I finally finished and released the Arrows video.
It was supposed to take 3 months, but it took many many more. Why?
The vast majority of my time was spent on coding the damn thing.
Early on, I built a layer system that lets me layer different visuals on top of each other. I can place down a canvas, or a shader, or some web stuff, or anything.
The system itself is good. It lets me pick the technology that’s best for each scenario, and combine them together. I can layer up all sorts of different things.
Except I didn’t actually do that. Instead, I just built everything out of normal web elements because it felt new and shiny. And I suffered greatly for it.
I even built another layer system on top of that layer, instead of using the in-built one! Waste of time.
I spent a lot of time building a signals system. It let me make my code ‘reactive’ so that it responds to changes in the right way.
For some reason, I really wanted it to handle everything automatically. I thought it would be neater, or something.
I found it really hard to make automatic signals work well, so I eventually added a way of doing it manually instead. I should have done that from the beginning.
Because of these new systems I was using, I had to spend a lot of time making arroost faster.
It was new unfamiliar territory for me, so it took ages. I should have stuck with what I knew.
I also built a new entity-component-system. It’s like a more advanced version of the approach I used in cellpond.
I made it way too complicated, and I eventually stripped out almost everything.
When I started, I thought I could make everything within one system. I incorrectly thought that my new systems would be powerful enough to support everything I needed. I thought wrong.
I had to split arroost into two separate projects. Arroost became just the user interface. And I made Nogan the underlying machine.
This means that I had to abandon work that I’d already done, and rebuild it in a new way.
And I had to spend extra time getting them to work together.
Essentially, Arroost is just a ‘projectional editor’ for Nogan. In theory, it should have a 1:1 mapping to it. It’s just another way of representing the same information.
But it doesn’t work like that under-the-hood. The two engines have to do a lot of back-and-forth communication to try to stay in sync. And they often got out of sync.
The biggest remaining bugs are caused by this.
When I continue work on Arroost, I will probably try to make them more tightly coupled.
I wasted time trying to handle audio myself.
Things sped up when I switched to the tone.js library.
At first, I built nogan without types. It was all very fiddly, and I knew I would make mistakes, so I added validation. If anything was incorrect, it yelled at me.
Eventually, I added types, and it did a much better job at pointing out my mistakes.
The validation was still useful, but I didn’t need it for everything, and it often got in the way.
I turned it off for certain things.
I spent too much time figuring out how to get a satisfyingly curvy arrow for wires.
In the rewrite, I didn’t bother re-implementing it.
There’s more!
Initial plans for Arroost looked very different. The video was originally going to focus on a grid-based system, like a classic cellular automata.
Early on, I realised that I needed a node-wire system in place to be able to support that. This massively grew the scope of the project.
I was emotionally attached to the grid plans, so I didn’t drop them. This meant that I couldn’t take certain shortcuts when engineering features.
I dropped those initial plans far too late in the video’s development.
I switched from adobe software to davinci software. It took a long time to learn everything again.
I re-recorded my voiceover two times due to script changes. I was too certain about certain things.
I should have figured out more of the engineering and design before committing to a video plan.
Normally it helps to have a voice take done early, because it helps me to arrange the music and structure the video. But there wasn’t any music to arrange this time, so it was pointless.
I had to make all the music myself, which took longer. Luckily, everything else took even longer, so I had plenty of time to figure it out.
Still, it took a lot more work than usual.
There’s more!
My joke language went viral. I received a lot of attention from it. I allowed myself to become distracted by it - doing talks, interviews, and so on.
I got a lot more patrons.
I used to message every new patron individually, and I manually added them to the credits screen. I did this for way too long.
Eventually, I setup an automated system, which took some time as well.
I started doing my weekly update as a podcast. Although I am really pleased to be doing it, it did take up a lot of my time.
I agreed to do a talk about cellpond.
This involved a lot of preparation and planning, as well as some coding work on cellpond itself.
All this time, I’ve been working at tldraw four days a week.
I made a lot of mistakes.
But something got me through in the end…
I picked up a new mindset along the way. I’ve been trying out a relentless embrace of simplicity and minimalism in my work. I call it the “slippy mindset” or the “tadi web”.
And it helped me get the video across the line.
Thank you tadi web!
And thank YOU for your patience.
Hopefully the next video won’t take so long.
Back to the wikib.