[MLton-user] IEEEReal rounding not working

[Florian Weimer]

* Matthew Fluet:

> My question is whether the comparison operators are pure; it seems
> clear that all of the arithmetic and trancendental operations depend
> on the rounding mode.  And, I guess the comparison operators are pure;
> the values have already been computed (at whatever rounding mode), and
> the comparison can't change that.

In GCC's case, excess precision and spilling can lead to impure
semantics of the comparison operators.  I'm not sure if this also
applies in MLton's case.

On the other hand, GCC should be pretty good at representing floating
point constants correctly.