The following program, despite looking correct, is a race condition and has unpredictable behaviour:
The proof is in the output of running this code several times:
sizeof(S) = 2, 511, 127
sizeof(S) = 2, 511, 127
sizeof(S) = 2, 0, 127
sizeof(S) = 2, 511, 0
sizeof(S) = 2, 511, 127
sizeof(S) = 2, 511, 127
sizeof(S) = 2, 511, 127
sizeof(S) = 2, 0, 127
sizeof(S) = 2, 0, 127
sizeof(S) = 2, 511, 127
The race condition and the problem here is that the C++ standard states that adjacent bit fields are a single object in memory and may be packed to share the same byte. The CPU cannot operate on single bits, it must load at least 1 byte at a time, and because of the overlap, loading bits
The fix is to use one mutex to protect all adjacent bit fields. I say all because you have no guarantee that the CPU will be able to load 1 byte at a time. It may be limited to working on 32-bit values at a time; depending on the architecture.
Corrected program:
good point
Nice