Segfault using result of place_back() as a Place

[andy-hanson]

When trying to write a data structure that wraps a Vec and is a Placer, I got a segfault when trying to use the result of place_back() as a Place.

I tried this code:

#![feature(collection_placement)]
#![feature(placement_in_syntax)]
#![feature(placement_new_protocol)]

use std::ops::Placer;
use std::vec::PlaceBack;

fn main() {
	let mut stack = Stack::new();
	&mut stack <- 1;
	println!("Doesn't get here!");
}

pub struct Stack(Vec<u32>);
impl Stack {
	pub fn new() -> Self {
		Stack(Vec::new())
	}
}
impl<'a> Placer<u32> for &'a mut Stack {
	type Place = PlaceBack<'a, u32>;

	fn make_place(self) -> Self::Place {
		self.0.place_back() //.make_place()
	}
}

I expected to see this happen: Prints "Doesn't get here!", and exits with code 0.

Instead, this happened: Prints "Segmentation fault", and exits with code 139.

The error is fixed by uncommenting .make_place().

Meta

rustc --version --verbose:

rustc 1.22.0-nightly (fd4bef5 2017-09-15)
binary: rustc
commit-hash: fd4bef5
commit-date: 2017-09-15
host: x86_64-unknown-linux-gnu
release: 1.22.0-nightly
LLVM version: 4.0

Backtrace:

N/A?

[arielb1]

I don't think it's an essential unsoundness - PlaceBack is for some reason both a Placer and a Place, and if you call make_place and pointer in the wrong order, you can access the vec out of bounds and segfault.

[alexcrichton]

cc #27779, the tracking issue for this as well

[mattico]

Is the fix for this just to make separate types for Placer and Place so that you have to call make_place to actually get a Place?

[leoyvens]

Triage: placement new has been unaccepted, this issue can be closed.