btCompoundShape change child after it attached to rigidbody

[GuoLei1990]

hello! I found some problem when I change the child of btCompoundShape(add or romove), if it attatched
to a rigidbody have problem, if not attatched to a rigidbody is Ok. but i want to dynamic change child weather it have attached . I know change a shape must remove rigidbody from the world and then add again and should compute setMassProps and calculateLocalInertia. I have do this. but still not work。Maybe some special thing need to do with btCompoundShape. Maybe my bullet verison is a litte old 2.82. Or I must destroy old btCompoundShape and new another btCompoundShape to set rigidbody again?（I know this can work）.

[BluePrintRandom]

I use unparent / transform / reparent each frame.

…

On Wed, Jun 6, 2018, 8:31 PM gl3336563 ***@***.***> wrote: hello! I found some problem when I change the child of btCompoundShape(add or romove), if it attatched to a rigidbody have problem, if not attatched to a rigidbody is Ok. but i want to dynamic change child weather it have attached . I know change a shape must remove rigidbody from the world and then add again and should compute setMassProps and calculateLocalInertia. I have do this. but still not work。Maybe some special thing need to do with btCompoundShape. Maybe my bullet verison is a litte old 2.82. Or I must destroy old btCompoundShape and new another btCompoundShape to set rigidbody again? I know this can work. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1737>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AG25Waw_zbR1bEh3CmmkqQVfqvUyR15iks5t6J56gaJpZM4Udrnn> .

[erwincoumans]

Yes, please update to latest Bullet version from github (not a release, just latest trunk/master).

Please zip a small modified Bullet example to reproduce your problem and attach to this topic. Also, try understanding/debugging the code for btCompoundCollisionShape/Algorithm to track down the issue yourself.

[BluePrintRandom]

I think what he is trying to do is have the center of mass change when a compound parent occurred, (I could be wrong*) I have done this manually in bge using a empty physics shape and parenting all children, moving the center of mass obj, parent all child shapes to empty(compound root) and setting it's mass to reflect its children.

…

On Thu, Jun 7, 2018, 9:25 AM erwincoumans ***@***.***> wrote: Please zip a small modified Bullet example to reproduce your problem. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#1737 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AG25WZ5_qhvygoEq0EH-t7NzSZ_OHkwtks5t6VP4gaJpZM4Udrnn> .

[GuoLei1990]

Oh! Thanks to reply me @BluePrintRandom and @erwincoumans. I am sorry my english is bad. In fact I am use the javascript version of bullet。 it work great. I will try the C++ sample later.
I found problem now. if I setLocalScale of btCompoundShape is not (1,1,1).for example is (3,3,3) and then addChild(childTransform,childShape)。the physic collider is small(have problem). if focre set setLocalScale of btCompoundShape(1,1,1) and then addChild(childTransform,childShape),and revert LocalScale(3,3,3) is OK(No problem).
In fact. I think
btCompoundShape.setLocalScale(3,3,3)
2.btCompoundShape.addChild(childTransform,childShape)
And
btCompoundShape.setLocalScale(1,1,1)
2.btCompoundShape.addChild(childTransform,childShape)
3.btCompoundShape.setLocalScale(3,3,3)
should same,because btCompoundShape.setLocalScale function is set the parent's LocalScale and childTransform is the child's local transform they are indepent. two ways world Bound should same.
Maybe I am not deep understand it.