ꙮ tatybara ꙮ

I flatter myself that I'm quite an accomplished programmer with a number of complex, impressive, successful projects under my belt. And every time I've tried to pick up video game programming for anything more complex than a Twine game my brain runs straight into a brick wall. There's something about the deep simulation-y nature of it that just doesn't work for my thought processes the way programming something to do a task does.

i'm a web app developer and not a game developer, but this is broadly applicable. using a lot of conditional logic ALWAYS feels slimy, even when it's necessary. it feels unsustainable and error prone. additionally i've absolutely had moments where i was relieved to discover that authoritative voices on the frameworks i use were relying on the same shaky-seeming solutions to problems i had come up with. i think this is part of what contributes to the imposter syndrome epidemic in this field.

anyway, relevant xkcd: https://xkcd.com/2030/

Professional gameplay programmer. I LIVE for this kinda stuff, and yeah, it's a very different ballgame than most other programming fields. You basically have to live by a few paradigms.

Programmer time IS iteration time, and iteration time has to be hyper low in video games. The faster you get a working solution, the better.

Gameplay gets nothing for CPU cycles. If I want to hit 60 frames a second, my whole game has to happen in 16ms, and gameplay maybe gets 1 whole ms if I'm lucky some projects. The stupidest most literal operations are usually the cheapest, and the heavy weight math solutions aren't performant enough.

On a related note, stupid code is easier to work with. Video games are pivoting huge swaths of "functionality" in programming terms sometimes daily. If you try and make every piece of the game a clean, usable, correct API, you will never finish a game. If your script just adds 5 to the position of an object every frame, it's stupid easy to change that with almost no technical knock on effects. Again, iteration and performance are everything. Gameplay programming is the bottleneck of most design iteration.

And lastly, the thing that trips up newer gameplay programmers the most, is that you should never solve problems you aren't actually having. You might think "oh if I design in this more general way I might be able to use it for X, Y, and Z in the future" and that gets you killed in gameplay programming. You're doing a ton of work for a hypothetical you don't actually know exists yet, and the majority case is that you discover more with faster iteration that disproves or negates those hypotheticals left and right. The most abstract you get, the more difficult it is to change when your model is proven wrong, changed by designers, changed by technical limitations, etc, etc.

It's great, I love it here, gameplay programmers are a horror to other programming disciplines!

This describes my exact frustration with game programming when I'm operating in "technical brain" mode! Which is like, half the time. When I'm operating on that wavelength, I become so fixated on doing things "the right way" that I often don't begin at all, instead falling into the "beginner trap" of endless tutorials and preparing for edge cases that will never happen.

The other half the time, I'm in "art brain" creating elaborate systems and things that I later don't quite understand when it comes time to debug them.

The only trick that has helped me is to not think about it just long enough that I start on something, but not get in too deep that I can still follow what's going on when "technical brain" kicks in. This is a big reason why I gravitate towards WYSIWYG or visual programming driven game engines and construction kits, since it takes care of the systems part so I don't have to worry about doing things "right".

Having no formal computer science background I am compelled to say: why do math, when if do trick?

I felt this way when writing the camera routines for ROCKFORCE, an in-progress Doom-engine 2D platformer run-n-gun type.

"ok, the camera always being centered at the player is kinda really jank... great, I'm gonna have to write up a spring tension solver and have the viewport tied to a sproingy tether with a low k-factor... wait.

so like... what if the player character 'pushes' a box surrounding them, that sits on the bottom of the level, and it's tied to the level camera. no, that can't possibly be-"

that's exactly how it works.

hard agree honestly. I think it helps to remember that video game concepts were mostly invented when computers were really slow, and doing anything the "right way" was pretty much impossible. So everything was hacks, just trying to get something working and it turns out that the hacks were general purpose enough that things kept working that way.

honestly yeah, teaching myself how to do any amount of programming has always felt Draw The Rest Of The Owl to me. great way to put it. for me, there's just some fundamental Stuff i'm missing in the whole process. and sometimes a lack of knowing when i'm missing something. and how to find things i don't know i'm missing...

