From: thomas at tada.se (Thomas Hallgren)
Date: Thu, 18 May 2006 09:35:33 +0200
Subject: [Pljava-dev] Problem with primary key
In-Reply-To: <446C1EE3.2030605@insa-toulouse.fr>
References: <4469E7C0.4040502@insa-toulouse.fr> <4469F3E6.6070802@tada.se>
	<446C1EE3.2030605@insa-toulouse.fr>
Message-ID: <446C23C5.207@tada.se>

Eric BOYER wrote:
> Thomas Hallgren a ?crit :
>
>> Eric,
>> You say that you re-throw the exception in your catch. Yet, in your 
>> example you don't do that. Also, to do an e.printStackTrace() might 
>> be a bad idea sometimes. If the postmaster runs as a service, there 
>> will be no System.out assigned.
>>  
>>
> It was a simple test. In reality, we use log4j !
Really? Why is that? You are aware that PL/Java is adapting the standard 
java.util.logging stuff on top of the PostgreSQL logger? More info here: 
http://wiki.tada.se/display/pljava/Logging


> OK but we wan't catch exception for different reasons ...

You must use savepoints if you plan to catch SQLExceptions without 
throwing them again (see: 
http://wiki.tada.se/display/pljava/Exception+handling). If you don't it 
really shouldn't matter if you catch a ServerException (the subclass of 
SQLException that you must have encountered). That exception will always 
be re-thrown and cause a rollback once your method returns. And that's 
regardless of if you catch it or not.

The fact that the backend reached an unstable state actually indicates a 
bug in PL/Java. My guess is that the endless recursion that you caused 
somehow messed up the internal error handling. I'll look into that.

Regards,
Thomas Hallgren