
* Insertions are always made by inserting a new bracketed block
144
into a single point in the previous weave. This implies they
145
can nest but not overlap, and the nesting must always have later
146
insertions on the inside.
148
* It doesn't seem very useful to have an active insertion
149
inside an inactive insertion, but it might happen.
151
* Therefore, all instructions are always"considered"; that
152
is passed onto and off the stack. An outer inactive block
153
doesn't disable an inner block.
155
* Lines are enabled if the most recent enclosing insertion is
156
active and none of the enclosing deletions are active.
158
* There is no point having a deletion directly inside its own
159
insertion; you might as well just not write it. And there
160
should be no way to get an earlier version deleting a later
164
Text of the weave; list of control instruction tuples and strings.
167
List of parents, indexed by version number.
168
It is only necessary to store the minimal set of parents for
169
each version; the parent's parents are implied.
172
List of hex SHA-1 of each version.
175
List of symbolic names for each version. Each should be unique.
178
For each name, the version number.
181
Descriptive name of this weave; typically the filename if known.
185
__slots__ = ['_weave', '_parents', '_sha1s', '_names', '_name_map',
186
'_weave_name', '_matcher']
188
def __init__(self, weave_name=None, access_mode='w', matcher=None, get_scope=None):
189
super(Weave, self).__init__(access_mode)
195
self._weave_name = weave_name
197
self._matcher = bzrlib.patiencediff.PatienceSequenceMatcher
199
self._matcher = matcher
200
if get_scope is None:
201
get_scope = lambda:None
202
self.get_scope = get_scope
203
self.scope = get_scope()
204
self._access_mode = access_mode
207
return "Weave(%r)" % self._weave_name
209
def _check_write_ok(self):
210
"""Is the versioned file marked as 'finished' ? Raise if it is."""
211
if self.get_scope() != self.scope:
212
raise errors.OutSideTransaction()
213
if self._access_mode != 'w':
214
raise errors.ReadOnlyObjectDirtiedError(self)
217
"""Return a deep copy of self.
219
The copy can be modified without affecting the original weave."""
221
other._weave = self._weave[:]
222
other._parents = self._parents[:]
223
other._sha1s = self._sha1s[:]
224
other._names = self._names[:]
225
other._name_map = self._name_map.copy()
226
other._weave_name = self._weave_name
229
def __eq__(self, other):
230
if not isinstance(other, Weave):
232
return self._parents == other._parents \
233
and self._weave == other._weave \
234
and self._sha1s == other._sha1s
236
def __ne__(self, other):
237
return not self.__eq__(other)
239
def _idx_to_name(self, version):
240
return self._names[version]
242
def _lookup(self, name):
243
"""Convert symbolic version name to index."""
244
self.check_not_reserved_id(name)
246
return self._name_map[name]
248
raise RevisionNotPresent(name, self._weave_name)
251
"""See VersionedFile.versions."""
252
return self._names[:]
254
def has_version(self, version_id):
255
"""See VersionedFile.has_version."""
256
return (version_id in self._name_map)
258
__contains__ = has_version
260
def get_parent_map(self, version_ids):
261
"""See VersionedFile.get_parent_map."""
263
for version_id in version_ids:
265
result[version_id] = tuple(
266
map(self._idx_to_name, self._parents[self._lookup(version_id)]))
267
except RevisionNotPresent:
271
def get_parents_with_ghosts(self, version_id):
272
raise NotImplementedError(self.get_parents_with_ghosts)
274
def _check_repeated_add(self, name, parents, text, sha1):
275
"""Check that a duplicated add is OK.
277
If it is, return the (old) index; otherwise raise an exception.
279
idx = self._lookup(name)
280
if sorted(self._parents[idx]) != sorted(parents) \
281
or sha1 != self._sha1s[idx]:
282
raise RevisionAlreadyPresent(name, self._weave_name)
285
def _add_lines(self, version_id, parents, lines, parent_texts,
286
left_matching_blocks, nostore_sha, random_id, check_content):
287
"""See VersionedFile.add_lines."""
