Race condition: Min and Max range of an integer

Question

I was recently asked this question in an interview.

Given the following code, what will be the min and max possible value of the static integer num?

import java.util.ArrayList; import java.util.List;  public class ThreadTest {     private static int num = 0;      public static void foo() {         for (int i = 0; i < 5; i++) {             num++;         }     }      public static void main(String[] args) throws Exception{         List<Thread> threads = new ArrayList<Thread>();         for (int i = 0; i < 5; i++) {             Thread thread = new Thread(new Task());             threads.add(thread);             thread.start();         }         for (int i = 0; i < 5; i++) {             threads.get(i).join();         }         // What will be the range of num ???         System.out.println(ThreadTest.num);     } }  class Task implements Runnable {     @Override     public void run() {         ThreadTest.foo();     }  } 

I told them that the max value would be 25 (in case there is no race condition), and min would be 5 (in case of a race condition between all the threads at every iteration).
But the interviewer said the the min value can go even below 5.
How is that possible?

Answer 1

I claim the minimum value possible is 2.

The key to this is the non-atomicity of num++, i.e., it is a read and a write, which may have other operations in between.

Call the threads T1..T5:

• T1 reads 0, T2 reads 0;
• T1 writes 1, and then reads and writes 3 times.
• Then T2 writes 1;
• Then T1 reads 1;
• Then T2-5 do all of their work
• Then, finally, T1 writes 2.

(Note: the result 2 is not dependent either on the number of threads, or the number of iterations, provided there are at least 2 of each.)

But the honest answer to this is: it really doesn't matter. There is a data race, as defined in JLS 17.4.5:

When a program contains two conflicting accesses (§17.4.1 ["Two accesses to (reads of or writes to) the same variable are said to be conflicting if at least one of the accesses is a write."]) that are not ordered by a happens-before relationship, it is said to contain a data race.

(There is an absence of happens-before relationships between the actions in the threads)

So you can't usefully rely on whatever it does. It is simply incorrect code.

(Moreover, I know the answer to this not because of some hard-won battle debugging multithreaded code, or deep technical reading: I know this because I have read this answer before elsewhere. It's a parlour trick, nothing more, and so asking the minimum value isn't a very good interview question).