[clang] Propagate lifetimebound from formal parameters to those in the canonical declaration and use that for analysis (PR #107627)

[Ilya Biryukov via cfe-commits]

================
@@ -3250,26 +3285,17 @@ static void mergeParamDeclAttributes(ParmVarDecl *newDecl,
            diag::note_carries_dependency_missing_first_decl) << 1/*Param*/;
   }
 
-  if (!oldDecl->hasAttrs())
-    return;
-
-  bool foundAny = newDecl->hasAttrs();
-
-  // Ensure that any moving of objects within the allocated map is
-  // done before we process them.
-  if (!foundAny) newDecl->setAttrs(AttrVec());
-
-  for (const auto *I : oldDecl->specific_attrs<InheritableParamAttr>()) {
-    if (!DeclHasAttr(newDecl, I)) {
-      InheritableAttr *newAttr =
-        cast<InheritableParamAttr>(I->clone(S.Context));
-      newAttr->setInherited(true);
-      newDecl->addAttr(newAttr);
-      foundAny = true;
-    }
-  }
-
-  if (!foundAny) newDecl->dropAttrs();
+  propagateAttributes(
+      newDecl, oldDecl, [&S](ParmVarDecl *toDecl, const ParmVarDecl *fromDecl) {
+        unsigned found = 0;
+        found += propagateAttribute<InheritableParamAttr>(toDecl, fromDecl, S);
+        // Propagate the lifetimebound attribute from parameters to the
+        // most recent declaration. Note that this doesn't include the implicit
+        // 'this' parameter, as the attribute is applied to the function type in
+        // that case.
+        found += propagateAttribute<LifetimeBoundAttr>(toDecl, fromDecl, S);
----------------
ilya-biryukov wrote:

It feels that LifetimeBoundAttr might need to be an `InheritableParamAttr` itself, which would allow to avoid having a template altogether.

Are there any downsides to that?

(I'm assuming that `InheritableParamAttr` represents the attributes that should be added from redeclarations, not just from the base methods of overridden members)

https://github.com/llvm/llvm-project/pull/107627