[Bug c++/79104] [DR 696] wrong semantics for odr-use of constant variable from lambda

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79104

Jason Merrill <jason at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P1                          |P3
             Status|ASSIGNED                    |NEW
            Summary|[7 Regression] ambiguity    |[DR 696] wrong semantics
                   |calling std::begin on a     |for odr-use of constant
                   |local constexpr array of    |variable from lambda
                   |structs                     |

--- Comment #4 from Jason Merrill <jason at gcc dot gnu.org> ---
The handling of this testcase went from wrong-code to rejects-valid, so I don't
think it qualifies as a regression.  The issue here is that we don't implement
DR 696 properly; we should capture a, but instead replace it with its initial
value.  Given that, r240819 changed this to be correctly rejected; a temporary
should not bind to a non-const reference.

Here's a simpler testcase to illustrate our non-conformance on DR 696:

void f ()
{
  const int i = 42;

  [&]() {
    &i;  // OK, captures i
  };
}