Date difference when group by value change - sql

Question

I have a table like this
|---------------------|------------------|---------------------|------------------|
| FILE | UNIT | DATE | ID SEQUENCE |
|---------------------|------------------|---------------------|------------------|
| 10 | 34 | 01/02/2000 | 10 |
|---------------------|------------------|---------------------|------------------|
| 10 | 34 | 01/05/2000 | 11 |
|---------------------|------------------|---------------------|------------------|
| 10 | 40 | 01/05/2000 | 12 |
|---------------------|------------------|---------------------|------------------|
| 10 | 40 | 01/02/2000 | 13 |
|---------------------|------------------|---------------------|------------------|
| 10 | 40 | 01/02/2000 | 14 |
|---------------------|------------------|---------------------|------------------|
| 10 | 40 | 01/15/2000 | 15 |
|---------------------|------------------|---------------------|------------------|
| 10 | 34 | 01/16/2000 | 16 |
|---------------------|------------------|---------------------|------------------|
| 10 | 70 | 01/17/2000 | 17 |
|---------------------|------------------|---------------------|------------------|
| 10 | 70 | 01/28/2000 | 18 |
|---------------------|------------------|---------------------|------------------|
I need to build a view like this (get the amount of days a file stay in every unit)
|---------------------|------------------|---------------------|
| FILE | UNIT | DAYS IN UNITY |
|---------------------|------------------|---------------------|
| 10 | 34 | 3 |
|---------------------|------------------|---------------------|
| 10 | 40 | 10 |
|---------------------|------------------|---------------------|
| 10 | 34 | 1 |
|---------------------|------------------|---------------------|
| 10 | 70 | 11 |
|---------------------|------------------|---------------------|
Any advice
Thanks in advance

Answer 1

This is a form of gaps-and-islands. For this purpose, I am thinking difference of row numbers:
select file, unit,
(lead(min(date), 1, max(date)) over (partition by file, unit) -
min(date)
) as days_in_unity
from (select t.*,
row_number() over (partition by file order by id_sequence) as seqnum,
row_number() over (partition by file, unit order by id_sequence) as seqnum_2
from t
) t
group by file, unit, (seqnum - seqnum_2)
Why this works is a little tricky to explain. If you look at the results of the subquery, you will see how the difference of the two row numbers is constant for the rows with the same unit value.

Answer 2

If you are using a newer version of Oracle (12c and above), it may be an idea to use MATCH_RECOGNIZE (pattern matching). In the following query, we define 2 patterns:
{1} multiple days, with a start date and an end date. This pattern may contain "iffy" dates eg the 01/02/2000 in between 01/05/2000 and 01/15/2000 for unit 40.
{2} single day: this occurs when the "start" and "end" are on one and the same day. In the MEASURES clause we pick up all the columns we need, and COALESCE them in the SELECT clause (MATCH_RECOGNIZE documentation here).
Table
create table fileandunit( FILE_, UNIT, DATE_, ID_SEQUENCE )
as
select 10, 34, to_date( '02-JAN-2000', 'DD-MON-YYYY'), 10 from dual union all
select 10, 34, to_date( '05-JAN-2000', 'DD-MON-YYYY'), 11 from dual union all
select 10, 40, to_date( '05-JAN-2000', 'DD-MON-YYYY'), 12 from dual union all
select 10, 40, to_date( '02-JAN-2000', 'DD-MON-YYYY'), 13 from dual union all
select 10, 40, to_date( '02-JAN-2000', 'DD-MON-YYYY'), 14 from dual union all
select 10, 40, to_date( '15-JAN-2000', 'DD-MON-YYYY'), 15 from dual union all
select 10, 34, to_date( '16-JAN-2000', 'DD-MON-YYYY'), 16 from dual union all
select 10, 70, to_date( '17-JAN-2000', 'DD-MON-YYYY'), 17 from dual union all
select 10, 70, to_date( '28-JAN-2000', 'DD-MON-YYYY'), 18 from dual ;
Data and patterns
select * from fileandunit order by id_sequence ;
FILE_ UNIT DATE_ ID_SEQUENCE
10 34 02-JAN-00 10 -- start
10 34 05-JAN-00 11 -- end
10 40 05-JAN-00 12 -- start
10 40 02-JAN-00 13 -- iffy
10 40 02-JAN-00 14 -- iffy
10 40 15-JAN-00 15 -- end
10 34 16-JAN-00 16 -- single day
10 70 17-JAN-00 17 -- start
10 70 28-JAN-00 18 -- end
Query
select
coalesce( RP.m_file, RP.s_file ) file_
, coalesce( RP.m_unit, RP.s_unit ) unit_
, coalesce( ( RP.m_end - RP.m_start ), 1 ) days_
from fileandunit
match_recognize(
partition by file_ order by id_sequence
measures
enddt.file_ as m_file -- m_: for multiple days
, enddt.unit as m_unit
, startdt.date_ as m_start
, enddt.date_ as m_end
, singledt.file_ as s_file -- s_: single day
, singledt.unit as s_unit
, singledt.date_ as s_date
one row per match
pattern ( ( startdt iffydt* enddt ) | singledt ) -- multiple days (or) single day
define
startdt as ( prev( date_ ) <= date_ or prev( date_ ) is null )
and ( prev( unit ) <> unit or prev( unit ) is null )
--
, enddt as ( next( date_ ) >= date_ or next( date_ ) is null )
and ( next( unit ) <> unit or next( unit ) is null )
--
, iffydt as ( prev( date_ ) >= date_ ) -- detect incorrect dates inside a multiple day block
and ( prev( unit ) = unit )
--
, singledt as ( prev( date_ ) = date_ - 1 and next( date_ ) = date_ + 1 )
and ( prev( unit ) <> unit and next( unit ) <> unit )
) RP ;
Result
FILE_ UNIT_ DAYS_
10 34 3
10 40 10
10 34 1
10 70 11
Tested with Oracle 18c. DBfiddle here.