Skip to content

fix(gnome-e2e): check out testsuite from the right repository - #1139

Merged
hivecommons-hive[bot] merged 4 commits into
mainfrom
fix/gnome-e2e-checkout
Nov 11, 2026
Merged

hivecommons-hive[bot] merged 4 commits into
mainfrom
fix/gnome-e2e-checkout

Conversation

@Danathar

@Danathar Danathar commented Nov 10, 2026 •

Copy link
Copy Markdown
Contributor

Summary: the GNOME OS desktop canary (#908) can't check out testsuite. The gnome-e2e action reads github.action_repository inside its checkout step, where it names actions/checkout, so it fetches testsuite's commit from actions/checkout and fails with "not our ref" (run 42919533096). The "extension-validation gate script missing" error that follows comes from that empty checkout. Once the checkout works, the canary would reach a bootc loopback install, which fails on the Ubuntu 26.04 runners ubuntu-latest now gives (#1123), so it's pinned to 24.04 as in #1131.

Changes

  • .github/actions/gnome-e2e/action.yml: a "Resolve testsuite source" run step works out the repository and ref (falling back to the caller's repository for a local ./ call), and the checkout uses those outputs.
  • .github/workflows/gnomeos-canary.yml: runs-on: ubuntu-24.04, with the same comment as ci(e2e): run the VM jobs on ubuntu-24.04 #1131.
  • tests/unit/test_gnome_e2e_action_summary.py: a regression test that the checkout step doesn't read github.action_repository / github.action_ref directly.
  • docs/skills/ci-ops/e2e-workflow/references/gnomeos-desktop-canary.md: notes both.

Further fixes found by the canary runs on this branch

Each canary run got one stage further:

  1. Checkout from the wrong repository (the original fix above). Run 42919533096 failed with "not our ref".
  2. --pid=host on the install's podman run. Run 42923546804: "This command must be run with the podman --pid=host flag".
  3. The BLS lookup no longer ends the step silently under pipefail when bootc's late bootloader failure leaves no loader/entries (run 42929881520).
  4. The real ostree boot entry is used when there's no BLS entry, as in e2e.yml and startup-qualification.yml, instead of a guessed boot.1 path. Run 42932368882 failed with ostree-prepare-root "Couldn't find specified OSTree root".

With all four, run 42934968230 boots, reaches SSH (102 s), starts the GNOME session and installs the test stack. behave then fails on two remaining problems: tests.shared can't be imported in the guest, and qecore-headless writes under read-only /usr. Those are tracked in a follow-up issue. So the canary is still red, but every change here is proven by the next stage it reached.

How to test

The canary only runs on manual dispatch. From this branch:
gh workflow run gnomeos-canary.yml -R projectbluefin/testsuite --ref fix/gnome-e2e-checkout

Checked locally: the 22 tests in test_gnome_e2e_action_summary.py pass, including the new one, and ruff is clean. Neither file is a release workflow.

Refs #908

🤖 Generated with Claude Code

https://claude.ai/code/session_01WcWm7qQa538Zjcuvv8jBtn

The action read github.action_repository inside its checkout step,
where it names actions/checkout, so the GNOME OS canary fetched
testsuite's commit from actions/checkout and failed with "not our ref"
(run 42919533096). The extension-validation "script missing" error that
followed came from that empty checkout. Resolve the repository and ref
in a run step first, falling back to the caller's repository for a
local ./ call, and add a unit test that keeps the checkout step from
reading them directly.

Also run the canary on ubuntu-24.04: the action does a bootc loopback
install, which fails on 26.04 runners (#1123, #1131).

Assisted-by: Claude Opus 5.5 via Claude Code
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WcWm7qQa538Zjcuvv8jBtn
Signed-off-by: Danathar <Danathar@users.noreply.github.com>
@hivecommons-hive hivecommons-hive Bot added author/clanker A contributor's own hive authored this PR under their login dco All commits are DCO signed-off role/trusted Author is trusted for unattended merge (trusted-author role or registry/claim tier in require_tiers) stage/050-review-pending Review swarm requested; awaiting a verdict for the current head labels Nov 10, 2026
@codecov-commenter

Copy link
Copy Markdown

⚠️ Please install the 'codecov app svg image' to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

@hivecommons-hive

hivecommons-hive Bot commented Nov 10, 2026 •

Copy link
Copy Markdown
Contributor
a76291d

ℹ️ Sentinel notice — informational; author is trusted (owner), merge is not blocked

Hive flagged this PR (author @Danathar, head a76291d4ba86) because it matches behaviors that can override security controls, escalate privileges or damage the codebase. This is a heuristic, not an accusation — a maintainer should confirm the change is intended before it merges.

