Possible VS2012 compiler bug (maybe in Whole Program Optimization?)

Question

Could this be a compiler error? My environment is:

Win7 pro (64-bit)
VS2012 (update 3)

I compile the tiny console program below. Things work fine for x64 bit release/debug builds. The x32 debug build also works just fine. The x32 release build, however displays 'BUG!'.

If i disable 'Whole Program Optimization' that will fix the issue.

Any ideas?

-

#include <string>
#include <iostream>


int main()
{
    std::string const buffer = "hello, world";
    std::string::size_type pos = 0;
    std::string::size_type previous_pos;


    while (pos != std::string::npos)
    {
        previous_pos = ++pos;
        pos = buffer.find('w', pos);
    } 


    if (previous_pos == std::string::npos)
    {
        std::cout << "BUG!!"<< std::endl;
    }

    return 0;
}

If the loop goes around zero times, then previous_pos is uninitialized. — brian beuning
– brian beuning, Commented Aug 4, 2013 at 23:40
What happens when you initialize previous_pos to zero? Perhaps the compiler is making some assumption based on the fact that it's uninitialized? — SeaBass
– SeaBass, Commented Aug 5, 2013 at 0:15
also, changing the pre-increment of pos to post-increment seems to make the bug go away. — markv
– markv, Commented Aug 5, 2013 at 0:31
while (pos != std::string::npos) { previous_pos = ++pos; pos = buffer.find('w', pos); std::cout << "Prev Pos: " << previous_pos << std::endl; } This is interesting. Adding a cout statement using previous_pos before the end of the loop also causes the bug to dissapear... Interesting question! I'm guessing this is something to do with compiler optimization? — David
– David, Commented Aug 5, 2013 at 3:23
Also, changing previous_post to = pos and pos = buffer.find('w', pos++) has some interesting results. In debug previous_pos will alternate between 0 and 7. In release 0 and 8.... Again, I'm using a cout in the loop here. — David
– David, Commented Aug 5, 2013 at 3:37

mattw · Accepted Answer · 2013-08-05 23:44:49Z

4

I can reproduce this too. When the bug manifests, the code is testing eax to deterimine whether to output "BUG", which is the same register it's using for 'pos'.

17:         previous_pos = ++pos;

013C12E5 inc eax
...

21:     if (previous_pos == std::string::npos)

00301345 cmp eax,eax
00301347 jne main+0F6h (0301366h)

However if you make a change to try to get the optimzer to realise they are distinct then the test is different. If I add ++previous_pos at the end of the loop body then it uses ecx for previous_pos and the bug goes away:

22:     if (previous_pos == std::string::npos)

00361349 cmp ecx,eax
0036134B jne main+0FAh (036136Ah)

If I change the find to 'pos = buffer.find('w', previous_pos);' (searching from previous_pos instead of pos, which has the same value) then it uses ebx, and again the bug goes away:

21:     if (previous_pos == std::string::npos)

00191345 cmp ebx,eax
00191347 jne main+0F6h (0191366h)

So it seems in the original that the optimiser is erroneously deciding that it can use eax for both of those variables, despite the line 'pos = buffer.find('w', pos);' that can set pos to a different value than previous_pos.

answered Aug 5, 2013 at 23:44

mattw

562 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

David Over a year ago

So adding the cout statement causes the programme to realize it cannot use the same register? What about my last comment above about the register alternating between 0 and 7 in debug and 0 and 8 in release?

markv Over a year ago

Thanks to all you guys for having a look at this issue. Seems like everyone is in agreement that this is indeed a compiler bug, so I think i'll mark this as 'answered' now to close it off.

Collectives™ on Stack Overflow

Possible VS2012 compiler bug (maybe in Whole Program Optimization?)

1 Answer 1

2 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Your Answer

Sign up or log in

Post as a guest

Related