How Code Fails in the Real World
summary
Adds direct crash and error reporting to a large cross‑platform desktop app using CrashRpt then a century backend that wraps crashpad/brakepad and performs symbolication, metadata capture and grouping. Finds 90% Windows users and a surprising root cause split: real crashes are often GPU/driver or filesystem/slow‑IO issues, not pointer bugs, while many failures stem from user impatience and bad data; one allocation site (the image loader causes 90% of OOMs) dominates. The pipeline works but only after building a symbol DB, aggregation and actionable UI.
AI-written summary · about
chapters (23)
- 0:00Start
- 0:04Background
- 3:05Flight Simulators
- 5:26Answering the Bug Report Tickets
- 6:10Anger
- 6:32Depression
- 9:29Crash Reporting Libraries
- 13:12Symbolication
- 14:56Crash Context
- 23:55Impatience
- 32:37Set New Handler
- 35:25Opengl Drivers
- 38:12Non-Crashing Failures
- 40:36Gross Configuration Errors
- 46:59Old Paths
- 51:35Updating New Versions
- 53:26Normal Error Handling
- 54:48Memory Allocation Errors
- 1:00:42Capturing the Last Rendered Frame
- 1:02:43Lessons Learned
- 1:09:31What Kind of Automated Testing Regime Do You Have
- 1:10:38Meta Reports
- 1:13:09Crash Free Rate
related talks
37mYour Software is Broken — Pay Attention
39mThe Good Bad Bug: Fail Your Way to Better Code
32mLove Your Crash Dumps
31mLet it crash!
50mPractical aspects of crash reporting, analytics and connectivity
37mPracticing Linux Crash/Panic Issue on Production and Cloud Server: Using K...
39mThe Good Bad Bug: Fail Your Way to Better Code
29mCrash-reporting taken to next-level