Hi this is the answer from Sparx about that:
„The Select Composite Diagram command is only shown for element types that can be composite, which excludes ports and parts. The composite structure of a port or part is the composite structure of its type, so you will need to go to the type element to select its composite diagram.“
In principle I say the argument is somehow valid. However instead of telling unexperienced users, you have to do this and that to see the detail of that element in EA and you have to do something else in Web EA, it would be a much better usability and therefore better user acceptance to say, just click or double click onto an element with the glasses or fork symbol to go to the details.
Furtheron, an Call Behavior Action, is as well just an action and is not intended to be composite. It is the called Activity which is composed.
So for call behavior action the clicking works.
For classified state (A state I assume is as well not intended to have a composite diagram) it works if the composite diagram is selected before classified by the state machine
For Sysml properties it does not works at all.
If you add a composite diagram under a requirement the “glasses link” is provided automatically. But later you are allowed to select any diagram in the model.
If you drag a use case having a composite diagram in a use case diagram, it shows the glasses and provide the link automatically as well. But later you are allowed to select any diagram in the model. However when you classify that composite use case by an other composite use case the linked diagram can not be changed any more. However when you classified a none composite use case by another use case again you can link to any diagram.
For objects and object nodes you are not allowed to add a composite diagram within the project bowser but you are allowed to link those objects to diagrams as long as it is not classified.
…
Whoever understands this, at least it is not me.

From my perspective that click and double click behavior is EA usability and not defined by any standard. A simple implementation could lead to better EA usability and would make it less hard to understand the magic,
Such a simple implementation could e.g. look like this.
I an element in a diagram has a composite diagram provide the “click” link to it.
I an element in a diagram has a composite diagram provide the “click” link to it and remain that link even it is classified later.
I an element in a diagram is classified by a classifier having a composite diagram provide the “click” link to it.
I an element in a diagram is classified by a classifier having a composite diagram provide the “click” link to it and remain it even a composite diagram is later added to that element. (or maybe switch the link to the new composite diagram
In all cases allow the use to select another diagram if needed.
Of cause the user can do odd things, but the user can do that anyway with the current implementation.