I’ve got a new saying that I quite like. It’s probably not original, but that’s true of a lot of stuff that I do.
If what you are observing is impossible it is either not happening, or not impossible.
I’ve found and fixed my impossible bug. Of course it wasn’t where I thought it was, because if it was, I’d have looked there and mended it. The fault wasn’t even in the code doing the job, it was in the code telling me what the system was doing.
I’d missed a break out of a C switch statement (a frighteningly easy thing to do in that language) so that my reporting code was printing out the wrong messages. Which led me to concoct all kinds of fanciful theories about memory corruption and race conditions - rather than simply checking that what I was being told was happening was actually happening.
The important thing about situations like these is that you change the way that you work to reduce the chances of them happening again. So, in future I’ll make sure that I test all the possible outputs of my diagnostic system.