Add initial prototype for implementation of

-Wuninitialized based on CFG dataflow analysis.  WIP.

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@123512 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/clang/Analysis/Analyses/UninitializedValuesV2.h b/include/clang/Analysis/Analyses/UninitializedValuesV2.h
new file mode 100644 (file)
index 0000000..65390b6
--- /dev/null
+++ b/include/clang/Analysis/Analyses/UninitializedValuesV2.h
@@ -0,0 +1,38 @@
+//= UninitializedValuesV2.h - Finding uses of uninitialized values --*- C++ -*-=
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines APIs for invoking and reported uninitialized values
+// warnings.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_CLANG_UNINIT_VALS_H
+#define LLVM_CLANG_UNINIT_VALS_H
+
+namespace clang {
+
+class CFG;  
+class DeclContext;
+class DeclRefExpr;
+class VarDecl;
+  
+class UninitVariablesHandler {
+public:
+  UninitVariablesHandler() {}
+  virtual ~UninitVariablesHandler();
+  
+  virtual void handleUseOfUninitVariable(const DeclRefExpr *dr,
+                                         const VarDecl *vd) {}
+};
+  
+void runUninitializedVariablesAnalysis(const DeclContext &dc, const CFG &cfg,
+                                       UninitVariablesHandler &handler);
+
+}
+#endif

diff --git a/include/clang/Basic/DiagnosticSemaKinds.td b/include/clang/Basic/DiagnosticSemaKinds.td
index a5a3d83833149d6f2e6e7a2f6ad1594cad16f705..aaccaabd7529f72d2dcb1ca20f83325fd0286103 100644 (file)
--- a/include/clang/Basic/DiagnosticSemaKinds.td
+++ b/include/clang/Basic/DiagnosticSemaKinds.td
@@ -815,6 +815,8 @@ def note_uninit_reference_member : Note<
   "uninitialized reference member is here">;
 def warn_field_is_uninit : Warning<"field is uninitialized when used here">,
   InGroup<DiagGroup<"uninitialized">>;
+def warn_var_is_uninit : Warning<"use of uninitialized variable %0">,
+  InGroup<DiagGroup<"uninitialized-experimental">>, DefaultIgnore;
 def err_init_incomplete_type : Error<"initialization of incomplete type %0">;
 
 def err_temp_copy_no_viable : Error<

diff --git a/lib/Analysis/CMakeLists.txt b/lib/Analysis/CMakeLists.txt
index 1709f6d6b78fc85c1d776a31478afe4590537971..a7bd8c193fc74fca29d64564990a1b17edab07fa 100644 (file)
--- a/lib/Analysis/CMakeLists.txt
+++ b/lib/Analysis/CMakeLists.txt
@@ -14,6 +14,7 @@ add_clang_library(clangAnalysis
   ReachableCode.cpp
   ScanfFormatString.cpp
   UninitializedValues.cpp
+  UninitializedValuesV2.cpp
   )
 
