assertion failed: !bcx.ccx().maps.moves_map.contains(&arg_expr.id)

[ssbr]

Sorry, since I don't know what causes this, can't be more specific, and don't know if the other bug reports (currently 4 others) are related or have distinct causes.

Other issues with the same error:

• unique pointers + macros → rust: task failed at 'assertion failed: !bcx.ccx().maps.moves_map.contains(&arg_expr.id)' #6000
• ICE with compound assignment and Drop::finalize #5959
• Assignment operator for BigInt causes ICE #6102
• NumCast::from() crashes compiler #6440

This is as small as I can get the code:

use core::num::Zero;

pub trait MyTrait {
    pub fn method(&mut self);
}

pub struct MyStruct<U> {
    pub attr: U,
}

pub impl<U: Num> MyStruct<U> {
    pub fn other_method(&mut self) {
        self.attr += Zero::zero();
    }
}

impl MyTrait for MyStruct<f64> {
    pub fn method(&mut self) {
        self.other_method();
    }
}

fn main() {
    let x = MyStruct {attr: 0.0f64};
    let _ = ~x as ~MyTrait;
}

[osaut]

The ICE seems to be related to the line

self.attr+= Zero::zero();

as replacing it with

self.attr=self.attr+Zero::zero();

works.

The code no longer compiles on recent master, here is an updated version

use std::num::Zero;

pub trait MyTrait {
    pub fn method(&mut self);
}

pub struct MyStruct<U> {
    pub attr: U,
}

impl<U: Num> MyStruct<U> {
   pub fn other_method(&mut self) {
        self.attr = self.attr+ Zero::zero();
   }
}

impl MyTrait for MyStruct<f64> {
   pub fn method(&mut self) {
        self.other_method();
    }
}

fn main() {
    let x = MyStruct {attr: 0.0f64};
    let _ = ~x as ~MyTrait;
}

[huonw]

This and #6000 both seem to have the common property of evaluating a generic/non-implicitly-copiable value twice (if I understand how += currently works). I don't know the details, but, in both cases, I'd guess the problem is because the things should be duplicated (with new id's and so) to be legally evaluated twice, and this isn't happening.

[emberian]

@osaut's updated version compiles and runs, and since then += has been removed

[pnkfelix]

it seems like this bug is no longer actionable, since the original test case relies on +=, and the updated version does not exhibit the bug according to @cmr. Closing.