  • sensitive_path — changes 2 sensitive path(s) (Touches a sensitive path (OWNERS, workflows, policies, hive config, security docs, …))
    • .github/actions/gnome-e2e/action.yml
    • .github/workflows/gnomeos-canary.yml
  • ci_gate_weakening — CI configuration is deleted, softened or has gate checks removed (Disables or softens CI gates (continue-on-error, || true, removed checks))
    • .github/actions/gnome-e2e/action.yml

Hive did not add the sentinel-alert label because this author is trusted. The finding remains recorded for audit and the dashboard Security tab; set sentinel.trusted_authors_block: true to require the blocking label for trusted authors too.

@hivecommons-hive hivecommons-hive Bot added stage/072-returned-to-contributor Contributor PR: CI red, changes requested or conflicting; diagnosis posted, author must act and removed stage/050-review-pending Review swarm requested; awaiting a verdict for the current head labels Nov 10, 2026
@hivecommons-hive

hivecommons-hive Bot commented Nov 10, 2026 •

Copy link
Copy Markdown
Contributor

@Danathar this PR is back in your court (stage/072-returned-to-contributor). The hive will not merge it until the head changes.

Failing checks:

  • canary

Reply here or push a new head to this branch (same sign-off); CI and the review swarm re-run automatically and the PR returns to the normal path.

Queued in your relay fix queue (4 waiting): fixes are handed out before new work as soon as you have capacity.

With the checkout fixed, the canary reached the install and bootc
refused: "This command must be run with the podman --pid=host flag"
(run 42923546804). e2e.yml's install step already passes --pid=host.

Assisted-by: Claude Opus 5.5 via Claude Code
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WcWm7qQa538Zjcuvv8jBtn
Signed-off-by: Danathar <Danathar@users.noreply.github.com>
@hivecommons-hive hivecommons-hive Bot added stage/050-review-pending Review swarm requested; awaiting a verdict for the current head stage/095-pre-merge-soak Lane admitted; the pre-merge soak window is running before the merge is attempted and removed stage/072-returned-to-contributor Contributor PR: CI red, changes requested or conflicting; diagnosis posted, author must act stage/050-review-pending Review swarm requested; awaiting a verdict for the current head labels Nov 11, 2026
After bootc's expected late bootloader failure (bootupd) the ESP has no
loader/entries directory. find then exits non-zero, and under the
step's pipefail the command substitution ended the step silently right
after the guest preflight, before the fallback that builds the kernel
arguments without a BLS entry (canary run 42929881520).

Assisted-by: Claude Opus 5.5 via Claude Code
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WcWm7qQa538Zjcuvv8jBtn
Signed-off-by: Danathar <Danathar@users.noreply.github.com>
@hivecommons-hive hivecommons-hive Bot added stage/072-returned-to-contributor Contributor PR: CI red, changes requested or conflicting; diagnosis posted, author must act and removed stage/095-pre-merge-soak Lane admitted; the pre-merge soak window is running before the merge is attempted labels Nov 11, 2026
@hivecommons-hive

hivecommons-hive Bot commented Nov 11, 2026 •

Copy link
Copy Markdown
Contributor

Hive status — updated in place; edits do not notify. Full history is in the Activity Monitor.

stage 095-pre-merge-soak → 072-returned-to-contributor: optional-check-failing:canary

Evidence bundle `#1139a76291d` sha256:e132fe56e1834246706f0680697b7785cb4429fb012d2caf4f72b21381aef956 — download from the dashboard or `hivectl review evidence #1139`.

@hivecommons-hive hivecommons-hive Bot added stage/050-review-pending Review swarm requested; awaiting a verdict for the current head stage/095-pre-merge-soak Lane admitted; the pre-merge soak window is running before the merge is attempted stage/072-returned-to-contributor Contributor PR: CI red, changes requested or conflicting; diagnosis posted, author must act and removed stage/072-returned-to-contributor Contributor PR: CI red, changes requested or conflicting; diagnosis posted, author must act stage/050-review-pending Review swarm requested; awaiting a verdict for the current head stage/095-pre-merge-soak Lane admitted; the pre-merge soak window is running before the merge is attempted labels Nov 11, 2026
With no BLS entry the action guessed ostree=/ostree/boot.1/default/<hash>/0,
which does not exist on the installed disk, so ostree-prepare-root failed
("Couldn't find specified OSTree root") and SSH never came up (canary run
42932368882). Find the deployment's boot entry the way e2e.yml and
startup-qualification.yml do: use the existing boot.N/default/<hash>/0
link (mapping a versioned boot.N.M to boot.N for ostree-system-generator)
or create boot.0/default/<hash>/0 pointing at the deployment.

