PostgreSQL Functions – How to Pass Collation to UDF

functionspostgresql

How can we pass the collation to a user-defined function?

e.g. I can use a hardcoded collation (e.g. "de-AT-x-icu") like this:

CREATE OR REPLACE FUNCTION test(agg_1 jsonb, agg_2 jsonb) RETURNS jsonb
    LANGUAGE SQL
AS
$$
SELECT  jsonb_build_object(
  'count', (agg_1 -> 'count')::int8 + (agg_2 -> 'count')::int8
, 'max', greatest(agg_1 ->> 'max' collate "de-AT-x-icu", agg_2 ->> 'max' collate "de-AT-x-icu")
)
$$;

and the test works:

select test(
  jsonb_build_object('count', 2 , 'max', 'H'),
  jsonb_build_object('count', 0, 'max', 'i')
);

We tried to use a text-argument,

CREATE OR REPLACE FUNCTION test2(agg_1 jsonb, agg_2 jsonb, text_collation text) RETURNS jsonb
    LANGUAGE SQL
AS
$$
SELECT  jsonb_build_object(
                'count', (agg_1 -> 'count')::int8 + (agg_2 -> 'count')::int8
            , 'max', greatest(agg_1 ->> 'max' collate text_collation, agg_2 ->> 'max' collate text_collation)
            )
$$;

but this fails with:

ERROR: collation "text_collation" for encoding "UTF8" does not exist

What is the best way to pass/use a collation in a UDF in such a case?

Best Answer

Collations are database objects, so their name is an identifier.

You cannot use a parameter for an identifier, you'd have to use dynamic SQL in a PL/pgSQL function:

RETURN QUERY EXECUTE
   format(
      'SELECT ... COLLATE %I ...'
      text_collation
   );

Tricky detail

The crosstab() function takes query strings as parameters. Function parameters are not visible inside crosstab(). So you need to pass in the date values as string literals!
I suggest the function format() for convenience. For example, the second parameter becomes:

,format('VALUES(%L::date), (%L), (%L), (%L)'
        , $1, $1 - 1, $1 - 7, ($1 - interval '1 month')::date
       )

instead of:

,$$
 VALUES
       (now()::date)
     , (now()::date - 1)
     , (now()::date - 7)
     , ((now() - interval '1 month')::date)
 $$

Complete code

SQL function

CREATE OR REPLACE FUNCTION foo_sql(_date date)
  RETURNS TABLE (
   name       text
 , keyword_id int
 , project_id int
 , the_date   date
 , today      int
 , yesterday  int
 , week       int
 , month      int) AS
$func$
BEGIN

SELECT k.name, f.keyword_id, f.project_id, _date
     , f.t, f.y, f.w, f.m
FROM   crosstab (
   $$
   SELECT rn
        , pk.keyword_id
        , pk.project_id
        , d.created_at
        , COALESCE(pr.pos, 0)
   FROM  (SELECT *, row_number() OVER () AS rn FROM project_keyword) pk
   CROSS  JOIN (
       $$
    || format('VALUES(%L::date), (%L), (%L), (%L)'
              , $1, $1 - 1, $1 - 7, ($1 - interval '1 month')::date)
   || $$
     ) d(created_at)
   LEFT JOIN (
      SELECT keyword_id
           , project_id
           , created_at::date AS created_at
           , min(position) AS pos
      FROM   project_report
      GROUP  BY keyword_id, project_id, created_at::date
      ) pr USING (keyword_id, project_id, created_at)
   ORDER  BY pk.rn, d.created_at
   $$

  ,format('VALUES(%L::date), (%L), (%L), (%L)'
              , $1, $1 - 1, $1 - 7, ($1 - interval '1 month')::date)
   ) f (rn int, keyword_id int, project_id int
       , t int, y int, w int, m int)
JOIN   keyword k USING (keyword_id);

END
$func$  LANGUAGE sql;

PL/pgSQL function

Shorter, reusing the VALUES expression. Probably a bit faster, too.

