fix: disable ANSI mode in benchmarks to avoid exceptions on invalid input

parthchandra · claude · parthchandra · commit 4ba0bcf4b128 · 2026-03-20T15:33:05.000-07:00
comet-test-apache-spark defaults spark.sql.ansi.enabled to true, causing
CAST on intentionally invalid benchmark data to throw instead of returning NULL.

Co-Authored-By: Claude Sonnet 4.6 &lt;noreply@anthropic.com&gt;
diff --git a/spark/src/test/scala/org/apache/spark/sql/benchmark/CometBenchmarkBase.scala b/spark/src/test/scala/org/apache/spark/sql/benchmark/CometBenchmarkBase.scala
@@ -63,6 +63,8 @@ trait CometBenchmarkBase
     sparkSession.conf.set(SQLConf.WHOLESTAGE_CODEGEN_ENABLED.key, "true")
     sparkSession.conf.set(CometConf.COMET_ENABLED.key, "false")
     sparkSession.conf.set(CometConf.COMET_EXEC_ENABLED.key, "false")
+    // Benchmarks use invalid input values that should produce NULL, not exceptions
+    sparkSession.conf.set(SQLConf.ANSI_ENABLED.key, "false")
 
     sparkSession
   }

Original file line number	Diff line number	Diff line change
`@@ -63,6 +63,8 @@ trait CometBenchmarkBase`
`63`	`63`	`sparkSession.conf.set(SQLConf.WHOLESTAGE_CODEGEN_ENABLED.key, "true")`
`64`	`64`	`sparkSession.conf.set(CometConf.COMET_ENABLED.key, "false")`
`65`	`65`	`sparkSession.conf.set(CometConf.COMET_EXEC_ENABLED.key, "false")`
	`66`	`+ // Benchmarks use invalid input values that should produce NULL, not exceptions`
	`67`	`+ sparkSession.conf.set(SQLConf.ANSI_ENABLED.key, "false")`
`66`	`68`
`67`	`69`	`sparkSession`
`68`	`70`	`}`