> > (1) that should be __mips__, because __mips shouldn't be defined by
> > properly configured NetBSD compilers (in-tree or out).
> 
> I was going to say that it didn't define it, but I looked at it again
> and it turns out that it does.   So you're right - it can test that
> instead.   However, the NetBSD/pmax compiler currently does define
> __mips.

"it shouldn't."  8-)


> > (2) if it applies to the mips, it probably applies to the alpha as
> > well.  (Indeed, this looks like it would apply to the alpha.)
> 
> Almost certainly true.
> 
> > (3) it might be better to key on __ELF__, so that when building with
> > and ELF toolchain (regardless of platform) the right thing happens.
> > of course, that has its own set of problems.
> 
> There's no such predefine.   I suppose we could make there be one,
> though.

"there should be one, probably."  8-)


> Actually, I think for now the ifdef should read:
> 
> #if !defined(__svr4__) && !defined(__SVR4) && \
> 	!(defined (__NetBSD__) && (defined (__mips__) || defined (__alpha__)))
> 
> Does that work for you?

That could be fine.

I might suggest

	(defined(__NetBSD__) && (defined(__mips__) || defined(__ELF__)))

that should do the right thing on e.g. the PPC as well.  once the
compiler on the mips-based ports is fixed, the special case can go
away.


chris