A better way of raising errors from this might be to introduce a new RuleResult variant that immediately terminates parsing instead of carrying on with alternatives. I think that's a great future improvement, but didn't consider for this PR because it would expand the scope quite a bit.

I don't think it's workable to treat reaching the stack limit as a simple parse failure because it will cause /, &, and ! operators to misbehave when reaching the limit. That's going to end up sometimes resulting in "successful" parses with unexpected parse trees. Possibly could even be weaponized to let an attacker skip parts of a grammar that involve deeper rule nesting, which could bypass data validation and produce a result that shouldn't be allowed.

For example, with the grammar

peg::parser! { grammar test() for str {
    stack_limit 2;

    rule keyword() -> Expr
        = "foo" { Expr::Foo }
        / "bar" { Expr::Bar }

    pub rule expr() -> Expr
        = "(" e: expr() ")" { e }
        / keyword()
        / v:$(['a'..='z']+) { Expr::Variable(v.to_string()) }
}}

the input ((foo)) incorrectly returns Expr::Variable("foo") instead of Expr::Foo because the call to keyword() fails with the stack limit.