What programming/technical projects have you been working on?

In an attempt to look more hirable in the Austrian tech sector (and therefore achieve a work Visa so I can actually live with my partner), I've been playing around with spinning up Kubernetes (K3S on a set of spare RPI5 devices) instances- a tool that I've never really had cause to use in any real capacity as my primary experience in DevOps is largely personal projects. This is limited to plenty of self-hosting, Bash & Python scripting, Linux system administration and network configuration.

Unfortunately I'm finding the excess of weirdly named tools that the industry fixates on a little overwhelming, as well as the collection of paradigms that everybody appears to be obsessed with at the moment. Kubernetes, Docker, Jenkins, Terraform, Tailwind, Angular, Infrastructure as Code, etc. It feels like everybody is just ticking boxes and making stack acronyms (stackronyms?) instead of writing any actual solutions themselves. This is completely at-odds to my approach to computers (as an example, I'm perfectly happy writing a basic WebGL renderer from scratch using vanilla JS in ~26kB of code).

I'm hesitant to jump back into typical software engineering, which is closer to my industry experience, but I'm also struggling to convince anybody that the experience that I do have is worth enough (the games industry doesn't really seem to be taken seriously by the rest of the tech sector). There also aren't a large amount of games companies in Vienna, and the games industry as a whole is not particularly healthy right now.

I guess this is less of a "what am I doing with technology right now" and more of a slightly disillusioned rant about my current situation. Maybe I'll go back to university and get a Masters?

Casually working my way through Crafting Interpreters, though I'm substituting a language of my own design in place of Lox. It's a macro language for code generation. Think assembly language macros: including files in the middle of other files, deduplicating repetitive code, evaluating template strings with numerical expressions, etc.

As part of designing that macro language, I've been studying languages that solve similar problems, such as actual assembly languages. My limited experience with assembler macros has been that simple things are easy, while everything else is difficult/impossible. Contrast with Tcl, which offers extreme flexibility — but don't goof up your syntax, because you'll have a heck of a time finding the issue. For my own language, I'm trying to strike a pleasant balance between these extremes.

I'm pretty far from my ultimate goal (the side project for which I'm inventing this macro language), but I'm enjoying the journey, and that's 90% of the satisfaction I'm hoping for. But I do hope to reach the destination someday! I'll have to fight perfectionist urges to get there, though. The pragmatic Rust mantra of "Keep calm and call clone" is extremely relevant here, as is the general attitude of not overcomplicating things. For example, do I really need parser errors to have line numbers? Probably not. But stuff like that is tempting.

Per @creesch 's comment last week, I began developing an app for myself and some friends to use for organizing our Bonnaroo experience (an upcoming music festival). So far I've been able to distinguish and/or surpass some main hurdles:

1. Configuring the lineup screen for listing artists - finished
2. Configuring the daily schedule - in progress. Right now it's just a matter of configuring a custom layout that looks and feels like a daily view of the Google calendar app, but dividing it between the 5 stages for better organization
3. Include an artist bio page without adding it to the main navigation bar at the bottom. I understand this to be possible using navigation stacking, but the implementation of it has me scratching my head currently.

Overall I've made a lot more progress in a week than I thought possible, and I hope the main hurdles can be surpassed here soon, as the festival is in a month and some change.

Can I ask for some help from any shell wizards?

Right now, I backup my btrfs drives with btrbk to a RAID-1 volume that lives in my house. It's amazing, I love the flexibility the snapshots provide. To protect against theft and fire, I'd like to now schedule a periodic backup to the cloud as well, using the backblaze b2 cli. I'd like to keep everything as snapshots, so my plan is to use btrfs send to send the compressed snapshot to gpg, encrypt it, and pipe it to b2. Since the b2 cli now supports reading data from stdin, I can do something like this:

btrfs send /btrfs_roots/data/snapshots/some-snapshot --compressed-data | gpg --batch --passphrase 'test password, please ignore' -c --output - | tee >(sha1sum) >(backblaze-b2 upload_file my-bucket - test-snapshot) > /dev/null

That way I don't need any additional drive space to warehouse the data and I can get the sha1sum on the way out. The problem is however, backblaze doesn't compute sha1sums for files that are sufficiently large. These snapshots are ~500GB and will get larger so I need to provide my own sha1sum, which I need to do when I first invoke the cli. Obviously, I can't do unless I dedicate a large drive to being a staging area where I can dump the encrypted snapshot to. Alternatively I could not encrypt them but I don't want to do that either.

What I'd like to do is basically loop through the contents of the file and stream different chunks of the snapshot each time. Then I can compute the sha1sum them on my end, and make sure the sha1sums in b2 match. But I'm not sure exactly how to do that in bash. I think I can just invoke the above set of processes multiple times if I can filter the bytes in some way, but I'm not sure what tool will let me do that.

Finishing moving all Dockerfiles to Nix* stuff. It's painful but the result is great.

*) I'll see where the project mess ends up, I'm sure there will be something in the end.

So I've more or less completed the RFID Jukebox I started last week. It turns out I didn't need a seperate metadata system or a fancy player...good old mpd to the rescue!

