Migrate all Python builds from scikit-build to scikit-build-core

[vyasr]

RAPIDS libraries are generally built with CMake. To facilitate better integration of the C++ builds with Python builds, we switched from using pure setuptools builds to using scikit-build. This change was crucial to enabling wheels by providing a single standard entrypoint (all the usual Python pip [install|wheel|etc] machinery) for building a Python package while also compiling the required C++ components. However, scikit-build's approach to enabling this is fundamentally limited because it relies on plugging into setuptools directly in ways that setuptools only marginally supports. The result is a tool that works most of the time, but has various sharp edges (e.g. incomplete support for MANIFEST.in, broken installations in certain cases, etc) and limitations (an inability to support true editable installations, mixed support for pyproject.toml/setup.py, etc).

The solution is to switch to the newer scikit-build-core builder, a modern standards-based builder that offers the same class of functionality as scikit-build (integrating a Python build with CMake) in a more reliable manner. Doing so will allow us to completely removed deprecated components of our build systems (various uses of setup.py), get rid of workarounds for scikit-build (e.g. the languages we must specify at the CMake level), and get full support for critical features like editable installs.

PRs Contributing To This Effort

Preview Give feedback

1. Switch to scikit-build-core raft#2051
   CMake breaking ci improvement python +5
2. Switch to scikit-build-core kvikio#325
   breaking improvement +2
3. Switch to scikit-build-core cuspatial#1313
   Python breaking ci cmake conda improvement +6
4. Switch to scikit-build-core cuml#5693
   CMake Cython / Python breaking ci conda improvement +6
5. Switch to scikit-build-core cugraph#4053
   CMake breaking ci conda improvement python +6
6. Switch to scikit-build-core cudf#13531
   CMake Python breaking ci conda improvement libcudf +7
   
7. Switch to scikit-build-core rmm#1287
   0 - Blocked CMake Python breaking conda improvement +6
   
8. Switch to scikit-build-core ucxx#146
   breaking improvement +2
   

[jameslamb]

I'm +1 on this.

For reference, the README has a thorough and direct list of differences between scikit-build and scikit-build-core: https://github.com/scikit-build/scikit-build-core

[bdice]

@vyasr Would it be possible to migrate cugraph-ops as well? It seems that repo was missed. It'd be good to handle this while it's fresh on our minds.

https://github.com/rapidsai/cugraph-ops/blob/96a388691c8d42a9f3b7a56bf6f29d24d8ffe166/pylibcugraphops/pyproject.toml#L20
https://github.com/rapidsai/cugraph-ops/blob/96a388691c8d42a9f3b7a56bf6f29d24d8ffe166/docker/Dockerfile#L63

[bdice]

Also, this might not need migrated, but rapids-cmake also relies on scikit-build as a test dependency (and cython>=0.29,<0.30 which might be possible to clean up at the same time).

https://github.com/rapidsai/rapids-cmake/blob/208b0d1240caaca41603fccc10b7ac44aea47477/dependencies.yaml#L146

[vyasr]

I chose not to include cugraph-ops in this work because that package uses nanobind instead of Cython for its bindings, so it's not sensitive to rapids-cython changes. I also don't know much about its nanobind usage, and it looks like they do something bespoke with setting the CMake install directory. I'd rather leave that one to the devs to manage the transition since it doesn't impact rapids-cython at all.

[vyasr]

Regarding rapids-cmake, I'm pretty sure that dependency is superfluous and can be removed altogether. We never actually launch builds via Python in rapids-cmake. Removed that dependency in rapidsai/rapids-cmake#512.

[jameslamb]

I think there's one more block of work that should be done as part of this.

There are some projects with configuration in environment variables that scikit-build recognized but scikit-build-core does not. From scikit-build-core's migration guide (docs)

The SKBUILD_CONFIGURE_OPTIONS environment variable is now named SKBUILD_CMAKE_ARGS for consistency.

The SKBUILD_BUILD_OPTIONS environment variable is not supported. Some specific features are accessible using alternative variables. In particular, use CMAKE_BUILD_PARALLEL_LEVEL or SKBUILD_CMAKE_VERBOSE to control build parallelism or CMake verbosity directly.

So I see 3 small pieces of additional work:

• convert all uses of SKBUILD_CONFIGURE_OPTIONS to SKBUILD_CMAKE_ARGS
• convert stuff like SKBUILD_BUILD_OPTIONS="-j4" to CMAKE_BUILD_PARALLEL_LEVEL=4
• convert stuff like SKBUILD_BUILD_OPTIONS="-v" to SKBUILD_CMAKE_VERBOSE="true"

For all projects that have recently cut over to scikit-build-core.

---

I didn't find a ton of these from GitHub search (e.g. this link), but there are a few.

For example, cuml cut over to scikit-build-core (rapidsai/cuml#5693) but is still using SKBUILD_BUILD_OPTIONS to control parallelism (code link).

[vyasr]

Good catches @jameslamb! Let me work through these. As far as build options we've just been removing the build options everywhere, so I'll do that, and hopefully the configure options are mostly fixed already since those need to be right for everything to work correctly.

[vyasr]

These are the remaining issues in repos that have made the scikit-build-core transition. We can close this issue once the following PRs are merged:

• Remove extraneous SKBUILD_BUILD_OPTIONS cuml#5714
• https://github.com/rapidsai/cuopt/pull/1615
• Fix CMake args cudf#14746

[vyasr]

Closing this now. All the associated work has been merged.