Thanks. I know about the stack trace, I can see where the memory was allocated, but we are allocating thousands of them per second and deallocation just slightly less then thousands per second (in the same path). So examining the actual leaked objects (in this case I think they are string fragments) seems like the best corse to track down the root cause of not deallocating them (I'm expecting one of the error paths doesn't deallocate everything it had going on, but that code is all C++ and it has been a while since I've crawled around in a big C++ mess and I figured having more information about the error state would be helpful...).