Обсуждение: UPDATE on subclass

Поиск
Список
Период
Сортировка

UPDATE on subclass

От
Chris Bitmead
Дата:
Been doing more tracing... 

The problem with UPDATE on inheritance hierarchies is that when it gets
down into ExecSeqScan, the value of...

node->scanstate->css_currentScanDesc->rs_rd->rd_id

is not equal to the value of...

node->plan.state->es_result_relation_info->ri_RelationDesc->rd_id

On the first scan, the former is equal to the relation for the base
class
and the latter is equal to the relation for the subclass.

Any thoughts anyone?

Re: [HACKERS] UPDATE on subclass

От
Chris Bitmead
Дата:
Hmm. In exec_append_initialize_next

nth(whichplan, rtable) 
refers to the subclass and ...
nth(whichplan, appendstate->as_result_relation_info_list);
refers to the baseclass.

Is there something that is constructing one of these 
structures in reverse order?

Chris Bitmead wrote:
> 
> Been doing more tracing...
> 
> The problem with UPDATE on inheritance hierarchies is that when it gets
> down into ExecSeqScan, the value of...
> 
> node->scanstate->css_currentScanDesc->rs_rd->rd_id
> 
> is not equal to the value of...
> 
> node->plan.state->es_result_relation_info->ri_RelationDesc->rd_id
> 
> On the first scan, the former is equal to the relation for the base
> class
> and the latter is equal to the relation for the subclass.
> 
> Any thoughts anyone?
> 
> ************

Is this it?

От
Chris Bitmead
Дата:
In ExecInitAppend it has a loop...


foreach(rtentryP, rtable) {
resultList = lcons(rri, resultList);}appendstate->as_result_relation_info_list = resultList;

If I'm not mistaken this will generate the as_result_relation_info_list
in the reverse order to the rtentry list, which is wrong... right?

Re: [HACKERS] Is this it?

От
Chris Bitmead
Дата:
Chris Bitmead wrote:
> 
> In ExecInitAppend it has a loop...
> 
>         foreach(rtentryP, rtable)
>         {
> 
>         resultList = lcons(rri, resultList);
>         }
> 
>         appendstate->as_result_relation_info_list = resultList;


This seems to be the problem. I'm going to change the above line to...
  appendstate->as_result_relation_info_list = lreverse(resultList);

After I do this, UPDATE and DELETE start working for me on subclasses.

I'll prepare a full patch for inclusion in 7.1 (Unless you want it for
7.0 :-).

Re: [HACKERS] Is this it?

От
Tom Lane
Дата:
Chris Bitmead <chrisb@nimrod.itg.telstra.com.au> writes:
> In ExecInitAppend it has a loop...
>     foreach(rtentryP, rtable) 
>     {
>     resultList = lcons(rri, resultList);
>     }
>       appendstate->as_result_relation_info_list = resultList;

> If I'm not mistaken this will generate the as_result_relation_info_list
> in the reverse order to the rtentry list,

Check ...

> which is wrong... right?

Maybe.  Is there code elsewhere that assumes these lists are ordered
alike?

You could change the lcons call to "lappend(resultList, rri)" if
you just want to try the experiment.
        regards, tom lane