Обсуждение: Long-time 7.4 contrib failure Mac OS X 10.3.8
The 7.4 tree has never built cleanly on Wallaroo, a Mac OS X 10.3.8 member of the build farm. Currently it's failing in the make contrib stage. I'd like to get it to build properly, but I don't know enough to be able to make sense of the log output. I'd be grateful if someone could spare a few minutes to take a look at the log and give me an idea of what it would take to fix it. http://pgbuildfarm.org/cgi-bin/show_log.pl?nm=wallaroo&dt=2005-11-10% 2021:15:00 It looks to me like there's a linking error with GIST. Does this mean it's looking for libraries in the wrong place? Perhaps the libraries it's looking for aren't installed on this box? Perhaps it's a configure problem? Thanks for any insight! Michael Glaesemann grzm myrealbox com
Michael Glaesemann <grzm@myrealbox.com> writes:
> The 7.4 tree has never built cleanly on Wallaroo, a Mac OS X 10.3.8
> member of the build farm. Currently it's failing in the make contrib
> stage.
I'm afraid 7.4 will probably never build completely cleanly on OS X.
The failure you're showing is related to the dblink -> libpq -> libssl
dependency chain. It works OK on PG 8.0 and later, and AFAICT the only
relevant difference is that 8.0 builds libpq as a dynamiclib (.dylib)
instead of a bundle (.so). Apparently the linker follows libpq's
dependency on libssl in one case but not the other. Unfortunately,
changing this in 7.4 would be an ABI breakage and so is not likely to
happen.
I believe you could get 7.4 contrib to build if you take out
--with-openssl in that branch's configure arguments.
You may find you have to back off on --with-python and/or --with-tcl
as well :-( I don't recall when we first got those PLs to work on
OS X, but it wasn't all that long ago.
regards, tom lane
On Nov 15, 2005, at 11:34 , Tom Lane wrote: > Michael Glaesemann <grzm@myrealbox.com> writes: >> The 7.4 tree has never built cleanly on Wallaroo, a Mac OS X 10.3.8 >> member of the build farm. Currently it's failing in the make contrib >> stage. <snip /> > I believe you could get 7.4 contrib to build if you take out > --with-openssl in that branch's configure arguments. > > You may find you have to back off on --with-python and/or --with-tcl > as well :-( I don't recall when we first got those PLs to work on > OS X, but it wasn't all that long ago. Thanks for the explanation. Is making this change to the build farm machine's config worth doing? Is it more useful on the build farm to see what works, or to see what fails? I'm thinking the latter. It'd be nice to see green, but green for green's sake isn't the point. Michael Glaesemann grzm myrealbox com
Michael Glaesemann <grzm@myrealbox.com> writes:
> Thanks for the explanation. Is making this change to the build farm
> machine's config worth doing? Is it more useful on the build farm to
> see what works, or to see what fails? I'm thinking the latter. It'd
> be nice to see green, but green for green's sake isn't the point.
True, but if we're writing off openssl-on-OSX-on-7.4 as "never will
happen", as I think we must, it's probably better to configure the
buildfarm to avoid the problem. Otherwise the failure may simply
mask breakages that we *would* like to know about. In particular,
we presently have no idea whether wallaroo would expose any other
contrib problems if it were able to get past building dblink;
and the fact that it's red and always has been red discourages
people from noticing if the failure changes to something else.
In general, I like build machines that usually are green. I've learned
to totally tune out failures from tuna and carp, for instance, because
they fail every couple of days due to local problems such as lack of
SHMMAX space. If they ever show a real problem, it's unlikely anyone
would notice for a long while --- ye old "cry wolf too much" problem.
regards, tom lane
On Nov 15, 2005, at 12:20 , Tom Lane wrote: > In particular, > we presently have no idea whether wallaroo would expose any other > contrib problems if it were able to get past building dblink; > and the fact that it's red and always has been red discourages > people from noticing if the failure changes to something else. Makes sense. I've stripped --with-openssl, and --with-tcl and --with- python for good measure. It still fails, and to my eye, with the same failure. http://pgbuildfarm.org/cgi-bin/show_log.pl?nm=wallaroo&dt=2005-11-15% 2006:29:06 Michael Glaesemann grzm myrealbox com
Michael Glaesemann <grzm@myrealbox.com> writes:
> Makes sense. I've stripped --with-openssl, and --with-tcl and --with-
> python for good measure. It still fails, and to my eye, with the same
> failure.