Assisted-by: Claude Opus 5.5 via Claude Code
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WcWm7qQa538Zjcuvv8jBtn
Signed-off-by: Danathar <Danathar@users.noreply.github.com>
@hivecommons-hive hivecommons-hive Bot added stage/050-review-pending Review swarm requested; awaiting a verdict for the current head and removed stage/072-returned-to-contributor Contributor PR: CI red, changes requested or conflicting; diagnosis posted, author must act stage/050-review-pending Review swarm requested; awaiting a verdict for the current head labels Nov 11, 2026
@hivecommons-hive hivecommons-hive Bot added stage/095-pre-merge-soak Lane admitted; the pre-merge soak window is running before the merge is attempted stage/072-returned-to-contributor Contributor PR: CI red, changes requested or conflicting; diagnosis posted, author must act and removed stage/095-pre-merge-soak Lane admitted; the pre-merge soak window is running before the merge is attempted labels Nov 11, 2026
@hivecommons-hive
hivecommons-hive Bot merged commit 467f16e into main Nov 11, 2026
7 of 8 checks passed
@hivecommons-hive

Copy link
Copy Markdown
Contributor

Merged by hive contributor-lane auto-merge: CI green for ≥30m, reviewer verdict approve, @Danathar was auto-promoted after 5 clean merged PRs in this hive's repositories.

@hivecommons-hive hivecommons-hive Bot removed role/trusted Author is trusted for unattended merge (trusted-author role or registry/claim tier in require_tiers) stage/072-returned-to-contributor Contributor PR: CI red, changes requested or conflicting; diagnosis posted, author must act labels Nov 11, 2026
kylerankin added a commit to kylerankin/testsuite that referenced this pull request Nov 11, 2026
The gnomeos-canary.yml projectbluefin#908 lane boots a GNOME OS VM and runs the
gnomeos-canary suite through .github/actions/gnome-e2e, but behave
aborts before any scenario runs for two reasons, both inside the
action:

1. tests.shared import fails. The "Run behave via qecore-headless"
   step copies tests/<suite> and tests/shared to the guest but not
   tests/__init__.py. Without it the guest's tests/ is only a
   namespace package, so an installed "tests" package on PYTHONPATH
   shadows tests.shared. Copy tests/__init__.py alongside the others,
   matching how .github/workflows/e2e.yml already does it.

2. qecore-headless writes /usr. GNOME OS ships a read-only /usr, and
   qecore-headless rewrites the GNOME Shell unit for --unsafe-mode.
   qecore's immutable-image fallback targets /etc/systemd/user/ but
   never creates that directory, so the rewrite fails and the run
   aborts. Pre-create /etc/systemd/user before qecore runs so the
   fallback lands instead of /usr.

Also read the QEMU pidfile with sudo (QEMU runs as root) to silence
the harmless "Permission denied" on cleanup.

- Add tests/__init__.py to the sparse-checkout list.
- Copy tests/__init__.py to the guest in the run step.
- sudo mkdir -p /etc/systemd/user before qecore-headless.
- sudo cat the QEMU pidfile.

Add tests/unit/test_gnome_e2e_action_summary.py asserting the copy,
sparse-checkout, systemd/user, and pidfile changes so they cannot
regress.

Closes projectbluefin#1144

Hive-Run: gh api -X POST projectbluefin/issues/1144/comments -f body='Canary dispatch from this fix branch: gh workflow run gnomeos-canary.yml --ref fix/gnome-e2e-canary-blockers. A maintainer must start it (the projectbluefin#908 lane is not merged into main yet).'
Hive-Plan: Rebased onto current projectbluefin/testsuite:main (projectbluefin#1139 added canary checkout support). Both blockers still present in .github/actions/gnome-e2e/action.yml; fixed there. Added a YAML-assertion regression test.
Hive-Spec: See PR body. Acceptance: gnomeos-canary.yml dispatched on the fix branch boots, runs the gnomeos-canary suite, and its "Gate suite has passing scenarios" step passes.
Signed-off-by: kylerankin <kylerankin@users.noreply.github.com>
kylerankin added a commit to kylerankin/testsuite that referenced this pull request Nov 11, 2026
The gnomeos-canary.yml projectbluefin#908 lane boots a GNOME OS VM and runs the
gnomeos-canary suite through .github/actions/gnome-e2e, but behave
aborts before any scenario runs for two reasons, both inside the
action:

1. tests.shared import fails. The "Run behave via qecore-headless"
   step copies tests/<suite> and tests/shared to the guest but not
   tests/__init__.py. Without it the guest's tests/ is only a
   namespace package, so an installed "tests" package on PYTHONPATH
   shadows tests.shared. Copy tests/__init__.py alongside the others,
   matching how .github/workflows/e2e.yml already does it.

