Question

I’d like to perform a group-concat in SQLite only on those records where there is more than one row to concatenate. It seems like you could do this beforehand (count records using a group by then remove those singleton rows before proceeding with the group_concat); after (complete the group_concat then remove rows where nothing was concatenated); or possibly even during?

My question: what’s the fastest way for SQLite to accomplish this?

I’ve worked out an “after” example using APSW in Python, but am not happy with it:

#set up a table with data
c.execute("create table foo(x,y)")
def getvals():
    a = [1, 1, 2, 3, 3]
    b = ['a','b','c','d','e']
    for i in range(5):
        yield a[i],b[i]
c.executemany("insert into foo values(?,?)",getvals())
c.execute('''create table fooc(a,b);
    insert into fooc(a,b) select x, group_concat(y) from foo group by x''')
c.execute('select * from fooc')
c.fetchall()  ## reports three records
c.execute("select * from fooc where b like '%,%'")
c.fetchall() ## reports two records .. what I want

It seems crazy (and slow?) to use LIKE for this kind of need.

Answer 1

Add a HAVING clause to your query:

INSERT INTO fooc(a,b) 
     SELECT x, group_concat(y) 
       FROM foo 
   GROUP BY x
     HAVING COUNT(*) > 1