I suspect the C behaviour is for a very specific reason, namely to create correct rounding for interval arithmetic. (There are quite a few pitfalls for that in

I suspect the C behaviour is for a very specific reason, namely to create correct rounding for interval arithmetic. (There are quite a few pitfalls for that in initialization). "Rounding modes" might seem like the most boring irrelevant part of floating point numbers. Who cares if your result is off by the tiniest ammount? It is "wrong" in any case. But actually rounding modes allow you to create floating point software with mathematically true results, with a few very clever tricks.