Demi Marie Obenour
@ska @dalias I suspect that in the future, high-performance code will use io_uring via a library that abstracts it away. I don’t think making it easy to use kernel APIs directly needs to be a goal.
It’s like OpenGL vs Vulkan: Vulkan makes it very hard to write a trivial application, but for large, complex applications it turns out to be a better fit.
Rich Felker
@alwayscurious @ska Good analogy because just like there I dont want to be using either. I just want read and write and poll, and I just want a "frame submit for swap-in on next refresh" primitive.
Demi Marie Obenour
@dalias @ska That makes sense if you don’t care about performance. I suspect that companies that pay developers of new kernel or graphics APIs generally do care.
Right now, I care a lot about software rendering, but as soon as the security problems with GPUs are solved I won’t care anymore.
Rich Felker
@alwayscurious @ska What I'm getting at is that they use this kind of "performance" machinery as excuses to expand resource consumption to make up for it.
When we got GPUs, suddenly every desktop had to have transparency and shadows and animated transitions all over the place, making it impossible to get by without one and impossible to use simple software stacks. Things have settled down a little but it's still bad.
io_uring is likely to do the same with gratuitous io all over the place. "Hey, it's fine if we do this because it's cheap now". 🤦
GNU social JP is a social network, courtesy of GNU social JP管理人. It runs on GNU social, version 2.0.2-dev, available under the GNU Affero General Public License.
All GNU social JP content and data are available under the Creative Commons Attribution 3.0 license.