What programming/technical projects have you been working on?

Just got the MVP of a raspberry pi-based Home Assistant tablet up and running!

It wasn't too bad in the end. I mostly ended up struggling with the poor performance of my model 3B+, but I had an underutilized 4B I've been using for another project. I swapped those out and most of the issues are gone.

Now I need to work on all the details - a dashboard specific to the tablet, a case/stand, etc.

Wish I had something interesting to share in one of these threads, but all I have is not really a project per sé. Recently I dusted off my Codecademy account and thought I'd try learning Lua.

Then I remembered why I never tried it before, except for when I briefly delved into WoW addon development (I tried creating an addon designed for sharing/distributing player blacklists to tackle in-game toxicity, but the amount of tutorials around are dogshit and I could never get my head around the Blizzard API or Ace3 framework.) Installing Lua on anything other than a Unix-based OS is a pain in the arse, and there is no official installer for Windows.

Also, I don't like what Codecademy has become. Every single time you complete a lesson, it keeps trying to upsell you with multiple ads about going for the Pro package and claiming that you could use the site to build a career. Like... bullshit can I pay £200 for an annual Pro subscription and somehow within a matter of months end up good enough to land a tech job. My brother is a software engineer working in a FAANG company, and I know what it took him to get into his career path (years of web dev experience as a teenager, a Computer Science degree and climbing up the corporate ladder.)

I feel very shy posting about this on a site where I get the impression that many (most?) users have technical/coding backgrounds. I just started learning how to "code" with Godot Script (GDScript) so that I can make a card game in Godot Engine.

It's a spiritual successor to "Underhand" which existed long ago on the Google Play store and I'm not sure how many (if any) folks might remember it. My game will be built from scratch with all new assets but the mechanics will remain largely the same. You are prompted with choices that either reward or take away resource cards with the end goal of summoning a god if you are able to meet all their summoning criteria. That's not the best explanation but I'm trying to keep it brief.

Anyway, from a non-technical background, even opening the engine was very intimidating. There's so many things I don't understand and I haven't found a guide that will help me understand how to do all the things I want. I've considered using AI to explain things to me but I'm worried that it will teach me how to do things incorrectly and I'll have no way to verify it. There are tutorials for understanding the basics of coding but how do I take those things and implement the mechanics I want in my game?

I wrote all the new cards with their options and rewards/penalties in an Excel file because when I started I didn't know what a JSON did and now I'm not sure how to transfer everything I wrote into the format it will need to be...All this to say, it's been a very humbling and frustrating experience haha. I hope I can find a way forward to continue creating while learning.

I've been designing my own keyboard layout. Most existing layouts largely assume a one-key one-character correspondence, and there's tons of info on the consequences of the placement of those keys. With seeing keyboard layouts with a thorn key or a magic key with few macros, I wanted to see what a keyboard layout could be like if I embraced the idea of one or two key press(es) entering common sequences of characters - not just bigrams like "th" but perhaps trigrams, 4-grams, etc. (think words ending in "tion" or even common words like "which"). How far can this be taken? How many key strokes can I reduce typing by? Would it even be comfortable to use? There's far fewer resources on this.

So my programming/technical project I've been working on is analyzing a dataset of already tabulated 1-grams through 9-grams. I'm in the process of identifying which N-grams I should consider as definitions of my macros. One idea is to dedicate a thumb key on my Ergodox as a one-shot layer containing the macros. This would open the door for defining something like 30 macros which can each be entered by a sequence of two key presses. I'm also considering demoting some letters to this layer so that I could have some single-key macros (think how the bigram "th" is more common than many single letters). I'm trying to be systematic and thorough though - defining a new "th" key, for example, will reduce the number of occurrences in which I need to use the "t" key. Same applies if I have 29 other macros, some which will overlap. I've been breaking my brain on how to do the math correctly on this recently without having the corpus itself.

I’m always a little nervous posting here will cause people to deanonymize me but I think this project is broad enough.

I’ve always envied those Linux power users that can boot up a new machine, run a script, and have the operating system customized exactly to their preferences.
However, I’ve tried Linux as my working OS a few times and never could quite click with it; too many good GUI apps on MacOS that I like to daily drive.

Recently, my computer broke and I needed a fresh install of MacOS. So, I’m building a script to do all of my MacOS customizations from the command line.

