« back to all changes in this revision
Viewing changes to bzrlib/weave.py
- browse files at revision 879
- compare with another revision
- download diff
- download tarball
- view history from revision 879
- Committer: Martin Pool
- Date: 2005-07-11 03:12:34 UTC
- Revision ID: mbp@sourcefrog.net-20050711031233-e5cc726bc78362e1
- fix typo
- files added:
- bzrlib/hashcache.py
- bzrlib/selftest
- effbot
- effbot/org
- plugins
- plugins/changeset
- tools
- files removed:
- bzrlib/statcache.py
- files renamed:
- bzrlib/selftest.py => bzrlib/selftest/__init__.py
- bzrlib/whitebox.py => bzrlib/selftest/whitebox.py
- files modified:
- .bzrignore
added
removed
bzrlib/weave.py
1
#! /usr/bin/python
2
3
# Copyright (C) 2005 Canonical Ltd
4
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.
9
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.
14
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
18
19
# Author: Martin Pool <mbp@canonical.com>
20
21
22
"""Weave - storage of related text file versions"""
23
24
# TODO: Perhaps have copy method for Weave instances?
25
26
# XXX: If we do weaves this way, will a merge still behave the same
27
# way if it's done in a different order? That's a pretty desirable
28
# property.
29
30
# TODO: How to write these to disk? One option is cPickle, which
31
# would be fast but less friendly to C, and perhaps not portable. Another is
32
33
# TODO: Nothing here so far assumes the lines are really \n newlines,
34
# rather than being split up in some other way. We could accomodate
35
# binaries, perhaps by naively splitting on \n or perhaps using
36
# something like a rolling checksum.
37
38
# TODO: Perhaps track SHA-1 in the header for protection? This would
39
# be redundant with it being stored in the inventory, but perhaps
40
# usefully so?
41
42
# TODO: Track version names as well as indexes.
43
44
# TODO: Probably do transitive expansion when specifying parents?
45
46
# TODO: Separate out some code to read and write weaves.
47
48
# TODO: End marker for each version so we can stop reading?
49
50
# TODO: Check that no insertion occurs inside a deletion that was
51
# active in the version of the insertion.
52
53
# TODO: Perhaps a special slower check() method that verifies more
54
# nesting constraints and the MD5 of each version?
55
56
57
58
try:
59
set
60
frozenset
61
except NameError:
62
from sets import Set, ImmutableSet
63
set = Set
64
frozenset = ImmutableSet
65
del Set, ImmutableSet
66
67
68
class WeaveError(Exception):
69
"""Exception in processing weave"""
70
71
72
class WeaveFormatError(WeaveError):
73
"""Weave invariant violated"""
74
75
76
class Weave(object):
77
"""weave - versioned text file storage.
78
79
A Weave manages versions of line-based text files, keeping track
80
of the originating version for each line.
81
82
To clients the "lines" of the file are represented as a list of strings.
83
These strings will typically have terminal newline characters, but
84
this is not required. In particular files commonly do not have a newline
85
at the end of the file.
86
87
Texts can be identified in either of two ways:
88
89
* a nonnegative index number.
90
91
* a version-id string.
92
93
Typically the index number will be valid only inside this weave and
94
the version-id is used to reference it in the larger world.
95
96
The weave is represented as a list mixing edit instructions and
97
literal text. Each entry in _l can be either a string (or
98
unicode), or a tuple. If a string, it means that the given line
99
should be output in the currently active revisions.
100
101
If a tuple, it gives a processing instruction saying in which
102
revisions the enclosed lines are active. The tuple has the form
103
(instruction, version).
104
105
The instruction can be '{' or '}' for an insertion block, and '['
106
and ']' for a deletion block respectively. The version is the
107
integer version index. There is no replace operator, only deletes
108
and inserts.
109
110
Constraints/notes:
111
112
* A later version can delete lines that were introduced by any
113
number of ancestor versions; this implies that deletion
114
instructions can span insertion blocks without regard to the
115
insertion block's nesting.
116
117
* Similarly, deletions need not be properly nested with regard to
118
each other, because they might have been generated by
119
independent revisions.
