Deriving from Integrator class in Python

[mkloczko-tango]

Hi

I'm trying to create a custom integrator in Python that derives from the Integrator class directly (not from Sampling Integrator).

class MyIntegrator(mi.Integrator):
    def __init__(self, props):
        #mi.Integrator.__init__(self, props)
        pass
    def aov_names(self):
        return []
    def should_stop(self):
        return False
    def render(self, scene, sensor, seed=0, spp=0, develop=True, evaluate=True):
        no_channels = sensor.m_film.base_channels_count()
        size = sensor.m_film.size()
        ret = dr.TensorXf(shape=(size.x, size.y, no_channels))
        return ret
    def cancel(self):
        print("Bye!")

mi.register_integrator("myintegrator", lambda props: MyIntegrator(props))
test = mi.load_dict({'type' : 'myintegrator'})

The issue is that I cannot use the mi.Integrator.__init__ since it is a pure virtual class, and for other Mitsuba objects if I do not call the base class constructor I get an issue with nanobind:

RuntimeWarning: nanobind: attempted to access an uninitialized instance of type '__main__.MyIntegrator'!

Can I derive from the Integrator class directly, and if so, how can I make it work? Do I have to call some nanobind functions?

[merlinND]

Hello @mkloczko-tango,

As far as I know, you can only inherit from the classes for which trampoline classes have been bound, i.e. SamplingIntegrator and AdjointIntegrator:

mitsuba3/src/render/python/integrator_v.cpp

Line 308 in 95ff1c1

MI_PY_TRAMPOLINE_CLASS(PySamplingIntegrator, SamplingIntegrator, Integrator)

mitsuba3/src/render/python/integrator_v.cpp

Line 375 in 95ff1c1

MI_PY_TRAMPOLINE_CLASS(PyAdjointIntegrator, AdjointIntegrator, Integrator)