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?
87
# TODO: Is there any potential performance win by having an add()
88
# variant that is passed a pre-cooked version of the single basis
94
from difflib import SequenceMatcher
97
from bzrlib.osutils import sha_strings
100
class WeaveError(Exception):
101
"""Exception in processing weave"""
104
class WeaveFormatError(WeaveError):
105
"""Weave invariant violated"""
109
"""weave - versioned text file storage.
111
A Weave manages versions of line-based text files, keeping track
112
of the originating version for each line.
114
To clients the "lines" of the file are represented as a list of strings.
115
These strings will typically have terminal newline characters, but
116
this is not required. In particular files commonly do not have a newline
117
at the end of the file.
119
Texts can be identified in either of two ways:
121
* a nonnegative index number.
123
* a version-id string.
125
Typically the index number will be valid only inside this weave and
126
the version-id is used to reference it in the larger world.
128
The weave is represented as a list mixing edit instructions and
129
literal text. Each entry in _weave can be either a string (or
130
unicode), or a tuple. If a string, it means that the given line
131
should be output in the currently active revisions.
133
If a tuple, it gives a processing instruction saying in which
134
revisions the enclosed lines are active. The tuple has the form
135
(instruction, version).
137
The instruction can be '{' or '}' for an insertion block, and '['
138
and ']' for a deletion block respectively. The version is the
139
integer version index. There is no replace operator, only deletes
140
and inserts. For '}', the end of an insertion, there is no
141
version parameter because it always closes the most recently
146
* A later version can delete lines that were introduced by any
147
number of ancestor versions; this implies that deletion
148
instructions can span insertion blocks without regard to the
149
insertion block's nesting.
151
* Similarly, deletions need not be properly nested with regard to
152
each other, because they might have been generated by
153
independent revisions.
155
* Insertions are always made by inserting a new bracketed block
156
into a single point in the previous weave. This implies they
157
can nest but not overlap, and the nesting must always have later
158
insertions on the inside.
160
* It doesn't seem very useful to have an active insertion
161
inside an inactive insertion, but it might happen.
163
* Therefore, all instructions are always"considered"; that
164
is passed onto and off the stack. An outer inactive block
165
doesn't disable an inner block.
167
* Lines are enabled if the most recent enclosing insertion is
168
active and none of the enclosing deletions are active.
170
* There is no point having a deletion directly inside its own
171
insertion; you might as well just not write it. And there
172
should be no way to get an earlier version deleting a later
176
Text of the weave; list of control instruction tuples and strings.
179
List of parents, indexed by version number.
180
It is only necessary to store the minimal set of parents for
181
each version; the parent's parents are implied.
184
List of hex SHA-1 of each version.
187
List of symbolic names for each version. Each should be unique.
190
For each name, the version number.
193
Descriptive name of this weave; typically the filename if known.
197
__slots__ = ['_weave', '_parents', '_sha1s', '_names', '_name_map',
200
def __init__(self, weave_name=None):
206
self._weave_name = weave_name
209
def __eq__(self, other):
210
if not isinstance(other, Weave):
212
return self._parents == other._parents \
213
and self._weave == other._weave \
214
and self._sha1s == other._sha1s
217
def __ne__(self, other):
218
return not self.__eq__(other)
221
def maybe_lookup(self, name_or_index):
222
"""Convert possible symbolic name to index, or pass through indexes."""
223
if isinstance(name_or_index, (int, long)):
226
return self.lookup(name_or_index)
229
def lookup(self, name):
230
"""Convert symbolic version name to index."""
232
return self._name_map[name]
234
raise WeaveError("name %r not present in weave %r" %
235
(name, self._weave_name))
238
def idx_to_name(self, version):
239
return self._names[version]
242
def _check_repeated_add(self, name, parents, text, sha1):
243
"""Check that a duplicated add is OK.
245
If it is, return the (old) index; otherwise raise an exception.
247
idx = self.lookup(name)
248
if sorted(self._parents[idx]) != sorted(parents):
249
raise WeaveError("name \"%s\" already present in weave "
250
"with different parents" % name)
251
if sha1 != self._sha1s[idx]:
252
raise WeaveError("name \"%s\" already present in weave "
253
"with different text" % name)
258
def add(self, name, parents, text, sha1=None):
259
"""Add a single text on top of the weave.
