On Thu, Aug 31, 2006 at 02:29:04AM +0200, Roland Mainz wrote: > BTW: Would it be possible to get the following two macros (and derivates > for |unsigned long|) into OS/Net ? IMO they should provide a quite clean > way to do such conversions and would provide an easy way to search for > such stuff: > -- snip -- > #define UTS_LONG_TO_PTR(l) ((void*)((char*)0 + (l))) > #define UTS_PTR_TO_LONG(ptr) ((long)((char *)(ptr) - (char *0))) > -- snip --
This doesn't solve the problem; the interesting case is a 32-bit integer casted to a pointer (or a 64-bit integer casted to a 32-bit pointer; the exact combinations that cause problems depend on how you're compiling). You could do something like #define UTS_TO_POINTER(x) ((void *)(uintptr_t)(x)) But that saves you very little, and also implies that this cast is invariably safe and correct, which it is not. If the argument to the macro is actually a signed type, sign extension will be performed by the cast to uintptr_t, which may or may not be desired - that is, the original variable may be signed because it really is signed or it may be signed because it comes from some other type that just happens to be defined that way. Basically, you have to know whether you want sign- or zero-extension and cast appropriately. Going the other way, you may want to cast the pointer to a signed or unsigned 32- or 64-bit quantity. Of course, more macros could be written to address every possibility, or a single pair of extremely complex macros using __typeof() could be written, but in neither case would the macros do much to improve readability, save typing, or reduce the incidence of programmer error. If you want to solve this problem in a useful way, change the compiler. -- Keith M Wesolowski "Sir, we're surrounded!" FishWorks "Excellent; we can attack in any direction!" _______________________________________________ tools-discuss mailing list tools-discuss@opensolaris.org
