Re: magical eref alias names

On Thu, Nov 7, 2024 at 2:05 AM Peter Eisentraut <peter(at)eisentraut(dot)org> wrote:
> On 06.11.24 20:06, Robert Haas wrote:
> > I can make nothing of*TLOCRN* or*TROCRN*, even
> > after looking at the relevant source code,
>
> These are from the SQL standard text. So they are more guidance to the
> implementer than anything else. I think something had to be put there,
> because erefs are required. I'm also interested in the discussion you
> raise about that.
>
> (I think they are meant to be something like "table {left|right} operand
> cycle reference name".)

Interesting. I guess we could maybe document that a bit better, but
it's not the main thing to focus on.

I tried replacing all of those fake RTE entry names with NULL and
fixing whatever broke in the regression tests. I attach the result. It
seems like this is basically a problem that applies to subqueries
rather than any other form of RTE, and mostly subqueries that are
named *SELECT* or *SELECT* %d. The main place where the names are
user-visible in EXPLAIN output. With this patch, the name displayed by
EXPLAIN changed to unnamed_subquery, but that's what we already use
for subqueries in some cases, so I don't see a problem with it.
There's one case where *SELECT* 2 actually showed up in an error
message. I chose to add a second variant of the message rather than
displaying the name as unnamed_subquery, but either could be done:

-DETAIL: There is a column named "q2" in table "*SELECT* 2", but it
cannot be referenced from this part of the query.
+DETAIL: There is a column named "q2", but it cannot be referenced
from this part of the query.

It's quite possible that this patch isn't completely correct and it's
also possible that I'm missing some deeper problem with this idea that
just isn't exercised by the regression tests. But overall I find these
results fairly encouraging -- it just wasn't very hard to make the
regression tests pass.

I did a bit of historical checking at the comment that eref->aliasname
is required to be present was added in commit
3a94e789f5c9537d804210be3cb26f7fb08e3b9e, Tom Lane, 2000. I can't
immediately tell what the reasoning was. Copying Tom in case he has
thoughts.

--
Robert Haas
EDB: http://www.enterprisedb.com