Dynamic change of a component Data field

[Camille-K]

Hi everyone,

This topic has been discussed briefly on the SOFA gitter, but it can be interesting to keep track of it in here.

I'm trying to change a Data field of a component (in the Cosserat plugin), dynamically during the simulation, with a Python controller, and I need to update internal variables of the component whenever this happens.
The way I'm trying to do this is by using the addUpdateCallback method in the component constructor :

this->addUpdateCallback("updateFrames", {&d_curv_abs_section, &d_curv_abs_frames}, [this](const core::DataTracker& t)
    {
        SOFA_UNUSED(t);
        this->initialize(); // code to be executed when input Data are changed
        return sofa::core::objectmodel::ComponentState::Valid;
    }, {});

where the first parameters &d_curv_abs_section, &d_curv_abs_frames are the input Data whose change would trigger the callback, and the second parameter is a lambda function whose code is executed when the input Data are modified. The last parameter, which is empty ({}) in my example, indicates the data fields which are modified by the lambda function. It seems however that by default, if the output Data parameter is empty, then d_componentState is automatically added, c.f. the definition of addUpdateCallback in Base.cpp :

if(std::find(engine.getOutputs().begin(), engine.getOutputs().end(), &d_componentState) == engine.getOutputs().end())
        engine.addOutput(&d_componentState);

It's seems that the callbacks follow a 'lazy update' mechanism, meaning that the lambda function will be called only if one of the output Data is used (e.g. with a output.getValue()). In my case, to make sure that the component is up-to-date when the input Data are changed, I'm checking on d_componentState in every method of my component which need to be up-to-date with the input Data, with :

if (this->d_componentState.getValue() != sofa::core::objectmodel::ComponentState::Valid)
        return;

The return;not leading to the simulation crash in this case.

I haven't yet succeeded in doing what I wanted with the Python controller, but the issue could come from the rest of the code, and not from my use of the update mechanism.
Is this analysis of what is going on is right?

[hugtalbot]

sounds correct to me @Camille-K
@damienmarchal agreed?

@Camille-K keep us posted!

[Camille-K]

Sure ! Part of my instabilities were solved by adding the callback, and others by setting to false an option of the MechanicalMatrixMapper (following Alex's analysis in #2687), but I still have minor issues I'm trying to understand. I'll investigate a bit more before setting this discussion to answered

[damienmarchal]

Yes, @Camille-K did a nice summary.