diff options
Diffstat (limited to 'contrib/llvm/tools/clang/lib/Analysis')
12 files changed, 2180 insertions, 1305 deletions
diff --git a/contrib/llvm/tools/clang/lib/Analysis/AnalysisDeclContext.cpp b/contrib/llvm/tools/clang/lib/Analysis/AnalysisDeclContext.cpp index 659cc6d..7de7f39 100644 --- a/contrib/llvm/tools/clang/lib/Analysis/AnalysisDeclContext.cpp +++ b/contrib/llvm/tools/clang/lib/Analysis/AnalysisDeclContext.cpp @@ -12,6 +12,7 @@ // //===----------------------------------------------------------------------===// +#include "clang/AST/ASTContext.h" #include "clang/AST/Decl.h" #include "clang/AST/DeclObjC.h" #include "clang/AST/DeclTemplate.h" @@ -34,11 +35,9 @@ typedef llvm::DenseMap<const void *, ManagedAnalysis *> ManagedAnalysisMap; AnalysisDeclContext::AnalysisDeclContext(AnalysisDeclContextManager *Mgr, const Decl *d, - idx::TranslationUnit *tu, const CFG::BuildOptions &buildOptions) : Manager(Mgr), D(d), - TU(tu), cfgBuildOptions(buildOptions), forcedBlkExprs(0), builtCFG(false), @@ -50,11 +49,9 @@ AnalysisDeclContext::AnalysisDeclContext(AnalysisDeclContextManager *Mgr, } AnalysisDeclContext::AnalysisDeclContext(AnalysisDeclContextManager *Mgr, - const Decl *d, - idx::TranslationUnit *tu) + const Decl *d) : Manager(Mgr), D(d), - TU(tu), forcedBlkExprs(0), builtCFG(false), builtCompleteCFG(false), @@ -184,8 +181,16 @@ void AnalysisDeclContext::dumpCFG(bool ShowColors) { } ParentMap &AnalysisDeclContext::getParentMap() { - if (!PM) + if (!PM) { PM.reset(new ParentMap(getBody())); + if (const CXXConstructorDecl *C = dyn_cast<CXXConstructorDecl>(getDecl())) { + for (CXXConstructorDecl::init_const_iterator I = C->init_begin(), + E = C->init_end(); + I != E; ++I) { + PM->addStmt((*I)->getInit()); + } + } + } return *PM; } @@ -195,11 +200,10 @@ PseudoConstantAnalysis *AnalysisDeclContext::getPseudoConstantAnalysis() { return PCA.get(); } -AnalysisDeclContext *AnalysisDeclContextManager::getContext(const Decl *D, - idx::TranslationUnit *TU) { +AnalysisDeclContext *AnalysisDeclContextManager::getContext(const Decl *D) { AnalysisDeclContext *&AC = Contexts[D]; if (!AC) - AC = new AnalysisDeclContext(this, D, TU, cfgBuildOptions); + AC = new AnalysisDeclContext(this, D, cfgBuildOptions); return AC; } @@ -209,6 +213,14 @@ AnalysisDeclContext::getStackFrame(LocationContext const *Parent, const Stmt *S, return getLocationContextManager().getStackFrame(this, Parent, S, Blk, Idx); } +const BlockInvocationContext * +AnalysisDeclContext::getBlockInvocationContext(const LocationContext *parent, + const clang::BlockDecl *BD, + const void *ContextData) { + return getLocationContextManager().getBlockInvocationContext(this, parent, + BD, ContextData); +} + LocationContextManager & AnalysisDeclContext::getLocationContextManager() { assert(Manager && "Cannot create LocationContexts without an AnalysisDeclContextManager!"); @@ -239,7 +251,7 @@ void ScopeContext::Profile(llvm::FoldingSetNodeID &ID) { } void BlockInvocationContext::Profile(llvm::FoldingSetNodeID &ID) { - Profile(ID, getAnalysisDeclContext(), getParent(), BD); + Profile(ID, getAnalysisDeclContext(), getParent(), BD, ContextData); } //===----------------------------------------------------------------------===// @@ -288,6 +300,24 @@ LocationContextManager::getScope(AnalysisDeclContext *ctx, return getLocationContext<ScopeContext, Stmt>(ctx, parent, s); } +const BlockInvocationContext * +LocationContextManager::getBlockInvocationContext(AnalysisDeclContext *ctx, + const LocationContext *parent, + const BlockDecl *BD, + const void *ContextData) { + llvm::FoldingSetNodeID ID; + BlockInvocationContext::Profile(ID, ctx, parent, BD, ContextData); + void *InsertPos; + BlockInvocationContext *L = + cast_or_null<BlockInvocationContext>(Contexts.FindNodeOrInsertPos(ID, + InsertPos)); + if (!L) { + L = new BlockInvocationContext(ctx, parent, BD, ContextData); + Contexts.InsertNode(L, InsertPos); + } + return L; +} + //===----------------------------------------------------------------------===// // LocationContext methods. //===----------------------------------------------------------------------===// @@ -302,19 +332,6 @@ const StackFrameContext *LocationContext::getCurrentStackFrame() const { return NULL; } -const StackFrameContext * -LocationContext::getStackFrameForDeclContext(const DeclContext *DC) const { - const LocationContext *LC = this; - while (LC) { - if (const StackFrameContext *SFC = dyn_cast<StackFrameContext>(LC)) { - if (cast<DeclContext>(SFC->getDecl()) == DC) - return SFC; - } - LC = LC->getParent(); - } - return NULL; -} - bool LocationContext::isParentOf(const LocationContext *LC) const { do { const LocationContext *Parent = LC->getParent(); diff --git a/contrib/llvm/tools/clang/lib/Analysis/CFG.cpp b/contrib/llvm/tools/clang/lib/Analysis/CFG.cpp index 2f1f1cb..05c5385 100644 --- a/contrib/llvm/tools/clang/lib/Analysis/CFG.cpp +++ b/contrib/llvm/tools/clang/lib/Analysis/CFG.cpp @@ -1,4 +1,4 @@ -//===--- CFG.cpp - Classes for representing and building CFGs----*- C++ -*-===// + //===--- CFG.cpp - Classes for representing and building CFGs----*- C++ -*-===// // // The LLVM Compiler Infrastructure // @@ -14,6 +14,7 @@ #include "llvm/Support/SaveAndRestore.h" #include "clang/Analysis/CFG.h" +#include "clang/AST/ASTContext.h" #include "clang/AST/DeclCXX.h" #include "clang/AST/StmtVisitor.h" #include "clang/AST/PrettyPrinter.h" @@ -312,19 +313,6 @@ private: CFGBlock *VisitAddrLabelExpr(AddrLabelExpr *A, AddStmtChoice asc); CFGBlock *VisitBinaryOperator(BinaryOperator *B, AddStmtChoice asc); CFGBlock *VisitBreakStmt(BreakStmt *B); - CFGBlock *VisitCXXCatchStmt(CXXCatchStmt *S); - CFGBlock *VisitExprWithCleanups(ExprWithCleanups *E, - AddStmtChoice asc); - CFGBlock *VisitCXXThrowExpr(CXXThrowExpr *T); - CFGBlock *VisitCXXTryStmt(CXXTryStmt *S); - CFGBlock *VisitCXXForRangeStmt(CXXForRangeStmt *S); - CFGBlock *VisitCXXBindTemporaryExpr(CXXBindTemporaryExpr *E, - AddStmtChoice asc); - CFGBlock *VisitCXXConstructExpr(CXXConstructExpr *C, AddStmtChoice asc); - CFGBlock *VisitCXXFunctionalCastExpr(CXXFunctionalCastExpr *E, - AddStmtChoice asc); - CFGBlock *VisitCXXTemporaryObjectExpr(CXXTemporaryObjectExpr *C, - AddStmtChoice asc); CFGBlock *VisitCallExpr(CallExpr *C, AddStmtChoice asc); CFGBlock *VisitCaseStmt(CaseStmt *C); CFGBlock *VisitChooseExpr(ChooseExpr *C, AddStmtChoice asc); @@ -332,31 +320,47 @@ private: CFGBlock *VisitConditionalOperator(AbstractConditionalOperator *C, AddStmtChoice asc); CFGBlock *VisitContinueStmt(ContinueStmt *C); + CFGBlock *VisitCXXBindTemporaryExpr(CXXBindTemporaryExpr *E, + AddStmtChoice asc); + CFGBlock *VisitCXXCatchStmt(CXXCatchStmt *S); + CFGBlock *VisitCXXConstructExpr(CXXConstructExpr *C, AddStmtChoice asc); + CFGBlock *VisitCXXForRangeStmt(CXXForRangeStmt *S); + CFGBlock *VisitCXXFunctionalCastExpr(CXXFunctionalCastExpr *E, + AddStmtChoice asc); + CFGBlock *VisitCXXTemporaryObjectExpr(CXXTemporaryObjectExpr *C, + AddStmtChoice asc); + CFGBlock *VisitCXXThrowExpr(CXXThrowExpr *T); + CFGBlock *VisitCXXTryStmt(CXXTryStmt *S); CFGBlock *VisitDeclStmt(DeclStmt *DS); CFGBlock *VisitDeclSubExpr(DeclStmt *DS); CFGBlock *VisitDefaultStmt(DefaultStmt *D); CFGBlock *VisitDoStmt(DoStmt *D); - CFGBlock *VisitLambdaExpr(LambdaExpr *E, AddStmtChoice asc); + CFGBlock *VisitExprWithCleanups(ExprWithCleanups *E, AddStmtChoice asc); CFGBlock *VisitForStmt(ForStmt *F); CFGBlock *VisitGotoStmt(GotoStmt *G); CFGBlock *VisitIfStmt(IfStmt *I); CFGBlock *VisitImplicitCastExpr(ImplicitCastExpr *E, AddStmtChoice asc); CFGBlock *VisitIndirectGotoStmt(IndirectGotoStmt *I); CFGBlock *VisitLabelStmt(LabelStmt *L); - CFGBlock *VisitLambdaExpr(LambdaExpr *L); + CFGBlock *VisitLambdaExpr(LambdaExpr *E, AddStmtChoice asc); + CFGBlock *VisitLogicalOperator(BinaryOperator *B); + std::pair<CFGBlock *, CFGBlock *> VisitLogicalOperator(BinaryOperator *B, + Stmt *Term, + CFGBlock *TrueBlock, + CFGBlock *FalseBlock); CFGBlock *VisitMemberExpr(MemberExpr *M, AddStmtChoice asc); CFGBlock *VisitObjCAtCatchStmt(ObjCAtCatchStmt *S); - CFGBlock *VisitObjCAutoreleasePoolStmt(ObjCAutoreleasePoolStmt *S); CFGBlock *VisitObjCAtSynchronizedStmt(ObjCAtSynchronizedStmt *S); CFGBlock *VisitObjCAtThrowStmt(ObjCAtThrowStmt *S); CFGBlock *VisitObjCAtTryStmt(ObjCAtTryStmt *S); + CFGBlock *VisitObjCAutoreleasePoolStmt(ObjCAutoreleasePoolStmt *S); CFGBlock *VisitObjCForCollectionStmt(ObjCForCollectionStmt *S); - CFGBlock *VisitReturnStmt(ReturnStmt *R); CFGBlock *VisitPseudoObjectExpr(PseudoObjectExpr *E); - CFGBlock *VisitUnaryExprOrTypeTraitExpr(UnaryExprOrTypeTraitExpr *E, - AddStmtChoice asc); + CFGBlock *VisitReturnStmt(ReturnStmt *R); CFGBlock *VisitStmtExpr(StmtExpr *S, AddStmtChoice asc); CFGBlock *VisitSwitchStmt(SwitchStmt *S); + CFGBlock *VisitUnaryExprOrTypeTraitExpr(UnaryExprOrTypeTraitExpr *E, + AddStmtChoice asc); CFGBlock *VisitUnaryOperator(UnaryOperator *U, AddStmtChoice asc); CFGBlock *VisitWhileStmt(WhileStmt *W); @@ -772,13 +776,12 @@ void CFGBuilder::addAutomaticObjDtors(LocalScope::const_iterator B, // If this destructor is marked as a no-return destructor, we need to // create a new block for the destructor which does not have as a successor // anything built thus far: control won't flow out of this block. - QualType Ty; - if ((*I)->getType()->isReferenceType()) { + QualType Ty = (*I)->getType(); + if (Ty->isReferenceType()) { Ty = getReferenceInitTemporaryType(*Context, (*I)->getInit()); - } else { - Ty = Context->getBaseElementType((*I)->getType()); } - + Ty = Context->getBaseElementType(Ty); + const CXXDestructorDecl *Dtor = Ty->getAsCXXRecordDecl()->getDestructor(); if (cast<FunctionType>(Dtor->getType())->getNoReturnAttr()) Block = createNoReturnBlock(); @@ -1070,9 +1073,6 @@ CFGBlock *CFGBuilder::Visit(Stmt * S, AddStmtChoice asc) { case Stmt::LambdaExprClass: return VisitLambdaExpr(cast<LambdaExpr>(S), asc); - case Stmt::AttributedStmtClass: - return Visit(cast<AttributedStmt>(S)->getSubStmt(), asc); - case Stmt::MemberExprClass: return VisitMemberExpr(cast<MemberExpr>(S), asc); @@ -1166,55 +1166,111 @@ CFGBlock *CFGBuilder::VisitUnaryOperator(UnaryOperator *U, return Visit(U->getSubExpr(), AddStmtChoice()); } -CFGBlock *CFGBuilder::VisitBinaryOperator(BinaryOperator *B, - AddStmtChoice asc) { - if (B->isLogicalOp()) { // && or || - CFGBlock *ConfluenceBlock = Block ? Block : createBlock(); - appendStmt(ConfluenceBlock, B); +CFGBlock *CFGBuilder::VisitLogicalOperator(BinaryOperator *B) { + CFGBlock *ConfluenceBlock = Block ? Block : createBlock(); + appendStmt(ConfluenceBlock, B); - if (badCFG) - return 0; + if (badCFG) + return 0; - // create the block evaluating the LHS - CFGBlock *LHSBlock = createBlock(false); - LHSBlock->setTerminator(B); + return VisitLogicalOperator(B, 0, ConfluenceBlock, ConfluenceBlock).first; +} - // create the block evaluating the RHS - Succ = ConfluenceBlock; - Block = NULL; - CFGBlock *RHSBlock = addStmt(B->getRHS()); +std::pair<CFGBlock*, CFGBlock*> +CFGBuilder::VisitLogicalOperator(BinaryOperator *B, + Stmt *Term, + CFGBlock *TrueBlock, + CFGBlock *FalseBlock) { - if (RHSBlock) { - if (badCFG) - return 0; - } else { - // Create an empty block for cases where the RHS doesn't require - // any explicit statements in the CFG. - RHSBlock = createBlock(); + // Introspect the RHS. If it is a nested logical operation, we recursively + // build the CFG using this function. Otherwise, resort to default + // CFG construction behavior. + Expr *RHS = B->getRHS()->IgnoreParens(); + CFGBlock *RHSBlock, *ExitBlock; + + do { + if (BinaryOperator *B_RHS = dyn_cast<BinaryOperator>(RHS)) + if (B_RHS->isLogicalOp()) { + llvm::tie(RHSBlock, ExitBlock) = + VisitLogicalOperator(B_RHS, Term, TrueBlock, FalseBlock); + break; + } + + // The RHS is not a nested logical operation. Don't push the terminator + // down further, but instead visit RHS and construct the respective + // pieces of the CFG, and link up the RHSBlock with the terminator + // we have been provided. + ExitBlock = RHSBlock = createBlock(false); + + if (!Term) { + assert(TrueBlock == FalseBlock); + addSuccessor(RHSBlock, TrueBlock); + } + else { + RHSBlock->setTerminator(Term); + TryResult KnownVal = tryEvaluateBool(RHS); + addSuccessor(RHSBlock, KnownVal.isFalse() ? NULL : TrueBlock); + addSuccessor(RHSBlock, KnownVal.isTrue() ? NULL : FalseBlock); } - // Generate the blocks for evaluating the LHS. - Block = LHSBlock; - CFGBlock *EntryLHSBlock = addStmt(B->getLHS()); + Block = RHSBlock; + RHSBlock = addStmt(RHS); + } + while (false); - // See if this is a known constant. - TryResult KnownVal = tryEvaluateBool(B->getLHS()); - if (KnownVal.isKnown() && (B->getOpcode() == BO_LOr)) - KnownVal.negate(); + if (badCFG) + return std::make_pair((CFGBlock*)0, (CFGBlock*)0); + + // Generate the blocks for evaluating the LHS. + Expr *LHS = B->getLHS()->IgnoreParens(); + + if (BinaryOperator *B_LHS = dyn_cast<BinaryOperator>(LHS)) + if (B_LHS->isLogicalOp()) { + if (B->getOpcode() == BO_LOr) + FalseBlock = RHSBlock; + else + TrueBlock = RHSBlock; - // Now link the LHSBlock with RHSBlock. - if (B->getOpcode() == BO_LOr) { - addSuccessor(LHSBlock, KnownVal.isTrue() ? NULL : ConfluenceBlock); - addSuccessor(LHSBlock, KnownVal.isFalse() ? NULL : RHSBlock); - } else { - assert(B->getOpcode() == BO_LAnd); - addSuccessor(LHSBlock, KnownVal.isFalse() ? NULL : RHSBlock); - addSuccessor(LHSBlock, KnownVal.isTrue() ? NULL : ConfluenceBlock); + // For the LHS, treat 'B' as the terminator that we want to sink + // into the nested branch. The RHS always gets the top-most + // terminator. + return VisitLogicalOperator(B_LHS, B, TrueBlock, FalseBlock); } - return EntryLHSBlock; + // Create the block evaluating the LHS. + // This contains the '&&' or '||' as the terminator. + CFGBlock *LHSBlock = createBlock(false); + LHSBlock->setTerminator(B); + + Block = LHSBlock; + CFGBlock *EntryLHSBlock = addStmt(LHS); + + if (badCFG) + return std::make_pair((CFGBlock*)0, (CFGBlock*)0); + + // See if this is a known constant. + TryResult KnownVal = tryEvaluateBool(LHS); + + // Now link the LHSBlock with RHSBlock. + if (B->getOpcode() == BO_LOr) { + addSuccessor(LHSBlock, KnownVal.isFalse() ? NULL : TrueBlock); + addSuccessor(LHSBlock, KnownVal.isTrue() ? NULL : RHSBlock); + } else { + assert(B->getOpcode() == BO_LAnd); + addSuccessor(LHSBlock, KnownVal.isFalse() ? NULL : RHSBlock); + addSuccessor(LHSBlock, KnownVal.isTrue() ? NULL : FalseBlock); } + return std::make_pair(EntryLHSBlock, ExitBlock); +} + + +CFGBlock *CFGBuilder::VisitBinaryOperator(BinaryOperator *B, + AddStmtChoice asc) { + // && or || + if (B->isLogicalOp()) + return VisitLogicalOperator(B); + if (B->getOpcode() == BO_Comma) { // , autoCreateBlock(); appendStmt(Block, B); @@ -1284,7 +1340,7 @@ static bool CanThrow(Expr *E, ASTContext &Ctx) { const FunctionType *FT = Ty->getAs<FunctionType>(); if (FT) { if (const FunctionProtoType *Proto = dyn_cast<FunctionProtoType>(FT)) - if (Proto->getExceptionSpecType() != EST_Uninstantiated && + if (!isUnresolvedExceptionSpec(Proto->getExceptionSpecType()) && Proto->isNothrow(Ctx)) return false; } @@ -1435,6 +1491,12 @@ CFGBlock *CFGBuilder::VisitConditionalOperator(AbstractConditionalOperator *C, if (badCFG) return 0; + // If the condition is a logical '&&' or '||', build a more accurate CFG. + if (BinaryOperator *Cond = + dyn_cast<BinaryOperator>(C->getCond()->IgnoreParens())) + if (Cond->isLogicalOp()) + return VisitLogicalOperator(Cond, C, LHSBlock, RHSBlock).first; + // Create the block that will contain the condition. Block = createBlock(false); @@ -1471,11 +1533,10 @@ CFGBlock *CFGBuilder::VisitDeclStmt(DeclStmt *DS) { CFGBlock *B = 0; - // FIXME: Add a reverse iterator for DeclStmt to avoid this extra copy. - typedef SmallVector<Decl*,10> BufTy; - BufTy Buf(DS->decl_begin(), DS->decl_end()); - - for (BufTy::reverse_iterator I = Buf.rbegin(), E = Buf.rend(); I != E; ++I) { + // Build an individual DeclStmt for each decl. + for (DeclStmt::reverse_decl_iterator I = DS->decl_rbegin(), + E = DS->decl_rend(); + I != E; ++I) { // Get the alignment of the new DeclStmt, padding out to >=8 bytes. unsigned A = llvm::AlignOf<DeclStmt>::Alignment < 8 ? 8 : llvm::AlignOf<DeclStmt>::Alignment; @@ -1645,6 +1706,19 @@ CFGBlock *CFGBuilder::VisitIfStmt(IfStmt *I) { } } + // Specially handle "if (expr1 || ...)" and "if (expr1 && ...)" by + // having these handle the actual control-flow jump. Note that + // if we introduce a condition variable, e.g. "if (int x = exp1 || exp2)" + // we resort to the old control-flow behavior. This special handling + // removes infeasible paths from the control-flow graph by having the + // control-flow transfer of '&&' or '||' go directly into the then/else + // blocks directly. + if (!I->getConditionVariable()) + if (BinaryOperator *Cond = + dyn_cast<BinaryOperator>(I->getCond()->IgnoreParens())) + if (Cond->isLogicalOp()) + return VisitLogicalOperator(Cond, I, ThenBlock, ElseBlock).first; + // Now create a new block containing the if statement. Block = createBlock(false); @@ -1795,75 +1869,26 @@ CFGBlock *CFGBuilder::VisitForStmt(ForStmt *F) { SaveAndRestore<JumpTarget> save_break(BreakJumpTarget); BreakJumpTarget = JumpTarget(LoopSuccessor, ScopePos); - // Because of short-circuit evaluation, the condition of the loop can span - // multiple basic blocks. Thus we need the "Entry" and "Exit" blocks that - // evaluate the condition. - CFGBlock *ExitConditionBlock = createBlock(false); - CFGBlock *EntryConditionBlock = ExitConditionBlock; - - // Set the terminator for the "exit" condition block. - ExitConditionBlock->setTerminator(F); - - // Now add the actual condition to the condition block. Because the condition - // itself may contain control-flow, new blocks may be created. - if (Stmt *C = F->getCond()) { - Block = ExitConditionBlock; - EntryConditionBlock = addStmt(C); - if (badCFG) - return 0; - assert(Block == EntryConditionBlock || - (Block == 0 && EntryConditionBlock == Succ)); - - // If this block contains a condition variable, add both the condition - // variable and initializer to the CFG. - if (VarDecl *VD = F->getConditionVariable()) { - if (Expr *Init = VD->getInit()) { - autoCreateBlock(); - appendStmt(Block, F->getConditionVariableDeclStmt()); - EntryConditionBlock = addStmt(Init); - assert(Block == EntryConditionBlock); - } - } - - if (Block) { - if (badCFG) - return 0; - } - } - - // The condition block is the implicit successor for the loop body as well as - // any code above the loop. - Succ = EntryConditionBlock; - - // See if this is a known constant. - TryResult KnownVal(true); - - if (F->getCond()) - KnownVal = tryEvaluateBool(F->getCond()); + CFGBlock *BodyBlock = 0, *TransitionBlock = 0; // Now create the loop body. { assert(F->getBody()); - // Save the current values for Block, Succ, and continue targets. - SaveAndRestore<CFGBlock*> save_Block(Block), save_Succ(Succ); - SaveAndRestore<JumpTarget> save_continue(ContinueJumpTarget); + // Save the current values for Block, Succ, continue and break targets. + SaveAndRestore<CFGBlock*> save_Block(Block), save_Succ(Succ); + SaveAndRestore<JumpTarget> save_continue(ContinueJumpTarget); - // Create a new block to contain the (bottom) of the loop body. - Block = NULL; - - // Loop body should end with destructor of Condition variable (if any). - addAutomaticObjDtors(ScopePos, LoopBeginScopePos, F); + // Create an empty block to represent the transition block for looping back + // to the head of the loop. If we have increment code, it will + // go in this block as well. + Block = Succ = TransitionBlock = createBlock(false); + TransitionBlock->setLoopTarget(F); if (Stmt *I = F->getInc()) { // Generate increment code in its own basic block. This is the target of // continue statements. Succ = addStmt(I); - } else { - // No increment code. Create a special, empty, block that is used as the - // target block for "looping back" to the start of the loop. - assert(Succ == EntryConditionBlock); - Succ = Block ? Block : createBlock(); } // Finish up the increment (or empty) block if it hasn't been already. @@ -1874,11 +1899,13 @@ CFGBlock *CFGBuilder::VisitForStmt(ForStmt *F) { Block = 0; } - ContinueJumpTarget = JumpTarget(Succ, ContinueScopePos); + // The starting block for the loop increment is the block that should + // represent the 'loop target' for looping back to the start of the loop. + ContinueJumpTarget = JumpTarget(Succ, ContinueScopePos); + ContinueJumpTarget.block->setLoopTarget(F); - // The starting block for the loop increment is the block that should - // represent the 'loop target' for looping back to the start of the loop. - ContinueJumpTarget.block->setLoopTarget(F); + // Loop body should end with destructor of Condition variable (if any). + addAutomaticObjDtors(ScopePos, LoopBeginScopePos, F); // If body is not a compound statement create implicit scope // and add destructors. @@ -1887,20 +1914,79 @@ CFGBlock *CFGBuilder::VisitForStmt(ForStmt *F) { // Now populate the body block, and in the process create new blocks as we // walk the body of the loop. - CFGBlock *BodyBlock = addStmt(F->getBody()); + BodyBlock = addStmt(F->getBody()); - if (!BodyBlock) - BodyBlock = ContinueJumpTarget.block;//can happen