Thanks!
You can watch the talk at Ada Developers Workshop Vasteras 2026, where I explain some of your questions:
https://www.youtube.com/watch?v=d-RISfK9Sy8
I should write more details in the repo's README file.
Basically, I was quite frustrated with the available solutions vs. my requirements and ended up making my own, and this process took me many years.
Ada is a language that I always admired, but at the time I didn't feel comfortable depending on the available open-source toolchains. GNAT has actually been open source for decades, but a few years ago, with the increasingly accessible FSF GNAT ecosystem, things changed dramatically.
I think people in the embedded world should really consider Ada/SPARK, where you can formally prove program correctness in Ada itself, going much beyond simple memory safety. Also, Ada's mapping of records directly to registers with bit and read/write semantics granularity is unbeatable. I could go on, but plenty of information is readily available online. BTW, I still consider myself a beginner in Ada; I'm perhaps leveraging 10% of its potential. I really recommend giving Ada (or yourself) a chance.
Why not Rust? I considered it, but dropping support for versions earlier than Windows 10 scared me away. It is not an option for my use case, where industrial customers have old operating systems.
SDL3 was a perfect fit as a backend since it abstracts away the hardware perfectly and allows me to focus on my own business. It remains to be seen whether I can abstract it away and let the user choose the backend. I use as few features as I can; for example, for rendering, I use the good old SDL_Renderer API, which does not require specific GPU drivers or features and falls back nicely to software rendering. Drop shadows are simply software-rendered and cached; the same applies to SVG images and Lottie animations. In this way, I simplified the requirements a lot. Then the SDL3_ttf was god send for text.