261
Returns the index number of the newly added version.
264
Symbolic name for this version.
265
(Typically the revision-id of the revision that added it.)
268
List or set of direct parent version numbers.
271
Sequence of lines to be added in the new version.
273
sha -- SHA-1 of the file, if known. This is trusted to be
277
assert isinstance(name, basestring)
279
sha1 = sha_strings(text)
280
if name in self._name_map:
281
return self._check_repeated_add(name, parents, text, sha1)
283
parents = map(self.maybe_lookup, parents)
284
self._check_versions(parents)
285
## self._check_lines(text)
286
new_version = len(self._parents)
289
# if we abort after here the (in-memory) weave will be corrupt because only
290
# some fields are updated
291
self._parents.append(parents[:])
292
self._sha1s.append(sha1)
293
self._names.append(name)
294
self._name_map[name] = new_version
298
# special case; adding with no parents revision; can do
299
# this more quickly by just appending unconditionally.
300
# even more specially, if we're adding an empty text we
301
# need do nothing at all.
303
self._weave.append(('{', new_version))
304
self._weave.extend(text)
305
self._weave.append(('}', None))
309
if len(parents) == 1:
310
pv = list(parents)[0]
311
if sha1 == self._sha1s[pv]:
312
# special case: same as the single parent
316
ancestors = self.inclusions(parents)
320
# basis a list of (origin, lineno, line)
323
for origin, lineno, line in self._extract(ancestors):
324
basis_lineno.append(lineno)
325
basis_lines.append(line)
327
# another small special case: a merge, producing the same text
329
if text == basis_lines:
332
# add a sentinal, because we can also match against the final line
333
basis_lineno.append(len(self._weave))
335
# XXX: which line of the weave should we really consider
336
# matches the end of the file? the current code says it's the
337
# last line of the weave?
339
#print 'basis_lines:', basis_lines
340
#print 'new_lines: ', lines
342
s = SequenceMatcher(None, basis_lines, text)
344
# offset gives the number of lines that have been inserted
345
# into the weave up to the current point; if the original edit instruction
346
# says to change line A then we actually change (A+offset)
349
for tag, i1, i2, j1, j2 in s.get_opcodes():
350
# i1,i2 are given in offsets within basis_lines; we need to map them
351
# back to offsets within the entire weave
352
#print 'raw match', tag, i1, i2, j1, j2
356
i1 = basis_lineno[i1]
357
i2 = basis_lineno[i2]
359
assert 0 <= j1 <= j2 <= len(text)
361
#print tag, i1, i2, j1, j2
363
# the deletion and insertion are handled separately.
364
# first delete the region.
366
self._weave.insert(i1+offset, ('[', new_version))
367
self._weave.insert(i2+offset+1, (']', new_version))
371
# there may have been a deletion spanning up to
372
# i2; we want to insert after this region to make sure
373
# we don't destroy ourselves
375
self._weave[i:i] = ([('{', new_version)]
378
offset += 2 + (j2 - j1)
382
def add_identical(self, old_rev_id, new_rev_id, parents):
383
"""Add an identical text to old_rev_id as new_rev_id."""
384
old_lines = self.get(self.lookup(old_rev_id))
385
self.add(new_rev_id, parents, old_lines)
387
def inclusions(self, versions):
388
"""Return set of all ancestors of given version(s)."""
390
for v in xrange(max(versions), 0, -1):
392
# include all its parents
393
i.update(self._parents[v])
395
## except IndexError:
396
## raise ValueError("version %d not present in weave" % v)
399
def parents(self, version):
400
return self._parents[version]
403
def minimal_parents(self, version):
404
"""Find the minimal set of parents for the version."""
