Tim,
Speaking with my polpack hat on, I've no opinion on this - never
having looked inside startcl very much, and knowing nothing about NBS. If
it looks a sensible thing to you, I'd say go ahead and try it.
David
On Sun, 12 Sep 2004, Tim Jenness wrote:
> NBS is a bit of a nightmare on AMD64. Everything is an int.
>
> Looking at the link scripts it seems to be used by the following packages:
>
> polpack [via startcl]
> icl
> startcl [simply to provide an interface to nbs]
> Starlink::NBS [perl modules]
>
> One of the problems with nbs is the way the fortran and C interface are
> the same code and clever macro definitions are used to switch between them
> - makes it hard to work out what is really going on.
>
> Also, lots of pointer arithmetic with ints rather than longs. I've made
> some progress but I'm beginning to wonder how significant NBS really is.
>
> For example, if I just had the C version to worry about life would be a
> lot easier...and it looks like no starlink code actually uses the fortran
> interface to NBS. Any objections if the fortran interface gets hobbled a
> little? [for example by returning a null pointer in the cases where the
> fortran interface requests a FILE* to the file on disk].
>
> --
> Tim Jenness
> JAC software
> http://www.jach.hawaii.edu/~timj
>
|