Subject: Re: [ublas] [bindings] New traits system
From: Rutger ter Borg (rutger_at_[hidden])
Date: 2010-01-22 09:32:25
Thomas Klimpel wrote:
> I can update them, if you don't mind. I guess blas1.cpp will fail under
> some configurations because of the BIND_FORTRAN_NO_F2C_RETURN_CONVENTIONS
> stuff, but this can be sorted out later.
> We should already be able to remove the old "atlas" bindings.
Ok, moving on, done.
> Next steps would be to port the umfpack and mumps bindings to the new
> traits system, so that the old traits stuff can be removed. I won't start
> this work before end of next week.
I'll try to have the traits supporting the Yale sparse (compressed) data
structure to be operational by then. A request for those bindings: can they
be made more like the other bindings? I.e., with a directory/include
structure, or something? Especially the UMFPACK bindings seems to be one big
file. In addition, is there a lot to do for the const-overloads of those?
> I wouldn't do the same for the tnt or eigen2 frontends, because I'm not
> sure how much of the bindings functionality they can support at all. It
> would certainly be interesting to see how far eigen2 can get when enough
> adapters are provided for it. For tnt, I wonder whether it wouldn't be
> better to simply remove the bindings, because tnt seems to be quite dead.
Eigen should be able to do algorithms that require a linear array as data
structure. It also has a Sparse module including a container for it.
TNT... looks quite dead indeed. I don't mind having ~5 kB of bindings code
for it, though. I also won't mind if you feel like deleting it :-).