In attempting to port the Persistence of Vision 1.0 ray tracing
package, I encountered one problem with Linux (MCC-Intermin 0.96c
with gcc 2.2.2).  There appears to be no support for the matherr()
function that is normally part of math library libm.a.  There was
at least a stub function (matherr.o) in the libm.a library of all 
the Unix systems I tried (SunOS 4.1.1, HP-UX 8.05, and even Apollo 
DomainOS 10.3).  Try a 'man matherr' for detailed information on 
how to use matherr().

This user supplied function is used to trap floating point exceptions. 
Also all these other systems contain lines like the following in
<math.h>:

    struct exception {
        int type;
            char *name;
            double arg1;
            double arg2;
            double retval;
    };

    #define DOMAIN      1      /* argument domain error        */
    #define     SING        2      /* argument singularity         */
    #define     OVERFLOW    3      /* overflow range error         */
    #define     UNDERFLOW   4      /* underflow range error        */
    #define     TLOSS       5      /* total loss of significance   */
    #define     PLOSS       6      /* partial loss of significance */

By adding these lines to <math.h> I could get the program to 
compile, but the executable gets floating point exception errors
and core dumps on many of the provided input files.  Others input
files will run to completion without problems.  I even tried 
changing all references of 'exception' to 'libm_exception' since
I had seen this used by gcc on some ports, but the results were
the same - floating point exceptions still occur.

Here's an example of using matherr() that I extracted from the
man page:

    #include <math.h>

    int
    matherr(x)
    register struct exception *x;
    {
        switch (x->type) {
        case DOMAIN:
            /* change sqrt to return sqrt(-arg1), not 0 */
            if (!strcmp(x->name, "sqrt")) {
                x->retval = sqrt(-x->arg1);
                return (0); /* print message and set errno */
            }
        case SING:
            /* all other domain or sing errors, print message, abort */
            fprintf(stderr, "domain error in %s\n", x->name);
            abort( );
        case PLOSS:
            /* print detailed error message */
            fprintf(stderr, "loss of significance in %s(%g) = %g\n",
                x->name, x->arg1, x->retval);
            return (1); /* take no other action */
        }
        return (0); /* all other errors, execute default procedure */
    }

Performance of Persistence of Vision 1.0 under Linux is very good (at
least I'm not complaining).  I have a 25MHz 386DX with 8MB of RAM and a
Cyrix 387.  The program compiled under Linux 0.96c with gcc 2.2.2 runs
about 1.8 times faster than the DOS executable (32-bit 386 protected 
mode) compiled with Intel's Code Builder.  The only problem is I can't
do all the ray traces under Linux because many of the input files 
will abort with floating point exceptions.

By the way Persistence of Vision 1.0 is available from:
        alfred.ccs.carleton.ca (134.117.1.1) in pub/pov-ray/POV-Ray1.0.
The following files are for Unix systems:
        povdoc.tar.Z      Documentation
        povscn.tar.Z      Scene data files
        povsrc.tar.Z      Source code

Is support for matherr() provided in newer versions of gcc for Linux?
I see there is a 2.2.2d out now which I don't have yet.  If not, could
it be added without to much trouble?  In the interim, does anyone know
of a reasonable work around?  Thanks.





-- 
Steve Snyder           
snyde_sl@otsc.eds.com  
uunet!tsca02!snyde_sl