Complex bakeoff

[Jim Thomas]

Complex bakeoff                                          10/13/95     2:37 PM

Frank Farance:
> The question for Jim is:
>
>         Does the CCE proposal require all implementations to
>         support signed zeroes, or just IEEE implementations?

I don't think the complex specification can (or should) attempt to dictate
such aspects of the underlying arithmetic.  The specification should be
applicable to IEEE and non-IEEE implementations.  This has been the approach
for FPCE too.  So, no, CCE does not propose to require all implementations to
support signed zeros.

What's missing from the discussion is the additional specification of complex
arithmetic for IEEE implementations.  This specification will guarantee the
properties such as  finite / zero = infinite  and  finite / infinite = zero ,
and the consistency in treatment of signed infinities and zeros that programs
like the bakeoff examples depend on.  (For the floating-point specification
for real arithmetic, IEEE implementations have the proposed IEEE Annex,
derived from the "For IEEE Implementations" parts of FPCE, and of course the
floating-point standard itself.)

CCE is designed to facilitate the specification for IEEE implementations.  It
makes some "desired" behaviors hard not to get right.  Some are not automatic.
 More work is needed.  

Meanwhile, Tom MacDonald is right in saying CCE (without the additional
specification) does not guarantee the bakeoff example programs will behave as
intended.  I believe the examples do illustrate the utility of complex
infinities and signed zeros and the value of a programming model that supports
them. 

-Jim