Linux regression tests

Use GCC 14, matching gcov 14, CMake, Ninja and gcovr 8.6. The CI installer supports Ubuntu 24.04. From the repository root:

export CC=gcc-14 GCOV=gcov-14 TX_COVERAGE=ON
./scripts/build.sh
./scripts/test.sh

The runner verifies the dependency commits recorded in threadx-revision.txt and filex-revision.txt. The FileX revision is the tested head of eclipse-threadx/filex#106, which is still open. Updating a dependency requires changing its pin and rerunning all ten configurations. An existing checkout with another revision or modified tracked files is rejected.

All configurations need the FileX source tree. Non-standalone configurations link the shared ThreadX and FileX libraries; standalone configurations build FileX with standalone support. LevelX has no Windows regression port.

TX_COVERAGE=ON instruments every LevelX configuration. A complete test run clears old coverage first, runs every configuration, and collects coverage even when a test fails. Reports under coverage_report/per_configuration contain JSON, XML and HTML for each configuration. The merged reports combine all ten JSON inputs and use repository-relative source names. Missing, empty or unmeasured inputs fail collection. Both line and branch floors are enforced by coverage.sh; the workflow also enforces its integer line floor.

After a successful full run, ./test/cmake/check_coverage.sh verifies rejection of missing, empty and unmeasured coverage inputs, then restores and remerges the valid reports.

The reusable workflow pin supplies bounded install, build and test steps and retains test and coverage artifacts on failure. Only master push or manual runs can deploy coverage. Dependabot updates target dev; GitHub activates this configuration once it is present on the default branch.

Coverage target

The measured GCC 14 union is 1,632/2,400 lines (68.00%) and 1,596/2,439 branches (65.44%). The enforced floors are 68.0% lines and 65.4% branches. The 100% target still needs 768 lines and 843 branches covered. The largest line gaps are the FileX simulator adapters (164), NOR block reclaim (69), NOR extended open (65), and NAND block data movement (38). Other gaps include media-error paths, metadata allocation, sector release and simulator failures. These sources remain in the denominator.

The ECC regression checks every single-bit position in a 512-byte page, corrections in both halves, and uncorrectable errors. All four ECC helper files have full line coverage. The full, driver-interface and combined NOR cache configurations contribute 97, 51 and 65 source lines absent from the default configuration respectively. Standalone configurations select a subset of the same source lines; their reports still participate in the union. Function merging uses the earliest declaration line because driver-interface macros place otherwise identical function declarations on different lines.

The ECC test views aligned USHORT storage through a character pointer because the ECC implementation accesses words. This is a deviation from advisory MISRA C:2004 Rule 11.4; character access preserves alignment and is explicitly permitted by the character-pointer exception in MISRA C:2012/2023 Rule 11.3.