for "for (...;...;...);" + if (!BodyBlock) { + // In the case of "for (...;...;...);" we can have a null BodyBlock. + // Use the continue jump target as the proxy for the body. + BodyBlock = ContinueJumpTarget.block; + } else if (badCFG) return 0; + } + + // Because of short-circuit evaluation, the condition of the loop can span + // multiple basic blocks. Thus we need the "Entry" and "Exit" blocks that + // evaluate the condition. + CFGBlock *EntryConditionBlock = 0, *ExitConditionBlock = 0; + + do { + Expr *C = F->getCond(); + + // Specially handle logical operators, which have a slightly + // more optimal CFG representation. + if (BinaryOperator *Cond = + dyn_cast_or_null<BinaryOperator>(C ? C->IgnoreParens() : 0)) + if (Cond->isLogicalOp()) { + llvm::tie(EntryConditionBlock, ExitConditionBlock) = + VisitLogicalOperator(Cond, F, BodyBlock, LoopSuccessor); + break; + } - // This new body block is a successor to our "exit" condition block. + // The default case when not handling logical operators. + EntryConditionBlock = ExitConditionBlock = createBlock(false); + ExitConditionBlock->setTerminator(F); + + // See if this is a known constant. + TryResult KnownVal(true); + + if (C) { + // Now add the actual condition to the condition block. + // Because the condition itself may contain control-flow, new blocks may + // be created. Thus we update "Succ" after adding the condition. + Block = ExitConditionBlock; + EntryConditionBlock = addStmt(C); + + // If this block contains a condition variable, add both the condition + // variable and initializer to the CFG. + if (VarDecl *VD = F->getConditionVariable()) { + if (Expr *Init = VD->getInit()) { + autoCreateBlock(); + appendStmt(Block, F->getConditionVariableDeclStmt()); + EntryConditionBlock = addStmt(Init); + assert(Block == EntryConditionBlock); + } + } + + if (Block && badCFG) + return 0; + + KnownVal = tryEvaluateBool(C); + } + + // Add the loop body entry as a successor to the condition. addSuccessor(ExitConditionBlock, KnownVal.isFalse() ? NULL : BodyBlock); - } + // Link up the condition block with the code that follows the loop. (the + // false branch). + addSuccessor(ExitConditionBlock, KnownVal.isTrue() ? NULL : LoopSuccessor); - // Link up the condition block with the code that follows the loop. (the - // false branch). - addSuccessor(ExitConditionBlock, KnownVal.isTrue() ? NULL : LoopSuccessor); + } while (false); + + // Link up the loop-back block to the entry condition block. + addSuccessor(TransitionBlock, EntryConditionBlock); + + // The condition block is the implicit successor for any code above the loop. + Succ = EntryConditionBlock; // If the loop contains initialization, create a new block for those // statements. This block can also contain statements that precede the loop. @@ -2108,74 +2194,30 @@ CFGBlock *CFGBuilder::VisitWhileStmt(WhileStmt *W) { return 0; LoopSuccessor = Block; Block = 0; - } else + } else { LoopSuccessor = Succ; - - // Because of short-circuit evaluation, the condition of the loop can span - // multiple basic blocks. Thus we need the "Entry" and "Exit" blocks that - // evaluate the condition. - CFGBlock *ExitConditionBlock = createBlock(false); - CFGBlock *EntryConditionBlock = ExitConditionBlock; - - // Set the terminator for the "exit" condition block. - ExitConditionBlock->setTerminator(W); - - // Now add the actual condition to the condition block. Because the condition - // itself may contain control-flow, new blocks may be created. Thus we update - // "Succ" after adding the condition. - if (Stmt *C = W->getCond()) { - Block = ExitConditionBlock; - EntryConditionBlock = addStmt(C); - // The condition might finish the current 'Block'. - Block = EntryConditionBlock; - - // If this block contains a condition variable, add both the condition - // variable and initializer to the CFG. - if (VarDecl *VD = W->getConditionVariable()) { - if (Expr *Init = VD->getInit()) { - autoCreateBlock(); - appendStmt(Block, W->getConditionVariableDeclStmt()); - EntryConditionBlock = addStmt(Init); - assert(Block == EntryConditionBlock); - } - } - - if (Block) { - if (badCFG) - return 0; - } } - // The condition block is the implicit successor for the loop body as well as - // any code above the loop. - Succ = EntryConditionBlock; - - // See if this is a known constant. - const TryResult& KnownVal = tryEvaluateBool(W->getCond()); + CFGBlock *BodyBlock = 0, *TransitionBlock = 0; // Process the loop body. { assert(W->getBody()); - // Save the current values for Block, Succ, and continue and break targets + // Save the current values for Block, Succ, continue and break targets. SaveAndRestore<CFGBlock*> save_Block(Block), save_Succ(Succ); SaveAndRestore<JumpTarget> save_continue(ContinueJumpTarget), - save_break(BreakJumpTarget); + save_break(BreakJumpTarget); // Create an empty block to represent the transition block for looping back // to the head of the loop. - Block = 0; - assert(Succ == EntryConditionBlock); - Succ = createBlock(); - Succ->setLoopTarget(W); + Succ = TransitionBlock = createBlock(false); + TransitionBlock->setLoopTarget(W); ContinueJumpTarget = JumpTarget(Succ, LoopBeginScopePos); // All breaks should go to the code following the loop. BreakJumpTarget = JumpTarget(LoopSuccessor, ScopePos); - // NULL out Block to force lazy instantiation of blocks for the body. - Block = NULL; - // Loop body should end with destructor of Condition variable (if any). addAutomaticObjDtors(ScopePos, LoopBeginScopePos, W); @@ -2185,22 +2227,69 @@ CFGBlock *CFGBuilder::VisitWhileStmt(WhileStmt *W) { addLocalScopeAndDtors(W->getBody()); // Create the body. The returned block is the entry to the loop body. - CFGBlock *BodyBlock = addStmt(W->getBody()); + BodyBlock = addStmt(W->getBody()); if (!BodyBlock) BodyBlock = ContinueJumpTarget.block; // can happen for "while(...) ;" - else if (Block) { - if (badCFG) - return 0; + else if (Block && badCFG) + return 0; + } + + // Because of short-circuit evaluation, the condition of the loop can span + // multiple basic blocks. Thus we need the "Entry" and "Exit" blocks that + // evaluate the condition. + CFGBlock *EntryConditionBlock = 0, *ExitConditionBlock = 0; + + do { + Expr *C = W->getCond(); + + // Specially handle logical operators, which have a slightly + // more optimal CFG representation. + if (BinaryOperator *Cond = dyn_cast<BinaryOperator>(C->IgnoreParens())) + if (Cond->isLogicalOp()) { + llvm::tie(EntryConditionBlock, ExitConditionBlock) = + VisitLogicalOperator(Cond, W, BodyBlock, + LoopSuccessor); + break; + } + + // The default case when not handling logical operators. + EntryConditionBlock = ExitConditionBlock = createBlock(false); + ExitConditionBlock->setTerminator(W); + + // Now add the actual condition to the condition block. + // Because the condition itself may contain control-flow, new blocks may + // be created. Thus we update "Succ" after adding the condition. + Block = ExitConditionBlock; + Block = EntryConditionBlock = addStmt(C); + + // If this block contains a condition variable, add both the condition + // variable and initializer to the CFG. + if (VarDecl *VD = W->getConditionVariable()) { + if (Expr *Init = VD->getInit()) { + autoCreateBlock(); + appendStmt(Block, W->getConditionVariableDeclStmt()); + EntryConditionBlock = addStmt(Init); + assert(Block == EntryConditionBlock); + } } + if (Block && badCFG) + return 0; + + // See if this is a known constant. + const TryResult& KnownVal = tryEvaluateBool(C); + // Add the loop body entry as a successor to the condition. addSuccessor(ExitConditionBlock, KnownVal.isFalse() ? NULL : BodyBlock); - } + // Link up the condition block with the code that follows the loop. (the + // false branch). + addSuccessor(ExitConditionBlock, KnownVal.isTrue() ? NULL : LoopSuccessor); - // Link up the condition block with the code that follows the loop. (the - // false branch). - addSuccessor(ExitConditionBlock, KnownVal.isTrue() ? NULL : LoopSuccessor); + } while(false); + + // Link up the loop-back block to the entry condition block. + addSuccessor(TransitionBlock, EntryConditionBlock); // There can be no more statements in the condition block since we loop back // to this block. NULL out Block to force lazy creation of another block. @@ -3203,8 +3292,8 @@ CFGImplicitDtor::getDestructorDecl(ASTContext &astContext) const { } bool CFGImplicitDtor::isNoReturn(ASTContext &astContext) const { - if (const CXXDestructorDecl *cdecl = getDestructorDecl(astContext)) { - QualType ty = cdecl->getType(); + if (const CXXDestructorDecl *decl = getDestructorDecl(astContext)) { + QualType ty = decl->getType(); return cast<FunctionType>(ty)->getNoReturnAttr(); } return false; @@ -3631,8 +3720,7 @@ static void print_elem(raw_ostream &OS, StmtPrinterHelper* Helper, const Type* T = VD->getType().getTypePtr(); if (const ReferenceType* RT = T->getAs<ReferenceType>()) T = RT->getPointeeType().getTypePtr(); - else if (const Type *ET = T->getArrayElementTypeNoTypeQual()) - T = ET; + T = T->getBaseElementTypeUnsafe(); OS << ".~" << T->getAsCXXRecordDecl()->getName().str() << "()"; OS << " (Implicit destructor)\n"; @@ -3644,11 +3732,7 @@ static void print_elem(raw_ostream &OS, StmtPrinterHelper* Helper, } else if (const CFGMemberDtor *ME = E.getAs<CFGMemberDtor>()) { const FieldDecl *FD = ME->getFieldDecl(); - - const Type *T = FD->getType().getTypePtr(); - if (const Type *ET = T->getArrayElementTypeNoTypeQual()) - T = ET; - + const Type *T = FD->getType()->getBaseElementTypeUnsafe(); OS << "this->" << FD->getName(); OS << ".~" << T->getAsCXXRecordDecl()->getName() << "()"; OS << " (Member object destructor)\n"; diff --git a/contrib/llvm/tools/clang/lib/Analysis/CallGraph.cpp b/contrib/llvm/tools/clang/lib/Analysis/CallGraph.cpp index 96a16c3..6b75956 100644 --- a/contrib/llvm/tools/clang/lib/Analysis/CallGraph.cpp +++ b/contrib/llvm/tools/clang/lib/Analysis/CallGraph.cpp @@ -25,12 +25,11 @@ namespace { /// given function body. class CGBuilder : public StmtVisitor<CGBuilder> { CallGraph *G; - const Decl *FD; CallGraphNode *CallerNode; public: - CGBuilder(CallGraph *g, const Decl *D, CallGraphNode *N) - : G(g), FD(D), CallerNode(N) {} + CGBuilder(CallGraph *g, CallGraphNode *N) + : G(g), CallerNode(N) {} void VisitStmt(Stmt *S) { VisitChildren(S); } @@ -99,7 +98,7 @@ void CallGraph::addNodeForDecl(Decl* D, bool IsGlobal) { Root->addCallee(Node, this); // Process all the calls by this function as well. - CGBuilder builder(this, D, Node); + CGBuilder builder(this, Node); if (Stmt *Body = D->getBody()) builder.Visit(Body); } diff --git a/contrib/llvm/tools/clang/lib/Analysis/CocoaConventions.cpp b/contrib/llvm/tools/clang/lib/Analysis/CocoaConventions.cpp index 7e9e38f..ce973af 100644 --- a/contrib/llvm/tools/clang/lib/Analysis/CocoaConventions.cpp +++ b/contrib/llvm/tools/clang/lib/Analysis/CocoaConventions.cpp @@ -17,6 +17,8 @@ #include "clang/AST/DeclObjC.h" #include "llvm/ADT/StringExtras.h" #include "llvm/Support/ErrorHandling.h" +#include <cctype> + using namespace clang; using namespace ento; diff --git a/contrib/llvm/tools/clang/lib/Analysis/FormatString.cpp b/contrib/llvm/tools/clang/lib/Analysis/FormatString.cpp index 51fac49..ff2f777 100644 --- a/contrib/llvm/tools/clang/lib/Analysis/FormatString.cpp +++ b/contrib/llvm/tools/clang/lib/Analysis/FormatString.cpp @@ -15,7 +15,7 @@ #include "FormatStringParsing.h" #include "clang/Basic/LangOptions.h" -using clang::analyze_format_string::ArgTypeResult; +using clang::analyze_format_string::ArgType; using clang::analyze_format_string::FormatStringHandler; using clang::analyze_format_string::FormatSpecifier; using clang::analyze_format_string::LengthModifier; @@ -229,18 +229,34 @@ clang::analyze_format_string::ParseLengthModifier(FormatSpecifier &FS, } //===----------------------------------------------------------------------===// -// Methods on ArgTypeResult. +// Methods on ArgType. //===----------------------------------------------------------------------===// -bool ArgTypeResult::matchesType(ASTContext &C, QualType argTy) const { +bool ArgType::matchesType(ASTContext &C, QualType argTy) const { + if (Ptr) { + // It has to be a pointer. + const PointerType *PT = argTy->getAs<PointerType>(); + if (!PT) + return false; + + // We cannot write through a const qualified pointer. + if (PT->getPointeeType().isConstQualified()) + return false; + + argTy = PT->getPointeeType(); + } + switch (K) { case InvalidTy: - llvm_unreachable("ArgTypeResult must be valid"); + llvm_unreachable("ArgType must be valid"); case UnknownTy: return true; case AnyCharTy: { + if (const EnumType *ETy = argTy->getAs<EnumType>()) + argTy = ETy->getDecl()->getIntegerType(); + if (const BuiltinType *BT = argTy->getAs<BuiltinType>()) switch (BT->getKind()) { default: @@ -255,7 +271,10 @@ bool ArgTypeResult::matchesType(ASTContext &C, QualType argTy) const { } case SpecificTy: { + if (const EnumType *ETy = argTy->getAs<EnumType>()) + argTy = ETy->getDecl()->getIntegerType(); argTy = C.getCanonicalType(argTy).getUnqualifiedType(); + if (T == argTy) return true; // Check for "compatible types". @@ -265,10 +284,9 @@ bool ArgTypeResult::matchesType(ASTContext &C, QualType argTy) const { break; case BuiltinType::Char_S: case BuiltinType::SChar: - return T == C.UnsignedCharTy; case BuiltinType::Char_U: case BuiltinType::UChar: - return T == C.SignedCharTy; + return T == C.UnsignedCharTy || T == C.SignedCharTy; case BuiltinType::Short: return T == C.UnsignedShortTy; case BuiltinType::UShort: @@ -319,20 +337,21 @@ bool ArgTypeResult::matchesType(ASTContext &C, QualType argTy) const { } case WIntTy: { - // Instead of doing a lookup for the definition of 'wint_t' (which - // is defined by the system headers) instead see if wchar_t and - // the argument type promote to the same type. - QualType PromoWChar = - C.getWCharType()->isPromotableIntegerType() - ? C.getPromotedIntegerType(C.getWCharType()) : C.getWCharType(); + QualType PromoArg = argTy->isPromotableIntegerType() ? C.getPromotedIntegerType(argTy) : argTy; - PromoWChar = C.getCanonicalType(PromoWChar).getUnqualifiedType(); + QualType WInt = C.getCanonicalType(C.getWIntType()).getUnqualifiedType(); PromoArg = C.getCanonicalType(PromoArg).getUnqualifiedType(); - return PromoWChar == PromoArg; + // If the promoted argument is the corresponding signed type of the + // wint_t type, then it should match. + if (PromoArg->hasSignedIntegerRepresentation() && + C.getCorrespondingUnsignedType(PromoArg) == WInt) + return true; + + return WInt == PromoArg; } case CPointerTy: @@ -358,40 +377,63 @@ bool ArgTypeResult::matchesType(ASTContext &C, QualType argTy) const { } } - llvm_unreachable("Invalid ArgTypeResult Kind!"); + llvm_unreachable("Invalid ArgType Kind!"); } -QualType ArgTypeResult::getRepresentativeType(ASTContext &C) const { +QualType ArgType::getRepresentativeType(ASTContext &C) const { + QualType Res; switch (K) { case InvalidTy: - llvm_unreachable("No representative type for Invalid ArgTypeResult"); + llvm_unreachable("No representative type for Invalid ArgType"); case UnknownTy: - return QualType(); + llvm_unreachable("No representative type for Unknown ArgType"); case AnyCharTy: - return C.CharTy; + Res = C.CharTy; + break; case SpecificTy: - return T; + Res = T; + break; case CStrTy: - return C.getPointerType(C.CharTy); + Res = C.getPointerType(C.CharTy); + break; case WCStrTy: - return C.getPointerType(C.getWCharType()); + Res = C.getPointerType(C.getWCharType()); + break; case ObjCPointerTy: - return C.ObjCBuiltinIdTy; + Res = C.ObjCBuiltinIdTy; + break; case CPointerTy: - return C.VoidPtrTy; + Res = C.VoidPtrTy; + break; case WIntTy: { - QualType WC = C.getWCharType(); - return WC->isPromotableIntegerType() ? C.getPromotedIntegerType(WC) : WC; + Res = C.getWIntType(); + break; } } - llvm_unreachable("Invalid ArgTypeResult Kind!"); + if (Ptr) + Res = C.getPointerType(Res); + return Res; } -std::string ArgTypeResult::getRepresentativeTypeName(ASTContext &C) const { +std::string ArgType::getRepresentativeTypeName(ASTContext &C) const { std::string S = getRepresentativeType(C).getAsString(); - if (Name && S != Name) - return std::string("'") + Name + "' (aka '" + S + "')"; + + std::string Alias; + if (Name) { + // Use a specific name for this type, e.g. "size_t". + Alias = Name; + if (Ptr) { + // If ArgType is actually a pointer to T, append an asterisk. + Alias += (Alias[Alias.size()-1] == '*') ? "*" : " *"; + } + // If Alias is the same as the underlying type, e.g. wchar_t, then drop it. + if (S == Alias) + Alias.clear(); + } + + if (!Alias.empty()) + return std::string("'") + Alias + "' (aka '" + S + "')"; return std::string("'") + S + "'"; } @@ -400,7 +442,7 @@ std::string ArgTypeResult::getRepresentativeTypeName(ASTContext &C) const { // Methods on OptionalAmount. //===----------------------------------------------------------------------===// -ArgTypeResult +ArgType analyze_format_string::OptionalAmount::getArgType(ASTContext &Ctx) const { return Ctx.IntTy; } @@ -686,3 +728,37 @@ bool FormatSpecifier::hasStandardLengthConversionCombination() const { } return true; } + +bool FormatSpecifier::namedTypeToLengthModifier(QualType QT, + LengthModifier &LM) { + assert(isa<TypedefType>(QT) && "Expected a TypedefType"); + const TypedefNameDecl *Typedef = cast<TypedefType>(QT)->getDecl(); + + for (;;) { + const IdentifierInfo *Identifier = Typedef->getIdentifier(); + if (Identifier->getName() == "size_t") { + LM.setKind(LengthModifier::AsSizeT); + return true; + } else if (Identifier->getName() == "ssize_t") { + // Not C99, but common in Unix. + LM.setKind(LengthModifier::AsSizeT); + return true; + } else if (Identifier->getName() == "intmax_t") { + LM.setKind(LengthModifier::AsIntMax); + return true; + } else if (Identifier->getName() == "uintmax_t") { + LM.setKind(LengthModifier::AsIntMax); + return true; + } else if (Identifier->getName() == "ptrdiff_t") { + LM.setKind(LengthModifier::AsPtrDiff); + return true; + } + + QualType T = Typedef->getUnderlyingType(); + if (!isa<TypedefType>(T)) + break; + + Typedef = cast<TypedefType>(T)->getDecl(); + } + return false; +} diff --git a/contrib/llvm/tools/clang/lib/Analysis/LiveVariables.cpp b/contrib/llvm/tools/clang/lib/Analysis/LiveVariables.cpp index ff6607d..38f8199 100644 --- a/contrib/llvm/tools/clang/lib/Analysis/LiveVariables.cpp +++ b/contrib/llvm/tools/clang/lib/Analysis/LiveVariables.cpp @@ -284,6 +284,14 @@ void TransferFunctions::Visit(Stmt *S) { } break; } + case Stmt::ObjCMessageExprClass: { + // In calls to super, include the implicit "self" pointer as being live. + ObjCMessageExpr *CE = cast<ObjCMessageExpr>(S); + if (CE->getReceiverKind() == ObjCMessageExpr::SuperInstance) + val.liveDecls = LV.DSetFact.add(val.liveDecls, + LV.analysisContext.getSelfDecl()); + break; + } case Stmt::DeclStmtClass: { const DeclStmt *DS = cast<DeclStmt>(S); if (const VarDecl *VD = dyn_cast<VarDecl>(DS->getSingleDecl())) { @@ -455,6 +463,12 @@ LiveVariablesImpl::runOnBlock(const CFGBlock *block, for (CFGBlock::const_reverse_iterator it = block->rbegin(), ei = block->rend(); it != ei; ++it) { const CFGElement &elem = *it; + + if (const CFGAutomaticObjDtor *Dtor = dyn_cast<CFGAutomaticObjDtor>(&elem)){ + val.liveDecls = DSetFact.add(val.liveDecls, Dtor->getVarDecl()); + continue; + } + if (!isa<CFGStmt>(elem)) continue; @@ -486,6 +500,11 @@ LiveVariables::computeLiveness(AnalysisDeclContext &AC, if (!cfg) return 0; + // The analysis currently has scalability issues for very large CFGs. + // Bail out if it looks too large. + if (cfg->getNumBlockIDs() > 300000) + return 0; + LiveVariablesImpl *LV = new LiveVariablesImpl(AC, killAtAssign); // Construct the dataflow worklist. Enqueue the exit block as the diff --git a/contrib/llvm/tools/clang/lib/Analysis/PrintfFormatString.cpp b/contrib/llvm/tools/clang/lib/Analysis/PrintfFormatString.cpp index 4b2a19e..2b350ce 100644 --- a/contrib/llvm/tools/clang/lib/Analysis/PrintfFormatString.cpp +++ b/contrib/llvm/tools/clang/lib/Analysis/PrintfFormatString.cpp @@ -15,7 +15,7 @@ #include "clang/Analysis/Analyses/FormatString.h" #include "FormatStringParsing.h" -using clang::analyze_format_string::ArgTypeResult; +using clang::analyze_format_string::ArgType; using clang::analyze_format_string::FormatStringHandler; using clang::analyze_format_string::LengthModifier; using clang::analyze_format_string::OptionalAmount; @@ -249,20 +249,20 @@ bool clang::analyze_format_string::ParsePrintfString(FormatStringHandler &H, // Methods on PrintfSpecifier. //===----------------------------------------------------------------------===// -ArgTypeResult PrintfSpecifier::getArgType(ASTContext &Ctx, - bool IsObjCLiteral) const { +ArgType PrintfSpecifier::getArgType(ASTContext &Ctx, + bool IsObjCLiteral) const { const PrintfConversionSpecifier &CS = getConversionSpecifier(); if (!CS.consumesDataArgument()) - return ArgTypeResult::Invalid(); + return ArgType::Invalid(); if (CS.getKind() == ConversionSpecifier::cArg) switch (LM.getKind()) { case LengthModifier::None: return Ctx.IntTy; case LengthModifier::AsLong: - return ArgTypeResult(ArgTypeResult::WIntTy, "wint_t"); + return ArgType(ArgType::WIntTy, "wint_t"); default: - return ArgTypeResult::Invalid(); + return