// Regression test for Issue 25199: Check that one cannot hide a // destructor's access to borrowed data behind a boxed trait object. // // Prior to fixing Issue 25199, this example was able to be compiled // with rustc, and thus when you ran it, you would see the `Drop` impl // for `Test` accessing state that had already been dropping (which is // marked explicitly here with checking code within the `Drop` impl // for `VecHolder`, but in the general case could just do unsound // things like accessing memory that has been freed). // // Note that I would have liked to encode my go-to example of cyclic // structure that accesses its neighbors in drop (and thus is // fundamentally unsound) via this trick, but the closest I was able // to come was dropck_trait_cycle_checked.rs, which is not quite as // "good" as this regression test because the encoding of that example // was forced to attach a lifetime to the trait definition itself // (`trait Obj<'a>`) while *this* example is solely use std::cell::RefCell; trait Obj { } struct VecHolder { v: Vec<(bool, &'static str)>, } impl Drop for VecHolder { fn drop(&mut self) { println!("Dropping Vec"); self.v[30].0 = false; self.v[30].1 = "invalid access: VecHolder dropped already"; } } struct Container<'a> { v: VecHolder, d: RefCell>>, } impl<'a> Container<'a> { fn new() -> Container<'a> { Container { d: RefCell::new(Vec::new()), v: VecHolder { v: vec![(true, "valid"); 100] } } } fn store(&'a self, val: T) { self.d.borrow_mut().push(Box::new(val)); } } struct Test<'a> { test: &'a Container<'a>, } impl<'a> Obj for Test<'a> { } impl<'a> Drop for Test<'a> { fn drop(&mut self) { for e in &self.test.v.v { assert!(e.0, e.1); } } } fn main() { let container = Container::new(); let test = Test{test: &container}; //~^ ERROR `container` does not live long enough println!("container.v[30]: {:?}", container.v.v[30]); container.store(test); }