Github user kiszk commented on the issue: https://github.com/apache/spark/pull/19447 @gatorsmile I am happy to take it. We have [one test case](https://github.com/apache/spark/pull/16648/files#diff-c7f041fda7fd9aa1a5225f86bab4b1b0R69) that cause the error `JVM limit of 0xFFFF`. I am not sure this new option may help this or not. I will try it this long weekend (Monday is a holiday in Japan). Ideally, we want to find better metric rather than string length of the class file. Since constant pool entry is used for [various purposes](https://docs.oracle.com/javase/specs/jvms/se7/html/jvms-4.html#jvms-4.4), I imagine that the majorities in this cause is `CONSTANT_Fieldref`. (i.e. instance variable). It may be helpful to watch # of instance variables in `CodegenContext.mutableStates`. cc: @rednaxelafx @maropu it would be good to make `currClassSize` more precise.
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org