Author	SHA1	Message	Date
Enrico Weigelt, metux IT consult	79dee7378e	.github: retry-transient: only once an hour ... reduce the interval to once an hour instead of every quarter hour. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-10-08 17:41:16 +02:00
Enrico Weigelt, metux IT consult	fffa0070b9	.github: move qemu-user builds to separate pipeline ... These builds just taking too long and slowing down all other builds. Let's only run them on a few branches (eg. baselines) and not on each single PR or arbitrary pushes. It's only running on: - master - release/* - qemu/* Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-10-08 11:59:45 +02:00
Enrico Weigelt, metux IT consult	5b2804aa28	.github: ubuntu: do extra apt update (github mirrors not in sync) ... Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-10-08 11:48:48 +02:00
Enrico Weigelt, metux IT consult	226ed53422	.github: don't limit builds to certain branches ... Several important lanes had been silently skipped, thus several broken commits could pass, which never should have passed. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-10-06 19:00:54 +02:00
Enrico Weigelt, metux IT consult	ecf8c20611	.github: netbsd: upgrade to 11.0 ... NetBSD's CDN doesn't seem to deliver older versions anymore, so we need to upgrade to 10.2. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-10-05 16:40:22 +02:00
Enrico Weigelt, metux IT consult	1d4c4208ac	ci: use release/** pattern for PR CI to cover future release branches	2026-10-01 15:07:53 +02:00
Enrico Weigelt, metux IT consult	41c06aa72e	ci: restrict push CI to non-PR branches; PR CI only on master and release/*	2026-10-01 15:02:55 +02:00
Enrico Weigelt, metux IT consult	71bf6fcdc6	ci: bump dragonflybsd-vm from v1.1.4 to v1.3.4 ... The DragonFlyBSD lane reports success without ever building. Measured in two independent runs: Run 36739123551 (failed overall, 18:25:20 -> 18:47:32, complete log): the guest-side shell was Killed twice (SIGKILL, OOM on the runner while a nested QEMU runs), and immediately after each kill the action still emitted ##[end-action ... outcome=success;conclusion=success]. No build ran in any of the three attempts: "meson" 0 occurrences, "running xserver build" 0. That is why attempts 1 and 2 were recorded as green. Run 36763250919 (reported success, 19:11:48 -> 19:15:29, complete log): the build script started and meson setup completed ("Build targets in project"), then nothing: "Compiling C" 0, "Linking" 0, "meson install" 0. Conclusion: success. So the lane swallows the exit code of the inner command. v1.1.4 is also twelve tag releases behind; the newest is v1.3.4. Not a host-load problem, which is why no load metric would have found it: in run 36762972575 netbsd, freebsd, hurd, gentoo and the image job all succeeded while dragonflybsd failed, on the same runner class at the same time. Fixing this restores an actual signal rather than making a broken one louder. Whether v1.3.4 also fixes the swallowed exit code is to be measured, not assumed - that is what the CI run on this PR is for. The other VM lanes are already current: freebsd floats on @v1 and is the most reliable lane, netbsd was bumped to v1.5.0 when it broke the same way. Signed-off-by: Enterprise <info@metux.net>	2026-10-01 13:31:15 +02:00
Enrico Weigelt, metux IT consult	78717e2d52	CI: bump vmactions/netbsd-vm v1.2.3 -> v1.5.0 ... The NetBSD lane occasionally fails before the build even starts: the QEMU VM does not come up SSH-reachable (`ssh: Could not resolve hostname netbsd`), exhausting the three retry attempts. v1.5.0 updates the vmactions runner (also drops the deprecated Node 20 runtime) and is expected to bring more reliable VM boot. Nothing else touched. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-09-18 20:58:19 +02:00
Enrico Weigelt, metux IT consult	b6384684a1	cygwin CI: install python312-lxml instead of python39-lxml ... The Cygwin meson runtime resolves to /usr/bin/python3.12.exe (python312 is pulled in transitively as a meson dependency), while the workflow only installed python39-lxml. hw/xwin/glx/meson.build therefore failed with "python3 lxml module not found" on every PR. Install the lxml module for the Python interpreter meson actually uses. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-09-11 20:41:21 +02:00
