Re: [patch] MIPS/Linux: Clean up cpp ABI dependencies

["Maciej W. Rozycki" <macro at mips dot com>]

On Mon, 8 Nov 2004, Richard Sandiford wrote:

> This is no coincidence.  It's very much a requirement that _ABIO32 ==
> _MIPS_SIM_ABI32, etc.

 Good.  I guess it means I can replace bits like:

#undef _MIPS_SIM_ABI32
#ifndef _ABIO32
# define _ABIO32	1
#endif
#define _MIPS_SIM_ABI32	_ABIO32

with:

#ifndef _ABIO32
# define _ABIO32	_MIPS_SIM_ABI32
#endif

in <sgidefs.h>.

> The issue here is that it's up to <sgidefs.h>, not the compiler,
> to define the _MIPS_SIM_* macros, and you therefore have to include
> <sgidefs.h> to access them.  But the compiler-defined _MIPS_SIM is
> designed to be comparable with macros like _MIPS_SIM_ABI32: that's
> the whole point.

 What's the point in such an awkward setup?

> To quote from SGI's implementation of <sgidefs.h>:
> 
>  *      If you use the _MIPS_* defines, you should include this header file.
>  *      In order to avoid problems when sgidefs.h is not included, the driver
>  *      predefines _MIPS_ISA and _MIPS_SIM to be the actual values (1-4) 
>  *      rather than the _MIPS_* names (otherwise, if sgidefs.h is not 
>  *      included, we would be defining _MIPS_SIM to be an undefined value).
> 
> Not an ideal arrangement perhaps, since comparisons against an undefined
> _MIPS_SIM_* macro won't generate an error.  But that's how the interface
> is defined (by SGI, I mean, not gcc maintainers ;).

 Well, it's for their systems -- IRIX, I presume.  Linux is not IRIX, so 
it does not have to follow something else's braindamage.

> GCC also defines _ABIO32 & co. as a convenience for those who don't
> want to depend on a header file and who don't care about non-gcc
> (or past-gcc) compatibility.  But...

 How about adding a few words on the matter to GCC info pages e.g. 
somewhere around the description of the "-mabi=" option?  I could cook 
some text, but you seem to have a better expertise and access to reference 
documentation in this area.

> The _MIPS_SIM_* macros are definitely more backwards-compatible though.
> FWIW, I think gcc should continue using them.

 Well, I don't ask for removing the _MIPS_SIM_* macros, so compatibility 
for user code should won't be affected.  For glibc itself backward 
compatibility seems not be much of an issue -- it already requires GCC >= 
3.2 and bumpig that requirement is likely just a matter of time.

  Maciej