Tests failing for mixture model from sasview model editor

[pkienzle]

Mixture model tests in SasView 4.3 is failing with because Fq is not defined in mixture.py. Similarly for product.py. Could either override Fq or _call_kernel. The latter is cleaner but a little more work. In either case, Iq won't be needed. Third option is to skip the Fq test in model_test.py if not model.info.have_Fq.

This is related to #205, but we don't need to support beta approximation to remove the error in the testing infrastructure.

Example model:

from sasmodels.core import load_model_info

model_info = load_model_info('sphere+cylinder')
model_info.name = 'sphere_plus_cylinder'

Test output

Running model 'sphere_plus_cylinder.py'...

Success:
Note: sphere_plus_cylinder has no user defined tests.
test_sphere+cylinder_opencl (sasmodels.model_test._hide_model_case_from_nose.<locals>.ModelTestCase) ... ERROR
Traceback (most recent call last):
  File "/Users/pkienzle/Source/sasmodels/sasmodels/model_test.py", line 236, in run_all
    results = [self.run_one(model, test) for test in P_tests]
  File "/Users/pkienzle/Source/sasmodels/sasmodels/model_test.py", line 236, in <listcomp>
    results = [self.run_one(model, test) for test in P_tests]
  File "/Users/pkienzle/Source/sasmodels/sasmodels/model_test.py", line 336, in run_one
    F, Fsq, R_eff, volume, volume_ratio = call_Fq(kernel, pars)
  File "/Users/pkienzle/Source/sasmodels/sasmodels/direct_model.py", line 82, in call_Fq
    return calculator.Fq(call_details, values, cutoff, is_magnetic, R_eff_type)
  File "/Users/pkienzle/Source/sasmodels/sasmodels/kernel.py", line 155, in Fq
    radius_effective_mode)
  File "/Users/pkienzle/Source/sasmodels/sasmodels/kernel.py", line 189, in _call_kernel
    raise NotImplementedError()
NotImplementedError: sphere_plus_cylinder-opencl

---

It is curious that the test registered an error but reported success.

[RichardHeenan]

Does this matter? Is sasview 4.3 going to use current sasmodels master or stick at sasmodels v0.99 ?

[pkienzle]

Tests ought to be failing for 5.0 as well, assuming it is running the standard 'smoke tests' for each model to make sure it runs and doesn't return NaN. I haven't checked that this is the case.