auto-completions for bb.cli/dispatch

[Sohalt]

You can test it with

# install test script
bbin install io.github.sohalt/bbct

# install completions

# bash
bbct --org.babashka.cli/completion-snippet bash >> ~/.bashrc

# zsh
# ~/.zsh needs to be on the $fpath variable for zsh
# if it's not, add this to your ~/.zshrc before a line with "compinit":
# fpath+=(~/.zsh)
bbct --org.babashka.cli/completion-snippet zsh > ~/.zsh/_bbct

# fish
bbct --org.babashka.cli/completion-snippet fish > ~/.config/fish/completions/bbct.fish

[borkdude]

@Sohalt Trying to test:

bbct --babashka.cli/completion-snippet zsh > /usr/local/share/zsh/site-functions/_bbct

I'm getting:

zsh: no such file or directory: /usr/local/share/zsh/site-functions/_bbct

[Sohalt]

The installation for zsh can vary a bit, depending on how your distro packages zsh.
Try echo $fpath from within zsh and put the _bbct file somewhere in a directory on that path.

[Sohalt]

Also I just rebased on main and force pushed a new version of bbct with the latest sha

[borkdude]

What is the recommended approach for this Rust-based CLI package? I suppose they must have some user-friendly docs for this?

[Sohalt]

They use
$ your_program --bpaf-complete-style-zsh > ~/.zsh/_your_program (https://docs.rs/bpaf/latest/bpaf/_documentation/_2_howto/_1_completion/index.html)
Though that depends on ~/.zsh/ being on your $fpath (e.g. on my system ~/.zsh/ doesn't exist).

[Sohalt]

What definitely should work is this:

1. mkdir ~/.zsh
2. Add fpath+=($HOME/.zsh)" in your .zshrc before a line that says compinit or similar.
3. Then bbct --babashka.cli/completion-snippet zsh > ~/.zsh/_bbct.

[borkdude]

Nice! First working result after doing that:

[Sohalt]

Tests are failing due to line endings on Windows. Does any of bash, zsh or fish even work on Windows? What's the situation there when using something like Cygwin?

[Sohalt]

Current limitations:

• only works with cli/dispatch
• only works with gnu-style options (starting with -or --)
• doesn't work well with options that take lists (e.g. :coerce [:string])

[borkdude]

We might want to add Powershell completions or so in the future, but let's not worry about that for now. Does the example Rust library support this? You can disable those tests for Windows

[Sohalt]

Does the example Rust library support this?

no afaics

You can disable those tests for Windows

how?

[borkdude]

By wrapping the whole test body in:

(when-not (fs/windows?) ...)

[Sohalt]

By wrapping the whole test body in:

(when-not (fs/windows?) ...)

Sometimes things can be so simple :) I thought there was some yaml programming involved to configure the test runner or similar 😅

[borkdude]

I prefer not to call System/exit in function that are supposed to be non-side-effecting. I'd rather throw exceptions or whatever

[Sohalt]

We need to short circuit all following execution in this case though, since we're printing the shell completion snippet and any other code running after this might print stuff that messes this up. It only happens when you pass the internal --babashka.cli/completion-snippet opt, so regular user code should never hit this.

[borkdude]

You don't need to short circuit if you just put the logic after all the parsing logic though?
Also removing the System/exit stuff makes it more REPL friendly

[Sohalt]

If I put the code after the parsing logic the consuming code could still print extra stuff. e.g.

(defn -main [& args]
  (cli/parse-opts args opts)
  (println "this does not belong in the completions script"))

[Sohalt]

The alternative would be to make it the responsibility of the user to call print-completion-shell-snippet (and make sure nothing else is printed in that case)

[borkdude]

Well, we control format-opts so we can decide to filter those options out, right?

[Sohalt]

Sure, but then you'll have a format-opts, that magically swallows some options.
I also don't particularly like the magic in parse-opts, but at least that means you get completions without any setup on your part.
Changing format-opts means you still have magic and need to do setup.

[Sohalt]

I'm also ok with just providing the complete * functions and having the user wire everything up themselves. Then it's even more work, but absolutely no magic.

[Sohalt]

Well, we control format-opts so we can decide to filter those options out, right?

Or do you mean the caller filters them out?

[borkdude]

magically swallows some options

The fact that org.babashka.cli/... options are internal, makes this non-magic imo. Users of this library should never make up options with that namespace themselves

[borkdude]

Same here