Re: REINDEX INDEX results in a crash for an index of pg_class since 9.6

I wrote:
> Andres Freund <andres(at)anarazel(dot)de> writes:
>> On 2019-05-02 11:41:28 -0400, Tom Lane wrote:
>>> But don't we need to worry about resetting relfrozenxid?

>> Indexes don't have that though? We couldn't do it for pg_class itself,
>> but that's not a problem here.

> Hmm. Again, that seems like the sort of assumption that could bite
> us later. But maybe we could add some assertions that the new values
> match the old? I'll experiment.

Huh, this actually seems to work. The attached is a quick hack for
testing. It gets through check-world straight up and with
xxx_FORCE_RELEASE, and I've verified reindexing pg_class works with
CLOBBER_CACHE_ALWAYS, but it'll be a few hours before I have a full CCA
run.

One interesting thing that turns up in check-world is that if wal_level
is minimal, we have to manually force an XID to be assigned, else
reindexing pg_class fails with "cannot commit a transaction that deleted
files but has no xid" :-(. Perhaps there's some other cleaner place to
do that?

If we go this path, we should remove RelationSetIndexList altogether
(in HEAD), but I've not done so here. The comments likely need more
work too.

I have to go out and do some errands for the next few hours, so I can't
push this forward any more right now.

regards, tom lane