CREATE OR REPLACE FUNCTION foo_plpgsql(_date date)
  RETURNS TABLE (name text, keyword_id int, project_id int, the_date date
               , today int, yesterday int, week int, month int) AS
$func$
DECLARE
   _dates text := format('VALUES(%L::date), (%L), (%L), (%L)'
                        , $1, $1 - 1, $1 - 7, ($1 - interval '1 month')::date);
BEGIN

SELECT k.name, f.keyword_id, f.project_id, _date, f.t, f.y, f.w, f.m
FROM   crosstab (
  'SELECT rn
        , pk.keyword_id
        , pk.project_id
        , d.created_at
        , COALESCE(pr.pos, 0)
   FROM  (SELECT *, row_number() OVER () AS rn FROM project_keyword) pk
   CROSS  JOIN (
       ' || _dates || '
     ) d(created_at)
   LEFT JOIN (
      SELECT keyword_id
           , project_id
           , created_at::date AS created_at
           , min(position) AS pos
      FROM   project_report
      GROUP  BY keyword_id, project_id, created_at::date
      ) pr USING (keyword_id, project_id, created_at)
   ORDER  BY pk.rn, d.created_at'

  ,_dates
   ) f (rn int, keyword_id int, project_id int, t int, y int, w int, m int)
JOIN   keyword k USING (keyword_id);

END
$func$  LANGUAGE plpgsql;

CTE

For completeness, the "same" without persisting a function, with CTEs:

WITH d(day)   AS (SELECT '2014-07-07'::date)  -- provide your date here
,    v(dates) AS (
   SELECT format('VALUES(%L::date), (%L), (%L), (%L)'
                 , day, day - 1, day - 7
                 ,(day - interval '1 month')::date)
   FROM d
   )
SELECT k.name, f.keyword_id, f.project_id, d.day AS the_date
     , f.t AS today, f.y As yesterday, f.w AS week, f.m AS month
FROM   crosstab (
  'SELECT rn
        , pk.keyword_id
        , pk.project_id
        , d.created_at
        , COALESCE(pr.pos, 0) AS pos
   FROM  (SELECT *, row_number() OVER () AS rn FROM project_keyword) pk
   CROSS  JOIN (
       ' || (SELECT dates FROM v) || '
     ) d(created_at)
   LEFT JOIN (
      SELECT keyword_id
           , project_id
           , created_at::date AS created_at
           , min(position) AS pos
      FROM   project_report
      GROUP  BY keyword_id, project_id, created_at::date
      ) pr USING (keyword_id, project_id, created_at)
   ORDER  BY pk.rn, d.created_at'

  ,(SELECT dates FROM v)
   ) f (rn int, keyword_id int, project_id int
       , t int, y int, w int, m int)
JOIN   keyword k USING (keyword_id);

MySQL UDF – wrong ELF class: ELFCLASS32

Unfortunately, mysql_config contains quite a bit of black magic and quite a bit of blatant sketchiness. It's notoriously untrustworthy if you have more than one copy of it or the version doesn't match your server build.

Unfortunately, too, if you're using one from a distro (not official binaries) then that utility may not even be in the same package as the server... it's bundled with the client, which could easily be 32 bit and work just fine with a 64-bit server.

$ mysql_config
The program 'mysql_config' can be found in the following packages:
 * libmysqlclient-dev
 * libmariadbclient-dev
Try: sudo apt-get install <selected package>

Not confidence-inspiring.

$ mysql_config --version
5.5.40

Neat, I don't even have server on this particular system, and never have.

So, that gives you a sense of how much hocus pocus this little utility can sometimes offer up.

Fundamentally, though, there's this: wrong ELF class: ELFCLASS32 means you compiled the plugin for 32 bit, but your server is indeed 64 bit.

The missing piece from your compiler flags is -m64. If you add that and remove the -m32 (I'm not sure where that's coming from... mysql_config doesn't include either one on my test system) then you should be able to get that plugin to compile and install.

You have the plugin in the correct directory, or the system wouldn't have found the file to complain about the architecture it was compiled for.

Best Answer

Related Solutions

How to Pass a Parameter into a Function in PostgreSQL

Tricky detail

Complete code

SQL function

PL/pgSQL function

CTE

MySQL UDF – wrong ELF class: ELFCLASS32

Related Question