bitset counting as a user defined function in postgresql 9.2? - Mailing list pgsql-hackers

As an initial attempt to try hands-on with postgresql, I've learned on user created functions and some other internals of postgresql. 

Few materials on postgresql indicate that the aggregate functions are slow due to some internal reasons (index covering, null etc). 

I'm looking forward to implement my own bit counting or counting mechanism (I've gone through some bit twiddling as well) via user defined functions (in C) and in cases I may need to create user defined data types as well.

Are there any possible scenarios where user defined count would perform better than the underlying implementation? Is it worth a try? 

And are there any better ways to evaluate, compare both? (other than EXPLAIN ANALYZIS)

Please guide.