Jonathan S. Shapiro wrote: >The basis of their work is automated region inference. Region inference >isn't going to work for the applications we are really concerned about, >because the inference engine will conclude (correctly) that the region >remains live nearly everywhere that we might call free(). > >So: this is definitely a useful auxiliary technique for BitC, but it >doesn't address the core issues of memory (de)allocation that we need to >solve. >
I see. I wonder, is it possible to have 'compilation profiles/environments' such that you can restrict compilation strategy to a definable set of techniques? For instance, let's suppose I wanted a program to be statically verifiable under the above region inference technique, could I specify this somehow? I imagine this sort of facility might be useful. I hope I'm being clear; I'm struggling a little for the words to express precisely what I mean. Also, have you looked at LLVM as a foundation for bitc, or do you have a some other foundation you're using? I'm curious what others think of it. Sandro _______________________________________________ bitc-dev mailing list [email protected] http://www.coyotos.org/mailman/listinfo/bitc-dev
