On Fri, Nov 22, 2013 at 04:38:53PM -0800, Kevin Grittner wrote:
> Andres Freund <andres@2ndquadrant.com> wrote:
> > On 2013-11-22 13:34:18 -0800, Kevin Grittner wrote:
> >> Oddly, it didn't complain about creating users within a read-only
> >> transaction. That seems like a potential bug.
> >
> > There's lots of things that escape XactReadOnly. I've thought (and I
> > think suggested) before that we should put in another layer of defense
> > by also putting a check in AssignTransactionId(). Imo the compatibility
> > woes (like not being able to run SELECT txid_current();) are well worth
> > the nearly ironclad guarantee that we're not writing.
>
> I agree that something like that is would be a good idea; however,
> I'm sure you would agree that would not be material for a
> back-patch to a stable branch.
I am not a fan of backpatching any of this. We have learned the fix is
more complex than thought, and the risk of breakage and having pg_dump
diffs change between minor releases doesn't seem justified.
-- Bruce Momjian <bruce@momjian.us> http://momjian.us EnterpriseDB
http://enterprisedb.com
+ Everyone has their own god. +