Re: cached plan issue in trigger func

Tom Lane wrote:
> Andrew Dunstan <andrew@dunslane.net> writes:
>   
>> I thought we had fixed this in 8.3:
>>     
>
> I think that behavior is intentional: plancache.c can deal with the plan
> changing internally, but it doesn't expect that its callers could
> survive the plan's argument datatypes changing underneath them.
>   

How do we reconcile that with this advertised feature of 8.3?:

  * Automatically re-plan cached queries when table definitions change    or statistics are updated


How is a user to know when s/he can rely on this and when they can't? I 
at least was expecting the plan to be invalidated by the table changes.


cheers

andrew