[MLton] starting builds

[Wesley W. Terpstra]

On Sep 9, 2005, at 9:05 PM, Stephen Weeks wrote:
>> Bug filed: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=327351
> Great.  From the looks of the discussion it seems it will be fixed
> soon.

Are you kidding?
Libc bugs last forever. :-P

> In any case, I don't think it makes sense to rewrite MLton's
> Real.{ceil,floor,round,trunc} to work around what is clearly a libc
> bug.

The point is that on some architectures you will have to.
I suppose this can be deferred until MLton is ported to them.

For example, arm can take the floor, but has no concept of
a rounding mode in the processor. The ML basis requires a
working floor (as does C) on these platforms, but both allow
for rounding modes to be missing on some platforms.

> It's not clear to me that on a platform where rint/festeround is
> broken that floor is any more likely to work.  If I were implementing
> libc I would do just as we did with MLton and implement floor in terms
> of rint/festround when possible.

That's a fair point, except that floor is much older than fesetround.