Launching Rocket without codegen

[bubelov]

Are there any examples of manually starting a server, without code generators? I'd like to make an unified binary which accepts and parses command line arguments before deciding what to do. Starting a server is only one of the possible actions, so I need full control over the main function in order to figure out if user wants to start a server or do something else

[baptiste0928]

You can manually launch a Rocket instance with the launch() method :

#[rocket::main]
async fn main() {
    rocket::build()
        .mount("/hello", routes![world])
        .launch()
        .await;
}

More information : https://rocket.rs/v0.5-rc/guide/overview/#launching

[bubelov]

@the10thWiz yea, that sounds like a path of least resistance. Thank you both for your suggestions!

[SergioBenitez]

It is specified in the source code that the function is unstable and is not intended to be used outside Rocket. Otherwise, it is always possible to build yourself the tokio runtime man_shrugging

Please, please don't do this. This is a private function that is pub because Rust forces us to make it so. Don't ignore the docs:

/// WARNING: This is unstable! Do not use this method outside of Rocket!

Please don't suggest contradicting the docs.

[baptiste0928]

Please don't suggest contradicting the docs.

Okay, sorry about that. I edited my answer to avoid confusion.

[cmcqueen]

Is it possible to use #[rocket::main] on a function called by main() and not on main() itself?

[the10thWiz]

Is it possible to use #[rocket::main] on a function called by main() and not on main() itself?

Yes, however I believe the function generated by rocket::main is still called main. Placing it in a submodule allows you to use a different function as the actual main function, and call the generated function.

Something like this is the best option:

fn main() {
   // Check CLI Args, etc
   server::main();
}

mod server {
   #[launch]
   fn rocket() -> _ {
      rocket::build()
   }
}

If the rocket server isn't the main function of the program, it should probably be placed in a separate file anyway, which makes this type of change much simpler. Perhaps a parameter could be added to the main annotation to specify a different function name to generate, if this is intended to be a supported use case. Something like #[rocket::main(name = "start_rocket")] would make this easier to handle.

The other issue is that Rust only allows a single async runtime to be started. This means that this approach only works for running Rocket 0.5 from within sync code - meaning async Rust applications can't support this type of use case. A more complete solution would depend on Rust making async runtimes easier to swap out, which I believe is currently dependent on standardizing the AsyncWrite and AsyncRead traits. I don't expect this to happen anytime soon, and I don't see a way to support starting Rocket from within async code. Perhaps in the future Rocket could support running on an existing tokio runtime, but I don't think this is worthwhile yet.