IEEE special values

[Norbert Nemec]

John Roth wrote:
>> is there any simple reason why IEEE special values are so poorly
>> supported in python?
> 
> Like a lot of things, Python's math functions are a rather thin layer
> on top of the native C library's math functions. If the C library's
> math functions, to use the vernacular, suck, then Python's math
> functions will do likewise.

That's what I would have expected, BUT: gcc and glibc behave perfectly well
(2.95.4) so obviously, the layer in between is rather too thick than too
thin.

p.e. log(-1.0) gives a plain "nan" in C, while python throws an exception.

> One of the essential problems is that every library does certain
> things differently, like deciding whether it's going to start up with
> silent or signaling NaN's, or what the actual string values are for
> NaN, infinity and so forth.

I do understand that string representations may be a problem here (that's
exactly what PEP 754 would address) but I don't touch strings anywhere so
far. And the behaviour for binary representations is perfectly well defined
by the IEEE standard. Maybe there are old libraries out there that don't
implement this correctly, but why let them spoil the fun to all others?

> It takes developer time to clean up that situation. I suspect
> that volunteers would be very welcome.

Of course - I will not debate about that. :-)

Ciao,
Nobbi