Best way to get rid of unwanted sql subselects? - tsql

Question

I have a table called Registrations with the following fields:
Id
DateStarted (not null)
DateCompleted (nullable)
I have a bar chart which shows the number of registrations started and completed by date.
My query looks like:
;
WITH Initial(DateStarted, StartCount)
as (
select Datestarted, COUNT(*)
FROM Registrations
GROUP BY DateStarted
)
select I.DateStarted, I.StartCount, COUNT(DISTINCT R.RegistrationId) as CompleteCount
from Initial I
inner join Registrations R
ON (I.DateStarted = R.DateCompleted)
GROUP BY I.DateStarted, I.StartCount
which returns a table that looks like:
DateStarted StartCount CompleteCount
2009-08-01 1033 903
2009-08-02 540 498
The query just has one of those code smell problems. What is a better way of doing this?

Answer 1

EDIT: so why wont the below work? you could throw coalesce() statements around the counts in the last select statement if you wanted to make the counts zero instead of null. it will also include dates that have completed (or ended in the example below) registrations even though that date doesn't have started registrations.
I am assuming the following table structure (roughly).
create table temp
(
id int,
start_date datetime,
end_date datetime
)
insert into temp values (1, '8/1/2009', '8/1/2009')
insert into temp values (2, '8/1/2009', '8/2/2009')
insert into temp values (3, '8/1/2009', null)
insert into temp values (4, '8/2/2009', '8/2/2009')
insert into temp values (5, '8/2/2009', '8/3/2009')
insert into temp values (6, '8/2/2009', '8/4/2009')
insert into temp values (7, '8/4/2009', null)
Then you could do the following to get what you want.
with start_helper as
(
select start_date, count(*) as count from temp group by start_date
),
end_helper as
(
select end_date, count(*) as count from temp group by end_date
)
select coalesce(a.start_date, b.end_date) as date, a.count as start_count, b.count as end_count
from start_helper a full outer join end_helper b on a.start_date = b.end_date
where coalesce(a.start_date, b.end_date) is not null
I would think the full outer join is necessary since a record can be completed today that started yesterday but we may have not started a new record today so you would lose a day from your results.

Answer 2

Off-hand, I think this does it:
SELECT
DateStarted
, COUNT(*) as StartCount
, SUM(CASE
WHEN DateCompleted = DateStated THEN 1
ELSE 0 END
) as CompleteCount
FROM Registration
GROUP BY DateStarted
OK, apparently I had the requirements wrong before. Given that the CompleteCounts are independent of the StartDate, then this is how I would do it:
;WITH StartDays AS
(
SELECT DateStarted
, Count(*) AS CompleteCount
FROM Registration
GROUP BY DateStarted
)
, CompleteDays AS
(
SELECT DateCompleted
, Count(*) AS StartCount
FROM Registration
GROUP BY DateCompleted
)
SELECT
DateStarted
, COALESCE(StartCount, 0) AS StartCount
, COALESCE(CompleteCount, 0) AS CompleteCount
FROM StartDays
FULL OUTER JOIN CompleteDays ON DateStarted = DateCompleted
Which actually is pretty close to what you had.

Answer 3

I don't see a problem. I see a common table expression being used.
You didn't provide DDL for the tables, so I'm not going to try to reproduce this. However, I think you may be able to directly substitute the SELECT for the use of Initial.

Answer 4

I believe the following is identical in function to what you have:
select DS.DateStarted
, count(distinct DS.RegistrationId) as StartCount
, count(distinct DC.RegistrationId) as CompleteCount
from Registrations DS
inner join Registrations DC on DS.DateStarted = DC.DateCompleted
group by Ds.DateStarted
I'm a bit confused by the name of the column DateStarted in the results. It looks to just be a date where both some things started and some things ended. And the counts are the number or registrations started and completed that day.
The inner join is throwing away any date where there is either 0 starts or 0 completes. To get all:
select coalesce(DS.DateStarted, DC.DateCompleted) as "Date"
, count(distinct DS.RegistrationId) as StartCount
, count(distinct DC.RegistrationId) as CompleteCount
from Registrations DS
full outer join Registrations DC on DS.DateStarted = DC.DateCompleted
group by Ds.DateStarted, DC.DateCompleted
If you wanted to include dates that are neither DateStarted nor DateCompleted, with counts of 0 and 0, then you will need a source of dates and I think it would be clearer to use two correlated sub-queries in select clause instead of joins and count distinct:
select DateSource."Date"
, (select count(*)
from Registrations
where DateStarted = DateSource."Date") as StartCount
, (select count (*)
from Registrations
where DateCompleted = DateSource."Datge") as CompleteCount
from DateSource -- implementation of date source left as exercise
where DateSource.Date between #LowDate and #HighDate