Should operations resulting in NaN or Inf cause an exception? about hafar

[Soupstraw]

I figured that currently it makes sense to throw an exception in cases where the function is not defined. That would allow to handle those cases manually.
I don't have any good ideas about what to do about infinite values either, e.g. it is not clear what the semantics of AF (Inf) [] (Inf) would be, same with NaN.

[folivetti]

I'm not really sure as well, in Numeric.Interval you get:

recip (-4 ... 4)
> (-0.25 ... 0.25)
recip (0 ... 0)
> (-Infinity ... Infinity)
log (-4 ... 4)
> (-Infinity ... 1.3862943611198906)

With the C++ libaffa library they also throw exceptions. But I think it should be better, when dealing with Floating, to return NaN or intervals containing Infinity. The reason is that if we have a sequence of operators that can fail, the failure would propagate (with NaN) and we would have to treat only the final result instead of each step.