Re: Latest patches break one of our unit-test, related to RLS

On Fri, 2025-09-12 at 14:22 +0200, Dominique Devienne wrote:
> On Thu, Sep 4, 2025 at 5:03 PM Dominique Devienne <ddevienne(at)gmail(dot)com> wrote:
> > OK with 16.9 and 17.5 (we cannot test on beta2 anymore)
> > KO with 16.10 and 17.6 (and beta3 too, released at the same time)
>
> I've tracked down the regression to this particular query, FWIW:
>
> select rolname, rolsuper, rolinherit, rolcreaterole,
> rolcreatedb, rolcanlogin, rolreplication, rolbypassrls,
> oid, shobj_description(oid, 'pg_authid')
> from pg_roles
> where rolname SIMILAR TO $1 AND pg_has_role(oid, 'SET')
> order by rolname
>
> In 17.5, returns 3 rows.
> In 17.6, returns 0 rows.

That must be commit e3ffc3e91d.

That commit fixed a bug in the conversion from SIMILAR TO
expressions to POSIX regular expressions.

You don't show us that data that match the pattern in 17.5, but
not in 17.6. Unless you show us a counterexample, I'd say that
the behavior in 17.6 is correct.

Minor releases shouldn't change the behavior EXCEPT when the
behavior is buggy.

Yours,
Laurenz Albe