CNA-VGAP-030: Model drawing on SOFTWARE, SDL_GPU, WEBGPU, FNA3D and DIRECTX9 is checked in CI at most at the stride draw boundary; the shared Model programs and the whole-corpus L7 rung run in CI for OPENGLES3 only

CNA snapshot 3b43bd86  ·  Known Issues › Verification gaps  ·  source links pinned to 3b43bd86

The EasyGL Model draw, hierarchy, two-mesh, skinned-playback and .cnj reader programs run verbatim as CTests on SOFTWARE and SDL_GPU, but no workflow runs them; the SOFTWARE whole-corpus L7 campaign is a committed report from an earlier commit.

Identifier

CNA-VGAP-030

Category

Verification gap

Subsystem

Testing & evidence

Status

Open

Verified against

CNA 3b43bd86 (3b43bd8687ed46523d873f9bec4b32bd8481dcb5)

Evidence basis

Source-verified: read at TARGET, not executed

Tests touching this area

Yes: see Current tests

Affected contract

ModelMesh::Draw / Model::Draw on each 3D renderer; the glTF L7 image comparison

Expected behaviour

A renderer that draws Model graphs should have repeatable evidence that it does so correctly, ideally in CI.

Actual behaviour at TARGET

The easygl_model_* sources are compiled unchanged by CMakeLists.txt (Software_ModelDraw_RedQuad and Software_Model_*) and CMakeLists.txt (SdlGpu_EasyGLOracle_model_*); HEADLESS has a Model::Draw check in headless_effects_test.cpp. No workflow runs any of these non-EasyGL registrations: gltf-renderer-stride-ci.yml configures SOFTWARE, SDL_GPU and WEBGPU but runs only ctest -L gltf-conformance, and platform-ci.yml filters ctest to platform tests. What CI does run on those renderers is RendererStrideConformance (a real Model::Draw plus Present per glTF stride that passes on no throw or a named refusal, with no pixel check) and the L0-L6 ladder; FNA3D and DIRECTX9 are configured by no automatic workflow. The l7-corpus job runs the whole-corpus image rung for OPENGLES3 only. The SOFTWARE whole-corpus result is the committed report gltf-l7-software-corpus-report.json, captured on 2026-08-18 at CNA commit b6c180fd.

Source locations

• .github/workflows/gltf-renderer-stride-ci.yml — conformance matrix (ctest -L gltf-conformance) and the OPENGLES3-only l7-corpus job
• docs/gltf-l7-software-corpus-report.json — capturedOn 2026-08-18, cnaCommit b6c180fd
• modules/renderers/sdl-gpu/examples/CMakeLists.txt — SdlGpu_EasyGLOracle_model_* registrations
• modules/renderers/software/examples/CMakeLists.txt — Software_Model_* registrations
• modules/renderers/easygl/examples/easygl_model_draw_test.cpp — one of the shared Model programs

Evidence

Checked by reading each renderer's CMake registrations of the shared Model programs, every workflow's ctest selection and the report header at 3b43bd86; not executed. The report is CNA's own record for its commit, not for 3b43bd86.

Re-verified at c6dec91f: the renderers retired on 2026-10-13 and 2026-10-14 took their Model registrations and their two committed L7 reports with them; the lavapipe cells of the glTF-conformance workflow moved to SDL_GPU and WEBGPU, which run the stride boundary only.

Focused reproduction

No focused reproduction is known. Nothing has been invented here; the evidence above is what exists.

Current tests

L0–L6 oracles and the stride test run per renderer in CI; the EasyGL Model programs run where EasyGL is built.

Regression test

A CI job that runs the SdlGpu_EasyGLOracle_model_* and Software_Model_* CTests (lavapipe serves SDL_GPU) and reruns the SOFTWARE L7 rung against its committed goldens.

Blast radius

Evidence only: regressions in Model drawing on SOFTWARE, SDL_GPU, WEBGPU, FNA3D and DIRECTX9 would not be caught by CI image comparison.

Workaround

No workaround is known.