This project has been involved. Some sidequests include:

• Sketching a dependency tree to figure out what order my configurations need to happen in, since I start with just a Z shell with the default environment.
• Deciding on a semi-secure way to avoid many superuser authentication prompts without running the entire script as the superuser (for MacOS reasons).
• Integrating other objectives into this project (like dotfile version control).
• (In-progress) Building a small CLI tool to instantiate a bunch of the same environment variables in a shell-agnostic way.
• Learning my way around osascript and a couple other MacOS command line tools to do all the OS-specific customizations like system preferences.
• Trying to minimize my use of osascript by just putting app-specific configuration files into Application Support.

Normally, I’d get exhausted learning this much. However, this project has activated my sense of “holy shit, I’m a wizard” in a way no other project has.
Finishing a piece of the project, running it, and watching my OS just transform in front of me is super fun.

I’ve already committed to show my work to a friend by factory resetting my computer when I’m finished. The stakes are high.

My current project revolves around my esoteric programming language, Funciton. It consists of Unicode box-drawing characters. Here's what the factorial function looks like:

           ╓───╖
           ║ ! ║
           ╙─┬─╜   ┌───╖  ╔═══╗
      ┌──────┴─────┤ > ╟──╢ 2 ║
   ┌──┴──┐         ╘═╤═╝  ╚═══╝
 ┌─┴─╖   │    ╔═══╗  │
 │ ♭ ║   │    ║ 1 ║  │
 ╘═╤═╝   │    ╚═╤═╝  │
 ┌─┴─╖ ┌─┴─╖  ┌─┴─╖  │
 │ ! ║ │ × ╟──┤ ‽ ╟──┘
 ╘═╤═╝ ╘═╤═╝  ╘═╤═╝
   └─────┘      │

I invented the language in 2011 and only posted about it on esolangs.org, but now I'm returning to it and making YouTube videos about it.

My first video series was about the language itself and how I built useful functionality in it (such as addition, multiplication, and yes: the factorial function lol).

I am now doing Advent of Code 2024 in it.

Another libgen mirror went down earlier this year, and as a result I have been exploring the rabbit hole of "can I download and store all of library genesis for a reasonable cost?". This comes out to around 4TiB of fiction (which I can fit on my current hard drives), 45TiB of nonfiction (which I cannot, but conceivably could), and 81TB of science papers and magazines (which crosses the line into "maybe start looking into alternative storage formats").

So I've landed on LTO-6 tape for now(*), and by spending entirely too much time scrolling through ebay, I have acquired a tape drive, the extra computer-bits required to use a tape drive, and 62.5TiB(**) of blank tapes, for a total of around $400. (Used-but-still-working hard drive prices are usually around $10/TiB, for scale.)

(* At some point in the future, I want to experiment with DIYing optical tape, which is a very cool thing that the industry seems to have abandoned for unclear reasons, but that can be a separate project later.)

(** Uncompressed size. LTO tapes have this bizarre marketing thing going on where they're advertised with sizes that assume some fixed compression ratio, which is Not How Data Compression Works At All. This is 25 tapes, each of which holds 2.5 real TiB or 6.25 imaginary marketing TiB.)

The other side of this is figuring out which things to download. The lazy answer is "literally all of it", but this is a big and messy dataset with a lot of duplicates. If there are five copies of exactly the same book and four are potato-quality scans I see no reason to waste the extra space on them. If there are hardcover and paperback editions with different ISBNs but the same content, those can also be considered equivalent, since I'm only storing the content. I've been playing around with smushing the libgen and openlibrary database dumps together to try to deduplicate to a reasonable level.

I'm glad someone else in my friend group has already taken on the role of Plex Server Person, else this would start to get silly.

Been putting in extra hours at work due to deadlines, and it's really eaten into any motivation to work on my side projects even when I have free time. To give you an idea, a lot of video transcoding for a live feature.

Those side projects aren't that interesting, but I've been thinking about getting a flipper zero and tinkering around with it. Anyone have one / have any opinions? I'd be interested to hear what people are doing with them

I’m launching my own thing for the first time. Feel pretty nervous about it, but just gotta go for it.

It’s a DNS-based load balancer (“Global Site Load Balancer”). It does healthchecking of various endpoints you tell it to, and will update DNS based on the results of the health checks. Right now I’ve got weighted round robin implemented, but will be expanding into geolocation-based soon.

As far as I can tell, it’s the only GLSB that is all of: open source, supports IPv6/AAAA records, supports DNSSEC, is dynamically reconfigurable and multi-user.

Anyhow, just started closed beta this week. Nervous and excited!

