Fix a bug with the scheduler and destructor order

The PausibleIdleCallback must have some handle into the event loop, and because
struct destructors are run in order of top-to-bottom in order of fields, this
meant that the event loop was getting destroyed before the idle callback was
getting destroyed.

I can't confirm that this fixes a problem in how we use libuv, but it does
semantically fix a problem for usage with other event loops.

diff --git a/src/libstd/rt/sched.rs b/src/libstd/rt/sched.rs
index 6e661884616175dd0d603ccc8cf9b9e4e022b238..1a6529dab189962d8ddcd17e7063854ecec4e0b5 100644 (file)
--- a/src/libstd/rt/sched.rs
+++ b/src/libstd/rt/sched.rs
@@ -85,7 +85,17 @@ pub struct Scheduler {
     priv yield_check_count: uint,
     /// A flag to tell the scheduler loop it needs to do some stealing
     /// in order to introduce randomness as part of a yield
-    priv steal_for_yield: bool
+    priv steal_for_yield: bool,
+
+    // n.b. currently destructors of an object are run in top-to-bottom in order
+    //      of field declaration. Due to its nature, the pausible idle callback
+    //      must have some sort of handle to the event loop, so it needs to get
+    //      destroyed before the event loop itself. For this reason, we destroy
+    //      the event loop last to ensure that any unsafe references to it are
+    //      destroyed before it's actually destroyed.
+
+    /// The event loop used to drive the scheduler and perform I/O
+    event_loop: ~EventLoopObject,
 }
 
 /// An indication of how hard to work on a given operation, the difference

diff --git a/src/test/run-pass/field-destruction-order.rs b/src/test/run-pass/field-destruction-order.rs
new file mode 100644 (file)
index 0000000..1d4c08f
--- /dev/null
+++ b/src/test/run-pass/field-destruction-order.rs
@@ -0,0 +1,52 @@
+// Copyright 2013 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// In theory, it doesn't matter what order destructors are run in for rust
+// because we have explicit ownership of values meaning that there's no need to
+// run one before another. With unsafe code, however, there may be a safe
+// interface which relies on fields having their destructors run in a particular
+// order. At the time of this writing, std::rt::sched::Scheduler is an example
+// of a structure which contains unsafe handles to FFI-like types, and the
+// destruction order of the fields matters in the sense that some handles need
+// to get destroyed before others.
+//
+// In C++, destruction order happens bottom-to-top in order of field
+// declarations, but we currently run them top-to-bottom. I don't think the
+// order really matters that much as long as we define what it is.
+
+struct A;
+struct B;
+struct C {
+    a: A,
+    b: B,
+}
+
+static mut hit: bool = false;
+
+impl Drop for A {
+    fn drop(&mut self) {
+        unsafe {
+            assert!(!hit);
+            hit = true;
+        }
+    }
+}
+
+impl Drop for B {
+    fn drop(&mut self) {
+        unsafe {
+            assert!(hit);
+        }
+    }
+}
+
+pub fn main() {
+    let _c = C { a: A, b: B };
+}