2. qecore-headless writes /usr. GNOME OS ships a read-only /usr, and
   qecore-headless rewrites the GNOME Shell unit for --unsafe-mode.
   qecore's immutable-image fallback targets /etc/systemd/user/ but
   never creates that directory, so the rewrite fails and the run
   aborts. Pre-create /etc/systemd/user before qecore runs so the
   fallback lands instead of /usr.

Also read the QEMU pidfile with sudo (QEMU runs as root) to silence
the harmless "Permission denied" on cleanup.

- Add tests/__init__.py to the sparse-checkout list.
- Copy tests/__init__.py to the guest in the run step.
- sudo mkdir -p /etc/systemd/user before qecore-headless.
- sudo cat the QEMU pidfile.

Add tests/unit/test_gnome_e2e_action_summary.py asserting the copy,
sparse-checkout, systemd/user, and pidfile changes so they cannot
regress.

Closes projectbluefin#1144

Hive-Run: gh api -X POST projectbluefin/issues/1144/comments -f body='Canary dispatch from this fix branch: gh workflow run gnomeos-canary.yml --ref fix/gnome-e2e-canary-blockers. A maintainer must start it (the projectbluefin#908 lane is not merged into main yet).'
Hive-Plan: Rebased onto current projectbluefin/testsuite:main (projectbluefin#1139 added canary checkout support). Both blockers still present in .github/actions/gnome-e2e/action.yml; fixed there. Added a YAML-assertion regression test.
Hive-Spec: See PR body. Acceptance: gnomeos-canary.yml dispatched on the fix branch boots, runs the gnomeos-canary suite, and its "Gate suite has passing scenarios" step passes.
Signed-off-by: kylerankin <kylerankin@users.noreply.github.com>
kylerankin added a commit to kylerankin/testsuite that referenced this pull request Nov 11, 2026
The gnomeos-canary.yml projectbluefin#908 lane boots a GNOME OS VM and runs the
gnomeos-canary suite through .github/actions/gnome-e2e, but behave
aborts before any scenario runs for two reasons, both inside the
action:

1. tests.shared import fails. The "Run behave via qecore-headless"
   step copies tests/<suite> and tests/shared to the guest but not
   tests/__init__.py. Without it the guest's tests/ is only a
   namespace package, so an installed "tests" package on PYTHONPATH
   shadows tests.shared. Copy tests/__init__.py alongside the others,
   matching how .github/workflows/e2e.yml already does it.

2. qecore-headless writes /usr. GNOME OS ships a read-only /usr, and
   qecore-headless rewrites the GNOME Shell unit for --unsafe-mode.
   qecore's immutable-image fallback targets /etc/systemd/user/ but
   never creates that directory, so the rewrite fails and the run
   aborts. Pre-create /etc/systemd/user before qecore runs so the
   fallback lands instead of /usr.

Also read the QEMU pidfile with sudo (QEMU runs as root) to silence
the harmless "Permission denied" on cleanup.

- Add tests/__init__.py to the sparse-checkout list.
- Copy tests/__init__.py to the guest in the run step.
- sudo mkdir -p /etc/systemd/user before qecore-headless.
- sudo cat the QEMU pidfile.

Add tests/unit/test_gnome_e2e_action_summary.py asserting the copy,
sparse-checkout, systemd/user, and pidfile changes so they cannot
regress.

Closes projectbluefin#1144

Hive-Run: gh api -X POST projectbluefin/issues/1144/comments -f body='Canary dispatch from this fix branch: gh workflow run gnomeos-canary.yml --ref fix/gnome-e2e-canary-blockers. A maintainer must start it (the projectbluefin#908 lane is not merged into main yet).'
Hive-Plan: Rebased onto current projectbluefin/testsuite:main (projectbluefin#1139 added canary checkout support). Both blockers still present in .github/actions/gnome-e2e/action.yml; fixed there. Added a YAML-assertion regression test.
Hive-Spec: See PR body. Acceptance: gnomeos-canary.yml dispatched on the fix branch boots, runs the gnomeos-canary suite, and its "Gate suite has passing scenarios" step passes.
Signed-off-by: kylerankin <kylerankin@users.noreply.github.com>
kylerankin added a commit to kylerankin/testsuite that referenced this pull request Nov 11, 2026
The gnomeos-canary.yml projectbluefin#908 lane boots a GNOME OS VM and runs the
gnomeos-canary suite through .github/actions/gnome-e2e, but behave
aborts before any scenario runs for two reasons, both inside the
action:

1. tests.shared import fails. The "Run behave via qecore-headless"
   step copies tests/<suite> and tests/shared to the guest but not
   tests/__init__.py. Without it the guest's tests/ is only a
   namespace package, so an installed "tests" package on PYTHONPATH
   shadows tests.shared. Copy tests/__init__.py alongside the others,
   matching how .github/workflows/e2e.yml already does it.

