dolfin team mailing list archive

["Martin Sandve Alnæs" <martinal@xxxxxxxxx>]

2008/8/23 Garth N. Wells <gnw20@xxxxxxxxx>:
>
>
> Martin Sandve Alnæs wrote:
>>
>> 2008/8/23 Garth N. Wells <gnw20@xxxxxxxxx>:
>>>
>>> Martin Sandve Alnæs wrote:
>>>>
>>>> 2008/8/23 Garth N. Wells <gnw20@xxxxxxxxx>:
>>>>>
>>>>> Martin Sandve Alnæs wrote:
>>>>>>
>>>>>> Add an instantiation of Array<DirichletBC*> to the .i files beside
>>>>>> ArrayFunctionPtr.
>>>>>
>>>>> I added this.
>>>>>
>>>>>> An use cpp_DirichletBC.
>>>>>
>>>>> I tried this is demo/pde/sym-dirichlet-bc/python/demo.py, but I get
>>>>>
>>>>> Traceback (most recent call last):
>>>>>  File "demo.py", line 55, in <module>
>>>>>  bc = cpp_DirichletBC(u0, mesh, boundary)
>>>>> NameError: name 'cpp_DirichletBC' is not defined
>>>>
>>>> from dolfin.dolfin import cpp_DirichletBC
>>>>
>>>> Note that cpp_DirichletBC won't hold on to python referenced objects,
>>>> so you need to keep some python references to your meshfunctions
>>>> and vectors etc. to avoid garbage collection.
>>>>
>>>>> As you say, this is a problem with this
>>>>>>
>>>>>> "inheritance pattern".
>>>>>>
>>>>> Is there a way to 'up cast' Python objects. For example, DirichletBC is
>>>>> derived from cpp_DirichletBC, so can I get a cast a DirichletBC object
>>>>> to
>>>>> a
>>>>> cpp_DirichletBC object?
>>>>>
>>>>> Garth
>>>>
>>>> The DirichletBC object is already a cpp_DirichletBC object, it obviously
>>>> works fine for single objects, so the problem is in the vector typemap.
>>>> Maybe you can create an empty ArrayDirichletBC() and add one item
>>>> at a time to it?
>>>>
>>>> a = ArrayDirichletBC()
>>>> for b in bcs:
>>>>   a.push_back(b)
>>>>
>>> No luck. I've pushed some code, but it doesn't work. Try running
>>>
>>>  demo/pde/sym-dirichlet-bc/python/demo.py
>>>
>>> Garth
>>
>> I'll try it. But this should work, we do the exact same thing with
>> functions in assemble.py.
>>
>>    _coefficients = ArrayFunctionPtr()
>> ...
>>            for c in form_data.coefficients:
>>                _coefficients.push_back(c.f)
>>
>> form_data is from ffc.jit, so c.f is a dolfin.Function, the kind that
>> inherits ffc.Function as well as cpp_Function.
>>
>
> I included Array.h in assemble.cpp and that did the trick (I've since moved
> the #include to Assemble.h). I have no idea why this makes it work.
>
> Garth

Are you sure it wasn't the change I checked in?
SWIG doesn't look at the .cpp files.

I instantiated std::vector<...> in addition to the
dolfin::Array<...>line you added first.
Otherwise SWIG wouldn't know the details of the base class for dolfin::Array.

--
Martin