 add_dependencies(clangAnalysis ClangAttrClasses ClangAttrList

diff --git a/lib/Analysis/UninitializedValuesV2.cpp b/lib/Analysis/UninitializedValuesV2.cpp
new file mode 100644 (file)
index 0000000..da94490
--- /dev/null
+++ b/lib/Analysis/UninitializedValuesV2.cpp
@@ -0,0 +1,362 @@
+//==- UninitializedValuesV2.cpp - Find Uninitialized Values -----*- C++ --*-==//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements uninitialized values analysis for source-level CFGs.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/ADT/Optional.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/BitVector.h"
+#include "llvm/ADT/DenseMap.h"
+#include "clang/AST/Decl.h"
+#include "clang/Analysis/CFG.h"
+#include "clang/Analysis/Visitors/CFGRecStmtDeclVisitor.h"
+#include "clang/Analysis/Analyses/UninitializedValuesV2.h"
+
+using namespace clang;
+
+//------------------------------------------------------------------------====//
+// DeclToBit: a mapping from Decls we track to bitvector indices.
+//====------------------------------------------------------------------------//
+
+namespace {
+class DeclToBit {
+  llvm::DenseMap<const VarDecl *, unsigned> map;
+public:
+  DeclToBit() {}
+  
+  /// Compute the actual mapping from declarations to bits.
+  void computeMap(const DeclContext &dc);
+  
+  /// Return the number of declarations in the map.
+  unsigned size() const { return map.size(); }
+  
+  /// Returns the bit vector index for a given declaration.
+  llvm::Optional<unsigned> getBitVectorIndex(const VarDecl *d);
+};
+}
+
+void DeclToBit::computeMap(const DeclContext &dc) {
+  unsigned count = 0;
+  DeclContext::specific_decl_iterator<VarDecl> I(dc.decls_begin()),
+                                               E(dc.decls_end());
+  for ( ; I != E; ++I) {
+    const VarDecl *vd = *I;
+    if (vd->isLocalVarDecl() && !vd->hasGlobalStorage())
+      map[vd] = count++;
+  }
+}
+
+llvm::Optional<unsigned> DeclToBit::getBitVectorIndex(const VarDecl *d) {
+  llvm::DenseMap<const VarDecl *, unsigned>::iterator I = map.find(d);
+  if (I == map.end())
+    return llvm::Optional<unsigned>();
+  return I->second;
+}
+
+//------------------------------------------------------------------------====//
+// CFGBlockValues: dataflow values for CFG blocks.
+//====------------------------------------------------------------------------//
+
+namespace {
+class CFGBlockValues {
+  const CFG &cfg;
+  llvm::BitVector **vals;
+  llvm::BitVector scratch;
+  DeclToBit declToBit;
+public:
+  CFGBlockValues(const CFG &cfg);
+  ~CFGBlockValues();
+  
+  void computeSetOfDeclarations(const DeclContext &dc);  
+  llvm::BitVector &getBitVector(const CFGBlock *block);
+  void mergeIntoScratch(llvm::BitVector const &source, bool isFirst);
+  bool updateBitVectorWithScratch(const CFGBlock *block);
+  
+  bool hasNoDeclarations() const {
+    return declToBit.size() == 0;
+  }
+  
+  void resetScratch();
+  llvm::BitVector::reference operator[](const VarDecl *vd);
+};  
+}
+
+CFGBlockValues::CFGBlockValues(const CFG &c) : cfg(c), vals(0) {
+  unsigned n = cfg.getNumBlockIDs();
+  if (!n)
+    return;
+  vals = new llvm::BitVector*[n];
+  bzero(vals, sizeof(*vals) * n);
+}
+
+CFGBlockValues::~CFGBlockValues() {
+  unsigned n = cfg.getNumBlockIDs();
+  if (n == 0)
+    return;
+  for (unsigned i = 0; i < n; ++i)
+    delete vals[i];
+  delete [] vals;
+}
+
+void CFGBlockValues::computeSetOfDeclarations(const DeclContext &dc) {
+  declToBit.computeMap(dc);
+  scratch.resize(declToBit.size());
+}
+
+llvm::BitVector &CFGBlockValues::getBitVector(const CFGBlock *block) {
+  unsigned idx = block->getBlockID();
+  llvm::BitVector *bv = vals[idx];
+  if (!bv) {
+    bv = new llvm::BitVector(declToBit.size());
+    vals[idx] = bv;
+  }
+  return *bv;
+}
+
+void CFGBlockValues::mergeIntoScratch(llvm::BitVector const &source,