stefan11111	bd6f4c6eee	CI: Enable DRI2 and DRI3 on Solaris ... Signed-off-by: stefan11111 <stefan11111github@gmail.com>	2026-09-09 11:33:33 +02:00
Enrico Weigelt, metux IT consult	5102fcb4b9	ci: native QEMU user-mode build lanes for foreign CPU architectures ... Add CI build lanes for ppc64el, armhf, mipsel (Debian bookworm) and sparc64, alpha (Debian-ports unstable) using QEMU user-mode emulation with debootstrap rootfs. The lanes run the real target-architecture gcc under qemu-user-static, so arch-specific inline asm (include/compiler.h, include/xlibre_membarrier.h) is genuinely compiled and tested — the x86_64-host lanes never exercise these code paths (the __asm__ "missing trailing ;" bugs in PR #3484/#3485 were invisible to every existing lane). Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-08-08 17:30:06 +02:00
Enrico Weigelt, metux IT consult	a3b152c165	CI/hurd: pin git/git-man to unreleased when the sid toolchain is unresolvable ... debian-ports can briefly hold an inconsistent sid/unreleased state for hurd-amd64: the arch:all git-man moves ahead (currently 2.55.0-1) while the hurd git binary lags (2.51.0-1 in sid, 2.53.0-1+hurd.1 in unreleased). Plain `apt-get install git` then fails hard: git:hurd-amd64=1:2.53.0-1+hurd.1 Depends git-man (< 1:2.53.0-.) but none of the choices are installable E: Unable to satisfy dependencies. apt picks the newer arch:all git-man from sid over the coherent unreleased pair, so git's `git-man (<< 1:2.53.0-.)` constraint can never be satisfied. This breaks every xserver-build-hurd run until the ports buildd catches up (affects all PRs, not any specific change). Work around it: retry the toolchain install with git/git-man pinned to unreleased when the plain install fails. Remove this fallback once the ports buildd rebuilds git for hurd-amd64. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-08-04 13:30:25 +02:00
callmetango	856d325dea	01-bug-report.yml: Add 25.2.x to version selector	2026-07-31 16:18:49 -04:00
Enrico Weigelt, metux IT consult	87dffaf30e	CI: work around Ubuntu mirror removing libc6-dbg (404 on fetch) ... libc6-dbg was pulled in as a dependency of valgrind, but the Ubuntu mirror no longer has the exact version matching the runner image's libc6. This caused all ubuntu-* CI lanes to fail with a 404 error. Fix: remove valgrind from the shared install-pkg.sh (it was only needed by the valgrind lane anyway) and install it separately in the xserver-build-ubuntu-valgrind job with a fresh apt-get update to resolve the latest package versions. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-07-29 16:35:25 +02:00
Enrico Weigelt, metux IT consult	06d4b6d0cc	licensing: add SPDX AGPL-3.0-or-later headers to .github/ files authored by Enrico Weigelt ... All 47 files in .github/ (actions, arch, docker, scripts, workflows) that were originally authored by Enrico Weigelt and don't end up in the final X server binary now carry SPDX-License-Identifier headers. File .github/scripts/ubuntu/run-pyxtest-valgrind.sh already had the AGPL-3.0-or-later header — left as-is. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-07-28 17:53:55 +02:00
Enrico Weigelt, metux IT consult	4ebdb499dd	Xext: remove the XF86-BigFont extension ... XF86-BigFont lets local clients share font-metrics arrays via SysV shared memory (and compresses them for remote clients) -- a local-only optimization dating back to ancient XFree86. It hasn't been built by default for really long time. Enabling it in our CI immediately exposed that the code has bit-rotted on 7 non-Linux CI lanes (macOS/BSD: including a nonexistent Linux-only header; Solaris: an ARRAY_SIZE redefinition pulled in via its system headers; mingw32: implicit geteuid()/getegid() plus an unused variable). Concrete evidence that this extension currently has no real users exercising it, since none of that breakage was ever noticed. Therefore no practical need for carrying it around any longer. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-07-27 18:31:44 +02:00
Enrico Weigelt, metux IT consult	995d4979fe	CI: solaris: drop non-existent x11/library/libxt from pkg list ... The FMRI does not exist in the OpenIndiana Hipster catalog (both the old per-package loop and the new batched install failed on it; the xserver does not link libXt, so the build succeeds without it). A single unmatched FMRI aborts the whole IPS transaction, so leaving it in broke the batched install and forced the per-package fallback. Removing it lets the single pkg install transaction succeed. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-07-14 13:56:46 +02:00
