5

Here we have two sets of numbers. The problem is that I can't figure out how to get from the input to the output of numbers (DDL and DML below and also in the fiddle here).

current_data
1,2,3
1,4
1,5,7
8,9,10
10,11,15

expected_outcome
1,2,3,4,5,7
8,9,10,11,15

We are simply trying to match a group of numbers based on if a single number matches in any other group. Then merge all of these groups.

For instance.

If we have:

('1,2,3'),
('1,4'),
('1,5,7')

We want:

(1,2,3,4,5,7)

We consolidate them into a single line in PostgreSQL.

or (another example):

('8,9,10'), 
('10,11,15')

desired output:

(8,9,10,11,15)

The query would group these numbers because they have the number 10 in common. But it wouldn't group with the previous row (i.e. (1,2,3,4,5,7)) that don't share a number.

When we have these groups in a table. They will only group together if they at least have one matching number in each group.

======== DDL and DML =============

create table current (current_data text not null);
create table expected_output (expected_outcome text not null);

insert into current (current_data) values ('1,2,3'),('1,4'),('1,5,7'),('8,9,10'), ('10,11,15');
insert into expected_output (expected_outcome) values ('1,2,3,4,5,7'),('8,9,10,11,15');
4
  • Hi, and welcome to dba.se! I edited your question to make the data in the question consistent with the data in the fiddle - which is what I used for my answer. I also rearranged it slightly - hope this is OK? You can reverse the changes if you so wish! BTW (+1) for a) providing a fiddle with your first question - if only all posters did this and b) for asking a question which I found interesting and tricky at the same time!
    – Vérace
    Aug 17 at 8:53
  • 1
    What if there are three groups, '2,3,4' '4,5,6' and '6,7,8'? Do you expect to have two groups in the output, '2,3,4,5,6' (common is 4) and '4,5,6,7,8' (common is 6)? Or a single group, '2,3,4,5,6,7,8'? Aug 18 at 14:29
  • Will there only ever be one value that is common between groups that merge? In the sample data the first three group share '1', the second two '10'. Would it be possible to have, say, (1,2,3,4,5), (2,4,6,8) where both 2 and 4 are common to both? Aug 19 at 13:49
  • Think of the data as a graph. Input values are nodes. Values in a group are connected by edges. You're looking for disjoint subgraphs in the input. Aug 22 at 20:29
3

If you want to see how this problem should NOT be tackled, take a look at the edits to see my first two solutions.

TL;DR (fiddle)

SELECT num AS "Token", STRING_AGG(val::TEXT, ',' ORDER BY val::INT) AS "The string"
FROM
(
  SELECT DISTINCT t02.d AS num, UNNEST(t03.d2) AS val FROM
  (
    SELECT d, COUNT(d) AS cnt
    FROM
    (
      SELECT 
        UNNEST(STRING_TO_ARRAY(c.current_data, ','))::INT AS d,
        ROW_NUMBER() OVER () AS rn
      FROM current c
    ) AS t01
    GROUP BY d
    HAVING COUNT(d) > 1
  ) AS t02
  JOIN
  (
    SELECT 
      STRING_TO_ARRAY(c.current_data, ',')::INT[] AS d2
    FROM current c
  ) AS t03
  ON t02.d = ANY(t03.d2)
) AS t04
GROUP BY num
ORDER BY num;

Result:

Token             The string
   11                 1,11,99,1203,2222,6666
   44                  13,44,1005,1110,10078
 1005    13,44,992,1005,1007,1008,1110,10078

/TL;DR

Introduction:

SQL was not designed for work with strings and, in particular, looking inside strings and manipulating individual elements of that string (unlike, say, C) - and although most RDBMS suppliers now have a healthy stable of string functions, it's not SQL's forté (or strong point)!

I changed the data quite a bit (even a lot!) so that I could test my solution - and when you're dealing with row numbers that are similar to your actual data, things quickly become confusing!