Much smaller list of symbols though:
ld: Undefined symbols:
_bindtextdomain
_dgettext
Apparently you don't get to have --enable-nls either :-(
regards, tom lane
On Nov 15, 2005, at 23:19 , Tom Lane wrote: > Apparently you don't get to have --enable-nls either :-( And we've got green! :) http://pgbuildfarm.org/cgi-bin/show_log.pl?nm=wallaroo&dt=2005-11-15% 2022:55:15 Thanks, Tom, for taking the time to look at this and explaining it a bit to me. Michael Glaesemann grzm myrealbox com
Michael Glaesemann <grzm@myrealbox.com> writes: > On Nov 15, 2005, at 23:19 , Tom Lane wrote: >> Apparently you don't get to have --enable-nls either :-( > And we've got green! :) > http://pgbuildfarm.org/cgi-bin/show_log.pl?nm=wallaroo&dt=2005-11-15% > 2022:55:15 > Thanks, Tom, for taking the time to look at this and explaining it a > bit to me. [ looks at buildfarm... ] It looks like you dropped the extra options on all of wallaroo's branches, which is not good. They should work in 8.0 and later, and you should test 'em where they're claimed to work. I'm pretty sure you can have per-branch configure options in buildfarm, but I don't recall how. Whack Andrew Dunstan about it if you can't see how from the documentation... regards, tom lane
On Tue, Nov 15, 2005 at 11:53:11PM -0500, Tom Lane wrote: > Michael Glaesemann <grzm@myrealbox.com> writes: > > On Nov 15, 2005, at 23:19 , Tom Lane wrote: > >> Apparently you don't get to have --enable-nls either :-( > > > And we've got green! :) > > http://pgbuildfarm.org/cgi-bin/show_log.pl?nm=wallaroo&dt=2005-11-15% > > 2022:55:15 > > Thanks, Tom, for taking the time to look at this and explaining it a > > bit to me. > > [ looks at buildfarm... ] It looks like you dropped the extra options > on all of wallaroo's branches, which is not good. They should work in > 8.0 and later, and you should test 'em where they're claimed to work. > > I'm pretty sure you can have per-branch configure options in buildfarm, > but I don't recall how. Whack Andrew Dunstan about it if you can't see > how from the documentation... You can. Sample from by mac: if ($branch eq 'HEAD') { push(@{$conf{config_opts}},"--enable-depend"); } if ($branch !~ /^REL7/) { push(@{$conf{config_opts}}, qw( --with-openssl ) ); } if ($branch !~ /^REL7_[23]/) { push(@{$conf{config_opts}},"--enable-nls"); } There's also some way you can pull an item out of an array, which could make that --with-openssl bit cleaner... -- Jim C. Nasby, Sr. Engineering Consultant jnasby@pervasive.com Pervasive Software http://pervasive.com work: 512-231-6117 vcard: http://jim.nasby.net/pervasive.vcf cell: 512-569-9461
On Nov 16, 2005, at 13:53 , Tom Lane wrote: > [ looks at buildfarm... ] It looks like you dropped the extra options > on all of wallaroo's branches, which is not good. They should work in > 8.0 and later, and you should test 'em where they're claimed to work. [looks at build-farm.conf, blushes] Actually, just on HEAD. I'm using per-branch configurations, but missed HEAD. I'll get that fixed. Thanks for the catch. Michael Glaesemann grzm myrealbox com
On Nov 16, 2005, at 15:37 , Michael Glaesemann wrote: > I'm using per-branch configurations, but missed HEAD. I'll get that > fixed. Thanks for the catch. Back to normal: http://pgbuildfarm.org/cgi-bin/show_log.pl?nm=wallaroo&dt=2005-11-16% 2009:12:09 Michael Glaesemann grzm myrealbox com