On Sat, Nov 17, 2007 at 04:31:33PM +0000, Andrew Coppin wrote: >> Both. A curious feature of the STG machine is that constructor thunks >> and evaluated data are represented identically in memory. > > Ooo... As per the Lambdacats "Boxed cat has a uniform representation"? > > Well, presumably the guys who designed STG did it this way for a really > good reason, and they know far more than me, so... ;-)
The STG-machine was brilliant when it was designed, but times have changed. In particular, indirect jumps are no longer cheap. Pointer tagging has allowed STG to hobble into the 21st century, but really the air is ripe for a new abstract machine. Stefan
signature.asc
Description: Digital signature
_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
