[Bug middle-end/78501] [7 regression] SEGV in vrp_val_max

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78501

--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #3)
> (In reply to prathamesh3492 from comment #2)
> > Hi,
> > From the trace, it seems ptrdiff_type_node is set to NULL for ada ?
> > Either we can guard setting range for CFN_BUILTIN_STRLEN by checking
> > ptrtdiff_type_node is non null, or make vrp_val_max return NULL if type is
> > NULL.
> > I am trying to reproduce on x86_64-unknown-linux-gnu.
> > 
> > Thanks,
> > Prathamesh
> 
> Looks like gimple-ssa-sprintf.c also unconditionally uses it.
> 
> I suppose we should move its initialization to build_common_tree_nodes,
> it's built from the PTRDIFF_TYPE target macro.  See how size_type_node is
> built.
> Also move unsigned_ptrdiff_type_node.

The way gimple-ssa-sprintf.c uses it looks just wrong to me.
    case FMT_LEN_z:
      dirtype = sign ? ptrdiff_type_node : size_type_node;
      break;

    case FMT_LEN_t:
      dirtype = sign ? ptrdiff_type_node : size_type_node;
      break;

From what I can see in POSIX, it should be
    case FMT_LEN_z:
      dirtype = sign ? signed_type_for (size_type_node) : size_type_node;
      break;

    case FMT_LEN_t:
      dirtype = sign ? ptrdiff_type_node : unsigned_type_for
(ptrdiff_type_node);
      break;