How to tell when a statement is in data/transformed data/parameters/transformed parameters, etc?

[SteveBronder]

I'm looking at pp_statement trying to make some changes when a Decl is happening, but I only want to make this change whenever I am declaring parameters. Is there information within pp_statement that I can use to tell that I am in the parameters block? If I can then I think I can do a small trick like #865 to make one copy of the parameters on the AD stack. So if someone did someting like

matrix[N, M] transform_x = multiply(data_matrix,, param_vector)

The only memory creation that happens here is for the new matrix transform_x

[nhuurre]

There's not. That information is in program.output_vars only.
You could either have a new function like

let pp_toplevel_statement (params : String.Set.t) ppf stmt =
  match stmt.Stmt.Fixed.pattern with
  | Decl {decl_adtype; decl_id; decl_type} when Set.mem params decl_id ->
      pp_decl ppf (decl_id, decl_type, decl_adtype)
  | _ -> pp_statement ppf stmt

or it might make sense to add a new is_mappable : bool field to Decl in middle/Stmt.ml.

[SteveBronder]

I sort of got something that does this, but it feels hacky

https://github.com/stan-dev/stanc3/compare/master...SteveBronder:feature/no-param-setup?expand=1

And it also breaks transform_inits() so need to wait on Ryan's PR for cleaning that up before we think more about this (which is waiting on a PR in the Stan repo).

I think what I actually want is a DeclAssign that can take where stanc breaks up the deceleration and assignment into a Decl and Assign right now like

      Eigen::Matrix<local_scalar_t__, -1, 1> p_vec;      
      current_statement__ = 9;
      p_vec = in__.template read_constrain_lb<Eigen::Matrix<local_scalar_t__, -1, 1>, jacobian__>(
                0, lp__, N);

So we can can just do this

Eigen::Matrix<local_scalar_t__, -1, 1> p_vec = 
  in__.template read_constrain_lb<Eigen::Matrix<local_scalar_t__, -1, 1>, jacobian__>(0, lp__, N);

Then once I can tell I'm in the parameters block with that DeclAssign I can do

auto p_vec = in__.template read_constrain_lb<Eigen::Matrix<local_scalar_t__, -1, 1>, jacobian__>(0, lp__, N);

Then in deserializer when the input type is a var I can just return an arena matrix so no copies happen for matrix and vector parameters when it's used in other functions.