https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119624
--- Comment #3 from Joseph S. Myers <jsm28 at gcc dot gnu.org> --- This raises a question of whether the interpretation of the type of the controlling expression "as if it had undergone an lvalue conversion" (per the resolution of issue 0481) also engages the (translation-time) undefined behavior (constraint violation in C2Y after the adoption of N3481 in Graz) for such a conversion of an lvalue with incomplete type. I previously (reflector message 28292, 30 Jan 2025) noted a similar question with unselected generic associations. I would argue that the particular case here is invalid independent of those questions, because of the conditional expression: the lvalues in question are operands of ?:, which means they get converted to rvalues, and the fact that the conditional expression is not evaluated is irrelevant, because the lvalues are evaluated *relative to that expression*, much as if the conditional expression were contained inside if (0) - the controlling expression is subject to all normal rules on validity of expressions even though it's not evaluated at execution time. (The "Discarded" series of papers would say the controlling expression is "discarded relative to the generic selection". This avoids issues with "not evaluated" being an execution-time concept, so not suitable for translation-time undefined behavior or constraints.)
