I was quiet for few last days. We were not idle - we were hunting.

Hunting for runtime errors.

Race condition is a elusive bug to trace. It happens when two parts of the code run simultaneously and interact in unexpected manner. It could lead to crashes, strange behaviour and more weird effects - but the worst part is that the bug is dependent on computer speed, and that can change by having a browser open, second monitor connected, updates in background and myriad different things. The one we traced caused crashes and things disappearing from the screen. It was really rare on Windows, but quire common on Linux systems - which led me initially to believe that it was a platform issue. It was not, it just happened that Linux run part of the code fast enough that the bug resurfaced more often.

Debugger was not helpful.

Usually I would trace this kind of a bug with debugger or debug build - but it didn't work this time, since they both slow down the execution, and crashes were not happening. I needed a full, optimized release build running Linux to crash reliably in few minutes, nothing else seemed to trigger that reliably. And in order to fix a bug, you need to see it happening. Fortunately I found out that I could use Windows Subsystem for Linux to trigger exactly same problem. That was a godsend - now I could debug the game without switching the system, which allowed me to iterate much faster.

Finally crushed the crash.

And now we have the bug pinned. Crashes were triggered by a transponder on your Lidar screen, appearing and disappearing as a ship flew past asteroids and out of your line of sight. With just right timing it could crash the game. It happened on all systems, but it was just easier to spot on Linux. If I didn't support that system, finding this bug would be nearly impossible. 0.193.6 release fixes that - along with couple of other unrelated bugs that just happened along the way.

[ 2020-05-27 13:54:12 CET ] [ Original post ]