The ancient Tektronix 2232 has a momentary push rotary control and uses variable programmed hysteresis in the rotation which is both very simple and very effective. If I try, I cannot even make it glitch.
Well, we'll just have to accept your claim that the Tek knob is impossible to glitch, with smooth movement over all speeds it can possibly be turned at. Nevertheless, from a UI point of view, having to push something which rotates (except for the simplest toggling - e.g. coarse/fine) is a stupid idea on many levels and should be dropped completely from design.
When used to make a menu selection, it appears to use about 5 or 6 steps per rotation with matching hysteresis. When used for cursor control with push selecting between cursors, it appears to use about 15 steps per revolution also with matching hysteresis but response is non-linear for fast rotations. I wonder how old the patent is on control acceleration.
Native resolution is about 1200 counts per revolution using a 10 bit ADC which is a lot higher than the expected mechanical resolution; two potentiometers locked together and rotated 180 degrees are encoded separately to remove the dead zone of a single potentiometer in lieu of an optical or mechanical rotary encoder.
I do not disagree that it is a bad idea but good implementations work well enough. I
have run across poor implementations before where it was difficult to press the control without changing the selection or position which is why I was pleasantly surprised that Tektronix managed to implement it without problems on such an old instrument. They used the same rotary control implementation for a decade or more in different instruments but the 2232 and close cousins are the only ones I know from that era which had this rotary control combined with a push control. Maybe they decided it was a bad idea at the time but their current models do it.