Keith Kraus

This still doesn't work unfortunately. I.E. here's a clearer example from @mrocklin above: ``` # foo.py def do_nothing(): pass ``` ``` # bar.py print("I was imported!") ``` ``` # main.py...

> `multiprocessing` always re-imports the main module in child processes. I'm not sure why that is, but there's probably a reason (it's been like that for as long as I...

Why are we planning on handling this at the Python level? DLPack is a C header and is usable from C libraries as well. If we solve the problem for...

> The concern is breaking the ABI, you can't just add a field. Can you think of a way to do it in C in a reasonable manner @kkraus14? It's...

Workaround that fixed the issue locally for me was adding `-DUNIT_TESTS_CXX_VERSION=17` to force the unit tests to use C++17 instead of C++20.

Does this happen with Pandas? Have you been able to narrow it down to a specific function in streamz / cudf that causes this?

cc @thomcom sounds like we have a memory leak somewhere in the `MultiIndex` / groupby codebase.

Rapidz (https://github.com/xpdAcq/rapidz/) is different than Rapids (https://rapids.ai). @jsmaupin I believe you were referring to **Rapids** as opposed to **Rapidz**.

> Well its also exciting streamz is going to support arrow through rapids too?! @kkraus14 At that point Arrow is more of an implementation detail for GPU execution, but yes...

Moving this issue to the new cuSpatial library at https://github.com/rapidsai/cuspatial