lib/Analysis/CallGraph.cpp

//===- CallGraph.cpp - AST-based Call graph -------------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
//  This file defines the AST-based CallGraph.
//
//===----------------------------------------------------------------------===//

#include "clang/Analysis/CallGraph.h"
#include "clang/AST/Decl.h"
#include "clang/AST/DeclBase.h"
#include "clang/AST/DeclObjC.h"
#include "clang/AST/Expr.h"
#include "clang/AST/ExprObjC.h"
#include "clang/AST/Stmt.h"
#include "clang/AST/StmtVisitor.h"
#include "clang/Basic/IdentifierTable.h"
#include "clang/Basic/LLVM.h"
#include "llvm/ADT/PostOrderIterator.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/DOTGraphTraits.h"
#include "llvm/Support/GraphWriter.h"
#include "llvm/Support/raw_ostream.h"
#include <cassert>
#include <memory>
#include <string>

using namespace clang;

#define DEBUG_TYPE "CallGraph"

STATISTIC(NumObjCCallEdges, "Number of Objective-C method call edges");
STATISTIC(NumBlockCallEdges, "Number of block call edges");

namespace {

/// A helper class, which walks the AST and locates all the call sites in the
/// given function body.
class CGBuilder : public StmtVisitor<CGBuilder> {
  CallGraph *G;
  CallGraphNode *CallerNode;

public:
  CGBuilder(CallGraph *g, CallGraphNode *N) : G(g), CallerNode(N) {}

  void VisitStmt(Stmt *S) { VisitChildren(S); }

  Decl *getDeclFromCall(CallExpr *CE) {
    if (FunctionDecl *CalleeDecl = CE->getDirectCallee())
      return CalleeDecl;

    // Simple detection of a call through a block.
    Expr *CEE = CE->getCallee()->IgnoreParenImpCasts();
    if (BlockExpr *Block = dyn_cast<BlockExpr>(CEE)) {
      NumBlockCallEdges++;
      return Block->getBlockDecl();
    }

    return nullptr;
  }

  void addCalledDecl(Decl *D, Expr *CallExpr) {
    if (G->includeCalleeInGraph(D)) {
      CallGraphNode *CalleeNode = G->getOrInsertNode(D);
      CallerNode->addCallee({CalleeNode, CallExpr});
    }
  }

  void VisitCallExpr(CallExpr *CE) {
    if (Decl *D = getDeclFromCall(CE))
      addCalledDecl(D, CE);
    VisitChildren(CE);
  }

  void VisitLambdaExpr(LambdaExpr *LE) {
    if (FunctionTemplateDecl *FTD = LE->getDependentCallOperator())
      for (FunctionDecl *FD : FTD->specializations())
        G->VisitFunctionDecl(FD);
    else if (CXXMethodDecl *MD = LE->getCallOperator())
      G->VisitFunctionDecl(MD);
  }

  void VisitCXXNewExpr(CXXNewExpr *E) {
    if (FunctionDecl *FD = E->getOperatorNew())
      addCalledDecl(FD, E);
    VisitChildren(E);
  }

  void VisitCXXConstructExpr(CXXConstructExpr *E) {
    CXXConstructorDecl *Ctor = E->getConstructor();
    if (FunctionDecl *Def = Ctor->getDefinition())
      addCalledDecl(Def, E);
    VisitChildren(E);
  }

  // Include the evaluation of the default argument.
  void VisitCXXDefaultArgExpr(CXXDefaultArgExpr *E) {
    Visit(E->getExpr());
  }

  // Include the evaluation of the default initializers in a class.
  void VisitCXXDefaultInitExpr(CXXDefaultInitExpr *E) {
    Visit(E->getExpr());
  }

  // Adds may-call edges for the ObjC message sends.
  void VisitObjCMessageExpr(ObjCMessageExpr *ME) {
    if (ObjCInterfaceDecl *IDecl = ME->getReceiverInterface()) {
      Selector Sel = ME->getSelector();

      // Find the callee definition within the same translation unit.
      Decl *D = nullptr;
      if (ME->isInstanceMessage())
        D = IDecl->lookupPrivateMethod(Sel);
      else
        D = IDecl->lookupPrivateClassMethod(Sel);
      if (D) {
        addCalledDecl(D, ME);
        NumObjCCallEdges++;
      }
    }
  }

  void VisitChildren(Stmt *S) {
    for (Stmt *SubStmt : S->children())
      if (SubStmt)
        this->Visit(SubStmt);
  }
};

