Re: [NOVICE] Question About Aggregate Functions - Mailing list pgsql-general

Ah, I did not know what was in your fields, so I did not assume they were Boolean values.  It looked to me like you were trying to use IS TRUE to substitute for the lack of a GROUP BY, so I didn't know what to do.

Yes, count() will include all non-NULL values.  Sorry if I sounded unclear there. 

 

If you do typecasting the value zero is false (and non-zero is true).  NULL in an expression always returns NULL, and many programs will interpret that result as false.  So I'm not sure of what results you might get with a Boolean test against a non-Boolean field, especially if it's an integer field.

 

postgres=# select 0::boolean = FALSE;

 ?column?

----------

 t

(1 row)

 

You should just be able to take the previous query and add in your WHERE clauses:

 

SELECT count(t1.fielda), count(t2.fielda), count(t2.fieldb), AVG(t2.fieldc)

FROM t1 JOIN t2 ON t1.item_id = t2.item_id

WHERE t1.fielda = TRUE AND t2.fielda = TRUE AND t2.fieldb = TRUE

GROUP BY NULL;

 

Now, the INNER JOIN you're using is only selecting fields where both t1.item_id and t2.item_id exist and the respective fields are TRUE.  That is, it's only going to run the count and average functions against the results of this query:

SELECT *

FROM t1 JOIN t2 ON ON t1.item_id = t2.item_id

WHERE t1.fielda = TRUE AND t2.fielda = TRUE AND t2.fieldb = TRUE;

 

If that's what you want, that's great.  

However, you might want a count of each field where that field is TRUE.  In that case, I would use either temporary tables, compound queries and derived tables, or multiple simple queries. 

It's also possible that you might want a count of fields where t1.item_id and t2.item_id exist, but where only each respective field is TRUE.  That is, you want a count of t1.fielda where it is TRUE no matter what t2.fielda and t2.fieldb are as long as t1.item_id matches t2.item_id.  In that case you have to do even more joins, and that could take a fair bit of time especially if you haven't indexed your item_id fields.

You really have to look at your result sets.  Sometimes it is better to run multiple simple queries instead of one big complex query to be sure you're getting the data you want and the query executes in a reasonable amount of time.

 

Also, consider that NULL values are generally considered bad to purposefully enter.  Logically, It would be better to create one table for each field and then create a record for each item_id as you need it so you never have NULLs.  The problem with that is one of performance if you end up doing large number of JOINs.  In that case, it might be better to use integers instead of Boolean fields, since you have three explicit states of TRUE, FALSE, and NOT YET DETERMINED.