405
included = self._parents[version]
410
li.sort(reverse=True)
418
gotit.update(self.inclusions(pv))
420
assert mininc[0] >= 0
421
assert mininc[-1] < version
426
def _check_lines(self, text):
427
if not isinstance(text, list):
428
raise ValueError("text should be a list, not %s" % type(text))
431
if not isinstance(l, basestring):
432
raise ValueError("text line should be a string or unicode, not %s"
437
def _check_versions(self, indexes):
438
"""Check everything in the sequence of indexes is valid"""
443
raise IndexError("invalid version number %r" % i)
446
def annotate(self, name_or_index):
447
return list(self.annotate_iter(name_or_index))
450
def annotate_iter(self, name_or_index):
451
"""Yield list of (index-id, line) pairs for the specified version.
453
The index indicates when the line originated in the weave."""
454
incls = [self.maybe_lookup(name_or_index)]
455
for origin, lineno, text in self._extract(incls):
463
(lineno, insert, deletes, text)
464
for each literal line.
470
lineno = 0 # line of weave, 0-based
472
for l in self._weave:
473
if isinstance(l, tuple):
486
raise WeaveFormatError('unexpected instruction %r'
489
assert isinstance(l, basestring)
491
yield lineno, istack[-1], dset, l
496
def _extract(self, versions):
497
"""Yield annotation of lines in included set.
499
Yields a sequence of tuples (origin, lineno, text), where
500
origin is the origin version, lineno the index in the weave,
501
and text the text of the line.
503
The set typically but not necessarily corresponds to a version.
506
if not isinstance(i, int):
509
included = self.inclusions(versions)
514
lineno = 0 # line of weave, 0-based
520
WFE = WeaveFormatError
522
for l in self._weave:
523
if isinstance(l, tuple):
527
assert v not in istack
541
assert isinstance(l, basestring)
543
isactive = (not dset) and istack and (istack[-1] in included)
545
result.append((istack[-1], lineno, l))
549
raise WFE("unclosed insertion blocks at end of weave",
552
raise WFE("unclosed deletion blocks at end of weave",
559
def get_iter(self, name_or_index):
560
"""Yield lines for the specified version."""
561
incls = [self.maybe_lookup(name_or_index)]
562
for origin, lineno, line in self._extract(incls):
566
def get_text(self, name_or_index):
567
return ''.join(self.get_iter(name_or_index))
568
assert isinstance(version, int)
571
def get_lines(self, name_or_index):
572
return list(self.get_iter(name_or_index))
578
def mash_iter(self, included):
579
"""Return composed version of multiple included versions."""
580
included = map(self.maybe_lookup, included)
581
for origin, lineno, text in self._extract(included):
585
def dump(self, to_file):
586
from pprint import pprint
587
print >>to_file, "Weave._weave = ",
588
pprint(self._weave, to_file)
589
print >>to_file, "Weave._parents = ",
590
pprint(self._parents, to_file)
594
def numversions(self):
595
l = len(self._parents)
596
assert l == len(self._sha1s)
601
return self.numversions()
604
def check(self, progress_bar=None):
605
# check no circular inclusions
606
for version in range(self.numversions()):
607
inclusions = list(self._parents[version])
610
if inclusions[-1] >= version:
611
raise WeaveFormatError("invalid included version %d for index %d"
612
% (inclusions[-1], version))
614
# try extracting all versions; this is a bit slow and parallel
615
# extraction could be used
616
nv = self.numversions()
617
for version in range(nv):
619
progress_bar.update('checking text', version, nv)
621
for l in self.get_iter(version):
624
expected = self._sha1s[version]
626
raise WeaveError("mismatched sha1 for version %d; "
627
"got %s, expected %s"
628
% (version, hd, expected))
630
# TODO: check insertions are properly nested, that there are
631
# no lines outside of insertion blocks, that deletions are
632
# properly paired, etc.
636
def merge(self, merge_versions):
637
"""Automerge and mark conflicts between versions.
639
This returns a sequence, each entry describing alternatives
640
for a chunk of the file. Each of the alternatives is given as
643
If there is a chunk of the file where there's no diagreement,
644
only one alternative is given.
647
# approach: find the included versions common to all the
649
raise NotImplementedError()
653
def _delta(self, included, lines):
654
"""Return changes from basis to new revision.
656
The old text for comparison is the union of included revisions.
658
This is used in inserting a new text.
