可以将文章内容翻译成中文,广告屏蔽插件可能会导致该功能失效(如失效，请关闭广告屏蔽插件后再试):

问题:

I keep all my functions in a text file with 'CREATE OR REPLACE FUNCTION somefunction'. So if I add or change some function I just feed the file to psql.

Now if I add or remove parameters to an existing function, it creates an overload with the same name and to delete the original I need type in all the parameter types in the exact order which is kind of tedious.

Is there some kind of wildcard I can use to DROP all functions with a given name so I can just add DROP FUNCTION lines to the top of my file?

回答1:

You would need to write a function that took the function name, and looked up each overload with its parameter types from information_schema, then built and executed a DROP for each one.

EDIT: This turned out to be a lot harder than I thought. It looks like information_schema doesn't keep the necessary parameter information in its routines catalog. So you need to use PostgreSQL's supplementary tables pg_proc and pg_type:

CREATE OR REPLACE FUNCTION udf_dropfunction(functionname text)
  RETURNS text AS
$BODY$
DECLARE
    funcrow RECORD;
    numfunctions smallint := 0;
    numparameters int;
    i int;
    paramtext text;
BEGIN
FOR funcrow IN SELECT proargtypes FROM pg_proc WHERE proname = functionname LOOP

    --for some reason array_upper is off by one for the oidvector type, hence the +1
    numparameters = array_upper(funcrow.proargtypes, 1) + 1;

    i = 0;
    paramtext = '';

    LOOP
        IF i < numparameters THEN
            IF i > 0 THEN
                paramtext = paramtext || ', ';
            END IF;
            paramtext = paramtext || (SELECT typname FROM pg_type WHERE oid = funcrow.proargtypes[i]);
            i = i + 1;
        ELSE
            EXIT;
        END IF;
    END LOOP;

    EXECUTE 'DROP FUNCTION ' || functionname || '(' || paramtext || ');';
    numfunctions = numfunctions + 1;

END LOOP;

RETURN 'Dropped ' || numfunctions || ' functions';
END;
$BODY$
  LANGUAGE plpgsql VOLATILE
  COST 100;

I successfully tested this on an overloaded function. It was thrown together pretty fast, but works fine as a utility function. I would recommend testing more before using it in practice, in case I overlooked something.

回答2:

Basic query

This query creates all necessary DDL statements (simplified with cast to regprocedure):

SELECT 'DROP FUNCTION ' || oid::regprocedure
FROM   pg_proc
WHERE  proname = 'my_function_name'  -- name without schema-qualification
AND    pg_function_is_visible(oid);  -- restrict to current search_path ..
                                     -- .. you may or may not want this

Output:

DROP FUNCTION my_function_name(string text, form text, maxlen integer);
DROP FUNCTION my_function_name(string text, form text);
DROP FUNCTION my_function_name(string text);

Execute the commands (after a plausibility check).

The function name is case-sensitive and with no added double-quotes when passed as text parameter to match against pg_proc.proname.

The cast to the object identifier type regprocedure (oid::regprocedure) makes all identifiers safe against SQL injection (by way of maliciously malformed identifiers). When converting to text, the function name is double-quoted and schema-qualified according to the current search_path automatically where needed.

pg_function_is_visible(oid) restricts the selection to functions in the current search_path. You may or may not want that. With the condition pg_function_is_visible(oid) in place, the function is guaranteed to be visible.

If you have multiple functions of the same name in multiple schemas, or overloaded functions with various function arguments, all of those will be listed separately. You may want to restrict to specific schema(s) or specific function parameter(s) after all.

When / how are default value expression functions bound with regard to search_path?

Function

You can build a plpgsql function around this to execute the statements immediately with EXECUTE. For Postgres 9.1 or later: Careful! It drops your functions!

