Hello John, Tuesday, April 11, 2006, 2:43:49 AM, you wrote:
true. in any case, deepseq is not always a win.
don't forget that Andy don't plan to apply deepSeq to any expression. in his program, there is a LARGE datastructure with a couple of unevaluated thunks what may be simplified by call to deepSeq. your example is based exclusively on the syntax transformations of source code, i think that in his program the logic is so complex that such syntax transformations is entirely impossible anyway i think that the easisest way for Andy to get what he need is to write ghc-specific `deepSeq` implementation that should just walk unevaluated parts of datastructure and evaluate them all. as i understand, he don't need to evaluate arguments of partially applied functions - there is just no such beasts in his data -- Best regards, Bulat mailto:Bulat.Ziganshin@gmail.com