Re: Bug with views and defaults - Mailing list pgsql-hackers

From Tom Lane
Subject Re: Bug with views and defaults
Date
Msg-id 9399.961198990@sss.pgh.pa.us
Whole thread Raw
In response to Bug with views and defaults  ("Mark Hollomon" <mhh@nortelnetworks.com>)
Responses Re: Bug with views and defaults  (Bruce Momjian <pgman@candle.pha.pa.us>)
List pgsql-hackers
"Mark Hollomon" <mhh@nortelnetworks.com> writes:
> CREATE TABLE foo (
>     name TEXT,
>     type CHAR(1),
>     when_added TIMESTAMP DEFAULT 'now'
> );

> CREATE VIEW mytype AS 
>     SELECT name, when_added FROM foo WHERE type = 'M';

> CREATE RULE mytype_insert AS
>     ON INSERT TO mytype DO INSTEAD
>     INSERT INTO foo (name, type) VALUES (NEW.name, 'M');

> Inserting directly into foo sets when_added to the current time.
> Inserting through the view sets it to what looks like the time of
> view creation.

This is a known and not readily fixable problem.  It's far safer
to write the default for a timestamp column as now(), rather than
relying on a string literal not getting coerced to timestamp form
too soon.  See
http://www.postgresql.org/mhonarc/pgsql-hackers/1999-10/msg00036.html

BTW, Bruce: it probably would be wise to have the FAQ's item 4.22
recommend now() and nothing else.  'now' has nothing much to recommend
it and there are still pitfalls like this one.
        regards, tom lane


pgsql-hackers by date:

Previous
From: Tom Lane
Date:
Subject: Re: Big 7.1 open items
Next
From: Bruce Momjian
Date:
Subject: Re: Bug with views and defaults