Virtual Judge and the NoYesApp are cool. AI is real and obviously with some guidance and good vibes, you can build something pretty big.
Still, it is not exactly what this blog is supposed to be about. We are embedded people, not web developers. We live off of C code and register bit banging.
My searches on using AI with embedded turn up a lot of MCU companies trying to sell an on chip accelerator. There is not much about using AI code gen tools to build actual systems.
To figure out the best way to use the AI, keep in from running off the road, and get a better outcome than hand coding, we need a training wheel project. Something small to get the tools working before we take on a bigger project. You have a number of ideas, I’m sure, but doing something brand new on top of a new tech stack is going to get us in trouble.
Here is my thinking. The book, “Mastering Algorithms in C” was and still is a goto for my C coding. Let’s start with section II, all about data structures in C. In my college data structures class I learnt nothing. This book taught me what I needed to know. I did not need to know Pascal.
Second, is “Test-Driven Development for Embedded C“. I agree with the book and the ideas set out therein. However, in the world of work there is never time, the code is old and creaky, and we never actually get to build a test bench. So, this time we will.
What I want to do is use Claude Code for code gen, and mash the two books together. This is what AI thinks that should look like.
Note to both publishers, I will never ever, ever publish this book. If I do, I will spell “Algorhped” correctly. What!? Thanks Nano Banna. I also tried with Google Gemini. For some weird reason it thought this was a fantasy project with dragons over a city. I could not get it to forget the dragon wings. I gave up. Actually, the cover reminds us of cool things AI can do, and do them wrong. Perfect!
This will be a series of posts implementing the Section II data structures. My copy is so old, it comes with the diskette. (Yes, there is still a 3 1/4″ floppy in the pocket bound into the last page. I have no computer able to read the floppy.) The coding style may be a bit ’99, so the AI can help us modernize and get things secure.
Each post I will try to add some tools that we all know are good, but seldom actually use in production.
- Make file
- Full unit tests
- Doxygen
- Multiple compilers
- Clean lint
- Code formatting tools
- Memory checking
- Logging and tracing
There is the plan for the next few weeks. Thoughts or comments?
Leave a Reply