Skip to content

armgit / kiln

Public Imported from GitHub
0

Build and revise procedural 3D assets with your coding agent. Local MCP server, CLI and TypeScript engine with rendered review, editable source and GLB export.

ci: run the render service's tests, which ran nowhere (#87)

Matthew Kissinger4e1cca0d89bb
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 2bee96f0..7e7259ed 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -139,6 +139,38 @@ jobs:
node -p "'node ' + process.version"
node -e "console.log('sharp versions:', JSON.stringify(require('sharp').versions))"
+ render-service-checks:
+ name: render service tests
+ runs-on: ubuntu-latest
+ timeout-minutes: 10
+ # `render-service/` shipped with 37 tests that ran in no workflow. It is a separate
+ # npm project with a native dependency, which is presumably why -- but the tests do
+ # not need it. All 37 are pure: framing arithmetic, PNG readback packing, cache
+ # identity, contract and preset validation. Not one acquires a device.
+ #
+ # So `--ignore-scripts`, which skips the `webgpu` package's native Dawn build and
+ # leaves the package directory in place for the one module that only resolves its
+ # path. Verified by running the suite against exactly this install, with no binding
+ # built: 37 pass, 0 fail. The GPU smoke and the two conformance runs stay manual,
+ # because those genuinely need a device.
+ steps:
+ - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
+
+ - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4
+ with:
+ node-version: 22.23.2
+
+ - name: Setup npm
+ run: npm install --global [email protected]
+
+ - name: Install the render service
+ run: npm ci --ignore-scripts
+ working-directory: render-service
+
+ - name: Test the render service
+ run: npm test
+ working-directory: render-service
+
package-candidate:
name: build portable Node package
runs-on: ubuntu-22.04
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 16f9e1c3..0620676f 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -31,8 +31,15 @@ GitHub. The package is not published on the npm registry.
and the only shadowing on that surface is drei's `ContactShadows` — its own depth pass to
a texture, which never consults `shadowMap.type`. Between the two paths, r186's PCFSoft
removal changes nothing anyone can see in this repository.
-- Worth knowing while reading this: `render-service` runs in **no** CI workflow. Its 37
- tests and its GPU conformance runs are manual only.
+- **`render-service` is in CI now**, which it never was. Thirty-seven tests guarding a
+ shipped subsystem ran on nobody's machine but a maintainer's — including, until this
+ commit, the change above it. It is a separate npm project with a native dependency,
+ which is presumably the reason; but the tests do not need it. All 37 are pure: framing
+ arithmetic, PNG readback packing, cache identity, contract and preset validation, and
+ not one acquires a device. So the job installs with `--ignore-scripts`, skipping the
+ `webgpu` package's Dawn build. Verified against exactly that install with no binding
+ built: 37 pass, 0 fail. The GPU smoke and the two conformance runs stay manual, because
+ those genuinely need a device.
## The lint baseline is zero, and stays zero — 2026-09-12
diff --git a/docs/plans/repo-size-and-r2-migration-2026-09-10.md b/docs/plans/repo-size-and-r2-migration-2026-09-10.md
index 7b401a1e..14157865 100644
--- a/docs/plans/repo-size-and-r2-migration-2026-09-10.md
+++ b/docs/plans/repo-size-and-r2-migration-2026-09-10.md
@@ -1577,6 +1577,8 @@ than for more care.
| 13.5 | Clear the one substantive lint finding: an unused import in `src/__tests__/optimize.test.ts`. Everything else in the 14/11 baseline is `useTemplate` and `useOptionalChain` style | **Done 2026-09-12, wider than written.** The 14/11 split was not all style: `isFinite` in the render inspector was the coercing global, `forEach` in the Bradley-Terry fit returned a value its callback discards, and three `any` bounds in `buildSandboxGlobals` were guarded by `eslint-disable` comments in a repository that lints with Biome and has no ESLint config -- suppressing nothing while reading as accepted. All 25 findings are fixed and the tree reports zero. `lint` now runs `--error-on-warnings`, and since that does not reach Biome's info severity the seven rules involved carry an explicit `error`; each was verified by reintroducing its defect. A baseline kept in prose is one the next finding arrives invisible against, which is why this went past the single unused import |
| 13.6 | Decide shadows by comparison, not assumption. r186 deleted the `PCFSoftShadowMap` implementation; `render-service` now names `PCFShadowMap`, and `VSMShadowMap` is the soft filter that survives. Render gallery presets under both and choose by eye, because this is the only user-visible regression in the r186 upgrade and it lands on marketing surface | **Premise was wrong; the real defect is fixed 2026-09-12.** **No shipped preset enables shadows** -- `neutral-studio-v1` sets `shadows.enabled: false` deliberately, `gallery-studio-v1` spreads it, and every light has `castsShadow: false`. So r186 is not a visible regression here and there is nothing to compare by eye: both filters render the same image, which has no shadow in it. Yesterday's `PCFShadowMap` edit corrected a line that never executes, and its claim about keeping the log clean was overstated -- the assignment is inside `if (preset.shadows.enabled)`, so the deprecation warning never fired either. What WAS wrong: `shadows.type` described a filter and selected nothing, and after r186 the one string the validator accepted was the one filter that no longer exists. It now selects from `basic`/`pcf`/`vsm`, the renderer maps each to its three constant, and the renderer checks at load that the map covers the schema's list -- verified by adding `pcf-soft` back and watching the service refuse to start. The gallery is unaffected for a second, independent reason: neither `<Canvas>` in `site/` sets r3f's `shadows` prop, so `shadowMap.enabled` is false there as well, and the only shadowing on that surface is drei's `ContactShadows`, which runs its own depth pass to a texture and never reads `shadowMap.type`. Across both paths r186's PCFSoft removal changes nothing anyone can see. **Open, and the founder's call:** whether a gallery preset should cast shadows at all -- a visual decision, not a repair, and the only thing a by-eye comparison would actually be for |
+| 13.7 | `render-service` in CI. Found while fixing 13.6: a shipped subsystem with 37 tests ran in no workflow, so the 13.6 change itself would have had no CI coverage. Every one of those tests is pure -- framing arithmetic, PNG readback packing, cache identity, contract and preset validation -- and none acquires a device, so the job installs with `--ignore-scripts` and skips the `webgpu` package's native Dawn build | **Done 2026-09-12.** Verified against exactly that install with no binding built: 37 pass, 0 fail. The GPU smoke and the two conformance runs stay manual, because those do need a device |
+
### Explicitly not in this phase
- **7.12 / SEP-2640.** Still `In Review` on the Skills Over MCP working group's own