Various fixes to the manifest #422
Conversation
It wasn't removed, instead the scripts added a feature to include dependencies recursively. So if cudf depends on kvikio, and kvikio depends on rmm, then cudf gets both |
That seems like a good thing to do but also error-prone to rely on exclusively. I would rather follow a similar practice to how you would handle Python dependencies, C++ includes, or Python imports: if you only use something transitively then the recursive solution is good, but if you use something directly you should still include it. Otherwise you could wind up needing to make nonlocal (e.g. if raft removes a dependency that cugraph needs you'd have to trace that) or seemingly unrelated (when we remove cugraph-ops as a dependency of cugraph here, we lose every dependency of cugraph) changes. |
|
Note that in this case what happened was that #414 removed the rmm dependency from kvikio because it does not in fact depend on rmm, but I didn't realize that was going to impact cudf. |
|
@trxcllnt what do you want to do with this PR? |
This PR fixes a number of issues in the manifest:
devcontainer-utilsandrapids-build-utilsupdates for DLFW #221. Since rmm is header-only I guess nobody noticed the issue since it doesn't increase build times and only shows up if you attempt to make local changes to rmm and test them in cudf.-D<package_name>_ROOTvariables in the CMake configure don't cause any problems so I made everything explicit.