Re: [Geomview-devel] [geomview-users] x86_64 build crash

[Lloyd W. <L....@su...>]

At Saturday 24/06/2006 22:16 +0200, Claus-Justus Heine wrote:
>There are indeed some bugs in the build-system; some dependencies 
>are wrong; also: the
>wa.yacc.c stuff is slightly out of date and will not work with 
>recent flex/bison
>combinations. Also, geomview.texi needed some polishing to work with 
>recent texinfo
>releases. Also, the building of the documentation needs the 
>psfree.fig TeX style-file,
>which has a licence incompatible with the GPL and so cannot be 
>packed with Geomview.
>
>I think I have fixed that stuff. BTW, writing Makefiles etc., that 
>is not black magic. If
>I can do that, then others can do it, too.

my comment was only that reconf was black magic (because it was 
unique to Geomview yet undocumented within Geomview -- and leads to 
somewhat unpredictable output!).
I've attempted to improve the documentation files in the top level of 
CVS to discuss reconf (both pre- and post- your recent rewriting of reconf.)


>However, I have prepared a new Geomview "distro". I have labeled it
>geomview-1.8.2-beta
>The tar-achives can be found at
>
>http://www.mathematik.uni-freiburg.de/IAM/homepages/claus/geomview-1.8.2-beta.tar.gz
>http://www.mathematik.uni-freiburg.de/IAM/homepages/claus/geomview-1.8.2-beta.tar.bz2

This can indeed be compiled, at least on Cygwin -- many thanks for 
that. (I had to install a Cygwin-base snapshot to get around 
src/bin/animate/glob.c presuming d_ino is always supported, when it's 
optional under POSIX. I've updated
http://www.ee.surrey.ac.uk/Personal/L.Wood/software/SaVi/building-under-Windows/
with instructions for installing cygwin snapshots. Also, compilation 
success is only when not installing xforms first. The xforms make 
process seems to generate many errors.)

However, when using this with any verison of SaVi, this Geomview 
1.8.2 beta throws a problem with drawing a simple sphere; SaVi still 
works, but the user can't see the Earth sphere that is central to 
SaVi. The error comes from SphereCreate in Geomview's 
src/lib/gprim/sphere/spherecreate.c (which uses va_args in 
interesting ways, I see...)

This downwards trends in drawing simple spheres concerns me -- in 
1.8.1 drawing simple spheres and texturemapping worked fine, in the 
1.8.2 "alpha", drawing simple spheres works but turning on 
texturemapping leads to an invisible texturemapped sphere, so I'd 
have to use 1.8.1 to get screenshots of texturemapping, and now with 
this 1.8.2 beta there's no sphere at all, even without 
texturemapping! (You can turn off 'Central Body' in SaVi's Rendering 
menu to view a ghost continents outline I deliberately included to 
support spherical/hyperbolic space views, but that's all the Earth 
you'll see. No spheres, no texturemaps.) I haven't tried the Orrery, 
but imagine it will be similiarly affected.

I should add that compiling this beta is only successful is when first doing
configure -with-motif=/usr/X11R6 -with-opengl=/usr/X11R6

I attempted to see if this was an opengl-only problem by make clean, then
configure -with-motif=/usr/X11R6
then make. This produced an initial error:
make[5]: Entering directory `/home/lloyd/geomview-1.8.2-beta/src/lib/mg/opengl'
if gcc -DHAVE_CONFIG_H -I. -I. -I../../../.. 
-I/usr/local/Geomview  -I../../../.
./include     -g -O2 -MT mgopengl.o -MD -MP -MF ".deps/mgopengl.Tpo" 
-c -o mgope
ngl.o mgopengl.c; \
then mv -f ".deps/mgopengl.Tpo" ".deps/mgopengl.Po"; else rm -f 
".deps/mgopengl.
Tpo"; exit 1; fi
In file included from mgopengl.c:32:
mgopenglP.h:33:21: GL/glx.h: No such file or directory
In file included from mgopengl.c:32:
mgopenglP.h:82: error: parse error before "GLXContext"
[..]

suggesting that this Geomview beta can't actually be built without OpenGL.

now, here's that Sphere error:

$ cd /home/lloyd/geomview-1.8.2-beta
lwood@lwood-wxp01 /home/lloyd/geomview-1.8.2-beta
$ ./geomview -run ../savi-dev/savi
[2] 4864
lwood@lwood-wxp01 /home/lloyd/geomview-1.8.2-beta
$ SphereCreate: Undefined option: 2279636
"read geometry" in "../savi-dev/oogl/earth_vect_h.oogl": error 
reading geometry's:
 >       appearance {shading smooth material {diffuse 0.2 0.4 1} }
 >       SPHERE 1.0
--------------^
 >       0.0 0.0 0.0

(the linefeed in the above doesn't matter, and the same error occurs 
without it -- oogl is freeform.) Fiddling with SaVi's oogl doesn't help, e.g.

