Front page | perl.perl5.porters |
Postings from April 2014
On Sunday-201404-27, 20:13, Jarkko Hietaniemi wrote:
> On Sunday-201404-27, 18:37, Karl Williamson wrote:
>> The makers of gcc refuse to consider casting to (void) to be a valid way
>> of handling the return value of a function which is declared as having
>> its return value not ignored.
>
> Had to see this myself... I started up a Linux image [1] and now I see
> somewhat more clearly... oh, so, yes it's not so much a gcc issue alone,
> though, but an interaction of gcvt() being marked up with the gcc
> attribute warn_unused_result (in glibc, I guess). And (void) in gcc
> failing to overcome that attribute.
>
> Maybe something like:
>
> #ifdef __GNUC__
> #define UNUSED_RESULT(v) ({ typeof(v) z = (v); (void)sizeof(z); })
> #else
> #define UNUSED_RESULT(v) ((void)(v))
> #endif
>
> Or something. No idea how typeof fares across GCC versions (that
> solution nicked from stackoverflow, we are not the first to have this
> problem).
Looks like __typeof__() is preferable, compilers that fake being gcc
(like clang in OS X) get cranky with typeof(). Suggested patch
attached. If adding PERL_UNUSED_RESULT to perl.h is too bold, a patch
smaller in scope can go (again) to sv.c, though that would seem kind of
silly.
Thread Previous