next »

[ChrisE]

Marvin,

So, I'm near my sanity's end on this issue.

Background

My scene graph consists of TransformGroups, each of which contains some number of Shape3D-derived objects (Line, Sphere, etc.) and TransformGroups as defined above (so, tree of TransformGroups which occasional Shape3Ds).

Problem

When I try to select certain groups by using a NearestPickListener (think that's the name), I'm unable to get any selection. Now, if I have no lines, selection works fine, and I am able to pick objects in the scene. However, when there are Line objects, in some cases I am unable to pick anything in the scene, including objects which are above the afflicted nodes in the scenegraph (which should be unaffected if it were just a bad local transform or bounds or something).

Any ideas about this?
Also, what is a PickHost and what is it for?

[Marvin Fröhlich]

PickHost: IF you have a complex object O built from several Shape3Ds, you might only want to get notified of O to get picked. So if you set O as the PickHost to all the contained shapes and one of these shapes gets picked, the pick listeners are notified, that O was picked.

It doesn't matter for the picking algorithm, if an object is "above" another. It won't get tested first. The order is computed later. So, if there is a bug, that keeps the picking algorithm from working correctly with a specific object, if might mess up picking before a closer object gets tested.

Lines have been an issue for picking in the past, because they have no geometrical size. They need a special pick test. This might be buggy.

Could you please build a small testcase for me?

[ChrisE]

So, funny thing... if one of the coordinates of the line is a NaN, the bounding volume for the line becomes filled with NaNs as well. And during the recursive calculation of the BVH for the scene, you get even more NaNs.

So, I fixed my code to reject malformed points, and all works again. However, we might want to consider skipping over the inclusion of NaN bounds in the BVH code.

-Chris