Repository navigation
Conversation
The two Windows legs were a placeholder under TRACE_CI_RUNNERS=self-hosted, so nothing exercised the WindowsDX catalog smoke, the FNA video smoke or the Windows render parity while the fleet handled the rest. No Windows machine is part of the local fleet, and this repository is public, where standard GitHub-hosted runners (Windows included) do not consume billed minutes, so the legs can run for real there. The matrix now always carries windows-latest; the flag still selects the self-hosted labels for the Linux and macOS legs, and windows-latest maps to itself either way. Job names are unchanged, so the required checks still resolve.
zignd
force-pushed
the
ci/windows-always-hosted
branch
from
October 24, 2026 21:40
4f79430 to
ff037fe
Compare
The Linux legs started failing with
XCB failed to connect to the X server due to error:1
ERROR at ./vulkaninfo/./vulkaninfo.h:737: AppCreateXcbSurface failed to establish connection
on the very first X program of the job. xvfb-run -a hands the display number to the payload as soon
as Xvfb creates its socket, and Xvfb 2:21.1.12-1ubuntu1.9 still needs a moment before it listens, so
an immediate client gets ECONNREFUSED. A fresh container on the server reproduces it: 4 of 5 plain
xvfb-run runs failed, 0 of 5 with a readiness wait.
scripts/xvfb-run.sh wraps xvfb-run with that wait (xdpyinfo, from x11-utils) and every call site now
uses it. This is an environment change (the xvfb package is installed at job time), not a code one:
it would have hit main on the next Linux job regardless of this branch.
zignd
changed the title
The Xvfb readiness wrapper documents the probe it uses, and cspell flags the name.
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Two changes, both needed for the checks to be meaningful.
Windows legs run for real. Under
TRACE_CI_RUNNERS=self-hostedthey were a placeholder that only reported they did not run, so the WindowsDX catalog smoke, the FNA video smoke and the Windows render parity never executed. No Windows machine is part of the local fleet, and Forma is public, where standard GitHub-hosted runners (Windows included) are not billed, so they now always run onwindows-latest. The flag still selects the self-hosted labels for Linux and macOS, and the job names are unchanged so the required checks keep resolving.Xvfb readiness. The first green run showed the Linux leg failing with
XCB failed to connect to the X server due to error:1from the first X program of the job.xvfb-run -ahands the display number to the payload as soon as Xvfb creates its socket, and Xvfb2:21.1.12-1ubuntu1.9still needs a moment before it listens. In a fresh container on the server, 4 of 5 plainxvfb-runruns failed and 0 of 5 failed behindscripts/xvfb-run.sh, which waits forxdpyinfobefore starting the payload. Every call site now uses it. This is a package update picked up at job time, so it would have hit main on the next Linux job regardless of this branch.