120
121
* Insertions are always made by inserting a new bracketed block
122
into a single point in the previous weave. This implies they
123
can nest but not overlap, and the nesting must always have later
124
insertions on the inside.
125
126
* It doesn't seem very useful to have an active insertion
127
inside an inactive insertion, but it might happen.
128
129
* Therefore, all instructions are always"considered"; that
130
is passed onto and off the stack. An outer inactive block
131
doesn't disable an inner block.
132
133
* Lines are enabled if the most recent enclosing insertion is
134
active and none of the enclosing deletions are active.
135
136
* There is no point having a deletion directly inside its own
137
insertion; you might as well just not write it. And there
138
should be no way to get an earlier version deleting a later
139
version.
140
141
_l
142
Text of the weave.
143
144
_v
145
List of versions, indexed by index number.
146
147
For each version we store the set (included_versions), which
148
lists the previous versions also considered active; the
149
versions included in those versions are included transitively.
150
So new versions created from nothing list []; most versions
151
have a single entry; some have more.
152
153
_sha1s
154
List of hex SHA-1 of each version, or None if not recorded.
155
"""
156
def __init__(self):
157
self._l = []
158
self._v = []
159
self._sha1s = []
160
161
162
def __eq__(self, other):
163
if not isinstance(other, Weave):
164
return False
165
return self._v == other._v \
166
and self._l == other._l
167
168
169
def __ne__(self, other):
170
return not self.__eq__(other)
171
172
173
def add(self, parents, text):
174
"""Add a single text on top of the weave.
175
176
Returns the index number of the newly added version.
177
178
parents
179
List or set of parent version numbers. This must normally include
180
the parents and the parent's parents, or wierd things might happen.
181
182
text
183
Sequence of lines to be added in the new version."""
184
## self._check_versions(parents)
185
## self._check_lines(text)
186
idx = len(self._v)
187
188
import sha
189
s = sha.new()
190
for l in text:
191
s.update(l)
192
sha1 = s.hexdigest()
193
del s
194
195
if parents:
196
delta = self._delta(self.inclusions(parents), text)
197
198
# offset gives the number of lines that have been inserted
199
# into the weave up to the current point; if the original edit instruction
200
# says to change line A then we actually change (A+offset)
201
offset = 0
202
203
for i1, i2, newlines in delta:
204
assert 0 <= i1
205
assert i1 <= i2
206
assert i2 <= len(self._l)
207
208
# the deletion and insertion are handled separately.
209
# first delete the region.
210
if i1 != i2:
211
self._l.insert(i1+offset, ('[', idx))
212
self._l.insert(i2+offset+1, (']', idx))
213
offset += 2
214
# is this OK???
215
216
if newlines:
217
# there may have been a deletion spanning up to
218
# i2; we want to insert after this region to make sure
219
# we don't destroy ourselves
220
i = i2 + offset
221
self._l[i:i] = [('{', idx)] \
222
+ newlines \
223
+ [('}', idx)]
224
offset += 2 + len(newlines)
225
226
self._addversion(parents)
227
else:
228
# special case; adding with no parents revision; can do this
229
# more quickly by just appending unconditionally
230
self._l.append(('{', idx))
231
self._l += text
232
self._l.append(('}', idx))
233
234
self._addversion(None)
235
236
self._sha1s.append(sha1)
237
238
return idx
239
240
241
def inclusions(self, versions):
242
"""Expand out everything included by versions."""
243
i = set(versions)
244
for v in versions:
245
try:
246
i.update(self._v[v])
247
except IndexError:
248
raise ValueError("version %d not present in weave" % v)
249
return i
250
251
252
def _addversion(self, parents):
253
if parents:
254
self._v.append(frozenset(parents))
255
else:
256
self._v.append(frozenset())
257
258
259
def _check_lines(self, text):
260
if not isinstance(text, list):
261
raise ValueError("text should be a list, not %s" % type(text))
262
263
for l in text:
264
if not isinstance(l, basestring):
265
raise ValueError("text line should be a string or unicode, not %s"
266
% type(l))
267
268
269
270
def _check_versions(self, indexes):
271
"""Check everything in the sequence of indexes is valid"""
272
for i in indexes:
273
try:
274
self._v[i]
275
except IndexError:
276
raise IndexError("invalid version number %r" % i)
277
278
279
def annotate(self, index):
280
return list(self.annotate_iter(index))
281
282
283
def annotate_iter(self, version):
284
"""Yield list of (index-id, line) pairs for the specified version.
