DOLFIN not attaching spatial dim to UFL functions (from Python)

[Garth Wells]

I'm getting a UFL error from the Python interface when using the UFL commands lhs/rhs because UFL is not getting the spatial dim for coefficient functions in forms. lhs/rhs calls expand_derivatives, and then this bails out because the spatial dim is unknown (although I'm taking derivatives of the coefficient function). I expect that DOLFIN is failing to attach this data.

I'll post a short piece of code soon and look into, unless someone sees a quick fix.

[Garth Wells (garth-wells)]

On closer inspection, I do have a derivative. The problem is in taking spatial derivatives of a Constant, which behind the scenes is a 'Real'. Can UFL be updated to handle this case?

[Martin Sandve Alnæs (martinal)]

I need a concrete example to figure that out.

Martin
Den 8. sep. 2011 11.31 skrev "Garth Wells" <email address hidden>
følgende:
> On closer inspection, I do have a derivative. The problem is in taking
> spatial derivatives of a Constant, which behind the scenes is a 'Real'.
> Can UFL be updated to handle this case?
>
> --
> You received this bug notification because you are a member of DOLFIN
> Core Team, which is subscribed to DOLFIN.
> https://bugs.launchpad.net/bugs/844642
>
> Title:
> DOLFIN not attaching spatial dim to UFL functions (from Python)
>
> Status in DOLFIN:
> Confirmed
>
> Bug description:
> I'm getting a UFL error from the Python interface when using the UFL
> commands lhs/rhs because UFL is not getting the spatial dim for
> coefficient functions in forms. lhs/rhs calls expand_derivatives, and
> then this bails out because the spatial dim is unknown (although I'm
> taking derivatives of the coefficient function). I expect that DOLFIN
> is failing to attach this data.
>
> I'll post a short piece of code soon and look into, unless someone
> sees a quick fix.
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/dolfin/+bug/844642/+subscriptions

[Martin Sandve Alnæs (martinal)]

Anyway, the workaround is to provide a cell to the Constant.

Martin
Den 8. sep. 2011 13.59 skrev "Martin Sandve Alnæs" <email address hidden>
følgende:
> I need a concrete example to figure that out.
>
> Martin
> Den 8. sep. 2011 11.31 skrev "Garth Wells" <email address hidden>
> følgende:
>> On closer inspection, I do have a derivative. The problem is in taking
>> spatial derivatives of a Constant, which behind the scenes is a 'Real'.
>> Can UFL be updated to handle this case?
>>
>> --
>> You received this bug notification because you are a member of DOLFIN
>> Core Team, which is subscribed to DOLFIN.
>> https://bugs.launchpad.net/bugs/844642
>>
>> Title:
>> DOLFIN not attaching spatial dim to UFL functions (from Python)
>>
>> Status in DOLFIN:
>> Confirmed
>>
>> Bug description:
>> I'm getting a UFL error from the Python interface when using the UFL
>> commands lhs/rhs because UFL is not getting the spatial dim for
>> coefficient functions in forms. lhs/rhs calls expand_derivatives, and
>> then this bails out because the spatial dim is unknown (although I'm
>> taking derivatives of the coefficient function). I expect that DOLFIN
>> is failing to attach this data.
>>
>> I'll post a short piece of code soon and look into, unless someone
>> sees a quick fix.
>>
>> To manage notifications about this bug go to:
>> https://bugs.launchpad.net/dolfin/+bug/844642/+subscriptions

[Martin Sandve Alnæs (martinal)]

This is either:
- Defined as not a bug but a feature (since coefficients without cells is a wanted feature by Anders even though it is not well formed UFL)
- Maybe a bug, if it is fixable, but to reproduce I still need the context.

