Install toplevel printers using [@@toplevel_printer]

[nojb]

Following discussion in #10557, this PR is to discuss the patch proposed in #7770 written originally by @jeremiedimino to automatically install values annotated with [@@toplevel_printer] as toplevel printers.

Fixes #7770

[nojb]

One issue I noticed while playing with this is that that the printer only takes effect after the first time the module (or rather, its .cmi) is loaded:

nojebar@PERVERSESHEAF:~/ocaml/local/bin$ cat foo.mli
type t
val t : t
val print : Format.formatter -> t -> unit [@@toplevel_printer]
nojebar@PERVERSESHEAF:~/ocaml/local/bin$ cat foo.ml
type t = int * int
let t = (12,34)
let print ppf (a, b) = Format.fprintf ppf "(%d,%d)" a b
nojebar@PERVERSESHEAF:~/ocaml/local/bin$ ./ocamlc -c foo.mli foo.ml
nojebar@PERVERSESHEAF:~/ocaml/local/bin$ rlwrap ./ocaml foo.cmo
        OCaml version 4.14.0+dev0-2021-06-03

# Foo.t;;
- : Foo.t = <abstr>
# Foo.t;;
- : Foo.t = (12,34)

This is because the .cmi is loaded when evaluating the first phrase and is only "acknowledged" afterwards (see the second call to acknowledge_new_cmis in the patch).

[nojb]

A second issue, when using #show to show the signature of a module that has not been loaded:

nojebar@PERVERSESHEAF:~/ocaml/local/bin$ rlwrap ./ocaml
        OCaml version 4.14.0+dev0-2021-06-03

# #show Foo;;
module Foo :
  sig type t val t : t val print : Format.formatter -> t -> unit end
Fatal error: exception Topcommon.Undefined_global("Foo")

[Drup]

I'm highly in favor of the feature. I haven't looked at the implementation just yet.

@nojb I don't remember this limitation in the utop implementation, is it there ? If it isn't, why ?

[nojb]

@nojb I don't remember this limitation in the utop implementation, is it there ? If it isn't, why ?

I confess I am not a utop user :) I'll give it a try and report back.

[nojb]

@nojb I don't remember this limitation in the utop implementation, is it there ? If it isn't, why ?

I confess I am not a utop user :) I'll give it a try and report back.

It turns out that in order to implement completion utop eagerly evaluates what you are typing, so by the time you "actually" evaluate the phrase (ie you type ;;) the required .cmi has already been loaded.

[xavierleroy]

I agree it would be nice to have .cma libraries that auto-install printers when loaded in the toplevel REPL. However, I think attaching an attribute to the declaration of the printer function is not the best way to go about it, if only because .cmi files are loaded on demand.

An example for an alternative design: extend the .cma file format (and the ocamlc -a command) so that it can contain a script (source code?) to be executed by the toplevel REPL when the library is loaded.

[dbuenzli]

An example for an alternative design: extend the .cma file format (and the ocamlc -a command) so that it can contain a script (source code?) to be executed by the toplevel REPL when the library is loaded.

The problem with scripts is that they easily go uncompiled. I played with such a scheme in omod where after loading an object file with basename o an existing corresponding o_top_init.ml side file is loaded. Since I never got to complicate my build systems to check that these files evaluate correctly I did sometimes release broken *_top_init.ml files.

I think it's also nicer if these things can be contained to the compilation units they concern. So pushing what you propose a bit further what about:

1. Add an API in Sys that can register printers (in the spirit of Printexc.register_printer) and that the toplevel machinery can redefine with it's own fun to do its magic.
2. Introduce a special function name and sig say val ocaml_toplevel_init : unit -> unit that toplevels calls after they have loaded the compilation unit (and a new cli flag to prevent its evaluation).

This for a module M that would end-up looking something like:

type t = …
…
let pp ppf v = …
let ocaml_toplevel_init () = Sys.register_printer pp

[dbuenzli]

2. Introduce a special function name and sig say val ocaml_toplevel_init : unit -> unit that toplevels calls after they have loaded the compilation unit (and a new cli flag to prevent its evaluation).

Actually if we only care about printers and not making other effects specially for the toplevel we could even eschew the special symbol and boil down the thing to 1. and maybe a new flag in ocaml and load directives to prevent printer install. Then things boil down to:

type t = …
…
let pp ppf v = …
let () = Sys.register_printer pp

And that Sys.register_printer is a nop in regular programs.

[nojb]

Actually if we only care about printers and not making other effects specially for the toplevel we could even eschew the special symbol and boil down the thing to 1. and maybe a new flag in ocaml and load directives to prevent printer install. Then things boil down to:

type t = …
…
let pp ppf v = …
let () = Sys.register_printer pp

And that Sys.register_printer is a nop in regular programs.

This seems like a simple solution that does not require any changes to the infrastructure.

[nojb]

1. Add an API in Sys that can register printers

One issue is that the printer-registering machinery in the Toplevel needs to have access to the concrete type of the printer being installed, and Sys.register_printer would have a polymorphic type: (Format.formatter -> 'a -> unit) -> unit.

An example for an alternative design: extend the .cma file format

What about extending the .cmo file format to record the list of toplevel printers (assuming they are annotated with an attribute like [@@toplevel_printer] in the implementation file)?

[dbuenzli]

an attribute like [@@toplevel_printer] in the implementation file)?

It seems that's what @let-def does in https://github.com/let-def/autoprinter

[nojb]

I would still like to have something like this, but the patch doesn't work in its current form. Closing.