Robert Kawulak skrev:
From: Thorsten Ottosen
It's not portability that has been the main concern here. It is the very small set of guarantees given by the standard regarding FP calculations, lacking some of the natural assumptions everybody has when working with FP. Also lacking some of the guarantees that would make preserving the invariant of constrained float an easy task. It's not the library that can take this decision for the users. The library does as much as it can to help the users. It still better than status quo.
Sorry, I'm not sure if I understand your point here -- could you please rephrase this?
There is nothing profound here. It's the same thing I been trying to communicate for some time. The fact that floating points are surprising and tricky is not an argument against putting support for bounded_float in your library. User deal with these problems in ad hoc manners (the status quo), and your library can help them with easier error-checking and clipping etc. -Thorsten