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
# TODO: Perhaps have copy method for Weave instances?
27
# XXX: If we do weaves this way, will a merge still behave the same
28
# way if it's done in a different order? That's a pretty desirable
31
# TODO: Nothing here so far assumes the lines are really \n newlines,
32
# rather than being split up in some other way. We could accomodate
33
# binaries, perhaps by naively splitting on \n or perhaps using
34
# something like a rolling checksum.
36
# TODO: End marker for each version so we can stop reading?
38
# TODO: Check that no insertion occurs inside a deletion that was
39
# active in the version of the insertion.
41
# TODO: In addition to the SHA-1 check, perhaps have some code that
42
# checks structural constraints of the weave: ie that insertions are
43
# properly nested, that there is no text outside of an insertion, that
44
# insertions or deletions are not repeated, etc.
46
# TODO: Parallel-extract that passes back each line along with a
47
# description of which revisions include it. Nice for checking all
48
# shas or calculating stats in parallel.
50
# TODO: Using a single _extract routine and then processing the output
51
# is probably inefficient. It's simple enough that we can afford to
52
# have slight specializations for different ways its used: annotate,
53
# basis for add, get, etc.
55
# TODO: Perhaps the API should work only in names to hide the integer
56
# indexes from the user?
58
# TODO: Is there any potential performance win by having an add()
59
# variant that is passed a pre-cooked version of the single basis
65
from difflib import SequenceMatcher
70
class WeaveError(Exception):
71
"""Exception in processing weave"""
74
class WeaveFormatError(WeaveError):
75
"""Weave invariant violated"""
79
"""weave - versioned text file storage.
81
A Weave manages versions of line-based text files, keeping track
82
of the originating version for each line.
84
To clients the "lines" of the file are represented as a list of strings.
85
These strings will typically have terminal newline characters, but
86
this is not required. In particular files commonly do not have a newline
87
at the end of the file.
89
Texts can be identified in either of two ways:
91
* a nonnegative index number.
93
* a version-id string.
95
Typically the index number will be valid only inside this weave and
96
the version-id is used to reference it in the larger world.
98
The weave is represented as a list mixing edit instructions and
99
literal text. Each entry in _weave can be either a string (or
100
unicode), or a tuple. If a string, it means that the given line
101
should be output in the currently active revisions.
103
If a tuple, it gives a processing instruction saying in which
104
revisions the enclosed lines are active. The tuple has the form
105
(instruction, version).
107
The instruction can be '{' or '}' for an insertion block, and '['
108
and ']' for a deletion block respectively. The version is the
109
integer version index. There is no replace operator, only deletes
110
and inserts. For '}', the end of an insertion, there is no
111
version parameter because it always closes the most recently
116
* A later version can delete lines that were introduced by any
117
number of ancestor versions; this implies that deletion
118
instructions can span insertion blocks without regard to the
119
insertion block's nesting.
121
* Similarly, deletions need not be properly nested with regard to
122
each other, because they might have been generated by
123
independent revisions.
125
* Insertions are always made by inserting a new bracketed block
126
into a single point in the previous weave. This implies they
127
can nest but not overlap, and the nesting must always have later
128
insertions on the inside.
130
* It doesn't seem very useful to have an active insertion
131
inside an inactive insertion, but it might happen.
133
* Therefore, all instructions are always"considered"; that
134
is passed onto and off the stack. An outer inactive block
135
doesn't disable an inner block.
137
* Lines are enabled if the most recent enclosing insertion is
138
active and none of the enclosing deletions are active.
140
* There is no point having a deletion directly inside its own
141
insertion; you might as well just not write it. And there
142
should be no way to get an earlier version deleting a later
146
Text of the weave; list of control instruction tuples and strings.
149
List of parents, indexed by version number.
150
It is only necessary to store the minimal set of parents for
151
each version; the parent's parents are implied.
154
List of hex SHA-1 of each version.
157
List of symbolic names for each version. Each should be unique.
160
For each name, the version number.
163
Descriptive name of this weave; typically the filename if known.
167
__slots__ = ['_weave', '_parents', '_sha1s', '_names', '_name_map',
170
def __init__(self, weave_name=None):
176
self._weave_name = weave_name
179
def __eq__(self, other):
180
if not isinstance(other, Weave):
182
return self._parents == other._parents \
183
and self._weave == other._weave \
184
and self._sha1s == other._sha1s
187
def __ne__(self, other):
188
return not self.__eq__(other)
191
def maybe_lookup(self, name_or_index):
192
"""Convert possible symbolic name to index, or pass through indexes."""