285
286
The index indicates when the line originated in the weave."""
287
included = self.inclusions([version])
288
for origin, lineno, text in self._extract(included):
289
yield origin, text
290
291
292
def _extract(self, included):
293
"""Yield annotation of lines in included set.
294
295
Yields a sequence of tuples (origin, lineno, text), where
296
origin is the origin version, lineno the index in the weave,
297
and text the text of the line.
298
299
The set typically but not necessarily corresponds to a version.
300
"""
301
istack = [] # versions for which an insertion block is current
302
303
dset = set() # versions for which a deletion block is current
304
305
isactive = None
306
307
lineno = 0 # line of weave, 0-based
308
309
# TODO: Probably only need to put included revisions in the istack
310
311
# TODO: Could split this into two functions, one that updates
312
# the stack and the other that processes the results -- but
313
# I'm not sure it's really needed.
314
315
# TODO: In fact, I think we only need to store the *count* of
316
# active insertions and deletions, and we can maintain that by
317
# just by just counting as we go along.
318
319
WFE = WeaveFormatError
320
321
for l in self._l:
322
if isinstance(l, tuple):
323
isactive = None # recalculate
324
c, v = l
325
if c == '{':
326
if istack and (istack[-1] >= v):
327
raise WFE("improperly nested insertions %d>=%d on line %d"
328
% (istack[-1], v, lineno))
329
istack.append(v)
330
elif c == '}':
331
try:
332
oldv = istack.pop()
333
except IndexError:
334
raise WFE("unmatched close of insertion %d on line %d"
335
% (v, lineno))
336
if oldv != v:
337
raise WFE("mismatched close of insertion %d!=%d on line %d"
338
% (oldv, v, lineno))
339
elif c == '[':
340
# block deleted in v
341
if v in dset:
342
raise WFE("repeated deletion marker for version %d on line %d"
343
% (v, lineno))
344
if istack:
345
if istack[-1] == v:
346
raise WFE("version %d deletes own text on line %d"
347
% (v, lineno))
348
# XXX
349
dset.add(v)
350
elif c == ']':
351
if v in dset:
352
dset.remove(v)
353
else:
354
raise WFE("unmatched close of deletion %d on line %d"
355
% (v, lineno))
356
else:
357
raise WFE("invalid processing instruction %r on line %d"
358
% (l, lineno))
359
else:
360
assert isinstance(l, basestring)
361
if not istack:
362
raise WFE("literal at top level on line %d"
363
% lineno)
364
if isactive == None:
365
isactive = (istack[-1] in included) \
366
and not included.intersection(dset)
367
if isactive:
368
origin = istack[-1]
369
yield origin, lineno, l
370
lineno += 1
371
372
if istack:
373
raise WFE("unclosed insertion blocks at end of weave",
374
istack)
375
if dset:
376
raise WFE("unclosed deletion blocks at end of weave",
377
dset)
378
379
380
def get_iter(self, version):
381
"""Yield lines for the specified version."""
382
for origin, lineno, line in self._extract(self.inclusions([version])):
383
yield line
384
385
386
def get(self, index):
387
return list(self.get_iter(index))
388
389
390
def mash_iter(self, included):
391
"""Return composed version of multiple included versions."""
