Content-Disposition: inline
Content-Transfer-Encoding: binary
Content-Type: text/plain
MIME-Version: 1.0
From: "Phil Endecott" <spam_from_postgresql_sql@chezphil.org>
To: Tom Lane <tgl@sss.pgh.pa.us>,
	"Riccardo G. Facchini" <abief_ag_-postgresql@yahoo.com>,
	Phil Endecott <spam_from_postgresql_sql@chezphil.org>,
	pgsql-sql@postgresql.org
Subject: Re: plpgsql - Insert from a record variable? 
Date: Mon, 21 Jun 2004 13:23:33 -0400
Message-Id: <4357626@chezphil.org>
In-Reply-To: <17840.1087670348@sss.pgh.pa.us>
References: <20040619060330.83941.qmail@web13901.mail.yahoo.com>
	<17840.1087670348@sss.pgh.pa.us>

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.