I am trying to implement a custom catalyst expression in Spark, which will parse each column in a dataframe into a string array. The toy example is attached.

case class CustomExpression(child: Expression)
  extends UnaryExpression {
  override def dataType: DataType = ArrayType(StringType)

  override protected def withNewChildInternal(newChild: Expression): Expression =
    copy(child = newChild)

  override def doGenCode(ctx: CodegenContext, ev: ExprCode): ExprCode = {
    // Generate the code to produce the desired output
    val digits = ctx.freshName("digits")
    val value = ctx.freshName("value")
    val code =
      code"""
         String[] $digits = new String[]{"1", "2", "3"};
         ${ev.value} = $digits;
         """

    ev.copy(code = code, isNull = FalseLiteral)
  }

and below is how to use it

val df = currentDf.withColumn(colName, new Column(new CustomExpression(col(colName).expr)))

when I want to display df with df.show(), the compiler give error File ‘generated.java’, Line 32, Column 1: Expression “value_1” is not an rvalue. When I check the generated code using df.explain(“codegen”), it gives the error File ‘generated.java’, Line 56, Column 1: Expression “project_value_0” is not an rvalue. The generateed code snippet is

/* 055 */         String[] project_digits_0 = new String[]{"1", "2", "3"};
/* 056 */         **project_value_0** = project_digits_0;
/* 057 */         String[] project_digits_1 = new String[]{"1", "2", "3"};
/* 058 */         project_value_2 = project_digits_1;
/* 059 */         columnartorow_mutableStateArray_3[1].reset();
/* 060 */
/* 061 */         if (false) {
/* 062 */           columnartorow_mutableStateArray_3[1].setNullAt(0);
/* 063 */         } else {
/* 064 */           // Remember the current cursor so that we can calculate how many bytes are
/* 065 */           // written later.
/* 066 */           final int project_previousCursor_0 = columnartorow_mutableStateArray_3[1].cursor();
/* 067 */
/* 068 */           final ArrayData project_tmpInput_0 = **project_value_0;**
/* 069 */           if (project_tmpInput_0 instanceof UnsafeArrayData) {
/* 070 */             columnartorow_mutableStateArray_3[1].write((UnsafeArrayData) project_tmpInput_0);
/* 071 */           } else {
/* 072 */             final int project_numElements_0 = project_tmpInput_0.numElements();

Acooding to the doGenCode() implementation, project_value_0 is the variable of the expression code value ${ev.value}. How to fix the error?

In the above toy example, I am expecting the output df with one column of a string array.