Fix metaclass check classcell

[yonmilk]

• Fix __build_class__ check __class__ cell
• Fix type.__new__ set classcell

When a class object is created with metaclass, the following operation is performed.
https://docs.python.org/3/reference/datamodel.html?highlight=classcell#:~:text=CPython implementation detail%3A In,a RuntimeError in Python 3.8.

CPython implementation detail: In CPython 3.6 and later, the __class__ cell is passed to the metaclass as a __classcell__ entry in the class namespace. If present, this must be propagated up to the type.__new__ call in order for the class to be initialised correctly. Failing to do so will result in a [RuntimeError] in Python 3.8.

1. Error handling when __class__ (the class it belongs to) and metaclass are different (bltinmodule.c)
2. If the result in step 1 is the same, set class in type.__new__ (typeobject.c)

But in Rust Python it sets classcell without comparing __class__.

if let Some(ref classcell) = classcell {
	  classcell.set(Some(class.clone()));
}

[Snowapril]

Unless we locate the code for setting type to classcell in type.__new__, build failure will not be disappeared 😂

[Snowapril]

Suggested change

	let __classcell__ = identifier!(vm, __classcell__);
	// It is correct to get it from tp_dict, not attributes.
	let cell = typ
	.attributes
	.write()
	.get(__classcell__)
	.map(|cell| PyCellRef::try_from_object(vm, cell.clone()));
	if let Some(Ok(cell)) = cell {
	cell.set(Some(typ.clone().to_pyobject(vm)));
	}
	typ.attributes
	.write()
	.get(identifier!(vm, __classcell__))
	.map(|cell| {
	if let Some(cell) = PyCellRef::try_from_object(vm, cell.clone()).ok() {
	cell.set(Some(typ.clone().to_pyobject(vm)));
	Ok(())
	} else {
	Err(vm.new_type_error(format!(
	"__classcell__ must be a nonlocal cell, not {}",
	cell.class()
	)))
	}
	})
	.transpose()?;

It seems that we miss to raise type error when the given __classcell__ is actually not a cell type

[youknowone]

I think previous version is better. cell.set has side effect and the result of map is not used. This is more like control flow than data stream. Using if-let will show the flow better and will not hide side-effect of cell.set.

[Snowapril]

Oh I agree. Then we only need to add type error handling

[Snowapril]

BTW, without that change request, looks great !! 😊

[youknowone]

This line has side-effect and the closure doesn't return anything. Using if-let statement would be better than map

[youknowone]

The logic looks good, but I changed error-handling style in edb4a96

PyObject is not expected to be used as Display object because it generates not very helpful error messages.
This was on TODO list for long time but forgotten. Once #4045 is merged, please rebase this PR and fix error formatting arguments.

Thank you for fixing this hair-pulling problem! Now I am try to getting how it is working.