aabb class should manage AABB rotation, not rotate_y

[hollasch]

Currently the rotate_y::rotate_y() constructor takes a hittable and a rotation angle, gets the unrotated hittable AABB, and then computes the new AABB post rotation. This is undesirable because it means that the rotate_y class has to get all of the guts of the AABB object, then perform AABB computation to get the components of the new AABB, and then generate a new AABB from these pieces.

Instead, this should all be in a aabb::rotate_y() method, which returns a new transformed AABB.

[trevordblack]

Would you also propose a aabb:translate() function to replace the translate::translate()

How does time affect these?
Will there still be a rotate_y::hit and translate::hit?

[hollasch]

Yes, similar for translations.

The hittable::bounding_box() method takes time start and time end, and it's up to the objects derived from hittable to compute the bounds of all bounding boxes at the extremes (since motion depends on the particular class). The aabb class should support a way to compute bounds of transformed bounds, since they don't act like regular geometry. Translate & scale are trivial, rotation and shear, for example, are unique to AABBs (which is why we shouldn't require all hittables that use AABBs to also manage their bounds).

[hollasch]

Punting out of v4.