Hello, I am pursuing my quest to convert standalone bpf tests (from tools/testing/selftests/bpf) into the test_progs framework so they can be executed automatically by the CI tooling.
I would like to continue on the bpftool tests, especially test_bpftool.py ([1]). This one involves quite a lot of json parsing on bpftool output (to validate that some entries are present in the output, depending on the used command), which does not seem to be a use case currently in bpf selftests. The first tests may be handled by some manual parsing (eg strstr'ing keys in the output, that's what I've done for the recently converted bpftool_metadata test, see [2]), but that's a bit fragile, and there are more complex subtests for which this loosy strategy will get even more fragile (eg test_feature_kernel_full_vs_not_full generates json output with two different commands, and tests the diff). I then face the need to have some proper json parsing in test_progs. I kind of understand that there is a will to keep the dependency list small for test_progs, so I'm asking here what could be the best option for this: - is it ok to make test_progs depend on a new json parsing library (eg: cJSON) ? and so add the library to CI images ? - or some implicit dependency on some CLI tooling (eg: jq) ? and so, add the cli tool to CI images ? - test_progs is reusing json_writer.c from bpftool ([2]), should we rather write a custom json_reader.c as well (even if not needed at that point by bpftool itself) ? Any opinion on this ? Alexis [1] https://elixir.bootlin.com/linux/v6.19.3/source/tools/testing/selftests/bpf/test_bpftool.py [2] https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next.git/tree/tools/testing/selftests/bpf/prog_tests/bpftool_metadata.c#n49 [3] https://elixir.bootlin.com/linux/v6.19.3/source/tools/bpf/bpftool/json_writer.c -- Alexis Lothoré, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
