Re: Transactional DDL, but not Serializable

* Joshua Berkus (josh@agliodbs.com) wrote:
> That seemed unnecessary.  Whether or not you approve of Stephen's solution, he is dealing with a real issue.

The solution felt, to me at least, to have a lot of parallel to an
index's indcheckxmin.  We've dealt with this issue there and have a
precedent for how to deal with it.  Based on discussions with other
folks it sounds like we may be forced to do it for constraints also, and
I think we'd want to try to deal with all of them in a similar way.

Perhaps the current solution for indexes is a hack and should be tossed
out with a wholesale replacment which solves all these problems, which
would certainly be quite a bit of work, but if that's necessary then
let's discuss it and get an idea down on a wiki somewhere about what
that should look like.
Thanks,
    Stephen