Re: [Koha-zebra] zebraidx: corrupted double-linked list, when ICU enabled

[LAURENT Henri-Damien]

Le 09/06/2011 08:09, Frédéric Demians a écrit :
> Hello,
> 
> When indexing a Koha set of biblio records containing Arabic characters,
> zebraidx crashes with this error message:
> 
> *** glibc detected *** zebraidx: corrupted double-linked list:
> 0x00000000017feb20 ***
> 
> If the same set of records are indexed by zebraidx one by one, it works.
> 
> If ICU is disabled is also works.
> 
> So this bug seems to be Zebra + ICU related.
> 
> I suspect there are 'invalid' Arabic characters somewhere but I can't
> see anything special.
> 
> Does it sound familiar to anybody? Does anyone find a solution for this
> bug?
> 
> Thanks.
Hi Frederic
I would personnaly export as iso2709 if possible and use yaz-marcdump to
split the file into smaller pieces... and index the smaller parts
in order to get the data indexed... And track the culprit.... You will
then have the biblios which make indexing fail.... (I think it is a list
of biblios...)
paul posted a rebuild_zebra_sliced.sh to export smaller parts of biblios
and index.
You may take ideas there.
Hope that helps.
-- 
Henri-Damien LAURENT
BibLibre