- If look_at and position are set and up is unset, then we should default up to be orthogonal to look_at - position. - There are other malformed camera cases that we should be able to detect early before they lead to errors. - Using position to control the size of the view frustum is hard. - We should consider dumping the camera as a blueprint hdf5 file earlier and separately - There is a bug with multi-domain compositing. It is easy to see with a cube which is decomposed into thin slabs. - We would like to optionally dump a partial image for each domain and include the depth buffer with each. - We should provide parallel scale as an output. We can calculate it.