Albert Zeyer

We can actually mostly share the code. I.e. move that over to the base `EngineBase`. Just `delete_model` might need to be specialized. That can also cover the optimizer states.

Note that PyCharm also covers type checks. I think the most common other tool for that is mypy.

Note that PyCharm also covers certain error checks, e.g. that a function must exists, or an object attrib, object method, and many more things. This is a particular important feature...

> > Note that PyCharm also covers type checks. I think the most common other tool for that is mypy. > > At least for me I have type check...

Note that I tried both pylint and flake8 at some point and it was in a totally different league compared to PyCharm. Maybe the flake8 extensions you linked can somewhat...

I now stumbled upon [Ruff - an extremely fast Python linter, written in Rust](https://github.com/charliermarsh/ruff). From a first glance, this looks interesting.

Almost all of it looks like false positives. I think also this issue is too generic. If you find some specific unused code, we should make separate issues about it,...

`extra_train` in `__main__`: This is valid per Python convention to leave this unused, when it is a tuple result. You often also have this when you iterate over dict items....

Many other things there are probably also used by some users, e.g. `MultiEndFastBaumWelchOp`, `SegmentFastBaumWelchOp`. So even if the tool is right that there is no use within RETURNN, it does...

Note that the API around `Cond`/`Loop` (#1282) is very related here. I am currently thinking that the best way forward is to avoid too much magic on RETURNN side and...