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
54
# TODO: Perhaps have copy method for Weave instances?
56
# XXX: If we do weaves this way, will a merge still behave the same
57
# way if it's done in a different order? That's a pretty desirable
60
# TODO: Nothing here so far assumes the lines are really \n newlines,
61
# rather than being split up in some other way. We could accomodate
62
# binaries, perhaps by naively splitting on \n or perhaps using
63
# something like a rolling checksum.
65
# TODO: End marker for each version so we can stop reading?
67
# TODO: Check that no insertion occurs inside a deletion that was
68
# active in the version of the insertion.
70
# TODO: In addition to the SHA-1 check, perhaps have some code that
71
# checks structural constraints of the weave: ie that insertions are
72
# properly nested, that there is no text outside of an insertion, that
73
# insertions or deletions are not repeated, etc.
75
# TODO: Parallel-extract that passes back each line along with a
76
# description of which revisions include it. Nice for checking all
79
# TODO: Using a single _extract routine and then processing the output
80
# is probably inefficient. It's simple enough that we can afford to
81
# have slight specializations for different ways its used: annotate,
82
# basis for add, get, etc.
84
# TODO: Perhaps the API should work only in names to hide the integer
85
# indexes from the user?
93
class WeaveError(Exception):
94
"""Exception in processing weave"""
97
class WeaveFormatError(WeaveError):
98
"""Weave invariant violated"""
102
"""weave - versioned text file storage.
104
A Weave manages versions of line-based text files, keeping track
105
of the originating version for each line.
107
To clients the "lines" of the file are represented as a list of strings.
108
These strings will typically have terminal newline characters, but
109
this is not required. In particular files commonly do not have a newline
110
at the end of the file.
112
Texts can be identified in either of two ways:
114
* a nonnegative index number.
116
* a version-id string. (not implemented yet)
118
Typically the index number will be valid only inside this weave and
119
the version-id is used to reference it in the larger world.
121
The weave is represented as a list mixing edit instructions and
122
literal text. Each entry in _weave can be either a string (or
123
unicode), or a tuple. If a string, it means that the given line
124
should be output in the currently active revisions.
126
If a tuple, it gives a processing instruction saying in which
127
revisions the enclosed lines are active. The tuple has the form
128
(instruction, version).
130
The instruction can be '{' or '}' for an insertion block, and '['
131
and ']' for a deletion block respectively. The version is the
132
integer version index. There is no replace operator, only deletes
133
and inserts. For '}', the end of an insertion, there is no
134
version parameter because it always closes the most recently
139
* A later version can delete lines that were introduced by any
140
number of ancestor versions; this implies that deletion
141
instructions can span insertion blocks without regard to the
142
insertion block's nesting.
144
* Similarly, deletions need not be properly nested with regard to
145
each other, because they might have been generated by
146
independent revisions.
148
* Insertions are always made by inserting a new bracketed block
149
into a single point in the previous weave. This implies they
150
can nest but not overlap, and the nesting must always have later
151
insertions on the inside.
153
* It doesn't seem very useful to have an active insertion
154
inside an inactive insertion, but it might happen.
156
* Therefore, all instructions are always"considered"; that
157
is passed onto and off the stack. An outer inactive block
158
doesn't disable an inner block.
160
* Lines are enabled if the most recent enclosing insertion is
161
active and none of the enclosing deletions are active.
163
* There is no point having a deletion directly inside its own
164
insertion; you might as well just not write it. And there
165
should be no way to get an earlier version deleting a later
169
Text of the weave; list of control instruction tuples and strings.
172
List of parents, indexed by version number.
173
It is only necessary to store the minimal set of parents for
174
each version; the parent's parents are implied.
177
List of hex SHA-1 of each version.
180
List of symbolic names for each version. Each should be unique.
183
For each name, the version number.
186
__slots__ = ['_weave', '_parents', '_sha1s', '_names', '_name_map']
196
def __eq__(self, other):
197
if not isinstance(other, Weave):
199
return self._parents == other._parents \
200
and self._weave == other._weave \
201
and self._sha1s == other._sha1s
204
def __ne__(self, other):
205
return not self.__eq__(other)
208
def lookup(self, name):
210
return self._name_map[name]
212
raise WeaveError("name %s not present in weave" % name)
215
def add(self, name, parents, text):
216
"""Add a single text on top of the weave.
218
Returns the index number of the newly added version.
221
Symbolic name for this version.
222
(Typically the revision-id of the revision that added it.)
225
List or set of direct parent version numbers.
228
Sequence of lines to be added in the new version."""
