LuciferYang opened a new issue, #12892: URL: https://github.com/apache/gluten/issues/12892
### What Gluten has no ANSI-on cast coverage on Spark 3.4 and 3.5, and the workflow that would run it does not cover those versions. On 4.0 and 4.1, `VeloxTestSettings` enables both `GlutenCastWithAnsiOffSuite` and `GlutenCastWithAnsiOnSuite`. On 3.4 and 3.5 only the ANSI-off one is enabled, under its older name `GlutenCastSuite`. ### The order matters Enabling an ANSI-on suite on 3.4/3.5 by itself buys nothing. `FallbackOnANSIMode` (`gluten-substrait/.../FallbackRules.scala:30`) tags the whole plan when `enableAnsiMode && enableAnsiFallback`, and `spark.gluten.sql.ansiFallback.enabled` defaults to true, so with session ANSI on the query falls back to vanilla Spark and the suite tests nothing about Velox. The only place that flag is turned off is `.github/workflows/velox_backend_ansi.yml`, and that workflow has jobs for 4.0 and 4.1 only. The comment on the 4.0 suite says as much: ANSI fallback has to be off for Velox to actually execute. So this needs two steps, in order: 1. Add 3.4 and 3.5 jobs to `velox_backend_ansi.yml`, mirroring the existing `spark-test-spark-ut-ansi-spark40` shape, including `-Dspark.gluten.sql.ansiFallback.enabled=false` in the argLine. 2. Enable the ANSI-on cast suite on those two modules and settle whatever it turns up, with per-case exclusions carrying a stated reason. ### Not a blocker for removing Spark 3.3 Recorded during the audit for #12807. It is unrelated to deleting the 3.3 module: the three ANSI cast suites that only exist on `gluten-ut/spark33` never reach Velox either, for the reasons written up in #12863, so nothing is lost by deleting them and nothing is gained by keeping them. Related: #12807, #12863. -- 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]