[Marie Rognes (meg-simula)]

Sample test case:

class FormOperations(unittest.TestCase):

    def test_lhs_rhs_simple(self):
        # Test taking lhs/rhs of forms involving constants. Most else
        # should be testable in UFL itself.

        mesh = UnitSquare(2, 2)
        V = FunctionSpace(mesh, "CG", 1)
        f = Constant(1.0)
        v = TestFunction(V)
        u = TrialFunction(V)

        F = inner(grad(f*v), grad(u))*dx + f*v*dx
        a, L = system(F)

        assert(len(a._integrals) == 1)
        assert(len(L._integrals) == 1)

[Martin Sandve Alnæs (martinal)]

A simpler case:

from dolfin import *

mesh = UnitSquare(2, 2)
V = FunctionSpace(mesh, "CG", 1)
f = Constant(1.0)
v = TestFunction(V)

F = grad(f*v)[0]*dx
a, L = system(F)

This is expected. Just pass cell to Constant to create
a well formed UFL expression and it will work fine:
f = Constant(1.0, cell=mesh.ufl_cell())

[Marie Rognes (meg-simula)]

On 11/10/2011 02:16 PM, Martin Sandve Alnæs wrote:
> A simpler case:
>
>
> from dolfin import *
>
> mesh = UnitSquare(2, 2)
> V = FunctionSpace(mesh, "CG", 1)
> f = Constant(1.0)
> v = TestFunction(V)
>
> F = grad(f*v)[0]*dx
> a, L = system(F)
>
>
> This is expected. Just pass cell to Constant to create
> a well formed UFL expression and it will work fine:
> f = Constant(1.0, cell=mesh.ufl_cell())
>

Yes.

So, let's mark the bug as a no fix, and add the "fixed" version as a
unit test to please the blueprint. Ok?

--
Marie

[Martin Sandve Alnæs (martinal)]

Actually, I think I've fixed it instead :)

Martin

On 10 November 2011 15:10, Marie Rognes <email address hidden> wrote:
> On 11/10/2011 02:16 PM, Martin Sandve Alnæs wrote:
>> A simpler case:
>>
>>
>> from dolfin import *
>>
>> mesh = UnitSquare(2, 2)
>> V = FunctionSpace(mesh, "CG", 1)
>> f = Constant(1.0)
>> v = TestFunction(V)
>>
>> F = grad(f*v)[0]*dx
>> a, L = system(F)
>>
>>
>> This is expected. Just pass cell to Constant to create
>> a well formed UFL expression and it will work fine:
>> f = Constant(1.0, cell=mesh.ufl_cell())
>>
>
> Yes.
>
> So, let's mark the bug as a no fix, and add the "fixed" version as a
> unit test to please the blueprint. Ok?
>
> --
> Marie
>
> --
> You received this bug notification because you are a member of DOLFIN
> Core Team, which is subscribed to DOLFIN.
> https://bugs.launchpad.net/bugs/844642
>
> Title:
>  DOLFIN not attaching spatial dim to UFL functions (from Python)
>
> Status in DOLFIN:
>  Confirmed
>
> Bug description:
>  I'm getting a UFL error from the Python interface when using the UFL
>  commands lhs/rhs because UFL is not getting the spatial dim for
>  coefficient functions in forms. lhs/rhs calls expand_derivatives, and
>  then this bails out because the spatial dim is unknown (although I'm
>  taking derivatives of the coefficient function). I expect that DOLFIN
>  is failing to attach this data.
>
>  I'll post a short piece of code soon and look into, unless someone
>  sees a quick fix.
>
> To manage notifications about this bug go to:
> https://bugs.launchpad.net/dolfin/+bug/844642/+subscriptions
>

[Martin Sandve Alnæs (martinal)]

I don't want to push my fixes until I can test it towards a green buildbot.
Can whoever feels responsible fix? Then I can push tomorrow.
My fix is available on lp:~martinal/ufl/work/ now.

