Simplifying (aliasing) T-SQL CASE statements. Any improvement possible?
https://stackoverflow.com/questions/951836
-
11-09-2019 - |
italiano
english
français
española
中国
日本の
العربية
Deutsch
한국어
Português
RussianQuestion
As you can see, this sucks big time. Any alternative? I've tried using the column alias in the group by clause to no avail.
select count(callid) ,
case
when callDuration > 0 and callDuration < 30 then 1
when callDuration >= 30 and callDuration < 60 then 2
when callDuration >= 60 and callDuration < 120 then 3
when callDuration >= 120 and callDuration < 180 then 4
when callDuration >= 180 and callDuration < 240 then 5
when callDuration >= 240 and callDuration < 300 then 6
when callDuration >= 300 and callDuration < 360 then 7
when callDuration >= 360 and callDuration < 420 then 8
when callDuration >= 420 and callDuration < 480 then 9
when callDuration >= 480 and callDuration < 540 then 10
when callDuration >= 540 and callDuration < 600 then 11
when callDuration >= 600 then 12
end as duration
from callmetatbl
where programid = 1001 and callDuration > 0
group by case
when callDuration > 0 and callDuration < 30 then 1
when callDuration >= 30 and callDuration < 60 then 2
when callDuration >= 60 and callDuration < 120 then 3
when callDuration >= 120 and callDuration < 180 then 4
when callDuration >= 180 and callDuration < 240 then 5
when callDuration >= 240 and callDuration < 300 then 6
when callDuration >= 300 and callDuration < 360 then 7
when callDuration >= 360 and callDuration < 420 then 8
when callDuration >= 420 and callDuration < 480 then 9
when callDuration >= 480 and callDuration < 540 then 10
when callDuration >= 540 and callDuration < 600 then 11
when callDuration >= 600 then 12
end
EDIT: I really meant to ask how to have a single case source, but case modifications are welcome anyway (although less useful because the intervals probably will be modified and might even be automatically generated).
As has been considered by some people, callDuration is indeed a float so some listed solutions are not valid for my use case, by leaving values out of the intervals.
Lessons:
Look for patterns in the case expression to reduce it if possible and worthwhile
case when callDuration > 0 AND callDuration < 30 then 1 when callDuration > 600 then 12 else floor(callDuration/60) + 2 end end as durationUse inline views to have a single source of the case
select count(d.callid), d.duration from ( select callid , case when callDuration > 0 AND callDuration < 30 then 1 when callDuration > 600 then 12 else floor(callDuration/60) + 2 end end as duration from callmetatbl where programid = 1001 and callDuration > 0 ) d group by d.durationOr use common table expressions
with duration_case as ( select callid , case when callDuration > 0 AND callDuration < 30 then 1 when callDuration > 600 then 12 else floor(callDuration/60) + 2 end end as duration from callmetatbl where programid = 1001 and callDuration > 0 ) select count(callid), duration from duration_case group by durationOr use an user defined function (no example so far :-) )
Or use a lookup table and a join
DECLARE @t TABLE(durationFrom float, durationTo float, result INT) --populate table with values so the query works select count(callid) , COALESCE(t.result, 12) from callmetatbl JOIN @t AS t ON callDuration >= t.durationFrom AND callDuration < t.durationTo where programid = 1001 and callDuration > 0
Thanks to everybody and I'm having a very difficult time choosing an accepted answer, as many covered different parts of the question (and I was there thinking it was a simple question with a straightforward answer :-), sorry for the confusion).
Solution
Q: how to get an alias to use in the GROUP BY clause
One approach is to use an inline view. [EDIT] The answer from Remus Rusanu (+1!) gives an example of a Common Table Expression to accomplish the same thing. [/EDIT]
The inline view gets you a simple "alias" for the complex expression which you can then reference in a GROUP BY clause in an outer query:
select count(d.callid)
, d.duration
from (select callid
, case
when callDuration >= 600 then 12
when callDuration >= 540 then 11
when callDuration >= 480 then 10
when callDuration >= 420 then 9
when callDuration >= 360 then 8
when callDuration >= 300 then 7
when callDuration >= 240 then 6
when callDuration >= 180 then 5
when callDuration >= 120 then 4
when callDuration >= 60 then 3
when callDuration >= 30 then 2
when callDuration > 0 then 1
--else null
end as duration
from callmetatbl
where programid = 1001
and callDuration > 0
) d
group by d.duration
Let's unpack that.
