Re: ghc 6.0.1 and Mac OS X 10.2.6 build

Alastair Reid wrote:
> This still feels like a brittle solution to me.

But an ABI is one of the most solid conventions on any platform, so I really
can't see a problem here.

> I believe that testing what your actual machine does and does not support is 
> the right approach when not cross-compiling.

Granted (almost :-).

> I appreciate that some people  may disagree here but if you do agree, then
> surely the same applies when cross-compiling and the only issue is how to
> implement such an approach?
>
> So, I'll ask again:
>
> Can we split our autoconf scripts into two parts: one to be run on the host 
> machine and one to be run on the test machine?

:-P * * * Seriously: Should the configure script run tests via rsh/ssh? And
what if the target platform can't be reached remotely? I've never seen such
an autoconf trick before...

My suggestion is:

 * If the test can be done via compilation (for the target platform!) alone,
   just do it that way (e.g. compile-time constants like EAGAIN etc.)

 * If the test can't be done via compilation alone, try the following in order:

    * If the test is very closely tied to the target host/platform (e.g.
      ABI-related stuff like underscores on external names) and it is a known
      host/target platform, use a case/switch on the host/target platform name.

    * If we are cross-compiling, we are out of luck: Abort! (with a suggestion,
      if possible, like "you probably don't need an underscore because the
      host/target platform looks like ELF").

    * Use a runtime test.

Cheers,
   S.