Link tests after GCC_NO_EXECUTABLES

[Richard Sandiford]

OK, I'm not doing a good job of staying out of this discussion as
promised, but I was hoping someone else would raise this point...

My main concern with applying this patch in its current state is that,
while we have (intentionally) only been talking about *-elf so far,
newlib != *-elf.  Cygwin is also an important newlib user, and I don't
remember anyone explicitly mentioning it so far in this thread.  Is this
patch really safe there?  Cygwin's newlib provides all sorts of POSIXy
stuff that you don't get in your typical *-elf target (see libc/posix)
so things like:

> +ac_cv_func_execl=${ac_cv_func_execl=no}
> +ac_cv_func_execve=${ac_cv_func_execve=no}
> +ac_cv_func_execvp=${ac_cv_func_execvp=no}

(just a sample) might unintentionally limit the functionality of
libgfortran on that target.  I think the same thing could apply
to targets like Newlib/Linux, if GCC supports it.

While I'm here, I noticed a very minor nit.  The patch has:

> +ac_cv_func_fork=${ac_cv_func_fork=no}
> +ac_cv_func_fork=${ac_cv_func_fork=yes}

where the last line is redundant (though harmless).

Richard