288
idx = self._add(version_id, lines, map(self._lookup, parents),
289
nostore_sha=nostore_sha)
290
return sha_strings(lines), sum(map(len, lines)), idx
292
def _add(self, version_id, lines, parents, sha1=None, nostore_sha=None):
293
"""Add a single text on top of the weave.
295
Returns the index number of the newly added version.
298
Symbolic name for this version.
299
(Typically the revision-id of the revision that added it.)
302
List or set of direct parent version numbers.
305
Sequence of lines to be added in the new version.
307
:param nostore_sha: See VersionedFile.add_lines.
309
assert isinstance(version_id, basestring)
310
self._check_lines_not_unicode(lines)
311
self._check_lines_are_lines(lines)
313
sha1 = sha_strings(lines)
314
if sha1 == nostore_sha:
315
raise errors.ExistingContent
316
if version_id in self._name_map:
317
return self._check_repeated_add(version_id, parents, lines, sha1)
319
self._check_versions(parents)
320
## self._check_lines(lines)
321
new_version = len(self._parents)
323
# if we abort after here the (in-memory) weave will be corrupt because only
324
# some fields are updated
325
# XXX: FIXME implement a succeed-or-fail of the rest of this routine.
326
# - Robert Collins 20060226
327
self._parents.append(parents[:])
328
self._sha1s.append(sha1)
329
self._names.append(version_id)
330
self._name_map[version_id] = new_version
334
# special case; adding with no parents revision; can do
335
# this more quickly by just appending unconditionally.
336
# even more specially, if we're adding an empty text we
337
# need do nothing at all.
339
self._weave.append(('{', new_version))
340
self._weave.extend(lines)
341
self._weave.append(('}', None))
344
if len(parents) == 1:
345
pv = list(parents)[0]
346
if sha1 == self._sha1s[pv]:
347
# special case: same as the single parent
351
ancestors = self._inclusions(parents)
355
# basis a list of (origin, lineno, line)
358
for origin, lineno, line in self._extract(ancestors):
359
basis_lineno.append(lineno)
360
basis_lines.append(line)
362
# another small special case: a merge, producing the same text
364
if lines == basis_lines:
367
# add a sentinel, because we can also match against the final line
368
basis_lineno.append(len(self._weave))
370
# XXX: which line of the weave should we really consider
371
# matches the end of the file? the current code says it's the
372
# last line of the weave?
374
#print 'basis_lines:', basis_lines
375
#print 'new_lines: ', lines
377
s = self._matcher(None, basis_lines, lines)
379
# offset gives the number of lines that have been inserted
380
# into the weave up to the current point; if the original edit instruction
381
# says to change line A then we actually change (A+offset)
384
for tag, i1, i2, j1, j2 in s.get_opcodes():
385
# i1,i2 are given in offsets within basis_lines; we need to map them
386
# back to offsets within the entire weave
387
#print 'raw match', tag, i1, i2, j1, j2
391
i1 = basis_lineno[i1]
392
i2 = basis_lineno[i2]
394
assert 0 <= j1 <= j2 <= len(lines)
396
#print tag, i1, i2, j1, j2
398
# the deletion and insertion are handled separately.
399
# first delete the region.
401
self._weave.insert(i1+offset, ('[', new_version))
402
self._weave.insert(i2+offset+1, (']', new_version))
406
# there may have been a deletion spanning up to
407
# i2; we want to insert after this region to make sure
408
# we don't destroy ourselves
410
self._weave[i:i] = ([('{', new_version)]
413
offset += 2 + (j2 - j1)
416
def _clone_text(self, new_version_id, old_version_id, parents):
417
"""See VersionedFile.clone_text."""