Martin

On 10 November 2011 19:01, Martin Sandve Alnæs <email address hidden> wrote:
> Actually, I think I've fixed it instead :)
>
> Martin
>
>
> On 10 November 2011 15:10, Marie Rognes <email address hidden> wrote:
>> On 11/10/2011 02:16 PM, Martin Sandve Alnæs wrote:
>>> A simpler case:
>>>
>>>
>>> from dolfin import *
>>>
>>> mesh = UnitSquare(2, 2)
>>> V = FunctionSpace(mesh, "CG", 1)
>>> f = Constant(1.0)
>>> v = TestFunction(V)
>>>
>>> F = grad(f*v)[0]*dx
>>> a, L = system(F)
>>>
>>>
>>> This is expected. Just pass cell to Constant to create
>>> a well formed UFL expression and it will work fine:
>>> f = Constant(1.0, cell=mesh.ufl_cell())
>>>
>>
>> Yes.
>>
>> So, let's mark the bug as a no fix, and add the "fixed" version as a
>> unit test to please the blueprint. Ok?
>>
>> --
>> Marie
>>
>> --
>> You received this bug notification because you are a member of DOLFIN
>> Core Team, which is subscribed to DOLFIN.
>> https://bugs.launchpad.net/bugs/844642
>>
>> Title:
>>  DOLFIN not attaching spatial dim to UFL functions (from Python)
>>
>> Status in DOLFIN:
>>  Confirmed
>>
>> Bug description:
>>  I'm getting a UFL error from the Python interface when using the UFL
>>  commands lhs/rhs because UFL is not getting the spatial dim for
>>  coefficient functions in forms. lhs/rhs calls expand_derivatives, and
>>  then this bails out because the spatial dim is unknown (although I'm
>>  taking derivatives of the coefficient function). I expect that DOLFIN
>>  is failing to attach this data.
>>
>>  I'll post a short piece of code soon and look into, unless someone
>>  sees a quick fix.
>>
>> To manage notifications about this bug go to:
>> https://bugs.launchpad.net/dolfin/+bug/844642/+subscriptions
>>
>

[Anders Logg (logg)]

I'm working on it. The problem is that I enabled a bunch of unit tests
that had somehow gone missing from the top level driver file. So
new bugs have been introduced - it's only showing already existing
bugs or possibly broken unit tests.

--
Anders

On Thu, Nov 10, 2011 at 06:28:54PM -0000, Martin Sandve Alnæs wrote:
> I don't want to push my fixes until I can test it towards a green buildbot.
> Can whoever feels responsible fix? Then I can push tomorrow.
> My fix is available on lp:~martinal/ufl/work/ now.
>
> Martin
>
> On 10 November 2011 19:01, Martin Sandve Alnæs <email address hidden> wrote:
> > Actually, I think I've fixed it instead :)
> >
> > Martin
> >
> >
> > On 10 November 2011 15:10, Marie Rognes <email address hidden> wrote:
> >> On 11/10/2011 02:16 PM, Martin Sandve Alnæs wrote:
> >>> A simpler case:
> >>>
> >>>
> >>> from dolfin import *
> >>>
> >>> mesh = UnitSquare(2, 2)
> >>> V = FunctionSpace(mesh, "CG", 1)
> >>> f = Constant(1.0)
> >>> v = TestFunction(V)
> >>>
> >>> F = grad(f*v)[0]*dx
> >>> a, L = system(F)
> >>>
> >>>
> >>> This is expected. Just pass cell to Constant to create
> >>> a well formed UFL expression and it will work fine:
> >>> f = Constant(1.0, cell=mesh.ufl_cell())
> >>>
> >>
> >> Yes.
> >>
> >> So, let's mark the bug as a no fix, and add the "fixed" version as a
> >> unit test to please the blueprint. Ok?
> >>
> >>
> >>
> >> Title:
> >>  DOLFIN not attaching spatial dim to UFL functions (from Python)
> >>
> >> Status in DOLFIN:
> >>  Confirmed
> >>
> >> Bug description:
> >>  I'm getting a UFL error from the Python interface when using the UFL
> >>  commands lhs/rhs because UFL is not getting the spatial dim for
> >>  coefficient functions in forms. lhs/rhs calls expand_derivatives, and
> >>  then this bails out because the spatial dim is unknown (although I'm
> >>  taking derivatives of the coefficient function). I expect that DOLFIN
> >>  is failing to attach this data.
> >>
> >>  I'll post a short piece of code soon and look into, unless someone
> >>  sees a quick fix.
> >>
> >> To manage notifications about this bug go to:
> >> https://bugs.launchpad.net/dolfin/+bug/844642/+subscriptions
> >>
> >
>