Hi Mykola, hi Domenico,
On 11/1/22 12:44 AM, Mykola Lysenko wrote:
Hi John,
Test FAILs when there is an unexpected condition during test/subtest execution, developer does not control it. Hence we propagate FAIL subtest result to be the test result, test_progs result and consequently CI result.
On the other hand, SKIP state is fully controlled by us. E.g. we decide when particular subtest/test should be skipped. We do not propagate SKIP state to the test_progs result. test_progs result can either be OK or FAIL. Also, SKIPPED subtest is not an indication of a problem in a test. Hence, I do not think one SKIPPED subtest should mark the whole test as SKIPPED.
For example, core_reloc_btfgen has 77 subtests (https://github.com/kernel-patches/bpf/actions/runs/3349035937/jobs/5548924891#step:6:4895). Some of them are skipped right now. However, most of them are passing. It is a normal state. For me, marking core_reloc_btfgen as SKIP would mean that something is not right with the whole test. Also, I do not think we are reviewing SKIP tests / subtests right now. Maybe we should. But this would be orthogonal discussion to this patch.
I think parts of the above should probably be incorporated into the below
commit description to better explain the rationale for the change.
On Oct 28, 2022, at 4:24 PM, John Fastabend <john.fastabend@xxxxxxxxx> wrote:
Domenico Cerasuolo wrote:
From: Domenico Cerasuolo <dceras@xxxxxxxx>
When showing the result of a test group, if one
of the subtests was skipped, while still having
passing subtets, the group result was marked as
nit: subtets
SKIPPED.
#223/1 usdt/basic:SKIP
#223/2 usdt/multispec:OK
#223 usdt:SKIP
With this change only if all of the subtests
were skipped the group test is marked as SKIPPED.
#223/1 usdt/basic:SKIP
#223/2 usdt/multispec:OK
#223 usdt:OK
I'm not sure don't you want to know that some of the tests
were skipped? With this change its not knowable from output
if everything passed or one passed.
I would prefer the behavior: If anything fails return
FAIL, else if anything is skipped SKIP and if _everything_
passes mark it OK.
My preference is to drop this change.
I guess for manual testing you could just grep for usdt and see all subtest
results. I think changing from SKIP to OK is fine, but could we indicate e.g.
"usdt:OK (SKIP:1/2)" to differ from "usdt:OK" where nothing had to be skipped?
Presumably this would address John's concern, too.
Signed-off-by: Domenico Cerasuolo <dceras@xxxxxxxx>
---
tools/testing/selftests/bpf/test_progs.c | 11 +++++++++--
1 file changed, 9 insertions(+), 2 deletions(-)
diff --git a/tools/testing/selftests/bpf/test_progs.c b/tools/testing/selftests/bpf/test_progs.c
index 0e9a47f97890..14b70393018b 100644
--- a/tools/testing/selftests/bpf/test_progs.c
+++ b/tools/testing/selftests/bpf/test_progs.c
@@ -222,6 +222,11 @@ static char *test_result(bool failed, bool skipped)
return failed ? "FAIL" : (skipped ? "SKIP" : "OK");
}
+static char *test_group_result(int tests_count, bool failed, int skipped)
+{
+ return failed ? "FAIL" : (skipped == tests_count ? "SKIP" : "OK");
+}
+
static void print_test_log(char *log_buf, size_t log_cnt)
{
log_buf[log_cnt] = '\0';
@@ -308,7 +313,8 @@ static void dump_test_log(const struct prog_test_def *test,
}
print_test_name(test->test_num, test->test_name,
- test_result(test_failed, test_state->skip_cnt));
+ test_group_result(test_state->subtest_num,
+ test_failed, test_state->skip_cnt));
}
static void stdio_restore(void);
@@ -1071,7 +1077,8 @@ static void run_one_test(int test_num)
if (verbose() && env.worker_id == -1)
print_test_name(test_num + 1, test->test_name,
- test_result(state->error_cnt, state->skip_cnt));
+ test_group_result(state->subtest_num,
+ state->error_cnt, state->skip_cnt));
reset_affinity();
restore_netns();
--
2.30.2
