I've been introspecting for most of 2021 on where I wanted to go with Mu. I'd never had any expectation that it would ever be mainstream popular, but I had hoped to coalesce a small community around it, of the scale of say suckless. In 2021, after 5 years of prototypes I found myself asking myself the hard question: why are so few people making the leap from reading about it and starring/liking/boosting/upvoting it to actually trying it out and playing with it? Which led me to the question of, what should attract them to build things atop Mu? The tough answer to face up to was: nothing. Mu today can't really do much. In particular, not having network drivers is a killing limitation. All you can build with it are toys, and I didn't set out to build toys.

Though I do have an older variant of Mu that runs on a Linux kernel. That should in principle provide easy network access. I always felt ambivalent about relying on the kernel, though. What's the point of being so draconian about avoiding C in Mu and building everything from scratch, if my stack includes C from the kernel? And then I learned about how the firmware sausage was made. I'd built Mu to advocate against the constant push to build languages atop other languages, but I started to realize that complexity grows not just upward but also downward. I'd built Mu with some idea of minimizing total complexity of a computing stack, but along the way I started to realize that reality has inextricable complexity that's independent of what we're trying to use it for. The goal shifted from minimizing total complexity to "finding a narrow waist of reality," something that provides a consistent experience above it while relying on very little below it. In this new framing it stopped to matter how deep things extended below. Because really, we're always going to depend on a deep stack going all the way down to electrons and quarks.

Ok, so networking is hard to recreate and some C is ok but not too much. I started casting about for a minimal stack that's built directly in C (because minimizing layers of abstraction is still just good engineering). Lua fits the bill quite nicely. Linux kernel + libc + 12kLoC for Lua is arguably as minimal an implementation of networking as I'm going to get. And Lua packs an impressive amount of technical depth in 12kLoC to make its programs go fast. Graphics is still too much, though. The Lua eco-system already has https://love2d.org for anyone wanting to do graphics. I have nothing to offer there. But I hope to make progress on the original goal of Mu, before all the feature creep:

• Apps are easy for anyone to run. The infrastructure they need is easy to build.
• Apps are easy for anyone to run.
• Once you start getting used to running an app, it's easy to poke inside the hood. Small changes can be made in an afternoon.
• When modifying an app, it's difficult to end up in irrecoverable situations or bugs that are too hard to track down. (That's really what all of Mu's memory safety was aiming for.)

So I'm going to spend the next 5 years wondering why nobody's trying this out :D It's quite possible that text mode is just a bridge too far for everyone else. We'll see.

(original conversation with Andrew Owen)

This post is part of my Freewheeling Apps Devlog.

Comments gratefully appreciated. Please send them to me by any method of your choice and I'll include them here.