Bruce Momjian <pgman@candle.pha.pa.us> writes:
> I have updated the TODO to:
> o Abort all or commit all SET changes made in an aborted transaction
> I don't think our current behavior is defended by anyone.
Hiroshi seems to like it ...
However, "commit SETs even after an error" is most certainly NOT
acceptable. It's not even sensible --- what if the SETs themselves
throw errors, or are depending on the results of failed non-SET
commands; will you try to commit them anyway?
It seems to me that the choices we realistically have are
(a) leave the behavior the way it is
(b) cause all SETs in an aborted transaction to roll back.
regards, tom lane