Question

I have three related tables:

• Course
• Student
• Teacher

Each course is given by a teacher to many students.

I can find the number of students attending a course:

SELECT c.id, count(*) FROM course as c, student as s
WHERE c.id = s.course_id
GROUP BY c.id;

I can find all the courses given by a teacher:

SELECT t.id, c.id FROM course as c, teacher as t
WHERE t.id = c.teacher_id;

I want to find how many students were taught by each teacher. Is the following query correct one?

SELECT t.id, sum(x.count_s) 
FROM 
   (SELECT count(*) AS count_s FROM course as c2, student as s
      WHERE c2.id = s.course_id AND c2.id = c.id
      GROUP BY c2.id) as x, 
   course as c, teacher as t
WHERE t.id = c.teacher_id;

Unfortunately, I cannot test it directly because this is actually a simplification of the real problem at hand. I need to find out which solution works for the simplified problem.

Answer 1

To answer your question, no. You cannot reference c.id inside the inline view aliased as x. That should throw an error.

But if you remove that, then your query has the potential to return an inflated count, due to a semi-Cartesian product, between the inline view aliased as x and c.

So that predicate needs to be relocated, and you’d need to return c2.id from x (i.e. add it to the SELECT list, you already have it referenced in the GROUP BY).

This is equivalent to your query, just rewritten to replace the comma join operators and relocate join predicates to ON clause. This statement is equivalent to yours, and is invalid):

SELECT t.id
     , SUM(x.count_s) 
  FROM ( SELECT count(*) AS count_s 
           FROM course c2
           JOIN student s
             ON c2.id = s.course_id
            AND c2.id = c.id        -- INVALID here
          GROUP
             BY c2.id
       ) x
 CROSS                              -- no join predicate 
  JOIN course c
  JOIN teacher t
    ON t.id = c.teacher_id

To fix that, add c2.id to the SELECT list in x, and relocate that predicate. Something like this:

SELECT t.id
     , SUM(x.count_s) 
  FROM ( SELECT count(*) AS count_s
              , c2.id                 -- added
           FROM course c2
           JOIN student s
             ON c2.id = s.course_id
       --   AND c2.id = c.id          -- relocated (removed from here)
          GROUP
             BY c2.id
       ) x
  JOIN course c
    ON x.id = c.id                    -- relocated (added here)
  JOIN teacher t
    ON t.id = c.teacher_id

Assuming that id is UNIQUE and NOT NULL in course, that query will return a reasonable count (although counts of zero will be “missing”).

To return the “zero” counts, you’d need to use an OUTER join. And as I always prefer to use LEFT JOIN, the tables/inline views in the outermost query would need to be re-ordered:

SELECT t.id
     , IFNULL(SUM(x.count_s),0)
  FROM teacher t
  LEFT
  JOIN course c
    ON c.teacher_id = t.id
  LEFT
  JOIN ( SELECT count(*) AS count_s
              , c2.id                 -- added
           FROM course c2
           JOIN student s
             ON c2.id = s.course_id
       --   AND c2.id = c.id          -- relocated (removed from here)
          GROUP
             BY c2.id
       ) x
    ON x.id = c.id                    -- relocated (added here)

Assuming that id is a PRIMARY KEY (or equivalent UNIQUE and NOT NULL) on each table, then that will return a “correct” count.

It’s not necessary to include the course table in the inline view aliased as x. It would be sufficient to GROUP BY s.course_id.

SELECT t.id
     , IFNULL(SUM(x.count_s),0)
  FROM teacher t
  LEFT
  JOIN course c
    ON c.teacher_id = t.id
  LEFT
  JOIN ( SELECT count(*) AS count_s
              , s.course_id 
           FROM student s
          GROUP 
             BY s.course_id
       ) x
    ON x.course_id = c.id                 -- relocated (added here)

That query will return a valid “count”.

---

A simpler statement would easier to understand. Here’s how I would get the count:

SELECT t.id        AS teacher_id
     , COUNT(s.id) AS how_many_students_taught
  FROM teacher t
  LEFT
  JOIN course c
    ON c.id = t.course_id
  LEFT
  JOIN student s
    ON s.course_id = c.id
 GROUP
    BY t.id

---