2. qecore-headless writes /usr. GNOME OS ships a read-only /usr, and
   qecore-headless rewrites the GNOME Shell unit for --unsafe-mode.
   qecore's immutable-image fallback targets /etc/systemd/user/ but
   never creates that directory, so the rewrite fails and the run
   aborts. Pre-create /etc/systemd/user before qecore runs so the
   fallback lands instead of /usr.

Also read the QEMU pidfile with sudo (QEMU runs as root) to silence
the harmless "Permission denied" on cleanup.

- Add tests/__init__.py to the sparse-checkout list.
- Copy tests/__init__.py to the guest in the run step.
- sudo mkdir -p /etc/systemd/user before qecore-headless.
- sudo cat the QEMU pidfile.

Add tests/unit/test_gnome_e2e_action_summary.py asserting the copy,
sparse-checkout, systemd/user, and pidfile changes so they cannot
regress.

Closes projectbluefin#1144

Hive-Run: gh api -X POST projectbluefin/issues/1144/comments -f body='Canary dispatch from this fix branch: gh workflow run gnomeos-canary.yml --ref fix/gnome-e2e-canary-blockers. A maintainer must start it (the projectbluefin#908 lane is not merged into main yet).'
Hive-Plan: Rebased onto current projectbluefin/testsuite:main (projectbluefin#1139 added canary checkout support). Both blockers still present in .github/actions/gnome-e2e/action.yml; fixed there. Added a YAML-assertion regression test.
Hive-Spec: See PR body. Acceptance: gnomeos-canary.yml dispatched on the fix branch boots, runs the gnomeos-canary suite, and its "Gate suite has passing scenarios" step passes.
Signed-off-by: kylerankin <kylerankin@users.noreply.github.com>
hivecommons-hive Bot pushed a commit that referenced this pull request Nov 12, 2026
* ci: boot the suites a pull request affects before it merges

e2e.yml only runs when something calls it, and contributor agents may
not dispatch workflows, so a VM-related change got its first VM run
after merge (as a v1 hold) or when a maintainer started a manual run.
On 2026-11-10 every VM question (#827, #838, #908, #1138, #1139) waited
on a person starting a run.

Add "PR VM check": on every pull request, plan the PR's own changes with
the v1 gate's planner (vm_validation_plan.py + v1_gate.py) and, when a
VM is needed, boot those suites through e2e.yml on dakota:testing pinned
by digest, testing the merge commit. Docs-only and tooling PRs boot
nothing; a change the gate cannot boot (the gnome-e2e composite action)
gets a notice. Advisory, not required. This PR changes the workflow
itself, so it boots common and smoke as its own proof.

Assisted-by: Claude Opus 5.5 via Claude Code
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WcWm7qQa538Zjcuvv8jBtn
Signed-off-by: Danathar <Danathar@users.noreply.github.com>

* docs: say how to validate what the PR VM check cannot boot

The manual run calls e2e.yml, so it cannot cover a hold. Point
composite_action changes (the gnome-e2e action and the extension
validation gate it runs) at gnomeos-canary.yml on a branch at the PR's
head, and no_hosted_lane suites at their own hardware or a hand check.
Also drop "advisory": the hive merges no PR while any check is red.

Addresses the hive reviewer's docs-currency finding on #1142.

Assisted-by: Claude Opus 5.5 via Claude Code
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WcWm7qQa538Zjcuvv8jBtn
Signed-off-by: Danathar <Danathar@users.noreply.github.com>

---------

Signed-off-by: Danathar <Danathar@users.noreply.github.com>
Co-authored-by: Danathar <Danathar@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
kylerankin added a commit to kylerankin/testsuite that referenced this pull request Nov 12, 2026
The gnomeos-canary.yml projectbluefin#908 lane boots a GNOME OS VM and runs the
gnomeos-canary suite through .github/actions/gnome-e2e, but behave
aborts before any scenario runs for two reasons, both inside the
action:

1. tests.shared import fails. The "Run behave via qecore-headless"
   step copies tests/<suite> and tests/shared to the guest but not
   tests/__init__.py. Without it the guest's tests/ is only a
   namespace package, so an installed "tests" package on PYTHONPATH
   shadows tests.shared. Copy tests/__init__.py alongside the others,
   matching how .github/workflows/e2e.yml already does it.

2. qecore-headless writes /usr. GNOME OS ships a read-only /usr, and
   qecore-headless rewrites the GNOME Shell unit for --unsafe-mode.
   qecore's immutable-image fallback targets /etc/systemd/user/ but
   never creates that directory, so the rewrite fails and the run
   aborts. Pre-create /etc/systemd/user before qecore runs so the
   fallback lands instead of /usr.

