latest CVS vs. Solaris 2.5.1

["Brandon S. Allbery" <allbery ece cmu edu>]

Here's the list of problems I've run in to with the latest anon-CVS build.  
It's a lot cleaner than the "release" version....

* gnome-libs still checks for gettext() in -lc before -lintl.  We use GNU 
gettext for everything, so we end up with the GNU includes from 
/usr/local/include and the Solaris libc libraries --- a losing combination.

* Various parts of GNOME (notably, libgtop) still don't comprehend that 
Solaris passes everything through a name service switch and therefore 
"-lresolv -lsocket -lnsl" is needed almost everywhere.  (gnome-libs' check for 
-lghttp also fails if -lresolv et al. aren't in $LIBS.  Or is this an instance 
of the next problem?)

* Some cases which look like the above problem are actually something more 
troublesome:  configuration-parameter scripts such as glib-config don't record 
the value of $LIBS in effect when they were configured and therefore don't 
include it when run --- but callers of these scripts often trust them to be 
complete specifications.  For example, gtk+ expects glib-config to return the 
full set of libraries required, but that set ends up not including the 
aforementioned -lsocket, et al., which I force using LIBS=, and also omits 
-lroken which I borrowed from KTH Kerberos to provide snprintf() and friends.  
The result is that I have to edit glib-config after regenerating glib or gtk+ 
will refuse to build because it can't link a simple glib program using only 
the information glib-config provides.  (Question:  should the configure script 
be adding $LIBS, or should glib-config have recorded it?)

* "checking for main in -lrx" succeeds.  But librx.a is the AFS RPC library on CMU systems.  It should check for symbols which are part of the GNU regex library instead of just blindly assuming that -lrx is *always* GNU regex....

* gnome-guile, as of Friday (haven't rebuilt it yet from this morning's update) still required explicit GNOME_INCLUDEDIR and GNOME_LIBDIR in order to find guile when the build-guile script doesn't exist.  This seems wrong; at minimum, shouldn't it check to see if it's already under $prefix?

* Also as of Friday, gnome-libs wanted flock().  Pardon?  Last I heard, portable file locking used fcntl().  (Flames about BSD vs. System V to /dev/null.  Solaris -lucb sucks regardless of which side you're on.)

I'm omitting the indent stuff because I see a FIXME in ORBit's configure.in about this already.  If it gets annoying enough I'll probably contribute a test for it.  I'll also leave out comments on gdm because it barely exists as yet....