Olly Betts
2009-Sep-03 00:43 UTC
[Xapian-devel] [Xapian-commits] 13440: trunk/xapian-core/ trunk/xapian-core/matcher/ trunk/xapian-core/tests/
On Thu, Sep 03, 2009 at 01:32:48AM +0100, richard wrote:> matcher/andmaybepostlist.cc: For AndMaybe, if the RHS has a > maximum possible weight of 0, we now decay to just the LHS. > tests/Makefile.am,tests/api_queryopt.cc: Test coverage that this > optimisation doesn't break anything.I can see two problems with this optimisation: Firstly, it means that count_matching_subqs() will return one less when the RHS would have matched, giving lower than expected percentage weights in some cases. Secondly, if under OP_SYNONYM, the wdf of the RHS will get ignored. Cheers, Olly