FR: add a subcommand for generating completions

[ethanpailes]

Apparently clap can do this automatically. We should look into setting it up and then add a usage section about how to set completions up.

discussion with some more context: #289

[ethanpailes]

This issue is pretty stand alone, so it would be a good one to start with if someone wanted to take it on. Does not really require knowing anything about shpool's internals.

[injust]

See fish-shell/fish-shell#12302 (comment) for a suggested patch (in the collapsible "Details" section).

[ethanpailes]

I haven't reviewed that patch super closely, but I think I noticed some use of unstable features, which makes me uncomfortable. shpool is maintained on a best-effort basis and I don't want to be in a position where the only options are dropping a feature and breaking backwards compatibility or re-implementing a now-deprecated or dropped feature that an upstream library decided to abandon. Maybe it's possible to use clap's completion support without relying on unstable features though.

[krobelus]

Right, clap's dynamic completions are still unstable.
Dynamic means that the shell will run something like SHPOOL_COMPLETE=fish shpool -- $arg1 $arg2 to compute completions,
i.e. it more or less minimizes shell-specific code.
IIRC clap's static completions are stable and can generate completions for --someflag but I don't think they can complete session names.
But that can be provided with a little bit of shell-specific glue logic, so I don't see a blocker to using static completions for now.

I haven't worked on it yet because I wanted a more general solution clap-rs/clap#5784 .. but static completions would be a strict improvement, and we could even keep the interface stable when we move to dynamic ones later.

[ethanpailes]

Sounds like static completions for now and then dynamic completions once those are stable in clap is the way to go then.