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?
90
from cStringIO import StringIO
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
Descriptive name of this weave; typically the filename if known.
190
__slots__ = ['_weave', '_parents', '_sha1s', '_names', '_name_map',
193
def __init__(self, weave_name=None):
199
self._weave_name = weave_name
202
def __eq__(self, other):
203
if not isinstance(other, Weave):
205
return self._parents == other._parents \
206
and self._weave == other._weave \
207
and self._sha1s == other._sha1s
210
def __ne__(self, other):
211
return not self.__eq__(other)
214
def lookup(self, name):
216
return self._name_map[name]
218
raise WeaveError("name %s not present in weave %s" %
219
(name, self._weave_name))
222
def add(self, name, parents, text):
223
"""Add a single text on top of the weave.
225
Returns the index number of the newly added version.
228
Symbolic name for this version.
229
(Typically the revision-id of the revision that added it.)
232
List or set of direct parent version numbers.
235
Sequence of lines to be added in the new version."""
237
assert isinstance(name, basestring)
238
if name in self._name_map:
239
raise WeaveError("name %r already present in weave" % name)
241
self._check_versions(parents)
242
## self._check_lines(text)
243
new_version = len(self._parents)
250
# if we abort after here the (in-memory) weave will be corrupt because only
251
# some fields are updated
252
self._parents.append(parents[:])
253
self._sha1s.append(sha1)
254
self._names.append(name)
255
self._name_map[name] = new_version
259
# special case; adding with no parents revision; can do
260
# this more quickly by just appending unconditionally.
261
# even more specially, if we're adding an empty text we
262
# need do nothing at all.
264
self._weave.append(('{', new_version))
265
self._weave.extend(text)
266
self._weave.append(('}', None))
270
if len(parents) == 1:
271
pv = list(parents)[0]
272
if sha1 == self._sha1s[pv]:
273
# special case: same as the single parent
277
ancestors = self.inclusions(parents)
281
# basis a list of (origin, lineno, line)
284
for origin, lineno, line in self._extract(ancestors):
285
basis_lineno.append(lineno)
286
basis_lines.append(line)
288
# another small special case: a merge, producing the same text
290
if text == basis_lines:
293
# add a sentinal, because we can also match against the final line
294
basis_lineno.append(len(self._weave))
296
# XXX: which line of the weave should we really consider
297
# matches the end of the file? the current code says it's the
298
# last line of the weave?
300
#print 'basis_lines:', basis_lines
301
#print 'new_lines: ', lines
303
from difflib import SequenceMatcher
304
s = SequenceMatcher(None, basis_lines, text)
306
# offset gives the number of lines that have been inserted
307
# into the weave up to the current point; if the original edit instruction
308
# says to change line A then we actually change (A+offset)
311
for tag, i1, i2, j1, j2 in s.get_opcodes():
312
# i1,i2 are given in offsets within basis_lines; we need to map them
313
# back to offsets within the entire weave
314
#print 'raw match', tag, i1, i2, j1, j2
318
i1 = basis_lineno[i1]
319
i2 = basis_lineno[i2]
321
assert 0 <= j1 <= j2 <= len(text)
323
#print tag, i1, i2, j1, j2
325
# the deletion and insertion are handled separately.
326
# first delete the region.
328
self._weave.insert(i1+offset, ('[', new_version))
329
self._weave.insert(i2+offset+1, (']', new_version))
333
# there may have been a deletion spanning up to
334
# i2; we want to insert after this region to make sure
335
# we don't destroy ourselves
337
self._weave[i:i] = ([('{', new_version)]
340
offset += 2 + (j2 - j1)
345
def inclusions(self, versions):
346
"""Return set of all ancestors of given version(s)."""
352
# include all its parents
353
i.update(self._parents[v])
357
raise ValueError("version %d not present in weave" % v)
360
def minimal_parents(self, version):
361
"""Find the minimal set of parents for the version."""
362
included = self._parents[version]
367
li.sort(reverse=True)
375
gotit.update(self.inclusions(pv))
377
assert mininc[0] >= 0
378
assert mininc[-1] < version
383
def _check_lines(self, text):
384
if not isinstance(text, list):
385
raise ValueError("text should be a list, not %s" % type(text))
388
if not isinstance(l, basestring):
389
raise ValueError("text line should be a string or unicode, not %s"
394
def _check_versions(self, indexes):
395
"""Check everything in the sequence of indexes is valid"""
400
raise IndexError("invalid version number %r" % i)
403
def annotate(self, index):
404
return list(self.annotate_iter(index))
407
def annotate_iter(self, version):
408
"""Yield list of (index-id, line) pairs for the specified version.
410
The index indicates when the line originated in the weave."""
411
for origin, lineno, text in self._extract([version]):
419
(lineno, insert, deletes, text)
420
for each literal line.
426
lineno = 0 # line of weave, 0-based
428
for l in self._weave:
429
if isinstance(l, tuple):
442
raise WeaveFormatError('unexpected instruction %r'
445
assert isinstance(l, basestring)
447
yield lineno, istack[-1], dset, l
452
def _extract(self, versions):
453
"""Yield annotation of lines in included set.
455
Yields a sequence of tuples (origin, lineno, text), where
456
origin is the origin version, lineno the index in the weave,
457
and text the text of the line.
459
The set typically but not necessarily corresponds to a version.
462
if not isinstance(i, int):
465
included = self.inclusions(versions)
470
lineno = 0 # line of weave, 0-based
476
WFE = WeaveFormatError
478
for l in self._weave:
479
if isinstance(l, tuple):
483
assert v not in istack
497
assert isinstance(l, basestring)
499
isactive = (not dset) and istack and (istack[-1] in included)
501
result.append((istack[-1], lineno, l))
505
raise WFE("unclosed insertion blocks at end of weave",
508
raise WFE("unclosed deletion blocks at end of weave",
515
def get_iter(self, version):
516
"""Yield lines for the specified version."""
517
for origin, lineno, line in self._extract([version]):
521
def get_text(self, version):
522
assert isinstance(version, int)
524
s.writelines(self.get_iter(version))
528
def get(self, index):
529
return list(self.get_iter(index))
532
def mash_iter(self, included):
533
"""Return composed version of multiple included versions."""
534
for origin, lineno, text in self._extract(included):
538
def dump(self, to_file):
539
from pprint import pprint
540
print >>to_file, "Weave._weave = ",
541
pprint(self._weave, to_file)
542
print >>to_file, "Weave._parents = ",
543
pprint(self._parents, to_file)
547
def numversions(self):
548
l = len(self._parents)
549
assert l == len(self._sha1s)
554
return self.numversions()
557
def check(self, progress_bar=None):
558
# check no circular inclusions
559
for version in range(self.numversions()):
560
inclusions = list(self._parents[version])
563
if inclusions[-1] >= version:
564
raise WeaveFormatError("invalid included version %d for index %d"
565
% (inclusions[-1], version))
567
# try extracting all versions; this is a bit slow and parallel
568
# extraction could be used
569
nv = self.numversions()
570
for version in range(nv):
572
progress_bar.update('checking text', version, nv)
574
for l in self.get_iter(version):
577
expected = self._sha1s[version]
579
raise WeaveError("mismatched sha1 for version %d; "
580
"got %s, expected %s"
581
% (version, hd, expected))
583
# TODO: check insertions are properly nested, that there are
584
# no lines outside of insertion blocks, that deletions are
585
# properly paired, etc.
589
def merge(self, merge_versions):
590
"""Automerge and mark conflicts between versions.
592
This returns a sequence, each entry describing alternatives
593
for a chunk of the file. Each of the alternatives is given as
596
If there is a chunk of the file where there's no diagreement,
597
only one alternative is given.
600
# approach: find the included versions common to all the
602
raise NotImplementedError()
606
def _delta(self, included, lines):
607
"""Return changes from basis to new revision.
609
The old text for comparison is the union of included revisions.
611
This is used in inserting a new text.
613
Delta is returned as a sequence of
614
(weave1, weave2, newlines).
616
This indicates that weave1:weave2 of the old weave should be
617
replaced by the sequence of lines in newlines. Note that
618
these line numbers are positions in the total weave and don't
619
correspond to the lines in any extracted version, or even the
620
extracted union of included versions.
622
If line1=line2, this is a pure insert; if newlines=[] this is a
623
pure delete. (Similar to difflib.)
628
def plan_merge(self, ver_a, ver_b):
629
"""Return pseudo-annotation indicating how the two versions merge.
631
This is computed between versions a and b and their common
634
Weave lines present in none of them are skipped entirely.
636
inc_a = self.inclusions([ver_a])
637
inc_b = self.inclusions([ver_b])
638
inc_c = inc_a & inc_b
640
for lineno, insert, deleteset, line in self._walk():
641
if deleteset & inc_c:
642
# killed in parent; can't be in either a or b
643
# not relevant to our work
644
yield 'killed-base', line
645
elif insert in inc_c:
646
# was inserted in base
647
killed_a = bool(deleteset & inc_a)
648
killed_b = bool(deleteset & inc_b)
649
if killed_a and killed_b:
650
yield 'killed-both', line
652
yield 'killed-a', line
654
yield 'killed-b', line
656
yield 'unchanged', line
657
elif insert in inc_a:
658
if deleteset & inc_a:
659
yield 'ghost-a', line
663
elif insert in inc_b:
664
if deleteset & inc_b:
665
yield 'ghost-b', line
669
# not in either revision
670
yield 'irrelevant', line
672
yield 'unchanged', '' # terminator
676
def weave_merge(self, plan):
681
for state, line in plan:
682
if state == 'unchanged' or state == 'killed-both':
683
# resync and flush queued conflicts changes if any
684
if not lines_a and not lines_b:
686
elif ch_a and not ch_b:
688
for l in lines_a: yield l
689
elif ch_b and not ch_a:
690
for l in lines_b: yield l
691
elif lines_a == lines_b:
692
for l in lines_a: yield l
695
for l in lines_a: yield l
697
for l in lines_b: yield l
704
if state == 'unchanged':
707
elif state == 'killed-a':
710
elif state == 'killed-b':
713
elif state == 'new-a':
716
elif state == 'new-b':
720
assert state in ('irrelevant', 'ghost-a', 'ghost-b', 'killed-base',
731
"""Show the weave's table-of-contents"""
732
print '%6s %50s %10s %10s' % ('ver', 'name', 'sha1', 'parents')
733
for i in (6, 50, 10, 10):
736
for i in range(w.numversions()):
739
parent_str = ' '.join(map(str, w._parents[i]))
740
print '%6d %-50.50s %10.10s %s' % (i, name, sha1, parent_str)
744
def weave_stats(weave_file):
745
from bzrlib.progress import ProgressBar
746
from bzrlib.weavefile import read_weave
750
wf = file(weave_file, 'rb')
752
# FIXME: doesn't work on pipes
753
weave_size = wf.tell()
757
for i in range(vers):
758
pb.update('checking sizes', i, vers)
759
for line in w.get_iter(i):
764
print 'versions %9d' % vers
765
print 'weave file %9d bytes' % weave_size
766
print 'total contents %9d bytes' % total
767
print 'compression ratio %9.2fx' % (float(total) / float(weave_size))
770
print 'average size %9d bytes' % avg
771
print 'relative size %9.2fx' % (float(weave_size) / float(avg))
775
print """bzr weave tool
777
Experimental tool for weave algorithm.
781
Create an empty weave file
782
weave get WEAVEFILE VERSION
783
Write out specified version.
784
weave check WEAVEFILE
785
Check consistency of all versions.
787
Display table of contents.
788
weave add WEAVEFILE NAME [BASE...] < NEWTEXT
789
Add NEWTEXT, with specified parent versions.
790
weave annotate WEAVEFILE VERSION
791
Display origin of each line.
792
weave mash WEAVEFILE VERSION...
793
Display composite of all selected versions.
794
weave merge WEAVEFILE VERSION1 VERSION2 > OUT
795
Auto-merge two versions and display conflicts.
799
% weave init foo.weave
801
% weave add foo.weave ver0 < foo.txt
804
(create updated version)
806
% weave get foo.weave 0 | diff -u - foo.txt
807
% weave add foo.weave ver1 0 < foo.txt
810
% weave get foo.weave 0 > foo.txt (create forked version)
812
% weave add foo.weave ver2 0 < foo.txt
815
% weave merge foo.weave 1 2 > foo.txt (merge them)
816
% vi foo.txt (resolve conflicts)
817
% weave add foo.weave merged 1 2 < foo.txt (commit merged version)
826
from weavefile import write_weave, read_weave
827
from bzrlib.progress import ProgressBar
842
return read_weave(file(argv[2], 'rb'))
848
# at the moment, based on everything in the file
850
parents = map(int, argv[4:])
851
lines = sys.stdin.readlines()
852
ver = w.add(name, parents, lines)
853
write_weave(w, file(argv[2], 'wb'))
854
print 'added version %r %d' % (name, ver)
857
if os.path.exists(fn):
858
raise IOError("file exists")
860
write_weave(w, file(fn, 'wb'))
861
elif cmd == 'get': # get one version
863
sys.stdout.writelines(w.get_iter(int(argv[3])))
865
elif cmd == 'mash': # get composite
867
sys.stdout.writelines(w.mash_iter(map(int, argv[3:])))
869
elif cmd == 'annotate':
871
# newline is added to all lines regardless; too hard to get
872
# reasonable formatting otherwise
874
for origin, text in w.annotate(int(argv[3])):
875
text = text.rstrip('\r\n')
877
print ' | %s' % (text)
879
print '%5d | %s' % (origin, text)
893
print '%d versions ok' % w.numversions()
895
elif cmd == 'inclusions':
897
print ' '.join(map(str, w.inclusions([int(argv[3])])))
899
elif cmd == 'parents':
901
print ' '.join(map(str, w._parents[int(argv[3])]))
903
elif cmd == 'plan-merge':
905
for state, line in w.plan_merge(int(argv[3]), int(argv[4])):
907
print '%14s | %s' % (state, line),
911
p = w.plan_merge(int(argv[3]), int(argv[4]))
912
sys.stdout.writelines(w.weave_merge(p))
914
elif cmd == 'mash-merge':
920
v1, v2 = map(int, argv[3:5])
922
basis = w.inclusions([v1]).intersection(w.inclusions([v2]))
924
base_lines = list(w.mash_iter(basis))
925
a_lines = list(w.get(v1))
926
b_lines = list(w.get(v2))
928
from bzrlib.merge3 import Merge3
929
m3 = Merge3(base_lines, a_lines, b_lines)
931
name_a = 'version %d' % v1
932
name_b = 'version %d' % v2
933
sys.stdout.writelines(m3.merge_lines(name_a=name_a, name_b=name_b))
935
raise ValueError('unknown command %r' % cmd)
939
def profile_main(argv):
940
import tempfile, hotshot, hotshot.stats
942
prof_f = tempfile.NamedTemporaryFile()
944
prof = hotshot.Profile(prof_f.name)
946
ret = prof.runcall(main, argv)
949
stats = hotshot.stats.load(prof_f.name)
951
stats.sort_stats('cumulative')
952
## XXX: Might like to write to stderr or the trace file instead but
953
## print_stats seems hardcoded to stdout
954
stats.print_stats(20)
959
if __name__ == '__main__':
961
if '--profile' in sys.argv:
963
args.remove('--profile')
964
sys.exit(profile_main(args))
966
sys.exit(main(sys.argv))