Re: [HACKERS] pg_upgrade and missing loadable libraries

Bruce Momjian <bruce@momjian.us> writes:
> On Sun, Jun  4, 2017 at 01:20:12PM -0400, Alvaro Herrera wrote:
>> I think it'd be better to be exhaustive about the report, i.e. report
>> all problems in all databases, if possible.  Doing repeated pg_upgrade
>> attempts until you've nailed all the problems is boring ...

> Well, I think there are three open items:

> *  should we print all the database names involved

Yes.

> *  should we print all the pg_proc.pronames that are involved, not just
> the unique library names
> *  should we output a query helping people find the pg_proc entries

> I think there are many cases where DROP EXTENSION XXX fixes the problem,

Yes.  I think in most cases nowadays there's a one-for-one correlation
between extensions and libraries; drilling down to the level of individual
functions would just be confusing clutter.  I think if you just print
a report saying "these libraries are referenced in these databases",
that would be sufficiently usable in most cases.

You could think about printing a script full of DROP EXTENSION commands,
but aside from the sheer difficulty of doing that, it doesn't seem all
that helpful.  Simply dropping every extension is usually *not* the
right answer, and it could easily lead to data loss if done blindly.
Usually people are going to need to stop and think anyway.
        regards, tom lane