193
if isinstance(name_or_index, (int, long)):
196
return self.lookup(name_or_index)
199
def lookup(self, name):
200
"""Convert symbolic version name to index."""
202
return self._name_map[name]
204
raise WeaveError("name %r not present in weave %r" %
205
(name, self._weave_name))
208
def idx_to_name(self, version):
209
return self._names[version]
212
def _check_repeated_add(self, name, parents, text, sha1):
213
"""Check that a duplicated add is OK.
215
If it is, return the (old) index; otherwise raise an exception.
217
idx = self.lookup(name)
218
if sorted(self._parents[idx]) != sorted(parents):
219
raise WeaveError("name \"%s\" already present in weave "
220
"with different parents" % name)
221
if sha1 != self._sha1s[idx]:
222
raise WeaveError("name \"%s\" already present in weave "
223
"with different text" % name)
228
def add(self, name, parents, text, sha1=None):
229
"""Add a single text on top of the weave.
231
Returns the index number of the newly added version.
234
Symbolic name for this version.
235
(Typically the revision-id of the revision that added it.)
238
List or set of direct parent version numbers.
241
Sequence of lines to be added in the new version.
243
sha -- SHA-1 of the file, if known. This is trusted to be
246
from bzrlib.osutils import sha_strings
248
assert isinstance(name, basestring)
250
sha1 = sha_strings(text)
251
if name in self._name_map:
252
return self._check_repeated_add(name, parents, text, sha1)
254
parents = map(self.maybe_lookup, parents)
255
self._check_versions(parents)
256
## self._check_lines(text)
257
new_version = len(self._parents)
260
# if we abort after here the (in-memory) weave will be corrupt because only
261
# some fields are updated
262
self._parents.append(parents[:])
263
self._sha1s.append(sha1)
264
self._names.append(name)
265
self._name_map[name] = new_version
269
# special case; adding with no parents revision; can do
270
# this more quickly by just appending unconditionally.
271
# even more specially, if we're adding an empty text we
272
# need do nothing at all.
274
self._weave.append(('{', new_version))
275
self._weave.extend(text)
276
self._weave.append(('}', None))
280
if len(parents) == 1:
281
pv = list(parents)[0]
282
if sha1 == self._sha1s[pv]:
283
# special case: same as the single parent
287
ancestors = self.inclusions(parents)
291
# basis a list of (origin, lineno, line)
294
for origin, lineno, line in self._extract(ancestors):
295
basis_lineno.append(lineno)
296
basis_lines.append(line)
298
# another small special case: a merge, producing the same text
300
if text == basis_lines:
303
# add a sentinal, because we can also match against the final line
304
basis_lineno.append(len(self._weave))
306
# XXX: which line of the weave should we really consider
307
# matches the end of the file? the current code says it's the
308
# last line of the weave?
310
#print 'basis_lines:', basis_lines
311
#print 'new_lines: ', lines
313
s = SequenceMatcher(None, basis_lines, text)
315
# offset gives the number of lines that have been inserted
316
# into the weave up to the current point; if the original edit instruction
317
# says to change line A then we actually change (A+offset)
320
for tag, i1, i2, j1, j2 in s.get_opcodes():
321
# i1,i2 are given in offsets within basis_lines; we need to map them
322
# back to offsets within the entire weave
323
#print 'raw match', tag, i1, i2, j1, j2
327
i1 = basis_lineno[i1]
328
i2 = basis_lineno[i2]
330
assert 0 <= j1 <= j2 <= len(text)
332
#print tag, i1, i2, j1, j2
334
# the deletion and insertion are handled separately.
335
# first delete the region.
337
self._weave.insert(i1+offset, ('[', new_version))
338
self._weave.insert(i2+offset+1, (']', new_version))
342
# there may have been a deletion spanning up to
343
# i2; we want to insert after this region to make sure
344
# we don't destroy ourselves
346
self._weave[i:i] = ([('{', new_version)]
349
offset += 2 + (j2 - j1)
353
def add_identical(self, old_rev_id, new_rev_id, parents):
354
"""Add an identical text to old_rev_id as new_rev_id."""
