buggy code generated when initializing a [Vec<_>;..]

[steven807]

The following code non-deterministically either works, goes into an infinite loop, or seg-faults, on both my mac environment and on https://play.rust-lang.org/, in all three channels. However, only in Debug mode -- it works fine in all three cases in Release. It also works fine if the "uninitialized()" is replaced by "zeroed()".

use std::mem;

fn main() {
    let _:[Vec<i32>;2] = {
        let mut arr: [Vec<i32>; 2] = unsafe { mem::uninitialized() };
        arr[0] = Vec::new();
        arr[1] = Vec::new();
        arr
    };
}

Meta

rustc 1.1.0 (35ceea399 2015-06-19)
binary: rustc
commit-hash: 35ceea3997c79a3b7562e89b462ab76af5b86b22
commit-date: 2015-06-19
host: x86_64-apple-darwin
release: 1.1.0

rustc 1.2.0-beta.4 (089f99c1c 2015-07-31)
binary: rustc
commit-hash: 089f99c1c40bd13d432fba7928bfefbbac13b3fa
commit-date: 2015-07-31
host: x86_64-apple-darwin
release: 1.2.0-beta.4

rustc 1.3.0-nightly (83b9deaf2 2015-08-01)
binary: rustc
commit-hash: 83b9deaf24200f6e41b1aec6fe090094f0190a7a
commit-date: 2015-08-01
host: x86_64-apple-darwin
release: 1.3.0-nightly

[Gankra]

Your code is invoking undefined behaviour. You can't overwrite mem::uninitialized() types that need_drop without using ptr::write/copy/copy_nonoverlapping. Otherwise Rust will try to drop the "old" values which are just uninitialized memory.

[steven807]

Ah, of course. Sorry for wasting time. I suspected the issue was that it was related to "unsafe", but I'm still new to all of this.

[Gankra]

See https://doc.rust-lang.org/nightly/adv-book/unchecked-uninit.html for detailed docs on this precise issue.

[Gankra]

(note to future readers, if that link becomes broken, it's because the URL has changed. It might be at https://doc.rust-lang.org/nightly/nomicon/unchecked-uninit.html someday)

[Gankra]

I think there's something to be said for fleshing out the docs of https://doc.rust-lang.org/std/mem/fn.uninitialized.html