Comment 4 for bug 1558206

The self-deletion sounds right, I think Jeff uses that inside CA client code.

What's wrong with moving the pointer-clear line up inside the guard block? It appears that whole purpose of the waitRelease flag is to tell epicsThreadCallEntryPoint() whether the epicsThread object still exists after the pthread->runnable.run() routine has returned (or thrown). Given that, I don't see why applying this patch on top of your commit shouldn't be correct:

=== modified file 'src/libCom/osi/epicsThread.cpp'
--- src/libCom/osi/epicsThread.cpp 2016-03-16 18:40:37 +0000
+++ src/libCom/osi/epicsThread.cpp 2016-03-16 21:20:43 +0000
@@ -104,12 +104,12 @@
     }
     if ( ! waitRelease ) {
         epicsGuard < epicsMutex > guard ( pThread->mutex );
+ pThread->pWaitReleaseFlag = NULL;
         pThread->terminated = true;
         pThread->exitEvent.signal ();
         // once the terminated flag is set and we release the lock
         // then the "this" pointer must not be touched again
     }
- pThread->pWaitReleaseFlag = NULL;
 }