Errors thrown from runZonedExperimental's onDone handler are crazy confusing.

[jbdeboer]

The following code snippet:

var nul;
var c;
runZonedExperimental(() {
  c = new Completer.sync();
  c.future.then((_) {});
}, onDone: () => nul.sad());
c.complete("ss");

produces an error:

Caught 'dart:async/zone.dart': Failed assertion: line 115 pos 12: '! _children.isEmpty' is not true.
    _ZoneBase._removeChild dart:async/zone.dart 115:12
    _ZoneBase._dispose dart:async/zone.dart 48:31
    _WaitForCompletionZone._dispose dart:async/zone.dart 169:19
    _ZoneBase._checkIfDone dart:async/zone.dart 53:15
    _ZoneBase._runInZone dart:async/zone.dart 95:19
    _ZoneBase._runUnguarded dart:async/zone.dart 102:22
    _ZoneBase.executeCallback dart:async/zone.dart 58:23
    _TransformFuture._sendValue dart:async/future_impl.dart 350:26
    _FutureImpl._setValueUnchecked dart:async/future_impl.dart 186:26
    _FutureImpl._setValue dart:async/future_impl.dart 176:23
    _SyncCompleter._setFutureValue dart:async/future_impl.dart 50:21
    _Completer.complete dart:async/future_impl.dart 17:20
    main.<anonymous closure>.<anonymous closure> /home/deboer/github/angular-dart/test/http_spec.dart 119:17 (e.g. the c.complete() call)

when it should throw an error like:

The null object does not have a method 'sad'.

NoSuchMethodError : method not found: 'sad'
Receiver: null
Arguments: []

­0 Object.noSuchMethod (dart:core-patch/object_patch.dart:19:25)

­1 main.<anonymous closure>.<anonymous closure>.<anonymous closure> (http://localhost:9876/base/test/http_spec.dart:116:36)

­2 _WaitForCompletionZone._dispose (dart:async/zone.dart:170:12)

­3 _ZoneBase._checkIfDone (dart:async/zone.dart:53:15)

­4 _ZoneBase._runInZone (dart:async/zone.dart:95:19)

[dgrove]

Set owner to @floitschG.
Added Area-Library, Triaged labels.

[floitschG]

Added Accepted label.

[floitschG]

Added Started label.

[floitschG]

The example works fine with newest revision, but I suspect that some refactorings just made the bug not manifest itself (instead of being fixed).
Now switched back to revision 24946 where the assert triggers. (Thanks for the nice test-case).

[floitschG]

Turns out this is a VM bug, where finally clauses are sometimes executed twice.
Filed issue #11972.

---

Added Duplicate label.
Marked as being merged into #11972.

[floitschG]

Issue #12195 has been merged into this issue.

---

cc @justinfagnani.