355
old_lines = self.get(self.lookup(old_rev_id))
356
self.add(new_rev_id, parents, old_lines)
358
def inclusions(self, versions):
359
"""Return set of all ancestors of given version(s)."""
361
for v in xrange(max(versions), 0, -1):
363
# include all its parents
364
i.update(self._parents[v])
366
## except IndexError:
367
## raise ValueError("version %d not present in weave" % v)
370
def parents(self, version):
371
return self._parents[version]
374
def minimal_parents(self, version):
375
"""Find the minimal set of parents for the version."""
376
included = self._parents[version]
381
li.sort(reverse=True)
389
gotit.update(self.inclusions(pv))
391
assert mininc[0] >= 0
392
assert mininc[-1] < version
397
def _check_lines(self, text):
398
if not isinstance(text, list):
399
raise ValueError("text should be a list, not %s" % type(text))
402
if not isinstance(l, basestring):
403
raise ValueError("text line should be a string or unicode, not %s"
408
def _check_versions(self, indexes):
409
"""Check everything in the sequence of indexes is valid"""
414
raise IndexError("invalid version number %r" % i)
417
def annotate(self, name_or_index):
418
return list(self.annotate_iter(name_or_index))
421
def annotate_iter(self, name_or_index):
422
"""Yield list of (index-id, line) pairs for the specified version.
424
The index indicates when the line originated in the weave."""
425
incls = [self.maybe_lookup(name_or_index)]
426
for origin, lineno, text in self._extract(incls):
434
(lineno, insert, deletes, text)
435
for each literal line.
441
lineno = 0 # line of weave, 0-based
443
for l in self._weave:
444
if isinstance(l, tuple):
457
raise WeaveFormatError('unexpected instruction %r'
460
assert isinstance(l, basestring)
462
yield lineno, istack[-1], dset, l
467
def _extract(self, versions):
468
"""Yield annotation of lines in included set.
470
Yields a sequence of tuples (origin, lineno, text), where
471
origin is the origin version, lineno the index in the weave,
472
and text the text of the line.
474
The set typically but not necessarily corresponds to a version.
477
if not isinstance(i, int):
480
included = self.inclusions(versions)
485
lineno = 0 # line of weave, 0-based
491
WFE = WeaveFormatError
493
for l in self._weave:
494
if isinstance(l, tuple):
498
assert v not in istack
512
assert isinstance(l, basestring)
514
isactive = (not dset) and istack and (istack[-1] in included)
516
result.append((istack[-1], lineno, l))
520
raise WFE("unclosed insertion blocks at end of weave",
523
raise WFE("unclosed deletion blocks at end of weave",
530
def get_iter(self, name_or_index):
531
"""Yield lines for the specified version."""
532
incls = [self.maybe_lookup(name_or_index)]
533
for origin, lineno, line in self._extract(incls):
537
def get_text(self, name_or_index):
538
return ''.join(self.get_iter(name_or_index))
539
assert isinstance(version, int)
542
def get_lines(self, name_or_index):
543
return list(self.get_iter(name_or_index))
549
def mash_iter(self, included):
550
"""Return composed version of multiple included versions."""
551
included = map(self.maybe_lookup, included)
552
for origin, lineno, text in self._extract(included):
556
def dump(self, to_file):
557
from pprint import pprint
558
print >>to_file, "Weave._weave = ",
559
pprint(self._weave, to_file)
560
print >>to_file, "Weave._parents = ",
561
pprint(self._parents, to_file)
565
def numversions(self):
566
l = len(self._parents)
567
assert l == len(self._sha1s)
572
return self.numversions()
575
def check(self, progress_bar=None):
576
# check no circular inclusions
577
for version in range(self.numversions()):
578
inclusions = list(self._parents[version])
581
if inclusions[-1] >= version:
582
raise WeaveFormatError("invalid included version %d for index %d"
583
% (inclusions[-1], version))
585
# try extracting all versions; this is a bit slow and parallel
586
# extraction could be used
587
nv = self.numversions()
588
for version in range(nv):
590
progress_bar.update('checking text', version, nv)
592
for l in self.get_iter(version):
595
expected = self._sha1s[version]
597
raise WeaveError("mismatched sha1 for version %d; "
598
"got %s, expected %s"
599
% (version, hd, expected))
601
# TODO: check insertions are properly nested, that there are
602
# no lines outside of insertion blocks, that deletions are
603
# properly paired, etc.
607
def merge(self, merge_versions):
608
"""Automerge and mark conflicts between versions.
