We created this package as a way to provide a common set of vendored dependencies for RAPIDS projects to help avoid conda clobbering warnings. While it does accomplish that goal, it means that we are relying on 1) being very careful in our patches and 2) having correct packaging constraints in place (which are never directly tested if we rely on CPM downloads to get these dependencies). If either of those is done incorrectly, we could produce binaries that are not ABI-compatible with third-party libraries that compile against the stock versions of our patched dependencies. With the resolution of rapidsai/build-planning#104, we no longer need to ship fmt and spdlog because they are completely hidden from our public APIs. We are actively working on getting RAPIDS building against the latest version of CCCL without patches. Once we accomplish that, we will no longer have any reason to use patches. As such, we have no need for rapids-core-dependencies and can get rid of the package altogether.
We created this package as a way to provide a common set of vendored dependencies for RAPIDS projects to help avoid conda clobbering warnings. While it does accomplish that goal, it means that we are relying on 1) being very careful in our patches and 2) having correct packaging constraints in place (which are never directly tested if we rely on CPM downloads to get these dependencies). If either of those is done incorrectly, we could produce binaries that are not ABI-compatible with third-party libraries that compile against the stock versions of our patched dependencies. With the resolution of rapidsai/build-planning#104, we no longer need to ship fmt and spdlog because they are completely hidden from our public APIs. We are actively working on getting RAPIDS building against the latest version of CCCL without patches. Once we accomplish that, we will no longer have any reason to use patches. As such, we have no need for rapids-core-dependencies and can get rid of the package altogether.