sync::Arc is failing to free memory

Quick test to reproduce, check against d79fbba

extern crate sync;

use sync::Arc;

fn main()
{
    let _ = Arc::new(~"failed to drop");
}

valgrind results

colin@Colin-ubuntu:~/Source$ valgrind --leak-check=full ./failed_to_drop
Memcheck, a memory error detector
Copyright (C) 2002-2011, and GNU GPL'd, by Julian Seward et al.
Using Valgrind-3.7.0 and LibVEX; rerun with -h for copyright info
Command: ./failed_to_drop


HEAP SUMMARY:
    in use at exit: 54 bytes in 2 blocks
  total heap usage: 16 allocs, 14 frees, 950 bytes allocated

54 (24 direct, 30 indirect) bytes in 1 blocks are definitely lost in loss record 2 of 2
   at 0x4C2B6CD: malloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
   by 0x40507F: rt::global_heap::malloc_raw::h9abfb11fdd4fbd35cba::v0.0 (in /home/colin/Source/failed_to_drop)
   by 0x405007: rt::global_heap::exchange_malloc::h1a731dc6869c0c73Vaa::v0.0 (in /home/colin/Source/failed_to_drop)
   by 0x404E99: arc::Arc$LT$T$GT$::new::haa6209d493a9d577raa::v0.0 (in /home/colin/Source/failed_to_drop)
   by 0x404DE4: main::hed1e648887e7476fgaa::v0.0 (in /home/colin/Source/failed_to_drop)
   by 0x438642: start::closure.7865 (in /home/colin/Source/failed_to_drop)
   by 0x4D3452: rt::task::Task::run::closure.41627 (in /home/colin/Source/failed_to_drop)
   by 0x4DE5AB: rust_try (in /home/colin/Source/failed_to_drop)
   by 0x4D32B1: rt::task::Task::run::h19b1ba3b1fab6483os9::v0.10.pre (in /home/colin/Source/failed_to_drop)
   by 0x438433: start::h7087cc56d053c8b1zvd::v0.10.pre (in /home/colin/Source/failed_to_drop)
   by 0x438223: lang_start::hcbb6c5fc89f3553bTud::v0.10.pre (in /home/colin/Source/failed_to_drop)
   by 0x404E4E: main (in /home/colin/Source/failed_to_drop)

LEAK SUMMARY:
   definitely lost: 24 bytes in 1 blocks
   indirectly lost: 30 bytes in 1 blocks
     possibly lost: 0 bytes in 0 blocks
   still reachable: 0 bytes in 0 blocks
        suppressed: 0 bytes in 0 blocks

For counts of detected and suppressed errors, rerun with: -v
ERROR SUMMARY: 1 errors from 1 contexts (suppressed: 2 from 2)

Looking through arc.rs it looks like the bug with how the reference counting was used in drop. fetch_sub returns the old value, not the new one. Meaning you need to check for 1 and not 0.

[alexcrichton]

@brson, this is a strong indication to me that valgrind is not running on the snapshot bots, even though we believe it to be running.