@karolherbst@chaos.social

☃️karolherbst☃️ @karolherbst@chaos.social · Jul 30, 2026

Guess #rusticl on #panfrost is also #OpenCL 3.0 conformant as of yesterday: https://www.khronos.org/conformance/adopters/conformant-products/opencl#submission_474

OpenCL 3.1 is in the works

26

2

6

1

☃️karolherbst☃️ @karolherbst@chaos.social · Jul 30, 2026

who decided to send harassment over to freedesktop again...

2

0

☃️karolherbst☃️ @karolherbst@chaos.social · Jul 30, 2026

me in metroidvanias:

80% of my life I lose in the first 20% of the boss fight

4

2

1

0

☃️karolherbst☃️ @karolherbst@chaos.social · Jul 27, 2026

Soo.. guess I can be more open about it: If anybody wants to hire me to do FOSS stuff in Mesa, please let me know. I haven't made any final decision yet, but it's always good to have more options.

28

0

57

0

☃️karolherbst☃️ @karolherbst@chaos.social · Jul 25, 2026

I wished most "let's stay apolitical" discussions could be abruptly ended by simply stating that "politics" means "affairs of the community" or "what concerns the community" (strictly speaking it's "city", but... you know) And if your project, any project, involves more than a handful of people, then it's a political project, because it has grown a community around it, and so because where are people there is politics...

17

0

4

0

☃️karolherbst☃️ @karolherbst@chaos.social · Jul 17, 2026

Replying to @mupuf@social.treehouse.systems

@mupuf@social.treehouse.systems though my point, or at least that I tried to make is, that it's a bit silly to argue that a persons opinion has automatically more weight due to their position like there are sometimes incompetent people, or sometimes people with skills make bad predictions of things Or sometimes a random person isn't heard on a subject they have a more correct opinion on than somebody in a position of power and we have to take that in mind when discussing what a famous dude said or didn't say

0

2

0

☃️karolherbst☃️ @karolherbst@chaos.social · Jul 17, 2026

Replying to @eniko@mastodon.gamedev.place

climate change, ableism, covid, negative Hover or focus to reveal Sensitive

@eniko@mastodon.gamedev.place the heat wave in June here caused a new record on excess deaths in a single day. Yes, even more than during COVID on any single day 😢 We don't have ACs rolled out across the entire country, soo.... And the conservative government currently instead of using the money that was supposed to be used for climate change projects (e.g. making ACs affordable to those who need it) they want to build more natural gas power plants instead with it 😭

2

0

1

0

☃️karolherbst☃️ @karolherbst@chaos.social · Jul 17, 2026

Replying to @pac85@social.treehouse.systems

@pac85@social.treehouse.systems @mupuf@social.treehouse.systems

That's not true especially in the open source field where lots of people are hired because do the work they do voluntarily.

That's luck.

What are poor people doing that would love to spend more time doing FOSS, but can't because their parents don't have the money and they need to do shitty jobs they don't want to do?

Is the lack of skill preventing them? Technically yes, because they don't have the time to build up those skills, but actually it's just bad luck.

13

1

3

0

☃️karolherbst☃️ @karolherbst@chaos.social · Jul 17, 2026

Replying to @mupuf@social.treehouse.systems

@mupuf@social.treehouse.systems Yeah, I'm not arguing that nobody has skills or there isn't merit in what one is doing However, why isn't somebody better suited in your position? There always is somebody like that somewhere, so why not them? Why do you have the interests you have? Did you make a conscious decision on where you want to develop skills? Sure people work hard to achieve something, and I don't want to take that away, but that doesn't mean it's the reason for where we are

3

1

0

☃️karolherbst☃️ @karolherbst@chaos.social · Jul 16, 2026

My most extreme believe is that any person in a position of power is there not due their skills but due to luck alone "skills", "intelligence", etc.. are constructs to defend the status quo, not to enable the more unfortunate people I have no problem with acknowledging that I'm where I'm today because of luck, not because of my "skills" or "intelligence" or whatever else fake construct we create Just because a person has power doesn't mean what they say has any weight or merit

11

1

2

0

☃️karolherbst☃️ @karolherbst@chaos.social · Jul 13, 2026

Replying to @mld@social.mldchan.dev

@mld@social.mldchan.dev for extra swap, there were those CD auto changer devices that could move to a different CD swap device when needed

