[Visual Test] Label bootstrap failure prevents menu screenshot run

[MichaelFisher1997]

Workflow

https://github.com/OpenStaticFish/ZigCraft/actions/runs/29634181508

Observed output

The checked-out workspace contains no build-output.log; the game command was never started. The failed job step status was:

Ensure visual-test label exists    completed  failure
Setup Lavapipe Vulkan              completed  skipped
Run menu screenshot capture        completed  skipped
Check screenshot exists             completed  success
Check build log exists              completed  success

The only available log retrieval response while the run was still active was:

run 29634181508 is still in progress; logs will be available when it is complete

weston.log contains only normal compositor shutdown:

[06:36:17.128] caught signal 15

There are no Vulkan instance/device, swapchain, shader, staging-buffer, fence, PPM/PNG, panic, or segfault errors from this run because the label bootstrap stopped execution before Lavapipe setup.

Diagnosis

The failure originates in .github/workflows/visual-test.yml:55-68, in the Ensure visual-test label exists step, not in the renderer. Both labels already exist in the repository. However, gh label list uses its default limit of 30 labels. The current label ordering places visual-test at position 21 and run-visual-test at position 41. The first probe can see visual-test, while the second probe cannot see run-visual-test, so the workflow attempts to create an already-existing label and the step fails.

The job declares issues: write at .github/workflows/visual-test.yml:20-24 and supplies GH_TOKEN at .github/workflows/visual-test.yml:67-68, so this is a label-discovery/idempotency bug rather than a Vulkan or Lavapipe environment failure.

The requested runtime paths were not exercised, but the source confirms the expected flow:

• src/game/app.zig:260-276 selects screenshot mode and loads createHomeScreen; src/game/app.zig:495-587 applies ZIGCRAFT_SMOKE_FRAMES and requests the capture before ending the frame.
• The actual HomeScreen implementation is modules/game-ui/src/screens/home.zig:33-41 and draws the preview/menu at modules/game-ui/src/screens/home.zig:62-98.
• RHI initialization enters modules/engine-graphics/src/rhi_vulkan.zig:36-38, then modules/engine-graphics/src/vulkan/rhi_init_deinit.zig:26-41; screenshot requests are wired by modules/engine-graphics/src/rhi_vulkan.zig:415-419.
• The headless image path is modules/engine-graphics/src/vulkan_swapchain.zig:127-174, where the offscreen image uses COLOR_ATTACHMENT | TRANSFER_SRC. Capture recording/completion is in modules/engine-graphics/src/vulkan/screenshot.zig:24-86 and :91-222.

One additional configuration mismatch is visible but was not reached: the checked-in workflow invokes -Dscreenshot-path=screenshot.png at .github/workflows/visual-test.yml:73-87, while a .ppm path would be rejected by modules/engine-graphics/src/vulkan/screenshot.zig:39-40 and :262-267, which currently supports PNG/JPEG/GIF/WebP only.

Suggested fix

Make label setup independent of pagination and safe when labels already exist. The simplest fix is to use --force and remove the list/grep probes:

      - name: Ensure visual-test label exists
        run: |
          gh label create "visual-test" \
            --description "Issues from automated visual regression tests" \
            --color "E06C75" \
            --force
          gh label create "run-visual-test" \
            --description "Run deterministic visual regression workflow on a PR" \
            --color "E06C75" \
            --force
        env:
          GH_TOKEN: ${{ secrets.OPENCODE_PAT }}

Alternatively, use gh label view <name> for existence checks or pass an explicit --limit greater than the repository label count, but --force makes the setup reliably idempotent as labels grow.

[github-actions]

Classification: Bug — CI workflow infrastructure, not a Vulkan or renderer failure. No additional reproduction data is needed; the run metadata and checked-in workflow are sufficient.

Validation: Confirmed against the repository: gh label list returns 30 labels by default, while the repository has 41; visual-test is item 21 and run-visual-test is item 41. The unbounded probes at .github/workflows/visual-test.yml:57 and :62 therefore produce a false negative for run-visual-test, and the duplicate gh label create fails. The run metadata confirms Lavapipe and screenshot capture were skipped. build-output.log is correctly absent because .github/actions/run-with-log/action.yml:33-34 is only reached by the skipped capture step.

Duplicates / scope: This is a recurrence of the same root cause reported in #916, #931, #935, and #942; all remain open. No open PR currently addresses the label bootstrap. The fix is narrow and should be one CI-only change, not split across engine subsystems.

Recommended action: Make label setup idempotent and independent of pagination, preferably replacing the list/grep guards with gh label create ... --force for both labels, or use direct gh label view <name> checks. Then rerun the visual workflow; only that rerun can provide meaningful Lavapipe/screenshot evidence. The stale .github/prompts/visual-test-diagnose.md:7-9 PPM command is a separate documentation/configuration follow-up—the checked-in workflow currently requests PNG, which the implementation supports.

[github-actions]

Posted the analysis comment: #944 (comment)

opencode session  |  github run