39 lines
2.9 KiB
Plaintext
39 lines
2.9 KiB
Plaintext
[video output=day176 member=cmuratori stream_platform=twitch stream_username=handmade_hero project=code title="Introduction to Debug Infrastructure" vod_platform=youtube id=vhSIwxAWDIY annotator=Miblo annotator=debiatan]
|
|
[1:07][Plan for today]
|
|
[2:34][Test-driven development doesn't help solving the harder problems of game programming]
|
|
[4:59][Purpose of debug services]
|
|
[6:33][1) Coaxing bugs to the surface]
|
|
[8:00][2) Locating bugs that are clearly present but difficult to pinpoint]
|
|
[11:58][We'll focus on versatile debug services aimed at solving difficult bugs]
|
|
[13:21][Logs. Example of debugging framerate problems]
|
|
[15:49][We could start by recording our debug timers on each frame]
|
|
[16:35][Simplifying the usage of our debug timers]
|
|
[17:38][We could combine the log and the looped live code editing replay system]
|
|
[20:01][Limitations of this approach - Is not particularly good for multithreading problems]
|
|
[21:23][Necessary components: counter log, replay system, log of memory consumption and diagramming]
|
|
[22:16][Diagramming]
|
|
[25:57][We want all the debugging subsystems going through the same log]
|
|
[27:14][We want to avoid having to change the code in order to debug it...]
|
|
[29:11][...and also avoid having to remove debug calls once we're done]
|
|
[31:44][Review of desired features for our debug system]
|
|
[32:56][Will we need tuning/fiddling support?]
|
|
[36:40][Let's start building the debug system]
|
|
[38:05][The old debug cycle counters were lousy on purpose]
|
|
[43:00][Improving the interface of the cycle counters]
|
|
[43:27][Abusing constructor/destructor pairs to achieve that better interface]
|
|
[54:49][We shouldn't need to store values anywhere to make paired calls at the beginning and end of a frame]
|
|
[57:11]["I just kind of ruined The Variable"][quote 202]
|
|
[1:01:03][Review of the inner workings of the new cycle counter interface]
|
|
[1:02:15][Testing that the code still runs after today's changes]
|
|
[1:02:37][Q&A][:speech]
|
|
[1:03:22][@SoysauceTheKid][Is there a way to determine how much overhead the debug takes when it is enabled? (Or is there such a thing as debug levels?)]
|
|
[1:04:48][@Psi9][Why not use a simple template? That would remove storing the ID, e.g. template<int ID> struct timed_block {};]
|
|
[1:05:50][@insofaras][Would you agree that there is a similarity between test driven development and "writing the usage code first"?]
|
|
[1:06:32][@ChaiLatteQuinoa][Would you consider functional programming for game development?]
|
|
[1:08:01][@ingenero][How does the StartCycleCount get stored in the struct when you don't explicitly set StartCycleCount = StartCycleCountInit or something what you did with the ID?]
|
|
[1:08:55][@jfcatalan][Do you get Heisenbugs, where the bug happens only when there's not any debug code running?]
|
|
[1:10:18][@InstantKafka][So the destructor gets called right when it goes out of scope?][quote 203]
|
|
[1:14:08][@abnercoimbre][Community fight]
|
|
[1:14:25][Close this down and wrap up another episode of Handmade Hero][:speech]
|
|
[/video]
|