Not a good idea – too many related changes. Better directly revert 58bb31b

I tried reverting and starting to fix test errors but I think this is not worth the effort.

True, we should have communicated this more clearly in the changelog.

But I think the benefit of reverting this is not worth the effort of potentially breaking already-adapted code.

So unless there is no workaround, I would prefer to retain isinstance(local_filename, Path).

But I think the benefit of reverting this is not worth the effort of potentially breaking already-adapted code.

So unless there is no workaround, I would prefer to retain isinstance(local_filename, Path).

Well, ... here's my arguments in favour of reverting to str:

• it's called local_filename, not local_path
• it is set via str, not via Path, so why should one not expect to get a str back?
• local_filename is also used as field/parameter on the CLI (which necessarily means str)
• I'm pretty sure it's easier to fix in core than in a multitude of external code bases (and we can never be sure we did not miss one, think Eynollah).
• Already adapted (external) code can easily be identified: search for local_filename. and change there. That's way more specific than the string local_filename in the other direction (adapting to new). Any str(*.local_filename) can be kept, because str(str(.)) == str(.).

I'm pretty sure it's easier to fix in core than in a multitude of external code bases

OK, then I will do the revert. And add typing information to avoid such a situation in the future.

Superseded by #1182