`Instruction does not dominate all uses` LLVM error in a pattern match with a ref pattern and dereference of a box

[omasanori]

The functions below are OK:

fn foo1(x: Option<Box<int>>) -> Option<int> {
    let res = match x {
        Some(ref v) => v.clone(),
        None => return None
    };
    Some(*res)
}
fn foo2(x: Option<Box<int>>) -> Option<int> {
    let res = match x {
        Some(ref v) => Some(*v.clone()),
        None => return None
    };
    return res;
}

but:

fn foo3(x: Option<Box<int>>) -> Option<int> {
    let res = match x {
        Some(ref v) => *v.clone(),
        None => return None
    };
    Some(res)
}

produces a LLVM ERROR. The message is:

Instruction does not dominate all uses!
%8 = load i64** %2
%12 = bitcast i64* %8 to i8*
LLVM ERROR: Broken function found, compilation aborted!

I'm sorry if it isn't a minimum example of this issue.

[nixpulvis]

On a related note (I think)

This works:

  fn tail(&self) -> LinkedList {
    match *self {
      Cons(_, ref l) => {
        *l.clone()
      },
      _ => fail!("called `tail` on Empty"),
    }
  }

But this doesn't with the same LLVM error.

  fn tail(&self) -> LinkedList {
    match *self {
      Cons(_, ref l) => *l.clone(),
      _ => fail!("called `tail` on Empty"),
    }
  }

Instruction does not dominate all uses!
  %12 = load %"enum.data::list::LinkedList<[]>"** %2
  %23 = bitcast %"enum.data::list::LinkedList<[]>"* %12 to i8*
LLVM ERROR: Broken function found, compilation aborted!

cc @luqmana

It's very subtle.

The following ICEs:

#[deriving(Clone)]
enum LinkedList { Cons(Box<uint>), Nil }

impl LinkedList {
    fn tail(self) -> uint {
        match self {
          Nil => 1u,
          Cons(l) => *l.clone()
        }
    }
}

fn main() { }

But this is okay:

#[deriving(Clone)]
enum LinkedList { Cons(Box<uint>), Nil }

impl LinkedList {
    fn tail(self) -> uint {
        match self {
          Nil => 1u,
          Cons(l) => (*l.clone()) // Notice the extra parens.
        }
    }
}

fn main() { }

Seems specific to Box that still has special treatment in cleanup.

[steveklabnik]

#18845 is a dup that has a smaller test case, so I'm giving this a close in favor of that.