Also read the QEMU pidfile with sudo (QEMU runs as root) to silence
the harmless "Permission denied" on cleanup.

- Add tests/__init__.py to the sparse-checkout list.
- Copy tests/__init__.py to the guest in the run step.
- sudo mkdir -p /etc/systemd/user before qecore-headless.
- sudo cat the QEMU pidfile.

Add tests/unit/test_gnome_e2e_action_summary.py asserting the copy,
sparse-checkout, systemd/user, and pidfile changes so they cannot
regress.

Closes projectbluefin#1144

Hive-Run: gh api -X POST projectbluefin/issues/1144/comments -f body='Canary dispatch from this fix branch: gh workflow run gnomeos-canary.yml --ref fix/gnome-e2e-canary-blockers. A maintainer must start it (the projectbluefin#908 lane is not merged into main yet).'
Hive-Plan: Rebased onto current projectbluefin/testsuite:main (projectbluefin#1139 added canary checkout support). Both blockers still present in .github/actions/gnome-e2e/action.yml; fixed there. Added a YAML-assertion regression test.
Hive-Spec: See PR body. Acceptance: gnomeos-canary.yml dispatched on the fix branch boots, runs the gnomeos-canary suite, and its "Gate suite has passing scenarios" step passes.
Signed-off-by: kylerankin <kylerankin@users.noreply.github.com>
Danathar added a commit that referenced this pull request Nov 13, 2026
…1181)

* fix(ci): gnome-e2e canary fix tests.shared import and read-only /usr

The gnomeos-canary.yml #908 lane boots a GNOME OS VM and runs the
gnomeos-canary suite through .github/actions/gnome-e2e, but behave
aborts before any scenario runs for two reasons, both inside the
action:

1. tests.shared import fails. The "Run behave via qecore-headless"
   step copies tests/<suite> and tests/shared to the guest but not
   tests/__init__.py. Without it the guest's tests/ is only a
   namespace package, so an installed "tests" package on PYTHONPATH
   shadows tests.shared. Copy tests/__init__.py alongside the others,
   matching how .github/workflows/e2e.yml already does it.

2. qecore-headless writes /usr. GNOME OS ships a read-only /usr, and
   qecore-headless rewrites the GNOME Shell unit for --unsafe-mode.
   qecore's immutable-image fallback targets /etc/systemd/user/ but
   never creates that directory, so the rewrite fails and the run
   aborts. Pre-create /etc/systemd/user before qecore runs so the
   fallback lands instead of /usr.

Also read the QEMU pidfile with sudo (QEMU runs as root) to silence
the harmless "Permission denied" on cleanup.

- Add tests/__init__.py to the sparse-checkout list.
- Copy tests/__init__.py to the guest in the run step.
- sudo mkdir -p /etc/systemd/user before qecore-headless.
- sudo cat the QEMU pidfile.

Add tests/unit/test_gnome_e2e_action_summary.py asserting the copy,
sparse-checkout, systemd/user, and pidfile changes so they cannot
regress.

Closes #1144

Hive-Run: gh api -X POST /issues/1144/comments -f body='Canary dispatch from this fix branch: gh workflow run gnomeos-canary.yml --ref fix/gnome-e2e-canary-blockers. A maintainer must start it (the #908 lane is not merged into main yet).'
Hive-Plan: Rebased onto current projectbluefin/testsuite:main (#1139 added canary checkout support). Both blockers still present in .github/actions/gnome-e2e/action.yml; fixed there. Added a YAML-assertion regression test.
Hive-Spec: See PR body. Acceptance: gnomeos-canary.yml dispatched on the fix branch boots, runs the gnomeos-canary suite, and its "Gate suite has passing scenarios" step passes.
Signed-off-by: kylerankin <kylerankin@users.noreply.github.com>

* fix(ci): read canary cleanup pidfile with sudo and correct canary docs

The review swarm for #1144 found the shutdown fix incomplete: the action's
boot message read the root-owned QEMU pidfile with sudo, but the workflow's
cleanup step still read it unprivileged and aborted on "Permission denied"
before signalling QEMU. Wrap the cleanup read in sudo too, and guard it in
tests so it cannot regress alongside the action's boot-message guard.

Also drop the obsolete "workflow must first exist on the default branch"
claim from the GNOME OS desktop canary reference; the workflow is already on
main and offers manual dispatch directly from any ref.

