Re: FYI: PR 15713 - interpret.cc aliasing

[Bryce McKinlay <mckinlay at redhat dot com>]

Ulrich Weigand wrote:

> Bryce McKinley wrote:
>
>  
>
> >       PR libgcj/15713
> >       * include/jvm.h (_Jv_value): New union type.
> >       * gcj/field.h (_Jv_Field): Add new _addr union field variants 
> >       * interperet.cc (run): Use _Jv_value union type and *_addr _Jv_Field 
> >       union members.
> >    
>
> This breaks s390(x) and presumably all other big-endian platforms, 
> because this:
>
>  
>
> > -                 jbyte value = (*(jint*)&rvalue) & 0xff;
> > -                 PUSHI (value);
> >    
>
> is not at all the same as this:
>
>  
>
> > +               PUSHI (rvalue.byte_value);
> >    
>
> except by chance on little-endian platforms.
>  

Yeah - I find it a rather annoying feature of libffi that 
smaller-than-int return values are promoted to int. Its not the first 
time I've been tripped up by this ;-)

> The patch below fixes this (and two other similiarly wrong changes).
> Tested on s390-ibm-linux and s390x-ibm-linux, fixes a bunch of gij failures.
>
> OK?
>  
Yes, thanks for fixing this...

> Bye,
> Ulrich
>
>
> ChangeLog:
>
> 	* interpret.cc (run): Fix bug introduced by last change.
>  

... however this ChangeLog could be a little more descriptive?

Thanks

Bryce