On Fri, Apr 22, 2011 at 12:55 PM, Stephen Tetley
There isn't much value to automatic memoization as people have already pointed out on SO - if you believe otherwise, you're probably better off proving a case on paper first before attempting to implement it in GHC.
Hi Stephen, I think there are already many many examples where memoization does indeed give much better performance (very simple and common one is the Fibonacci sequence). So the prove was already done. I am not sure what to prove otherwise. Demonstrating that it does indeed perform well when you apply it globally to some real-world application is not really something you prove on a paper. You do the actual benchmarking instead. I.e., you implement the memoization in GHC and benchmark some programs with it. Implementing it shouldn't also be complicated -- I already presented a 60 line Python implementation which is trivial to be applied globally and took 10 minutes to be implemented. I'm not sure though how complicated it would be to add that to GHC. I have no idea where to start. Regards, Albert