gaoyangxiaozhu commented on PR #6464: URL: https://github.com/apache/incubator-gluten/pull/6464#issuecomment-2230087232
> > > > could you share the details of the exception stack ? > > > > > > > > > Sorry, I can't get the detail of exception stack from log, but I get the following log. `INFO Driver GlutenFallbackReporter: Validation failed for plan: Scan orc xxx, due to: native check failure:native validation failed due to: in ReadRel, constructSubfieldFilters not support for HUGEINT type.` > > > So I can write some parts of the exception stack: > > > ``` > > > SubstraitToVeloxPlanConverter::constructSubfieldFilters > > > SubstraitToVeloxPlanConverter::mapToFilters > > > SubstraitToVeloxPlanConverter::createSubfieldFilters > > > SubstraitToVeloxPlanConverter::toVeloxPlan > > > ``` > > > > > > that's fine. Actually if my understanding is right, with this PR , instead of fallback to vinalia spark due to unsupported type, the filter would be keep in remaining filter and still offload to native part right ? > > If yes, then this change is fine with me. The only consideration is we would have 2 duplicate check for `hugint` type. one is introduced by this PR. Antoher is here https://github.com/apache/incubator-gluten/blob/main/cpp/velox/substrait/SubstraitToVeloxPlan.cc#L2144-L2146, i would like to keep only one check in this PR and remove the check in line 2144- 2146. > > @PHILO-HE / @rui-mo for review and also double check if we still need this check or actually `hugint` type is supported in native and can be removed. > > This PR actually moves the check earlier, and we will remove the previous judgment logic. ok. thanks for clarify, then If there is always a fallback, curious what benefit does the check eailer bring? -- 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]