ArgType::Invalid(); } if (CS.isIntArg()) @@ -271,22 +271,22 @@ ArgTypeResult PrintfSpecifier::getArgType(ASTContext &Ctx, // GNU extension. return Ctx.LongLongTy; case LengthModifier::None: return Ctx.IntTy; - case LengthModifier::AsChar: return ArgTypeResult::AnyCharTy; + case LengthModifier::AsChar: return ArgType::AnyCharTy; case LengthModifier::AsShort: return Ctx.ShortTy; case LengthModifier::AsLong: return Ctx.LongTy; case LengthModifier::AsLongLong: case LengthModifier::AsQuad: return Ctx.LongLongTy; case LengthModifier::AsIntMax: - return ArgTypeResult(Ctx.getIntMaxType(), "intmax_t"); + return ArgType(Ctx.getIntMaxType(), "intmax_t"); case LengthModifier::AsSizeT: // FIXME: How to get the corresponding signed version of size_t? - return ArgTypeResult(); + return ArgType(); case LengthModifier::AsPtrDiff: - return ArgTypeResult(Ctx.getPointerDiffType(), "ptrdiff_t"); + return ArgType(Ctx.getPointerDiffType(), "ptrdiff_t"); case LengthModifier::AsAllocate: case LengthModifier::AsMAllocate: - return ArgTypeResult::Invalid(); + return ArgType::Invalid(); } if (CS.isUIntArg()) @@ -302,16 +302,16 @@ ArgTypeResult PrintfSpecifier::getArgType(ASTContext &Ctx, case LengthModifier::AsQuad: return Ctx.UnsignedLongLongTy; case LengthModifier::AsIntMax: - return ArgTypeResult(Ctx.getUIntMaxType(), "uintmax_t"); + return ArgType(Ctx.getUIntMaxType(), "uintmax_t"); case LengthModifier::AsSizeT: - return ArgTypeResult(Ctx.getSizeType(), "size_t"); + return ArgType(Ctx.getSizeType(), "size_t"); case LengthModifier::AsPtrDiff: // FIXME: How to get the corresponding unsigned // version of ptrdiff_t? - return ArgTypeResult(); + return ArgType(); case LengthModifier::AsAllocate: case LengthModifier::AsMAllocate: - return ArgTypeResult::Invalid(); + return ArgType::Invalid(); } if (CS.isDoubleArg()) { @@ -320,37 +320,90 @@ ArgTypeResult PrintfSpecifier::getArgType(ASTContext &Ctx, return Ctx.DoubleTy; } + if (CS.getKind() == ConversionSpecifier::nArg) { + switch (LM.getKind()) { + case LengthModifier::None: + return ArgType::PtrTo(Ctx.IntTy); + case LengthModifier::AsChar: + return ArgType::PtrTo(Ctx.SignedCharTy); + case LengthModifier::AsShort: + return ArgType::PtrTo(Ctx.ShortTy); + case LengthModifier::AsLong: + return ArgType::PtrTo(Ctx.LongTy); + case LengthModifier::AsLongLong: + case LengthModifier::AsQuad: + return ArgType::PtrTo(Ctx.LongLongTy); + case LengthModifier::AsIntMax: + return ArgType::PtrTo(ArgType(Ctx.getIntMaxType(), "intmax_t")); + case LengthModifier::AsSizeT: + return ArgType(); // FIXME: ssize_t + case LengthModifier::AsPtrDiff: + return ArgType::PtrTo(ArgType(Ctx.getPointerDiffType(), "ptrdiff_t")); + case LengthModifier::AsLongDouble: + return ArgType(); // FIXME: Is this a known extension? + case LengthModifier::AsAllocate: + case LengthModifier::AsMAllocate: + return ArgType::Invalid(); + } + } + switch (CS.getKind()) { case ConversionSpecifier::sArg: if (LM.getKind() == LengthModifier::AsWideChar) { if (IsObjCLiteral) return Ctx.getPointerType(Ctx.UnsignedShortTy.withConst()); - return ArgTypeResult(ArgTypeResult::WCStrTy, "wchar_t *"); + return ArgType(ArgType::WCStrTy, "wchar_t *"); } - return ArgTypeResult::CStrTy; + return ArgType::CStrTy; case ConversionSpecifier::SArg: if (IsObjCLiteral) return Ctx.getPointerType(Ctx.UnsignedShortTy.withConst()); - return ArgTypeResult(ArgTypeResult::WCStrTy, "wchar_t *"); + return ArgType(ArgType::WCStrTy, "wchar_t *"); case ConversionSpecifier::CArg: if (IsObjCLiteral) return Ctx.UnsignedShortTy; - return ArgTypeResult(Ctx.WCharTy, "wchar_t"); + return ArgType(Ctx.WCharTy, "wchar_t"); case ConversionSpecifier::pArg: - return ArgTypeResult::CPointerTy; + return ArgType::CPointerTy; case ConversionSpecifier::ObjCObjArg: - return ArgTypeResult::ObjCPointerTy; + return ArgType::ObjCPointerTy; default: break; } // FIXME: Handle other cases. - return ArgTypeResult(); + return ArgType(); } bool PrintfSpecifier::fixType(QualType QT, const LangOptions &LangOpt, ASTContext &Ctx, bool IsObjCLiteral) { - // Handle strings first (char *, wchar_t *) + // %n is different from other conversion specifiers; don't try to fix it. + if (CS.getKind() == ConversionSpecifier::nArg) + return false; + + // Handle Objective-C objects first. Note that while the '%@' specifier will + // not warn for structure pointer or void pointer arguments (because that's + // how CoreFoundation objects are implemented), we only show a fixit for '%@' + // if we know it's an object (block, id, class, or __attribute__((NSObject))). + if (QT->isObjCRetainableType()) { + if (!IsObjCLiteral) + return false; + + CS.setKind(ConversionSpecifier::ObjCObjArg); + + // Disable irrelevant flags + HasThousandsGrouping = false; + HasPlusPrefix = false; + HasSpacePrefix = false; + HasAlternativeForm = false; + HasLeadingZeroes = false; + Precision.setHowSpecified(OptionalAmount::NotSpecified); + LM.setKind(LengthModifier::None); + + return true; + } + + // Handle strings next (char *, wchar_t *) if (QT->isPointerType() && (QT->getPointeeType()->isAnyCharacterType())) { CS.setKind(ConversionSpecifier::sArg); @@ -367,6 +420,10 @@ bool PrintfSpecifier::fixType(QualType QT, const LangOptions &LangOpt, return true; } + // If it's an enum, get its underlying type. + if (const EnumType *ETy = QT->getAs<EnumType>()) + QT = ETy->getDecl()->getIntegerType(); + // We can only work with builtin types. const BuiltinType *BT = QT->getAs<BuiltinType>(); if (!BT) @@ -429,24 +486,11 @@ bool PrintfSpecifier::fixType(QualType QT, const LangOptions &LangOpt, } // Handle size_t, ptrdiff_t, etc. that have dedicated length modifiers in C99. - if (isa<TypedefType>(QT) && (LangOpt.C99 || LangOpt.CPlusPlus0x)) { - const IdentifierInfo *Identifier = QT.getBaseTypeIdentifier(); - if (Identifier->getName() == "size_t") { - LM.setKind(LengthModifier::AsSizeT); - } else if (Identifier->getName() == "ssize_t") { - // Not C99, but common in Unix. - LM.setKind(LengthModifier::AsSizeT); - } else if (Identifier->getName() == "intmax_t") { - LM.setKind(LengthModifier::AsIntMax); - } else if (Identifier->getName() == "uintmax_t") { - LM.setKind(LengthModifier::AsIntMax); - } else if (Identifier->getName() == "ptrdiff_t") { - LM.setKind(LengthModifier::AsPtrDiff); - } - } + if (isa<TypedefType>(QT) && (LangOpt.C99 || LangOpt.CPlusPlus0x)) + namedTypeToLengthModifier(QT, LM); // If fixing the length modifier was enough, we are done. - const analyze_printf::ArgTypeResult &ATR = getArgType(Ctx, IsObjCLiteral); + const analyze_printf::ArgType &ATR = getArgType(Ctx, IsObjCLiteral); if (hasValidLengthModifier() && ATR.isValid() && ATR.matchesType(Ctx, QT)) return true; diff --git a/contrib/llvm/tools/clang/lib/Analysis/ProgramPoint.cpp b/contrib/llvm/tools/clang/lib/Analysis/ProgramPoint.cpp index 3f711b4..7d67e8a 100644 --- a/contrib/llvm/tools/clang/lib/Analysis/ProgramPoint.cpp +++ b/contrib/llvm/tools/clang/lib/Analysis/ProgramPoint.cpp @@ -36,8 +36,10 @@ ProgramPoint ProgramPoint::getProgramPoint(const Stmt *S, ProgramPoint::Kind K, return PreStore(S, LC, tag); case ProgramPoint::PostLValueKind: return PostLValue(S, LC, tag); - case ProgramPoint::PostPurgeDeadSymbolsKind: - return PostPurgeDeadSymbols(S, LC, tag); + case ProgramPoint::PostStmtPurgeDeadSymbolsKind: + return PostStmtPurgeDeadSymbols(S, LC, tag); + case ProgramPoint::PreStmtPurgeDeadSymbolsKind: + return PreStmtPurgeDeadSymbols(S, LC, tag); } } diff --git a/contrib/llvm/tools/clang/lib/Analysis/PseudoConstantAnalysis.cpp b/contrib/llvm/tools/clang/lib/Analysis/PseudoConstantAnalysis.cpp index c8b491a..5d659ce 100644 --- a/contrib/llvm/tools/clang/lib/Analysis/PseudoConstantAnalysis.cpp +++ b/contrib/llvm/tools/clang/lib/Analysis/PseudoConstantAnalysis.cpp @@ -17,6 +17,7 @@ #include "clang/AST/Decl.h" #include "clang/AST/Expr.h" #include "clang/AST/Stmt.h" +#include "llvm/ADT/SmallPtrSet.h" #include <deque> using namespace clang; diff --git a/contrib/llvm/tools/clang/lib/Analysis/ScanfFormatString.cpp b/contrib/llvm/tools/clang/lib/Analysis/ScanfFormatString.cpp index 6bc4adb..2942400 100644 --- a/contrib/llvm/tools/clang/lib/Analysis/ScanfFormatString.cpp +++ b/contrib/llvm/tools/clang/lib/Analysis/ScanfFormatString.cpp @@ -15,12 +15,11 @@ #include "clang/Analysis/Analyses/FormatString.h" #include "FormatStringParsing.h" -using clang::analyze_format_string::ArgTypeResult; +using clang::analyze_format_string::ArgType; using clang::analyze_format_string::FormatStringHandler; using clang::analyze_format_string::LengthModifier; using clang::analyze_format_string::OptionalAmount; using clang::analyze_format_string::ConversionSpecifier; -using clang::analyze_scanf::ScanfArgTypeResult; using clang::analyze_scanf::ScanfConversionSpecifier; using clang::analyze_scanf::ScanfSpecifier; using clang::UpdateOnReturn; @@ -194,37 +193,42 @@ static ScanfSpecifierResult ParseScanfSpecifier(FormatStringHandler &H, return ScanfSpecifierResult(Start, FS); } -ScanfArgTypeResult ScanfSpecifier::getArgType(ASTContext &Ctx) const { +ArgType ScanfSpecifier::getArgType(ASTContext &Ctx) const { const ScanfConversionSpecifier &CS = getConversionSpecifier(); if (!CS.consumesDataArgument()) - return ScanfArgTypeResult::Invalid(); + return ArgType::Invalid(); switch(CS.getKind()) { // Signed int. case ConversionSpecifier::dArg: case ConversionSpecifier::iArg: switch (LM.getKind()) { - case LengthModifier::None: return ArgTypeResult(Ctx.IntTy); + case LengthModifier::None: + return ArgType::PtrTo(Ctx.IntTy); case LengthModifier::AsChar: - return ArgTypeResult(ArgTypeResult::AnyCharTy); - case LengthModifier::AsShort: return ArgTypeResult(Ctx.ShortTy); - case LengthModifier::AsLong: return ArgTypeResult(Ctx.LongTy); + return ArgType::PtrTo(ArgType::AnyCharTy); + case LengthModifier::AsShort: + return ArgType::PtrTo(Ctx.ShortTy); + case LengthModifier::AsLong: + return ArgType::PtrTo(Ctx.LongTy); case LengthModifier::AsLongLong: case LengthModifier::AsQuad: - return ArgTypeResult(Ctx.LongLongTy); + return ArgType::PtrTo(Ctx.LongLongTy); case LengthModifier::AsIntMax: - return ScanfArgTypeResult(Ctx.getIntMaxType(), "intmax_t *"); + return ArgType::PtrTo(ArgType(Ctx.getIntMaxType(), "intmax_t")); case LengthModifier::AsSizeT: // FIXME: ssize_t. - return ScanfArgTypeResult(); + return ArgType(); case LengthModifier::AsPtrDiff: - return ScanfArgTypeResult(Ctx.getPointerDiffType(), "ptrdiff_t *"); + return ArgType::PtrTo(ArgType(Ctx.getPointerDiffType(), "ptrdiff_t")); case LengthModifier::AsLongDouble: // GNU extension. - return ArgTypeResult(Ctx.LongLongTy); - case LengthModifier::AsAllocate: return ScanfArgTypeResult::Invalid(); - case LengthModifier::AsMAllocate: return ScanfArgTypeResult::Invalid(); + return ArgType::PtrTo(Ctx.LongLongTy); + case LengthModifier::AsAllocate: + return ArgType::Invalid(); + case LengthModifier::AsMAllocate: + return ArgType::Invalid(); } // Unsigned int. @@ -233,25 +237,31 @@ ScanfArgTypeResult ScanfSpecifier::getArgType(ASTContext &Ctx) const { case ConversionSpecifier::xArg: case ConversionSpecifier::XArg: switch (LM.getKind()) { - case LengthModifier::None: return ArgTypeResult(Ctx.UnsignedIntTy); - case LengthModifier::AsChar: return ArgTypeResult(Ctx.UnsignedCharTy); - case LengthModifier::AsShort: return ArgTypeResult(Ctx.UnsignedShortTy); - case LengthModifier::AsLong: return ArgTypeResult(Ctx.UnsignedLongTy); + case LengthModifier::None: + return ArgType::PtrTo(Ctx.UnsignedIntTy); + case LengthModifier::AsChar: + return ArgType::PtrTo(Ctx.UnsignedCharTy); + case LengthModifier::AsShort: + return ArgType::PtrTo(Ctx.UnsignedShortTy); + case LengthModifier::AsLong: + return ArgType::PtrTo(Ctx.UnsignedLongTy); case LengthModifier::AsLongLong: case LengthModifier::AsQuad: - return ArgTypeResult(Ctx.UnsignedLongLongTy); + return ArgType::PtrTo(Ctx.UnsignedLongLongTy); case LengthModifier::AsIntMax: - return ScanfArgTypeResult(Ctx.getUIntMaxType(), "uintmax_t *"); + return ArgType::PtrTo(ArgType(Ctx.getUIntMaxType(), "uintmax_t")); case LengthModifier::AsSizeT: - return ScanfArgTypeResult(Ctx.getSizeType(), "size_t *"); + return ArgType::PtrTo(ArgType(Ctx.getSizeType(), "size_t")); case LengthModifier::AsPtrDiff: // FIXME: Unsigned version of ptrdiff_t? - return ScanfArgTypeResult(); + return ArgType(); case LengthModifier::AsLongDouble: // GNU extension. - return ArgTypeResult(Ctx.UnsignedLongLongTy); - case LengthModifier::AsAllocate: return ScanfArgTypeResult::Invalid(); - case LengthModifier::AsMAllocate: return ScanfArgTypeResult::Invalid(); + return ArgType::PtrTo(Ctx.UnsignedLongLongTy); + case LengthModifier::AsAllocate: + return ArgType::Invalid(); + case LengthModifier::AsMAllocate: + return ArgType::Invalid(); } // Float. @@ -264,12 +274,14 @@ ScanfArgTypeResult ScanfSpecifier::getArgType(ASTContext &Ctx) const { case ConversionSpecifier::gArg: case ConversionSpecifier::GArg: switch (LM.getKind()) { - case LengthModifier::None: return ArgTypeResult(Ctx.FloatTy); - case LengthModifier::AsLong: return ArgTypeResult(Ctx.DoubleTy); + case LengthModifier::None: + return ArgType::PtrTo(Ctx.FloatTy); + case LengthModifier::AsLong: + return ArgType::PtrTo(Ctx.DoubleTy); case LengthModifier::AsLongDouble: - return ArgTypeResult(Ctx.LongDoubleTy); + return ArgType::PtrTo(Ctx.LongDoubleTy); default: - return ScanfArgTypeResult::Invalid(); + return ArgType::Invalid(); } // Char, string and scanlist. @@ -277,37 +289,65 @@ ScanfArgTypeResult ScanfSpecifier::getArgType(ASTContext &Ctx) const { case ConversionSpecifier::sArg: case ConversionSpecifier::ScanListArg: switch (LM.getKind()) { - case LengthModifier::None: return ScanfArgTypeResult::CStrTy; + case LengthModifier::None: + return ArgType::PtrTo(ArgType::AnyCharTy); case LengthModifier::AsLong: - return ScanfArgTypeResult(ScanfArgTypeResult::WCStrTy, "wchar_t *"); + return ArgType::PtrTo(ArgType(Ctx.getWCharType(), "wchar_t")); case LengthModifier::AsAllocate: case LengthModifier::AsMAllocate: - return ScanfArgTypeResult(ArgTypeResult::CStrTy); + return ArgType::PtrTo(ArgType::CStrTy); default: - return ScanfArgTypeResult::Invalid(); + return ArgType::Invalid(); } case ConversionSpecifier::CArg: case ConversionSpecifier::SArg: // FIXME: Mac OS X specific? switch (LM.getKind()) { case LengthModifier::None: - return ScanfArgTypeResult(ScanfArgTypeResult::WCStrTy, "wchar_t *"); + return ArgType::PtrTo(ArgType(Ctx.getWCharType(), "wchar_t")); case LengthModifier::AsAllocate: case LengthModifier::AsMAllocate: - return ScanfArgTypeResult(ArgTypeResult::WCStrTy, "wchar_t **"); + return ArgType::PtrTo(ArgType(ArgType::WCStrTy, "wchar_t *")); default: - return ScanfArgTypeResult::Invalid(); + return ArgType::Invalid(); } // Pointer. case ConversionSpecifier::pArg: - return ScanfArgTypeResult(ArgTypeResult(ArgTypeResult::CPointerTy)); + return ArgType::PtrTo(ArgType::CPointerTy); + + // Write-back. + case ConversionSpecifier::nArg: + switch (LM.getKind()) { + case LengthModifier::None: + return ArgType::PtrTo(Ctx.IntTy); + case LengthModifier::AsChar: + return ArgType::PtrTo(Ctx.SignedCharTy); + case LengthModifier::AsShort: + return ArgType::PtrTo(Ctx.ShortTy); + case LengthModifier::AsLong: + return ArgType::PtrTo(Ctx.LongTy); + case LengthModifier::AsLongLong: + case LengthModifier::AsQuad: + return ArgType::PtrTo(Ctx.LongLongTy); + case LengthModifier::AsIntMax: + return ArgType::PtrTo(ArgType(Ctx.getIntMaxType(), "intmax_t")); + case LengthModifier::AsSizeT: + return ArgType(); // FIXME: ssize_t + case LengthModifier::AsPtrDiff: + return ArgType::PtrTo(ArgType(Ctx.getPointerDiffType(), "ptrdiff_t")); + case LengthModifier::AsLongDouble: + return ArgType(); // FIXME: Is this a known extension? + case LengthModifier::AsAllocate: + case LengthModifier::AsMAllocate: + return ArgType::Invalid(); + } default: break; } - return ScanfArgTypeResult(); + return ArgType(); } bool ScanfSpecifier::fixType(QualType QT, const LangOptions &LangOpt, @@ -315,7 +355,16 @@ bool ScanfSpecifier::fixType(QualType QT, const LangOptions &LangOpt, if (!QT->isPointerType()) return false; + // %n is different from other conversion specifiers; don't try to fix it. + if (CS.getKind() == ConversionSpecifier::nArg) + return false; + QualType PT = QT->getPointeeType(); + + // If it's an enum, get its underlying type. + if (const EnumType *ETy = QT->getAs<EnumType>()) + QT = ETy->getDecl()->getIntegerType(); + const BuiltinType *BT = PT->getAs<BuiltinType>(); if (!BT) return false; @@ -377,25 +426,12 @@ bool ScanfSpecifier::fixType(QualType QT, const LangOptions &LangOpt, } // Handle size_t, ptrdiff_t, etc. that have dedicated length modifiers in C99. - if (isa<TypedefType>(PT) && (LangOpt.C99 || LangOpt.CPlusPlus0x)) { - const IdentifierInfo *Identifier = QT.getBaseTypeIdentifier(); - if (Identifier->getName() == "size_t") { - LM.setKind(LengthModifier::AsSizeT); - } else if (Identifier->getName() == "ssize_t") { - // Not C99, but common in Unix. - LM.setKind(LengthModifier::AsSizeT); - } else if (Identifier->getName() == "intmax_t") { - LM.setKind(LengthModifier::AsIntMax); - } else if (Identifier->getName() == "uintmax_t") { - LM.setKind(LengthModifier::AsIntMax); - } else if (Identifier->getName() == "ptrdiff_t") { - LM.setKind(LengthModifier::AsPtrDiff); - } - } + if (isa<TypedefType>(PT) && (LangOpt.C99 || LangOpt.CPlusPlus0x)) + namedTypeToLengthModifier(PT, LM); // If fixing the length modifier was enough, we are done. - const analyze_scanf::ScanfArgTypeResult &ATR = getArgType(Ctx); - if (hasValidLengthModifier() && ATR.isValid() && ATR.matchesType(Ctx, QT)) + const analyze_scanf::ArgType &AT = getArgType(Ctx); + if (hasValidLengthModifier() && AT.isValid() && AT.matchesType(Ctx, QT)) return true; // Figure out the conversion specifier. @@ -452,48 +488,3 @@ bool clang::analyze_format_string::ParseScanfString(FormatStringHandler &H, assert(I == E && "Format string not exhausted"); return false; } - -bool ScanfArgTypeResult::matchesType(ASTContext& C, QualType argTy) const { - switch (K) { - case InvalidTy: - llvm_unreachable("ArgTypeResult must be valid"); - case UnknownTy: - return true; - case CStrTy: - return ArgTypeResult(ArgTypeResult::CStrTy).matchesType(C, argTy); - case WCStrTy: - return ArgTypeResult(ArgTypeResult::WCStrTy).matchesType(C, argTy); - case PtrToArgTypeResultTy: { - const PointerType *PT = argTy->getAs<PointerType>(); - if (!PT) - return false; - return A.matchesType(C, PT->getPointeeType()); - } - } - - llvm_unreachable("Invalid ScanfArgTypeResult Kind!"); -} - -QualType ScanfArgTypeResult::getRepresentativeType(ASTContext &C) const { - switch (K) { - case InvalidTy: - llvm_unreachable("No representative type for Invalid ArgTypeResult"); - case UnknownTy: - return QualType(); - case CStrTy: - return C.getPointerType(C.CharTy); - case WCStrTy: - return C.getPointerType(C.getWCharType()); - case PtrToArgTypeResultTy: - return C.getPointerType(A.getRepresentativeType(C)); - } - - llvm_unreachable("Invalid ScanfArgTypeResult Kind!"); -} - -std::string ScanfArgTypeResult::getRepresentativeTypeName(ASTContext& C) const { - std::string S = getRepresentativeType(C).getAsString(); - if (!Name) - return std::string("'") + S + "'"; - return std::string("'") + Name + "' (aka '" + S + "')"; -} diff --git a/contrib/llvm/tools/clang/lib/Analysis/ThreadSafety.cpp b/contrib/llvm/tools/clang/lib/Analysis/ThreadSafety.cpp index 2f7e794..5954682 100644 --- a/contrib/llvm/tools/clang/lib/Analysis/ThreadSafety.cpp +++ b/contrib/llvm/tools/clang/lib/Analysis/ThreadSafety.cpp @@ -26,6 +26,7 @@ #include "clang/AST/StmtVisitor.h" #include "clang/Basic/SourceManager.h" #include "clang/Basic/SourceLocation.h" +#include "clang/Basic/OperatorKinds.h" #include "llvm/ADT/BitVector.h" #include "llvm/ADT/FoldingSet.h" #include "llvm/ADT/ImmutableMap.h" @@ -45,8 +46,15 @@ ThreadSafetyHandler::~ThreadSafetyHandler() {} namespace { -/// \brief A MutexID object uniquely identifies a particular mutex, and -/// is built from an Expr* (i.e. calling a lock function). +/// SExpr implements a simple expression language that is used to store, +/// compare, and pretty-print C++ expressions. Unlike a clang Expr, a SExpr +/// does not capture surface syntax, and it does not distinguish between +/// C++ concepts, like pointers and references, that have no real semantic +/// differences. This simplicity allows SExprs to be meaningfully compared, +/// e.g. +/// (x) = x +/// (*this).foo = this->foo +/// *&a = a /// /// Thread-safety analysis works by comparing lock expressions. Within the /// body of a function, an expression such as "x->foo->bar.mu" will resolve to @@ -59,41 +67,194 @@ namespace { /// /// The current implementation assumes, but does not verify, that multiple uses /// of the same lock expression satisfies these criteria. -/// -/// Clang introduces an additional wrinkle, which is that it is difficult to -/// derive canonical expressions, or compare expressions directly for equality. -/// Thus, we identify a mutex not by an Expr, but by the list of named -/// declarations that are referenced by the Expr. In other words, -/// x->foo->bar.mu will be a four element vector with the Decls for -/// mu, bar, and foo, and x. The vector will uniquely identify the expression -/// for all practical purposes. Null is used to denote 'this'. -/// -/// Note we will need to perform substitution on "this" and function parameter -/// names when constructing a lock expression. -/// -/// For example: -/// class C { Mutex Mu; void lock() EXCLUSIVE_LOCK_FUNCTION(this->Mu); }; -/// void myFunc(C *X) { ... X->lock() ... } -/// The original expression for the mutex acquired by myFunc is "this->Mu", but -/// "X" is substituted for "this" so we get X->Mu(); -/// -/// For another example: -/// foo(MyList *L) EXCLUSIVE_LOCKS_REQUIRED(L->Mu) { ... } -/// MyList *MyL; -/// foo(MyL); // requires lock MyL->Mu to be held -class MutexID { - SmallVector<NamedDecl*, 2> DeclSeq; - - /// Build a Decl sequence representing the lock from the given expression. +class SExpr { +private: + enum ExprOp { + EOP_Nop, //< No-op + EOP_Wildcard, //< Matches anything. + EOP_This, //< This keyword. + EOP_NVar, //< Named variable. + EOP_LVar, //< Local variable. + EOP_Dot, //< Field access + EOP_Call, //< Function call + EOP_MCall, //< Method call + EOP_Index, //< Array index + EOP_Unary, //< Unary operation + EOP_Binary, //< Binary operation + EOP_Unknown //< Catchall for everything else + }; + + + class SExprNode { + private: + unsigned char Op; //< Opcode of the root node + unsigned char Flags; //< Additional opcode-specific data + unsigned short Sz; //< Number of child nodes + const void* Data; //< Additional opcode-specific data + + public: + SExprNode(ExprOp O, unsigned F, const void* D) + : Op(static_cast<unsigned char>(O)), + Flags(static_cast<unsigned char>(F)), Sz(1), Data(D) + { } + + unsigned size() const { return Sz; } + void setSize(unsigned S) { Sz = S; } + + ExprOp kind() const { return static_cast<ExprOp>(Op); } + + const NamedDecl* getNamedDecl() const { + assert(Op == EOP_NVar || Op == EOP_LVar || Op == EOP_Dot); + return reinterpret_cast<const NamedDecl*>(Data); + } + + const NamedDecl* getFunctionDecl() const { + assert(Op == EOP_Call || Op == EOP_MCall); + return reinterpret_cast<const NamedDecl*>(Data); + } + + bool isArrow() const { return Op == EOP_Dot && Flags == 1; } + void setArrow(bool A) { Flags = A ? 