Tom Lane <tgl@sss.pgh.pa.us> writes:
> Completely. This is a user-visible behavior that we have encouraged
> people to rely on, and for which there is no easy substitute.
Excited to have self-healing tables (against bloat), I parse this as an
opening. Previously on this thread you say:
> (Actually, the ctid is only being used for fast access here; the xmin
> is what is really needed to detect that someone else updated the row.
> But the proposed tuple-mover would break the xmin check too.)
So to have the impossible feature, we need a way not to break existing
code relying on ctid and xmin. How stretching would you consider the
idea of taking a (maybe new) table lock as soon as a SELECT output
contains system columns, this lock preventing the magic utility to
operate?
Regards,
--
dim
