You might be entirely correct. I noticed a small perf issue with the backwards scanning and had it rewrite it not using rfind("_"). First time I prompted it, though, it didn't actually manage to fix the problem and gave me some mild gaslighting until I told it to explicitly remove the rfind() call. Now I think the result should be comparable to or beat a regexp, but it is getting quite a bit more complex:

    fn strip_suffix(s: &str) -> String {
        let mut idx = s.len();
        for (i, c) in s.char_indices().rev() {
            if c == '_' {
                idx = i;
                break;
            } else if !c.is_ascii_digit() {
                return s.to_string();
            }
        }
        if idx == s.len() {
            return s.to_string();
        }
        if s[idx+1..].chars().all(|c| c.is_ascii_digit()) {
            return s[..idx].to_string();
        }
        s.to_string()
    }

.char_indexes().rev() there worries me a bit as well now that I look at it... haven't tested that at all.