Am 17.10.2006 um 03:45 schrieb Ken Dickey:

> Greetings,
>
> I'd like to raise the idea of changing the complex number code in  
> 3.10/4.0.
>
> See change set at:
> 	http://bugs.impara.de/view.php?id=3311
>
>
> Current (3.8/3.9):
>   2i isNumber. "false"
>   -4 ln. "NaN"
>   -4 sqrt. "exception"
>
>  Alternate Code:
>   2i isNumber. "true"
>   -4 ln. "(1.38629436111989 +3.141592653589793i)"
>   -4 sqrt. "2.0i"
>
>
> I consider this a "community issue".
>
> Questions:
>   - Are there users of complex numbers (does anyone care)?
>   - Assuming yes, are there objective criteria for choosing between  
> alternate
> implementations?
>    + behavior/completeness/test-cases
>    + performance (I suspect that "the wrong answer fast" is not the  
> Smalltalk
> way as we can always augment the primOps)
>    + complex number user community vote
>    + other...?
>
> I am actually agnostic as to which code base gets chosen, but we  
> really should
> get the answers right.

Well, in best Smalltalk tradition I would vote for the alternate  
code. It's similar to fractional arithmetic ( 3 / 4 ). Yes, fractions  
bite some newbies, but they learn to use #// or #asFloat fast enough.  
Same here - if you need an error when taking a negative's square  
root, send #asFloat.

It's also a bit similar to Smalltalk being one of the few systems  
with correct integers, where a bug like this is not possible:

http://googleresearch.blogspot.com/2006/06/extra-extra-read-all-about- 
it-nearly.html

I'd like to have no arbitrary limits, like forbidding to take the  
logarithm of negative numbers.

- Bert -