Sorry for the delay.
On Mon, Jan 27, 2020 at 10:03 PM SeongJae Park sj38.park@gmail.com wrote:
On Mon, 27 Jan 2020 16:02:48 -0800 Brendan Higgins brendanhiggins@google.com wrote:
On Sat, Jan 25, 2020 at 5:59 PM sj38.park@gmail.com wrote:
From: SeongJae Park sjpark@amazon.de
Deletions of configs in the '.kunitconfig' is not applied because kunit rebuilds '.config' only if the '.config' is not a subset of the '.kunitconfig'. To allow the deletions to applied, this commit modifies the '.config' rebuild condition to addtionally check the modified times of those files.
The reason it only checks that .kunitconfig is a subset of .config is because we don't want the .kunitconfig to remove options just because it doesn't recognize them.
It runs `make ARCH=um olddefconfig` on the .config that it generates from the .kunitconfig, and most of the time that means you will get a .config with lots of things in it that aren't in the .kunitconfig. Consequently, nothing should ever be deleted from the .config just because it was deleted in the .kunitconfig (unless, of course, you change a =y to a =n or # ... is not set), so I don't see what this change would do.
Can you maybe provide an example?
Sorry for my insufficient explanation. I added a kunit test (SYSCTL_KUNIT_TEST) to '.kunitconfig', ran the added test, and then removed it from the file. However, '.config' is not generated again due to the condition and therefore the test still runs.
For more detail:
$ ./tools/testing/kunit/kunit.py run --defconfig --build_dir ../kunit.out/ $ echo "CONFIG_SYSCTL_KUNIT_TEST=y" >> ../kunit.out/.kunitconfig $ ./tools/testing/kunit/kunit.py run --build_dir ../kunit.out/ $ sed -i '4d' ../kunit.out/.kunitconfig $ ./tools/testing/kunit/kunit.py run --build_dir ../kunit.out/The 2nd line command adds sysctl kunit test and the 3rd line shows it runs the added test as expected. Because the default kunit config contains only 3 lines, The 4th line command removes the sysctl kunit test from the .kunitconfig. However, the 5th line still run the test.
This patch is for such cases. Of course, this might make more false positives but I believe it would not be a big problem because .config generation takes no long time. If I missed something, please let me know.
I think I understand.
It is intentional - currently - that KUnit doesn't generate a new .config with every invocation. The reason is basically to support interaction with other methods of generating .configs. Consider that you might want to use make menuconfig to turn something on. It is a pretty handy interface if you work on vastly different parts of the kernel. Or maybe you have a defconfig that you always use for some platform, I think it is easier to run
make foo_config; tools/testing/kunit/kunit.py run
Then having to maintain both your defconfig and a .kunitconfig which is a superset of the defconfig.
Your change would make it so that you have to have a .kunitconfig for every test environment that you care about, and you could not as easily take advantage of menuconfig.
I think what we do now is a bit janky, and the use cases I mentioned are not super well supported. So I am sympathetic to what you are trying to do, maybe we could have a config option for it?
I think Ted and Bjorn might have opinions on this; they had some related opinions in the past.