Stephan Szabo <sszabo(at)megazone23(dot)bigpanda(dot)com> writes:
> Of course, I also really only noticed it when I ran the two really
> close together or the alter table inside a transaction.
Right, the problem only comes up if the report of the table change
arrives at the other backend just as it's preparing to start a new
transaction on that same table. If the other backend's first table
lock after the ALTER commits is on some other table, no problem.
We do need to fix it but I think there are higher-priority problems...
regards, tom lane