Panic in rune-internal tracing logging while formatting anyhow `Result` wrapped in rune `Value`

[VorpalBlade]

Panic in rune-internal tracing logging while formatting anyhow `Result` wrapped in rune `Value`

[VorpalBlade]

Before I switched from log to tracing for my project I would get this when attempting to unwrap the returned result in my user code. I came up with this monstrosity as a workaround. But this doesn't help now that it crashes internally in tracing.

/// Attempt to format the error in the best way possible.
///
/// Unfortunately this is awkward with dynamic Rune values.
fn try_format_error(phase: Phase, e: &rune::Value) -> anyhow::Result<()> {
    match e.clone().into_any() {
        rune::runtime::VmResult::Ok(any) => {
            match any.clone().downcast_into_ref::<anyhow::Error>() {
                Ok(err) => {
                    anyhow::bail!("Got error result from {phase}: {:?}", *err);
                }
                Err(err) => {
                    let ty = try_get_type_info(e, "error");
                    anyhow::bail!(
                        "Got error result from {phase}, but it was not an anyhow error: {err:?}, type info: {ty}: {any:?}",
                    );
                }
            }
        }
        rune::runtime::VmResult::Err(not_any) => {
            tracing::error!("Got error result from {phase}, it was not an Any: {not_any:?}. Trying other approches at printing the error.");
        }
    }
    // Attempt to format the error
    let formatted = catch_unwind(AssertUnwindSafe(|| {
        format!("Got error result from {phase}: {e:?}")
    }));
    match formatted {
        Ok(str) => anyhow::bail!(str),
        Err(_) => {
            let ty = try_get_type_info(e, "error");
            anyhow::bail!(
                "Got error result from {phase}, but got a panic while attempting to format said error for printing, {ty}",
            );
        }
    }
}

/// Best effort attempt at gettint the type info and printing it
fn try_get_type_info(e: &rune::Value, what: &str) -> String {
    match e.type_info() {
        rune::runtime::VmResult::Ok(ty) => format!("type info for {what}: {ty:?}"),
        rune::runtime::VmResult::Err(err) => {
            format!("failed getting type info for {what}: {err:?}")
        }
    }
}

[udoprog]

Well, Value returns fmt::Error if debug printing errors. For non-builtin values, this would be the case if it's called outside of a vm.

For some reason we've also opted to instrument the return value here. This is not correct and should be removed since formatting might fail.

It is convenient for Value to implement fmt::Debug, but maybe it shouldn't return fmt::Error and instead debug print the errors that occurred to avoid issues like these. It's also a footgun since it definitely will not work unless called inside a vm externally with Vm::with.

[VorpalBlade]

It is convenient for Value to implement fmt::Debug, but maybe it shouldn't return fmt::Error and instead debug print the errors that occurred to avoid issues like these. It's also a footgun since it definitely will not work unless called inside a vm externally with Vm::with.

My understanding is that std::fmt::Error is for when the formatting itself fails. And the Rust standard library will in fact panic in format!() when this happen.

So if I understand you correctly and it error when printing an error (as opposed when to the formatting fails) this seems incorrect and like a misunderstanding.

[udoprog]

Yeah, we should fix it.

[VorpalBlade]

I'll make a PR later today