1 : 0; } + + unsigned arity() const { + switch (Op) { + case EOP_Nop: return 0; + case EOP_Wildcard: return 0; + case EOP_NVar: return 0; + case EOP_LVar: return 0; + case EOP_This: return 0; + case EOP_Dot: return 1; + case EOP_Call: return Flags+1; // First arg is function. + case EOP_MCall: return Flags+1; // First arg is implicit obj. + case EOP_Index: return 2; + case EOP_Unary: return 1; + case EOP_Binary: return 2; + case EOP_Unknown: return Flags; + } + return 0; + } + + bool operator==(const SExprNode& Other) const { + // Ignore flags and size -- they don't matter. + return (Op == Other.Op && + Data == Other.Data); + } + + bool operator!=(const SExprNode& Other) const { + return !(*this == Other); + } + + bool matches(const SExprNode& Other) const { + return (*this == Other) || + (Op == EOP_Wildcard) || + (Other.Op == EOP_Wildcard); + } + }; + + + /// \brief Encapsulates the lexical context of a function call. The lexical + /// context includes the arguments to the call, including the implicit object + /// argument. When an attribute containing a mutex expression is attached to + /// a method, the expression may refer to formal parameters of the method. + /// Actual arguments must be substituted for formal parameters to derive + /// the appropriate mutex expression in the lexical context where the function + /// is called. PrevCtx holds the context in which the arguments themselves + /// should be evaluated; multiple calling contexts can be chained together + /// by the lock_returned attribute. + struct CallingContext { + const NamedDecl* AttrDecl; // The decl to which the attribute is attached. + Expr* SelfArg; // Implicit object argument -- e.g. 'this' + bool SelfArrow; // is Self referred to with -> or .? + unsigned NumArgs; // Number of funArgs + Expr** FunArgs; // Function arguments + CallingContext* PrevCtx; // The previous context; or 0 if none. + + CallingContext(const NamedDecl *D = 0, Expr *S = 0, + unsigned N = 0, Expr **A = 0, CallingContext *P = 0) + : AttrDecl(D), SelfArg(S), SelfArrow(false), + NumArgs(N), FunArgs(A), PrevCtx(P) + { } + }; + + typedef SmallVector<SExprNode, 4> NodeVector; + +private: + // A SExpr is a list of SExprNodes in prefix order. The Size field allows + // the list to be traversed as a tree. + NodeVector NodeVec; + +private: + unsigned makeNop() { + NodeVec.push_back(SExprNode(EOP_Nop, 0, 0)); + return NodeVec.size()-1; + } + + unsigned makeWildcard() { + NodeVec.push_back(SExprNode(EOP_Wildcard, 0, 0)); + return NodeVec.size()-1; + } + + unsigned makeNamedVar(const NamedDecl *D) { + NodeVec.push_back(SExprNode(EOP_NVar, 0, D)); + return NodeVec.size()-1; + } + + unsigned makeLocalVar(const NamedDecl *D) { + NodeVec.push_back(SExprNode(EOP_LVar, 0, D)); + return NodeVec.size()-1; + } + + unsigned makeThis() { + NodeVec.push_back(SExprNode(EOP_This, 0, 0)); + return NodeVec.size()-1; + } + + unsigned makeDot(const NamedDecl *D, bool Arrow) { + NodeVec.push_back(SExprNode(EOP_Dot, Arrow ? 1 : 0, D)); + return NodeVec.size()-1; + } + + unsigned makeCall(unsigned NumArgs, const NamedDecl *D) { + NodeVec.push_back(SExprNode(EOP_Call, NumArgs, D)); + return NodeVec.size()-1; + } + + unsigned makeMCall(unsigned NumArgs, const NamedDecl *D) { + NodeVec.push_back(SExprNode(EOP_MCall, NumArgs, D)); + return NodeVec.size()-1; + } + + unsigned makeIndex() { + NodeVec.push_back(SExprNode(EOP_Index, 0, 0)); + return NodeVec.size()-1; + } + + unsigned makeUnary() { + NodeVec.push_back(SExprNode(EOP_Unary, 0, 0)); + return NodeVec.size()-1; + } + + unsigned makeBinary() { + NodeVec.push_back(SExprNode(EOP_Binary, 0, 0)); + return NodeVec.size()-1; + } + + unsigned makeUnknown(unsigned Arity) { + NodeVec.push_back(SExprNode(EOP_Unknown, Arity, 0)); + return NodeVec.size()-1; + } + + /// Build an SExpr from the given C++ expression. /// Recursive function that terminates on DeclRefExpr. - /// Note: this function merely creates a MutexID; it does not check to + /// Note: this function merely creates a SExpr; it does not check to /// ensure that the original expression is a valid mutex expression. - void buildMutexID(Expr *Exp, const NamedDecl *D, Expr *Parent, - unsigned NumArgs, Expr **FunArgs) { - if (!Exp) { - DeclSeq.clear(); - return; - } + /// + /// NDeref returns the number of Derefence and AddressOf operations + /// preceeding the Expr; this is used to decide whether to pretty-print + /// SExprs with . or ->. + unsigned buildSExpr(Expr *Exp, CallingContext* CallCtx, int* NDeref = 0) { + if (!Exp) + return 0; if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(Exp)) { NamedDecl *ND = cast<NamedDecl>(DRE->getDecl()->getCanonicalDecl()); @@ -103,144 +264,246 @@ class MutexID { cast<FunctionDecl>(PV->getDeclContext())->getCanonicalDecl(); unsigned i = PV->getFunctionScopeIndex(); - if (FunArgs && FD == D->getCanonicalDecl()) { + if (CallCtx && CallCtx->FunArgs && + FD == CallCtx->AttrDecl->getCanonicalDecl()) { // Substitute call arguments for references to function parameters - assert(i < NumArgs); - buildMutexID(FunArgs[i], D, 0, 0, 0); - return; + assert(i < CallCtx->NumArgs); + return buildSExpr(CallCtx->FunArgs[i], CallCtx->PrevCtx, NDeref); } // Map the param back to the param of the original function declaration. - DeclSeq.push_back(FD->getParamDecl(i)); - return; + makeNamedVar(FD->getParamDecl(i)); + return 1; } // Not a function parameter -- just store the reference. - DeclSeq.push_back(ND); - } else if (MemberExpr *ME = dyn_cast<MemberExpr>(Exp)) { - NamedDecl *ND = ME->getMemberDecl(); - DeclSeq.push_back(ND); - buildMutexID(ME->getBase(), D, Parent, NumArgs, FunArgs); + makeNamedVar(ND); + return 1; } else if (isa<CXXThisExpr>(Exp)) { - if (Parent) - buildMutexID(Parent, D, 0, 0, 0); + // Substitute parent for 'this' + if (CallCtx && CallCtx->SelfArg) { + if (!CallCtx->SelfArrow && NDeref) + // 'this' is a pointer, but self is not, so need to take address. + --(*NDeref); + return buildSExpr(CallCtx->SelfArg, CallCtx->PrevCtx, NDeref); + } else { - DeclSeq.push_back(0); // Use 0 to represent 'this'. - return; // mutexID is still valid in this case + makeThis(); + return 1; } + } else if (MemberExpr *ME = dyn_cast<MemberExpr>(Exp)) { + NamedDecl *ND = ME->getMemberDecl(); + int ImplicitDeref = ME->isArrow() ? 1 : 0; + unsigned Root = makeDot(ND, false); + unsigned Sz = buildSExpr(ME->getBase(), CallCtx, &ImplicitDeref); + NodeVec[Root].setArrow(ImplicitDeref > 0); + NodeVec[Root].setSize(Sz + 1); + return Sz + 1; } else if (CXXMemberCallExpr *CMCE = dyn_cast<CXXMemberCallExpr>(Exp)) { - DeclSeq.push_back(CMCE->getMethodDecl()->getCanonicalDecl()); - buildMutexID(CMCE->getImplicitObjectArgument(), - D, Parent, NumArgs, FunArgs); + // When calling a function with a lock_returned attribute, replace + // the function call with the expression in lock_returned. + if (LockReturnedAttr* At = + CMCE->getMethodDecl()->getAttr<LockReturnedAttr>()) { + CallingContext LRCallCtx(CMCE->getMethodDecl()); + LRCallCtx.SelfArg = CMCE->getImplicitObjectArgument(); + LRCallCtx.SelfArrow = + dyn_cast<MemberExpr>(CMCE->getCallee())->isArrow(); + LRCallCtx.NumArgs = CMCE->getNumArgs(); + LRCallCtx.FunArgs = CMCE->getArgs(); + LRCallCtx.PrevCtx = CallCtx; + return buildSExpr(At->getArg(), &LRCallCtx); + } + // Hack to treat smart pointers and iterators as pointers; + // ignore any method named get(). + if (CMCE->getMethodDecl()->getNameAsString() == "get" && + CMCE->getNumArgs() == 0) { + if (NDeref && dyn_cast<MemberExpr>(CMCE->getCallee())->isArrow()) + ++(*NDeref); + return buildSExpr(CMCE->getImplicitObjectArgument(), CallCtx, NDeref); + } unsigned NumCallArgs = CMCE->getNumArgs(); + unsigned Root = + makeMCall(NumCallArgs, CMCE->getMethodDecl()->getCanonicalDecl()); + unsigned Sz = buildSExpr(CMCE->getImplicitObjectArgument(), CallCtx); Expr** CallArgs = CMCE->getArgs(); for (unsigned i = 0; i < NumCallArgs; ++i) { - buildMutexID(CallArgs[i], D, Parent, NumArgs, FunArgs); + Sz += buildSExpr(CallArgs[i], CallCtx); } + NodeVec[Root].setSize(Sz + 1); + return Sz + 1; } else if (CallExpr *CE = dyn_cast<CallExpr>(Exp)) { - buildMutexID(CE->getCallee(), D, Parent, NumArgs, FunArgs); + if (LockReturnedAttr* At = + CE->getDirectCallee()->getAttr<LockReturnedAttr>()) { + CallingContext LRCallCtx(CE->getDirectCallee()); + LRCallCtx.NumArgs = CE->getNumArgs(); + LRCallCtx.FunArgs = CE->getArgs(); + LRCallCtx.PrevCtx = CallCtx; + return buildSExpr(At->getArg(), &LRCallCtx); + } + // Treat smart pointers and iterators as pointers; + // ignore the * and -> operators. + if (CXXOperatorCallExpr *OE = dyn_cast<CXXOperatorCallExpr>(CE)) { + OverloadedOperatorKind k = OE->getOperator(); + if (k == OO_Star) { + if (NDeref) ++(*NDeref); + return buildSExpr(OE->getArg(0), CallCtx, NDeref); + } + else if (k == OO_Arrow) { + return buildSExpr(OE->getArg(0), CallCtx, NDeref); + } + } unsigned NumCallArgs = CE->getNumArgs(); + unsigned Root = makeCall(NumCallArgs, 0); + unsigned Sz = buildSExpr(CE->getCallee(), CallCtx); Expr** CallArgs = CE->getArgs(); for (unsigned i = 0; i < NumCallArgs; ++i) { - buildMutexID(CallArgs[i], D, Parent, NumArgs, FunArgs); + Sz += buildSExpr(CallArgs[i], CallCtx); } + NodeVec[Root].setSize(Sz+1); + return Sz+1; } else if (BinaryOperator *BOE = dyn_cast<BinaryOperator>(Exp)) { - buildMutexID(BOE->getLHS(), D, Parent, NumArgs, FunArgs); - buildMutexID(BOE->getRHS(), D, Parent, NumArgs, FunArgs); + unsigned Root = makeBinary(); + unsigned Sz = buildSExpr(BOE->getLHS(), CallCtx); + Sz += buildSExpr(BOE->getRHS(), CallCtx); + NodeVec[Root].setSize(Sz); + return Sz; } else if (UnaryOperator *UOE = dyn_cast<UnaryOperator>(Exp)) { - buildMutexID(UOE->getSubExpr(), D, Parent, NumArgs, FunArgs); + // Ignore & and * operators -- they're no-ops. + // However, we try to figure out whether the expression is a pointer, + // so we can use . and -> appropriately in error messages. + if (UOE->getOpcode() == UO_Deref) { + if (NDeref) ++(*NDeref); + return buildSExpr(UOE->getSubExpr(), CallCtx, NDeref); + } + if (UOE->getOpcode() == UO_AddrOf) { + if (DeclRefExpr* DRE = dyn_cast<DeclRefExpr>(UOE->getSubExpr())) { + if (DRE->getDecl()->isCXXInstanceMember()) { + // This is a pointer-to-member expression, e.g. &MyClass::mu_. + // We interpret this syntax specially, as a wildcard. + unsigned Root = makeDot(DRE->getDecl(), false); + makeWildcard(); + NodeVec[Root].setSize(2); + return 2; + } + } + if (NDeref) --(*NDeref); + return buildSExpr(UOE->getSubExpr(), CallCtx, NDeref); + } + unsigned Root = makeUnary(); + unsigned Sz = buildSExpr(UOE->getSubExpr(), CallCtx); + NodeVec[Root].setSize(Sz); + return Sz; } else if (ArraySubscriptExpr *ASE = dyn_cast<ArraySubscriptExpr>(Exp)) { - buildMutexID(ASE->getBase(), D, Parent, NumArgs, FunArgs); - buildMutexID(ASE->getIdx(), D, Parent, NumArgs, FunArgs); + unsigned Root = makeIndex(); + unsigned Sz = buildSExpr(ASE->getBase(), CallCtx); + Sz += buildSExpr(ASE->getIdx(), CallCtx); + NodeVec[Root].setSize(Sz); + return Sz; } else if (AbstractConditionalOperator *CE = - dyn_cast<AbstractConditionalOperator>(Exp)) { - buildMutexID(CE->getCond(), D, Parent, NumArgs, FunArgs); - buildMutexID(CE->getTrueExpr(), D, Parent, NumArgs, FunArgs); - buildMutexID(CE->getFalseExpr(), D, Parent, NumArgs, FunArgs); + dyn_cast<AbstractConditionalOperator>(Exp)) { + unsigned Root = makeUnknown(3); + unsigned Sz = buildSExpr(CE->getCond(), CallCtx); + Sz += buildSExpr(CE->getTrueExpr(), CallCtx); + Sz += buildSExpr(CE->getFalseExpr(), CallCtx); + NodeVec[Root].setSize(Sz); + return Sz; } else if (ChooseExpr *CE = dyn_cast<ChooseExpr>(Exp)) { - buildMutexID(CE->getCond(), D, Parent, NumArgs, FunArgs); - buildMutexID(CE->getLHS(), D, Parent, NumArgs, FunArgs); - buildMutexID(CE->getRHS(), D, Parent, NumArgs, FunArgs); + unsigned Root = makeUnknown(3); + unsigned Sz = buildSExpr(CE->getCond(), CallCtx); + Sz += buildSExpr(CE->getLHS(), CallCtx); + Sz += buildSExpr(CE->getRHS(), CallCtx); + NodeVec[Root].setSize(Sz); + return Sz; } else if (CastExpr *CE = dyn_cast<CastExpr>(Exp)) { - buildMutexID(CE->getSubExpr(), D, Parent, NumArgs, FunArgs); + return buildSExpr(CE->getSubExpr(), CallCtx, NDeref); } else if (ParenExpr *PE = dyn_cast<ParenExpr>(Exp)) { - buildMutexID(PE->getSubExpr(), D, Parent, NumArgs, FunArgs); + return buildSExpr(PE->getSubExpr(), CallCtx, NDeref); + } else if (ExprWithCleanups *EWC = dyn_cast<ExprWithCleanups>(Exp)) { + return buildSExpr(EWC->getSubExpr(), CallCtx, NDeref); + } else if (CXXBindTemporaryExpr *E = dyn_cast<CXXBindTemporaryExpr>(Exp)) { + return buildSExpr(E->getSubExpr(), CallCtx, NDeref); } else if (isa<CharacterLiteral>(Exp) || - isa<CXXNullPtrLiteralExpr>(Exp) || - isa<GNUNullExpr>(Exp) || - isa<CXXBoolLiteralExpr>(Exp) || - isa<FloatingLiteral>(Exp) || - isa<ImaginaryLiteral>(Exp) || - isa<IntegerLiteral>(Exp) || - isa<StringLiteral>(Exp) || - isa<ObjCStringLiteral>(Exp)) { - return; // FIXME: Ignore literals for now + isa<CXXNullPtrLiteralExpr>(Exp) || + isa<GNUNullExpr>(Exp) || + isa<CXXBoolLiteralExpr>(Exp) || + isa<FloatingLiteral>(Exp) || + isa<ImaginaryLiteral>(Exp) || + isa<IntegerLiteral>(Exp) || + isa<StringLiteral>(Exp) || + isa<ObjCStringLiteral>(Exp)) { + makeNop(); + return 1; // FIXME: Ignore literals for now } else { - // Ignore. FIXME: mark as invalid expression? + makeNop(); + return 1; // Ignore. FIXME: mark as invalid expression? } } - /// \brief Construct a MutexID from an expression. + /// \brief Construct a SExpr from an expression. /// \param MutexExp The original mutex expression within an attribute /// \param DeclExp An expression involving the Decl on which the attribute /// occurs. /// \param D The declaration to which the lock/unlock attribute is attached. - void buildMutexIDFromExp(Expr *MutexExp, Expr *DeclExp, const NamedDecl *D) { - Expr *Parent = 0; - unsigned NumArgs = 0; - Expr **FunArgs = 0; + void buildSExprFromExpr(Expr *MutexExp, Expr *DeclExp, const NamedDecl *D) { + CallingContext CallCtx(D); // If we are processing a raw attribute expression, with no substitutions. if (DeclExp == 0) { - buildMutexID(MutexExp, D, 0, 0, 0); + buildSExpr(MutexExp, 0); return; } - // Examine DeclExp to find Parent and FunArgs, which are used to substitute + // Examine DeclExp to find SelfArg and FunArgs, which are used to substitute // for formal parameters when we call buildMutexID later. if (MemberExpr *ME = dyn_cast<MemberExpr>(DeclExp)) { - Parent = ME->getBase(); + CallCtx.SelfArg = ME->getBase(); + CallCtx.SelfArrow = ME->isArrow(); } else if (CXXMemberCallExpr *CE = dyn_cast<CXXMemberCallExpr>(DeclExp)) { - Parent = CE->getImplicitObjectArgument(); - NumArgs = CE->getNumArgs(); - FunArgs = CE->getArgs(); + CallCtx.SelfArg = CE->getImplicitObjectArgument(); + CallCtx.SelfArrow = dyn_cast<MemberExpr>(CE->getCallee())->isArrow(); + CallCtx.NumArgs = CE->getNumArgs(); + CallCtx.FunArgs = CE->getArgs(); } else if (CallExpr *CE = dyn_cast<CallExpr>(DeclExp)) { - NumArgs = CE->getNumArgs(); - FunArgs = CE->getArgs(); + CallCtx.NumArgs = CE->getNumArgs(); + CallCtx.FunArgs = CE->getArgs(); } else if (CXXConstructExpr *CE = dyn_cast<CXXConstructExpr>(DeclExp)) { - Parent = 0; // FIXME -- get the parent from DeclStmt - NumArgs = CE->getNumArgs(); - FunArgs = CE->getArgs(); + CallCtx.SelfArg = 0; // FIXME -- get the parent from DeclStmt + CallCtx.NumArgs = CE->getNumArgs(); + CallCtx.FunArgs = CE->getArgs(); } else if (D && isa<CXXDestructorDecl>(D)) { // There's no such thing as a "destructor call" in the AST. - Parent = DeclExp; + CallCtx.SelfArg = DeclExp; } // If the attribute has no arguments, then assume the argument is "this". if (MutexExp == 0) { - buildMutexID(Parent, D, 0, 0, 0); + buildSExpr(CallCtx.SelfArg, 0); return; } - buildMutexID(MutexExp, D, Parent, NumArgs, FunArgs); + // For most attributes. + buildSExpr(MutexExp, &CallCtx); } -public: - explicit MutexID(clang::Decl::EmptyShell e) { - DeclSeq.clear(); + /// \brief Get index of next sibling of node i. + unsigned getNextSibling(unsigned i) const { + return i + NodeVec[i].size(); } +public: + explicit SExpr(clang::Decl::EmptyShell e) { NodeVec.clear(); } + /// \param MutexExp The original mutex expression within an attribute /// \param DeclExp An expression involving the Decl on which the attribute /// occurs. /// \param D The declaration to which the lock/unlock attribute is attached. /// Caller must check isValid() after construction. - MutexID(Expr* MutexExp, Expr *DeclExp, const NamedDecl* D) { - buildMutexIDFromExp(MutexExp, DeclExp, D); + SExpr(Expr* MutexExp, Expr *DeclExp, const NamedDecl* D) { + buildSExprFromExpr(MutexExp, DeclExp, D); } /// Return true if this is a valid decl sequence. /// Caller must call this by hand after construction to handle errors. bool isValid() const { - return !DeclSeq.empty(); + return !NodeVec.empty(); } /// Issue a warning about an invalid lock expression @@ -255,44 +518,144 @@ public: Handler.handleInvalidLockExp(Loc); } - bool operator==(const MutexID &other) const { - return DeclSeq == other.DeclSeq; + bool operator==(const SExpr &other) const { + return NodeVec == other.NodeVec; } - bool operator!=(const MutexID &other) const { + bool operator!=(const SExpr &other) const { return !