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"""
24
# before intset (r923) 2000 versions in 41.5s
25
# with intset (r926) 2000 versions in 93s !!!
26
# better to just use plain sets.
28
# making _extract build and return a list, rather than being a generator
31
# with python -O, r923 does 2000 versions in 36.87s
33
# with optimizations to avoid mutating lists - 35.75! I guess copying
34
# all the elements every time costs more than the small manipulations.
35
# a surprisingly small change.
37
# r931, which avoids using a generator for extract, does 36.98s
39
# with memoized inclusions, takes 41.49s; not very good
41
# with slots, takes 37.35s; without takes 39.16, a bit surprising
43
# with the delta calculation mixed in with the add method, rather than
44
# separated, takes 36.78s
46
# with delta folded in and mutation of the list, 36.13s
48
# with all this and simplification of add code, 33s
51
# TODO: Perhaps have copy method for Weave instances?
53
# XXX: If we do weaves this way, will a merge still behave the same
54
# way if it's done in a different order? That's a pretty desirable
57
# TODO: Nothing here so far assumes the lines are really \n newlines,
58
# rather than being split up in some other way. We could accomodate
59
# binaries, perhaps by naively splitting on \n or perhaps using
60
# something like a rolling checksum.
62
# TODO: Track version names as well as indexes.
64
# TODO: End marker for each version so we can stop reading?
66
# TODO: Check that no insertion occurs inside a deletion that was
67
# active in the version of the insertion.
69
# TODO: In addition to the SHA-1 check, perhaps have some code that
70
# checks structural constraints of the weave: ie that insertions are
71
# properly nested, that there is no text outside of an insertion, that
72
# insertions or deletions are not repeated, etc.
74
# TODO: Parallel-extract that passes back each line along with a
75
# description of which revisions include it. Nice for checking all
81
class WeaveError(Exception):
82
"""Exception in processing weave"""
85
class WeaveFormatError(WeaveError):
86
"""Weave invariant violated"""
90
"""weave - versioned text file storage.
92
A Weave manages versions of line-based text files, keeping track
93
of the originating version for each line.
95
To clients the "lines" of the file are represented as a list of strings.
96
These strings will typically have terminal newline characters, but
97
this is not required. In particular files commonly do not have a newline
98
at the end of the file.
100
Texts can be identified in either of two ways:
102
* a nonnegative index number.
104
* a version-id string.
106
Typically the index number will be valid only inside this weave and
107
the version-id is used to reference it in the larger world.
109
The weave is represented as a list mixing edit instructions and
110
literal text. Each entry in _weave can be either a string (or
111
unicode), or a tuple. If a string, it means that the given line
112
should be output in the currently active revisions.
114
If a tuple, it gives a processing instruction saying in which
115
revisions the enclosed lines are active. The tuple has the form
116
(instruction, version).
118
The instruction can be '{' or '}' for an insertion block, and '['
119
and ']' for a deletion block respectively. The version is the
120
integer version index. There is no replace operator, only deletes
125
* A later version can delete lines that were introduced by any
126
number of ancestor versions; this implies that deletion
127
instructions can span insertion blocks without regard to the
128
insertion block's nesting.
130
* Similarly, deletions need not be properly nested with regard to
131
each other, because they might have been generated by
132
independent revisions.
134
* Insertions are always made by inserting a new bracketed block
135
into a single point in the previous weave. This implies they
136
can nest but not overlap, and the nesting must always have later
137
insertions on the inside.
139
* It doesn't seem very useful to have an active insertion
140
inside an inactive insertion, but it might happen.
142
* Therefore, all instructions are always"considered"; that
143
is passed onto and off the stack. An outer inactive block
144
doesn't disable an inner block.
146
* Lines are enabled if the most recent enclosing insertion is
147
active and none of the enclosing deletions are active.
149
* There is no point having a deletion directly inside its own
150
insertion; you might as well just not write it. And there
151
should be no way to get an earlier version deleting a later
155
Text of the weave; list of control instruction tuples and strings.
158
List of parents, indexed by version number.
159
It is only necessary to store the minimal set of parents for
160
each version; the parent's parents are implied.
163
List of hex SHA-1 of each version, or None if not recorded.
166
__slots__ = ['_weave', '_parents', '_sha1s']
174
def __eq__(self, other):
175
if not isinstance(other, Weave):
177
return self._parents == other._parents \
178
and self._weave == other._weave
181
def __ne__(self, other):
182
return not self.__eq__(other)
185
def add(self, parents, text):
186
"""Add a single text on top of the weave.
188
Returns the index number of the newly added version.
191
List or set of direct parent version numbers.
