Thread: Difference in postgres9.0.4 and postgres9.1beta1 when displaying error lines in functions with comments

Postgres 9.0.4 and postgres9.1beta1 give different line of error numbers on
functions that raise runtime errors (it seems).

How to reproduce, given the next SQL file:

$ cat /tmp/sql
begin;
CREATE FUNCTION pg_temp.commchecksql() returns boolean strict language
plpgsql as
$$
begin

-- SQL comment
perform 'aoeuaoeuaeou'::regclass;
return true;
end;
$$;

select pg_temp.commchecksql();
rollback;

begin;
CREATE FUNCTION pg_temp.commcheckC() returns boolean strict language plpgsql as
$$
begin

/* C-like comment */
perform 'aoeuaoeuaeou'::regclass;
return true;
end;
$$;

select pg_temp.commcheckC();

rollback;

------------------

If I run it on postgres9.0.4:

$ psql -U postgres template1 -f/tmp/sql
SET
BEGIN
CREATE FUNCTION
psql:/tmp/sql:12: ERROR:  relation "aoeuaoeuaeou" does not exist
LINE 1: SELECT 'aoeuaoeuaeou'::regclass              ^
QUERY:  SELECT 'aoeuaoeuaeou'::regclass
CONTEXT:  PL/pgSQL function "commchecksql" line 4 at PERFORM
ROLLBACK
BEGIN
CREATE FUNCTION
psql:/tmp/sql:26: ERROR:  relation "aoeuaoeuaeou" does not exist
LINE 1: SELECT 'aoeuaoeuaeou'::regclass              ^
QUERY:  SELECT 'aoeuaoeuaeou'::regclass
CONTEXT:  PL/pgSQL function "commcheckc" line 4 at PERFORM
ROLLBACK

--------------------

On postgres9.1beta1:

$ psql -U postgres template1 -f/tmp/sql

SET
BEGIN
CREATE FUNCTION
psql:/tmp/sql:12: ERROR:  relation "aoeuaoeuaeou" does not exist
LINE 1: SELECT 'aoeuaoeuaeou'::regclass              ^
QUERY:  SELECT 'aoeuaoeuaeou'::regclass
CONTEXT:  PL/pgSQL function "commchecksql" line 5 at PERFORM
ROLLBACK
BEGIN
CREATE FUNCTION
psql:/tmp/sql:26: ERROR:  relation "aoeuaoeuaeou" does not exist
LINE 1: SELECT 'aoeuaoeuaeou'::regclass              ^
QUERY:  SELECT 'aoeuaoeuaeou'::regclass
CONTEXT:  PL/pgSQL function "commcheckc" line 5 at PERFORM
ROLLBACK

--------------------

The line numbers of where the error is have changed:

--- /tmp/out9.0.4    Thu Jun  9 20:47:45 2011
+++ /tmp/out9.1beta1    Thu Jun  9 20:49:02 2011
@@ -5,7 +5,7 @@LINE 1: SELECT 'aoeuaoeuaeou'::regclass               ^QUERY:  SELECT 'aoeuaoeuaeou'::regclass
-CONTEXT:  PL/pgSQL function "commchecksql" line 4 at PERFORM
+CONTEXT:  PL/pgSQL function "commchecksql" line 5 at PERFORMROLLBACKBEGINCREATE FUNCTION
@@ -13,5 +13,5 @@LINE 1: SELECT 'aoeuaoeuaeou'::regclass               ^QUERY:  SELECT 'aoeuaoeuaeou'::regclass
-CONTEXT:  PL/pgSQL function "commcheckc" line 4 at PERFORM
+CONTEXT:  PL/pgSQL function "commcheckc" line 5 at PERFORMROLLBACK

---------------------

I care about this because I have some custom regression tests that check the
failing of functions coded like this... and they suddenly failed with that
difference when I checked them against postgres9.0.4.

Is this an intended behaviour?

Thank you.

Abel Abraham Camarillo Ojeda <acamari@verlet.org> writes:
> Postgres 9.0.4 and postgres9.1beta1 give different line of error numbers on
> functions that raise runtime errors (it seems).

This is not a bug.  See the release notes:
Adjust PL/pgSQL's error line numbering code to be consistentwith other PLs (Pavel Stehule)
Previously, PL/pgSQL would ignore (not count) an empty line atthe start of the function body. Since this was
inconsistentwithall other languages, the special case was removed.
 
        regards, tom lane

Thanks... and sorry for sending it to hackers list... I had not read the bug
reporting guides... nor the release notes... it seems.

Again, thanks.

On Thu, Jun 9, 2011 at 9:22 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Abel Abraham Camarillo Ojeda <acamari@verlet.org> writes:
>> Postgres 9.0.4 and postgres9.1beta1 give different line of error numbers on
>> functions that raise runtime errors (it seems).
>
> This is not a bug.  See the release notes:
>
>        Adjust PL/pgSQL's error line numbering code to be consistent
>        with other PLs (Pavel Stehule)
>
>        Previously, PL/pgSQL would ignore (not count) an empty line at
>        the start of the function body. Since this was inconsistent with
>        all other languages, the special case was removed.
>
>                        regards, tom lane
>