Andres Freund <andres@anarazel.de> writes:
> On 2018-02-21 10:08:12 -0500, Tom Lane wrote:
>> Yeah, that does look suspiciously like a set of facts matching this
>> problem :-(
> I'd personally say ;), given that it's one less potentially data
> corrupting bug to worry about ;). And it's one that can be fixed without
> dataloss to boot.
I've pushed a fix for this. I concluded after further study that there
wasn't really much call for a unified notion of applicable relkinds,
because actually what this code is dealing with is "relkinds that need
preservation of dropped columns" and "relkinds that need preservation
of relfrozenxid", both of which are somewhat different from "relkinds
that have storage". What's really lacking is documentation, which
I attempted to supply. For the code change proper, I did what you
suggested and split the relfrozenxid hacking into an independent
if-block.
regards, tom lane