"task ... ran out of stack" error

[skinner]

I get a "task .. ran out of stack" error with this program:

use core::hashmap::HashMap;

fn main() {
    let mut map : HashMap<~str, int> = HashMap::new();
    let keys = ~[~"foo", ~"bar", ~"foo"];
    do keys.consume |_, key| {
        let count;
        {
            let count_option = map.find(&key);
            match count_option {
                Some(x) => { count = *x; },
                None => { count = 0; }
            }
        }
        map.insert(key, count + 1);
    }

    for map.each |&key, &count| {
        println(fmt!("%?: %?", key, count));
    }
}

I was originally trying to work around the lack of flow-sensitivity in borrowck, so the code is a little contorted from what would be natural. But I imagine that the "out of stack" error is a separate issue.

rustc is:
rustc 0.6 (ee3789b 2013-04-24 13:33:29 -0700)
host: x86_64-unknown-linux-gnu

running on Fedora 17.

[kud1ing]

rustsqlite's unit test tests::step_row_basics recently has a similar error with rust incoming:

rust: task 7fb9a14107e0 ran out of stack

[skinner]

Okay, I reproduced this on a different machine (also fedora 17 x86_64), and I cut the test case down. I'm a little surprised by how simple it ended up being:

use core::hashmap::HashMap;

fn main() {
    let mut map : HashMap<~str, int> = HashMap::new();
    map.insert(~"foo", 1);
}

I ran gdb on it; I set a breakpoint in rust_task::fail per pcwalton's suggestion. The backtrace from there contains a few hundred thousand calls to rand::next, but I went all the way to the bottom of the backtrace, and this is what it looks like:

#283966 0x0000000000405db5 in rand::__extensions__::next_2755::_90bd14656eb73ce::_00 ()
#283967 0x0000000000405db5 in rand::__extensions__::next_2755::_90bd14656eb73ce::_00 ()
#283968 0x0000000000405db5 in rand::__extensions__::next_2755::_90bd14656eb73ce::_00 ()
#283969 0x0000000000405d31 in rand::__extensions__::rand_2752::_da34413e5895a8b::_00 ()
#283970 0x0000000000405cd3 in rand::__extensions__::gen_2749::_67c43c75b977a99f::_00 ()
#283971 0x0000000000403aa9 in hashmap::linear_map_with_capacity_2584::_e74c518b1fb540de::_00 ()
#283972 0x0000000000403a3a in hashmap::__extensions__::with_capacity_2581::_e74c518b1fb540de::_00 ()
#283973 0x00000000004039d3 in hashmap::__extensions__::new_2578::_b4139432f8d437b0::_00 ()
#283974 0x00000000004037c6 in main::_28d936a8bfce3ea3::_00 ()
#283975 0x00000000004038f4 in _rust_main ()
#283976 0x00007ffff78343d4 in task_start_wrapper (a=0x61e060) at /home/mitch/sw/rust/src/rt/rust_task.cpp:163
#283977 0x0000000000000000 in ?? ()

rustc for this is a slightly newer version from incoming:

rustc 0.6 (0604468 2013-04-25 12:42:41 -0700)
host: x86_64-unknown-linux-gnu

[gifnksm]

I also reproduce this. My test case is almost the same as skinner's.

use core::hashmap::{ HashMap };

fn main() {
  let mut map = HashMap::new();
  map.insert(1u, 1u);
}

HashMap::new calls IsaacRng#gen(), and following code causes stack overflow.

use core::rand::{ IsaacRng, RngUtil };

fn main() {
  let r = @IsaacRng::new();
  io::println(fmt!("%u", r.gen()));
}

But following code doesn't cause stack overflow (@IsaacRng vs IsaacRng).

use core::rand::{ IsaacRng, RngUtil };

fn main() {
  let r = IsaacRng::new();
  io::println(fmt!("%u", r.gen()));
}

My machine information is below.

$ rustc --version
rustc 0.6 (5f7947a 2013-04-27 19:18:33 -0700)
host: x86_64-unknown-linux-gnu
$ ulimit -s
8192
$ uname -a
Linux arch-vbox 3.8.4-1-ARCH #1 SMP PREEMPT Wed Mar 20 22:10:25 CET 2013 x86_64 GNU/Linux

[pnkfelix]

See also #6061

[skinner]

I'm not seeing the "ran out of stack" error anymore with incoming, so I think I'll close this issue. There may still be a problem, per @pnkfelix's comments on #6061, but maybe that should be a separate issue?