Question

I have been intrigued by a problem on SQLZoo. It is a “greatest-n-per-group” problem. I would like to understand how the engine is operating.

A table called bbc contains the name, region of the world and population of each country:

bbc( name, region, population)

The given task is to select the most populous country of each region, showing its name, the region and population.

The solution provided is:

SELECT region, name, population FROM bbc x
  WHERE population >= ALL
    (SELECT population FROM bbc y
        WHERE y.region=x.region
          AND population>0)

1. Main Question. I am finding this a bit of a mind twister. I would like to understand how the engine processes this, because at first blush it seems there is some kind of co-dependence (x depending on y, and y depending on x). Does the engine follow some kind of recursion to produce the final selection? Or am I missing something, such that either x or y is actually fixed?

2. Secondary Question. Oddly, when I pull the “AND population>0” out of the parenthesis and leave it on its own at the bottom, one of the regions (Europe / Russia) goes missing from the 8 results. Why? I don’t understand that.

And indeed, when I try the query on the world database (available from the mySQL website on the same page as Sakila), the behavior is different:
With population > 0 out of the parentheses, I get 6 regions. Six is the right number in this database, because “SELECT continent FROM country GROUP BY continent” reveals seven continents, of which one is Antarctica, which includes 5 “countries”, all with a 0 population.

So that seems right.

SELECT continent, `name`, population FROM country X
WHERE population >= ALL 
(SELECT population FROM country Y
WHERE Y.`Continent` = X.`Continent`)
AND population>0

On the other hand, when I pull “population > 0” back into the parentheses as on SQLZoo, I also get 5 countries with a zero (the countries “belonging to Antarctica”). It doesn’t matter if I specify x.population or y.population, I get zeroes.

continent      name                                          population  
-------------  --------------------------------------------  ------------
Antarctica     Antarctica                                               0
Antarctica     French Southern territories                              0
Oceania        Australia                                         18886000
South America  Brazil                                           170115000
Antarctica     Bouvet Island                                            0
Asia           China                                           1277558000
Antarctica     Heard Island and McDonald Islands                        0
Africa         Nigeria                                          111506000
Europe         Russian Federation                               146934000
Antarctica     South Georgia and the South Sandwich Islands             0
North America  United States                                    278357000

Very much looking for insights on these questions!

Wishing you all a beautiful week.

🙂

Notes:

1. For reference, the problem is number 3a on this page:
   http://old.sqlzoo.net/1a.htm?answer=1

2. A thread mentioning the “greatest-n-per-group” problem for the same query:
   MySQL world database Trying to avoid subquery

3. The world database is available here: http://dev.mysql.com/doc/index-other.html

Answer 1

1. Main Question. I am finding this a bit of a mind twister. I would like to understand how the engine processes this, because at first
   blush it seems there is some kind of co-dependence (x depending on y,
   and y depending on x). Does the engine follow some kind of recursion
   to produce the final selection? Or am I missing something, such that
   either x or y is actually fixed?

This isn’t recursion. See this from the MySQL docs. Their solution to the problem is equivalent to this

SELECT region, name, population FROM bbc x
  WHERE population = 
    (SELECT max(population) FROM bbc y
        WHERE y.region=x.region
          )

1. Secondary Question. Oddly, when I pull the “AND population>0” out of the parenthesis and leave it on its own at the bottom, one of the
   regions (Europe / Russia) goes missing from the 8 results. Why? I
   don’t understand that.

Slight changes (as suggested by ypercube above) work

SELECT region, name, population FROM bbc x
  WHERE population >= ALL
    (SELECT population FROM bbc y
        WHERE y.region=x.region
          AND population IS NOT NULL)

This query

SELECT region, name, population FROM bbc x
  WHERE population is null

Returns a row. Not sure why population should be nullable, but didn’t take a good look at the rest of it. Otherwise, the query should work fine without the >0

Also, this is different from the greatest-n-per-group. In that problem you seek to find the top N items instead of just the top one.