Yet Another CLI Library (well, not really)
https://discuss.ocaml.org/t/cmdlang-yet-another-cli-library-well-not-really/15258
Greetings fellow camlers,
I hope you had a nice summer! Mine took an unexpected turn when, roughly at the same time, I wrote my first cmdliner subcommand and heard about climate for the first time. My experience with OCaml CLI so far had been centered around core.command.
When I read climate's terminology section and how it defines Terms, Arguments, and Parameters, something clicked. Seeing how climate's API managed to make positional and named arguments fit so nicely together, I thought: "Wow, for the first time, it seems I'll be able to write a CLI spec on a whiteboard without referring to some code I never seem to get right (I am looking at you, core.command's anonymous arguments)."
I got quite excited and thought: "Can I switch to climate today?" But reality checked: it's not on opam yet, still under construction, I'm not sure what the community will do, etc.
Implementing my own engine for an API resembling climate felt like a wasted effort, knowing about the work happening in climate. Still, having a 'a Param.t, 'a Arg.t, and 'a Command.t type that I would get to know and love felt too good to pass up.
I stared at the climate types for a while, and filled with happy thoughts about a bright CLI future, it occurred to me: can I use an API like climate but compile it down to existing libraries such as cmdliner or core.command? (and climate too!). I wrote down the following types:
climate
'a Param.t -> 'a Climate.Arg_parser.conv
'a Arg.t -> 'a Climate.Arg_parser.t
'a Command.t -> 'a Climate.Command.t
cmdliner
'a Param.t -> 'a Cmdliner.Arg.conv
'a Arg.t -> 'a Cmdliner.Term.t
'a Command.t -> 'a Cmdliner.Cmd.t
core.command
'a Param.t -> 'a core.Command.Arg_type.t
'a Arg.t -> 'a core.Command.Param.t
unit Command.t -> core.Command.t
... which I interpreted as stating the following theorem:
There exists an abstraction to encode OCaml CLIs that lives in the intersection of what's expressible in other well established libraries.
"One EDSL to command them all," so to speak. I couldn't resist the temptation to build actual terms for these types. That gave birth to cmdlang.
As a test, I switched one of my projects to cmdlang, with cmdliner as a backend. I liked the changes I made in the process. The 1-line bin/main.ml is now the only place that specifies which backend I want to use; the rest of the code is programmed solely against the cmdlang API. This means I'll be able to easily experiment with compiling down to climate in the future.
I am not against the multiplicity of solutions in general, but I tend to feel uneasy when incompatible libraries emerge, partitioning the ecosystem. As a community, we know too many examples of this. In this instance, I want to call the core.command vs cmdliner situation a ... cli-vage.
I don't see my work on cmdlang as competing with these other libraries. Quite the contrary, it makes it easier for me to experiment with them without much changes while exploring the subject of CLI in general. Also, as a library author, if you wish to expose CLI helpers to your users, a library like cmdlang will give you a pleasant way to do so, as you can express your helpers with it, knowing your users will have the choice to translate them to the backend of their choice.
Before writing this post, I had a very pleasant chat with @gridbugs. I want to make it clear that I do not think cmdlang is competing with climate either. I think climate is a very promising library and I believe it will, in due time, deliver auto-completion to many - this has been a highly anticipated feature within the community. I wish to dedicate the initial work that I did on cmdlang to @gridbugs due to the impactful influence climate had on my work, and how it helped me improve my general understanding of declarative CLI libraries.
These are very early days for cmdlang. There are still areas I am fuzzy on, and I haven't really validated the whole design yet. I have put some thoughts in this Future Plans page. One thing that I did not initially include there would be to explore the feasibility of writing a mini-compiler for cmdlang targeting stdlib.arg as a runner. I am not sure how much you'd end up restricting cmdlang for this to work. I thought that'd be a fun project to tackle at a future point, as it would make a nice addition to the overall architecture of the project.
I'd welcome your input to help me shape the future of cmdlang if you have an interest in this project.
Thanks for reading!