Experimental Physics and
Industrial Control System

"Hoff, Lawrence" <[email protected]> · Thu, 14 Apr 2005 10:35:48 -0400


 To play devil's advocate........
	Does anyone know what happens in VxWorks
if the C++ "new" operator throws an unhandled exception
if there is insufficient heap space?
	*If* it suspends the task, then taskwd will work
in an intuitive fashion. *If*, however, the task exits,
then (it sounds like) it will escape the notice of taskwd.
	In the latter case, or in any case where a library
function might call "exit()", then taskwd ought to check for
a missing task and treat it the same as a suspended task.
	Alternatively, all code under taskwd supervision
must handle all exceptions, and never use library functions
which might call "exit" (a fairly tall order).
	As Kay says, if a monitored task *decides* to exit,
it should intentionally unregister itself. Under those rules,
there seems little harm in taskwd checking for missing tasks
as well as suspended tasks (as Dave T. suggested).
-- Larry
P.S. I now know the answer to my hypothetical, at least for
the SNS VxWorks kernel. The task is suspended (so taskwd works
properly). Try this code to see what happens in your O/S:
#include <ostream.h> // cout
extern "C" void newTest(void){
  // ask for too much memory
  char *tooBig = new char[(1<<24)];
  // do we get here?
  cout << "Value of tooBig is: " << (unsigned long) tooBig << endl;
}

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System