4

0

☃️karolherbst☃️ @karolherbst@chaos.social · Jun 25, 2026

Replying to @ShadowJonathan@tech.lgbt

@ShadowJonathan@tech.lgbt I recommend heading out 30 minutes before the store closes lol

1

0

☃️karolherbst☃️ @karolherbst@chaos.social · Jun 16, 2026

Replying to @thephd@pony.social

@thephd@pony.social can recommend. Which is funny, because I don't think most people who get into GPU driver development do it because they want to write a compiler, but once you are there, you basically work on a compiler full time 🙃

3

0

☃️karolherbst☃️ @karolherbst@chaos.social · May 28, 2026

Replying to @airlied@floss.social

@airlied@floss.social 100.000? Rookie numbers, I've seen 10.000.000

1

0

☃️karolherbst☃️ @karolherbst@chaos.social · May 24, 2026

What Linux drama did I miss this time....

3

6

0

☃️karolherbst☃️ @karolherbst@chaos.social · May 24, 2026

I do kinda wonder how this years push for AI use for accessibility is perceived by people having to rely on related technology, especially around GAAD.

I do have my own opinions, but I'm also interested how others, who rely on the tech, see it.

0

2

0

☃️karolherbst☃️ @karolherbst@chaos.social · May 22, 2026

self OH: though, compiling shaders inside the kernel build system would be kinda wild

2

0

☃️karolherbst☃️ @karolherbst@chaos.social · May 20, 2026

RIP google

3

1

0

☃️karolherbst☃️ @karolherbst@chaos.social · May 15, 2026

Replying to @xerz@masfloss.net

@xerz@soc.masfloss.net @ShadowJonathan@tech.lgbt @cadey@pony.social so in case those are ai-found bugs, the reasoning for publishing those immediately is, that anybody else with access to an AI could exploit them, so everybody is probably on the safer side to assume that everything found with those tools are zero-days now.

0

☃️karolherbst☃️ @karolherbst@chaos.social · May 15, 2026

Replying to @cadey@pony.social

@cadey@pony.social happy daily reboot your system for a new kernel update day :)

0

☃️karolherbst☃️ @karolherbst@chaos.social · May 14, 2026

Replying to @cadey@pony.social

@cadey@pony.social weird flex to admit they haven't in the past...

0

☃️karolherbst☃️ @karolherbst@chaos.social · May 13, 2026

Replying to @sovtechfund@mastodon.social

@sovtechfund@mastodon.social @kde@floss.social We (as in Freedesktop) should probably also apply for funding at some point. I was thinking about some cross Desktop accessibility work could fit the bill. Or trying to get off gitlab/rework our CI infrastructure. But.. I don't think anybody of us has any experience with applying for grants in general, so it's a bit scary.

6

1

0

☃️karolherbst☃️ @karolherbst@chaos.social · May 11, 2026

Replying to @ash@gts.spooky.academy

@ash@gts.spooky.academy @aks@scalie.zone oh yeah, this looks close enough. Good to know this is possible in go!

0

☃️karolherbst☃️ @karolherbst@chaos.social · May 11, 2026

Replying to @aks@scalie.zone

@aks@scalie.zone I don't really know what's the cool thing in go there, and my experience is mostly Java based, but there are a couple of libraries that map the same concept to other languages, like: #[get("/some/path/{param}")] fn my_handler(param: u32) -> ResponseType { ... } and then have a central place where you register all of those. https://actix.rs/ has an example of this concept on their webpage, just to give you an idea.

3

0

☃️karolherbst☃️ @karolherbst@chaos.social · May 11, 2026

Replying to @aks@scalie.zone

@aks@scalie.zone Sooooo I was doing like actual web development and I have.... _opinions_ on your approach 🙃 But generally you kinda want to use something more higher level that deals with the URL handling in a less noisy way. Like URL handlers that map parameters in the URL to function arguments and then you don't have to deal with any of the "what page is it trying to load" part of the misery that web development is.

