> I adjusted the set_attnotnull() API and removed the added > queue_validation parameter. Rather, the function start using wqueue > input parameter as a check. > If wqueue is NULL, skip the queue_validation. Attaching patch here, > but not sure how clear it is, in term of understanding the API. Your > thoughts/inputs?
Yeah, I think this makes sense, if it supports all the cases correctly. (If in your code you have any calls of set_attnotnull that pass a NULL wqueue during ALTER TABLE in order to avoid queueing a check, you had better have comments to explain this.)
Done.
Kindly do not send patches standalone, because the CFbot doesn't understand that it needs to apply it on top of the three previous patches. You need to send all 4 patches every time. You can see the result here: https://commitfest.postgresql.org/patch/5554/ If you click on the "need rebase!" label you'll see that it tried to apply patch 0004 to the top of the master branch, and that obviously failed. (FWIW if you click on the "summary" label you'll also see that the patch has been failing the CI tests since Feb 27.)
Sure, attaching the rebased patch here.
> Looking further for AdjustNotNullInheritance() I don't see need of > rename this API as it's just adding new error check now for an > existing NOT NULL constraint.
OK.
> JFYI, I can reproduce the failure Ashutosh Bapat reported, while > running the pg_upgrade test through meson commands. I am > investigating that further.
Ah good, thanks.
Somehow, I am now not able to reproduce after the clean build. Yesterday
I was able to reproduce, so I was happy, but again trying to analyze the issue
