I just did this using PG 14.2:
create procedure p(a out int)
language plpgsql
as $body$
begin
a := 42;
end;
$body$;
do $body$
declare
a constant int := 0;
begin
call p(a);
raise info '%', a::text;
end;
$body$;
The DO block runs without error and reports "INFO: 42". This is an unambiguous
semantic error because "a" is declared "constant".
Is this a known issue?
Replace "a" with the literal "37" in this test:
do $body$
begin
call p(37);
raise info '%', a::text;
end;
$body$;
This causes the expected runtime error 42601:
procedure parameter "a" is an output parameter but corresponding argument is
not writable.
Bt.w., error 42601 is mapped to the name "syntax_error" in PL/pgSQL. I'd say
that this is its own distinct bug. The syntax is fine. It's a semantic error.
Notice that the test can be trivially transcribed to Oracle Database's PL/SQL
as this SQL*Plus script:
create procedure p(a out integer)
authid definer
as
begin
a := 42;
end;
/
declare
a /*constant*/ int := 0;
begin
p(a);
DBMS_Output.put_line('a: '||to_char(a));
end;
/
When "constant" is commented out (as presented), the anonymous block runs
without error and outputs "a: 42". But when "constant" is uncommented, the
attempt causes this error:
PLS-00363: expression 'A' cannot be used as an assignment target
This is the proper report of what clearly is a semantic error. PG should do the
same.
B.t.w., this happens to be a compilation error in ORCL and not a run-time
error. But that's an entirely different story and reflects the fundamentally
different compilation and execution models for anonymous blocks, user-defined
functions, and user-defined procedures between ORCL and PG.
