Question

My problem is Play! Framework / JPA specific. But I think it’s applicable to general SQL syntax.

Here is a sample query with a simple JOIN:

return Post.find(
    "select distinct p from Post p join p.tags as t where t.name = ?", tag
).fetch();

It’s simple and works well.

My question is: What if I want to JOIN on more values in the same table?

Example (Doesn’t work. It’s a pseudo-syntax I created):

return Post.find(
    "select distinct p from Post p join p.tags1 as t, p.tags2 as u, p.tags3 as v where t.name = ?, u.name = ?, v.name = ?", tag1, tag2, tag3,
).fetch();

Answer 1

Your programming logic seems okay, but the SQL statement needs some work. Seems you’re new to SQL, and as you pointed out, you don’t seem to understand what a JOIN is.

You’re trying to select data from 4 tables named POST, TAG1, TAG2, and TAG3.

I don’t know what’s in these tables, and it’s hard to give sample SQL statements without that information. So, I’m going to make something up, just for the purposes of discussion. Let’s say that table POST has 6 columns, and there’s 8 rows of data in it.

P Fname Lname   Country Color Headgear
- ----- -----   ------- ----- --------
1 Alex  Andrews 1       1     0
2 Bob   Barker  2       3     0
3 Chuck Conners 1       5     0
4 Don   Duck    3       6     1
5 Ed    Edwards 2       4     2
6 Frank Farkle  4       2     1
7 Geoff Good    1       1     0
8 Hank  Howard  1       3     0

We’ll say that TAG1, TAG2, and TAG3 are lookup tables, with only 2 columns each. Table TAG1 has 4 country codes:

C Name
- -------
1 USA
2 France
3 Germany
4 Spain

Table TAG2 has 6 Color codes:

C Name
- ------
1 Red
2 Orange
3 Yellow
4 Green
5 Blue
6 Violet

Table TAG3 has 4 Headgear codes:

C Name
- -------
0 None
1 Glasses
2 Hat
3 Monacle

Now, when you select data from these 4 tables, for P=6, you’re trying to get something like this:

Fname Lname  Country Color  Headgear
----- ------ ------- ------ -------
Frank Farkle Spain   Orange None

First thing, let’s look at your WHERE clause:

where t.name = ?, u.name = ?, v.name = ?

Sorry, but using commas like this is a syntax error. Normally you only want to find data where all 3 conditions are true; you do this by using AND:

where t.name=? AND u.name=? AND v.name=?

Second, why are you joining tables together? Because you need more information. Table POST says that Frank’s COUNTRY value is 4; table TAG1 says that 4 means Spain. So we need to “join” these tables together.

The ancient (before 1980, I think) way to join tables is to list more than one table name in the FROM clause, separated by commas. This gives us:

SELECT P.FNAME, P.LNAME, T.NAME As Country, U.NAME As Color, V.NAME As Headgear
FROM POST P, TAG1 T, TAG2 U, TAG3 V

The trouble with this query is that you’re not telling it WHICH rows you want, or how they relate to each other. So the database generates something called a “Cartesian Product”. It’s extremely rare that you want a Cartesian Product – normally this is a HUGE MISTAKE. Even though your database only has 22 rows in it, this SELECT statement is going to return 768 rows of data:

Alex Andrews USA   Red    None
Alex Andrews USA   Red    Glasses
Alex Andrews USA   Red    Hat
Alex Andrews USA   Red    Monacle
Alex Andrews USA   Orange None
Alex Andrews USA   Orange Glasses
...
Hank Howard  Spain Violet Monacle

That’s right, it returns every possible combination of data from the 4 tables. Imagine for a second that the POST table eventually grows to 20000 rows, and the three TAG tables have 100 rows each. The whole database would be less than a megabyte, but the Cartesian Product would have 20,000,000,000 rows of data — probably about 120 GB of data. Any database engine would choke on that.

So if you want to use the Ancient way of specifying tables, it is VERY IMPORTANT to make sure that your WHERE clause shows the relationship between every table you’re querying. This makes a lot more sense:

SELECT P.FNAME, P.LNAME, T.NAME As Country, U.NAME As Color, V.NAME As Headgear
FROM POST P, TAG1 T, TAG2 U, TAG3 V
WHERE P.Country=T.C AND P.Color=U.C AND P.Headgear=V.C

This only returns 8 rows of data.

Using the Ancient way, it’s easy to accidentally create Cartesian Products, which are almost always bad. So they revised SQL to make it harder to do. That’s the JOIN keyword. Now, when you specify additional tables you can specify how they relate at the same time. The New Way is:

SELECT P.FNAME, P.LNAME, T.NAME As Country, U.NAME As Color, V.NAME As Headgear
FROM POST P
INNER JOIN TAG1 T ON P.Country=T.C
INNER JOIN TAG2 U ON P.Color=U.C
INNER JOIN TAG3 V ON P.Headgear=V.C

You can still use a WHERE clause, too.

SELECT P.FNAME, P.LNAME, T.NAME As Country, U.NAME As Color, V.NAME As Headgear
FROM POST P
INNER JOIN TAG1 T ON P.Country=T.C
INNER JOIN TAG2 U ON P.Color=U.C
INNER JOIN TAG3 V ON P.Headgear=V.C
WHERE P.P=?

If you call this and pass in the value 6, you get only one row back:

Fname Lname  Country Color  Headgear
----- ------ ------- ------ --------
Frank Farkle Spain   Orange None