Oracle: How functional table works in cross join? - oracle

Question

I explored technique to unfold comma separated list in column into rows:
with tbl as (
select 1 id, 'a,b' lst from dual
union all select 2 id, 'c' lst from dual
union all select 3 id, 'e,f,g' lst from dual)
select
tbl.ID
, regexp_substr(tbl.lst, '[^,]+', 1, lvl.column_value) elem
, lvl.column_value lvl
from
tbl
, table(cast(multiset(
select level from dual
connect by level <= regexp_count(tbl.lst, ',')+1) as sys.odcinumberlist)) lvl;
Result is:
ID ELEM LVL
1 a 1
1 b 2
2 c 1
3 e 1
3 f 2
3 g 3
As you can see LVL depends on value of regexp_count, so second functional table in cross join is parametrized by first table.
How is it working? How is it called? Can I paramertize third table based on two preceding in cross join and so forth?
Is parametrization limited to cross join or can be applied in join syntax too?
Reference: Splitting string into multiple rows in Oracle

Answer 1

From the documentation:
LATERAL
Specify LATERAL to designate subquery as a lateral inline
view. Within a lateral inline view, you can specify tables that appear
to the left of the lateral inline view in the FROM clause of a query.
You can specify this left correlation anywhere within subquery (such
as the SELECT, FROM, and WHERE clauses) and at any nesting level.
-- a variation of the query in your question ...
select
dt.id
, dt.list
, regexp_substr( dt.list, '[^,]+', 1, dt2.lvl ) elements
, dt2.lvl
from (
select 1 id, 'a,b' list from dual union all
select 2, 'c' from dual union all
select 3, 'e,f,g' from dual
) dt, lateral (
select level lvl from dual
connect by level <= regexp_count(dt.list, ',') + 1
) dt2
;
-- output
ID LIST ELEMENTS LVL
1 a,b a 1
1 a,b b 2
2 c c 1
3 e,f,g e 1
3 e,f,g f 2
3 e,f,g g 3
Example with 3 tables:
--drop table t1 ;
--drop table t2 ;
--drop table t3 ;
-- tables/data
create table t1
as
select 1 id, 'a' letter from dual union all
select 2, 'b' from dual union all
select 3, 'c' from dual ;
create table t2
as
select 1 id, 'd' letter from dual union all
select 2, 'e' from dual union all
select 3, 'f' from dual ;
create table t3
as
select 1 id, 'g' letter from dual union all
select 2, 'h' from dual union all
select 3, 'i' from dual ;
-- query
select *
from
t1
, lateral ( select letter from t2 where id = t1.id ) t2
, lateral ( select letter from t3 where id = t2.id )
;
-- output
ID LETTER LETTER LETTER
1 a d g
2 b e h
3 c f i
Also (using the same tables)
-- reference t1 <- t2,
-- reference t1 and t2 <- t3
select *
from
t1
, lateral ( select letter from t2 where id = t1.id ) t2
, lateral ( select letter || t1.letter from t3 where id = t2.id )
;
-- output
ID LETTER LETTER LETTER||T1.LETTER
1 a d ga
2 b e hb
3 c f ic
Whereas a "standard" cross join would give us ...
select *
from
t1 cross join t2 cross join t3
;
ID LETTER ID LETTER ID LETTER
1 a 1 d 1 g
1 a 1 d 2 h
1 a 1 d 3 i
1 a 2 e 1 g
1 a 2 e 2 h
1 a 2 e 3 i
...
-- 27 rows
Related topics: CROSS APPLY (see documentation and examples here).