I had already bought this USB-powered Aux-in speaker for something else. When the project bubbled to my mind, I cracked this puppy open to find that a Raspberry Pi 3B+ can easily fit inside....which I also happened to have lying around. The lights on the one side of the speaker also readily detach, and I was able to bust open the RFID reader I bought and fit the board in its place

Needed to get a right-angle headphone extension and a right-angle microusb extension in order to fit everything in the case cleanly. I probably could have hacked up my existing cables, but for $5 I was able to use these and be able to repurpose stuff later if needed. Added bonus, the microusb extension cable was able to slip out where the old cables used to come out, allowing me to use any microusb cable/charger that had sufficient amperage to power it all (5v3a works well).

Plug speaker + rfid reader into pi, cram all the cables in, I only needed to make one small internal cut to let some more wires into the center cavity. Everything fits into the package, and bam, all the hardware is there...just the poweron/poweroff is plugging/unplugging it.

For sofware, I installed regular raspbian, set it up as usual. Then changed to auto-login to CLI, configured pulseaudio to run as a system daemon instead of per-user. Installed mpd, cursed while troubleshooting (can provide more detailed fixes if anybody desires), and then ginned up a quick python script using python3-mpd2 that runs on startup.

It uses MPD's sticker functionality to assign the RFID card to the song, and then accepts input from the RFID reader (it just reads the code then hits enter) to play the scanned song. Child is young so more sophisticated functionality has been disabled/gated, but is now trivially simple to upgrade.

#!/usr/bin/env python3
# coding: utf-8
from mpd import MPDClient
from sys import argv

def setSong(cli,rfid):
  song = cli.sticker_find('song','','rfid','=',rfid)[0]['file']
  in_playlist = cli.playlistfind('file',song)
  if len(in_playlist) > 0:
    print('Playing existing song')
    cli.play(in_playlist[0]['pos'])
  else:
    id = cli.addid(song)
    cli.playid(id)

def assign_tags():
  cli = MPDClient()
  cli.connect('localhost', 6600)
  library = cli.listall('/')
  for song in library:
    print(song)
    rfid = input("Assign RFID: ")
    cli.sticker_set('song',song['file'],'rfid',rfid)
  cli.disconnect()

def main():
  cli = MPDClient()
  cli.connect('localhost', 6600)
  cli.repeat(0)
  cli.disconnect()
  while True:
    rfid_scan = input("Scan: ")
    cli.connect('localhost', 6600)
    # Just wipe the playlist
    cli.clear()
    setSong(cli,rfid_scan)
    cli.disconnect()

# Script starts here
if __name__ == '__main__':
  if len(argv) > 1:
    assign_tags()
  else:
    main()

You'll notice if you launch the script with any argument, it'll put it in 'batch assign' mode. The cards came in a neat 10-long sleeve, so I just scan the cards as fast as possible and it'll assign them in alphabetical order.

Because I wanted a tightly curated list of songs, and we use Youtube music, this handy one-liner let me download a private playlist, and keep track of what was downloaded so it doesn't re-download.

yt-dlp -x $PLAYLISTURL --embed-thumbnail --convert-thumbnail jpg -o "%(artist)s - %(track)s [%(id)s].%(ext)s" --download-archive downloaded.txt

This left me with a bunch of .opus files, with (terrible) embedded album art metadata, perfect for this use case. The whole system is now functional, and added bonus since it's both a pulseaudio and a mpd server to boot I now have dozens of potential future projects (casting it to other linux devices, vice versa, or running an internet radio station)!

Very pleased with how it turned out physically, and since I didn't need to make any big changes to anything, I can easily repurpose it later if/when desired. I'm sure they'll want better speakers at some point...

After a bunch of canoodling I have finally got a virtual Windows 11 system running on my server to play old games that aren't well supported by Wine/Proton. I just tried remote SPICE for the first time and I was rather disappointed in the performance; it's actually more responsive if I RDP to the GUI I have set up on the server.

My server is actually my old gaming PC, and it has the best graphics card I own in it right now. From what I understand I can have the card diverted so that the VM has full access to it. I'd be curious to see if anyone else has done this and if they would recommend it.

This week I spent several hours attempting to fix a Powerapps bug that affected our production environment that makes 0 fucking sense.

Suddenly, out of no where, I can no longer "query" our database if I try to do any sort of date filtering. Functions that have been working for 4 years have started to fail, and no changes have been made to anything related.

EXTRA stupid, is that if I copy any of the failing functions into a brand new app, it works!.

Even more extra stupid, the workaround has been to remove the date part of the filter while keeping the rest, wrap the entire function in a sort by descending (on date), then wrap that in a FirstN so I only get a few records, and THEN i can do the date filtering.

I have got to find time to really learn frontend in F# land because I am utterly done with powerapps. This is just totally unacceptable behavior in a production environment. Especially when I opened a ticket with MS on the 29th and they've basically just sat on their hands wasting my time, only responding when i prod them.

Does anyone have any suggestions for hosting a server? More specifically a Node.js rest API. I was using some of the free ones, but eventually their VC funding runs out and it turns to a paid service instead. It doesn't have to be free, but if you have any to suggest, great! I guess I'm looking for a cheap option for an API that will probably be rarely hit.

I've been hacking together my own customized static site generator for my blog! See context here

The end goal is to replace Hugo, so that whenever I want to extend or add functionality, I can just adjust my script rather than spending hours pouring through Hugo docs and forums hoping it supports my special use-cases.

It's written in Go, using text/template for gemini & email outputs, and html/template for html, rss, and atom outputs. Attempting to make yet another general-purpose SSG is a non-goal. It's why I'm planning to move off Hugo in the first place.