Skip to content

Latest commit

 

History

History
71 lines (60 loc) · 4.38 KB

README.md

File metadata and controls

71 lines (60 loc) · 4.38 KB

Y2Linux host test profiles

The production release check also runs python3 -m unittest tests.test_reborn_splash -v on the packaging host with native libdrm development files and Pillow. The minimal locked kernel container does not include those dependencies. This separate pass is mandatory when packaging the splash helper; a container dependency error is not a passing protocol test. Product PM single-write error handling is covered by the suspend helper's Rust tests (platform/crates/y2-platform/src/commands/suspend/helper_tests.rs) in tools/production/tests.sh.

Retained Baseline02 validation

Fresh kernel/config/DT/module/ABI, Buildroot ARM and Reborn ARM checks pass. The integrated suite has 391 cases: 386 passed and 5 dependency skips; 27 native filesystem/GIO and 3 ALSA checks plus installed ARM checks cover the missing minimal-environment dependencies. All 125 focused source tests and 238 Reborn tests, formatting/lint, Cortex-A7 QEMU, installed ARM applications, ELF/dependencies and preserving-package checks pass. These are retained build receipts, not tests rerun by this documentation update.

The boot-policy fault tests are Rust (platform/crates/y2-platform/src/commands/cpu_idle_tests.rs); exact MT6582 UART admission/ACK cases are in tests/test_cpu_c3_uart.py. Existing CPU, cpuidle, timer, hotplug, CIRQ/SPM, clock, MMC and USB guards remain in their production/source suites. No implementation tests are rerun for documentation alone: use python3 tools/development/check-docs.py, JSON/matrix/manifest consistency checks and git diff --check for this synchronization.

Physical acceptance is separate. UART01 already passed 21 guarded real C3 reset-and-return trials and normal-policy residency. The packaged SSH harness must verify Baseline02's exact identity and automatic cold-boot receipt before foundation checks and bounded C3 trials. See Baseline02 and UART01 hardware.

tools/production/tests.sh is the release-facing production-profile runner. tools/development/host-tests.sh runs the same module list on the host for a quick check (it does not replace the container run). Run it inside the locked, device-isolated environment with tools/build/prepare.py and tools/build/run.py. It generates the pinned Linux UAPI headers needed by native host checks, runs the curated production manifest/storage and source-contract tests, then checks ARM shell scripts and the target ABI with Buildroot's ARM tools. Its QEMU checks do not access a Y2.

The explicitly listed tests.test_* modules are the current production gate; the module list is intentional and new correctness tests belong there when they validate the package path. The retained top-level test modules outside that list are subsystem or historical/profile-specific regression tests. They remain available for focused runs, but are not treated as a current production qualification result merely because broad unittest discovery collected them. test_system_update_fallback uses a real temporary ext4 image to check the root-overlay fallback's kernel, filesystem, version and preservation contract. It requires native e2fsprogs and blkid on the packaging host. The locked kernel environment lacks these tools; its explicit skip must be accompanied by a passing packaging-host run, plus validation of the actual fallback image. In particular, test_baseline*, test_dev*, and profile-named hardware contracts preserve evidence for their named source/profile boundaries.

Host prerequisites are explicit: pinned source/header inputs, Buildroot's ARM rootfs and qemu-arm, and the production test environment marker. A missing prerequisite is an environment result, not a passing test and not a reason to silently skip the associated assertions. Physical-device tests are not part of this runner.

test_filesystems adds scratch-image checks for ext4, FAT32 and exFAT. It needs native e2fsprogs, dosfstools, exfatprogs and mtools (mcopy). Missing tools are explicit dependency skips and require a packaging-host run. The storage pass ran all three on new host images; it did not format or repair device media. Actual card mounts, UTF-8 names and large-file behavior remain separate physical checks; see the storage receipt.