Logical Analysis...

Question

Okay,
I have two different SQL statements, that logically I think should be identical, but then again I'm not the expert so I've come here.
Prep Table:
create table count_table as (
  pcount number(12),
  ccount number(12)
insert into count_table
  (pcount, ccount)
  values (0,0);
update count_table
  set pcount = (select count(*) from ptable);
update count_table
  set ccount = (select count(*) from ctable);This is a table I create to simply my count percentages for the following two select statements:
1:
select /*+ USE_HASH(c inp) */
  count(distinct c.c_id)
from
  customer c inner join (
    select
      c_ID,
      pnum1,
      row_number() over (order by pnum1 desc) rna,
      pnum2,
      row_number() over (order by pnum2 desc) rnb,
    from ptable
  ) inp on (i.c_id = inp.c_id)
where
  (pnum1 > 0 and rna <= (select pcount * 0.15 from count_table)) or   
  (pnum2 > 0 and rnb <= (select pcount * 0.15 from count_table));2:
select /*+ USE_HASH(c pt1 pt2) */
  count(distinct c.c_id)
from
  customer c left join (
    select c_id from (
      select c_id, row_number() over (order by pnum1 desc) rn
      from ptable
      where pnum1 > 0
    ) where rn <= (select pcount * 0.15 from count_table)
  ) pt1 on (c.c_id = pt1.c_id)
  left join (
    select c_id from (
      select c_id, row_number() over (order by pnum2 desc) rn
      from ptable
      where pnum2 > 0
    ) where rn <= (select pcount * 0.15 from count_table)
  ) pt2 on (c.c_id = pt2.c_id)
where
  pt1.c_id is not null or
  pt2.c_id is not null;Now, #2 is very straight forward, in that in two different sub-select statements, i retrieve the top 15% of entries in the ptable, but then I Left join them to the customer table. So I should have all of the entries in Customer table, all the entries of PT1 that match Customer, and all the entries of PT2 that match customer. By limiting the select to where either pt1.c_id or pt2.c_id is not null, it is the same as doing to seperate select statemetns with INNER JOINs between Customer and the individual PT# tables, and then UNIONing the two results.
However, in #1 I was trying to eliminate some of these steps by taking advantage of the analytical function row_number(). i was trying to basically grab the top 15% of each number in the table, without the need for multiple select statements. No real reason but because i could, and wanted to try. However, the two select statements yield slightly different counts each time. I'm wondering if anyone could give me some insight into the logic as to why.
(as always, I deal mostly in theory and work out the specifics on my own)
Thanks
Jaeden "SIfo Dyas" al'Raec Ruiner

Answer 1

Was I the only Programmer to start using SQL?
I am really confused why every time i post a question, people start asking about data. What does Data matter? I've stated it's two numerical fields, and the lack of quotes around other values mean they are all numerical. So the data could be anything. Just imaging that pcount is 10000, and pnum1 and pnum2 are 10000 random numbers between 0 and 1000.
What I'm looking for is an analysis of the logic, which operates on the data, but as a Programmer you never care about the data, as long as the logic is pure no matter what the data is it will work the way the logic was design. So far, SQL has followed suit in this axiom, so I keep thinking of SQL as a process of logic, not data.
Now, Granted:
    select c_id from (
      select c_id, row_number() over (order by pnum1 desc) rn
      from ptable
      where pnum1 > 0
    ) where rn <= (select pcount * 0.1 from count_table)In this code, I am removing the Zero (0) values from the query, before the analytic function is calculated. Meaning, that if i have 10000 values, and say 200 of them are 0, then the resulting values of RN will be 1 to 9800, or i can't add, and since we have it limited to 10% of 10000, i should (and do) get 1000 entries returned.
Second Set:
    select c_id from (
      select c_id, row_number() over (order by pnum2 desc) rn
      from ptable
      where pnum2 > 0
    ) where rn <= (select pcount * 0.1 from count_table)Let's say that pnum2 also has 200 that are 0, but 50 of them are different than pnum1, and of the 1000 rows returned by pnum2, 500 of the c_id's are returned by pnum1, so we should get a ballpark base count of 1500, 500 shared, and 500 each individual.
Now in the other one, i've updated it somewhat, thinking about what you said that the analytical function fires before the 0's are removed.
select * from ( 
  select
      c_id,
      pnum1,
      case pnum1 when 0 then 0 else row_number() over (order by pnum1 desc) end rna,
      pnum2,
      case pnum2 when 0 then 0 else row_number() over (order by pnum2 desc) end rnb
  from
      ptable
where
  (rna > 0 and rna <= (select pcount * 0.1 from count_table)) or
  (rnb > 0 and rnb <= (select pcount * 0.1 from count_table))Now, logically, given the same values from above, rna should again equal 1-9800 for both of the rna and rnb columns, with the remainder just equal to 0, so I still have 10000 values. RNA should be 1 to 9800 based upon the descending order of pnum1 and RNB should be the same for the descending order of pnum2. Yes they'll be different per each c_id, but i would still get the same cross section of two values. the 0s are always the largest row numbers, since the order is desc based upon the appropriate field.
So by limiting what i return from the inner select statement to be only when RNA is not 0 and within the 10% of the total count, and the same for RNB, logically that says the OR acts like a UNION. 500 of pnum1/RNA's c_id match 500 of pnum2/RNB's c_id, and then 500 a piece should still equal 1500.
What I need to know is why it doesn't?
      row_number() over (order by X desc)that calculates 1,2,3,4 based upon the values of X going 5,4,3,2,1,0,0,0,0,0. So why would putting one row_number() in the same select as another have any effect on the final result?
Effectively I don't see a difference between the first two select statements and:
    select c_id from (
      select c_id, pnum2, row_number() over (order by pnum2 desc) rn
      from ptable
    ) where
            rn <= (select pcount * 0.1 from count_table) and
            pnum2 > 0Primarily because i'm counting via row number, 1-10000, and the last rows, 9800-10000 are are all pnum=0, but since 9800-10000 is NOT <= 1000 why should it care?
So, I actually tested it, and i was right, either having the pnum2 >0 on the inside or outside yields the same results. So why does it not quite match when there are multiple analytical functions in the same select statement?
Maybe i'm not exactly understanding what row_number() does, and that might be my problem.
Thanks for your time,
Regards,
Jaeden "Sifo Dyas" al'Raec Ruiner