Type analysis on branches is inconsistent depending on whether an explicit return is used

[illicitonion]

I would expect the following two snippets of code to behave the same:

extern crate futures;
use futures::future::{self, Future};

fn main() {
    println!("{}", foo().wait().unwrap());
}

fn foo() -> Box<Future<Item=String, Error=String>> {
    (|| {
        if true {
            Box::new(future::ok("foo".to_owned()))
        } else {
            bar()
        }   
    })()
}

fn bar() -> Box<Future<Item=String, Error=String>> {
    Box::new(future::ok("bar".to_owned()))
}

extern crate futures;
use futures::future::{self, Future};

fn main() {
    println!("{}", foo().wait().unwrap());
}

fn foo() -> Box<Future<Item=String, Error=String>> {
    (|| {
        if true {
            return Box::new(future::ok("foo".to_owned()));
        } else {
            bar()
        }   
    })()
}

fn bar() -> Box<Future<Item=String, Error=String>> {
    Box::new(future::ok("bar".to_owned()))
}

But in reality, the first snippet works fine, and the second fails because the typechecker cannot unify std::boxed::Box<futures::FutureResult<std::string::String, _>> with std::boxed::Box<futures::Future<Item=std::string::String, Error=std::string::String>>. Given that the typechecker can and does unify them in the first case, it seems it should be able to do the same in the second case. Whether I use an implicit or explicit return shouldn't change the type analysis being done.