Date: Thu, 21 Feb 2019 10:30:30 -0600
From: Justin Pryzby <pryzby@telsasoft.com>
To: Christoph Berg <myon@debian.org>
Cc: pgsql-pkg-yum@postgresql.org
Subject: Re: [PostgreSQL YUM Repository - Bug #3710] (Resolved) RHEL7
 postgresql11 postgis2.5 fails with /usr/pgsql-11/lib/postgis-2.5.so:
 undefined symbol: geod_polygon_init
Message-ID: <20190221163030.GY28750@telsasoft.com>
References: <redmine.issue-3710.20181018173810@postgresql.org>
 <redmine.journal-14074.20181210113352.5ccd98437e7314f6@postgresql.org>
 <20181210150643.GF4848@telsasoft.com>
 <20181219202107.GA5406@telsasoft.com>
 <20190107215410.GA22493@telsasoft.com>
 <20190221132606.GC5238@msg.df7cb.de>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20190221132606.GC5238@msg.df7cb.de>
User-Agent: Mutt/1.5.23 (2014-03-12)
Precedence: bulk

On Thu, Feb 21, 2019 at 02:26:06PM +0100, Christoph Berg wrote:
> Why does postgis-2.4.so even "see" the geos37 package?
> 
> > [pryzbyj@template0 ~]$ ldd /usr/pgsql-11/lib/postgis-2.4.so
> >         libgeos_c.so.1 => /usr/geos36/lib64/libgeos_c.so.1 (0x00007f3e50e6e000)
> >         libgeos-3.6.2.so => /usr/geos36/lib64/libgeos-3.6.2.so (0x00007f3e4f12e000)
> > 
> > Fixed with: rpm -e geos36-3.6.2-3.1.rhel7.x86_64
> > Should declare the RPM equivalent of debian's "Breaks" ??

Note that the problem was postgis24 was "seeing" geos 36 INSTEAD OF 37, not
that it was seeing 37.

I don't know about debian (from which I've been absent for a decade), but I
think for centos it's because of this:

[pryzbyj@template0 ~]$ rpm -ql geos37-3.7.0-1.rhel7.1.x86_64
/etc/ld.so.conf.d/geos37-pgdg-libs.conf
...
[pryzbyj@template0 ~]$ cat /etc/ld.so.conf.d/geos37-pgdg-libs.conf
/usr/geos37/lib64/

[pryzbyj@database ~]$ rpm -ql geos36
/etc/ld.so.conf.d/geos36-pgdg-libs.conf
...
[pryzbyj@database ~]$ cat /etc/ld.so.conf.d/geos36-pgdg-libs.conf 
/usr/geos36/lib64/

Presumably, if both are installed it finds geos36-pgdg-libs.conf first, thus
linker uses to geos36.

I think to fully resolve this it's necessary to *prohibit* co-installation of
new postgis with old geos.

But note that's not adequately resolved by releasing a new geos36, since
someone might run rpm/dpkg -i on a single package, without installing all the
other updates from the repo.

Justin