Skip to content

Overloaded operators can copy self when self is noncopyable. #2548

Closed
eholk opened this Issue Jun 8, 2012 · 9 comments

4 participants

@eholk
eholk commented Jun 8, 2012
resource r<T> (_x: T) { io::println("Goodbye, World!") }

fn main() {
    let mut res = r(5);

    let mut v = [mut];
    v <- [mut res] + v;
}

This outputs Goodbye, World! twice, when it should only do it once.

@eholk
eholk commented Jun 8, 2012

This small variant has the correct behavior.

resource r<T> (_x: T) { io::println("Goodbye, World!") }

fn main() {
    let mut res = r(5);

    let mut v = [mut];
    v <- [mut res];
}
@eholk
eholk commented Jun 8, 2012

This happens because trans does not know to move in tvec::trans_add. I'm attempting to fix this by moving vector addition to libcore.

@eholk eholk was assigned Jun 11, 2012
@eholk
eholk commented Jun 21, 2012

I just landed the "move vector-append out of trans" code, so this should be mostly fixed. Unfortunately, there's a bug in checking self types that means it's still not too hard to copy an uncopyable.

@hatahet
hatahet commented Jul 18, 2012

Are there 2 instances being created?

class r {
  let mut x: int;
  new() {
    self.x = 1;
  }
  drop {
    self.x += 1;
    io::println("self.x: " + int::str(self.x));
  }
}

fn main() {
  let mut res = r();

  let mut _v = ~[mut];
  _v <- ~[mut res] + _v;
}

Output:

self.x: 2
self.x: 2
@eholk
eholk commented Jul 18, 2012

Yeah, that's the problem with this issue. Classes with a destructor are supposed to be non-copyable, but this shows how to trick the type system into copying one for you.

@nikomatsakis

I don't understand, what leads to the copy of self? @eholk can you update the title and maybe add a comment on what precisely is still wrong?

@eholk
eholk commented Aug 3, 2012

@nikomatsakis I'll add an updated test case on this. After thinking about it for a bit, I think this might be a dup of #2587.

@eholk eholk added a commit that referenced this issue Aug 3, 2012
@eholk eholk Adding a test case for #2548 b3933b8
@eholk
eholk commented Aug 3, 2012

Hmm, that test case should actually be a compile-fail test (unless + took ownership of self), since + is defined on vectors of copyable things.

@catamorphism

Should be fixed as of 46990ad

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Something went wrong with that request. Please try again.