BUG #14754: ecpg SQL parsing error

On 21 July 2017 at 15:16, <zuberre@gmail.com> wrote:

The following bug has been logged on the website:

Bug reference: 14754
Logged by: Richard Zuber
Email address: zuberre@gmail.com
PostgreSQL version: 9.5.7
Operating system: CentOS 7.3.1611
Description:

Hello,

I’ve been using ecpg as part of my automated test framework to ensure my
various SQL migrations have proper syntax. I believe I have run into an
error where ECPG is reporting a syntax error that does not in fact exist.

Assume the following table is defined (though this should not be relevant to
the issue):

CREATE TABLE foo
(
id SERIAL NOT NULL,
name text NOT NULL,
PRIMARY KEY (id)
);

There are two files to be fed into ecpg (version - ecpg (PostgreSQL 9.5.7)
4.11.0) :

--doesnt_work.sql
EXEC SQL WITH cte AS
(
INSERT INTO foo(name)
VALUES
('bar')
RETURNING id
)
SELECT * FROM foo;
--EOF

The output of “doesn’t_work.sql” is:

$ ecpg doesnt_work.sql
doesnt_work.sql:7: ERROR: syntax error at or near ")"
$ echo $?
3

Since this is a valid statement from a syntax perspective (ignoring what the
statement does), I would expect no error.

The following appears to work as expected:

--works.sql
EXEC SQL WITH cte AS
(
INSERT INTO foo(name)
VALUES
('bar')
)
SELECT * FROM foo;
--EOF

$ ecpg works.sql
$ echo $?
0

The difference between the two is the use of the “RETURNING” statement in
the CTE. I have reviewed section 33.5 of the manual. While I took note of
33.5.3 in the manual, this statement does not actually return a result set.
That said, I may be misinterpreting the documentation in this matter.

I have reviewed release notes from later versions of postgres as well as the
open TODOs, but have not seen anything that appears to relate to this
behavior.

I appreciate your attention,
Richard Zuber

--
Sent via pgsql-bugs mailing list (pgsql-bugs@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-bugs

--
greg

--
Sent via pgsql-bugs mailing list (pgsql-bugs@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-bugs

Thanks for the report Richard, sorry it took me a while to find time to
debug.

I’ve been using ecpg as part of my automated test framework to ensure
my
various SQL migrations have proper syntax.  I believe I have run into
an
error where ECPG is reporting a syntax error that does not in fact
exist.

I guess I agree, after all ecpg should accept everything that the
server accepts, right? :)

The problem is that ecpg expects every non-empty returning clause to
end with an "into <variable>" sectio to store the data in C. This
obviously does not make sense. I have to do more tests to see if making
the into clause optional breaks other things. If you have a large base
of test cases you're welcome to try it out too, once the patch is done.

Michael
--
Michael Meskes
Michael at Fam-Meskes dot De, Michael at Meskes dot (De|Com|Net|Org)
Meskes at (Debian|Postgresql) dot Org
Jabber: michael at xmpp dot meskes dot org
VfL Borussia! Força Barça! SF 49ers! Use Debian GNU/Linux, PostgreSQL

--
Sent via pgsql-bugs mailing list (pgsql-bugs@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-bugs

I’ve been using ecpg as part of my automated test framework to ensure my
various SQL migrations have proper syntax. I believe I have run into an
error where ECPG is reporting a syntax error that does not in fact exist.

Fixed in master, if I don't hear of any problems I'll backport in a
couple of days.

Michael
--
Michael Meskes
Michael at Fam-Meskes dot De, Michael at Meskes dot (De|Com|Net|Org)
Meskes at (Debian|Postgresql) dot Org
Jabber: michael at xmpp dot meskes dot org
VfL Borussia! Força Barça! Go SF 49ers! Use Debian GNU/Linux, PostgreSQL

--
Sent via pgsql-bugs mailing list (pgsql-bugs@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-bugs