230
assert isinstance(name, basestring)
231
if name in self._name_map:
232
raise WeaveError("name %r already present in weave" % name)
234
self._check_versions(parents)
235
## self._check_lines(text)
236
new_version = len(self._parents)
243
# if we abort after here the (in-memory) weave will be corrupt because only
244
# some fields are updated
245
self._parents.append(parents[:])
246
self._sha1s.append(sha1)
247
self._names.append(name)
248
self._name_map[name] = new_version
252
# special case; adding with no parents revision; can do
253
# this more quickly by just appending unconditionally.
254
# even more specially, if we're adding an empty text we
255
# need do nothing at all.
257
self._weave.append(('{', new_version))
258
self._weave.extend(text)
259
self._weave.append(('}', None))
263
if len(parents) == 1:
264
pv = list(parents)[0]
265
if sha1 == self._sha1s[pv]:
266
# special case: same as the single parent
270
ancestors = self.inclusions(parents)
274
# basis a list of (origin, lineno, line)
277
for origin, lineno, line in self._extract(ancestors):
278
basis_lineno.append(lineno)
279
basis_lines.append(line)
281
# another small special case: a merge, producing the same text
283
if text == basis_lines:
286
# add a sentinal, because we can also match against the final line
287
basis_lineno.append(len(self._weave))
289
# XXX: which line of the weave should we really consider
290
# matches the end of the file? the current code says it's the
291
# last line of the weave?
293
#print 'basis_lines:', basis_lines
294
#print 'new_lines: ', lines
296
from difflib import SequenceMatcher
297
s = SequenceMatcher(None, basis_lines, text)
299
# offset gives the number of lines that have been inserted
300
# into the weave up to the current point; if the original edit instruction
301
# says to change line A then we actually change (A+offset)
304
for tag, i1, i2, j1, j2 in s.get_opcodes():
305
# i1,i2 are given in offsets within basis_lines; we need to map them
306
# back to offsets within the entire weave
307
#print 'raw match', tag, i1, i2, j1, j2
311
i1 = basis_lineno[i1]
312
i2 = basis_lineno[i2]
314
assert 0 <= j1 <= j2 <= len(text)
316
#print tag, i1, i2, j1, j2
318
# the deletion and insertion are handled separately.
319
# first delete the region.
321
self._weave.insert(i1+offset, ('[', new_version))
322
self._weave.insert(i2+offset+1, (']', new_version))
326
# there may have been a deletion spanning up to
327
# i2; we want to insert after this region to make sure
328
# we don't destroy ourselves
330
self._weave[i:i] = ([('{', new_version)]
333
offset += 2 + (j2 - j1)
338
def inclusions(self, versions):
339
"""Return set of all ancestors of given version(s)."""
345
# include all its parents
346
i.update(self._parents[v])
350
raise ValueError("version %d not present in weave" % v)
353
def minimal_parents(self, version):
354
"""Find the minimal set of parents for the version."""
355
included = self._parents[version]
360
li.sort(reverse=True)
368
gotit.update(self.inclusions(pv))
370
assert mininc[0] >= 0
371
assert mininc[-1] < version
376
def _check_lines(self, text):
377
if not isinstance(text, list):
378
raise ValueError("text should be a list, not %s" % type(text))
381
if not isinstance(l, basestring):
382
raise ValueError("text line should be a string or unicode, not %s"
387
def _check_versions(self, indexes):
388
"""Check everything in the sequence of indexes is valid"""
393
raise IndexError("invalid version number %r" % i)
396
def annotate(self, index):
397
return list(self.annotate_iter(index))
400
def annotate_iter(self, version):
401
"""Yield list of (index-id, line) pairs for the specified version.
403
The index indicates when the line originated in the weave."""
404
for origin, lineno, text in self._extract([version]):
412
(lineno, insert, deletes, text)
413
for each literal line.
419
lineno = 0 # line of weave, 0-based
421
for l in self._weave:
422
if isinstance(l, tuple):
435
raise WeaveFormatError('unexpected instruction %r'
438
assert isinstance(l, basestring)
440
yield lineno, istack[-1], dset, l
445
def _extract(self, versions):
446
"""Yield annotation of lines in included set.
448
Yields a sequence of tuples (origin, lineno, text), where
449
origin is the origin version, lineno the index in the weave,
450
and text the text of the line.
452
The set typically but not necessarily corresponds to a version.
454
included = self.inclusions(versions)
459
lineno = 0 # line of weave, 0-based
465
WFE = WeaveFormatError
467
for l in self._weave:
468
if isinstance(l, tuple):
472
assert v not in istack
486
assert isinstance(l, basestring)
488
isactive = (not dset) and istack and (istack[-1] in included)
490
result.append((istack[-1], lineno, l))
494
raise WFE("unclosed insertion blocks at end of weave",
497
raise WFE("unclosed deletion blocks at end of weave",
504
def get_iter(self, version):
505
"""Yield lines for the specified version."""
506
for origin, lineno, line in self._extract([version]):
510
def get(self, index):
511
return list(self.get_iter(index))
514
def mash_iter(self, included):
515
"""Return composed version of multiple included versions."""
