Conversation
|
I removed the unnecessary kernel and LXC patches and addressed the remaining nitpicks. |
|
Maybe of interest to you, when I was working on that (I abandoned all that because it didn't work in any usable manner for my personal use-case/opinion) I had built an anbox image using robotnix
YMMV, not sure it all works still. |
There was a problem hiding this comment.
This means it is still racy though...
There was a problem hiding this comment.
I think the purpose of this patcheset is to get anbox to a more usable state. I wouldn't spend more energy on doing proper things as upstream is dead.
This patch is used actively by postmarketOS, which in turn *just* uses Alpine Linux packages for anbox.
This helps on slower devices, at the cost of making the command "resolve" less quickly. It replaces commands "failing because things weren'd ready", but things actually still succeeding by commands *possibly* failing in 10× more time.
The `anbox-application-manager` helper, which is actually broken as it's not installed as +x, starts "su" by default, which might not be desirable, depending on the end-user's wishes.
It now *strictly* is a shortcut to launch the application manager.
This fixes two issues - store path is hardcoded in desktop files - `.anbox-wrapped` is used in desktop files
SDL defaults to reading the executable from the running app - https://github.com/libsdl-org/SDL/blob/64724db0a1ffac7c3332df66a942d3ba27b2dc0b/src/video/x11/SDL_x11video.c#L69-L71 Which in turn gives this from xprop: ``` WM_CLASS(STRING) = ".anbox-wrapped", ".anbox-wrapped" ``` Meaning that any special casing a window manager does will be broken. This will also fallback correctly on Wayland: - https://github.com/libsdl-org/SDL/blob/d956636c85aafc055d37153d52370e9b1c2c5929/src/video/wayland/SDL_waylandvideo.c#L93-L99
Currently the windows don't provide icons. It's rather inconvenient for many reasons. This patch forwards the icons from the Android application to the window. Patch origin: https://fjordtek.com/git/Fincer/anbox-install/commit/89226287860cf8ced3d41868c80cfa216b5237af Rebased on top of the current tip of anbox master See also: anbox/anbox#1514
As noted in NixOS#102341 this is not actually running as a forked process. It only tells the process that it is running "as a daemon, so shut the warning up". See `daemon_` here - https://github.com/anbox/anbox/blob/9de4e87cdd05135e1c71e6eadb68bf82719cebdf/src/anbox/cmds/container_manager.cpp#L38-L79 It is **strictly** used to hide that message. Co-authored-by: Matt Votava <mvnetbiz@gmail.com>
|
Doesn't download at least for aarch64-linux, apparently? (just FYI, discovered by accident) |
|
Yeah, I wanted to mention this as well. |
|
That would have been pretty visible if that failure would be red and not grey which github treats as ignoreable when collapsing the CI checks. |
|
The problem is just that build.anbox.io is dead too, only the x86_64 test works because the image has been saved by cache.nixos.org. There's not much I can do besides replacing them with the postmarket OS version: the ARM images aren't even on the internet archive. Anyway, both x86 and ARM were broken before this PR, so it's not a big deal. |
Description of changes
Continuation of PR #125600, which was abandoned.
This brings anbox into a somewhat working state, though it will be eventually removed.
That is because upstream development is dead and newer kernels removed a module needed by anbox.
Things done
sandbox = trueset innix.conf? (See Nix manual)anbox.tests