Update multiple columns by loop? - sql

Question

I have a select statement which I want to convert into an update statement for all the columns in the table which have the name Variable[N].
For example, I want to do these things:
I want to be able to convert the SQL below into an update statement.
I have n columns with the name variable[N]. The example below only updates column variable63, but I want to dynamically run the update on all columns with names variable1 through variableN without knowing how many variable[N] columns I have in advance. Also, in the example below I get the updated result into NewCol. I actually want to update the respective variable column with the results if possible, variable63 in my example below.
I want to have a wrapper that loops over column variable1 through variableN and perform the same respective update operation on all those columns:
SELECT
projectid
,documentid
,revisionno
,configurationid
,variable63
,ISNULL(Variable63,
(SELECT TOP 1
variable63
FROM table1
WHERE
documentid = t.documentid
and projectid=t.projectid
and configurationid=t.configurationid
and cast(revisionno as int) < cast(t.revisionno as int)
AND Variable63 is NOT NULL
ORDER BY
projectid desc
,documentid desc
,revisionno desc
,configurationid desc
)) as NewCol
FROM table1 t;

Answer 1

There's no general way to loop through variables in SQL, you're supposed to know exactly what you want to modify. In some databases, it will be possible to query system tables to dynamically build an update statement (I know how to do that in InterBase and it's decessor Firebird), but you haven't told us anything which database engine you're using.
Below is a way you could update several fields that are null, COALESCE and CASE are two way of doing the same thing, as is using LEFT JOIN or NOT EXISTS. Use the ones you and your database engine is most comfortable with. Beware that all records will be updated, so this is not a good solution if your database contains millions of records, each record is large and you want this query to be executed lots of times.
UPDATE table1 t
SET t.VARIABLE63 =
COALESCE(t.VARIABLE63,
(SELECT VARIABLE63
FROM table1 t0
LEFT JOIN table1 tNot
ON tNot.documentid = t.documentid
AND tNot.projectid=t.projectid
AND tNot.configurationid=t.configurationid
AND cast(tNot.revisionno as int) > cast(t0.revisionno as int)
AND cast(tNot.revisionno as int) < cast(t.revisionno as int)
AND tNot.Variable63 is NOT NULL
WHERE t0.documentid = t.documentid
AND t0.projectid=t.projectid
AND t0.configurationid=t.configurationid
AND cast(t0.revisionno as int) < cast(t.revisionno as int)
AND t0.Variable63 is NOT NULL
AND tNot.Variable63 is NULL)),
t.VARIABLE64 = CASE WHEN t.VARIABLE64 IS NOT NULL then t.VARIABLE64
ELSE (SELECT VARIABLE64
FROM table1 t0
WHERE t0.documentid = t.documentid
AND t0.projectid=t.projectid
AND t0.configurationid=t.configurationid
AND cast(t0.revisionno as int) < cast(t.revisionno as int)
AND t0.Variable64 is NOT NULL
AND NOT EXISTS(SELECT 1
FROM table1 tNot
WHERE tNot.documentid = t.documentid
AND tNot.projectid=t.projectid
AND tNot.configurationid=t.configurationid
AND cast(tNot.revisionno as int) > cast(t0.revisionno as int)
AND cast(tNot.revisionno as int) < cast(t.revisionno as int)
AND tNot.Variable64 is NOT NULL)) END

Answer 2

OK I think I got it. Function that loops through columns and runs an update command per column.
DECLARE #sql NVARCHAR(1000),
#cn NVARCHAR(1000)--,
--#r NVARCHAR(1000),
--#start INT
DECLARE col_names CURSOR FOR
SELECT column_name
FROM information_schema.columns
WHERE table_name = 'PIVOT_TABLE'
ORDER BY ordinal_position
--SET #start = 0
DECLARE #op VARCHAR(max)
SET #op=''
OPEN col_names FETCH next FROM col_names INTO #cn
WHILE ##FETCH_STATUS = 0
BEGIN
--print #cn
IF UPPER(#cn)<> 'DOCUMENTID' and UPPER(#cn)<> 'CONFIGURATIONID' and UPPER(#cn)<> 'PROJECTID' and UPPER(#cn)<> 'REVISIONNO'
BEGIN
SET #sql = 'UPdate pt
set pt.' + #cn + ' = ((SELECT TOP 1 t.' + #cn + ' FROM pivot_table t WHERE t.documentid = pt.documentid and t.projectid=pt.projectid
and t.configurationid=pt.configurationid and cast(t.revisionno as int) < cast(pt.revisionno as int) AND t.' + #cn + ' is NOT NULL
ORDER BY revisionno desc)) from PIVOT_TABLE pt where pt.' + #cn + ' is NULL;'
EXEC Sp_executesql
#sql
--print #cn
END
FETCH next FROM col_names INTO #cn
END
CLOSE col_names
DEALLOCATE col_names;