Beyond the Stack Trace: A Crash in Three Acts

When a shipped game crashes on a player’s machine, you usually get a stack trace back. Sometimes you only get an error code like DXGI_ERROR_DEVICE_REMOVED. Either way, you learn where the game died but not what it was doing, and you can’t get your hands on the machine to find out.

I’ll go through three problems we ran into while building crash reporting for games and consoles at Sentry. First, why we wrote our own out-of-process crash backend, and what that made possible on locked-down consoles. Second, how we capture crashes that happen on the GPU or inside the OS. Third, how we turn a raw memory dump back into readable variables and automatically flag memory corruption.