516
for origin, lineno, text in self._extract(included):
520
def dump(self, to_file):
521
from pprint import pprint
522
print >>to_file, "Weave._weave = ",
523
pprint(self._weave, to_file)
524
print >>to_file, "Weave._parents = ",
525
pprint(self._parents, to_file)
529
def numversions(self):
530
l = len(self._parents)
531
assert l == len(self._sha1s)
536
return self.numversions()
539
def check(self, progress_bar=None):
540
# check no circular inclusions
541
for version in range(self.numversions()):
542
inclusions = list(self._parents[version])
545
if inclusions[-1] >= version:
546
raise WeaveFormatError("invalid included version %d for index %d"
547
% (inclusions[-1], version))
549
# try extracting all versions; this is a bit slow and parallel
550
# extraction could be used
551
nv = self.numversions()
552
for version in range(nv):
554
progress_bar.update('checking text', version, nv)
556
for l in self.get_iter(version):
559
expected = self._sha1s[version]
561
raise WeaveError("mismatched sha1 for version %d; "
562
"got %s, expected %s"
563
% (version, hd, expected))
565
# TODO: check insertions are properly nested, that there are
566
# no lines outside of insertion blocks, that deletions are
567
# properly paired, etc.
571
def merge(self, merge_versions):
572
"""Automerge and mark conflicts between versions.
574
This returns a sequence, each entry describing alternatives
575
for a chunk of the file. Each of the alternatives is given as
578
If there is a chunk of the file where there's no diagreement,
579
only one alternative is given.
582
# approach: find the included versions common to all the
584
raise NotImplementedError()
588
def _delta(self, included, lines):
589
"""Return changes from basis to new revision.
591
The old text for comparison is the union of included revisions.
593
This is used in inserting a new text.
595
Delta is returned as a sequence of
596
(weave1, weave2, newlines).
598
This indicates that weave1:weave2 of the old weave should be
599
replaced by the sequence of lines in newlines. Note that
600
these line numbers are positions in the total weave and don't
601
correspond to the lines in any extracted version, or even the
602
extracted union of included versions.
604
If line1=line2, this is a pure insert; if newlines=[] this is a
605
pure delete. (Similar to difflib.)
610
def plan_merge(self, ver_a, ver_b):
611
"""Return pseudo-annotation indicating how the two versions merge.
613
This is computed between versions a and b and their common
616
Weave lines present in none of them are skipped entirely.
618
inc_a = self.inclusions([ver_a])
619
inc_b = self.inclusions([ver_b])
620
inc_c = inc_a & inc_b
622
for lineno, insert, deleteset, line in self._walk():
623
if deleteset & inc_c:
624
# killed in parent; can't be in either a or b
625
# not relevant to our work
626
yield 'killed-base', line
627
elif insert in inc_c:
628
# was inserted in base
629
killed_a = bool(deleteset & inc_a)
630
killed_b = bool(deleteset & inc_b)
631
if killed_a and killed_b:
632
yield 'killed-both', line
634
yield 'killed-a', line
636
yield 'killed-b', line
638
yield 'unchanged', line
639
elif insert in inc_a:
640
if deleteset & inc_a:
641
yield 'ghost-a', line
645
elif insert in inc_b:
646
if deleteset & inc_b:
647
yield 'ghost-b', line
651
# not in either revision
652
yield 'irrelevant', line
654
yield 'unchanged', '' # terminator
658
def weave_merge(self, plan):
663
for state, line in plan:
664
if state == 'unchanged' or state == 'killed-both':
665
# resync and flush queued conflicts changes if any
666
if not lines_a and not lines_b:
668
elif ch_a and not ch_b:
670
for l in lines_a: yield l
671
elif ch_b and not ch_a:
672
for l in lines_b: yield l
673
elif lines_a == lines_b:
674
for l in lines_a: yield l
677
for l in lines_a: yield l
679
for l in lines_b: yield l
686
if state == 'unchanged':
689
elif state == 'killed-a':
692
elif state == 'killed-b':
695
elif state == 'new-a':
698
elif state == 'new-b':
702
assert state in ('irrelevant', 'ghost-a', 'ghost-b', 'killed-base',
713
"""Show the weave's table-of-contents"""
714
print '%6s %50s %10s %10s' % ('ver', 'name', 'sha1', 'parents')
715
for i in (6, 50, 10, 10):
718
for i in range(w.numversions()):
721
parent_str = ' '.join(map(str, w._parents[i]))
722
print '%6d %-50.50s %10.10s %s' % (i, name, sha1, parent_str)
726
def weave_stats(weave_file):
727
from bzrlib.progress import ProgressBar
728
from bzrlib.weavefile import read_weave
732
wf = file(weave_file, 'rb')
734
# FIXME: doesn't work on pipes
735
weave_size = wf.tell()
739
for i in range(vers):
740
pb.update('checking sizes', i, vers)
741
for line in w.get_iter(i):
746
print 'versions %9d' % vers
747
print 'weave file %9d bytes' % weave_size
748
print 'total contents %9d bytes' % total
749
print 'compression ratio %9.2fx' % (float(total) / float(weave_size))
752
print 'average size %9d bytes' % avg
753
print 'relative size %9.2fx' % (float(weave_size) / float(avg))
757
print """bzr weave tool
759
Experimental tool for weave algorithm.
