iiinova — where your digital self keeps evolving

Andre Weissflog @flohofwoe.bsky.social · 12/10/2026

New sokol_gfx.h feature: fragment shaders can now write into storage buffers (previously writing to storage resources was only allowed in compute passes). Changelog item: github.com/floooh/sokol... New sample (needs WebGPU): floooh.github.io/sokol-webgpu...

0132

Andre Weissflog @flohofwoe.bsky.social · 08/10/2026

Also macOS system headers have a similar thing with `Ptr`, and don't even get me started about Xlib headers on Linux squatting on commonly used names :D

110

Andre Weissflog @flohofwoe.bsky.social · 08/10/2026

Wait until you want to use min or max ;) I'm not sure if 'small' has similar opt-outs like minmax, e.g. I do:

110

Andre Weissflog @flohofwoe.bsky.social · 08/10/2026

Or you can shoot questions my way, depending on topic I might also be able to help. I can also try to hook you up with Rene, I don't think he's on social media.

020

Andre Weissflog @flohofwoe.bsky.social · 08/10/2026

I have been summoned? :D René Meyer would probably be best to help with all sorts of information about computing in East Germany, he's also a veteran and wrote a couple of books about the topic too: www.amazon.de/-/en/Compute...

amazon.de

Amazon.de

120

Andre Weissflog @flohofwoe.bsky.social · 06/10/2026

Yeah, local LLMs is my future escape hatch when I can't simply burn tokens on my employer's bill :D And tbh: "Opus is good enough for everything". E.g. hopefully we're near the plataeu already and local LLMs can catch up quickly.

110

Andre Weissflog @flohofwoe.bsky.social · 06/10/2026

...e.g. I'm basically trying too find use cases that are entirely outside the current pro- vs anti-AI discussion and all the philosophical/esoterical bullshit coming with it ;)

100

Andre Weissflog @flohofwoe.bsky.social · 06/10/2026

...interestingly this sort of 'LLM-assisted-manual' workflow makes me *slower* for the actual implementation work (because using an LLM for code reviewing almost always brings up esoteric edge cases which then need to be explored and decided on), but this seems to increase robustness for first impl.

220

Andre Weissflog @flohofwoe.bsky.social · 06/10/2026

I'd recommend an LLM workflow for prototyping, planning, static code analysis and 'boring' automation work on the fuzzy edge of a project (tests, build scripts...). But at least for my hobby work I'll write all 'actual' code by hand (because I want to stay deeply familiar with the code base).

140

Reposted by Andre Weissflog

Aras Pranckevičius @aras-p.bsky.social · 05/10/2026

"Unroll a loop, make shader 10x faster", where I accidentally speed up Blender's Kuwahara compositor node on Apple GPUs (and elsewhere). aras-p.info/blog/2026/10...

26616

Andre Weissflog @flohofwoe.bsky.social · 05/10/2026

Critical regression bugfix in the sokol_app.h GL backend on Windows with AMD drivers (framebuffer format selection failed when a GL driver only advertises SRGB capable formats, apparently AMD does that). Details in the changelog: github.com/floooh/sokol...

github.com

071

Andre Weissflog @flohofwoe.bsky.social · 04/10/2026

PS: the sokol-samples repository isn't merged yet and the samples webpage hasn't been updated yet, e.g. the link in the changelog doesn't work yet - Github Actions has a slow day again

040

Andre Weissflog @flohofwoe.bsky.social · 04/10/2026

Ok, next "big" sokol update is out, this is the next step of the resource-update API rewrite (breaking changes in sokol_gfx.h and sokol_imgui.h). Details in the changelog (which turned into a small blog post this time): github.com/floooh/sokol...

1151

Andre Weissflog @flohofwoe.bsky.social · 04/10/2026

Ah good to know, I'll need to check why it doesn't work out of the box when just switching to the KC85/3 emulator (I guess the M011 module needs to be manually activated first, IIRC only the 16 KB RAM module is activated automatically by the /3)

000

Andre Weissflog @flohofwoe.bsky.social · 04/10/2026

...the simple-sounding idea of "let's add a handful functions to copy between GPU resources" turned out to be a pretty massive quagmire, because the various 3D APIs have all sorts of feature gaps, restrictions and bugs in that area :/

060

Andre Weissflog @flohofwoe.bsky.social · 04/10/2026

Hrmpf, turns out I'll need to write a shitty workaround for the broken pixel unpack offset in Apple's GL driver after all. Being able to copy from different portions of a 'staging buffer' into different target textures is very useful for Dear ImGui's "new" texture atlas code...

