PPC ICE

[Steve Kargl]

On Wed, Aug 19, 2009 at 10:30:41AM +0200, Janus Weil wrote:
> 
> >> > troutmask:sgk[206] gfc4x -c -std=f95 r.f90
> >> > r.f90:4.38:
> >> >
> >> > Â  Â  procedure (), pointer, nopass :: f
> >> > Â  Â  Â  Â  Â  Â  Â  Â  Â  Â  Â  Â  Â  Â  Â  Â  Â  Â  Â  1
> >> > Error: Fortran 2003: Procedure pointer at (1)
> >> > f951: internal compiler error: Segmentation fault: 11
> 
> thanks for reporting this.
> 
> 
> > The ICE goes away with the following patch.
> >
> > 2009-08-18  Steven G. Kargl  <kargl@gcc.gnu.org>
> >
> > Â  Â  Â  Â * resolve.c (resolve_fl_derived): Â Don't dereference a NULL pointer.
> >
> > Index: resolve.c
> > ===================================================================
> > --- resolve.c   (revision 150866)
> > +++ resolve.c   (working copy)
> > @@ -9494,7 +9494,8 @@ resolve_fl_derived (gfc_symbol *sym)
> > Â  Â  Â  Â }
> >
> > Â  Â  Â  /* Procedure pointer components: Check PASS arg. Â */
> > - Â  Â  Â if (c->attr.proc_pointer && !c->tb->nopass && c->tb->pass_arg_num == 0)
> > + Â  Â  Â if (c->attr.proc_pointer
> > + Â  Â  Â  Â  && (c->tb && !c->tb->nopass && c->tb->pass_arg_num == 0))
> > Â  Â  Â  Â {
> > Â  Â  Â  Â  Â gfc_symbol* me_arg;
> 
> 
> In principle the patch is ok. An alternative solution would be to
> check for -std=... earlier (before the PPC is even added to the
> derived type, so that we don't have to resolve it later on). I'd
> propose the following:
> 
> Index: decl.c
> ===================================================================
> --- decl.c	(revision 150888)
> +++ decl.c	(working copy)
> @@ -4448,6 +4448,10 @@ match_ppc_decl (void)
>        return MATCH_ERROR;
>      }
> 
> +  if (gfc_notify_std (GFC_STD_F2003, "Fortran 2003: Procedure pointer
> component"
> +		      " at %C") == FAILURE)
> +    return MATCH_ERROR;
> +
>    /* Match PPC names.  */
>    ts = current_ts;
>    for(num=1;;num++)
> 
> We do have such a check already (for all procedure pointers), but it
> comes too late in the case at hand.
> 
> Steve, please feel free to commit whatever version you like best
> (together with the dejagnuified test case).
> 

I'll use your patch, degagnu the testcase, and commit on Saturday.
Thanks.

-- 
Steve