Derived Memo not triggering correctly when not rendered

[fdietze]

Describe the bug
A derived Memo doesn't recompute correctly. Rendering it makes it trigger correctly.

Leptos Dependencies

leptos = { version = "0.7.0", features = ["csr", "nightly"] }

To Reproduce

#[component]
fn App() -> impl IntoView {
    let (input, set_input) = signal(0);
    let first_memo = Memo::new(move |_| input() == 1);
    let second_memo = Memo::new(move |_| first_memo());

    let (submitted_value, set_submitted_value) = signal(false);

    let submit = move || {
        set_submitted_value(second_memo());
    };

    view! {
        Press these buttons in order to reproduce the bug:
        <div>
            <button on:click=move |_| { set_input(1) }>1</button>
            <button on:click=move |_| { submit() }>submit</button>
            <button on:click=move |_| { set_input(2) }>2</button>
            <button on:click=move |_| { set_input(3) }>3</button>
            <button on:click=move |_| { submit() }>submit</button>
        </div>
        Input:
        {move || input()}
        <br />
        First Memo:
        {move || first_memo()}
        <br />
        <Show when=|| {
            false
        }>
            Second Memo: {move || second_memo()}<br />
            Enabling this Show, makes the second memo trigger correctly.
        </Show>
        <br />
        Submitted:
        {move || submitted_value()}
    }
}

1. Press the buttons in order.
2. BUG: the submitted value is true, but should be false
3. Change the false in <Show> to true, to render second_memo
4. Press the buttons in order.
5. The submitted value is correct: false

Expected behavior
Derived memos should always have the latest state available when their inputs change.

[gbj]

Thanks!

As a note to myself: the problem occurs specifically when the first memo is triggered twice before the second memo is ever read again, so that by the time .update_if_necessary() runs on the second memo during its read, its parent has already marked itself clean again. The second memo should probably have been marked dirty when the first one changed.

[gbj]

The issue here turns out to be super simple and a regression in my rewrite of the reactive graph for 0.7: the Memo::mark_check function should only mark a Memo as being in the Check state if it isn't already Dirty, which is a stronger claim. The one-line change in the PR (plus regression test) close this issue.