patch for PR/23522

[Ian Lance Taylor]

Alexey Starovoytov <alexey.starovoytov@sun.com> writes:

> On Wed, 28 Sep 2005, Ian Lance Taylor wrote:
> 
> > Alexey Starovoytov <alexey.starovoytov@sun.com> writes:
> >
> > > > What if we change the condition to
> > > >     if (TREE_CODE (arg1_unw) != INTEGER_CST
> > > >         || !int_fits_type_p (arg1_unw, shorter_type))
> > > >       return NULL_TREE;
> > > >
> > > > Does that fix the problem?
> > >
> > > Yes.
> > > and going even further the 'if' should probably look like:
> > >
> > >   if (TREE_CODE (arg1_unw) != INTEGER_CST
> > >       || TREE_CODE (shorter_type) != INTEGER_TYPE
> > >       || !int_fits_type_p (arg1_unw, shorter_type))
> > >     return NULL_TREE;
> > >
> > > Certianly solves the problem and more safe.
> >
> > I think you can reasonably use INTEGRAL_TYPE_P (shorter_type) instead
> > of TREE_CODE (shorter_type) != INTEGER_TYPE.
> 
> Then we should probably change the couple lines above that spot
> to use INTEGRAL_TYPE_P as well.

Oh, sorry, you're right.  != INTEGER_TYPE is correct.  Forget what I
said.  Please test your patch above:

   if (TREE_CODE (arg1_unw) != INTEGER_CST
       || TREE_CODE (shorter_type) != INTEGER_TYPE
       || !int_fits_type_p (arg1_unw, shorter_type))
     return NULL_TREE;

Thanks!

Ian