maizi <maizi@lirmm.fr> writes:
> CREATE LANGUAGE plpythonu ;
> CREATE FUNCTION wrong() RETURNS trigger AS $wrong$
> from mx import DateTime
> TD['new']['modif_time'] = DateTime.now()
> return 'MODIFY'
> $wrong$ LANGUAGE plpythonu;
> CREATE TABLE pb ( a TEXT, modif_time TIMESTAMP(0) WITHOUT TIME ZONE ) ;
> CREATE TRIGGER wrong BEFORE UPDATE ON pb FOR EACH ROW EXECUTE PROCEDURE
> wrong() ;
> INSERT INTO pb VALUES ( 'a', now() ) ;
> SELECT * FROM pb ;
> UPDATE pb SET a = 'b' ;
> SELECT * FROM pb ;
> produces this result:
> #========================
> INSERT 0 1
> a | modif_time
> ---+---------------------
> a | 2010-10-10 18:30:30
> (1 ligne)
> UPDATE 1
> a | modif_time
> ---+------------------------
> b | 2010-10-10 18:30:29.74
> (1 ligne)
> #========================
> notice the ".74" and the fact that in this example the modif_time of the
> modified tuple is earlier than the preceding one!
It looks to me like this used to work and got broken by this patch:
http://git.postgresql.org/gitweb?p=postgresql.git;a=commitdiff;h=3ab8b7fa6f9ac2fb04096f8860261dc42d59a570
which lobotomized plpython to not care about passing the right typmod to
I/O functions. This is a regression, and it's particularly annoying
because the other three standard PLs get this case right. Peter?
regards, tom lane