IMHO, most of the Porter-Duff compositing operations are fairly useless. There is a nice mathematical purity to the model from which these operations are derived, but that's also their weakness: they are based on an ultra-simplistic abstraction instead of real world use cases. Punching out alpha mattes in various ways just isn't very interesting in practice.

Porter and Duff can be forgiven because the "real world" didn't exist when they designed this model in the early '80s. At the time, a frame buffer add-on to a computer could cost as much as a car...!

When designing a modern graphics API, I think it would be more useful to look at the operations that people are actually familiar with, e.g. Photoshop blending modes, and design the interface to match those expectations.