Classes

[andyfriesen]

Rendered

We jumped the gun on merging this RFC. Some light edits are necessary.

Most notably, creating new class instances is now always done via a static .new() method.

Class constructors are described separately at #210
Inheritance is described at #230

[deviaze]

As extensively discussed off-RFC, I still feel banning users from defining new and forcing users to define __init to define new (except when inheritance complicates things) is a confusing mental model.

As someone who's used the WIP experimental classes feature for useful code, I very much appreciate that POD structs (classes with default constructor) just look like named tables.

I still feel T() (with T { } syntax) makes more sense for POD nominal structs but T.new() (or another factory function that acts like a constructor) with user defined arguments and user defined behavior makes more sense for classes with actual constructor logic.

[deviaze]

We had a discussion on forced .new vs. constructors that I was recommended to post here for publicity/transparency:

[TenebrisNoctua]

As extensively discussed off-RFC, I still feel banning users from defining new and forcing users to define __init to define new (except when inheritance complicates things) is a confusing mental model.

As someone who's used the WIP experimental classes feature for useful code, I very much appreciate that POD structs (classes with default constructor) just look like named tables.

I still feel T() (with T { } syntax) makes more sense for POD nominal structs but T.new() (or another factory function that acts like a constructor) with user defined arguments and user defined behavior makes more sense for classes with actual constructor logic.

I proposed an alternative solution for this on this comment here, for anyone to see:
#210 (comment)