We create and destroy instances all the time with high frequency (hundreds of avatars moving in and out of high fidelity ranges, and so on), and the InstancedMesh we use for this have mesh.doNotSyncBoundingInfo = true and mesh.alwaysSelectAsActiveMesh = true, since anything else is too costly.
Every time we create an InstancedMesh, in its constructor this.refreshBoundingInfo(true, true) is called ( Babylon.js/packages/dev/core/src/Meshes/instancedMesh.ts at aeec5b9613b6d15942860ae6aba11b6792900b97 · BabylonJS/Babylon.js · GitHub), which is very costly, especially since it can happen several times per frame. We never use this bounding info and it’s identical for all instances (bounding info of a proto mesh), since our displacement happens entirely in a vertex shader.
We’ve now approached this by mocking the InstancedMesh.prototype.refreshBoundingInfo while we create our instances, and then restoring the original function when done:
InstancedMesh.prototype.refreshBoundingInfo = function () {
return this;
};
It works well, despite the prototype mangling. We do wrap the mocking in a try { } catch { } finally { }, so it’s not possible for the mocked function to stick, even if a runtime error occurs while it’s mocked. Does anyone see any additional side effects/issues with this approach? Thanks!