[petsc-dev] [petsc-maint] Complementarity problems: vinewton?sls or tao_?s?ls

[Lisandro Dalcin]

CC-ing petsc-dev

On 10 March 2015 at 03:08, Patrick Farrell
<patrick.farrell at maths.ox.ac.uk> wrote:
> I personally think that if the user hasn't supplied a getSubMatrix method in
> Python, that the Mat shouldn't have a mat->ops->getsubmatrix --- but maybe
> that
> isn't feasible or easy.
>
> Lisandro, what do you think?

Well, the thing is that Python is a dynamic language, so I designed
things to let users add methods even after the matrix was created.
This definitely plays bad with the way MatHasOperation() is
implemented.

If we remove that feature (I would not really care too much), then we
could defer the filling of most mat->ops pointers up to the point
MatSetUp() is called. What do you think? And BTW, perhaps we should
follow the same approach in the other Python subtypes (KSP, SNES, TS,
etc).

Or, we could change MatHasOperation() to have a slot in mat->ops, i.e.
if mat-ops->hasoperation is available, we delegate the call to the
subtype. This approach requires more coding in the PETSc side, some
careful check of places where mat->ops->xxx is accessed  where
MatHasOperation() should be used instead, and some minor extra code in
petsc4py (basically, a dict operation:methodname and a hasattr check).


-- 
Lisandro Dalcin
============
Research Scientist
Computer, Electrical and Mathematical Sciences & Engineering (CEMSE)
Numerical Porous Media Center (NumPor)
King Abdullah University of Science and Technology (KAUST)
http://numpor.kaust.edu.sa/

4700 King Abdullah University of Science and Technology
al-Khawarizmi Bldg (Bldg 1), Office # 4332
Thuwal 23955-6900, Kingdom of Saudi Arabia
http://www.kaust.edu.sa

Office Phone: +966 12 808-0459