On 2014/10/29 11:23, Tom Lane wrote:
> Michael Paquier <michael.paquier@gmail.com> writes:
>> On Wed, Oct 29, 2014 at 10:44 AM, Romu Hu <huruomu@gmail.com> wrote:
>>> I ran the test against an existing installation (redhat enterprise linux
>>> software collection postgresql92). The postgres server and the tests are
>>> from the same source package.
>> Well, your diffs are telling us the contrary.
> Yeah, this is definitely some kind of version skew problem.
>
>> So you may be indeed running the tests on a 9.2 server, but what is
>> sure is that you are comparing the results with the regression output
>> of a 9.4 server.
> No, that's not quite right I think. I can't find anything in a quick
> look that is clearly different server behavior. There are a bunch of
> differences in output of \d commands, and a bunch of different formatting
> of query results containing newlines, but both of those things are on
> psql's head not the server's. The psql being used is clearly older than
> 9.0, which is where the display of newlines changed. I'd bet on it being
> the 8.4.something version shipped by Red Hat with their regular RHEL6
> postgresql package.
You are right, the test was using /usr/bin/psql from RHEL6
postgresql-8.4.20 package, hence the diffs. I removed postgresql-8.4.20
and reran the regression tests from the postgresql92 software
collection, it failed the run because it couldn't find /usr/bin/psql, I
linked /opt/rh/postgresql92/root/usr/bin/psql to /usr/bin/psql and all
regression passed.
Thanks
Romu
> I managed to leave Red Hat before they shipped any of those "software
> collections" packages, but when I was there I was pretty unimpressed
> with that packaging technology. You have to use the packages just so
> or things fall apart, because for example their programs aren't in the
> system default PATH. I think something like that happened here:
> somehow or other the test process is invoking /usr/bin/psql and not
> the psql included in the software-collections PG package.
>
> Take a close re-read of the documentation for the software-collections
> PG package and see if you missed a setup setup (scenable or whatever it
> was called), or maybe they have a special recipe for running the
> regression tests. If you can't find anything, file a bug with Red Hat
> (not us), to the effect that their regression test packaging is either
> broken or underdocumented.
>
> regards, tom lane
