> On Jul 6, 2022, at 12:51, Matthias Apitz <g...@unixarea.de> wrote:
> it is uniqu to identify a row in a table once
> known.
I think the point that we are trying to make here is that a ctid *isn't* that.
There is no guarantee, at all, at any level, that the ctid of a row will remain
stable, not even between two SELECT statements. (Although it doesn't right
now, I could easily image some kind of repack logic in PostgreSQL that runs on
read operations, not just write.) It shouldn't be considered an API. I
understand that it might be painful to change to a generated primary key, but I
think that will be less painful in the long run than having to stay ahead of
PostgreSQL's internals.
