Обсуждение: Bug(s) or not?

Hi!

Simple test suite for current CVS:
%initdb ...
%createdb q
%psql q
q=# select * from pg_class limit 1;
SELECT
q=# select * from pg_class limit 2;
server sent data ("D" message) without prior row description ("T" message)
SELECT
q=# select * from pg_class limit 3;
server sent data ("D" message) without prior row description ("T" message)
server sent data ("D" message) without prior row description ("T" message)
SELECT

Why can I select from pg_class? Other system tables are working properly...

And one more:
q=# create table q(i int);
CREATE TABLE
q=# insert into q values (1);
INSERT 17450 1
q=# insert into q values (2);
INSERT 17451 1

q=# select count(*) from q; count
-------     2
(1 row)

q=# select count(*) from q limit 1; count
-------     2
(1 row)

q=# select count(*) from q limit 1 offset 1; count
-------
(0 rows)

q=# select count(*) from q  offset 1; count
-------
(0 rows)

q=# select count(*) from q  offset 0; count
-------     2
(1 row)

I understand that it isn't correct query, but why pgsql do something strange 
instead of say 'error'?


-- 
Teodor Sigaev                                  E-mail: teodor@sigaev.ru

On Mon, May 26, 2003 at 14:17:52 +0400, Teodor Sigaev <teodor@sigaev.ru> wrote:
> 
> q=# select count(*) from q;
>  count
> -------
>      2
> (1 row)
> 
> q=# select count(*) from q limit 1;
>  count
> -------
>      2
> (1 row)
> 
> q=# select count(*) from q limit 1 offset 1;
>  count
> -------
> (0 rows)
> 
> q=# select count(*) from q  offset 1;
>  count
> -------
> (0 rows)
> 
> q=# select count(*) from q  offset 0;
>  count
> -------
>      2
> (1 row)
> 
> I understand that it isn't correct query, but why pgsql do something 
> strange instead of say 'error'?

The above queries appear to be correct. The count returns a single row
with the number of rows in the table. If you skip over that row using
offset, you aren't going to get any rows.

On Mon, May 26, 2003 at 02:17:52PM +0400, Teodor Sigaev wrote:

> q=# select * from pg_class limit 3;
> server sent data ("D" message) without prior row description ("T" message)
> server sent data ("D" message) without prior row description ("T" message)
> SELECT
> 
> Why can I select from pg_class? Other system tables are working properly...

Doh... I've been seen the same behavior, but thought it was some bug in
my local copy.  In fact I'm having a hard time finding a bug with
unreleased locks with code that seems functionally identical to CVS head
and I'm starting to think that it's not really a bug with my
modifications.

-- 
Alvaro Herrera (<alvherre[@]dcc.uchile.cl>)
Major Fambrough: You wish to see the frontier?
John Dunbar: Yes sir, before it's gone.

Teodor Sigaev <teodor@sigaev.ru> writes:
> Why can I select from pg_class? Other system tables are working properly...

That's darn bizarre.  I see it too --- will look into it.

> And one more:

That all looks like correct behavior to me.  LIMIT/OFFSET are with
respect to the final query result, ie the single aggregate row.
        regards, tom lane

> Teodor Sigaev <teodor@sigaev.ru> writes:
>> Why can I select from pg_class? Other system tables are working properly...

> That's darn bizarre.  I see it too --- will look into it.

Found it --- fix committed.  Looks like libpq still needs some more work
for error recovery, too --- it was losing the original error message
about "unexpected field count in D message".
        regards, tom lane