[analyzer][GSoC] Re-implemente current virtual calls checker in a path-sensitive way

Patch by: Xin Wang

Differential Revision: https://reviews.llvm.org/D34275

llvm-svn: 311877
diff --git a/clang/lib/StaticAnalyzer/Checkers/VirtualCallChecker.cpp b/clang/lib/StaticAnalyzer/Checkers/VirtualCallChecker.cpp
index b47762b..ea1976e 100644
--- a/clang/lib/StaticAnalyzer/Checkers/VirtualCallChecker.cpp
+++ b/clang/lib/StaticAnalyzer/Checkers/VirtualCallChecker.cpp
@@ -14,279 +14,272 @@
 
 #include "ClangSACheckers.h"
 #include "clang/AST/DeclCXX.h"
-#include "clang/AST/StmtVisitor.h"
 #include "clang/StaticAnalyzer/Core/BugReporter/BugReporter.h"
+#include "clang/StaticAnalyzer/Core/BugReporter/BugType.h"
 #include "clang/StaticAnalyzer/Core/Checker.h"
-#include "clang/StaticAnalyzer/Core/PathSensitive/AnalysisManager.h"
-#include "llvm/ADT/SmallString.h"
-#include "llvm/Support/SaveAndRestore.h"
-#include "llvm/Support/raw_ostream.h"
+#include "clang/StaticAnalyzer/Core/PathSensitive/CallEvent.h"
+#include "clang/StaticAnalyzer/Core/PathSensitive/CheckerContext.h"
+#include "clang/StaticAnalyzer/Core/PathSensitive/ProgramStateTrait.h"
+#include "clang/StaticAnalyzer/Core/PathSensitive/SValBuilder.h"
 
 using namespace clang;
 using namespace ento;
 
 namespace {
+enum class ObjectState : bool { CtorCalled, DtorCalled };
+} // end namespace
+  // FIXME: Ascending over StackFrameContext maybe another method.
 
