Need to automate sorting while accounting for overlapping values in SQL - sql

Question

I'm facing a challenging request that's had me beating my head against the keyboard. I need to implement a script which will sort and summarize a dataset while accounting for overlapping values which are associated with different identifiers. The table from which I am selecting contains the following columns:
BoxNumber (Need to group by this value, which serves as the identifier)
ProdBeg (Contains the first 'page number' for the document/record)
ProdEnd (Contains the last 'page number' for the document/record)
DateProduced (Date the document was produced)
ArtifactID (Unique identifier for each document)
NumPages (Contains the number of pages associated with each document)
Selecting a sample of the data with no conditions resembles the following (sorry for lousy formatting):
BoxNumber | ProdBeg | ProdEnd | DateProduced | ArtifactID | NumPages
1200 | ABC01 | ABC10 | 12/4/2013 | 1564589 | 10
1201 | ABC11 | ABC20 | 12/4/2013 | 1498658 | 10
1200 | ABC21 | ABC30 | 12/4/2013 | 1648596 | 10
1200 | ABC31 | ABC40 | 12/4/2013 | 1489535 | 10
Using something like the following effectively groups and sorts the data by box number while accounting for different DateProduced dates, but does not account for overlapping ProdBeg/ProdEnd values between different BoxNumbers:
SELECT BoxNumber, MIN(ProdBeg) AS 'ProdBeg', MAX(ProdEnd) AS 'ProdEnd', DateProduced, COUNT(ArtifactID) AS 'Documents', SUM(NumPages) AS 'Pages'
FROM MyTable
GROUP BY BoxNumber, DateProduced
ORDER BY ProdBeg, ProdEnd
This yields:
BoxNumber | ProdBeg | ProdEnd| DateProduced | Documents| Pages
1200 | ABC01 | ABC40 | 12/4/2013 | 3 | 30
1201 | ABC11 | ABC20 | 12/4/2013 | 1 | 10
Here, it becomes apparent that the ProdBeg/ProdEnd values for box 1201 overlap those for box 1200. No variation on the script above will work, as it will inherently ignore any overlaps and only select the min/max. We need something which will produce the following result:
BoxNumber | ProdBeg | ProdEnd | DateProduced | Documents| Pages
1200 | ABC01 | ABC10 | 12/4/2013 | 1 | 10
1201 | ABC11 | ABC20 | 12/4/2013 | 1 | 10
1200 | ABC21 | ABC40 | 12/4/2013 | 2 | 20
I'm just not sure how we can group by box number without showing only distinct values (which can result in overlaps for ProdBeg/ProdEnd). Any suggestions would be greatly appreciated! The environment version is SQL 2008 R2 (SP1).

Answer 1

Yuch. This would be helped at least if you had lead()/lag() as in SQL Server 2012. But it is doable.
The idea is the following:
Add a variable that is the number part of the code (the last two digits).
Calculate the next number in the sequence.
Calculate a flag if there is a gap to the next number. This is the start of a "group".
Calculate the cumulative sum of the "start of a group" flag. This is a group id.
Do the aggregation.
The following query follows this logic. I didn't include the date produced. This seems redundant with the number, unless a box can appear on multiple days. (Adding the date produced is just a matter of adding the condition to the where clauses.) The resulting query is:
with bp as (
select t.*,
cast(right(prodbeg, 2) as int) as pbeg,
cast(right(prodend, 2) as int) as pend
from mytable t
),
bp1 as (
select bp.*,
(select top 1 pbeg
from bp bp2
where bp2.pbeg < bp.pbeg and pb2.BoxNumber = pb.BoxNumber
order by bp2.pbeg desc
) as prevpend
from bp
),
bp2 as (
select bp1.*,
(select sum(case when prevpend = pbeg - 1 then 0 else 1 end)
from bp1 bp1a
where bp1a.pbeg < bp1.pbeg and pb1a.BoxNumber = pb1.BoxNumber
) as groupid
from bp1
)
select BoxNumber, MIN(ProdBeg) AS ProdBeg, MAX(ProdEnd) AS ProdEnd, DateProduced,
COUNT(ArtifactID) AS 'Documents', SUM(NumPages) AS 'Pages'
FROM bp2
GROUP BY BoxNumber, groupid
ORDER BY ProdBeg, ProdEnd;