在 Oracle SQL 中选择唯一行并忽略空值

Question

There is a table with four columns: client, city, postcode and street. For each client I want to count the number of unique addresses. Unfortunately, there can be nulls in some of the columns: city, postcode or street. I have to ignore them when comparing for count distinct. So this can't be solved by simple group by and count distinct.

For example, there are

'client1', 'city1', 'postcode1', 'street1'
'client1', 'city1', 'postcode1', null
'client1', 'city1', null, 'street1'
'client1', null, null, 'street2'

'client1', 'city2', null, 'street1'
'client1', 'city2', null, 'street2'

For my task, the unique addresses should be ( edited )

'client1', 'city1', 'postcode1', 'street1'

'client1', 'city2', null, 'street1'
'client1', 'city2', null, 'street2'

(so the answer is 3 unique addresses for client1),
but for a standard distinct clause these are all unique, eg, rows

'client1', 'city1', 'postcode1', 'street1'
'client1', 'city1', 'postcode1', null
'client1', 'city1', null, 'street1'

are treated as different, whereas for my task, these are not different and I want to count them as 1.

Edit after some comments: If we had

'client1', null, null, 'street3'

then this is a unique address (since there are no other addresses with 'street3') and should be counted in.

Answer 1

You can use the min analytical function as follows:

Select distinct t.client,
       t.city,
       Coalesce(t.postcode,Min(t.postcode) over (partition by t.client, t.city)) as postcode,
       Coalesce(t.street,Min(t.street) over (partition by t.client, t.city)) as street
  From your table
 Where city is not null;

-- update

I can think of self hoin solution, check if it works for you.

Select distinct a.client,
       Coalesce(a.city, b.city) as city,
       Coalesce(a.postcode, b.postcode) as postcode,
       Coalesce(a.street, b.street) as street
  From your_table a left join your_table b
    On a.client = b.client
   And (a.city = b.city or (a.city is null or b.city is null))
   And (a.postcode = b.postcode or (a.postcode is null or b.postcode is null))
   And (a.street = b.street or (a.street is null or b.street is null))
   And a.rowid <> b.rowid
       

Answer 2

I've solved my problem in PL/SQL. The code is quite lengthy, so I will give only an outline of the idea in case someone is interested.

1. We create a table with distinct tuples. This is our input table.

2. We insert to our output table tuples with three not null values as they will not be covered by any other tuples with missing values.

3. We insert to our output table tuples with two not null values. Here we filter the ones that are covered by tuples from 2.

4. We insert to our output table tuples with one not null value. Here we filter the ones that are covered by tuples from 2 and 3.

I think that there are no counterexamples for this solution.