418
old_lines = self.get_text(old_version_id)
419
self.add_lines(new_version_id, parents, old_lines)
421
def _inclusions(self, versions):
422
"""Return set of all ancestors of given version(s)."""
423
if not len(versions):
426
for v in xrange(max(versions), 0, -1):
428
# include all its parents
429
i.update(self._parents[v])
431
## except IndexError:
432
## raise ValueError("version %d not present in weave" % v)
434
def get_ancestry(self, version_ids, topo_sorted=True):
435
"""See VersionedFile.get_ancestry."""
436
if isinstance(version_ids, basestring):
437
version_ids = [version_ids]
438
i = self._inclusions([self._lookup(v) for v in version_ids])
439
return [self._idx_to_name(v) for v in i]
441
def _check_lines(self, text):
442
if not isinstance(text, list):
443
raise ValueError("text should be a list, not %s" % type(text))
446
if not isinstance(l, basestring):
447
raise ValueError("text line should be a string or unicode, not %s"
452
def _check_versions(self, indexes):
453
"""Check everything in the sequence of indexes is valid"""
458
raise IndexError("invalid version number %r" % i)
460
def _compatible_parents(self, my_parents, other_parents):
461
"""During join check that other_parents are joinable with my_parents.
463
Joinable is defined as 'is a subset of' - supersets may require
464
regeneration of diffs, but subsets do not.
466
return len(other_parents.difference(my_parents)) == 0
468
def annotate_iter(self, version_id):
469
"""Yield list of (version-id, line) pairs for the specified version.
471
The index indicates when the line originated in the weave."""
472
incls = [self._lookup(version_id)]
473
for origin, lineno, text in self._extract(incls):
474
yield self._idx_to_name(origin), text
476
def iter_lines_added_or_present_in_versions(self, version_ids=None,
478
"""See VersionedFile.iter_lines_added_or_present_in_versions()."""
479
if version_ids is None:
480
version_ids = self.versions()
481
version_ids = set(version_ids)
482
for lineno, inserted, deletes, line in self._walk_internal(version_ids):
483
# if inserted not in version_ids then it was inserted before the
484
# versions we care about, but because weaves cannot represent ghosts
485
# properly, we do not filter down to that
486
# if inserted not in version_ids: continue
488
yield line + '\n', inserted
492
def _walk_internal(self, version_ids=None):
493
"""Helper method for weave actions."""
498
lineno = 0 # line of weave, 0-based
500
for l in self._weave:
501
if l.__class__ == tuple:
505
istack.append(self._names[v])
509
assert self._names[v] not in dset
510
dset.add(self._names[v])
512
dset.remove(self._names[v])
514
raise WeaveFormatError('unexpected instruction %r' % v)
516
assert l.__class__ in (str, unicode)
518
yield lineno, istack[-1], frozenset(dset), l
522
raise WeaveFormatError("unclosed insertion blocks "
523
"at end of weave: %s" % istack)
525
raise WeaveFormatError("unclosed deletion blocks at end of weave: %s"
528
def plan_merge(self, ver_a, ver_b):
529
"""Return pseudo-annotation indicating how the two versions merge.
531
This is computed between versions a and b and their common
534
Weave lines present in none of them are skipped entirely.
536
inc_a = set(self.get_ancestry([ver_a]))
537
inc_b = set(self.get_ancestry([ver_b]))
538
inc_c = inc_a & inc_b
540
for lineno, insert, deleteset, line in self._walk_internal([ver_a, ver_b]):
541
if deleteset & inc_c:
542
# killed in parent; can't be in either a or b
543
# not relevant to our work
544
yield 'killed-base', line
545
elif insert in inc_c:
546
# was inserted in base
547
killed_a = bool(deleteset & inc_a)
548
killed_b = bool(deleteset & inc_b)
549
if killed_a and killed_b:
550
yield 'killed-both', line
552
yield 'killed-a', line
554
yield 'killed-b', line
556
yield 'unchanged', line
557
elif insert in inc_a:
558
if deleteset & inc_a:
559
yield 'ghost-a', line
563
elif insert in inc_b:
564
if deleteset & inc_b:
565
yield 'ghost-b', line
569
# not in either revision
570
yield 'irrelevant', line
572
yield 'unchanged', '' # terminator
574
def _extract(self, versions):
575
"""Yield annotation of lines in included set.
