Can you expand on a use case where this bites a user? Do we actually end up passing an empty list to supports_feature in some case?

Can you expand on a use case where this bites a user? Do we actually end up passing an empty list to supports_feature in some case?

No, this doesn't affect functionality in any way.
There are only two places where this function is currently used:

1. In get_grammars, which simply returns all grammars if target_versions is empty - no call to supports_feature is made.
2. In _format_str_once, where supports_feature is called with versions obtained from detect_target_versions. detect_target_versions cannot possibly return an empty set (if it does that would indicate some issue elsewhere), so again, no issue here.

The misleading behavior of this function ended up wasting quite a bit of my time while working on #4720, so I figured it’s worth fixing to save others the trouble down the line.

If the function is never called with an empty value, and is never intended to be called with an empty value, let's throw a meaningful error instead of returning False.

P.S. The CI failure does seem unrelated.