[clang][Sema] Avoid guessing unexpanded packs' size in getFullyPackExpandedSize (#87768)

There has been an optimization for `SizeOfPackExprs` since c5452ed, in
which
we overlooked a case where the template arguments were not yet
formed into a `PackExpansionType` at the token annotation stage. This
led to a problem in that a template involving such expressions may
lose its nature of being dependent, causing some false-positive
diagnostics.

Fixes #84220

Author: zyn0217

clang/docs/ReleaseNotes.rst

@@ -526,6 +526,7 @@ Bug Fixes to C++ Support
 - Fix crash when inheriting from a cv-qualified type. Fixes:
   (`#35603 <https://github.com/llvm/llvm-project/issues/35603>`_)
 - Fix a crash when the using enum declaration uses an anonymous enumeration. Fixes (#GH86790).
+- Handled an edge case in ``getFullyPackExpandedSize`` so that we now avoid a false-positive diagnostic. (#GH84220)
 - Clang now correctly tracks type dependence of by-value captures in lambdas with an explicit
   object parameter.
   Fixes (#GH70604), (#GH79754), (#GH84163), (#GH84425), (#GH86054), (#GH86398), and (#GH86399).

clang/lib/Sema/SemaTemplateVariadic.cpp

@@ -1243,6 +1243,17 @@ std::optional<unsigned> Sema::getFullyPackExpandedSize(TemplateArgument Arg) {
     // expanded this pack expansion into the enclosing pack if we could.
     if (Elem.isPackExpansion())
       return std::nullopt;
+    // Don't guess the size of unexpanded packs. The pack within a template
+    // argument may have yet to be of a PackExpansion type before we see the
+    // ellipsis in the annotation stage.
+    //
+    // This doesn't mean we would invalidate the optimization: Arg can be an
+    // unexpanded pack regardless of Elem's dependence. For instance,
+    // A TemplateArgument that contains either a SubstTemplateTypeParmPackType
+    // or SubstNonTypeTemplateParmPackExpr is always considered Unexpanded, but
+    // the underlying TemplateArgument thereof may not.
+    if (Elem.containsUnexpandedParameterPack())
+      return std::nullopt;
   }
   return Pack.pack_size();
 }

clang/test/SemaTemplate/alias-templates.cpp

@@ -236,6 +236,29 @@ namespace PR14858 {
   void test_q(int (&a)[5]) { Q<B, B, B>().f<B, B>(&a); }
 }
 
+namespace PR84220 {
+
+template <class...> class list {};
+
+template <int> struct foo_impl {
+  template <class> using f = int;
+};
+
+template <class... xs>
+using foo = typename foo_impl<sizeof...(xs)>::template f<xs...>;
+
+// We call getFullyPackExpandedSize at the annotation stage
+// before parsing the ellipsis next to the foo<xs>. This happens before
+// a PackExpansionType is formed for foo<xs>.
+// getFullyPackExpandedSize shouldn't determine the value here. Otherwise,
+// foo_impl<sizeof...(xs)> would lose its dependency despite the template
+// arguments being unsubstituted.
+template <class... xs> using test = list<foo<xs>...>;
+
+test<int> a;
+
+}
+
 namespace redecl {
   template<typename> using A = int;
   template<typename = void> using A = int;