577
Yields a sequence of tuples (origin, lineno, text), where
578
origin is the origin version, lineno the index in the weave,
579
and text the text of the line.
581
The set typically but not necessarily corresponds to a version.
584
if not isinstance(i, int):
587
included = self._inclusions(versions)
593
lineno = 0 # line of weave, 0-based
599
WFE = WeaveFormatError
602
# 449 0 4474.6820 2356.5590 bzrlib.weave:556(_extract)
603
# +285282 0 1676.8040 1676.8040 +<isinstance>
604
# 1.6 seconds in 'isinstance'.
605
# changing the first isinstance:
606
# 449 0 2814.2660 1577.1760 bzrlib.weave:556(_extract)
607
# +140414 0 762.8050 762.8050 +<isinstance>
608
# note that the inline time actually dropped (less function calls)
609
# and total processing time was halved.
610
# we're still spending ~1/4 of the method in isinstance though.
611
# so lets hard code the acceptable string classes we expect:
612
# 449 0 1202.9420 786.2930 bzrlib.weave:556(_extract)
613
# +71352 0 377.5560 377.5560 +<method 'append' of 'list'
615
# yay, down to ~1/4 the initial extract time, and our inline time
616
# has shrunk again, with isinstance no longer dominating.
617
# tweaking the stack inclusion test to use a set gives:
618
# 449 0 1122.8030 713.0080 bzrlib.weave:556(_extract)
619
# +71352 0 354.9980 354.9980 +<method 'append' of 'list'
621
# - a 5% win, or possibly just noise. However with large istacks that
622
# 'in' test could dominate, so I'm leaving this change in place -
623
# when its fast enough to consider profiling big datasets we can review.
628
for l in self._weave:
629
if l.__class__ == tuple:
637
iset.remove(istack.pop())
648
assert l.__class__ in (str, unicode)
650
isactive = (not dset) and istack and (istack[-1] in included)
652
result.append((istack[-1], lineno, l))
655
raise WeaveFormatError("unclosed insertion blocks "
656
"at end of weave: %s" % istack)
658
raise WeaveFormatError("unclosed deletion blocks at end of weave: %s"
662
def _maybe_lookup(self, name_or_index):
663
"""Convert possible symbolic name to index, or pass through indexes.
667
if isinstance(name_or_index, (int, long)):
670
return self._lookup(name_or_index)
672
def get_lines(self, version_id):
673
"""See VersionedFile.get_lines()."""
674
int_index = self._maybe_lookup(version_id)
675
result = [line for (origin, lineno, line) in self._extract([int_index])]
676
expected_sha1 = self._sha1s[int_index]
677
measured_sha1 = sha_strings(result)
678
if measured_sha1 != expected_sha1:
679
raise errors.WeaveInvalidChecksum(
680
'file %s, revision %s, expected: %s, measured %s'
681
% (self._weave_name, version_id,
682
expected_sha1, measured_sha1))
685
def get_sha1(self, version_id):
686
"""See VersionedFile.get_sha1()."""
687
return self._sha1s[self._lookup(version_id)]
689
def get_sha1s(self, version_ids):
690
"""See VersionedFile.get_sha1s()."""
691
return [self._sha1s[self._lookup(v)] for v in version_ids]
693
def num_versions(self):
694
"""How many versions are in this weave?"""