Enrico Weigelt, metux IT consult	408a4a06bd	CI: speed up solaris build (pre-baked -build image, batched pkg install) ... Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-07-14 13:56:46 +02:00
Enrico Weigelt, metux IT consult	bad0d97865	CI: pin go-x11proto to v0.0.8 instead of master ... Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-07-14 12:55:06 +02:00
Enrico Weigelt, metux IT consult	0229438f09	ci: run pyxtest under valgrind memcheck on Ubuntu ... Add a dedicated xserver-build-ubuntu-valgrind job that builds a debug Xvfb and runs the pyxtest suite (test/pyxtest/) with --valgrind, so memory errors (invalid reads/writes, use-after-free, uninitialised value use) are caught in CI, not just OOB bugs ASAN happens to catch locally. The pytest-based harness (test/pyxtest/conftest.py, xserver.py, valgrind.py) already implements valgrind support; it was simply never wired into CI. pytest itself was also not installed in the Ubuntu CI image, so the existing pyxtest suite has been silently skipped (not built, since meson's find_program('pytest', required: false) found nothing) on every other Ubuntu lane too. Installing python3-pytest fixes that as a side effect: the normal xserver-build-ubuntu* jobs now also actually run pyxtest (without valgrind) as part of "meson test", exercising the existing protocol-fuzzing regression tests that were previously dead weight. Kept as a separate lane rather than adding --valgrind to the existing run-xserver-build-and-test.sh: valgrind is slow, and pyxtest only needs Xvfb, not the full Xorg/Xephyr/Xnest/XTS/piglit build the other lanes carry. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-07-07 11:16:03 +02:00
Enrico Weigelt, metux IT consult	a959773be9	ci: freebsd: synthesize a bzip2.pc if the bzip2 package doesn't ship one ... Installing bzip2 (previous commit) wasn't enough: on this CI image the package provides libbz2 + headers but no bzip2.pc, so pkgconf's dependency resolution for freetype2 (Requires.private: bzip2, pulled in transitively by xfont2) still fails outright even though the library itself is present and usable - confirmed via the job log still showing 'Package bzip2 was not found in the pkg-config search path' with env[PKG_CONFIG_PATH] correctly unset (default search path), ruling out a path problem. This is missing package metadata, not a missing library: write a minimal bzip2.pc if pkg-config --exists bzip2 fails, locating the actual installed lib/header paths via 'pkg info -l bzip2' rather than hardcoding them. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-07-06 19:36:54 +02:00
Enrico Weigelt, metux IT consult	f95ab701fe	ci: freebsd: install bzip2 (freetype2's pkgconfig Requires) ... xfont2's pkg-config dependency lookup failed once the IGNORE_OSVERSION fix (previous commit) let the package-version check through to the actual build: freetype2's .pc file Requires: bzip2, but bzip2 wasn't in the install list, so pkgconfig couldn't generate cflags for xfont2 and meson bailed. Pre-existing gap, unmasked (not caused) by the previous fix - this lane never got far enough to hit it before. The 'ld: error: undefined symbol: main' lines in the same log are unrelated noise from meson's own linker-capability probe (compiling a main-less test program to detect LLD), not a real link failure. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-07-06 19:36:54 +02:00
Enrico Weigelt, metux IT consult	aaab164c0e	ci: freebsd: set IGNORE_OSVERSION=yes to survive pkg repo/VM release skew ... FreeBSD's official package repo tracks the '14' ABI branch, not the exact point release pinned for the CI VM (release: "14.3" in build-xserver.yml). When the repo rolls forward to packages built against a newer point release, pkg refuses to proceed non-interactively ("repository FreeBSD contains packages for wrong OS version", no TTY in CI to answer the y/N prompt), so pkg update -f fails every time and the job never reaches the actual build - currently breaking every FreeBSD CI run on master and on open PRs alike. IGNORE_OSVERSION=yes is pkg's own documented escape hatch for exactly this case; a one-point-release ABI skew is expected to be compatible in practice. Confirmed via job logs: 'pkg: ... wrong OS version: FreeBSD:14:amd64' / 'Newer FreeBSD version for package zycore-c: package: 1404000, running userland: 1403000' immediately precedes the generic 'pkg update -f' retry-loop failure that was previously misdiagnosed as a network flake. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-07-06 19:36:54 +02:00
