0.0f

The old mesh Interface is the new m_triangleIndexVertexArray shared between physics controllers so we can delete the it.

We have the same behavior in RefreshCcdPhysicsController, and this loop is totaly wrong because each physics controllers has a unique shape.

It was commented out because (i think) previously we didn't support shape update for softbody, indeed we need a special operation for softbody in ReplaceControllerShape.

Oops i forgot this o.O

"an indexes", probably a typo. Also include an article ("the" or "a") between "for" and "triangle". Capitalize "Bullet" whenever you refer to the physics engine, and end sentences with a comma.

For gimpact the mesh interface is the triangle vertex array, but this triangle vertex array is shared between gimpact shape and store in CcdShapeConstructionInfo, so free one time line 2661~2.

You need to call RcreateControllerShape when you want to 1) remove and free the current Bullet shape 2) create an new Bullet shape 3) set the new Bullet shape. while ReplaceCOntrollerShape only set the new Bullet shape.

Is this also not true for TRIANGLE_MESH_SHAPE_PROXYTYPE? If not, we really should have a comment here explaining why we specifically don't handle GIMPACT_SHAPE_PROXYTYPE.

Why does a "recreate" function need to be called to "free" data? Also, your comment really doesn't answer the bulk of the questions/concerns raised in my original comment.

Is this the same memory as m_triangleIndexVertexArray?

Do we no longer have to clear collisions? I know we're removing the softbody, but what about any objects currently involved in a collision with a softbody?

I'm thinking this line could be moved to ReplaceControllerShape(). We could have the first argument be optional, and if it is NULL, we can create a btCollisionShape from the m_shapeInfo.

Why does this branch not also use m_triangleIndexVertexArray?

yes, meshInterface is m_triangleIndexVertexArray.

see line 683 in CcdPhysicsEnvironement : RefreshCcdPhysicsController
this old code is wrong (we check all rigid bodies with same Bullet shape, but Bullet shape are unique) and do the same work in RefreshCcdPhysicsController.

But the name will be not in harmony with the operation, replace not means create a new shape in certain case. and how we will do for arguments margin, gimpact ect ?

Okay.

I was thinking of having this in ReplaceControllerShape().

if (newShape == NULL)
    m_shapeInfo->CreateBulletShape(m_cci.m_margin, m_cci.m_bGimpact, !m_cci.m_bSoft);

It seems to make sense to me to have ReplaceControllerShape(NULL) recreate the existing shape.

This is still unanswered.

m_cci is own by CcdPhysicsController not by CcdShapeContrustionInfo.

in the case of softbody we have to use an unique vertex array to be deformed so we can use m_triangleIndexVertexArray which is reserved for sharing Bullet shape. We know that is a softbody when m_weldingThreshold1 is to 1.0f. it's also explicated in DeleteControllerShape.

ReplaceControllerShape() is a member function of CcdPhysicsController, not CcdShapeConstructionInfo. Why would CcdShapeConstructionInfo need to access CcdPhysicsController::m_cci?

This branch isn't for soft bodies, but it still doesn't use m_triangleIndexVertexArray.

for btScaledBvhTriangleMeshShape we don't use m_triangleIndexVertexArray because we free the shape which contains the triangle vertex array here only during the recreating of m_unscaledShape.
The equivalent way of gimpact shape but with a proxy : the class btBvhTriangleMeshShape.

With the changes to ReplaceControllerShape(), we no longer need this member function. We'll need to make ReplaceControllerShape() public again, and maybe DeleteControllerShape().