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?
91
from cStringIO import StringIO
93
from bzrlib.osutils import sha_strings
96
class WeaveError(Exception):
97
"""Exception in processing weave"""
100
class WeaveFormatError(WeaveError):
101
"""Weave invariant violated"""
105
"""weave - versioned text file storage.
107
A Weave manages versions of line-based text files, keeping track
108
of the originating version for each line.
110
To clients the "lines" of the file are represented as a list of strings.
111
These strings will typically have terminal newline characters, but
112
this is not required. In particular files commonly do not have a newline
113
at the end of the file.
115
Texts can be identified in either of two ways:
117
* a nonnegative index number.
119
* a version-id string. (not implemented yet)
121
Typically the index number will be valid only inside this weave and
122
the version-id is used to reference it in the larger world.
124
The weave is represented as a list mixing edit instructions and
125
literal text. Each entry in _weave can be either a string (or
126
unicode), or a tuple. If a string, it means that the given line
127
should be output in the currently active revisions.
129
If a tuple, it gives a processing instruction saying in which
130
revisions the enclosed lines are active. The tuple has the form
131
(instruction, version).
133
The instruction can be '{' or '}' for an insertion block, and '['
134
and ']' for a deletion block respectively. The version is the
135
integer version index. There is no replace operator, only deletes
136
and inserts. For '}', the end of an insertion, there is no
137
version parameter because it always closes the most recently
142
* A later version can delete lines that were introduced by any
143
number of ancestor versions; this implies that deletion
144
instructions can span insertion blocks without regard to the
145
insertion block's nesting.
147
* Similarly, deletions need not be properly nested with regard to
148
each other, because they might have been generated by
149
independent revisions.
151
* Insertions are always made by inserting a new bracketed block
152
into a single point in the previous weave. This implies they
153
can nest but not overlap, and the nesting must always have later
154
insertions on the inside.
156
* It doesn't seem very useful to have an active insertion
157
inside an inactive insertion, but it might happen.
159
* Therefore, all instructions are always"considered"; that
160
is passed onto and off the stack. An outer inactive block
161
doesn't disable an inner block.
163
* Lines are enabled if the most recent enclosing insertion is
164
active and none of the enclosing deletions are active.
166
* There is no point having a deletion directly inside its own
167
insertion; you might as well just not write it. And there
168
should be no way to get an earlier version deleting a later
172
Text of the weave; list of control instruction tuples and strings.
175
List of parents, indexed by version number.
176
It is only necessary to store the minimal set of parents for
177
each version; the parent's parents are implied.
180
List of hex SHA-1 of each version.
183
List of symbolic names for each version. Each should be unique.
186
For each name, the version number.
189
Descriptive name of this weave; typically the filename if known.
193
__slots__ = ['_weave', '_parents', '_sha1s', '_names', '_name_map',
196
def __init__(self, weave_name=None):
202
self._weave_name = weave_name
205
def __eq__(self, other):
206
if not isinstance(other, Weave):
208
return self._parents == other._parents \
209
and self._weave == other._weave \
210
and self._sha1s == other._sha1s
213
def __ne__(self, other):
214
return not self.__eq__(other)
217
def maybe_lookup(self, name_or_index):
218
"""Convert possible symbolic name to index, or pass through indexes."""
219
if isinstance(name_or_index, (int, long)):
222
return self.lookup(name_or_index)
225
def lookup(self, name):
226
"""Convert symbolic version name to index."""
228
return self._name_map[name]
230
raise WeaveError("name %r not present in weave %r" %
231
(name, self._weave_name))
234
def idx_to_name(self, version):
235
return self._names[version]
238
def _check_repeated_add(self, name, parents, text):
239
"""Check that a duplicated add is OK.
241
If it is, return the (old) index; otherwise raise an exception.
