[meta] Enable wayland testing in Gecko
Categories
(Firefox Build System :: Task Configuration, task, P1)
Tracking
(Not tracked)
People
(Reporter: ahal, Assigned: ahal)
References
(Depends on 11 open bugs, Blocks 2 open bugs)
Details
(Keywords: meta)
Dave House has done some initial investigative work on what it would take to stand up native Wayland testing. This bug will track the in-tree changes to docker images and setup scripts necessary to get these running.
Initially this will be for investigation / knowledge transfer, but may turn into a meta bug to track enabling all of the various test suites there.
Assignee | ||
Comment 1•1 year ago
|
||
Let's turn this into a tracking bug.
Assignee | ||
Comment 2•1 year ago
|
||
Quick status update, :aerickson has some standalone workers with Wayland enabled in a virtualbox vm. The plan will be to migrate these to a worker-manager managed pool eventually (he did standalone for now to get around some issues there). In the meantime, we can start getting patches going to support disabling tests running in wayland in the manifests, as well as start filing bugs for early issues we come across.
Assignee | ||
Comment 3•1 year ago
|
||
Also bug 1578640 already tracks the issues with Wayland, so let's keep this bug more focused on the test harness bits and CI tasks.
Updated•1 year ago
|
Comment 4•11 months ago
|
||
(In reply to Andrew Halberstadt [:ahal] from comment #2)
Quick status update, :aerickson has some standalone workers with Wayland enabled in a virtualbox vm.
So apparently this means that we can't use pernosco with these jobs. I tried and reported the failure on matrix in the pernosco channel:
khuey
Pernosco has its own infrastructure to repro because Mozilla's VMs don't support perf counters
and that infrastructure is docker based because everything taskcluster based is (was) docker based
Is there a timeline on moving these to be docker based?
Assignee | ||
Comment 5•11 months ago
|
||
There's some discussion on this in Slack. As of now we're blocked by a kernel bug, though it's unclear whether GCP or Canonical is the one that needs to fix it. More details in this Taskcluster issue and here's the Canonical ticket. It's possible we can find a workaround, though AIUI there's no clear path forward yet.
Comment 6•10 months ago
|
||
Andrew I've seen a test running on Wayland/headless. From browser perspective that config is the same as X11/headless as headless means completely without display connection. On headless target gtk_init() isn't called and both GdkIsWaylandDisplay()/GdkIsX11Display() returns false.
So if we run headless tasks on X11 and Wayland we waste resources as it's identical environment.
Assignee | ||
Comment 7•10 months ago
|
||
Thanks, I'll take a look. I don't think I've enabled any Wayland headless tasks.
Assignee | ||
Comment 8•10 months ago
|
||
I can confirm there are no Wayland headless tasks configured to run on trunk. It's possible you saw some on try which could have happened if someone used --full
which would allow them to schedule any task, not just the ones that are configured to run.
Description
•