Re: AC_PROG_CC not working

[Pavel Roskin]

Hello, Peter!

> One real example is PostgreSQL on AIX, where you need to use xlc instead
> of whatever else configure might think.  (AC_PROG_CC_STDC is less than
> ideal for related reasons.)  I still want to check for gcc first, though,
> so I'd use AC_PROG_CC([gcc xlc cc]).  Then we have older versions of
> BSD/OS, where you need to take "gcc2".  There are several more examples on
> the C++ side, e.g., aCC on HP-UX, xlC on AIX.
> 
> "Fix the list" is not a good answer, because you might break programs that
> actually needed some other (K&R?) compiler.
> 
> Of course the final solution is to implement support for C dialects, as I
> have mumbled several times before around these parts (and it even made it
> to the TODO list).  But this feature is a good stop-gap.

I agree with your arguments, except that I don't believe in the "final
solution" and I don't want to reduce the problem to C dialects.

There are more requirements to the compiler than just its ability to
support some language features. Other package-specific requirements may
include:

- Ability to produce dependencies
- Faster code
- Smaller code
- Ability to eliminate use of some CPU registers
- Ability to pass structures to the native libraries

It would be wrong to think that every system will have one "best" compiler
suitable for everything.

To sum up:
- the real live is more complicated that we think it is
- some kind of hints to Autoconf is needed
- what we have now is not ideal but should be kept

Regards,
Pavel Roskin