I am in a training class right now (TCI, Al Neumann). He states that Perpendicularity can be referenced to datums that are not perpendicular to the surface being held perpendicular. He has an example in his workbook showing it that way also. He refers to Perpendicularity, Parallelism and Angularity as Orientation controls (which they are), but seems to indicate that they are really flatness controls with orientation and that Perp, Par and Ang are just our perception of the direction of orientation.
In Alex Krulikowski's book "Fundamentals of GD&T" he explicitely states that the surface being called Perpendicular must be perpendicular to the datums it is being referenced to in the feature control frame.
Which is it? I've always gone with Krulikowski's interpretation. If we didn't, we could just get rid of Perp, Ang and Par and call it all Orientation.
What says you?