- the inner (indented) query is called and inline view (we given it an alias
d) - in the outer query, we can reference the alias
durationfromd
That should be sufficient to answer your question. If you're looking for an equivalent replacement expression, the one from tekBlues (+1 !) is the right answer (it works on the boundary and for non-integers.)
With the replacement expression from tekBlues (+1!):
select count(d.callid)
, d.duration
from (select callid
, case
when callduration >=30 and callduration<600
then floor(callduration/60)+2
when callduration>0 and callduration< 30
then 1
when callduration>=600
then 12
end as duration
from callmetatbl
where programid = 1001
and callDuration > 0
) d
group by d.duration
(This should be sufficient to answer your question.)
[UPDATE:] sample user defined function (a replacement for inline CASE expression)
CREATE FUNCTION [dev].[udf_duration](@cd FLOAT)
RETURNS SMALLINT
AS
BEGIN
DECLARE @bucket SMALLINT
SET @bucket =
CASE
WHEN @cd >= 600 THEN 12
WHEN @cd >= 540 THEN 11
WHEN @cd >= 480 THEN 10
WHEN @cd >= 420 THEN 9
WHEN @cd >= 360 THEN 8
WHEN @cd >= 300 THEN 7
WHEN @cd >= 240 THEN 6
WHEN @cd >= 180 THEN 5
WHEN @cd >= 120 THEN 4
WHEN @cd >= 60 THEN 3
WHEN @cd >= 30 THEN 2
WHEN @cd > 0 THEN 1
--ELSE NULL
END
RETURN @bucket
END
select count(callid)
, [dev].[udf_duration](callDuration)
from callmetatbl
where programid = 1001
and callDuration > 0
group by [dev].[udf_duration](callDuration)
NOTES: be aware that the user defined function will add overhead, and (of course) add a dependency on another database object.
This example function is equivalent to the original expression. The OP CASE expression doesn't have any gaps, but it does reference each "breakpoint" twice, I prefer to test only the lower bound. (CASE returns when a condition is satisfied. Doing the tests in reverse lets the unhandled case (<=0 or NULL) fall through without test, an ELSE NULL is not necessary, but could be added for completeness.
ADDITIONAL DETAILS
(Be sure to check the performance and the optimizer plan, to make sure it's the same as (or not significantly worse than) the original. In the past, I've had problems getting predicates pushed into the inline view, doesn't look like it will be a problem in your case.)
stored view
Note that the inline view could also be stored as view definition in the database. But there's no reason to do that, other than to "hide" the complex expression from your statement.
simplifying the complex expression
Another way to make a complex expression "simpler" is to use a user defined function. But a user defined function comes with its own set of issues (including degraded performance.)
add database "lookup" table
Some answers recommend adding a "lookup" table to the database. I don't see that this is really necessary. It could be done of course, and could make sense if you want to be able to derive different values for duration from callDuration, on the fly, without having to modify your query and without having to run any DDL statements (e.g. to alter a view definition, or modify a user defined function).
With a join to a "lookup" table, one benefit is that you could make the query return different result sets by just performing DML operations on the "lookup" table.
But that same advantage may actually be a drawback as well.
Consider carefully if the benefit actually outweighs the downside. Consider the impact that new table will have on unit testing, how to verify the contents of the lookup table are valid and not changed (any overlaps? any gaps?), impact on ongoing maintenance to the code (due to the additional complexity).
some BIG assumptions
A lot of the answers given here seem to assume that callDuration is an INTEGER datatype. It seems they have overlooked the possibility that it's not an integer, but maybe I missed that nugget in the question.
It's fairly simple test case to demonstrate that:
callDuration BETWEEN 0 AND 30
is NOT equivalent to
callDuration > 0 AND callDuration < 30
OTHER TIPS
Is there any reason you're not using between? The case statements themselves don't look too bad. If you really hate it you could throw all this into a table and map it.
Durations
------------------
low high value
0 30 1
31 60 2
etc...
(SELECT value FROM Durations WHERE callDuration BETWEEN low AND high) as Duration
EDIT: Or, in a case where floats are being used and between becomes cumbersome.
