Being careless - data races

A data race, also known as a race condition, occurs when two or more threads attempt to write to the same shared memory simultaneously. As a result, the state of the shared memory during and at the end of the sequence of instructions executed by each thread is by definition, non-deterministic.

As we saw in Chapter 6, Debugging Multithreaded Code, data races are reported quite often by tools used to debug multi-threaded applications. For example:

==6984== Possible data race during write of size 1 at 0x5CD9260 by thread #1
 ==6984== Locks held: none
 ==6984==    at 0x40362C: Worker::stop() (worker.h:37)
 ==6984==    by 0x403184: Dispatcher::stop() (dispatcher.cpp:50)
 ==6984==    by 0x409163: main (main.cpp:70)
 ==6984== 
 ==6984== This conflicts with a previous read of size 1 by thread #2
 ==6984== Locks held: none
 ==6984==    at 0x401E0E: Worker::run() (worker.cpp:51)
 ==6984==    by 0x408FA4: void std::_Mem_fn_base<void (Worker::*)(), true>::operator()&lt...