forming reference to nullptr is not rejected in constexpr

[ldalessa]

Bugzilla Link	49321
Version	trunk
OS	All
CC	@randomnetcat,@zygoloid

Extended Description

I think this UB should be rejected.

constexpr bool foo(int* i) {
    int& j = *i;
    return true;
}

static_assert(foo(nullptr));

https://godbolt.org/z/cKo3jf

[zygoloid]

Similarly we should reject:

int arr[3];
constexpr bool f() {
  int &r = arr[3];
  return true;
}
static_assert(f(), "");

[Endilll]

Both examples are still accepted, but we issue -Warray-bounds diagnostic in Richard's example: https://godbolt.org/z/5ocT47qMG

[llvmbot]

@llvm/issue-subscribers-clang-frontend

Author: Luke D'Alessandro (ldalessa)

| | | | --- | --- | | Bugzilla Link | [49321](https://llvm.org/bz49321) | | Version | trunk | | OS | All | | CC | @randomnetcat,@zygoloid |

Extended Description

I think this UB should be rejected.

constexpr bool foo(int* i) {
    int& j = *i;
    return true;
}

static_assert(foo(nullptr));

https://godbolt.org/z/cKo3jf

[cor3ntin]

@zygoloid Probably stupid question, what wording makes that ill-formed?
I can't find anything here https://eel.is/c++draft/dcl.init.ref

[Endilll]

I can't find anything here https://eel.is/c++draft/dcl.init.ref

Since Tokyo we have a note in [dcl.ref] about this:

Because a null pointer value or a pointer past the end of an object does not point to an object, a reference in a well-defined program cannot refer to such things; see 7.6.2.2 [expr.unary.op].

Before resolution of CWG453 was adopted in Tokyo, for the longest time we had the following normative wording in [dcl.ref]/5:

A reference shall be initialized to refer to a valid object or function.