I get this brainworm so bad that I wrote my own game engine just to be more "correct" and "functional" like I am in my other fields of development ... and I still ended up sometimes just having to go "look it collides if it's 5 pixels from center, do not make me do box math" in places. And the underlying graphics code to make that beautiful, elegant, functional-declarative design is a mess, because Canvas is just Like That:tm:.

On the other hand, monkeying about in PICO-8 was kinda liberating, just jabbing stuff in variables and brute force flinging pixels and sprites around ... it's just that kinda dev I never get too far with because the mental model required for all that game state starts to exceed my grasp.

I am an experienced programmer but have always struggled with trying to make games, because my math is quite poor and my spatial reasoning is pretty bad too, and it turns out those help a lot for making games.

Anyway, one thing I found that attempts to address the "draw the rest of the owl" problem is Handmade Hero, where he starts from scratch to make a game. It's like 600+ videos long at this point, but is very incremental, because you see literally all of the code he ever types.

really cool post, thanks for writing this up. i'm sort of on the opposite end of the spectrum -- i love the wild west nature of game programming and all the hacky nonsense feels natural to me, but application programming has always felt overly clinical and claustrophobic to me.

learning game programming also means learning to love hacky ugly solutions that make you go "Seriously? 50 Years Of Progress And This Is Still How We Do It?" and there's not a single game engine in the world that can make that okay

i feel like a lot of disciplines and skills quietly delegate The Rest of the Owl as something to be taught by someone else. which creates a huge gap that is very difficult to cross, unless you know someone who knows the skill, can teach you, and do so in a way that is helpful to you.

(gamedev is specifically egregious because of the aggressive deadlines forcing you to compromise on reasonable solutions)

god i feel like this also applies to the simulation stuff i do

a huge amount of computing is like this, not just games.

• how does a kernel decide what pages to put in the page file vs keep in RAM? just drop the least recently used ones.
• how does raytracing work? just toss a ray at the scene and randomly sample where it bounces. average the results.
  • this techique is actually so robust that "do random bullshit and average the result" is a Named Algorithm (Monte Carlo Simulation).
• how do we lay out digital logic chips? scatter all the cells on the board and tug them around with virtual springs until they fit, with several layers of correction and postprocessing to slightly make up for the fact that this is stupid.
• how does an OpenGL driver decide if a buffer should be in vidmem or sysmem? just count how many times it's used from both places, and move it to the most frequent one.
• it keeps going.

never be afraid of the dumb solution, it's usually fast, and it's easy to debug and make more complex later. doing the complex one first is only good for job security.

Oh my god, as someone who knows nothing about programming and has to learn on the spot for shool, sometimes even during exams (they were open book exams, I had access to internet), it feel so accurate. I still know almost nothing about programming, it is a constent struggle, and it seem like other people have it so easy! And that in-between case 1 and case 1 is exactly where I have issues.

Implementing collision systems, platformer mechanics, camera systems, rendering systems, dialog systems, etc. is actually just quite hard, programming-wise.

From the outside they're seen as basic, fundamental parts of games, which makes it seem like they ought to be easy and intuitive to build too, but unfortunately they're just not. You have to develop a lot of knowledge, intuition and experience to make complex games and that takes time and practice. There's no tutorial that can replace that.

One (free) ebook I've found extremely helpful is Game Programming Patterns by Robert Nystrom. It covers the architectural side of game programming, which is rarely touched in tutorials but frequently the hidden wall people run into (it was for me at least).

This rings hauntingly with my struggles at video game programming, having failed to make several at this point. I've settled myself with the question of how to make sense of it (though not how to do it) kind of like this:

Games as a concept are fundamentally analog, because the point is to be interacted with by squishy bodies. But they all have rules so that there's a mix of skill and chance to keep things interesting. Some of those rules involve taking turns.

As we know, a digital computer works by dividing everything into numbers, including time. So even in a supposedly real-time analog-like game, every frame (or tick, world update, etc.) is represented by a discrete slice of time where everything happens at once.

In other words, every computer game is turn-based, just usually with most of the decisions being hidden away from the player. And every pixel is a board square. And then you get into the scripted sequences, physics hacks, asset streaming, client-side prediction...