CREATE OR REPLACE FUNCTION f_delfunc(_name text, OUT func_dropped int) AS
$func$
DECLARE
   _sql text;
BEGIN
   SELECT count(*)::int
        , 'DROP FUNCTION ' || string_agg(oid::regprocedure::text, '; DROP FUNCTION ')
   FROM   pg_proc
   WHERE  proname = _name
   AND    pg_function_is_visible(oid)
   INTO   func_dropped, _sql;  -- only returned if trailing DROPs succeed

   IF func_dropped > 0 THEN    -- only if function(s) found
     EXECUTE _sql;
   END IF;
END
$func$ LANGUAGE plpgsql;

Call:

SELECT * FROM f_delfunc('my_function_name');

Or just:

SELECT f_delfunc('my_function_name');

This way you don't get the column name func_dropped for the result column. May not matter to you.

The function returns the number of functions found and dropped (no exception raised) - 0 if none were found.

It assumes a (default) search_path where pg_catalog has not been moved around.
More in these related answers:

How does the search_path influence identifier resolution and the "current schema"
Truncating all tables in a Postgres database
PostgreSQL parameterized Order By / Limit in table function

For Postgres versions older than 9.1 or older variants of the function using regproc and pg_get_function_identity_arguments(oid) check the edit history of this answer.

回答3:

Improving original answer in order to take schema into account, ie. schema.my_function_name,

select
    format('DROP FUNCTION %s(%s);',
      p.oid::regproc, pg_get_function_identity_arguments(p.oid))
FROM pg_catalog.pg_proc p
LEFT JOIN pg_catalog.pg_namespace n ON n.oid = p.pronamespace
WHERE
    p.oid::regproc::text = 'schema.my_function_name';

回答4:

Slightly enhanced version of Erwin's answer. Additionally supports following

'like' instead of exact function name match
can run in 'dry-mode' and 'trace' the SQL for removing of the functions

Code for copy/paste:

/**
 * Removes all functions matching given function name mask
 *
 * @param p_name_mask   Mask in SQL 'like' syntax
 * @param p_opts        Combination of comma|space separated options:
 *                        trace - output SQL to be executed as 'NOTICE'
 *                        dryrun - do not execute generated SQL
 * @returns             Generated SQL 'drop functions' string
 */
CREATE OR REPLACE FUNCTION mypg_drop_functions(IN p_name_mask text,
                                               IN p_opts text = '')
    RETURNS text LANGUAGE plpgsql AS $$
DECLARE
    v_trace boolean;
    v_dryrun boolean;
    v_opts text[];
    v_sql text;
BEGIN
    if p_opts is null then
        v_trace = false;
        v_dryrun = false;
    else
        v_opts = regexp_split_to_array(p_opts, E'(\\s*,\\s*)|(\\s+)');
        v_trace = ('trace' = any(v_opts)); 
        v_dryrun = ('dry' = any(v_opts)) or ('dryrun' = any(v_opts)); 
    end if;

    select string_agg(format('DROP FUNCTION %s(%s);', 
        oid::regproc, pg_get_function_identity_arguments(oid)), E'\n')
    from pg_proc
    where proname like p_name_mask
    into v_sql;

    if v_sql is not null then
        if v_trace then
            raise notice E'\n%', v_sql;
        end if;

        if not v_dryrun then
            execute v_sql;
        end if;
    end if;

    return v_sql;
END $$;

select mypg_drop_functions('fn_dosomething_%', 'trace dryrun');

回答5:

Here is the query I built on top of @Сухой27 solution that generates sql statements for dropping all the stored functions in a schema:

WITH f AS (SELECT specific_schema || '.' || ROUTINE_NAME AS func_name 
        FROM information_schema.routines
        WHERE routine_type='FUNCTION' AND specific_schema='a3i')
SELECT
    format('DROP FUNCTION %s(%s);',
      p.oid::regproc, pg_get_function_identity_arguments(p.oid))
FROM pg_catalog.pg_proc p
LEFT JOIN pg_catalog.pg_namespace n ON n.oid = p.pronamespace
WHERE
    p.oid::regproc::text IN (SELECT func_name FROM f);

回答6:

As of Postgres 10 you can drop functions by name only, as long as the names are unique to their schema. Just place the following declaration at the top of your function file: