Question

I have 2 tables: items and items_purchased. The tag name and other descriptive information is in the items table and the # of purchases for each item is saved in the items_purchased table. The tables are JOINed on item_id.

Here’s my query so far:

(select *, sum(purchaseyesno) as tots from items join items_purchased on items.item_id=items_purchased.item_id where tag = 'History' and deleted!=1 and pub_priv!=1    order by tots desc limit 1)
UNION ALL
(select *, sum(purchaseyesno) as tots from items join items_purchased on items.item_id=items_purchased.item_id where tag = 'Medicine' and deleted!=1 and pub_priv!=1  order by tots desc limit 1)
UNION ALL
(select *, sum(purchaseyesno) as tots from items join items_purchased on items.item_id=items_purchased.item_id where tag = 'Biology' and deleted!=1 and pub_priv!=1  order by tots desc limit 1)
//... more tags (20 in all) would follow

The problem is that items with certain tags haven’t been purchased yet so the query throws an error of Column "item_id" cannot be null.

here’s the sample tables:

            items              |     items_purchased   |   
  item_id   tag      title     | item_id  purchaseyesno|
    1     Biology  DNA is cool |    1         1        | 
    2     Medicine  Doctors    |    2         1        |  
    3      Law    Laws are cool|    4         1        |    
    4     Biology DNA NOT cool |    1         1        |

Sample results:

item_id   tag       title      tots
  1      Biology   DNA is cool  2
  2      Medicine  Doctors      1
  3       Law   Laws are cool   0

2 questions:

1. How can I exclude one of these SELECT statements if the results of it will be NULL?
2. Is there a better way to do this query rather than 20 select statements joined by 19 UNION ALLs?

Answer 1

You simply want to group your results—nothing more complicated than that (no UNION in sight!):

SELECT   items.*, SUM(purchaseyesno) AS tots
FROM     items JOIN items_purchased USING (item_id)
GROUP BY item_id

---

UPDATE

Following the comments beneath and the update to the OP’s question, it’s clear that the problem is more complicated than first thought: it is, in fact, a special case of the group-wise maximum problem.

However, the column over which the maximum must be taken is itself the result of another group-wise aggregation (the summation of purchaseyesno).

Therefore, one’s query becomes rather inelegant with subqueries:

SELECT *
FROM (
  SELECT   items.*, IFNULL(SUM(purchaseyesno),0) AS tots
  FROM     items LEFT JOIN items_purchased ON folder_id = item_id
  GROUP BY item_id
) AS t NATURAL JOIN (
  SELECT tag, MAX(tots) AS tots
  FROM (
    SELECT   items.*, IFNULL(SUM(purchaseyesno),0) AS tots
    FROM     items LEFT JOIN items_purchased ON folder_id = item_id
    GROUP BY item_id
  ) AS u
  GROUP BY tag
) AS v
GROUP BY tag

See it on sqlfiddle.

What’s happening here is that (using an outer join to include records such as Law that have no references in the items_purchased table), we take the sum of purchases for each item (materialised table u) and then determine the maximum number of purchases for items with the same tag name.

We then join the result (materialised table v) back with the first table (u, materialised again as table t) on both tag name and number of purchases.

Finally we group the results again by tag to ensure that, if multiple items having the same tag have the same number of maximal purchases, only one of them is returned indeterminately.