You might well be better off (performance wise) using C (or another language of your choice) to do this on the app side! Just because it can be done in "pure" SQL, that doesn't mean it should be done!

With regards to the word "pure", this solution makes use of a few non-standard extensions - so does a SQL Server solution and I imagine that it would probably be nigh-on impossible to solve this without these handy tools!

I found these answers helpful:

The table is as created in the question, the input data is as follows:

INSERT INTO current (current_data) VALUES 
('992,1005,1007,1008'),
('44,1005,1110'),
('13,44,1005,10078'),
('11,1203,6666'),
('1,11,99,2222'),
('1234');           -- note the singleton!

and the desired output is:

Token                       The string
   11           1,11,99,1203,2222,6666
   44            13,44,1005,1110,10078
 1005     13,44,992,1005,1007,1008,100

The 1234 singleton is not included in the desired resultset because the question asks:

if one value from the group matches another group.

My reading of this (if .... matches...) is that singletons are excluded!

The final SQL is shown above but also (and the whole process step-by-step) can be found on the fiddle:

Step 1:

We need to find a way of identifying which numbers in the strings belong together - we do this using the ROW_NUMBER() function! If you're unclear on SQL window functions, I would urge you to watch Bruce Momjian's YouTube "The Magic of Window Functions in Postgres" presentation. They are extremely powerful, part of the SQL standard, and will repay any effort spent learning how to use them many times over!

So, we run this query:

SELECT 
  c.current_data AS d,
  ROW_NUMBER() OVER () AS rn
FROM current c;

Result:

         d             rn
992,1005,1007,1008      1
      44,1005,1110      2
  13,44,1005,10078      3
      11,1203,6666      4
      1,11,99,2222      5
              1234      6

Step 2:

Above, I mentioned that strings are not SQL's forté - what is their forté is tables and records - so, we turn our data into a table by running:

SELECT 
  UNNEST(STRING_TO_ARRAY(c.current_data, ','))::INT AS d,
  ROW_NUMBER() OVER () AS rn
FROM current c
ORDER BY rn, d;

Result:

    d         rn
  992          1
 1005          1
 1007          1
 1008          1
   44          2
 1005          2
  ...
  ... snipped for brevity
  ...

This makes use of two non-standard functions - STRING_TO_ARRAY() and UNNEST()!

STRING_TO_ARRAY does what it says on the tin! It takes the string and the provided delimiter (',') and turns it into an array. PostgreSQL makes use of arrays "under the hood" so this is quite an efficient type!

The :: operator is PostgreSQL shorthand for CASTing - way more elegant than any other RDBMS - so we now have an array of integers.

We then use the UNNEST function which turns the elements of these arrays into rows in a "virtual" table - so we now have a table which contains our original combined string data as integer rows in a table!

Step 3:

Now, we determine the tokens in which we are interested - i.e. the ones which are repeated across groups! We make use of the SQL above and the aggregate COUNT() function (which can also be a window function to operate on subsets - i.e. "windows" - of your data) and the HAVING clause, since we're only interested in those which are repeated - i.e. COUNT(x) > 1 as follows:

SELECT 
  token, COUNT(token)  -- the COUNT(token) here is not required and can be omitted
FROM
(
  SELECT 
    t.rn AS rn,
    UNNEST(STRING_TO_ARRAY(t.d, ',')::INT[]) AS token
  FROM
  (
    SELECT 
      c.current_data AS d,
      ROW_NUMBER() OVER () AS rn
    FROM current c
  ) AS t
) AS u
GROUP BY token
HAVING COUNT(token) > 1;

Result:

token   count
   11       2
   44       2
 1005       3

Step 4:

Now, we take our duplicate tokens and join these back to our original data (converted to an integer array using STRING_TO_ARRAY without UNNESTing) to pick out individual string elements as integers which occur in the same range as our duplicate tokens.

