Re: Type casts

Nix> On Mon, 22 Jul 2002, Martin Buchholz uttered the following: >>Nix> On Sat, 20 Jul 2002, Martin Buchholz moaned: Nix> In many cases, GCC will notice things like Nix> (foo *) ((void *) bar) Nix> and elide the useless cast to `void *'. >> C is not Elisp. There is no "code" there to be elided. Nix> Quite so: but the compiler is permitted to *ignore* that cast to `void Nix> *' when it comes to determining what the expression is permitted to Nix> alias. I see what you mean now. Yes, the compiler is allowed to (and probably should) always treat (foo *) ((void *) bar) identically to (foo *) bar Deterring the compiler from making aliasing optimizations involves making many annoying code changes, typically involving the use of unions. My attempt to inhibit aliasing optimizations by accessing data using a type like the following typedef union { char c; void *p; } *dfc_aliasing_voidpp; works with gcc, but not with some other compilers. It is technically insufficient. This is why the DFC_ macros use memcpy.