2010YOUY01 commented on issue #27082: URL: https://github.com/apache/datafusion/issues/27082#issuecomment-6760344633
> Hi [@2010YOUY01](https://github.com/2010YOUY01), I'm working on [#24243](https://github.com/apache/datafusion/issues/24243) (adding NULLs to fuzz test data). Most of its remaining items are also on this tracker. Like limit, sort, merge, sort-preserving repartition, join, and the old aggregate path. > > Would you like me to migrate each one to `FuzzTest` as I add NULLs, or keep the two efforts separate? I'd start with `limit_fuzz.rs`. > > If migrating makes sense: these tests use `SessionContext` and the DataFrame API, so the `datafusion-fuzz` crate would need `datafusion` as a dev-dependency. Is that OK with you? I think #24243 is a valid improvement, but I don't think expanding the existing internal fuzz tests is the highest-ROI way to improve fuzzing coverage. My preference is to: - Keep existing internal fuzz tests in maintenance mode. The intention of this tracking issue is to simplify the infrastructure through mechanical refactoring, rather than expand coverage. - Invest in [datafusion-fuzzer](https://github.com/datafusion-contrib/datafusion-fuzzer) for new coverage. Currently, internal fuzz tests (sort/join/aggr) each have their own data generation framework, so even adding NULL support requires changes across multiple suites. In contrast, datafusion-fuzzer provides unified data/query generation across operators and can exercise more complex queries. I think this would give us better coverage in the long term, with less duplicated maintenance effort. -- 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]
