Last modified: 2008-05-02 03:15:37 UTC
Wikipedia is occasionally (very rarely?) showing diffs that are inconsistent with the article content. Upon purging the diff cache, everything looks OK again. Unfortunately, the cache of this incident has been purged already, so it cannot be reproduced any more. Screenshot at the time of my particular incident: http://juffo.org/stuff/mediawiki-wtf.png Discussion on #wikimedia-tech, confirming other people also seeing this, and suggesting that it's not an isolated incident: [08 21:24'31] <intgr> WTF, see this edit: http://en.wikipedia.org/w/index.php?title=RTT&diff=prev&oldid=121266608 [08 21:24'52] <intgr> It appears as if the person added three links to the article, but the preview only shows one at the end. [08 21:27'41] <intgr> Can anyone confirm that it doesn't just appear that way to me? [08 21:27'41] <Duesentrieb> intgr: that's... scarry. [08 21:28'37] <Duesentrieb> domas: yea, right [08 21:28'54] <Duesentrieb> intgr: yes, me. [08 21:28'55] <Splarka> intgr: that does look like a funky diff [08 21:29'08] <Duesentrieb> domas: look at intgr's diff link [08 21:29'48] <domas> whats in it? [08 21:29'57] <Simetrical> domas: Diff doesn't match the resultant text. [08 21:30'04] <Duesentrieb> domas: the diff shows bogus changes [08 21:30'11] <domas> hehe [08 21:30'13] <domas> interesting [08 21:32'30] <Splarka> another funky diff: http://en.wikipedia.org/w/index.php?title=Ghostface_Killah&diff=prev&oldid=120004683 (reported in VP/T April 6th) [08 21:32'39] <domas> refresh [08 21:32'52] <domas> hehe [08 21:33'13] <domas> 'unknown exception' [08 21:33'24] <domas> ok, I guess I destroyed traces of previous diff corruption [08 21:33'43] <Splarka> and another one from 3 days ago: http://en.wikipedia.org/wiki/Wikipedia:Help_desk#Really_confusing [08 21:34'34] <Splarka> (but that one seems fix-ed now) [08 21:34'39] <Duesentrieb> domas: tnx for fixing it [08 21:34'50] <domas> its not fixing [08 21:34'52] <intgr> Sounds like duct tape. [08 21:34'54] <domas> I just purged the diff cache [08 21:35'09] <Duesentrieb> domas: well, as long as the funky stuff doesn't get in there again... [08 21:35'22] <intgr> Well, it'll happen again sooner or later. [08 21:35'28] <Splarka> the quicker differ cacher purger
Could be a bug or inconsistency in the wikidiff2 extension... All live Apaches seem to have consistent copies (all 32-bit same, all 64-bit same, and all same date).
Another instance of this bug: http://en.wikipedia.org/w/index.php?title=User_talk:Boothy443&diff=prev&oldid=134912276 If you try to "undo" it, the change turns out to be completely different: http://en.wikipedia.org/w/index.php?title=User_talk:Boothy443&curid=8879613&diff=134936206&oldid=134912276 I took screenshots of this just in case someone resets the diff cache again: http://juffo.org/stuff/wikibug_diff1.png http://juffo.org/stuff/wikibug_diff2.png
As of r24326, users can use "action=purge" on a difference page to force the cache to be skipped and regenerated (if applicable) for that request.
*** Bug 11352 has been marked as a duplicate of this bug. ***
*** Bug 9901 has been marked as a duplicate of this bug. ***
*** Bug 13129 has been marked as a duplicate of this bug. ***
Changing component to wikidiff2, there's really no other explanation.
*** Bug 13704 has been marked as a duplicate of this bug. ***
*** Bug 13575 has been marked as a duplicate of this bug. ***
Looks like I have seen this bug too and reported it as a comment under Bug 10511 where other folks are also reporting it. Then I noticed Bug 9112 and if that description is accurate then there may be a very serious problem.
*** Bug 13851 has been marked as a duplicate of this bug. ***
Have hopefully resolved this with fresh recompile. Once we got some logging in, the buggy output seemed to be limited to Fedora Core 4 machines; on the assumption that an FC3-compiled binary (which worked on FC3 machines) was misbehaving on FC4, I recompiled on FC4 and haven't had any logged failures since.
*** Bug 10000 has been marked as a duplicate of this bug. ***
*** Bug 2305 has been marked as a duplicate of this bug. ***
*** Bug 10511 has been marked as a duplicate of this bug. ***