On Tue, 2008-10-07 at 19:09 -0400, Sandro Magi wrote: > Well if the two options are a) specify effect variables on all source, > or b) have the analysis in a separate tool, why not consider some middle > ground, like the ability to specify an effect signature separately from > the type signature.
This is not the proposal. We will do "pure" effects within the language. The question here only concerns what to do about NOALLOC. As to effect signatures, see my previous response about why this is "all or nothing". > The compiler obviously needs both to call f, but I'm assuming that the > effects can be fully inferred so most developers need not bother with > them unless they're interested in declaring some property. Okay. Now hiding in here you said something interesting: the compiler can infer a lot of this stuff. Perhaps one answer is to let the compiler do the inference at compile time and store the result into the object file somehow. The result can then be composed and checked at link time. Technically, this will work. The problem with it is that it defers detection of effect violations to link time. shap _______________________________________________ bitc-dev mailing list bitc-dev@coyotos.org http://www.coyotos.org/mailman/listinfo/bitc-dev