695
l = len(self._parents)
696
assert l == len(self._sha1s)
699
__len__ = num_versions
701
def check(self, progress_bar=None):
702
# TODO evaluate performance hit of using string sets in this routine.
703
# TODO: check no circular inclusions
704
# TODO: create a nested progress bar
705
for version in range(self.num_versions()):
706
inclusions = list(self._parents[version])
709
if inclusions[-1] >= version:
710
raise WeaveFormatError("invalid included version %d for index %d"
711
% (inclusions[-1], version))
713
# try extracting all versions; parallel extraction is used
714
nv = self.num_versions()
719
# For creating the ancestry, IntSet is much faster (3.7s vs 0.17s)
720
# The problem is that set membership is much more expensive
721
name = self._idx_to_name(i)
722
sha1s[name] = sha.new()
724
new_inc = set([name])
725
for p in self._parents[i]:
726
new_inc.update(inclusions[self._idx_to_name(p)])
728
assert set(new_inc) == set(self.get_ancestry(name)), \
729
'failed %s != %s' % (set(new_inc), set(self.get_ancestry(name)))
730
inclusions[name] = new_inc
732
nlines = len(self._weave)
734
update_text = 'checking weave'
736
short_name = os.path.basename(self._weave_name)
737
update_text = 'checking %s' % (short_name,)
738
update_text = update_text[:25]
740
for lineno, insert, deleteset, line in self._walk_internal():
742
progress_bar.update(update_text, lineno, nlines)
744
for name, name_inclusions in inclusions.items():
745
# The active inclusion must be an ancestor,
746
# and no ancestors must have deleted this line,
747
# because we don't support resurrection.
748
if (insert in name_inclusions) and not (deleteset & name_inclusions):
749
sha1s[name].update(line)
752
version = self._idx_to_name(i)
753
hd = sha1s[version].hexdigest()
754
expected = self._sha1s[i]
756
raise errors.WeaveInvalidChecksum(
757
"mismatched sha1 for version %s: "
758
"got %s, expected %s"
759
% (version, hd, expected))
761
# TODO: check insertions are properly nested, that there are
762
# no lines outside of insertion blocks, that deletions are
763
# properly paired, etc.
765
def _join(self, other, pb, msg, version_ids, ignore_missing):
766
"""Worker routine for join()."""
767
if not other.versions():
768
return # nothing to update, easy
771
# versions is never none, InterWeave checks this.
774
# two loops so that we do not change ourselves before verifying it
776
# work through in index order to make sure we get all dependencies
779
# get the selected versions only that are in other.versions.
780
version_ids = set(other.versions()).intersection(set(version_ids))
781
# pull in the referenced graph.
782
version_ids = other.get_ancestry(version_ids)
783
pending_parents = other.get_parent_map(version_ids)
784
pending_graph = pending_parents.items()
785
if len(pending_graph) != len(version_ids):
786
raise RevisionNotPresent(
787
set(version_ids) - set(pending_parents.keys()), self)
788
for name in topo_sort(pending_graph):
789
other_idx = other._name_map[name]
790
# returns True if we have it, False if we need it.
791
if not self._check_version_consistent(other, other_idx, name):
792
names_to_join.append((other_idx, name))
800
for other_idx, name in names_to_join:
801
# TODO: If all the parents of the other version are already
802
# present then we can avoid some work by just taking the delta
803
# and adjusting the offsets.
804
new_parents = self._imported_parents(other, other_idx)
805
sha1 = other._sha1s[other_idx]
810
pb.update(msg, merged, len(names_to_join))
812
lines = other.get_lines(other_idx)
813
self._add(name, lines, new_parents, sha1)
815
mutter("merged = %d, processed = %d, file_id=%s; deltat=%d"%(
816
merged, processed, self._weave_name, time.time()-time0))
818
def _imported_parents(self, other, other_idx):
819
"""Return list of parents in self corresponding to indexes in other."""
