https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110635
Arsen Arsenović <arsen at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Last reconfirmed| |2024-07-15
Ever confirmed|0 |1
CC| |arsen at gcc dot gnu.org
Status|UNCONFIRMED |NEW
Keywords| |ice-on-valid-code
--- Comment #1 from Arsen Arsenović <arsen at gcc dot gnu.org> ---
confirmed, I've debugged it a bit also, in the following (coroutines.cc:4224
vicinity):
if (initial_await != error_mark_node)
{
/* Build a compound expression that sets the
initial-await-resume-called variable true and then calls the
initial suspend expression await resume.
In the case that the user decides to make the initial await
await_resume() return a value, we need to discard it and, it is
a reference type, look past the indirection. */
if (INDIRECT_REF_P (initial_await))
initial_await = TREE_OPERAND (initial_await, 0);
tree vec = TREE_OPERAND (initial_await, 3);
tree aw_r = TREE_VEC_ELT (vec, 2);
aw_r = convert_to_void (aw_r, ICV_STATEMENT, tf_warning_or_error);
tree update = build2 (MODIFY_EXPR, boolean_type_node, i_a_r_c,
boolean_true_node);
aw_r = cp_build_compound_expr (update, aw_r, tf_warning_or_error);
TREE_VEC_ELT (vec, 2) = aw_r;
}
... the code does not anticipate that initial_await is a TARGET_EXPR, and so it
fails because it tries to alter a non-existent element 2 (which is meant to
correspond to the await_resume of the initial_suspend_awaiter)