Closes #1144

Signed-off-by: kylerankin <kylerankin@users.noreply.github.com>

* fix(docs): name the correct GNOME OS canary mandatory gate

The gnomeos-desktop-canary reference described the mandatory result gate
only generically. Name `Gate extension-validation run` as the applicable
gate for the gnomeos-canary suite and note that `Gate suite has passing
scenarios` explicitly excludes it, matching the reviewer's docs-currency
finding on #1145 and the corrected PR Test plan.

Signed-off-by: kylerankin <kylerankin@users.noreply.github.com>

* fix(ci): run the GNOME OS canary's behave inside the live session

qecore-headless 4.19.3 clears its environment and rebuilds it from the first
gnome-session process's environ plus a short allowlist, then runs the user
script with `bash -c`. That environ has no WAYLAND_DISPLAY and no usable
session bus, so the canary's first step failed with "Missing Wayland
display" (#1144).

For the gnomeos-canary suite the action now runs behave through
scripts/gnomeos_canary_behave.py, called with the harness Python because
qecore requires an executable first word. After qecore's GDM restart the
wrapper reads WAYLAND_DISPLAY from the user manager (falling back to the
wayland-N socket), the AT-SPI address from org.a11y.Bus, sets
XDG_RUNTIME_DIR and the session bus, and execs behave. It waits up to 60s
for the restarted shell and execs regardless, so a missing value is still
reported by the canary's own first step. Other suites are unchanged.

Supersedes the environment.d drop-in approach in #1145, which feeds the
systemd user manager rather than the session leader qecore reads.

Assisted-by: Claude Opus 5.5 via Claude Code
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WcWm7qQa538Zjcuvv8jBtn
Signed-off-by: Danathar <Danathar@users.noreply.github.com>

* test(canary): record what the shell saw when the window wait times out

With the session environment fixed, the canary's first step passes and the
probe now times out in ponytail.waitFor. Record why: the shell's own window
list (org.gnome.Shell.Introspect.GetWindows, which Ponytail relies on and
which only answers under --unsafe-mode), the gnome-shell command line, and
whether the entry window process is still alive. Written to the evidence
JSON and the log, so a timeout explains itself.

Assisted-by: Claude Opus 5.5 via Claude Code
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WcWm7qQa538Zjcuvv8jBtn
Signed-off-by: Danathar <Danathar@users.noreply.github.com>

* fix(canary): focus the canary window with a real click before typing

Diagnostics from the last canary run showed the window mapped and listed by
org.gnome.Shell.Introspect.GetWindows ('GNOME OS input canary', 480x157),
the entry process alive and the shell in --unsafe-mode, but 'has-focus'
false. Ponytail's waitFor() only accepts a focused window, and on Wayland
the shell doesn't give focus to a window a background process maps without
an activation token, so waitFor timed out with the window on screen.

Wait for the window to be listed (focus not required), connect Ponytail to
it, then focus it with a virtual left click low in the window (inside the
entry, clear of any title bar) and wait for the shell to report focus
before typing. The text still arrives only as Ponytail keysyms and is read
back only over AT-SPI. Evidence records how focus was obtained; a window
that never maps or never takes focus fails with the shell's view attached.

Assisted-by: Claude Opus 5.5 via Claude Code
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WcWm7qQa538Zjcuvv8jBtn
Signed-off-by: Danathar <Danathar@users.noreply.github.com>

* fix(canary): press Escape to leave the startup overview before focusing

The click-to-focus run still reported the canary window unfocused. A GNOME
session starts in the Activities overview, which covers a newly mapped
window, so a click at the window's normal position doesn't reach it. Press
Escape through Ponytail first (as a user would), then fall back to the
click. Diagnostics now also record whether the overview is visible (a
read-only Main.overview.visible query; unsafe mode is on) and save a
screenshot of what was on screen when focus never arrived.

Assisted-by: Claude Opus 5.5 via Claude Code
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WcWm7qQa538Zjcuvv8jBtn
Signed-off-by: Danathar <Danathar@users.noreply.github.com>

* test(canary): record whether AT-SPI found the entry when typed text isn't read back

With Escape-first focus the window takes focus ('focus': 'escape'), but the
16 typed characters read back as ''. Record whether AT-SPI found the entry
at all, whether the window still had focus after typing, and a screenshot,
so the next run separates 'text never arrived' from 'entry not found'.

Assisted-by: Claude Opus 5.5 via Claude Code
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WcWm7qQa538Zjcuvv8jBtn
Signed-off-by: Danathar <Danathar@users.noreply.github.com>

* fix(canary): type the probe text with XKB keycodes

The last run had the canary window focused (overview left with Escape), the
entry found over AT-SPI and the text cursor in it (screenshot), yet the 16
characters sent as keysyms never arrived. Escape, a compositor action,
worked, so the loss is in keysym-to-key translation. Send physical keys
instead: Ponytail's GenerateKeycodeEvent with XKB codes (evdev + 8, which the
daemon converts) for the hex digits token_hex() produces. Evidence records
the input method.