} // namespace

void CallGraph::addNodesForBlocks(DeclContext *D) {
  if (BlockDecl *BD = dyn_cast<BlockDecl>(D))
    addNodeForDecl(BD, true);

  for (auto *I : D->decls())
    if (auto *DC = dyn_cast<DeclContext>(I))
      addNodesForBlocks(DC);
}

CallGraph::CallGraph() {
  ShouldWalkTypesOfTypeLocs = false;
  ShouldVisitTemplateInstantiations = true;
  ShouldVisitImplicitCode = true;
  Root = getOrInsertNode(nullptr);
}

CallGraph::~CallGraph() = default;

bool CallGraph::includeInGraph(const Decl *D) {
  assert(D);
  if (!D->hasBody())
    return false;

  return includeCalleeInGraph(D);
}

bool CallGraph::includeCalleeInGraph(const Decl *D) {
  if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(D)) {
    // We skip function template definitions, as their semantics is
    // only determined when they are instantiated.
    if (FD->isDependentContext())
      return false;

    IdentifierInfo *II = FD->getIdentifier();
    if (II && II->getName().starts_with("__inline"))
      return false;
  }

  return true;
}

void CallGraph::addNodeForDecl(Decl* D, bool IsGlobal) {
  assert(D);

  // Allocate a new node, mark it as root, and process its calls.
  CallGraphNode *Node = getOrInsertNode(D);

  // Process all the calls by this function as well.
  CGBuilder builder(this, Node);
  if (Stmt *Body = D->getBody())
    builder.Visit(Body);

  // Include C++ constructor member initializers.
  if (auto constructor = dyn_cast<CXXConstructorDecl>(D)) {
    for (CXXCtorInitializer *init : constructor->inits()) {
      builder.Visit(init->getInit());
    }
  }
}

CallGraphNode *CallGraph::getNode(const Decl *F) const {
  FunctionMapTy::const_iterator I = FunctionMap.find(F);
  if (I == FunctionMap.end()) return nullptr;
  return I->second.get();
}

CallGraphNode *CallGraph::getOrInsertNode(Decl *F) {
  if (F && !isa<ObjCMethodDecl>(F))
    F = F->getCanonicalDecl();

  std::unique_ptr<CallGraphNode> &Node = FunctionMap[F];
  if (Node)
    return Node.get();

  Node = std::make_unique<CallGraphNode>(F);
  // Make Root node a parent of all functions to make sure all are reachable.
  if (F)
    Root->addCallee({Node.get(), /*Call=*/nullptr});
  return Node.get();
}

void CallGraph::print(raw_ostream &OS) const {
  OS << " --- Call graph Dump --- \n";

  // We are going to print the graph in reverse post order, partially, to make
  // sure the output is deterministic.
  llvm::ReversePostOrderTraversal<const CallGraph *> RPOT(this);
  for (const CallGraphNode *N : RPOT) {
    OS << "  Function: ";
    if (N == Root)
      OS << "< root >";
    else
      N->print(OS);

    OS << " calls: ";
    for (CallGraphNode::const_iterator CI = N->begin(),
                                       CE = N->end(); CI != CE; ++CI) {
      assert(CI->Callee != Root && "No one can call the root node.");
      CI->Callee->print(OS);
      OS << " ";
    }
    OS << '\n';
  }
  OS.flush();
}

LLVM_DUMP_METHOD void CallGraph::dump() const {
  print(llvm::errs());
}

void CallGraph::viewGraph() const {
  llvm::ViewGraph(this, "CallGraph");
}

void CallGraphNode::print(raw_ostream &os) const {
  if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(FD))
      return ND->printQualifiedName(os);
  os << "< >";
}

LLVM_DUMP_METHOD void CallGraphNode::dump() const {
  print(llvm::errs());
}

namespace llvm {

template <>
struct DOTGraphTraits<const CallGraph*> : public DefaultDOTGraphTraits {
  DOTGraphTraits (bool isSimple = false) : DefaultDOTGraphTraits(isSimple) {}

  static std::string getNodeLabel(const CallGraphNode *Node,
                                  const CallGraph *CG) {
    if (CG->getRoot() == Node) {
      return "< root >";
    }
    if (const NamedDecl *ND = dyn_cast_or_null<NamedDecl>(Node->getDecl()))
      return ND->getNameAsString();
    else
      return "< >";
  }
};

} // namespace llvm