On Wed, 4 Nov 2009, Kevin Grittner wrote:
> Having received no feedback on the preliminary patch I previously
> posted, I'll suggest this (fairly minimal) patch. I didn't attempt to
> deal with the distributed transaction issue, since it seemed pretty
> messy and I wasn't sure the fix was worth the mess it would make. As
> far as I can see, it should fix the OP's problem, although it would be
> nice to get a confirmation of that.
>
I've commited a modified version of this patch to HEAD only, because
changing this behavior could break peoples applications.
I extended the check for the connection already being closed to the other
connection methods. There's no reason commit/rollback are special here.
I also needed to fix a regression testcase which was violating this rule.
Kris Jurka