Test RigidLinearDeformationMappings_test failing sometimes

[hugtalbot]

As you can see in the dashboard (ex: commit 2b9daa), the test RigidLinearDeformationMappings_test is sometime failing ... and succeeds the next commit.

Quite strange.
This might be due to time stepping (convergence) issue.

[guparan]

SOFA-dev: we could add a build option DEBUG_RIGIDLINERDEFORMATIONMAPPINGS to log out the test in CI.

[epernod]

just to be sure, is it this test we are speeking about?
./Flexible_test --gtest_filter=RigidLinearDeformationMappings_test*

[damienmarchal]

Oui

[epernod]

Ok... I'm stuck in my investigations.
I tracked the error in the class: plugins/Flexible/deformationMapping/BaseDeformationMapping.inl
In method: void BaseDeformationMappingT<JacobianBlockType>::apply(const core::MechanicalParams * /*mparams*/ , Data<OutVecCoord>& dOut, const Data<InVecCoord>& dIn)

The VecCoord dIn seems to be corrupted, see log below. How do I know from where comes the vectore dIn and which component touched it before the "apply" method ? I don't know that part of Sofa visitors.

[WARNING] [LinearMapping()] out[  0 ]:  0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
[WARNING] [LinearMapping()] in[  0 ]: 3.47071e-311 3.47071e-311 0 0 0 0 1
[WARNING] [LinearMapping()] out[0] + jacobian[0][0].addapply(out[0],in[0]) =  -1.73535e-312 0 0 -1.73535e-312 0.5 0 0 0 0.5 -0.1 0 0 0 -0.05 0 0 0 -0.05 0 0 0 -0.05 0 0 0 0 0 0 0 0 0 0 0 -0.05 0 0

[WARNING] [LinearMapping()] in[  1 ]: 20 0 0 -nan -nan -nan -nan
[WARNING] [LinearMapping()] out[0] + jacobian[0][1].addapply(out[0],in[1]) =  -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan -nan

[maxime-tournier]

Hi all,

I'll ping Benjamin and Matthieu about this one, they might know more about it.

IIRC randomly failing mapping tests are mostly due to automatic differentiation step size/error threshold being incoherent with curvature, but this one should definitely not have NaNs in its input.

[guparan]

This will unfortunately not be fixed before v17.12
Reminder: activate SOFA_DUMP_VISITOR_INFO for further investigations.

[guparan]

May be linked to Quater problem raised in sofa-framework/sofa#399

[damienmarchal]

This will unfortunately not be fixed before v19.06 :)

[guparan]

To be closed @fredroy ?

[fredroy]

Now that Flexibile is no more supported, it should be closed here (sofa). But I will transfer this issue to the Flexible plugin, to keep a trace...