`for` loops allow moving out of references

[steveklabnik]

Updated description

for loops allow moving out of fixed size arrays. This code compiles and should not compile.

struct Foo {
    a: [Box<int>, ..3],
}

fn main() {
    let f = Foo { a: [box 3, box 4, box 5] };
    for &a in f.a.iter() {
    }
}

Original description

segfault with channels

I don't know how to make this test case smaller, but...

https://github.com/steveklabnik/dining_philosophers/commit/52ca03637a750f603f70c1d3b17d4326a2ae5eaa

This commit introduces a segfault.

steve@computer:~/src/dining_philosophers$ cargo run
   Compiling dining_philosophers v0.0.1 (file:/home/steve/src/dining_philosophers)
     Running `target/dining_philosophers`
Baruch Spinoza has sat down to eat.
An unknown error occurred

To learn more, run the command again with --verbose.  
steve@computer:~/src/dining_philosophers$ ./target/dining_philosophers 
Karl Marx has sat down to eat.
Karl Marx is thinking.
Baruch Spinoza has sat down to eat.
Baruch Spinoza is thinking.
Segmentation fault (core dumped)
steve@computer:~/src/dining_philosophers$

Seems bad.

[JCRPaquin]

From what I'm seeing here(debugger wise) the problem looks to stem from the call to channel.try_recv in the have_dinner function.

Inside try_recv we enter the loop, make it through the first unsafe match and then the problem appears to occur in the Oneshot branch when we try to match again.

[alexcrichton]

Can you describe the system that you're running on as well as the cargo/rustc versions you're using? I've been unable to reproduce the segfault on the most recent nightly on both linux and osx.

[JCRPaquin]

Oh sorry, I'm running Windows here. Latest installer version.

C:\Windows\System32>rustc --version
rustc 0.12.0-pre-nightly (d7cfc34a2 2014-08-02 00:31:03 +0000)

After restarting into Xubuntu and updating I recompiled there and got a segfault. I get rust from the hansjorg/rust PPA.

magisun@magimobile-ubuntu:~$ rustc --version
rustc 0.12.0-pre

I'm running off of a Toshiba Satellite L875D-S7210 with an AMD A6-4400M APU. Windows is 64bit and Xubuntu is 32bit.

[alexcrichton]

Ah, found it.

Nominating and cc @pcwalton, this seems bad and related to the recent removal of for-loop desugaring.

[pnkfelix]

Assigning P-backcompat-lang, 1.0 milestone.

[huonw]

Simpler reproduction (it's not related to fixed sized arrays, updating title):

fn main() {
    let x = Some(box 1i);
    for &a in x.iter() {
    }
}

[reem]

If you want to see a segfault you have to do something in the loop and iterate twice:

fn main() {
    let x = [box 1i, box 2i];
    for &a in x.iter() {
        println!("{}", a);
    }
    for &a in x.iter() {
        println!("{}", a);
    }
}

This prints 1 and 2, then stalls a bit, then segfaults.