Re: Draft minutes for 2011-01-12

On Wed, 12 Jan 2011 18:55:01 +0100, John Boyer <boyerj@ca.ibm.com> wrote:

> I also believe the following attribution to Steven should be changed:
>
> 4) "That sounds a bit kludgey. In our system, we discovered not
> circularity but math precision errors..."  => "That sounds a bit kludgey
> but we did also find a reason to stop infinite processing in our system
> from the eighties. We discovered not circularity but...

Circularity is actually OK, as long as there is a fixed point.

It wasn't exactly math precision errors we had, but the fact that computer  
arithmetic has slightly different properties from real arithmetic
(for instance (a+b)-c can give different results to (a-c)+b) and we had  
constraints where it looked right, but if one value got changed that would  
change the another, and it should have stopped there, but that caused the  
original number to change, and they would both ping-pong between two very  
close numbers. So instead of:

 celsius = (fahrenheit-32)*5/9
 fahrenheit = (celsius * 9/5) +32

we had to effectively do:

 celsius = if (fahrenheit != (celsius * 9/5) +32) then (fahrenheit-32)*5/9  
else celsius
 fahrenheit = if (celsius != (fahrenheit-32)*5/9) then (celsius * 9/5) +32  
else fahrenheit

(the details were different, but this was the basis of the idea).

Steven

Received on Wednesday, 19 January 2011 15:53:14 UTC