194
Sequence of lines to be added in the new version."""
196
self._check_versions(parents)
197
## self._check_lines(text)
198
new_version = len(self._parents)
206
# if we abort after here the weave will be corrupt
207
self._parents.append(frozenset(parents))
208
self._sha1s.append(sha1)
212
# special case; adding with no parents revision; can do
213
# this more quickly by just appending unconditionally.
214
# even more specially, if we're adding an empty text we
215
# need do nothing at all.
217
self._weave.append(('{', new_version))
218
self._weave.extend(text)
219
self._weave.append(('}', new_version))
223
if len(parents) == 1:
224
pv = list(parents)[0]
225
if sha1 == self._sha1s[pv]:
226
# special case: same as the single parent
230
ancestors = self.inclusions(parents)
234
# basis a list of (origin, lineno, line)
237
for origin, lineno, line in self._extract(ancestors):
238
basis_lineno.append(lineno)
239
basis_lines.append(line)
241
# another small special case: a merge, producing the same text
243
if text == basis_lines:
246
# add a sentinal, because we can also match against the final line
247
basis_lineno.append(len(self._weave))
249
# XXX: which line of the weave should we really consider
250
# matches the end of the file? the current code says it's the
251
# last line of the weave?
253
#print 'basis_lines:', basis_lines
254
#print 'new_lines: ', lines
256
from difflib import SequenceMatcher
257
s = SequenceMatcher(None, basis_lines, text)
259
# offset gives the number of lines that have been inserted
260
# into the weave up to the current point; if the original edit instruction
261
# says to change line A then we actually change (A+offset)
264
for tag, i1, i2, j1, j2 in s.get_opcodes():
265
# i1,i2 are given in offsets within basis_lines; we need to map them
266
# back to offsets within the entire weave
267
#print 'raw match', tag, i1, i2, j1, j2
271
i1 = basis_lineno[i1]
272
i2 = basis_lineno[i2]
274
assert 0 <= j1 <= j2 <= len(text)
276
#print tag, i1, i2, j1, j2
278
# the deletion and insertion are handled separately.
279
# first delete the region.
281
self._weave.insert(i1+offset, ('[', new_version))
282
self._weave.insert(i2+offset+1, (']', new_version))
286
# there may have been a deletion spanning up to
287
# i2; we want to insert after this region to make sure
288
# we don't destroy ourselves
290
self._weave[i:i] = ([('{', new_version)]
292
+ [('}', new_version)])
293
offset += 2 + (j2 - j1)
298
def inclusions(self, versions):
299
"""Return set of all ancestors of given version(s)."""
305
# include all its parents
306
i.update(self._parents[v])
310
raise ValueError("version %d not present in weave" % v)
313
def minimal_parents(self, version):
314
"""Find the minimal set of parents for the version."""
315
included = self._parents[version]
320
li.sort(reverse=True)
328
gotit.update(self.inclusions(pv))
330
assert mininc[0] >= 0
331
assert mininc[-1] < version
336
def _check_lines(self, text):
337
if not isinstance(text, list):
338
raise ValueError("text should be a list, not %s" % type(text))
341
if not isinstance(l, basestring):
342
raise ValueError("text line should be a string or unicode, not %s"
347
def _check_versions(self, indexes):
348
"""Check everything in the sequence of indexes is valid"""
353
raise IndexError("invalid version number %r" % i)
356
def annotate(self, index):
357
return list(self.annotate_iter(index))
360
def annotate_iter(self, version):
361
"""Yield list of (index-id, line) pairs for the specified version.
363
The index indicates when the line originated in the weave."""
364
for origin, lineno, text in self._extract([version]):
372
(lineno, insert, deletes, text)
373
for each literal line.
379
lineno = 0 # line of weave, 0-based
381
for l in self._weave:
382
if isinstance(l, tuple):
395
raise WeaveFormatError('unexpected instruction %r'
398
assert isinstance(l, basestring)
400
yield lineno, istack[-1], dset, l
405
def _extract(self, versions):
406
"""Yield annotation of lines in included set.
408
Yields a sequence of tuples (origin, lineno, text), where
409
origin is the origin version, lineno the index in the weave,
410
and text the text of the line.
412
The set typically but not necessarily corresponds to a version.
414
included = self.inclusions(versions)
419
lineno = 0 # line of weave, 0-based
425
WFE = WeaveFormatError
427
for l in self._weave:
428
if isinstance(l, tuple):
432
assert v not in istack
447
assert isinstance(l, basestring)
449
isactive = (not dset) and istack and (istack[-1] in included)
451
result.append((istack[-1], lineno, l))
455
raise WFE("unclosed insertion blocks at end of weave",
458
raise WFE("unclosed deletion blocks at end of weave",
465
def get_iter(self, version):
466
"""Yield lines for the specified version."""
