Doug, Chuck,
thanks for your feedback, proposals and explanations. The way to proceed seems quite clear to me now. Due to other obligations it will take probably about two to three weeks until I've implemented a new patch. I'll get back to you as soon as it's finished.
--Wolf
I was not able to follow the whole discussion while it was going on but I tried to catch up now. I did not look into Chuck´s and Wolf´s implementation. From the discussion I think what I propose is new but I am not completely sure. So sorry if I bother you with already discussed issues.
As far as I understand the solution you are currently working on is the one proposed by Doug. This implies
*) Rewriting the query in MultiSearcher making calls to Searchables. *) Creating a Weight in MultiSearcher *) Changing Searchables API so that the Weight is handed over as parameter to search methods
I also think this is the best way to fix this bug. However there may be a
way to implement this while avoiding to change the Weight and Searchable
API. The idea is to rewrite the query in MultiSearcher and while rewriting
compile the global idf into the query boosts. For this step calls to all Searchables are needed and the optimization already in this thread is
applicable. The search methods in Searchables then have to be told to not
use their local idf during search. This could e.g. be made by giving them
a special Similarity. It should be a subclass of the similarity specified for
the current search with idf() always returning 1. Similarity would have to be
serializable and Query would have to contain a member of type similarity.
Query´s getSimilarity(Searcher) could be implement to either return Searcher´s
Similarity or Query´s private similarity if not null.
This is just an idea that looks a little bit simpler to me on the first sight. I hope it makes sense.
Christoph
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
