T-SQL to generate expect result - sql

Question

I have a case in SQL: Source table have three columns: ID, Cate, Type.
With same Cate, Type (A, A-) (B, B-) eliminate each other, and return rows have MAX(Id).
eg:
With cate = AM0001 : Id = 1,2,3 then Id 1,2 eliminate each other --> keep id =3.
With cate = AM003: Id= 4,6 , type = B --> keep both.
With cate = AM005: Id= 7,8,9 row: 7,8 eliminate each other --> keep: Id =9
With cate = AM0006: Id= 10,11, type = A -->keep both.
Expected result:
I'm using cursor to resolve it quite hard to resolve. Is there any clue for resolving it in T-SQL

Answer 1

Assuming that I understand the problem:
you have a number of rows with sections ("Cates") and symbols ("Type");
if there are any symbols ending in a minus sign then these indicate a row without a minus sign should be removed;
symbols are never "mixed" per section, i.e. a section can never have "A" and "B-";
there will always be a row to remove if there is a type with a minus;
rows should be removed starting with the lowest Id.
Then this should work:
DECLARE #data TABLE (
Id INT,
Cate VARCHAR(5),
[Type] VARCHAR(2));
INSERT INTO #data SELECT 1, 'AM001', 'A';
INSERT INTO #data SELECT 2, 'AM001', 'A-';
INSERT INTO #data SELECT 3, 'AM001', 'A';
INSERT INTO #data SELECT 4, 'AM003', 'B';
INSERT INTO #data SELECT 6, 'AM003', 'B';
INSERT INTO #data SELECT 7, 'AM005', 'B';
INSERT INTO #data SELECT 8, 'AM005', 'B-';
INSERT INTO #data SELECT 9, 'AM005', 'B';
INSERT INTO #data SELECT 10, 'AM006', 'A';
INSERT INTO #data SELECT 11, 'AM006', 'A';
INSERT INTO #data SELECT 12, 'AM011', 'B';
INSERT INTO #data SELECT 13, 'AM011', 'B-';
INSERT INTO #data SELECT 14, 'AM011', 'B';
WITH NumberToRemove AS (
SELECT
Cate,
COUNT(*) AS TakeOff
FROM
#data
WHERE
[Type] LIKE '_-'
GROUP BY
Cate),
Ordered AS (
SELECT
Id,
Cate,
[Type],
ROW_NUMBER() OVER (PARTITION BY Cate ORDER BY Id) AS RowId
FROM
#data
WHERE
[Type] NOT LIKE '_-')
SELECT
d.*
FROM
#data d
LEFT JOIN NumberToRemove m ON m.Cate = d.Cate
INNER JOIN Ordered o ON o.Id = d.Id
WHERE
o.RowId > ISNULL(m.TakeOff, 0);
The query works by first counting the number of rows to remove from each section ("Cate") by tallying up the number of symbols with a minus sign per section. Next it sorts the rows where the symbols don't have a minus sign and assigns each row a number in Id order ("row number"), starting back at 1 for each new section ("Cate").
Finally I just pick the rows without a minus sign symbol, where the row number is greater than the number that were to be removed. Note that if a section has no rows to remove then it will return NULL rows to remove, so I transform this to 0, because ALL rows in that section with have a row number greater than 0.
My results were:
Id Cate Type
3 AM001 A
4 AM003 B
6 AM003 B
9 AM005 B
10 AM006 A
11 AM006 A
14 AM011 B
If my assumptions were incorrect then this script could easily be amended to suit...