Enrico Weigelt, metux IT consult	410e0ac3af	Xext: namespace: add Xnamespace management protocol extension ... Add a superPower-only X11 extension (X-NAMESPACE) for managing namespaces at runtime, the settings that were previously only provisionable from the namespace.conf file at startup. Requests: QueryVersion, {List,Create,Delete,Query}Namespace, SetNamespaceFlags, {Add,Remove,List}AuthToken and GetClientNamespace. Names are the wire identifier; capabilities/attributes are bitmasks. Transient namespaces auto-remove when their last client exits. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-07-03 20:05:08 +02:00
Enrico Weigelt, metux IT consult	812c0238ad	ci: bump go-x11proto (go-xts) pin v0.0.4 -> v0.0.7 ... The go-xts CI suite is built against the go-x11proto client library, pinned via PKG_GOXPROTO_REF/GOXPROTO_REF. Raise it from v0.0.4 to the current release v0.0.7 (fresh tag pushed upstream). Both pin sites must move together: install-prereq.sh consumes PKG_GOXPROTO_REF (conf.sh), which defaults to $GOXPROTO_REF from the workflow env. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-07-01 20:12:07 +02:00
Enrico Weigelt, metux IT consult	5c9ba0bd89	.github: build xf86bigfont on the CI lanes, keep it off by default ... PR #3202 both fixed the extension's cross-platform build *and* flipped the meson default on. Keep the compile fixes, but restore the meson default to off: xf86bigfont is a low-relevance legacy optimisation (server-side core fonts; modern toolkits render client-side) that was deliberately off, and re-enabling it for all downstreams isn't wanted. Instead pass -Dxf86bigfont=true on the CI build lanes, so the extension stays compiled and validated across all platforms in CI without changing the shipped default. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-07-01 17:50:11 +02:00
Enrico Weigelt, metux IT consult	96e5bf81a6	.github: enable -Dwerror on the solaris/gentoo/openbsd build lanes ... These lanes were pinned -Dwerror=false as "freshly added"; a probe run shows they now build clean under -Dwerror=true once the two warnings fixed in the preceding commits are addressed (gentoo: bigreq write unused-result; openbsd: unused /proc vars). gentoo's old format-truncation blocker is already gone (fixed in set_sun_path). solaris was clean already. alpine stays -Dwerror=false: on musl, libbsd's <bsd/sys/cdefs.h> pulls the system <sys/cdefs.h> whose #warning ("non-standard ... deprecated") -Werror=cpp turns fatal in every libbsd-using TU — a toolchain quirk, not an xserver warning, so it can't be fixed in-tree. Documented inline. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-07-01 11:40:27 +02:00
Enrico Weigelt, metux IT consult	b37fd20a2d	hurd: build every Hurd-capable X server by default in CI ... The GNU/Hurd lane added earlier built only the virtual DDXes (Xvfb + Xnest). The other servers that build on GNU/Hurd have since been confirmed in CI, so make them all part of the lane's FATAL build, in one meson run in the existing xserver-build-hurd job (parallel with the other platform lanes, no separate workflow; ~12 min in the VM): - Xvfb, Xnest virtual DDXes - Xorg the physical xfree86 server (libpciaccess has a Hurd backend, so the PCI layer configures/links) - Xephyr kdrive server running as an X client over XCB - glx the GLX extension (builds without libdrm) A regression breaking any of them now fails CI. All five link cleanly (hw/xfree86/Xorg, hw/kdrive/ephyr/Xephyr, hw/vfb/Xvfb, hw/xnest/Xnest, hw/xfree86/dixmods/libglx.so). Kept off because Hurd has no DRM kernel interface: - dri1/dri2/dri3 — libdrm's <drm.h> pulls a nonexistent mach/x86_64/ioccom.h - glamor — glamor_egl.c needs DRM_FORMAT_MOD_INVALID + GBM, and GBM needs DRM; confirmed non-buildable on Hurd And Xfbdev (kdrive fbdev): it builds hw/kdrive/linux, which needs Linux VTs (<linux/vt.h>) and would require a dedicated Hurd kdrive backend. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-07-01 09:57:08 +02:00