821
for parent_idx in other._parents[other_idx]:
822
parent_name = other._names[parent_idx]
823
if parent_name not in self._name_map:
824
# should not be possible
825
raise WeaveError("missing parent {%s} of {%s} in %r"
826
% (parent_name, other._name_map[other_idx], self))
827
new_parents.append(self._name_map[parent_name])
830
def _check_version_consistent(self, other, other_idx, name):
831
"""Check if a version in consistent in this and other.
833
To be consistent it must have:
836
* the same direct parents (by name, not index, and disregarding
839
If present & correct return True;
840
if not present in self return False;
841
if inconsistent raise error."""
842
this_idx = self._name_map.get(name, -1)
844
if self._sha1s[this_idx] != other._sha1s[other_idx]:
845
raise errors.WeaveTextDiffers(name, self, other)
846
self_parents = self._parents[this_idx]
847
other_parents = other._parents[other_idx]
848
n1 = set([self._names[i] for i in self_parents])
849
n2 = set([other._names[i] for i in other_parents])
850
if not self._compatible_parents(n1, n2):
851
raise WeaveParentMismatch("inconsistent parents "
852
"for version {%s}: %s vs %s" % (name, n1, n2))
858
def _reweave(self, other, pb, msg):
859
"""Reweave self with other - internal helper for join().
861
:param other: The other weave to merge
862
:param pb: An optional progress bar, indicating how far done we are
863
:param msg: An optional message for the progress
865
new_weave = _reweave(self, other, pb=pb, msg=msg)
866
self._copy_weave_content(new_weave)
868
def _copy_weave_content(self, otherweave):
869
"""adsorb the content from otherweave."""
870
for attr in self.__slots__:
871
if attr != '_weave_name':
872
setattr(self, attr, copy(getattr(otherweave, attr)))
875
class WeaveFile(Weave):
876
"""A WeaveFile represents a Weave on disk and writes on change."""
878
WEAVE_SUFFIX = '.weave'
880
def __init__(self, name, transport, filemode=None, create=False, access_mode='w', get_scope=None):
881
"""Create a WeaveFile.
883
:param create: If not True, only open an existing knit.
885
super(WeaveFile, self).__init__(name, access_mode, get_scope=get_scope)
886
self._transport = transport
887
self._filemode = filemode
889
_read_weave_v5(self._transport.get(name + WeaveFile.WEAVE_SUFFIX), self)
890
except errors.NoSuchFile:
896
def _add_lines(self, version_id, parents, lines, parent_texts,
897
left_matching_blocks, nostore_sha, random_id, check_content):
898
"""Add a version and save the weave."""
899
self.check_not_reserved_id(version_id)
900
result = super(WeaveFile, self)._add_lines(version_id, parents, lines,
901
parent_texts, left_matching_blocks, nostore_sha, random_id,
906
def _clone_text(self, new_version_id, old_version_id, parents):
907
"""See VersionedFile.clone_text."""
908
super(WeaveFile, self)._clone_text(new_version_id, old_version_id, parents)
911
def copy_to(self, name, transport):
912
"""See VersionedFile.copy_to()."""
913
# as we are all in memory always, just serialise to the new place.
915
write_weave_v5(self, sio)
917
transport.put_file(name + WeaveFile.WEAVE_SUFFIX, sio, self._filemode)
920
"""Save the weave."""
921
self._check_write_ok()
923
write_weave_v5(self, sio)
925
self._transport.put_file(self._weave_name + WeaveFile.WEAVE_SUFFIX,
931
"""See VersionedFile.get_suffixes()."""
932
return [WeaveFile.WEAVE_SUFFIX]
934
def join(self, other, pb=None, msg=None, version_ids=None,
935
ignore_missing=False):
936
"""Join other into self and save."""
937
super(WeaveFile, self).join(other, pb, msg, version_ids, ignore_missing)
941
def _reweave(wa, wb, pb=None, msg=None):
942
"""Combine two weaves and return the result.
944
This works even if a revision R has different parents in
945
wa and wb. In the resulting weave all the parents are given.
