I wrote:
> I don't intend 0003 for back-patch, but the rest of this has to go
> back as far as the bug can be observed, which I didn't test yet.
Ah: this bug is new in v17, because neither brininsertcleanup nor
the whole aminsertcleanup infrastructure exist before that.
So that leaves us with a question of whether 0004 should be
back-patched any further than v17. There's some attraction
to doing so to keep the branches in sync; but there's already
quite a lot of cross-branch churn in worker.c, so on the whole
I'm inclined to just do v17.
regards, tom lane
