On Jul25, 2011, at 20:37 , Bernd Helmle wrote:
> Ah, but i got now what's wrong here: configure is confusing both libxml2
> installations, and a quick look into config.log proves that: it uses the
> xml2-config from the OSX libs (my $PATH has /usr in front of the bindir of
> MacPorts, though i seem to recall to have changed this in the past....).
>
> So, all i need to do is
>
> XML2_CONFIG=/opt/local/bin/xml2-config ./configure --with-libxml
> --with-includes=/opt/local/include/ --with-libraries=/opt/local/lib
>
> and everything is smooth:
>
> % grep HAVE_XMLSTRUCTUREDERRORCONTEXT src/include/pg_config.h
> #define HAVE_XMLSTRUCTUREDERRORCONTEXT 1
>
> Regression tests passes now. This was too obvious...
Hm, but I still think there's a bug lurking there. Using a different libxml2
version for the configure checks than for actual builds surely isn't good...
From looking at configure.in, it seems that we use xml2-config to figure out
the CFLAGS and LDFLAGS required to build and link against libxml. I guess we
somehow end up not using these flags when we later test for
xmlStructuredErrorContext, but do use them during the actual build. Or maybe
the order of the -I and -L flags just ends up being different in the two cases.
My skills in the black art that are autotools are severely lacking, so it's
quite likely that I somehow botched the incantations we use to test for
xmlStructuredErrorContext. I don't really know where to start looking for the
error, though. Ideas, anyone?
best regards,
Florian Pflug