392
included = frozenset(included)
393
for origin, lineno, text in self._extract(included):
394
yield text
395
396
397
def dump(self, to_file):
398
from pprint import pprint
399
print >>to_file, "Weave._l = ",
400
pprint(self._l, to_file)
401
print >>to_file, "Weave._v = ",
402
pprint(self._v, to_file)
403
404
405
406
def numversions(self):
407
l = len(self._v)
408
assert l == len(self._sha1s)
409
return l
410
411
412
def check(self):
413
# check no circular inclusions
414
for version in range(self.numversions()):
415
inclusions = list(self._v[version])
416
if inclusions:
417
inclusions.sort()
418
if inclusions[-1] >= version:
419
raise WeaveFormatError("invalid included version %d for index %d"
420
% (inclusions[-1], version))
421
422
# try extracting all versions; this is a bit slow and parallel
423
# extraction could be used
424
import sha
425
for version in range(self.numversions()):
426
s = sha.new()
427
for l in self.get_iter(version):
428
s.update(l)
429
hd = s.hexdigest()
430
expected = self._sha1s[version]
431
if hd != expected:
432
raise WeaveError("mismatched sha1 for version %d; "
433
"got %s, expected %s"
434
% (version, hd, expected))
435
436
437
438
def merge(self, merge_versions):
439
"""Automerge and mark conflicts between versions.
440
441
This returns a sequence, each entry describing alternatives
442
for a chunk of the file. Each of the alternatives is given as
443
a list of lines.
444
445
If there is a chunk of the file where there's no diagreement,
446
only one alternative is given.
447
"""
448
449
# approach: find the included versions common to all the
450
# merged versions
451
raise NotImplementedError()
452
453
454
455
def _delta(self, included, lines):
456
"""Return changes from basis to new revision.
457
458
The old text for comparison is the union of included revisions.
459
460
This is used in inserting a new text.
461
462
Delta is returned as a sequence of
463
(weave1, weave2, newlines).
464
465
This indicates that weave1:weave2 of the old weave should be
466
replaced by the sequence of lines in newlines. Note that
467
these line numbers are positions in the total weave and don't
468
correspond to the lines in any extracted version, or even the
469
extracted union of included versions.
470
471
If line1=line2, this is a pure insert; if newlines=[] this is a
472
pure delete. (Similar to difflib.)
473
"""
474
# basis a list of (origin, lineno, line)
475
basis_lineno = []
476
basis_lines = []
477
for origin, lineno, line in self._extract(included):
478
basis_lineno.append(lineno)
479
basis_lines.append(line)
480
481
# add a sentinal, because we can also match against the final line
482
basis_lineno.append(len(self._l))
483
484
# XXX: which line of the weave should we really consider
485
# matches the end of the file? the current code says it's the
486
# last line of the weave?
487
488
from difflib import SequenceMatcher
489
s = SequenceMatcher(None, basis_lines, lines)
490
491
# TODO: Perhaps return line numbers from composed weave as well?
492
493
for tag, i1, i2, j1, j2 in s.get_opcodes():
494
##print tag, i1, i2, j1, j2
495
496
if tag == 'equal':
497
continue
498
499
# i1,i2 are given in offsets within basis_lines; we need to map them
500
# back to offsets within the entire weave
501
real_i1 = basis_lineno[i1]
502
real_i2 = basis_lineno[i2]
503
504
assert 0 <= j1
505
assert j1 <= j2
506
assert j2 <= len(lines)
507
508
yield real_i1, real_i2, lines[j1:j2]
509
510
511
512
def weave_info(filename, out):
513
"""Show some text information about the weave."""
514
from weavefile import read_weave
515
wf = file(filename, 'rb')
516
w = read_weave(wf)
517
# FIXME: doesn't work on pipes
518
weave_size = wf.tell()
519
print >>out, "weave file size %d bytes" % weave_size
520
print >>out, "weave contains %d versions" % len(w._v)
521
522
total = 0
523
print '%6s %6s %8s %40s %20s' % ('ver', 'lines', 'bytes', 'sha1', 'parents')
524
for i in (6, 6, 8, 40, 20):
525
print '-' * i,
526
print
527
for i in range(len(w._v)):
528
text = w.get(i)
529
lines = len(text)
530
bytes = sum((len(a) for a in text))
531
sha1 = w._sha1s[i]
532
print '%6d %6d %8d %40s' % (i, lines, bytes, sha1),
533
print ', '.join(map(str, w._v[i]))
534
total += bytes
535
536
print >>out, "versions total %d bytes" % total
537
print >>out, "compression ratio %.3f" % (float(total)/float(weave_size))
538
539
540
def usage():
541
print """bzr weave tool
542
543
Experimental tool for weave algorithm.
