Re: Dumping/restoring fails on inherited generated column

Peter Eisentraut <peter(dot)eisentraut(at)2ndquadrant(dot)com> writes:
> On 2019-12-04 15:14, Tom Lane wrote:
>> It looks like gtest1_1 inherits column "b" from gtest1, so possibly
>> pg_dump is just confused about the combination of inheritance and
>> generated columns.

> Yeah, there was some stuff about the "separate" dumping of defaults that
> I apparently forgot to address. The attached patch fixes it. I'll see
> about adding a test case for it, too.

I don't think this is right. It will probably misbehave if the
"generated" expression has any interesting dependencies:

1. You didn't duplicate the behavior of the existing separate=false
case, where it adds a dependency to try to force the default's
dependencies to exist before the table is created.

2. If that dependency turns out to create a dependency loop, the
later code will break the loop by setting separate=true anyway.
Then what?

I also find it improbable that overriding the !shouldPrintColumn
test is really the right thing. That test is what distinguishes
the is-a-parent-table from the is-a-child-table cases, and the
core of the issue here seems to be that we need to treat those
differently.

I wonder if the right fix is to not generate a DO_ATTRDEF
object at all for generated columns in child tables. Am
I right in guessing that we propagate generated-ness to
child tables automatically?

regards, tom lane