CNA-BUG-200: The registered CTest CNAEXT_NoPosixSetenv fails at this snapshot: twelve Wayland setenv/unsetenv call lines trip the gate

CNA snapshot 009d40f5  ·  Known Issues › Current bugs  ·  source links pinned to 009d40f5

scripts/check_no_posix_setenv.py forbids POSIX setenv/unsetenv under modules/, tools/ and tests/, but the native Wayland backend and its tests added twelve such lines on 2026-09-15 and 2026-09-16, so the gate CNA registers in every test-enabled tree has exited 1 since 2026-09-15.

Identifier

CNA-BUG-200

Category

Bug

Subsystem

Build & CI

Status

Open

Verified against

CNA 009d40f5 (009d40f5dd085c4e674d3479675fac84b12b3e0a)

Severity

Medium (a triage suggestion, not a project priority)

Evidence basis

Reproduced: executed for this entry (the Evidence section names exactly what was run)

Tests touching this area

Yes: see Current tests

Affected contract

CTest CNAEXT_NoPosixSetenv (modules/graphics-ext/examples/CMakeLists.txt, label CnaExt)

Expected behaviour

A gate CNA registers passes on its own tree. The rule it enforces (use System::Environment::SetEnvironmentVariable, because MinGW-w64 has no setenv/unsetenv) was introduced with the gate on 2026-08-19.

Actual behaviour at TARGET

check_no_posix_setenv.py reports twelve lines: ::unsetenv("XDG_ACTIVATION_TOKEN") in WaylandPlatform.cpp, and setenv/unsetenv of WAYLAND_SOCKET, WAYLAND_DISPLAY, LC_ALL, DBUS_SESSION_BUS_ADDRESS and XDG_ACTIVATION_TOKEN in WaylandDesktopPortalTests.cpp, WaylandProtocolTests.cpp, WaylandTestCompositor.cpp and WaylandTestEnvironment.cpp, all under modules/platform/tests/CNA/Platform/. The test is added in CMakeLists.txt inside if(CNA_BUILD_TESTS) whenever a Python 3 interpreter is found, so it is red in every test-enabled tree. The Wayland code never reaches a Windows build (CMakeLists.txt compiles src/Wayland only for CNA_PLATFORM=WAYLAND, and UnitTests.cmake drops the Wayland tests otherwise), so the MinGW breakage the gate guards against does not occur; the defect is the permanently red gate. general-tests-ci.yml runs an unfiltered ctest and its known-failure list names only EasyGL_GraphicsDevice_ReferenceStencil, so that job would classify this failure as new.

Source locations

• scripts/check_no_posix_setenv.py — CALL regex and the modules/tools/tests scan
• modules/graphics-ext/examples/CMakeLists.txt — add_test NAME CNAEXT_NoPosixSetenv
• modules/platform/src/Wayland/WaylandPlatform.cpp — ::unsetenv("XDG_ACTIVATION_TOKEN")
• modules/platform/tests/CNA/Platform/WaylandProtocolTests.cpp — setenv/unsetenv of WAYLAND_SOCKET, XDG_ACTIVATION_TOKEN, LC_ALL
• modules/platform/tests/CNA/Platform/WaylandTestEnvironment.cpp — setenv of DBUS_SESSION_BUS_ADDRESS, WAYLAND_DISPLAY; unsetenv WAYLAND_SOCKET
• modules/platform/tests/CNA/Platform/WaylandDesktopPortalTests.cpp — unsetenv("WAYLAND_SOCKET")
• modules/platform/tests/CNA/Platform/WaylandTestCompositor.cpp — setenv("WAYLAND_SOCKET", ...)
• .github/workflows/general-tests-ci.yml — unfiltered ctest and the KNOWN_FAILURES list
• docs/opengl4-renderer.md — CNA's own record of the CNAEXT_NoPosixSetenv failure

Evidence

Executed: python3 scripts/check_no_posix_setenv.py . against the read-only extracted snapshot tree (a text scan, nothing built); it printed the twelve lines and exited 1. CNA's own records agree: opengl4-renderer.md lists CNAEXT_NoPosixSetenv as the one CNAEXT example failure “on every renderer”, and several plans record it as a standing failure. The Wayland sources arrived in commits 678c6a51b and 227157d2c (2026-09-15), after the gate (c9f1359e1, 2026-08-19). Not executed: ctest itself or any CI run.

Independent re-verification: Executed: python3 check_no_posix_setenv.py . against the read-only extracted snapshot tree (a text scan, nothing built); it printed the twelve lines and exited 1, and the same scan repeated for this review gave the same result. CNA's own records agree: opengl4-renderer.md lists CNAEXT_NoPosixSetenv as the one CNAEXT example failure "on every renderer", and several plans record it as a standing failure. The first Wayland lines arrived in commits 678c6a51b and 227157d2c (2026-09-15) and the last four in c6e125083 and a271ae0ef (2026-09-16), all after the gate (c9f1359e1, 2026-08-19). Not executed: ctest itself or any CI run.

Focused reproduction

# Executed against the snapshot tree (read-only text scan):
python3 scripts/check_no_posix_setenv.py .
# POSIX setenv/unsetenv found ... (12 lines under modules/platform) ; exit status 1

Current tests

The gate itself; it has been red since 2026-09-15.

Regression test

The gate becomes green when the Wayland sources use System::Environment::SetEnvironmentVariable (an empty value unsets), or when the gate scopes out POSIX-only platform directories with a stated reason.

Blast radius

Every test-enabled tree reports one failing CTest; an unfiltered CI run cannot go green, which hides new failures behind a known red one. No runtime behaviour is affected.

Workaround

ctest -E '^CNAEXT_NoPosixSetenv$' for local runs.