What programming/technical projects have you been working on?

A few months back, I found my old metal 160GB iPod Classic 7 and thought it was great to experience this time capsule from the 2010s. However, the battery life only allowed for 30 minutes of playback, and I suspected a swollen battery.

Then I got sucked into the iPod modding community and tried to pry open the iPod... Apparently, the 7th edition is notorious for being difficult to open. Long story short, I scratched up the front and broke the screen.

I then purchased a busted iPod Classic 5th, which was way easier to open, and I installed a new (larger) battery and a 256GB SD card. Now, playback on the stock firmware sometimes lags, which is to be expected since the database file for this type of iPod isn’t designed for my 100+GB of songs.

So, I installed Rockbox on it, but it constantly freezes.

In the end, this has taken way more time and caused more headaches than I expected, but it's nice to tinker around with it.

I am once again here to talk about the progress of my game.

Missing Update from Last Week - Save Files

I wasn't able to post last week when I was working on setting up save files for my game. Right now, there's just auto-saves at the end of each battle, then the player can opt to either continue from the latest save file they have, or to load one of the 5 latest auto-saves (I discard the oldest and limit it to 5... it's arbitrary, but also I'm too lazy to implement scrolling right now).

Eventually, I want the player to be able to save scum to their heart's content, but that's for later (serializing the battle state is intimidating because I didn't plan ahead for that, my bad).

Failure to Communicate

I attempted to make a video demonstrating this functionality (along with another one), but I bit off more than I could chew and ended up with a long video with lots of sped-up portions that nobody could understand, so I'm not going to offend your eyeballs with that here. I have to keep my videos bite-sized or no one will want to watch 'em.

Overwriting Save Files

From my understanding, the standard practice when dealing with save files seems to be to use atomic operations whenever possible. So... when there's a pre-existing save file that you're overwriting:

1. Write the new save file on a temp path (like a .tmp)
2. Move the old save file to a backup path (like turn it into a .bak)
3. Move the new save file to the correct path (the path you wanted to write to all along)

Some VFX Work

This week was more about setting up VFX for a "summon field" ability. The idea is that the player can place fields with customized effects that get applied on characters that stay on them at the end of the turn. Here's what it looks like so far (just 11 secs): https://imgur.com/qJGEvoK. The icons that float around in the field indicate what effect gets applied (or at least, I'm hoping players would understand it that way).

I started a new wikipedia backup last weekend.

During college I had used kiwix to make myself an offline backup just for funsies, but these days it seems that project might actually be of some use.

I used a backup from jan 2022, I cant decide if I want any of the post gen-ai backups or not.

This is a very minor thing, but I massively improved the LaTeX "build process" for a (pretty non-technical) friend currently working on their thesis. Now it consists of just running make, whereas before, well…

.PHONY: all clean

all: thesis.pdf

thesis.pdf: thesis.tex
	mkdir -p build
	xelatex thesis.tex > build/full.log
	bibtex thesis > build/bibtex.log
	makeglossaries thesis > build/makegls.log
	xelatex thesis.tex > build/full.log
	mv thesis* build/
	mv build/thesis.{tex,pdf} .

clean:
	rm -rf build

That was all it really took to "automate" this process away, plus a new .gitignore entry:

# Ignore (xe-)LaTeX build output files
build/*

First of all, for some (probably arcane 80s leftover) reason, using BibTeX seems to require at minimum two compilation passes, and you have to do them manually due to the in-between step.
Secondly, this effort is also made to have the nearly dozen auxiliary files "disappear" as best as possible from cluttering their directory (and the logging messages from cluttering their Terminal window). Today’s computers are fast enough that taking the performance "hit" from not reusing them (I’m assuming this is what they are for, at least partially?) is forgettable.

P.S.: Should anyone be interested in getting started with TeX, I can do nothing but highly recommend the short-ish Modern LaTeX book. It’s really not all that difficult, and if you’ve ever been annoyed by Word and the like, well, you already have a reason to try it out!

Still working on Gametje. Previously posted Show Tildes. It is a bit like a jackbox games. I had taken a break creating new games for a while to focus on the main page branding and basic UX experience but now I have started tinkering with a couple ideas in the new year.

