Error estimation DOLFIN wrapper for FFC broken

[Garth Wells]

The demos

    demo/undocumented/auto-adaptive-poisson/cpp
    demo/undocumented/auto-adaptive-navier-stokes/cpp

fail to build after re-generating the UFL input using FFC. I get errors like:

    demo/undocumented/auto-adaptive-poisson/cpp/AdaptivePoisson.h:14200:9: error: ‘Form_8’ does not name a type

At the end of AdaptivePoisson.h there is:

   typedef Form_8 BilinearForm;

but there is no class 'Form_8'. There is a

    class adaptivepoisson_form_8

This might come from some recent UFL changes.

The broken demos are presently commented out.

[Martin Sandve Alnæs (martinal)]

The name adaptivepoisson_form_8 is not made by UFL, neither is the counting of forms (form #8), this must be an FFC issue. Probably caused by the regression fix I did in late august to make ffc use the form names that have always been provided by ufl (ffc -r1754).

How is this supposed to be compiled? Compiling with ffc -l dolfin I get different errors on make:

martinal@martinal-mac:~/dev/fenics/dolfin/work/demo/undocumented/auto-adaptive-poisson/cpp$ make
...
/home/martinal/dev/fenics/dolfin/work/demo/undocumented/auto-adaptive-poisson/cpp/main.cpp:68:3: error: ‘LinearForm’ is not a member of ‘AdaptivePoisson’

This is because the forms L and M in the .ufl file are both linear forms, and therefore no single LinearForm typedef can be made.

[Martin Sandve Alnæs (martinal)]

It was documented in the navier stokes demo, ffc -e -l dolfin.

[Marie Rognes (meg-simula)]

Yes, to generate error control forms and wrappers, compile with

  ffc -e -l dolfin

I'll add this comment to the auto-adaptive-poisson demo too, and fix the naming issue
by using Form_a/L instead of Form_8/9 (there are already some assumptions
on the forms for the error control stuff).

But Martin, the form 'M' does not appear as Form_M, do you know why?

[Martin Sandve Alnæs (martinal)]

It does if you do just ffc -l dolfin:

martinal@martinal-mac:~/dev/fenics/dolfin/work/demo/undocumented/auto-adaptive-poisson/cpp$ egrep 'class Form_' *.h
class Form_a_FunctionSpace_0: public dolfin::FunctionSpace
class Form_a_FunctionSpace_1: public dolfin::FunctionSpace
class Form_a: public dolfin::Form
class Form_L_FunctionSpace_0: public dolfin::FunctionSpace
class Form_L: public dolfin::Form
class Form_M_FunctionSpace_0: public dolfin::FunctionSpace
class Form_M: public dolfin::Form

so my guess is that maybe the error control code overwrites form_data.name 'M' with 10 in this case?

martinal@martinal-mac:~/dev/fenics/dolfin/work/demo/undocumented/auto-adaptive-poisson/cpp$ egrep 'class Form_' *.h | grep -v FunctionSpace
class Form_0: public dolfin::Form
class Form_1: public dolfin::Form
class Form_2: public dolfin::Form
class Form_3: public dolfin::Form
class Form_4: public dolfin::Form
class Form_5: public dolfin::Form
class Form_6: public dolfin::Form
class Form_7: public dolfin::Form
class Form_a: public dolfin::Form
class Form_L: public dolfin::Form
class Form_10: public dolfin::GoalFunctional

[Marie Rognes (meg-simula)]

Ok, thanks, will dig.

[Martin Sandve Alnæs (martinal)]

I think I see why. In the poisson demo the form M is a linear form M=u*dx where u is a TrialFunction, so I guess you create a new form and replace u with a Coefficient, and the name does not carry over.

Haven't looked at the code though. We have a workshop soon so I don't have more time for this now.

[Marie Rognes (meg-simula)]

This should be fixed now in the branch lp:~dolfin-core/dolfin/agec-fixes-051212. However, at the moment my dolfin doesn't seem to compile..., so I'll have to sort out some issues before testing and pushing.

[Marie Rognes (meg-simula)]

This should be fixed now, but I have not regenerated the form files because of dolfin::uint issues. I didn't see where these demos were commented out, so I haven't commented them back in.