[clang] [Clang][Sema] Do not mark template parameters in the exception specification as used during partial ordering (PR #91534)

[Erich Keane via cfe-commits]

================
@@ -5485,20 +5485,40 @@ static bool isAtLeastAsSpecializedAs(Sema &S, SourceLocation Loc,
   switch (TPOC) {
   case TPOC_Call:
     for (unsigned I = 0, N = Args2.size(); I != N; ++I)
-      ::MarkUsedTemplateParameters(S.Context, Args2[I], false,
-                                   TemplateParams->getDepth(),
-                                   UsedParameters);
+      ::MarkUsedTemplateParameters(S.Context, Args2[I], /*OnlyDeduced=*/false,
+                                   TemplateParams->getDepth(), UsedParameters);
     break;
 
   case TPOC_Conversion:
-    ::MarkUsedTemplateParameters(S.Context, Proto2->getReturnType(), false,
+    ::MarkUsedTemplateParameters(S.Context, Proto2->getReturnType(),
+                                 /*OnlyDeduced=*/false,
                                  TemplateParams->getDepth(), UsedParameters);
     break;
 
   case TPOC_Other:
-    ::MarkUsedTemplateParameters(S.Context, FD2->getType(), false,
-                                 TemplateParams->getDepth(),
-                                 UsedParameters);
+    // We do not deduce template arguments from the exception specification
+    // when determining the primary template of a function template
+    // specialization or when taking the address of a function template.
+    // Therefore, we do not mark template parameters in the exception
+    // specification as used during partial ordering to prevent the following
+    // from being ambiguous:
+    //
+    //   template<typename T, typename U>
+    //   void f(U) noexcept(noexcept(T())); // #1
+    //
+    //   template<typename T>
+    //   void f(T*) noexcept; // #2
+    //
+    //   template<>
+    //   void f<int>(int*) noexcept; // explicit specialization of #2
+    //
+    // Although there is no corresponding wording in the standard, this seems
----------------
erichkeane wrote:

Please file a core issue here!  It would be great to get this fixed in CWG.  @Endilll or @shafik  can help with that if you need it.  Also, put the CWG issue # in this comment once we have it.

https://github.com/llvm/llvm-project/pull/91534