A complete overhaul of the sanitycheck script and how we build and run
tests. This new version of sanitycheck uses python for job distribution and
drops use of Make.
In addition to the move to python threading library, the following has
- All handlers now run in parallel, meaning that any simulator will run
in parallel and when testing on multiple devices (using
--device-testing) the tests are run in parallel.
- Lexical filtering (using the filter keyword in yaml files) is now
evaluated at runtime and is no long being pre-processed. This will allow
us to immediately start executing tests and skip the wait time that was
needed for filtering.
- Device testing now supports multiple devices connected at the same
time and is managed using a hardware map that needs to be generated and
maintained for every test environment. (using --generate-hardware-map
- Reports are not long stored in the Zephyr tree and instead stored in
the output directory where all build artefacts are generated.
- Each tested target now has a junit report in the output directory.
- Recording option for performance data and other metrics is now
available. This will allow us to record the output from the console and
store the data for later processing. For example benchmark data can be
captured and uploaded to a tracking server.
- Test configurations (or instances) are no longer being sorted, this
will help with balancing the load when we run sanitycheck on multiple
hosts (as we do in CI).
And many other cleanups and improvements...