Edge case conversions, exceptions to IEEE FPA
Van Snyder wrote:
> If the standard only provides one function to convert textual numbers to
> FP intervals, should the argument be considered to be exact or inexact?
The standard may provide explicit syntax for indicating whether a
literal is exact or not (and even encode an uncertainty bound).
You do raise an interesting point however: suppose the program gets
number strings from somewhere, and those numbers are in some plain
decimal format. Must the program use string operations to decorate
each number according to the aforementioned syntax, or should there
be a two-operand conversion function
stringToInterval(text, conversionSpecification)
where details such as abolute or relative uncertainty could be given?
One might also augment my
absUncertainFloatToInterval(value, abserr)
relUncertainFloatToInterval(value, relerr)
with
absUncertainStringToInterval(value, abserr)
relUncertainStringToInterval(value, relerr)
Michel.
Sent: 2008-11-11 23:30:54 UTC