[libcxx] [test] std::get<0>([std::variant constant expression]) *is* noexcept

[CaseyCarter]

Summary

noexcept(foo) is true when foo is a constant expression.

[CaseyCarter]

test/std/utilities/variant/variant.get/get_index.pass.cpp

39

THAT'S RIGHT, A CLANG BUG (https://llvm.org/bugs/show_bug.cgi?id=15481) THAT C1XX GETS CORRECT. You saw it here first.

[STL_MSFT]

I also saw get_type.pass.cpp failing.

I would recommend commenting the ifndef with the clang bug URL you found.

[EricWF]

test/std/utilities/variant/variant.get/get_index.pass.cpp

39

Ha A clang bug indeed! Although IDK how sane the other behavior is, specifically:

constexpr bool foo() noexcept(false) { return true; }
static_assert(!noexcept(foo()));

This part of the change LGTM, but could you also add a test that the function is not noexcept when it cannot do constant evaluation?

[CaseyCarter]

Address review comments:

• Annotate the clang bug which the conditional compilation is avoiding.
• Assure the expression in question is not noexcept when not a constant expression.
• Cover get_type

[mclow.lists]

I'd appreciate it if someone made a note to revisit this test when the clang bug is fixed, and change the #ifndef __clang__ to something like #ifndef __clang__ || clang_version < XXXX".

This LGTM.

test/std/utilities/variant/variant.get/get_index.pass.cpp

 #include <type_traits>
 #include <utility>
 #include <variant>
 
 void test_const_lvalue_get() {
   {
     using V = std::variant<int, const long>;
     constexpr V v(42);
-    ASSERT_NOT_NOEXCEPT(std::get<0>(v));
+#ifndef __clang__
+    ASSERT_NOEXCEPT(std::get<0>(v));
+#endif
     ASSERT_SAME_TYPE(decltype(std::get<0>(v)), const int &);
     static_assert(std::get<0>(v) == 42, "");
   }
   {
     using V = std::variant<int, const long>;
     constexpr V v(42l);
     ASSERT_SAME_TYPE(decltype(std::get<1>(v)), const long &);
     static_assert(std::get<1>(v) == 42, "");