Re: Proposal to suppress errors thrown by to_reg*()

Hello.

At Thu, 14 Mar 2019 13:37:00 +0900, Takuma Hoshiai <hoshiai(at)sraoss(dot)co(dot)jp> wrote in <20190314133700(dot)c271429ddc00ddab3aac2619(at)sraoss(dot)co(dot)jp>
> Hi, hackers,
>
> According to the document, "to_reg* functions return null rather than
> throwing an error if the name is not found", but this is not the case
> if the arguments to those functions are schema qualified and the
> caller does not have access permission of the schema even if the table
> (or other object) does exist -- we get an error.

You explicitly specified the namespace and I think that it is not
the case of not-found. It is right that the error happens since
you explicitly tried to access a unprivileged schema.

> For example, to_regclass() throws an error if its argument is
> 'schema_name.table_name'' (i.e. contains schema name) and caller's
> role doesn't have access permission of the schema. Same thing can be
> said to Other functions,
>
> I get complain from Pgpool-II users because it uses to_regclass()
> internally to confirm table's existence but in the case above it's
> not useful because the error aborts user's transaction.

I'm not sure how such unaccessible table names are given to the
function there, but it is also natural that any user trying to
access unprivileged objects gets an error.

> To be more consistent with the doc and to make those functions more
> useful, I propose to change current implementation so that they do not
> throw an error if the name space cannot be accessible by the caller.

Since it doesn't seem a bug, I think that changing the existing
behavior is not acceptable. Maybe we can live with another
signature of the function like to_regproc(name text, noerror
bool).

regards.

--
Kyotaro Horiguchi
NTT Open Source Software Center