Re: [3.1] Re: ICE in do_SUBST when compiling

[Franz Sirl <Franz dot Sirl-kernel at lauterbach dot com>]

On Tuesday 30 April 2002 19:35, Jan Hubicka wrote:
> > Hi,
> >
> > We have an ICE in do_SUBST on x86_64, even with the current CVS of gcc
> > 3.1.
> >
> > If compiled with gcc -O2 foo.i we get:
> > /suse/meissner/foo.i: In function `g':
> > /suse/meissner/foo.i:5: Internal compiler error in do_SUBST, at
> > combine.c:439
> >
> > I have attached foo.i since it is very short.
> >
> > Ciao, Marcus
> >
> > unsigned short
> > g (unsigned short x)
> > {
> >   return (x < 32767) ? x : 32767;
> > }
>
> Hi,
> this patch is backported form 3.2.  I've bootstrapped it, but I am not sure
> whether it can go to 3.1.  It is fairly obvious, but I am not aware of
> other target where the bug reproduce.  Mark?
>
> I've had approval from Richard for it, but due to changes in plan for
> diploma I missed my chance to put it on the branch.

Just a comment here on how we handled the 
GEN_INT/trunc_int_for_mode->gen_int_mode conversion between mainline and 
branch. Could we have some kind of policy for the branch that would allow us 
to add the gen_int_mode function to the branch without using it initially 
(not converting existing code), but allowing us to use it if we apply a 
bugfix to both mainline and branch?

These minimal differences between mainline and branch look really silly to me 
and add additional complexity to backports of patches.

Franz.