For each PR, we also want to have quick feedback if compilation and QEMU-based tests fail, but we can't do hardware-based tests for every PR on all platforms.
In addition, for hardware tests, we don't have a centralized CI - similar to the majority of the open-source programs. The expectation is that every contributor company takes care in depth of their own platforms, using GitHub triggers to run the tests as needed. Intel folks are not going to run any tests on i.MX hardware nor would we expect anyone to run tests for each PR on Intel hardware, though it'd be nice to know if the tests fail with the results visible.
So short answer, we can add some compilation checks for i.MX8 on both Travis CI and the Intel-hosted one, but don't expect more.
Was this your understanding?
This is my understanding too. We need to add compilation check for i.MX the rest we will try to handle it internally.
ok, we'll work with Xiuli to make it happen.
- Send Linux kernel side patches.
For the Linux part, we try to build from a minimal kernel, see the kconfig subproject on SOF. If we can have a similar solution for i.MX it'd be great. Maintaining entire configs is too complicated, using a default + SOF-required additions is a lot simpler.
default + SOF required additions is good ans simple. The one problem is that arm64 use one single config for all supported boards, hence the compilation takes sometimes more than we want.
For me this is not a problem because I don't build from fresh code very often but for CI indeed I need to find a way to have a smaller config.
Yes, I spent a bit of time to remove the stuff we didn't need for SOF in the x64_64_defconfig while still keeping the minimal support to actually test (network, i915, storage, etc). It's much faster to compile and load.