On Mon, 2023-10-23 at 22:43 -0400, Tom Lane wrote:
> Laurenz Albe <laurenz.albe@cybertec.at> writes:
> > On Mon, 2023-10-23 at 11:37 -0700, David G. Johnston wrote:
> > > I do believe that we should be against exposing, like in this case, any internal
> > > implementation detail that encodes something (e.g., default privileges) as NULL
> > > in the catalogs, to the user of the psql meta-commands.
>
> > Sure, it would be best to hide this implementation detail from the user.
> > The correct way to do that would be to fake an ACL entry like "laurenz=arwdDxt/laurenz"
> > if there is a NULL in the catalog, but that would add a ton of special-case
> > code to psql, which does not look appealing at all.
>
> For better or worse, that *is* the backend's catalog representation,
> and I don't think that psql would be doing our users a service by
> trying to obscure the fact. They'd run into it anyway the moment
> they look at the catalogs with anything but a \d-something command.
... for example with a client like pgAdmin, which is a frequent choice
of many PostgreSQL beginners (they display empty privileges).
Yes, it is "(default)" or NULL. The former is friendlier for beginners,
the latter incurs less backward incompatibility.
I could live with either solution, but I am still leaning towards NULL.
I ran the regression tests with a patch that displays "(default)",
and I counted 22 failures, excluding the one added by my patch.
The tests can of course be fixed, but perhaps that serves as a measure
of the backward incompatibility.
Yours,
Laurenz Albe