RE: [xml] Test comparison fails on value (-2147483648)

[Chris_Poblete Dell com]

> -----Original Message-----
> From: Daniel Veillard [mailto:veillard redhat com]
> Sent: Sunday, August 26, 2001 3:54 PM
> To: Chris_Poblete exchange dell com
> Cc: xml gnome org
> Subject: Re: [xml] Test comparison fails on value (-2147483648)
>
>
> On Sun, Aug 26, 2001 at 12:37:55PM -0500, 
> Chris_Poblete Dell com wrote:
> > VERSIONS: libxml2-2.4.3
> >           libxslt-1.0.3
> >
> > PROBLEM:
> > The tag <xsl:when test="OutputWatts &lt; 0"> fails to 
> evaluate correctly 
> > when OutputWatts is equal to -2147483648.  


Consider the ff. for the tag <xsl:when test="OutputWatts &lt; 0">
     OutputWatts            Evaluation
        -21474836                pass
        -214748364               pass
        -2147483648             (FAIL)
        -21474836489             pass
        -2147483648999           pass
        -2147483648999999999999  pass
The above values of OutputWatts are all negative numbers and
no matter how many digits there are, it fails at -2147483648
since an int is used temporary buffer.


>   Well OutputWatts will evaluate to a nodeset which then once
> used in the comparison will end up being casted to a float. 
> OutputWatts
> cannot be equal to -2147483648 but it's float value may be 
> equal to this
>  value.
>   I hope you understand that this compare a set of XML nodes 
> with a value.
>
> > PROPOSED RESOLUTION:
> > xpath:xmlXPathStringEvalNumber(6567)
> > ------------------------------------------------------------
> > <   ret = (double) tmp;
> > ----------
> > >   ret = (double) tmp;
> > >   // Variable tmp is declared as signed integer which may 
> evaluate to
> > a negative
> > >   // number.  We only need to calculate the absolute value of the
> > whole number since
> > >   // any negative character in the input string has been flagged
> > separately.
> > >   if (tmp < 0)
> > >           ret = -ret;
> > ------------------------------------------------------------
>
>   This looks wrong to me. The sign is caught when beginning to parse
> the string:
>
>     if (*cur == '-') {
>       isneg = 1;
>       cur++;
>     }
>
>   Then before returning the value
>     if (isneg) ret = -ret;
>
>  If we apply you suggested change we change the sign twice in 
> all cases,
> looks wrong to me.
>  The problem you're facing is an integer overflow. If you look at the
> comment 5 lines before your suggested change you will get the 
> explanation:
>
>     /*
>      * tmp is a workaroudn against a gcc compiler bug
>      */
>
> I spent a number of hours chasing it down and finding a 
> workaround, your


Is there a reason why the function "atof" or "strtod" is not used instead?
These functions are supported in Linux, Windows, and Netware.
Sample code is:
{
  double retval;
  retval = strtod(str, (char**)NULL);
  if (retval == HUGE_VAL)
    return xmlXPathNAN;
  else
    return retval;
}


> problem case don't doesn't seems realistic and I would prefer 
> not change
> that part of the code.
>
> Daniel
>
> -- 
> Daniel Veillard      | Red Hat Network 
> http://redhat.com/products/network/
> veillard redhat com  | libxml Gnome XML XSLT toolkit