Ensure we pass the has_nulls tparam to mixed_join kernels

[abellina]

Fixes #16706

I'll build/test our stack with this change, but it looks like a typo.

If there's a quick unit test we can add I'd be happy to hear recommendations or for someone else to follow on with such a test.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[abellina]

/ok to test

[davidwendt]

Do we not have a simple gtest for this?

[abellina]

I was able to build our plugin with a cuDF that has this change and it fixes our issue.

[abellina]

Do we not have a simple gtest for this?

There is code in mixed_join_tests that should test the null path: https://github.com/rapidsai/cudf/blob/branch-24.10/cpp/tests/join/mixed_join_tests.cu#L711, and the not null path: https://github.com/rapidsai/cudf/blob/branch-24.10/cpp/tests/join/mixed_join_tests.cu#L680. I think..., I am not familiar with this code, so I could be missing it.

That said, these tests are not failing with or without my change. We have been able to repro it from spark, and its apparent from the linked PR that this was missed. That said I do think we should add a test or figure out why it isn't working.

I tried setting --rmm_mode=cuda because a pool could have hidden this, but it doesn't repro in the JOIN_TEST suite.

[ttnghia]

I think it is hard to reproduce a test that "fails" because this is a memory access UB issue rather than a validity issue.

[abellina]

/ok to test

[mythrocks]

Piling on to the 👍s here. Good catch, @abellina.

[abellina]

/merge