467
for origin, lineno, line in self._extract([version]):
471
def get(self, index):
472
return list(self.get_iter(index))
475
def mash_iter(self, included):
476
"""Return composed version of multiple included versions."""
477
for origin, lineno, text in self._extract(included):
481
def dump(self, to_file):
482
from pprint import pprint
483
print >>to_file, "Weave._weave = ",
484
pprint(self._weave, to_file)
485
print >>to_file, "Weave._parents = ",
486
pprint(self._parents, to_file)
490
def numversions(self):
491
l = len(self._parents)
492
assert l == len(self._sha1s)
497
return self.numversions()
500
def check(self, progress_bar=None):
501
# check no circular inclusions
502
for version in range(self.numversions()):
503
inclusions = list(self._parents[version])
506
if inclusions[-1] >= version:
507
raise WeaveFormatError("invalid included version %d for index %d"
508
% (inclusions[-1], version))
510
# try extracting all versions; this is a bit slow and parallel
511
# extraction could be used
513
nv = self.numversions()
514
for version in range(nv):
516
progress_bar.update('checking text', version, nv)
518
for l in self.get_iter(version):
521
expected = self._sha1s[version]
523
raise WeaveError("mismatched sha1 for version %d; "
524
"got %s, expected %s"
525
% (version, hd, expected))
527
# TODO: check insertions are properly nested, that there are
528
# no lines outside of insertion blocks, that deletions are
529
# properly paired, etc.
533
def merge(self, merge_versions):
534
"""Automerge and mark conflicts between versions.
536
This returns a sequence, each entry describing alternatives
537
for a chunk of the file. Each of the alternatives is given as
540
If there is a chunk of the file where there's no diagreement,
541
only one alternative is given.
544
# approach: find the included versions common to all the
546
raise NotImplementedError()
550
def _delta(self, included, lines):
551
"""Return changes from basis to new revision.
553
The old text for comparison is the union of included revisions.
555
This is used in inserting a new text.
557
Delta is returned as a sequence of
558
(weave1, weave2, newlines).
560
This indicates that weave1:weave2 of the old weave should be
561
replaced by the sequence of lines in newlines. Note that
562
these line numbers are positions in the total weave and don't
563
correspond to the lines in any extracted version, or even the
564
extracted union of included versions.
566
If line1=line2, this is a pure insert; if newlines=[] this is a
567
pure delete. (Similar to difflib.)
572
def plan_merge(self, ver_a, ver_b):
573
"""Return pseudo-annotation indicating how the two versions merge.
575
This is computed between versions a and b and their common
578
Weave lines present in none of them are skipped entirely.
580
inc_a = self.inclusions([ver_a])
581
inc_b = self.inclusions([ver_b])
582
inc_c = inc_a & inc_b
584
for lineno, insert, deleteset, line in self._walk():
585
if deleteset & inc_c:
586
# killed in parent; can't be in either a or b
587
# not relevant to our work
588
yield 'killed-base', line
589
elif insert in inc_c:
590
# was inserted in base
591
killed_a = bool(deleteset & inc_a)
592
killed_b = bool(deleteset & inc_b)
593
if killed_a and killed_b:
594
yield 'killed-both', line
596
yield 'killed-a', line
598
yield 'killed-b', line
600
yield 'unchanged', line
601
elif insert in inc_a:
602
if deleteset & inc_a:
603
yield 'ghost-a', line
607
elif insert in inc_b:
608
if deleteset & inc_b:
609
yield 'ghost-b', line
613
# not in either revision
614
yield 'irrelevant', line
616
yield 'unchanged', '' # terminator
620
def weave_merge(self, plan):
625
for state, line in plan:
626
if state == 'unchanged' or state == 'killed-both':
627
# resync and flush queued conflicts changes if any
628
if not lines_a and not lines_b:
630
elif ch_a and not ch_b:
632
for l in lines_a: yield l
633
elif ch_b and not ch_a:
634
for l in lines_b: yield l
635
elif lines_a == lines_b:
636
for l in lines_a: yield l
639
for l in lines_a: yield l
641
for l in lines_b: yield l
648
if state == 'unchanged':
651
elif state == 'killed-a':
654
elif state == 'killed-b':
657
elif state == 'new-a':
660
elif state == 'new-b':
664
assert state in ('irrelevant', 'ghost-a', 'ghost-b', 'killed-base',
675
"""Show some text information about the weave."""
