A watchpoint tracks a memory location—it does not track a variable. This means that a watchpoint might not perform as you would expect it to when watching stack or automatic variables. For example, suppose that you want to watch a variable in a subroutine. When control exits from the subroutine, the memory allocated on the stack for this subroutine is deallocated. At this time, TotalView is watching unallocated stack memory. When the stack memory is reallocated to a new stack frame, TotalView is still watching this same position. This means that TotalView triggers the watchpoint when something changes this newly allocated memory.
Also, if your program reinvokes a subroutine, it usually executes in a different stack location. TotalView cannot monitor changes to the variable because it is at a different memory location.
All of this means that in most circumstances, you shouldn’t place a watchpoint on a stack variable. If you need to watch a stack variable, you will need to create and delete the watchpoint each time your program invokes the subroutine.
This doesn’t mean you can’t place a watchpoint on a stack or heap variable. It just means that what happens is undefined after this memory is released. For example, after you enter a routine, you can be assured that memory locations are always tracked accurately until the memory is released.
|
In some circumstances, a subroutine may be called from the same location. This means that its local variables might be in the same location. So, you might want to try.
|
If you place a watchpoint on a global or static variable that is always accessed by reference (that is, the value of a variable is always accessed using a pointer to the variable), you can set a watchpoint on it because the memory locations used by the variable are not changing.