Question

It appears that when using a having clause a count(*) is evaluated before the having but the number of rows returned is evaluated after. The fix is probably sub-queries but if it’s possible to avoid it I’d like to. I’m using count(*) to avoid no_data_found.

This behaviour occurs in 11.2.0.1.0, 10.2.0.1.0, 9.2.0.7.0 so it’s obviously intended but I don’t quite understand why. An easily replicable example is below.

Does anyone know why this happens? I would have expected the count(*) to return 1.

create table tmp_test1 as
  select level as id, level as val
    from dual
 connect by level <= 1000
         ;

Table created.

create table tmp_test2 as
  select level as id, level as val
    from dual
 connect by level <= 1000
         ;

Table created.

select count(*) as count
  from tmp_test1 a
  join tmp_test2 b
    on a.id = b.id
having max(a.val) = max(b.val)
       ;

     COUNT
----------
      1000

select 1 as num_rows
  from tmp_test1 a
  join tmp_test2 b
    on a.id = b.id
having max(a.val) = max(b.val)
       ;

  NUM_ROWS
----------
         1

Answer 1

The having clause operates at the group level – where no group by is specified, this means that it is operating across the entire dataset.

This means that in any query with a having clause and no group by clause, the results can only return 0 rows (if the having condition is false) or 1 row (if the having condition is true).