Bugs that are never 100% reproducible are the worst?
I was troubleshooting a regular Heisenbug today.
One of our application's features is exporting some metadata in XML form. Pretty straightforward. However, the unit test for this was failing. So I started looking into it, first by running it.
Immediately, I see something is wrong. The XML being returned from the program is very, very different from the test reference. I try debugging, step through some breakpoints, nothing sticks out. So I look at the output again.
The mother trucker has changed. Yep. The output of our XML generator is completely random. I'm not even kidding when I say that I can barely get it to return the same thing twice.
At this point I call my boss over. He's equally confused, but we dig a little deeper. We tested it on code two weeks old. Test passes flawlessly. That means sometime between Monday and August 4, this bug crept in through one of our commits.
At this point, he suggests we send a team-wide email to see if any of the devs at our satellite offices know what's going on. We're still waiting on a response.