170

Andre Weissflog @flohofwoe.bsky.social · 26/09/2026

Vielleicht in den letzten 20 Jahren, die Goldene Ära der deutschen Synchronkunst war in den 70ern und 80ern (auch wenn die deutsche Version teilweise wenig mit dem Original zu tun hatte, siehe "Die Zwei"). Bei der DEFA gabs auch paar echte Perlen, zum Beispiel "Adolars phantastische Abenteuer"

020

Andre Weissflog @flohofwoe.bsky.social · 26/09/2026

Boilerplate (C64): floooh.github.io/tiny8bit/pla... 8 Bit Breaker (KC85/4): floooh.github.io/tiny8bit/pla... Dream On (KC85/4): floooh.github.io/tiny8bit/play

floooh.github.io

Tiny Emulators

000

Andre Weissflog @flohofwoe.bsky.social · 26/09/2026

Direct links: Bits Brew (CPC): floooh.github.io/tiny8bit/pla... Neonpulse (C64): floooh.github.io/tiny8bit/pla... Traveller (C64): floooh.github.io/tiny8bit/pla... ...

floooh.github.io

Tiny Emulators

100

Andre Weissflog @flohofwoe.bsky.social · 26/09/2026

New demo haul on the Tiny Emulators page (2x KC85/4, 1x CPC, 3x C64): floooh.github.io/tiny8bit/ind... The KC85/4 demos by Moods Plateau are especially impressive: sampled music on a simple beeper.

281

Andre Weissflog @flohofwoe.bsky.social · 24/09/2026

While sokol-procrastinating I gave the Tiny Emulators a little design- and UX-refresh :D (disclaimer: cobbled together via Claude Design, which actually did a pretty ok job to build an efficient page) floooh.github.io/tiny8bit/ind...

050

Reposted by Andre Weissflog

Mattias Ljungström @spacesofplay.com · 23/09/2026

minc compiler 0.9.15 now available from minc.dev More correctness, optimizations fixes, updated sokol headers to latest, two new sokol-samples (cmdbuf, sparse_ub) at (spacesofplay.github.io/sokol-sample...). Also re-transpiled latest box3d api with new features (spacesofplay.github.io/box3d-minc/).

193

Andre Weissflog @flohofwoe.bsky.social · 20/09/2026

Yeah, in an ideal timeline there would be a third edition of the K&R book covering C99, and C99 would have gotten a completely revamped C stdlib which makes use of the new features introduced in C99.

010

Andre Weissflog @flohofwoe.bsky.social · 16/09/2026

Tbf, the regional connections work mostly fine, at least IME. E.g. I can't complain at all about the S5/S5X between Halle - Leipzig - Zwickau, or Dresden-Chemnitz-Zwickau-Hof, and my "regional service" Erzgebirgsbahn. Even when there's construction work, the SEV works flawlessly.

110

Andre Weissflog @flohofwoe.bsky.social · 16/09/2026

...but those 5 minutes then become 10, 15, 30, 45 minutes, and that's always only announced right before the last time window becomes invalid. ...sometimes I'm lucky though and catch the *previous* train that was delayed and that way even save a bit of time ;)

100

Andre Weissflog @flohofwoe.bsky.social · 16/09/2026

OTH I can't remember when the ICE train between Hamburg and Munich was actually on time in the last two years or so. I wouldn't actually mind if the information would be better, but it usually starts with '5 minutes late' so you think "ok that doesn't using an alternative connection"...

110

Andre Weissflog @flohofwoe.bsky.social · 14/09/2026

...also in the same changelog update some words about improved test coverage via LLM-generated mocking libraries for D3D11, GL and Metal (and soon-ish: WebGPU and Vulkan). Those actually uncovered a couple lingering (but mostly uncritical) bugs in the sokol-gfx Metal backend :)

020

Andre Weissflog @flohofwoe.bsky.social · 14/09/2026

Minor sokol_app.h update: on Linux, mouse locking now works like in the Windows and Web backend (e.g. sapp_lock_mouse() only sets a 'lock-requested' flag, and the actual lock/unlock happens at most once per frame). Details in changelog: github.com/floooh/sokol...

github.com

160

Andre Weissflog @flohofwoe.bsky.social · 12/09/2026

No plans so far, also currently I'd hesitate to add more bindings without first "somehow" reducing the maintenance overhead per binding, and I don't have a good idea yet how to do that (LLMs can help, but ideally the bindings generation scripts would first be 'harmonized' a bit more).

