Re: Vector class
> I am currently working on a modification of the package which allows
> for memory re-use (important for high-performance or time-critical
> applications) instead of creating a new Matrix with almost every operation.
Yes that would be nice as well. I have a library of APL-like methods that
will probably evolve into a small APL-in-Java implementation and all
relevant methods have in-place and "new" variants.
What's really irritating, though, is that Java does not dispatch based
on return type, only on arguments types, so the elegant and intuitive
solution of having both methods bear the same name and be distinguished by
their return type (void vs some handle) is not allowed. Instead you have
to give them different names or use some other kludge, like an extra
dummy argument. How do you handle that design issue?
A really tough one, too, is how to coax the library into dispatching based
on shape (i.e, atoms, 1D, 2D, 3D etc) without writing a zillion specialized
methods. There must have been some work on these issues in a C++ context
already, what with all those C++ numeric libraries. Do you know of any
interesting papers?
Regards,
-- O.L.
Date Index |
Thread Index |
Problems or questions? Contact list-master@nist.gov