-class WalkAST : public StmtVisitor<WalkAST> {
-  const CheckerBase *Checker;
-  BugReporter &BR;
-  AnalysisDeclContext *AC;
-
-  /// The root constructor or destructor whose callees are being analyzed.
-  const CXXMethodDecl *RootMethod = nullptr;
-
-  /// Whether the checker should walk into bodies of called functions.
-  /// Controlled by the "Interprocedural" analyzer-config option.
-  bool IsInterprocedural = false;
-
-  /// Whether the checker should only warn for calls to pure virtual functions
-  /// (which is undefined behavior) or for all virtual functions (which may
-  /// may result in unexpected behavior).
-  bool ReportPureOnly = false;
-
-  typedef const CallExpr * WorkListUnit;
-  typedef SmallVector<WorkListUnit, 20> DFSWorkList;
-
-  /// A vector representing the worklist which has a chain of CallExprs.
-  DFSWorkList WList;
-
-  // PreVisited : A CallExpr to this FunctionDecl is in the worklist, but the
-  // body has not been visited yet.
-  // PostVisited : A CallExpr to this FunctionDecl is in the worklist, and the
-  // body has been visited.
-  enum Kind { NotVisited,
-              PreVisited,  /**< A CallExpr to this FunctionDecl is in the
-                                worklist, but the body has not yet been
-                                visited. */
-              PostVisited  /**< A CallExpr to this FunctionDecl is in the
-                                worklist, and the body has been visited. */
-  };
-
-  /// A DenseMap that records visited states of FunctionDecls.
-  llvm::DenseMap<const FunctionDecl *, Kind> VisitedFunctions;
-
-  /// The CallExpr whose body is currently being visited.  This is used for
-  /// generating bug reports.  This is null while visiting the body of a
-  /// constructor or destructor.
-  const CallExpr *visitingCallExpr;
-
-public:
-  WalkAST(const CheckerBase *checker, BugReporter &br, AnalysisDeclContext *ac,
-          const CXXMethodDecl *rootMethod, bool isInterprocedural,
-          bool reportPureOnly)
-      : Checker(checker), BR(br), AC(ac), RootMethod(rootMethod),
-        IsInterprocedural(isInterprocedural), ReportPureOnly(reportPureOnly),
-        visitingCallExpr(nullptr) {
-    // Walking should always start from either a constructor or a destructor.
-    assert(isa<CXXConstructorDecl>(rootMethod) ||
-           isa<CXXDestructorDecl>(rootMethod));
+namespace llvm {
+template <> struct FoldingSetTrait<ObjectState> {
+  static inline void Profile(ObjectState X, FoldingSetNodeID &ID) {
+    ID.AddInteger(static_cast<int>(X));
   }
-
-  bool hasWork() const { return !WList.empty(); }
-
-  /// This method adds a CallExpr to the worklist and marks the callee as
-  /// being PreVisited.
-  void Enqueue(WorkListUnit WLUnit) {
-    const FunctionDecl *FD = WLUnit->getDirectCallee();
-    if (!FD || !FD->getBody())
-      return;
-    Kind &K = VisitedFunctions[FD];
-    if (K != NotVisited)
-      return;
-    K = PreVisited;
-    WList.push_back(WLUnit);
-  }
-
-  /// This method returns an item from the worklist without removing it.
-  WorkListUnit Dequeue() {
-    assert(!WList.empty());
-    return WList.back();
-  }
-
-  void Execute() {
-    while (hasWork()) {
-      WorkListUnit WLUnit = Dequeue();
-      const FunctionDecl *FD = WLUnit->getDirectCallee();
-      assert(FD && FD->getBody());
-
-      if (VisitedFunctions[FD] == PreVisited) {
-        // If the callee is PreVisited, walk its body.
-        // Visit the body.
-        SaveAndRestore<const CallExpr *> SaveCall(visitingCallExpr, WLUnit);
-        Visit(FD->getBody());
-
-        // Mark the function as being PostVisited to indicate we have
-        // scanned the body.
-        VisitedFunctions[FD] = PostVisited;
-        continue;
-      }
-
-      // Otherwise, the callee is PostVisited.
-      // Remove it from the worklist.
-      assert(VisitedFunctions[FD] == PostVisited);
-      WList.pop_back();
-    }
-  }
-
-  // Stmt visitor methods.
-  void VisitCallExpr(CallExpr *CE);
-  void VisitCXXMemberCallExpr(CallExpr *CE);
-  void VisitStmt(Stmt *S) { VisitChildren(S); }
-  void VisitChildren(Stmt *S);
-
-  void ReportVirtualCall(const CallExpr *CE, bool isPure);
-
 };
-} // end anonymous namespace
-
-//===----------------------------------------------------------------------===//
-// AST walking.
-//===----------------------------------------------------------------------===//
-
-void WalkAST::VisitChildren(Stmt *S) {
-  for (Stmt *Child : S->children())
-    if (Child)
-      Visit(Child);
-}
-
-void WalkAST::VisitCallExpr(CallExpr *CE) {
-  VisitChildren(CE);
-  if (IsInterprocedural)
-    Enqueue(CE);
-}
-
-void WalkAST::VisitCXXMemberCallExpr(CallExpr *CE) {
-  VisitChildren(CE);
-  bool callIsNonVirtual = false;
-
-  // Several situations to elide for checking.
-  if (MemberExpr *CME = dyn_cast<MemberExpr>(CE->getCallee())) {
-    // If the member access is fully qualified (i.e., X::F), then treat
-    // this as a non-virtual call and do not warn.
-    if (CME->getQualifier())
-      callIsNonVirtual = true;
-
-    if (Expr *base = CME->getBase()->IgnoreImpCasts()) {
-      // Elide analyzing the call entirely if the base pointer is not 'this'.
-      if (!isa<CXXThisExpr>(base))
-        return;
-
-      // If the most derived class is marked final, we know that now subclass
-      // can override this member.
-      if (base->getBestDynamicClassType()->hasAttr<FinalAttr>())
-        callIsNonVirtual = true;
-    }
-  }
-
-  // Get the callee.
-  const CXXMethodDecl *MD =
-      dyn_cast_or_null<CXXMethodDecl>(CE->getDirectCallee());
-  if (MD && MD->isVirtual() && !callIsNonVirtual && !MD->hasAttr<FinalAttr>() &&
-      !MD->getParent()->hasAttr<FinalAttr>())
-    ReportVirtualCall(CE, MD->isPure());
-
-  if (IsInterprocedural)
-    Enqueue(CE);
-}
-
-void WalkAST::ReportVirtualCall(const CallExpr *CE, bool isPure) {
-  if (ReportPureOnly && !isPure)
-    return;
-
-  SmallString<100> buf;
-  llvm::raw_svector_ostream os(buf);
-
-  // FIXME: The interprocedural diagnostic experience here is not good.
-  // Ultimately this checker should be re-written to be path sensitive.
-  // For now, only diagnose intraprocedurally, by default.
-  if (IsInterprocedural) {
-    os << "Call Path : ";
-    // Name of current visiting CallExpr.
-    os << *CE->getDirectCallee();
-
-    // Name of the CallExpr whose body is current being walked.
-    if (visitingCallExpr)
-      os << " <-- " << *visitingCallExpr->getDirectCallee();
-    // Names of FunctionDecls in worklist with state PostVisited.
-    for (SmallVectorImpl<const CallExpr *>::iterator I = WList.end(),
-         E = WList.begin(); I != E; --I) {
-      const FunctionDecl *FD = (*(I-1))->getDirectCallee();
-      assert(FD);
-      if (VisitedFunctions[FD] == PostVisited)
-        os << " <-- " << *FD;
-    }
-
-    os << "\n";
-  }
-
-  PathDiagnosticLocation CELoc =
-    PathDiagnosticLocation::createBegin(CE, BR.getSourceManager(), AC);
-  SourceRange R = CE->getCallee()->getSourceRange();
-
-  os << "Call to ";
-  if (isPure)
-    os << "pure ";
-
-  os << "virtual function during ";
-
-  if (isa<CXXConstructorDecl>(RootMethod))
-    os << "construction ";
-  else
-    os << "destruction ";
-
-  if (isPure)
-    os << "has undefined behavior";
-  else
-    os << "will not dispatch to derived class";
-
-  BR.EmitBasicReport(AC->getDecl(), Checker,
-                     "Call to virtual function during construction or "
-                     "destruction",
-                     "C++ Object Lifecycle", os.str(), CELoc, R);
-}
-
-//===----------------------------------------------------------------------===//
-// VirtualCallChecker
-//===----------------------------------------------------------------------===//
+} // end namespace llvm
 
