As of version 10.3.1, Item_func_case (when handling a simple CASE) stores the predicant argument (in the args[] array) after the WHEN..THEN arguments and before the ELSE argument.

For example, this expression:

CASE pred WHEN search1 THEN res1 WHEN search2 THEN res2 ELSE resE END

stores the arguments as follows:

• args[0]=search1
• args[1]=res1
• args[2]=search2
• args[3]=res2
• args[4]=pred
• args[6]=resE

Under terms of this task we'll do the following:
1. Change Item_func_case to store the arguments in the order of their appearance in the parser: the predicant argument in args[0], followed by the WHEN..THEN..ELSE arguments
2. Split Item_func_case into two separate classes Item_func_case_simple and Item_func_case_searched (for CASE expressions with and without predicant respectively).
3. Change the constructors of the affected classes just to accept a List<Item> argument (without additional first_expr_arg and else_expr_arg.

Advantages:

a. #1 and #3 will help to simplify the code in sql_yacc_ora.yy

| DECODE_SYM '(' expr ',' decode_when_list ')'

  {

    // 30 lines of the code, extracting {{else_expr_arg}} from the list.

  }

to something as simple as:

| DECODE_SYM '(' expr ',' decode_when_list ')'

  {

    $5->push_front($3, thd->mem_root);

    if (!($$= new (thd->mem_root) Item_func_case_simple(thd, *$5)))

      MYSQL_YYABORT;

  }

b. #1 and #3 will help to implement MDEV-13863 easier. MDEV-13836 will introduce a new class Item_func_decode_oracle. Without #1, we'd have to use the same complex code from sql_yacc_ora.yy, now for both DECODE() and DECODE_ORACLE().
c. #2 will slightly reduce the memory size required to handle searched CASE expression, because only Item_func_case_simple will need to derive from Predicant_to_list_comparator
d. #2 will make the code more readable by removing a lot of if statements like this:

if (first_expr_num != -1)

{

  ...

}

else

{

  ...

}

as the branches for the searched and the simple CASE expressions will reside in methods of different classes.
e. #1 will make it easier to debug the code in gdb, as printing the arguments in their syntactic order will be much easier than now.