[clang] [clang] SFINAE context refactor (PR #164703)

[Younan Zhang via cfe-commits]

================
@@ -12394,45 +12391,65 @@ class Sema final : public SemaBase {
   /// failures rather than hard errors.
   bool AccessCheckingSFINAE;
 
+  class SFINAETrap;
+
+  struct SFINAEContextBase {
+    SFINAEContextBase(Sema &S, SFINAETrap *Cur)
+        : S(S), Prev(std::exchange(S.CurrentSFINAEContext, Cur)) {}
+
+  protected:
+    Sema &S;
+    ~SFINAEContextBase() { S.CurrentSFINAEContext = Prev; }
+
+  private:
+    SFINAETrap *Prev;
+  };
----------------
zyn0217 wrote:

It looks weird to have a sub-class's pointer in a base class. Though this is not a blocker

https://github.com/llvm/llvm-project/pull/164703