Posts by qyliss@chaos.social
 (DIR) Post #ASVgcVSStZ7h8q6Fxg by qyliss@chaos.social
       2023-02-09T13:07:29Z
       
       1 likes, 0 repeats
       
       i think it is finally time to learn how to write a nixos test
       
 (DIR) Post #ASVgcX4msZN69z2NKi by qyliss@chaos.social
       2023-02-09T17:01:35Z
       
       0 likes, 0 repeats
       
       Going well so far…
       
 (DIR) Post #ATYEuhWTQ3f4gWPCt6 by qyliss@chaos.social
       2023-03-12T19:46:26Z
       
       0 likes, 0 repeats
       
       I spent this weekend writing a test for the crosvm functionality Spectrum relies on (Wayland over vhost-user-gpu).  crosvm upstream moves very fast, and things change/break very frequently, so it's really hard to stay up to date, and hopefully this will help.  I'm especially looking forward to much easier bisects now that it won't be a manual process.#spectrumos
       
 (DIR) Post #ATYEuiK6RXMPARcqFE by qyliss@chaos.social
       2023-03-12T19:58:41Z
       
       0 likes, 0 repeats
       
       Testing this sounds easy, but isn't.  The idea is to run a Wayland compositor, and crosvm, with a VM image that opens a Wayland window over virtio-gpu, and then check that that window is displayed in the compositor.But how do you find out when the compositor shows a new window?  Sway has config hooks for it, but wlroots doesn't support software rendering in a crosvm-friendly way, which would mean the test couldn't be run in CI.#spectrumos
       
 (DIR) Post #ATYEuiucFmZEzgi09o by qyliss@chaos.social
       2023-03-12T20:01:10Z
       
       0 likes, 0 repeats
       
       So the solution I came up with was to use Weston, with a custom plugin that listens for the expected window appearing, and writes into a fifo when it does.  The test runner can then just wait to receive data over the fifo, with a timeout in case something in the setup is broken and the window never appears.#SpectrumOS
       
 (DIR) Post #ATYEujQsJqN6bjnlRI by qyliss@chaos.social
       2023-03-12T20:27:27Z
       
       0 likes, 0 repeats
       
       Other things I did in this, the most productive weekend ever:• Fix logs not being stored in the Spectrum binary cache.• Sent v2 of a kernel patch to fix userspace notifications for loop device partitions.• Some new/improved documentation.:D#SpectrumOS
       
 (DIR) Post #ATYEujyuHJasJHiwU4 by qyliss@chaos.social
       2023-03-12T20:53:14Z
       
       2 likes, 0 repeats
       
       Here's the test I'm so proud of.  There's a #NixOS test that runs #Weston and #crosvm and listens on the fifo, and the custom Weston plugin to write into the fifo when the window appears.One really cool thing about the NixOS test runner is that it can output screenshots, so I can also confirm visually that the test worked.#SpectrumOShttps://spectrum-os.org/git/spectrum/tree/release/checks/wayland?h=wayland
       
 (DIR) Post #AU98tIOk77lmND4RU0 by qyliss@chaos.social
       2023-03-30T16:15:01Z
       
       2 likes, 0 repeats
       
       got my first functional change to the kernel accepted :)(I've done typo fixes and stuff before)
       
 (DIR) Post #AUibRYIyClhKWocO4u by qyliss@chaos.social
       2023-04-16T18:16:39Z
       
       0 likes, 0 repeats
       
       do you use appimages? what applications? is appimage your preferred choice, or more of an "if there's no other option" sort of thing? :BoostOK:
       
 (DIR) Post #AUibRZyTzuUxhr33QG by qyliss@chaos.social
       2023-04-16T18:33:52Z
       
       0 likes, 0 repeats
       
       I've been looking into what it would take to add easy AppImage support to #SpectrumOS.  AppImage is an attractive thing to integrate first because there's no shared file store to worry about.  But it seems like most popular applications are not distributing AppImages (any more), and very few of them include the libraries they need for Wayland, even when the application supports Wayland.
       
