Add basic wlr-output-management support

[joggee-fr]

I have tried to resume work from @dimkr in PR #213.

• Rebased on current HEAD
• Implements output manager test (wlr-randr --dryrun)
• Add output layout change monitor to ensure views are well positioned according to new layout

For the moment, I have only played with Wayland as backend.
I am still unsure about current_mode test, see discussion.
Furthermore, trying to disable an output returns an error. I don't know if it specific to Wayland backend. Here are the logs from wlroots (0.16.2):

00:00:04.247 [types/output/output.c:656] Tried to modeset a disabled output
00:00:04.247 [types/output/output.c:734] Basic output test failed for WL-2

[joggee-fr]

When disabling an output, we must avoid to set other parameters (scale, mode, etc.) as it will cause the error described in my first comment. Now, with a Wayland backend, we still observe an error but a kind of expected one:
[backend/wayland/output.c:254] Unsupported output state fields: 0x8.

I have also tried with DRM backend. Disabling output is OK but re-enabling next is not yet very stable.

[joggee-fr]

I have also tried with DRM backend. Disabling output is OK but re-enabling next is not yet very stable.

Simply because I forgot Cage terminates when no more output is available.

[emersion]

with a Wayland backend, we still observe an error but a kind of expected one

Right. With wlroots 0.16, only the DRM backend supports disabling outputs. With wlroots 0.17, all backends support disabling outputs.

[joggee-fr]

I have added commits (to be squashed if OK) to answer @emersion comments.
By the way, I have also modified the access to previous or next item in output list causing segfaults when using -m last option.

[emersion]

Yeah, the changes you made LGTM.

[emersion]

I'm still not super sure about "output: do not terminate when last output is destroyed". It sounds like this is useful mainly when running nested under the X11 and Wayland backends, to stop running when the last cage window is closed.

[joggee-fr]

Squashed some commits and removed extra parentheses.

[joggee-fr]

I'm still not super sure about "output: do not terminate when last output is destroyed". It sounds like this is useful mainly when running nested under the X11 and Wayland backends, to stop running when the last cage window is closed.

I don't have done any tests with X11 backend for now.
When using Wayland backend, how can we destroy output? I understand with next version of wlroots, we will be able to disable outputs, but how about output destruction?

Below a short description of the issue I observed with terminate when last output available is destroyed:

• Cage started with a monitor plugged in HDMI output. DRM backend is selected, GPU is integrated Intel Iris with i915 kernel driver ;
• HDMI-A-1 output is disabled using wlr-randr --output HDMI-A-1 --off ;
• Monitor lost signal and automatically enters a standby mode ;
• HDMI output is re-enabled using wlr-randr --output HDMI-A-1 --on ;
• I think it is due to monitor exiting "slowly" from standby but Cage receives a destroy event just before handling again a new HDMI output. Consequently, things are broken if terminating here.

[emersion]

I don't have done any tests with X11 backend for now.

It's fine, it should behave exactly just like the Wayland backend.

When using Wayland backend, how can we destroy output? I understand with next version of wlroots, we will be able to disable outputs, but how about output destruction?

Closing windows should destroy the output, regardless of wlroots version.

Below a short description of the issue I observed with terminate when last output available is destroyed

Yeah. Maybe it would make sense to special-case Wayland and X11.

[joggee-fr]

Closing windows should destroy the output, regardless of wlroots version.

As I understand it (and according to tests), in such a case, whatever the selected backend, Cage receives sigchld_handler() as the started/forked child process has been closed and terminates. Last available output patch does not modify this behavior.

[emersion]

No, I'm talking about closing the wlroots Wayland/X11 backend window. The nested child process is not involved here.

I don't think Cage ever sends a xdg_toplevel.close event to the nested Wayland client.

[emersion]

Come to think of it, maybe it would be better to not force-terminate when the wlroots Wayland/X11 backend window is closed by the user. Instead we could just relay the close request to the nested Wayland client. That way "do you want to save your work before closing?" kind of dialogs would work properly. But this requires some wlroots changes: https://gitlab.freedesktop.org/wlroots/wlroots/-/issues/3589

[joggee-fr]

No, I'm talking about closing the wlroots Wayland/X11 backend window. The nested child process is not involved here.

Ok, now I get it. I have just given it a try with Wayland backend and indeed, Cage is now running without any output available. Kind of weird 😅 .

[emersion]

If you want to get this PR merged, maybe we can leave out that commit for now and fix it in a separate PR?

[joggee-fr]

If you want to get this PR merged, maybe we can leave out that commit for now and fix it in a separate PR?

Just dropped the commit from this PR.
Who get rights to merge it (if OK)? @Hjdskes does not seem to be (publicly) active here since several months.

[emersion]

Thanks!