Failed asserting on Select drop

[ghost]

This is reproduce able 100% with the following code. Looking at the disassembly the problem is that one of the handles drop function is called after the select drop is called.

use std::comm::{Port, Chan, Select};

fn wait_commands(cmd: &mut Port<Port<()>>, ports: &mut ~[Port<()>]) -> Port<()>
{
    let mut handles = ~[];
    let select = Select::new();
    let mut cmd_handle = select.add(cmd);
    for port in ports.mut_iter() {
        handles.push(select.add(port));
    }
    loop {
        let id = select.wait();
        if id == cmd_handle.id {
            let port = cmd_handle.recv();
            println!("{:?}", port);
            return port;
        }
        for h in handles.mut_iter() {
            if h.id == id {
                break;
            }
        }
    }
}

fn thread(cmd: Port<Port<()>>)
{
    let mut cmd = cmd;
    let mut ports: ~[Port<()>] = ~[];
    loop {
        let port = wait_commands(&mut cmd, &mut ports);
        println!("{:?}", port);
        ports.push(port);
    }
}

fn main()
{
    let (p, c) = Chan::new();
    spawn(proc() {
        thread(p);
    });
    let (new_p0, new_c0): (Port<()>, Chan<()>) = Chan::new();
    let (new_p1, new_c1): (Port<()>, Chan<()>) = Chan::new();
    c.send(new_p0);
    c.send(new_p1);
    new_c0.send(());
    new_c1.send(());
}

output

std::comm::Port<()>{queue: SPSC(std::sync::spsc_queue::Consumer<(),std::comm::Packet>{state: std::sync::arc::UnsafeArc<std::sync::spsc_queue::State<(),std::comm::Packet>>{data: (0x7f435c001640 as *mut ())}}), marker: std::kinds::marker::NoFreeze}
std::comm::Port<()>{queue: SPSC(std::sync::spsc_queue::Consumer<(),std::comm::Packet>{state: std::sync::arc::UnsafeArc<std::sync::spsc_queue::State<(),std::comm::Packet>>{data: (0x7f435c001640 as *mut ())}}), marker: std::kinds::marker::NoFreeze}
std::comm::Port<()>{queue: SPSC(std::sync::spsc_queue::Consumer<(),std::comm::Packet>{state: std::sync::arc::UnsafeArc<std::sync::spsc_queue::State<(),std::comm::Packet>>{data: (0x7f435c001740 as *mut ())}}), marker: std::kinds::marker::NoFreeze}
task '<unnamed>' failed at 'assertion failed: self.head.is_null()', /build/buildd/rust-nightly-201402020405~3e39e3e~precise/src/libstd/comm/select.rs:298

[alexcrichton]

My previous thoughts about destructors are not providing the soundness that I thought I was getting. I would expect code like this to not compile:

struct Foo {
    a: int
}

struct Bar<'a> {
    a: &'a Foo,
}

impl Drop for Foo {
    fn drop(&mut self) {
        println!("dropping foo");
    }
}

#[unsafe_destructor]
impl<'a> Drop for Bar<'a> {
    fn drop(&mut self) {
        println!("dropping bar {}", self.a.a);
    }
}

fn main() {
    let mut a = ~[];
    let b = Foo { a: 2 };
    a.push(Bar { a: &b });
}

The output is

dropping foo
dropping bar 2

Which is clearly a use-after-free. I presume that this is why I had to write #[unsafe_destructor]. If you re-order

    let mut handles = ~[];
    let select = Select::new();

to

    let select = Select::new();
    let mut handles = ~[];

your code should work alright.

cc @nikomatsakis, just want to make sure that this isn't an unknown bug in destructors right now. Also, could any of the new marker types help me with this bug?

[alexcrichton]

Nominating.

[nikomatsakis]

This is why you had to write unsafe destructor. This is not an unknown bug -- you should never write unsafe destructor on a type that "escapes" to the wild. It is only valid if you control exactly where that type is instantiated. #8861 would possibly help here.

[pnkfelix]

Not a 1.0 blocker, P-low. #8861 subsumes the high priority dtor semanitcs issue implicit here.

[reem]

If this is superseded by #8861, can it be closed? It seems like this is mostly a mistake and not really a bug.

[pnkfelix]

man this is some old code in this example I need to forward port here!

[pnkfelix]

Okay, here is my attempt to forward port the original code: http://is.gd/kk3Fb9

It now produces an error that select does not live long enough.

And if you do the suggested rewrite from #12042 (comment) illustrated here: http://is.gd/aSNKtF then it compiles (and times out in the playpen)! Score one for #8861! It only took us a year to get there!

So, i am closing.