Why so many body types?

[khvzak]

Hello,
Out of curiosity, why ntex (+actix-web) has many different body types?
Eg:
http::body::Body
http::client::MessageBody
http::Payload

maybe the same question for request/response types:
http::Request
http::client::ClientRequest
web::WebRequest

http::Response
web::WebResponse
http::client::ClientResponse

Just to compare, axum/hyper uses a single type, from the http crate.

Thanks!

[fakeshadow]

because real world http types are complicated. In order to express them you either have a single type with compromise or use specialized types for low overhead.

let's take a very common use case for example:

struct Reject<S> {
    service: S,
}

// ignore the unconstraint B type.
impl<S,  B> Service<Request> for Reject<S>  
where
    S: Service<Request, Response = Response<B>>
{
    async fn call(&self, req: Request) -> Result<Response<B>, Error> {
        if req.path != "/" {
            return Ok(Response::new(<what_type_is_this>))
        }  
        self.service.call(req).await
    }
}

in ntex you can express it with Body type as it has explicit variants for Bytes. You can also nest Body types with the help of MessageBody trait object variant if you want to.
but in axum you would likely be told to use Box<dyn HttpBody> for the Bytes which is an additional heap allocation.

[fakeshadow]

As for Payload type the benefit of separating request/response body type is mainly you can relax the Unpin bound. From server's view request body type can be always Unpin while response body type can be !Unpin for special body use. This is not something often used but it leaves the possibility open. From client's view this still holds in reverse order.

[khvzak]

Thank you!