Question

this is a follow on question to a previously asked question.

I have the following data in a single db table.

Name                LeftId    RightId
------------------------------------------
Cat                     1  
Cat                     1
Dog                     2
Dog                     2
Dog                               3
Dog                               3
Gerbil                  4         5 
Cat                
Bird
Cow                     6
Cow
Cow                               7
Dog                     8         9

Note that some rows have NO data for for LeftId and RightId.

Now, what i wish to do is find two different queries

1. All rows which have at least 1 Id in one of the two columns AND row with NO data in both of those two Id columns.

eg.

Cat     1
Cow     6 (or 7 .. i'm not worried)

2. All the rows where LeftId and RightId are NULL grouped by the same name. If another row (with the same name) has a value in the LeftId or RightId, this this name will not be returned.

eg.

Bird

hmm..

EDIT: Reworded the first question properly.

Answer 1

For the first query, you want rows that answer to both of the following criteria:

1. The Name in the row appears in the table in the same row in which LeftId and RightId are both NULL.
2. The Name in the row appears in the table in same row where at at least one of LeftId and RightId is not NULL.

Well, #1 is done by:

SELECT Name FROM Tbl WHERE (LeftId IS NULL) AND (RightId IS NULL)

And #2 is done by:

SELECT Name FROM Tbl WHERE (LeftId IS NOT NULL) OR (RightId IS NOT NULL)

You could intersect them to see which Names appear in both lists:

SELECT Name FROM Tbl WHERE (LeftId IS NULL) AND (RightId IS NULL)
INTERSECT
SELECT Name FROM Tbl WHERE (LeftId IS NOT NULL) OR (RightId IS NOT NULL)

Which returns:

Name
----
Cat
Cow

But you want the LeftId and RightId, and you don’t care which, so I guess we’ll aggregate on the Name:

SELECT Name, MIN(LeftId) AS LeftId, MIN(RightId) AS RightId 
    FROM Tbl WHERE Tbl.Name IN (
      SELECT Name FROM Tbl WHERE (LeftId IS NULL) AND (RightId IS NULL)
      INTERSECT
      SELECT Name FROM Tbl WHERE (LeftId IS NOT NULL) OR (RightId IS NOT NULL)
    )
GROUP BY Name

Which returns

Name  LeftId  RightId
----  ------  -------
Cat   1
Cow   6       7

lc already suggested using COALESE to turn those two IDs to a single one. So how about this:

SELECT Name, COALESCE(MIN(LeftId),MIN(RightId)) AS Id 
    FROM Tbl WHERE Tbl.Name IN (
      SELECT Name FROM Tbl WHERE (LeftId IS NULL) AND (RightId IS NULL)
      INTERSECT
      SELECT Name FROM Tbl WHERE (LeftId IS NOT NULL) OR (RightId IS NOT NULL)
    )
GROUP BY Name

Which returns:

Name  Id
----  --
Cat   1
Cow   6

---

For the second query, you want rows that obey the following criteria:

1. The Name appears only in rows that have no LeftId and RightId

I can’t think of a way to do that sort of self-referencing query in SQL in a single set of criteria, so I’ll break it down to two criteria. Both must be obeyed to be acceptable:

1. The Name appears in rows that have no LeftId and RightId
2. The Name does not appear in rows that have either LeftId or RightId

Doing #1 is simply:

SELECT Name FROM Tbl WHERE (LeftId IS NULL) AND (RightId IS NULL)

But #2 is tricky. Of course doing the opposite of #2 (“all the Name that appear in rows that have either LeftId or RightId) is just like before:

SELECT Name FROM Tbl WHERE (LeftId IS NOT NULL) OR (RightId IS NOT NULL)

Now comes the tricky bit – we want all the rows that obey #1 but don’t obey the opposite of #2. This is where EXCEPT is useful:

SELECT Name FROM Tbl WHERE (LeftId IS NULL) AND (RightId IS NULL)
EXCEPT
SELECT Name FROM Tbl WHERE (LeftId IS NOT NULL) OR (RightId IS NOT NULL)

Which returns:

Name
----
Bird

Which is what we wanted!