I've been looking forward to dynamic completion support for a long time. I see ZSH support has landed as an unstable feature so I started playing around with a POC on a small project to get an idea how it all goes together.

So far I have...

1. Enabled features = [ "unstable-dynamic" ] for clap_complete in both my dependencies and build-dependencies.

2. Added a new subcommand to my CLI using the derive macros similar to the example in the docs:

   /// Used internally by Clap to generate dynamic completions
   #[command(subcommand)]
   pub _complete: Option<CompleteCommand>,

3. Edited my main function to intercept ARG[1] being 'complete' and handle dynamic completions:

   let cli = Cli::parse();
   if let Some(completions) = cli._complete {
       completions.complete(&mut Cli::command());
   } else {
       // my original app logic
   }

4. Swapped out the completer on one of my Args:

   /// Optional list of paths to operate on instead of default which is all files tracked by Git
   -#[arg(value_hint = clap::ValueHint::FilePath)]
   +#[arg(add = ArgValueCompleter::new(|| { vec![
   +        CompletionCandidate::new("foo"),
   +        CompletionCandidate::new("bar"),
   +        CompletionCandidate::new("baz")] }))]
   pub paths: Option<Vec<String>>,

This got me as far as being abe to use myapp complete --shell zsh to generate a shell completion function that I could dump to a file manually that later enable dynamic completions.

What I can't figure out is how to use generate_to() (or similar) to output the same completion file with dynamic support at build time:

Am I missing some updated way to do that, or has it just not been implemented yet?