Re: pg_upgrade (12->14) fails on aggregate

On Sat, Jun 25, 2022 at 03:34:49PM +0500, Andrey Borodin wrote:
> > On 25 Jun 2022, at 01:28, Justin Pryzby <pryzby@telsasoft.com> wrote:
> 
> >  this is my latest.
> > <0001-WIP-pg_upgrade-check-detect-old-polymorphics-from-pr.patch>
> 
> Let's rename "databases_with_old_polymorphics.txt" to somthing like "old_polymorphics.txt" or maybe even "incompatible_polymorphics_usage.txt"?
> I think you will come up with a better name, my point is here everythin is in "databases", and "old" doesn't describe essence of the problem.

> Also, let's check that oid of used functions belong to system catalog (<16384)? We don't care about user-defined functions with the same name.

Right now, we test
  =ANY(ARRAY['array_remove(anyarray,anyelement)',...]::regprocedure)

..which will find the system's array_remove, and not some other one, due to
ALWAYS_SECURE_SEARCH_PATH_SQL (which is also why ::regprocedure prints a
namespace for the non-system functions we're interested in displaying).

I had "transnsp.nspname='pg_catalog'", which was redundant, so I removed it.

I tested that this allows upgrades with aggregates on top of non-system
functions of the same name/args:

postgres=# CREATE FUNCTION array_append(anyarray, anyelement) RETURNS ANYARRAY LANGUAGE SQL AS $$ $$;
postgres=# CREATE AGGREGATE foo(anyelement) (sfunc=public.array_append, stype=anyarray, initcond='{}');

> And, probably, we can do this unconditionally:
> if (old_cluster.major_version >= 9500)
>         appendPQExpBufferStr(&old_polymorphics,
> Nothing bad will happen if we blacklist usage of nonexistent functions.

Nope, it's as I said: this would break pg_upgrade from older versions.

> I realized that my latest patch would break upgrades from old servers, which do
> not have array_position/s nor width_bucket, so ::reprocedure would fail.  Maybe
> Andrey's way is better (checking proname rather than its OID).

This fixes several error with the version test.

-- 
Justin