Logging checks

Logging acceptance covers owned destinations, queue loss, callback reentry, retained resources and supported teardown.

Configure a separate directory for each row. Use Release as the build configuration for this matrix so profile changes remain independent of the Debug sanitizer configuration. The Engine-only selection below avoids unrelated native/display prerequisites; use the Component and SDK checks when module composition also changes.

Row	CHERYL_LOG_PROFILE	CHERYL_LOG_COMPILED_MASK
Developer	developer	empty
Support	support	empty
Release	release	empty
Minimal	release	0x07
Disabled	release	0

For the developer row:

(
  set -e
  cd "$(git rev-parse --show-toplevel)"
  cmake -S . -B build/log-developer -DCMAKE_BUILD_TYPE=Release -G Ninja \
    -DCMAKE_POLICY_VERSION_MINIMUM=3.5 -DCMAKE_CXX_SCAN_FOR_MODULES=OFF \
    -DCMAKE_GTEST_DISCOVER_TESTS_DISCOVERY_MODE=PRE_TEST \
    -DCHERYL_BUILD_TESTS=ON -DCHERYL_BUILD_CONSUMER_TESTS=OFF \
    -DCHERYL_BUILD_NATIVE_GLFW=OFF -DCHERYL_BUILD_OPENGL=OFF \
    -DCHERYL_BUILD_DEBUG_TERMINAL_LINUX=OFF \
    -DCHERYL_BUILD_UI_TGUI=OFF -DCHERYL_BUILD_UI_RMLUI=OFF \
    -DCHERYL_BUILD_AUDIO_MINIAUDIO=OFF \
    -DCHERYL_BUILD_WORKER_TOPOLOGY_HWLOC=OFF -DCHERYL_BUILD_WORKER_TOPOLOGY_JSON=OFF \
    -DCHERYL_BUILD_DOCS=OFF -DCHERYL_BUILD_SDK=OFF \
    -DCHERYL_BUILD_DEMO=OFF \
    -DCHERYL_LOG_PROFILE=developer
  cmake --build build/log-developer --parallel \
    --target tests-logging acceptance-logging
  python3 projects/engine/tests/logging-acceptance/logging.py build/log-developer
)

The runner accepts multiple build directories and runs them serially. It neither configures nor builds. Its focused filter selects logging.*; the first-include logging-policy translation unit is compiled as part of tests-logging (output tests-engine-logging); isolated cases use acceptance-logging (output tests-acceptance-engine-logging). Severity cases check selected-profile output, disabled guarded side effects, logger/delegate gates, and include-order stability. In the disabled profile, facade queue cases explicitly skip; isolated native queue cases still exercise the backend independently of compile stripping. Record these distinctions instead of claiming zero skips.

Independent processes exercise standard/non-standard sink log/flush failures, failed rotation leaving a closed file, saturated recursive native submission, Block and DiscardNew loss behavior, discarded flush requests, retained backend ownership, static singleton/pool teardown, and fatal final owner destruction from a callback. Named-routing acceptance separately checks the five engine files, formatted/stream/ direct named writes, category filtering and closed/reopened behavior after the application replaces its default logger. Category creation must not replace that default, and engine records must never appear in the application destinations. The runner enforces a 30-second timeout, checks surviving record order/content, checks shared loss output, and verifies the fatal case’s deliberate exit code. Timeouts and nonzero results are failures, not accepted limitations. Native GPU or real-font fixtures are not prerequisites for this logging matrix.

When the runner fails, use the scenario to choose the next investigation:

Result	Check next
A required executable is missing	Build both targets for that row; the Python driver does not configure or build.
Expected severity records are absent	Compare the row’s compiled profile/mask, runtime gates and selected destination; compile stripping cannot be undone at runtime.
A child times out or returns an unexpected status	Inspect that isolated scenario’s output and teardown path. The deliberate fatal-child exit is accepted only when it matches the driver’s expectation.
A fixture or platform case skips	Preserve it as missing coverage and supply its prerequisites; a successful remaining selection does not accept the skipped case.

For concurrency changes, run relevant cases under separate sanitizer configurations. Debug currently enables ASan/UBSan. A TSan configuration must disable that combination before enabling TSan; adding it is a separate build configuration task, not a reason to combine incompatible sanitizers. Source-only review and this finite matrix cannot prove every interleaving or arbitrary callback termination. Assess selected profiles, failures/skips and host limits separately from source completion; an unselected or skipped scenario supplies no acceptance.