One is a more child friendly game (to play with my kids) involving sort of a digital hide-and-seek + memory game. One player is the seeker, and all other players are trying to hide. The hiders will have objects they can hide behind and move between. The seeker must watch each player's avatar carefully and try to remember where they are hiding. The hiders are free to move about before the timer runs out. After the hiding is done, the seeker is asked to tell where each specific player is hiding. If they are right, they score points, if they are wrong, the hider scores points. Each person gets a chance at being the seeker (where the most potential points are scored). So far, I have only the frontend stuff somewhat working using some "borrowed" assets. If they weren't borrowed I'd share s screenshot/gif. I still need to design a few maps and assets to hide behind etc. Then I need to design the backend db and APIs.

Also working on another game. It is a bit farther along and is a different take on the mind meld game concept. In my version, 2-10 people can play. First, two random words are chosen (from player submissions). Each player is given the same 2 words and tries to come up with a new word/phrase that associates/relates to both words. Then everyone's submissions are checked to see if anyone matched. Players that match with each other (sometimes more than 2 at a time) score points and are done for the round. Players that did not match will continue. When at least 2 people haven't matched, two new words are selected from the previous guesses and used for the next "level". If everyone has matched or 1 person is left unmatched then the round ends. A new round starts with random submissions from all players.

Scoring is based on matching with at least 1 other player and doing it quickly. The longer it takes you to match, the less points you score. It's roughly: Base round score / (Math.max(1, (level - 1) * 2)) = net score.
e.g. Base round score: 1000
Level 1 match: 1000
Level 2 match 500
Level 3 match 250
Level 4 match 125

The backend is nearly completely done (API/cache/db etc), still working on the frontend. Should be playable by alpha testers in the next couple weeks if I find time. (To become an alpha tester, you need to signup and enable it in your profile).

Overall happy with my progress.

It's not that recent but I haven't posted it before. A while ago, I made https://whendnd.com/ for helping my D&D group schedule times. The pitch is that it's good at scheduling for a group with very low friction. There's no login required and it's totally anonymous. For an example game, see https://whendnd.com/g/tildes

I did this project partly to learn about SolidJS, which I've found that I like quite a bit. I still have gripes with it, but less so than Svelte 5 and SvelteKit (I used to use Svelte wherever I could). Recently, though, I turned the whole thing into an Astro site, and was impressed! Astro lets you add in components with SSR support, which is very handy, and the site got a lot faster with Astro than as only a SolidJS SPA. I'm leaning towards using Astro+Solid in other projects - we'll see how it goes...

A week or two ago I had said jokingly that the hardest part of my game engine was done, and naturally things have immediately gotten more difficult and I have made almost no visible progress.

So when I started the project, I had picked up three libraries that I thought I would need. The first one is called classic, which was supposed to be a simplistic implementation of classes and objects in Lua, which intentionally doesn’t have it. But in my struggles to understand the essentials of game logic and as I started to understand some of the less obvious Lua-isms, I actually ended up more or less recreating objects from first principles, so I haven’t really used it.

Actually the first one I decided to use was a löve specific Tiled map library. And while I don’t think I could have gotten this far without it, the library ended up having a critical bug in it that prevented referencing tile properties on “infinite” maps due to their chunked data structure. I submitted an issue and said that maybe the library should flatten it to the same format as fixed maps during loading to simplify the library, and a day later the author said he’d like to see my PR. A few hours later I did - a rather simple patch that only took about 10 to 20 lines that will probably be rejected because I was lazy and didn’t remove the rest of the chunk-related code that I had obsoleted.

Honestly the real reason I am upset about this was because it lead to me realizing that the codebase I had been doing these tests on had both an off-by-one error and in one place I had actually confused X and Y coordinates, and that took three times as long to fix.

The third library is for pathfinding. And now that I’m approaching the point of implementing it I am realizing that since I only need to traverse small areas the much bigger problem is figuring out the AI to determine where to actually move to and because of the somewhat weird data structures I have been working with it might be easier to write my own pathfinding algorithm.

My writing about this probably sounds like I’m bitching about it. I can’t help but see it like an outsider and look at the visuals and be disappointed. But the reality of it is that the more I look at my code the more I see the ugliness of it, and I think that’s fantastic. Yes, it’s a bummer because that means I’m almost certainly going to refactor it. But every single bit of ugliness that I see is proof of a lesson learned. When I started I was rather desperately looking for resources to learn the patterns necessary for game development. Now I am that resource. And I have come to realize that this is the first programming project I have really felt any pride in in years.

