VecDeque::push_back(()) does not work correctly.

[dpc]

http://is.gd/NY4tCl

use std::collections::VecDeque;

fn main() {
    let mut v = VecDeque::new();
    v.push_back(());                                                                                                         
    assert!(v.len() > 0 );   
}

thread '<main>' panicked at 'assertion failed: v.len() > 0', <anon>:6

[agbaroni]

() is void type / value, so it should be correct; if you add nothinh to an empty collection, the collection remains empty ...

[utkarshkukreti]

VecDeque::push_front() too:

use std::collections::VecDeque;

fn main() {
    let mut v = VecDeque::new();
    v.push_front(());                                                                                                         
    assert_eq!(v.len(), 1);   
}

fails with

thread '<main>' panicked at 'assertion failed: `(left == right)` (left: `2`, right: `1`)', <anon>:6

[jonas-schievink]

@ituxbag No, the collection just contains one nothing ;)

() is just a zero-sized type, adding it to collections stores 0 bytes, but 1 element. For example, it works fine with aVec<()>:

fn main() {
    let mut v = Vec::new();
    v.push(());
    v.push(());
    assert_eq!(v.len(), 2);
}

[steveklabnik]

/cc @rust-lang/libs this is interesting.

[apasel422]

This variant appears to loop forever, presumably in drop since it outputs 2 (!) first:

use std::collections::VecDeque;

fn main() {
    let mut v = VecDeque::new();
    v.push_front(());
    println!("{}", v.len());
}

[bluss]

Let's aim to improve the testsuite with general collections + Zero sized types tests too.

[bluss]

Ok, Zero sized types (ZSTs) are a can of worms, potentially there may be several problems, but so far I found that the VecDeque in Rust 1.3 with ZST uses capacity = !0 which is not a power of two. Now how to fix this..

[Gankra]

I would recommend isize::MAX + 1, giving a capacity of isize::MAX elements -- I'm becoming more swayed that this is a reasonable upper bound to select unconditionally.

[Gankra]

Oh, right this capacity is inherited from RawVec! Annoying, but not a big deal.

[bluss]

This used to work in Rust 1.0, 1.1, 1.2. Regression report, ad-hoc via travis

First broken release: Rust 1.3.0.

[bluss]

@dpc As soon as a new nightly is built, it would be great if you could verify that the bug was fixed for the original problem.

[dpc]

Sure. I'll try that tomorrow.

[bluss]

@dpc Any luck?

[dpc]

The original problem reported confirmed it's fixed in: https://github.com/dpc/mioco/issues/55#issuecomment-142004340

[bluss]

Thank you! Good to have some confidence so that we can backport the fix to beta.