Re: select record versus dbpr() versus SPARC (at least)

[Marty Kraimer <mrk@aps.anl.gov>]

Benjamin Franksen wrote:
> Jim Thomas wrote:
>
> > Consider a select record with SELM == "Specified" (SELECTED in the code).
> > The select record initializes non-constant inputs to 1e+30 .
>
>
> Does anyone know the rationale behind this (IMHO rather strange)
> initialization?
>
> Why this (very large?) number?



It's been a long time since this was done so the following is just a guess.

Looking at the algorithms for  SELECT_HIGH, SELECT_LOW, SELECT_MEDIAN the code 
has to know if a fields was never given a value. For some reason the value 1e+30 
was chosen to represent "never given a value". What else could be used? Big 
problem unless IEEE floating point is assumed. Today perhaps a valid assumption 
(even today what about VAX and alpha?). 12 years ago this assumption was not 
considered safe to make.

> Why - if deemed necessary or at least useful - is initialization not
> done in selRecord.dbd, but somewhat hidden inside selRecord.c?

Since the code for the algorithm is hidden in the code why not also hide the 
initization in the code? At least it all appears in one source module.

> Ben



Marty