Hello Pavel,
>> After giving it some thoughts, I see three possible solutions:
>>
>> 0. Do nothing about it.
>> I would prefer the prepare is cleaned up.
>>
>> 1. assign a special name, eg "_psql_gdesc_", so that
>> DEALLOCATE "_psql_gdesc_" can be issued afterwards.
>>
>> [...]
>
> The doc says about unnamed prepared statements - any new unnamed prepared
> statement deallocates previous by self. From psql environment is not
> possible to create unnamed prepared statement.
That is a good point. It seems that it is not possible to execute it
either.
> I prefer @0 agaisnt @1 because workflow is simple and robust. If unnamed PP
> doesn't exists, then it will be created, else it will be replaced. @1 has
> little bit more complex workflow, because there is not command like
> DEALLOCATE IF EXISTS, so I have to ensure deallocation in all possible
> ways.
ISTM That it is only of the PQprepare succeeded, so there should be only
one point, at the end of the corresponding OK condition?
> Another reason for @0 is not necessity to generate some auxiliary
> name.
Yes. I do not like special names either. But I do not like keeping objects
lives if they are dead. Not sure which is worst.
> My opinion in this case is not too strong - just I see the advantages of @0
> (robust and simple) nice. The question is about cost of unwanted allocated
> PP to end of session.
My opinion is not strong either, it is more the principle that I do not
like to let things forever live in the server while they are known dead.
Hmmm. Strange. For some obscure reason, the unnamed prepared statement
does not show in pg_prepared_statements:
calvin=# PREPARE test AS SELECT 2; calvin=# EXECUTE test; -- 2 calvin=# SELECT 1 AS one \gdesc -- one |
integer calvin=# SELECT * FROM pg_prepared_statements ; -- just one row: -- test | PREPARE test AS SELECT 2;
│7..
Conclusion: Maybe let it as solution 0 for the time being, with a comment
telling that it will be cleaned on the next unnamed PQprepare, and the
committer will have its opinion.
--
Fabien.
