Fix for issue 1184

[mguijarr]

Following comment by @cappel89 , this MR proposes to define PositionerBase as an abstract class.

This way, it would not be needed to mention move() method needs to be overwritten in a specific way, while still
having a default implementation that can be called in subclasses (must be done explicitely). Same with .egu property,
no need to ask developer to overwrite it -> it would be done automatically by Python...

What do you think ?

[tacaswell]

If I recall correctly, when we started working on ophyd we were using a metaclass for Device and multi-inherentance involving multiple-metaclasses results in Python telling you it is too complicated and refusing to do it. I think you could do it by making a MI meta-class of the meta-classes and then using that as the meta-class on the normally-MI'd class or something like that....so we chose to not use the second meta-class! When the subclass init hook became available we switch to using that (in the spirit of using the simplest language features we needed to get what we wanted) so I think that is why this now works.

noticed API issue on final look

[tacaswell]

I'm not sure what to do about the argument order here.

Adding wait as the second positional arguement matches the order of EpicmMotor.move and PseudoPositioner.move but doing so is a major API breakage on this method (as if anyone was passing moved_cb positionally then they would both not get there callback run but wait for the motion to finish!

[mguijarr]

I had the same dilemma. I thought already implemented move show how everybody is supposed to have it.

What about adding * after position to force use of keyword args ?

I agree it is an API change. On the other hand current situation is not ideal neither.

Another possibility is to move wait argument at the end...