as I'm sure you're aware, bytecode interpretation typically implies a superior performing model than AST interpretation, and compiling into bytecode produces a

as I'm sure you're aware, bytecode interpretation typically implies a superior performing model than AST interpretation, and compiling into bytecode produces a lot of opportunities for optimization that are not typically feasible when working with an AST directly. Of course it's all bits and anything is possible, but it's assumed to be a better approach in a generally non-subtle way.