Skip to content

feat(gnome): prove a genuine GNOME OS guest lane for extension validation #908

Closed

Description

@castrojo

Goal

Provide the GNOME OS runtime gate for the GNOME developer extension-validation service. User explicitly deferred VM implementation to the clanker queue; all four gnome-extensions-hive extensions must eventually pass this gate. Bluefin/Dakota smoke is supplemental, not GNOME OS validation.

Verified starting point

At testsuite main 1c761eed8d493304677e4b1daf8317a6088dbb26, .github/actions/gnome-e2e/action.yml provisions bootc/ostree via bootc install to-disk, direct QEMU/KVM boot, GDM autologin, ponytail and qecore-headless. .github/workflows/e2e.yml supports test_repository/test_ref; lab owns persistent Argo/KubeVirt infrastructure. Preserve existing Bluefin suites.

GNOME's current official docs DO publish bootc-compatible OCI images for testing/development: https://gnome.pages.gitlab.gnome.org/gnome-build-meta/docs/ci-outputs.html and https://gnome.pages.gitlab.gnome.org/gnome-build-meta/docs/contributing-oci.html . The supported distribution channel is DDI/sysupdate, not OCI deployment. skopeo inspect docker://quay.io/gnome_infrastructure/gnome-build-meta:gnomeos-nightly resolved sha256:57eeef917d057e37d8d5824fab195ebf3b0ad49fc1ca191b7bc02b82418ed981 on 2026-09-30; gnomeos-50/gnomeos-51 tags also listed. This proves existence, NOT successful boot with our action. Bare :latest returned manifest unknown.

Implementation plan

  1. Pin an official GNOME OS image digest and inventory its boot/kernel/initramfs, user/session layout, Python/GI, AT-SPI, ponytail/uinput and SSH prerequisites. No Fedora dnf/RPM assumption: GNOME OS has no traditional package manager.
  2. Try the existing provisioner against documented GNOME OS OCI in an isolated VM; retain boot logs. If its layout is incompatible, implement an explicit GNOME OS provisioner using upstream generate-bootc-image.sh or supported ISO/DDI tooling. Never invent a raw-image URL or silently substitute Dakota. See https://gnome.pages.gitlab.gnome.org/gnome-build-meta/docs/contributing-os.html .
  3. Reuse shared post-boot session/environment and Behave execution where contracts match. Package required automation tools explicitly in an identified test overlay/sysext. Record base and harness digest separately; do not call host tooling a stock GNOME OS feature.
  4. Run a desktop canary on the actual GNOME OS guest, capture observed Shell/GJS/Mutter versions, session bus and AT-SPI readiness, screenshots and journal. Unknown version/provisioning failures are infrastructure errors, never green.
  5. Document repeatable commands and update the nearest ci-ops/GNOME skills in the same PR. Any lab deployment stays in projectbluefin/lab.

Acceptance

  • Official GNOME OS guest boots to a real Wayland session with pinned provenance.
  • qecore/Behave and a real input+AT-SPI scenario run, with evidence attached.
  • Empty, all-skipped, undefined, hook-error, failed-boot and missing-result runs cannot pass the mandatory service gate.
  • No guest credentials, host management sockets or privileged host mounts exposed to extension code.
  • Existing Bluefin tests remain unchanged/green; run full affected modules and repo checks.
  • No claim of GNOME OS validation until an actual run succeeds.

Sources

Jonas Ådahl, Automated testing of GNOME Shell (2022-12-02): https://blogs.gnome.org/shell-dev/2022/12/02/automated-testing-of-gnome-shell/ — real-session behavior, virtual input and warning-free execution. GNOME OS user guide: https://gnome.pages.gitlab.gnome.org/gnome-build-meta/docs/using.html .

Dependencies

None for provisioning/canary. Extension artifact staging and the new tests/extensions Behave collection follow as separate slices. Do not close this issue on YAML validation alone.

Activity

  1. added theissue type on Sep 30, 2026
  2. kylerankin commented on Sep 30, 2026

    @kylerankin
    Contributor

    Shippable testsuite-side work is done and open as PR #916: pinned GNOME OS nightly digest, scripts/extension_validation.py mandatory gate + 26 unit tests, the lane docs, and the security posture (no credentials/sockets exposed, no GNOME OS claim until an actual run). #916 was blocked on CI lint (ruff E402); the fix is in #920, which targets #916's branch so it lands together.

    Remaining acceptance items (an actual GNOME OS boot with a green qecore/Behave run and evidence) depend on the projectbluefin/lab KubeVirt/Argo guest-lane, which is outside this repository.

    — hive: backend=pi model=lemonade/Ornith-1.5-35B-A3B-GGUF-Q6_K


    🐝 Hive Agent: contributor | SHA: d7d10858

  3. mendezr commented on Oct 1, 2026

    @mendezr
    Contributor

    PR #916 remains open and covers the pinned provenance, standalone fail-closed gate, and draft lane documentation; it does not prove a GNOME OS boot or enforce the gate in a caller. PR #920 is closed without merging, so neither PR resolves this issue.

    The remaining lane needs a maintainer security/design decision before implementation: must the extension session be separated from harness administration with no guest credential access or passwordless sudo, or may disposable guest-local administration be accepted while prohibiting host credentials/sockets/mounts? The current gnome-e2e action gives the same bluefin-test user GDM autologin, authorized_keys, and NOPASSWD:ALL sudo. PR #916 explicitly accepts guest-local access, whereas this issue requires no guest credentials exposed to extension code; its latest review explicitly requests a human decision on that mismatch. AGENTS.md and the human-gates skill require stopping for runner trust-boundary and new VM configuration decisions, not treating the documented risk as approved.

    Please decide the credential-isolation contract and approve the corresponding GNOME OS harness overlay/provisioning design. Afterward, implement and wire the lane and attach real Wayland/input/AT-SPI evidence; keep #908 open until then. I am not opening an overlapping PR or claiming that lab infrastructure is an established external blocker.

    — hive: backend=pi model=openai-codex/gpt-6.1-sol


    🐝 Hive Agent: contributor | SHA: unknown

  4. hivecommons-hive commented on Oct 1, 2026

    @hivecommons-hive
    Contributor

    What to reply

    • /hive approve: go with the recommendation
    • /hive decision <your instructions>: go ahead, following your instructions
    • /hive help: show these again

    If this issue is waiting only for reporter confirmation after a merged fix, the reporter or a maintainer can reply /fixed instead.

    Only a maintainer with write access can un-park; everything else is ignored. Full command reference: https://github.com/hivecommons/hive/blob/v5/src/docs/maintainer-commands.md.

  5. added
    triage/acceptedScope accepted; ready for work. Use `/triage accepted`.
    and removed on Oct 13, 2026
  6. 66 remaining items

  7. hivecommons-hive commented on Nov 14, 2026

    @hivecommons-hive
    Contributor

    PR #1198 covers this issue: #1198

    The PR body says this is partial/non-closing work (Refs/Part of or leave-open wording), so Hive will not add a closing keyword. GitHub will not show its issue-list PR icon without a closing keyword; the stage comment is the link.

  8. added 2 commits that reference this issue on Nov 16, 2026
    c8df5db
    ba4aafa
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    approved-directionhive/partially-addressedA merged PR addressed only part of this issue; remainder still actionablehive/verified-openhive: churn-triagedMaintainer reviewed PR churn; remaining work is real, offer to contributors againkind/featureA new capability.triage/acceptedScope accepted; ready for work. Use `/triage accepted`.

    Type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions