Re: Failed float self test on OS X 10.5 with PowerPC

[Jeffrey Walton]

On Sun, May 3, 2020 at 3:04 AM Bruno Haible <address@hidden> wrote:
>
> Hi Jeffrey,
>
> > I think Bruno and I were talking about this for a while back in 2018
> > or 2019. Gnulib was failing one self test due to floats on OS X 10.5
> > on PowerPC....
>
> Do you have a pointer to the mail archives? What was my reply back then?
> Probably it was a reference to
>   https://www.gnu.org/software/gnulib/manual/html_node/Target-Platforms.html
> and "I know, I'm seeing this too, but Mac OS X/PowerPC is a platform of the
> past, not of the future".
>
> > $ gcc -O1 test.c -o test.exe
> > $ ./test.exe
> > sizeof(long double): 16
> > LDBL_MANT_DIG: 106
> > LDBL_MIN_EXP: -968
> > LDBL_MAX_EXP: 1024
> > LDBL_DIG: 31
> > LDBL_MIN_10_EXP: -291
> > LDBL_MAX_10_EXP: 308
>
> That's what I'm seeing as well on
>   - Mac OS X 10.4/PowerPC
>   - Mac OS X 10.5 with CC="gcc -arch ppc"
>
> > $ gcc -O1 -mlong-double-64 test.c -o test.exe
> > $ ./test.exe
> > sizeof(long double): 8
> > LDBL_MANT_DIG: 53
> > LDBL_MIN_EXP: -1021
> > LDBL_MAX_EXP: 1024
> > LDBL_DIG: 15
> > LDBL_MIN_10_EXP: -307
> > LDBL_MAX_10_EXP: 308
>
> We really don't have time to worry about ABIs that are binary incompatible
> to the default ABI of a platform.
>
> There are *tons* of GCC options that cause binary incompatible code to be
> generated by GCC:
> https://gcc.gnu.org/onlinedocs/gcc-9.3.0/gcc/Submodel-Options.html
> There's no point in spending time on supporting any of these in gnulib.

Maybe configure should die if LDBL_MANT_DIG is not expected. That
should avoid the problem of silent failures. When configure fails
folks will have to tend to the problem.

Jeff