(SELECT value FROM Durations WHERE callDuration >= low AND callDuration <= high) as Duration
the case can be written like this:
case
when callduration >=30 and callduration<600 then floor(callduration/60)+2
when callduration>0 and callduration< 30 then 1
when callduration>=600 then 12
end
The having is not needed, replace it by a "where callduration>0"
I like the translate table answer given before! that's the best solution
You need to push the CASE further down the query tree so that its projection is visible to the GROUP BY. This can be achieve in two ways:
- Use a derived table (already Spencer, Adam and Jeremy showed how)
Use a common table expressions
with duration_case as ( select callid , case when callDuration > 0 and callDuration < 30 then 1 when callDuration >= 30 and callDuration < 60 then 2 when callDuration >= 60 and callDuration < 120 then 3 when callDuration >= 120 and callDuration < 180 then 4 when callDuration >= 180 and callDuration < 240 then 5 when callDuration >= 240 and callDuration < 300 then 6 when callDuration >= 300 and callDuration < 360 then 7 when callDuration >= 360 and callDuration < 420 then 8 when callDuration >= 420 and callDuration < 480 then 9 when callDuration >= 480 and callDuration < 540 then 10 when callDuration >= 540 and callDuration < 600 then 11 when callDuration >= 600 then 12 end as duration from callmetatbl where programid = 1001 and callDuration > 0 ) select count(callid), duration from duration_case group by duration
Both solutions are equivalent in every respect. I find CTEs more readable, some prefer derived tables as more portable.
Divide callDuration by 60:
case
when callDuration between 1 AND 29 then 1
when callDuration > 600 then 12
else (callDuration /60) + 2 end
end as duration
Note that between is inclusive of the bounds, and I'm assuming callDuration will be treated as an integer.
Update:
Combine this with some of the other answers, and you can get the entire query down to this:
select count(d.callid), d.duration
from (
select callid
, case
when callDuration between 1 AND 29 then 1
when callDuration > 600 then 12
else (callDuration /60) + 2 end
end as duration
from callmetatbl
where programid = 1001
and callDuration > 0
) d
group by d.duration
select count(callid), duration from
(
select callid ,
case
when callDuration > 0 and callDuration < 30 then 1
when callDuration >= 30 and callDuration < 60 then 2
when callDuration >= 60 and callDuration < 120 then 3
when callDuration >= 120 and callDuration < 180 then 4
when callDuration >= 180 and callDuration < 240 then 5
when callDuration >= 240 and callDuration < 300 then 6
when callDuration >= 300 and callDuration < 360 then 7
when callDuration >= 360 and callDuration < 420 then 8
when callDuration >= 420 and callDuration < 480 then 9
when callDuration >= 480 and callDuration < 540 then 10
when callDuration >= 540 and callDuration < 600 then 11
when callDuration >= 600 then 12
end as duration
from callmetatbl
where programid = 1001 and callDuration > 0
) source
group by duration
Untested:
select count(callid) , duracion
from
(select
callid,
case
when callDuration > 0 and callDuration < 30 then 1
when callDuration >= 30 and callDuration < 60 then 2
when callDuration >= 60 and callDuration < 120 then 3
when callDuration >= 120 and callDuration < 180 then 4
when callDuration >= 180 and callDuration < 240 then 5
when callDuration >= 240 and callDuration < 300 then 6
when callDuration >= 300 and callDuration < 360 then 7
when callDuration >= 360 and callDuration < 420 then 8
when callDuration >= 420 and callDuration < 480 then 9
when callDuration >= 480 and callDuration < 540 then 10
when callDuration >= 540 and callDuration < 600 then 11
when callDuration >= 600 then 12
else 0
end as duracion
from callmetatbl
where programid = 1001) GRP
where duracion > 0
group by duracion
Add all the cases into a table variable and do an outer join
DECLARE @t TABLE(durationFrom INT, durationTo INT, result INT)
-- when callDuration > 0 and callDuration < 30 then 1
INSERT INTO @t VALUES(1, 30, 1);
-- when callDuration >= 30 and callDuration < 60 then 2
INSERT INTO @t VALUES(30, 60, 2);
select count(callid) , COALESCE(t.result, 12)
from callmetatbl JOIN @t AS t ON callDuration >= t.durationFrom AND callDuration < t.durationTo
where programid = 1001 and callDuration > 0
Here's my shot at it. All of the components you need can be done in straight SQL.
