Even if the code would still be bad for 32-bit machines, I'm not sure they would suffer from the bug (if they had 32-bit signed int as well).

1. Your uint8_t would get promoted to a 32-bit int

2. Your int would get shifted, getting a 1 in the signing bit

3. The result would get stored in a uint32_t when being casted

Now I don't know how uint64_t works on a 32-bit machine, I'd have to compile the C code with -m32 to see.

EDIT: just tested and you're right, it's still a problem on 32-bit machines.