1

5

0

☃️karolherbst☃️ @karolherbst@chaos.social · May 07, 2026

Replying to @portaloffreedom@social.linux.pizza

@portaloffreedom@social.linux.pizza I have to wake up at like 7:30 or so 😭

0

1

0

☃️karolherbst☃️ @karolherbst@chaos.social · May 06, 2026

Replying to @ShadowJonathan@tech.lgbt

@ShadowJonathan@tech.lgbt 🫡

1

0

☃️karolherbst☃️ @karolherbst@chaos.social · May 05, 2026

Replying to @aks@scalie.zone

@aks@scalie.zone most cursed user: "your promised me LTS and now you don't even want to backport this fix" on a report against an upstream EoL branch. I think we need to talk more about what kind of emotional abuse this enables.

1

0

1

0

☃️karolherbst☃️ @karolherbst@chaos.social · May 03, 2026

Replying to @blinry@chaos.social

@blinry I think when looking at what other projects are deciding here that it's important to keep in mind the size of those projects. A small team will have an easy time to decide if they want to ban AI or not. But if you have a big community of like 100+ active developers it's going to become really difficult to navigate this because you don't want to risk pushing valuable contributors away with the decision. And you won't be able to convince everybody to ban AI.

3

0

☃️karolherbst☃️ @karolherbst@chaos.social · May 03, 2026

Replying to @thesamesam@social.treehouse.systems

@thesamesam @mgorny @fmoessbauer seems like I looked at the wrong date in the end.

1

0

☃️karolherbst☃️ @karolherbst@chaos.social · May 03, 2026

Replying to @thesamesam@social.treehouse.systems

@thesamesam @mgorny @fmoessbauer "CVE-2026-31431" or was there another one I'm not aware of?

0

3

0

☃️karolherbst☃️ @karolherbst@chaos.social · May 03, 2026

Replying to @mgorny@social.treehouse.systems

@mgorny @fmoessbauer the CVE was assigned after the fact. And the backports happened as regularly fixes. What you are asking for here is that you request from the upstream kernel project to backport _all_ the fixes _all_ the time. The CVE entry was created after all that become public. But also, there is no difference between security fixes and "normal" fixes besides there is a CVE entry for one and not the other 🙃 those "normal" fixes are the unfixed CVEs Greg talks about here.

0

1

0

☃️karolherbst☃️ @karolherbst@chaos.social · Apr 28, 2026

Replying to @austriancoder@fosstodon.org

@austriancoder@fosstodon.org let me guess, something with pack?

0

1

0

☃️karolherbst☃️ @karolherbst@chaos.social · Apr 19, 2026

Replying to @karolherbst@chaos.social

@thomholwerda commercial software was also awful 20 years ago, and I don't agree with the narrative of things becoming worse in _that_ sense. You just see what was behind doors in the public, because Open Source is a hot thing now, which it wasn't 20 years ago. And I'm in this industry for almost that long already.

0

☃️karolherbst☃️ @karolherbst@chaos.social · Apr 19, 2026

Replying to @thomholwerda@exquisite.social

@thomholwerda Well initially only "nerds" got into computers, because it was kinda niche and also niche in a different sense. And then it became this well paying job and a lot of people are in for the money. And I don't think this is a bad thing in itself, because many people have different things they take pride in. E.g. providing for their family. I doubt it's any different in other professions, just more transparent in FOSS. Some are in for the job, others for the money.

0

1

0

☃️karolherbst☃️ @karolherbst@chaos.social · Apr 18, 2026

Replying to @ekuber@hachyderm.io

@ekuber I kinda hoped the baseline understanding of art is around that it doesn't start at "this is a masterpiece that changes everybody's life" or "what this brilliant person is doing is art and not what those peasants are doing"... Which I think is the more fundamental disagreement, maybe even misunderstanding. Though I do know that most of programmers aren't doing programming for artistic or self-expression reasons.

0

☃️karolherbst☃️ @karolherbst@chaos.social · Apr 15, 2026

Replying to @jcm@wafrn.jcm.re

@jcm I think then you also should provide a validation mechanism to verify the usage is actually correct in whatever form. The way Vulkan does it is quite good e.g. and then if you ignore validation errors it's on you.

1

0

☃️karolherbst☃️ @karolherbst@chaos.social · Apr 15, 2026

Replying to @etchedpixels@mastodon.social

@etchedpixels@mastodon.social but changing APIs in a breaking way also introduces more work on the consumer side. I don't know what's worse in overall maintenance cost, but e.g. if the Linux kernel would go "we break things and you gotta deal with the outfall" I'm sure we'd have to need more developers time to deal with this in total, even if it would be "cheaper" on the kernel side. And I'm not saying "maintain the old path forever", if you remove it in 6 months, so be it.

