Timo Heister

Using a custom data type with more than 2^31 elements works correctly, but using the MPI_File_*_c functions with a bigger count directly does not.

I would argue that code simplification is another reason to merge this PR. :+1:

I think what Marc argues makes sense. I would like to add that it makes sense to allow the user to specify values any time the FE changes, not only...

What compiler is this? We should keep this in mind for 9.4.1.

> I suggest to: I fully agree (we came up with this together on zoom). To add to this: ThreadGroup/Thread is not useful for any part inside the library as...

> but it needs to be done. ah, okay. I will take a look at that as well.

> would need to somewhere sketch the dependencies. That would be helpful. We can also video chat about it, if you think that helps.

@tamiko I don't understand https://github.com/dealii/dealii/blob/bbd65bc0527d84c18670ff7014e42aae1768ac77/cmake/config/CMakeLists.txt#L202 can you explain? does that explain why test with ``with_threads=on`` always run?

> If you want to have a different behavior for now We now always support creation of threads, so I don't see a situation where this would be necessary. Maybe...

I am looking for some initial feedback for an implementation of Workstream::run() using taskflow. I have made 4 different implementations so far and benchmark them against TBB. v4 seems to...