Global arguments are silently ignored with nested subcommands

[kuecks]

Please complete the following tasks

• I have searched the discussions
• I have searched the open and rejected issues

Rust Version

rustc 1.58.0-nightly (65f3f8b22 2021-11-21)

Clap Version

3.0.14

Minimal reproducible code

use clap::{Parser, Args, Subcommand}; // 3.0.14

#[derive(Parser, Debug)]
struct Options {
    #[clap(subcommand)]
    command: Command1,
}

#[derive(Subcommand, Debug)]
enum Command1 {
    A(Command1AOptions),
}

#[derive(Args, Debug)]
struct Command1AOptions {
    #[clap(subcommand)]
    command: Command2,
    #[clap(long, global = true)]
    flag1: bool,
}

#[derive(Subcommand, Debug)]
enum Command2 {
    B {
        #[clap(subcommand)]
        command: Command3,
        #[clap(long, global = true)]
        flag2: bool,
    },
}

#[derive(Subcommand, Debug)]
enum Command3 {
    C(Command3COptions),
}

#[derive(Args, Debug)]
struct Command3COptions {}

fn main() {
    let options = Options::parse();
    dbg!(options);
}

Steps to reproduce the bug with the above code

cargo run -- a b c --flag2 --flag1

This passes both global flags at the lowest level (flag1 is declared with a, flag2 is declared with b)

Actual Behaviour

flag1 is correctly set to true, but flag2 is false!

    Finished dev [unoptimized + debuginfo] target(s) in 0.04s
     Running `target/debug/clap_test a b c --flag2 --flag1`
[src/main.rs:42] options = Options {
    command: A(
        Command1AOptions {
            command: B {
                command: C(
                    Command3COptions,
                ),
                flag2: false,
            },
            flag1: true,
        },
    ),
}

Expected Behaviour

Both flag1 and flag2 should be true

Additional Context

No response

Debug Output

debug.txt

[epage]

I think I've identified root cause and am working on the fix. Our recursive discovery of global flags appears to be broken,

[kuecks]

@epage Wow that was very fast! Thank you very much!

[epage]

The fix was fast but the part that might be slow is the release. I generally aim to release as soon as a user-facing change is in but I'm batching up some deprecations, so focusing on finish that first. Sorry for the delay.