Pushed 0001.
I confirm the finding that 0002 makes TestForOldSnapshot blow up. I
suppose the right fix is to pass the buffer, not just the page, rather
than trying to reverse-engineer the buffer from the page.
Regarding adding the proposed checks, which I think we should clearly
do, my preferred fix would be to split PageGetLSN in two: one that is
applied to shared buffers, which always runs the assertion, which
retains the current name, and another that is applied to other buffers,
which never does but it does check that the page is not in
shared_buffers and uses another name.
--
Álvaro Herrera https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
