RAGUMAGU
Casey Muratori and Reinventing the Wheel
I present two cases where I thought
I have to make an app for that
, but ended up using
solutions that already existed. Then, I present one case
where I could not use existing solutions, so I had to make
my own.
Managing software processes
I had to deploy a project with some analytics (built with Python), a backend (built with Flask), and a frontend (built with React.Js). I needed a way to keep them running, with simple triggers to start, stop, and restart it as needed.
At first, I made many experiments with Python subprocesses and PowerShell scripts and executables, writing a driver to start and keep all the programs running. But that became a hairy mess quite fast: how would you restart the driver if it goes down? Its turtles all the way.
One way to do this is by running the program as a service. This was a Windows server, and Windows Services can be written in Python using the PyWin32 extensions. However, getting it to work, debugging it and maintaining it was quite hard.
I realized I could use a process manager. A process manager
like pm2 lets you specify all the programs and scripts to
run in a config file. You can also set up restart
strategies
Windows got in the way again: there is no init system
and running pm2 startup raises an error at
the time of writing this post. To have pm2 start on
reboot, you can add a PowerShell script to start up. If
pm2 crashes, you have to just restart it manually.
and cron jobs if needed.
Making vector images
I wanted to digitally re-create some figures from a scanned document, so I could view them in more detail and play around. Some of the figures were simple bar and line plots, and I could make them easily.
Some were diagrams of complex machinery, and making them isn't a straight-forward task. I had to get the scan as a vector image.
PNG to vector convertors didn't work well.
I tried extracting contours from the image with a Python script, and then tried converting them into straight lines and bezier curves. But the source image was noisy, and I could not extract meaningful contours from it.
I thought I would have to build an app that let me trace and save a vector image. Later, I realized I could just use Inkscape to do this. It seems obvious now, but it did not occur to me that I could do it this way, till I listed out the desired features I would have to capture mouse-x and y coordinates and handle click and drag events. I would also need an algorithm that would fit a bezier curve to pass through the points I selected, and I needed a way to move the nodes, to wiggle the curve, for fitting it better. I could do this in JavaScript, and show the diagram in the background with the HTML Canvas element. I also had to figure out how to convert the coordinates into an SVG and also learn how to save this. I ignored the need for an undo stack, zoom and hundreds of other issues that I would encounter when building this. for the app I would build. Inkscape even lets you draw in layers, so you can work on detailed parts individually.
Stepping through a video frame by frame
I was working on a remote server, and I had to step through a lot of video, frame by frame. I needed an app which was fast enough to whiz through lots of footage and gave me keyboard shortcuts to play/pause and goto next/previous frames.
VLC supports going through a video frame by frame. And many other solutions would do this out of the box. Surely I would not have to write an app for this, right? Right?
Wrong. The server didn't have an app for that. I could not install any apps on it, because of firewalls and other restrictions. I could not download to my local machine either, because the files were too big. :/
I ended up saving the frames to separate image files on disk, and wrote a small app (with Tkinter and Pillow) with a slider to scroll through.
⚜ ⚜ ⚜
If you know the fundamentals involved in a problem, you will know what is possible. You will know when to reach for an existing solution; you will appreciate and criticise it, and you will know when it is wrong. You will also know how to build a solution when you can't reach for the existing one. You might even come up with a better solution when building it.
This resonates with what Casey Muratori said at the start of
his
Handmade Hero
project, literally in episode 1. When a viewer asked the
question:
Why not use an engine? Why reinvent the wheel?
, he
replied:
If you start with an engine, then it changes what you're learning, from the fundamental truth of how to implement a game, to someone else's version of that. What you're really learning is that engine. You haven't learned how to make games, you've learned how to make games in that engine. Right? And if that engine were to disappear, for example, you would no longer know how to make a game at all. I'm not exaggerating this, that's just the truth.
It always bothers me when people say
why reinvent the wheel, because the wheel is an amazing invention. It is almost perfect -- or quite possibly perfect -- for what it does.It is something that so elegant and beautiful, that if you were to ask why we would want to reinvent it, I would say of course we do not want to reinvent it, the wheel is amazing. It has worked for thousands of years, and has been unchanged for thousands of years.
Nothing we have developed in the past 30 years of game-development, is a wheel. In the future, we will have so much more powerful tools for game development, that people will laugh at the things we did to make games. If we want to get to a wheel, we need people that will attempt to make the wheel.
Should you reinvent the wheel? It depends.
Should you know how to reinvent it? Yes. Wheels can always be made rounder.
TL;DR: Context is King.
Shrinidhi Raghunandan
Published: 1 March, 2024.