[clang] [libcxx] [Clang] Add __common_type builtin (PR #99473)

[Nikolas Klauser via cfe-commits]

================
@@ -3058,6 +3058,133 @@ void Sema::NoteAllFoundTemplates(TemplateName Name) {
   }
 }
 
+static QualType commonTypeImpl(Sema &S, TemplateName BaseTemplate,
+                               SourceLocation TemplateLoc,
+                               ArrayRef<TemplateArgument> Ts) {
+  auto lookUpCommonType = [&](TemplateArgument T1,
+                              TemplateArgument T2) -> QualType {
+    // Don't bother looking for other specializations if both types are
+    // builtins - users aren't allowed to specialize for them
+    if (T1.getAsType()->isBuiltinType() && T2.getAsType()->isBuiltinType())
+      return commonTypeImpl(S, BaseTemplate, TemplateLoc, {T1, T2});
+
+    TemplateArgumentListInfo Args;
+    Args.addArgument(TemplateArgumentLoc(
+        T1, S.Context.getTrivialTypeSourceInfo(T1.getAsType())));
+    Args.addArgument(TemplateArgumentLoc(
+        T2, S.Context.getTrivialTypeSourceInfo(T2.getAsType())));
+    QualType BaseTemplateInst =
+        S.CheckTemplateIdType(BaseTemplate, TemplateLoc, Args);
+    if (S.RequireCompleteType(TemplateLoc, BaseTemplateInst,
+                              diag::err_incomplete_type))
+      return QualType();
+    return S.getTypeMember(BaseTemplateInst, "type");
----------------
philnik777 wrote:

Nice idea! Looks like it works just fine and doesn't have any impact on the performance. The only downside is that we don't diagnose incomplete `common_type`s anymore, since they SFINAE away now. OTOH http://eel.is/c++draft/type.traits#meta.trans.other-4.2 could be read as requiring that behaviour so who knows.

https://github.com/llvm/llvm-project/pull/99473