Thread: Bad link Makefile patch

I built the v8.0.3 product from postgresql-8.0.3-1PGDG.src.rpm on
RedHat9 (I'm thinking the source RPM for RH9 should not have exactly the
same name as the FC3 version, since they are different files).  When I
tried to roll it into an RPM CD builder transaction, I got 'RPM
dependency errors': "CRITICAL ERROR: Unable to resolve dependency
libecpg.so.3 for postgresql-libs" and "CRITICAL ERROR: Unable to resolve
dependency libpq.so.3 for postgresql-libs".  I was only including the
base rpm, -server, and -lib in the RPM transaction set.

The culprit was a nasty bit at $TOP/src/Makefile.shlib:243.  This piece
insures that for the link step of libecpg.so.5.0 and
libecpg_compat.so.2.0, /usr/lib is searched for libpq and libecpg.so
_before_ the locally built copy is searched.

At least for libecpg.so.5.0 and libecpg_compat.so.2.0, the attached
patch fixes the problem.  Not sure if there would be other instances in
-contrib, etc.

Hope this helps and that I'm not redundant with your other fans.

Regards,
Mark

So, my co-workers chided me mercilessly to get the -contrib RPM working
as well; so the full patch is now attached.

BTW, I did search the archive and this problem did not stick out; but it
could be my crummy reference skills.  And, no, I didn't read all 3500
emails since the v8.0.3 release.

Also, there's some good chance that the binary RPMs on the
postgresql.org FTP site have the same flaw that I encountered as
described below.

Regards,
Mark

On Wed, 6 Jul 2005 09:08:50 -0700
Mark Deric <laguna@laguna1.com> wrote:

> I built the v8.0.3 product from postgresql-8.0.3-1PGDG.src.rpm on
> RedHat9 (I'm thinking the source RPM for RH9 should not have exactly
> the same name as the FC3 version, since they are different files).
> When I tried to roll it into an RPM CD builder transaction, I got 'RPM
> dependency errors': "CRITICAL ERROR: Unable to resolve dependency
> libecpg.so.3 for postgresql-libs" and "CRITICAL ERROR: Unable to
> resolve dependency libpq.so.3 for postgresql-libs".  I was only
> including the base rpm, -server, and -lib in the RPM transaction set.
>
> The culprit was a nasty bit at $TOP/src/Makefile.shlib:243.  This
> piece insures that for the link step of libecpg.so.5.0 and
> libecpg_compat.so.2.0, /usr/lib is searched for libpq and libecpg.so
> _before_ the locally built copy is searched.
>
> At least for libecpg.so.5.0 and libecpg_compat.so.2.0, the attached
> patch fixes the problem.  Not sure if there would be other instances
> in-contrib, etc.
>
> Hope this helps and that I'm not redundant with your other fans.
>
> Regards,
> Mark
>

Mark Deric <laguna@laguna1.com> writes:
> So, my co-workers chided me mercilessly to get the -contrib RPM working
> as well; so the full patch is now attached.

I think this overlaps or may be redundant with the patch that Peter E.
recently proposed; have you been following the thread about buildfarm
failures?

http://archives.postgresql.org/pgsql-hackers/2005-07/msg00178.php

            regards, tom lane

Tom Lane wrote:
> Mark Deric <laguna@laguna1.com> writes:
> > So, my co-workers chided me mercilessly to get the -contrib RPM working
> > as well; so the full patch is now attached.
>
> I think this overlaps or may be redundant with the patch that Peter E.
> recently proposed; have you been following the thread about buildfarm
> failures?
>
> http://archives.postgresql.org/pgsql-hackers/2005-07/msg00178.php

I am going to assume this issue was resolved by Peter's patch.

--
  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 359-1001
  +  If your life is a hard drive,     |  13 Roberts Road
  +  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073