Is this a JVM bug or "expected behavior"?

Question

I noticed some unexpected behavior (unexpected relative to my personal expectations), and I'm wondering if something if there is a bug in the JVM or if perhaps this is a fringe case where I don't understand some of the details of what exactly is supposed to happen. Suppose we had the following code in a main method by itself:

int i;
int count = 0;
for(i=0; i < Integer.MAX_VALUE; i+=2){
  count++;
}
System.out.println(i++);

A naive expectation would be that this would print Integer.MAX_VALUE-1, the largest even representable int. However, I believe integer arithmetic is supposed to "rollover" in Java, so adding 1 to Integer.MAX_VALUE should result in Integer.MIN_VALUE. Since Integer.MIN_VALUE is still less than Integer.MAX_VALUE, the loop would keep iterating through the negative even ints. Eventually it would get back to 0, and this process should repeat as an infinite loop.

When I actually run this code, I get non-deterministic results. The result that gets printed tends to be on the order of half a million, but the exact value varies. So not only is the loop terminating when I believe it should be an infinite loop, but it seems to terminate randomly. What's going on?

My guess is that this is either a bug in the JVM, or there is a lot of funky optimization going on that makes this expected behavior. Which is it?

@Michael: Just checking there wasn't some funky threading going on. On my machine it's always printing out 2147483640, but that's still unexpected. — Jon Skeet, Mar 03 '11 at 16:26
I tried it 3 times with different results between 300,000 and 500,000 - so looks like it is VM-specific (build 1.6.0_24-b07, 32bit linux) — Matthew Gilliard, Mar 03 '11 at 16:27
The program never terminates on 32-bit Windows XP with HotSpot 1.6.0_24 — Jeremy, Mar 03 '11 at 16:34
Looks like a bug to me. bugs.sun.com would be the appropriate place to report it. (And yes, infinite loops should not terminate.) — Tom Hawtin - tackline, Mar 03 '11 at 16:34
Mine also never terminates. Java HotSpot 1.6.0_13-b03 @ WinXp 32 bit. — mdrg, Mar 03 '11 at 16:38
What JVM are you using? Sun JVM 1.6.0_23-b05 and 1.5.0_14-b03 both loop forever. — Richard Miskin, Mar 03 '11 at 16:49
Java never ceases to amaze me. +1 to the OP and +1 to the answer linking to the Oracle/Sung bug(s). My latest favorite was this one: http://stackoverflow.com/questions/4949057 :) — SyntaxT3rr0r, Mar 03 '11 at 19:40

score 48 · Accepted Answer · edited Oct 22 '12 at 16:39

48

Known bug. Related to

http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6196102

http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6357214

and others.

I think they're considered low-priority to fix because they don't come up in the real world.

edited Oct 22 '12 at 16:39

Michael McGowan

6,528
8
42
70

answered Mar 03 '11 at 16:44

Kevin Bourrillion

40,336
12
74
87

Jon Skeet · Answer 2 · 2011-03-03T16:48:01.833

This is bizarre. It certainly looks like a bug somewhere. I get the same results every time with the same code, but trivial changes to the code change the result. For example:

public class Test {
  public static void main(String[] args) {
    int i;
    int count = 0;
    for (i = 0; i < Integer.MAX_VALUE; i+=2) {
      count++;
    }
    System.out.println(i);
    System.out.println(i < Integer.MAX_VALUE);
  }
}

... always prints 2147483640 and true

whereas this:

public class Test {
  public static void main(String[] args) {
    int i;
    for (i = 0; i < Integer.MAX_VALUE; i+=2) {
    }
    System.out.println(i);
    System.out.println(i < Integer.MAX_VALUE);
  }
}

always prints -2147483648 and true.

Very, very weird.

(That's running an OpenJDK 1.6 VM on Linux.)

EDIT: Running OpenJDK 1.7 on Windows 7, I don't see the problem:

java version "1.7.0-ea"
Java(TM) SE Runtime Environment (build 1.7.0-ea-b78)
Java HotSpot(TM) Client VM (build 17.0-b05, mixed mode, sharing)

Matthew · Answer 3 · 2011-03-03T17:19:21.833

4

Try adding System.out.println(count);

I wonder if there is optimization occurring because count is never read from.

Edit - another answer gave the link to bugs in Oracle's bug tracker. Drawing from that:

6196102 in particular mentions that there is a canonicalization bug where Integer.MAX_VALUE is concerned.
Java must be trying to optimize the loop because count is never read from.

However, this is unlikely to occur in practice, because:

Integer.MAX_VALUE is an unlikely loop guard
Usually loops do work that wouldn't allow this optimization in the first place

edited Mar 03 '11 at 17:19

answered Mar 03 '11 at 16:27

Matthew

44,826
10
98
87

This makes my test runs consistent, at least. So it is an optimization issue? – Matthew Gilliard Mar 03 '11 at 16:30

score 2 · Answer 4 · answered Mar 03 '11 at 16:37

This seems to be a loop optimizations as I observe the same result but IF I also print out count then the result changes.

I.e.

    int i;
    int count = 0;
    for(i=0; i < Integer.MAX_VALUE; i+=2){
      count++;
    }
    System.out.println(count);
    System.out.println(i++);

Produces 2147483638 while the original code produces 457158 (or similar)

score 0 · Answer 5 · answered Mar 03 '11 at 17:26

0

java version "1.6.0_22"
Java(TM) SE Runtime Environment (build 1.6.0_22-b04)
Java HotSpot(TM) Client VM (build 17.1-b03, mixed mode, sharing)

working as expected. infinite loop

answered Mar 03 '11 at 17:26

jmj

237,923
42
401
438

Is this a JVM bug or "expected behavior"?

5 Answers5

Linked