[clang] Accepts-invalid (?) in discarded statement with return type deduction

[efriedma-quic]

Consider:

auto f() {
    if constexpr (false) return;
    return 3;
}

Missing error on the first return statement.

I think this violates [stmt.return]p2? Or maybe it's an oversight in the wording.

[llvmbot]

@llvm/issue-subscribers-clang-frontend

Author: Eli Friedman (efriedma-quic)

Details

Consider:

auto f() {
    if constexpr (false) return;
    return 3;
}

Missing error on the first return statement.

I think this violates [stmt.return]p2? Or maybe it's an oversight in the wording.

[zygoloid]

I think that's supposed to be valid, but the wording in [stmt.return] doesn't seem to take if constexpr into account. I think the intent is that each occurence of "shall be used only in" in [stmt.return]/2 means "shall appear as a non-discarded statement only in". For example, [dcl.type.auto.deduct]/2.1.1 wouldn't make much sense if return; as a discarded statement were ill-formed if the return type gets deduced as something else.

GCC and MSVC also accept. EDG rejects because it deduces the return type as both void and int, but that's clearly in violation of [decl.spec.auto]/10 so I don't think we should take that as guidance.

Filed cplusplus/CWG#737 for the presumed wording issue.