1

0

☃️karolherbst☃️ @karolherbst@chaos.social · Apr 15, 2026

Replying to @pac85@social.treehouse.systems

@pac85@social.treehouse.systems well we don't maintain our own APIs...so not much we can do besides following the spec.

0

1

0

☃️karolherbst☃️ @karolherbst@chaos.social · Apr 15, 2026

Replying to @etchedpixels@mastodon.social

@etchedpixels@mastodon.social yeah, and I'm just saying we need to be responsible. If you break behavior, make a new API, call it v2 and drop the old one after X years. Not only if parameter change, but for behavior. There are ways of doing this properly.

1

3

0

☃️karolherbst☃️ @karolherbst@chaos.social · Apr 14, 2026

Hot take: more projects need a regression policy like the linux kernel.

It doesn't matter if what a change is doing is "technically" correct, if it breaks a bunch of other stuff, it's that projects responsibility that it doesn't.

No "this was technically illegal" or "you just misused the API", "you relied on UB behavior" nonsense.

If you want to push forward a breaking change, makes sure the users had reasonable amount of time to adjust before pushing breaking changes forward.

20

29

6

0

☃️karolherbst☃️ @karolherbst@chaos.social · Apr 13, 2026

Replying to @pac85@social.treehouse.systems

@pac85 they keep some gfx. It's pretty rare for nvidia to entirely remove GFX, because it's useful for certain types of in-memory data transfers. But yeah... anyway, I think all PCIe cards besides the Hopper ones have all gfx units still enabled. For quadro I'm 100% certain, for Tesla/GRID 90%.

0

1

0

☃️karolherbst☃️ @karolherbst@chaos.social · Apr 13, 2026

Replying to @karolherbst@chaos.social

@pac85 ohh seems like Blackwell DCs as well.. but those aren't PCIe cards anyway..

0

3

0

☃️karolherbst☃️ @karolherbst@chaos.social · Apr 13, 2026

Replying to @pac85@social.treehouse.systems

@pac85 Well the only GPU where I'm aware they've done it recently are the Hopper DC cards. But the ISA normally stays the same.

0

4

0

☃️karolherbst☃️ @karolherbst@chaos.social · Apr 13, 2026

Replying to @cadey@pony.social

@cadey This kinda depends highly on the specific models though. There are many DC cards that have full 3D units, because they are also useful for e.g. desktop virtualization. And e.g. high-end Nvidia Quadro cards are really just normal GPUs with much EEC VRAM you could also use in DCs due to other features they also come with.

0

6

0

☃️karolherbst☃️ @karolherbst@chaos.social · Apr 09, 2026

Replying to @decathorpe@mastodon.social

@decathorpe I got into programming when MB still meant MiB

3

0

☃️karolherbst☃️ @karolherbst@chaos.social · Apr 09, 2026

Replying to @decathorpe@mastodon.social

@decathorpe @hughsie have you seen the LZ4 comment? I saw it after my reply and was like: ah yes, they don't run into the same issue than my suggestion would: https://github.com/lz4/lz4/blob/9da37b2eebf082bfab6e57c49be71cc41119a40d/programs/bench.c#L74C9-L80

1

2

0

☃️karolherbst☃️ @karolherbst@chaos.social · Apr 09, 2026

Replying to @hughsie@mastodon.social

@hughsie #include #define MB << 20 int main() { printf("0x%x\n", 5 MB); }

6

5

4

0

☃️karolherbst☃️ @karolherbst@chaos.social · Apr 02, 2026

Replying to @ShadowJonathan@tech.lgbt

@ShadowJonathan @MiaWinter it's funny, how google translates this sentence to English in a very... bad way? Like I'm no dutch expert and I might be wrong, but it feels like that "destroy what can be destroyed" has a different meaning than "maak kapot wat kapot kan", but maybe that's just due to how "das kann" is used in German 🙃 so I _think_ your sentence means "destroy what we don't need it anymore" rather than "destroy what we are able to" and I'm wondering how much of that is correct.

3

0

☃️karolherbst☃️ @karolherbst@chaos.social · Apr 02, 2026

Replying to @ShadowJonathan@tech.lgbt

@ShadowJonathan @MiaWinter yes

3

0

☃️karolherbst☃️

Posts

Remote instance