RE: [xml] XPath question w/ libxml2

[Mark_Vakoc peoplesoft com]

First of all, I want a better compiler....
> > I'm afraid you might be seeing some weird floating point troubles.

You have no idea...

> > Can you try to see if the basic tests from test/XPath/expr/floats
> > work correctly first ? Then it might be number('NaN') failing or
> > the index tests.

Ok, I traced things down to xmlXPathEvaluatePredicateResult.
xmlXPathNAN is being initialized properly.  However with this compiler
comparing any value to xmlXPathNAN returns true.  Here's a sample
program to show the strangeness


#include <libxml/xpath.h>
#include <libxml/xpathInternals.h>
#include <stdio.h>

void main() {
        xmlXPathObjectPtr obj1,obj2;
        xmlInitParser();

        obj1 = xmlXPathNewString("NaN");
        xmlXPathDebugDumpObject(stdout,obj1,1);
        obj1 = xmlXPathConvertNumber(obj1);
        xmlXPathDebugDumpObject(stdout,obj1,1);

        obj2 = xmlXPathNewFloat(1.0);
        xmlXPathDebugDumpObject(stdout,obj2,1);

        fprintf(stdout,"obj1(%f) is equal to obj2(%f) : %d",
                        obj1->floatval, obj2->floatval, 
                   obj1->floatval == obj2->floatval);
        xmlCleanupParser();
}

And the output:

C:\>test
  Object is a string : NaN
  Object is a number : NaN
  Object is a number : 1
obj1(1.#QNAN0) is equal to obj2(1.000000) : 1
C:\>

Seems bizarre to me.  The attached patch fixes this specific problem
with msvc, but I wonder where else this might bite.

Attachment: xpath.diff
Description: Binary data