660
Delta is returned as a sequence of
661
(weave1, weave2, newlines).
663
This indicates that weave1:weave2 of the old weave should be
664
replaced by the sequence of lines in newlines. Note that
665
these line numbers are positions in the total weave and don't
666
correspond to the lines in any extracted version, or even the
667
extracted union of included versions.
669
If line1=line2, this is a pure insert; if newlines=[] this is a
670
pure delete. (Similar to difflib.)
675
def plan_merge(self, ver_a, ver_b):
676
"""Return pseudo-annotation indicating how the two versions merge.
678
This is computed between versions a and b and their common
681
Weave lines present in none of them are skipped entirely.
683
inc_a = self.inclusions([ver_a])
684
inc_b = self.inclusions([ver_b])
685
inc_c = inc_a & inc_b
687
for lineno, insert, deleteset, line in self._walk():
688
if deleteset & inc_c:
689
# killed in parent; can't be in either a or b
690
# not relevant to our work
691
yield 'killed-base', line
692
elif insert in inc_c:
693
# was inserted in base
694
killed_a = bool(deleteset & inc_a)
695
killed_b = bool(deleteset & inc_b)
696
if killed_a and killed_b:
697
yield 'killed-both', line
699
yield 'killed-a', line
701
yield 'killed-b', line
703
yield 'unchanged', line
704
elif insert in inc_a:
705
if deleteset & inc_a:
706
yield 'ghost-a', line
710
elif insert in inc_b:
711
if deleteset & inc_b:
712
yield 'ghost-b', line
716
# not in either revision
717
yield 'irrelevant', line
719
yield 'unchanged', '' # terminator
723
def weave_merge(self, plan):
728
for state, line in plan:
729
if state == 'unchanged' or state == 'killed-both':
730
# resync and flush queued conflicts changes if any
731
if not lines_a and not lines_b:
733
elif ch_a and not ch_b:
735
for l in lines_a: yield l
736
elif ch_b and not ch_a:
737
for l in lines_b: yield l
738
elif lines_a == lines_b:
739
for l in lines_a: yield l
742
for l in lines_a: yield l
744
for l in lines_b: yield l
751
if state == 'unchanged':
754
elif state == 'killed-a':
757
elif state == 'killed-b':
760
elif state == 'new-a':
763
elif state == 'new-b':
767
assert state in ('irrelevant', 'ghost-a', 'ghost-b', 'killed-base',
778
"""Show the weave's table-of-contents"""
779
print '%6s %50s %10s %10s' % ('ver', 'name', 'sha1', 'parents')
780
for i in (6, 50, 10, 10):
783
for i in range(w.numversions()):
786
parent_str = ' '.join(map(str, w._parents[i]))
787
print '%6d %-50.50s %10.10s %s' % (i, name, sha1, parent_str)
791
def weave_stats(weave_file):
792
from bzrlib.progress import ProgressBar
793
from bzrlib.weavefile import read_weave
797
wf = file(weave_file, 'rb')
799
# FIXME: doesn't work on pipes
800
weave_size = wf.tell()
804
for i in range(vers):
805
pb.update('checking sizes', i, vers)
806
for line in w.get_iter(i):
811
print 'versions %9d' % vers
812
print 'weave file %9d bytes' % weave_size
813
print 'total contents %9d bytes' % total
814
print 'compression ratio %9.2fx' % (float(total) / float(weave_size))
817
print 'average size %9d bytes' % avg
818
print 'relative size %9.2fx' % (float(weave_size) / float(avg))
822
print """bzr weave tool
824
Experimental tool for weave algorithm.
828
Create an empty weave file
829
weave get WEAVEFILE VERSION
830
Write out specified version.
831
weave check WEAVEFILE
832
Check consistency of all versions.
834
Display table of contents.
835
weave add WEAVEFILE NAME [BASE...] < NEWTEXT
836
Add NEWTEXT, with specified parent versions.
837
weave annotate WEAVEFILE VERSION
838
Display origin of each line.
839
weave mash WEAVEFILE VERSION...
840
Display composite of all selected versions.
841
weave merge WEAVEFILE VERSION1 VERSION2 > OUT
842
Auto-merge two versions and display conflicts.
