*.ml & interface file *.mli

[Fourchaux]

Let's have a foo.ml file and no corresponding foo.mli.
Being able to obtain this interface file
(as the result of ocamlc -i foo.ml)
would be nice.

[raphael-proust]

Can't you just set up your editor to do that?
In vim:
:command GenerateMli !ocamlc -i % > %:r.mli

A more advance/complete solution would check if the interface file exists before writing to it. Also it could use ocamlfind to make the necessary packages available.

Do you have a use case where you specifically want Merlin to do that rather than your editor?

[Fourchaux]

Sure we can set up every editor usable with Merlin (the list is growing ... )
but if Merlin can give this directly it's even better, no ?
The tuareg mode (I'm using Spacemacs) has a function 'tuareg-find-alternate-file'
coming from 'caml-find-alternate-file' of the caml mode, I suppose,
which ,in our case, does a checking and creates a file foo.mli.
This new file is always empty, so having the result of 'ocamlc -i' seems
eventually more interesting.

[raphael-proust]

If implemented in Merlin, each of the editor support file would need to include bindings for the functionality. It is probably easier to code the functionality in OCaml and provide bindings for each editor. In this sense, it is probably better.

(Although I'm not sure what would happen when editing a remote file (by which I do not mean remote editing a file, but rather have the local editor open a remote file). That could be very editor specific.)

[let-def]

But Merlin knows more about build setup, ocamlc -i will fail in all but trivial cases.

[MiloDavis]

I've started some work on implementing this for values. I currently envision this as a way to add/update a toplevel definition in the mli. I'm envisioning something that operates as follows (in emacs at least):

• User puts cursor on a toplevel definition
• User triggers the command (keybinding etc)
• Merlin generates a type and jumps to the location of the exported definition in the mli file if it exists. If it does not exist, the user will be asked if they want to create one.
• Merlin will write the val <identifier> : syntax to the buffer
• Merlin will then prompt the user for each type in the definition that has aliases available in scope. This means that if a type can be written as int or t, the user will be asked to select between them. I'll have at least an option to use only the local types and possibly some other heuristics.
• Any type definitions from the module referenced in the type will be automatically added to the mli. The user can choose to make them abstract (using yes-or-no-p in emacs)

Does this sound like a reasonable approach? Any advice on implementing this would be appreciated.

[rgrinberg]

What about we just add custom completion to the .mli? E.g. after a user types in val |, merlin will complete values and their types from the implementation.