Pass an option to ActiveRecordAssociations compiler for how to generate association types

[stathis-alexander]

Motivation

With the introduction of the column type option for the ActiveRecordColumn, we are finally almost able to migrate from the now deprecated Sorbet Rails onto Tapioca for our AR model class RBIs. However, the option only affects active record columns that have a non-null DB index. (A separate related issue is that these should respect validations as well; addressed in #1996).

Another common, related pattern that makes Tapioca difficult to adopt is that has_one or belongs_to associations can also be declared as non-nil through the required and optional kwargs, respectively. Presumably the justification for this behavior is identical to that of the active record columns; i.e., newly instantiated models may not have these relationships setup yes, despite the standard case being that the model is actually persisted and fetched from the DB.

This PR introduces a new ActiveRecordAssociationTypes option with possible values nilable and persisted. The default is nilable and retains the existing, backwards compatible behavior of always generating nil-able types for associations.

With the persisted option set, a belongs_to relationship would generate:

belongs_to :widget, class_name: 'Widget'
# sig { returns(::Widget) }
# def widget; end

belongs_to :doohickey, class_name: 'Doohickey', optional: true
# sig { returns(T.nilable(::Doohickey))
# def doohickey; end

Implementation

We follow the implementation in #1888 to add a new compiler option.

Tests

Tests have been added to address any basic cases.

[stathis-alexander]

cc: @paracycle who implemented the original behavior.

[KaanOzkan]

I think this is fine to go ahead with the implementation. I don't have context on the reflection options but it looks normal.

[stathis-alexander]

@KaanOzkan Thank you for the comments! I have rewritten this for clarity, splitting the type option out into its own proper option. I have chosen a default of "nilable", since this retains the existing default behavior.

[stathis-alexander]

I felt it best to introduce this and then iterate later, rather than try to do to much at once. I left these notes for posterity, but I can remove them if it's unwanted.

If we do improve / extend the behavior here, we will likely want to dry up the code with that in the AR columns compiler.

[KaanOzkan]

This LGTM (I changed the title) but we need compiler documentation similar to 4c458b6. You can update the comment in the compiler file and then run bin/docs to update the corresponding markdown file.

You could also mention this compiler option in the relevant README section but it's not necessary.

[stathis-alexander]

@KaanOzkan , apologies for the delay here. I was traveling last week. I have updated the documentation. I believe this is ready for release.

[KaanOzkan]

Opened a PR to fix CI #2012

[stathis-alexander]

Hello @paracycle ! Thank you so much for your careful review and comments. I believe I have addressed them all.

[paracycle]

Thanks for this, looks great to me!

[egiurleo]

Hi, @stathis-alexander, we just fixed CI on Ruby 3.3. Would you mind rebasing on main and force pushing your branch?

[stathis-alexander]

Hi @egiurleo , I was under the impression based on the conversation here that this pull request would not be accepted.

[egiurleo]

Oh I apologize, I hadn't seen that other conversation. Let me catch up on it now.

[stathis-alexander]

Closing due to the conversation here. AL is using https://github.com/angellist/boba now.