select
count(1) as total
,(fixedDuration / divisor) + adder as duration
from
(
select
case/*(30s_increments_else_60s)*/when(callDuration<60)then(120)else(60)end as divisor
,case/*(increment_by_1_else_2)*/when(callDuration<30)then(1)else(2)end as adder
,(/*duration_capped@600*/callDuration+600-ABS(callDuration-600))/2 as fixedDuration
,callDuration
from
callmetatbl
where
programid = 1001
and
callDuration > 0
) as foo
group by
(fixedDuration / divisor) + adder
Here's the SQL I used for testing. (I don't have my own personal callmetatbl ;)
select
count(1) as total
,(fixedDuration / divisor) + adder as duration
from
(
select
case/*(30s_increments_else_60s)*/when(callDuration<60)then(120)else(60)end as divisor
,case/*(increment_by_1_else_2)*/when(callDuration<30)then(1)else(2)end as adder
,(/*duration_capped@600*/callDuration+600-ABS(callDuration-600))/2 as fixedDuration
,callDuration
from -- callmetatbl -- using test view below
(
select 1001 as programid, 0 as callDuration union
select 1001 as programid, 1 as callDuration union
select 1001 as programid, 29 as callDuration union
select 1001 as programid, 30 as callDuration union
select 1001 as programid, 59 as callDuration union
select 1001 as programid, 60 as callDuration union
select 1001 as programid, 119 as callDuration union
select 1001 as programid, 120 as callDuration union
select 1001 as programid, 179 as callDuration union
select 1001 as programid, 180 as callDuration union
select 1001 as programid, 239 as callDuration union
select 1001 as programid, 240 as callDuration union
select 1001 as programid, 299 as callDuration union
select 1001 as programid, 300 as callDuration union
select 1001 as programid, 359 as callDuration union
select 1001 as programid, 360 as callDuration union
select 1001 as programid, 419 as callDuration union
select 1001 as programid, 420 as callDuration union
select 1001 as programid, 479 as callDuration union
select 1001 as programid, 480 as callDuration union
select 1001 as programid, 539 as callDuration union
select 1001 as programid, 540 as callDuration union
select 1001 as programid, 599 as callDuration union
select 1001 as programid, 600 as callDuration union
select 1001 as programid,1000 as callDuration
) as callmetatbl
where
programid = 1001
and
callDuration > 0
) as foo
group by
(fixedDuration / divisor) + adder
The SQL output is shown below, as 2 records counted for each duration (bucket) 1 through 12.
total duration
2 1
2 2
2 3
2 4
2 5
2 6
2 7
2 8
2 9
2 10
2 11
2 12
Here are the results from the "foo" sub-query:
divisor adder fixedDuration callDuration
120 1 1 1
120 1 29 29
120 2 30 30
120 2 59 59
60 2 60 60
60 2 119 119
60 2 120 120
60 2 179 179
60 2 180 180
60 2 239 239
60 2 240 240
60 2 299 299
60 2 300 300
60 2 359 359
60 2 360 360
60 2 419 419
60 2 420 420
60 2 479 479
60 2 480 480
60 2 539 539
60 2 540 540
60 2 599 599
60 2 600 600
60 2 600 1000
Cheers.
What's so wrong with a User Defined Function here? You could both visually clean up the code and centralize the functionality that way. Performance-wise, I can't see the hit being too horrible unless you are doing something really retarded within said UDF.
Create a lookup table for duration
Using a look up table will speed up the SELECT statement as well.
Here is the end result of how it will look with lookup table.
select count(a.callid), b.ID as duration
from callmetatbl a
inner join DurationMap b
on a.callDuration >= b.Minimum
and a.callDuration < IsNUll(b.Maximum, a.CallDuration + 1)
group by b.ID
Here is the look up table.
create table DurationMap (
ID int identity(1,1) primary key,
Minimum int not null,
Maximum int
)
insert DurationMap(Minimum, Maximum) select 0,30
insert DurationMap(Minimum, Maximum) select 30,60
insert DurationMap(Minimum, Maximum) select 60,120
insert DurationMap(Minimum, Maximum) select 120,180
insert DurationMap(Minimum, Maximum) select 180,240
insert DurationMap(Minimum, Maximum) select 240,300
insert DurationMap(Minimum, Maximum) select 300,360
insert DurationMap(Minimum, Maximum) select 360,420
insert DurationMap(Minimum, Maximum) select 420,480
insert DurationMap(Minimum, Maximum) select 480,540
insert DurationMap(Minimum, Maximum) select 540,600
insert DurationMap(Minimum) select 600
