One-to-many selfjoin removing records with the same ranking or with a substitute

Question

Sorry for my bad choice of discussion title, feel free to suggest me a more pertinent one
I've rewritten post for clarity and following the FAQ.
DB Version
I'm using Oracle Enterprise 10g 10.2.0.1.0 64bit
Tables involved
CREATE TABLE wrhwr (
wr_id INTEGER PRIMARY KEY,
eq_id VARCHAR2(50) NULL,
date_completed DATE NULL,
status VARCHAR2(20) NOT NULL,
pmp_id VARCHAR2(20) NOT NULL,
description VARCHAR2(20) NULL);
Sample data
INSERT into wrhwr  VALUES (1,'MI-EXT-0001',date'2013-07-16','Com','VER-EXC','Revisione')
INSERT into wrhwr VALUES  (2,'MI-EXT-0001',date'2013-07-01','Com','VER-EXC','Verifica')
INSERT into wrhwr  VALUES (3,'MI-EXT-0001',date'2013-06-15','Com','VER-EXC','Revisione')
INSERT into wrhwr  VALUES (4,'MI-EXT-0001',date'2013-06-25','Com','VER-EXC','Verifica')
INSERT into wrhwr  VALUES (5,'MI-EXT-0001',date'2013-04-14','Com','VER-EXC','Revisione')
INSERT into wrhwr  VALUES (6,'MI-EXT-0001',date'2013-04-30','Com','VER-EXC','Verifica')
INSERT into wrhwr  VALUES (7,'MI-EXT-0001',date'2013-03-14','Com','VER-EXC','Collaudo')
Query used
SELECT *
  FROM (SELECT eq_id,
               date_completed,
               RANK ()
               OVER (PARTITION BY eq_id
                     ORDER BY date_completed DESC NULLS LAST)
                  rn
          FROM wrhwr
         WHERE     status != 'S'
               AND pmp_id LIKE 'VER-EX%'
               AND description LIKE '%Verifica%') table1,
       (SELECT eq_id,
               date_completed,      
               RANK ()
               OVER (PARTITION BY eq_id
                     ORDER BY date_completed DESC NULLS LAST)
                  rn
          FROM wrhwr
         WHERE     status != 'S'
               AND pmp_id LIKE 'VER-EX%'
               AND description LIKE '%Revisione%') table2,
       (SELECT eq_id,
               date_completed,           
               RANK ()
               OVER (PARTITION BY eq_id
                     ORDER BY date_completed DESC NULLS LAST)
                  rn
          FROM wrhwr
         WHERE     status != 'S'
               AND pmp_id LIKE 'VER-EX%'
               AND description LIKE '%Collaudo%') table3
WHERE     table1.eq_id = table3.eq_id
       AND table2.eq_id = table3.eq_id
       AND table1.eq_id = table2.eq_id
Purpose of the above query is to selfjoin wrhwr table 3 times in order to have for every row:
eq_id;
completition date of a work request of type Verifica for this eq_id (table1 alias);
completition date of a wr of type Revisione (table2 alias) for this eq_id;
completition date of a wr of type Collaudo (table3 alias) for this eq_id;
A distinct eq_id:
can have many work requests (wrhwr records) with different completition dates or without completition date (date_completed column NULL);
in a date range can have all the types of wrhwr ('Verifica', 'Revisione', 'Collaudo') or some of them (ex. Verifica, Revisione but not Collaudo, Collaudo but not Verifica and Revisione, etc.);
substrings in description shouldn't repeat;
(eq_id,date_completed) aren't unique but (eq_id,date_completed,description,pmp_id) should be unique;
Expected output
Using sample data above I expect this output:
eq_id,table1.date_completed,table2.date_completed,table3.date_completed
MI-EXT-001,2013-07-01,2013-07-16,2013-03-14 <--- for this eq_id table3 doesn't have 3 rows but only 1. I want to repeat the most ranked value of table3 for every result row
MI-EXT-001,2013-07-01,2013-06-15,2013-03-14 <-- I don't wanna this row because table1 and table2 have both 3 rows so the match must be in rank terms (1st,1st) (2nd,2nd) (3rd,3rd)
MI-EXT-001,2013-06-25,2013-06-15,2013-03-14 <-- 2nd rank of table1 joins 2nd rank of table2
MI-EXT-001,2013-04-30,2013-04-14,2013-03-14 <-- 1st rank table1, 1st rank table2, 1st rank table3
In vector style syntax, expected tuple output must be:
ix = i-th ranking of tableX
(i1, i2, i3) IF EXISTS an i-th ranking row in every table
ELSE
(i1, b, b)
where b is the first available lower ranking of table2 OR NULL if there isn't any row  of lower ranking.
Any clues?
With the query I'm unable to remove "spurius" rows.
I'm thinking at a solution based on analytic functions like LAG() and LEAD(), using ROLLUP() or CUBE(), using nested query but I would find a solution elegant, easy, fast and easy to maintain.
Thanks

