Class inheritance

[andyfriesen]

Rendered

See #229 for the main discussion thread on classes as a whole.

[Bottersnike]

Why is field redefinition with subtypes disallowed but methods are allowed to be redefined with a subtype? nvm I can see why it would cause issues depending on where it's accessed from

[deviaze]

I like that inheritable classes must be prefixed by open. It's a great indicator of when a class is supposed to be inherited upon, makes inheritance opt in instead of a default behavior, and prevents library authors from inadvertently adding a prop to a class that wasn't meant to be inherited and breaking downstream users who extend it.

[TenebrisNoctua]

This looks pretty good, except that I think Java's step of using final is better than open. An opt-out is better IMO, where I can extend any class, except the ones the developer does not want, versus I cannot extend any class except the ones the developer specifically goes out of their way to allow for inheritance, which is a thought that doesn't occur most of the time, even when the developer isn't against inheritance of their classes.

That aside, the reasoning for super is quite strong especially within the presence of inheriting constructors. It should be re-considered.

[MagmaBurnsV]

Again, going to mention this. Implicit hoisting is a hack and doesn't even work for 90% of the cases it would be useful for. Get rid of it.

[alexmccord]

Orthogonal to this RFC, but could we also have syntax for abstract methods? Using error("override me") as the implementation in the base class won't scale since that can error at runtime, but an abstract method can error during class prototyping.

[Bottersnike]

It's worth potentially also considering interfaces, and deciding on abstract vs interfaces vs both (both can be quite useful to have, but also potentially too much stuff going on depending on the desired design direction).

[TenebrisNoctua]

an abstract keyword seems to be the best option personally.