+                                      bool isFirst) {
+  if (isFirst)
+    scratch = source;
+  else
+    scratch &= source;  
+}
+
+bool CFGBlockValues::updateBitVectorWithScratch(const CFGBlock *block) {
+  llvm::BitVector &dst = getBitVector(block);
+  bool changed = (dst != scratch);
+  if (changed)
+    dst = scratch;
+  return changed;
+}
+
+void CFGBlockValues::resetScratch() {
+  scratch.reset();
+}
+
+llvm::BitVector::reference CFGBlockValues::operator[](const VarDecl *vd) {
+  const llvm::Optional<unsigned> &idx = declToBit.getBitVectorIndex(vd);
+  assert(idx.hasValue());
+  return scratch[idx.getValue()];
+}
+
+//------------------------------------------------------------------------====//
+// Worklist: worklist for dataflow analysis.
+//====------------------------------------------------------------------------//
+
+namespace {
+class DataflowWorklist {
+  llvm::SmallVector<const CFGBlock *, 20> worklist;
+  llvm::BitVector enqueuedBlocks;
+public:
+  DataflowWorklist(const CFG &cfg) : enqueuedBlocks(cfg.getNumBlockIDs()) {}
+  
+  void enqueue(const CFGBlock *block);
+  void enqueueSuccessors(const CFGBlock *block);
+  const CFGBlock *dequeue();
+  
+};
+}
+
+void DataflowWorklist::enqueue(const CFGBlock *block) {
+  unsigned idx = block->getBlockID();
+  if (enqueuedBlocks[idx])
+    return;
+  worklist.push_back(block);
+  enqueuedBlocks[idx] = true;
+}
+
+void DataflowWorklist::enqueueSuccessors(const clang::CFGBlock *block) {
+  for (CFGBlock::const_succ_iterator I = block->succ_begin(),
+       E = block->succ_end(); I != E; ++I) {
+    enqueue(*I);
+  }
+}
+
+const CFGBlock *DataflowWorklist::dequeue() {
+  if (worklist.empty())
+    return 0;
+  const CFGBlock *b = worklist.back();
+  worklist.pop_back();
+  enqueuedBlocks[b->getBlockID()] = false;
+  return b;
+}
+
+//------------------------------------------------------------------------====//
+// Transfer function for uninitialized values analysis.
+//====------------------------------------------------------------------------//
+
+static const bool Initialized = true;
+static const bool Uninitialized = false;
+
+namespace {
+class FindVarResult {
+  const VarDecl *vd;
+  const DeclRefExpr *dr;
+public:
+  FindVarResult(VarDecl *vd, DeclRefExpr *dr) : vd(vd), dr(dr) {}
+  
+  const DeclRefExpr *getDeclRefExpr() const { return dr; }
+  const VarDecl *getDecl() const { return vd; }
+};
+  
+class TransferFunctions : public CFGRecStmtVisitor<TransferFunctions> {
+  CFGBlockValues &vals;
+  const CFG &cfg;
+  UninitVariablesHandler *handler;
+public:
+  TransferFunctions(CFGBlockValues &vals, const CFG &cfg,
+                    UninitVariablesHandler *handler)
+    : vals(vals), cfg(cfg), handler(handler) {}
+  
+  const CFG &getCFG() { return cfg; }
+  void reportUninit(const DeclRefExpr *ex, const VarDecl *vd);
+  
+  void VisitDeclStmt(DeclStmt *ds);
+  void VisitUnaryOperator(UnaryOperator *uo);
+  void VisitBinaryOperator(BinaryOperator *bo);
+  void VisitCastExpr(CastExpr *ce);
+};
+}
+
+void TransferFunctions::reportUninit(const DeclRefExpr *ex,
+                                     const VarDecl *vd) {
+  if (handler) handler->handleUseOfUninitVariable(ex, vd);
+}
+
+void TransferFunctions::VisitDeclStmt(DeclStmt *ds) {
+  for (DeclStmt::decl_iterator DI = ds->decl_begin(), DE = ds->decl_end();
+       DI != DE; ++DI) {
+    if (VarDecl *vd = dyn_cast<VarDecl>(*DI)) {
+      if (vd->isLocalVarDecl() && !vd->hasGlobalStorage()) {
+        if (Stmt *init = vd->getInit()) {
+          vals[vd] = Initialized;
+          Visit(init);
+        }
+      }
+    }
+  }
+}
+
+static FindVarResult findBlockVarDecl(Expr* ex) {
+  if (DeclRefExpr* dr = dyn_cast<DeclRefExpr>(ex->IgnoreParenCasts()))
+    if (VarDecl *vd = dyn_cast<VarDecl>(dr->getDecl()))
