I was always unhappy because I couldn't search CJK characters well. So I decided to patch it myself. I looked at source code. Surprisingly CJK search was already implemented beautifully. Asian "words" are already searched as phrases. But it didn't work well.
I realized that preg_match_all NEVER matches asian characters at all. The following is code line 337--341 in fulltext.php (2007-06-26).
337 // asian "words" need to be searched as phrases
338 if(@preg_match_all('/('.IDX_ASIAN.'+)/u',$w,$matches)){
339 $q['phrases'] = array_merge($q['phrases'],$matches[1]);
340
341 }
IDX_ASIAN was not defined anywhere but IDX_ASIAN1, IDX_ASIAN2, IDX_ASIAN3 are!
So I upload a simple patch for this. I think defining IDX_ASIAN will be the best solution. Applying this patch it works as the developer originally intended. The patch file patches dokuwiki/inc/indexer.php.