Still working on my aws hosted pihole for ad blocking. I’ve automated deployment via terraform and a couple scripts. I’ve automated Tailscale auth key rotation every 90 days. And Tailscale api key rotation. Now in working on automating defining it as the nameserver for my tailnet.

Learning a lot of terraform and GitHub actions. Any advice welcome!

Edit: Fudge. Just realized I have no permanent tfstate and therefore cannot destroy.

I've been working on a music library management Python script. This is my first Python code since university, I've been doing C++ and embedded stuff professionally ever since.

The problem:
My music library is mostly in FLAC format (lossless compressed for those unfamiliar). Lossless media takes up a ton of space, so it isn't ideal for streaming from a home server or keeping on a smartphone. The solution I came up with is to keep a mirror copy of my library, but transcoded (a.k.a. converted) to the Opus* format (lossy compression that is much better than mp3). I'm constantly adding to my library, so I need a good method to mirror any changes I make in the lossless library over to the lossy one. Transcoding the entire library to Opus every time I make a tiny change would take way too long, and I don't want to manually keep track of what changes I've made since I last transcoded.

The solution:
A Python script that scans the lossless library, saving info about each file (e.g. filepath, fingerprint in the form of either the last modification time or a hash, the fingerprint from the last time the file was mirrored to the lossy library, etc.) into a yaml file. Whenever I make a change in my lossless library I can scan it with the script to detect changes and mirror only the changes to the lossy side. All non-FLAC files are copied (or symlinked or hardlinked) and FLAC files are transcoded to Opus. It is multithreaded for extra speed.

It supports libflac, ffmpeg, and opusenc codecs (ffmpeg was very annoying to handle signals for). It can also test FLAC files for corruption. No third-party libraries were used scratch that, I forgot about PyYAML. It maybe works cross-platform, though I've only tested with Linux.

Comments:
I've used AI at work to help debug embedded issues in the past, but this is the first time I've used AI to help build something from scratch. I started with Gemini since that was the only one I could use without creating an account, then moved to Kagi Assistant recently since it's now available for all tiers. I'm not vibe coding, I'm asking questions to:

• get a starting point on researching some Python concept
• debug issues if I know the problem is in just a few lines of code I can copy and paste into a prompt
• learn what's available in the standard library for whatever problem I'm facing

This approach has worked well to avoid hallucinations. The answers that AI gives to these kinds of questions are super easy to verify with an internet search. I can tell that if I hadn't used AI I would have burned through all of my Kagi searches in the first few days just learning Python again.

If anyone has an idea for what to name this script I'd love to hear it!

*Opus is awesome. FLAC music typically has a bit rate of 1000-1200 Kbps, while the Opus bit rate that people generally agree is impossible to distinguish from lossless is 96-128 Kbps. That's a 10x reduction with no audible difference. For comparison, an mp3 would have to be 192-320 Kbps to reach approximately the same quality.

I wanted to write a simulation for a deck idea I had for Magic The Gathering.

Gyruda has a trigger that mills 4 cards, and puts an even Mana Value creature into play. Well lots of clones are even Mana Value, so you can re buy the trigger!

I wanted to know how many cards on average I would mill over depending on the density of clones.

There is a point where it becomes "deterministic" which is kind of cool!

I implemented a sparse array with a dense index data structure in Zig, with only three mutations: reserve (appends to the end of the array), release (releases any element in the array, leaving a tombstone in the index) and compact (removes all the tombstone and compacts the index). Release is O(1). When the array is compact, append is O(1). If the index is not compact, append may call compact, becoming O(n),

Because for my application the deletions are mostly batched, the cost of compacting (which is O(n)) can be amortized after deletions in this way. I'd call it once after a deletion pass, and then it would remain compact until the next deletion pass.

The next step is to benchmark it. Specifically, I want to compare it to the solution I use in my game: pool-backed linked lists. Performance is very good for the game already, but I just finished and released it so I have some time to play around with ideas that were harder to justify before release.

I have been starting to develop a playdate game. I have never used Lua before, so I am learning that at the same time. I quite like it, but I wish it had a type and error system as mature as typescript. There have already been multiple times where something wasn't working because I mistyped something. In my past typescript projects, those sorts of errors are nearly impossible, because you will get a ton of warnings that you did something wrong. Despite that, I think I quite enjoy Lua.

Right now, I am building a clone/spiritual successor to Adventure from the Atari. I didn't have an Atari, but I did have some "12 games in one" type systems with a joystick that had RCA connectors you plug into your TV. I really enjoyed adventure, even if I wasn't very good at it as a kid. I know it spawned the entire genre of adventure games, but I don't think anything really does what it did well.

So far, I have just been working on assets and tooling. I am using Aseprite, which is also scripted in Lua. There have already been some interesting challenges, most notably color. Adventure used color to differentiate rooms and castles. That is, of course, not possible on the playdate. I have been drawing castles with different designs, since they can't be differentiated with color. I wanted to play with asymmetrical castle designs, and decided to try making one look like it had fallen into disrepair. Here is what I ended up with. Let me know what you think.

Next up is getting collision working and implementing moving between maps. It has been so fun getting back into programming.

I've just finished repairing our microwave oven. We have it for around 10 years by now. It developed fading 7-segmebt LED display segments right around two years mark (we have 2 year warranty here) and started complaining about door open or just started microwaving on its own out of nowhere. Since I have some background in electronics ad programming and I know enough to open microwave oven, I went in.

I don't encourage anyone to do that! You can kill yourself just by touching bad wire even if the oven is disconnected!

There was one integrated circuit for everything including driving the LEDs of said display directly out of the chip's pins. It was predestined to fail at the time they designed it.

I threw out old electronics, made my own 7-segment LED display (using shift registers as I don't like multiplexing when it's not really needed), wired in Arduino and programmed my own software for the microwave. Luckily all he power electronics is switched by 5V signal to switch relays and there is safety wiring so that you can't run the oven with door open - the wiring is open as well (meaning no current can run through it). This was done 7-8 years ago.

It started misbehaving in the last month or so, so I had to go into the oven once more. Since then I own an oscilloscope which I used as the symptoms were pointing to "signal noise" - false readings of signal when pushing buttons or turning the time set wheel. Indeed, that was the problem. I had to make new protoboard PCB to decouple the signal (make it less noisy, more precise/reliable) and I even manged to find my Arduino sketch if I needed to make software fixes. Debouncing made zhe situation better. Not good, but actually good enough to go with it. I put the oven back together and it's working better - it still gets false reading here and there, but it's more like once every 50 clicks instead of every other one. I still could go for software tweaks but I chose not to (for now).

It's unbelievable how dumb the actuall hardware of the microwave oven is - you could just put a throw switch there and run the whole appliance by one switch manually. The electronics is there only to set/show time and do the defrost or power level stuff (which is actually just PWM but real slow, like 10 seconds on, 5 seconds off). And to break the appliance so you have to buy new one. Not me, though!

I've been working with my first client with my new business, and have the project in a pre-launch state I'm starting to be happy with:

Live site: https://TheatreBunch.com/

Dev site: https://dev.theatrebunch.com/ - note that entire sections have placeholder content, it's playing around with the design/colours for the most part.

I'm really happy with where this is going. I have a friend who does graphic design and SEO and I outsourced the logo design to her, then customized and tweaked it. (It was originally a light design and I made it dark, did the text and more).

I've recently started learning how to play the keyboard so I've been building some small web apps that use webmidi to help with that. There's options out there but they're either above my current level of knowledge or covered with ads and frankly I just wanted a project to play around with.

The first one I did over the last few weeks was just to remember the notes by name, it prompts me the note name and times how long it takes me to press the corresponding key on my midi keyboard and gives me some stats to track progress.

It's mostly evolving as my very basic level of piano knowledge improves so the next two enhancements are kind of in line with that:

1. I want to add prompts in music notation rather than just note names (e.g. show a staff with a G# note on it rather than just prompt G#4)
2. Also prompt with keys. First phase would probably be: "Using E major, press the F key" so that I can e.g. remember that E major actually uses F#, and then later on integrate that with the music notation so I get a staff with a key signature on it and a note.

I decided to sit down and learn zig, starting over the Easter weekend. I was planning to do advent of code 2024 in zig, but life got in the way. Ive gone through the ziglings course, and have started setting up the zig build system for advent of code. Its... not a bad build system, not my favorite so far, but so many leagues ahead of cmake and the like.

I've been learning elisp by making an automatic file namer for emacs. It just cuts each word, sorts them based on frequency, then sorts out the top 100 most common words.

The idea is from this horrid one liner from one of my bash scripts

FILENAME=$(tr '[:upper:]' '[:lower:]' < "$FILE"| tr -s '[:space:]' '\n' | tr -d '[:punct:]' | 
grep -vwFf .stopwords.txt | grep -E '.{5,}' | sort |
 uniq -c | sort -nrk1,1 | awk '{print $2}' | head -n1)