Hi Brian and Patrick,

The attached patch moves the derivative functionality into a separate
workspace.  Users allocate a gsl_bspline_workspace then a
gsl_bspline_deriv_workspace if they need the new capabilities.  All
the derivative-related functions are named like gsl_bspline_deriv_* to
help make the separation clear.  The patch includes updates to
doc/bspline.texi.

Applies cleanly to savannah and Brian's bspline-compat branch.  Brian,
I'm not sure if I missed running a specific compatibility test in your
branch or not...  If you had one in there, I missed it and haven't run
it.

- Rhys

On Fri, Dec 5, 2008 at 4:40 PM, Rhys Ulerich <rhys.ulerich@gmail.com> wrote:
> I'll break out the bspline derivative workspace and get back to you guys
> with a patch.
>
> FYI, the derivative workspace storage is O(order^2) in contrast to the
> evaluation workspace being only O(order).  Leading constant on the
> derivative storage term is small, but I could see where later joining the
> workspaces is a disadvantage for someone needing only evaluation
> capabilities.
>
> - Rhys
>
> On Fri, Dec 5, 2008 at 1:10 PM, Patrick Alken <patrick.alken@colorado.edu>
> wrote:
>>
>> Ah I hadn't considered this issue. If there are other areas of
>> GSL which have the same issues maybe we should keep a list so
>> when there is a major new release like 2.0 we could integrate
>> these "extra" workspaces? Of course one problem with that is
>> if we have a separate workspace for bspline derivatives it may
>> never be convenient to integrate it with the normal bspline
>> workspace.
>>
>> On Fri, Dec 05, 2008 at 06:56:15PM +0000, Brian Gough wrote:
>> > I noticed a problem when preparing the new release.  The extra entries
>> > in the bspline derivatives workspace break binary compatibility.
>> >
>> > This isn't necessarily a huge problem, but it seems undesirable to
>> > force an upgrade of all applications just for those two struct
>> > entries.  If we're going to break binary compatibility we may as well
>> > do it for all other functions that need it at the same time.
>> >
>> > So I'd propose using a separate workspace for the derivative functions
>> > in this release.
>> >
>> > See e.g.  branch bspline-compat at
>> > http://src.network-theory.com/gsl.git/
>> > which is basically finished, I just need to add some error checking.
>
>