Last modified: 2014-05-02 11:48:56 UTC
Created attachment 15243 [details] screenshot of massive excerpt I searched for "Flow browser" and got a few very large excerpts. See screenshot of a 15,901 character excerpt. This link includes the screenshotted result, as of today: https://www.mediawiki.org/w/index.php?title=Special:Search&limit=10&offset=27&profile=default&search=flow+browser A few of the other results, were also larger than the expected ~100 characters.
Thanks for filing this. I'm almost 100% sure this is caused by a new component I just cut mediawiki.org over to. I'll get a solution for this soon.
// Now we have to pick a start and end offset, but there are // actually four cases that can happen given the above for start // offset and four for end. I'll show the start offset here // because the end is just the mirror image: // // Case 1: // --------+------[-------+-------]--+---------------------- // min expand max // Case 2: // ----[---+-------+--------+------------------------------- // min expand max // Case 3: // --------+---[----+-------+---]--------------------------- // min expand max // Case 4: // --------+-------+---------+------------------------------ // expand min max // // Case 1 is "normal", there are no obstructions and we pick // the boundary by looking from expand to [, the max scan, and if // that doesn't find anything looking from expand to the ], and if // that doesn't find anything defaulting to expand. // // Case 2 is almost normal. We look from expand to min but if that // doesn't find anything we deem min a valid boundary. Min is // generally the beginning of the source or the end of the last // segment and therefore a valid boundary. The case where expand // is right on top of min is pretty much a variant of this case. // // Case 3 is like case 2 but in reverse. We look to [ as in case // 1 and if we don't find anything we look to max. If that doesn't // find anything then we declare max a valid boundary. Max is // generally the beginning of the first hit, so very likely a valid // boundary. // // Case 4 is different. We'd like to expand past min which isn't // allowed so instead we deem min the boundary and try to shift the // whole segment forward some to make up for it.
https://gerrit.wikimedia.org/r/#/c/130615/
Deployed the fix yesterday. Looks to be fixed now.