+      if (vd->isLocalVarDecl() && !vd->hasGlobalStorage())
+        return FindVarResult(vd, dr);
+
+  return FindVarResult(0, 0);
+}
+
+void TransferFunctions::VisitBinaryOperator(clang::BinaryOperator *bo) {
+  Visit(bo->getRHS());
+  Visit(bo->getLHS());
+  if (bo->isAssignmentOp()) {
+    const FindVarResult &res = findBlockVarDecl(bo->getLHS());
+    if (const VarDecl* vd = res.getDecl()) {
+      llvm::BitVector::reference bit = vals[vd];
+      if (bit == Uninitialized) {
+        if (bo->getOpcode() != BO_Assign)
+          reportUninit(res.getDeclRefExpr(), vd);
+        bit = Initialized;
+      }
+    }
+  }
+}
+
+void TransferFunctions::VisitUnaryOperator(clang::UnaryOperator *uo) {
+  Visit(uo->getSubExpr());
+  switch (uo->getOpcode()) {
+    case clang::UO_AddrOf:
+      if (const VarDecl *vd = findBlockVarDecl(uo->getSubExpr()).getDecl())
+        vals[vd] = Initialized;
+      break;
+    case clang::UO_PostDec:
+    case clang::UO_PostInc:
+    case clang::UO_PreDec:
+    case clang::UO_PreInc: {
+      const FindVarResult &res = findBlockVarDecl(uo->getSubExpr());
+      if (const VarDecl *vd = res.getDecl()) {
+        llvm::BitVector::reference bit = vals[vd];
+        if (bit == Uninitialized) {
+          reportUninit(res.getDeclRefExpr(), vd);
+          bit = Initialized;
+        }
+      }
+      break;
+    }
+    default:
+      break;
+  }
+}
+
+void TransferFunctions::VisitCastExpr(clang::CastExpr *ce) {
+  Visit(ce->getSubExpr());
+  if (ce->getCastKind() == CK_LValueToRValue) {
+    const FindVarResult &res = findBlockVarDecl(ce->getSubExpr());
+    if (const VarDecl *vd = res.getDecl())
+      if (vals[vd] == Uninitialized)
+        reportUninit(res.getDeclRefExpr(), vd);
+  }
+}
+
+//------------------------------------------------------------------------====//
+// High-level "driver" logic for uninitialized values analysis.
+//====------------------------------------------------------------------------//
+
+static void runOnBlock(const CFGBlock *block, const CFG &cfg,
+                       CFGBlockValues &vals,
+                       UninitVariablesHandler *handler = 0) {
+  // Merge in values of predecessor blocks.    
+  vals.resetScratch();
+  bool isFirst = true;
+  for (CFGBlock::const_pred_iterator I = block->pred_begin(),
+       E = block->pred_end(); I != E; ++I) {
+    vals.mergeIntoScratch(vals.getBitVector(*I), isFirst);
+    isFirst = false;
+  }
+  // Apply the transfer function.
+  TransferFunctions tf(vals, cfg, handler);
+  for (CFGBlock::const_iterator I = block->begin(), E = block->end(); 
+       I != E; ++I) {
+    if (const CFGStmt *cs = dyn_cast<CFGStmt>(&*I)) {
+      tf.BlockStmt_Visit(cs->getStmt());
+    }
+  }
+}
+
+void clang::runUninitializedVariablesAnalysis(const DeclContext &dc,
+                                              const CFG &cfg,
+                                              UninitVariablesHandler &handler) {
+  CFGBlockValues vals(cfg);
+  vals.computeSetOfDeclarations(dc);
+  if (vals.hasNoDeclarations())
+    return;
+  DataflowWorklist worklist(cfg);
+  llvm::BitVector previouslyVisited(cfg.getNumBlockIDs());
+  
+  worklist.enqueueSuccessors(&cfg.getEntry());
+
+  while (const CFGBlock *block = worklist.dequeue()) {
+    runOnBlock(block, cfg, vals);    
+    // Did the block change?
+    bool changed = vals.updateBitVectorWithScratch(block);    
+    if (changed || !previouslyVisited[block->getBlockID()])
+      worklist.enqueueSuccessors(block);    
+    previouslyVisited[block->getBlockID()] = true;
+  }
+  
+  // Run through the blocks one more time, and report uninitialized variabes.
+  for (CFG::const_iterator BI = cfg.begin(), BE = cfg.end(); BI != BE; ++BI) {
+    runOnBlock(*BI, cfg, vals, &handler);
+  }
+}
+
+UninitVariablesHandler::~UninitVariablesHandler() {}
+

