ChatGPT解决这个技术问题 Extra ChatGPT

Postgresql GROUP_CONCAT equivalent?

I have a table and I'd like to pull one row per id with field values concatenated.

In my table, for example, I have this:

TM67 | 4  | 32556
TM67 | 9  | 98200
TM67 | 72 | 22300
TM99 | 2  | 23009
TM99 | 3  | 11200

And I'd like to output:

TM67 | 4,9,72 | 32556,98200,22300
TM99 | 2,3    | 23009,11200

In MySQL I was able to use the aggregate function GROUP_CONCAT, but that doesn't seem to work here... Is there an equivalent for PostgreSQL, or another way to accomplish this?


L
Lukas Eder

Since 9.0 this is even easier:

SELECT id, 
       string_agg(some_column, ',')
FROM the_table
GROUP BY id

Note that the syntax also allows you to specify the order of values in the string (or array, using array_agg) e.g. string_agg(some_column, ',' ORDER BY some_column) or even string_agg(surname || ', ' || forename, '; ' ORDER BY surname, forename)
It's awesome that distinct works with string_agg, so one can use string_agg(distinct some_solumn, ',')
Note that you may need to cast the column value to TEXT if it is a non-stringable value (ie. uuid). This would look like string_agg(some_column::text, ',')
If column type is integer, don't forget to convert or use concat(column, '') for implicit conversion
PS, string_agg(value, ',') can be use only if the value is string. For example, if the value is integer value need to be casted to string like value::character varying
A
Ariel Allon

This is probably a good starting point (version 8.4+ only):

SELECT id_field, array_agg(value_field1), array_agg(value_field2)
FROM data_table
GROUP BY id_field

array_agg returns an array, but you can CAST that to text and edit as needed (see clarifications, below).

Prior to version 8.4, you have to define it yourself prior to use:

CREATE AGGREGATE array_agg (anyelement)
(
    sfunc = array_append,
    stype = anyarray,
    initcond = '{}'
);

(paraphrased from the PostgreSQL documentation)

Clarifications:

The result of casting an array to text is that the resulting string starts and ends with curly braces. Those braces need to be removed by some method, if they are not desired.

Casting ANYARRAY to TEXT best simulates CSV output as elements that contain embedded commas are double-quoted in the output in standard CSV style. Neither array_to_string() or string_agg() (the "group_concat" function added in 9.1) quote strings with embedded commas, resulting in an incorrect number of elements in the resulting list.

The new 9.1 string_agg() function does NOT cast the inner results to TEXT first. So "string_agg(value_field)" would generate an error if value_field is an integer. "string_agg(value_field::text)" would be required. The array_agg() method requires only one cast after the aggregation (rather than a cast per value).


And in 9.0 you will have listagg()
To get CSV the query should be: SELECT id_field, array_to_string(array_agg(value_field1), ','), array_to_string(array_agg(value_field2),', ') FROM data_table GROUP BY id_field
You can't use array_to_string in all cases here. If your value_field contains an embedded comma, the resulting CSV is incorrect. Using array_agg() and casting to TEXT properly quotes strings with embedded commas. The only caveat is that it also includes the starting and ending curly braces, hence my statement "and edit as needed". I will edit to clarify that point.
FYI: here's a link to docs on array_agg in 8.4
K
Kornel
SELECT array_to_string(array(SELECT a FROM b),', ');

Will do as well.


Is it possible to do something like in this comment, where you aggregate in a certain order? How would you handle grouping by one column and ordering by another one (for example, to concatenate variables within a longitudinal data set)?
m
max_spy

Try like this:

select field1, array_to_string(array_agg(field2), ',')
from table1
group by field1;

K
Kérdezösködő Indián

Assuming that the table your_table has three columns (name, id, value), the query is this one:

  select name, 
         array_to_string(array_agg(id), ','), 
         array_to_string(array_agg(value), ',')
    from your_table
group by name
order by name
;

"TM67"  "4,9,72"    "32556,98200,22300"
"TM99"  "2,3"       "23009,11200"

KI


S
Sławomir Lenart

and the version to work on the array type:

select
  array_to_string(
    array(select distinct unnest(zip_codes) from table),
    ', '
);

Duplicate answer, @max_spy said the same thing five years ago
@EmilVikström: you have right to be wrong, but read carefully. It's not only different, but I gave an example, which works with array type - like zip_codes being character varying(5)[]. Also, I've verified that for my purpose - unnest is needed, otherwise you will see ERROR: cannot accumulate arrays of different dimensionality.
m
mypetlion

My sugestion in postgresql

SELECT cpf || ';' || nome || ';' || telefone  
FROM (
      SELECT cpf
            ,nome
            ,STRING_AGG(CONCAT_WS( ';' , DDD_1, TELEFONE_1),';') AS telefone 
      FROM (
            SELECT DISTINCT * 
            FROM temp_bd 
            ORDER BY cpf DESC ) AS y
      GROUP BY 1,2 ) AS x   

Why are you doing ORDER BY in an inner query? Won't the ordering get lost anyway?
C
Chris Catignani

Hope below Oracle query will work.

Select First_column,LISTAGG(second_column,',') 
    WITHIN GROUP (ORDER BY second_column) as Sec_column, 
    LISTAGG(third_column,',') 
    WITHIN GROUP (ORDER BY second_column) as thrd_column 
FROM tablename 
GROUP BY first_column

I tested it on rextester.com/l/postgresql_online_compiler and didn't work: 42883: function listagg(text, unknown, text) does not exist
Oracle has different syntax and functions than postgres.