On 07/May/2021 07:19, aotto wrote:
> Thanks,
>
> first:
> The "too late" was just an assumption because the "ac_XXX" definition
> use the CFLAGS
> before the CFLAGS is set by me, I understand that my try to force 32
> bit is not
> quite common but if the source-code is ONLY written and tested for
> 32bit it would
> be quite "unfriendly" to let the user build 64bit code which is
> probably broken.
>
> second:
> I think the hint about -m32 and -mx32 is good because -mx32 is
> probably the thing I really
> want. A "autoconf" option is a good approach to let the used decide
> between -m32 and
> -mx32
>
> third:
> The alternative would be to write a "wrapper-script" for "configure"
> to force the thing
> I want, I think this would be the most-worse thing.
>
> what I really missing is the proper (--host or --target) option which
> set the 32bit code
> correct and portable. Any --host option seems to search for compiler
> with "host" prefix
> -> this does not work together with gnu "gcc" because the option
> approach uses the same
> compiler front-end as 64bit.
>
> If I use my second approach the configure output start with:
>
> configure: loading site script /usr/share/site/x86_64-unknown-linux-gnu
> checking build system type... x86_64-pc-linux-gnu
> checking host system type... x86_64-pc-linux-gnu
> checking target system type... x86_64-pc-linux-gnu
>
> > there is NOTHING about 32 bit, even the -m32/-mx32 flag is not recognized
> > in the "host/target system type message.
>
> I feel uncomfortable about the "32bit" support in autoconf.
Forcing 32-bit is inappropriate, instead, the configure script should
do what it's told, if it can, or abort otherwise.
Example:
AC_CANONICAL_HOST
AS_CASE(["$host_os"],
[i?86-*], []
[*], [
AC_MSG_ERROR([Unsupported host architecture.])
])
# and here you could check the compiler output to see it matches
# host_os too.
