How to insert rows which optionally have parent rows in the same table at once

Question

Given is the following table:

CREATE TABLE public.parenttest (
  id      bigserial NOT NULL PRIMARY KEY,
  mydata  varchar(30),
  parent  bigint
  ) WITH (
    OIDS = FALSE
); 

I'd like to insert a bunch of rows. Some of these rows should take the sequence-generated id of a row inserted before as value for the column parent.

For example:

INSERT INTO parenttest (mydata,parent) VALUES ('rootnode',null); 
INSERT INTO parenttest (mydata,parent) VALUES ('child1', /*id of rootnode*/); 
INSERT INTO parenttest (mydata,parent) VALUES ('child2', /*id of rootnode*/); 
INSERT INTO parenttest (mydata,parent) VALUES ('child2.1', /*id of child 2*/);
INSERT INTO parenttest (mydata.parent) VALUES ('child2.2', /*id of child 2*/);

...should result in the following datasets (id, mydata,parent)

1,'rootnode',null
2,'child1',1
3,'child2',1
4,'child2.1',3
5,'child2.2',3

UNTIL child2.2, everything is fine when i'm using

SELECT currval('parenttest_id_seq');

to get the parent's id but then i'm getting the id of child2.2, of course.

It's important to me that i can do all necessary stuff with as few client-side requests as possible - and i'd like to do all id generation on the server-side.

Answer 1

Here's how I'd do it, if it's acceptable for the rows to briefly have a NULL parent:

INSERT INTO parenttest (mydata) VALUES
('rootnode'), 
('child1'),
('child2'), 
('child2.1'),
('child2.2');

UPDATE parenttest SET parent = (select id from parenttest pt where
(pt.mydata = 'rootnode' and parenttest.mydata in ('child1','child2')) or
(pt.mydata = 'child2' and parenttest.mydata in ('child2.1','child2.2')))
WHERE
  mydata in ('child1',
             'child2', 
             'child2.1',
             'child2.2');

fiddle

Answer 2

You could do something as ugly as:

INSERT INTO parenttest (mydata,parent) VALUES ('rootnode',null); 
INSERT INTO parenttest (mydata,parent) SELECT 'child1', id FROM parenttest WHERE mydata='rootnode';
INSERT INTO parenttest (mydata,parent) SELECT 'child2', id FROM parenttest WHERE mydata='rootnode';
INSERT INTO parenttest (mydata,parent) SELECT 'child2.1', id FROM parenttest WHERE mydata='child2';
INSERT INTO parenttest (mydata.parent) SELECT 'child2.2', id FROM parenttest WHERE mydata='child2';

I think the right thing to do here is go with client-side, retrieving the id of 'rootnode' after inserting it, and then sending the following two statements with its id explicitly in the INSERT statement.

Answer 3

After some experiments, i've found one solution using an anonymous PL/PgSQL code block:

DO $$DECLARE
  parentid bigint;
BEGIN  
  INSERT INTO parenttest (mydata) VALUES ('rootnode');  
  parentid:=currval('parenttest_id_seq');  
  INSERT INTO parenttest (mydata,parent) VALUES ('child 1',parentid);
  INSERT INTO parenttest (mydata,parent) VALUES ('child 2',parentid);
  parentid:=currval('parenttest_id_seq');
  INSERT INTO parenttest (mydata,parent) VALUES ('child 2.1',parentid),('child 2.2',parentid);
END$$;

It's possible to use it within a single request from the client and i don't have to generate anything at the client-side, so this may be one possible solution for my problem.

Edit: It can be expressed shorter like this:

DO $$DECLARE
  parentid bigint;
BEGIN
  INSERT INTO parenttest (mydata) VALUES ('Flat') RETURNING id INTO parentid;
  INSERT INTO parenttest (mydata, parent) VALUES ('Subflat 1', parentid);
  INSERT INTO parenttest (mydata, parent) VALUES ('Subflat 2', parentid)RETURNING id INTO parentid;
  INSERT INTO parenttest (mydata, parent) VALUES ('Subflat 2.1', parentid), ('Subflat 2.2', parentid);
END$$;