Why does HttpServer take a closure instead of an App instance?

[weezy20]

#[actix_rt::main]
async fn main() -> std::io::Result<()> {
    HttpServer::new(
        || App::new()
            .service(web::resource("/").to(|| HttpResponse::Ok())))
        .bind("127.0.0.1:59090")?
        .run()
        .await
}

Why does HttpServer take a App factory and calls the very same closure on each worker thread? I am trying to understand this design, as one could very well instantiate one App struct and then Clone it across the various worker threads. Since app_data requires all data to be Copy, Clone, Send, Sync, I was under the impression that App could be cloned and sent across threads as well.

/// The top-level builder for an Actix Web application.
pub struct App<T> {
    endpoint: T,
    services: Vec<Box<dyn AppServiceFactory>>,
    default: Option<Rc<BoxedHttpServiceFactory>>,
    factory_ref: Rc<RefCell<Option<AppRoutingFactory>>>,
    data_factories: Vec<FnDataFactory>,
    external: Vec<ResourceDef>,
    extensions: Extensions,
}

Is it because the vecs in here have something that aren't Cloneable or Sendable?

[fakeshadow]

actix-web uses Send + Clone factory types to construct !Send + !Sync types.
For example:

use std::rc::Rc;

use actix_web::{App, HttpServer};

fn main() {
    let rc = Rc::new(123);
    HttpServer::new(move || {
        let _ = rc.clone(); // this would not work.
        let rc = Rc::new(123); // this would.

        App::new()
    });
}

app_data also does not require any bound you just mentioned. This code works just fine:

use std::sync::Mutex;

use actix_web::{App, HttpServer};

fn main() {
    HttpServer::new(|| {
        App::new()
            .app_data(ptr())// !Send + !Sync
            .app_data(Mutex::new(())) // !Clone + !Copy
    });
}

fn ptr() -> *mut String {
    todo!()
}

In other word App is used to construct a per thread application instance. You can opt in to use thread safe smart pointers to make these application instances point to the same data.

[weezy20]

Are you sure about the app_data not requiring Clone/Copy? Last time I checked, app_data(arg) would not accept a type that was arg: !Clone type and so we would have to use something like an Arc.

More specifically arg is either a PgConnection/PgPool from sqlx.

[fakeshadow]

Just run the code I posted and see for yourself.

[Abdillah]

Are you sure about the app_data not requiring Clone/Copy?

It is the HttpServer closure that require Clone. I'm searching the discussion about how to pass from the outside non cloneable (like Mutex above) to the inside.