I have the feeling it is an race condition/timing problem.
I was able to recreate the problem also in release after rapidly clicking arround (at least 50 times, machine gun style). In debug I clicked for minutes without anything appening.

In T29567#247266, @a178n wrote:

Is @wasserth active here? Or we goto GitHub and make an issue over there?

I would also open an issue at GitHub. As you already analyzed the current versions, we could also think of providing a pull request (and have the fork with that changes as back up plan for the release).

@a178n could you check if this task is solved with T29572?

I can do it, if you want.

I think it would be more efficient /sufficient to advise @wasserth to use explicitly defined versions for total segmentator dependencies. That is something all projects should regard for productive versions to have a better control on their supply chain.

grow cut (as all preview tools) does not generate any undo point right now. What you witness is the undo of your seed painting up front.

I am not sure I would put it a member function. This can be done without the need of accessing any internals.

We should discuss if it is still worth it to get it back online after all this time and with the new tools.

We should discuss if we realy try to get region growing back online with all the other tools ...

You should do a compbination (to keep the 4D support).

@a178n the changes in D816 should offer everything to solve the problem. I also adapted TotalSegmentator Tool. But you should verify that it works as expected.
Basically the tool should now change the labels in the preview as they like (see Otsu and TotalSegmentator for example) and use the TransferMode AddLabel. Then the labels (with all there settings) of the preview image will be added instead of mapped.

Pushed new branch to rMITK MITK: bugfix/T29458-Optimize_QmitkMultiLabelManager_buttons.

Serialization error: I already renamed LabelSetImageSerializer into MultiLabelSegmentationSerializer, but the data class still names LabeSetImage. I forgot that our scene serializer logic deduces the serializer by <className>+Serializer.
Thus it breaks now. I guess the simpelest solution for now would be to rename the serializer class back.

Crash seems to occur because the SurfaceInterpolationController observer logic is broken. It seems like registered observers ar never removed only added.
E.g. also if the segmentation is set to nullptr, nothing is removed from the old segmentation. That leads to the crash, because the events are triggered by the old segmentation when session is closed, but the controller has only a null ptr at this point.

right. it is now broken since we allow modification on none selected elements.

Definetly the interpolator has to rework and the different accept function refactored/streamlined.