Kevin Grittner-5 wrote
>> [ Examples shows that both SERIALIZABLE and REPEATABLE READ
>> transactions could see an empty table which was not empty as of
>> the point the snapshot was taken. For that matter, it was not
>> empty at any later point, either. ]
>
> Why don't we rewrite tuples with their existing xid in such cases?
> The current state of affairs seem to me to be a pretty clear bug.
In the ADD COLUMN scenario this results in the new column being visible when
it technically should not be but that is not likely a huge concern.
In the DROP COLUMN scenario you can no longer see data which should
technically be present. I guess if you go to use that data and it is not
present you'd get an error which is the technically correct response anyway
so probably not a huge concern either.
The idea of altering a record but not updating its xid sounds unclean but
I'm not able to evaluate any potential pitfalls of such an action.
David J.
--
View this message in context:
http://postgresql.1045698.n5.nabble.com/ALTER-TABLE-transaction-isolation-problem-tp5769289p5769393.html
Sent from the PostgreSQL - general mailing list archive at Nabble.com.