Re: [pure-lang-users] complex 0

[Albert G. <Dr....@t-...>]

Albert Graef wrote:
> Good point! In fact gcc seems to do that, too (I get nan+:nan for /0.0 
> just as well as for /(0.0+I*0.0).

As a matter of fact, Haskell gives the same results, too (at least Hugs 
does), and the Haskell report uses the same formulas as math.pure in the 
complex/complex case.

I'm beginning to wonder whether it's really worth the hassle to follow 
the ISOC99 recommendations concerning the treatment of infinities and 
NaNs for complex * and /. The algorithms in Annex G do look like a kind 
of kludge to me, and they employ C99 functions for extracting and 
scaling the exponents of floating point numbers which might not be 
available on some systems (such as Windows), at least not directly. 
Hence special support in the runtime would be needed to implement these 
in a portable way across all platforms.

That seems like an awful lot of kludges to just make those nan+:nan's go 
away, considering that Pure also offers the polar representation where 
you'll get the correct infinite results for division by zero and similar 
cases, as long as the phase angles are finite.

So actually John's suggestion seems to be the most reasonable: just 
promote real operands of * and / to complex, then the results will at 
least be consistent. Or does anyone here really need the ISOC 
recommended behaviour?

Albert

-- 
Dr. Albert Gr"af
Dept. of Music-Informatics, University of Mainz, Germany
Email:  Dr....@t-..., ag...@mu...
WWW:    http://www.musikinformatik.uni-mainz.de/ag