> -----Original Message-----
> From: [EMAIL PROTECTED] 
> [mailto:[EMAIL PROTECTED] On Behalf Of Rico Jansen
> 
> The solution, as Michiel says, probably lies in the relm of 
> full-text search engines. Like Verity, Excalibur, Lucene (the 
> last one is open source btw). I know that Excalibur has a 
> concept called overlap-iso which maps all accented characters 
> to their non-accented couter-part. All searches take place on 
> the non-accented indexes it creates in that situation. I 
> suspect a solution like that is better for the long run. 
> However it means quite a bit of work to solve, especially if 
> there is no support in the database for this stuff. However 
> it was taken into account in the query-project to support 
> external indexes. I don't know how much of that support has 
> been fleshed out.

Correct. Options depend largely on what the database supports. 
E.g. Excalibur (an Informix-extension) supports full-text-search
indices, that can be configured to ignore the differences between the
character-families as described by Henk. 

In order for MMBase to take advantage of such indices, the
StringSearchConstraint can be used. In order to enable these constraints
the functionality of one of the existing SqlHandler implementations
needs to be extendended, which can be done by implementing and
configuring a ChainedSqlHandler. 

Package
org.mmbase.storage.search.implementation.database.informix.excalibur
contains an example that does just that for the Excalibur-Informix
setup. 
This code has not yet been fully tested due to stability problems with
Excalibur, but it is expected that these issues will be adressed
sometime soon.

Rob van Maris
Technical Consultant

Quantiq
xmedia & communication solutions

Koninginneweg 11-13
1217 KP Hilversum
 
T    +31 (0)356257211
M    +31 (0)651444006
E    [EMAIL PROTECTED]
 


Reply via email to