So most of what the computer's being used for is to pretend the rules are different than they are. Hmm.

I've only ever shipped one video game, and I didn't make the movement or combat parts, but one thing that sticks out for me was learning that invincibility frames were invented to keep characters from taking infinity damage when they get hit, and that's basically still how that's done.

"the idea of skipping pixels seemed, and still seems, completely wrong to me. surely that would make the gameplay jittery??"

It does! In fact, in Super Mario World, speedrunners have to evaluate how jittery the screen is to determine whether they're flying at the fastest possible speed, and try to adjust to the optimal speed by tapping right while in flight which cycles the sub-pixels-per-frame at which mario flies, hunting for the fastest option.

Game dev is wild, and game devs are geniuses, and the clever ways things end up finally coming together to form a cohesive system is just staggeringly cool a lot of the time.

explanations of glitches in video games often do a great job of showing what assumptions the game’s systems are making and in what situations those fail. someday i want to make a compilation of such glitch explanations in an educational format.

in this overview of Jellycar’s softbody physics, at 1:38 they explain a primary assumption that the collision detection makes, and then show how it can fail. while technically a “bug”, it’s not something they consider a flaw with the system, because keeping the system acceptably functional, simple, and efficient is a greater priority. there are a ton of other great “wait, that’s it??” realizations throughout this video too.

No kidding - I have a math degree and a lot of experience in programming, and game programming in every game programming framework was hell for me.

I wasn't able to actually start building a video game until I started programming in Rust.

this is one thing I really enjoy about (my beginner attempts at, must underline and bold caveat here!!) reverse-engineering code, the journey from "there's no way it's doing THAT, I must have interpreted this wrong" or "I don't understand this at all so it must be something elegant and complex" to "...no it really is doing THAT."

re: the Rest Of The Owl issue: the missing middle is most likely because everything between beginner and advanced is dealing with specific sub-problems. so say you want to run a webserver. the beginner guide says "here's how to do a basic setup with Python/Django, good luck!" the advanced guide says "here's an innovative and terrifying way to improve the performance of Django," but all the intermediate info is in "why am i getting a type error here?"

most of this info exists but most programmers make the leap from beginner to intermediate via school, and exit school into a world where they have specific error messages they want help with. that's my theory, anyway.

On the subject of "I found exactly two kinds of tutorials..."

One thing I've noticed is that as languages have got 'simpler'/almost usable by normal people. The tutorials have got much worse.

in the ages of C/C++/Java/etc., the most beginner tutorials tended to be "Install IDE ABC_XYZ, then we'll start doing some printing and looping. We'll get to the complex stuff later!"

where Python/Node/etc. tend to be "DO NOT EVEN THINK ABOUT DOING ANY CODING UNTIL YOU'VE SET UP VIRTUAL ENVIRONMENTS, PACKAGE MANAGERS, A VPN TO THE REMOTE SERVER YOU OBVIOUSLY HAVE ACCESS TO, REMOTE GIT REPOSITORY WHICH YOU CLON...."

I'm reading this and thinking back to how all those old computer magazines would just have the example games where you hard code in every aspect of it and I'm genuinely starting to wonder if that's just the entire field. Is it really just hack hardcoded workarounds all the way down?

see at least for me i think that “mathematical beauty” and those things you wrote are not exclusive. mario.x += mario.speed is great when the only thing that can happen to mario’s x is it can be changed by speed but stuff like Player.cs is imo a direct product of that mentality, just adding more and more special cases. and it probably does work but it ends up being harder to use than a generalized system once you get to a certain level of complexity or resource limitation.

a lot of programmers have been burned too many times

it's surprisingly easy for a handful of simple solutions to pile up and tangle together into some unmaintainable mess of hellcode that can't be replaced because every single piece of the company's software somehow ended up dependent on that hellcode and tied to its particular implementation details

working on old B2B software is more like archaeology than programming. you spend a ton of time digging ancient fossils out of the code, and trying to figure out if they're gonna come to life and bite your arm off when you try to remove them