SQL does not allow aggregates of aggregates directly.
However, if you write the inner aggregate in a sub-query in the FROM clause (or use a WITH clause and a Common Table Expression, CTE), you can achieve the result:
SELECT gc1.CategoryName, gc1.SubCategoryName, gc1.BookCount
FROM (SELECT cat.CategoryName, sb.SubCategoryName,
COUNT(bs.BookID) AS BookCount
FROM Category AS cat
JOIN SubCategory AS sb ON cat.CategoryID = sb.CategoryID
JOIN Book_Subcategory AS bs ON sb.SubCategoryID = bs.SubCategoryID
GROUP BY cat.CategoryName, sb.SubCategoryName
) AS gc1
WHERE gc1.BookCount = (SELECT MAX(gc2.BookCount)
FROM (SELECT cat.CategoryName, sb.SubCategoryName,
COUNT(bs.BookID) AS BookCount
FROM Category AS cat
JOIN SubCategory AS sb
ON cat.CategoryID = sb.CategoryID
JOIN Book_Subcategory AS bs
ON sb.SubCategoryID = bs.SubCategoryID
GROUP BY cat.CategoryName, sb.SubCategoryName
) AS gc2
)
This is complex because it doesn't use a CTE, and there is a common table expression that must be written out twice.
Using the CTE form (possibly with syntax errors):
WITH gc1 AS (SELECT cat.CategoryName, sb.SubCategoryName,
COUNT(bs.BookID) AS BookCount
FROM Category AS cat
JOIN SubCategory AS sb
ON cat.CategoryID = sb.CategoryID
JOIN Book_Subcategory AS bs
ON sb.SubCategoryID = bs.SubCategoryID
GROUP BY cat.CategoryName, sb.SubCategoryName
)
SELECT gc1.CategoryName, gc1.SubCategoryName, gc1.BookCount
FROM gc1
WHERE gc1.BookCount = SELECT MAX(gc1.BookCount) FROM gc1);
Much tidier!
You can simulate a CTE with a temporary table if your DBMS makes it easy to create them. For example, IBM Informix Dynamic Server could use:
SELECT cat.CategoryName, sb.SubCategoryName,
COUNT(bs.BookID) AS BookCount
FROM Category AS cat
JOIN SubCategory AS sb ON cat.CategoryID = sb.CategoryID
JOIN Book_Subcategory AS bs ON sb.SubCategoryID = bs.SubCategoryID
GROUP BY cat.CategoryName, sb.SubCategoryName
INTO TEMP gc1;
SELECT gc1.CategoryName, gc1.SubCategoryName, gc1.BookCount
FROM gc1
WHERE gc1.BookCount = (SELECT MAX(gc1.BookCount) FROM gc1);
DROP TABLE gc1; -- Optional: table will be deleted at end of session anyway
Given the following tables and data, the main query (copied and pasted from this answer) gave the result I expected when run against IBM Informix Dynamic Server 11.50.FC6 on MacOS X 10.6.4, namely:
Non-Fiction SQL Theory 4
Fiction War 4
That doesn't prove that it 'must work' when run against Oracle - I don't have Oracle and can't demonstrate either way. It does show that there is at least one SQL DBMS that handles the query without problems. (Since IDS does not support the WITH clause and CTEs, I can't show whether that formulation works.)
Schema
CREATE TABLE Category
(
CategoryID INTEGER NOT NULL PRIMARY KEY,
CategoryName VARCHAR(20) NOT NULL
);
CREATE TABLE SubCategory
(
CategoryID INTEGER NOT NULL REFERENCES Category,
SubCategoryID INTEGER NOT NULL PRIMARY KEY,
SubCategoryName VARCHAR(20) NOT NULL
);
CREATE TABLE Book_SubCategory
(
SubCategoryID INTEGER NOT NULL REFERENCES SubCategory,
BookID INTEGER NOT NULL PRIMARY KEY
);
Data
INSERT INTO Category VALUES(1, 'Fiction');
INSERT INTO Category VALUES(2, 'Non-Fiction');
INSERT INTO SubCategory VALUES(2, 1, 'SQL Theory');
INSERT INTO SubCategory VALUES(2, 2, 'Mathematics');
INSERT INTO SubCategory VALUES(1, 3, 'Romance');
INSERT INTO SubCategory VALUES(1, 4, 'War');
INSERT INTO Book_SubCategory VALUES(1, 10);
INSERT INTO Book_SubCategory VALUES(2, 11);
INSERT INTO Book_SubCategory VALUES(3, 12);
INSERT INTO Book_SubCategory VALUES(3, 13);
INSERT INTO Book_SubCategory VALUES(4, 14);
INSERT INTO Book_SubCategory VALUES(1, 15);
INSERT INTO Book_SubCategory VALUES(1, 16);
INSERT INTO Book_SubCategory VALUES(2, 17);
INSERT INTO Book_SubCategory VALUES(1, 18);
INSERT INTO Book_SubCategory VALUES(3, 19);
INSERT INTO Book_SubCategory VALUES(4, 20);
INSERT INTO Book_SubCategory VALUES(4, 21);
INSERT INTO Book_SubCategory VALUES(4, 22);