Re: plpgsql - Insert from a record variable?

Dear Postgresql experts,

I'm writing a simple plpgsql function that selects a row from a table, modifies it slightly, and inserts the modified version. This sounds simple, but I can't find how to do the insert in a simple generic fashion: although SELECT has a form that puts the result in a record variable, INSERT doesn't seem to have anything similar.

What I'd like to be able to write is something like this:

DECLARE
R RECORD;
BEGIN
SELECT * INTO R FROM TABLE WHERE id=n;
R.something := x;
INSERT INTO TABLE R;
END

But as far as I can see the only syntax that is allowed for INSERT is the normal INSERT INTO TABLE (x,y,z) VALUES (a,b,c). To achieve what I want to do I'd need to iterate over the fields of the record (how?) and build up the string of the INSERT query.

It seems odd that SELECT can use a record variable but INSERT can't, so I wonder if I have missed something. Any suggestions?

Thanks in advance for any help anyone can offer.

Regards,

--Phil.

--- Phil Endecott <__> wrote:
> Dear Postgresql experts,
>
> I'm writing a simple plpgsql function that selects a row from a
> table, modifies it slightly, and inserts the modified version. This
> sounds simple, but I can't find how to do the insert in a simple
> generic fashion: although SELECT has a form that puts the result in a
> record variable, INSERT doesn't seem to have anything similar.
>
> What I'd like to be able to write is something like this:
>
> DECLARE
> R RECORD;
> BEGIN
> SELECT * INTO R FROM TABLE WHERE id=n;
> R.something := x;
> INSERT INTO TABLE R;
> END

I'm not sure if it will work, but have you tried either two of these
forms?
--
declare r record;

begin
select * into r from table where id = n;
r.something := x;
insert into table select r;
end;
--
declare r record;

begin
select * into r from table where id = n;
r.something := x;
insert into table select r.*;
end;

>
> But as far as I can see the only syntax that is allowed for INSERT is
> the normal INSERT INTO TABLE (x,y,z) VALUES (a,b,c). To achieve what
> I want to do I'd need to iterate over the fields of the record (how?)
> and build up the string of the INSERT query.
>
> It seems odd that SELECT can use a record variable but INSERT can't,
> so I wonder if I have missed something. Any suggestions?
>
> Thanks in advance for any help anyone can offer.
>
> Regards,
>
> --Phil.
>
> ---------------------------(end of
> broadcast)---------------------------
> TIP 7: don't forget to increase your free space map settings

regards,

=====
Riccardo G. Facchini

"Riccardo G. Facchini" <abief_ag_-postgresql(at)yahoo(dot)com> writes:
> --- Phil Endecott <__> wrote:
>> I'm writing a simple plpgsql function that selects a row from a
>> table, modifies it slightly, and inserts the modified version. This
>> sounds simple, but I can't find how to do the insert in a simple
>> generic fashion: although SELECT has a form that puts the result in a
>> record variable, INSERT doesn't seem to have anything similar.

> I'm not sure if it will work, but have you tried [snip]

> declare r record;
> insert into table select r.*;

I don't know of any way to handle this in existing releases ---
plpgsql's support for rowtype variables is just too limited.
However, in 7.5 you will be able to do either of

insert into table select r.*;
insert into table values(r.*);

so long as r is declared to be of a named rowtype (not just RECORD;
that doesn't have a known field list to plan the query with). For
instance I was just experimenting with this test case:

create or replace function fooey(int) returns void as '
declare r foo;
begin
select * into r from foo where id = $1;
r.f1 = ''baz'';
insert into foo select r.*;
return;
end' language plpgsql;

regards, tom lane

Phil> Insert from a record variable?

Riccardo> Try insert into table select r.*;

Tom> in 7.5
Tom> insert into table select r.*;
Tom> insert into table values(r.*);
Tom> so long as r is declared to be of a named rowtype (not just
Tom> RECORD)

Thanks! Unfortunately I need record, rather than %rowtype. See my later email where I describe how I am trying to use this with inheritance; the function looks up a row in a base table, finds the derived table in which it really exists using pg_class.relname, and then inserts a modified copy of the row in the derived table.

I'm not concerned about the performance issues to do with pre-planning the queries. I think what I really need is an introspection mechanism so that I can loop over each element of the record and construct the insert as a string. Maybe this is possible using a different server-side language? I've just had an idea: perhaps rather than inspecting the record variable to see what fields it contains, I can look at the table to see what columns it contains (since this amounts to the same thing). Presumably I can do this using information_schema.columns. I'll have a go.

--Phil.

I wrote:
> perhaps rather than inspecting the record variable to see what fields
> it contains, I can look at the table to see what columns it contains

This is my effort. It doesn't work, because I can only EXECUTE
SQL statements, not PLPGSQL statements. I need to EXECUTE an
assignment statement to accumulate the string of column values.

I have a feeling that I can EXECUTE a CREATE FUNCTION statement,
and then call the function, but this seems over-the-top. I just
want to insert a record into a table! Any better ideas?

-- Simply insert record r into table t.

-- Doesn't work, because EXECUTE takes an SQL command, not
-- a plpgsql statement.

create function insert_record ( record, text ) as '
-- probably ought to pass schema as well as table name, since
-- information_schema.columns query doesn't use search_path.
declare
r as alias for $1;
t as alias for $2;

cr information_schema.columns%rowtype;
first boolean;
column_names text;
column_values text;
begin

first := true;
for cr in select * from information_schema.columns
where table_name=t loop
if not first then
column_names := column_names || '', '';
column_values := column_values || '', '';
first := false;
end if;
column_names := column_names || quote_ident(cr.column_name);
!! execute ''column_values :=
!! column_values || quote_literal(r.'' || cr.column_name || '')'';
end loop;

execute ''insert into '' || t || ''('' || column_names ||
'') values ('' || column_values || '')'';

end;
' language plpgsql;

--Phil.

Phil Endecott wrote:
> !! execute ''column_values :=
> !! column_values || quote_literal(r.'' || cr.column_name || '')'';
I'm guessing you want something like

FOR rec IN EXECUTE ''select column_values || quote_literal(r.'' || cr.column_name || '') alias column_values''; LOOP
column_values := rec.column_values;
END LOOP;

another suggestion, rather than do
> if not first then
> column_names := column_names || '', '';
> column_values := column_values || '', '';
> first := false;
> end if;
why not use two arrays and append the name/values using array_append() and use array_to_string() to join them after that?

--
basic

"Phil Endecott" <spam_from_postgresql_sql(at)chezphil(dot)org> writes:
> I think what I really need is an introspection mechanism
> so that I can loop over each element of the record and construct the
> insert as a string. Maybe this is possible using a different
> server-side language?

pltcl can probably handle this; I'm less sure about plperl or plpython.
(No reflection on the languages, but pltcl has the most complete
Postgres interface.)

regards, tom lane

Phil> execute ''column_values :=
Phil> column_values || quote_literal(r.'' || cr.column_name || '')'';

basic> FOR rec IN EXECUTE
basic> ''select column_values || quote_literal(r.'' ||
basic> cr.column_name || '') alias column_values''; LOOP
basic> column_values := rec.column_values;
basic> END LOOP;

I think your code will try to execute a query like this:

select column_values || quote_literal(r.something) alias column_values

This will fail because column_values and r are both plpgsql variables, and so are not visible to the SQL interpreter.

Any other suggestions?

--Phil.