Is your feature request related to a problem? Please describe.
Published ai.rapids:cudf classifier JARs are not smoke-tested the way a Maven Central consumer would use them. Packaging issues (e.g. classifiers, static libcudf, nvcomp symbols, native logging) can slip past Java tests that run against an in-tree build.
Describe the solution you'd like
Add smoke tests that load published classifier JARs (cudf + slf4j only) on amd64/arm64 × CUDA 12/13, and wire them into the Java publish/release path as a gate. The goal is to validate the artifact consumers get, not only the tree we built from.
Describe alternatives you've considered
- Rely only on existing cudf-java unit tests. They validate the in-tree build, not a published classifier classpath.
- One-off manual checks against RC bundles. Useful for a single release, but not a lasting gate.
Additional context
Is your feature request related to a problem? Please describe.
Published
ai.rapids:cudfclassifier JARs are not smoke-tested the way a Maven Central consumer would use them. Packaging issues (e.g. classifiers, static libcudf, nvcomp symbols, native logging) can slip past Java tests that run against an in-tree build.Describe the solution you'd like
Add smoke tests that load published classifier JARs (cudf + slf4j only) on amd64/arm64 × CUDA 12/13, and wire them into the Java publish/release path as a gate. The goal is to validate the artifact consumers get, not only the tree we built from.
Describe alternatives you've considered
Additional context