Segmentation fault on QPInverseKinematicsUnitTests with iDynTree 14

[traversaro]

See conda-forge/bipedal-locomotion-framework-feedstock#141 . iDynTree 14 was released ~2 weeks ago, but we only caught it lately with the conda-forge rebuild of blf with iDynTree 14.

[traversaro]

fyi @Nicogene @GiulioRomualdi @LoreMoretti @xela-95

[Nicogene]

I managed to reproduce it locally, the test that is failing is:

https://github.com/ami-iit/bipedal-locomotion-framework/blob/3ab9c6bdda1f8cf3a42d5fafd54d49ede910c8a3/src/IK/tests/QPInverseKinematicsTest.cpp#L376-L387

In particular the implementation of customGetRandomModelWithNoPrismaticJoints

https://github.com/ami-iit/bipedal-locomotion-framework/blob/3ab9c6bdda1f8cf3a42d5fafd54d49ede910c8a3/src/IK/tests/QPInverseKinematicsTest.cpp#L332-L355

uses addRandomLinkToModel that in:

• Add SphericalJoint and generalize CRBA to handle multiple-DOF joints idyntree#1249

Changed and now does not have a boolean for handling the joint types but a bit mask:

enum JointTypes : unsigned int
{
    JOINT_FIXED = 1 << 0,        // 0001 - FixedJoint
    JOINT_REVOLUTE = 1 << 1,     // 0010 - RevoluteJoint
    JOINT_PRISMATIC = 1 << 2,    // 0100 - PrismaticJoint
    JOINT_REVOLUTE_SO2 = 1 << 3, // 1000 - RevoluteSO2Joint
    JOINT_SPHERICAL = 1 << 4     // 10000 - SphericalJoint
};

Probably the model created does not only contain revolute joints

[traversaro]

uses addRandomLinkToModel that in:

*

Argh, I guess that changed in gbionics/idyntree#1255 ? That is a breaking change of a public function that we should have documented, I had missed that.