fix(codegen): unwrap NewTypes before classifying return types for field selection

[mike1o1]

tl;dr

Fix missing call to unwrap new types, so fields get properly added/generated. Basically a one-line change - everything else is just tests.

Closes #59

Summary

When a generic action returns a NewType that wraps :map with field constraints (e.g., {:array, MyApp.Types.MySuggestion}), the generated TypeScript function is missing the fields parameter. However, the runtime RPC pipeline correctly requires field selection, resulting in a "requires_field_selection" error at runtime.

Root Cause

ActionIntrospection.check_action_returns/1 in action_introspection.ex does not call Introspection.unwrap_new_type/2 before classifying the return type. The NewType module (e.g., MyApp.Types.MySuggestion) is not in @field_constrained_types ([Ash.Type.Map, Ash.Type.Keyword, Ash.Type.Tuple]), so classify_return_type/2 returns {:error, :not_field_selectable_type}.

Meanwhile, the runtime field processing in field_selector.ex (line 116) does properly unwrap NewTypes before classification, sees the underlying Ash.Type.Map with field constraints, and demands field selection — creating the mismatch.

How to Reproduce

1. Define a NewType wrapping a map with field constraints

defmodule MyApp.Types.MySuggestion do
  use Ash.Type.NewType,
    subtype_of: :map,
    constraints: [
      fields: [
        name: [type: :string, allow_nil?: false],
        category: [type: :string, allow_nil?: true],
        score: [type: :integer, allow_nil?: false]
      ]
    ]
end

2. Define a generic action returning an array of this type

defmodule MyApp.MyResource do
  use Ash.Resource, ...

  actions do
    action :search, {:array, MyApp.Types.MySuggestion} do
      argument :query, :string, default: ""

      run fn input, _context ->
        {:ok, [%{name: "test", category: nil, score: 1}]}
      end
    end
  end
end

3. Expose via typescript_rpc

typescript_rpc do
  resource MyApp.MyResource do
    rpc_action :search, :search
  end
end

4. Run codegen and observe

The generated TypeScript function will NOT include a fields parameter:

// Generated — missing `fields`
export async function search(config: {
  input?: SearchInput;
  // No fields parameter here!
}): Promise<SearchResult> { ... }

5. Call the RPC and get runtime error

{
  "type": "requires_field_selection",
  "message": "%{field_type} requires field selection",
  "shortMessage": "Field selection required",
  "vars": { "fieldType": "Field_constrained_type" }
}

Proposed Fix

Add Introspection.unwrap_new_type/2 call in check_action_returns/1 before classification:

# lib/ash_typescript/rpc/codegen/helpers/action_introspection.ex

defp check_action_returns(action) do
  {base_type, constraints, is_array} = unwrap_return_type(action)

  # FIX: Unwrap NewTypes before classification, matching field_selector.ex behavior
  {unwrapped_type, unwrapped_constraints} =
    Introspection.unwrap_new_type(base_type, constraints)

  case classify_return_type(unwrapped_type, unwrapped_constraints) do
    {:resource, resource} ->
      type = if is_array, do: :array_of_resource, else: :resource
      {:ok, type, resource}

    {:typed_map, fields} ->
      type = if is_array, do: :array_of_typed_map, else: :typed_map
      {:ok, type, fields}

    {:typed_struct, value} ->
      type = if is_array, do: :array_of_typed_struct, else: :typed_struct
      {:ok, type, value}

    :unconstrained_map ->
      type = if is_array, do: :array_of_unconstrained_map, else: :unconstrained_map
      {:ok, type, nil}

    {:error, reason} ->
      {:error, reason}
  end
end

This mirrors the pattern already used in field_selector.ex (line 116) and ensures codegen and runtime agree on whether field selection is required.

---

Contributor checklist

Leave anything that you believe does not apply unchecked.

• I accept the AI Policy, or AI was not used in the creation of this PR.
• Bug fixes include regression tests
• Chores
• Documentation changes
• Features include unit/acceptance tests
• Refactoring
• Update dependencies

[Torkan]

Thanks a bunch! 🤩