I think #define volatile is by far the sneakiest.
The rest would probably be found quite quickly, but omitting volatiles could lead to hard to detect bugs that will only manifest themselves under specific circumstances.
The rest would probably be found quite quickly, but omitting volatiles could lead to hard to detect bugs that will only manifest themselves under specific circumstances.
The only legitimate reasons for using volatile generally involve memory-mapped IO and POSIX signal handlers. C's volatile is nothing like Java's volatile and has very weak semantics. With lock-based code you generally have nothing to worry about because lock/unlock has load-acquire/store-release semantics. But if you're writing lock-free code, volatile is useless in general and you need to think carefully about how atomicity is affected by memory alignment and operand size (e.g. a byte write that might look atomic implicitly turns into a non-atomic read-modify-write), explicit memory barriers, etc.
So defining away volatile won't actually impact most applications unless they're relying on very compiler-specific behavior.
The values of automatic variables are unspecified after a call to longjmp() if they meet all the following criteria:
· they are local to the function that made the corresponding setjmp(3) call;
· their values are changed between the calls to setjmp(3) and longjmp(); and
· they are not declared as volatile.
// region 1:
// x is allocated to register r1
int x = 0; // store 0 to r1
jmpbuf jb;
if (setjmp(jb) == 0) {
// region 2:
// x is re-allocated to r2, so r1 is copied to r2
x = 42; // store 42 to r2
longjmp(jb, 1);
}
// region 3:
// x is allocated to r1
printf("%d\n", x); // load from r1
Actually, this would still occur even if x was always allocated to r1. The reason is that before calling setjmp() the compiler spills x to the stack, so the callee can use the registers for its own use, and on returning (either via longjmp() or the first time through) it will restore x from the value on the stack.
I tried the example above with GCC. At the default optimization level it printed 42. But with -O2 it printed 0. Looking at the assembly code, it's actually even worse than I described. The compiler has treated the longjmp() akin to an exit() and so has determined that the x = 42 is actually a no-op that can be eliminated. This is similar to how in printf("1"); exit(0); printf("2"); the compiler will actually remove the second printf() as dead code from the executable.
So, in conclusion, there seems to be a whole bunch of ways in which the compiler could screw this up while staying within the bounds of standards-acceptable behavior.
My knowledge of proper usage of volatile in C is weak. Does it have something to do with the fact that sizeof bool < sizeof int on most platforms?
In that case I mentioned, yes, the issue is that sizeof(bool) = 1 by default on most compilers. You can force it to be 4 on MSVC with a compiler directive but that has its own problems (e.g. you are using a library which uses bools in its API and the library was compiled with sizeof(bool) = 1). The solution is to never use bool, int, etc, directly for variables that will be written to in lock-free code. Use typedefs (or even better for type safety, wrapper structs) that are explicit about size, make all lock-free reads and writes go through library functions/macros that provide the right atomic semantics.
Here's the specific data race with the bools. Assume that sizeof(bool) = 1 and that the two aforementioned bool bytes (call them A and B) wind up next to each other in the same 4-byte aligned word of memory. Thread 1 writes to A and thread 2 writes to B concurrently. It looks like there should be no data race because they're writing to separate variables. But unlike storing a full word to memory, storing a byte involves an implicit read-modify-write, so the only way for this situation to be safe is if thread 1 and 2 use CAS operations to write A and B.
I'm not sure how much you'd have to mark volatile to get a compiler to behave the way you want, so you could use int and dodge the issue. Or better, the standard sigatomic_t type is provided and should work.
I'd personally go with while -> if, that's very sneaky.
int main(void)
{
while (1) break;
return 0;
} int main(void)
{
while (1) {
char c = getchar();
if (c == 'q') break;
process(c);
}
return 0;
}