243
idx = self.lookup(name)
244
if sorted(self._parents[idx]) != sorted(parents):
245
raise WeaveError("name \"%s\" already present in weave "
246
"with different parents" % name)
247
new_sha1 = sha_strings(text)
248
if new_sha1 != self._sha1s[idx]:
249
raise WeaveError("name \"%s\" already present in weave "
250
"with different text" % name)
255
def add(self, name, parents, text):
256
"""Add a single text on top of the weave.
258
Returns the index number of the newly added version.
261
Symbolic name for this version.
262
(Typically the revision-id of the revision that added it.)
265
List or set of direct parent version numbers.
268
Sequence of lines to be added in the new version."""
270
assert isinstance(name, basestring)
271
if name in self._name_map:
272
return self._check_repeated_add(name, parents, text)
274
parents = map(self.maybe_lookup, parents)
275
self._check_versions(parents)
276
## self._check_lines(text)
277
new_version = len(self._parents)
279
sha1 = sha_strings(text)
281
# if we abort after here the (in-memory) weave will be corrupt because only
282
# some fields are updated
283
self._parents.append(parents[:])
284
self._sha1s.append(sha1)
285
self._names.append(name)
286
self._name_map[name] = new_version
290
# special case; adding with no parents revision; can do
291
# this more quickly by just appending unconditionally.
292
# even more specially, if we're adding an empty text we
293
# need do nothing at all.
295
self._weave.append(('{', new_version))
296
self._weave.extend(text)
297
self._weave.append(('}', None))
301
if len(parents) == 1:
302
pv = list(parents)[0]
303
if sha1 == self._sha1s[pv]:
304
# special case: same as the single parent
308
ancestors = self.inclusions(parents)
312
# basis a list of (origin, lineno, line)
315
for origin, lineno, line in self._extract(ancestors):
316
basis_lineno.append(lineno)
317
basis_lines.append(line)
319
# another small special case: a merge, producing the same text
321
if text == basis_lines:
324
# add a sentinal, because we can also match against the final line
325
basis_lineno.append(len(self._weave))
327
# XXX: which line of the weave should we really consider
328
# matches the end of the file? the current code says it's the
329
# last line of the weave?
331
#print 'basis_lines:', basis_lines
332
#print 'new_lines: ', lines
334
from difflib import SequenceMatcher
335
s = SequenceMatcher(None, basis_lines, text)
337
# offset gives the number of lines that have been inserted
338
# into the weave up to the current point; if the original edit instruction
339
# says to change line A then we actually change (A+offset)
342
for tag, i1, i2, j1, j2 in s.get_opcodes():
343
# i1,i2 are given in offsets within basis_lines; we need to map them
344
# back to offsets within the entire weave
345
#print 'raw match', tag, i1, i2, j1, j2
349
i1 = basis_lineno[i1]
350
i2 = basis_lineno[i2]
352
assert 0 <= j1 <= j2 <= len(text)
354
#print tag, i1, i2, j1, j2
356
# the deletion and insertion are handled separately.
357
# first delete the region.
359
self._weave.insert(i1+offset, ('[', new_version))
360
self._weave.insert(i2+offset+1, (']', new_version))
364
# there may have been a deletion spanning up to
365
# i2; we want to insert after this region to make sure
366
# we don't destroy ourselves
368
self._weave[i:i] = ([('{', new_version)]
371
offset += 2 + (j2 - j1)
376
def inclusions(self, versions):
377
"""Return set of all ancestors of given version(s)."""
383
# include all its parents
384
i.update(self._parents[v])
388
raise ValueError("version %d not present in weave" % v)
391
def parents(self, version):
392
return self._parents[version]
395
def minimal_parents(self, version):
396
"""Find the minimal set of parents for the version."""
