Nested result sets as with dynamic names - sql

Question

I'm trying to join the result of two referencing tables to get row values which are referencing different table names, which rows are selectable by their uuid.
my tables look like this:
table entry
table map
table cats
table dogs
nrrefInt
id name mapRef breed
mapRef breed
1 123
123'dogs'
456 'bengal'
123 'sheepdog'
2 456
456 'cats'
888 'birma' 999 'poodle'
3 789
789'dogs'
4 123
refInt of entry is referencing to map. the name of map is the reference to tables in addition with the field id which is also applied on the tables cats/dogs (dynamic tables loading).
// subset 1: list of numbers that needs to be loaded from entry table (1-4)
SELECT DISTINCT refInt FROM entry WHERE nr in (1,2,3,4)
// subset 2: get all names from map that have the same id like refInt from subset1
SELECT name FROM map WHERE id in subset1
// main query: load all rows from table with the given name
// from map table that have the same mapRef value on it
SELECT * FROM (subset2.names) WHERE mapRef IN (subset2.ids)
result should be the rows:
1) 456 bengal
2) 123 sheepdog
I also made a SQLFiddle of it.
Is there a way to combine this to one query?

Answer 1

It's going to look something like:
SqlFiddle
select
sub.nr,
sub.breed
from (
select e.nr, e.refInt,
case
when c.breed is not null then c.breed
when d.breed is not null then d.breed
else null
end as breed
from (
select e.nr, e.refInt, m.name
from entry e
inner join map m on e.refInt = m.id
) e
left join cats c on e.refInt = c.mapRef and e.name = 'cats'
left join dogs d on e.refInt = d.mapref and e.name = 'dogs'
) sub
where sub.breed is not null
This is going to be very poor in performance.
Now the IMO the correct schema would be:
table entry
nr refint
1 123
2 456
3 789
4 124 (duplicate?)
table breed
mapRef breed species
123 sheepdog 1
999 poodle 1
456 bengal 2
888 birma 2
table species
id species
1 dogs
2 cats
This is normalized and has very good performance.

Answer 2

Note how the following query fully achieves the desired result set and fully demonstrates how the tables cats and dogs are truly just partitions of a single entity animals. The schema should be reworked to reflect this new understanding. This query is also efficient because the inclusion test id pushed to the depths of the innermost CTE's, at the level where actual table rows are being read, without relying on the engine to discover this potential optimization (which can be problematic with UNIONs).
with
cats2 as (
select species='cat', mapref, breed
from cats animals
join entry on entry.refint = animals.mapref
where entry.nr in (1,2,3,4)
),
dogs2 as (
select species='dog', mapref, breed
from dogs animals
join entry on entry.refint = animals.mapref
where entry.nr in (1,2,3,4)
),
animals as (
select species, mapref, breed from cats2
union all
select species, mapref, breed from dogs2
)
select species, mapref, breed
from animals
group by species, mapref, breed
This test script:
declare #entry table (nr int, refint int );
declare #map table (id int, name varchar(20) );
declare #cats table (mapRef int, breed varchar(20));
declare #dogs table (mapRef int, breed varchar(20));
insert #entry(nr,refint) values
(1,123)
,(2,456)
,(3,789)
,(4,123);
insert #map(id,name) values
(123,'dogs')
,(456,'cats')
,(789,'dogs');
insert #cats(mapRef,breed) values
(456,'bengal'),(888,'burma');
insert #dogs(mapRef,breed) values
(123,'sheepdog'), (999,'poodle');
with
cats2 as (
select species='cat', mapref, breed
from #cats animals
join #entry entry on entry.refint = animals.mapref
where entry.nr in (1,2,3,4)
),
dogs2 as (
select species='dog', mapref, breed
from #dogs animals
join #entry entry on entry.refint = animals.mapref
where entry.nr in (1,2,3,4)
),
animals as (
select species, mapref, breed from cats2
union all
select species, mapref, breed from dogs2
)
select species, mapref, breed
from animals
group by species, mapref, breed
yields as desired:
species mapref breed
------- ----------- --------------------
cat 456 bengal
dog 123 sheepdog