baibaichen commented on issue #7139: URL: https://github.com/apache/gluten/issues/7139#issuecomment-5077423614
This issue is still relevant, and the problem is broader than avoiding a duplicate Arrow build step. ### Why this is needed In vcpkg mode, `--build_arrow=OFF` only skips Gluten's standalone `dev/build-arrow.sh`. Velox still used `Arrow_SOURCE=AUTO`, so the result depended on machine state: - a host with Arrow under `/usr/local` could silently use that installation; - a clean host could fall back to Velox's bundled Arrow 18 ExternalProject; - headers and static libraries from different Arrow versions could be mixed; - bundled Arrow expected shared zstd while Gluten's vcpkg triplet installs static zstd; - Arrow's testing library also lost required Boost.Process usage requirements with newer Boost versions after the Velox Arrow testing patch. This makes `package-vcpkg.sh` non-reproducible on a clean machine and means vcpkg does not actually own the complete native dependency graph. ### Expected behavior When vcpkg is enabled: - vcpkg builds native Arrow and `libarrow_testing.a`; - Velox and Gluten consume Arrow and its dependencies only from `vcpkg_installed`; - missing packages fail instead of falling back to bundled or host libraries; - `dev/build-arrow.sh` is not used; - Java Arrow artifacts continue to be resolved by Maven. Drafted with assistance from GitHub Copilot CLI 1.0.75. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
