CNA-BUG-012: GraphicsDeviceCapabilityTest's catch-all expectations are still wrong for STUB and for Linux SDL_GPU builds without libshaderc

CNA snapshot c1c316b9  ·  Known Issues › Current bugs  ·  source links pinned to c1c316b9

ExpectedCapabilities() now has an SDL_GPU arm, but on Linux it keeps the catch-all custom-effects expectation true although an SDL_GPU build without libshaderc answers false, and STUB, which answers false to every capability, still takes the all-true default.

Identifier

CNA-BUG-012

Category

Bug

Subsystem

Testing & evidence

Status

Narrowed (partially fixed; describes only what survives)

Verified against

CNA c1c316b9 (c1c316b9c7a846ce8002809c151fcd1af14942c9)

Severity

Low (a triage suggestion, not a project priority)

Evidence basis

Source-verified: read at TARGET, not executed

Tests touching this area

Yes: see Current tests

Affected contract

GraphicsDeviceCapabilityTest.SupportsOcclusionQuery, SupportsCustomEffects and SupportsMultipleRenderTargets (ExpectedCapabilities())

Expected behaviour

A renderer that truthfully reports a capability as false should pass the shared capability suite. ExpectedCapabilities() in GraphicsDeviceCapabilityTests.cpp exists to state each renderer's accepted boundary, so every renderer whose answers differ from the all-true catch-all needs an arm that says so.

Actual behaviour at TARGET

ExpectedCapabilities() has explicit arms for FNA3D, SOFTWARE, SDL_RENDERER and SDL_GPU; every other renderer takes default: return {true, true, true}. The SDL_GPU arm expects occlusion queries false everywhere and custom effects false on Apple, Windows and Emscripten, but on other hosts it keeps the default true, because whether the build found libshaderc is visible only to the renderer target (CNA_SDL_GPU_SHADER_EFFECTS is a private definition) and SdlGpuRenderer::SupportsCapability(CustomEffects) answers false without it; SupportsCustomEffects therefore fails in a Linux SDL_GPU build without libshaderc. STUB answers false to every capability (StubRenderer::SupportsCapability) and still takes the catch-all, so SupportsOcclusionQuery, SupportsCustomEffects and the HiDef half of SupportsMultipleRenderTargets expect true on STUB.

Source locations

• modules/graphics/tests/Microsoft/Xna/Framework/Graphics/GraphicsDeviceCapabilityTests.cpp — ExpectedCapabilities(): the SDL_GPU arm and the default arm; SupportsOcclusionQuery, SupportsCustomEffects, SupportsMultipleRenderTargets
• modules/renderers/sdl-gpu/CMakeLists.txt — CNA_SDL_GPU_SHADER_EFFECTS: a private definition, only with a target-native libshaderc
• modules/renderers/sdl-gpu/src/SdlGpuRenderer.cpp — SdlGpuRenderer::SupportsCapability: CustomEffects follows CNA_SDL_GPU_SHADER_EFFECTS
• modules/renderers/stub/include/CNA/Internal/Renderers/Stub/StubRenderer.hpp — StubRenderer::SupportsCapability returns false for every capability

Evidence

Checked by reading at c1c316b9; not executed. CNA's records show the catch-all failing five cases under STUB (integration/BATCH_1_STABILIZATION.md); no CNA record of GraphicsDeviceCapabilityTest.* in a Linux SDL_GPU build without libshaderc was found.

Narrowed at c1c316b9: bb9250261 (AM4-070) rewrote sdlgpu_occlusionquery_limitation_test.cpp to assert the public constructor's NotSupportedException and the renderer's exact message at its own query factory, so that test no longer contradicts the renderer; e5a943e0d (AM4-175) gave ExpectedCapabilities() its SDL_GPU arm (occlusion queries false everywhere, custom effects false on Apple, Windows and Emscripten); and METAL now answers true for all three capabilities (MetalPolicy.hpp), so the catch-all describes it. Read at c1c316b9; not executed.

Focused reproduction

# Not executed for this entry. A Linux SDL_GPU build with tests and without libshaderc, or a STUB build with tests:
ctest --test-dir <build> -R 'GraphicsDeviceCapabilityTest\.Supports(CustomEffects|OcclusionQuery|MultipleRenderTargets)$' --output-on-failure

Current tests

GraphicsDeviceCapabilityTests.cpp (per-renderer expectations; its MRT case also compares the capability with a real two-target bind) and sdlgpu_occlusionquery_limitation_test.cpp (now consistent with the renderer). Nothing checks that ExpectedCapabilities() has an arm for every renderer whose answers differ from the catch-all.

Regression test

Give ExpectedCapabilities() an explicit STUB arm (all false), and let the SDL_GPU arm see whether the shader-effect route was built (a public compile definition, or ask the renderer's custom-effect intake directly). A structural guard in the style of TheMultipleRenderTargetCapabilityMatchesWhatBindingActuallyDoes for CustomEffects would catch the next mismatch.

Blast radius

Test evidence only; renderer behaviour is correct. STUB runs of the shared capability suite and Linux SDL_GPU runs without libshaderc report failures that are not renderer defects, which inflates their failure counts and can hide real regressions among them.

Workaround

When triaging such a run, treat these failures as test defects.