Message-ID: <467C0011.8000805@archonet.com>
Date: Fri, 22 Jun 2007 18:00:01 +0100
From: Richard Huxton <dev@archonet.com>
User-Agent: Thunderbird 2.0.0.0 (X11/20070326)
MIME-Version: 1.0
To: Bart Degryse <Bart.Degryse@indicator.be>
Cc: pgsql-sql@postgresql.org
Subject: Re: Transactions and Exceptions
References: <467BE1BB.A3DD.0030.0@indicator.be><467BCDA1.9070606@archonet.co
	m>
	<467BF045.A3DD.0030.0@indicator.be>
In-Reply-To: <467BF045.A3DD.0030.0@indicator.be>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Bart Degryse wrote:
>> 2. Using dblink / dbi-link to reconnect to the database, which means 
>> your logging will take place in its own transaction.
> This I like more. Though I don't use either dblink nor dbi-link, I do use this kind of 
> 'double' connections already for other purposes in which transactions were never an 
> issue. So it never crossed my mind that I could use them for this too.
> For the time being that's the way I'll walk. Thanks a lot!
>  
> Reading your suggestions I assume PostgreSQL lacks something like Oracle's
> PRAGMA AUTONOMOUS_TRANSACTION

It might well be a useful feature (particularly for logging-style 
interactions as you have). I'm not sure it's particularly simple to do 
(other than by having a whole new backend as dblink will give you).

> Shouldn't it be added, are there any plans in that direction? Returning a status code
> is not always an option and using some dbi variant certainly isn't because of the 
> need for perlu. So then you're stuck?!

Well, dblink is pure C iirc. I think the advantage with dbi-link is that 
you can connect to any type of database DBI can.

-- 
   Richard Huxton
   Archonet Ltd