On Tue, 2008-10-07 at 15:16 -0700, Raoul Duke wrote: > > Not really, because they require syntactic support. The issue isn't > > compile time or computational complexity of analysis. The issue is the > > need for syntax cruft. > > not sure i explained myself well :-) or i'm mis-interpreting your > note. i was meaning a nuance about when there is a "need for syntax > cruft": if you could make the use of manually written effect > annotations in the code optional for the programmer...
Unfortunately, things like noalloc and pure are non-local properties. You cannot (in general) look at a single procedure that calls other procedures and determine what the answer is. If we required whole-program compilation, this would not be a problem. When all source code is available the answers (at least in these two cases) can be inferred without user assistance. But when separate compilation is required, it becomes necessary to introduce annotations on functions so that the necessary dependency information can be represented. This isn't optional, because if *any* procedure is missing the annotations, and that procedure is called, the analysis fails for the entire program. shap _______________________________________________ bitc-dev mailing list [email protected] http://www.coyotos.org/mailman/listinfo/bitc-dev
