Using &mut Trait requires mut local variable for some cases

[mukilan]

Looks like some cases were not tested in #8398

The following code:

use std::rt::io::Reader;
use std::rt::io::extensions::ReaderUtil;


fn parse(reader: &mut Reader) {
    reader.read_byte();
}

fn main() {
}

still yields a compilation error:

test.rs:6:4: 6:10 error: cannot borrow immutable argument as mutable
test.rs:6     reader.read_byte();

[mukilan]

Issue still present in master. Updating code to compile.

trait Foo {
   fn foo(&mut self);
}

impl<'a> Foo for &'a mut Foo {
    fn foo(&mut self) { self.foo(); }
}

struct S {
    i :int
}


impl Foo for S {
    fn foo(&mut self) {
        self.i = self.i + 1;
    }
}

trait Bar {
    fn bar(&mut self);
}

impl<T: Foo> Bar for T {
    fn bar(&mut self) {
        self.foo();
    }
}

fn main() {
    let s = &mut S{i:0} as &mut Foo;
    s.foo(); // works fine
    s.bar(); // error: cannot borrow immutable local variable as mutable

    let mut s = &mut S{i:0} as &mut Foo;
    s.bar(); // works fine

}

[pnkfelix]

reading the updated example, I do wonder if this is "working as designed" ... I'm not 100% sure yet (I haven't tried playing around with it locally).

(I'm also not sure if that impl<'a> Foo for &'a mut Foo is quite right ... does it infinite loop in its current state, once you get main to compile? You might need (**self).foo() or something along those lines instead.)

[mukilan]

@pnkfelix : You are right. It does "work as designed". Here's my explanation as to why. Please correct me if I'm wrong. When s.foo() invoked, since s is &mut Foo, the compiler can invoke foo() using the vtable pointer in s. However, when s.bar() is invoked, the compiler has no way of knowing the right trait Bar vtable to use, given that "s" is just &mut Foo . Hence the need for the impl<'a> Foo for &'a mut Foo, which can now be used with impl<T: Foo> Bar for T to create the Bar vtable with the self type as &mut Foo .But since the method is bar(&mut self) , the actual type of the first parameter is &mut &mut Foo and hence s is required to be mutable.

For the original code I submitted in the issue (the one with the &mut Reader), I was told in IRC that it was probably a bug and was asked to raise an issue. But even that was working per design, if the the above explanation is valid and correct.

Shall I close this issue?

[steveklabnik]

Looks like this should be closed.