[RFC] Opt-in generation of empty interfaces of modules without .mli ?

[nojb]

When a module is used solely for their side-effects, one typically wants to add an empty interface file to benefit from the various "unused" warnings, faster incremental compilation, etc.

In our case we have some libraries where we have dozens and dozens of such files. It would be nice if we could have dune generate the empty interface files on demand, eg having a field (modules_without_interface ...) that would tell dune to generate empty interface files for the listed modules (both in libraries and executables).

(In case you are curious, we currently patch the compiler to avoid having to actually create all these empty interfaces, but it seems like a task that would fit nicely to Dune.)

Opinions?

[nojb]

xref #3768

[nojb]

See also the second bullet point in https://discuss.ocaml.org/t/dealing-with-warning-70-missing-mli/8542 where a similar need is expressed.

[bobot]

I think it would be nice. Perhaps another name (modules_with_empty_interface ...). Perhaps even more useful would be a mode where dune automatically put an empty interface if there is no .mli : something like (modules_without_interface_are_empty)

[nojb]

I think it would be nice. Perhaps another name (modules_with_empty_interface ...). Perhaps even more useful would be a mode where dune automatically put an empty interface if there is no .mli : something like (modules_without_interface_are_empty)

Yes, this would work for me as well.

[ghost]

For me, "empty module" means "struct end".

[nojb]

For me, "empty module" means "struct end".

What about (empty_module_interface_if_absent) ? Anyway, regardless of the exact name, is there agreement on the general feature?

[ghost]

The general feature seems fine to me. (empty_module_interface_if_absent) makes sense to me.