Re: PATCH: fix gcc bug and big-endian issues for GPT

[Richard Hirst]

On Sat, Apr 06, 2002 at 06:20:44PM +1000, Andrew Clausen wrote:
> On Fri, Apr 05, 2002 at 10:17:37PM +0100, Richard Hirst wrote:
> > to config.h.  Now, when asm/types.h gets pulled in, the line
> > 
> > > typedef __signed__ char __s8;
> > 
> > gets pre-processed in to another typedef for int8_t, hence the above error.
> 
> Ah.  Subtle.
> 
> Does editting configure.in (and rerunning aclocal, autoconf, configure)
> help:
> 
>       AC_CHECK_TYPE(__u8, ,
>                     [AC_DEFINE(__s8, int8_t)
>                      AC_DEFINE(__u8, u_int8_t)
>                      AC_DEFINE(__s16, int16_t)
>                      AC_DEFINE(__u16, u_int16_t)
>                      AC_DEFINE(__s32, int32_t)
>                      AC_DEFINE(__u32, u_int32_t)
>                      AC_DEFINE(__s64, int64_t)
>                      AC_DEFINE(__u64, u_int64_t)],
>                     /usr/include/sys/types.h)
> 
> (replace the old AC_CHECK_TYPE commands...)

Last line of that edit needs to be

        [#include "/usr/include/sys/types.h"])

I think.  Anyway, it doesn't help.  I assume you expected that to make
configure find __u8 was defined, and so not define them in config.h?
It doesn't because the typedef is in asm/types.h.  If I specify /asm/ in
place of /sys/ above, then configure does find __u8 is defined, but then
of course the build breaks on every file that was relying on __s8 being
defined in config.h.  It probably would let ui.c compile though.

An alternative conclusion might be that glibc headers are broken on
ia64, as presumably no other arch is hitting this problem.  I don't know
whether it is reasonable for <signal.h> to pull in <asm/types.h>, but
<sys/types.h> does not pull in <asm/types.h>, so maybe not.  Anyway,
enough rambling ;)

Richard