On Fri, Jan 16, 2026 at 08:22:48AM +0900, Michael Paquier wrote:
> I have a bunch of other changes for the rest of the patch set, that
> needs to be refreshed based on the choices we had to deal with for the
> clear function. I'll send that in a couple of hours, before moving on
> with more reviews next week..
And attached is the rebased patch set, that includes all the changes
coming from HEAD, as well as the embarrassing 395b73c045e0 fixed for
the restore function. This is still heavily WIP, but I have spent
some time learning and tweaking the code, completing it more,
including for the restore function:
- Tests for MAINTAIN.
- Tests for invalid values, invalid combinations and non-existing
objects.
- Regression test style. The thing is still bloated, particularly at
the end. We should improve that more.
- errmsg reworded to be in line with the normal style guideline.
- More comments.
- Fixed the RangeVar lookup to be consistent with the clear function.
- Adjusted the restore function to take a relation and its schema in
input.
Note that the restore function needs tests for the inherited record
case, and that we had the same error as the clear function when
grabbing the argument value for inherited. I have fixed the bug, did
not include any tests yet. Another thing that I find lacking is
coverage for the many paths that refer to incorrect input formats. A
lot of them are not covered at all, which is not good. I did not look
at the pg_dump portion yet.
--
Michael
