Joining sql tables with no common columns without ordering - sql

Question

I have my data in a form of 2 coma separated strings
DECLARE #ids nvarchar(max) = '1,2,3'
DECLARE #guids nvarchar(max) =
'0000000-0001-0000-0000-000000000000,
`0000000-0022-0000-0000-000000000000`,
`0000000-0013-0000-0000-000000000000'`
I need them in a table as separate columns based on their position in the string
Table1
| Id | Guid |
| 1 | 0000000-0001-0000-0000-000000000000 |
| 2 | 0000000-0022-0000-0000-000000000000 |
| 3 | 0000000-0013-0000-0000-000000000000 |
I can split both strings into separate tables by using
DECLARE #split_ids
(value nvarchar(max))
DECLARE #xml xml
SET #xml = N'<root><r>' + replace(#ids, ',' ,'</r><r>') + '</r></root>'
INSERT INTO #split_ids(Value)
SELECT r.value('.','nvarchar(max)')
FROM #xml.nodes('//root/r') as records(r)
I've tried
SELECT t1.*, t2.*
FROM (SELECT t1.*, row_number() OVER (ORDER BY [Value]) as seqnum
from cte_Ids t1
) t1 FULL OUTER JOIN
(SELECT t2.*, row_number() OVER (ORDER BY [Value]) as seqnum
from cte_barcodes t2
) t2
ON t1.seqnum = t2.seqnum;
But that orders the tables by Value and my data is random and can't be ordered.
Is there a way of joining tables based on their row numbers without ordering them first?
Or is there another way of inserting data from a string to a table?

Answer 1

You do not need to split and/or insert the input data into separate tables. In this situation you simply need to parse the input strings and get the substrings and their ordinal positions (an XML-based approach or a splitter function are possible solutions).
But if you use SQL Server 2016+, a JSON-based approach is also an option. The idea is to transform the strings into valid JSON arrays (1,2,3 into [1,2,3]), parse the arrays with OPENJSON() and join the tables returned from OPENJSON() calls. As is explained in the documentation, the columns that OPENJSON() function returns (when the default schema is used) are key, value and type and in case of JSON array, the key column holds the index of the element in the specified array.
DECLARE #ids nvarchar(max) = N'1,2,3'
DECLARE #guids nvarchar(max) = N'0000000-0001-0000-0000-000000000000,0000000-0022-0000-0000-000000000000,0000000-0013-0000-0000-000000000000'
SELECT j1.[value] AS Id, j2.[value] AS Guid
FROM OPENJSON(CONCAT('[', #ids, ']')) j1
JOIN OPENJSON(CONCAT('["', REPLACE(#guids, ',', '","'), '"]')) j2 ON j1.[key] = j2.[key]
Result:
Id Guid
1 0000000-0001-0000-0000-000000000000
2 0000000-0022-0000-0000-000000000000
3 0000000-0013-0000-0000-000000000000

Answer 2

You need row numbering over initial order, this means that you should use some constant expression in window function order_by clause.
SQL server does not allow use constants directly, but over(order_by (select 1)) is allowed:
SELECT t1.*, t2.*
FROM (SELECT t1.*, row_number() OVER (ORDER BY (select 1)) as seqnum
from cte_Ids t1
) t1 FULL OUTER JOIN
(SELECT t2.*, row_number() OVER (ORDER BY (select 1)) as seqnum
from cte_barcodes t2
) t2
ON t1.seqnum = t2.seqnum;
Note that this doesn't guarantee initial order (it will be unspecified), but often it behaves correctly :)

Answer 3

One of solutions is to parse your comma separated values in a loop (using WHILE) from both variables. Then you could insert those extracted in the same iteration values at once as one row to a table.

Answer 4

One solution uses recursive CTEs:
with cte as (
select cast(null as nvarchar(max)) as id, cast(null as nvarchar(max)) as guid, #ids + ',' as rest_ids, #guids + ',' as rest_guids, 0 as lev
union all
select left(rest_ids, charindex(',', rest_ids) - 1),
left(rest_guids, charindex(',', rest_guids) - 1),
stuff(rest_ids, 1, charindex(',', rest_ids), ''),
stuff(rest_guids, 1, charindex(',', rest_guids), ''),
lev + 1
from cte
where rest_ids <> ''
)
select id, guid
from cte
where lev > 0;
Here is a db<>fiddle.