asked on

Order of Precedence in T-SQL

In the code below, it appears that I have a different result whenever I change the order of the element above the denominator. Since these are just simple addition and subtraction operators why would it make a difference how they are ordered. I was finally able to get it woking properly but I am a bit confused about the operatior order of precedence.

(SELECT DV.ForecastYear, @Metric1,
	(    Sum(case when MetricName=@Metric1 then MetricValue end)
	+  Sum(case when MetricName=@Metric3 then MetricValue end)
	+  Sum(case when MetricName=@Metric4 then MetricValue end)
	-   Sum(case when MetricName=@Metric2 then MetricValue end)
	+  Sum(case when MetricName=@Metric5 then MetricValue end)
	-   Sum(case when MetricName=@Metric6 then MetricValue end)
	) / Sum(case when MetricName=@Metric7 then MetricValue end) CalcResults, @spid

	FROM Metrics M2 
					INNER JOIN DataVersion DV 
					ON M2.DataVersionRowID = DV.DataVersionRowID 
					WHERE DV.DataVersionID = @DVID 
							AND MetricName in (@Metric1, @Metric2, @Metric3, @Metric4, @Metric5, @Metric6, @Metric7)
	Group by DV.ForecastYear)

Open in new window

ASKER CERTIFIED SOLUTION

vdr1620

membership

This solution is only available to members.

To access this solution, you must be a member of Experts Exchange.

Start Free Trial

cyberkiwi

Are you sure you have no other changes except the order?
There should be no difference whatsoever however you order these lines:

Sum(case when MetricName=@Metric1 then MetricValue end)
      + Sum(case when MetricName=@Metric3 then MetricValue end)
      + Sum(case when MetricName=@Metric4 then MetricValue end)
      - Sum(case when MetricName=@Metric2 then MetricValue end)
      + Sum(case when MetricName=@Metric5 then MetricValue end)
      - Sum(case when MetricName=@Metric6 then MetricValue end)

cyberkiwi

Just a thought, if @Metric1 through 6 are always DISTINCT, this should simplify it

Sum(case when MetricName in (@Metric1,@Metric3,@Metric4,@Metric5) then MetricValue end)
- Sum(case when MetricName in (@Metric2,@Metric6) then MetricValue end)

But re your question, the order should not have mattered as all if that is your full query.

Sharath S

I suggest you to do the aggregations first and then apply the calculations.

SELECT ForecastYear, 
       Metric1, 
       (sum_1 + sum_2 + sum_3 - sum_4 + sum_5 - sum_6) / sum_7, 
       spid 
  FROM (  SELECT DV.ForecastYear, 
                 @Metric1 Metric1, 
                 SUM(CASE 
                       WHEN MetricName = @Metric1 THEN MetricValue 
                     END) sum_1, 
                 SUM(CASE 
                       WHEN MetricName = @Metric3 THEN MetricValue 
                     END) sum_2, 
                 SUM(CASE 
                       WHEN MetricName = @Metric4 THEN MetricValue 
                     END) sum_3, 
                 SUM(CASE 
                       WHEN MetricName = @Metric2 THEN MetricValue 
                     END) sum_4, 
                 SUM(CASE 
                       WHEN MetricName = @Metric5 THEN MetricValue 
                     END) sum_5, 
                 SUM(CASE 
                       WHEN MetricName = @Metric6 THEN MetricValue 
                     END) sum_6, 
                 SUM(CASE 
                       WHEN MetricName = @Metric7 THEN MetricValue 
                     END) sum_7, 
                 @spid    spid 
            FROM Metrics M2 
                 INNER JOIN DataVersion DV 
                   ON M2.DataVersionRowID = DV.DataVersionRowID 
           WHERE DV.DataVersionID = @DVID 
                 AND MetricName IN (@Metric1,@Metric2,@Metric3,@Metric4, 
                                    @Metric5,@Metric6,@Metric7) 
        GROUP BY DV.ForecastYear) t1

Open in new window

Mark Wills

It really shouldnt matter. With divides, you need to be careful of using integers rather than decimals, so sometimes do a * 1.0 to ensure it uses decimal arithmatic.

The only thing I can think of off the top of my head is maybe NULL values upsetting the aggregate functions.

Using your original, and by the way both cyberkiwi and sharath have good suggestions above...

