As promised I downloaded 0.92.1 and tested with UnixWare's cc instead of
gcc. Things actually went mostly well after I hit a snag. I was getting
symbol undefined errors for _Xconst in various libXm files. I went
looking and found that _Xconst is defined in <X11/Xfuncproto.h> as follows:
------------------
#ifndef NeedFunctionPrototypes
#if defined(FUNCPROTO) || __STDC__ || defined(__cplusplus) || defined(c_plusplus)
#define NeedFunctionPrototypes 1
#else
#define NeedFunctionPrototypes 0
#endif
#endif /* NeedFunctionPrototypes */
...
#if NeedFunctionPrototypes
#ifndef _Xconst
#if __STDC__ || defined(__cplusplus) || defined(c_plusplus) || (FUNCPROTO&4)
#define _Xconst const
#else
#define _Xconst
#endif
#endif /* _Xconst */
...
------------------
I found that NeedFunctionPrototypes was not being set to 1, therefore
the _Xconst definition wasn't being exposed. I looked further and found
that UnixWare's cc compiler defines __STDC__, but sets its value to 0.
It's set to 1 only if cc -Xc option is used. This is in contrast to gcc
where the default has __STDC__ defined to be 1.
I also checked the <X11/Xfuncproto.h> on Linux and found the code is like
this:
------------------
#ifndef NeedFunctionPrototypes
#if defined(FUNCPROTO) || defined(__STDC__) || defined(__cplusplus) ||
defined(c_plusplus)
^^^^^^^^^^^^^^^^^ note the difference
#define NeedFunctionPrototypes 1
...
------------------
Argh. That would have worked. I don't know why USL/SCO made this change.
But more basically, I think this may be a X11 base problem. Why should
_Xconst be defined only when NeedFunctionPrototypes is set to 1? I would
think _Xconst should be defined to the appropriate thing under all conditions.
At any rate, I found that if I use cc -Xc I get too many compile warnings,
because the compiler enforces strict ANSI and there is a lot of code in
various places that seems to violate that. I finally decided to add
-DFUNCPROTO instead, and that made everything compile fine. I don't know
whether this is something lesstif's configure script should deal with.
Maybe it should be in the README or release notes for platform-specific stuff.
The one last snag is when linking lesstif's mwm, both the 1.2 and 2.0
version of mwm exhibit this problem:
-------------------
cc -O -DFUNCPROTO -I/usr/X/include -o mwm colormaps.o cursors.o decorate.o defaults.o
desktop.o events.o functions.o icons.o menus.o misc.o move.o mwm.o mwmerr.o mwmparse.o
mwmlex.o pager.o pan.o props.o resize.o resources.o screens.o windows.o
../../../lib/Xm/.libs/libXm.a -L/usr/X/lib -lXt -lXext -lX11 -lsocket
Undefined first referenced
symbol in file
_XtWaitForSomething ../../../lib/Xm/.libs/libXm.a(Vendor.o)
UX:ld: ERROR: mwm: fatal error: Symbol referencing errors. No output written to mwm
-------------------
Hmm. Deja vu. This time, though, I am trying to build lesstif based on
UnixWare's X11R5 headers and libraries. Did someone reverse the logic
since 0.92.0? Because it worked before, after I corrected my mistake,
-Ti (author of xmcd, xmmix)
--
/// Ti Kan Vorsprung durch Technik
/// AMB Research Laboratories, Sunnyvale, CA. USA
/// [EMAIL PROTECTED]
////// http://www.ibiblio.org/tkan/
///
