On 2019-03-09 02:35, Michael Paquier wrote:
> On Fri, Mar 08, 2019 at 12:00:40PM +0100, Peter Eisentraut wrote:
>> OK, let's go with this. Attached is your patch with my comment wording.
>
> Thanks for double-checking. Your change looks fine to me.
>
>> Also a test case, but I don't suggest committing that, it's a bit too
>> weird. (The effect of the test is to cause catalog corruption that will
>> cause a later test to fail.) We can leave it in the archive for posterity.
>
> Letting the test case out is fine for me too.
I looked into backpatching this, but the test case doesn't fail in PG10.
I ran a round of bisecting but didn't arrive at a sensible result. The
test case appears to be a bit random in its failure modes. The issue
could in principle extend further back, since the code in question isn't
really new. Any ideas or suggestions?
--
Peter Eisentraut http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services