[PATCH] fix KJS crash on some more strict platforms

[Helge Deller]

On Tuesday 16 December 2003 23:13, Shaheed wrote:
> On Tuesday 16 December 2003 21:08, Helge Deller wrote:
> > Some platforms are pretty strict about the alignment for some types
> > of variables.
> >
> > Code like this (taken from kdelibs/kjs/internal.cpp):
> > const unsigned char NaN_Bytes[] = { 0x7f, 0xf8, 0, 0, 0, 0, 0, 0 };
> > ....
> > double x = * (double*) NaN_Bytes;
> 
> Isn't this also broken in an endianness sense? 

Yes and No - in the real code the endianess is taken into account.
Take a look at the patch I attached (or the original source), the 
#ifdef BIGENDIAN... preprocessor switches are there.

> Hmmm. I'm not an expert, but  according to:
> 
> http://research.microsoft.com/~hollasch/cgindex/coding/ieeefloat.html
> 
> I would suggest something like this (not tested):
> 
> const Q_INT64 qNaN = 0x7ff8000000000001LL; // Quiet NaN
> const Q_INT64 sNaN = 0x7ff0000000000001LL; // SignallingNaN

I didn't wanted to recode the code. Of course using Q_INT64 is cleaner,
but I think KJS is generic code which should not be bound to Qt and is
used for Safari too. But I'm no maintainer or expert of this code....

Helge