Tom Lane <tgl@sss.pgh.pa.us> writes:
> I've got to say that I think this is fundamentally the wrong approach:
> rather than fixing the generic problem of ALTER EXTENSION not coping
> with multiple dependency paths to the same object, it hacks the specific
> case of owned sequences, and what's more it does that by assuming that
> every owned sequence *will* have a dependency on the extension. That's
> not a safe assumption.
In general, agreed.
> Still, this might be the best approach for the back branches, given that
> we do not know of any existing multiple-dependency scenarios other than
> the owned-sequence case. A real fix is looking mighty invasive.
That's what I was aiming at, best approach for the back branches.
>> Even for TIP I don't want us to change how pg_depend tracking is done,
>
> Agreed. Quite aside from backwards-compatibility concerns, I think that
> trying to avoid multiple dependency paths is doomed to failure.
For a =E2=80=9CDIRTT=E2=80=9D approach to the problems, I think =C3=81lvaro=
's work is in the
right direction, and should be pursued without trying to address the
back branches too. I don't remember now if his tracking attempt was also
trying to change pg_depend entries, I think that was in two separate
patches versions.
DIRTT: Do It Right This Time
=C3=81lvaro, do you want to be working on a master only version of the fix
or do you want me to?
Regards,
--=20
Dimitri Fontaine 06 63 07 10 78
http://2ndQuadrant.fr PostgreSQL : Expertise, Formation et Support