544
545
usage:
546
weave init WEAVEFILE
547
Create an empty weave file
548
weave get WEAVEFILE VERSION
549
Write out specified version.
550
weave check WEAVEFILE
551
Check consistency of all versions.
552
weave info WEAVEFILE
553
Display table of contents.
554
weave add WEAVEFILE [BASE...] < NEWTEXT
555
Add NEWTEXT, with specified parent versions.
556
weave annotate WEAVEFILE VERSION
557
Display origin of each line.
558
weave mash WEAVEFILE VERSION...
559
Display composite of all selected versions.
560
weave merge WEAVEFILE VERSION1 VERSION2 > OUT
561
Auto-merge two versions and display conflicts.
562
563
example:
564
565
% weave init foo.weave
566
% vi foo.txt
567
% weave add foo.weave < foo.txt
568
added version 0
569
570
(create updated version)
571
% vi foo.txt
572
% weave get foo.weave 0 | diff -u - foo.txt
573
% weave add foo.weave 0 < foo.txt
574
added version 1
575
576
% weave get foo.weave 0 > foo.txt (create forked version)
577
% vi foo.txt
578
% weave add foo.weave 0 < foo.txt
579
added version 2
580
581
% weave merge foo.weave 1 2 > foo.txt (merge them)
582
% vi foo.txt (resolve conflicts)
583
% weave add foo.weave 1 2 < foo.txt (commit merged version)
584
585
"""
586
587
588
589
def main(argv):
590
import sys
591
import os
592
from weavefile import write_weave, read_weave
593
cmd = argv[1]
594
595
def readit():
596
return read_weave(file(argv[2], 'rb'))
597
598
if cmd == 'help':
599
usage()
600
elif cmd == 'add':
601
w = readit()
602
# at the moment, based on everything in the file
603
parents = map(int, argv[3:])
604
lines = sys.stdin.readlines()
605
ver = w.add(parents, lines)
606
write_weave(w, file(argv[2], 'wb'))
607
print 'added version %d' % ver
608
elif cmd == 'init':
609
fn = argv[2]
610
if os.path.exists(fn):
611
raise IOError("file exists")
612
w = Weave()
613
write_weave(w, file(fn, 'wb'))
614
elif cmd == 'get': # get one version
615
w = readit()
616
sys.stdout.writelines(w.get_iter(int(argv[3])))
617
618
elif cmd == 'mash': # get composite
619
w = readit()
620
sys.stdout.writelines(w.mash_iter(map(int, argv[3:])))
621
622
elif cmd == 'annotate':
623
w = readit()
624
# newline is added to all lines regardless; too hard to get
625
# reasonable formatting otherwise
626
lasto = None
627
for origin, text in w.annotate(int(argv[3])):
628
text = text.rstrip('\r\n')
629
if origin == lasto:
630
print ' | %s' % (text)
631
else:
632
print '%5d | %s' % (origin, text)
633
lasto = origin
634
635
elif cmd == 'info':
636
weave_info(argv[2], sys.stdout)
637
638
elif cmd == 'check':
639
w = readit()
640
w.check()
641
642
elif cmd == 'merge':
643
if len(argv) != 5:
644
usage()
645
return 1
646
647
w = readit()
648
v1, v2 = map(int, argv[3:5])
649
650
basis = w.inclusions([v1]).intersection(w.inclusions([v2]))
651
652
base_lines = list(w.mash_iter(basis))
653
a_lines = list(w.get(v1))
654
b_lines = list(w.get(v2))
655
656
from bzrlib.merge3 import Merge3
657
m3 = Merge3(base_lines, a_lines, b_lines)
658
659
name_a = 'version %d' % v1
660
name_b = 'version %d' % v2
661
sys.stdout.writelines(m3.merge_lines(name_a=name_a, name_b=name_b))
662
else:
663
raise ValueError('unknown command %r' % cmd)
664
665
666
if __name__ == '__main__':
667
import sys
668
sys.exit(main(sys.argv))
Loggerhead is a web-based interface for Breezy
Version: 2.0.1