 namespace {
-class VirtualCallChecker : public Checker<check::ASTDecl<CXXRecordDecl> > {
+class VirtualCallChecker
+    : public Checker<check::BeginFunction, check::EndFunction, check::PreCall> {
+  mutable std::unique_ptr<BugType> BT;
+
 public:
-  DefaultBool isInterprocedural;
-  DefaultBool isPureOnly;
+  // The flag to determine if pure virtual functions should be issued only.
+  DefaultBool IsPureOnly;
 
-  void checkASTDecl(const CXXRecordDecl *RD, AnalysisManager& mgr,
-                    BugReporter &BR) const {
-    AnalysisDeclContext *ADC = mgr.getAnalysisDeclContext(RD);
+  void checkBeginFunction(CheckerContext &C) const;
+  void checkEndFunction(CheckerContext &C) const;
+  void checkPreCall(const CallEvent &Call, CheckerContext &C) const;
 
-    // Check the constructors.
-    for (const auto *I : RD->ctors()) {
-      if (!I->isCopyOrMoveConstructor())
-        if (Stmt *Body = I->getBody()) {
-          WalkAST walker(this, BR, ADC, I, isInterprocedural, isPureOnly);
-          walker.Visit(Body);
-          walker.Execute();
-        }
+private:
+  void registerCtorDtorCallInState(bool IsBeginFunction,
+                                   CheckerContext &C) const;
+  void reportBug(StringRef Msg, bool PureError, const MemRegion *Reg,
+                 CheckerContext &C) const;
+
+  class VirtualBugVisitor : public BugReporterVisitorImpl<VirtualBugVisitor> {
+  private:
+    const MemRegion *ObjectRegion;
+    bool Found;
+
+  public:
+    VirtualBugVisitor(const MemRegion *R) : ObjectRegion(R), Found(false) {}
+
+    void Profile(llvm::FoldingSetNodeID &ID) const override {
+      static int X = 0;
+      ID.AddPointer(&X);
+      ID.AddPointer(ObjectRegion);
     }
 