( Sum(case when MetricName=@Metric1 then MetricValue else 0.0 end)
+ Sum(case when MetricName=@Metric3 then MetricValue else 0.0 end)
+ Sum(case when MetricName=@Metric4 then MetricValue else 0.0 end)
- Sum(case when MetricName=@Metric2 then MetricValue else 0.0 end)
+ Sum(case when MetricName=@Metric5 then MetricValue else 0.0 end)
- Sum(case when MetricName=@Metric6 then MetricValue else 0.0 end)
)

-- should be the same as

( Sum(case when MetricName=@Metric1 then MetricValue else 0.0 end)
- Sum(case when MetricName=@Metric2 then MetricValue else 0.0 end)
+ Sum(case when MetricName=@Metric3 then MetricValue else 0.0 end)
+ Sum(case when MetricName=@Metric4 then MetricValue else 0.0 end)
+ Sum(case when MetricName=@Metric5 then MetricValue else 0.0 end)
- Sum(case when MetricName=@Metric6 then MetricValue else 0.0 end)
)

also your denominator should also be checked, not just for "else 0" but also that it is in fact not zero.

Juan Velasquez

ASKER

Thanks for the useful reference

Mark Wills

Interesting choice - does that help explain your question ? Or just a useful reference ?

Whilst a good link, think all it was saying is what the other experts actually said - it should not matter.

The only time it might matter is starting off with subtracting - you have to subtract from something (or could " + (something * -1) " )

check below for what I was saying about NULLs - two scenarios, one is NULL metricvalue and the other is missing metricnames.

There are settings, and if running in batch mode might be needed to avoid an abort.

But anyway, just wanted to make sure you knew the problems with handling NULL values.


if object_id('tempdb..#mydata_source') is not null drop table #mydata_source
create table #mydata_source (Forecastyear int, MetricName varchar(20), metricValue decimal (18,6))

insert #mydata_source values (2009,'M2',222)
insert #mydata_source values (2009,'M5',555)
insert #mydata_source values (2009,'M6',666)
insert #mydata_source values (2009,'M7',777)
insert #mydata_source values (2010,'M1',111.111)
insert #mydata_source values (2010,'M2',222.222)
insert #mydata_source values (2010,'M3',333.333)
insert #mydata_source values (2010,'M4',444.444)
insert #mydata_source values (2010,'M5',555.555)
insert #mydata_source values (2010,'M7',777.777)
--insert #mydata_source values (2010,'M6',666.66)
insert #mydata_source values (2010,'M6',NULL)

declare @metric1 varchar(20)
declare @metric2 varchar(20)
declare @metric3 varchar(20)
declare @metric4 varchar(20)
declare @metric5 varchar(20)
declare @metric6 varchar(20)
declare @metric7 varchar(20)

set @metric1 = 'M1'
set @metric2 = 'M2'
set @metric3 = 'M3'
set @metric4 = 'M4'
set @metric5 = 'M5'
set @metric6 = 'M6'
set @metric7 = 'M7'

select forecastyear,
	(  Sum(case when MetricName=@Metric1 then MetricValue end)
	+  Sum(case when MetricName=@Metric3 then MetricValue end)
	+  Sum(case when MetricName=@Metric4 then MetricValue end)
	-  Sum(case when MetricName=@Metric2 then MetricValue end)
	+  Sum(case when MetricName=@Metric5 then MetricValue end)
	-  Sum(case when MetricName=@Metric6 then MetricValue end)
	)  as old_way,
    (  Sum(case when MetricName=@Metric1 then MetricValue else 0.0 end)
    -  Sum(case when MetricName=@Metric2 then MetricValue else 0.0 end)
    +  Sum(case when MetricName=@Metric3 then MetricValue else 0.0 end)
    +  Sum(case when MetricName=@Metric4 then MetricValue else 0.0 end)
    +  Sum(case when MetricName=@Metric5 then MetricValue else 0.0 end)
    -  Sum(case when MetricName=@Metric6 then MetricValue else 0.0 end)
    )  as new_way,
    (  0
    -  Sum(case when MetricName=@Metric6 then MetricValue else 0.0 end)
    -  Sum(case when MetricName=@Metric2 then MetricValue else 0.0 end)
    +  Sum(case when MetricName=@Metric1 then MetricValue else 0.0 end)
    +  Sum(case when MetricName=@Metric3 then MetricValue else 0.0 end)
    +  Sum(case when MetricName=@Metric4 then MetricValue else 0.0 end)
    +  Sum(case when MetricName=@Metric5 then MetricValue else 0.0 end)
    )  as new_way_jumbled
from #mydata_source
group by forecastyear

Open in new window