--- a/commsfwsupport/commselements/meshmachine/src/mm_node.cpp Tue Apr 27 17:53:34 2010 +0300
+++ b/commsfwsupport/commselements/meshmachine/src/mm_node.cpp Tue May 11 17:20:19 2010 +0300
@@ -458,25 +458,31 @@
TInt idx = aContext.iNodeActivity->FindOriginator(aCommsId);
if (KErrNotFound!=idx)
{
+ TBool canSend = ETrue;
if(aContext.iNodeActivity->iOriginators.Count() == 1) // only if this is the final originator
{
aContext.iNodeActivity->SetError(KErrAbort);
aContext.iNodeActivity->Cancel(aContext);
+ //This is a workaround for CCommsBinderRequest. The proper fix is to abolish the concept of aborting activities.
+ //Aborting activities is a bad idea as an aborted activity isn't given a chance to perform graceful cleanup.
+ //Today activities get aborted because their orinators urgently leave. I.e.: they are trully leaving now! Last orders!
+ //It is then incorrect to leave the activity d'tor to finish the wrap up - because the node will be gone by then.
+ //So whether and when to send an error must be decided here, by this generic code that has no clue on the subtleties
+ //of individual activities. If there is no abort - there is urgent leavers. They send TLeaveRequest and they politely
+ //wait for the completion and all this code is unnecessary.
+ canSend = (aContext.iNodeActivity->Error() != KErrNone);
}
-
-
+
//In the "quiet mode", when the hosting node is being destroyed, we can not afford sending
//an error to the node as it would hit void.
TNodePeerId& originator = aContext.iNodeActivity->iOriginators[idx];
- TBool canSend = !((aIsNodeBeingDestroyed && originator == aContext.NodeId())
- || aContext.iMessage.IsMessage<TEChild::TLeft>());
+ canSend &= !((aIsNodeBeingDestroyed && originator == aContext.NodeId())
+ || aContext.iMessage.IsMessage<TEChild::TLeft>());
if (canSend)
{
aContext.iNodeActivity->PostToOriginator(originator, TEBase::TError(aContext.iMessage.MessageId(), KErrAbort).CRef());
}
-
-
- aContext.iNodeActivity->RemoveOriginator(idx);
+ aContext.iNodeActivity->RemoveOriginator(idx);
}
}
}