On Fri, Feb 19, 2021 at 12:28:16PM -0600, Alex G. wrote:
Hi,
I keep being hit by two frustrating issues when trying to run tests. My expectation is that I can run a test at any time when working on something -- usually to check the correctness of that something. That's not the case today.
The first one is an error message about mrproper:
$ ./test/py/test.py --bd sandbox_spl .. is not clean, please run 'make mrproper'
I don't think this message is very useful. Running 'mrproper' is out of the question, as it would wipe out the branch I'm currently working on. So what I end up doing is deleting the mrimproper check from the Makefile. And while it's a fast workaround, it dirties the tree and gets in the way of rebasing patches or working with git.
I can continue just fine without 'mrproper death' so I really don't understand the over-abundance of caution in breaking the build.
So, I looked a bit here. The problem is that our logic in the top-level Makefile for "Is this a separate srctree/objtree build?" is very out of date / wrong. Re-syncing this with a current Linux Makefile should resolve this problem.
The second issue is the how the tests are trying to build graphics:
$ ./test/py/test.py --bd sandbox_spl -k "whatever" make[1]: sdl2-config: Command not found ../arch/sandbox/cpu/sdl.c:10:10: fatal error: SDL2/SDL.h: No such file or directory 10 | #include <SDL2/SDL.h>
The solution is to set NO_SDL=1, but this is far from obvious. The first instinct is always to look up the command help:
$ ./test/py/test.py --help
But this really isn't at all useful. A simple grep for sdl2-config also doesn't immediately reveal the solution. Most u-boot tools get their arguments via the commandline. Thus, a reasonably competent u-boot developer will not think that environment variables are the solution. Ergo, environment variables are not the ideal way to solve this.
Maybe we could have a commandline option, and at the very least, catch this error and print something useful on the console.
I'm not sure what the preferred way would be to solve the above. For me, these issues cause a significant enough disruption to my workflow, that I am very likely to not run tests regularly. I suspect I'm not alone.
Part of the problem is that if you don't have graphics enabled in sandbox, you in turn don't get other tests run I believe. A check in arch/sandbox/config.mk for NO_SDL being unset AND not having $(SDL_CONFIG) exist would provide a better user experience, yes.
Thanks for the feedback!