feat(completions): Dynamic fish completions

[ModProg]

Supersedes #4177.

Part of #3917.

[ModProg]

@epage is there any e2e tests for dynamic completions yet? I only found them for the "legacy" completions, bash dynamic completions only asserts the generated completion script AFAICT:

clap/clap_complete/tests/testsuite/bash.rs

Lines 99 to 115 in 74109e5

	#[cfg(feature = "unstable-dynamic")]
	#[test]
	fn register_minimal() {
	use clap_complete::dynamic::Completer;
	let name = "my-app";
	let bin = name;
	let completer = name;
	let mut buf = Vec::new();
	clap_complete::dynamic::shells::Bash
	.write_registration(name, bin, completer, &mut buf)
	.unwrap();
	snapbox::Assert::new()
	.action_env("SNAPSHOTS")
	.matches_path("tests/snapshots/register_minimal.bash", buf);
	}

And the tests for dynamic only includes tests without the shells:
https://github.com/clap-rs/clap/blob/master/clap_complete/tests/testsuite/dynamic.rs

I'd extend the utilities in common for dynamic completion testing? And add some basic tests for bash and fish?

[ModProg]

I joined the example into one file, though some of the workarounds feel a bit hacky.

One thing is, that when the unstable-dynamic feature is removed we'd need to use a different cfg for the example, but that is probably fine.

To not conflict with the non dynamic build, the dynamic binary needs to be compiled in a different folder.

[ModProg]

I also had to rename test because of it's name conflict.

[ModProg]

https://github.com/clap-rs/clap/actions/runs/5694718476/job/15436420390#step:9:125

[epage]

My PR has been merged

[ModProg]

Not sure about the sorting to make bash test work, probably a bit hacky.

[ModProg]

1. I'd like to add help messages for the completions, this could probably be done by having the complete function return pairs of completion and help message.

2. I'd like to only complete flags when the current argument starts with --. Should I add an argument to complete that enables this? Or should I just enable that by default?

3. Concerning path completions. I'm not sure what the best way to handle these in a fish compatible way would be. I could enable glob support in our path completions, though that is pretty complex for fish completions:

   • expand globs: */ba would complete to */bash.rs and */banana.rs (given the paths a/bash.rs, a/banana.rs, and b/bash.rs exist)
   • complete substrings (not only prefixes) rc/he/ba would complete to src/shell/bash.rs.

Or we could query the shell to handle this for us, and we'd only do filtering afterwards, i.e. complete can take an IntoIterable<PathBuf> and then filter them accordingly.

[ModProg]

Is there a place to document degradations?

I.e. ways the dynamic completions worsen user experience compared to static completions.

For fish taht would currently be:

• Globs not working in path completions
• short flags not displaying their long counter parts can only be solved by convincing fish (Dynamic Options for Completion fish-shell/fish-shell#7943)

[epage]

Not sure about the sorting to make bash test work, probably a bit hacky.

Can you remove it from this PR? This is focused on fish support and as we've both seen, there seem to be complications with bash testing.

... list of changes...

I'd recommend creating issues for these so we can discuss and resolve them independently

Is there a place to document degradations?

The tracking issue.

[epage]

Thanks!

[epage]

btw I rewrote the PR description to not close #3917 since it looked like you were treating that as a mini-tracking issue. I leave it to you on whether you actually want to close that (and use the main tracking issue) or keep it open after this is merged

[ModProg]

looked like you were treating that as a mini-tracking issue.

I think it makes sense to collect the shell specific things in the fish issue, though the generic things such as the option filtering or help displaying should probably also be added to the main tracking issue.

[epage]

Thanks for your patience through all of the back and forth in getting this in!