Thread: Changes for AIX buildfarm
Need to have these patches applied to get AIX 5.2 to build as a member of the buildfarm. The Makefile.shlib patch allows all the contrib modules to build (this seems to be the same problem that cygwin is facing). AIX needs -lm (to pick up math routines for atleast cube module). I am not sure why the other platforms build without this. There are C++ style comments in contrib/pgcrypto which is giving cc problems. Thanks, -rocco
Attachment
"Rocco Altier" <RoccoA@Routescape.com> writes: > Need to have these patches applied to get AIX 5.2 to build as a member > of the buildfarm. Er, which buildfarm report is prompting this? > There are C++ style comments in contrib/pgcrypto which is giving cc > problems. Definitely contrary to project policy, will fix. regards, tom lane
Tom Lane wrote: >"Rocco Altier" <RoccoA@Routescape.com> writes: > > >>Need to have these patches applied to get AIX 5.2 to build as a member >>of the buildfarm. >> >> > >Er, which buildfarm report is prompting this? > > > Rocco is probably doing what I normally ask, which is to get builds working before starting to upload results to the buildfarm server (a request which is more honored in the breach than the observance). cheers andrew
Andrew Dunstan <andrew@dunslane.net> writes: > Tom Lane wrote: >> Er, which buildfarm report is prompting this? > Rocco is probably doing what I normally ask, which is to get builds > working before starting to upload results to the buildfarm server (a > request which is more honored in the breach than the observance). Fair enough. But I'm concerned about the proposed patch because it seems to revert a deliberate change made some time ago: http://developer.postgresql.org/cvsweb.cgi/pgsql/src/Makefile.shlib#rev1.65 http://archives.postgresql.org/pgsql-committers/2002-10/msg00054.php so I think we need more eyeballs on the problem before deciding this is a good fix. regards, tom lane
It looks like I jumped the gun a bit on sending out the patches, since I was intending to have the failed build report show up in the buildfarm at the same time. I am running another build with send so that the error will show on the build farm. The animal name is appropiately: kookaburra -rocco > -----Original Message----- > From: Andrew Dunstan [mailto:andrew@dunslane.net] > Sent: Monday, July 18, 2005 12:42 PM > To: Tom Lane > Cc: Rocco Altier; pgsql-patches@postgresql.org > Subject: Re: [PATCHES] Changes for AIX buildfarm > > > > > Tom Lane wrote: > > >"Rocco Altier" <RoccoA@Routescape.com> writes: > > > > > >>Need to have these patches applied to get AIX 5.2 to build > as a member > >>of the buildfarm. > >> > >> > > > >Er, which buildfarm report is prompting this? > > > > > > > > Rocco is probably doing what I normally ask, which is to get builds > working before starting to upload results to the buildfarm server (a > request which is more honored in the breach than the observance). > > cheers > > andrew >
Marko Kreen <marko@l-t.ee> writes: > On Mon, Jul 18, 2005 at 12:32:21PM -0400, Tom Lane wrote: >> Definitely contrary to project policy, will fix. > Don't, they are fixed in "small cleanups" patch that I sent. Too late, already committed ;-) I will do something to clean up the mess... regards, tom lane
On Mon, Jul 18, 2005 at 12:32:21PM -0400, Tom Lane wrote: > "Rocco Altier" <RoccoA@Routescape.com> writes: > > There are C++ style comments in contrib/pgcrypto which is giving cc > > problems. > > Definitely contrary to project policy, will fix. Don't, they are fixed in "small cleanups" patch that I sent. -- marko
I have looked more closely at what is happening during the link of the contrib modules, and I agree that my original patch is too broad. I am seeing that cube fails, but userlocks works. I have noticed that cube is using a MODULE_big directive (which is failing to pick up all libraries), but that userlocks is using MODULES which is working correctly. I will dig further and come up with another patch. -rocco > -----Original Message----- > From: Tom Lane [mailto:tgl@sss.pgh.pa.us] > Sent: Monday, July 18, 2005 12:47 PM > To: Andrew Dunstan > Cc: Rocco Altier; Zeugswetter Andreas DAZ SD; > pgsql-patches@postgresql.org > Subject: Re: [PATCHES] Changes for AIX buildfarm > > > Andrew Dunstan <andrew@dunslane.net> writes: > > Tom Lane