(*this == other); } - // SmallVector overloads Operator< to do lexicographic ordering. Note that - // we use pointer equality (and <) to compare NamedDecls. This means the order - // of MutexIDs in a lockset is nondeterministic. In order to output - // diagnostics in a deterministic ordering, we must order all diagnostics to - // output by SourceLocation when iterating through this lockset. - bool operator<(const MutexID &other) const { - return DeclSeq < other.DeclSeq; + bool matches(const SExpr &Other, unsigned i = 0, unsigned j = 0) const { + if (NodeVec[i].matches(Other.NodeVec[j])) { + unsigned n = NodeVec[i].arity(); + bool Result = true; + unsigned ci = i+1; // first child of i + unsigned cj = j+1; // first child of j + for (unsigned k = 0; k < n; + ++k, ci=getNextSibling(ci), cj = Other.getNextSibling(cj)) { + Result = Result && matches(Other, ci, cj); + } + return Result; + } + return false; } - /// \brief Returns the name of the first Decl in the list for a given MutexID; - /// e.g. the lock expression foo.bar() has name "bar". - /// The caret will point unambiguously to the lock expression, so using this - /// name in diagnostics is a way to get simple, and consistent, mutex names. - /// We do not want to output the entire expression text for security reasons. - std::string getName() const { + /// \brief Pretty print a lock expression for use in error messages. + std::string toString(unsigned i = 0) const { assert(isValid()); - if (!DeclSeq.front()) - return "this"; // Use 0 to represent 'this'. - return DeclSeq.front()->getNameAsString(); + if (i >= NodeVec.size()) + return ""; + + const SExprNode* N = &NodeVec[i]; + switch (N->kind()) { + case EOP_Nop: + return "_"; + case EOP_Wildcard: + return "(?)"; + case EOP_This: + return "this"; + case EOP_NVar: + case EOP_LVar: { + return N->getNamedDecl()->getNameAsString(); + } + case EOP_Dot: { + if (NodeVec[i+1].kind() == EOP_Wildcard) { + std::string S = "&"; + S += N->getNamedDecl()->getQualifiedNameAsString(); + return S; + } + std::string FieldName = N->getNamedDecl()->getNameAsString(); + if (NodeVec[i+1].kind() == EOP_This) + return FieldName; + + std::string S = toString(i+1); + if (N->isArrow()) + return S + "->" + FieldName; + else + return S + "." + FieldName; + } + case EOP_Call: { + std::string S = toString(i+1) + "("; + unsigned NumArgs = N->arity()-1; + unsigned ci = getNextSibling(i+1); + for (unsigned k=0; k<NumArgs; ++k, ci = getNextSibling(ci)) { + S += toString(ci); + if (k+1 < NumArgs) S += ","; + } + S += ")"; + return S; + } + case EOP_MCall: { + std::string S = ""; + if (NodeVec[i+1].kind() != EOP_This) + S = toString(i+1) + "."; + if (const NamedDecl *D = N->getFunctionDecl()) + S += D->getNameAsString() + "("; + else + S += "#("; + unsigned NumArgs = N->arity()-1; + unsigned ci = getNextSibling(i+1); + for (unsigned k=0; k<NumArgs; ++k, ci = getNextSibling(ci)) { + S += toString(ci); + if (k+1 < NumArgs) S += ","; + } + S += ")"; + return S; + } + case EOP_Index: { + std::string S1 = toString(i+1); + std::string S2 = toString(i+1 + NodeVec[i+1].size()); + return S1 + "[" + S2 + "]"; + } + case EOP_Unary: { + std::string S = toString(i+1); + return "#" + S; + } + case EOP_Binary: { + std::string S1 = toString(i+1); + std::string S2 = toString(i+1 + NodeVec[i+1].size()); + return "(" + S1 + "#" + S2 + ")"; + } + case EOP_Unknown: { + unsigned NumChildren = N->arity(); + if (NumChildren == 0) + return "(...)"; + std::string S = "("; + unsigned ci = i+1; + for (unsigned j = 0; j < NumChildren; ++j, ci = getNextSibling(ci)) { + S += toString(ci); + if (j+1 < NumChildren) S += "#"; + } + S += ")"; + return S; + } + } + return ""; } +}; - void Profile(llvm::FoldingSetNodeID &ID) const { - for (SmallVectorImpl<NamedDecl*>::const_iterator I = DeclSeq.begin(), - E = DeclSeq.end(); I != E; ++I) { - ID.AddPointer(*I); - } + + +/// \brief A short list of SExprs +class MutexIDList : public SmallVector<SExpr, 3> { +public: + /// \brief Return true if the list contains the specified SExpr + /// Performs a linear search, because these lists are almost always very small. + bool contains(const SExpr& M) { + for (iterator I=begin(),E=end(); I != E; ++I) + if ((*I) == M) return true; + return false; + } + + /// \brief Push M onto list, bud discard duplicates + void push_back_nodup(const SExpr& M) { + if (!contains(M)) push_back(M); } }; + /// \brief This is a helper class that stores info about the most recent /// accquire of a Lock. /// @@ -307,14 +670,18 @@ struct LockData { /// /// FIXME: add support for re-entrant locking and lock up/downgrading LockKind LKind; - MutexID UnderlyingMutex; // for ScopedLockable objects + bool Managed; // for ScopedLockable objects + SExpr UnderlyingMutex; // for ScopedLockable objects - LockData(SourceLocation AcquireLoc, LockKind LKind) - : AcquireLoc(AcquireLoc), LKind(LKind), UnderlyingMutex(Decl::EmptyShell()) + LockData(SourceLocation AcquireLoc, LockKind LKind, bool M = false) + : AcquireLoc(AcquireLoc), LKind(LKind), Managed(M), + UnderlyingMutex(Decl::EmptyShell()) {} - LockData(SourceLocation AcquireLoc, LockKind LKind, const MutexID &Mu) - : AcquireLoc(AcquireLoc), LKind(LKind), UnderlyingMutex(Mu) {} + LockData(SourceLocation AcquireLoc, LockKind LKind, const SExpr &Mu) + : AcquireLoc(AcquireLoc), LKind(LKind), Managed(false), + UnderlyingMutex(Mu) + {} bool operator==(const LockData &other) const { return AcquireLoc == other.AcquireLoc && LKind == other.LKind; @@ -331,10 +698,102 @@ struct LockData { }; -/// A Lockset maps each MutexID (defined above) to information about how it has +/// \brief A FactEntry stores a single fact that is known at a particular point +/// in the program execution. Currently, this is information regarding a lock +/// that is held at that point. +struct FactEntry { + SExpr MutID; + LockData LDat; + + FactEntry(const SExpr& M, const LockData& L) + : MutID(M), LDat(L) + { } +}; + + +typedef unsigned short FactID; + +/// \brief FactManager manages the memory for all facts that are created during +/// the analysis of a single routine. +class FactManager { +private: + std::vector<FactEntry> Facts; + +public: + FactID newLock(const SExpr& M, const LockData& L) { + Facts.push_back(FactEntry(M,L)); + return static_cast<unsigned short>(Facts.size() - 1); + } + + const FactEntry& operator[](FactID F) const { return Facts[F]; } + FactEntry& operator[](FactID F) { return Facts[F]; } +}; + + +/// \brief A FactSet is the set of facts that are known to be true at a +/// particular program point. FactSets must be small, because they are +/// frequently copied, and are thus implemented as a set of indices into a +/// table maintained by a FactManager. A typical FactSet only holds 1 or 2 +/// locks, so we can get away with doing a linear search for lookup. Note +/// that a hashtable or map is inappropriate in this case, because lookups +/// may involve partial pattern matches, rather than exact matches. +class FactSet { +private: + typedef SmallVector<FactID, 4> FactVec; + + FactVec FactIDs; + +public: + typedef FactVec::iterator iterator; + typedef FactVec::const_iterator const_iterator; + + iterator begin() { return FactIDs.begin(); } + const_iterator begin() const { return FactIDs.begin(); } + + iterator end() { return FactIDs.end(); } + const_iterator end() const { return FactIDs.end(); } + + bool isEmpty() const { return FactIDs.size() == 0; } + + FactID addLock(FactManager& FM, const SExpr& M, const LockData& L) { + FactID F = FM.newLock(M, L); + FactIDs.push_back(F); + return F; + } + + bool removeLock(FactManager& FM, const SExpr& M) { + unsigned n = FactIDs.size(); + if (n == 0) + return false; + + for (unsigned i = 0; i < n-1; ++i) { + if (FM[FactIDs[i]].MutID.matches(M)) { + FactIDs[i] = FactIDs[n-1]; + FactIDs.pop_back(); + return true; + } + } + if (FM[FactIDs[n-1]].MutID.matches(M)) { + FactIDs.pop_back(); + return true; + } + return false; + } + + LockData* findLock(FactManager& FM, const SExpr& M) const { + for (const_iterator I=begin(), E=end(); I != E; ++I) { + if (FM[*I].MutID.matches(M)) return &FM[*I].LDat; + } + return 0; + } +}; + + + +/// A Lockset maps each SExpr (defined above) to information about how it has /// been locked. -typedef llvm::ImmutableMap<MutexID, LockData> Lockset; -typedef llvm::ImmutableMap<NamedDecl*, unsigned> LocalVarContext; +typedef llvm::ImmutableMap<SExpr, LockData> Lockset; +typedef llvm::ImmutableMap<const NamedDecl*, unsigned> LocalVarContext; class LocalVariableMap; @@ -345,15 +804,15 @@ enum CFGBlockSide { CBS_Entry, CBS_Exit }; /// maintained for each block in the CFG. See LocalVariableMap for more /// information about the contexts. struct CFGBlockInfo { - Lockset EntrySet; // Lockset held at entry to block - Lockset ExitSet; // Lockset held at exit from block + FactSet EntrySet; // Lockset held at entry to block + FactSet ExitSet; // Lockset held at exit from block LocalVarContext EntryContext; // Context held at entry to block LocalVarContext ExitContext; // Context held at exit from block SourceLocation EntryLoc; // Location of first statement in block SourceLocation ExitLoc; // Location of last statement in block. unsigned EntryIndex; // Used to replay contexts later - const Lockset &getSet(CFGBlockSide Side) const { + const FactSet &getSet(CFGBlockSide Side) const { return Side == CBS_Entry ? EntrySet : ExitSet; } SourceLocation getLocation(CFGBlockSide Side) const { @@ -361,14 +820,12 @@ struct CFGBlockInfo { } private: - CFGBlockInfo(Lockset EmptySet, LocalVarContext EmptyCtx) - : EntrySet(EmptySet), ExitSet(EmptySet), - EntryContext(EmptyCtx), ExitContext(EmptyCtx) + CFGBlockInfo(LocalVarContext EmptyCtx) + : EntryContext(EmptyCtx), ExitContext(EmptyCtx) { } public: - static CFGBlockInfo getEmptyBlockInfo(Lockset::Factory &F, - LocalVariableMap &M); + static CFGBlockInfo getEmptyBlockInfo(LocalVariableMap &M); }; @@ -398,21 +855,21 @@ public: public: friend class LocalVariableMap; - NamedDecl *Dec; // The original declaration for this variable. - Expr *Exp; // The expression for this variable, OR - unsigned Ref; // Reference to another VarDefinition - Context Ctx; // The map with which Exp should be interpreted. + const NamedDecl *Dec; // The original declaration for this variable. + const Expr *Exp; // The expression for this variable, OR + unsigned Ref; // Reference to another VarDefinition + Context Ctx; // The map with which Exp should be interpreted. bool isReference() { return !Exp; } private: // Create ordinary variable definition - VarDefinition(NamedDecl *D, Expr *E, Context C) + VarDefinition(const NamedDecl *D, const Expr *E, Context C) : Dec(D), Exp(E), Ref(0), Ctx(C) { } // Create reference to previous definition - VarDefinition(NamedDecl *D, unsigned R, Context C) + VarDefinition(const NamedDecl *D, unsigned R, Context C) : Dec(D), Exp(0), Ref(R), Ctx(C) { } }; @@ -430,7 +887,7 @@ public: } /// Look up a definition, within the given context. - const VarDefinition* lookup(NamedDecl *D, Context Ctx) { + const VarDefinition* lookup(const NamedDecl *D, Context Ctx) { const unsigned *i = Ctx.lookup(D); if (!i) return 0; @@ -441,7 +898,7 @@ public: /// Look up the definition for D within the given context. Returns /// NULL if the expression is not statically known. If successful, also /// modifies Ctx to hold the context of the return Expr. - Expr* lookupExpr(NamedDecl *D, Context &Ctx) { + const Expr* lookupExpr(const NamedDecl *D, Context &Ctx) { const unsigned *P = Ctx.lookup(D); if (!P) return 0; @@ -476,7 +933,7 @@ public: llvm::errs() << "Undefined"; return; } - NamedDecl *Dec = VarDefinitions[i].Dec; + const NamedDecl *Dec = VarDefinitions[i].Dec; if (!Dec) { llvm::errs() << "<<NULL>>"; return; @@ -488,7 +945,7 @@ public: /// Dumps an ASCII representation of the variable map to llvm::errs() void dump() { for (unsigned i = 1, e = VarDefinitions.size(); i < e; ++i) { - Expr *Exp = VarDefinitions[i].Exp; + const Expr *Exp = VarDefinitions[i].Exp; unsigned Ref = VarDefinitions[i].Ref; dumpVarDefinitionName(i); @@ -504,7 +961,7 @@ public: /// Dumps an ASCII representation of a Context to llvm::errs() void dumpContext(Context C) { for (Context::iterator I = C.begin(), E = C.end(); I != E; ++I) { - NamedDecl *D = I.getKey(); + const NamedDecl *D = I.getKey(); D->printName(llvm::errs()); const unsigned *i = C.lookup(D); llvm::errs() << " -> "; @@ -528,7 +985,7 @@ protected: // Adds a new definition to the given context, and returns a new context. // This method should be called when declaring a new variable. - Context addDefinition(NamedDecl *D, Expr *Exp, Context Ctx) { + Context addDefinition(const NamedDecl *D, Expr *Exp, Context Ctx) { assert(!Ctx.contains(D)); unsigned newID = VarDefinitions.size(); Context NewCtx = ContextFactory.add(Ctx, D, newID); @@ -537,7 +994,7 @@ protected: } // Add a new reference to an existing definition. - Context addReference(NamedDecl *D, unsigned i, Context Ctx) { + Context addReference(const NamedDecl *D, unsigned i, Context Ctx) { unsigned newID = VarDefinitions.size(); Context NewCtx = ContextFactory.add(Ctx, D, newID); VarDefinitions.push_back(VarDefinition(D, i, Ctx)); @@ -546,7 +1003,7 @@ protected: // Updates a definition only if that definition is already in the map. // This method should be called when assigning to an existing variable. - Context updateDefinition(NamedDecl *D, Expr *Exp, Context Ctx) { + Context updateDefinition(const NamedDecl *D, Expr *Exp, Context Ctx) { if (Ctx.contains(D)) { unsigned newID = VarDefinitions.size(); Context NewCtx = ContextFactory.remove(Ctx, D); @@ -559,7 +1016,7 @@ protected: // Removes a definition from the context, but keeps the variable name // as a valid variable. The index 0 is a placeholder for cleared definitions. - Context clearDefinition(NamedDecl *D, Context Ctx) { + Context clearDefinition(const NamedDecl *D, Context Ctx) { Context NewCtx = Ctx; if (NewCtx.contains(D)) { NewCtx = ContextFactory.remove(NewCtx, D); @@ -569,7 +1026,7 @@ protected: } // Remove a definition entirely frmo the context. - Context removeDefinition(NamedDecl *D, Context Ctx) { + Context removeDefinition(const NamedDecl *D, Context Ctx) { Context NewCtx = Ctx; if (NewCtx.contains(D)) { NewCtx = ContextFactory.remove(NewCtx, D); @@ -586,9 +1043,8 @@ protected: // This has to be defined after LocalVariableMap. -CFGBlockInfo CFGBlockInfo::getEmptyBlockInfo(Lockset::Factory &F, - LocalVariableMap &M) { - return CFGBlockInfo(F.getEmptyMap(), M.getEmptyContext()); +CFGBlockInfo CFGBlockInfo::getEmptyBlockInfo(LocalVariableMap &M) { + return CFGBlockInfo(M.getEmptyContext()); } @@ -655,7 +1111,7 @@ LocalVariableMap::Context LocalVariableMap::intersectContexts(Context C1, Context C2) { Context Result = C1; for (Context::iterator I = C1.begin(), E = C1.end(); I != E; ++I) { - NamedDecl *Dec = I.getKey(); + const NamedDecl *Dec = I.getKey(); unsigned i1 = I.getData(); const unsigned *i2 = C2.lookup(Dec); if (!i2) // variable doesn't exist on second path @@ -672,7 +1128,7 @@ LocalVariableMap::intersectContexts(Context C1, Context C2) { LocalVariableMap::Context LocalVariableMap::createReferenceContext(Context C) { Context Result = getEmptyContext(); for (Context::iterator I = C.begin(), E = C.end(); I != E; ++I) { - NamedDecl *Dec = I.getKey(); + const NamedDecl *Dec = I.getKey(); unsigned i = I.getData(); Result = addReference(Dec, i, Result); } @@ -684,7 +1140,7 @@ LocalVariableMap::Context LocalVariableMap::createReferenceContext(Context C) { // createReferenceContext. void LocalVariableMap::intersectBackEdge(Context C1, Context C2) { for (Context::iterator I = C1.begin(), E = C1.end(); I != E; ++I) { - NamedDecl *Dec = I.getKey(); + const NamedDecl *Dec = I.getKey(); unsigned i1 = I.getData(); VarDefinition *VDef = &VarDefinitions[i1]; assert(VDef->isReference()); @@ -725,7 +1181,7 @@ void LocalVariableMap::intersectBackEdge(Context C1, Context C2) { // incoming back edge, it duplicates the context, creating new definitions // that refer back to the originals. (These correspond to places where SSA // might have to insert a phi node.) On the second pass, these definitions are -// set to NULL if the the variable has changed on the back-edge (i.e. a phi +// set to NULL if the variable has changed on the back-edge (i.e. a phi // node was actually required.) E.g. // // { Context | VarDefinitions } @@ -869,24 +1325,294 @@ static void findBlockLocations(CFG *CFGraph, class ThreadSafetyAnalyzer { friend class BuildLockset; - ThreadSafetyHandler &Handler; - Lockset::Factory LocksetFactory; - LocalVariableMap LocalVarMap; + ThreadSafetyHandler &Handler; + LocalVariableMap LocalVarMap; + FactManager FactMan; + std::vector<CFGBlockInfo> BlockInfo; public: ThreadSafetyAnalyzer(ThreadSafetyHandler &H) : Handler(H) {} - Lockset intersectAndWarn(const CFGBlockInfo &Block1, CFGBlockSide Side1, - const CFGBlockInfo &Block2, CFGBlockSide Side2, - LockErrorKind LEK); + void addLock(FactSet &FSet, const SExpr &Mutex, const LockData &LDat); + void removeLock(FactSet &FSet, const SExpr &Mutex, + SourceLocation UnlockLoc, bool FullyRemove=false); + + template <typename AttrType> + void getMutexIDs(MutexIDList &Mtxs, AttrType *Attr, Expr *Exp, + const NamedDecl *D); - Lockset addLock(Lockset &LSet, Expr *MutexExp, const NamedDecl *D, - LockKind LK, SourceLocation Loc); + template <class AttrType> + void getMutexIDs(MutexIDList &Mtxs, AttrType *Attr, Expr *Exp, + const NamedDecl *D, + const CFGBlock *PredBlock, const CFGBlock *CurrBlock, + Expr *BrE, bool Neg); + + const CallExpr* getTrylockCallExpr(const Stmt *Cond, LocalVarContext C, + bool &Negate); + + void getEdgeLockset(FactSet &Result, const FactSet &ExitSet, + const CFGBlock* PredBlock, + const CFGBlock *CurrBlock); + + void intersectAndWarn(FactSet &FSet1, const FactSet &FSet2, + SourceLocation JoinLoc, + LockErrorKind LEK1, LockErrorKind LEK2, + bool Modify=true); + + void intersectAndWarn(FactSet &FSet1, const FactSet &FSet2, + SourceLocation JoinLoc, LockErrorKind LEK1, + bool Modify=true) { + intersectAndWarn(FSet1, FSet2, JoinLoc, LEK1, LEK1, Modify); + } void runAnalysis(AnalysisDeclContext &AC); }; +/// \brief Add a new lock to the lockset, warning if the lock is already there. +/// \param Mutex -- the Mutex expression for the lock +/// \param LDat -- the LockData for the lock +void ThreadSafetyAnalyzer::addLock(FactSet &FSet, const SExpr &Mutex, + const LockData &LDat) { + // FIXME: deal with acquired before/after annotations. + // FIXME: Don't always warn when we have support for reentrant locks. + if (FSet.findLock(FactMan, Mutex)) { + Handler.handleDoubleLock(Mutex.toString(), LDat.AcquireLoc); + } else { + FSet.addLock(FactMan, Mutex, LDat); + } +} + + +/// \brief Remove a lock from the lockset, warning if the lock is not there. +/// \param LockExp The lock expression corresponding to the lock to be removed +/// \param UnlockLoc The source location of the unlock (only used in error msg) +void ThreadSafetyAnalyzer::removeLock(FactSet &FSet, + const SExpr &Mutex, + SourceLocation UnlockLoc, + bool FullyRemove) { + const LockData *LDat = FSet.findLock(FactMan, Mutex); + if (!LDat) { + Handler.handleUnmatchedUnlock(Mutex.toString(), UnlockLoc); + return; + } + + if (LDat->UnderlyingMutex.isValid()) { + // This is scoped lockable object, which manages the real mutex. + if (FullyRemove) { + // We're destroying the managing object. + // Remove the underlying mutex if it exists; but don't warn. + if (FSet.findLock(FactMan, LDat->UnderlyingMutex)) + FSet.removeLock(FactMan, LDat->UnderlyingMutex); + } else { + // We're releasing the underlying mutex, but not destroying the + // managing object. Warn on dual release. + if (!FSet.findLock(FactMan, LDat->UnderlyingMutex)) { + Handler.handleUnmatchedUnlock(LDat->UnderlyingMutex.toString(), + UnlockLoc); + } + FSet.removeLock(FactMan, LDat->UnderlyingMutex); + return; + } + } + FSet.removeLock(FactMan, Mutex); +} + + +/// \brief Extract the list of mutexIDs from the attribute on an expression, +/// and push them onto Mtxs, discarding any duplicates. +template <typename AttrType> +void ThreadSafetyAnalyzer::getMutexIDs(MutexIDList &Mtxs, AttrType *Attr, + Expr *Exp, const NamedDecl *D) { + typedef typename AttrType::args_iterator iterator_type; + + if (Attr->args_size() == 0) { + // The mutex held is the "this" object. + SExpr Mu(0, Exp, D); + if (!Mu.isValid()) + SExpr::warnInvalidLock(Handler, 0, Exp, D); + else + Mtxs.push_back_nodup(Mu); + return; + } + + for (iterator_type I=Attr->args_begin(), E=Attr->args_end(); I != E; ++I) { + SExpr Mu(*I, Exp, D); + if (!Mu.isValid()) + SExpr::warnInvalidLock(Handler, *I, Exp, D); + else + Mtxs.push_back_nodup(Mu); + } +} + + +/// \brief Extract the list of mutexIDs from a trylock attribute. If the +/// trylock applies to the given edge, then push them onto Mtxs, discarding +/// any duplicates. +template <class AttrType> +void ThreadSafetyAnalyzer::getMutexIDs(MutexIDList &Mtxs, AttrType *Attr, + Expr *Exp, const NamedDecl *D, + const CFGBlock *PredBlock, + const CFGBlock *CurrBlock, + Expr *BrE, bool Neg) { + // Find out which branch has the lock + bool branch = 0; + if (CXXBoolLiteralExpr *BLE = dyn_cast_or_null<CXXBoolLiteralExpr>(BrE)) { + branch = BLE->getValue(); + } + else if (IntegerLiteral *ILE = dyn_cast_or_null<IntegerLiteral>(BrE)) { + branch = ILE->getValue().getBoolValue(); + } + int branchnum = branch ? 