dolfin team mailing list archive

[Johan Hake <johan.hake@xxxxxxxxx>]

On Thursday May 5 2011 13:46:29 Kristian Ølgaard wrote:
> On 5 May 2011 20:04, Johan Hake <johan.hake@xxxxxxxxx> wrote:
> > On Thursday May 5 2011 10:16:11 Neilen Marais wrote:
> >> Hi,
> >> 
> >> On Thu, Apr 21, 2011 at 6:18 PM, Johan Hake <johan.hake@xxxxxxxxx> wrote:
> >> >> > We should probably ignore all the ufc::cell versions, and add unit
> >> >> > test for the one we do expose...
> >> >> 
> >> >> Please excuse my naivete, but what is the essential difficulty in
> >> >> exposing the functions to the python side?
> >> > 
> >> > We have not paid too much attention making methods taking generated
> >> > code, ufc objects, available from Python. This is too low level, and
> >> > some of the ufc objects gets reordered when the dolfin objects gets
> >> > instantitated. This is mostly the case when run in parrallel.
> >> > 
> >> >> I ask since from a perspective of both testability of my own code and
> >> >> of learning what FEniCS does it is quite usefull to be able to call
> >> >> the lower level functions from python. For instance, I also want to
> >> >> use finite_element.map_to_reference_cell(), but it seems to give the
> >> >> same error that evaluate_basis_all() did before you fixed it :)
> >> > 
> >> > Yes, this would be the same case here. We also need to consider the
> >> > interface. It would clutter the interface to duplicate too many
> >> > functions. But if Garth or Anders is not opposed to it we could expose
> >> > that one too.
> >> 
> >> For me it would make sense to allow these things from python, since
> >> that is usually where I'd want to experiment with a new technique
> >> before trying to implement it fast in C++. I'm a little out of my
> >> depth here, but to me seems that "dolfin" is an easier to use wrapper
> >> around the lower level UFL stuff.
> >> 
> >> I would potentially be happy to code to a slightly lower level, but
> >> let me sketch out what I'm looking for in more detail. I don't want to
> >> have to implement mesh interconnectivity handling and DOF mapping from
> >> scratch. At the same time I want the ability to formulate things at
> >> the level of basis function evaluations as needed. In part this is
> >> simply because the "standard" dolfin doesn't cater quite that well for
> >> stuff I need for electromagnetic simulation (yet?), and in part since
> >> I may want to try new and/or hare-brained things :)
> >> 
> >> I'm quite happy to work with FEniCS developers to get the needed
> >> functionality into dolfin, and that is really first prize for me. In
> >> the meantime, however, my slave-driving overlords want a working EM
> >> FEM solver with academic outputs, so I can't neccesarily wait for that
> >> to happen first :)
> >> 
> >> >> If it is not to difficult you could teach me how to fix such
> >> >> problems. Then I could go fishing rather than asking for fish
> >> >> periodically :)
> >> > 
> >> > We need more fishermen! Just look at the last commit, where i added
> >> > the dolfin version. It should be pretty straight forward to implement
> >> > a similar method for the other methods which takes a ufc::cell.
> >> 
> >> OK, as far as I can tell the trick is just to cast the dolfin cell to
> >> a UFC cell. I tried the patch below to get mapt_to_reference_cell()
> >> working, but receive:
> >> 
> >> RuntimeError: map_to_reference_cell not yet implemented (introduced in
> >> UFC 2.0).
> > 
> > This you need to address to the FFC list, as FFC has not generated code
> > for that ufc method.
> > 
> >> Am I right in assuming that this is a limitation in the current dolfin
> >> C++ codebase rather than an issue with the code I added?
> >> 
> >> On another note, if it simply down to casting a dolfin cell to a UFC
> >> cell, could one not define a python casting function that returns a
> >> "UFC compatible" python object, and then tell the user that if they
> >> want to access the lower level UFC functionality they need to do that
> >> first?
> > 
> > I think this should be OK for UFCCell. But we should typically not expose
> > ufc::dof_map, as this is might be reordered.
> > 
> > One way to handle this is to add a typemap at the Python side, so you can
> > send a dolfin::Cell to any method accepting a ufc::cell. This will fix
> > the Python side of this request.
> > 
> > But I wonder why we need a dolfin version of any of these methods as one
> > in C++ can just create a UFCCell and pass that one instead.
> 
> Because it's easier when you iterate over cells in the mesh to just use
> that --> less code to be written by me.
> 
> > So a question to all C++ users. Would you mind if I removed all
> > dolfin::Cell versions of the dolfin::FiniteElement::foo and force you to
> > call these methods using a create UFCCell instead of dolfin::Cell.
> 
> Yes.

The point is that it is not consistant. Some methods have a dolfin::Cell 
version and some not. If we do not remove them I suggest we implement a 
dolfin::Cell version of all methods.

> If not I think we need to add
> 
> > dolfin::Cell version of all ufc::cell methods in dolfin::FiniteElement,
> > otherwise we will get new requests in the future.
> 
> Why isn't it possible to create a UFCCell in Python and pass that as
> argument? Everything is so easy in Python, I've heard.

That is possible, and will be done using typemaps, if we do not implement a 
dolfin::Cell version of all methods.

Johan

> Kristian
> 
> > Johan
> > 
> >> Cheers
> >> Neilen
> >> 
> >> 
> >> --- dolfin/fem/FiniteElement.h        2011-04-20 23:21:30 +0000
> >> +++ dolfin/fem/FiniteElement.h        2011-05-05 16:49:15 +0000
> >> @@ -192,6 +192,16 @@
> >>        _ufc_element->evaluate_basis_all(values, coordinates, ufc_cell);
> >>      }
> >> 
> >> +    /// Map from coordinate x in cell to coordinate xhat in reference
> >> cell +    void map_to_reference_cell(double* xhat,
> >> +                               const double* x,
> >> +                            const Cell& cell) const
> >> +    {
> >> +      assert(_ufc_element);
> >> +      UFCCell ufc_cell(cell);
> >> +      _ufc_element->map_to_reference_cell(xhat, x, ufc_cell);
> >> +    }
> >> +
> >> 
> >> > Johan
> >> > 
> >> >> Cheers
> >> >> Neilen
> >> >> 
> >> >> _______________________________________________
> >> >> Mailing list: https://launchpad.net/~dolfin
> >> >> Post to     : dolfin@xxxxxxxxxxxxxxxxxxx
> >> >> Unsubscribe : https://launchpad.net/~dolfin
> >> >> More help   : https://help.launchpad.net/ListHelp
> >> 
> >> _______________________________________________
> >> Mailing list: https://launchpad.net/~dolfin
> >> Post to     : dolfin@xxxxxxxxxxxxxxxxxxx
> >> Unsubscribe : https://launchpad.net/~dolfin
> >> More help   : https://help.launchpad.net/ListHelp
> > 
> > _______________________________________________
> > Mailing list: https://launchpad.net/~dolfin
> > Post to     : dolfin@xxxxxxxxxxxxxxxxxxx
> > Unsubscribe : https://launchpad.net/~dolfin
> > More help   : https://help.launchpad.net/ListHelp