The BagPlog hint brought me on the probably right track:
I copied these BagPlot classes (inheriting from the plot matrix view/representation) and made them “empty” -> the resulting view/representation generated a plot matrix
removed the empty BagPlot classes and moved them one inheritance level “up”, so my view and representation classes became 1:1 renamed copies of the plot matrix classes -> test: still getting plot matrix from little test table
finally did the same with the vtkScatterPlotMatrix class - because I realized that this is actually the “heart” of the plot matrix classes -> same test ok
reformatted all the new clone classes with my own code formatting style: not because it is “better”, but for me this is the best way to learn what the code is actually doing!
Now I am ready to gradually change EVERYTHING in that plot matrix view!
(The question is whether I should start kicking out functionalities that I do not need - like the rather complex animation logic -, or probably better just leave it in for the moment…)
Bottom line: for me this is a way how source code becomes a “teacher” - after finding the right point to start with!
A learning about often used programming style in VTK/ParaView: I had to clone all these classes, instead of subclassing them, because they are all wrapping the “real thing” in some opaque internal class, only exposing an API that very often does not allow more than very little cosmetic changes. Ok, a second means to “hide the core” is of course private variables: the plot matrix view for example only needs to be cloned because the vtkScatterPlotMatrix pointer is “hidden” as private, so I need to clone the class in order to use my own clone of the vtkScatterPlotMatrix - which only gives me access to the “real working” of the plot matrix.
I assume that maybe it was not really the intention to make functions hard to subclass (for achieving major changes), but at the moment I am first of all happy it is working like it should! And if I look at what I want to achieve it is indeed a different thing than a scatter plot matrix, but rather some kind of “report with charts”