0 : 1; + if (Neg) branchnum = !branchnum; + + // If we've taken the trylock branch, then add the lock + int i = 0; + for (CFGBlock::const_succ_iterator SI = PredBlock->succ_begin(), + SE = PredBlock->succ_end(); SI != SE && i < 2; ++SI, ++i) { + if (*SI == CurrBlock && i == branchnum) { + getMutexIDs(Mtxs, Attr, Exp, D); + } + } +} + + +bool getStaticBooleanValue(Expr* E, bool& TCond) { + if (isa<CXXNullPtrLiteralExpr>(E) || isa<GNUNullExpr>(E)) { + TCond = false; + return true; + } else if (CXXBoolLiteralExpr *BLE = dyn_cast<CXXBoolLiteralExpr>(E)) { + TCond = BLE->getValue(); + return true; + } else if (IntegerLiteral *ILE = dyn_cast<IntegerLiteral>(E)) { + TCond = ILE->getValue().getBoolValue(); + return true; + } else if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(E)) { + return getStaticBooleanValue(CE->getSubExpr(), TCond); + } + return false; +} + + +// If Cond can be traced back to a function call, return the call expression. +// The negate variable should be called with false, and will be set to true +// if the function call is negated, e.g. if (!mu.tryLock(...)) +const CallExpr* ThreadSafetyAnalyzer::getTrylockCallExpr(const Stmt *Cond, + LocalVarContext C, + bool &Negate) { + if (!Cond) + return 0; + + if (const CallExpr *CallExp = dyn_cast<CallExpr>(Cond)) { + return CallExp; + } + else if (const ParenExpr *PE = dyn_cast<ParenExpr>(Cond)) { + return getTrylockCallExpr(PE->getSubExpr(), C, Negate); + } + else if (const ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(Cond)) { + return getTrylockCallExpr(CE->getSubExpr(), C, Negate); + } + else if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(Cond)) { + const Expr *E = LocalVarMap.lookupExpr(DRE->getDecl(), C); + return getTrylockCallExpr(E, C, Negate); + } + else if (const UnaryOperator *UOP = dyn_cast<UnaryOperator>(Cond)) { + if (UOP->getOpcode() == UO_LNot) { + Negate = !Negate; + return getTrylockCallExpr(UOP->getSubExpr(), C, Negate); + } + return 0; + } + else if (const BinaryOperator *BOP = dyn_cast<BinaryOperator>(Cond)) { + if (BOP->getOpcode() == BO_EQ || BOP->getOpcode() == BO_NE) { + if (BOP->getOpcode() == BO_NE) + Negate = !Negate; + + bool TCond = false; + if (getStaticBooleanValue(BOP->getRHS(), TCond)) { + if (!TCond) Negate = !Negate; + return getTrylockCallExpr(BOP->getLHS(), C, Negate); + } + else if (getStaticBooleanValue(BOP->getLHS(), TCond)) { + if (!TCond) Negate = !Negate; + return getTrylockCallExpr(BOP->getRHS(), C, Negate); + } + return 0; + } + return 0; + } + // FIXME -- handle && and || as well. + return 0; +} + + +/// \brief Find the lockset that holds on the edge between PredBlock +/// and CurrBlock. The edge set is the exit set of PredBlock (passed +/// as the ExitSet parameter) plus any trylocks, which are conditionally held. +void ThreadSafetyAnalyzer::getEdgeLockset(FactSet& Result, + const FactSet &ExitSet, + const CFGBlock *PredBlock, + const CFGBlock *CurrBlock) { + Result = ExitSet; + + if (!PredBlock->getTerminatorCondition()) + return; + + bool Negate = false; + const Stmt *Cond = PredBlock->getTerminatorCondition(); + const CFGBlockInfo *PredBlockInfo = &BlockInfo[PredBlock->getBlockID()]; + const LocalVarContext &LVarCtx = PredBlockInfo->ExitContext; + + CallExpr *Exp = + const_cast<CallExpr*>(getTrylockCallExpr(Cond, LVarCtx, Negate)); + if (!Exp) + return; + + NamedDecl *FunDecl = dyn_cast_or_null<NamedDecl>(Exp->getCalleeDecl()); + if(!FunDecl || !FunDecl->hasAttrs()) + return; + + + MutexIDList ExclusiveLocksToAdd; + MutexIDList SharedLocksToAdd; + + // If the condition is a call to a Trylock function, then grab the attributes + AttrVec &ArgAttrs = FunDecl->getAttrs(); + for (unsigned i = 0; i < ArgAttrs.size(); ++i) { + Attr *Attr = ArgAttrs[i]; + switch (Attr->getKind()) { + case attr::ExclusiveTrylockFunction: { + ExclusiveTrylockFunctionAttr *A = + cast<ExclusiveTrylockFunctionAttr>(Attr); + getMutexIDs(ExclusiveLocksToAdd, A, Exp, FunDecl, + PredBlock, CurrBlock, A->getSuccessValue(), Negate); + break; + } + case attr::SharedTrylockFunction: { + SharedTrylockFunctionAttr *A = + cast<SharedTrylockFunctionAttr>(Attr); + getMutexIDs(ExclusiveLocksToAdd, A, Exp, FunDecl, + PredBlock, CurrBlock, A->getSuccessValue(), Negate); + break; + } + default: + break; + } + } + + // Add and remove locks. + SourceLocation Loc = Exp->getExprLoc(); + for (unsigned i=0,n=ExclusiveLocksToAdd.size(); i<n; ++i) { + addLock(Result, ExclusiveLocksToAdd[i], + LockData(Loc, LK_Exclusive)); + } + for (unsigned i=0,n=SharedLocksToAdd.size(); i<n; ++i) { + addLock(Result, SharedLocksToAdd[i], + LockData(Loc, LK_Shared)); + } +} + + /// \brief We use this class to visit different types of expressions in /// CFGBlocks, and build up the lockset. /// An expression may cause us to add or remove locks from the lockset, or else @@ -895,50 +1621,31 @@ public: class BuildLockset : public StmtVisitor<BuildLockset> { friend class ThreadSafetyAnalyzer; - ThreadSafetyHandler &Handler; - Lockset::Factory &LocksetFactory; - LocalVariableMap &LocalVarMap; - - Lockset LSet; + ThreadSafetyAnalyzer *Analyzer; + FactSet FSet; LocalVariableMap::Context LVarCtx; unsigned CtxIndex; // Helper functions - void addLock(const MutexID &Mutex, const LockData &LDat); - void removeLock(const MutexID &Mutex, SourceLocation UnlockLoc); + const ValueDecl *getValueDecl(Expr *Exp); - template <class AttrType> - void addLocksToSet(LockKind LK, AttrType *Attr, - Expr *Exp, NamedDecl *D, VarDecl *VD = 0); - void removeLocksFromSet(UnlockFunctionAttr *Attr, - Expr *Exp, NamedDecl* FunDecl); + void warnIfMutexNotHeld(const NamedDecl *D, Expr *Exp, AccessKind AK, + Expr *MutexExp, ProtectedOperationKind POK); - const ValueDecl *getValueDecl(Expr *Exp); - void warnIfMutexNotHeld (const NamedDecl *D, Expr *Exp, AccessKind AK, - Expr *MutexExp, ProtectedOperationKind POK); void checkAccess(Expr *Exp, AccessKind AK); void checkDereference(Expr *Exp, AccessKind AK); - void handleCall(Expr *Exp, NamedDecl *D, VarDecl *VD = 0); - - template <class AttrType> - void addTrylock(LockKind LK, AttrType *Attr, Expr *Exp, NamedDecl *FunDecl, - const CFGBlock* PredBlock, const CFGBlock *CurrBlock, - Expr *BrE, bool Neg); - CallExpr* getTrylockCallExpr(Stmt *Cond, LocalVariableMap::Context C, - bool &Negate); - void handleTrylock(Stmt *Cond, const CFGBlock* PredBlock, - const CFGBlock *CurrBlock); + void handleCall(Expr *Exp, const NamedDecl *D, VarDecl *VD = 0); /// \brief Returns true if the lockset contains a lock, regardless of whether /// the lock is held exclusively or shared. - bool locksetContains(const MutexID &Lock) const { - return LSet.lookup(Lock); + bool locksetContains(const SExpr &Mu) const { + return FSet.findLock(Analyzer->FactMan, Mu); } /// \brief Returns true if the lockset contains a lock with the passed in /// locktype. - bool locksetContains(const MutexID &Lock, LockKind KindRequested) const { - const LockData *LockHeld = LSet.lookup(Lock); + bool locksetContains(const SExpr &Mu, LockKind KindRequested) const { + const LockData *LockHeld = FSet.findLock(Analyzer->FactMan, Mu); return (LockHeld && KindRequested == LockHeld->LKind); } @@ -946,7 +1653,7 @@ class BuildLockset : public StmtVisitor<BuildLockset> { /// passed in locktype. So for example, if we pass in LK_Shared, this function /// returns true if the lock is held LK_Shared or LK_Exclusive. If we pass in /// LK_Exclusive, this function returns true if the lock is held LK_Exclusive. - bool locksetContainsAtLeast(const MutexID &Lock, + bool locksetContainsAtLeast(const SExpr &Lock, LockKind KindRequested) const { switch (KindRequested) { case LK_Shared: @@ -958,12 +1665,10 @@ class BuildLockset : public StmtVisitor<BuildLockset> { } public: - BuildLockset(ThreadSafetyAnalyzer *analyzer, CFGBlockInfo &Info) + BuildLockset(ThreadSafetyAnalyzer *Anlzr, CFGBlockInfo &Info) : StmtVisitor<BuildLockset>(), - Handler(analyzer->Handler), - LocksetFactory(analyzer->LocksetFactory), - LocalVarMap(analyzer->LocalVarMap), - LSet(Info.EntrySet), + Analyzer(Anlzr), + FSet(Info.EntrySet), LVarCtx(Info.EntryContext), CtxIndex(Info.EntryIndex) {} @@ -976,104 +1681,6 @@ public: void VisitDeclStmt(DeclStmt *S); }; -/// \brief Add a new lock to the lockset, warning if the lock is already there. -/// \param Mutex -- the Mutex expression for the lock -/// \param LDat -- the LockData for the lock -void BuildLockset::addLock(const MutexID &Mutex, const LockData& LDat) { - // FIXME: deal with acquired before/after annotations. - // FIXME: Don't always warn when we have support for reentrant locks. - if (locksetContains(Mutex)) - Handler.handleDoubleLock(Mutex.getName(), LDat.AcquireLoc); - else - LSet = LocksetFactory.add(LSet, Mutex, LDat); -} - -/// \brief Remove a lock from the lockset, warning if the lock is not there. -/// \param LockExp The lock expression corresponding to the lock to be removed -/// \param UnlockLoc The source location of the unlock (only used in error msg) -void BuildLockset::removeLock(const MutexID &Mutex, SourceLocation UnlockLoc) { - const LockData *LDat = LSet.lookup(Mutex); - if (!LDat) - Handler.handleUnmatchedUnlock(Mutex.getName(), UnlockLoc); - else { - // For scoped-lockable vars, remove the mutex associated with this var. - if (LDat->UnderlyingMutex.isValid()) - removeLock(LDat->UnderlyingMutex, UnlockLoc); - LSet = LocksetFactory.remove(LSet, Mutex); - } -} - -/// \brief This function, parameterized by an attribute type, is used to add a -/// set of locks specified as attribute arguments to the lockset. -template <typename AttrType> -void BuildLockset::addLocksToSet(LockKind LK, AttrType *Attr, - Expr *Exp, NamedDecl* FunDecl, VarDecl *VD) { - typedef typename AttrType::args_iterator iterator_type; - - SourceLocation ExpLocation = Exp->getExprLoc(); - - // Figure out if we're calling the constructor of scoped lockable class - bool isScopedVar = false; - if (VD) { - if (CXXConstructorDecl *CD = dyn_cast<CXXConstructorDecl>(FunDecl)) { - CXXRecordDecl* PD = CD->getParent(); - if (PD && PD->getAttr<ScopedLockableAttr>()) - isScopedVar = true; - } - } - - if (Attr->args_size() == 0) { - // The mutex held is the "this" object. - MutexID Mutex(0, Exp, FunDecl); - if (!Mutex.isValid()) - MutexID::warnInvalidLock(Handler, 0, Exp, FunDecl); - else - addLock(Mutex, LockData(ExpLocation, LK)); - return; - } - - for (iterator_type I=Attr->args_begin(), E=Attr->args_end(); I != E; ++I) { - MutexID Mutex(*I, Exp, FunDecl); - if (!Mutex.isValid()) - MutexID::warnInvalidLock(Handler, *I, Exp, FunDecl); - else { - addLock(Mutex, LockData(ExpLocation, LK)); - if (isScopedVar) { - // For scoped lockable vars, map this var to its underlying mutex. - DeclRefExpr DRE(VD, false, VD->getType(), VK_LValue, VD->getLocation()); - MutexID SMutex(&DRE, 0, 0); - addLock(SMutex, LockData(VD->getLocation(), LK, Mutex)); - } - } - } -} - -/// \brief This function removes a set of locks specified as attribute -/// arguments from the lockset. -void BuildLockset::removeLocksFromSet(UnlockFunctionAttr *Attr, - Expr *Exp, NamedDecl* FunDecl) { - SourceLocation ExpLocation; - if (Exp) ExpLocation = Exp->getExprLoc(); - - if (Attr->args_size() == 0) { - // The mutex held is the "this" object. - MutexID Mu(0, Exp, FunDecl); - if (!Mu.isValid()) - MutexID::warnInvalidLock(Handler, 0, Exp, FunDecl); - else - removeLock(Mu, ExpLocation); - return; - } - - for (UnlockFunctionAttr::args_iterator I = Attr->args_begin(), - E = Attr->args_end(); I != E; ++I) { - MutexID Mutex(*I, Exp, FunDecl); - if (!Mutex.isValid()) - MutexID::warnInvalidLock(Handler, *I, Exp, FunDecl); - else - removeLock(Mutex, ExpLocation); - } -} /// \brief Gets the value decl pointer from DeclRefExprs or MemberExprs const ValueDecl *BuildLockset::getValueDecl(Expr *Exp) { @@ -1093,11 +1700,12 @@ void BuildLockset::warnIfMutexNotHeld(const NamedDecl *D, Expr *Exp, ProtectedOperationKind POK) { LockKind LK = getLockKindFromAccessKind(AK); - MutexID Mutex(MutexExp, Exp, D); + SExpr Mutex(MutexExp, Exp, D); if (!Mutex.isValid()) - MutexID::warnInvalidLock(Handler, MutexExp, Exp, D); + SExpr::warnInvalidLock(Analyzer->Handler, MutexExp, Exp, D); else if (!locksetContainsAtLeast(Mutex, LK)) - Handler.handleMutexNotHeld(D, POK, Mutex.getName(), LK, Exp->getExprLoc()); + Analyzer->Handler.handleMutexNotHeld(D, POK, Mutex.toString(), LK, + Exp->getExprLoc()); } /// \brief This method identifies variable dereferences and checks pt_guarded_by @@ -1116,8 +1724,9 @@ void BuildLockset::checkDereference(Expr *Exp, AccessKind AK) { if(!D || !D->hasAttrs()) return; - if (D->getAttr<PtGuardedVarAttr>() && LSet.isEmpty()) - Handler.handleNoMutexHeld(D, POK_VarDereference, AK, Exp->getExprLoc()); + if (D->getAttr<PtGuardedVarAttr>() && FSet.isEmpty()) + Analyzer->Handler.handleNoMutexHeld(D, POK_VarDereference, AK, + Exp->getExprLoc()); const AttrVec &ArgAttrs = D->getAttrs(); for(unsigned i = 0, Size = ArgAttrs.size(); i < Size; ++i) @@ -1134,8 +1743,9 @@ void BuildLockset::checkAccess(Expr *Exp, AccessKind AK) { if(!D || !D->hasAttrs()) return; - if (D->getAttr<GuardedVarAttr>() && LSet.isEmpty()) - Handler.handleNoMutexHeld(D, POK_VarAccess, AK, Exp->getExprLoc()); + if (D->getAttr<GuardedVarAttr>() && FSet.isEmpty()) + Analyzer->Handler.handleNoMutexHeld(D, POK_VarAccess, AK, + Exp->getExprLoc()); const AttrVec &ArgAttrs = D->getAttrs(); for(unsigned i = 0, Size = ArgAttrs.size(); i < Size; ++i) @@ -1153,68 +1763,68 @@ void BuildLockset::checkAccess(Expr *Exp, AccessKind AK) { /// and check that the appropriate locks are held. Non-const method calls with /// the same signature as const method calls can be also treated as reads. /// -/// FIXME: We need to also visit CallExprs to catch/check global functions. -/// -/// FIXME: Do not flag an error for member variables accessed in constructors/ -/// destructors -void BuildLockset::handleCall(Expr *Exp, NamedDecl *D, VarDecl *VD) { - AttrVec &ArgAttrs = D->getAttrs(); +void BuildLockset::handleCall(Expr *Exp, const NamedDecl *D, VarDecl *VD) { + const AttrVec &ArgAttrs = D->getAttrs(); + MutexIDList ExclusiveLocksToAdd; + MutexIDList SharedLocksToAdd; + MutexIDList LocksToRemove; + for(unsigned i = 0; i < ArgAttrs.size(); ++i) { - Attr *Attr = ArgAttrs[i]; - switch (Attr->getKind()) { + Attr *At = const_cast<Attr*>(ArgAttrs[i]); + switch (At->getKind()) { // When we encounter an exclusive lock function, we need to add the lock // to our lockset with kind exclusive. case attr::ExclusiveLockFunction: { - ExclusiveLockFunctionAttr *A = cast<ExclusiveLockFunctionAttr>(Attr); - addLocksToSet(LK_Exclusive, A, Exp, D, VD); + ExclusiveLockFunctionAttr *A = cast<ExclusiveLockFunctionAttr>(At); + Analyzer->getMutexIDs(ExclusiveLocksToAdd, A, Exp, D); break; } // When we encounter a shared lock function, we need to add the lock // to our lockset with kind shared. case attr::SharedLockFunction: { - SharedLockFunctionAttr *A = cast<SharedLockFunctionAttr>(Attr); - addLocksToSet(LK_Shared, A, Exp, D, VD); + SharedLockFunctionAttr *A = cast<SharedLockFunctionAttr>(At); + Analyzer->getMutexIDs(SharedLocksToAdd, A, Exp, D); break; } // When we encounter an unlock function, we need to remove unlocked // mutexes from the lockset, and flag a warning if they are not there. case attr::UnlockFunction: { - UnlockFunctionAttr *UFAttr = cast<UnlockFunctionAttr>(Attr); - removeLocksFromSet(UFAttr, Exp, D); + UnlockFunctionAttr *A = cast<UnlockFunctionAttr>(At); + Analyzer->getMutexIDs(LocksToRemove, A, Exp, D); break; } case attr::ExclusiveLocksRequired: { - ExclusiveLocksRequiredAttr *ELRAttr = - cast<ExclusiveLocksRequiredAttr>(Attr); + ExclusiveLocksRequiredAttr *A = cast<ExclusiveLocksRequiredAttr>(At); for (ExclusiveLocksRequiredAttr::args_iterator - I = ELRAttr->args_begin(), E = ELRAttr->args_end(); I != E; ++I) + I = A->args_begin(), E = A->args_end(); I != E; ++I) warnIfMutexNotHeld(D, Exp, AK_Written, *I, POK_FunctionCall); break; } case attr::SharedLocksRequired: { - SharedLocksRequiredAttr *SLRAttr = cast<SharedLocksRequiredAttr>(Attr); + SharedLocksRequiredAttr *A = cast<SharedLocksRequiredAttr>(At); - for (SharedLocksRequiredAttr::args_iterator I = SLRAttr->args_begin(), - E = SLRAttr->args_end(); I != E; ++I) + for (SharedLocksRequiredAttr::args_iterator I = A->args_begin(), + E = A->args_end(); I != E; ++I) warnIfMutexNotHeld(D, Exp, AK_Read, *I, POK_FunctionCall); break; } case attr::LocksExcluded: { - LocksExcludedAttr *LEAttr = cast<LocksExcludedAttr>(Attr); - for (LocksExcludedAttr::args_iterator I = LEAttr->args_begin(), - E = LEAttr->args_end(); I != E; ++I) { - MutexID Mutex(*I, Exp, D); + LocksExcludedAttr *A = cast<LocksExcludedAttr>(At); + for (LocksExcludedAttr::args_iterator I = A->args_begin(), + E = A->args_end(); I != E; ++I) { + SExpr Mutex(*I, Exp, D); if (!Mutex.isValid()) - MutexID::warnInvalidLock(Handler, *I, Exp, D); + SExpr::warnInvalidLock(Analyzer->Handler, *I, Exp, D); else if (locksetContains(Mutex)) - Handler.handleFunExcludesLock(D->getName(), Mutex.getName(), - Exp->getExprLoc()); + Analyzer->Handler.handleFunExcludesLock(D->getName(), + Mutex.toString(), + Exp->getExprLoc()); } break; } @@ -1224,102 +1834,50 @@ void BuildLockset::handleCall(Expr *Exp, NamedDecl *D, VarDecl *VD) { break; } } -} - - -/// \brief Add lock to set, if the current block is in the taken branch of a -/// trylock. -template <class AttrType> -void BuildLockset::addTrylock(LockKind LK, AttrType *Attr, Expr *Exp, - NamedDecl *FunDecl, const CFGBlock *PredBlock, - const CFGBlock *CurrBlock, Expr *BrE, bool Neg) { - // Find out which branch has the lock - bool branch = 0; - if (CXXBoolLiteralExpr *BLE = dyn_cast_or_null<CXXBoolLiteralExpr>(BrE)) { - branch = BLE->getValue(); - } - else if (IntegerLiteral *ILE = dyn_cast_or_null<IntegerLiteral>(BrE)) { - branch = ILE->getValue().getBoolValue(); - } - int branchnum = branch ? 0 : 1; - if (Neg) branchnum = !branchnum; - // If we've taken the trylock branch, then add the lock - int i = 0; - for (CFGBlock::const_succ_iterator SI = PredBlock->succ_begin(), - SE = PredBlock->succ_end(); SI != SE && i < 2; ++SI, ++i) { - if (*SI == CurrBlock && i == branchnum) { - addLocksToSet(LK, Attr, Exp, FunDecl, 0); + // Figure out if we're calling the constructor of scoped lockable class + bool isScopedVar = false; + if (VD) { + if (const CXXConstructorDecl *CD = dyn_cast<const CXXConstructorDecl>(D)) { + const CXXRecordDecl* PD = CD->getParent(); + if (PD && PD->getAttr<ScopedLockableAttr>()) + isScopedVar = true; } } -} - -// If Cond can be traced back to a function call, return the call expression. -// The negate variable should be called with false, and will be set to true -// if the function call is negated, e.g. if (!mu.tryLock(...)) -CallExpr* BuildLockset::getTrylockCallExpr(Stmt *Cond, - LocalVariableMap::Context C, - bool &Negate) { - if (!Cond) - return 0; - - if (CallExpr *CallExp = dyn_cast<CallExpr>(Cond)) { - return CallExp; - } - else if (ImplicitCastExpr *CE = dyn_cast<ImplicitCastExpr>(Cond)) { - return getTrylockCallExpr(CE->getSubExpr(), C, Negate); + // Add locks. + SourceLocation Loc = Exp->getExprLoc(); + for (unsigned i=0,n=ExclusiveLocksToAdd.size(); i<n; ++i) { + Analyzer->addLock(FSet, ExclusiveLocksToAdd[i], + LockData(Loc, LK_Exclusive, isScopedVar)); } - else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(Cond)) { - Expr *E = LocalVarMap.lookupExpr(DRE->getDecl(), C); - return getTrylockCallExpr(E, C, Negate); - } - else if (UnaryOperator *UOP = dyn_cast<UnaryOperator>(Cond)) { - if (UOP->getOpcode() == UO_LNot) { - Negate = !Negate; - return getTrylockCallExpr(UOP->getSubExpr(), C, Negate); - } + for (unsigned i=0,n=SharedLocksToAdd.size(); i<n; ++i) { + Analyzer->addLock(FSet, SharedLocksToAdd[i], + LockData(Loc, LK_Shared, isScopedVar)); } - // FIXME -- handle && and || as well. - return NULL; -} - - -/// \brief Process a conditional branch from a previous block to the current -/// block, looking for trylock calls. -void BuildLockset::handleTrylock(Stmt *Cond, const CFGBlock *PredBlock, - const CFGBlock *CurrBlock) { - bool Negate = false; - CallExpr *Exp = getTrylockCallExpr(Cond, LVarCtx, Negate); - if (!Exp) - return; - NamedDecl *FunDecl = dyn_cast_or_null<NamedDecl>(Exp->getCalleeDecl()); - if(!FunDecl || !FunDecl->hasAttrs()) - return; + // Add the managing object as a dummy mutex, mapped to the underlying mutex. + // FIXME -- this doesn't work if we acquire multiple locks. + if (isScopedVar) { + SourceLocation MLoc = VD->getLocation(); + DeclRefExpr DRE(VD, false, VD->getType(), VK_LValue, VD->getLocation()); + SExpr SMutex(&DRE, 0, 0); - // If the condition is a call to a Trylock function, then grab the attributes - AttrVec &ArgAttrs = FunDecl->getAttrs(); - for (unsigned i = 0; i < ArgAttrs.size(); ++i) { - Attr *Attr = ArgAttrs[i]; - switch (Attr->getKind()) { - case attr::ExclusiveTrylockFunction: { - ExclusiveTrylockFunctionAttr *A = - cast<ExclusiveTrylockFunctionAttr>(Attr); - addTrylock(LK_Exclusive, A, Exp, FunDecl, PredBlock, CurrBlock, - A->getSuccessValue(), Negate); - break; - } - case attr::SharedTrylockFunction: { - SharedTrylockFunctionAttr *A = - cast<SharedTrylockFunctionAttr>(Attr); - addTrylock(LK_Shared, A, Exp, FunDecl, PredBlock, CurrBlock, - A->getSuccessValue(), Negate); - break; - } - default: - break; + for (unsigned i=0,n=ExclusiveLocksToAdd.size(); i<n; ++i) { + Analyzer->addLock(FSet, SMutex, LockData(MLoc, LK_Exclusive, + ExclusiveLocksToAdd[i])); } + for (unsigned i=0,n=SharedLocksToAdd.size(); i<n; ++i) { + Analyzer->addLock(FSet, SMutex, LockData(MLoc, LK_Shared, + SharedLocksToAdd[i])); + } + } + + // Remove locks. + // FIXME -- should only fully remove if the attribute refers to 'this'. + bool Dtor = isa<CXXDestructorDecl>(D); + for (unsigned i=0,n=LocksToRemove.size(); i<n; ++i) { + Analyzer->removeLock(FSet, LocksToRemove[i], Loc, Dtor); } } @@ -1351,7 +1909,7 @@ void BuildLockset::VisitBinaryOperator(BinaryOperator *BO) { return; // adjust the context - LVarCtx = LocalVarMap.getNextContext(CtxIndex, BO, LVarCtx); + LVarCtx = Analyzer->LocalVarMap.getNextContext(CtxIndex, BO, LVarCtx); Expr *LHSExp = BO->getLHS()->IgnoreParenCasts(); checkAccess(LHSExp, AK_Written); @@ -1383,13 +1941,17 @@ void BuildLockset::VisitCXXConstructExpr(CXXConstructExpr *Exp) { void BuildLockset::VisitDeclStmt(DeclStmt *S) { // adjust the context - LVarCtx = LocalVarMap.getNextContext(CtxIndex, S, LVarCtx); + LVarCtx = Analyzer->LocalVarMap.getNextContext(CtxIndex, S, LVarCtx); DeclGroupRef DGrp = S->getDeclGroup(); for (DeclGroupRef::iterator I = DGrp.begin(), E = DGrp.end(); I != E; ++I) { Decl *D = *I; if (VarDecl *VD = dyn_cast_or_null<VarDecl>(D)) { Expr *E = VD->getInit(); + // handle constructors that involve temporaries + if (ExprWithCleanups *EWC = dyn_cast_or_null<ExprWithCleanups>(E)) + E = EWC->getSubExpr(); + if (CXXConstructExpr *CE = dyn_cast_or_null<CXXConstructExpr>(E)) { NamedDecl *CtorD = dyn_cast_or_null<NamedDecl>(CE->getConstructor()); if (!CtorD || !CtorD->hasAttrs()) @@ -1401,6 +1963,7 @@ void BuildLockset::VisitDeclStmt(DeclStmt *S) { } + /// \brief Compute the intersection of two locksets and issue warnings for any /// locks in the symmetric difference. /// @@ -1409,58 +1972,80 @@ void BuildLockset::VisitDeclStmt(DeclStmt *S) { /// A; if () then B; else C; D; we need to check that the lockset after B and C /// are the same. In the event of a difference, we use the intersection of these /// two locksets at the start of D. -Lockset ThreadSafetyAnalyzer::intersectAndWarn(const CFGBlockInfo &Block1, - CFGBlockSide Side1, - const CFGBlockInfo &Block2, - CFGBlockSide Side2, - LockErrorKind LEK) { - Lockset LSet1 = Block1.getSet(Side1); - Lockset LSet2 = Block2.getSet(Side2); - - Lockset Intersection = LSet1; - for (Lockset::iterator I = LSet2.begin(), E = LSet2.end(); I != E; ++I) { - const MutexID &LSet2Mutex = I.getKey(); - const LockData &LSet2LockData = I.getData(); - if (const LockData *LD = LSet1.lookup(LSet2Mutex)) { - if (LD->LKind != LSet2LockData.LKind) { - Handler.handleExclusiveAndShared(LSet2Mutex.getName(), - LSet2LockData.AcquireLoc, - LD->AcquireLoc); - if (LD->LKind != LK_Exclusive) - Intersection = LocksetFactory.add(Intersection, LSet2Mutex, - LSet2LockData); +/// +/// \param LSet1 The first lockset. +/// \param LSet2 The second lockset. +/// \param JoinLoc The location of the join point for error reporting +/// \param LEK1 The error message to report if a mutex is missing from LSet1 +/// \param LEK2 The error message to report if a mutex is missing from Lset2 +void ThreadSafetyAnalyzer::intersectAndWarn(FactSet &FSet1, + const FactSet &FSet2, + SourceLocation JoinLoc, + LockErrorKind LEK1, + LockErrorKind LEK2, + bool Modify) { + FactSet FSet1Orig = FSet1; + + for (FactSet::const_iterator I = FSet2.begin(), E = FSet2.end(); + I != E; ++I) { + const SExpr &FSet2Mutex = FactMan[*I].MutID; + const LockData &LDat2 = FactMan[*I].LDat; + + if (const LockData *LDat1 = FSet1.findLock(FactMan, FSet2Mutex)) { + if (LDat1->LKind != LDat2.LKind) { + Handler.handleExclusiveAndShared(FSet2Mutex.toString(), + LDat2.AcquireLoc, + LDat1->AcquireLoc); + if (Modify && LDat1->LKind != LK_Exclusive) { + FSet1.removeLock(FactMan, FSet2Mutex); + FSet1.addLock(FactMan, FSet2Mutex, LDat2); + } } } else { - Handler.handleMutexHeldEndOfScope(LSet2Mutex.getName(), - LSet2LockData.AcquireLoc, - Block1.getLocation(Side1), LEK); + if (LDat2.UnderlyingMutex.isValid()) { + if (FSet2.findLock(FactMan, LDat2.UnderlyingMutex)) { + // If this is a scoped lock that manages another mutex, and if the + // underlying mutex is still held, then warn about the underlying + // mutex. + Handler.handleMutexHeldEndOfScope(LDat2.UnderlyingMutex.toString(), + LDat2.AcquireLoc, + JoinLoc, LEK1); + } + } + else if (!LDat2.Managed) + Handler.handleMutexHeldEndOfScope(FSet2Mutex.toString(), + LDat2.AcquireLoc, + JoinLoc, LEK1); } } - for (Lockset::iterator I = LSet1.begin(), E = LSet1.end(); I != E; ++I) { - if (!LSet2.contains(I.getKey())) { - const MutexID &Mutex = I.getKey(); - const LockData &MissingLock = I.getData(); - Handler.handleMutexHeldEndOfScope(Mutex.getName(), - MissingLock.AcquireLoc, - Block2.getLocation(Side2), LEK); - Intersection = LocksetFactory.remove(Intersection, Mutex); + for (FactSet::const_iterator I = FSet1.begin(), E = FSet1.end(); + I != E; ++I) { + const SExpr &FSet1Mutex = FactMan[*I].MutID; + const LockData &LDat1 = FactMan[*I].LDat; + + if (!FSet2.findLock(FactMan, FSet1Mutex)) { + if (LDat1.UnderlyingMutex.isValid()) { + if (FSet1Orig.findLock(FactMan, LDat1.UnderlyingMutex)) { + // If this is a scoped lock that manages another mutex, and if the + // underlying mutex is still held, then warn about the underlying + // mutex. + Handler.handleMutexHeldEndOfScope(LDat1.UnderlyingMutex.toString(), + LDat1.AcquireLoc, + JoinLoc, LEK1); + } + } + else if (!LDat1.Managed) + Handler.handleMutexHeldEndOfScope(FSet1Mutex.toString(), + LDat1.AcquireLoc, + JoinLoc, LEK2); + if (Modify) + FSet1.removeLock(FactMan, FSet1Mutex); } } - return Intersection; } -Lockset ThreadSafetyAnalyzer::addLock(Lockset &LSet, Expr *MutexExp, - const NamedDecl *D, - LockKind LK, SourceLocation Loc) { - MutexID Mutex(MutexExp, 0, D); - if (!Mutex.isValid()) { - MutexID::warnInvalidLock(Handler, MutexExp, 0, D); - return LSet; - } - LockData NewLock(Loc, LK); - return LocksetFactory.add(LSet, Mutex, NewLock); -} + /// \brief Check a function's CFG for thread-safety violations. /// @@ -1472,6 +2057,8 @@ void ThreadSafetyAnalyzer::runAnalysis(AnalysisDeclContext &AC) { if (!CFGraph) return; const NamedDecl *D = dyn_cast_or_null<NamedDecl>(AC.getDecl()); + // AC.dumpCFG(true); + if (!D) return; // Ignore anonymous functions for now. if (D->getAttr<NoThreadSafetyAnalysisAttr>()) @@ -1485,8 +2072,8 @@ void ThreadSafetyAnalyzer::runAnalysis(AnalysisDeclContext &AC) { if (isa<CXXDestructorDecl>(D)) return; // Don't check inside destructors. - std::vector<CFGBlockInfo> BlockInfo(CFGraph->getNumBlockIDs(), - CFGBlockInfo::getEmptyBlockInfo(LocksetFactory, LocalVarMap)); + BlockInfo.resize(CFGraph->getNumBlockIDs(), + CFGBlockInfo::getEmptyBlockInfo(LocalVarMap)); // We need to explore the CFG via a "topological" ordering. // That way, we will be guaranteed to have information about required @@ -1505,27 +2092,22 @@ void ThreadSafetyAnalyzer::runAnalysis(AnalysisDeclContext &AC) { // FIXME: is there a more intelligent way to check lock/unlock functions? if (!SortedGraph->empty() && D->hasAttrs()) { const CFGBlock *FirstBlock = *SortedGraph->begin(); - Lockset &InitialLockset = BlockInfo[FirstBlock->getBlockID()].EntrySet; + FactSet &InitialLockset = BlockInfo[FirstBlock->getBlockID()].EntrySet; const AttrVec &ArgAttrs = D->getAttrs(); + + MutexIDList ExclusiveLocksToAdd; + MutexIDList SharedLocksToAdd; + + SourceLocation Loc = D->getLocation(); for (unsigned i = 0; i < ArgAttrs.size(); ++i) { Attr *Attr = ArgAttrs[i]; - SourceLocation AttrLoc = Attr->getLocation(); - if (SharedLocksRequiredAttr *SLRAttr - = dyn_cast<SharedLocksRequiredAttr>(Attr)) { - for (SharedLocksRequiredAttr::args_iterator - SLRIter = SLRAttr->args_begin(), - SLREnd = SLRAttr->args_end(); SLRIter != SLREnd; ++SLRIter) - InitialLockset = addLock(InitialLockset, - *SLRIter, D, LK_Shared, - AttrLoc); - } else if (ExclusiveLocksRequiredAttr *ELRAttr - = dyn_cast<ExclusiveLocksRequiredAttr>(Attr)) { - for (ExclusiveLocksRequiredAttr::args_iterator - ELRIter = ELRAttr->args_begin(), - ELREnd = ELRAttr->args_end(); ELRIter != ELREnd; ++ELRIter) - InitialLockset = addLock(InitialLockset, - *ELRIter, D, LK_Exclusive, - AttrLoc); + Loc = Attr->getLocation(); + if (ExclusiveLocksRequiredAttr *A + = dyn_cast<ExclusiveLocksRequiredAttr>(Attr)) { + getMutexIDs(ExclusiveLocksToAdd, A, (Expr*) 0, D); + } else if (SharedLocksRequiredAttr *A + = dyn_cast<SharedLocksRequiredAttr>(Attr)) { + getMutexIDs(SharedLocksToAdd, A, (Expr*) 0, D); } else if (isa<UnlockFunctionAttr>(Attr)) { // Don't try to check unlock functions for now return; @@ -1535,8 +2117,24 @@ void ThreadSafetyAnalyzer::runAnalysis(AnalysisDeclContext &AC) { } else if (isa<SharedLockFunctionAttr>(Attr)) { // Don't try to check lock functions for now return; + } else if (isa<ExclusiveTrylockFunctionAttr>(Attr)) { + // Don't try to check trylock functions for now + return; + } else if (isa<SharedTrylockFunctionAttr>(Attr)) { + // Don't try to check trylock functions for now + return; } } + + // FIXME -- Loc can be wrong here. + for (unsigned i=0,n=ExclusiveLocksToAdd.size(); i<n; ++i) { + addLock(InitialLockset, ExclusiveLocksToAdd[i], + LockData(Loc, LK_Exclusive)); + } + for (unsigned i=0,n=SharedLocksToAdd.size(); i<n; ++i) { + addLock(InitialLockset, SharedLocksToAdd[i], + LockData(Loc, LK_Shared)); + } } for (PostOrderCFGView::iterator I = SortedGraph->begin(), @@ -1587,15 +2185,16 @@ void ThreadSafetyAnalyzer::runAnalysis(AnalysisDeclContext &AC) { int PrevBlockID = (*PI)->getBlockID(); CFGBlockInfo *PrevBlockInfo = &BlockInfo[PrevBlockID]; + FactSet PrevLockset; + getEdgeLockset(PrevLockset, PrevBlockInfo->ExitSet, *PI, CurrBlock); if (!LocksetInitialized) { - CurrBlockInfo->EntrySet = PrevBlockInfo->ExitSet; + CurrBlockInfo->EntrySet = PrevLockset; LocksetInitialized = true; } else { - CurrBlockInfo->EntrySet = - intersectAndWarn(*CurrBlockInfo, CBS_Entry, - *PrevBlockInfo, CBS_Exit, - LEK_LockedSomePredecessors); + intersectAndWarn(CurrBlockInfo->EntrySet, PrevLockset, + CurrBlockInfo->EntryLoc, + LEK_LockedSomePredecessors); } } @@ -1619,23 +2218,20 @@ void ThreadSafetyAnalyzer::runAnalysis(AnalysisDeclContext &AC) { const Stmt *Terminator = PrevBlock->getTerminator(); bool IsLoop = Terminator && isa<ContinueStmt>(Terminator); + FactSet PrevLockset; + getEdgeLockset(PrevLockset, PrevBlockInfo->ExitSet, + PrevBlock, CurrBlock); + // Do not update EntrySet. - intersectAndWarn(*CurrBlockInfo, CBS_Entry, *PrevBlockInfo, CBS_Exit, + intersectAndWarn(CurrBlockInfo->EntrySet, PrevLockset, + PrevBlockInfo->ExitLoc, IsLoop ? LEK_LockedSomeLoopIterations - : LEK_LockedSomePredecessors); + : LEK_LockedSomePredecessors, + false); } } BuildLockset LocksetBuilder(this, *CurrBlockInfo); - CFGBlock::const_pred_iterator PI = CurrBlock->pred_begin(), - PE = CurrBlock->pred_end(); - if (PI != PE) { - // If the predecessor ended in a branch, then process any trylocks. - // FIXME -- check to make sure there's only one predecessor. - if (Stmt *TCE = (*PI)->getTerminatorCondition()) { - LocksetBuilder.handleTrylock(TCE, *PI, CurrBlock); - } - } // Visit all the statements in the basic block. for (CFGBlock::const_iterator BI = CurrBlock->begin(), @@ -1665,7 +2261,7 @@ void ThreadSafetyAnalyzer::runAnalysis(AnalysisDeclContext &AC) { break; } } - CurrBlockInfo->ExitSet = LocksetBuilder.LSet; + CurrBlockInfo->ExitSet = LocksetBuilder.FSet; // For every back edge from CurrBlock (the end of the loop) to another block // (FirstLoopBlock) we need to check that the Lockset of Block is equal to @@ -1679,19 +2275,24 @@ void ThreadSafetyAnalyzer::runAnalysis(AnalysisDeclContext &AC) { continue; CFGBlock *FirstLoopBlock = *SI; - CFGBlockInfo &PreLoop = BlockInfo[FirstLoopBlock->getBlockID()]; - CFGBlockInfo &LoopEnd = BlockInfo[CurrBlockID]; - intersectAndWarn(LoopEnd, CBS_Exit, PreLoop, CBS_Entry, - LEK_LockedSomeLoopIterations); + CFGBlockInfo *PreLoop = &BlockInfo[FirstLoopBlock->getBlockID()]; + CFGBlockInfo *LoopEnd = &BlockInfo[CurrBlockID]; + intersectAndWarn(LoopEnd->ExitSet, PreLoop->EntrySet, + PreLoop->EntryLoc, + LEK_LockedSomeLoopIterations, + false); } } - CFGBlockInfo &Initial = BlockInfo[CFGraph->getEntry().getBlockID()]; - CFGBlockInfo &Final = BlockInfo[CFGraph->getExit().getBlockID()]; + CFGBlockInfo *Initial = &BlockInfo[CFGraph->getEntry().getBlockID()]; + CFGBlockInfo *Final = &BlockInfo[CFGraph->getExit().getBlockID()]; // FIXME: Should we call this function for all blocks which exit the function? - intersectAndWarn(Initial, CBS_Entry, Final, CBS_Exit, - LEK_LockedAtEndOfFunction); + intersectAndWarn(Initial->EntrySet, Final->ExitSet, + Final->ExitLoc, + LEK_LockedAtEndOfFunction, + LEK_NotLockedAtEndOfFunction, + false); } } // end anonymous namespace diff --git a/contrib/llvm/tools/clang/lib/Analysis/UninitializedValues.cpp b/contrib/llvm/tools/clang/lib/Analysis/UninitializedValues.cpp index 1c7e6b6..858be45 100644 --- a/contrib/llvm/tools/clang/lib/Analysis/UninitializedValues.cpp +++ b/contrib/llvm/tools/clang/lib/Analysis/UninitializedValues.cpp @@ -16,6 +16,7 @@ #include "llvm/ADT/SmallVector.h" #include "llvm/ADT/PackedVector.h" #include "llvm/ADT/DenseMap.h" +#include "clang/AST/ASTContext.h" #include "clang/AST/Decl.h" #include "clang/Analysis/CFG.h" #include "clang/Analysis/AnalysisContext.h" @@ -25,6 +26,8 @@ using namespace clang; +#define DEBUG_LOGGING 0 + static bool isTrackedVar(const VarDecl *vd, const DeclContext *dc) { if (vd->isLocalVarDecl() && !vd->hasGlobalStorage() && !vd->isExceptionVariable() && @@ -95,143 +98,79 @@ static bool isAlwaysUninit(const Value v) { namespace { typedef llvm::PackedVector<Value, 2> ValueVector; -typedef std::pair<ValueVector *, ValueVector *> BVPair; class CFGBlockValues { const CFG &cfg; - BVPair *vals; + std::vector<ValueVector*> vals; ValueVector scratch; DeclToIndex declToIndex; - - ValueVector &lazyCreate(ValueVector *&bv); public: CFGBlockValues(const CFG &cfg); ~CFGBlockValues(); - + unsigned getNumEntries() const { return declToIndex.size(); } void computeSetOfDeclarations(const DeclContext &dc); - ValueVector &getValueVector(const CFGBlock *block, - const CFGBlock *dstBlock); - - BVPair &getValueVectors(const CFGBlock *block, bool shouldLazyCreate); + ValueVector &getValueVector(const CFGBlock *block) { + return *vals[block->getBlockID()]; + } + void setAllScratchValues(Value V); void mergeIntoScratch(ValueVector const &source, bool isFirst); bool updateValueVectorWithScratch(const CFGBlock *block); - bool updateValueVectors(const CFGBlock *block, const BVPair &newVals); bool hasNoDeclarations() const { return declToIndex.size() == 0; } void resetScratch(); - ValueVector &getScratch() { return scratch; } ValueVector::reference operator[](const VarDecl *vd); + + Value getValue(const CFGBlock *block, const CFGBlock *dstBlock, + const VarDecl *vd) { + const llvm::Optional<unsigned> &idx = declToIndex.getValueIndex(vd); + assert(idx.hasValue()); + return getValueVector(block)[idx.getValue()]; + } }; } // end anonymous namespace -CFGBlockValues::CFGBlockValues(const CFG &c) : cfg(c), vals(0) { - unsigned n = cfg.getNumBlockIDs(); - if (!n) - return; - vals = new std::pair<ValueVector*, ValueVector*>[n]; - memset((void*)vals, 0, sizeof(*vals) * n); -} +CFGBlockValues::CFGBlockValues(const CFG &c) : cfg(c), vals(0) {} CFGBlockValues::~CFGBlockValues() { - unsigned n = cfg.getNumBlockIDs(); - if (n == 0) - return; - for (unsigned i = 0; i < n; ++i) { - delete vals[i].first; - delete vals[i].second; - } - delete [] vals; + for (std::vector<ValueVector*>::iterator I = vals.begin(), E = vals.end(); + I != E; ++I) + delete *I; } void CFGBlockValues::computeSetOfDeclarations(const DeclContext &dc) { declToIndex.computeMap(dc); - scratch.resize(declToIndex.size()); -} - -ValueVector &CFGBlockValues::lazyCreate(ValueVector *&bv) { - if (!bv) - bv = new ValueVector(declToIndex.size()); - return *bv; -} - -/// This function pattern matches for a '&&' or '||' that appears at -/// the beginning of a CFGBlock that also (1) has a terminator and -/// (2) has no other elements. If such an expression is found, it is returned. -static const BinaryOperator *getLogicalOperatorInChain(const CFGBlock *block) { - if (block->empty()) - return 0; - - CFGElement front = block->front(); - const CFGStmt *cstmt = front.getAs<CFGStmt>(); - if (!cstmt) - return 0; - - const BinaryOperator *b = dyn_cast_or_null<BinaryOperator>(cstmt->getStmt()); - - if (!b || !b->isLogicalOp()) - return 0; - - if (block->pred_size() == 2) { - if (block->getTerminatorCondition() == b) { - if (block->succ_size() == 2) - return b; - } - else if (block->size() == 1) - return b; - } - - return 0; -} - -ValueVector &CFGBlockValues::getValueVector(const CFGBlock *block, - const CFGBlock *dstBlock) { - unsigned idx = block->getBlockID(); - if (dstBlock && getLogicalOperatorInChain(block)) { - if (*block->succ_begin() == dstBlock) - return lazyCreate(vals[idx].first); - assert(*(block->succ_begin()+1) == dstBlock); - return lazyCreate(vals[idx].second); - } - - assert(vals[idx].second == 0); - return lazyCreate(vals[idx].first); -} - -BVPair &CFGBlockValues::getValueVectors(const clang::CFGBlock *block, - bool shouldLazyCreate) { - unsigned idx = block->getBlockID(); - lazyCreate(vals[idx].first); - if (shouldLazyCreate) - lazyCreate(vals[idx].second); - return vals[idx]; + unsigned decls = declToIndex.size(); + scratch.resize(decls); + unsigned n = cfg.getNumBlockIDs(); + if (!n) + return; + vals.resize(n); + for (unsigned i = 0; i < n; ++i) + vals[i] = new ValueVector(decls); } -#if 0 +#if DEBUG_LOGGING static void printVector(const CFGBlock *block, ValueVector &bv, unsigned num) { - llvm::errs() << block->getBlockID() << " :"; for (unsigned i = 0; i < bv.size(); ++i) { llvm::errs() << ' ' << bv[i]; } llvm::errs() << " : " << num << '\n'; } +#endif -static void printVector(const char *name, ValueVector const &bv) { - llvm::errs() << name << " : "; - for (unsigned i = 0; i < bv.size(); ++i) { - llvm::errs() << ' ' << bv[i]; - } - llvm::errs() << "\n"; +void CFGBlockValues::setAllScratchValues(Value V) { + for (unsigned I = 0, E = scratch.size(); I != E; ++I) + scratch[I] = V; } -#endif void CFGBlockValues::mergeIntoScratch(ValueVector const &source, bool isFirst) { @@ -242,30 +181,16 @@ void CFGBlockValues::mergeIntoScratch(ValueVector const &source, } bool CFGBlockValues::updateValueVectorWithScratch(const CFGBlock *block) { - ValueVector &dst = getValueVector(block, 0); + ValueVector &dst = getValueVector(block); bool changed = (dst != scratch); if (changed) dst = scratch; -#if 0 +#if DEBUG_LOGGING printVector(block, scratch, 0); #endif return changed; } -bool CFGBlockValues::updateValueVectors(const CFGBlock *block, - const BVPair &newVals) { - BVPair &vals = getValueVectors(block, true); - bool changed = *newVals.first != *vals.first || - *newVals.second != *vals.second; - *vals.first = *newVals.first; - *vals.second = *newVals.second; -#if 0 - printVector(block, *vals.first, 1); - printVector(block, *vals.second, 2); -#endif - return changed; -} - void CFGBlockValues::resetScratch() { scratch.reset(); } @@ -321,7 +246,7 @@ const CFGBlock *DataflowWorklist::dequeue() { } //------------------------------------------------------------------------====// -// Transfer function for uninitialized values analysis. +// Classification of DeclRefExprs as use or initialization. //====------------------------------------------------------------------------// namespace { @@ -329,106 +254,339 @@ class FindVarResult { const VarDecl *vd; const DeclRefExpr *dr; public: - FindVarResult(VarDecl *vd, DeclRefExpr *dr) : vd(vd), dr(dr) {} - + FindVarResult(const VarDecl *vd, const DeclRefExpr *dr) : vd(vd), dr(dr) {} + const DeclRefExpr *getDeclRefExpr() const { return dr; } const VarDecl *getDecl() const { return vd; } }; - + +static const Expr *stripCasts(ASTContext &C, const Expr *Ex) { + while (Ex) { + Ex = Ex->IgnoreParenNoopCasts(C); + if (const CastExpr *CE = dyn_cast<CastExpr>(Ex)) { + if (CE->getCastKind() == CK_LValueBitCast) { + Ex = CE->getSubExpr(); + continue; + } + } + break; + } + return Ex; +} + +/// If E is an expression comprising a reference to a single variable, find that +/// variable. +static FindVarResult findVar(const Expr *E, const DeclContext *DC) { + if (const DeclRefExpr *DRE = + dyn_cast<DeclRefExpr>(stripCasts(DC->getParentASTContext(), E))) + if (const VarDecl *VD = dyn_cast<VarDecl>(DRE->getDecl())) + if (isTrackedVar(VD, DC)) + return FindVarResult(VD, DRE); + return FindVarResult(0, 0); +} + +/// \brief Classify each DeclRefExpr as an initialization or a use. Any +/// DeclRefExpr which isn't explicitly classified will be assumed to have +/// escaped the analysis and will be treated as an initialization. +class ClassifyRefs : public StmtVisitor<ClassifyRefs> { +public: + enum Class { + Init, + Use, + SelfInit, + Ignore + }; + +private: + const DeclContext *DC; + llvm::DenseMap<const DeclRefExpr*, Class> Classification; + + bool isTrackedVar(const VarDecl *VD) const { + return ::isTrackedVar(VD, DC); + } + + void classify(const Expr *E, Class C); + +public: + ClassifyRefs(AnalysisDeclContext &AC) : DC(cast<DeclContext>(AC.getDecl())) {} + + void VisitDeclStmt(DeclStmt *DS); + void VisitUnaryOperator(UnaryOperator *UO); + void VisitBinaryOperator(BinaryOperator *BO); + void VisitCallExpr(CallExpr *CE); + void VisitCastExpr(CastExpr *CE); + + void operator()(Stmt *S) { Visit(S); } + + Class get(const DeclRefExpr *DRE) const { + llvm::DenseMap<const DeclRefExpr*, Class>::const_iterator I + = Classification.find(DRE); + if (I != Classification.end()) + return I->second; + + const VarDecl *VD = dyn_cast<VarDecl>(DRE->getDecl()); + if (!VD || !isTrackedVar(VD)) + return Ignore; + + return Init; + } +}; +} + +static const DeclRefExpr *getSelfInitExpr(VarDecl *VD) { + if (Expr *Init = VD->getInit()) { + const DeclRefExpr *DRE + = dyn_cast<DeclRefExpr>(stripCasts(VD->getASTContext(), Init)); + if (DRE && DRE->getDecl() == VD) + return DRE; + } + return 0; +} + +void ClassifyRefs::classify(const Expr *E, Class C) { + FindVarResult Var = findVar(E, DC); + if (const DeclRefExpr *DRE = Var.getDeclRefExpr()) + Classification[DRE] = std::max(Classification[DRE], C); +} + +void ClassifyRefs::VisitDeclStmt(DeclStmt *DS) { + for (DeclStmt::decl_iterator DI = DS->decl_begin(), DE = DS->decl_end(); + DI != DE; ++DI) { + VarDecl *VD = dyn_cast<VarDecl>(*DI); + if (VD && isTrackedVar(VD)) + if (const DeclRefExpr *DRE = getSelfInitExpr(VD)) + Classification[DRE] = SelfInit; + } +} + +void ClassifyRefs::VisitBinaryOperator(BinaryOperator *BO) { + // Ignore the