[Visual Test] Label preflight skipped screenshot capture on run #21

[MichaelFisher1997]

Workflow run

https://github.com/OpenStaticFish/ZigCraft/actions/runs/29395853245

Scheduled visual-test run #21 on commit d0e348c (dev).

Failure evidence

build-output.log is absent from the workspace. This is not an engine logging failure: current check-step metadata shows that the game command was never started.

Step	Result
Ensure visual-test label exists	failure (07:02:39-07:02:40 UTC)
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 relevant duplicate-label failure from this unchanged command path is:

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

The current action log could not be downloaded through REST because the workflow token is rate-limited, but GraphQL check-step metadata confirms the same step boundary, and the repository state still reproduces the exact guard condition: gh label list defaults to 30 labels, the repository has 41, visual-test is position 21, and run-visual-test is position 41. The identical error above was captured from the same step and command in #916.

The only local runtime log is weston.log, showing that Weston initialized successfully and was terminated immediately after the preflight failed:

[07:02:38.815] Command line: weston --socket=headless --backend=headless-backend.so --width=1280 --height=720
[07:02:38.820] Output 'headless' enabled with head(s) headless
[07:02:40.312] caught signal 15

There are therefore no Vulkan instance/device errors, swapchain errors, shader failures, screenshot staging/fence/write errors, panics, or segfaults in this run: none of those paths executed.

Root cause

The Ensure visual-test label exists shell step in .github/workflows/visual-test.yml:55-69 checks label existence with:

gh label list --json name --jq '.[].name'

gh label list has a default --limit 30. The repository currently has 41 labels and run-visual-test is item 41 in the returned ordering. The guard at .github/workflows/visual-test.yml:62 therefore incorrectly treats the existing label as missing and invokes gh label create "run-visual-test" at line 63. gh label create returns non-zero because that label already exists, aborting the step and skipping all subsequent non-always() steps.

Because .github/actions/run-with-log/action.yml:33-34 is only entered by the skipped screenshot step, its initial tee never creates build-output.log.

This is a recurrence of #916, #931, and #935; the recommended pagination/idempotency fix has still not landed.

Source trace

The failure originates in the workflow shell guard, not a Zig function:

• .github/workflows/visual-test.yml:55-69 — failing Ensure visual-test label exists step.
• .github/workflows/visual-test.yml:62-65 — false negative followed by duplicate gh label create.
• .github/actions/run-with-log/action.yml:33-34 — would create build-output.log, but was skipped.

The requested engine paths were inspected and are downstream of the skipped command:

• src/game/app.zig:243-251 loads the menu in screenshot mode; App.runSingleFrame requests capture at src/game/app.zig:467-520.
• modules/game-ui/src/screens/home.zig:33-41 initializes HomeScreen and its preview.
• modules/engine-graphics/src/vulkan_swapchain.zig:120-167 creates the offscreen transfer-source image.
• modules/engine-graphics/src/rhi_vulkan.zig:409-413 forwards capture to screenshot.requestCapture.
• modules/engine-graphics/src/vulkan/screenshot.zig:23-75 allocates the staging capture and :188-206 waits for completion and writes the image.
• modules/engine-graphics/src/vulkan/rhi_init_deinit.zig:26-43 initializes the Vulkan device, frame manager, swapchain, and pipelines.

None of these functions ran in workflow run #21.

Suggested fix

Minimal fix: fetch enough labels in both guards. gh label list supports --limit, not --paginate:

- 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

More robustly, make label maintenance idempotent and non-blocking so API quota or label metadata cannot disable visual testing:

- name: Ensure visual-test labels exist
  continue-on-error: true
  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

The label setup could also be removed from this workflow entirely because both labels already exist. After fixing the preflight, rerun the workflow to obtain the first meaningful Vulkan/screenshot result.

[github-actions]

I've validated the issue and confirmed the data. Posting a comment:

---

Classification: Bug (CI infrastructure regression) — not an engine bug.

Validation: Confirmed locally:

• gh label list --limit 100 returns 41 labels; run-visual-test is at position 41 and visual-test at position 21.
• .github/workflows/visual-test.yml:57,62 both call gh label list --json name --jq '.[].name' with no --limit, so they silently cap at 30 and miss run-visual-test.
• gh label create "run-visual-test" then exits non-zero (already exists), aborting the step before any screenshot/run path executes. build-output.log is never written because .github/actions/run-with-log/action.yml:33-34 only runs from the skipped Run menu screenshot capture step.
• weston.log only shows Weston startup + signal 15 — no Vulkan/swapchain/screenshot code ran, so the engine source traces listed in the issue body are not implicated in run opencode test #21.

Duplicates: This is the 4th recurrence of the same root cause — #916, #931, #935, and now #937 — and all three predecessors remain OPEN. No merged PR addresses the pagination/idempotency gap (the only merged PR returned for related searches is #262 about LPV lighting, unrelated).

Scope: Tight, single-file change. No engine modules touched.

Recommended fix (minimal, sufficient):

- name: Ensure visual-test labels exist
  run: |
    for L in visual-test run-visual-test; do
      if ! gh label list --limit 100 --json name --jq '.[].name' | grep -qx "$L"; then
        gh label create "$L" --description "$(case $L in visual-test) echo 'Issues from automated visual regression tests' ;; run-visual-test) echo 'Run deterministic visual regression workflow on a PR' ;; esac)" --color "E06C75"
      fi
    done
  env:
    GH_TOKEN: ${{ secrets.OPENCODE_PAT }}

Stronger alternative (idempotent + non-fatal, recommended since this is the 4th recurrence): drop the guard entirely and use gh label create ... --force with continue-on-error: true. Both labels already exist, so label setup can also be removed from this workflow and moved to a one-off bootstrap step.

Action items:

1. Land the workflow fix (single PR, target dev).
2. Close [Visual Test] Label preflight fails: gh label list only returns first 30 labels, run-visual-test is on page 2 #916, [Visual Test] Label preflight regressed: gh label list pagination skips run-visual-test on run #19 #931, [Visual Test] Label preflight regressed again: unpaginated gh label list still misses run-visual-test on run #20 #935, [Visual Test] Label preflight skipped screenshot capture on run #21 #937 as duplicates of the fix PR.
3. Rerun the visual-test workflow to obtain the first real Lavapipe + screenshot evidence — anything else has been masked by this preflight.

No reproduction steps, crash logs, or engine changes needed from the reporter; the issue is self-contained and reproducible from the workflow file alone.

opencode session  |  github run

[github-actions]

This issue has had no activity for 60 days. It will be closed in 14 days unless there is new activity or an exemption label is added.