Niklas Fehn

looks good from my side. @kronbichler do you also agree? I think then we could merge this.

I think you identified a place in the code that is currently not realized optimally. However, I am not sure if the proposed design should be chosen. The fact that...

> Regarding, computing quantities twice per time step: I thought about putting the quantities in a pair and recompute only if timestep_number differs. This has the advantage that not all...

@jh66637 You realized quite a good solution! Looking at the code, there is - in my opinion - one obvious deficiency in the code design that we need to fix...

With this PR, the day has come where my wishes have been granted! The problem is exactly the case with additional `FEEvaluation` objects (like in linearized operators; or convective operators...

I am not sure if it makes sense to merge a partial solution into ExaDG before we have a concrete plan of how to deal with the most difficult cases.

> However, this class would not store the actual FEEvaluation classes but rather return a wrapper around a vector/map of FEEvaluation object for the current cell batch. The wrapper class...

I thought once more about this topic. Maybe we should try to integrate this functionality into existing classes such as `FEEvaluation` and `MatrixFree`. Let me explain this ... In some...

@kronbichler If this is meant as a draft and not planned to be integrated into ExaDG in its current form/design, I think we should close this PR now.

Is the implementation restricted to certain types of finite elements, i.e. does the implementation assume an "isotropic" element in case of vectorial problems, where the different components all have the...