Add zsh and bash completions for x

[clubby789]

Suggest the x subcommands, then suggest the paths they may be invoked with. We can then point users to these scripts in the dev guide. There is probably a better way to implement these, but I'm unfamiliar with shell completion APIs.

The content of src/etc/completions/*_paths is from the output of x <cmd> -h -v, hardcoded to avoid the shell having to invoke x whenever the completion script is loaded.

@rustbot label +T-bootstrap

[rustbot]

r? @Mark-Simulacrum

(rustbot has picked a reviewer for you, use r? to override)

[jyn514]

The content of src/etc/completions/*_paths is from the output of x <cmd> -h -v, hardcoded to avoid the shell having to invoke x whenever the completion script is loaded.

Hmm, I don't know how I feel about this. I'd like to have an upgrade path in case we add more paths at a later date.

cc @thomcc, I think you had some plans around this.

[clubby789]

I did wonder if a tidy check might help with this. Alternatively, just changing the argument to use more basic patterns might make it simpler to maintain, as well as running a bit faster - tab completing the tests here is noticably slower than the default shell completion as it processes the globs.

[Mark-Simulacrum]

I don't think we should be creating lists like this - part of the reason we aim for paths to be the inputs is that shells already autocomplete those.

Can we generate these on the fly perhaps? It seems like e.g. a cache based on git HEAD might work okay here and not cause too many problems.

At minimum, if we do need hardcoded lists, they should be verified and bless'able like formatting.

[thomcc]

For fish-shell I started on this but never got around to completing it. That said, for fish a hard-coded path list wasn't needed, I just parsed the list that was in help output. It was fairly ad-hoc but worked. In general I figured that since not many people use fish-shell, the only way it would be upstreamable is if it were small and not problematic.

[jyn514]

another alternative is to autocomplete only the subcommands (build/test/...) and treat any following arguments as a path for the purposes of completion.

cc #107375 which proposed switching to clap, that would autogenerate these completions for us.

[clubby789]

Resolving as a path would probably be the easiest and quickest way to do this. Do shells have the ability complete files relative to the source root though? Otherwise tab completion wouldn't work from a subdirectory, which was the main motivation for doing it this way

[jyn514]

Can you spawn a subshell and generate completions from there? not sure if there's a way to pass env variables back to the parent shell but you could do COMPREPLY=(cd $(git rev-parse --show-toplevel) && ...) I think.

[clubby789]

This seems to be working better. Bash completion isn't great for large amounts of files (fills the screen and partially overwrites its own output) but I'm not sure if that's an issue with the completion or bash just not handling it well

[Mark-Simulacrum]

Bash completion isn't great for large amounts of files (fills the screen and partially overwrites its own output) but I'm not sure if that's an issue with the completion or bash just not handling it well

This seems pretty unfortunate. Maybe we can limit to a subset (e.g., head -n20) or something?

Are these completions enabled by default for x? What do users need to do to enable them?

[albertlarsan68]

Is it possible to also generate/have completions for NuShell please?

[clubby789]

Are these completions enabled by default for x? What do users need to do to enable them?
Enabling completions will need users to run a command to load the completions for the current shell instance, or added to bash/zshrc (which we can mention in the dev guide)

I was hoping to replace this with #108083, since clap_complete automatically generates completions (and there seems to be an extension for NuShell) but some bugs may need to be fixed in clap_complete to work with the layout of xs arguments

[Mark-Simulacrum]

@rustbot author

[clubby789]

Closed by #111388