The C++ FE holds class members in two vectors, a vector of overloaded methods
and a vector of fields and other stuff. This is bad
1) These vectors are unsorted until the class is complete
2) When sorted they have O(log N) performance, and N can get resonably big these
3) You have to look in two places
4) It won't play nice when we get using declarations right -- you can't tell
at parse time whether a using declaration from a dependent base will be a
We should have one hash table per class to hold everything.
12944 is for generic namelookup problems.