[Bug c++/92106] [8/9/10 Regression] ICE with structured bindings and -Wreturn-local-addr

[mpolacek at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92106

--- Comment #4 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #3)
> (In reply to Marek Polacek from comment #2)
> >  9351           /* When returning address of a structured binding, if the
> > structured
> >  9352              binding is not a reference, continue normally, if it is a
> >  9353              reference, recurse on the initializer of the structured
> >  9354              binding.  */
> >  9355           tree base = DECL_DECOMP_BASE (whats_returned);
> >  9356           if (TYPE_REF_P (TREE_TYPE (base)))
> >  9357             {
> >  9358               tree init = DECL_INITIAL (base);
> >  9359               return maybe_warn_about_returning_address_of_local
> > (init);
> >  9360             }
> > 
> > init is null here and cp_fold crashes on that.
> 
> I'd add
> && DECL_INITIAL (base)
> to line 9356.

I'm testing the below -- wanted to lose the { }s and avoid using DECL_INITIAL
twice.

--- a/gcc/cp/typeck.c
+++ b/gcc/cp/typeck.c
@@ -9353,11 +9353,10 @@ maybe_warn_about_returning_address_of_local (tree
retval)
         reference, recurse on the initializer of the structured
         binding.  */
      tree base = DECL_DECOMP_BASE (whats_returned);
-     if (TYPE_REF_P (TREE_TYPE (base)))
-       {
-         tree init = DECL_INITIAL (base);
-         return maybe_warn_about_returning_address_of_local (init);
-       }
+     tree init;
+     if (TYPE_REF_P (TREE_TYPE (base))
+         && (init = DECL_INITIAL (base)))
+       return maybe_warn_about_returning_address_of_local (init);
    }
       bool w = false;
       auto_diagnostic_group d;


> Maybe it would be better to add return nullptr;
> to find if the range for loop has no iterations.

Added -- it's not relevant to this bug.