Enrico Weigelt, metux IT consult	627b6da5a2	.github: content-address all deps images, drop standalone image builders ... Make every CI base image content-addressed and built-if-missing, and fold the three standalone image-builder workflows into the main "Build X servers" pipeline. Previously the gentoo / ubuntu base images and the SDK image were built by their own workflows (gentoo-image.yml, ubuntu-deps-image.yml, sdk-image.yml). Those rebuilt the full image on every branch that touched the relevant paths (and a brand-new branch matched the push path filter spuriously), wasting ~15-22 min per run; and a WIP branch that changed the deps still consumed master's :latest base, so its build could be stale. Now each deps image gets a tag derived from a hash of its build inputs and is built only when that exact tag is missing in GHCR: - gentoo-deps-image : tag = hash(gentoo Dockerfile) - ubuntu-deps-image : tag = hash(ubuntu Dockerfile + install-pkg.sh + install-prereq.sh + conf.sh + util.sh) An unchanged image is reused on ANY branch via a few-second `docker manifest inspect` existence check (no rebuild); a branch that edits the deps automatically gets its own image — no waste and no staleness. :latest is moved only by master. The per-commit SDK (build-sdk-image) now builds FROM the content-hashed ubuntu base: the SDK Dockerfile takes `ARG BASE` and the job passes `--build-arg BASE=…/xserver-ubuntu-build:<hash>`, so a WIP branch's SDK is built on exactly the deps that branch defines. This also makes the standalone sdk-image.yml redundant (build-sdk-image already produces the per-commit xserver-sdk:<sha> that the driver jobs consume), so it is removed too. ubuntu-deps-image is gated like build-sdk-image (its only consumer), i.e. it runs when the ABI changed or on a tag. prune-sdk-images.yml is kept (it still prunes accumulated xserver-sdk:<sha> tags). Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-06-30 17:47:36 +02:00
Enrico Weigelt, metux IT consult	18d94c4ced	.github: add GNU/Hurd VM build lane (Xvfb + Xnest) ... There is no vmactions GNU/Hurd VM, so this hand-rolls one on the ubuntu runner: it fetches Debian's pre-installed Hurd amd64 image, injects an ssh key and routes the guest console to serial, then boots it in QEMU and builds the X server inside via ssh. Key gotchas baked into the scripts: - The amd64 Hurd image uses rumpdisk for SATA, so the disk must be attached via q35 + an ich9-ahci AHCI controller (the default i440fx IDE disk yields an ext2fs I/O error and never boots). - gnumach only boots single-CPU (no -smp). - GRUB and gnumach log to VGA only; the grub.cfg is patched to add a serial console so the boot is visible under -nographic. - /dev/kvm exists on GH runners but isn't accessible to the runner user; chmod 666 it (falls back to TCG if unavailable). - VM boot can flake transiently, so boot is retried up to 3x. Inside the VM the toolchain (git, meson, ninja, pkg-config) install is fatal; the X protocol libs are best-effort, and the server is configured for the parts that build on Hurd today: -Dxvfb=true -Dxnest=true with xorg/xephyr/glx/dri/udev/logind disabled. Runs in ~10.5 min and is wired into the primary "Build X servers" workflow alongside the other platform lanes. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-06-30 14:08:02 +02:00
Enrico Weigelt, metux IT consult	b95ed8ebf2	.github: re-enable -Dwerror for the RHEL build job ... The -Werror=format-truncation false positive that forced -Dwerror=false on the RHEL job is now fixed at the source (set_sun_path), so tighten the job back to -Dwerror=true to match the other Linux build jobs. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-06-30 13:34:29 +02:00
Enrico Weigelt, metux IT consult	9a1618d66d	.github: add RHEL build job ... Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-06-30 09:07:54 +02:00
Enrico Weigelt, metux IT consult	59dcf88d66	.github: prune old xserver-sdk images (step 5 of driver-pipeline rework) ... build-sdk-image pushes a per-commit xserver-sdk:<sha> tag, so they accumulate in GHCR. Add a weekly/on-demand workflow that keeps the newest 20 versions and deletes the rest (best-effort; org package-version deletion may need a delete:packages PAT). Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-06-29 18:24:13 +02:00
