fix: Require validation not bypassed by mutually exclusive groups

[12yanogden]

Problem

Fixes #4707

Previously, arguments required via requires could be bypassed when they conflicted with other present arguments in mutually exclusive groups.

Example Issue:

#[derive(Parser)]
#[clap(group = ArgGroup::new("command").multiple(false))]
struct Args {
    #[clap(long, group = "command")]
    read: bool,
    
    #[clap(long, group = "command")]  
    write: bool,
    
    #[clap(long, requires = "read")]
    show_hex: bool,
}

[epage]

While this might fix the immediate problem, I'm worried that this is too brittle and we can change things in other code and not realize this needs to be update as well.

[epage]

Note that #4520 is the core issue and it hasn't been accepted yet, requiring further analysis for correctness and whether this would constitute a breaking change.

For future contributions, please keep in mind

• We prefer to resolve issues before moving on to PRs
• Commits should be atomic, including tests passing
• We do ask for tests to be added in commits before the behavior changes but that is to show the existing behavior
• derive tests should go in the derive testsuite, not the builder one
• tests should be grouped by and named for their functionality, not issue numbers

See also https://github.com/clap-rs/clap/blob/master/CONTRIBUTING.md

[12yanogden]

So wait, is 4707 not a workable issue? I have to wait until 4520 is resolved?

[epage]

They are the same concept (how should requires interact with conflicts) and that question needs to be addressed before either moves forward. Rather than split that conversation between two related issues, I recommend we centralize it on #4520. That doesn't mean that #4520 needs to be implemented first. In fact, its likely that a fix for one will fix the other or be just a one or two line change.