Item_sum_num has features of DOUBLE data type by default and implements DOUBLE-applicable methods:

• val_int()
• val_str()
• val_decimal()

These methods then are overriden to INT-applicable and hybrid-applicable as follows:

• Item_sum_int overrides these methods to INT-applicable.
• Item_sum_window_with_row_count derives DOUBLE-applicable implementations from Item_sum_num.
• Item_sum_ntile derives DOUBLE-applicable implementations from Item_sum_window_with_row_count and overrides val_real() to INT-based implementation. Note, Item_sum_ntile does not override val_decimal() and val_str(), which looks wrong because they work through val_real() by default and therefore may return truncated results on a huge data sets.

• Item_sum_cume_dist derives from Item_sum_window_with_row_count.

• Item_sum_percentile_disc derives from Item_sum_cume_dist and Type_handler_hybrid_field_type and overrides value methods to hybrid-applicable, except get_date().

• Item_sum_percentile_cont derives from Item_sum_cume_dist and Type_handler_hybrid_field_type, but in fact it can olny have the DOUBLE data type.

This multiple overriding is hard to follow. Let's introduce a new abstract class Item_sum_double and change the Item_sum_num hierarchy as follows:

Item_sum_num

  Item_sum_percentile_disc  -- hybrid type

  Item_sum_sum              -- hybrid type

    Item_sum_avg

  Item_sum_double           -- DOUBLE

    Item_sum_percent_rank

    Item_sum_cume_dist

    Item_sum_percentile_cont

    Item_sum_variance

      Item_sum_std

    Item_sum_udf_float

    Item_sum_udf_int

    Item_sum_udf_decimal

    Item_sum_udf_str

  Item_sum_int              -- INT

    Item_sum_row_number

    Item_sum_rank

    Item_sum_dense_rank

    Item_sum_ntile

    Item_sum_count

    Item_sum_bit

      Item_sum_or

      Item_sum_and

      Item_sum_xor

so that the method type_handler() and value methods like val_int(), val_real(), val_decimal(), get_date() are implemented inside the same class, without multiple nested overriding.