[Cfp-interest] exceptions and flags

[David Hough CFP]

> So 754-2008 does not explicitly recommendation a means to detect underflow exceptions, but its recommendation that languages provide alternate exception handling entails detecting underflow exceptions. Maybe Part 5 could leave detection of tiny exact underflow cases as a recommended option, which would allow the simpler slower implementation using flags.

I'd think it would be one of those situations where conformance to part 5
is optional, but if you do say you conform, then you conform to all of it.
Maybe we do need to have separate levels of conformance for different 
subparts.    The directives for optimization, however,
only give permission and don't require any implementation effort beyond
accepting the pragmas.

Having too many possible subsets is going to defeat portability.