Re: patch for EXECUTE .. INTO (from TODO)

Hello

I did small trivial patch (almost all was written) for storing
result from executing dynamic query into ROW or RECORD variable.

CREATE TABLE fxx(i integer, y integer);
CREATE TYPE fxt AS (i integer, z integer);

CREATE OR REPLACE FUNCTION foo(varchar) RETURNS RECORD AS $$
DECLARE _r RECORD; _f fxx%ROWTYPE; _t fxt; z fxx;
BEGIN
DELETE FROM fxx;
EXECUTE 'INSERT INTO fxx VALUES(10,15)';
EXECUTE 'SELECT (row).* from (select row(10,1)::fxx)s' INTO _r;
RAISE NOTICE '%', _r.i;
EXECUTE 'SELECT * FROM '||$1||' LIMIT 1' INTO _f;
RETURN _f;
END; $$ LANGUAGE plpgsql;
SELECT foo('fxx');

pokus=# NOTICE: 10
foo
---------
(10,15)
(1 row)

Best regards
Pavel Stehule

On Wed, 2005-06-01 at 11:30 +0200, Pavel Stehule wrote:
> I did small trivial patch (almost all was written) for storing
> result from executing dynamic query into ROW or RECORD variable.

Cool, this will be useful. A few minor comments:

The patch needs some regression tests.

I'm not sure the parser modifications are quite right -- these
statements are treated as identical:

EXECUTE 'SELECT (row).* from (select row(10,1)::fxx) s'; _r;
EXECUTE 'SELECT (row).* from (select row(10,1)::fxx) s' INTO _r;

Accepting the former as valid syntax might mean misinterpreting some
function definitions (for example, EXECUTE '...'; var := 5). If you use
read_sql_construct() directly, you can use the *endtoken out parameter
to check whether the parser saw an INTO or a semicolon, and only look
for a following variable in the former case. In any case there's not
much point in defining plpgsql_read_expression2(), the rest of gram.y
just uses read_sql_construct() directly.

-Neil