Re: SPU build break (Re: [PATCH v2] math: guard BSD finite/isinf/isnan functions properly in math.h)

[Yaakov Selkowitz <yselkowi at redhat dot com>]

On 2016-04-06 08:55, Ulrich Weigand wrote:
> Yaakov Selkowitz wrote:
>
> > diff --git a/newlib/libc/include/ieeefp.h b/newlib/libc/include/ieeefp.h
> > index 2ffa456..2c04284 100644
> > --- a/newlib/libc/include/ieeefp.h
> > +++ b/newlib/libc/include/ieeefp.h
> > @@ -241,19 +241,6 @@ typedef int fp_rdi;
> >   fp_rdi _EXFUN(fpgetroundtoi,(void));
> >   fp_rdi _EXFUN(fpsetroundtoi,(fp_rdi));
> >
> > -#undef isnan
> > -#undef isinf
> > -
> > -int _EXFUN(isnan, (double));
> > -int _EXFUN(isinf, (double));
> > -int _EXFUN(finite, (double));
>
>
> It seems this caused a build break on SPU:
>
> newlib/libm/common/s_isinf.c:17:9: error: expected identifier or '(' before 'sizeof'
>   _DEFUN (isinf, (x),
>           ^
> newlib/libm/common/s_isinf.c:17:9: error: expected ')' before '?' token
>   _DEFUN (isinf, (x),
>           ^
> because the SPU-specific #define of isinf from machine/ieeefp.h is still visible.
>
> Since common/s_isinf.c includes <ieeefp.h>, it seems this wasn't a problem in
> the past because of the above #undef that your patch removed.
>
> Should the #undef now move somewhere else, e.g. into s_isinf.c (and presumably
> s_isnan.c as well)?

Yes, those .c files would be the place for it; sf_isnan.c and sf_isinf.c 
already have such #undef's.

--
Yaakov Selkowitz
Associate Software Engineer, ARM
Red Hat, Inc.