Sema: Allow scoped enums as source type for integral conversion.

[pcc]

Summary

Fixes pr54158.

[pcc]

Ping.

[pcc]

Ping^2.

[shafik]

I commented on the bug report: https://github.com/llvm/llvm-project/issues/54158

I wanted to also add a note here, I don't believe this is a valid change. The example should be ill-formed as was clarified by defect report 2374

[pcc]

Does that DR apply retroactively to C++17? I get the impression that "Status: C++20" means that the issue was only fixed in C++20, which would make this well-formed with -std=c++17.

[shafik]

In D120862#3857340, @pcc wrote:

Does that DR apply retroactively to C++17? I get the impression that "Status: C++20" means that the issue was only fixed in C++20, which would make this well-formed with -std=c++17.

This was simply an oversite in the wording of the paper see discussion here and this was never intended to work.

We have some discretion about how far back to apply DRs and I believe as long as applying them back does not cause major disruption we do CC @aaron.ballman

[aaron.ballman]

In D120862#3857372, @shafik wrote:

In D120862#3857340, @pcc wrote:

Does that DR apply retroactively to C++17? I get the impression that "Status: C++20" means that the issue was only fixed in C++20, which would make this well-formed with -std=c++17.

This was simply an oversite in the wording of the paper see discussion here and this was never intended to work.

We have some discretion about how far back to apply DRs and I believe as long as applying them back does not cause major disruption we do CC @aaron.ballman

You're correct that "Status: C++20" means that the issue was fixed in C++20. And technically speaking, we only need to apply those changes in C++20 and later. However, the intent of the DR process is to fix mistakes in older language modes. ISO doesn't let us retroactively change a published standard, so we keep this side list of "oh you should also fix this stuff" with the intent that the fixes apply as far back as they're relevant. We try to follow that guidance whenever we can, but if a DR causes code to break and that turns out to be disruptive, we'll sometimes decide to not apply it farther back than the standard version it was fixed in. I don't think there's a problem applying that specific DR as far back as we can.

clang/lib/Sema/SemaOverload.cpp

NarrowingKind StandardConversionSequence::getNarrowingKind(
 
   // -- from an integer type or unscoped enumeration type to an integer type
   //    that cannot represent all the values of the original type, except where
   //    the source is a constant expression and the actual value after
   //    conversion will fit into the target type and will produce the original
   //    value when converted back to the original type.
   case ICK_Integral_Conversion:
   IntegralConversion: {
-    assert(FromType->isIntegralOrUnscopedEnumerationType());
+    assert(FromType->isIntegralOrEnumerationType());
     assert(ToType->isIntegralOrUnscopedEnumerationType());
     const bool FromSigned = FromType->isSignedIntegerOrEnumerationType();
     const unsigned FromWidth = Ctx.getIntWidth(FromType);
     const bool ToSigned = ToType->isSignedIntegerOrEnumerationType();
     const unsigned ToWidth = Ctx.getIntWidth(ToType);
 
     if (FromWidth > ToWidth ||
         (FromWidth == ToWidth && FromSigned != ToSigned) ||

clang/test/SemaCXX/cxx17-enum-scoped.cpp

+// RUN: %clang_cc1 -fsyntax-only -pedantic -std=c++17 -verify -triple x86_64-apple-darwin %s
+
+// expected-no-diagnostics
+
+namespace PR54158 {
+  enum class A : int;
+  enum class B : int;
+  B x{A{}};
+}