CNA-VGAP-011: No test exercises PlatformSelection.cmake's reserved-identifier refusal paths

CNA snapshot c1c316b9  ·  Known Issues › Verification gaps  ·  source links pinned to c1c316b9

PlatformSelection.cmake fails configure for SDL12, EMSCRIPTEN and TERMINAL on Windows, but no test asserts those refusals; the one platform-selection test, CnaPlatformSelection_Retired, covers only the retired and unknown names.

Identifier

CNA-VGAP-011

Category

Verification gap

Subsystem

Build & CI

Status

Narrowed (partially fixed; describes only what survives)

Verified against

CNA c1c316b9 (c1c316b9c7a846ce8002809c151fcd1af14942c9)

Evidence basis

Source-verified: read at TARGET, not executed

Tests touching this area

None

Affected contract

cmake/PlatformSelection.cmake configure-time refusals for reserved CNA_PLATFORM values

Expected behaviour

PlatformSelection.cmake makes these refusals a deliberate contract: it says the reserved-name failure is “a hard error on purpose”, because “falling back to SDL3 would build something other than what you asked for.” A contract that regresses silently -- a reserved name that starts to fall through to SDL3 -- deserves a test, the way the retired names now have one.

Actual behaviour at TARGET

The only platform-selection test registered in cmake/UnitTests.cmake is CnaPlatformSelection_Retired (RetiredPlatformSelection.cmake). It runs the selection file in script mode with WIN32 set to false and to true and asserts the refusal of every former selector: SDL2 as an unknown name, and the former direct window-system values as removed values that point to -DCNA_PLATFORM=SDL3. No test runs CNA_PLATFORM=SDL12, EMSCRIPTEN or TERMINAL on Windows and asserts the reserved-identifier FATAL_ERROR. The RendererDefaultCase and RendererRetiredIdentityCase tests cover renderer selection, not the reserved-identifier platform branch.

Source locations

• cmake/PlatformSelection.cmake — the FATAL_ERROR refusal branches: retired list, reserved list (TERMINAL reserved on Windows), unknown name
• cmake/UnitTests.cmake — CnaPlatformSelection_Retired is the only platform-selection configure test
• cmake/Tests/RetiredPlatformSelection.cmake — covers the retired values and the unknown-name refusal only

Evidence

Checked by reading PlatformSelection.cmake and every add_test in cmake/UnitTests.cmake and each fixture under cmake/Tests/ at c1c316b9. Not executed.

Narrowed at c1c316b9: 39e84add7 (PLAT-142) replaced the earlier SDL2-only fixture with CnaPlatformSelection_Retired, which drives the retired-value and unknown-name branches on a simulated Windows and non-Windows host, so the refusal of a Windows-only value off Windows that this entry listed is now covered (that value is retired everywhere). Nothing drives the reserved-identifier branch.

Focused reproduction

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

Current tests

CnaPlatformSelection_Retired (the retired and unknown names) and the renderer-selection fixtures. None runs a reserved CNA_PLATFORM.

Regression test

A fixture like RetiredPlatformSelection.cmake that runs the selection file in script mode with CNA_PLATFORM set to each reserved value (and TERMINAL with WIN32 true), asserting a non-zero exit and the message that names the reserved and available sets.

Blast radius

The guarantee that a reserved CNA_PLATFORM fails loudly instead of falling back to SDL3 is unverified; a regression there would ship silently. The offered platforms themselves are exercised by the platform CI cells.

Workaround

No workaround is known.