Hello! So spark is primarily designed for helping to build “structural” DSLs. I.e defining data.

For example, in Ash Framework, on an Ash.Resource

attributes do
  attribute :name, :string, allow_nil?: false
end

That is a section called :attributes, with an entity called :attribute. We then extract this using:

Ash.Resource.Info.attributes(module, :attributes)

which gives us, for example:

%Attribute{
  name: :name,
  type: :string,
  allow_nil?: false
}

(the examples are greatly simplified from real life).

So the DSL simply holds the description. It’s just “something else” (whatever else you want) that contains the “actual functionality” of the DSL. You can leverage it at compile time to do code generation, or at runtime to interpret the data structure however you like.

If you want things to happen “in the module”, then you can use transformers, i.e

defmodule YourTransformer do
  use Spark.Dsl.Transformer

  def transform(dsl_state) do
    {:ok, 
      Spark.Dsl.Transformer.eval(dsl_state, [], quote do
         ...your quoted code here
      end)
    }
  end
end

If you’re looking to write something like a single macro that contains some kind of customized imperative/declarative/expression language, then spark is unlikely to help you.

i.e if you’re looking to make something like NX or something like Ecto’s Ecto.Query macros.