On 5/22/19 9:46 AM, Tom Lane wrote:
> Michael Paquier <michael@paquier.xyz> writes:
>> Trying to do pg_dump[all] on a 9.5 or older server results in spurious
>> failures:
>> pg_dump: column number -1 is out of range 0..36
>> After looking around, the problem comes from
>> check_tuple_field_number(), more specifically from getTables() where
>> someone has forgotten to add NULL values for amname when querying
>> older server versions.
>> Attached is a patch to fix that. I am not seeing other failures with
>> an instance that includes all the contents of installcheck, so it
>> seems that the rest is fine.
> Looks like the right fix. I'm sad that the buildfarm did not catch
> this ... why wouldn't the cross-version-upgrade tests have seen it?
That's a good question.
It's in the output - see for example
<https://buildfarm.postgresql.org/cgi-bin/show_stage_log.pl?nm=crake&dt=2019-05-22%2017%3A47%3A30&stg=xversion-upgrade-REL9_4_STABLE-HEAD>
and scroll down a bit.
But since this doesn't cause pg_dumpall to fail, it doesn't on its own
cause the buildfarm to fail either, and this is apparently sufficiently
benign to allow the tests to succeed.
cheers
andrew
--
Andrew Dunstan https://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services