[Bug c++/106775] [10/11/12/13 regression] Subscipt operator with array rvalue inside brackets returns lvalue

["jakub at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106775

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jason at gcc dot gnu.org

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
So, perhaps either we need to in the
556           if (swapped
557               && flag_strong_eval_order == 2
558               && (TREE_SIDE_EFFECTS (array_expr) || TREE_SIDE_EFFECTS
(index_exp)))
559             expr = build_array_ref (input_location, index_exp, array_expr);
560           else
561             expr = build_array_ref (input_location, array_expr, index_exp);
spot if array_expr has ARRAY_TYPE and is xvalue rather than lvalue
choose some different representation which will ensure the side-effects are
evaluated in the right order, perhaps by using save_expr around index_exp and
using COMPOUND_EXPR for that save_expr and array_expr, or need to adjust
lvalue_kind so that it is able to recognize even what we lower the array ref to
in the line 559 case as array ref with ARRAY_TYPE xvalue base.  We actually
lower it as *(x + y) instead of x[y] with ensuring the side-effects being
evaluated in the right order.