Make Gauss-Seidel great again! The art of unrolling

Fortran community member loiseaujc described using AI to reimplement a PNG and GIF encoder and decoder in Fortran from scratch, saying the work would not have been feasible by hand given time constraints. The commenter argued that AI makes reimplementing software from scratch easier while still providing a better understanding of the details than relying on existing libraries, and suggested LFortran with Jupyter and JupyterLite as a browser-based, installation-free development setup.

@loiseaujc https://fortran-lang.discourse.group/u/loiseaujc thank you for this excellent answer I could spend hours discussing, but I’ll just say quickly that I think LFortran + Jupyter, lately including JupyterLite so that it runs in the browser on a statically-hosted webiste, no installation needed might be a great solution. Regarding why program at all when we have AI: I would say AI actually allowed me to reimplement from scratch many things that I had to use libraries for before. An example from just this month: PNG and GIF encoder and decoder in Fortran using AI, there is no way I would have time to do that by hand, but using AI and studying the sources, it’s actually not complicated at all, and I have a much better understanding of the details than I would have otherwise. Of course, if I did it by hand, I would understand it even better, but I don’t have time for that. But even before AI, I always liked to reimplement things from scratch, it’s the only way to understand them for me. But it’s sometimes hard to justify the time but even before AI often in the long run it would pay off to reimplement from scratch, I have done that many times in my career . But with AI it’s actually quite easy to reimplement many things from scratch.