676
print '%6s %40s %20s' % ('ver', 'sha1', 'parents')
677
for i in (6, 40, 20):
680
for i in range(w.numversions()):
682
print '%6d %40s %s' % (i, sha1, ' '.join(map(str, w._parents[i])))
686
def weave_stats(weave_file):
687
from bzrlib.progress import ProgressBar
688
from bzrlib.weavefile import read_weave
692
wf = file(weave_file, 'rb')
694
# FIXME: doesn't work on pipes
695
weave_size = wf.tell()
699
for i in range(vers):
700
pb.update('checking sizes', i, vers)
701
for line in w.get_iter(i):
706
print 'versions %9d' % vers
707
print 'weave file %9d bytes' % weave_size
708
print 'total contents %9d bytes' % total
709
print 'compression ratio %9.2fx' % (float(total) / float(weave_size))
712
print 'average size %9d bytes' % avg
713
print 'relative size %9.2fx' % (float(weave_size) / float(avg))
717
print """bzr weave tool
719
Experimental tool for weave algorithm.
723
Create an empty weave file
724
weave get WEAVEFILE VERSION
725
Write out specified version.
726
weave check WEAVEFILE
727
Check consistency of all versions.
729
Display table of contents.
730
weave add WEAVEFILE [BASE...] < NEWTEXT
731
Add NEWTEXT, with specified parent versions.
732
weave annotate WEAVEFILE VERSION
733
Display origin of each line.
734
weave mash WEAVEFILE VERSION...
735
Display composite of all selected versions.
736
weave merge WEAVEFILE VERSION1 VERSION2 > OUT
737
Auto-merge two versions and display conflicts.
741
% weave init foo.weave
743
% weave add foo.weave < foo.txt
746
(create updated version)
748
% weave get foo.weave 0 | diff -u - foo.txt
749
% weave add foo.weave 0 < foo.txt
752
% weave get foo.weave 0 > foo.txt (create forked version)
754
% weave add foo.weave 0 < foo.txt
757
% weave merge foo.weave 1 2 > foo.txt (merge them)
758
% vi foo.txt (resolve conflicts)
759
% weave add foo.weave 1 2 < foo.txt (commit merged version)
768
from weavefile import write_weave, read_weave
769
from bzrlib.progress import ProgressBar
777
return read_weave(file(argv[2], 'rb'))
783
# at the moment, based on everything in the file
784
parents = map(int, argv[3:])
785
lines = sys.stdin.readlines()
786
ver = w.add(parents, lines)
787
write_weave(w, file(argv[2], 'wb'))
788
print 'added version %d' % ver
791
if os.path.exists(fn):
792
raise IOError("file exists")
794
write_weave(w, file(fn, 'wb'))
795
elif cmd == 'get': # get one version
797
sys.stdout.writelines(w.get_iter(int(argv[3])))
799
elif cmd == 'mash': # get composite
801
sys.stdout.writelines(w.mash_iter(map(int, argv[3:])))
803
elif cmd == 'annotate':
805
# newline is added to all lines regardless; too hard to get
806
# reasonable formatting otherwise
808
for origin, text in w.annotate(int(argv[3])):
809
text = text.rstrip('\r\n')
811
print ' | %s' % (text)
813
print '%5d | %s' % (origin, text)
827
print '%d versions ok' % w.numversions()
829
elif cmd == 'inclusions':
831
print ' '.join(map(str, w.inclusions([int(argv[3])])))
833
elif cmd == 'parents':
835
print ' '.join(map(str, w._parents[int(argv[3])]))
837
elif cmd == 'plan-merge':
839
for state, line in w.plan_merge(int(argv[3]), int(argv[4])):
841
print '%14s | %s' % (state, line),
845
p = w.plan_merge(int(argv[3]), int(argv[4]))
846
sys.stdout.writelines(w.weave_merge(p))
848
elif cmd == 'mash-merge':
854
v1, v2 = map(int, argv[3:5])
856
basis = w.inclusions([v1]).intersection(w.inclusions([v2]))
858
base_lines = list(w.mash_iter(basis))
859
a_lines = list(w.get(v1))
860
b_lines = list(w.get(v2))
862
from bzrlib.merge3 import Merge3
863
m3 = Merge3(base_lines, a_lines, b_lines)
865
name_a = 'version %d' % v1
866
name_b = 'version %d' % v2
867
sys.stdout.writelines(m3.merge_lines(name_a=name_a, name_b=name_b))
869
raise ValueError('unknown command %r' % cmd)
872
if __name__ == '__main__':
874
sys.exit(main(sys.argv))