next »

[patico]

Hi there,
I'm wondering, how do you handle unexpected exceptions and crashes after game launched/released? How do you getting know if your game works fine or not on the player's side, especially when you release on few platforms?
Is there any best practices or solutions to catch exceptions after release?

or maybe "release-and-foget-it" is the best strategy? 

[Schrompf]

I used to check the Steam forums for complaints. Since the dawn of time I have code in place to write a stacktrace to the logfile when a crash or exception occurs. Whenever someone complains about crashes I ask for the log. Works mostly. Except when you support OSX or Linux - on these systems things tend to break easily and quickly, and especially Linux with it's million combinations of runtime, window manager, GPU drivers, OS zealots refusing to install the NVidia proprietary driver and so on is a pure maintenance nightmare.

If I ever do another project, I intend to not support anything else but Windows. The 2% of additional revenue does not justify being screamed at by entitled whiners who don't want to click "Refund" for unstated reasons.

[oahda]

pls support me i'm not windows

How are you getting the trace after the program already crashed? I've never tried catching arbitrary crashes before so I actually don't know how that works. Try-catch clause around the whole program?

[Schrompf]

There's a bunch of different things to take care of.

a) exceptions. Like throw Stuff(); - You can try-catch globally, but the stacktrace of the origin would be lost. I use a macro ThrowException(...) or similar, which then invokes some code I got from the internet like 15 years ago. Nowadays there's boost::stacktrace - maybe just use that.

b) crashes. Like SEGFAULT and friends. On Windows there's SEH - Structured Exception Handling. I copied some code off the internet 15 years ago, I think it was called WheatyExceptionHandler. Of course you need different code on Linux / OSX - GoogleTest does this, for example. It also catches crashes in your unit tests and allows you to handle or even EXPECT those, so it definitely has code to handle these situations. It might be a simpler source to rip out the code from than boost.

There's also All-In solutions like GoogleCrash, but those are huge and not easy to set up. What they do is basically run your game in a separate process and observe that process for crashes/stalls, firing off emails / web requests / you-name-it in case of distress. Alternatively on Windows, you can enable MiniDumps for your application via some method which escapes me right now. The resulting dump file is a few hundred kb in size - you can ask your players to email you that file instead.

GoogleCrash, the MiniDumps and so on also have the advantage that you don't have to deploy debug symbols along with your game. I did this - on Windows I put the .pdb files next to the executables, on Linux... erm... I did something different. Amounted to 50MB in addition to the 3MB executable, but it compresses well. A usual Splatter update on Steam only updating the executable was ~9MB in size.

Sorry for the many vague references, I'm off of game development for a while now, and haven't touched the associated code in years.