I built a game engine from scratch. Here’s what actually happened.
Not a tutorial follow-along. Not a game jam entry. A deliberate challenge — no engine, no shortcuts, just C++ and a question: am I actually a good engineer, or do I just consume knowledge?
The Origin
Not a tutorial follow-along. Not a game jam entry. A deliberate challenge — no engine, no shortcuts, just C++ and a question: am I actually a good engineer, or do I just consume knowledge?
How it’s built
Core loop
Poll → Update → Render
Fixed real-time tick. No abstractions between stages.
Physics
Manual, no library
Each object carries its own velocity vector; update step integrates directly.
Rendering
SFML at draw level
Object batching, state updates, and frame pacing are all manual.
Collision detection
The most algorithmically interesting problem. Naïve pair-checking gives you O(m × n) per frame — fine for a handful of objects, unusable at scale.
The honest timeline
Day 0 — The setup wall
SFML 3.0 was installed. Every guide online was written for 2.x. The API had changed, examples didn’t compile, and error messages weren’t helpful. The unblock came from a single article by someone who’d hit the exact same wall — and documented the 3.0 differences clearly.
Hour 7 — Almost quit
After a straight 7-hour session, I genuinely didn’t want to continue. Not because the problem was unsolvable — just deep fatigue. Took a break. The break was good. Going back was hard. Went back anyway. That’s the engineer’s mindset in its most unglamorous form.
The click
First time the loop ran smooth at a consistent frame rate — no stutters, no lag spikes, just clean ticks. It wasn’t just that it worked. It was that I understood why it worked, because I had built every part of the reason.
What changed
Before this project, I used tools. After it, I look at tools differently. When I see a game engine now, I have a mental model of what’s underneath — the loop, the collision pipeline, the render pass.
I understand why entity-component systems exist and what problem they actually solve. I understand what you give up when you reach for an abstraction too early. The O(m × n) → ~O(m + n) reduction wasn’t academic — it was something I debugged, measured, and felt in the frame rate.
Scope
A foundation for understanding, not a product. The goal was to reach the point where nothing in this space feels like a black box. That goal got achieved.