CUME_DIST is a bit special. It needs to either

1. remember a value for each peer group somewhere,
2. three cursors.

Details:

create table t1 (

  pk int, 

  a int

);

insert into t1 values

(1, 20),

(2, 20),

(3, 20),

(4, 20),

(5, 30),

(6, 30),

(7, 30),

(8, 40);

select pk,a,cume_dist() over (order by a) from t1;

+------+------+

| pk   | a    |

+------+------+

|    1 |   20 | 0.5

|    2 |   20 | 0.5

|    3 |   20 | 0.5

|    4 |   20 | 0.5

|    5 |   30 | 0.875

|    6 |   30 | 0.875

|    7 |   30 | 0.875

|    8 |   40 | 1

+------+------+

How to compute these?
1. The first pass needs to find how many rows are in the partition.
(Q: BTW, is it true that "any two-pass window function needs the first pass
only to find #rows in the partition"?)

2. On the second pass, we need
2.1 find how many peers are in the peer group (5 rows have a=20)
2.2 compute the value and set it for every peer member.

Possible ways to do step#2:
A: make it in two passes:

• first, let the first row of the group store #values in the group
• second, propagate this value to all group members.

B: Use two cursors.

• The front cursor reaches the end of the peer group, remembers how many values
  it saw and then stays there.
• The back cursor walks through the peer group and updates the values.