Assisted-by: Claude Opus 5.5 via Claude Code
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WcWm7qQa538Zjcuvv8jBtn
Signed-off-by: Danathar <Danathar@users.noreply.github.com>

* fix(canary): type through a monitor connection, as dogtail does

Keysyms and keycodes sent through Ponytail's window-scoped connection never
reached the focused entry (entry found over AT-SPI, window focused, text
empty, in both runs). The dakota smoke tests that do type successfully go
through dogtail, which connects Ponytail to the monitor for keyboard input
(ponytail_helper.ponytail_check_connection). Do the same: after focusing
through the window connection, reconnect to the monitor and send keysyms.

Assisted-by: Claude Opus 5.5 via Claude Code
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WcWm7qQa538Zjcuvv8jBtn
Signed-off-by: Danathar <Danathar@users.noreply.github.com>

* fix(canary): bound Ponytail's connection waits and log what stalls

Run 43328291065 hit the step's 90 s limit after switching to the monitor
connection: Ponytail's client waits for Connected and Disconnected on a GLib
main loop with no timeout, so nothing said which call stuck. The run before
it had the window focused, the caret in the entry and no text arrive.

Call the daemon's D-Bus methods directly and wait at most 15 s for each
signal, record the stage reached in desktop-canary.json, start the daemon
with G_MESSAGES_DEBUG=all so the guest journal shows each injected event,
and record the input-method state when typed text isn't read back.

Assisted-by: Claude Opus 5.5 via Claude Code
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WcWm7qQa538Zjcuvv8jBtn
Signed-off-by: Danathar <Danathar@users.noreply.github.com>

* fix(canary): type keycodes through the window connection; test the IM path

Run 43332133969 stalled reconnecting Ponytail to the monitor. Its journal
also explains the 'Unknown/invalid virtual device key 0x0 pressed' lines:
Ponytail 0.0.11 sends a bogus key 0 before every unmodified keysym (its
modifier guard tests key_keycode), then the real key, so Escape does work.
The keycode run before it reached the shell with no errors, the entry had
focus and still no text arrived, which points at the input-method path.

Type XKB keycodes through the window connection again. When the text isn't
read back, record the input-method state, reopen the entry with
GTK_IM_MODULE=gtk-im-context-simple, type again and record the result as
evidence; the canary still fails. The step allows 150 s for that retry.

Assisted-by: Claude Opus 5.5 via Claude Code
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WcWm7qQa538Zjcuvv8jBtn
Signed-off-by: Danathar <Danathar@users.noreply.github.com>

* fix(canary): type into an entry that uses GTK's simple input context

Run 43334774167 settled where the keys went. With IBus running
(ibus-daemon --panel disable, US layout), the entry focused and the keys
reaching the shell without errors, no text arrived. Reopened with
GTK_IM_MODULE=gtk-im-context-simple, the exact random value arrived and
was read back over AT-SPI.

Make the simple context the entry's only input path (no fallback, no second
attempt) and record the IBus state as evidence. The canary proves Ponytail,
Mutter, GTK and AT-SPI; the IBus path is tracked separately. Drop the
diagnostic retry and return the step bound to 90 s.

Assisted-by: Claude Opus 5.5 via Claude Code
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WcWm7qQa538Zjcuvv8jBtn
Signed-off-by: Danathar <Danathar@users.noreply.github.com>

* fix(canary): type into a window mapped after the overview is gone

Correction to df6be4d: IBus was not the cause. Run 43337304054 ran the
entry with GTK's simple input context and still read back nothing. What
differed in the one run where the text arrived (43334774167) was the
window: its retry opened a second window after the overview was gone, which
took focus by itself, and the keys arrived exactly. Every window mapped
under the startup overview and focused by pressing Escape lost them.

Open a throwaway window first, press Escape to leave the overview, close
it, and type once into a fresh canary window on the plain desktop. The
entry goes back to the session's normal input method; the input-method
state stays in the evidence. The doc no longer blames IBus.

Assisted-by: Claude Opus 5.5 via Claude Code
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01WcWm7qQa538Zjcuvv8jBtn
Signed-off-by: Danathar <Danathar@users.noreply.github.com>

---------

Signed-off-by: kylerankin <kylerankin@users.noreply.github.com>
Signed-off-by: Danathar <Danathar@users.noreply.github.com>
Co-authored-by: kylerankin <kylerankin@users.noreply.github.com>
Co-authored-by: Danathar <Danathar@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

author/clanker A contributor's own hive authored this PR under their login dco All commits are DCO signed-off

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants