ffc team mailing list archive

[Anders Logg <logg@xxxxxxxxx>]

On Sun, May 23, 2010 at 03:06:28PM +0100, Garth N. Wells wrote:
>
>
> On 23/05/10 14:59, Anders Logg wrote:
> >On Sun, May 23, 2010 at 03:32:51PM +0200, Kristian Oelgaard wrote:
> >>On 23 May 2010 15:08, Garth N. Wells<gnw20@xxxxxxxxx>  wrote:
> >>>
> >>>
> >>>On 23/05/10 13:58, Kristian Oelgaard wrote:
> >>>>
> >>>>On 23 May 2010 12:51, Kristian Oelgaard<k.b.oelgaard@xxxxxxxxx>    wrote:
> >>>>>
> >>>>>On 23 May 2010 12:14, Garth N. Wells<gnw20@xxxxxxxxx>    wrote:
> >>>>>>
> >>>>>>Where are we at with attaching data to integrals? The quadrature order
> >>>>>>can
> >>>>>>be attached, but I's like to do:
> >>>>>>
> >>>>>>    dS({"representation": "tensor"})
> >>>>>>
> >>>>>>Is this supported by FFC?
> >>>>
> >>>>It looks like this is already supported in FFC.
> >>>>Note that the representation must be the same on each subdomain (this
> >>>>is also a requirement for quadrature_degree).
> >>>>
> >>>>  dS(0, {"representation": "tensor"}) + dS(0, {"representation":
> >>>>"quadrature"})
> >>>>
> >>>>is not possible, but
> >>>>
> >>>>  dS(1, {"representation": "tensor"}) + dS(0, {"representation":
> >>>>"quadrature"})
> >>>>
> >>>
> >>>OK.
> >>>
> >>>    dS(0, {"representation": "tensor"}) \
> >>>  + dS(0, {"representation": "tensor"})
> >>>
> >>>works, but
> >>>
> >>>    dS({"representation": "tensor"}) \
> >>>  + dS({"representation": "tensor"})
> >>>
> >>>doesn't. Seems like a bug.
> >>
> >>It's because domain_id is not a keyword argument in the __call__
> >>function of Measure.
> >>I changed this such that you can do:
> >>
> >>  dS(metadata={"representation": "tensor"}) \
> >>  + dS(metadata={"representation": "tensor"})
> >>
> >>>Turns out I can't use this anyway because facet normals are not supported by
> >>>FFC with the tensor representation,
> >>>
> >>>    ffc.tensor.monomialextraction.MonomialException: No handler defined for
> >>>terminal FacetNormal
> >>
> >>I don't know if this is just because it is not implemented yet, or
> >>because it is not possible to support this with
> >>tensor representation.
> >
> >It's possible to support (and has been before), but I removed some
> >"extra features" of the tensor representation since those were
> >supported by quadrature anyway.
> >
>
> It would be nice to have the normal vector working. Does it work for
> exterior integrals? The exterior case is very common.

I would not object to it being added but it's not very high on my list
of priorities to add it.

--
Anders

Attachment: signature.asc
Description: Digital signature