[clang] [analyzer] MallocChecker – Fix false positive leak for smart pointers in temporary objects (PR #152751)

[Ivan Murashko via cfe-commits]

================
@@ -3068,12 +3124,242 @@ void MallocChecker::checkDeadSymbols(SymbolReaper &SymReaper,
   C.addTransition(state->set<RegionState>(RS), N);
 }
 
+// Helper function to check if a name is a recognized smart pointer name
+static bool isSmartPtrName(StringRef Name) {
+  return Name == "unique_ptr" || Name == "shared_ptr";
+}
+
+// Allowlist of owning smart pointers we want to recognize.
+// Start with unique_ptr and shared_ptr. (intentionally exclude weak_ptr)
+static bool isSmartOwningPtrType(QualType QT) {
+  QT = QT->getCanonicalTypeUnqualified();
+
+  // First try TemplateSpecializationType (for std smart pointers)
+  if (const auto *TST = QT->getAs<TemplateSpecializationType>()) {
+    const TemplateDecl *TD = TST->getTemplateName().getAsTemplateDecl();
+    if (!TD)
+      return false;
+
+    const auto *ND = dyn_cast_or_null<NamedDecl>(TD->getTemplatedDecl());
+    if (!ND)
+      return false;
+
+    // Check if it's in std namespace
+    if (!isWithinStdNamespace(ND))
+      return false;
+
+    return isSmartPtrName(ND->getName());
+  }
+
+  // Also try RecordType (for custom smart pointer implementations)
+  if (const auto *RD = QT->getAsCXXRecordDecl()) {
+    // Accept any custom unique_ptr or shared_ptr implementation
+    return isSmartPtrName(RD->getName());
+  }
----------------
ivanmurashko wrote:

Good catch! You're absolutely right. The RecordType fallback became redundant after the constructor-based detection was introduced in bb5eacc79dc69. The constructor mechanism (`handleSmartPointerConstructorArguments`) now properly handles both template and non-template smart pointers, making the RecordType branch unnecessary. That's why I removed it in 08a24effa0007135844fb37d0ecf6e8f1972a45b.

https://github.com/llvm/llvm-project/pull/152751