Enrico Weigelt, metux IT consult	957a77305c	.github: build drivers in the SDK image (steps 3+4 of driver-pipeline rework) ... Replace drivers-prepare-ubuntu (which built the xserver+SDK and published it through actions/cache) with a build-sdk-image job that pushes xserver-sdk:<sha> to GHCR, and run drivers-build-ubuntu in `container: xserver-sdk:<sha>`. Each driver job now only compiles the driver against the baked-in SDK at /opt/x11 — no apt, no prereq build, no cache restore (the cache could be evicted under load before the queued driver jobs restored it, forcing a full manual rerun; a registry image isn't subject to that LRU eviction). Verified locally: input-evdev builds cleanly inside the SDK image. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-06-29 18:23:51 +02:00
Enrico Weigelt, metux IT consult	2d43435b40	.github: add xserver SDK image (step 2 of driver-pipeline rework) ... Builds the X-server + SDK (-Dxorg-sdk=true) from the current commit on top of the xserver-ubuntu-build base image and pushes it to GHCR as xserver-sdk:<sha> (+ :latest). Multi-stage so the final image carries only the installed SDK in /opt/x11, not the source. This is the per-commit artifact the driver builds will consume in the next step (container: xserver-sdk:<sha>) instead of restoring the xserver+SDK from the actions/cache that can be evicted under load. No existing job consumes it yet. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-06-29 16:45:45 +02:00
Enrico Weigelt, metux IT consult	a3e366a787	.github: add ubuntu deps base image (step 1 of driver-pipeline rework) ... Adds a prebuilt base image with the apt build dependencies and the from-source prereqs (rendercheck, libdrm, libxcvt, xorgproto, xts, piglit, goxproto) baked in, plus a builder workflow that pushes it to GHCR (on demand / weekly / on input change). This is step 1 of reworking the driver pipeline: later steps build the xserver+SDK on top of this image and hand it to the driver builds via a per-commit image instead of the eviction-prone actions/cache (under high load the SDK cache can be evicted before the queued driver jobs restore it, forcing a full manual rerun). No existing job is changed yet — this only adds the image + builder. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-06-29 15:41:49 +02:00
Enrico Weigelt, metux IT consult	2bd571b95f	.github: drop GHCR credentials from gentoo job (image is now public) ... The xserver-gentoo-deps image is public, so the container pulls anonymously. Remove the now-unnecessary credentials block and the job-level packages:read permission; this also lets fork PRs pull it. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-06-26 22:46:12 +02:00
Enrico Weigelt, metux IT consult	171a1ac3ce	test: retry the XTS suite on timeout instead of just failing ... The xvfb and xephyr-glamor XTS suites on llvmpipe occasionally *hang* — a single subtest wedges and meson kills it at the suite timeout, reported as a per-test TIMEOUT (SIGTERM), e.g. "Ok: 15, Fail: 0, Timeout: 1". That is a flake, not a real failure, and raising the timeout would only delay it. Wrap `meson test` in a timeout-aware retry: on a TIMEOUT, re-run the tests with --no-rebuild (the existing build is reused, no recompile) for up to 3 attempts. A genuine failure (no TIMEOUT in the log) is not retried and fails the job immediately, so real regressions are never masked. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-06-26 21:35:23 +02:00
Enrico Weigelt, metux IT consult	86ab67a794	ci: bump go-x11proto test dependency to v0.0.4 ... v0.0.4 fixes a "send on closed channel" race in the go-x11proto client's event read loop (eventCh/errorCh closed by Close() while readLoop was still delivering an event), which surfaced as intermittent go-xts CI failures (e.g. TestXSettingsWatch). Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-06-26 21:34:40 +02:00
Enrico Weigelt, metux IT consult	d7d81b8bce	.github: add gentoo job ... Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-06-26 18:48:28 +02:00
Enrico Weigelt, metux IT consult	d6662255a4	.github: add OpenBSD build ... Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-06-25 20:35:25 +02:00
