Hi Mike, > Let's remove the .empty(), and let's switch to this "normal iterator" > semantics -- I think I had originally added that back when TermsEnum > didn't have a .term() method, ie, you had to keep the result returned > from calling .next() or .seek(). But that proved too much of hassle > (lots of places where we pass a TermsEnum down and the callee needs to > figure out what term it's on), so I've since added .term().
+1 > Actually TermEnum (the current API) already behaves like a normal > iterator, right? It's only FilteredTermEnum (and, FilteredTermsEnum) > that pre-next's itself on construction. So we should fix > FilteredTermsEnum to not pre-next, and to remove its empty() method? Not really, if you get the unitialized TermEnum from IR.terms(), it is like an iterator, but if you get it with IR.terms(Term), it is already placed on the first term. You have to then check, if it the current term is null (e.g. exhausted if you take the last term). You see this type of code everywhere in lucene, like in MTQ. In my opinion, it would be better here to always call next() first (and this was often done by users, then missing the first term). The new API does not have this problem, as you only get an terms iterator for a specific field that is per default always uninitialized and you have to call seek. But with trunk, that is not the case. > Uwe do you want to take a stab at it? Or, I can... let me know. I think, you know the code better, but I could find out what to do. I would like to have an IllegalStateException if you call docFreq() or term(), too. I think this would help people migrating from old to new code (and rewriting their do-while code to while(next)). Uwe --------------------------------------------------------------------- To unsubscribe, e-mail: java-dev-unsubscr...@lucene.apache.org For additional commands, e-mail: java-dev-h...@lucene.apache.org
