Forum Federato

@DarcMoughty Yup! "Why is my app blurry? This never happened under X11!" Yeah, dude, that's because X11 never had frame scaling. Like, have you seen the attempts at doing scaling with X11?!?

@mcc @ariadne

@mcc @ariadne I think maybe you can make an argument there for sysv init and how everything is just shell scripts and a lot of people know bash.

But have you actually tried fixing X11? I have. It's not easier. There is absolutely nothing more approachable about X11 as a technology. There's just a bunch of people who are angry that XYZ unmaintained tool they found in the 90s doesn't work anymore.

Why is there an AI chatbot in my Firefox? Did Google pay them to put it there?

AI is going to doom us all...

I just did a google search for "opencl image atomic", hoping to dig up information on whether or not OpenCL even has image atomics and, if so, what they look like in OpenCL C. At the very top of the page, Google's AI confidently says:

OpenCL supports atomic operations on image objects, but with specific limitations:

and then goes on to list the usual GL/Vulkan image atomic restrictions. It also includes some use cases for image atomics.

Then I read the spec. OpenCL does not support image atomics at all.

Gotta love it when what starts off as an honest, if confused spec question turns out to be a deep and subtle synchronization bug in the Mesa vkQueueSubmit() implementation.

When you plug a Turing into your Ampere...

Probably not going to make the news but this week is actually pretty big for nouveau users. @airlied and I finally tracked down a fencing bug that's been plaguing nouveau users for a while.

Sometimes the context will randomly wedge, resulting in a timeout message in dmesg and your app suddenly freezing and getting killed. When this happens to your compositor or X server, you lose your session. We've seen a number of bug reports about things like this over the course of the last year or so but until this week weren't able to track it down.

Thanks to a bit of unrelated Blackwell debugging I was doing this week, I finally figured out how to reliably reproduce it and Dave was able to hunt it down in the nouveau code. The final fix was a group effort:

https://lists.freedesktop.org/archives/dri-devel/2025-August/522088.html

The fix has been merged to drm-misc-fixes and will be backported to the stable Linux kernel branches. Users should get that fix soon through their regular distro update cycle.

Hopefully, this will significantly improve the overall robustness of the nouveau stack and relieve some of the annoyance users have been experiencing.