evaluation of a DeclRefExpr on the LHS of an assignment. If this + // is not a compound-assignment, we will treat it as initializing the variable + // when TransferFunctions visits it. A compound-assignment does not affect + // whether a variable is uninitialized, and there's no point counting it as a + // use. + if (BO->isCompoundAssignmentOp()) + classify(BO->getLHS(), Use); + else if (BO->getOpcode() == BO_Assign) + classify(BO->getLHS(), Ignore); +} + +void ClassifyRefs::VisitUnaryOperator(UnaryOperator *UO) { + // Increment and decrement are uses despite there being no lvalue-to-rvalue + // conversion. + if (UO->isIncrementDecrementOp()) + classify(UO->getSubExpr(), Use); +} + +void ClassifyRefs::VisitCallExpr(CallExpr *CE) { + // If a value is passed by const reference to a function, we should not assume + // that it is initialized by the call, and we conservatively do not assume + // that it is used. + for (CallExpr::arg_iterator I = CE->arg_begin(), E = CE->arg_end(); + I != E; ++I) + if ((*I)->getType().isConstQualified() && (*I)->isGLValue()) + classify(*I, Ignore); +} + +void ClassifyRefs::VisitCastExpr(CastExpr *CE) { + if (CE->getCastKind() == CK_LValueToRValue) + classify(CE->getSubExpr(), Use); + else if (CStyleCastExpr *CSE = dyn_cast<CStyleCastExpr>(CE)) { + if (CSE->getType()->isVoidType()) { + // Squelch any detected load of an uninitialized value if + // we cast it to void. + // e.g. (void) x; + classify(CSE->getSubExpr(), Ignore); + } + } +} + +//------------------------------------------------------------------------====// +// Transfer function for uninitialized values analysis. +//====------------------------------------------------------------------------// + +namespace { class TransferFunctions : public StmtVisitor<TransferFunctions> { CFGBlockValues &vals; const CFG &cfg; + const CFGBlock *block; AnalysisDeclContext ∾ + const ClassifyRefs &classification; UninitVariablesHandler *handler; - - /// The last DeclRefExpr seen when analyzing a block. Used to - /// cheat when detecting cases when the address of a variable is taken. - DeclRefExpr *lastDR; - - /// The last lvalue-to-rvalue conversion of a variable whose value - /// was uninitialized. Normally this results in a warning, but it is - /// possible to either silence the warning in some cases, or we - /// propagate the uninitialized value. - CastExpr *lastLoad; - - /// For some expressions, we want to ignore any post-processing after - /// visitation. - bool skipProcessUses; - + public: TransferFunctions(CFGBlockValues &vals, const CFG &cfg, - AnalysisDeclContext &ac, + const CFGBlock *block, AnalysisDeclContext &ac, + const ClassifyRefs &classification, UninitVariablesHandler *handler) - : vals(vals), cfg(cfg), ac(ac), handler(handler), - lastDR(0), lastLoad(0), - skipProcessUses(false) {} - - void reportUninit(const DeclRefExpr *ex, const VarDecl *vd, - bool isAlwaysUninit); + : vals(vals), cfg(cfg), block(block), ac(ac), + classification(classification), handler(handler) {} + void reportUse(const Expr *ex, const VarDecl *vd); + + void VisitObjCForCollectionStmt(ObjCForCollectionStmt *FS); void VisitBlockExpr(BlockExpr *be); + void VisitCallExpr(CallExpr *ce); void VisitDeclStmt(DeclStmt *ds); void VisitDeclRefExpr(DeclRefExpr *dr); - void VisitUnaryOperator(UnaryOperator *uo); void VisitBinaryOperator(BinaryOperator *bo); - void VisitCastExpr(CastExpr *ce); - void VisitObjCForCollectionStmt(ObjCForCollectionStmt *fs); - void Visit(Stmt *s); - + bool isTrackedVar(const VarDecl *vd) { return ::isTrackedVar(vd, cast<DeclContext>(ac.getDecl())); } - - FindVarResult findBlockVarDecl(Expr *ex); - - void ProcessUses(Stmt *s = 0); -}; -} -static const Expr *stripCasts(ASTContext &C, const Expr *Ex) { - while (Ex) { - Ex = Ex->IgnoreParenNoopCasts(C); - if (const CastExpr *CE = dyn_cast<CastExpr>(Ex)) { - if (CE->getCastKind() == CK_LValueBitCast) { - Ex = CE->getSubExpr(); - continue; + FindVarResult findVar(const Expr *ex) { + return ::findVar(ex, cast<DeclContext>(ac.getDecl())); + } + + UninitUse getUninitUse(const Expr *ex, const VarDecl *vd, Value v) { + UninitUse Use(ex, isAlwaysUninit(v)); + + assert(isUninitialized(v)); + if (Use.getKind() == UninitUse::Always) + return Use; + + // If an edge which leads unconditionally to this use did not initialize + // the variable, we can say something stronger than 'may be uninitialized': + // we can say 'either it's used uninitialized or you have dead code'. + // + // We track the number of successors of a node which have been visited, and + // visit a node once we have visited all of its successors. Only edges where + // the variable might still be uninitialized are followed. Since a variable + // can't transfer from being initialized to being uninitialized, this will + // trace out the subgraph which inevitably leads to the use and does not + // initialize the variable. We do not want to skip past loops, since their + // non-termination might be correlated with the initialization condition. + // + // For example: + // + // void f(bool a, bool b) { + // block1: int n; + // if (a) { + // block2: if (b) + // block3: n = 1; + // block4: } else if (b) { + // block5: while (!a) { + // block6: do_work(&a); + // n = 2; + // } + // } + // block7: if (a) + // block8: g(); + // block9: return n; + // } + // + // Starting from the maybe-uninitialized use in block 9: + // * Block 7 is not visited because we have only visited one of its two + // successors. + // * Block 8 is visited because we've visited its only successor. + // From block 8: + // * Block 7 is visited because we've now visited both of its successors. + // From block 7: + // * Blocks 1, 2, 4, 5, and 6 are not visited because we didn't visit all + // of their successors (we didn't visit 4, 3, 5, 6, and 5, respectively). + // * Block 3 is not visited because it initializes 'n'. + // Now the algorithm terminates, having visited blocks 7 and 8, and having + // found the frontier is blocks 2, 4, and 5. + // + // 'n' is definitely uninitialized for two edges into block 7 (from blocks 2 + // and 4), so we report that any time either of those edges is taken (in + // each case when 'b == false'), 'n' is used uninitialized. + llvm::SmallVector<const CFGBlock*, 32> Queue; + llvm::SmallVector<unsigned, 32> SuccsVisited(cfg.getNumBlockIDs(), 0); + Queue.push_back(block); + // Specify that we've already visited all successors of the starting block. + // This has the dual purpose of ensuring we never add it to the queue, and + // of marking it as not being a candidate element of the frontier. + SuccsVisited[block->getBlockID()] = block->succ_size(); + while (!Queue.empty()) { + const CFGBlock *B = Queue.back(); + Queue.pop_back(); + for (CFGBlock::const_pred_iterator I = B->pred_begin(), E = B->pred_end(); + I != E; ++I) { + const CFGBlock *Pred = *I; + if (vals.getValue(Pred, B, vd) == Initialized) + // This block initializes the variable. + continue; + + unsigned &SV = SuccsVisited[Pred->getBlockID()]; + if (!SV) { + // When visiting the first successor of a block, mark all NULL + // successors as having been visited. + for (CFGBlock::const_succ_iterator SI = Pred->succ_begin(), + SE = Pred->succ_end(); + SI != SE; ++SI) + if (!*SI) + ++SV; + } + + if (++SV == Pred->succ_size()) + // All paths from this block lead to the use and don't initialize the + // variable. + Queue.push_back(Pred); + } + } + + // Scan the frontier, looking for blocks where the variable was + // uninitialized. + for (CFG::const_iterator BI = cfg.begin(), BE = cfg.end(); BI != BE; ++BI) { + const CFGBlock *Block = *BI; + unsigned BlockID = Block->getBlockID(); + const Stmt *Term = Block->getTerminator(); + if (SuccsVisited[BlockID] && SuccsVisited[BlockID] < Block->succ_size() && + Term) { + // This block inevitably leads to the use. If we have an edge from here + // to a post-dominator block, and the variable is uninitialized on that + // edge, we have found a bug. + for (CFGBlock::const_succ_iterator I = Block->succ_begin(), + E = Block->succ_end(); I != E; ++I) { + const CFGBlock *Succ = *I; + if (Succ && SuccsVisited[Succ->getBlockID()] >= Succ->succ_size() && + vals.getValue(Block, Succ, vd) == Uninitialized) { + // Switch cases are a special case: report the label to the caller + // as the 'terminator', not the switch statement itself. Suppress + // situations where no label matched: we can't be sure that's + // possible. + if (isa<SwitchStmt>(Term)) { + const Stmt *Label = Succ->getLabel(); + if (!Label || !isa<SwitchCase>(Label)) + // Might not be possible. + continue; + UninitUse::Branch Branch; + Branch.Terminator = Label; + Branch.Output = 0; // Ignored. + Use.addUninitBranch(Branch); + } else { + UninitUse::Branch Branch; + Branch.Terminator = Term; + Branch.Output = I - Block->succ_begin(); + Use.addUninitBranch(Branch); + } + } + } } } - break; - } - return Ex; -} -void TransferFunctions::reportUninit(const DeclRefExpr *ex, - const VarDecl *vd, bool isAlwaysUnit) { - if (handler) handler->handleUseOfUninitVariable(ex, vd, isAlwaysUnit); + return Use; + } +}; } -FindVarResult TransferFunctions::findBlockVarDecl(Expr *ex) { - if (DeclRefExpr *dr = dyn_cast<DeclRefExpr>(ex->IgnoreParenCasts())) - if (VarDecl *vd = dyn_cast<VarDecl>(dr->getDecl())) - if (isTrackedVar(vd)) - return FindVarResult(vd, dr); - return FindVarResult(0, 0); +void TransferFunctions::reportUse(const Expr *ex, const VarDecl *vd) { + if (!handler) + return; + Value v = vals[vd]; + if (isUninitialized(v)) + handler->handleUseOfUninitVariable(vd, getUninitUse(ex, vd, v)); } -void TransferFunctions::VisitObjCForCollectionStmt(ObjCForCollectionStmt *fs) { +void TransferFunctions::VisitObjCForCollectionStmt(ObjCForCollectionStmt *FS) { // This represents an initialization of the 'element' value. - Stmt *element = fs->getElement(); - const VarDecl *vd = 0; - - if (DeclStmt *ds = dyn_cast<DeclStmt>(element)) { - vd = cast<VarDecl>(ds->getSingleDecl()); - if (!isTrackedVar(vd)) - vd = 0; - } else { - // Initialize the value of the reference variable. - const FindVarResult &res = findBlockVarDecl(cast<Expr>(element)); - vd = res.getDecl(); + if (DeclStmt *DS = dyn_cast<DeclStmt>(FS->getElement())) { + const VarDecl *VD = cast<VarDecl>(DS->getSingleDecl()); + if (isTrackedVar(VD)) + vals[VD] = Initialized; } - - if (vd) - vals[vd] = Initialized; } void TransferFunctions::VisitBlockExpr(BlockExpr *be) { @@ -442,231 +600,112 @@ void TransferFunctions::VisitBlockExpr(BlockExpr *be) { vals[vd] = Initialized; continue; } - Value v = vals[vd]; - if (handler && isUninitialized(v)) - handler->handleUseOfUninitVariable(be, vd, isAlwaysUninit(v)); + reportUse(be, vd); } } -void TransferFunctions::VisitDeclRefExpr(DeclRefExpr *dr) { - // Record the last DeclRefExpr seen. This is an lvalue computation. - // We use this value to later detect if a variable "escapes" the analysis. - if (const VarDecl *vd = dyn_cast<VarDecl>(dr->getDecl())) - if (isTrackedVar(vd)) { - ProcessUses(); - lastDR = dr; - } -} - -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 (isTrackedVar(vd)) { - if (Expr *init = vd->getInit()) { - // If the initializer consists solely of a reference to itself, we - // explicitly mark the variable as uninitialized. This allows code - // like the following: - // - // int x = x; - // - // to deliberately leave a variable uninitialized. Different analysis - // clients can detect this pattern and adjust their reporting - // appropriately, but we need to continue to analyze subsequent uses - // of the variable. - if (init == lastLoad) { - const DeclRefExpr *DR - = cast<DeclRefExpr>(stripCasts(ac.getASTContext(), - lastLoad->getSubExpr())); - if (DR->getDecl() == vd) { - // int x = x; - // Propagate uninitialized value, but don't immediately report - // a problem. - vals[vd] = Uninitialized; - lastLoad = 0; - lastDR = 0; - if (handler) - handler->handleSelfInit(vd); - return; - } - } - - // All other cases: treat the new variable as initialized. - // This is a minor optimization to reduce the propagation - // of the analysis, since we will have already reported - // the use of the uninitialized value (which visiting the - // initializer). - vals[vd] = Initialized; - } - } - } - } +void TransferFunctions::VisitCallExpr(CallExpr *ce) { + // After a call to a function like setjmp or vfork, any variable which is + // initialized anywhere within this function may now be initialized. For now, + // just assume such a call initializes all variables. + // FIXME: Only mark variables as initialized if they have an initializer which + // is reachable from here. + Decl *Callee = ce->getCalleeDecl(); + if (Callee && Callee->hasAttr<ReturnsTwiceAttr>()) + vals.setAllScratchValues(Initialized); } -void TransferFunctions::VisitBinaryOperator(clang::BinaryOperator *bo) { - if (bo->isAssignmentOp()) { - const FindVarResult &res = findBlockVarDecl(bo->getLHS()); - if (const VarDecl *vd = res.getDecl()) { - ValueVector::reference val = vals[vd]; - if (isUninitialized(val)) { - if (bo->getOpcode() != BO_Assign) - reportUninit(res.getDeclRefExpr(), vd, isAlwaysUninit(val)); - else - val = Initialized; - } - } +void TransferFunctions::VisitDeclRefExpr(DeclRefExpr *dr) { + switch (classification.get(dr)) { + case ClassifyRefs::Ignore: + break; + case ClassifyRefs::Use: + reportUse(dr, cast<VarDecl>(dr->getDecl())); + break; + case ClassifyRefs::Init: + vals[cast<VarDecl>(dr->getDecl())] = Initialized; + break; + case ClassifyRefs::SelfInit: + if (handler) + handler->handleSelfInit(cast<VarDecl>(dr->getDecl())); + break; } } -void TransferFunctions::VisitUnaryOperator(clang::UnaryOperator *uo) { - switch (uo->getOpcode()) { - 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()) { - assert(res.getDeclRefExpr() == lastDR); - // We null out lastDR to indicate we have fully processed it - // and we don't want the auto-value setting in Visit(). - lastDR = 0; - - ValueVector::reference val = vals[vd]; - if (isUninitialized(val)) - reportUninit(res.getDeclRefExpr(), vd, isAlwaysUninit(val)); - } - break; - } - default: - break; +void TransferFunctions::VisitBinaryOperator(BinaryOperator *BO) { + if (BO->getOpcode() == BO_Assign) { + FindVarResult Var = findVar(BO->getLHS()); + if (const VarDecl *VD = Var.getDecl()) + vals[VD] = Initialized; } } -void TransferFunctions::VisitCastExpr(clang::CastExpr *ce) { - if (ce->getCastKind() == CK_LValueToRValue) { - const FindVarResult &res = findBlockVarDecl(ce->getSubExpr()); - if (res.getDecl()) { - assert(res.getDeclRefExpr() == lastDR); - lastLoad = ce; - } - } - else if (ce->getCastKind() == CK_NoOp || - ce->getCastKind() == CK_LValueBitCast) { - skipProcessUses = true; - } - else if (CStyleCastExpr *cse = dyn_cast<CStyleCastExpr>(ce)) { - if (cse->getType()->isVoidType()) { - // e.g. (void) x; - if (lastLoad == cse->getSubExpr()) { - // Squelch any detected load of an uninitialized value if - // we cast it to void. - lastLoad = 0; - lastDR = 0; +void TransferFunctions::VisitDeclStmt(DeclStmt *DS) { + for (DeclStmt::decl_iterator DI = DS->decl_begin(), DE = DS->decl_end(); + DI != DE; ++DI) { + VarDecl *VD = dyn_cast<VarDecl>(*DI); + if (VD && isTrackedVar(VD)) { + if (getSelfInitExpr(VD)) { + // If the initializer consists solely of a reference to itself, we + // explicitly mark the variable as uninitialized. This allows code + // like the following: + // + // int x = x; + // + // to deliberately leave a variable uninitialized. Different analysis + // clients can detect this pattern and adjust their reporting + // appropriately, but we need to continue to analyze subsequent uses + // of the variable. + vals[VD] = Uninitialized; + } else if (VD->getInit()) { + // Treat the new variable as initialized. + vals[VD] = Initialized; + } else { + // No initializer: the variable is now uninitialized. This matters + // for cases like: + // while (...) { + // int n; + // use(n); + // n = 0; + // } + // FIXME: Mark the variable as uninitialized whenever its scope is + // left, since its scope could be re-entered by a jump over the + // declaration. + vals[VD] = Uninitialized; } } } } -void TransferFunctions::Visit(clang::Stmt *s) { - skipProcessUses = false; - StmtVisitor<TransferFunctions>::Visit(s); - if (!skipProcessUses) - ProcessUses(s); -} - -void TransferFunctions::ProcessUses(Stmt *s) { - // This method is typically called after visiting a CFGElement statement - // in the CFG. We delay processing of reporting many loads of uninitialized - // values until here. - if (lastLoad) { - // If we just visited the lvalue-to-rvalue cast, there is nothing - // left to do. - if (lastLoad == s) - return; - - const DeclRefExpr *DR = - cast<DeclRefExpr>(stripCasts(ac.getASTContext(), - lastLoad->getSubExpr())); - const VarDecl *VD = cast<VarDecl>(DR->getDecl()); - - // If we reach here, we may have seen a load of an uninitialized value - // and it hasn't been casted to void or otherwise handled. In this - // situation, report the incident. - if (isUninitialized(vals[VD])) - reportUninit(DR, VD, isAlwaysUninit(vals[VD])); - - lastLoad = 0; - - if (DR == lastDR) { - lastDR = 0; - return; - } - } - - // Any other uses of 'lastDR' involve taking an lvalue of variable. - // In this case, it "escapes" the analysis. - if (lastDR && lastDR != s) { - vals[cast<VarDecl>(lastDR->getDecl())] = Initialized; - lastDR = 0; - } -} - //------------------------------------------------------------------------====// // High-level "driver" logic for uninitialized values analysis. //====------------------------------------------------------------------------// static bool runOnBlock(const CFGBlock *block, const CFG &cfg, AnalysisDeclContext &ac, CFGBlockValues &vals, + const ClassifyRefs &classification, llvm::BitVector &wasAnalyzed, UninitVariablesHandler *handler = 0) { - wasAnalyzed[block->getBlockID()] = true; - - if (const BinaryOperator *b = getLogicalOperatorInChain(block)) { - CFGBlock::const_pred_iterator itr = block->pred_begin(); - BVPair vA = vals.getValueVectors(*itr, false); - ++itr; - BVPair vB = vals.getValueVectors(*itr, false); - - BVPair valsAB; - - if (b->getOpcode() == BO_LAnd) { - // Merge the 'F' bits from the first and second. - vals.mergeIntoScratch(*(vA.second ? vA.second : vA.first), true); - vals.mergeIntoScratch(*(vB.second ? vB.second : vB.first), false); - valsAB.first = vA.first; - valsAB.second = &vals.getScratch(); - } else { - // Merge the 'T' bits from the first and second. - assert(b->getOpcode() == BO_LOr); - vals.mergeIntoScratch(*vA.first, true); - vals.mergeIntoScratch(*vB.first, false); - valsAB.first = &vals.getScratch(); - valsAB.second = vA.second ? vA.second : vA.first; - } - return vals.updateValueVectors(block, valsAB); - } - - // Default behavior: merge in values of predecessor blocks. vals.resetScratch(); + // Merge in values of predecessor blocks. bool isFirst = true; for (CFGBlock::const_pred_iterator I = block->pred_begin(), E = block->pred_end(); I != E; ++I) { const CFGBlock *pred = *I; if (wasAnalyzed[pred->getBlockID()]) { - vals.mergeIntoScratch(vals.getValueVector(pred, block), isFirst); + vals.mergeIntoScratch(vals.getValueVector(pred), isFirst); isFirst = false; } } // Apply the transfer function. - TransferFunctions tf(vals, cfg, ac, handler); + TransferFunctions tf(vals, cfg, block, ac, classification, handler); for (CFGBlock::const_iterator I = block->begin(), E = block->end(); I != E; ++I) { if (const CFGStmt *cs = dyn_cast<CFGStmt>(&*I)) { tf.Visit(const_cast<Stmt*>(cs->getStmt())); } } - tf.ProcessUses(); return vals.updateValueVectorWithScratch(block); } @@ -683,17 +722,16 @@ void clang::runUninitializedVariablesAnalysis( stats.NumVariablesAnalyzed = vals.getNumEntries(); + // Precompute which expressions are uses and which are initializations. + ClassifyRefs classification(ac); + cfg.VisitBlockStmts(classification); + // Mark all variables uninitialized at the entry. const CFGBlock &entry = cfg.getEntry(); - for (CFGBlock::const_succ_iterator i = entry.succ_begin(), - e = entry.succ_end(); i != e; ++i) { - if (const CFGBlock *succ = *i) { - ValueVector &vec = vals.getValueVector(&entry, succ); - const unsigned n = vals.getNumEntries(); - for (unsigned j = 0; j < n ; ++j) { - vec[j] = Uninitialized; - } - } + ValueVector &vec = vals.getValueVector(&entry); + const unsigned n = vals.getNumEntries(); + for (unsigned j = 0; j < n ; ++j) { + vec[j] = Uninitialized; } // Proceed with the workist. @@ -705,7 +743,8 @@ void clang::runUninitializedVariablesAnalysis( while (const CFGBlock *block = worklist.dequeue()) { // Did the block change? - bool changed = runOnBlock(block, cfg, ac, vals, wasAnalyzed); + bool changed = runOnBlock(block, cfg, ac, vals, + classification, wasAnalyzed); ++stats.NumBlockVisits; if (changed || !previouslyVisited[block->getBlockID()]) worklist.enqueueSuccessors(block); @@ -716,7 +755,7 @@ void clang::runUninitializedVariablesAnalysis( for (CFG::const_iterator BI = cfg.begin(), BE = cfg.end(); BI != BE; ++BI) { const CFGBlock *block = *BI; if (wasAnalyzed[block->getBlockID()]) { - runOnBlock(block, cfg, ac, vals, wasAnalyzed, &handler); + runOnBlock(block, cfg, ac, vals, classification, wasAnalyzed, &handler); ++stats.NumBlockVisits; } } |
