Fix the test_valgrind.sh script running tests under valgrind

[ldorau]

The current script just checks if there is the
"ERROR SUMMARY: 0 errors from 0 contexts"
line in the output of a test.

It turned out that there can be many lines starting with
"ERROR SUMMARY:" in the output of one test - some with
errors and some without, so the current approach is wrong.
We have to check if there are no other lines starting with
"ERROR SUMMARY:" than this:
"ERROR SUMMARY: 0 errors from 0 contexts" instead.

[igchor]

@ldorau, the failures for protection_flag_* testcases in provider_os_memory_config` are expected since we are triggering segfaults intentionally. I think we need some kind of blacklist/filter for tests under valgrind.

[igchor]

But there are still some failures in numa_modes test (after commenting out protection_flag_* tests) that are not reported after your fixes:

./test/umf_test-provider_os_memory_config (valgrind failed) - OK
./test/umf_test-scalable_pool - OK
PASSED

The reason is that the test fails under valgrind - but the failure is not a memory leak, we just fail on an assert. It looks like get_mempolicy() does not work under valgrind.

[ldorau]

@ldorau, the failures for protection_flag_* testcases in provider_os_memory_config` are expected since we are triggering segfaults intentionally. I think we need some kind of blacklist/filter for tests under valgrind.

I will try to add suppressions for them.

[ldorau]

@ldorau, the failures for protection_flag_* testcases in provider_os_memory_config` are expected since we are triggering segfaults intentionally. I think we need some kind of blacklist/filter for tests under valgrind.

I will try to add suppressions for them.

@igchor Suppressions added.

[igchor]

@ldorau, the failures for protection_flag_* testcases in provider_os_memory_config` are expected since we are triggering segfaults intentionally. I think we need some kind of blacklist/filter for tests under valgrind.

I will try to add suppressions for them.

@igchor Suppressions added.

And what about the failures with get_mempolicy? In the logs, I still see:
./test/umf_test-provider_os_memory_config (valgrind failed) - OK

Are we just going to ignore this? I don't really know why get_mempolicy does not work under valgrind.

[ldorau]

And what about the failures with get_mempolicy? In the logs, I still see: ./test/umf_test-provider_os_memory_config (valgrind failed) - OK

Are we just going to ignore this? I don't really know why get_mempolicy does not work under valgrind.

Yes, I would ignore that. The most important is that this test passes without valgrind. Apparently valgrind changes behavior of this test. We cannot help this.

[ldorau]

The valgrind (Nightly) build passes: https://github.com/oneapi-src/unified-memory-framework/actions/runs/7731094933/job/21077971238?pr=189

[ldorau]

The corresponding Nightly CI job: https://github.com/ldorau/unified-memory-framework/actions/runs/7739459731

[DamianDuy]

LGTM