Re: rollback to savepoint issue

On 2023-Sep-04, Erik Wienhold wrote:

> On 04/09/2023 16:56 CEST David G. Johnston <david(dot)g(dot)johnston(at)gmail(dot)com> wrote:
>
> > On Monday, September 4, 2023, Erik Wienhold <ewie(at)ewie(dot)name> wrote:
> >
> > > On 04/09/2023 11:51 CEST Lorusso Domenico <domenico(dot)l76(at)gmail(dot)com> wrote:
> > >
> > > > The original code in embedded in a function, but the problem is the same:
> > >
> > > Transaction control is not possible in functions. Only in procedures (CALL)
> > > and DO blocks.
> >
> > Then explain why the original savepoint command wasn’t a syntax, or runtime,
> > error?
>
> I don't need to because CREATE {FUNCTION | PROCEDURE} already fails because of
> ROLLBACK TO SAVEPOINT. And without a function to execute there can't be any
> runtime error because of SAVEPOINT. My point was about transaction control in
> plpgsql in general.

You can create the function nonetheless if you set check_function_bodies
to OFF beforehand. Then it's a "syntax" error detected at runtime.

55432 17devel 1274266=# create function quarrel() returns void language plpgsql as $$ begin savepoint foo; insert into foo default values; rollback to savepoint foo; end $$;
ERROR: syntax error at or near "to"
LINE 1: ...int foo; insert into foo default values; rollback to savepoi...
^

55432 17devel 1274266=# set check_function_bodies to 0;
SET

55432 17devel 1274266=# create function quarrel() returns void language plpgsql as $$ begin savepoint foo; insert into foo default values; rollback to savepoint foo; end $$;
CREATE FUNCTION

55432 17devel 1274266=# select quarrel();
ERROR: syntax error at or near "to"
LINE 1: ...int foo; insert into foo default values; rollback to savepoi...
^
QUERY: begin savepoint foo; insert into foo default values; rollback to savepoint foo; end
CONTEXT: compilation of PL/pgSQL function "quarrel" near line 1

But it gets worse. If you create a procedure (no longer a function)
with a SAVEPOINT and ROLLBACK TO SAVEPOINT, it'll complain about the
ROLLBACK TO command, but not about SAVEPOINT; and if you remove that,
then it'll fail at runtime saying that SAVEPOINT is unsupported.

So it's not super consistent about how it report these various problems.

--
Álvaro Herrera 48°01'N 7°57'E — https://www.EnterpriseDB.com/
"El sudor es la mejor cura para un pensamiento enfermo" (Bardia)