wrote: > >> Er, which buildfarm report is prompting this? > > > Rocco is probably doing what I normally ask, which is to get builds > > working before starting to upload results to the buildfarm > server (a > > request which is more honored in the breach than the observance). > > Fair enough. But I'm concerned about the proposed patch because it > seems to revert a deliberate change made some time ago: > http://developer.postgresql.org/cvsweb.cgi/pgsql/src/Makefile.shlib#rev1 .65 http://archives.postgresql.org/pgsql-committers/2002-10/msg00054.php so I think we need more eyeballs on the problem before deciding this is a good fix. regards, tom lane
The problem trying to be fixed there seems to be related to libraries being linked into the backend vs. frontend. Since the contrib shared library modules are linked in a similar environment backend code I decided to use BE_DLLLIBS. Here is the new patch (which doesn't affect libpq). -rocco Index: src/makefiles/Makefile.aix =================================================================== RCS file: /projects/cvsroot/pgsql/src/makefiles/Makefile.aix,v retrieving revision 1.21 diff -c -r1.21 Makefile.aix *** src/makefiles/Makefile.aix 9 Oct 2002 16:21:54 -0000 1.21 --- src/makefiles/Makefile.aix 18 Jul 2005 19:59:59 -0000 *************** *** 15,20 **** --- 15,21 ---- LDFLAGS_SL = -Wl,-bnoentry -Wl,-H512 -Wl,-bM:SRE endif + BE_DLLLIBS= $(LIBS) EXPSUFF= .exp IMPSUFF= .imp > -----Original Message----- > From: Tom Lane [mailto:tgl@sss.pgh.pa.us] > Sent: Monday, July 18, 2005 12:47 PM > To: Andrew Dunstan > Cc: Rocco Altier; Zeugswetter Andreas DAZ SD; > pgsql-patches@postgresql.org > Subject: Re: [PATCHES] Changes for AIX buildfarm > > > Andrew Dunstan <andrew@dunslane.net> writes: > > Tom Lane wrote: > >> Er, which buildfarm report is prompting this? > > > Rocco is probably doing what I normally ask, which is to get builds > > working before starting to upload results to the buildfarm > server (a > > request which is more honored in the breach than the observance). > > Fair enough. But I'm concerned about the proposed patch because it > seems to revert a deliberate change made some time ago: > http://developer.postgresql.org/cvsweb.cgi/pgsql/src/Makefile.shlib#rev1 .65 http://archives.postgresql.org/pgsql-committers/2002-10/msg00054.php so I think we need more eyeballs on the problem before deciding this is a good fix. regards, tom lane
Tom Lane wrote: > Fair enough. But I'm concerned about the proposed patch because it > seems to revert a deliberate change made some time ago: > > http://developer.postgresql.org/cvsweb.cgi/pgsql/src/Makefile.shlib#r >ev1.65 > http://archives.postgresql.org/pgsql-committers/2002-10/msg00054.php > > so I think we need more eyeballs on the problem before deciding this > is a good fix. A potentially more correct solution would be to set SHLIB_LINK in the makefiles of the affected modules only. Certainly we don't want to link with all libraries all the time. -- Peter Eisentraut http://developer.postgresql.org/~petere/
Attached is a patch for HEAD to implement Peter's suggestion to set the SHLIB_LINK in the contrib modules that need it. Also, this changes Makefile.aix to use SHLIB_LINK instead of LIBS so that the changes to the contrib Makefiles are picked up correctly. Further this makes it match more closely the link rules from Makefile.shlib. -rocco > -----Original Message----- > From: Peter Eisentraut [mailto:peter_e@gmx.net] > Sent: Wednesday, July 20, 2005 2:12 PM > To: Rocco Altier > Cc: pgsql-patches@postgresql.org; Tom Lane; Andrew Dunstan; > Zeugswetter Andreas DAZ SD > Subject: Re: [PATCHES] Changes for AIX buildfarm > > > Tom Lane wrote: > > Fair enough. But I'm concerned about the proposed patch because it > > seems to revert a deliberate change made some time ago: > > > > > http://developer.postgresql.org/cvsweb.cgi/pgs> ql/src/Makefile.shlib#r > >ev1.65 > > http://archives.postgresql.org/pgsql-committers/2002-10/msg00054.php > > > > so I think we need more eyeballs on the problem before deciding this > > is a good fix. > > A potentially more correct solution would be to set SHLIB_LINK in the > makefiles of the affected modules only. Certainly we don't want to > link with all libraries all the time. > > -- > Peter Eisentraut > http://developer.postgresql.org/~petere/ >