-    // Check the destructor.
-    if (CXXDestructorDecl *DD = RD->getDestructor())
-      if (Stmt *Body = DD->getBody()) {
-        WalkAST walker(this, BR, ADC, DD, isInterprocedural, isPureOnly);
-        walker.Visit(Body);
-        walker.Execute();
-      }
-  }
+    std::shared_ptr<PathDiagnosticPiece> VisitNode(const ExplodedNode *N,
+                                                   const ExplodedNode *PrevN,
+                                                   BugReporterContext &BRC,
+                                                   BugReport &BR) override;
+  };
 };
+} // end namespace
+
+// GDM (generic data map) to the memregion of this for the ctor and dtor.
+REGISTER_MAP_WITH_PROGRAMSTATE(CtorDtorMap, const MemRegion *, ObjectState)
+
+std::shared_ptr<PathDiagnosticPiece>
+VirtualCallChecker::VirtualBugVisitor::VisitNode(const ExplodedNode *N,
+                                                 const ExplodedNode *PrevN,
+                                                 BugReporterContext &BRC,
+                                                 BugReport &BR) {
+  // We need the last ctor/dtor which call the virtual function.
+  // The visitor walks the ExplodedGraph backwards.
+  if (Found)
+    return nullptr;
+
+  ProgramStateRef State = N->getState();
+  const LocationContext *LCtx = N->getLocationContext();
+  const CXXConstructorDecl *CD =
+      dyn_cast_or_null<CXXConstructorDecl>(LCtx->getDecl());
+  const CXXDestructorDecl *DD =
+      dyn_cast_or_null<CXXDestructorDecl>(LCtx->getDecl());
+
+  if (!CD && !DD)
+    return nullptr;
+
+  ProgramStateManager &PSM = State->getStateManager();
+  auto &SVB = PSM.getSValBuilder();
+  const auto *MD = dyn_cast<CXXMethodDecl>(LCtx->getDecl());
+  if (!MD)
+    return nullptr;
+  auto ThiSVal =
+      State->getSVal(SVB.getCXXThis(MD, LCtx->getCurrentStackFrame()));
+  const MemRegion *Reg = ThiSVal.castAs<loc::MemRegionVal>().getRegion();
+  if (!Reg)
+    return nullptr;
+  if (Reg != ObjectRegion)
+    return nullptr;
+
+  const Stmt *S = PathDiagnosticLocation::getStmt(N);
+  if (!S)
+    return nullptr;
+  Found = true;
+
+  std::string InfoText;
+  if (CD)
+    InfoText = "This constructor of an object of type '" +
+               CD->getNameAsString() +
+               "' has not returned when the virtual method was called";
+  else
+    InfoText = "This destructor of an object of type '" +
+               DD->getNameAsString() +
+               "' has not returned when the virtual method was called";
+
+  // Generate the extra diagnostic.
+  PathDiagnosticLocation Pos(S, BRC.getSourceManager(),
+                             N->getLocationContext());
+  return std::make_shared<PathDiagnosticEventPiece>(Pos, InfoText, true);
+}
+
+// The function to check if a callexpr is a virtual function.
+static bool isVirtualCall(const CallExpr *CE) {
+  bool CallIsNonVirtual = false;
+
+  if (const MemberExpr *CME = dyn_cast<MemberExpr>(CE->getCallee())) {
+    // The member access is fully qualified (i.e., X::F).
+    // Treat this as a non-virtual call and do not warn.
+    if (CME->getQualifier())
+      CallIsNonVirtual = true;
+
+    if (const Expr *Base = CME->getBase()->IgnoreImpCasts()) {
+      // The most derived class is marked final.
+      if (Base->getBestDynamicClassType()->hasAttr<FinalAttr>())
+        CallIsNonVirtual = true;
+    }
+  }
+
+  const CXXMethodDecl *MD =
+      dyn_cast_or_null<CXXMethodDecl>(CE->getDirectCallee());
+  if (MD && MD->isVirtual() && !CallIsNonVirtual && !MD->hasAttr<FinalAttr>() &&
+      !MD->getParent()->hasAttr<FinalAttr>())
+    return true;
+  return false;
+}
+
+// The BeginFunction callback when enter a constructor or a destructor.
+void VirtualCallChecker::checkBeginFunction(CheckerContext &C) const {
+  registerCtorDtorCallInState(true, C);
+}
+
+// The EndFunction callback when leave a constructor or a destructor.
+void VirtualCallChecker::checkEndFunction(CheckerContext &C) const {
+  registerCtorDtorCallInState(false, C);
+}
+
+void VirtualCallChecker::checkPreCall(const CallEvent &Call,