763
Create an empty weave file
764
weave get WEAVEFILE VERSION
765
Write out specified version.
766
weave check WEAVEFILE
767
Check consistency of all versions.
769
Display table of contents.
770
weave add WEAVEFILE NAME [BASE...] < NEWTEXT
771
Add NEWTEXT, with specified parent versions.
772
weave annotate WEAVEFILE VERSION
773
Display origin of each line.
774
weave mash WEAVEFILE VERSION...
775
Display composite of all selected versions.
776
weave merge WEAVEFILE VERSION1 VERSION2 > OUT
777
Auto-merge two versions and display conflicts.
781
% weave init foo.weave
783
% weave add foo.weave ver0 < foo.txt
786
(create updated version)
788
% weave get foo.weave 0 | diff -u - foo.txt
789
% weave add foo.weave ver1 0 < foo.txt
792
% weave get foo.weave 0 > foo.txt (create forked version)
794
% weave add foo.weave ver2 0 < foo.txt
797
% weave merge foo.weave 1 2 > foo.txt (merge them)
798
% vi foo.txt (resolve conflicts)
799
% weave add foo.weave merged 1 2 < foo.txt (commit merged version)
808
from weavefile import write_weave, read_weave
809
from bzrlib.progress import ProgressBar
824
return read_weave(file(argv[2], 'rb'))
830
# at the moment, based on everything in the file
832
parents = map(int, argv[4:])
833
lines = sys.stdin.readlines()
834
ver = w.add(name, parents, lines)
835
write_weave(w, file(argv[2], 'wb'))
836
print 'added version %r %d' % (name, ver)
839
if os.path.exists(fn):
840
raise IOError("file exists")
842
write_weave(w, file(fn, 'wb'))
843
elif cmd == 'get': # get one version
845
sys.stdout.writelines(w.get_iter(int(argv[3])))
847
elif cmd == 'mash': # get composite
849
sys.stdout.writelines(w.mash_iter(map(int, argv[3:])))
851
elif cmd == 'annotate':
853
# newline is added to all lines regardless; too hard to get
854
# reasonable formatting otherwise
856
for origin, text in w.annotate(int(argv[3])):
857
text = text.rstrip('\r\n')
859
print ' | %s' % (text)
861
print '%5d | %s' % (origin, text)
875
print '%d versions ok' % w.numversions()
877
elif cmd == 'inclusions':
879
print ' '.join(map(str, w.inclusions([int(argv[3])])))
881
elif cmd == 'parents':
883
print ' '.join(map(str, w._parents[int(argv[3])]))
885
elif cmd == 'plan-merge':
887
for state, line in w.plan_merge(int(argv[3]), int(argv[4])):
889
print '%14s | %s' % (state, line),
893
p = w.plan_merge(int(argv[3]), int(argv[4]))
894
sys.stdout.writelines(w.weave_merge(p))
896
elif cmd == 'mash-merge':
902
v1, v2 = map(int, argv[3:5])
904
basis = w.inclusions([v1]).intersection(w.inclusions([v2]))
906
base_lines = list(w.mash_iter(basis))
907
a_lines = list(w.get(v1))
908
b_lines = list(w.get(v2))
910
from bzrlib.merge3 import Merge3
911
m3 = Merge3(base_lines, a_lines, b_lines)
913
name_a = 'version %d' % v1
914
name_b = 'version %d' % v2
915
sys.stdout.writelines(m3.merge_lines(name_a=name_a, name_b=name_b))
917
raise ValueError('unknown command %r' % cmd)
921
def profile_main(argv):
922
import tempfile, hotshot, hotshot.stats
924
prof_f = tempfile.NamedTemporaryFile()
926
prof = hotshot.Profile(prof_f.name)
928
ret = prof.runcall(main, argv)
931
stats = hotshot.stats.load(prof_f.name)
933
stats.sort_stats('cumulative')
934
## XXX: Might like to write to stderr or the trace file instead but
935
## print_stats seems hardcoded to stdout
936
stats.print_stats(20)
941
if __name__ == '__main__':
943
if '--profile' in sys.argv:
945
args.remove('--profile')
946
sys.exit(profile_main(args))
948
sys.exit(main(sys.argv))