How to use SQL LAG function with condition - sql

Question

I have a table as the following rows:
tipoProtocolo numeroProtocolo dataReferencia dataAssinatura dataVencimento
------------- --------------- -------------- -------------- --------------
1 47676 NULL 20150112 20151231
1 47676 20151231 20150209 NULL
1 47676 NULL 20150224 NULL
1 47676 NULL 20151005 NULL
1 47676 NULL 20151021 NULL
1 47676 NULL 20151026 NULL
1 47676 NULL 20151120 NULL
I've implemented a piece of code that gets the value from the dataVencimento column (previous row) to the dataRefencia column (red arrow in the image). However, I would like to check if the column dataVencimento (from the previous row) is NULL. If this condition is true I need to copy the value from the column dataReferencia from the previous row (blue arrow in the image).
SELECT tipoProtocolo,
numeroProtocolo,
LAG(dataVencimento, 1 ) OVER(
PARTITION BY numeroProtocolo, tipoProtocolo
ORDER BY dataAssinatura
) dataReferencia,
dataAssinatura,
dataVencimento
FROM cte_ContratoAditivo
Here is my SQL code:
SELECT tipoProtocolo, numeroProtocolo,
LAG(dataVencimento, 1) OVER(
PARTITION BY numeroProtocolo, tipoProtocolo
ORDER BY dataAssinatura
) dataReferencia,
dataAssinatura, dataVencimento
FROM cte_ContratoAditivo

Answer 1

What you want is lag(ignore nulls). Unfortunately, SQL Server does not support this.
If the dates are increasing, you can use a cumulative max:
select . . .,
max(dataVencimento) over (
partition by numeroProtocolo, tipoProtocolo
order by dataAssinatura
rows between unbounded preceding and 1 preceding
) as dataReferencia
If this is not the case, you can use two levels of aggregation:
select ca.*,
max(dataVencimento) over (
partition by numeroProtocolo, tipoProtocolo
order by dataAssinatura
) as dataReferencia
from (select ca.*,
count(dataVencimento) over (
partition by numeroProtocolo, tipoProtocolo
order by dataAssinatura
) as grouping
from cte_ContratoAditivo ca
) ca;
The subquery counts the number of valid values. This is really to assign a group number to the rows. The outer query then spreads the value over the entire group.

Answer 2

As the OP did respond I've gone with the literal and the guessed answer. The first is the literal answer; if the prior row is NULL use that the one prior:
WITH VTE AS (
SELECT *
FROM (VALUES(1,47676,CONVERT(date,NULL),CONVERT(date,'20150112'),CONVERT(date,'20151231')),
(1,47676,CONVERT(date,'20151231'),CONVERT(date,'20150209'),CONVERT(date,NULL)),
(1,47676,CONVERT(date,NULL),CONVERT(date,'20150224'),CONVERT(date,NULL)),
(1,47676,CONVERT(date,NULL),CONVERT(date,'20151005'),CONVERT(date,NULL)),
(1,47676,CONVERT(date,NULL),CONVERT(date,'20151021'),CONVERT(date,NULL)),
(1,47676,CONVERT(date,NULL),CONVERT(date,'20151026'),CONVERT(date,NULL)),
(1,47676,CONVERT(date,NULL),CONVERT(date,'20151120'),CONVERT(date,NULL))) V(tipoProtocolo,numeroProtocolo,dataReferencia,dataAssinatura,dataVencimento)),
CTE AS(
SELECT V.tipoProtocolo,
V.numeroProtocolo,
V.dataReferencia,
V.dataAssinatura,
V.dataVencimento,
LAG(dataVencimento) OVER (PARTITION BY numeroProtocolo, tipoProtocolo ORDER BY dataAssinatura) AS dataReferencia1,
LAG(dataVencimento,2) OVER (PARTITION BY numeroProtocolo, tipoProtocolo ORDER BY dataAssinatura) AS dataReferencia2
FROM VTE V)
SELECT C.tipoProtocolo,
C.numeroProtocolo,
C.dataReferencia,
C.dataAssinatura,
C.dataVencimento,
ISNULL(C.dataReferencia1,C.dataReferencia2) AS dataReferencia
FROM CTE C;
The other is what I suspect the OP really means and that they want the last non-NULLvalue. If this is the case, this is a "classic" gaps and islands problem:
WITH VTE AS (
SELECT *
FROM (VALUES(1,47676,CONVERT(date,NULL),CONVERT(date,'20150112'),CONVERT(date,'20151231')),
(1,47676,CONVERT(date,'20151231'),CONVERT(date,'20150209'),CONVERT(date,NULL)),
(1,47676,CONVERT(date,NULL),CONVERT(date,'20150224'),CONVERT(date,NULL)),
(1,47676,CONVERT(date,NULL),CONVERT(date,'20151005'),CONVERT(date,NULL)),
(1,47676,CONVERT(date,NULL),CONVERT(date,'20151021'),CONVERT(date,NULL)),
(1,47676,CONVERT(date,NULL),CONVERT(date,'20151026'),CONVERT(date,NULL)),
(1,47676,CONVERT(date,NULL),CONVERT(date,'20151120'),CONVERT(date,NULL))) V(tipoProtocolo,numeroProtocolo,dataReferencia,dataAssinatura,dataVencimento)),
Grps AS(
SELECT V.tipoProtocolo,
V.numeroProtocolo,
V.dataReferencia,
V.dataAssinatura,
V.dataVencimento,
COUNT(dataVencimento) OVER (PARTITION BY numeroProtocolo, tipoProtocolo ORDER BY dataAssinatura
ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS Grp
FROM VTE V)
SELECT G.tipoProtocolo,
G.numeroProtocolo,
G.dataReferencia,
G.dataAssinatura,
G.dataVencimento,
MAX(dataVencimento) OVER (PARTITION BY G.Grp) AS dataReferencia,
G.Grp
FROM Grps G
ORDER BY dataAssinatura;
I will note that is seems odd that you call the column with the LAG expression dataReferencia, despite that the expression is on dataVencimento (and there is already a column called dataReferencia).