Zero sized DST struct item causes invalid free

[talchas]

struct Node<T: ?Sized>(T);

fn main() {
   let x: Box<Node<[isize]>> = box Node([]);
}

seems to result in a call to foo::heap::exchange_free(1, 0, 1) in the drop glue (and no allocates), which results in a crash. Just doing Box<[isize]> works, as do non-empty input arrays.

[alexcrichton]

cc @nrc

[nrc]

cc @eddyb should this have been fixed by #23376? Or did #23376 cause a regression?

[eddyb]

@nrc it appears to have been fixed, this example doesn't crash in my local branch (assuming it's not something else I've done since).

[fhahn]

I have added 2 tests based on @talchas example, one using the box syntax and one using Box::new. Interestingly, using Box::new causes a compile time error:

issue-23491-2.rs:5:41: 5:49 error: mismatched types:
     expected `Node<[isize]>`,
     found `Node<[isize; 0]>`
 (expected slice,
    found array of 0 elements) [E0308]
issue-23491-2.rs:5    let x: Box<Node<[isize]>> = Box::new(Node([]));
                                                           ^~~~~~~~
issue-23491-2.rs:5:32: 5:40 error: the trait `core::marker::Sized` is not implemented for the type  `[isize]` [E0277]
issue-23491-2.rs:5    let x: Box<Node<[isize]>> = Box::new(Node([]));
                                                   ^~~~~~~~
issue-23491-2.rs:5:32: 5:40 note: `[isize]` does not have a constant size known at compile-time
issue-23491-2.rs:5    let x: Box<Node<[isize]>> = Box::new(Node([]));
                                                    ^~~~~~~~
error: aborting due to 2 previous errors

Shouldn't they both behave similar in this case?

[apasel422]

@fhahn's test with Box::new doesn't compile because the argument to that function must be Sized. However, the code does compile and run successfully with the box syntax, so a PR containing src/test/run-pass/issue-23491.rs
from fhahn@e8d7122 should close this issue.