Answer 1

FrankKulash ha scritto:
About duplicate dates: I was most interested in what you wanted when 2 (or more) rows with the same eq_id and row type (e.g. 'Collaudo') had exactly the same completed_date.
In the new results, did you get the columns mixed up?  It looks like the row with eq_id='MI-EXT-0002' has 'Collaudo' in the desciption, but the date appears in the verifica column of the output, not the collaudo  column.
Why don't you want 'MI-EXT-0001' in the results?  Is it realted to the non-unique date?
For all optimization questions, see the forum FAQ:https://forums.oracle.com/message/9362003
If you can explain what you need to do in the view (and post some sample data and output as examples) then someone might help you find a better way to do it.
It looks like there's a lot of repetition in the code.  Whatever you're trying to do, I suspect there's a simpler, more efficient way to do it.
About Duplicate dates: query must show ONLY one date_completed and ignore duplicated. Those records are "bad data". You can't have 2 collaudos with the same date completed.
Collaudo stands for equipment check. A craftperson does an equipment check once a day and, with a mobile app, update the work request related to equipment and procedure of preventive maintenance, so is impossibile that complete more than one check (Collaudo) in a day, by design.
In the new results, it's my fault: during digitation I've swapped columns
With "I don't want 'MI-EXT-0001'" I mean: "I don't want to show AGAIN MI-EXT-0001. In the previous post was correct the output including MI-EXT-0001.
Regarding optimisation...
repetition of
LAST_VALUE ( 
                        MIN (CASE WHEN r_type = THEN column_name END) IGNORE NULLS) 
                     OVER (PARTITION BY eq_id ORDER BY r_num)  AS alias_column_name
is because I don't know another feasible way to have all columns needed of table wrhwr in main query, maintaining the correct order. So i get them in got_r_type and propagate them in all the subquery.
In main query I join eq table (which contains all the information about a specific equipment) with "correct" dates and columns of wrhwr table.
I filter eq table for the specific equipment standard (eq_std column).
efm_eq_tablet table and where clause
AND e.eq_id = e2.eq_id 
          AND e2.is_active = 'S'; 
means: show only rows in eq table that have an equal row in efm_eq_tablet table AND are active (represented by 'S' value in is_active column).
About the tables v2, r2 and c2
          (SELECT doc_data, doc_data_rinnovo, eq_id 
             FROM efm_doc_rep edr 
            WHERE edr.csi_id = '1011503' AND edr.doc_validita_temp = 'LIM') v2, 
          (SELECT doc_data, doc_data_rinnovo, eq_id 
             FROM efm_doc_rep edr 
            WHERE     eq_id = edr.eq_id 
                  AND edr.csi_id = '1011504' 
                  AND edr.doc_validita_temp = 'LIM') r2, 
          (SELECT doc_data, doc_data_rinnovo, eq_id 
             FROM efm_doc_rep edr 
            WHERE     edr.csi_id IN ('1011505', '1011507') 
                  AND edr.doc_validita_temp = 'LIM' 
                  AND adempimento_ok = 'SI') c2, 
Those tables contains "alternate" dates of completition to be used when there isn't any wrhwr row for an eq_id OR when all date_completed are NULL.
NVL() and NVL2() functions are used in main query in order to impletement this.
The CASE/WHEN blocks inside main query implements the behavior of selecting the correct date based of the above conditions.