Re: questions about the configure script for building postgres 9.0.3 from source

Stephan Doliov <stephandoliov(at)yahoo(dot)com> writes:
> I trolled the archives as I was having difficulty in building Postgres 9.0.3
> from source. I was running configure with the following options (the pasted in
> text of this I had wrapped in a shell script). I apologize in advance that
> setting up the question takes a bit of context.

> ./configure CFLAGS='-O2 -I /usr/local/include:/usr/include -L
> /usr/local/lib64:/usr/local/lib:/usr/lib64:/usr/lib' --prefix=/usr/local
> --enable-nls --enable-profiling --with-wal-segsize=64 --with-perl
> --with-python --with-pam --with-ldap --with-openssl --with-ossp-uuid
> --with-libxml --with-libxslt --with-system-tzdata=/usr/share/zoneinfo
> --with-libs=/usr/local/lib64:/usr/local/lib:/usr/lib64:/usr/lib:/usr/apache2/lib
> --with-includes=/usr/local/include:/usr/include

> and I was constantly receiving the error that libreadline was not installed,
> although I had personally installed it, verified its installation and ensured
> that ldconfig could find it.

Are you sure it's complaining about readline's library, and not
readline's header files? That rather random --with-libs list draws my
suspicion, particularly since it doesn't parallel your --with-includes
list.

> Upon inspecting config.log, I found many references to an error in conftest.c,
> where various ifndef macros were suggesting a call to "choke me", which is both
> invalid syntax (no semi colon and of course choke is not defined either in any
> header files.

That's pretty normal --- configure likes to try to compile things like

#ifdef SYMBOL
choke me
#endif

to find out whether SYMBOL is defined. (If it is, this'll draw a
compile error, else not.) Elegance is not really something autoconf
puts high priority on ...

> So my questions are:
> 1) upon not finding libs such as xml, xslt or the msgfmt binary, is the
> configure script set (perhaps by bug) to report that libreadline is not
> installed (when in fact it is?).

No, I doubt that.

> Another way of phrasing this question: does
> configure misreport that which is missing?

And that too. You'd need to show some actual evidence, and not a highly
abstracted summary, to convince anyone that you've found a configure
bug.

regards, tom lane