947
This is done by just building up a new weave, maintaining ordering
948
of the versions in the two inputs. More efficient approaches
949
might be possible but it should only be necessary to do
950
this operation rarely, when a new previously ghost version is
953
:param pb: An optional progress bar, indicating how far done we are
954
:param msg: An optional message for the progress
958
queue_a = range(wa.num_versions())
959
queue_b = range(wb.num_versions())
960
# first determine combined parents of all versions
961
# map from version name -> all parent names
962
combined_parents = _reweave_parent_graphs(wa, wb)
963
mutter("combined parents: %r", combined_parents)
964
order = topo_sort(combined_parents.iteritems())
965
mutter("order to reweave: %r", order)
970
for idx, name in enumerate(order):
972
pb.update(msg, idx, len(order))
973
if name in wa._name_map:
974
lines = wa.get_lines(name)
975
if name in wb._name_map:
976
lines_b = wb.get_lines(name)
978
mutter('Weaves differ on content. rev_id {%s}', name)
979
mutter('weaves: %s, %s', wa._weave_name, wb._weave_name)
981
lines = list(difflib.unified_diff(lines, lines_b,
982
wa._weave_name, wb._weave_name))
983
mutter('lines:\n%s', ''.join(lines))
984
raise errors.WeaveTextDiffers(name, wa, wb)
986
lines = wb.get_lines(name)
987
wr._add(name, lines, [wr._lookup(i) for i in combined_parents[name]])
990
def _reweave_parent_graphs(wa, wb):
991
"""Return combined parent ancestry for two weaves.
993
Returned as a list of (version_name, set(parent_names))"""
995
for weave in [wa, wb]:
996
for idx, name in enumerate(weave._names):
997
p = combined.setdefault(name, set())
998
p.update(map(weave._idx_to_name, weave._parents[idx]))
1003
"""Show the weave's table-of-contents"""
1004
print '%6s %50s %10s %10s' % ('ver', 'name', 'sha1', 'parents')
1005
for i in (6, 50, 10, 10):
1008
for i in range(w.num_versions()):
1011
parent_str = ' '.join(map(str, w._parents[i]))
1012
print '%6d %-50.50s %10.10s %s' % (i, name, sha1, parent_str)
1016
def weave_stats(weave_file, pb):
1017
from bzrlib.weavefile import read_weave
1019
wf = file(weave_file, 'rb')
1021
# FIXME: doesn't work on pipes
1022
weave_size = wf.tell()
1026
for i in range(vers):
1027
pb.update('checking sizes', i, vers)
1028
for origin, lineno, line in w._extract([i]):
1033
print 'versions %9d' % vers
1034
print 'weave file %9d bytes' % weave_size
1035
print 'total contents %9d bytes' % total
1036
print 'compression ratio %9.2fx' % (float(total) / float(weave_size))
1039
print 'average size %9d bytes' % avg
1040
print 'relative size %9.2fx' % (float(weave_size) / float(avg))
1044
print """bzr weave tool
1046
Experimental tool for weave algorithm.
1049
weave init WEAVEFILE
1050
Create an empty weave file
1051
weave get WEAVEFILE VERSION
1052
Write out specified version.
1053
weave check WEAVEFILE
1054
Check consistency of all versions.
1056
Display table of contents.
1057
weave add WEAVEFILE NAME [BASE...] < NEWTEXT
1058
Add NEWTEXT, with specified parent versions.
1059
weave annotate WEAVEFILE VERSION
1060
Display origin of each line.
1061
weave merge WEAVEFILE VERSION1 VERSION2 > OUT
1062
Auto-merge two versions and display conflicts.
1063
weave diff WEAVEFILE VERSION1 VERSION2
1064
Show differences between two versions.
