Certain ascii characters crash WASM when received from server fn

[jollygreenlaser]

EDIT: Current fix is to revert to nightly-2024-02-14: rustup default nightly-2024-02-14, may need to rustup target add wasm32-unknown-unknown again and if using cranelift, rustup component add rustc-codegen-cranelift-preview --toolchain nightly-2024-02-14. If using one, eg as done by default with cargo-leptos, update the rust-toolchain.toml file as well.

---

Describe the bug

Latest nightly breaks receiving certain characters from a server function, showing a dmalloc error. For me, it was en dash (–) and likely includes others. Unclear what caused this to suddenly break or what the true origin is.

This happens either upon navigating to the page with the server function call or during hydration when refreshing on that page.

This is very annoying for anything that has to display user entered data, eg profile bios.

Luckily the current workaround is pretty simple, just use this util all over the place:

pub fn clean_ascii(v: &String) -> String {
    v.chars()
        .filter(|c| *c <= '~')
        .collect::<String>()
}

This started happening after upgrading my nightly from late January. Due to testing a separate bug I've been able to repro this bug on 0.6.5, 0.6.6 and the tip of main version I'm on. I have a deployed 0.6.5, late January nightly version without this issue.

It feels very likely that the issue is in a dependency rather than Leptos proper. However, I've not been able to track this down at all despite a lot of effort and I'm guessing others will start to hit it too.

Leptos Dependencies

$ rustup -V
rustup 1.26.0 (5af9b9484 2023-04-05)
info: This is the version for the rustup toolchain manager, not the rustc compiler.
info: The currently active `rustc` version is `rustc 1.78.0-nightly (fc3800f65 2024-02-26)`

[patch.crates-io]
leptos = { git = "https://github.com/leptos-rs/leptos.git", rev = "6d6019b956c81b2ec86e3aab7dd0e93db54dca3e" }
leptos_meta = { git = "https://github.com/leptos-rs/leptos.git", rev = "6d6019b956c81b2ec86e3aab7dd0e93db54dca3e" }
leptos_router = { git = "https://github.com/leptos-rs/leptos.git", rev = "6d6019b956c81b2ec86e3aab7dd0e93db54dca3e" }
leptos_axum = { git = "https://github.com/leptos-rs/leptos.git", rev = "6d6019b956c81b2ec86e3aab7dd0e93db54dca3e" }

https://github.com/leptos-rs/leptos/tree/6d6019b956c81b2ec86e3aab7dd0e93db54dca3e

Screenshot:

Text:

panicked at /rust/deps/dlmalloc-0.2.5/src/dlmalloc.rs:1182:13:
assertion failed: psize <= size + max_overhead

Stack:

Error
    at imports.wbg.__wbg_new_abda76e883ba8a5f (http://172.91.146.59:3000/pkg/local.js:530:13)
    at clepticle.wasm.console_error_panic_hook::hook::h8eb889a93dddf01a (http://172.91.146.59:3000/pkg/local.wasm:wasm-function[3411]:0x103d005)
    at clepticle.wasm.core::ops::function::Fn::call::hdb33dbd3bd9e95b1 (http://172.91.146.59:3000/pkg/local.wasm:wasm-function[15663]:0x130db10)
    at clepticle.wasm.std::panicking::rust_panic_with_hook::h8e9923a1fd637d6a (http://172.91.146.59:3000/pkg/local.wasm:wasm-function[8913]:0x123fd15)
    at clepticle.wasm.std::panicking::begin_panic_handler::{{closure}}::h6dea320ce1182e58 (http://172.91.146.59:3000/pkg/local.wasm:wasm-function[10394]:0x1285d4a)
    at clepticle.wasm.std::sys_common::backtrace::__rust_end_short_backtrace::ha27827269adf0111 (http://172.91.146.59:3000/pkg/local.wasm:wasm-function[21114]:0x134bc43)
    at clepticle.wasm.rust_begin_unwind (http://172.91.146.59:3000/pkg/local.wasm:wasm-function[16307]:0x1317b8b)
    at clepticle.wasm.core::panicking::panic_fmt::h97853b40b106d930 (http://172.91.146.59:3000/pkg/local.wasm:wasm-function[17276]:0x1324e73)
    at clepticle.wasm.core::panicking::panic::h11ea08863e3a8cec (http://172.91.146.59:3000/pkg/local.wasm:wasm-function[15642]:0x130d5b0)
    at clepticle.wasm.__rdl_dealloc (http://172.91.146.59:3000/pkg/local.wasm:wasm-function[13151]:0x12dd02e)

Uncaught (in promise) RuntimeError: unreachable
    at clepticle.wasm.__rust_start_panic (local.wasm:0x134c4c4)
    at clepticle.wasm.rust_panic (local.wasm:0x134abb1)
    at clepticle.wasm.std::panicking::rust_panic_with_hook::h8e9923a1fd637d6a (local.wasm:0x123fd42)
    at clepticle.wasm.std::panicking::begin_panic_handler::{{closure}}::h6dea320ce1182e58 (local.wasm:0x1285d4a)
    at clepticle.wasm.std::sys_common::backtrace::__rust_end_short_backtrace::ha27827269adf0111 (local.wasm:0x134bc43)
    at clepticle.wasm.rust_begin_unwind (local.wasm:0x1317b8b)
    at clepticle.wasm.core::panicking::panic_fmt::h97853b40b106d930 (local.wasm:0x1324e73)
    at clepticle.wasm.core::panicking::panic::h11ea08863e3a8cec (local.wasm:0x130d5b0)
    at clepticle.wasm.__rdl_dealloc (local.wasm:0x12dd02e)
    at clepticle.wasm.__rust_dealloc (local.wasm:0x134a953)

[benwis]

Stuff like this is why I switched to stable

[gbj]

Do you have a minimal reproduction of a server function that can trigger this panic?

[jollygreenlaser]

I'll have one soon.

Further investigation finds this works fine on nightly-2024-02-14 and fails on nightly-2024-02-15

The merged PR is rust-lang/rust#120777 which mentions doing unicode things

[OvermindDL1]

For note, the dash you put in the post here is not ascii (it is indeed a UTF-8 en-dash in this case, but github 'unifies' everything to utf-8 in posts anyway regardless of its source), copy pasted into a rust repl:

>> "–".as_bytes()
[226, 128, 147]

Are you sure that in the page it's utf-8 and not something else like utf-16 or so?

[jollygreenlaser]

Sorry, I'm bad with terminology on ascii vs utf. It is a character, that was given by a server function, that I used to be able to receive ok.

[OvermindDL1]

If you debug in sending it on the server side pre-serialization is it parsed into a rust String and then sent over or is it kept as a javascript string or a byte array or something then sent? Can you make a minimal example case?

[jollygreenlaser]

https://github.com/jollygreenlaser/ascii-death repro

In short:

#[derive(Clone, Debug, Serialize, Deserialize)]
pub struct AsciiDeath {
    pub killer: String, // Also works with Vec<String>
    pub after: bool,    // Can be anything
}

#[server]
pub async fn kill() -> Result<AsciiDeath, ServerFnError> {
    Ok(AsciiDeath {
        // Needs something after it in the string
        // Seems to die on anything € or higher, aka not in character code 32-127
        killer: "€a".to_string(),
        // Needs something after it in the struct
        after: true,
    })
}

#[component]
fn HomePage() -> impl IntoView {
    let (count, set_count) = create_signal(0);
    let on_click = move |_| set_count.update(|count| *count += 1);

    // Can be create_resource or create_local_resource
    // With create_resource the page dies on hydration - the "click me" stops working
    // With create_local_resource it dies on fetch - the "click me" still works
    let deadly_data = create_resource(move || (), |_| async move { kill().await });
    
    // Not technically needed - will die even if this is an empty div
    view! {
        <h1>"Welcome to Leptos!"</h1>
        <button on:click=on_click>"Click Me: " {count}</button>
        <Suspense
            fallback=move || view! { <p>Loading...</p> }
        >
            {move || deadly_data.get().map(|res| match res {
                Ok(res) => {
                    log!("Saw res: {res:?}");
                    view! {
                        <p>{format!("Got data: {res:?}")}</p>
                    }
                },
                Err(err) => {
                    view! {
                        <p>{format!("Had error: {err:?}")}</p>
                    }
                },
            })}
        </Suspense>
    }
}

@OvermindDL1 thanks for the suggestion, will try that next

[jollygreenlaser]

Also happens in Actix: https://github.com/jollygreenlaser/actix-death

[jollygreenlaser]

Further testing shows that string length may have a minimum. What I thought was "requires something after in the struct" may actually be this.

It seems all that matters is number of bytes after the unwanted character.

Replacing kill with:

#[server]
pub async fn kill() -> Result<String, ServerFnError> {
    Ok("€aaaaaaaaaaaaaaa".to_string())
}

Also causes the same crash. Even just one fewer a doesn't crash. No amount of characters before € will cause a crash on their own.

[purung]

For anyone else running into the problem, this seems like a good issue thread.

Wasm-pack issue summary

[SFBdragon]

@jollygreenlaser How can I get ascii-death/actix-death running locally? (I've gotten as far as getting wasm-pack to build it successfully with my instrumented allocator.)

I'm hoping it might be possible to isolate the code producing the bad allocation with such a small repro.