Re: GSL 2.0 roadmap (one man's view)

[Gerard Jungman <jungman@lanl.gov>]

regarding error handling:

On Mon, 2009-09-07 at 14:57 +0100, Brian Gough wrote:
> 
> The issue here is threading -- otherwise we could use a global
> variable to control the behaviour.  

> Either we have to 
> 
> - pass an extra argument to all(?) functions for error control (seems
>   a large impact for a relatively small problem)
> 
> - provide separate _impl non-aborting versions for functions called
>   internally, similar to the special functions (simplest option).

A combination of these seems right. We also have the usual problem
of error codes for functions with natural prototypes. We should
finally sort that out and do something coherent across the library.
The _e and natural versions of the specfunc functions are just one
example of that. Any ideas are welcome. We should think about this
now, since it affects the interfaces across the library.


> - use thread local storage for a global error control variable (not
>   100% portable).

I would like to know how non-portable this is. I mean, I understand it
is different for every platform, but if everybody has one then it
only becomes an issue of configuring the variability.

Roughly speaking, I would think that every thread implementation must
have a notion of thread-local storage, so we're never required to
provide something that doesn't exist...

--
G. Jungman