diff --git a/lib/Sema/AnalysisBasedWarnings.cpp b/lib/Sema/AnalysisBasedWarnings.cpp
index 20a503d6a56ff72d07b22cc2763476c7378ef348..14d75338e0d2d0f1eb07d0da7d07912da8aedb82 100644 (file)
--- a/lib/Sema/AnalysisBasedWarnings.cpp
+++ b/lib/Sema/AnalysisBasedWarnings.cpp
@@ -25,6 +25,7 @@
 #include "clang/Analysis/AnalysisContext.h"
 #include "clang/Analysis/CFG.h"
 #include "clang/Analysis/Analyses/ReachableCode.h"
+#include "clang/Analysis/Analyses/UninitializedValuesV2.h"
 #include "llvm/ADT/BitVector.h"
 #include "llvm/Support/Casting.h"
 
@@ -343,6 +344,23 @@ static void CheckFallThroughForBody(Sema &S, const Decl *D, const Stmt *Body,
   }
 }
 
+//===----------------------------------------------------------------------===//
+// -Wuninitialized
+//===----------------------------------------------------------------------===//
+
+namespace {
+class UninitValsDiagReporter : public UninitVariablesHandler {
+  Sema &S;
+public:
+  UninitValsDiagReporter(Sema &S) : S(S) {}
+  
+  void handleUseOfUninitVariable(const DeclRefExpr *dr, const VarDecl *vd) {
+    S.Diag(dr->getLocStart(), diag::warn_var_is_uninit)
+      << vd->getDeclName() << dr->getSourceRange();
+  }
+};
+}
+
 //===----------------------------------------------------------------------===//
 // AnalysisBasedWarnings - Worker object used by Sema to execute analysis-based
 //  warnings on a function, method, or block.
@@ -406,6 +424,18 @@ AnalysisBasedWarnings::IssueWarnings(sema::AnalysisBasedWarnings::Policy P,
   // Warning: check for unreachable code
   if (P.enableCheckUnreachable)
     CheckUnreachable(S, AC);
+  
+  if (Diags.getDiagnosticLevel(diag::warn_var_is_uninit, D->getLocStart())
+      != Diagnostic::Ignored) {
+    if (!S.getLangOptions().CPlusPlus) {
+      CFG *cfg = AC.getCFG();
+      if (cfg) {
+        UninitValsDiagReporter reporter(S);
+        runUninitializedVariablesAnalysis(*cast<DeclContext>(D), *cfg,
+                                          reporter);
+      }
+    }
+  }
 }
 
 void clang::sema::

diff --git a/test/Sema/uninit-variables.c b/test/Sema/uninit-variables.c
new file mode 100644 (file)
index 0000000..c2d98c4
--- /dev/null
+++ b/test/Sema/uninit-variables.c
@@ -0,0 +1,88 @@
+// RUN: %clang -Wuninitialized-experimental -fsyntax-only %s
+
+int test1() {
+  int x;
+  return x; // expected-warning{{use of uninitialized variable 'x'}}
+}
+
+int test2() {
+  int x = 0;
+  return x; // no-warning
+}
+
+int test3() {
+  int x;
+  x = 0;
+  return x; // no-warning
+}
+
+int test4() {
+  int x;
+  ++x; // expected-warning{{use of uninitialized variable 'x'}}
+  return x; 
+}
+
+int test5() {
+  int x, y;
+  x = y; // expected-warning{{use of uninitialized variable 'y'}}
+  return x;
+}
+
+int test6() {
+  int x;
+  x += 2; // expected-warning{{use of uninitialized variable 'x'}}
+  return x;
+}
+
+int test7(int y) {
+  int x;
+  if (y)
+    x = 1;
+  return x;  // expected-warning{{use of uninitialized variable 'x'}}
+}
+
+int test8(int y) {
+  int x;
+  if (y)
+    x = 1;
+  else
+    x = 0;
+  return x; // no-warning
+}
+
+int test9(int n) {
+  int x;
+  for (unsigned i = 0 ; i < n; ++i) {
+    if (i == n - 1)
+      break;
+    x = 1;    
+  }
+  return x; // expected-warning{{use of uninitialized variable 'x'}}
+}
+
+int test10(unsigned n) {
+  int x;
+  for (unsigned i = 0 ; i < n; ++i) {
+    x = 1;
+  }
+  return x; // expected-warning{{use of uninitialized variable 'x'}}
+}
+
+int test11(unsigned n) {
+  int x;
+  for (unsigned i = 0 ; i <= n; ++i) {
+    x = 1;
+  }
+  return x; // expected-warning{{use of uninitialized variable 'x'}}
+}
+
+void test12(unsigned n) {
+  for (unsigned i ; n ; ++i) ; // expected-warning{{use of uninitialized variable 'i'}}
+}
+
+int test13() {
+  static int i;
+  return i; // no-warning
+}
+
+