397
included = self._parents[version]
402
li.sort(reverse=True)
410
gotit.update(self.inclusions(pv))
412
assert mininc[0] >= 0
413
assert mininc[-1] < version
418
def _check_lines(self, text):
419
if not isinstance(text, list):
420
raise ValueError("text should be a list, not %s" % type(text))
423
if not isinstance(l, basestring):
424
raise ValueError("text line should be a string or unicode, not %s"
429
def _check_versions(self, indexes):
430
"""Check everything in the sequence of indexes is valid"""
435
raise IndexError("invalid version number %r" % i)
438
def annotate(self, name_or_index):
439
return list(self.annotate_iter(name_or_index))
442
def annotate_iter(self, name_or_index):
443
"""Yield list of (index-id, line) pairs for the specified version.
445
The index indicates when the line originated in the weave."""
446
incls = [self.maybe_lookup(name_or_index)]
447
for origin, lineno, text in self._extract(incls):
455
(lineno, insert, deletes, text)
456
for each literal line.
462
lineno = 0 # line of weave, 0-based
464
for l in self._weave:
465
if isinstance(l, tuple):
478
raise WeaveFormatError('unexpected instruction %r'
481
assert isinstance(l, basestring)
483
yield lineno, istack[-1], dset, l
488
def _extract(self, versions):
489
"""Yield annotation of lines in included set.
491
Yields a sequence of tuples (origin, lineno, text), where
492
origin is the origin version, lineno the index in the weave,
493
and text the text of the line.
495
The set typically but not necessarily corresponds to a version.
498
if not isinstance(i, int):
501
included = self.inclusions(versions)
506
lineno = 0 # line of weave, 0-based
512
WFE = WeaveFormatError
514
for l in self._weave:
515
if isinstance(l, tuple):
519
assert v not in istack
533
assert isinstance(l, basestring)
535
isactive = (not dset) and istack and (istack[-1] in included)
537
result.append((istack[-1], lineno, l))
541
raise WFE("unclosed insertion blocks at end of weave",
544
raise WFE("unclosed deletion blocks at end of weave",
551
def get_iter(self, name_or_index):
552
"""Yield lines for the specified version."""
553
incls = [self.maybe_lookup(name_or_index)]
554
for origin, lineno, line in self._extract(incls):
558
def get_text(self, version):
559
assert isinstance(version, int)
561
s.writelines(self.get_iter(version))
565
def get(self, name_or_index):
566
return list(self.get_iter(name_or_index))
569
def mash_iter(self, included):
570
"""Return composed version of multiple included versions."""
571
for origin, lineno, text in self._extract(included):
575
def dump(self, to_file):
576
from pprint import pprint
577
print >>to_file, "Weave._weave = ",
578
pprint(self._weave, to_file)
579
print >>to_file, "Weave._parents = ",
580
pprint(self._parents, to_file)
584
def numversions(self):
585
l = len(self._parents)
586
assert l == len(self._sha1s)
591
return self.numversions()
594
def check(self, progress_bar=None):
595
# check no circular inclusions
596
for version in range(self.numversions()):
597
inclusions = list(self._parents[version])
600
if inclusions[-1] >= version:
601
raise WeaveFormatError("invalid included version %d for index %d"
602
% (inclusions[-1], version))
604
# try extracting all versions; this is a bit slow and parallel
605
# extraction could be used
606
nv = self.numversions()
607
for version in range(nv):
609
progress_bar.update('checking text', version, nv)
611
for l in self.get_iter(version):
614
expected = self._sha1s[version]
616
raise WeaveError("mismatched sha1 for version %d; "
617
"got %s, expected %s"
618
% (version, hd, expected))
620
# TODO: check insertions are properly nested, that there are
621
# no lines outside of insertion blocks, that deletions are
622
# properly paired, etc.
626
def merge(self, merge_versions):
627
"""Automerge and mark conflicts between versions.
629
This returns a sequence, each entry describing alternatives
630
for a chunk of the file. Each of the alternatives is given as
633
If there is a chunk of the file where there's no diagreement,
634
only one alternative is given.
637
# approach: find the included versions common to all the
639
raise NotImplementedError()
643
def _delta(self, included, lines):
644
"""Return changes from basis to new revision.
646
The old text for comparison is the union of included revisions.
