Question

Consider these data:

ID       Year     Code1    Code2
-------- -------- -------- --------
ABC123   99/00    10010    A1121
ABC123   00/01    10010    A1131
ABC123   01/02    10010    A1141
XYZ567   06/07    12501    B2213
XYZ567   07/08    12501    B2223

These four fields make up the primary key and I need to find the earliest occurrence of each instance of ID, with the three other keys. So, in this case, I would want:

ID       Year     Code1    Code2
-------- -------- -------- --------
ABC123   99/00    10010    A1121
XYZ567   06/07    12501    B2213

In Oracle, my current solution is to aggregate over the fourth character of Code2, which is the year index, and return the minimum then recompose the key (i.e., the varying Year part); however, this is obviously ambiguous and may not map one-to-one. (Also, this method is slow!) An ordering won’t work either because, in general, there will be multiple values under ID and Year doesn’t include the century.

I was thinking that maybe I could do a subquery for each ID and then do an ordering and pick the first item (i.e. ROWNUM=1). However, this requires at least two levels of subquery and, thus, is both awkward and slow… So, any better ideas?! Thanks 🙂

Answer 1

This is what analytic functions were invented for! You want something like:

select id, year, code1, code2 from
( select id, year, code1, code2,
         row_number() over (partition by id order by ???) as rn
  from mytable
)
where rn = 1

I left the order by as ??? because I don’t fully understand what you want here. It could be:

order by year, code1, code2

or

order by substr(code2,4,1), year

or whatever.