On Sun, Sep 20, 2009 at 01:12:14PM -0400, David Roundy wrote:
I'm wondering how one would go about adding a Num instance to CLong?
Ah. types like this have some (annoying) history behind them I have been trying to rewrite, slowly but surely. Basically, when I first started writing jhc, I figured I'd start with a simple haskell 98 compiler then add extensions (like unboxed types) later. Which turned out to be a big mistake. unboxed types should have been there from day one. they made so many things way simpler. The whole PrimitiveOperators generated code mess is due to me having to have the compiler internally generate all these instances for Int,Int32,Word,etc... instances which would have been perfectly expressible in haskell had I had unboxed types. So, originally, primitive types were implemented as so
data Int
and the compiler had some magic to add instances and a representation to them. which was fairly complicated, because in order to have the compiler add a 'Num' instance, it meant 'Num' had to have special meaning to the compiler, so there was a cascade effect of adding way more library specific code to the compiler than I liked. The proper way to do such types nowadays would be
data Int = Int Bits32_
or
newtype Int = Int Int32 deriving(...)
then manually create all the instaces. (well, I have m4 macros to make this easy for many instances) So. Long answer short, types like CLong need to be rewritten to be concrete types. like Float and Double. If you just want to get something to work, you can change CLong to be
newtype CLong = CLong IntMax
and then you can get at the underlying IntMax which has the right instances defined. John -- John Meacham - ⑆repetae.net⑆john⑈ - http://notanumber.net/