On 10/7/14, 9:11 AM, Feike Steenbergen wrote:
> Perhaps I am the only one using autocommit-off mode
You most definitely aren't.
> and we shouldn't put effort
> into fixing this?
It's not clear to me that this is fixing a problem, to be honest. If
you're running autocommit=off, you have an expectation that you can roll
back commands at will. It's fine if I can't roll back a VACUUM, for
example, since I would practically never want to do that. But ALTER
TYPE .. ADD VALUE ..; is an entirely different beast. That one's
permanent; there's no DROP equivalent. If the command is just executed,
and I can't roll it back, wouldn't that be a serious violation of the
principle of least astonishment? DROP INDEX CONCURRENTLY has a bit of
the same problem. You can CREATE INDEX CONCURRENTLY, but it might take
days in some cases.
I think that just running the command is a bad idea, and if we want to
fix something here we should focus on just providing a better error message.
.marko