Port to wlroots 0.18.x #2452
Port to wlroots 0.18.x #2452
Conversation
|
Compiled 0.9 with this PR - works ok so far. Seems like the new wlroots is better in terms of battery life at least for firefox-youtube. |
Thanks for testing! |
soreau
force-pushed
the
track-wlroots
branch
28 times, most recently
from
0e5e05e
to
c9c1798
Compare
because i am using your branch in the pkgbuild... not reproducible, this is how me and ammen99 did in the past, he just had the crash reports with no clue, it's up to you to fix it or not, but ammen99 could consider this a really issue before merging. |
There was a problem hiding this comment.
Looking at https://gitlab.freedesktop.org/wlroots/wlroots/-/merge_requests/4345, I believe we need to change all places which use xdg_surface.destroy events to xdg_popup/xdg_toplevel.destroy.
Otherwise LGTM, thanks for the patience!
src/core/output-layout.cpp
Outdated
| LOGD("Changed adaptive sync on output: ", handle->name, " to ", current_state.vrr); | ||
| } else | ||
| { | ||
| LOGE("Failed to change adaptive sync on output: ", handle->name); | ||
| wlr_output_rollback(handle); | ||
| // wlr_output_rollback(handle); |
There was a problem hiding this comment.
We should here revert &state to the previous vrr state I think, otherwise, &state gets permanently 'broken' afaict
There was a problem hiding this comment.
Yes, see this later commit.
There was a problem hiding this comment.
I looked at this and what I think is best, is to call wlr_output_state_finish() and wlr_output_state_init() after each wlr_output_state_commit(), so I pushed a commit to do that. This way, we should never have to worry about rolling back states or potentially stale state data.
src/core/output-layout.cpp
Outdated
| { | ||
| wlr_output_commit(handle); | ||
| wlr_output_commit_state(handle, &state); |
There was a problem hiding this comment.
Same as above, we ought to reset in the other case .. In all honestly previously we probably had to do a rollback as well, but we didn't (which might have been a compositor bug that nobody happened to complain about ...)
| d["type"] = wlr_input_device_type_to_string(device->get_wlr_handle()->type); | ||
| d["id"] = (intptr_t)device->get_wlr_handle(); | ||
| d["name"] = nonull(device->get_wlr_handle()->name); | ||
| d["type"] = wlr_input_device_type_to_string(device->get_wlr_handle()->type); |
There was a problem hiding this comment.
Is there no way to get vendor/product? Maybe by asking libinput directly for libinput devices, and just setting 'unknown' or 'nil' or whatever is appropriate for others. I would like to avoid any breaking changes to the IPC if possible.
There was a problem hiding this comment.
From the 0.18 release notes:
"..drop wlr_input_device.{vendor,product}, these aren't super useful without the bus type. Compositors can still obtain this information from the libinput device if they want to."
There was a problem hiding this comment.
I could use some help finding the call(s) to get the vendor/product id from libinput.
There was a problem hiding this comment.
I tried libinput_device_get_id_{vendor|product}(wlr_libinput_get_device_handle(device->get_wlr_handle())); but it always crashes.
There was a problem hiding this comment.
Devices are not necessarily libinput devices (especially in the wayland backend, you won't find a libinput device). So this needs to be in a check, and in the other cases we can set 'unknown' or -1 as a value.
I thought I caught them all but apparently xdg-toplevel is still using destroy events from the base xdg_surface (see last link). xdg_popups from master vs. track-wlroots branch. Similarly for xdg_toplevel, master and track-wlroots EDIT: Pushed a fix for this. |
This leads to a crash on wlroots 0.18, as wlroots handles this already.
xdg wayland surfaces were not committing because compositors are expected to send a configure when wlr_xdg_surface.initial_commit is set.
There is nothing needed to do here when using wlr_output_state.
Without edge, latest only has 1.22 until possibly Nov '24, when alpine image 3.21 is released.
If a subsurface is created and then becomes smaller without repositioning, the old larger size was never damaged, leaving artifacts on the output.
This fixes the build.
update to the latest git commit from the 'port to wlroots 0.18.x' upstream branch, cf WayfireWM/wayfire#2452
Gotchas:
wl_drm was dropped, so clients must update. A notable one is libva, which stopped working with wlroots 0.18 until upgrading libva.
wlr_input_inhibit was dropped and as of now, there is no replacement.
Possibilities:
linux-drm-syncobj-v1