Do not allow a hot reload if the isolate has a sticky error

runtime/vm/isolate.cc

Index: runtime/vm/isolate.cc
diff --git a/runtime/vm/isolate.cc b/runtime/vm/isolate.cc
index e6e15646830edf2c38fb814aa40f7f46ed92aac1..3c6c88b8049299086efe71ec19e2c6e858493fbd 100644
--- a/runtime/vm/isolate.cc
+++ b/runtime/vm/isolate.cc
@@ -2192,12 +2192,10 @@ RawObject* Isolate::InvokePendingServiceExtensionCalls() {
           "[+%" Pd64 "ms] Isolate %s : _runExtension complete for %s\n",
           Dart::timestamp(), name(), method_name.ToCString());
     }
+    // Propagate the error.
     if (result.IsError()) {
-      if (result.IsUnwindError()) {
-        // Propagate the unwind error. Remaining service extension calls
-        // are dropped.
-        return result.raw();
-      } else {
+      // Remaining service extension calls are dropped.
+      if (!result.IsUnwindError()) {
         // Send error back over the protocol.
         Service::PostError(method_name,
                            parameter_keys,
@@ -2206,9 +2204,13 @@ RawObject* Isolate::InvokePendingServiceExtensionCalls() {
                            id,
                            Error::Cast(result));
       }
+      return result.raw();
     }
+    // Drain the microtask queue.
     result = DartLibraryCalls::DrainMicrotaskQueue();
+    // Propagate the error.
     if (result.IsError()) {
+      // Remaining service extension calls are dropped.

[siva, 2016/08/24 17:53:24]

Should this error also be posted back over the protocol so that clients
listening are aware of it?

[Cutch, 2016/08/24 18:30:49]

The code Service::PostError is actually responding to the JSON-RPC request. The
idea is that if the function that implements the service extension throws an
error that we should send back a JSON-RPC error so the client doesn't wait
forever. If the microtask queue generates an error we can't be certain that it
means the service request wasn't already responded to.

       return result.raw();
     }
   }