$ SphereCreate: Undefined option: 2279636
"read geometry" in "../savi-dev/oogl/earth_vect_h.oogl": error 
reading geometry's:
 >
 > (read geometry { define unit_sphere_h {
 >       SPHERE 1 0 0 0
--------------^
 >       appearance {shading smooth material {diffuse 0.2 0.4 1} }


and the pipe problems are still present with large LIST objects on 
cygwin, of course:

SaVi: loaded data/iridium-66.tcl
       changed coverage angle from 0.0 to 8.2.
Error reading "[1]../savi-dev/savi ": can't seek back far enough (on pipe?):
 > ... transform { 0.340785 0 0 0 0 0.340785 0 0 0 0 0.182152 0 0 0 0 
1 } geom: cone_h } } )
-----^
"read geometry" in "[1]../savi-dev/savi ": error reading geometry's:
 > ... transform { 0.340785 0 0 0 0 0.340785 0 0 0 0 0.182152 0 0 0 0 
1 } geom: cone_h } } )
-----^
gv_ready: error, got char: (-1)
crashing SaVi as before.

This sphere drawing problem does not appear to my eyes to be a 
configure/make-related problem, so I've skipped including the logs of 
configure/make here. Requiring opengl does, however, appear to be a 
make/configure problem, and more...

so, not ready for prime time, I think.

thanks,

L.

oh well, back to 1.8.1 to admire my dynamic texturemapping in SaVi. 
http://savi.sf.net/


>  The reason
>is that I have included the documentation, pdf, info and html, into 
>the tar-files. Don't
>bother me if you cannot access those files, if so, then some server 
>is down. I have
>checked the links for typos and tried that they work and are 
>accessible from the outside.
>
>I was using the following programs to build the stuff:
>
>gcc-4.1.1
>autoconf-2.59
>automake-1.9.6
>texinfo-4.8 (makeinfo stuff)
>bison-2.1
>flex-2.5.33
>pdftex, pstoepsi, epstopdf, convert
>openmotif-2.2.3, a recent xorg-x11 system, xforms-1.0.90
>
>For the compilation only the following _should_ be needed:
>
>- C-compiler/binutils
>- X-libraries, Motif-libs, xforms-libs (optional)
>
>In particular, the autoconf stuff should not be needed, bison/flex 
>should not be needed,
>"make" should not try to regenerate the documentation.
>
>The tar-balls were created with "make dist", I was able to run "make 
>distcheck"
>successfully (this means that the tarballs could be successfully 
>extracted, it was
>possible to do a VPATH build with the extracted sources, it was 
>possible to run "make
>dist" again, the combination of "make install" and "make uninstall" 
>did not leave any
>traces on the system (with the exception of empty directories), 
>"make distclean" did not
>leave any files in the build-directory).
>
>I was able to compile and run this version (i.e. the sources 
>extracted from the tarballs)
>of Geomview on the following systems:
>
>- Desktop AMD X86_64, Linux 2.6.16, Gentoo distro
>- Laptop Intel Pentium M, Linux 2.6.16.1, Gentoo distro
>- Desktop Intel P4 (at work),
>
>Please test the stuff; if it works -- basically -- then we should 
>label it 1.8.2 and move
>it to the sourceforge download area. NOTE: I did not change the 
>source code, just some
>small changes to the build-system, i.e. the wa.yacc.c stuff should 
>be resolved etc. Most
>of my "efforts" went into "porting" the TeXInfo documentation such 
>that recent texinfo
>release can eat it.
>
>In comparision with the "alpha" snapshot: the most remarkable 
>difference _should_ be that
>it is possible to use this beast, at least on Linux (X86 and 
>X86_64). At least I am able
>to use it there ;)
>
>Two year ago or so I have change the oogl2vrml program to output 
>VRML2, otherwise there
>are no new features compared to the "alpha" tar-ball.
>
>If you test it and feel inclined to report bugs:
>
>POINT ZERO: FIRST TRY TO FIX THE STUFF YOURSELF. If that should not 
>work out after
>spending a day or so:
>
>- include the output of "configure"
>- include "config.log"
>- include all the output of the "make command" you used to compile the stuff
>- include information about the program versions used for building 
>the stuff (i.e.
>   gcc --version, autoconf --version etc.)
>- COMPRESS the stuff before sending.
>
>Claus
>
>
>
>
>
>--
>   Claus-Justus Heine <ch...@do...>
>
>   Ftape - the Linux Floppy Tape Project           http://ftape.dot-heine.de/
>   NFS Swapping for 
> Linux                          http://nfs-swap.dot/heine.de/

<http://www.ee.surrey.ac.uk/Personal/L.Wood><L....@su...>