weno_order solver keyword argument is silently ignored when kernel_language='Python'

[ahmadia]

We should either raise an error when this occurs or fix it.

[ahmadia]

This is pretty simple to reproduce:

cd pyclaw/apps/acoustics_1d_homogeneous
ipython
from acoustics import acoustics
clawf = acoustics(use_petsc=False, kernel_language='Fortran', \
       solver_type='sharpclaw', weno_order=15)
clawp = acoustics(use_petsc=False, kernel_language='Python', \
       solver_type='sharpclaw', weno_order=15)
q0 = clawp.frames[0].state.get_q_global()
qfinal = clawp.frames[claw.num_output_times].state.get_q_global()
print "Python kernel error: %e" % (dx*np.linalg.norm(qfinal-q0,1))

q0 = clawf.frames[0].state.get_q_global()
qfinal = clawf.frames[claw.num_output_times].state.get_q_global()
print "Fortran kernel error: %e" % (dx*np.linalg.norm(qfinal-q0,1))

On my machine I see the following output:

Python kernel error: 1.236605e-05
Fortran kernel error: 9.067284e-06

The Python error aligns with the default weno order=5:

clawf = acoustics(use_petsc=False, kernel_language='Fortran', \
       solver_type='sharpclaw', weno_order=5)
q0 = clawf.frames[0].state.get_q_global()
qfinal = clawf.frames[claw.num_output_times].state.get_q_global()
print "Fortran (5) kernel error: %e" % (dx*np.linalg.norm(qfinal-q0,1))
Fortran (5) kernel error: 1.236605e-05

[ketch]

If you're raising this as an issue, there are a dozen other combinations to raise. This was never designed as a real API. The point of the apps/ directory is to give simple examples. If you wanted to check for all these things, you'd need to make all the scripts twice as long (definitely a bad idea) or add new code implementing a standardized API for the apps scripts (probably a bad thing, in my opinion, since it makes it seem more complicated for users to roll their own scripts).

[ketch]

To amend my last comment, after looking more closely: I agree that the solver should tell you something is wrong if you set solver.weno_order != 5 and solver.kernel_language='Python'. Or we should go in and call the PyWENO kernels.