1068
% weave init foo.weave
1070
% weave add foo.weave ver0 < foo.txt
1073
(create updated version)
1075
% weave get foo.weave 0 | diff -u - foo.txt
1076
% weave add foo.weave ver1 0 < foo.txt
1079
% weave get foo.weave 0 > foo.txt (create forked version)
1081
% weave add foo.weave ver2 0 < foo.txt
1084
% weave merge foo.weave 1 2 > foo.txt (merge them)
1085
% vi foo.txt (resolve conflicts)
1086
% weave add foo.weave merged 1 2 < foo.txt (commit merged version)
1098
# in case we're run directly from the subdirectory
1099
sys.path.append('..')
1101
from bzrlib.weavefile import write_weave, read_weave
1102
from bzrlib.progress import ProgressBar
1117
return read_weave(file(argv[2], 'rb'))
1123
# at the moment, based on everything in the file
1125
parents = map(int, argv[4:])
1126
lines = sys.stdin.readlines()
1127
ver = w.add(name, parents, lines)
1128
write_weave(w, file(argv[2], 'wb'))
1129
print 'added version %r %d' % (name, ver)
1132
if os.path.exists(fn):
1133
raise IOError("file exists")
1135
write_weave(w, file(fn, 'wb'))
1136
elif cmd == 'get': # get one version
1138
sys.stdout.writelines(w.get_iter(int(argv[3])))
1143
v1, v2 = map(int, argv[3:5])
1146
diff_gen = bzrlib.patiencediff.unified_diff(lines1, lines2,
1147
'%s version %d' % (fn, v1),
1148
'%s version %d' % (fn, v2))
1149
sys.stdout.writelines(diff_gen)
1151
elif cmd == 'annotate':
1153
# newline is added to all lines regardless; too hard to get
1154
# reasonable formatting otherwise
1156
for origin, text in w.annotate(int(argv[3])):
1157
text = text.rstrip('\r\n')
1159
print ' | %s' % (text)
1161
print '%5d | %s' % (origin, text)
1167
elif cmd == 'stats':
1168
weave_stats(argv[2], ProgressBar())
1170
elif cmd == 'check':
1175
print '%d versions ok' % w.num_versions()
1177
elif cmd == 'inclusions':
1179
print ' '.join(map(str, w.inclusions([int(argv[3])])))
1181
elif cmd == 'parents':
1183
print ' '.join(map(str, w._parents[int(argv[3])]))
1185
elif cmd == 'plan-merge':
1186
# replaced by 'bzr weave-plan-merge'
1188
for state, line in w.plan_merge(int(argv[3]), int(argv[4])):
1190
print '%14s | %s' % (state, line),
1191
elif cmd == 'merge':
1192
# replaced by 'bzr weave-merge-text'
1194
p = w.plan_merge(int(argv[3]), int(argv[4]))
1195
sys.stdout.writelines(w.weave_merge(p))
1197
raise ValueError('unknown command %r' % cmd)
1200
if __name__ == '__main__':
1202
sys.exit(main(sys.argv))
1205
class InterWeave(InterVersionedFile):
1206
"""Optimised code paths for weave to weave operations."""
1208
_matching_file_from_factory = staticmethod(WeaveFile)
1209
_matching_file_to_factory = staticmethod(WeaveFile)
1212
def is_compatible(source, target):
1213
"""Be compatible with weaves."""
1215
return (isinstance(source, Weave) and
1216
isinstance(target, Weave))
1217
except AttributeError:
1220
def join(self, pb=None, msg=None, version_ids=None, ignore_missing=False):
1221
"""See InterVersionedFile.join."""
1222
version_ids = self._get_source_version_ids(version_ids, ignore_missing)
1223
if self.target.versions() == [] and version_ids is None:
1224
self.target._copy_weave_content(self.source)
1226
self.target._join(self.source, pb, msg, version_ids, ignore_missing)
1229
InterVersionedFile.register_optimiser(InterWeave)
added
removed
bzrlib/weave.py
