How to use a thread-death notifier

To use a thread-death notifier, a thread which needs to know about the death of other threads:

• constructs an RUndertaker handle and then calls its Create() member function to create the associated Kernel side object. The RUndertaker handle is process-relative which means that the handle is not closed if the requesting thread dies.

• issues a notification request to the thread-death notifier, passing a reference to a TRequestStatus object and a reference to a TInt.

When any other thread dies, the requesting thread's request semaphore is signalled to indicate that the notification request is complete. The TRequestStatus object supplied by the requesting thread is set to KErrDied.

In addition, the Kernel opens a local thread-relative handle on the dying thread. (thread-relative, here, means relative to the requesting thread) and sets the TInt, supplied by the requesting thread, to the resulting handle-number. The requesting thread can construct an RThread from this handle-number. The following diagram helps visualise the situation.

The following code fragments demonstrate this:

{
...
RUndertaker     the_undertaker;
TRequestStatus  the_status;
TInt            the_dyingthread_handle_number;
...
the_undertaker.Create();
the_undertaker.Logon(the_status,the_dyingthread_handle_number);
User::WaitForRequest(the_status);
...
...// prepare for a long wait
...
RThread r;
r.SetHandle(the_dyingthread_handle_number);
...
...// Now have an open thread-relative handle to the
...// dying thread so we can, for example, ...
if (r.ExitReason()==0x666)
    {
    ...
    }
...// We have an open handle on the dying thread; the dying thread
...// remains in existence until we close it.
r.Close();
...
}

In practical code, an RUndertaker is used by an active object.

The RUndertaker handle also offers a LogonCancel() function. Calling this function, causes the requesting thread's request semaphore to be signalled and any wait to complete. The TRequestStatus is set to KErrCancel.

This technique of using an RUndertaker is not guaranteed to report the death of all threads. In particular, if a second thread dies while a requesting thread is handling the request completion due to the death of a first thread (but before it can make another notification request), the death of this second thread will go unnoticed.