https://gcc.gnu.org/bugzilla/show_bug.cgi?id=124456
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org,
| |ppalka at gcc dot gnu.org
--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
The problem is that when the structured bindings initializer is type dependent,
we really don't know how many elements or which ones the structured binding
pack has.
When cp_finish_decomp is called with a type dependent structured bindings,
DECL_VALUE_EXPR of each of the structured binding and the pack is set to
ARRAY_REF of the base and corresponding index. Only when cp_finish_decomp is
called on non-type dependent one it can actually figure everything up and then
DECL_VALUE_EXPR of the
structured binding pack is a TREE_VEC with the individual elements of the pack
etc.
I've tried:
--- gcc/cp/pt.cc.jj 2026-03-12 08:48:04.747569066 +0100
+++ gcc/cp/pt.cc 2026-03-12 09:48:43.698506293 +0100
@@ -14135,6 +14135,15 @@ tsubst_pack_expansion (tree t, tree args
return error_mark_node;
gcc_assert (DECL_HAS_VALUE_EXPR_P (orig_arg));
arg_pack = DECL_VALUE_EXPR (orig_arg);
+ if (TREE_CODE (arg_pack) == ARRAY_REF)
+ {
+ /* If the structured binding pack has type dependent
+ base, we can't expand it yet. */
+ tree base = TREE_OPERAND (arg_pack, 0);
+ gcc_assert (VAR_P (base)
+ && type_dependent_expression_p (base));
+ return t;
+ }
tree vec = make_tree_vec (TREE_VEC_LENGTH (arg_pack) - 2);
if (TREE_VEC_LENGTH (vec))
memcpy (TREE_VEC_BEGIN (vec), &TREE_VEC_ELT (arg_pack, 2),
and while it fixes the ICE on the above testcase, when I try to actually
instantiate the lambda, I get an ICE again.
So I'm afraid I'm stuck.
// PR c++/124456
struct A { int i; };
A
foo (auto...)
{
return A ();
}
void
bar (auto...)
{
}
auto
baz (auto... t)
{
return [=] (auto... a) { auto [... k] = foo (t..., a...); return bar (k...);
};
}
auto x = baz (1);
auto y = x (1, 2, 3);
