Error at compile-time when a non-subclassable class is being subclassed

[ChayimFriedman2]

Previously this crashed at runtime.

We even get to remove few tests. Nothing more fun than removing a no-longer-needed test!

Fixes #4451.

[davidhewitt]

Thanks very much for improving this; it's great to improve UX!

I think we might be able to keep the codebase a bit simpler by improving the existing PyClassBaseType trait instead of adding a new trait & generated code. At the moment all #[pyclass] types unconditionally implement it via a blanket impl, and most native types do via the pyobject_native_type_sized! macro. I guess we could add new bounds on the blanket, or even remove it and just emit implementations in the #[pyclass] macro if that improves diagnostics.

[ChayimFriedman2]

@davidhewitt Done.

[davidhewitt]

Overall this looks good to me, however there seem to be some feature combinations which aren't satisfied. I'd suggest running nox -s check-feature-powerset to help identify the broken cases and add #[cfg]s on them.

(Note that in the long run we should be able to remove a lot of the cfgs if we solved #1344. Which I think is very doable but just hasn't managed to top my priorities...)

[ChayimFriedman2]

@davidhewitt Fixed that. Those cfgs are a mess!

[ChayimFriedman2]

@davidhewitt How come this PR decreased coverage when all it did was fiddling with types? Do you have an idea?

[davidhewitt]

Pretty sure the coverage is due to other changes reducing coverage on main compared to the merge base on this PR. I think it can be fixed by #4472.

Let's merge this - thanks very much!

[davidhewitt]

I think the test-debug failure is another manifestation of #4185 (comment) 🤔

[ChayimFriedman2]

I can't reproduce the CI failure locally. Why does it even run the test under abi3 when it is explicitly cfg'ed out in this case?

[davidhewitt]

It's a mismatch between how the configuration and features are set up for that job. I have an idea how to make progress here that I'm going to play with tonight.

[davidhewitt]

@ChayimFriedman2 I pushed a commit which should fix the test-debug issue. Looks like I got clippy wrong; don't worry about fixing here. If test-debug passes we can drop the commit from this branch and merge it in #4497.

[davidhewitt]

I merged #4497 so I dropped the commit from here; hopefully CI is green and we can merge this shortly!

[davidhewitt]

Thanks again!