Enrico Weigelt, metux IT consult	e94eab9cd5	.github: use go-x11proto test suite in the CI ... Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-06-25 20:16:15 +02:00
Enrico Weigelt, metux IT consult	29ce34c6fa	ci: add explicit test stage to cygwin build ... Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-06-25 19:23:33 +02:00
Enrico Weigelt, metux IT consult	1e824a563e	.github: add Solaris/Illumos build ... Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-06-25 19:19:21 +02:00
Enrico Weigelt, metux IT consult	80ec128e1a	ci: add debug build job for ubuntu ... Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-06-25 16:35:04 +02:00
Enrico Weigelt, metux IT consult	c4252526cf	ci: make BSD VM build jobs resilient to transient failures ... The freebsd/netbsd/dragonfly jobs run in vmactions VMs whose boot and image provisioning flake transiently; a single hiccup aborted the whole run and had to be restarted by hand. The existing dragonfly retry was ineffective: it re-ran the VM action without tearing down attempt 1's leftover libvirt domain, so attempt 2 hit "Guest name 'dragonflybsd' is already in use" and timed out again. Two layers of hardening: - In-VM pkg installs (freebsd, dragonfly) now retry with backoff and an explicit catalogue refresh, so a transient mirror failure no longer fails the build (netbsd already had mirror fallback + curl --retry). - Each BSD VM step is retried up to 3x, with a new shared .github/actions/libvirt-cleanup composite action destroying the leftover domain + orphaned qemu between attempts. Only the final attempt can fail the job; a success short-circuits the rest. Rolled out to freebsd and netbsd too (previously single-attempt) and added timeout-minutes to bound a fully-broken job. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net> Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>	2026-06-25 15:00:36 +02:00
Enrico Weigelt, metux IT consult	2f22ffcad2	ci: skip driver builds unless ABI-relevant files changed ... The vast majority of our commits doesn't touch anything that's driver visible, thus no value in additional compile-checking of ~48 drivers here. Now only running those checks on SDK header changes or tags. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-06-25 09:56:10 +02:00
Enrico Weigelt, metux IT consult	123446d11d	ci: de-duplicate pipeline runs for pull requests ... A push to a branch that also has an open pull request triggered two identical workflow runs: one for the `push` event and one for the `pull_request` (synchronize) event. That doubles the runner cost and clutters the checks list, most painfully when a PR branch is re-pushed repeatedly. Drop the redundant run without changing normal-branch behaviour and without relying on any branch-name convention: - A top-level `concurrency` group keyed by `github.ref`, with `cancel-in-progress: true`, cancels a still-running run when the same ref is pushed again. Only the newest commit gets built; stale in-flight runs are aborted. - A guard on the four build jobs that have no `needs:` (ubuntu-fetch-pkg, xserver-build-macos, xserver-build-cygwin, xserver-build-arch): if: >- github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name != github.repository On the `pull_request` event for a branch in this same repository the second operand is false, so the job is skipped: the branch's own `push` event has already built it. Every other build job chains off one of these four roots, and a skipped `needs:` job skips its dependents, so the condition does not have to be repeated on each job. Fork PRs raise no `push` event in this repository, so the second operand is true and they keep building via the `pull_request` event exactly as before. Pushes to ordinary branches still run on `push`, unchanged. The PR-only `check-sign-off` job and the tag-only `release` job are deliberately left untouched. The PR checks indicator is unaffected: for GitHub Actions the required status-check context is the job name (independent of the triggering event), and for a same-repo PR the push run reports those jobs against the very commit the PR points at. Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-06-25 08:54:19 +02:00
Enrico Weigelt, metux IT consult	862a9333f8	.github: dragonfly: automatically try again on transient failure ... Signed-off-by: Enrico Weigelt, metux IT consult <info@metux.net>	2026-06-25 07:34:42 +02:00