Note - numbers do not relate to your questions. This pretty much answers 1
and 3.
1. I'd ensure that project name/id on a table can never be null by creating
a dummy project that means "none assigned".
2. I would avoid full join. To do so I'd cross join a distinct list of
companies with a distinct list of projects.
3. Against the join in 2 you then left join three times, once for each
product table.
4. In the final result any product not having a correspond company/project
would have its value coalesced to zero.
The end result is a table without any NULL and, in the case of your
explanatory data, 9 rows - two of which would be all zeros: (c3,p1) and
(c3,p,2)
If you cannot change the raw data I'd use CTE/WITH to normalize the data
according to 1 and the use these CTEs in the rest of the query. I would
also do 2 is a CTE then 3 would be normal subqueries or relation references
as necessary.
David J.
--
View this message in context:
http://postgresql.1045698.n5.nabble.com/Problem-with-duplicate-rows-when-FULL-OUTER-JOIN-ing-3-derived-tables-tp5806970p5807003.html
Sent from the PostgreSQL - sql mailing list archive at Nabble.com.
