Hi Cedric,
reading about your experience reports and plans is always a joy! Thank you for participating.
But -- I still met one of my goals for 2024 though, scoring a solid "yes" on the question "can I use h-on-Genode with Falkon on all my computers". I checked that again with the recent package upgrades prompted by an ABI change: I had to update all related computers in this house, and verified yet again that I can boot up any of them with my USB thumbdrive in seconds, click 'Download' on the relevant package(s), and run Falkon once the download is complete. Cannot ask for more as far as that (specific/narrow) functional aspect is concerned.
Reading that makes me very happy. It's gratifying to see our work paying off and being recognized.
The aspect that affects me most is the mixer transition, which occured many months ago already but which I'm only now near to tackling (hopefully in early January).
Since we introduced the new audio interfaces about one year ago, we gathered quite a bit of practical experience with using the new mixer, which revealed one point I wish to refine: To make the mixer robust against overflow/underflow issues, it adjusts the sample rate per client dynamically according to the client's behavior. This works well for clients that work periodically enough. But we figured that periodic behavior is not a given for all audio clients. Insufficiently periodic clients can exhibit audible droning artifacts. This is not always the client's fault but may be caused by sporadic CPU load induced by high-priority components. In cases like this, it would make sense to move the time-window allocation from the mixer to the client side and thereby weaken the requirement of periodicity.
Over the course of 2025, I'd like to try this out. Your feedback and validation would be very valuable for me.
Optimizations in the FS stack are something I keep an eye on. I'll never allow myself to say "pressing!" as I know it's a huge endeavour, and one should not rush to optimization before everything is clean and refactored. That said, I think I will be able to help a bit with those in terms of Q.A., seeing as my VFS use-cases are out of the beaten path a little and thus extend the coverage and verification, and given that my project's test suite tends to "push" Genode's vfs a little, in terms of coverage (corner cases) ad in terms of being demanding performance-wise, so as to reveal things of interest in that area.
Thank you for the offer. Your past feedback about the VFS was super helpful. So I'm happy about your willingness to help in the Q/A department for the changes to come. :-)
Cheers Norman