010

Andre Weissflog @flohofwoe.bsky.social · 10/09/2026

Yeah, it was a short sweet era when software development could be done entirely with free tools. And just when it started to be fun, big tech has found a way to monetize every line of code via "tokens" ;)

110

Reposted by Andre Weissflog

Why485 @why485.bsky.social · 10/09/2026

Got caught holding the bag on a significant and surprise expense hitting me at a bad time. If anybody needs short term (part time, 3 mo. max?) commission work to write or polish the physics/controllers for arcadish jeeps, planes, spaceships, vehicles, etc. send me a message!

15024

Andre Weissflog @flohofwoe.bsky.social · 08/09/2026

stb_vorbis.h + sokol_audio.h seems to become quite the iconic duo ;) @aras-p.bsky.social also used this combo in his Syntonic Dentiforms remake: aras-p.info/blog/2026/04...

aras-p.info

Syntonic Dentiforms redux · Aras' website

020

Reposted by Andre Weissflog

Mattias Ljungström @spacesofplay.com · 08/09/2026

Added a new audio example, showing stb_vorbis decodes and audio mixing: minc.dev/playground/g... (click on it to activate audio in browser)

141

Andre Weissflog @flohofwoe.bsky.social · 06/09/2026

You missed the only good John Wick movie ;)

110

Andre Weissflog @flohofwoe.bsky.social · 06/09/2026

New sokol header: sokol_cmdbuf.h, this is basically a simple software command buffer to record sokol-gfx render/compute calls outside of sokol-gfx passes and then 'submit' them inside a pass. Details in the changelog: github.com/floooh/sokol...

github.com

sokol/CHANGELOG.md at master · floooh/sokol

minimal cross-platform standalone C headers. Contribute to floooh/sokol development by creating an account on GitHub.

090

Andre Weissflog @flohofwoe.bsky.social · 06/09/2026

Well, compared to East German homebrew computer keyboards cobbled together from model railroad switchboards that looks outright spectacular :)

160

Andre Weissflog @flohofwoe.bsky.social · 05/09/2026

Emulators can also expose information to debuggers that real hardware can't :)

110

Andre Weissflog @flohofwoe.bsky.social · 05/09/2026

Nah, no human should ever have to deal with Rust's uglyness ;)

021

Andre Weissflog @flohofwoe.bsky.social · 05/09/2026

TL;DR is basically that I don't want to use LLMs to gain speed (e.g. add more features faster), but to increase robustness and quality (and I would even accept *losing* implementation speed for that).

141

Andre Weissflog @flohofwoe.bsky.social · 05/09/2026

...to early to say though whether the LLM maintained tests are too detailed. But that's better than my old approach to 'under test' because it's too much hassle ;)

210

Andre Weissflog @flohofwoe.bsky.social · 05/09/2026

Well yeah, when you let LLM build the implementation, then it makes sense to let tests drive the whole thing. But that's not what I'm doing, the library implementation is strictly old-school manual work, LLMs are only used to automate some other things that are *not* the actual library code.

100

Andre Weissflog @flohofwoe.bsky.social · 05/09/2026

Important to point out though that I'm not doing TDD. The tests are mostly there to trigger all execution paths, check things like error handling, or catch regressions. They're not meant to define implementation behaviour before the implementation even exists.

000

Andre Weissflog @flohofwoe.bsky.social · 05/09/2026

Writing tests manually while covering all edge cases isn't a lot of fun, and at the same time, LLMs are pretty good at it. E.g. it's boring 'clanker work'.

210

Andre Weissflog @flohofwoe.bsky.social · 05/09/2026

...also while at it, finally a simple CONTRIBUTING md to set expectations for what types or PRs will be accepted vs rejected: github.com/floooh/sokol...

050

Andre Weissflog @flohofwoe.bsky.social · 05/09/2026

Added an AGENTS.md to the sokol repo to clarify acceptable LLM usage 😬 TL;DR: - LLM generation forbidden for sokol headers and documentation, but allowed for 'adjacent' code (tests, etc) - 'passive' LLM usage allowed and encouraged (bug scanning, code analysis...) github.com/floooh/sokol...

4220

Andre Weissflog @flohofwoe.bsky.social · 30/08/2026

Since Github CI is so slow that a C64 looks fast it will take a while until all downstream projects will be uptodate (samples webpage and language bindings)...

030

Andre Weissflog @flohofwoe.bsky.social · 30/08/2026

New breaking sokol_gfx.h update: Adds new way to update buffers and images with CPU data via new usage flag `.write_transient` and two new functions `sg_write_buffer_transient()` and `sg_write_image_transient()`. Replaces the old '.stream_update`. Details in changelog: github.com/floooh/sokol...

github.com

160