User-defined function in order by?

Question

Hey guys.  We're trying to port our postgres db to Oracle, and I'm stumbling on this one issue.  We have columns called 'sortSequences', which are arbitrarily long arrays of integers.  For POC purposes, I am using varray(), but that probably won't work as an production implementation, because of it's strongly-defined size.  Anyway, the crux of the problem is that I need to be able to select * from foo order by sortSequence, where the ordering evaluation is custom code.  The array of integers essentially represent infinite-precision math, and we have a deterministic algorithm for comparing them, but one cannot simply produce a value from them (which I think prevents me from using a map member function in a user-defined object type.)  In other words, we don't really care what the values are, we are just using them collectively as a sorting index.  An aggregate function didn't look like the right answer, and a simple scalar function is out of the question.  We were looking at domain indexes, but it's not clear to me that even if we created an operator bound to our user-defined type that the db would use that function for order by evaluation...
Can anybody think of a solution for this problem?  If we absolutely have to live without this functionality, we can replicate it in app code, but we wish to avoid that at all costs...
thanks,
brian

Answer 1

Doesn't look very promising for a general case.
Even for short sequences and moderate number values the computed numbers to order by are quite large.
with
data_table as
(select '[10,1,1]' csv from dual union all
select '[1,5,-1]' from dual union all
select '[1,1,1,1]' from dual union all
select '[3,-5,1,3,2]' from dual union all
select '[1,1,2]' from dual union all
select '[1,20]' from dual union all
select '[2]' from dual union all
select '[1,7,1]' from dual
mapped as
(select csv,
        lvl,
        item - min(item) over (order by null rows between unbounded preceding and unbounded following) n,
        max(item) over (order by null rows between unbounded preceding and unbounded following) -
        min(item) over (order by null rows between unbounded preceding and unbounded following) m,
        max(lvl) over (order by null rows between unbounded preceding and unbounded following) x
   from (select csv,
                level lvl,
                to_number(regexp_substr(t_csv,'[^,]+',1,level)) item
           from (select csv,
                        ltrim(rtrim(csv,']'),'[') t_csv
                   from data_table
         connect by level <= regexp_count(csv,',') + 1
                and prior csv = csv
                and prior sys_guid() is not null
joined as
(select l.csv,l.lvl,nvl(m.n,0) n,l.m,row_number() over (partition by l.csv order by l.lvl desc) rn
   from mapped m
        right outer join
        (select csv,lvl,m
           from (select level lvl,
                        (select max(m) from mapped) m
                           from dual
                         connect by level <= (select max(x) from mapped))
                cross join
                (select csv from data_table)
        ) l
     on l.csv = m.csv
    and l.lvl = m.lvl
order by csv,lvl
select csv,o
  from (select csv,rn,o
          from joined
         model
         partition by (csv)
         dimension by (lvl)
         measures (rn,n,m,0 o)
         rules
         (o[lvl] = (nvl(o[cv(lvl) - 1],0) + n[cv(lvl)]) * m[cv(lvl)])
where rn = 1
order by o
CSV
O
[1,1,1,1]
61035000
[1,1,2]
61046875
[1,5,-1]
62562500
[1,7,1]
63375000
[1,20]
68359375
[2]
68359375
[3,-5,1,3,2]
78223925
[10,1,1]
148921875