[clang] [analyzer] MallocChecker: Recognize std::atomics in smart pointer suppression. (PR #90918)

[Gábor Horváth via cfe-commits]

================
@@ -3479,13 +3479,24 @@ PathDiagnosticPieceRef MallocBugVisitor::VisitNode(const ExplodedNode *N,
   // original reference count is positive, we should not report use-after-frees
   // on objects deleted in such destructors. This can probably be improved
   // through better shared pointer modeling.
-  if (ReleaseDestructorLC) {
+  if (ReleaseDestructorLC && (ReleaseDestructorLC == CurrentLC ||
+                              ReleaseDestructorLC->isParentOf(CurrentLC))) {
     if (const auto *AE = dyn_cast<AtomicExpr>(S)) {
+      // Check for manual use of atomic builtins.
       AtomicExpr::AtomicOp Op = AE->getOp();
       if (Op == AtomicExpr::AO__c11_atomic_fetch_add ||
           Op == AtomicExpr::AO__c11_atomic_fetch_sub) {
-        if (ReleaseDestructorLC == CurrentLC ||
-            ReleaseDestructorLC->isParentOf(CurrentLC)) {
+        BR.markInvalid(getTag(), S);
+      }
+    } else if (const auto *CE = dyn_cast<CallExpr>(S)) {
+      // Check for `std::atomic` and such. This covers both regular method calls
+      // and operator calls.
+      if (const auto *MD =
+              dyn_cast_or_null<CXXMethodDecl>(CE->getDirectCallee())) {
+        const CXXRecordDecl *RD = MD->getParent();
+        // A bit wobbly with ".contains()" because it may be like
+        // "__atomic_base" or something.
+        if (StringRef(RD->getNameAsString()).contains("atomic")) {
----------------
Xazax-hun wrote:

Do we need to do the heavy weight `getNameAsString` here? Could we get the identifier instead?

https://github.com/llvm/llvm-project/pull/90918