Re: x86_64 and x32 fail to build

[Jens Gustedt <jens.gustedt@inria.fr>]

[-- Attachment #1: Type: text/plain, Size: 1774 bytes --]

Am Montag, den 20.04.2015, 21:35 -0400 schrieb Rich Felker:
> On Tue, Apr 21, 2015 at 02:56:27AM +0300, Alexander Monakov wrote:
> > > When building without --enable-warnings, there are many false positives from
> > > -Wpointer-to-int-cast about x32 __scc(); at least from 4.5 onwards GCC enables
> > > this warning by default, so perhaps if musl really wants to silence it, it
> > > should test the corresponding -Wno-... flag outside of x$warnings == xyes
> > > test?
> > 
> > Could it be possible that implementation of __scc() can be adjusted to avoid
> > triggering the warning?  I hoped the following would achieve that:
> > 
> > // Cast X to signed integral type of corresponding size
> > #define __scc1(X) (__typeof__((X)-(__typeof__(1?(X):0))0)) (X)
> > #define __scc(X) sizeof(1?(X):0ULL) < 8 ? (unsigned long) __scc1(X) : (long long) __scc1(X)
> > 
> > .... but unfortunately it doesn't work with -Werror=pointer-arith when type of
> > X is 'void *'.
> 
> Indeed, introducing invalid C to make a warning go away is not a nice
> tradeoff.

Since such a solution would use a gcc'ish extension __typeof__,
anyhow, perhaps it would be preferable to use another one, namely
__builtin_choose_expr:

#define __scc(X) __builtin_choose_expr(sizeof(1?(X):0ULL) < 8, (unsigned long) (X), (long long) (X))

if that helps to quiencen the warning?

(and in any case __scc for x32 should get some () around, just to be
complete)

Jens


-- 
:: INRIA Nancy Grand Est ::: Camus ::::::: ICube/ICPS :::
:: ::::::::::::::: office Strasbourg : +33 368854536   ::
:: :::::::::::::::::::::: gsm France : +33 651400183   ::
:: ::::::::::::::: gsm international : +49 15737185122 ::
:: http://icube-icps.unistra.fr/index.php/Jens_Gustedt ::




[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 181 bytes --]