843
weave diff WEAVEFILE VERSION1 VERSION2
844
Show differences between two versions.
848
% weave init foo.weave
850
% weave add foo.weave ver0 < foo.txt
853
(create updated version)
855
% weave get foo.weave 0 | diff -u - foo.txt
856
% weave add foo.weave ver1 0 < foo.txt
859
% weave get foo.weave 0 > foo.txt (create forked version)
861
% weave add foo.weave ver2 0 < foo.txt
864
% weave merge foo.weave 1 2 > foo.txt (merge them)
865
% vi foo.txt (resolve conflicts)
866
% weave add foo.weave merged 1 2 < foo.txt (commit merged version)
875
from bzrlib.weavefile import write_weave, read_weave
876
from bzrlib.progress import ProgressBar
891
return read_weave(file(argv[2], 'rb'))
897
# at the moment, based on everything in the file
899
parents = map(int, argv[4:])
900
lines = sys.stdin.readlines()
901
ver = w.add(name, parents, lines)
902
write_weave(w, file(argv[2], 'wb'))
903
print 'added version %r %d' % (name, ver)
906
if os.path.exists(fn):
907
raise IOError("file exists")
909
write_weave(w, file(fn, 'wb'))
910
elif cmd == 'get': # get one version
912
sys.stdout.writelines(w.get_iter(int(argv[3])))
914
elif cmd == 'mash': # get composite
916
sys.stdout.writelines(w.mash_iter(map(int, argv[3:])))
919
from difflib import unified_diff
922
v1, v2 = map(int, argv[3:5])
925
diff_gen = unified_diff(lines1, lines2,
926
'%s version %d' % (fn, v1),
927
'%s version %d' % (fn, v2))
928
sys.stdout.writelines(diff_gen)
930
elif cmd == 'annotate':
932
# newline is added to all lines regardless; too hard to get
933
# reasonable formatting otherwise
935
for origin, text in w.annotate(int(argv[3])):
936
text = text.rstrip('\r\n')
938
print ' | %s' % (text)
940
print '%5d | %s' % (origin, text)
954
print '%d versions ok' % w.numversions()
956
elif cmd == 'inclusions':
958
print ' '.join(map(str, w.inclusions([int(argv[3])])))
960
elif cmd == 'parents':
962
print ' '.join(map(str, w._parents[int(argv[3])]))
964
elif cmd == 'plan-merge':
966
for state, line in w.plan_merge(int(argv[3]), int(argv[4])):
968
print '%14s | %s' % (state, line),
972
p = w.plan_merge(int(argv[3]), int(argv[4]))
973
sys.stdout.writelines(w.weave_merge(p))
975
elif cmd == 'mash-merge':
981
v1, v2 = map(int, argv[3:5])
983
basis = w.inclusions([v1]).intersection(w.inclusions([v2]))
985
base_lines = list(w.mash_iter(basis))
986
a_lines = list(w.get(v1))
987
b_lines = list(w.get(v2))
989
from bzrlib.merge3 import Merge3
990
m3 = Merge3(base_lines, a_lines, b_lines)
992
name_a = 'version %d' % v1
993
name_b = 'version %d' % v2
994
sys.stdout.writelines(m3.merge_lines(name_a=name_a, name_b=name_b))
996
raise ValueError('unknown command %r' % cmd)
1000
def profile_main(argv):
1001
import tempfile, hotshot, hotshot.stats
1003
prof_f = tempfile.NamedTemporaryFile()
1005
prof = hotshot.Profile(prof_f.name)
1007
ret = prof.runcall(main, argv)
1010
stats = hotshot.stats.load(prof_f.name)
1012
stats.sort_stats('cumulative')
1013
## XXX: Might like to write to stderr or the trace file instead but
1014
## print_stats seems hardcoded to stdout
1015
stats.print_stats(20)
1020
if __name__ == '__main__':
1022
if '--profile' in sys.argv:
1024
args.remove('--profile')
1025
sys.exit(profile_main(args))
1027
sys.exit(main(sys.argv))
added
removed
bzrlib/weave.py
