Thanks for the work on this!

There is one point I wasn't clear about in the expectations for this: we should probably keep a completion hidden unless there are no visible choices.

In reviewing this, I also noticed

• I didn't mention hidden aliases
• We aren't showing visible subcommand aliases

I do find it less than ideal how inconsistent we are with where we check is_hide_set. However, I assume cleaning that up would naturally fall out of the above as we'd need to start returning (value, is_hidden, help) (maybe turned into a struct? if so, ideally that refactor would be its own commit) so we can handle all of these cases.

With that said, this is strictly an improvement from where we are at and I'm willing to merge as-is. I would remove the "Closes" so we'd keep that issue open for the rest of this work.

Would you prefer me merging this or wrapping up the above?

If you do go, I would recommend putting the test fixture changes in their own commits, showing the behavior at the time of that commit, with the feature work done in a follow up commit. This makes it a lot easier to review the tests / behavior change as I can more easily see what behavior changed in the feature commits via the test changes.

@epage Thank you for your review! I prefer continuing this before merging.

If you do go, I would recommend putting the test fixture changes in their own commits, showing the behavior at the time of that commit, with the feature work done in a follow up commit.

Makes sense. I'll do that!

I'm curious, what led you to working on this?

@epage Actually I was trying to generate shell completion from clap with dynamic listing, as explained in #1232. Then I found the tracking issue #3166 and found what I can do for that.

As this is a bit stale, I'm going to go ahead and close this. Thank you for your work! Even if this wasn't merged as-is, it can serve as a starting point for you or someone else to pick back up the work!