610
This returns a sequence, each entry describing alternatives
611
for a chunk of the file. Each of the alternatives is given as
614
If there is a chunk of the file where there's no diagreement,
615
only one alternative is given.
617
# approach: find the included versions common to all the
619
raise NotImplementedError()
623
def _delta(self, included, lines):
624
"""Return changes from basis to new revision.
626
The old text for comparison is the union of included revisions.
628
This is used in inserting a new text.
630
Delta is returned as a sequence of
631
(weave1, weave2, newlines).
633
This indicates that weave1:weave2 of the old weave should be
634
replaced by the sequence of lines in newlines. Note that
635
these line numbers are positions in the total weave and don't
636
correspond to the lines in any extracted version, or even the
637
extracted union of included versions.
639
If line1=line2, this is a pure insert; if newlines=[] this is a
640
pure delete. (Similar to difflib.)
642
raise NotImplementedError()
645
def plan_merge(self, ver_a, ver_b):
646
"""Return pseudo-annotation indicating how the two versions merge.
648
This is computed between versions a and b and their common
651
Weave lines present in none of them are skipped entirely.
653
inc_a = self.inclusions([ver_a])
654
inc_b = self.inclusions([ver_b])
655
inc_c = inc_a & inc_b
657
for lineno, insert, deleteset, line in self._walk():
658
if deleteset & inc_c:
659
# killed in parent; can't be in either a or b
660
# not relevant to our work
661
yield 'killed-base', line
662
elif insert in inc_c:
663
# was inserted in base
664
killed_a = bool(deleteset & inc_a)
665
killed_b = bool(deleteset & inc_b)
666
if killed_a and killed_b:
667
yield 'killed-both', line
669
yield 'killed-a', line
671
yield 'killed-b', line
673
yield 'unchanged', line
674
elif insert in inc_a:
675
if deleteset & inc_a:
676
yield 'ghost-a', line
680
elif insert in inc_b:
681
if deleteset & inc_b:
682
yield 'ghost-b', line
686
# not in either revision
687
yield 'irrelevant', line
689
yield 'unchanged', '' # terminator
693
def weave_merge(self, plan):
698
for state, line in plan:
699
if state == 'unchanged' or state == 'killed-both':
700
# resync and flush queued conflicts changes if any
701
if not lines_a and not lines_b:
703
elif ch_a and not ch_b:
705
for l in lines_a: yield l
706
elif ch_b and not ch_a:
707
for l in lines_b: yield l
708
elif lines_a == lines_b:
709
for l in lines_a: yield l
712
for l in lines_a: yield l
714
for l in lines_b: yield l
721
if state == 'unchanged':
724
elif state == 'killed-a':
727
elif state == 'killed-b':
730
elif state == 'new-a':
733
elif state == 'new-b':
737
assert state in ('irrelevant', 'ghost-a', 'ghost-b', 'killed-base',
748
"""Show the weave's table-of-contents"""
749
print '%6s %50s %10s %10s' % ('ver', 'name', 'sha1', 'parents')
750
for i in (6, 50, 10, 10):
753
for i in range(w.numversions()):
756
parent_str = ' '.join(map(str, w._parents[i]))
757
print '%6d %-50.50s %10.10s %s' % (i, name, sha1, parent_str)
761
def weave_stats(weave_file, pb):
762
from bzrlib.weavefile import read_weave
764
wf = file(weave_file, 'rb')
766
# FIXME: doesn't work on pipes
767
weave_size = wf.tell()
771
for i in range(vers):
772
pb.update('checking sizes', i, vers)
773
for origin, lineno, line in w._extract([i]):
778
print 'versions %9d' % vers
779
print 'weave file %9d bytes' % weave_size
780
print 'total contents %9d bytes' % total
781
print 'compression ratio %9.2fx' % (float(total) / float(weave_size))
784
print 'average size %9d bytes' % avg
785
print 'relative size %9.2fx' % (float(weave_size) / float(avg))
789
print """bzr weave tool
791
Experimental tool for weave algorithm.
795
Create an empty weave file
796
weave get WEAVEFILE VERSION
797
Write out specified version.
798
weave check WEAVEFILE
799
Check consistency of all versions.
801
Display table of contents.
802
weave add WEAVEFILE NAME [BASE...] < NEWTEXT
803
Add NEWTEXT, with specified parent versions.
