Usability in utop: must manually load various stuff before using the ppx

[Armael]

Hi,

I think this is more a kind of "usability bug": in order to make the following declaration work in utop "type t = int * int [@@deriving sexp];;", I figured out by trial and error that I had to first, in this order:

#require "ppx_deriving";;
#require "sexplib";;
#require "ppx_sexp_conv";;
open Sexplib.Conv;;

The open Sexplib.Conv bit being the most puzzling imo (without it it complains that int_of_sexp is unbound). Could this be shortened/documented a bit? I my head, a #require "ppx_deriving";; #require "ppx_sexp_conv";; should be sufficient...?

Cheers,
Armaël

[ghost]

Note: you shouldn't have to require "ppx_deriving" manually, can you confirm it works without?

ppx_sexp_conv can probably require sexplib. I'm not sure what to do about the open Sexplib.Conv though, as something the basic converters will come from somewhere else. For instance if you are using core then open Core.Std will bring into scope all the standard converters.

[Armael]

Without requiring ppx_deriving, I get:

# #require "sexplib";;
# #require "ppx_sexp_conv";;
# open Sexplib.Conv;;
# type t = int * int [@@deriving sexp];; 
Cannot locate deriver sexp

[ghost]

Not sure what's happening. @whitequark, any idea why we need to load ppx_deriving first?

[whitequark]

Try utop -dsource:

# #require "ppx_sexp_conv";;
[@@@findlib.ppxopt ("ppx_deriving", "/home/whitequark/.opam/4.02.3/lib/ppx_optcomp/./ppx_optcomp.cma")];;

[@@@findlib.ppxopt ("ppx_deriving", "/home/whitequark/.opam/4.02.3/lib/ppx_driver/./ppx_driver.cma")];;

[@@@findlib.ppxopt ("ppx_deriving", "/home/whitequark/.opam/4.02.3/lib/ppx_sexp_conv/./ppx_sexp_conv_expander.cma")];;

[@@@findlib.ppxopt ("ppx_deriving", "/home/whitequark/.opam/4.02.3/lib/ppx_type_conv/./ppx_type_conv.cma")];;

[@@@findlib.ppxopt ("ppx_deriving", "/home/whitequark/.opam/4.02.3/lib/ppx_type_conv/./ppx_type_conv_deriving.cma")];;

[%%ocaml.error
  let _ =
    "Cannot load /home/whitequark/.opam/4.02.3/lib/ppx_sexp_conv/./ppx_sexp_conv.cmxs: error loading shared library: /home/whitequark/.opam/4.02.3/lib/ppx_sexp_conv/./ppx_sexp_conv.cmxs: undefined symbol: camlPpx_sexp_conv_expander"
  let _ = ""];;

#require  "ppx_sexp_conv";;

[whitequark]

Now if you #require "ppx_deriving" first:

# #require "ppx_deriving";;
#require  "ppx_deriving";;
# #require "ppx_sexp_conv";;
;;
;;
;;
;;
;;
;;
#require  "ppx_sexp_conv";;

[whitequark]

@diml oh, I see. Note how e.g. ppx_core supplies the ppxopt(-ppx_driver) = "ppx_deriving,./whatever.cma" option, but does not require ppx_deriving. This means that ppx_deriving might well not be loaded until right before ppx_sexp_conv itself is loaded, and then the ppxopt is lost.

[ghost]

I see, my bad. I'll add the dependency.

Actually that's a problem I had while creating the packages and wanted to come back to: ppx_core (and a few other packages) are general purpose libraries that are not related to ppx_deriving or ppx_type_conv, so they should have a META file similar to any library. However there is currently no way to specify a dependency from a ppx_deriving plugin to a normal library. For instance I can't write a ppx_deriving plugin that would use pcre.

[whitequark]

That's a very good point. I think ppx_deriving should use findlib to discover the package archives. I think I used to link findlib before? I don't remember what made me switch to pure-dynlink.

[ghost]

Yeah I think it would be easier if ppx_deriving was using findlib

[agarwal]

This is a still a bit frustrating. You have to load ppx_deriving first. If you load ppx_sexp_conv first, there's no way out except to restart utop.

utop # #require "ppx_sexp_conv";;
utop # type t = int [@@deriving sexp];;
Cannot locate deriver sexp

(* Now you google for this error and end up here. You realize you need to load ppx_deriving, so you try it. *)

#require "ppx_deriving";;
utop # type t = int [@@deriving sexp];;
Cannot locate deriver sexp

utop # #require "ppx_deriving";;
utop # #require "ppx_sexp_conv";;
utop # type t = int [@@deriving sexp];;
Cannot locate deriver sexp

Finally, restart utop and do the last sequence, and then it works.

[whitequark]

Yeah, unfortunately I haven't had time yet to fix that.

[ghost]

This is fixed with ppx_sexp_conv 113.24.00

[Armael]

Ok, thanks!