+                                      CheckerContext &C) const {
+  const auto MC = dyn_cast<CXXMemberCall>(&Call);
+  if (!MC)
+    return;
+
+  const CXXMethodDecl *MD = dyn_cast_or_null<CXXMethodDecl>(Call.getDecl());
+  if (!MD)
+    return;
+  ProgramStateRef State = C.getState();
+  const CallExpr *CE = dyn_cast_or_null<CallExpr>(Call.getOriginExpr());
+
+  if (IsPureOnly && !MD->isPure())
+    return;
+  if (!isVirtualCall(CE))
+    return;
+
+  const MemRegion *Reg = MC->getCXXThisVal().getAsRegion();
+  const ObjectState *ObState = State->get<CtorDtorMap>(Reg);
+  if (!ObState)
+    return;
+  // Check if a virtual method is called.
+  // The GDM of constructor and destructor should be true.
+  if (*ObState == ObjectState::CtorCalled) {
+    if (IsPureOnly && MD->isPure())
+      reportBug("Call to pure virtual function during construction", true, Reg,
+                C);
+    else if (!MD->isPure())
+      reportBug("Call to virtual function during construction", false, Reg, C);
+    else
+      reportBug("Call to pure virtual function during construction", false, Reg,
+                C);
+  }
+
+  if (*ObState == ObjectState::DtorCalled) {
+    if (IsPureOnly && MD->isPure())
+      reportBug("Call to pure virtual function during destruction", true, Reg,
+                C);
+    else if (!MD->isPure())
+      reportBug("Call to virtual function during destruction", false, Reg, C);
+    else
+      reportBug("Call to pure virtual function during construction", false, Reg,
+                C);
+  }
+}
+
+void VirtualCallChecker::registerCtorDtorCallInState(bool IsBeginFunction,
+                                                     CheckerContext &C) const {
+  const auto *LCtx = C.getLocationContext();
+  const auto *MD = dyn_cast_or_null<CXXMethodDecl>(LCtx->getDecl());
+  if (!MD)
+    return;
+
+  ProgramStateRef State = C.getState();
+  auto &SVB = C.getSValBuilder();
+
+  // Enter a constructor, set the corresponding memregion be true.
+  if (isa<CXXConstructorDecl>(MD)) {
+    auto ThiSVal =
+        State->getSVal(SVB.getCXXThis(MD, LCtx->getCurrentStackFrame()));
+    const MemRegion *Reg = ThiSVal.getAsRegion();
+    if (IsBeginFunction)
+      State = State->set<CtorDtorMap>(Reg, ObjectState::CtorCalled);
+    else
+      State = State->remove<CtorDtorMap>(Reg);
+
+    C.addTransition(State);
+    return;
+  }
+
+  // Enter a Destructor, set the corresponding memregion be true.
+  if (isa<CXXDestructorDecl>(MD)) {
+    auto ThiSVal =
+        State->getSVal(SVB.getCXXThis(MD, LCtx->getCurrentStackFrame()));
+    const MemRegion *Reg = ThiSVal.getAsRegion();
+    if (IsBeginFunction)
+      State = State->set<CtorDtorMap>(Reg, ObjectState::DtorCalled);
+    else
+      State = State->remove<CtorDtorMap>(Reg);
+
+    C.addTransition(State);
+    return;
+  }
+}
+
+void VirtualCallChecker::reportBug(StringRef Msg, bool IsSink,
+                                   const MemRegion *Reg,
+                                   CheckerContext &C) const {
+  ExplodedNode *N;
+  if (IsSink)
+    N = C.generateErrorNode();
+  else
+    N = C.generateNonFatalErrorNode();
+
+  if (!N)
+    return;
+  if (!BT)
+    BT.reset(new BugType(
+        this, "Call to virtual function during construction or destruction",
+        "C++ Object Lifecycle"));
+
+  auto Reporter = llvm::make_unique<BugReport>(*BT, Msg, N);
+  Reporter->addVisitor(llvm::make_unique<VirtualBugVisitor>(Reg));
+  C.emitReport(std::move(Reporter));
 }
 
 void ento::registerVirtualCallChecker(CheckerManager &mgr) {
   VirtualCallChecker *checker = mgr.registerChecker<VirtualCallChecker>();
-  checker->isInterprocedural =
-      mgr.getAnalyzerOptions().getBooleanOption("Interprocedural", false,
-                                                checker);
 
-  checker->isPureOnly =
-      mgr.getAnalyzerOptions().getBooleanOption("PureOnly", false,
-                                                checker);
+  checker->IsPureOnly =
+      mgr.getAnalyzerOptions().getBooleanOption("PureOnly", false, checker);
 }