TSQL Interleaved sequence without join operations or user defined objects - sql

Question

Is that possible to create an interleaved sequence with first and last values going in turns untill data set is empty without using joins and user defined functions or procedures?
code type model price
1 Hats 1298 700,00
1 Shoes 1232 600,00
1 Pants 1276 400,00
2 Hats 1321 970,00
2 Shoes 1121 850,00
2 Pants 1433 270,00
3 Hats 1750 1200,00
3 Shoes 1233 600,00
3 Pants 1434 290,00
4 Hats 1298 1050,00
4 Shoes 1121 850,00
4 Pants 1401 150,00
5 Hats 1752 1150,00
5 Shoes 1121 850,00
5 Pants 1408 270,00
6 Hats 1298 950,00
6 Shoes 1233 950,00
6 Pants 1288 400,00
7 Shoes 1232 400,00
8 Shoes 1232 350,00
9 Shoes 1232 350,00
10 Shoes 1260 350,00
11 Shoes 1233 980,00
12 Shoes 1233 970,00
I have added extra spaces between rows to get the interleaved sequence idea.
You want to get the odd values going from lowest coded items (asc) and even values with the highest coded items (desc). You also want to order type by Hats,Shoes and Pants.
code type model price
1 Hats 1298 700,00
1 Shoes 1232 600,00
1 Pants 1276 400,00
6 Hats 1298 950,00
12 Shoes 1233 970,00
6 Pants 1288 400,00
2 Hats 1321 970,00
2 Shoes 1121 850,00
2 Pants 1433 270,00
5 Hats 1752 1150,00
11 Shoes 1233 980,00
5 Pants 1408 270,00
3 Hats 1750 1200,00
3 Shoes 1233 600,00
3 Pants 1434 290,00
4 Hats 1298 1050,00
10 Shoes 1260 350,00
4 Pants 1401 150,00
4 Shoes 1121 850,00
9 Shoes 1232 350,00
5 Shoes 1121 850,00
8 Shoes 1232 350,00
6 Shoes 1233 950,00
7 Shoes 1232 400,00
Right now I came up with solution that includes joins but I am looking for something that would work without using it.
My solution with using joins:
with cteasc as
(
select
ROW_NUMBER() over(order by code,charindex(type, 'HatsShoesPants'))id
,(ROW_NUMBER() over(partition by type order by code,charindex(type, 'HatsShoesPants')) + 1) / 2 offsetasc
,code,model,price,type
from mydata
),
ctedsc as
(
select
ROW_NUMBER() over (partition by type order by code desc)id
,code,model,price,type
from cteasc
)
select t1.id
,case
when t1.code%2=1
then LAG(t1.type,t1.code-t1.offsetasc,t1.type)over(partition by t1.type order by t1.id)
else LAG(t2.type,t1.code-t1.offsetasc,t1.type)over(partition by t1.type order by t1.id)
end type
,case
when t1.code%2=1
then LAG(t1.model,t1.code-t1.offsetasc,t1.model)over(partition by t1.type order by t1.id)
else LAG(t2.model,t1.code-t1.offsetasc,t1.model)over(partition by t1.type order by t1.id)
end model
,case
when t1.code%2=1
then LAG(t1.price,t1.code-t1.offsetasc,t1.price)over(partition by t1.type order by t1.id)
else LAG(t2.price,t1.code-t1.offsetasc,t1.price)over(partition by t1.type order by t1.id)
end price
from
cteasc t1
join ctedsc t2
on t1.code = t2.id
and t1.type = t2.type

Answer 1

The idea would be to generate two sequences:
The ascending items with row numbers 1, 3, 5, ...
The descending items with row numbers 2, 4, 6, ...
Then, we simply need to UNION ALL the sequences and sort them.
with ascItems as (
select *, row_number() over (order by someCol ASC) r * 2 - 1 as r from T
)
, descItems as (
select *, row_number() over (order by someCol DESC) r * 2 as r from T
)
select * from ascItems
union all
select * from descItems
order by r
This should require the table T to be available sorted twice (either by index or by 2xsorting). The UNION ALL should manifest itself as a cheap merge concat.