Re: CXSparse v2.0

[David Bateman]

John W. Eaton wrote:
> On  4-May-2006, David Bateman wrote:
>
> | David Bateman wrote:
> | 
> | >There was a discussion a month or so ago about the interface changes in
> | >CXSparse v2.0 and how to treat these. In the end its not a major issue
> | >to allow both versions prior to and after v2.0.0 to be used, and the
> | >attached patch to the CVS allows this
> | >
> | >Regards
> | >David
> | >
> | >2006-05-04  David Bateman  <address@hidden>
> | >
> | >    * SparseQR.cc, SparseQR.h, SparseCmplxQR.cc, SparseCmplxQR.h,
> | >    sparse-dmsolve.cc : Allow compilation with versions v2.0.0 of
> | >    CXSparse or later
> | >
> | >2006-05-04  David Bateman  <address@hidden>
> | >
> | >    * DLD-FUNCTIONS/spqr.cc (Fdmperm): Allow compilation with versions
> | >    v2.0.0 of CXSparse or later.
> | 
> | Sorry, here is the same patch without the extraneous muck...
>
> Is there any reason we shouldn't just require version 2 or later?
>   
CXSparse v2.0 is available as a separate package at the moment but not 
part of UFsparse and so there is no appropriate upstream release for the 
distributions to build against yet. This minimizes the pain of the 
transition will adding a little cruft that can be removed in a few 
months time..
> Would it be worth having a feature check in configure instead of
> looking at the version number?
>   
Perhaps, but I don't see the interface will change again in the near 
future, so a check against the version number seems just as good and 
avoids changes to the autoconf stuff.
> What about using some macros to avoid repeating the #ifdefs
> everywhere?  Something like
>
>   #if defined(CS_VER) && (CS_VER >= 2)
>   #define CXSPARSE_SQR(ORDER, A, QR) \
>     CXSPARSE_ZNAME (_sqr) (ORDER, A, QR)
>   #else
>   #define CXSPARSE_SQR(ORDER, A, QR) \
>     CXSPARSE_ZNAME (_sqr) (A, ORDER-1, QR)
>   #endif
>
> and then
>
>   S = CXSPARSE_SQR (order, &A, 1);
>
> instead of
>
>   #if defined(CS_VER) && (CS_VER >= 2)
>     S = CXSPARSE_ZNAME (_sqr) (order, &A, 1);
>   #else
>     S = CXSPARSE_ZNAME (_sqr) (&A, order - 1, 1);
>   #endif
>
>   
Ok, this seems like a good idea. I'll do it this way..

D.