Process distinct comma-separated value in oracle - sql

Question

I have a column with the following data:
Brand
-------------
Audi, Opel, Ford
Skoda, Renault
Audi, BMW
Audi, Volkswagen, Opel
Toyota, Hyundai
I would like to have query which automates assign the data into group as following:
Brand
-------------------
Audi, Opel, Ford, BMW, Volkwagen
Skoda, Renault
Toyota, Hyundai
Note that if we insert another record into the table like this ...
Toyota, BMW
... the required output would be:
Brand
-------------------
Audi, Opel, Ford, BMW, Volkwagen, Toyota, Hyundai
Skoda, Renault

Answer 1

This is an interesting and difficult problem, obscured by your poor data model (which violates First Normal Form). Normalizing the data - and de-normalizing at the end - is trivial, it's just an annoyance (and it will make the query much slower). The interesting part: the input groups are the nodes of a graph, two nodes are connected if they have a "make" in common. You need to find the connected components of the graph; this is the interesting problem.
Here is a complete solution (creating the testing data on the fly, in the first factored subquery in the with clause). Question for you though: even assuming that this solution works for you and you put it in production, who is going to maintain it in the future?
EDIT It occurred to me that my original query can be simplified. Here is the revised version; you can click on the Edited link below the answer if you are curious to see the original version.
with
sample_data (brand) as (
select 'Audi, Opel, Ford' from dual union all
select 'Skoda, Renault' from dual union all
select 'Audi, BMW' from dual union all
select 'Audi, Volkswagen, Opel' from dual union all
select 'Toyota, Hyundai' from dual union all
select 'Tesla' from dual
)
, prep (id, brand) as (
select rownum, brand
from sample_data
)
, fnf (id, brand) as (
select p.id, ca.brand
from prep p cross apply
( select trim(regexp_substr(p.brand, '[^,]+', 1, level)) as brand
from dual
connect by level <= regexp_count(p.brand, '[^,]+')
) ca
)
, g (b1, b2) as (
select distinct fnf1.brand, fnf2.brand
from fnf fnf1 join fnf fnf2 on fnf1.id = fnf2.id
)
, cc (rt, brand) as (
select min(connect_by_root b1), b2
from g
connect by nocycle b1 = prior b2
group by b2
)
select listagg(brand, ', ') within group (order by null) as brand
from cc
group by rt;
Output:
BRAND
---------------------------------------------
Audi, BMW, Ford, Opel, Volkswagen
Hyundai, Toyota
Renault, Skoda
Tesla

Answer 2

That is standard Connected components problem. You can find fast pl/sql solution for production use here: http://orasql.org/2017/09/29/connected-components/
Or in case of just educational purposes, you can use SQL-only solution:
https://gist.github.com/xtender/b6e5cac4dec461c0121145b0e62c5cf5
with t(Brand) as (
select 'Audi, Opel, Ford' brand from dual union all
select 'Skoda, Renault' from dual union all
select 'Audi, BMW' from dual union all
select 'Audi, Volkswagen, Opel' from dual union all
select 'Toyota, Hyundai' from dual union all
select 'Tesla' from dual union all
select 'A'||level||', A'||(level+1) from dual connect by level<=500 union all
select 'B'||level||', B'||(level+1) from dual connect by level<=500 union all
select 'C'||level||', C'||(level+1) from dual connect by level<=500
)
,split_tab as (
select
dense_rank()over(order by t.brand) rn
,x.*
from t,
xmltable(
'ora:tokenize(concat(",",.),",")[position()>1]'
passing t.brand
columns
n for ordinality
,name varchar2(20) path 'normalize-space(.)'
) x
)
,pairs as (
select
t1.rn, t1.name name1, t2.name name2
from split_tab t1
,split_tab t2
where t1.rn=t2.rn
)
select listagg(x,',')within group(order by x)
from (
select x, min(root) grp
from (
select distinct connect_by_root(name1) root, name1 x
from pairs
connect by nocycle
prior name1 = name2
)
group by x
)
group by grp
/
PS. I've split my solution into smallest possible steps, so you can check each CTE separately step-by-step to view how to get results.