According to Oracle's documentation (SQL reference), GROUP BY and HAVING
clause cannot have scalar subquery. However, I tested with Oracle 9.2
personal edition and found that the following query can be executed without
error. The query does contain a scalar subquery in its HAVING clause.
This query is actually taken from Oreilly's Mastering Oracle SQL 2nd edition
chapter 5 section 5.2.1.

SELECT sales_emp_id, COUNT(*)
FROM cust_order
GROUP BY sales_emp_id
HAVING COUNT(*) = (SELECT MAX(COUNT(*))
FROM cust_order
GROUP BY sales_emp_id);

Can anyone help me to explain the discrepancy?

It is a definite error in the manuals. I am not sure but I had thought
that the term scalar subquery had been applied to select statements in
the select list of another query: select col1, (select colb from table
2 ...), col3 when Oracle first introduced it. Subqueries of the form
in the example were just normal subqueries. The definition that
appears in the 9.2 and 10g SQL manual under the topic "Scalar Subquery
Expressions" makes no such distiction. So either the list about where
the scalar subqueries are legal is wrong or the definition of a scalar
subquery requires clarification.

IMHO -- Mark D Powell --