Postgres trigger-based insert redirection without breaking RETURNING

Question

I'm using table inheritance in postgres, but the trigger I'm using to partition data into the child tables isn't quite behaving right. For example, this query returns nil, but I would like it to return the id of the new record.

INSERT INTO flags (flaggable_id, flaggable_type) 
VALUES (233, 'Thank') 
RETURNING id;

If I change the return value of the trigger function from NULL to NEW, I get the desired RETURNING behavior, but then two identical rows are inserted in the database. This makes sense, since a non-null return value from the trigger function causes the original INSERT statement execute, whereas returning NULL causes the statement to halt execution. A unique index might halt the second insertion, but would probably raise an error.

Any ideas how to make the INSERT with RETURNING work properly with a trigger like this?

CREATE TABLE flags (
  id integer NOT NULL,
  flaggable_type character varying(255) NOT NULL,
  flaggable_id integer NOT NULL,
  body text
);

ALTER TABLE ONLY flags
    ADD CONSTRAINT flags_pkey PRIMARY KEY (id);

CREATE TABLE "comment_flags" (
 CHECK ("flaggable_type" = 'Comment'),
 PRIMARY KEY ("id"),
 FOREIGN KEY ("flaggable_id") REFERENCES "comments"("id")
) INHERITS ("flags");

CREATE TABLE "profile_flags" (
 CHECK ("flaggable_type" = 'Profile'),
 PRIMARY KEY ("id"),
 FOREIGN KEY ("flaggable_id") REFERENCES "profiles"("id")
) INHERITS ("flags");

CREATE OR REPLACE FUNCTION flag_insert_trigger_fun() RETURNS TRIGGER AS $BODY$
BEGIN
  IF (NEW."flaggable_type" = 'Comment') THEN
    INSERT INTO comment_flags VALUES (NEW.*);
  ELSIF (NEW."flaggable_type" = 'Profile') THEN
    INSERT INTO profile_flags VALUES (NEW.*);
  ELSE
    RAISE EXCEPTION 'Wrong "flaggable_type"="%", fix flag_insert_trigger_fun() function', NEW."flaggable_type";
  END IF;
  RETURN NULL;
END; $BODY$ LANGUAGE plpgsql;

CREATE TRIGGER flag_insert_trigger
  BEFORE INSERT ON flags
  FOR EACH ROW EXECUTE PROCEDURE flag_insert_trigger_fun();

Answer 1

The only workaround I found, is to create a view for the base table & use INSTEAD OF triggers on that view:

CREATE TABLE flags_base (
    id integer NOT NULL,
    flaggable_type character varying(255) NOT NULL,
    flaggable_id integer NOT NULL,
    body text
);

ALTER TABLE ONLY flags_base
    ADD CONSTRAINT flags_base_pkey PRIMARY KEY (id);

CREATE TABLE "comment_flags" (
 CHECK ("flaggable_type" = 'Comment'),
 PRIMARY KEY ("id")
) INHERITS ("flags_base");

CREATE TABLE "profile_flags" (
 CHECK ("flaggable_type" = 'Profile'),
 PRIMARY KEY ("id")
) INHERITS ("flags_base");

CREATE OR REPLACE VIEW flags AS SELECT * FROM flags_base;

CREATE OR REPLACE FUNCTION flag_insert_trigger_fun() RETURNS TRIGGER AS $BODY$
BEGIN
  IF (NEW."flaggable_type" = 'Comment') THEN
    INSERT INTO comment_flags VALUES (NEW.*);
  ELSIF (NEW."flaggable_type" = 'Profile') THEN
    INSERT INTO profile_flags VALUES (NEW.*);
  ELSE
    RAISE EXCEPTION 'Wrong "flaggable_type"="%", fix flag_insert_trigger_fun() function', NEW."flaggable_type";
  END IF;
  RETURN NEW;
END; $BODY$ LANGUAGE plpgsql;

CREATE TRIGGER flag_insert_trigger
  INSTEAD OF INSERT ON flags
  FOR EACH ROW EXECUTE PROCEDURE flag_insert_trigger_fun();

But this way you must supply the id field on each insertion (even if flags_base's primary key has a default value / is a serial), so you must prepare your insert trigger to fix NEW.id if it is a NULL.

UPDATE: It seems views' columns can have a default values too, set with

ALTER VIEW [ IF EXISTS ] name ALTER [ COLUMN ] column_name SET DEFAULT expression

which is only used in views have an insert/update rule/trigger.

http://www.postgresql.org/docs/9.3/static/sql-alterview.html