Unfortunately DuckDB fts doesn't have a trigram tokenizer which is useful for some of my programs. But this isn't a complete blocker so I'm still evaluating it.

I made a script to make searching fts. It even can create an fts index for you:

duckdb_fts.py test.duckdb table_name --pk path col1,col2,col3 --create-index
duckdb_fts.py test.duckdb table_name --pk path col1,col2,col3 -s search -E exclude

DuckDB has a lot of friendly syntax, and I must say a very beautiful EXPLAIN ANALYZE, but there are still quite a few rough edges.

For example:

SELECT
    m.id AS id
    , SUM(CASE WHEN h.done = 1 THEN 1 ELSE 0 END) AS play_count
    , MIN(h.time_played) AS time_first_played
    , MAX(h.time_played) AS time_last_played
    , FIRST_VALUE(h.playhead) OVER (PARTITION BY h.media_id ORDER BY h.time_played DESC) AS playhead
    -- , *  -- SQLite even lets you do this... but I concede that is a bit extreme...
FROM media m
JOIN history h ON h.media_id = m.id
GROUP BY m.id;

I get this error until I add all the referenced columns to the GROUP BY:

Binder Error: column "playhead" must appear in the GROUP BY clause or must be part of an aggregate function.
Either add it to the GROUP BY list, or use "ANY_VALUE(playhead)" if the exact value of "playhead" is not important.
LINE 6:     FIRST_VALUE(h.playhead) OVER (PARTITION BY h.media_id ORDER BY h.time_played DESC) AS playhead

If I change the last line to:

GROUP BY m.id, h.media_id, h.time_played, h.playhead;

I get results... but this is not the same thing. I want to get the most recent playhead value, but I guess that's not possible without making it a separate CTE or subquery:

SELECT DISTINCT ON (media_id)
    media_id
    , playhead
FROM history
ORDER BY time_played DESC

I've also had quite a bit of trouble converting existing DBs to duckdb because some of my file paths are not UTF-8 so I end up with invalid duckdb files and partial parquet: https://github.com/duckdb/duckdb/discussions/16049

The Minecraft mod I've been working on over the past few weeks is released!

• Modrinth
• Curseforge

Someone from the Fabric modding server took the time to test it and found a couple of things I'd missed, so I'm working on a little patch release to follow. After that I want to backport it to 1.21.1 at which point it'll basically be complete. I do want to maintain it for at least a year or to, so here' hoping Mojang doesn't do any gigantic rewrites in between now and then.

Back in December I got a Banana Pi BPI-F3, a neat little RISC-V board, to mess around with. I haven't had too much time so I'd just been running Luca Barbato's Gentoo image for it off an SD card but I've spent a couple of days trying to get it set up with my own Gentoo install on the eMMC. I've been struggling reading through lots of (bad) pages of documentation trying to understand opensbi and u-boot to get it to actually boot, but it's been interesting so far.

I've been having trouble with a lot of email marketing platforms, so I have just been implementing my own workflow for launching a landing page, storing email subscriptions, and sending out emails.

So I finally have the screen and rotary encoder for my DIY MP3 player that can run Spotify. However, it is my first time using a GPIO display (or even just GPIO pins at all) and am realizing that the display I got was cheap because documentation is quite sparse. After doing the install process, it is not displaying and documentation is slim for me to try and troubleshoot it. I am able to turn the backlight on/off so I do have some progress, but not much. Hopefully after figuring out the display I can start moving into the final stages of this project.

Edit: I feel like the best approach to figuring this out is to switch back from a CLI only version of RaspberryOS to one with a DE, and use that to help figure out display issues before going back to RaspberryOS Lite for the final product.

Edit 2: Upon further reading, it seems like switching to the I2C may be a good path to go down, as it will use less of my GPIO pins, freeing them up for the clickwheel. The I2C pins are labelled well on the board, except the VCC does not specify whether it wants the 3.3 or 5V, and there is a fifth pin, labelled "int" which is probably an interrupt signal which is non-standard for I2C apparently. I read into I2C a bit more, seems like it is not what I would want. I think best approach is to return it since I ordered via Amazon due to poor documentation.