[2026-06-26]

- annotations do not work properly because macroexpansion creates an entirely new body for lambdas
- we realized that we are handling macro application in two separate places: pit_expand_macros expands lambda bodies, and pit_eval expands macros encountered while evaling. we ought to unify this so that only one is used (probably pit_expand_macros, because we need to expand macros eagerly to identify free variables to capture)
- we probably can make pit_expand_macros and pit_eval much nicer
- we can probably make pit_expand_macros operate in place
- if we want to be really smart, cool, happy, rich:
  let's just make stuff translate to a little VM before it evaluates, and let's store VM programs as functions instead of sexps

* little vm thing
let's have a "program" (representation of one function) be a cons-list of instructions
during execution, we maintain a stack of programs.
calling a lisp function entails pushing a new "frame" to this stack
we iterate over the frame by cdring as usual and we pop the frame when we reach the end

#+begin_src pit
(defun! add2 (x y) (+ x y 2))
#+end_src

#+begin_src pit
[
  (literal x)
  (get)
  (literal y)
  (get)
  (literal 2)
  (literal +)
  (fget)
  (apply 3)
]
#+end_src

#+begin_src pit
(defun! foo (x) (* (add2 x x) 4))
#+end_src

#+begin_src pit
[
  (literal x)
  (get)
  (literal x)
  (get)
  (literal add2)
  (fget)
  (apply 2) ;; evaluating this pushes a new frame with the code for add2. next "tick" of evaluator continues in that code. once it returns, result is on the stack (for free) 
  (literal 4)
  (literal *)
  (fget)
  (apply 2)
]
#+end_src