DAP server: Improved handling of Rust `panic` in the API.

[noppej]

DAP server: Improved handling of Rust `panic` in the API.

[noppej]

As an experiment I implemented first std::panic::catch_unwind and then std::thread::spawn in probe_rs::cmd::dap_server, and they both behave pretty much the same. It protects the debugger from exiting on a panic, but the information available to the end user isn't very useful. See the following code snippet, as well as the documented TODO for why this was abandoned.

pub fn run(cmd: Cmd, time_offset: UtcOffset) -> Result<()> {
    let log_info_message = setup_logging(time_offset)?;

    // The debug server runs as a background process that the VSCode extension depends on.
    // If the server panics, the VSCode extension will not be able to communicate with it,
    // and the user will have no idea what happened.
    // The debug server and associated code relies on standard Rust error handling, and will
    // report `Err` values to the user, in appropriately formatted messages.
    // However, the various crates used by the debug server may panic, so we catch those panics
    // here, and report them to the user via the logging mechanism, which is configured in `setup_logging`,
    // to ensure the VSCode extension can pick up the error and report it to the user.
    match thread::Builder::new()
        .name("probe-rs DAP server".to_string())
        .spawn(move || debug(cmd.port, cmd.vscode, &log_info_message, time_offset))
    {
        Ok(dap_server_handle) => {
            if let Err(err) = dap_server_handle.join() {
                tracing::error!(
                    "Debug server thread panicked. Please report this as a bug: {err:?}"
                );
                // TODO: The value reported for `err` is not very helpful, because it is a `Box<dyn Any + Send + 'static>`.
                // [This RFC](https://github.com/rust-lang/rust/issues/96024) when it has run its course,
                // we can something like the proposed `err.request_value()` on the backtrace.
            }
        }
        Err(e) => {
            tracing::error!("Failed to spawn probe-rs DAP server thread: {e:?}");
        }
    };

    Ok(())
}

Coincidentally, the work that was done in probe-rs/vscode#63 and #1699, effectively means that Rust's panic message and stack trace, will be displayed to the user. In other words, this opens up the possibility to allow the use of panic!, expect!, todo!etc. in the API, and still provides an acceptable user experience.

Based on that, I will close this issue as completed.

[Yatekii]

I think I would still implement this because it lets you shutdown gracefully instead of crashing the DAP server :)

[noppej]

The code posted above is done and tested and can be implemented. However, the current version of Rust obscures the original panic message and stack trace. Until this RFC completes, the current experience is still better (more informative for the user). We can keep this issue open until then :)

[Yatekii]

Uhm, I am pretty sure it should be possible to log the panic :)

[noppej]

It is currently being logged, but as soon as you hide it behind a spawn or a catch_unwind, it is lost. I imagine that is why the RFC is there. If I'm mistaken, I'd love you to show me the way :)