Re: GSL Error Handling

[Gerard Jungman <jungman at lanl dot gov>]

"Gregory R. Warnes" wrote:
> 
> On Sun, 16 Jul 2000, E. Robert Tisdale wrote:
> 
>   ERT>> GSL functions should NOT return error codes!
> 
> I STRONGLY disagree.  Error checking is a critical part of any reliable
> software system.

Suffice it to say that we, the GSL team (right Brian?),
agree completely with the second statement and with 
everything else that Warnes says. Tisdale's problem
may be that he spends too much time thinking about vectors
and matrices, which are really pretty simple objects from the
standpoint of error handling and related state management;
you either got a valid array reference or you don't...

Obviously managing other more involved computational
tasks is not that simple.

The problem with library design in this area is not that
there is too much "side-band" interaction with the application,
but too little. Applications need to know what is going on
in a numerical implementation; they have a right to know.
The problem is arranging some form of communication channel
without breaking encapsulation. This is not an easy problem,
and archaic languages make it nearly impossible. So we
are stuck with "error codes" for now.


-- 
G. Jungman