[Visual Test] Label preflight prevents screenshot capture on run #26

[MichaelFisher1997]

Workflow run

https://github.com/OpenStaticFish/ZigCraft/actions/runs/29478947848

Scheduled visual-test run #26 on commit 8e8eee20390193f2d4d8ecfb62f7144fdd54f8de.

Failure evidence

The requested build-output.log is not present in the workspace. The GitHub run is still in progress in the failure-diagnosis step, so the raw job log is not yet downloadable; however, the job step metadata identifies the failure boundary:

Ensure visual-test label exists  failure
Setup Lavapipe Vulkan            skipped
Run menu screenshot capture      skipped
Check screenshot exists          success (screenshot_exists=false)
Check build log exists           success (exists=false)
Upload build log artifact        skipped

The same unchanged shell path has previously emitted this duplicate-label error:

label with name "run-visual-test" already exists; use `--force` to update its color and description
##[error]Process completed with exit code 1.

The local Weston log confirms the compositor started normally and was stopped by cleanup after the preflight failure:

[07:12:39.203] Command line: weston --socket=headless --backend=headless-backend.so --width=1280 --height=720
[07:12:39.209] Output 'headless' enabled with head(s) headless
[07:12:42.484] caught signal 15

There are no Vulkan instance/device, swapchain, shader, staging-buffer, fence, image-write, panic, or segfault errors for this run because the game command never started.

Diagnosis

The failure originates in the Ensure visual-test label exists shell step in .github/workflows/visual-test.yml:55-69, specifically the second guard at lines 62-65:

if ! gh label list --json name --jq '.[].name' | grep -q '^run-visual-test$'; then
  gh label create "run-visual-test" \
    --description "Run deterministic visual regression workflow on a PR" \
    --color "E06C75"
fi

The repository currently has 41 labels. Running the exact unbounded check returns only 30 labels and does not include the existing run-visual-test label, while the same command with --limit 100 returns all 41 and does include it. The guard therefore treats an existing label as absent and gh label create exits non-zero on the duplicate. Since the step fails, all subsequent non-always() steps—including Lavapipe setup and screenshot capture—are skipped.

build-output.log is correctly absent: .github/actions/run-with-log/action.yml:22-34 creates it only when the skipped screenshot step runs. This is a CI workflow failure, not an engine failure.

The inspected engine paths are downstream and were not exercised:

• modules/engine-graphics/src/vulkan/rhi_init_deinit.zig:26-43 initializes Vulkan and the swapchain.
• modules/engine-graphics/src/vulkan_swapchain.zig:127-174 creates the headless offscreen transfer-source image.
• src/game/app.zig:268-276 selects the menu screenshot mode, and src/game/app.zig:492-565 counts frames and requests capture.
• modules/game-ui/src/screens/home.zig:33-41 initializes the HomeScreen preview.
• modules/engine-graphics/src/rhi_vulkan.zig:415-420 forwards capture to the Vulkan screenshot implementation.
• modules/engine-graphics/src/vulkan/screenshot.zig:24-86 validates and queues the readback.

None of these functions ran in this workflow attempt.

There is also a stale prompt/workflow inconsistency: .github/prompts/visual-test-diagnose.md:7-9 describes screenshot.ppm, while .github/workflows/visual-test.yml:73-81 currently requests screenshot.png. The screenshot implementation accepts PNG but not PPM (modules/engine-graphics/src/vulkan/screenshot.zig:39-40 and :262-267); that is a separate issue and was not reached here.

Suggested fix

Fetch enough labels before testing for existence:

- name: Ensure visual-test label exists
  run: |
    if ! gh label list --limit 100 --json name --jq '.[].name' | grep -q '^visual-test$'; then
      gh label create "visual-test" \
        --description "Issues from automated visual regression tests" \
        --color "E06C75"
    fi
    if ! gh label list --limit 100 --json name --jq '.[].name' | grep -q '^run-visual-test$'; then
      gh label create "run-visual-test" \
        --description "Run deterministic visual regression workflow on a PR" \
        --color "E06C75"
    fi

Alternatively, make label maintenance idempotent with gh label create ... --force, or remove this setup step because both labels already exist. After fixing the preflight, rerun the workflow to obtain a meaningful Vulkan/screenshot result.

[github-actions]

Read the visual-test.yml workflow and visual-test-diagnose.md prompt to validate the reporter's claim that gh label list defaults to 30 results, confirming the preflight false-negative on run-visual-test.

opencode session  |  github run