648
This is used in inserting a new text.
650
Delta is returned as a sequence of
651
(weave1, weave2, newlines).
653
This indicates that weave1:weave2 of the old weave should be
654
replaced by the sequence of lines in newlines. Note that
655
these line numbers are positions in the total weave and don't
656
correspond to the lines in any extracted version, or even the
657
extracted union of included versions.
659
If line1=line2, this is a pure insert; if newlines=[] this is a
660
pure delete. (Similar to difflib.)
665
def plan_merge(self, ver_a, ver_b):
666
"""Return pseudo-annotation indicating how the two versions merge.
668
This is computed between versions a and b and their common
671
Weave lines present in none of them are skipped entirely.
673
inc_a = self.inclusions([ver_a])
674
inc_b = self.inclusions([ver_b])
675
inc_c = inc_a & inc_b
677
for lineno, insert, deleteset, line in self._walk():
678
if deleteset & inc_c:
679
# killed in parent; can't be in either a or b
680
# not relevant to our work
681
yield 'killed-base', line
682
elif insert in inc_c:
683
# was inserted in base
684
killed_a = bool(deleteset & inc_a)
685
killed_b = bool(deleteset & inc_b)
686
if killed_a and killed_b:
687
yield 'killed-both', line
689
yield 'killed-a', line
691
yield 'killed-b', line
693
yield 'unchanged', line
694
elif insert in inc_a:
695
if deleteset & inc_a:
696
yield 'ghost-a', line
700
elif insert in inc_b:
701
if deleteset & inc_b:
702
yield 'ghost-b', line
706
# not in either revision
707
yield 'irrelevant', line
709
yield 'unchanged', '' # terminator
713
def weave_merge(self, plan):
718
for state, line in plan:
719
if state == 'unchanged' or state == 'killed-both':
720
# resync and flush queued conflicts changes if any
721
if not lines_a and not lines_b:
723
elif ch_a and not ch_b:
725
for l in lines_a: yield l
726
elif ch_b and not ch_a:
727
for l in lines_b: yield l
728
elif lines_a == lines_b:
729
for l in lines_a: yield l
732
for l in lines_a: yield l
734
for l in lines_b: yield l
741
if state == 'unchanged':
744
elif state == 'killed-a':
747
elif state == 'killed-b':
750
elif state == 'new-a':
753
elif state == 'new-b':
757
assert state in ('irrelevant', 'ghost-a', 'ghost-b', 'killed-base',
768
"""Show the weave's table-of-contents"""
769
print '%6s %50s %10s %10s' % ('ver', 'name', 'sha1', 'parents')
770
for i in (6, 50, 10, 10):
773
for i in range(w.numversions()):
776
parent_str = ' '.join(map(str, w._parents[i]))
777
print '%6d %-50.50s %10.10s %s' % (i, name, sha1, parent_str)
781
def weave_stats(weave_file):
782
from bzrlib.progress import ProgressBar
783
from bzrlib.weavefile import read_weave
787
wf = file(weave_file, 'rb')
789
# FIXME: doesn't work on pipes
790
weave_size = wf.tell()
794
for i in range(vers):
795
pb.update('checking sizes', i, vers)
796
for line in w.get_iter(i):
801
print 'versions %9d' % vers
802
print 'weave file %9d bytes' % weave_size
803
print 'total contents %9d bytes' % total
804
print 'compression ratio %9.2fx' % (float(total) / float(weave_size))
807
print 'average size %9d bytes' % avg
808
print 'relative size %9.2fx' % (float(weave_size) / float(avg))
812
print """bzr weave tool
814
Experimental tool for weave algorithm.
818
Create an empty weave file
819
weave get WEAVEFILE VERSION
820
Write out specified version.
821
weave check WEAVEFILE
822
Check consistency of all versions.
824
Display table of contents.
825
weave add WEAVEFILE NAME [BASE...] < NEWTEXT
826
Add NEWTEXT, with specified parent versions.
827
weave annotate WEAVEFILE VERSION
828
Display origin of each line.
829
weave mash WEAVEFILE VERSION...
830
Display composite of all selected versions.
831
weave merge WEAVEFILE VERSION1 VERSION2 > OUT
832
Auto-merge two versions and display conflicts.
836
% weave init foo.weave
838
% weave add foo.weave ver0 < foo.txt
841
(create updated version)
843
% weave get foo.weave 0 | diff -u - foo.txt
844
% weave add foo.weave ver1 0 < foo.txt
847
% weave get foo.weave 0 > foo.txt (create forked version)
849
% weave add foo.weave ver2 0 < foo.txt
852
% weave merge foo.weave 1 2 > foo.txt (merge them)
853
% vi foo.txt (resolve conflicts)
854
% weave add foo.weave merged 1 2 < foo.txt (commit merged version)
863
from weavefile import write_weave, read_weave
864
from bzrlib.progress import ProgressBar
879
return read_weave(file(argv[2], 'rb'))
885
# at the moment, based on everything in the file
887
parents = map(int, argv[4:])
888
lines = sys.stdin.readlines()
889
ver = w.add(name, parents, lines)
890
write_weave(w, file(argv[2], 'wb'))
891
print 'added version %r %d' % (name, ver)
894
if os.path.exists(fn):
895
raise IOError("file exists")
897
write_weave(w, file(fn, 'wb'))
898
elif cmd == 'get': # get one version
900
sys.stdout.writelines(w.get_iter(int(argv[3])))
902
elif cmd == 'mash': # get composite
904
sys.stdout.writelines(w.mash_iter(map(int, argv[3:])))
906
elif cmd == 'annotate':
908
# newline is added to all lines regardless; too hard to get
909
# reasonable formatting otherwise
911
for origin, text in w.annotate(int(argv[3])):
912
text = text.rstrip('\r\n')
914
print ' | %s' % (text)
916
print '%5d | %s' % (origin, text)
930
print '%d versions ok' % w.numversions()
932
elif cmd == 'inclusions':
934
print ' '.join(map(str, w.inclusions([int(argv[3])])))
936
elif cmd == 'parents':
938
print ' '.join(map(str, w._parents[int(argv[3])]))
940
elif cmd == 'plan-merge':
942
for state, line in w.plan_merge(int(argv[3]), int(argv[4])):
944
print '%14s | %s' % (state, line),
948
p = w.plan_merge(int(argv[3]), int(argv[4]))
949
sys.stdout.writelines(w.weave_merge(p))
951
elif cmd == 'mash-merge':
957
v1, v2 = map(int, argv[3:5])
959
basis = w.inclusions([v1]).intersection(w.inclusions([v2]))
961
base_lines = list(w.mash_iter(basis))
962
a_lines = list(w.get(v1))
963
b_lines = list(w.get(v2))
965
from bzrlib.merge3 import Merge3
966
m3 = Merge3(base_lines, a_lines, b_lines)
968
name_a = 'version %d' % v1
969
name_b = 'version %d' % v2
970
sys.stdout.writelines(m3.merge_lines(name_a=name_a, name_b=name_b))
972
raise ValueError('unknown command %r' % cmd)
976
def profile_main(argv):
977
import tempfile, hotshot, hotshot.stats
979
prof_f = tempfile.NamedTemporaryFile()
981
prof = hotshot.Profile(prof_f.name)
983
ret = prof.runcall(main, argv)
986
stats = hotshot.stats.load(prof_f.name)
988
stats.sort_stats('cumulative')
989
## XXX: Might like to write to stderr or the trace file instead but
990
## print_stats seems hardcoded to stdout
991
stats.print_stats(20)
996
if __name__ == '__main__':
998
if '--profile' in sys.argv:
1000
args.remove('--profile')
1001
sys.exit(profile_main(args))
1003
sys.exit(main(sys.argv))
added
removed
bzrlib/weave.py
