Re: Attribute of type record has wrong type error with MERGE ... WHEN NOT MATCHED BY SOURCE THEN DELETE

I wrote:
> Double ugh: that doesn't fix it, because we also do a round of
> WRITE_READ_PARSE_PLAN_TREES checks on the rewriter output.
> Not sure how to fix this, unless we lobotomize that write/read
> check somehow.

After some thought, I'm inclined to suggest that we just remove
the post-rewrite check in the affected branches. This is certainly
not an ideal solution, but unless someone can think of a
fundamentally different way to fix the original bug in these branches,
I don't see a better way. There are some mitigating points:

* Applying WRITE_READ_PARSE_PLAN_TREES here does not correspond to
any actual system functionality requirement. We have to be able
to write/read post-parse-analysis trees to store views and rules;
and we have to be able to write/read plan trees to transfer them
to parallel workers. But there's no real reason why we have to
be able to do it for the purely-transient output of rewriting.

* We aren't going to be changing the outfuncs/readfuncs code
in these branches anymore anyway, for precisely the same reasons
that this patchset can't do that. So it's not clear what the
test is doing for us.

So that's the best I've got for tonight. I've verified that the
attached patch for v15 fixes this failure.

regards, tom lane