804
weave annotate WEAVEFILE VERSION
805
Display origin of each line.
806
weave mash WEAVEFILE VERSION...
807
Display composite of all selected versions.
808
weave merge WEAVEFILE VERSION1 VERSION2 > OUT
809
Auto-merge two versions and display conflicts.
810
weave diff WEAVEFILE VERSION1 VERSION2
811
Show differences between two versions.
815
% weave init foo.weave
817
% weave add foo.weave ver0 < foo.txt
820
(create updated version)
822
% weave get foo.weave 0 | diff -u - foo.txt
823
% weave add foo.weave ver1 0 < foo.txt
826
% weave get foo.weave 0 > foo.txt (create forked version)
828
% weave add foo.weave ver2 0 < foo.txt
831
% weave merge foo.weave 1 2 > foo.txt (merge them)
832
% vi foo.txt (resolve conflicts)
833
% weave add foo.weave merged 1 2 < foo.txt (commit merged version)
845
# in case we're run directly from the subdirectory
846
sys.path.append('..')
848
from bzrlib.weavefile import write_weave, read_weave
849
from bzrlib.progress import ProgressBar
864
return read_weave(file(argv[2], 'rb'))
870
# at the moment, based on everything in the file
872
parents = map(int, argv[4:])
873
lines = sys.stdin.readlines()
874
ver = w.add(name, parents, lines)
875
write_weave(w, file(argv[2], 'wb'))
876
print 'added version %r %d' % (name, ver)
879
if os.path.exists(fn):
880
raise IOError("file exists")
882
write_weave(w, file(fn, 'wb'))
883
elif cmd == 'get': # get one version
885
sys.stdout.writelines(w.get_iter(int(argv[3])))
887
elif cmd == 'mash': # get composite
889
sys.stdout.writelines(w.mash_iter(map(int, argv[3:])))
892
from difflib import unified_diff
895
v1, v2 = map(int, argv[3:5])
898
diff_gen = unified_diff(lines1, lines2,
899
'%s version %d' % (fn, v1),
900
'%s version %d' % (fn, v2))
901
sys.stdout.writelines(diff_gen)
903
elif cmd == 'annotate':
905
# newline is added to all lines regardless; too hard to get
906
# reasonable formatting otherwise
908
for origin, text in w.annotate(int(argv[3])):
909
text = text.rstrip('\r\n')
911
print ' | %s' % (text)
913
print '%5d | %s' % (origin, text)
920
weave_stats(argv[2], ProgressBar())
927
print '%d versions ok' % w.numversions()
929
elif cmd == 'inclusions':
931
print ' '.join(map(str, w.inclusions([int(argv[3])])))
933
elif cmd == 'parents':
935
print ' '.join(map(str, w._parents[int(argv[3])]))
937
elif cmd == 'plan-merge':
939
for state, line in w.plan_merge(int(argv[3]), int(argv[4])):
941
print '%14s | %s' % (state, line),
945
p = w.plan_merge(int(argv[3]), int(argv[4]))
946
sys.stdout.writelines(w.weave_merge(p))
948
elif cmd == 'mash-merge':
954
v1, v2 = map(int, argv[3:5])
956
basis = w.inclusions([v1]).intersection(w.inclusions([v2]))
958
base_lines = list(w.mash_iter(basis))
959
a_lines = list(w.get(v1))
960
b_lines = list(w.get(v2))
962
from bzrlib.merge3 import Merge3
963
m3 = Merge3(base_lines, a_lines, b_lines)
965
name_a = 'version %d' % v1
966
name_b = 'version %d' % v2
967
sys.stdout.writelines(m3.merge_lines(name_a=name_a, name_b=name_b))
969
raise ValueError('unknown command %r' % cmd)
973
def profile_main(argv):
974
import tempfile, hotshot, hotshot.stats
976
prof_f = tempfile.NamedTemporaryFile()
978
prof = hotshot.Profile(prof_f.name)
980
ret = prof.runcall(main, argv)
983
stats = hotshot.stats.load(prof_f.name)
985
stats.sort_stats('cumulative')
986
## XXX: Might like to write to stderr or the trace file instead but
987
## print_stats seems hardcoded to stdout
988
stats.print_stats(20)
993
if __name__ == '__main__':
995
if '--profile' in sys.argv:
997
args.remove('--profile')
998
sys.exit(profile_main(args))
1000
sys.exit(main(sys.argv))