 (DIR) Post #AUibRbXcAmC8Z6Ucoy by qyliss@chaos.social
       2023-04-16T18:37:08Z
       
       0 likes, 0 repeats
       
       I was also surprised to learn that AppImages don't contain all of their dependencies.  They contain some of their dependencies, but it's up to the developer to guess which ones they should include based on what their users are likely to have installed.  So unlike with e.g. Flatpak it's impossible to guarantee compatibility.
       
 (DIR) Post #AUyRjyfL2p0KU7aRCS by qyliss@chaos.social
       2023-04-24T10:17:27Z
       
       2 likes, 0 repeats
       
       Linux 6.3 has two commits from me in it and I'm feeling proud. :3One fixes a small amount of debug info being included in kernel images that were supposed to be stripped on x86_64 and RISC-V (which was especially problematic for Nixpkgs because it caused an output cycle).The other fixes userspace not being notified about the new partitions in some cases when a loop device was set up.
       
 (DIR) Post #AVAzI3NXemRJKddcQa by qyliss@chaos.social
       2023-04-30T10:36:22Z
       
       1 likes, 0 repeats
       
       today is a day to chill and play supertuxkart
       
 (DIR) Post #AVaHbxLSwGhH6VauYK by qyliss@chaos.social
       2023-05-12T09:38:58Z
       
       0 likes, 0 repeats
       
       @erincandescent @galibert @marcan AIUI, the protocol also doesn't _technically_ require the two to be colocated, but we haven't really seen a system where they aren't. Closest is River (a compositor), which can outsource some WM decisions to another process using a custom Wayland protocol.
       
 (DIR) Post #AW55TvacHf1w2ZHyds by qyliss@chaos.social
       2023-05-27T11:38:48Z
       
       1 likes, 0 repeats
       
       IstillthinkwritingpackageinputslikethismakesitveryhardtoreadbutIamlosingmywilltokeepresisting.
       
 (DIR) Post #AWO3mlz3F0POjTj6Zc by qyliss@chaos.social
       2023-06-05T14:25:57Z
       
       1 likes, 0 repeats
       
       Working on NixOS modules is burnout fuel for me, because we don't have a good way to manage state migrations.  Modules either accumulate ad-hoc shell code to migrate old state from arbitrarily old Nixpkgs versions, or problems that would require state migrations just don't get fixed.It's a huge difference to pkgs/, where there's zero state to worry about.  We can aggressively refactor, and mistakes that weren't caught in review don't matter — they can just be fixed later.
       
 (DIR) Post #AblLUXq2SU7lAIXJAm by qyliss@chaos.social
       2023-11-13T11:41:51Z
       
       0 likes, 1 repeats
       
       #NLnet / @NGIZero grantees in Germany — how do you handle VAT?  They're pretty clear they don't expect their donations to be VATable, but I've been having trouble convincing a Steuerberater of that, so if I can't do that I'm going to be stuck paying the VAT myself.  (I know they wouldn't be in scope for VAT in the UK, because the UK's VAT rules are in English and I can read them.) #NextGenerationInternet
       
 (DIR) Post #Am01qWJYIXn3aLgBGq by qyliss@chaos.social
       2024-09-14T16:05:52Z
       
       0 likes, 0 repeats
       
       Coming soon to #SpectrumOS — double click an AppImage, and it runs in a temporary VM, with no other setup required.Just need to tidy up the commits a bit…
       
 (DIR) Post #Am01qXC91ZSWJfDmMa by qyliss@chaos.social
       2024-09-14T16:19:34Z
       
       1 likes, 0 repeats
       
       Why AppImage — it's not a very relevant format, right?I chose AppImage because it's very simple — I don't have to figure out what parts of an OSTree store like I'd have to with Flatpak.  As a result of how simple it is, I had to write almost no AppImage-specific code — the vast majority of the work was generic, like implementing support for creating VMs on the fly rather than requiring them to all be defined at boot.  So this is also step 1 towards supporting other app formats!
       
 (DIR) Post #AyOkhwqFUOcdDo24eG by qyliss@chaos.social
       2025-09-20T08:46:39Z
       
       1 likes, 0 repeats
       
       I don't think this is how COSMIC is supposed to look but wow I love it