The test for matching is the ANY() array function which:

The right-hand side is a parenthesized expression, which must yield an array value. The left-hand expression is evaluated and compared to each element of the array using the given operator, which must yield a Boolean result. The result of ANY is “true” if any true result is obtained. The result is “false” if no true result is found (including the case where the array has zero elements).

So, for each dupe, if there is ANY matching integer element in the array (constructed from the original string) which matches the integer dupe token, the dupe and that matching value will be returned!

The DISTINCT clause will eliminate records with the same values for both fields and makes the upstream SQL that bit clearer:

SELECT DISTINCT t02.d AS num, UNNEST(t03.d2) AS val FROM
(
  SELECT d, COUNT(d) AS cnt
  FROM
  (
    SELECT 
      UNNEST(STRING_TO_ARRAY(c.current_data, ','))::INT AS d,
      ROW_NUMBER() OVER () AS rn
    FROM current c
  ) AS t01
  GROUP BY d
  HAVING COUNT(d) > 1
) AS t02
JOIN
(
  SELECT 
    STRING_TO_ARRAY(c.current_data, ',')::INT[] AS d2
  FROM current c
) AS t03
ON t02.d = ANY(t03.d2);

Result:

  num        val
   44        1005
   44        1110
   11          99
   11        2222
 1005         992
 1005        1008
   11          11
  ...
  ... snipped for brevity
  ...

So, we can see that for every duplicate integer token, we have a record for every integer that also occurs in the same row - including the token itself.

Step 5 (final):

We now make use of the STRING_AGG() function to group the matching elements with their duplicates by grouping on the dupes. The use of DISTINCT above means that we don't have to use it in the STRING_AGG() function. Note the sorting - it is done by using the ::INT casting operator so we have the values in the strings sorted by their numerical value!

SELECT 
  num AS "Token", -- the token is not required and can be omitted
  STRING_AGG(val::TEXT, ',' ORDER BY val::INT) AS "The string"
FROM
(
  SELECT DISTINCT t02.d AS num, UNNEST(t03.d2) AS val FROM
  (
    SELECT d, COUNT(d) AS cnt
    FROM
    (
      SELECT 
        UNNEST(STRING_TO_ARRAY(c.current_data, ','))::INT AS d,
        ROW_NUMBER() OVER () AS rn
      FROM current c
    ) AS t01
    GROUP BY d
    HAVING COUNT(d) > 1
  ) AS t02
  JOIN
  (
    SELECT 
      STRING_TO_ARRAY(c.current_data, ',')::INT[] AS d2
    FROM current c
  ) AS t03
  ON t02.d = ANY(t03.d2)
) AS t04
GROUP BY num
ORDER BY num;

Result:

Token                              The string
   11                  1,11,99,1203,2222,6666
   44                   13,44,1005,1110,10078
 1005     13,44,992,1005,1007,1008,1110,10078

If you are unclear about any of the above - and it's a lot to digest, I suggest that you remove some of the function nesting and see what happens if you break down each composite function into its individual parts. At the end of the fiddle, I've included the output of

EXPLAIN (ANALYZE, BUFFERS, VERBOSE)
< the query >

As mentioned above, SQL wasn't designed for string manipulation, so I can make no guarantees about performance with a large resultset - anything from the fiddle is fairly meaningless other than to give a general impression! I would urge you to test any solution with a realistic dataset!

As a final word, again as mentioned above, just because it can be done, doesn't mean that it should be done! What you really should do is put your data into proper normalised data structures and database records.

An INTEGER is a number and should not be stored as TEXT and especially not as a concatenated .csv string - your query would be trivial if your database was correctly normalised! Furthermore, you have no way of enforcing any sort of data integrity - anybody could insert 34, 54, XXX, 73, Oh... as a string. You could have some sort of fancy REXEXP expression in a CHECK CONSTRAINT - but the time spent writing that would be better spent changing your schema!

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.