On 2010-10-05 <23:27:33>, Hans Hagen wrote:
> On 5-10-2010 11:17, Philipp Gesang wrote:
> 
> >I guess there is some testing going on in order to determine
> >whether to proceed with the current entry or switch to the next
> >one. The position is the same, however the comparison with the
> >last item fails and a new one is created instead. (Only
> >guessing.)
> 
> it's a sequence of tests per comparison, like
> 
> Polyneikes
>     polyneikes % lowercased
>     polyneikes % shapes

I assume by “shapes” you mean the base symbol (all diacritics
stripped).

>     Polyneikes % unicode
> 
> Πολυνείκης
>     Πολυνείκης % lowercased
>     polyneikes % shapes
>     Πολυνείκης % unicode
> 
> casing and shapes depends on the mapping vectors and the order can
> be influenced, you can see this in action with
> 
> \enabletrackers[sorters.tests]

Bingo! The tracker instantly revealed a really nasty flaw in the
German standard transcription for Greek: “υ” is normally
converted to Latin “y”, but is retained as Latin “u” in
diphthongs like “ευ” and “ηυ”. So with the sorting definition I
posted I get amongst the results:

sorters         > Kapaneys > Kapaneus

because all “υ” are lazily mapped to “y”. Thus, for those
occasional three words per book, determining the sorting position
by hand (e.g. “\index[Kapaneus]{Καπανεύς}”) might be less prone
to error.

Thanks for the hint and sorry for posting a non-solution,

Philipp


-- 
()  ascii ribbon campaign - against html e-mail
/\  www.asciiribbon.org   - against proprietary attachments